# OME-WIDE STUDIES OF GRAPEVINE FRUIT COMPOSITION AND RESPONSES TO AGRO-ENVIRONMENTAL FACTORS IN THE ERA OF SYSTEMS BIOLOGY

EDITED BY : José Tomás Matus, Simone Diego Castellarin and Giovanni Battista Tornielli PUBLISHED IN : Frontiers in Plant Science

#### Frontiers eBook Copyright Statement

The copyright in the text of individual articles in this eBook is the property of their respective authors or their respective institutions or funders. The copyright in graphics and images within each article may be subject to copyright of other parties. In both cases this is subject to a license granted to Frontiers. The compilation of articles constituting this eBook is the property of Frontiers.

Each article within this eBook, and the eBook itself, are published under the most recent version of the Creative Commons CC-BY licence. The version current at the date of publication of this eBook is CC-BY 4.0. If the CC-BY licence is updated, the licence granted by Frontiers is automatically updated to the new version.

When exercising any right under the CC-BY licence, Frontiers must be attributed as the original publisher of the article or eBook, as applicable.

Authors have the responsibility of ensuring that any graphics or other materials which are the property of others may be included in the CC-BY licence, but this should be checked before relying on the CC-BY licence to reproduce those materials. Any copyright notices relating to those materials must be complied with.

Copyright and source acknowledgement notices may not be removed and must be displayed in any copy, derivative work or partial copy which includes the elements in question.

All copyright, and all rights therein, are protected by national and international copyright laws. The above represents a summary only. For further information please read Frontiers' Conditions for Website Use and Copyright Statement, and the applicable CC-BY licence.

ISSN 1664-8714 ISBN 978-2-88963-211-4 DOI 10.3389/978-2-88963-211-4

#### About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

#### Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

#### Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

#### What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

# OME-WIDE STUDIES OF GRAPEVINE FRUIT COMPOSITION AND RESPONSES TO AGRO-ENVIRONMENTAL FACTORS IN THE ERA OF SYSTEMS BIOLOGY

Topic Editors:

José Tomás Matus, Institute for Integrative Systems Biology, Spain Simone Diego Castellarin, University of British Columbia, Canada Giovanni Battista Tornielli, University of Verona, Italy

Network representation of a grape cluster as a group of modules of highly-connected nodes. Image: José Tomás Matus.

Fruits play a substantial role in the human diet as a source of vitamins, minerals, dietary fiber and a wide range of molecules relevant to health promotion and disease prevention. The characterization of genes involved in the accumulation of these molecules during fruit development and ripening, and in the overall plant's response to the environment, constitutes a fundamental step for improving yield- and quality-related traits, and for predicting this crop's behavior in the field. This is certainly the case for grapevine (*Vitis vinifera* L.), one of the most largely cultivated fruit crops in the world. The cultivation of this species is facing challenging scenarios driven by climate change – including increases in atmospheric carbon dioxide (CO2 ), solar radiation, and earth surface temperature, and decreases of water and nutrient availability. All these events will potentially affect the grapevine phenology, physiology, and metabolism in many growing regions and ultimately affect the quality of their fruits and of the most important derived product, the wine.

The sequencing of the grapevine genome has given rise to a new era, characterized by the generation of large-scale data that requires complex computational analyses. Numerous transcriptomic and metabolomic studies have been performed in the past fifteen years, providing insights into the gene circuits that control the accumulation of all sorts of metabolites in grapevines. From now on, the integration of two or more 'omics' will allow depicting gene-transcript-metabolite networks from a more holistic (i.e. systems) perspective.

This eBook attempts to support this new direction, by gathering innovative studies that assess the impact of genotypes, the environment, and agronomical practices on fruits at the 'ome'-scale. The works hereby collected are part of a Research Topic covering the use of 'omics'-driven strategies to understand how environmental factors and agronomical practices – including microclimate modification (e.g. sunlight incidence or temperature), water availability and irrigation, and postharvest management – affect fruit development and composition. These studies report well-settled transcriptomic and metabolomic methods, in addition to newly-developed techniques addressing proteome profiles, genome methylation landscapes and ionomic signatures, some of which attempt to tackle the influence of terroir, i.e. the synergic effect of (micro) climate, soil composition, grape genotype, and vineyard practices. A few reviews and opinions are included that focus on the advantages of applying network theory in grapevine research. Studies on vegetative organs in their relation to fruit development and on fruit-derived cell cultures are also considered.

Citation: Matus, J. T., Castellarin, S. D., Tornielli, G. B., eds. (2019). Ome-wide Studies of Grapevine Fruit Composition and Responses to Agro-environmental Factors in the Era of Systems Biology. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-211-4

# Table of Contents

## SECTION 1

#### INTEGRATED RESPONSES TO FLUCTUATING ENVIRONMENTAL CONDITIONS 1.1 SUNLIGHT AND ULTRAVIOLET RADIATION


Run-Ze Sun, Guo Cheng, Qiang Li, Yan-Nan He, Yu Wang, Yi-Bin Lan, Si-Yu Li, Yan-Rong Zhu, Wen-Feng Song, Xue Zhang, Xiao-Di Cui, Wu Chen and Jun Wang


Stefan Czemmel, Janine Höll, Rodrigo Loyola, Patricio Arce-Johnson, José Antonio Alcalde, José Tomás Matus and Jochen Bogs

#### 1.2 TEMPERATURE


Gianluca Allegro, Gabriele Valentini, Ilaria Filippetti and Giovanni Battista Tornielli

#### 1.3 NUTRITIONAL STATE

*123 Constraint-Based Modeling Highlights Cell Energy, Redox Status and*  a*-Ketoglutarate Availability as Metabolic Drivers for Anthocyanin Accumulation in Grape Cells Under Nitrogen Limitation*

Eric Soubeyrand, Sophie Colombié, Bertrand Beauvoit, Zhanwu Dai, Stéphanie Cluzet, Ghislaine Hilbert, Christel Renaud, Lilly Maneta-Peyret, Martine Dieuaide-Noubhani, Jean-Michel Mérillon, Yves Gibon, Serge Delrot and Eric Gomès

#### SECTION 2

#### EFFECT OF AGRICULTURAL AND POST-HARVEST PRACTICES AND RELATION TO TERROIR

#### *137 Dissecting the Variations of Ripening Progression and Flavonoid Metabolism in Grape Berries Grown Under Double Cropping System* Wei-Kai Chen, Xian-Jin Bai, Mu-Ming Cao, Guo Cheng, Xiong-Jun Cao, Rong-Rong Guo, Yu Wang, Lei He, Xiao-Hui Yang, Fei He, Chang-Qing Duan and Jun Wang

*157 Multi-Omics and Integrated Network Analyses Reveal New Insights Into the Systems Relationships Between Metabolites, Structural Genes, and Transcriptional Regulators in Developing Grape Berries (*Vitis vinifera *L.) Exposed to Water Deficit*

Stefania Savoi, Darren C. J. Wong, Asfaw Degu, Jose C. Herrera, Barbara Bucchetti, Enrico Peterlunger, Aaron Fait, Fulvio Mattivi and Simone D. Castellarin

*176 Transcriptional Responses to Pre-flowering Leaf Defoliation in Grapevine Berry From Different Growing Sites, Years, and Genotypes*

Sara Zenoni, Silvia Dal Santo, Giovanni B. Tornielli, Erica D'Incà, Ilaria Filippetti, Chiara Pastore, Gianluca Allegro, Oriana Silvestroni, Vania Lanari, Antonino Pisciotta, Rosario Di Lorenzo, Alberto Palliotti, Sergio Tombesi, Matteo Gatti and Stefano Poni

*197 Global DNA Methylation Patterns Can Play a Role in Defining Terroir in Grapevine (*Vitis vinifera *cv. Shiraz)*

Huahan Xie, Moumouni Konate, Na Sai, Kiflu G. Tesfamicael, Timothy Cavagnaro, Matthew Gilliham, James Breen, Andrew Metcalfe, John R. Stephen, Roberta De Bei, Cassandra Collins and Carlos M. R. Lopez

*213 Prospect on Ionomic Signatures for the Classification of Grapevine Berries According to Their Geographical Origin* Youry Pii, Anita Zamboni, Silvia Dal Santo, Mario Pezzotti, Zeno Varanini and

Tiziana Pandolfini

*220 The Induction of Noble Rot (*Botrytis cinerea*) Infection During Postharvest Withering Changes the Metabolome of Grapevine Berries (*Vitis vinifera *L., cv. Garganega)*

Stefano Negri, Arianna Lovato, Filippo Boscaini, Elisa Salvetti, Sandra Torriani, Mauro Commisso, Roberta Danzi, Maurizio Ugliano, Annalisa Polverari, Giovanni B. Tornielli and Flavia Guzzo

# SECTION 3

#### GENOTYPE AND DEVELOPMENT-RELATED FACTORS

*232 Metabolite Profiling Reveals Developmental Inequalities in Pinot Noir Berry Tissues Late in Ripening*

Amanda M. Vondras, Mauro Commisso, Flavia Guzzo and Laurent G. Deluc

*246 System-Level and Granger Network Analysis of Integrated Proteomic and Metabolomic Dynamics Identifies Key Points of Grape Berry Development at the Interface of Primary and Secondary Metabolism*

Lei Wang, Xiaoliang Sun, Jakob Weiszmann and Wolfram Weckwerth


Darren C. J. Wong and José Tomás Matus


Stefania Pilati, Giorgia Bagagli, Paolo Sonego, Marco Moretto, Daniele Brazzale, Giulia Castorina, Laura Simoni, Chiara Tonelli, Graziano Guella, Kristof Engelen, Massimo Galbiati and Claudio Moser

*354 Insights Into the Role of the Berry-Specific Ethylene Responsive Factor*  VviERF045

Carmen Leida, Antonio Dal Rì, Lorenza Dalla Costa, Maria D. Gómez, Valerio Pompili, Paolo Sonego, Kristof Engelen, Domenico Masuero, Gabino Ríos and Claudio Moser

*371 Sequence Polymorphisms and Structural Variations Among Four Grapevine (*Vitis vinifera *L.) Cultivars Representing Sardinian Agriculture* Luca Mercenaro, Giovanni Nieddu, Andrea Porceddu, Mario Pezzotti and Salvatore Camiolo

#### SECTION 4

#### SUMMING-UP AND PERSPECTIVE VIEWS OF 'OME'-WIDE RESEARCH IN GRAPEVINE

*384 Omics Approaches for Understanding Grapevine Berry Development: Regulatory Networks Associated With Endogenous Processes and Environmental Responses*

Alejandra Serrano, Carmen Espinoza, Grace Armijo, Claudio Inostroza-Blancheteau, Evelyn Poblete, Carlos Meyer-Regueiro, Anibal Arce, Francisca Parada, Claudia Santibáñez and Patricio Arce-Johnson

*399 How Single Molecule Real-Time Sequencing and Haplotype Phasing Have Enabled Reference-Grade Diploid Genome Assembly of Wine Grapes* Andrea Minio, Jerry Lin, Brandon S. Gaut and Dario Cantu

*405 A Concise Review on Multi-Omics Data Integration for Terroir Analysis in*  Vitis vinifera

Pastor Jullian Fabres, Cassandra Collins, Timothy R. Cavagnaro and Carlos M. Rodríguez López

*413 Plant Stress Responses and Phenotypic Plasticity in the Epigenomics Era: Perspectives on the Grapevine Scenario, a Model for Perennial Crop Plants*

Ana M. Fortes and Philippe Gallusci

# The Transcriptional Responses and Metabolic Consequences of Acclimation to Elevated Light Exposure in Grapevine Berries

Kari du Plessis <sup>1</sup> , Philip R. Young<sup>1</sup> , Hans A. Eyéghé-Bickong1, 2 and Melané A. Vivier <sup>1</sup> \*

1 Institute for Wine Biotechnology, Department of Viticulture and Oenology, Stellenbosch University, Stellenbosch, South Africa, <sup>2</sup> Institute for Grape and Wine Sciences, Department of Viticulture and Oenology, Stellenbosch University, Stellenbosch, South Africa

An increasing number of field studies that focus on grapevine berry development and ripening implement systems biology approaches; the results are highlighting not only the intricacies of the developmental programming/reprogramming that occurs, but also the complexity of how profoundly the microclimate influences the metabolism of the berry throughout the different stages of development. In a previous study we confirmed that a leaf removal treatment to Sauvignon Blanc grapes, grown in a highly characterized vineyard, primarily affected the level of light exposure to the berries throughout their development. A full transcriptomic analysis of berries from this model vineyard details the underlying molecular responses of the berries in reaction to the exposure and show how the berries acclimated to the imposing light stress. Gene expression involved in the protection of the photosynthetic machinery through rapid protein-turnover and the expression of photoprotective flavonoid compounds were most significantly affected in green berries. Overall, the transcriptome analysis showed that the berries implemented multiple stress-mitigation strategies in parallel and metabolite analysis was used to support the main findings. Combining the transcriptome data and amino acid profiling provided evidence that amino acid catabolism probably contributed to the mitigation of a likely energetic deficit created by the upregulation of (energetically) costly stress defense mechanisms. Furthermore, the rapid turnover of essential proteins involved in the maintenance of primary metabolism and growth in the photosynthetically active grapes appeared to provide precursors for the production of protective secondary metabolites such as apocarotenoids and flavonols in the ripening stages of the berries. Taken together, these results confirmed that the green grape berries responded to light stress much like other vegetative organs and were able to acclimate to the increased exposure, managing their metabolism and energy requirements to sustain the developmental cycle toward ripening. The typical metabolic consequences of leaf removal on grape berries can therefore now be linked to increased light exposure through mechanisms of photoprotection in green berries that leads toward acclimation responses that remain intact until ripening.

Keywords: grape, microclimate, photosynthesis, RNAseq analysis, acclimation to stress

#### Edited by:

Giovanni Battista Tornielli, University of Verona, Italy

#### Reviewed by:

Claudio Moser, Fondazione Edmund Mach, Italy Etti Or, Agricultural Research Organization, Volcani Center, Israel

> \*Correspondence: Melané A. Vivier mav@sun.ac.za

#### Specialty section:

This article was submitted to Crop Science and Horticulture, a section of the journal Frontiers in Plant Science

> Received: 10 April 2017 Accepted: 04 July 2017 Published: 20 July 2017

#### Citation:

du Plessis K, Young PR, Eyéghé-Bickong HA and Vivier MA (2017) The Transcriptional Responses and Metabolic Consequences of Acclimation to Elevated Light Exposure in Grapevine Berries. Front. Plant Sci. 8:1261. doi: 10.3389/fpls.2017.01261

# INTRODUCTION

Plants show remarkable adaptability to environmental factors and/or stresses to ultimately ensure that their core metabolic functions are maintained. Although these aspects have been intensively studied in model plants under controlled conditions to establish the basic principles and underlying pathways, as technologies developed, our ability to study and understand crop plants in their cultivated natural environments are yielding important information regarding the processes of stress protection and specifically the concept of acclimation.

In plant biology, stress is generically defined as any unfavorable conditions that affect metabolism, growth and/or development (Lichtenthaler and Burkart, 1996). The relative tolerance/sensitivity of the affected plant subsequently determines if a stress factor will have a positive (eustress) or negative (distress) outcome (Kranner et al., 2010). Acclimation refers to the short-term responses of plants to adapt to unfavorable (stress) factors in their immediate environment (Lichtenthaler and Burkart, 1996; Lichtenthaler, 1998); whereas adaptation refers to plants' long-term survival strategy to stress factors that occurs via genetic changes such as mutations and subsequent natural selection over many generations within a population. When compared to adaptation, acclimation is a rapid response, occurs within individuals, is reversible, and does not involve any permanent genetic changes. Acclimation can involve transcriptional, metabolic and/or physiological responses to improve the performance and survival of the individual to the stress. The ability of biennial plants (e.g., onions, cabbages, and carrots) to survive winter (Andrews, 1996) and the accumulation of phenolic compounds in response to increased light exposure (Caldwell et al., 1983), are examples of acclimation to low temperature and UV-B, respectively.

In grapevine, acclimation to climatic conditions is particularly important and the plasticity of grapevine responses have been highlighted in a number of publications (overviewed recently in Kuhn et al., 2014). The transcriptomic and metabolic reprogramming occurring during grape berry development has been well studied (Zenoni et al., 2010; Sweetman et al., 2012; Palumbo et al., 2014; Pilati et al., 2014; Wong et al., 2016). Research on abiotic stress factors has focused on the dominant environmental factors either individually: temperature (Carbonell-Bejerano et al., 2013; Rienth et al., 2014), light (Wu et al., 2014; Reshef et al., 2017; Sun et al., 2017), UV (Martinez-Luscher et al., 2014; Suzuki et al., 2015; Matus, 2016), and water deficit (Ghan et al., 2015; Santo et al., 2016; Savoi et al., 2016) or collectively as terroir or vintage studies (e.g., Santo et al., 2013; Anesi et al., 2015).

Light has long been recognized as central to plant metabolism through photosynthesis, but recent studies have highlighted the importance of light as a source of information for plants (reviewed in Apel and Hirt, 2004; Eberhard et al., 2008; Li et al., 2009 and references within). In viticulture, many canopy management practices are performed to optimize light exposure to drive photosynthesis of the canopy (reviewed in Smart, 1985; Clingeleffer, 2010). Apart from leaves, other plant organs including the stems, flowers, tendrils and fruits contain functional chloroplasts, and are capable of photosynthesis (reviewed in Blanke and Lenz, 1989). The conditions under which photosynthesis occurs in these non-foliar organs, however, are markedly different to their foliar counterparts. In fruits, for example, the gradual disappearance of stomata and/or the development of an impermeable waxy cuticle during development results in an internal environment that is characterized by high CO<sup>2</sup> and low O<sup>2</sup> (hypoxic) levels (Blanke and Leyhe, 1987, 1988; Kyzeridou et al., 2015). Decreased photosynthesis in green fruits can be attributed to these physical/anatomical features, rather than a decrease in the photosystems. Kyzeridou et al. (2015) demonstrated that in comparison to leaves, the green fruits of Nerium oleander and Rosa sp. had higher Car/Chl ratio due to increased xanthophyll cycle components (violaxanthin, antheraxanthin and zeaxanthin) and a lower chlorophyll content. This resulted in a photoprotective xanthophyll cycle that is more functional under high light in green fruits than in leaves. This has also been reported for apple (Cheng and Ma, 2004) and grapevine (Young et al., 2016) and it is speculated that this exists in nonfoliar photosynthetic organs to reflect a common strategy for photosynthetic green tissues under similar low oxygen conditions (Kyzeridou et al., 2015).

Some canopy manipulations, such as leaf removal in the fruiting zones are, however, utilized to increase light penetration to the berries (reviewed in Reynolds, 2010). A significant number of studies have investigated the impacts of leaf removal on berry development and ripening. Depending on the cultivar, the objectives range from improving the acid balance (Hunter and Visser, 1990; Toda et al., 2013; Baiano et al., 2015); improving anthocyanin/color stability (Chorti et al., 2010; Sternad Lemut et al., 2011; Lee and Skinkis, 2013; Baiano et al., 2015; Song et al., 2015; Guan et al., 2016; Yu et al., 2016; Pastore et al., 2017); increasing specific secondary metabolites such as volatile aroma precursors (Staff et al., 1997; Tardaguila et al., 2010; Feng et al., 2015; Song et al., 2015; Suklje et al., 2016; Young et al., 2016) or lowering of metabolites that are perceived negatively in the grapes/wines (Sala et al., 2004; reviewed in Sidhu et al., 2015). One of the main outcomes of leaf removal in the bunch zones is the accumulation of protective phenolic compounds i.e., anthocyanins (Lee and Skinkis, 2013; Guan et al., 2016; Lee, 2017) and flavonols (Yu et al., 2016; Pastore et al., 2017), as well as changes to volatile aroma compounds i.e., the norisoprenoid, β-damascenone (Feng et al., 2015; Young et al., 2016) and monoterpenes (Song et al., 2015; Young et al., 2016). These studies have all highlighted the adaptability of the grapevine berries to the changed microclimate and have also provided scope to investigate mechanisms of perceiving and adapting to the stresses linked to changes in microclimate.

Taking advantage of a validated experimental setting where light exposure (to the bunch zone) was the major environmental factor significantly altered by a classic leaf removal treatment in a model Sauvignon Blanc vineyard, the mechanism of berry acclimation to increased light exposure (Young et al., 2016) was targeted in this study. A pertinent result from the phenotyping and metabolite profiling was that none of the parameters and metabolites measured indicated a compromised primary growth/development and ripening of the berries under the increased exposure. Metabolically, the berries responded to increased light exposure by producing specific secondary metabolites that have photo-protective and/or antioxidant functions. The data generated in the targeted metabolite profiling of the berries lead to the conclusion that the berries mitigated the stress with metabolite reprogramming to acclimate to the increased exposure and that the response was strongly influenced by developmental stage. Although sugars, organic acids, chlorophylls and major photosynthetic pigments (β-carotene and lutein) were not affected by the increased light exposure; specific monoterpenes and photoprotective xanthophylls (zeaxanthin, antheraxanthin, and lutein epoxide) were shown to be increased (Young et al., 2016). These results raised an important question: How were primary metabolism and developmental patterns maintained, despite the light stressresponse and metabolic reorganization activated in the exposed berries?

Our primary approach toward achieving these aims was to take a global transcriptional snapshot of gene expression at various berry developmental stages using RNA Sequencing (RNASeq) to thereby create an overview of the effects of elevated light exposure on berry development and ripening. Using this global overview, we were able to target specific metabolic pathways of which gene expression was most significantly affected by the treatment. We could further explore what affects these alterations in gene expression could have on accumulation of metabolites involved in these affected pathways to ultimately determine how berry growth and primary metabolism was maintained despite the activation of stress response mechanisms previously reported (Young et al., 2016).

#### MATERIALS AND METHODS

#### Experimental Design, Agronomical Treatments, and Sampling Strategy

The Vitis vinifera cv. Sauvignon blanc grapes that were the research materials for this study were harvested from an experimental vineyard located in Elgin region of South Africa during the 2010/2011-harvest season. The complete details pertaining to the climatic measurements, vineyard layout, viticultural practices and sampling strategy of the relevant samples have been performed according to an established fieldomics workflow (Alexandersson et al., 2014) and are available in Young et al. (2016). Briefly, grapes were sampled from twelve biological replicates (or panels with six panels per row; and six panels per treatment) in two adjacent vineyard rows (NW-SE row orientation). Each individual biological replicate (panel) consisted of four consecutive vines. The leaf-removal treatment included leaf and lateral shoot removal applied in the bunch zone on the SE-facing side of the canopy at EL29. This leafremoval treatment was applied to every alternate panel creating a "checkerboard" plot layout where a control panel was always adjacent to an exposed panel (both within a row, and between rows) (Young et al., 2016).

The berries were sampled at green- (pea-sized) (EL31) (Eichhorn and Lorenz, 1977), pre-véraison- (EL33), véraison- (EL35), and the ripe-stage (EL38; corresponding to the commercial harvest date) from control (shaded) and exposed vine panels after which it was frozen in liquid nitrogen in the field. The seeds were removed from the frozen berries in the laboratory and the whole berries, including skins and pulp, were kept at −80◦C until subsequent analyses were performed.

# Transcriptional Analysis

#### RNA Extraction and Sequencing

Total RNA was extracted from three out of the six biological replicates sampled at four developmental stages under both exposed and control conditions according to an established protocol (Reid et al., 2006). Each of the 24 samples was subjected to DNAse1 treatment (Sigma-Aldrich, Saint-Louis, MO, USA) to eliminate contamination with genomic DNA. The concentration and purity of the extracted RNA samples were established using a Nanodrop 2000 Spectrophotometer (Thermo Scientific, Wilmington, DE, USA) and the integrity of the samples were confirmed through analysis of a Bioanalyzer Chip RNA 7500 series II (Agilent, Santa Clara, CA, USA) according to the manufacturer's instructions.

After achieving each of the quality control criteria, poly(A) mRNA was prepared for each of the RNA samples and sequenced through an Illumina HiSeq 1000 sequencer according to manufacturing protocols (Illumina Inc., San Diego, CA, USA). The reads generated from the sequencing procedure were aligned to the V1 version of the V. vinifera genome (PN40024) using version 2.0 of the TopHat software (Trapnell et al., 2012), allowing a maximum of two nucleotide mismatches. Cufflinks software (version 2.0) was subsequently used in order to assemble transcripts from generated sequence reads (Trapnell et al., 2012), hereby calculating the transcript abundance of each gene in the form of an FPKM value (expected fragments per kilobase of transcript per million fragments mapped). For the purpose of determining which transcripts show differential expression between treatments, CuffDiff (version 2.0) was used after transcript abundances were determined (Trapnell et al., 2012).

#### RNASeq Expression Data Analysis

When comparing the entire transcriptomes of each of the samples included in this study, Pearson correlations were calculated using R (version 3.3.1) in RStudio (version 0.99.903) and the visualization of the results in the form of a correlation matrix were performed using Microsoft Excel (version 14.1.0).

Gene Ontology (GO) Enrichment analyses of the entire gene lists that showed non-significant differential expression between exposed and control samples at each phenological stage were performed in the BiNGO application in Cytoscape (version 3.4.0) using the Benjamini and Hochberg False Discovery Rate Correction metric. These genes will be further referred to as "unaffected." GO terms were considered significant with a pvalue smaller than 0.05.

In order to evaluate genes that were significantly affected by elevated light, the results generated from the differential expression analysis were implemented in a three step process according to different selection criteria. The first step was to perform GO enrichment analysis of all the genes that were significantly differentially expressed (q ≤ 0.05) under exposed conditions at each developmental stage in order to evaluate the effect that the treatment had on the berry transcriptome throughout development. Next, two distinct thresholds were chosen based upon the number of genes generated that would be most appropriate for subsequent analyses. The first threshold was set to include all differentially expressed genes with a log<sup>2</sup> fold change greater than 1.5 and smaller than −1.5 when comparing the expression of exposed to control genes in order to generate a large list of highly significantly affected genes for the purpose of clustering analysis. This would allow for the identification and evaluation of the most prominent expression profiles of the genes affected by increased exposure without specifically focusing on individual genes. The second threshold was set to include only differentially expressed genes with a log<sup>2</sup> fold change greater than 2 and smaller than −2 for the purpose of focusing on the individual genes that were most affected by increased exposure.

GO enrichment analysis of significantly enriched expression profile clusters of genes expressed at a log<sup>2</sup> fold change (log2FC) greater than 1.5 between exposed and control grapes during at least one of the phenological stages were performed using the online analysis tool, AgriGO (Du et al., 2010) using the Fisher statistical method with the Yekutieli False Discovery Rate multitest adjustment metric. Significantly enriched GO terms (p < 0.05) were further visualized and summarized using the Reduce + Visualize Gene Ontology Web Server (http://revigo. irb.hr; Supek et al., 2011).

For the purpose of performing clustering analysis to infer which genes conform significantly to predetermined gene expression profiles, the Short Time-Course Expression Miner (STEM) was implemented (Ernst et al., 2006). Visualizations of the abovementioned differential expression analyses were performed using Microsoft Excel and Powerpoint (version 14.1.0).

The putative developmental biomarkers were identified and further explored in a three step process. Firstly, the molecular biomarkers of the control grapes representing the two most distinct developmental phases (i.e., green stages vs. ripening stages) were identified by implementing a previously established method (Zamboni et al., 2010). Putative biomarkers that represent the transcriptional difference between the green and the ripening grape berry stages were identified. A two-class OPLS-DA model was generated by representing the expression of green, control berry samples (EL31 and EL33) as its own class as a reference against expression of ripening, control berry samples (EL35 and EL38) set as the second class using SIMCA (version 14.0). An S-plot was subsequently generated to identify the loading correlation coefficient of each gene as described by Zamboni et al. (2010; Wiklund et al., 2008). The aim of this investigation was to generate a broad overview of the developmental progression of the grapes included in this study and therefore, a less stringent correlation cut-off was implemented than in previous studies to identify genes with a loading correlation coefficient higher than 0.8 (positive biomarkers) and lower than −0.8 (negative biomarkers). The expression of positive biomarkers were significantly higher in ripening berries compared to green berries, whereas negative biomarker expression was significantly lower in ripening berries compared to green berries (according to the nomenclature adopted by Zamboni et al., 2010).

Secondly, to establish whether these identified control grape berry developmental biomarkers were comparable to those already established for grape developmental progression, molecular biomarkers identified in this investigation were compared to those published from two previous investigations. The first set of biomarkers included in this comparison was published by Zamboni et al. (2010) in which transcriptional elements unique to early berry development (EL33 and EL35) and late berry development (EL36 and EL38) were identified and named Class a and Class b genes, respectively. These biomarkers will be referred to as early and late developmental markers in subsequent sections of this publication. The second set of genes used to compare the development of the grapes included in this study was published by Palumbo et al. (2014) in which they identified so-called "switch genes" that are considered to characterize the unique transcriptional switch that occurs when grape berries transition from being green, photosynthesizing organs to becoming ripening, sink organs. This aforementioned study utilized transcriptional data generated from five red Italian grape cultivars as well as data generated from the grapevine transcription atlas (Fasoli et al., 2012). A Venn diagram was constructed using the Bioinformatics and Evolutionary Genomics platform (http://bioinformatics.psb. ugent.be/webtools/Venn/) by comparing the genes from the abovementioned studies and the molecular biomarkers identified in this study.

Finally, using the identified developmental biomarkers, the effect of the treatment on the progression of berry development was further explored. This was achieved by determining which of the identified biomarkers shared between this and previous studies were significantly affected by the leaf-removal treatment (and increased exposure) by evaluating the differential expression of these genes.

In order to determine how photosynthesis is affected on a transcriptional level by elevated light exposure, the appropriate gene accessions encoding proteins of PSI and PSII of the thylakoid membranes were obtained from the KEGG Pathway database for V. vinifera (http://www.kegg.jp/kegg-bin/highlight\_ pathway?scale=1.0&map=vvi00195&keyword).

#### Quantitative Real-Time Polymerase Chain Reaction (RT-PCR)

In order to validate the accuracy of the gene expression patterns observed in the results generated through RNASeq analysis, RT-PCR was performed using the Applied Biosystems 7500 Real-time PCR System. For these verification assays, total RNA was extracted from three of the six biological replicates originally harvested for metabolic and RNA Seq analyses using the SpectrumTM Plant Total RNA Kit (Sigma-Aldrich, Saint-Louis, MO, USA). cDNA was synthesized from the total RNA using the SensiFASTTM cDNA Synthesis Kit (Bioline, London, UK) and RT-PCR was performed using the KAPA SYBR <sup>R</sup> FAST qRT-PCR Kit according to the manufacturer's instructions (Kapa Biosystems, Cape Town, South Africa). Six genes were selected as targets for the PCR reactions based on their expression patterns in response to the treatment as reported by the RNASeq analysis. Four of these target genes were upregulated in response to the treatment by a log2FC greater than 2 at various developmental stages (VIT\_10s0116g00410, VIT\_18s0001g03470, VIT\_05s0020g04110, VIT\_02s0025g04060). The other two of the target genes were related to photosynthesis and were significantly upregulated by elevated light exposure in the green berries (VIT\_01s0010g03620, VIT\_19s0014g00160). Appropriate primers were designed using QuantPrime (Arvidsson et al., 2008). These primers, their sequences and their characteristics are summarized in **Table S1**. All PCR reactions were performed in triplicate. The normalization and absolute quantification of the expression levels of each of the six genes were performed using the Linear Regression Efficiency (LRE) method using LRE Analyzer software (Rutledge and Stewart, 2008; Rutledge, 2011).

#### Metabolite Analysis

Extractions and subsequent metabolite analyses were performed from three out of the six available biological repeats that represented the biological triplicates sampled at four developmental stages under both exposed and control conditions.

#### Amino Acid Analysis

The extraction and HPLC analysis of amino acids in berry samples was performed as described in Antalick et al. (2010), with minor changes. Frozen homogenized berry tissue (200 ± 10 mg) was weighed into 2 mL microfuge tubes and 0.5 mL of 70% (v/v) methanol [containing 25 mg/L of each of the two internal standards (IS), sarcosine and norvaline] was added. Samples were briefly vortexed and sonicated for 10 min at room temperature. After sonication, the samples were centrifuged at 1,250 rpm for 5 min and 200 µL of the supernatant was transferred to amber vials, crimp-sealed and if not analyzed immediately stored at −4 ◦C. Each biological replicate was extracted and analyzed in triplicate. The extracted amino acids were derivatized before analysis on HPLC as described in Suklje et al. (2016).

Major amino acids (AAs) were identified based on their retention times with respect to authentic standard elution and quantified using external standard calibration based on standard curves plotted using the peak areas vs. the standard concentrations. Concentrations were normalized to the IS amount and the sample fresh weight (FW) to obtain the AA concentrations per fresh berry weight (mg/g FW).

#### Quantification of Phenolic Compound Contents

All authentic standards namely quercetin-glucoside; catechin, epicatechin as well as caftaric acid and caffeic acid as well as the HPLC grade solvents used for sample extraction and separation such as methanol (MeOH, 99.0%), acetonitrile (99.0%), hydrochloric acid (HCl), and the orthophosphoric acid (H2PO4, 99.0%) were acquired from Sigma Aldrich (Steinheim, Germany).

Homogenized grapevine berries (200 ± 10 mg) were weighed and 0.5 ml of acidified MeOH (70%; adjusted to pH 1.5 with HCl) was added to each vial, which was then vortexed and sonicated for 15 min at room temperature. After sonication, the samples were centrifuged at 1400 rpm for 5 min and 200 µL of the supernatant was collected and added into amber vials, crimp-sealed for HPLC analysis. Extraction was done in triplicate, in a dark room away from direct light. Extracted flavonoids and phenolic acids in berries were separated and quantified using an Agilent 1100 series HPLC system (Agilent Technologies©, Palo Alto, California, USA) equipped with a diode array detector (DAD) and controlled by a ChemStation Rev. A.10.02 software (Agilent Technologies©). The column used was a Phenomenex Prodigy ODS-2 (4.6 × 150 mm, 5 µm) preloaded with Phenomenex Prodigy guard cartridge (2.1 mm × 100 mm, 1.7 um). The mobile phases were composed of 15% (v/v) H2PO<sup>4</sup> (A) and 80% acetonitrile containing 20% A (B) and the flow rate was 1 mL/min. The gradient elution conditions started with a linear gradient from 6 to 31% B for 68 min following with another linear gradient from 31 to 65% B for 5 min. Then, the gradient was kept constant at 65% B for 5 min and was decreased from 65%, back to the starting conditions at 6% B for 5 min. The system was re-equilibrated at 6% B for another 10 min before the next injection. The injection volume was set at 20 µL and the column temperature at 40◦C.

The major flavonoids and phenolic acids in grapevine berry samples were identified based on their retention times with respect to authentic standard elution and quantified using external standard calibration based on standard curves plotted using the peak areas vs. the standard concentrations. These chromatographic peaks were obtained using the following DAD wavelengths: 280 nm for flavan-3-ols; 360 nm for flavonols and 320 nm for the phenolic acids. Compounds without available standards were quantified using the calibration parameters from quercetin-glucoside (all flavonols) and caftaric acid. The concentrations in samples were normalized to the sample fresh weight (FW) to obtain the sample amount per berry FW (µg/gFW). **Table S2** summarizes the retention time and calibration parameters of all standards used in this analysis.

#### Lipophilic-Oxygen Radical Absorbance Capacity (L-ORAC) Assay

L-ORAC analysis was performed by the Antioxidant Research Unit (Cape Peninsula University of Technology, South Africa) on three biological replicates (in triplicate) harvested at EL33 and EL38, respectively.

#### Statistical Analysis

The concentrations generated from the analysis of amino acids and phenolic compounds of the grapes were subjected to multivariate data analysis using Statistica (version 13.0). A repeated measures analysis of variance (ANOVA) was performed to identify the relationship between the increased exposure treatment and the concentrations of the measured compounds (AAs and Phenolic compounds). A Fisher LSD Post-Hoc test was conducted for each compound to confirm whether the concentration of the compound was statistically significantly affected by the treatment (q-value).

Basic statistical analysis of data generated from the L-ORAC assay was conducted in Microsoft Excel (version 14.1.0) using a paired t-test to determine whether exposed grapes had significantly higher lipophilic antioxidant capacity than control grapes at EL33 and EL38.

#### RESULTS

#### Overview of the Transcriptional Data Generated

In this study, RNASeq was performed with 24 Sauvignon blanc berry samples representing grapes from shaded (control) and exposed (treatment) microclimates at four developmental stages from a highly characterized vineyard. A summary of the parsed reads from each of the samples and the number of reads that mapped onto the V. vinifera cv. Pinot noir reference genome (PN40024) are included in **Table S3**. The complete RNASeq dataset is available in the NCBI's GEO under the series accession, GSE98873.

In order to compare the complete transcriptomes generated for the 24 grape samples, a correlation matrix was generated by implementing a Pearson's correlation coefficient as a distance metric (**Figure S1**). The resulting matrix revealed that one sample harvested at EL38 did not correlate strongly to the rest of the EL38 samples, but rather to samples taken at EL33. Not only were the other 23 samples closely grouped according to their specific developmental stage, targeted metabolite profiling of the same grape samples previously confirmed the close grouping of all the EL38 samples (Figures 3, 4 in Young et al., 2016). This sample was treated as an outlier (anomaly) and excluded from all subsequent analyses.

The Pearson correlation matrix was reconstructed including only the 23 remaining samples and is presented in **Figure 1**; the matrix shows a strong correlation between grapes from the same developmental stage, regardless of the viticultural treatment implemented. Furthermore, gene expression of green berries was more closely correlated between EL31 and EL33 stages than with the two consecutive ripening stages, EL35 and EL38. The correlation matrix also provided confidence in the experimental design and sampling strategy since the biological replicates of the control and exposed treatments confirmed the repeatability of the effect that the leaf removal treatment had on the berry transcriptome at each developmental stage.

Out of the 29,970 genes represented in this version (V1) of the grapevine genome, the expression of 5,050 genes (16.5%) could not be detected across any of the observed developmental stages and treatments and the enriched GO terms representing these genes are summarized as **Figure S2** (as represented by Revigo). A further 4,715 genes with FPKM expression values lower than the recommended reliable RNASeq threshold of an FPKM = 1 (Warden et al., 2013; Massonnet, 2015) throughout all developmental stages and treatments were excluded from further analyses.

RT-PCR analysis of six genes that showed significant upregulation in response to the exposure treatment was conducted and validated the accuracy of the RNASeq results (**Figure S3**). Predominantly, the general expression trend throughout development of each of the genes was similar when comparing the RNASeq and RT-PCR results for control and exposed grapes. These initial analyses not only established confidence in the experimental design and the repeatability among biological replicates, but it further established the accuracy of the RNASeq method and subsequent results generated.

#### Developmental Biomarker Analysis

In total, the expression of 4,975 genes was identified as developmental phase-specific biomarkers responsible for the greatest transcriptional differences between the green and ripening developmental stages. 2,242 and 2,733 of these genes were positively and negatively correlated (Correlation value ≥ 0.8) to the separation, respectively (**Table S4**).

The expression of these markers was comparable to previously established markers for grape berry development (Zamboni


FIGURE 1 | Pearson correlation matrix of the entire transcriptomes of 23 samples representing exposed and control grapes at four developmental stages (EL31, EL33, EL35, and EL38).

et al., 2010; Palumbo et al., 2014). Furthermore, the expression of 81% of these shared markers developmental markers were not affected by the treatment. The remaining nine genes responsible for the 19% of developmental biomarkers that were affected by the treatment included an auxin-responsive gene (SAUR29; VIT\_16s0098g01150), two genes encoding protein subunits of photosystem I and II (VIT\_12s0028g01080, VIT\_05s0020g03180) and a calmodulin-binding heat shock protein (VIT\_14s0006g01030). The results are summarized in **Figure S3**.

#### Transcriptional Response of the Berries to Increased Exposure

#### Transcripts That Were Unaffected

The number of annotated genes that were either not expressed, unaffected by the leaf removal treatment or differentially expressed when comparing exposed to control grapes at each of the phenological stages are summarized in **Figure 2**.

GO enrichment analysis of the genes statistically unaffected by the light treatment revealed that GO terms associated with growth and development were enriched throughout development. Among these were GO terms related to "Biosynthetic process," "Signal transduction," "Protein metabolic process," "Translation," "Transport," and "Response to external stimulus" (**Figure 2**). Furthermore, during the developmental stages in which the berries were photosynthetically active and growing in size (EL31, EL33, and EL35), genes associated with the GO terms "Growth" and "Multicellular organismal development" were unaffected by the treatment at EL31 and EL33 as well.

#### Transcripts That Were Differentially Expressed as a Consequence of the Treatment

By implementing Cuffdiff software, transcripts that were significantly differentially expressed (q ≤ 0.05) when comparing exposed to control grapes could be identified. For each of the

FIGURE 2 | Pie charts representing the number of genes in the grapevine genome either not expressed, significantly unaffected (q ≥ 0.05) and significantly differentially expressed in response to elevated light (q ≤ 0.05) at the four phenological stages, respectively. The GO terms significantly enriched representing the genes unaffected by the treatment at each phenological stage are summarized in tables. Gray shading represents GO terms that were commonly unaffected throughout berry development.

four developmental stages being evaluated, the percentage of differentially expressed genes (DEGs) were calculated and the genes that were either significantly up or downregulated in response to the leaf removal treatment could be explored by implementing GO enrichment analysis. The results of these analyses are summarized in **Figure 3**.

These results revealed that grape berries were most significantly affected by the treatment on a transcriptional level during the early developmental stages (EL31 and EL33) and the global description of the biological processes these gene groups were involved with, shared a high degree of similarity between EL31 and EL33 grapes. GO terms associated with photosynthesis and the generation of precursor metabolites and energy were very highly upregulated in exposed grapes until véraison. In the green grapes, especially during EL33, genes associated with the GO terms "cell death" and "response to stress" were among the most significantly downregulated functional groups, exclusively representing genes associated with disease and nematode resistance.

Although ripe berries had the highest number of DEGs in response to the treatment, the enrichment of the GO terms affected by the treatment were lower in comparison to the preceding developmental stages. These enriched GO terms were further associated with genes that were significantly downregulated in response to the treatment as opposed to the preceding stages that were dominated by upregulation in response to increased exposure.

Out of the 29970 genes included in the grapevine genome, 723 genes showed either significant up or downregulation with a factor greater than 1.5 (log2FC) during at least one developmental stage in response to the elevated light treatment. Clustering analysis revealed that the expression of 431 of these genes could be grouped to seven expression profile clusters as predetermined by the STEM software (**Figure 4A**), with the GO subcategories provided in **Figure 4B** and the genes within each cluster summarized in **Table S6**.

Two of these identified expression clusters (clusters 1 and 2) were represented by genes that followed the predicted developmental progression whilst simultaneously being affected by the treatment. Cluster 1 (p = 5E−90; total of 128 genes) represented genes that were simultaneously

FIGURE 3 | Grape berry transcripts that are significantly differentially expressed in response to elevated light exposure at four phenological stages. Significantly enriched GO categories (q ≤ 0.05) at each phenological stage. Significance is represented as log10 P-values of each GO category with positive values indicating upregulation and negative values indicating downregulation.

in response to elevated light exposure during at least one of the four phenological stages. (A) The seven expression profiles to which a significant amount of genes aligned. Shaded columns (gray) indicate the expression of the genes in that specific cluster under control conditions and white columns indicate expression of the same genes under exposed conditions at the developmental stage indicated in the X-axis below. Significance is indicated in each cluster profile representation in the form of a P-value. (B) Functional GO subcategories of each significantly enriched expression cluster summarized within representative GO terms as summarized by ReviGO. Significance is represented as −log10 P-values of each subcategory; the size of each data circle indicates the number of genes that is represented within each enriched GO term.

driven by the increased exposure treatment as well as developmental cues. Several of the functional annotations were associated with the progression of grape berry development, but also secondary metabolic processes linked to abiotic stress responses. Examples of genes within cluster 1 included three Ethylene-responsive transcription factors (VIT\_07S0031G01980, VIT\_01S0150G00120, VIT\_14S0108G00050), a 2-oxoglutarate/malate carrier protein (UCP5; VIT\_18S0001G07320) that has been proposed to be involved with acid regulation in grape berries (Chen et al., 2015), a Galactinol synthase (GolS4; VIT\_01S0127G00470) involved in the synthesis of the osmoprotectant oligosaccharide, raffinose, a gene encoding a Gamma-aminobutrytic acid transporter (VIT\_13S0074G00570), two genes encoding enzymes involved in the phenylpropanoid/flavonoid pathway (anthocyanidin 3-O-glucosyltransferase, VIT\_12S0034G00130; Flavanone 3-hydroxylase, VIT\_16S0098G00860), as well as the early light-inducible protein (ELIP1, VIT\_05S0020G04110) involved in the inhibition of chlorophyll biosynthesis.

Interestingly, 64 of the genes represented by cluster 1 were also identified as developmental biomarkers (**Figure S4**) of which five were shared with the analyses of Zamboni et al. (2010) and Palumbo et al. (2014). One of these genes is a 9-cis-epoxycarotenoid dioxygenase encoding gene (NCED; VIT\_02S0087G00930) responsible for the degradation of carotenoids synthesized during the early developmental phases to produce the plant hormone, abscisic acid (ABA) that further plays a pivotal role in plant adaptation to stress.

Cluster 2 represented 80 genes that showed significant downregulation throughout development under exposed conditions, while simultaneously following the same developmental progression. Among the GO terms associated with this cluster were "lipid metabolic process" that represented two senescence-associated genes (SAG101, VIT\_14S0066G01830, VIT\_14S0066G01820) involved in stress-related signaling, as well as the GO terms "photosynthesis" and "generation of precursor metabolites and energy" that both represented genes that encode a photosystem II PsbO protein (VIT\_18S0001G11710), an LHB1B1 light harvesting protein (VIT\_12S0028G00320) and another a polyphenol oxidase chloroplast precursor (VIT\_10S0116G00560). Cluster 2 also contained an Alanine-glyoxylate aminotransferase encoding gene (Alanine-glyoxylate aminotransferase 2 3, mitochondrial, VIT\_08S0058G00930) that plays a central role in the photorespiratory pathway and a gene encoding a trehalose-6-phosphate phosphatase (VIT\_00S0304G00080) that is known to have an indispensible role in normal plant growth and development. Furthermore, 18 of the genes represented by cluster 2 have been identified as negative biomarkers in this study (**Figure S4**).

Clusters 3 and 4 (**Figure 4**) contained genes that were highly responsive to the elevated light exposure treatment regardless of the developmental profile. The genes represented by these clusters show strong functional associations to the activation of several protection mechanisms of the photosynthetic machinery, activated at either the first (EL31) or the second green developmental stage (EL33). Several heat shock protein (HSP) encoding genes, including the well-known abiotic stress signaling regulator, heat shock factor 2A (VIT\_04S0008G01110), alongside its putative co-activator, Multiprotein-bridging factor 1 (VIT\_11S0016G04080), as well as small HSPs formed part of these clusters.

Cluster 3 further represented several genes that contributed to the GO term, "photosynthesis." These included a gene encoding a chloroplastic carbonic anhydrase (VIT\_14S0066G01210) critical in the maintenance of the rate of photosynthetic CO<sup>2</sup> fixation, and a photosystem II protein encoding gene (PsbP, VIT\_13S0019G00320) that forms part of the oxygen evolving complex of PSII, specifically contributing toward its stabilization. Furthermore, a WUSCHEL encoding gene (VIT\_18S0001G10160) was present in this cluster that represents a member of a transcription factor gene family involved in reproductive organ development, hormone signaling and abiotic stress response in several plant species.

The 30 genes represented by expression cluster 4 show significantly higher expression from EL33 until véraison after which the expression of these genes was unaffected in ripe berries in response to the treatment. Among the 24 genes within this cluster that had been functionally annotated, an FtsH protease encoding gene (VIT\_14S0108G00590), known to be involved in the efficient turnover of the D1 protein of PSII in response to photooxidation, as well as a Calmodulin encoding gene (VIT\_18S0122G00180) known to be involved in stress perception and signaling related to cellular calcium ion (Ca2+) concentration in plants were included. Furthermore, this cluster represented genes encoding a galactinol synthase (VIT\_07S0005G01970), a Methyl jasmonate esterase (VIT\_00S0253G00150) and a 2-oxoglutarate-dependent dioxygenase (VIT\_05S0049G00220) among others. Clusters 3 and 4 therefore point toward the activation and maintenance of light stress mitigation strategies during the green developmental stages.

The remaining three clusters (clusters 5, 6, and 7) represented genes that were differentially affected by elevated light exposure according to neither a unique developmental pattern nor consistently by the treatment (**Figure 4**; **Table S6**). Due to the random and complex nature of their transcriptional responses, these gene clusters were not further investigated for the purpose of this study.

In the second step taken to elucidate which transcriptional elements are the most significantly affected by elevated light exposure at each individual stage, genes that show a Log<sup>2</sup> fold change (Log2FC) either higher than 2 or lower than −2 in exposed compared to control grapes were further explored. In total, 245 and 157 genes were up and downregulated in exposed compared to control grapes according to these criteria, respectively. These genes are listed in **Table S5** and their functional associations are summarized in **Figure 5**.

Among these 245 significantly upregulated genes, 185 were uniquely upregulated at very high levels at each developmental stage investigated with 12, 47, 61, and 65 genes upregulated (Log2FC ≥ 2) at EL31, EL33, EL35, and EL38, respectively. Out of the 157 genes that were most significantly downregulated (Log2FC ≤ −2), 156 of these genes were uniquely downregulated at either EL31 (28 genes), EL33 (29 genes), EL35 (12), or EL38 (87) in response to elevated light exposure. Several genes were similarly upregulated in various developmental stages (**Figure 5**). These genes, their functional annotations and the significance of their differential expression (q-values) are summarized in **Table S7**.

### Metabolic Processes Most Affected by Elevated Light Exposure

The global transcriptional analysis of Sauvignon blanc grape berries yielded insights into which metabolic processes are most affected by elevated light exposure. Gene expression involved in photosynthesis and the synthesis of flavonoid compounds were most significantly activated by the light treatment, which warranted further investigation into how subsequent primary and secondary metabolism of the grape berries was affected by the treatment. In order to investigate these metabolic processes, the synthesis and degradation of the amino acid transcription and composition was further investigated and explored in the context of how this AA metabolism may affect secondary metabolism in response to elevated light exposure in the berry bunch zone.

#### Protection of the Photosynthetic Machinery

The 24 genes included in the investigation of PSI and PSII, their functional annotations and the Log<sup>2</sup> fold change of the expression of each gene when comparing exposed to control grapes at each developmental stage is summarized in **Figure 6**. Every gene included in this analysis was significantly upregulated (q ≤ 0.05) in response to the leaf removal treatment at EL31.

Similarly, during EL33 and EL35, most of the genes of PSI and PSII remained significantly upregulated with the exception of one LHCA gene (LHCA5, VIT\_18s0001g10550), two Psb encoding genes (PsbP, VIT\_13s0019g00320; PsbZ, VIT\_12s0059g01810) that were unaffected from véraison onwards and a PsbQ (VIT\_19s0014g05080) that was unaffected by elevated light exposure from EL33 onwards. Thereafter, at EL38, with the exception of one CAB encoding gene (LHCII type I CAB-1, VIT\_19s0014g00160), all of the genes evaluated became either


FIGURE 5 | Venn diagram summarizing the functional associations of the genes up or downregulated by a Log2 fold-change greater than 2 and smaller than −2 when comparing exposed to control grapes at four phenological stages. The number of upregulated genes is indicated in bold and the number of downregulated genes is italicized in the Venn diagram and GO annotations of the genes uniquely highly differentially regulated at each developmental stage. GO descriptions indicate GO terms that were representative of each gene group, percentages indicate the percentage of genes that are represented by each GO description.

unaffected by the treatment or significantly downregulated in response to the treatment.

The genes putatively encoding enzymes involved in photoprotection mechanisms in grapevine have been acquired from Arabidopsis orthologs and the log2FC of their expression when comparing exposed to control grapes at each developmental stage and is also summarized in **Figure 6**. At EL31, all the genes encoding the enzymes of both nonphotochemical quenching (NPQ) and reversible photoinhibition were significantly upregulated with exception of one FtsH protease-encoding gene (VIT\_14S0108G00590). Although the abovementioned FtsH protease appeared to be highly upregulated (Log2FC = 7.32), it's expression proved to be highly variable among the biological replicates in this study and was therefore not significantly different when comparing exposed to control berries at EL31. The genes encoding NPQ associated proteins that include PsbS (VIT\_18s0001g02740) and one violaxanthin deepoxidase enzyme (VDE) encoding gene (VIT\_04s0043g01010) were strongly upregulated by the treatment at EL31. At EL33, however, the FtsH protease-encoding gene (VIT\_14S0108G00590), putatively responsible for the degradation of damaged copies of the D1 protein, was most significantly and highly upregulated until the berries were ripe (EL38).

#### The Effect of Elevated Light Exposure on Amino Acid Metabolism of Developing Grape Berries

HPLC analysis was performed that yielded the concentrations of 23 amino acids at the four developmental stages. The amino acid (AA) concentrations generated for each of the samples generated are included in **Table S8**. The log2FC values and statistical significance between exposed and control grapes are summarized in **Table 1**. Among these 23 amino acids, the concentrations of eight of these were not affected by the leaf removal treatment at any of the developmental stages evaluated. The only amino acid that was affected by the leaf removal treatment throughout the entire berry development was Gly that was present at significantly higher concentrations from EL31 until EL38.

Taken together these results revealed that, with the exception of Gly, most of the AA concentrations remained unaffected by the treatment until the onset of ripening, followed by the accumulation of significantly altered AA concentrations when comparing exposed to control grapes.

thylakoid membrane within the chloroplasts. (B) A table of representative candidate genes involved in photosystem I and II and two mechanisms of photoprotection in the form of non-photochemical quenching and reversible photoinhibition (RPI), their accessions and the log2 fold-change when comparing their expression levels (FPKM) between exposed and control grapes at each developmental stage. Significant differences in expression between exposed and control grapes are indicated in bold.

At véraison (EL35) 10 out of the 23 AAs measured were present at significantly lower concentrations in exposed grapes, including the four key nitrogen assimilation AAs, Asp, Asn, Glu, Gln, as well as Ala, Arg, Cys, Met and two aromatic AAs, Phe, and Trp. When the berries achieved ripeness at EL38, GABA, Met, Pro, and Val were present at significantly higher concentrations

TABLE 1 | The fold change (Log2) of the amino acid concentrations (mg/gFW) of developing grapes when comparing exposed to control berries at four phenological stages.


Values that are statistically different between exposed and control grapes (q ≤ 0.05) are colored according to either higher or lower concentrations. These colors indicate higher or lower log<sup>2</sup> fold changes between exposed and control grapes based on their concentrations but are not indicative of higher or lower concentrations themselves.

along with Gly whereas Arg, Asp, Phe, and Trp remained present at lower concentrations in exposed grapes. At this stage, His concentrations were also significantly lower when comparing exposed to control grapes.

To explore the transcriptional regulation of the synthesis and degradation of several of the AAs that were present at altered concentrations in response to the leaf removal treatment, four metabolic pathways including several of the altered AAs were targeted for further investigation. These metabolic pathways included Gly metabolism (**Figure 7A**), the superpathway of Lys, Met and Thr metabolism (**Figure 7B**), the superpathway of Trp, Phe, and Tyr metabolism (**Figure 7C**) and the pathway that involved Pro, Arg, and GABA metabolism (**Figure 7D**). The genes putatively involved in these metabolic pathways according to the current available gene annotation collection are indicated by numbers in the appropriate diagrams and are summarized in **Table S9**.

By evaluating these four AA metabolic pathways it became clear that transcription of the biosynthetic enzyme encoding genes were only marginally affected by the leaf removal treatment, whereas the genes encoding enzymes responsible for the degradation of many of the evaluated AAs were transcriptionally far more reactive to the treatment in the ripening period. The pathway depicting Trp, Phe, and Tyr metabolism (**Figure 7C**) is one example of this upregulation of AA catabolic enzyme encoding genes where the genes responsible for the synthesis of Phe and Tyr were not significantly affected by elevated light exposure at any of the berry developmental stages. The Phe ammonia lyase (PAL) encoding genes (VIT\_06s0004g02620, VIT\_08s0040g01710, VIT\_13s0019g04460) and the Tyr aminotransferase encoding genes (VIT\_00s0225g00230, VIT\_00s0394g00040) respectively responsible for the degradation of Phe and Tyr were, however, significantly differentially expressed in response to the increased exposure at various stages of berry development.

Increased exposure had distinctly different consequences on grape AA metabolism when comparing green to ripening berries. An example of this developmental, stage-specific metabolism was evident in the upregulation of AA catabolic enzymes in the pathways involved in Gly synthesis (**Figure 7A**) whereby Gly synthesis from the catabolism of both Ser and Glyoxylate were higher in exposed grapes during the green berry stages under elevated light conditions. Conversely, during the berry ripening stages, the synthesis of Gly from the degradation of Ser and Thr by the upregulation of catabolic enzyme encoding genes were higher in exposed grapes. The degradation of several of these AAs will make their constituents, whether secondary compounds or other AAs, available as substrates to secondary metabolic processes that warranted further investigation.

#### Metabolic Shifts between Primary and Secondary Metabolism in Response to Elevated Light Exposure throughout Berry Development

For the purpose of determining how elevated light exposure could shift developing grape primary and secondary metabolism, a summarized diagram was constructed to evaluate several metabolic branch points by integrating transcriptomic and metabolomic data generated from the same developing grape berries (**Figure 8**). The diagram overlays the concentrations of AAs, phenolic acids and flavonoid compounds in developing grapes with the expression levels of the transcripts known to be responsible for the enzymatic steps in the metabolic pathway between primary and secondary metabolism (**Table S10**). This integrated metabolic pathway focused on the branch point at which Shikimic acid could be either utilized toward the synthesis of hydrolysable tannins or toward the synthesis of chorismate, which serves as substrate for multiple downstream metabolic processes that include the synthesis of auxin from Trp or the synthesis of Tyr or Phe. Tyr in turn serves as a substrate for either the synthesis of the lipophilic antioxidants, tocopherol, or the synthesis of hydroxycinnamic acids from tyramine. Phe on the other hand is an aromatic AA that serves as a precursor for the synthesis of several secondary metabolites such as phenolic acids and flavonoid compounds that could serve as antioxidant molecules under abiotic stress conditions.

The synthesis of higher levels of hydroxycinnamic acids in green grapes were facilitated by both the upregulation of genes encoding the catabolism enzymes of Tyr (VIT\_07s0005g04480,

FIGURE 7 | A summarized schematic representation of the four amino acid metabolite and transcriptomic networks analyzed in this study. Enzymatic steps are indicated as black arrowed lines, spontaneous (non-enzymatic) metabolic processes are indicated by gray arrowed lines. (A) The network representing the various (Continued)

#### FIGURE 7 | Continued

pathways involved in Gly synthesis. (B) The superpathway of Lys, Met, and Thr synthesis from Asp. (C) The superpathway of Trp, Phe and Tyr synthesis from chorismate. (D) The superpathway of Pro, Arg, and GABA metabolism. Dotted lines represent feedback inhibition loops, whereas striped lines represent catabolic pathways of amino acids not included in this diagram. Blocks indicate the mean-centered log2 fold change of the FPKM expression value of the specific transcript encoding the particular enzymatic step at each berry developmental stage when comparing exposed to control samples. Significant differences between FPKM expression values between exposed and control grapes at a particular developmental stage is indicated by a bold frame around the specific gene. Amino acid concentrations [mg/g fresh weight (FW)] are represented as ANOVA line-plots where significant differences (q ≤ 0.05) between exposed and control grapes are indicated by an asterisks (\*). Line graphs representing exposed and control samples are staggered along the x-axis representing the respective developmental stages. The genes represented by numbers are listed in Table S9.

VIT\_13s0019g04540) and Phe (VIT\_06s0004g02620, VIT\_08s0040g01710, VIT\_13s0019g04460) while upregulation of the same Phe catabolism genes facilitated the accumulation of higher levels of flavonols. The upregulation of a different set of Tyr catabolic enzyme genes (VIT\_00s0394g00040, VIT\_00s0225g00230, VIT\_10s0116g01660, VIT\_12s0028g00710, VIT\_16s0039g01410) simultaneously contributed to the transcription of tocopherols that subsequently lead to the accumulation of elevated lipophilic antioxidant levels (L-ORAC) in green grapes exposed to elevated light (**Figure 8**).

## DISCUSSION

Molecular profiling tools provide sensitive and comprehensive snapshots of how a plant/organ/tissue is responding at a specific point in time. It is quite obvious that the value of these molecular snapshots is amplified if they are framed by an accurate understanding of the environmental cues, the developmental stage and general plant status of the plant. This has lead to a renewed focus on integrating accurate measurements of environmental impact factors with grapevine phenotypes observed, specifically in grapevine berries. Several recent studies have advanced our understanding of berry development, ripening and reactions to stress signals and have convincingly shown that berries throughout their growth curve react to their microclimatic environments, but with different responses (Zenoni et al., 2010; Sweetman et al., 2012; Palumbo et al., 2014; Pilati et al., 2014; Wong et al., 2016). Interestingly, many of these studies also showed the resilience of berries to mitigate mild stresses (Carbonell-Bejerano et al., 2013; Martinez-Luscher et al., 2014; Rienth et al., 2014; Wu et al., 2014; Ghan et al., 2015; Suzuki et al., 2015; Joubert et al., 2016; Santo et al., 2016; Savoi et al., 2016; Young et al., 2016; Sun et al., 2017) leading to minimal impacts on overall berry growth and development. How this is orchestrated/managed was the focus of this study, and an experimental system that was previously proven to render grape berries more exposed to light, with minimal changes in berry temperatures, was used (validation of light as the main experimental parameter in the vineyard experiment was described in Young et al., 2016).

#### The Grape Berry Developmental Profile Remained the Strongest Transcriptional Driver Despite Elevated Light Exposure

Our data confirmed that development remained the strongest driver for the statistical separation of the grape samples based on their transcriptomes, regardless of viticulture treatment implemented. On average, not more than 8% of the berry transcriptome was affected by the elevated exposure at any of the developmental stages evaluated. As expected, berries in the green developmental stages were transcriptionally more similar in the global sense to each other than to berries from the ripening stages. Developmental phase-specific biomarkers were identified as genes that were responsible for the greatest transcriptional differences observed between green and ripening grape berries. Not only were 48 of the biomarkers identified in this study (**Figure S4**) also previously established as biomarkers by other research groups (Zamboni et al., 2010; Palumbo et al., 2014), but all, except nine of these genes, were unaffected by elevated light exposure at the stages when the berries were either green, ripening or throughout development.

#### Green Grapes Maintain Growth and Development by Protecting the Photosynthetic Machinery under Light Stress Conditions

It was previously shown that the exposed grape berries were not different from their control counterparts in terms of size and weight, sugar accumulation and acid degradation patterns (Young et al., 2016) and the transcriptional data also showed that gene expression associated with growth and development, and primary metabolism was not altered by the leaf removal treatment (this study). Despite this fact, photosynthesis-related gene expression, that forms part of primary metabolism, proved to be (the most) significantly affected by the treatment in green grapes.

Our data confirmed that the green berries responded to the increased exposure to try and mitigate the light stress—the first line of defense against potentially damaging effects of photodamage, was the simultaneous activation of several avoidance strategies. One of the strong reactions was the transcription and synthesis of phenolic compounds and tocopherols that were activated, presumably to maintain the redox balance.

Among the phenolic compounds that accumulated at higher levels in response to elevated light were hydroxycinnamic acids and flavonols. Both hydroxycinnamic acids and flavonols can limit photodamage through their ability to scavenge free radicals and ROS, thereby contributing to the maintenance of oxidative homeostasis (Tattini et al., 2005; Agati et al., 2007, 2012, 2013). Flavonols, however, additionally possess the ability to act as sunscreen molecules themselves. They

FIGURE 8 | A summarized overview of the branch points between primary berry metabolism toward the phenylpropanoid pathway overlaying transcriptomic and metabolomic data generated from exposed and control grapes harvested at each phenological stage. Blocks indicate the mean-centered log2 fold change of (Continued)

#### FIGURE 8 | Continued

the FPKM expression value of the specific transcripts and metabolites involved in the particular enzymatic step at each berry developmental stage when comparing exposed to control samples. Significant differential expression (q ≤ 0.05) of genes and compounds are indicated by a bold contour (frame). Total concentrations (µg/g FW) of phenolic acids, flavonols and flavan-3-ols are represented by ANOVA line-plots where significant differences (q ≤ 0.05) between exposed and control samples are indicated by an asterisks (\*). Line graphs representing exposed and control samples are staggered along the x-axis representing the respective developmental stages. Gray circles represented compounds that were not measured, whereas black circles represent various possible compounds at the same enzymatic step. Striped gray arrows represent regulatory steps by associated transcription factors. The genes represented by numbers are listed in Table S10.

achieve this by absorbing highly energetic solar wavelengths, thereby limiting the generation of ROS due to photooxidation. Although flavonol levels have been found to be negligibly low in developing grape berries, the transcription and subsequent accumulation of these compounds in both a light-dependent and development-independent manner have been reported and extensively characterized in grapes (reviewed by Downey et al., 2006; Czemmel et al., 2009; Matus et al., 2009; Malacarne et al., 2016; Yu et al., 2016; Pastore et al., 2017).

The other avoidance mechanism activated in the exposed berries was non-photochemical quenching, the process by which a large part of excitation energy generated by excessive light exposure can be dissipated as heat (via carotenoids). Through this process, the xanthophyll cycle is activated in which the xanthophyll pigment, violaxanthin, is de-epoxidized to zeaxanthin through the activity of the violaxanthin de-epoxidase (VDE) enzyme, thereby limiting energy transfer from LHCII to PSII. Although the carotenoid metabolites and their ratio's, as well as the transcriptional activation and elevated synthesis of the VDE enzyme already confirmed that the green berries have activated the xanthophyll cycle pigments (Young et al., 2016), the transcriptional mechanism of NPQ activation could be further explored in this study. The PsbS subunit of PSII has been established as the enzyme responsible for "sensing" the impending light stress and initiating NPQ (Li et al., 2000; Gregan and Jordan, 2016). The gene encoding the grapevine PsbS enzyme was found to be significantly upregulated by the leaf removal treatment from the onset of green berry development, potentially linking to the activation and upregulation of the VDE enzyme and subsequent increase of the xanthophyll pool as reported in Young et al. (2016).

However, as high levels of light exposure were maintained throughout the season, it appears that damage to the photosynthetic machinery could no longer be avoided through NPQ alone. At the second green developmental stage (EL33); the process of reversible photoinhibition (RPI) was subsequently activated in an attempt to no longer avoid, but rather acclimate to the continuous light stress, while the synthesis of other antioxidant molecules such as tocopherol and flavonols remained transcriptionally and metabolically upregulated (**Figure 8**, **Table S10**). RPI is the process in which photodamage is actively concentrated to the reaction-center binding D1 protein that forms part of Photosystem II (Kyle et al., 1984; Powles, 1984). In doing so, the rapid and ongoing turnover of the D1 protein is ensured through the disorganization of the PSII-LCHII supercomplex in order to remove and replace the damaged D1 protein with a newly synthesized copy. This results in the protection of the photosynthetic machinery from photooxidative stress.

These photoprotective strategies have been well characterized and extensively reported in vegetative tissues (leaves and stems) of numerous plant species (Li et al., 2000; Crouchman et al., 2006; Kato et al., 2012; Niyogi and Truong, 2013; Gorecka et al., 2014). To our knowledge, NPQ and RPI have not been thoroughly investigated in the context of green grape development.

Young et al. (2016) showed higher carotenoid levels (especially xanthophylls) in the exposed berries, yet chlorophyll a: chlorophyll b and total carotene: chlorophyll ratios were maintained in the earlier stages (up until véraison). Total chlorophyll, and the levels of the major photosynthetic carotenoids (β-carotene and lutein) were also not significantly affected. The authors concluded that a pool of carotenoids (predominantly xanthophylls) were responsive to the treatment and increased in response to the increased exposure (light). Since the major carotenoids and chlorophylls were seemingly unaffected, the authors concluded that the increased pool of xanthophylls were able to protect the photosynthetic machinery for normal development to proceed (without damage). The data presented here shows that on a transcriptional level the structural proteins of photosynthesis were significantly upregulated and indicated that there was a higher demand for these proteins possibly due to an increased turnover (damage and repair cycle). Kyzeridou et al. (2015) demonstrated the green fruits of Nerium oleander and Rosa sp. have a higher cyclic electron flow activity around PSI, when compared to leaves. Kotakis et al. (2006) further showed that cyclic electron flow is enhanced (at the expense of the linear photosynthetic electron flow) in twig collenchyma to adjust potential ATP/NADPH ratios and/or to counteract the detrimental effects of hypoxia. This, combined with the increased activity of non-photochemical quenching via the xanthophyll cycle observed in apple (Cheng and Ma, 2004) and grapevine (Young et al., 2016), suggest that non-foliar photosynthesis is possibly required to produce ATP in organs where gas exchange is prevented (Kalachanis and Manetas, 2010).

In this study, the sequential and simultaneous transcriptional activation of light stress mitigation mechanisms proved to be effective in avoiding irreversible photoinhibition and maintaining the development and growth of grapes. This was evident in the global transcriptional responses and the accumulation of AAs that remained predominantly unaffected by the treatment in the green berries. Furthermore, the AAs considered as oxidative stress markers, Pro and GABA, remained unaffected by the treatment in the green grapes despite elevated exposure to light.

This combination of NPQ, RPI and developmentindependent flavonol synthesis, although effective in stress mitigation and acclimation, appeared to be energetically costly to the developing green grapes. Firstly, hydroxycinnamic acids were no longer differentially accumulated in response to elevated light, although the transcription and accumulation of flavonols remained dramatically higher in exposed grapes (**Figure 8**, **Table S10**). This might be explained by the fact that these compounds compete for the same aromatic AAs, Phe and Tyr, as precursors. Results to this effect were previously reported in tomato leaves exposed to various abiotic stresses (Martinez et al., 2016). The authors demonstrated that flavonols were more effective in the maintenance of oxidative homeostasis than hydroxycinnamic acids when precursors were limited. Furthermore, the MYB transcription factors known to regulate the transcription of several enzymatic steps involved in flavonoid synthesis (Czemmel et al., 2009) were significantly upregulated at each of the developmental stages (**Figure 8**). Secondly, the upregulation of several AA catabolic enzymes were further testament to the limitations placed on grape berry energetic resources as a consequence of photoprotection. AAs are involved in highly regulated metabolic networks and are crucial for the synthesis of proteins whilst also acting as precursors for a myriad of downstream metabolic processes. AAs have not only been implicated in normal growth and development but also in stress tolerance as their degradation may provide energetic advantage to maintain stress response mechanisms which prove to be energetically expensive to plant metabolism under suboptimal growing conditions. In Arabidopsis, evidence exist that transcription of AA catabolic enzymes, with the exception of Pro catabolic enzymes, were more sensitive to abiotic stresses than that of the enzymes responsible for AA synthesis (Less and Galili, 2008). Caldana et al. (2011) showed that amino acid catabolism serves as the main cellular energy supply under adverse environmental conditions as inferred by high-density kinetic analysis. The activity of these catabolic enzymes could therefore provide metabolic energy generated from the breakdown of AAs for utilization toward maintaining plant primary metabolism under stressful biotic and abiotic conditions. Additionally, it has been proposed that excessive accumulation of the branched chain amino acids, or rapid protein turnover induced by adverse environmental conditions could potentially lead to cellular apoptosis as a result of respiratory oxidation. The catabolic breakdown of these AAs is seen as a necessary detoxification mechanism under these conditions, as observed in Arabidopsis cell cultures (Taylor, 2004). Since, the branched chain AAs did not accumulate differentially in our investigation (**Figure S5**) we, however, did not consider it the likely metabolic driver for the differential transcription of AA catabolic enzyme encoding genes.

Genes characterized in one of the aforementioned studies (Less and Galili, 2008) were utilized to identify homologous grapevine genes and their expression analysis in our investigation yielded similar results to previous reports. Transcription of the enzymes responsible for AA synthesis was predominantly unaltered by the elevated light exposure treatment whereas genes encoding the AA catabolic enzymes were far more sensitive to the treatment in comparison (**Figure 7**).

The catabolism of AAs during the green berry developmental stages therefore could have provided the green grapes with substrates necessary for downstream metabolic reactions when energetically costly abiotic stress protection mechanisms were simultaneously activated. These included the maintenance of nitrogen fixation that lead to slightly shifted substrate utilization and lower accumulation of Asn, Asp, and Gln levels. The accumulation of lower levels of Phe that serves as the precursor for flavonols necessarily synthesized to protect the grapes against elevated light, were also evident, similar to the mechanisms implemented by vegetative plant organs.

Significantly higher concentrations of Gly in response to the light treatment further substantiate the notion that green grapes respond to light stress as vegetative, source organs. Gly and the enzymes responsible for its decarboxylation, Gly decarboxylase complex (GDC) play an integral part in the successful functioning of photorespiration system. Increased photosynthesis and subsequent elevated levels of electron flow through the photosystems as a means to protect the photosynthetic machinery from light stress, is proposed to cause an altered redox state that ultimately influences the rate of photorespiration (Hutchison et al., 2000; Wingler et al., 2000; Voss et al., 2013). Despite elevated expression levels of the GDC encoding genes reported in our investigation (**Figure 7A**), the GDC themselves are prone to oxidation, hereby causing the accumulation of Gly under high light. Furthermore, Gly is considered to be the rate-determining compound in the synthesis of the antioxidant, glutathione, that might contribute to maintaining the oxidative homeostasis within the developing grape berry. This effect that elevated light exposure had on photorespiration and subsequent high Gly accumulation were previously reported in Arabidopsis (Caldana et al., 2011; Florian et al., 2014). To further support this proposed link between Gly and protection of the photosynthetic machinery in green grapes, the difference in the concentration of Gly when comparing exposed to control grapes become less significant as photosynthetic activity declines throughout berry development.

These findings established that green grapes responded to elevated light exposure by activating and refining stress mitigation strategies to predominantly protect the photosynthetic machinery similar to vegetative plant organs. In an attempt to prioritize growth and development, green grapes utilized and combined several precursor substrates and mechanisms to maintain photoprotection and the synthesis of flavonols, regardless of limited energetic resources.

# Ripening Berries Do Not Effectively Mitigate the Effects of Light Stress

Véraison is the grape developmental stage during which the berry begins to transition from being a photosynthesizing, organ toward becoming a senescing organ while it retains metabolic characteristics of both berry developmental phases on a transcriptional level, as reported here. Véraison has further been extensively characterized by an oxidative burst that includes the production of ROS (particularly H2O2) that serves as a signaling molecule to signify the initiation of the ripening (Pilati et al., 2007). It would be reasonable to expect that the production of low-levels of H2O<sup>2</sup> as a consequence of light stress along with this developmentally driven oxidative burst could culminate toward a redox imbalance in berries exposed to elevated light. In contrast, the grapes that were exposed to elevated light at véraison did not accumulate higher levels of the known stress markers, Pro and GABA, however, at EL38, when the grapes were no longer photosynthetically active, these stress markers did accumulate at higher levels in exposed grapes. It would therefore be reasonable to speculate that this could be a reflection of the berries' successful limitation of the accumulation of ROS through the combination of NPQ, RPI and flavonol production until véraison (**Figures 4**, **6**, **8**).

The rapid accumulation of both Pro and the non-protein AA, GABA, have been extensively reported in plants exposed to abiotic stresses and the metabolism of these AAs are intimately linked (**Figure 7**). Pro has been shown to enhance primary photochemical activity of thylakoid membranes by limiting photoinhibition and its synthesis is highly sensitive to light (Alia et al., 1997). Furthermore, in grapevine leaves, it has been reported that Pro has the ability to limit inactivation of some antioxidant enzymes while further being capable of stimulating the expression of others (Agudelo-Romero et al., 2013). Therefore, the importance of Pro homeostasis, as opposed to its accumulation, in response to oxidative stress has gained particular interest in the context of plant abiotic stress response (Kavi Kishor and Sreenivasulu, 2014). The homeostasis of Pro levels was found to be imperative to actively dividing plant cells to sustain growth despite exposure to long-term stress. GABA, on the other hand, is capable of either contributing to plant abiotic stress response through its involvement as either a stress signal amplifier or in the maintenance of the carbon: nitrogen ratio under stressful conditions (Barbosa et al., 2010; Kinnersley and Turano, 2010). The accumulation of elevated levels of both Pro and GABA can therefore be associated with plants experiencing abiotic stress symptoms.

Similar to the earlier green developmental stages, the maintenance of photoprotective mechanisms throughout most of the berry development comes at an energetic cost to the grapes that are at this stage no longer accumulating precursors and energy at the rate that photosynthesizing organs are able to. This energetic strain on the grapes are reflected in lower levels of almost half of the AAs measured in these grapes as well as lower total AA concentrations overall measured in the grapes exposed to elevated light.

The transcription and accumulation of flavonols remained elevated in an attempt to protect the berries from light damage and at this stage, the antioxidant pool available to the ripe berries were additionally supplemented by higher levels of apocarotenoid accumulation as reported earlier (Young et al., 2016). Due to significantly higher transcription involved in photosynthesis-related proteins during the early developmental stages, combined with increased carotenoids provides a larger pool of substrates for the degradation via carotenoid cleavage enzymes (CCDs). This leads to an increased apocarotenoid pool in the later stages. Although these compounds are thought of as mere degradation products or volatile impact odorants; they also function as antioxidants and it is speculated that apocarotenoids may play an important signaling role in plant development and in responses to environmental stimuli (Avendaño-Vázquez et al., 2014; Hou et al., 2016).

Similarly, we hypothesize that higher concentrations of several AAs at EL38 (**Table 1**) in response to elevated light exposure may not be a consequence of transcription of the related biosynthetic enzyme genes at this late developmental stage, but rather due to the systematic degradation of higher protein levels synthesized during early development. The degradation of higher protein levels could therefore liberate higher levels of the respective AA constituents. The dramatic and consistent upregulation of numerous heat shock proteins throughout berry development (**Table S7**) further supports this hypothesis because of their wellestablished role as molecular chaperones associated with protein recycling in response to abiotic stress in other plant models as reviewed in Wang et al. (2004).

This systematic shut-down of the protection strategies as the grapes reach maturity were further evident by the fact that the lipophilic antioxidant capacity (L-ORAC) of these grapes were no longer elevated significantly and that Pro and GABA levels were significantly higher in exposed compared to control grapes at this stage. Although the oxidative homeostasis of these grapes were no longer entirely intact (as evident by elevated Pro and GABA levels), it is however important to consider that despite the lightinduced stress status of these grapes at EL38, the sole purpose of the fruit had been achieved in the successful development and maturation of the grape seed. The redox-balance and stress responses of the grape berry were no longer of critical importance to the final development of the fruit as evident by the fact that the exposed and control grapes were not physically distinguishable when they were ripe.

# CONCLUSION

In this study, we aimed to determine how developing Sauvignon blanc grapes manage to maintain primary metabolism and development despite being exposed and responding to nonlethal light stress. Our approach was to explore the global transcriptional response of grapes sampled from a highly characterized vineyard to determine how these grapes acclimated to light stress on a transcriptional level and to elucidate the metabolic consequences of these transcriptional changes. This approach allowed us to demonstrate that a leaf removal treatment in the berry bunch zone of developing Sauvignon blanc grape berries lead to the activation and refinement of several stress avoidance and tolerance strategies in parallel for the purpose of mitigating the effects of light stress whilst maintaining the normal developmental program of the grapes.

These results revealed that photosynthetically active berries are successful at mitigating the effects of light stress much like other vegetative plant organs by potentially limiting the synthesis and distribution of potentially harmful ROS through the continuous turnover of the photosynthetic machinery and the production of light-absorbing flavonoid compounds as well as higher levels of carotenoids in green berries and subsequent apocarotenoids in ripe berries. These grapes achieved a state of acclimation through the redistribution of energy resources in the form of AA catabolism that provided energy precursors and substrates that contributed to the maintenance of these energetically costly stress mitigation mechanisms. To this end, green, photosynthesizing grapes maintain growth and development at all costs to protect the development and maturation of the grape seed.

#### AUTHOR CONTRIBUTIONS

MV and PY conceptualized and planned the study. PY implemented and maintained the viticultural treatments and was responsible for the berry sampling. KdP performed RNA processing and RNASeq data analysis. HE performed HPLC analysis for AAs and phenolic compounds. KdP, PY, and MV drafted the original manuscript and all authors contributed and finalized the publication.

#### ACKNOWLEDGMENTS

The authors would like to acknowledge the following people for their invaluable contributions to this research: Dr. Dan Jacobson, Dr. Erik Alexandersson, Ms. Zelmari Coetzee, Mr. Cobus Steyn for assistance with the viticultural treatments and/or sampling, Dr. Liezel Gouws, Ms. Varsha Premsagar, and Dr. Jay Belli-Kullan for assistance with sample processing; Prof. Mario Pezzotti, Prof. Massimo Delledonne, and Dr. Alessandra Dal Molin for her contribution with analysis of the raw RNASeq data and differential expression analysis; Dr. Marianna Fasoli for guidance and support with the RNASeq data analysis; Mr. Fanie Rautenbach for assistance with L-ORAC assays; Ms. Chandré Joubert for her assistance with statistical analyses. This research was made possible by the financial contributions from Wine Industry Network for Expertise and Technology (Winetech), The South African Department of Science and Technology (DST), the Technology and Human Resources for Industry Program (THRIP) and the South African National Research Foundation (NRF).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2017. 01261/full#supplementary-material

Figure S1 | Pearson correlation matrix representing the entire transcriptomes of the initial 24 samples representing 3 biological replicates from control and exposed grapes at four phenological stages.

Figure S2 | ReviGO analysis output of GO enrichment data generated from the 5050 genes in the grapevine genome that was not expressed whatsoever in the grapes investigated in this study.

Figure S3 | Summarized results generated from Real-time PCR analysis.

Figure S4 | Venn diagram comparing the molecular biomarkers generated in this study to previously published biomarkers from Zamboni et al. (2010) and Palumbo et al. (2014) and differential expression analysis of biomarkers shared between this investigation and previously published biomarkers.

Figure S5 | The amino acid super pathway of Ile, Val, and Leu biosynthesis.

Table S1 | Primers used for Real-time PCR.

Table S2 | A table summarizing the retention times of phenolic compounds measured.

Table S3 | Summary of RNASeq reads and mapping.

Table S4 | List of positive and negative molecular biomarkers separating green (EL31 and EL33) from ripening (EL35 and EL38) berries.

Table S5 | Table listing the genes most significantly up and downregulated at each developmental stage (−2 > Log2FC > 2).

Table S6 | Table listing all significantly differentially expressed genes (q ≤ 0.05; 1.5 ≤ Log2FC ≤ −1.5) significantly correlated to predetermined gene expression clusters according to STEM analysis.

Table S7 | Functional annotation (Grimplet et al., 2012) of each of the genes that were highly upregulated (2 ≤ Log2FC ≤ −2) between two or more phenological stages indicated in color as represented in Figure 5. Q-values represent the level of significant difference between the expression of each indicated gene at the specific developmental stage. Asterisks (∗) indicate multiple genes represented by the same functional annotation with Q-values in this case indicative of the average value of the multiple genes sharing the same function.

Table S8 | The amino acid concentrations of all the exposed and control grapes sampled from EL31, EL33, EL35, and EL38. D.N.Q. refers to AA concentrations that were detected but were at concentrations below the limit of quantification.

Table S9 | Genes involved in amino acid synthesis and catabolism as indicated by the numbers assigned in Figure 7.

Table S10 | The metabolite concentrations and gene expression levels as indicated by the numbers assigned in Figure 8.

#### REFERENCES


important role of polyamine catabolism. Plant Physiol. Biochem. 67, 105–119. doi: 10.1016/j.plaphy.2013.02.024


on Cabernet Sauvignon, Nero d'Avola, Raboso Piave and Sangiovese Vitis vinifera L. cultivars. Sci. Hortic. (Amsterdam). 218, 147–155. doi: 10.1016/j.scienta.2017.01.048


L.) wines: a chemical and sensory investigation. J. Sci. Food Agric. 96, 915–926. doi: 10.1002/jsfa.7165


for identification of biochemically interesting compounds using OPLS class models. Anal. Chem. 80, 115–122. doi: 10.1021/ac0713510


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 du Plessis, Young, Eyéghé-Bickong and Vivier. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Light-induced Variation in Phenolic Compounds in Cabernet Sauvignon Grapes (Vitis vinifera L.) Involves Extensive Transcriptome Reprogramming of Biosynthetic Enzymes, Transcription Factors, and Phytohormonal Regulators

Run-Ze Sun1,5, Guo Cheng1,6, Qiang Li1,7, Yan-Nan He<sup>3</sup> , Yu Wang1,2, Yi-Bin Lan1,2 , Si-Yu Li1,2, Yan-Rong Zhu<sup>1</sup> , Wen-Feng Song<sup>1</sup> , Xue Zhang<sup>1</sup> , Xiao-Di Cui<sup>1</sup> , Wu Chen<sup>4</sup> and Jun Wang1,2 \*

#### Edited by:

José Tomás Matus, Centre for Research in Agricultural Genomics, Spain

#### Reviewed by:

Rodrigo Loyola, Pontificia Universidad Católica de Chile, Chile Chiara Pastore, University of Bologna, Italy

#### \*Correspondence:

Jun Wang jun\_wang@cau.edu.cn

#### Specialty section:

This article was submitted to Crop Science and Horticulture, a section of the journal Frontiers in Plant Science

Received: 01 December 2016 Accepted: 27 March 2017 Published: 19 April 2017

#### Citation:

Sun R-Z, Cheng G, Li Q, He Y-N, Wang Y, Lan Y-B, Li S-Y, Zhu Y-R, Song W-F, Zhang X, Cui X -D, Chen W and Wang J (2017) Light-induced Variation in Phenolic Compounds in Cabernet Sauvignon Grapes (Vitis vinifera L.) Involves Extensive Transcriptome Reprogramming of Biosynthetic Enzymes, Transcription Factors, and Phytohormonal Regulators. Front. Plant Sci. 8:547. doi: 10.3389/fpls.2017.00547 <sup>1</sup> Center for Viticulture and Enology, College of Food Science and Nutritional Engineering, China Agricultural University, Beijing, China, <sup>2</sup> Key Laboratory of Viticulture and Enology, Ministry of Agriculture, Beijing, China, <sup>3</sup> College of Enology, Northwest A&F University, Yangling, China, <sup>4</sup> CITIC Guoan Wine Co. Ltd., Xinjiang, China, <sup>5</sup> Key Laboratory of Plant Resources and Beijing Botanical Garden, Institute of Botany, Chinese Academy of Science, Beijing, China, <sup>6</sup> Grape and Wine Research Institute, Guangxi Academy of Agricultural Sciences, Nanning, China, <sup>7</sup> Institute of Vegetables and Flowers, Chinese Academy of Agricultural Sciences, Beijing, China

Light environments have long been known to influence grape (Vitis vinifera L.) berry development and biosynthesis of phenolic compounds, and ultimately affect wine quality. Here, the accumulation and compositional changes of hydroxycinnamic acids (HCAs) and flavonoids, as well as global gene expression were analyzed in Cabernet Sauvignon grape berries under sunlight exposure treatments at different phenological stages. Sunlight exposure did not consistently affect the accumulation of berry skin flavan-3-ol or anthocyanin among different seasons due to climatic variations, but increased HCA content significantly at véraison and harvest, and enhanced flavonol accumulation dramatically with its timing and severity degree trend. As in sunlight exposed berries, a highly significant correlation was observed between the expression of genes coding phenylalanine ammonia-lyase, 4-coumarate: CoA ligase, flavanone 3-hydroxylase and flavonol synthase family members and corresponding metabolite accumulation in the phenolic biosynthesis pathway, which may positively or negatively be regulated by MYB, bHLH, WRKY, AP2/EREBP, C2C2, NAC, and C2H2 transcription factors (TFs). Furthermore, some candidate genes required for auxin, ethylene and abscisic acid signal transductions were also identified which are probably involved in berry development and flavonoid biosynthesis in response to enhanced sunlight irradiation. Taken together, this study provides a valuable overview of the light-induced phenolic metabolism and transcriptome changes, especially the dynamic responses of TFs and signaling components of phytohormones, and contributes to the further understanding of sunlight-responsive phenolic biosynthesis regulation in grape berries.

Keywords: Vitis vinifera, sunlight exposure, phenolic compounds, transcriptome, transcription factor, phytohormone signaling

## INTRODUCTION

fpls-08-00547 April 19, 2017 Time: 17:49 # 2

Phenolic compounds, mainly hydroxycinnamic acids (HCAs) and flavonoids, are one of the most abundant secondary metabolites in grape (Vitis vinifera L.) berries and important to wine quality. HCAs accumulated in grape berry skin and flesh are p-coumaric, caffeic, ferulic, sinapic acid and their derivatives, usually in the form of esters (Baderschneider and Winterhalter, 2001). Three major classes of flavonoid compounds found in grapes include proanthocyanidins (PAs), anthocyanins and flavonols. PAs, also named condensed tannins, are polymers of flavan-3-ol monomeric units (such as catechin, epicatechin, epicatechin-3-O-gallte, and epigallocatechin) which located in both the grape skins and the seeds, with trace amounts also accumulated in the vasculature of berries, whereas flavonols and anthocyanins are detected only in berry skins (Downey et al., 2006). All these compounds have important physiological functions in diverse aspects of grape berry development, such as free radical scavenging, pigmentation and co-pigmentation, ultraviolet (UV) radiation protection and defense against microbial and fungal infections (Harborne and Williams, 2000; Winkel-Shirley, 2001). Furthermore, their contribution to the color, bitterness, astringency, and antioxidant properties of red wine and potential benefits for human health have gained much attention on elucidating the regulatory mechanism of phenolic biosynthesis in grapes over the years (Santos-Buelga and Scalbert, 2000; Conde et al., 2007). Phenolic compounds are derived from multiple branches of the phenylpropanoid biosynthetic pathway, one of the secondary metabolic routes well-characterized in diverse plant species (Hahlbrock and Scheel, 1989; Ferrer et al., 2008). The genes encoding the enzymes of the phenolic biosynthesis pathway in grapes have been isolated (Sparvoli et al., 1994) and also predicted from the complete genome sequence (Velasco et al., 2007; Da Silva et al., 2013), nearly all of which are composed of small gene families. The transcriptional regulation of some structural genes is mainly controlled by a ternary complex (MBW) involving transcription factors (TFs) from the R2R3-MYB, MYC-like basic helix-loop-helix (bHLH), and tryptophan-aspartic acid repeat (WDR, also known as WD40) proteins in several plant species, including V. vinifera (Hichri et al., 2011). Additional potential regulators of the phenolic biosynthesis pathway have also been identified in model and crop plants, such as the Arabidopsis WRKY, MADS (MCM1, Agamous, Deficiens, serum response factor) box and bZIP (basic domain/leucine zipper) TFs, as well as the maize R-Interacting Factor 1 (RIF1), an EMSY-related protein interacted with a certain bHLH protein, ZmR (Hichri et al., 2011). Furthermore, several other R2R3-MYB and single-repeat R3-MYB proteins, such as Arabidopsis AtMYBL2 and CPC (CAPRICE), gentian GtMYB1R1 and GtMYB1R9 and strawberry FaMYB1, act as transcriptional repressors which negatively regulate the biosynthesis of anthocyanins or PAs in plants (Aharoni et al., 2001; Matsui et al., 2008; Zhu et al., 2009; Nakatsuka et al., 2013). In the case of grapevine, VviMYB4a and its close homolog VviMYB4b have been characterized as important negative regulators of small-weight phenolic biosynthesis, whereas two other repressors, VviMYBC2-L1 and VviMYBC2-L3, were shown to fine tune flavonoid levels additionally (Huang et al., 2014; Cavallini et al., 2015).

Environmental factors (light, temperature, water status, and nutrients, etc.) and viticulture practices have been acknowledged to influence the development, ripening and phenolics composition of grape berries, and could thereby affect wine quality (Jackson and Lombard, 1993; Downey et al., 2006). Bunch shading and exposure treatments are regarded as influential practices that alter the accumulation and composition of phenolics and the expression of the corresponding biosynthetic genes by directly affecting the incidence of light on grape clusters and also changing other microclimatic aspects, such as temperature and humidity (Ristic et al., 2007; Koyama and Goto-Yamamoto, 2008; Chorti et al., 2010). Many studies have shown that artificial bunch shading resulted in greatly decreased flavonol concentrations, while the levels of PAs and anthocyanins were not significantly changed at harvest (Downey et al., 2004; Fujita et al., 2005; Cortell and Kennedy, 2006; Koyama and Goto-Yamamoto, 2008). On the other hand, enhanced sunlight exposure induced by basal leaf removal generally led to increased accumulation of flavonols, but did not alter anthocyanin concentration compared with the control, which might be correlated with the negative effects of elevated berry skin temperature (Downey et al., 2004; Chorti et al., 2010). In addition, grapes from sunlight exposure bunches had a higher proportion of B-ring trihydroxylation subunits within PAs and anthocyanins in comparison with normal and bunch shading fruit, which agree with the relative increase of flavonoid 3<sup>0</sup> ,50 -hydroxylase (VviF3<sup>0</sup> 5 <sup>0</sup>H) expression (Cortell and Kennedy, 2006; Koyama and Goto-Yamamoto, 2008).

In recent years, a considerable amount of effort has been devoted to investigating the impact of cluster sunlight exposure treatments during specific stages of berry development and ripening on the detailed phenolic profiles as well as the expression of related structural and regulatory genes in different grape varieties (Matus et al., 2009; Chorti et al., 2010; Lemut et al., 2011; Kotseridis et al., 2012; Lee and Skinkis, 2013; Matsuyama et al., 2014; Wu et al., 2014; Friedel et al., 2015). For instance, the expression of flavonol synthase 1 (VviFLS1, also known as VviFLS4) and its specific transcriptional activator VviMYB12 (also named VviMYBF1) was drastically increased following leaf removal treatment, which ultimately resulted in the quickly increased flavonol synthesis (Matus et al., 2009). Leaf removal also up-regulated anthocyanin synthesis related structural genes and regulators in grape skins, such as chalcone synthase (VviCHS), uridine diphosphate (UDP)-glucose:flavonoid 3-Oglucosyltransferase (VviUFGT), anthocyanin-O-methyltransferase (VviAOMT), flavonoid 3<sup>0</sup> -hydroxylase (VviF30H), VviF3<sup>0</sup> 5 0H, VviMYBA1, and VviMYB5a (Matus et al., 2009; Matsuyama et al., 2014; Wu et al., 2014). More recently, two bZIP TFs elongated hypocotyl 5 protein (HY5) orthologs, VviHY5 and VviHYH, were characterized as constituents of the UV-B response pathway in grapevine and mediated flavonol accumulation in response to high radiation exposure (Loyola et al., 2016; Matus, 2016). However, there are still pending questions regarding the complex underlying molecular mechanism of the phenolic metabolism regulation network involved in light response. In the present

study, accumulation and compositional changes of HCAs, flavan-3-ols, anthocyanins and flavonols were determined in V. vinifera L. cv. Cabernet Sauvignon grape berries from different fruit-zone light-exposure treatments in multiple phenological stages under field conditions over three successive seasons. To understand the regulation of phenolic biosynthesis under different irradiation conditions, the influences of light exposure on the transcription of phenolic biosynthetic genes and their putative upstream regulators, as well as the relationship between metabolism and transcription in grapes throughout berry development were also examined.

# MATERIALS AND METHODS

# Plant Material and Sunlight Exposure Treatment

Field experiments were conducted in a commercial vineyard of V. vinifera L. Cabernet Sauvignon located in Manas Country (44◦ 17<sup>0</sup> North, 86◦ 12<sup>0</sup> East, 475 m above sea level), the wine-producing region of Xinjiang province, China, for three consecutive growing seasons (2011, 2012, and 2013). The ownrooted vines in this vineyard were planted in 2000, managed on a modified Vertical-Shoot-Positioned (M-VSP) trellis system with a spur-pruned cordon retaining 15 nodes per linear meter, arranged in north-south rows with 2.5 m × 1 m vine spacing and equipped with a furrow irrigated system. Nutrition and pest management was carried out according to industry standards for this cultivar and the region as previously described (Cheng et al., 2014).

Sunlight exposure treatments were carried out as described by Matus et al. (2009), with some modifications. In three consecutive years, eight fruit-zone light exposure levels were established in the vines through artificial leaf removal, half leaf removal, or leaf moving (**Figure 1**): leaf removal at berry pepper-corn size (LR-PS); leaf removal at véraison (LR-V); leaf removal after véraison (LR-AV); half leaf removal at véraison (HLR-V); half leaf removal after véraison (HLR-AV); leaf moving at véraison (LM-V); leaf moving after véraison (LM-AV); and non-treated control (C). Leaf removal and half leaf removal treatments were carried out by removing the first one to six basal leaves from the main shoots with clusters and three basal leaves from the first, third, and fifth of each shoot with clusters, respectively. For leaf moving treatment, the first one to six basal leaves of each shoot with clusters were moved aside by the use of nylon zipties, in order to increase the sunlight exposure of grape clusters without affecting the photosynthetic carbon assimilation to the fruit. Each treatment was arranged in a completely randomized experimental design with three biological replicates. In each biological replicate, treatment was applied to 15 vines randomly selected from the vineyard's south and north sites.

The meteorological data during berry development in 2011, 2012, and 2013, including sunlight duration (h), growing degree days (◦C), temperature (◦C), rainfall (mm) and relative humidity (%), were gathered from the local meteorological administration (Supplementary Table S1). To determine the influence of sunlight exposure on canopy microclimatic conditions, photosynthetically active radiation (PAR) sensor (model S-LIA-M003, Onset Computer Corporation, Bourne, MA, USA) and total radiation sensor (model S-LIB-M003, Onset Computer Corporation, Bourne, MA, USA) were positioned parallel with the cordon at the bunch zone on the defoliated side (west) of the canopy of both exposure and control groups during grape berry development in 2012 and 2013. The air temperature and relative humidity (RH) inside the canopy of each group were also monitored via a Hobo temp/RH smart sensor (model S-THB-M002, Onset Computer Corporation, Bourne, MA, USA) placed at the fruit zone. Each measurement was performed at 5-min intervals (Supplementary Table S2).

Berries from each treatment and control group were sampled at the following developmental time points: 3 weeks after flowering (waf) (berry pepper-corn size; E-L 29), five waf (berry pea-size, E-L 31), seven waf (berry still hard and green, E-L 33), early-véraison (berries begin to color, E-L 35), mid-ripening stage (berries with intermediate Brix values, E-L 36), end of véraison (berries not quite ripe, E-L 37) and complete ripening stage (E-L 38) (Coombe, 1995). For each biological replicate, 600 berries were randomly separated from both sunny and shade sides of at least 100 clusters within 15 vines. The sampling time was fixed at 10:00 to 11:00 am, and three biological replicates were collected with the same method at each sampling date. After being washed with distilled water, a sub-sample of 100 berries from each biological replicate was subjected to the physiological measurements, including berry fresh weight, total soluble solids (TSS) content and titratable acidity (TA), the rest were frozen in liquid nitrogen immediately and transported to the lab in dry ice for the subsequent metabolites determination or transcriptional analysis. TSS concentrations of the juices were measured with digital pocket handheld refractometer (Digital Hand-held Pocket Refractometer PAL-1, Atago, Tokyo, Japan), and TA was determined by titration with NaOH to the end point of pH 8.2 and expressed as tartaric acid equivalent (Cheng et al., 2014).

# Isolation and Identification of Compounds

Phenolic acids were extracted from berries and analyzed as described by Song et al. (2013, in Chinese with English abstract). In detail, a sub-sample of 100 frozen berries randomly selected from each biological replicate was first ground into powder under liquid nitrogen after weighing and removing the seeds. For the analysis of monomeric phenolics, 5 g of ground powder was extracted with 25 mL of 1% (v/v) ascorbic acid and 10 mM EDTA in 4 M NaOH. The extraction mixture was then sonicated for 3 min and shaken in incubator shakers in dark for 8 h under a nitrogen atmosphere at 35◦C. After acidification to pH 2 using 6 M HCl and centrifugation at 8,000 rpm for 20 min, the clear supernatant was extracted four times with diethyl ether to obtain the free phenolic acids released from the soluble ester. The combined supernatant was evaporated to dryness, dispersed in 0.5 mL of methanol, and filtered through a 0.45 µm Millipore membrane filter (Millipore Co. Ltd, Billerica, MA, USA) prior to high performance liquid chromatography (HPLC) analysis.

Skin flavonoids were extracted from a sub-sample of 100 berries randomly selected from each biological replicate as described by Li et al. (2014), with some modifications. For flavan-3-ols preparation, 0.1 g sub-sample of skin powder was extracted in a solution of 50 g/L phloroglucinol (1 mL) containing 0.3 N HCl and 0.5% (v/v) ascorbic acid in darkness at 50◦C for 20 min. After terminating the reaction by addition of 1 ml NaAc (50 mM), the extraction mixture was centrifuged at 10,000 rpm for 15 min, and the clear supernatant was collected. The residues were re-extracted three times, and all the supernatants were mixed and stored at −40◦C. To extract anthocyanins, 0.5 g sub-sample of skin powder was extracted in 10 mL of methanol solution containing 1% formic acid under sonication for 10 min at room temperature, and then shaken in incubator shakers in dark at 25◦C for 30 min at a rate of 200 rpm. The extraction mixture was centrifuged at 8,000 rpm for 20 min and the clear supernatant was collected. The residues were re-extracted four times, and all the supernatants were pooled and evaporated to dryness in a rotary evaporator at 30◦C, and then dissolved in 10 mL of 10.8% (v/v) acetonitrile aqueous solution with 2% formic acid. All the extracts obtained above were filtered through 0.22 µm nylon membrane filters before HPLC analysis. For flavonols extraction, 5 g subsample of skin powder was immersed in 15 mL of 50% ethanol solution containing 1% acetic acid with the aid of ultrasonic vibrations for 35 min at room temperature and then centrifuged at 8,000 rpm for 10 min. The residues were re-extracted four times, and the pooled supernatants were macerated with 50 mL of distilled water and then extracted in 40 mL of ethyl acetate three times. The organic phase was collected and evaporated to dryness in a rotary evaporator at 30◦C, and then suspended in 2 mL of 25% methanol. Three independent extractions from three biological repeats were conducted for either the berry or the skin of each sample.

Phenolic acids were monitored on an Agilent 1100 series HPLC-MSD trap VL (Agilent, Santa Clara, CA, USA), equipped with a diode array detector (DAD) and a reversed phase column (Zorbax SB-C18, 250 × 4 mm, 5 µm). The injection volumes were 10 µL and the column thermostat was set at 30◦C. Mobile phase A consisted of methanol/acetic acid/water (10:2:88, v/v/v), and mobile phase B consisted of methanol/acetic acid/water

(900:15:85, v/v/v). The gradient was from 0 to 3.6% B for 7 min, from 3.6 to 15% B for 19 min, from 15 to 25.5% B for 6 min, from 25.5 to 29.7% B for 3 min, from 29.7 to 45.5% B for 10 min, from 45.5 to 0% B for 8 min, at a flow rate of 1 mL/min. All phenolic acid compounds were identified by matching the retention time and their spectral characteristics against those of standards (Sigma, St. Louis, MO, USA). Chlorogenic acid and caffeic acid were quantified at 325 nm while p-coumaric acid, ferulic acid and sinapic acid at 275 nm. The conditions for the mass spectrometry (MS) were as follows: electrospray ionization (ESI) interface; negative ion model; nebulizer pressure, 241.3 kPa; dry gas flow rate, 10 L/min; dry gas temperature, 350◦C; Trap ion charge control (ICC), 30,000 units; collision-induced dissociation (CID) voltage, 1.00 V; scan at m/z 100–1,000.

Qualitative and quantitative analyses of flavonoids were carried out on an Agilent 1200 series HPLC-MSD trap VL linked simultaneously to a DAD (for flavan-3-ols and anthocyanins) or a variable wavelength detector (for flavonols) as described previously (Cheng et al., 2014; Li et al., 2014; Zhu et al., 2014). Flavan-3-ols, anthocyanins and their derivatives were analyzed as in Cheng et al. (2014) and Li et al. (2014), respectively. Flavonols and their derivatives were eluted by using a selection of reverse phase column (Zorbax SB-C18, 50 × 3 mm, 1.8 µm) and binary gradient elution with mobile phase A consisted of acetonitrile/formic acid/water (50:85:865, v/v/v), and mobile phase B consisted of acetonitrile/methanol/formic acid/water (250:450:85:215, v/v/v/v), which was in accordance with Zhu et al. (2014, in Chinese with English abstract) with minor revision. Proportions of solvent B varied as follows: from 0 to 14.2% for 24.2 min, from 14.2 to 15.7% for 2.8 min, from 15.7 to 18.8% for 6.4 min, from 18.8 to 23.5% for 5.4 min, from 23.5 to 26% for 6 min, from 26 to 27.4% for 2 min, from 27.4 to 32% for 4.6 min, from 32 to 40% for 10.2 min, from 40 to 100% for 6 min, from 100 to 0% B for 10.6 min, at a flow rate of 1 mL/min. The injection volumes were 50 µL and the column thermostat was set at 40◦C. All flavonol compounds were identified by the UV spectrum and retention time of quercetin-3-O-glucoside (Sigma, St. Louis, MO, USA). The detector wavelength was 360 nm. The ESI parameters were as follows: negative ion model, nebulizer pressure, 30 psi; dry gas flow rate, 10 mL/min; dry gas temperature, 325◦C; Trap ICC, 30,000 units; CID voltage, 1.00 V; scan at m/z 100–1,000. Quantitative determination of flavonoids was performed using the external standard method with commercial standards. All analyses were run in replicate and averaged for each biological replicate. One-way ANOVA followed by the Duncan's new multiple range test was performed using SPSS 20.0 for windows (SPSS Inc., Chicago, IL, USA) to determine significant differences of the physicochemical indexes and phenolic accumulations among treatments at each sampling time point.

#### RNA Isolation, Sequencing, and Data Analysis

Based on biochemical parameters and metabolite profiles, we selected the berries of three developmental stages (E-L 36, 37, and 38) from the LR-V and LM-V treatments and the control group during the 2012 growing season to conduct the transcriptome profiling analysis. A sub-sample of 50 berries were randomly selected from each biological replicate for RNA extraction. Total RNAs for RNA-seq analysis were isolated from frozen deseeded berries using a Plant Total RNA Extraction Kit (Sigma, St. Louis, MO, USA), and further purified by DNase I (Promega, Madison, WI, USA) digestion. RNA integrity and concentration were analyzed using the Nanodrop 2000 spectrophotometer (Thermo Fisher Scientific Inc., Wilmington, DE, USA) and the Aglient 2100 Bioanalyzer (Agilent, Santa Clara, CA, USA). Following quality assessment, cDNA libraries constructed from three biological replicates of each sample were sequenced by Illumina HiseqTM2000 sequencer (Illumina Inc., San Diego, CA, USA) with a 50-bp single read module RNA-seq reads and then aligned against the reference grapevine genome V2<sup>1</sup> using the alignment software Bowtie (Langmead et al., 2009), allowing no more than two nucleotides mismatched. The FPKM (expected fragments per kilobase of transcript per million fragments mapped) method was used for calculating the transcript abundance of each gene (Trapnell et al., 2010). Transcripts were mapped to reference canonical pathways in the Kyoto Encyclopedia of Genes and Genomes (KEGG)<sup>2</sup> as described previously (Sun et al., 2015a). Prediction of TFs was performed by using the HMMsearch program. Identification of differentially expressed genes (DEGs) between samples was performed with R package 'NOISeq' (Tarazona et al., 2011). A threshold of fold-change ≥ 2 and divergence probability ≥ 0.8 was used for filtering the significance of the gene expression difference. Heatmap visualizations were performed using the R package 'pheatmap' (Kolde, 2012). Pearson correlation evaluation was conducted with R package 'Hmisc' using the rcorr function (Harrell and Dupont, 2012) and co-expression networks were visualized with the Cytoscape software version 3.2.0 (Shannon et al., 2003).

#### Quantitative Real-time PCR

Validation of the transcript quantification from the RNA-seq data was carried out through quantitative real-time PCR (qRT-PCR). For extraction of skin RNAs, berry skins from another subsample of 50 berries randomly selected from each biological replicate were manually separated from pulps. Total RNA from berry skins was isolated using the same method mentioned above. The subsequent cDNA synthesis and qRT-PCRs were performed as described by Sun et al. (2015a). Gene-specific primers used for qRT-PCR are listed in Supplementary Table S3 (Downey et al., 2003; Castellarin et al., 2006; Fujita et al., 2006; Reid et al., 2006; Bogs et al., 2007; Czemmel et al., 2009; Shimazaki et al., 2011; Azuma et al., 2012; Sun et al., 2015b, 2016). All reactions were run in triplicate, and the normalized relative expression levels of target genes were calculated by 2−δCt (1Ct = CtTarget – CtControl, Ct: cycle threshold). VviUbiquitin1 and Vviβ-Actin genes were selected as endogenous controls for normalization and CtControl was the geometric mean of their threshold cycles.

<sup>1</sup>http://genomes.cribi.unipd.it/grape/

<sup>2</sup>http://www.genome.ad.jp/kegg/

# RESULTS AND DISCUSSION

fpls-08-00547 April 19, 2017 Time: 17:49 # 6

### Berry Development and Ripening

Changes in berry fresh weight, TSS and TA of Cabernet Sauvignon grape berries collected from E-L 31 stage until harvest during the first growing season (2011) and from E-L 29 stage until harvest during the subsequent two growing seasons (2012 and 2013) are shown in Supplementary Figure S1. The date of véraison and harvest were set at approximately eight and sixteen waf, respectively, during all three seasons. Previous studies on several grape species and cultivars reported little or no effect of leaf removal treatments on fruit weights and juice soluble solid contents at harvest (Haselgrove et al., 2000; Main and Morris, 2004; Chorti et al., 2010; Kotseridis et al., 2012). In this study, weight of berries for LR-PS was significantly lower than that of the control at harvest in the 2011 season, while an opposite result was observed in the 2012 season. LM-V and LM-AV significantly increased berry weights from 1 week after each treatment until berry ripening during the 2011 and 2012 seasons, however, there were no discernible differences between each of the two treatments and the control during the third experimental season (Supplementary Table S4). Overall, there were no consistent trends in the differences of berry fresh weight between each of the light-exposure treatment group and the control group during berry development in all the three experimental seasons. Similarly, the differences in the level of juice soluble solids between the exposed and shaded berries during berry development were also inconsistent among different seasons. No treatment differences were found in soluble solid contents at harvest in all the 3 years, except for those of berries from LR-PS, which was higher than that of the control in the 2011 season (Supplementary Figure S1). Specifically, ripening berries from the control and all light-exposure treated clusters have lower soluble solid contents in the 2013 season compared with the other two seasons, which could be attributed to the relative lower level of flowering-to-harvest growing degree days (GDD) accumulation in 2013 than those in the other 2 years (Supplementary Table S1), similar with the results from a previous study (Spayd et al., 2002). The decrease of the TA in the berries from LR-PS treatment during véraison was slightly faster than that of the control during the 2011 and 2012 seasons. However, the influence of LR-PS treatment on TA was negligible during the 2013 season. At harvest, there was no significant difference in the TA among treatments in all the three seasons (Supplementary Figure S1), which was inconsistent with previous studies that leaf removal treatments reduced TA (Zoecklein et al., 1992; Percival et al., 1994), suggesting that TA was differently affected by sunlight exposure depending on cultivar and climate.

#### Influence of Cluster Sunlight Exposure Treatments on Phenolic Concentration and Composition

Of the many environmental factors that affect the phenolic biosynthesis in many plants, light has been regarded as one of the major influences (Downey et al., 2004; Cominelli et al., 2008; Wang et al., 2012; Jaakola, 2013). The present study shows that the concentration of total HCAs was significantly increased in the sunlight exposure berries in comparison with the control berries at both véraison and harvest in all the three experimental seasons, except for berries from LR-PS and LR-AV treatments, in which the level of HCAs was slightly higher during véraison while lower at the harvest stage compared with the control in both the 2011 and 2012 seasons (**Figure 2A** and Supplementary Table S4). The results obtained in our study are inconsistent with a previous study conducted on Pinot Noir, which showed that the concentration of HCAs throughout maturation was effectively enhanced by leaf removal at berry set but slight influenced by leaf removal at véraison (Lemut et al., 2011). Of the three classes of flavonoids, flavan-3-ols are present in the greatest proportion in grapes, followed by anthocyanins, with flavonols being present at relatively low levels (Downey et al., 2004). Sunlight exposure did not consistently affect the accumulation of total flavonoids or flavan-3-ols in the skins throughout berry development or at harvest among different seasons (**Figures 2B**, **3A**). The concentration of total flavonoids and flavan-3-ols in the skins was slightly lower in LR-PS, LR-V, LR-AV, and HLR-V, while the former was higher in HLR-AV and LM-V treated berries than those from control berries at harvest in the 2011 season. Each of the seven sunlight exposure treatments resulted in decrease in concentration of skin total flavonoids as well as flavan-3-ols at harvest in the 2012 season, while an opposite result was observed in the 2013 season, except for berries from LR-PS, in which the concentration of total flavonoids and flavan-3-ols showed a significant decrease and no discernible difference compared with the control, respectively (Supplementary Table S4). The seasonal variations in the level of flavan-3-ol compounds might be caused by differences of temperature, GDD or the level of PAR during berry ripening among seasons (Supplementary Tables S1, S2), which suggests an ambiguous effect of light-exposure on the flavan-3-ol biosynthesis in grape berry skins.

Anthocyanin biosynthesis has been found to be variably regulated in response to light conditions in a number of sunlight exposure studies, which is possibly confounded by varying experimental settings and other factors, such as cultivar, vineyard location, timing of leaf removal, and growing season. For instance, leaf removal at berry set increased skin anthocyanins in grapes of Merlot, Cabernet Sauvignon and Pinot noir (Lemut et al., 2011; Kotseridis et al., 2012), and pre-bloom leaf removal also substantially increased anthocyanin concentration in Barbera, Lambrusco salamino, Graciano, and Carignan grapes when compared with no leaf removal (Poni et al., 2009; Tardaguila et al., 2010). However, another exposure study indicated that no significant differences among treatments were observed in anthocyanin levels in Nebbiolo grapes at harvest, although leaf removal caused a temporary acceleration of anthocyanin accumulation throughout ripening (Chorti et al., 2010). In the present study, sunlight exposure increased the level of skin anthocyanins at the initiation of grape coloration compared to the control in at least two of the three experimental seasons, while no consistent differences among treatments were observed in the concentration of anthocyanins at harvest over the three experimental seasons (**Figure 3B**). In the 2011 season, almost all exposure treatments could significantly (LR-V, LR-AV,

biological replicates. Light gray background represents the phenological phase of véraison from 5 to 100% of colored berries.

HLR-AV, and LM-V) or slightly (LR-PS and HLR-V) increase skin anthocyanin amount at harvest, whereas no significant differences in the concentration of skin anthocyanins at harvest among treatments were observed in the 2013 season. On the contrary, slight or significant decreases in the skin anthocyanin amount were observed from all of the light exposed berries in the 2012 season (Supplementary Table S4). The diametrically opposed result was possibly due to increased levels of PAR, solar radiation and average temperature for exposed berries during coloration and the maturation phase (Supplementary Table S2), accompanied with a relative high temperature during berry ripening in this experimental season (Supplementary Table S1), which could lead to the inhibit formation or induce degradation of anthocyanins (Yamane et al., 2006; Tarara et al., 2008; Pastore et al., 2013).

Among the three major classes of flavonoid compounds, the accumulation of flavonols was most drastically affected in berry skins under the sunlight exposure treatments. Flavonols have been found to accumulate in sun-exposed tissue of grapes and are thought to act as UV protectants and free radical scavengers (Price et al., 1995; Downey et al., 2004). Leaf removal at all three phenological stages resulted in a dramatic increase in flavonol concentration in the grape skin throughout berry development during the three experimental seasons, similar to the results conducted on Sangiovese berries previously (Pastore et al., 2013), but the degree of their effect was variable among seasons in our experiments (**Figure 3C**). The level of flavonol compounds was also moderately increased by HLR-AV and LM-V treatments, except that there was no significant difference between the control and HLR-AV in the 2011 season. HLR-V and LM-AV did not significantly influence the concentration of skin flavonols at harvest compared with the control, although there was a temporary acceleration of flavonol accumulation throughout ripening in HLR-V treated berries (**Figure 3C** and Supplementary Table S4). However, opposite results regarding changes in flavonol contents under different sunlight exposure treatments have been found previously, in which leaf moving at véraison increased flavonol synthesis greater than leaf removal treatment (Matus et al., 2009), suggesting that the enhanced accumulation of flavonols under treatments is strongly associated with the severity degree of sunlight exposure, and also climate divergences in different years and regions.

In the flavonoid biosynthesis pathway, two metabolic branches leading to the biosynthesis of B-ring dihydroxylated (3,3<sup>0</sup> -OH) and trihydroxylated (3,3<sup>0</sup> ,50 -OH) subunits were reported to have different sensitivities in response to various lighting conditions in many previous researches (Downey et al., 2004; Azuma et al., 2012; Guan et al., 2015). Our results showed that the molar ratio of dihydroxylated to trihydroxylated flavonoids was continuously decreased throughout berry ripening, while changes of the ratio among treatments were inconsistent during the three seasons (**Figure 3D**). In the first experimental season, the ratio of dihydroxylated/trihydroxylated flavonoids in berries from almost all of the sunlight exposure treatments was decreased compared with the control across development, except for LR-PS at E-L 33 stage, LR-V at E-L 35 stage and LM-AV at harvest. In contrast, however, light-exposure caused a slight or marked increase in the ratio of dihydroxylated/trihydroxylated flavonoids during berry ripening in the 2012 growing season. In the last experimental season, no significant differences in the ratio of dihydroxylated/trihydroxylated flavonoids were observed among treatments except for berries from the leaf removal treatments, which was increase in LR-V and LR-AV treated berries at harvest and in LR-PS treated berries during coloration (**Figure 3D** and Supplementary Table S4). In contrast to the results obtained from

cluster shading studies (Koyama and Goto-Yamamoto, 2008; Guan et al., 2015), the effect of light-exposure was possibly influenced by the temperature or other climate variables of the year.

# Light-induced Transcriptional Changes of Phenolic Biosynthetic Pathway Genes

To investigate the responses of phenylpropanoid/flavonoid biosynthetic pathway related structural and regulatory genes to different light-exposure treatments, berries of three distinct development stages (E-L 36, 37, and 38) from the LR-V and LM-V treated and the control groups during the 2012 growing season were selected to characterize the changes in gene expression at the transcript level by RNA-seq. Results showed that the general structural genes of phenylpropanoid and flavonoid metabolic pathways, including some members of phenylalanine ammonialyase (PAL, EC 4.3.1.24), 4-coumarate: CoA ligase (4CL, EC 6.2.1.12) and flavanone 3-hydroxylase (F3H, EC 1.14.11.9), were significantly or moderately (fold-change ≥ 2 while divergence probability ≤ 0.8) up-regulated in berries from the LR-V and LM-V treatment groups at E-L 36 and 38 stages while downregulated at E-L 37 stage. Furthermore, almost all members of CHS (EC 2.3.1.74) and chalcone isomerase (CHI, EC 5.5.1.6) were moderately down-regulated in sunlight exposed grapes in comparison with those from the control group across the three developmental stages. The expression of members of specific structural genes required for HCA and flavonol biosynthesis, including cinnamyl-alcohol dehydrogenase (CAD, EC 1.1.1.195) and FLS (EC 1.14.11.23) across the three developmental stages as well as bifunctional UDP-glucose/UDP-galactose:flavonol-3- O-glucosyltransferase/galactosyltransferase (GT6, EC 2.1.1.76) at

E-L 38 stage, and members of dihydroflavonol reductase (DFR, EC 1.1.1.219) and UFGT (EC 2.4.1.115) required for anthocyanin biosynthesis at E-L 36 and 37 stages were significantly or moderately up-regulated in LR-V and LM-V treated berries, leading to the increased accumulation of corresponding phenolic products. The lower contents of flavan-3-ols in LR-V and LM-V treated berry skins compared with the control group at E-L 37 and 38 stages is supported by the moderately down-regulation of members of leucoanthocyanidin reductase (LAR, EC 1.17.1.3) and anthocyanidin reductase (ANR, EC 1.3.1.77), which are directly involved in flavan-3-ol biosynthesis (**Figures 3**, **4A**).

It was previously reported that VviFLS4 was the sole member of the grapevine FLS family which specifically responded to different light regimes and showed a clear expression pattern corresponding to the accumulation of flavonols in the berry skins (Fujita et al., 2006; Matus et al., 2009; Koyama et al., 2012; Pastore et al., 2013). In our study, the transcription of several other members of FLS family in addition to VviFLS4 (VIT\_218s0001g03470) was also drastically induced (fold-change ≥ 2) by LR-V and/or LM-V treatments in different developmental stages, such as VIT\_202s0012g00390, VIT\_202s0012g00400 and VIT\_202s0012g00450 in LR-V treated berries at E-L 36 stage, as well as VIT\_208s0007g00750 and VIT\_213s0067g01020 in LR-V and LM-V treated berries at both E-L 36 and 38 stages (**Figure 4A**). These results were consistent with the increased accumulation of flavonols in lightexposure berries, which indicates that grapevine FLS gene family may be functionally redundant in response to light signal. The hydroxylation pattern of flavonoids is known to be mediated by the enzyme activity of F30H (EC 1.14.13.21) and F3<sup>0</sup> 5 <sup>0</sup>H (EC 1.14.13.88), which catalyze the hydroxylation of naringenin and dihydrokaempferol at the 3<sup>0</sup> and 3<sup>0</sup> 5 <sup>0</sup> positions of the B-ring, respectively (Bogs et al., 2006; Guan et al., 2015). Our results showed that no significant differences in the expression level of genes encoding F30H among treatments were detected at each sampling point, but the transcription abundance of several F30 5 <sup>0</sup>H family members was significantly down-regulated by LR-V and moderately down-regulated by LM-V at both E-L 36 and 37 stages (**Figure 4A**), which were in fair agreement with the higher ratio of dihydroxylated/trihydroxylated flavonoids observed in berries from LR-V and LM-V treatments, in comparison with those in berries from the control group (**Figure 3D**).

To identify additional genes that might contribute to alterations in phenolic metabolism in berries grown under different light conditions, the transcription profile of phenolic biosynthesis-related genes was compared with the HCA, total flavonoid, flavonol, flavan-3-ol, and anthocyanin profiles of all samples, respectively. The correlation analysis based on pearson's coefficient revealed that the expression of four members of PAL (VIT\_216s0039g01100, VIT\_216s0039g01110, VIT\_216s0039g01120, and VIT\_216s0039g01130), the first committed enzyme in phenylpropanoid metabolism (Sparvoli et al., 1994), was highly significantly (p-value ≤ 0.01) correlated with the accumulation of flavonoids in berries from different light treated groups. The transcript of two other genes (VIT\_213s0047g00210 and VIT\_206s0061g00450) belonging to the 4CL and F3H families was also significantly (p-value ≤ 0.05) correlated with the changes of flavonoid content in each samples. In addition, there was a significant correlation (p-value ≤ 0.05) between flavonol accumulation and the expression of two members of FLS family (VviFLS4 and VIT\_208s0007g00750) mentioned above (Supplementary Table S5). However, no specific structural genes for phenolic biosynthesis pathway were found to be correlated with the accumulation of HCAs, flavan-3-ols or anthocyanins, which indicates that the biosynthesis or degradation of these compounds in berries under different light regimes might be controlled by the cooperation of multiple enzymes from the entrance to branches.

## Expression Analysis of Flavonoid Biosynthesis-related Transcription Factors

In grapes, some members of R2R3-MYB TF family and their co-activators belonging to other TF families (bHLH and WDR) which could regulate the transcription of downstream target genes required for phenolic biosynthesis pathways have been isolated and characterized recently (Hichri et al., 2011). The expression of these TFs involved in flavonoid metabolism has also been reported to be induced or suppressed by many environmental factors, such as light quality, temperature and water deficit conditions (Castellarin et al., 2007; Cominelli et al., 2008; Azuma et al., 2012). The transcript level of VviMYBF1, which acts as a direct regulator of VviFLS4 expression (Czemmel et al., 2009; Matus et al., 2009), was moderately greater in berries after light-exposure treatments than those in the control berries, except for a slight decrease in LM-V treated berries at E-L 37 stage compared with the control (**Figure 4B**). Expression of two UV-B-inducible grapevine flavonol synthesis regulators, VviHY5 and VviHYH (Loyola et al., 2016), was also significantly and moderately up-regulated in LR-V and LM-V treated berries at postvéraison berry developmental (E-L 38) stage, respectively. These results were well consistent with an increase in the transcript abundances of members of FLS as well as flavonol concentrations after light-exposed treatments, which indicates that light affects flavonol biosynthesis through transcript activation of a series of TFs and structural genes. Several regulators of the general branch and different branches of flavonoid synthesis, including VviMYB5a, VviMYB5b, VviMYBPA1, VviMYBPA2, VviMYBPAR, VviMYBC2-L1, VviMYBC2-L2, VviMYBC2-L3, and a TTG2-like homolog protein VviWRKY26 (Amato et al., 2016), showed large divergent changes in the transcript levels during berry development or under different light-exposure treatments (**Figure 4B**). Transcript abundances for VviMYB5a and VviMYB5b, VviWRKY26, as well as the negative regulator of PA accumulation VviMYBC2-L1 (Huang et al., 2014) in grape berries presented a high level during the three development stages but did not respond to changing light conditions, while low levels of VviMYBPA2, VviMYBPAR, VviMYBC2-L2, and VviMYBC2-L3 transcripts were detected in berries from all treatment groups. In addition, the expression of VviMYBPA1 was significantly and slightly down-regulated in LR-V and LM-V treated berries

FIGURE 4 | Effects of sunlight exposure on the transcript profile of the (A) enzymes and (B) regulatory factors involved in phenolic biosynthesis in grape berries. PAL, phenylalanine ammonia-lyase; C4H, trans-cinnamate 4-monooxygenase; CCR, cinnamoyl-CoA reductase; CAD, cinnamyl-alcohol dehydrogenase; COMT, caffeic acid 3-O-methyltransferase; 4CL, 4-coumarate: CoA ligase; CHS:,chalcone synthase; CHI, chalcone isomerase; F3H, flavanone 3-hydroxylase; F30H, flavonoid 3<sup>0</sup> -hydroxylase; F305 <sup>0</sup>H: flavonoid 3<sup>0</sup> ,50 -hydroxylase; FLS, flavonol synthase; GT5, uridine diphosphate (UDP)-glucuronic acid:flavonol-3-O-glucuronosyltransferase; GT6, bifunctional UDP-glucose/UDP-galactose:flavonol-3-O-glucosyltransferase/galactosyltransferase; DFR, dihydroflavonol reductase; LAR, leucoanthocyanidin reductase; LDOX, leucoanthocyanidin dioxygenase; ANR, anthocyanidin reductase; UFGT, UDP-glucose:flavonoid 3-O-glucosyltransferase. The number of expressed family members for each enzyme is indicated in green box. Each square in the heatmap located beside their gene names corresponds to the average FPKM value of the gene in each sample as illustrated in the legend. Genes with significant expression changes compared with the control groups in each developmental stage are indicated by asterisks (<sup>∗</sup> ) in the squares. Expression profiles of specific structural genes required for the biosynthesis of HCAs, flavonols, flavan-3-ols and anthocyanins are shown in pink, yellow, and violet dotted boxes, respectively. C, control group; LR-V, leaf removal at véraison; LM-V, leaf moving at véraison.

at E-L 38 stage, respectively, which correlated well with the changes of LDOX expression and total flavan-3-ol contents at this stage. Therefore, it may be speculated that the transcript level of VviMYBPA1 might lead to the difference in the flavan-3-ol contents of grape berries growing under different light conditions. Furthermore, transcript abundances of VviMYBA1 and VviMYBA2, two regulators of the anthocyanin branch (Walker et al., 2007), were slightly up-regulated at E-L 36 stage while down-regulated at E-L 37 or 38 stage in light exposed berries, correlating with the anthocyanin levels responded to light conditions in the grape skin. However, no significant changes in the expression of two bHLH factors, VviMYC1 and VviMYCA1, as well as two WDR proteins VviWDR1 and VviWDR2 among different light-exposure groups were observed (**Figure 4B**), although they have been reported to be involved in anthocyanin and/or PA synthesis (Hichri et al., 2010; Matus et al., 2010) and differentially modulated by different light qualities in other plant species (Sompornpailin et al., 2002; Cominelli et al., 2008).

# Validation of RNA-seq by Quantitative Real-time PCR

To validate the expression profiles obtained from the RNA-seq data, 15 genes relating to our biological focus were selected to subject to qRT-PCR analysis. They included 12 phenylpropanoid/flavonoid biosynthetic pathway related structural genes (VviPAL1, VviPAL2, VviPAL7, VviPAL15, VviF3H1, VviF3H2, VviF3'H, VviF3<sup>0</sup> 5 <sup>0</sup>H, VviFLS1, VviFLS2, VviFLS3, and VviFLS4), as well as VviMYBF1, VviMYBPA1 and VviMYBA1 TF genes involved in the regulation of flavonol, flavan-3-ol and anthocyanin biosynthesis, respectively (Supplementary Figure S2). Two housekeeping genes in V. vinifera, VviUbiquitin1 and Vviβ-Actin were used as endogenous controls for normalization as their relatively constant expression throughout grape berry development as well as in berries under various stress conditions (Downey et al., 2003; Reid et al., 2006). The results showed that the expression of 15 genes determined by qRT-PCR was significantly correlation with those from the RNA-seq data at the 0.01 level (r = 0.54), thus verifying the method.

### Co-expression Analysis between Metabolic Pathway Genes and Transcription Factor Genes

Transcriptome co-expression analysis, which is based on the assumption that genes with similar expression patterns are most likely to be functionally associated, has proven to be a powerful tool for revealing regulatory networks of genes involved in linked processes (Persson et al., 2005). In plants, this strategy has been applied to identify factors regulating several metabolic pathways, such as two Arabidopsis MYB TFs regulating aliphatic glucosinolate biosynthesis and a rice AP2/EREBP (APETALA 2/ethylene responsive element binding protein) family TF involved in starch biosynthesis (Hirai et al., 2007; Fu and Xue, 2010). To systemically identify unknown putative regulators that control the phenolic biosynthesis in grape berries in response to different light regimes, a genomewide co-expression analysis was employed between metabolic pathway genes and TF genes. Eight phenolic synthesis genes screened previously, including genes encoding PALs, 4CL, F3H, and FLSs, were selected as "guide genes" to identify co-expression relationships specific to the light-induced differentially expressed TF genes using expression data of all light treated and control samples from RNA-seq. Any two genes with an absolute value of the Pearson Correlation Coefficient (PCC) greater than 0.7 (or 0.8) and p-value less than or equal to 0.05 (or 0.01) between their expression profiles were considered as significant (or highly significant) co-expressed genes (Fu and Xue, 2010). The results showed that a total of 120 and 59 TFs were highly co-expressed with the six total flavonoid biosynthesis-related (group I) and the two flavonol biosynthesis-related (group II) guide genes, respectively (**Figure 5**). Among the identified group I co-expressed TFs, the most abundant positively correlated TFs were members of the MYB, WRKY, C2C2, AP2/EREBP, bHLH, and MADS-box families (p-value ≤ 0.01), whereas the most abundant negatively correlated TFs belonging to MYB, NAC (No apical meristem, ATAF 1,2, Cup-shaped cotyledon 2), Cys2/His2 (C2H2) type and CCCH type (C3H) zinc finger protein families (p-value ≤ 0.01). Similarly, specific members of MYB, AP2/EREBP and C2C2 families were also found to be the most abundant significantly positively co-expressed TFs with group II (p-value ≤ 0.05).

In all plant species analyzed to date, MYB TFs, together with bHLH and WDR proteins, act as common denominators in the regulation of flavonoid accumulation under various biotic or abiotic signals, such as high-light, UV, drought, and extreme temperatures (Koes et al., 2005; Xu et al., 2015). Some additional potential transcriptional regulators that belong to WRKY, MADS-box, and bZIP TF families have also been reported to be involved in specific branches of the phenylpropanoid metabolism (Hichri et al., 2011). Besides, several negative regulators of flavonoid synthesis, such as R2R3-MYB, single domain R3-MYB repressors and truncated bHLH, inhibit the formation of MBW complex or modify it, thereby actively repress transcription in plants (Liu et al., 2015). The positively or negatively co-expression of multiple members of MYB, bHLH, WRKY, and MADS-box TF families with those key genes in the present result suggests that light-regulated flavonoid biosynthesis in grape berries is maintained by a complex regulatory network involves both positive and negative feedback loops. By using over-expressing and antisense transgenic plant strategies, it was shown that some DOF (DNA-binding One Zinc Finger) genes from the C2C2 zinc finger-containing TF superfamily putatively involved in regulation of enzymes of the phenylpropanoid and flavonoid pathways in Arabidopsis (Noguero et al., 2013). The plant-specific AP2/EREBP TF family, which was composed of AP2, DREB (cis-acting dehydration responsive element-binding protein), RAV (related to ABI3/VP1), ERF (ethylene responsive factor) and other subfamilies, plays a major role in several developmental processes, and also participates in plant hormone signal transduction as well as plant's responses to pathogens and various environmental stresses (Dietz et al., 2010). Recent studies revealed that a class of repressor-type ERF-subfamily

TFs act as active or passive repressors of transcription via their ERF-associated amphiphilic repression (EAR) domain, which was also found in some C2H2 type zinc-finger proteins and R2R3-MYB repressors of flavonoid synthesis-related genes in various plant species (Aharoni et al., 2001; Ciftci-Yilmaz and Mittler, 2008; Huang et al., 2014). NAC proteins are another plant-specific TFs which have been shown to play an essential role in regulating senescence, cell division, and wood formation, and also participate in plant response to pathogens, viral infections, and various environmental stresses (Nakashima et al., 2012). It was also reported that a NAC protein in Arabidopsis, ANAC078, positively regulates the expression of genes related to the biosynthesis of flavonoids, subsequently leading to the accumulation of anthocyanins in response to high-light (Morishita et al., 2009). Furthermore, VviNAC29, a protein belonging to the grapevine NAC TF superfamily, was demonstrated to act as a cooperative regulator controlling the stress-responsive expression of VviF30H in our previous

study (Sun et al., 2015b). However, their roles in the negative regulation of the flavonoid synthesis-related genes have not been investigated previously, thus, the conclusion that whether they could directly or indirectly regulate the light-response of phenolic biosynthesis still needs to be further characterized.

# Light Response of Plant Hormone Signal Transduction Related Genes

Phytohormones have been implicated in controlling various aspects of grape berry development, in particular, the important processes of ripening and adaptation to adverse environmental conditions, including harmful UV radiation (Jeong et al., 2004). In some cases, hormone pathways act downstream of the light signal pathways to regulate growth, whereas in other cases they interact with each other reciprocally (Alabadí and Blázquez, 2009). In LR-V treated berries, the transcript abundances of some members of PYR/PYL (VIT\_208s0058g00470 and VIT\_210s0003g01335) and abscisic acid (ABA) responsive element binding factor (ABF; VIT\_208s0007g03420) involved in ABA signal transduction (Klingler et al., 2010), xyloglucan:xyloglucosyl transferase TCH4 (VIT\_211s0052g01190) involved in BR signal transduction (Clouse, 2015), jasmonate ZIM domain-containing protein (JAZ; VIT\_201s0146g00480) and MYC2 TF (VIT\_211s0052g00100) involved in jasmonic acid (JA) signal transduction (Fernández-Calvo et al., 2011), as well as TGA TF (VIT\_207s0031g02670 and VIT\_208s0007g06160) and pathogenesis-related protein 1 (PR-1; VIT\_203s0088g00780, VIT\_203s0088g00810, VIT\_203s0088g00910, and VIT\_203s0097g00700) involved in salicylic acid (SA) signal transduction (Eulgem, 2005) were significantly or moderately higher compared with that of the control at nearly all the three developmental stages (**Figure 6**). In addition, the transcription of some proteins involved in other plant hormone signal transductions was also significantly or moderately up-regulated in LR-V treated berries at specific developmental stages, such as members of the histidine kinase receptors CRE1 (VIT\_213s0019g01180 and VIT\_217s0000g04920) and histidine-containing phosphotransfer protein (AHP; VIT\_211s0016g03170) required for cytokinin (CTK) signal transduction (Choi and Hwang, 2007) and gibberellin receptor GID1 (VIT\_213s0084g00130) required for Gibberellin (GA) signal transduction (Hirano et al., 2008) at E-L 36 stage, members of the auxin influx carrier AUX1 (VIT\_208s0007g02030) and auxin-responsive protein IAA (AUX/IAA; VIT\_211s0016g03540) required for auxin signal transduction (Lau et al., 2008) at E-L 38 stage. Notably, a member of serine/threonine-protein kinase CTR1 (VIT\_218s0001g07700), a negative regulator of the ethylene (ETH) response pathway (Chen et al., 2005), was up-regulated in LR-V treated berries at E-L 38 stage. Meanwhile, a member of 1-aminocyclopropane-1-carboxylic acid oxidase (ACO, EC 1.14.17.4; VIT\_211s0016g02380), the last enzyme in the ETH production pathway which controlled the biosynthesis of ETH in plants (Chervin et al., 2004), was significantly down-regulated in the same developmental stage. In berries from LM-V treatment, a member of TCH4 (VIT\_211s0052g01190) was significantly up-regulated, while members of AUX/IAA (VIT\_214s0030g02310) and PR-1 (VIT\_203s0088g00710) were significantly down-regulated at E-L 37 stage. No significant differences in the expression of plant hormone signal transduction related genes between berries from LM-V treatment and the control group were detected at both E-L 36 and 38 stages. Nevertheless, a member of 9-cis-epoxycarotenoid dioxygenase (NCED, EC 1.13.11.51; VIT\_219s0093g00550), which limits the level of ABA in the biosynthesis pathway (Zhang et al., 2009), was up-regulated in LM-V treated berries at E-L 36 stage, while a member of ACO (VIT\_211s0016g02380) was down-regulated at E-L 38 stage (**Figure 6**).

Previous studies indicate that the accumulation of phenolics in berry skin during the ripening stage, as well as the expression of structural genes and their transcriptional regulators considered to be involved in the phenylpropanoid and flavonoid pathways, were enhanced by exogenous ABA and ETH treatments, while suppressed by synthetic auxins, NAA (Ban et al., 2003; El-Kereamy et al., 2003; Jeong et al., 2004; Fujita et al., 2006; Koyama et al., 2010). The acceleration of berry ripening and flavonoid accumulation in LR-V treated grape berries was well correlated with the enhancement of ABA signal transduction, which might act as a protective mechanism induced by enhanced light irradiation (Berli et al., 2011), while the increased auxin signal transduction and decreased biosynthesis of ETH might result in the suppression of flavonoid biosynthesis, especially flavan-3-ols and anthocyanins at harvest. Moreover, the different expression of metabolic enzymes of phytohormones at E-L 36 and 38 stages, and the transcriptional changes of AUX/IAA at E-L 37 stage, were shown to be perfectly correlated with changes in the accumulation of flavonoids in LM-V treated berries. A number of studies have indicated that light signaling affects the biosynthesis and/or signaling of multiple phytohormones such as auxin, GA, CTKs, ETH, and BRs (Carvalho et al., 2011). The interactions between light and hormones pathways operate through distinct molecular mechanisms in plants and play an important role in the adjustment of developmental programs and behavior of the plants to the environment (Alabadí and Blázquez, 2009). Taken together, our results suggest that phenolic metabolic in berry skins of the Cabernet Sauvignon grape is precisely controlled by a series of phytohormones in response to exchanged light irradiation.

# CONCLUSION

In the present study, the transcriptional profiles and metabolite profiles of phenolic biosynthesis pathway were analyzed in Cabernet Sauvignon grapes under different sunlight exposure treatments during berry development. Leaf removal or leaf moving at different berry development stages did not show consistent effects on the accumulation of flavan-3-ol, anthocyanin or total flavonoids in grape berries over three seasons. However, the concentrations of HCAs and flavonols were moderately and drastically increased in sunlight exposed grape berries, respectively, which is well correlated with changes

in transcriptional abundance of PAL, 4CL, F3H, and FLS family members as well as large amounts of regulatory genes. Furthermore, the transcriptional changes of genes required for the biosynthesis and signal transduction of auxin, ETH and ABA were found to be exactly in accordance with the accumulation of phenolics in light exposed berries during development, confirmed the importance of phytohormones on berry phenolic biosynthesis of grapes in response to light environment. Taken together, our results provide new valuable insights into understanding of the complex regulatory network of sunlightresponsive phenolic biosynthesis in grape berries, as well as theoretical foundations for cultivation management and wine production.

#### AUTHOR CONTRIBUTIONS

JW and WC conceived and designed the experiments. GC and Y-NH did the field experiments. R-ZS, GC, and QL analyzed the transcriptome sequencing data. GC, QL, YW, Y-RZ, W-FS, XZ, and X-DC performed the HPLC quantification. Y-BL and S-YL provided technical support. R-ZS and JW wrote the paper. All the authors revised it critically for important

#### REFERENCES


intellectual content and approved the final version of this manuscript.

#### FUNDING

This work was supported by China Agriculture Research System (CARS-30).

#### ACKNOWLEDGMENTS

The authors thank to Prof. Chang-Qing Duan, Prof. Qiu-Hong Pan, Dr. Fei He (China Agricultural University) and Dr. Bao-Qing Zhu (Beijing Forestry University) for information support. The authors are also grateful to CITIC Guoan Wine Co. Ltd for technical assistance.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2017.00547/ full#supplementary-material


of white-fleshed and teinturier grape berries. Planta 243, 23–41. doi: 10.1007/ s00425-015-2391-4


fpls-08-00547 April 19, 2017 Time: 17:49 # 16


of public microarray data sets. Proc. Natl. Acad. Sci. U.S.A. 102, 8633–8638. doi: 10.1073/pnas.0503392102


fpls-08-00547 April 19, 2017 Time: 17:49 # 17

heterozygous grapevine variety. PLoS ONE 2:e1326. doi: 10.1371/journal.pone. 0001326


during abscisic acid biosynthesis from peach and grape fruits. J. Plant Physiol. 166, 1241–1252. doi: 10.1016/j.jplph.2009.01.013


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Sun, Cheng, Li, He, Wang, Lan, Li, Zhu, Song, Zhang, Cui, Chen and Wang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Sunlight Modulates Fruit Metabolic Profile and Shapes the Spatial Pattern of Compound Accumulation within the Grape Cluster

#### Noam Reshef, Natasha Walbaum, Nurit Agam\* and Aaron Fait\*

*French Associates Institute for Agriculture and Biotechnology of Drylands, The Jacob Blaustein Institutes for Desert Research, Ben-Gurion University of the Negev, Sede Boqer, Israel*

Vineyards are characterized by their large spatial variability of solar irradiance (SI) and temperature, known to effectively modulate grape metabolism. To explore the role of sunlight in shaping fruit composition and cluster uniformity, we studied the spatial pattern of incoming irradiance, fruit temperature and metabolic profile within individual grape clusters under three levels of sunlight exposure. The experiment was conducted in a vineyard of Cabernet Sauvignon cv. located in the Negev Highlands, Israel, where excess SI and midday temperatures are known to degrade grape quality. Filtering SI lowered the surface temperature of exposed fruits and increased the uniformity of irradiance and temperature in the cluster zone. SI affected the overall levels and patterns of accumulation of sugars, organic acids, amino acids and phenylpropanoids, across the grape cluster. Increased exposure to sunlight was associated with lower accumulation levels of malate, aspartate, and maleate but with higher levels of valine, leucine, and serine, in addition to the stress-related proline and GABA. Flavan-3-ols metabolites showed a negative response to SI, whereas flavonols were highly induced. The overall levels of anthocyanins decreased with increased sunlight exposure; however, a hierarchical cluster analysis revealed that the members of this family were grouped into three distinct accumulation patterns, with malvidin anthocyanins and cyanidin-glucoside showing contrasting trends. The flavonol-glucosides, quercetin and kaempferol, exhibited a logarithmic response to SI, leading to improved cluster uniformity under high-light conditions. Comparing the within-cluster variability of metabolite accumulation highlighted the stability of sugars, flavan-3-ols, and cinnamic acid metabolites to SI, in contrast to the plasticity of flavonols. A correlation-based network analysis revealed that extended exposure to SI modified metabolic coordination, increasing the number of negative correlations between metabolites in both pulp and skin. This integrated study of micrometeorology and metabolomics provided insights into the grape-cluster pattern of accumulation of 70 primary and secondary metabolites as a function of spatial variations in SI. Studying compound-specific responses against an extended gradient of quantified conditions improved our knowledge regarding the modulation of berry metabolism by SI, with the aim of using sunlight regulation to accurately modulate fruit composition in warm and arid/semi-arid regions.

Keywords: solar irradiance, microclimate, spatial heterogeneity, grape composition, phenylpropanoids, metabolite profiling, climate change, shading

#### Edited by:

*Giovanni Battista Tornielli, University of Verona, Italy*

#### Reviewed by:

*Laurent Deluc, Oregon State University, USA Marianna Fasoli, E & J Gallo Winery, USA*

#### \*Correspondence:

*Nurit Agam agam@bgu.ac.il Aaron Fait fait@bgu.ac.il*

#### Specialty section:

*This article was submitted to Crop Science and Horticulture, a section of the journal Frontiers in Plant Science*

Received: *01 November 2016* Accepted: *12 January 2017* Published: *01 February 2017*

#### Citation:

*Reshef N, Walbaum N, Agam N and Fait A (2017) Sunlight Modulates Fruit Metabolic Profile and Shapes the Spatial Pattern of Compound Accumulation within the Grape Cluster. Front. Plant Sci. 8:70. doi: 10.3389/fpls.2017.00070*

# INTRODUCTION

Plasticity is the ability of an organism to adapt its phenotype in response to changes in the environment. For plants, it is an important adaptive strategy to cope with environmental variability (Svanback and Eklov, 2006) and facilitates the use of the same cultivar in a wide range of growing conditions (Dai et al., 2011). However, at the single-field scale, phenotypic plasticity may be regarded as a significant cause for within-field variation. Field systems, such as vineyards experience a large range of microclimate conditions manifested in significant spatial variation of SI and temperature (Oyarzun et al., 2007; Matese et al., 2014).

Grape and wine quality reflects the levels and composition of a large number of primary and secondary metabolites that shape the overall sensory experience of its derived product. Primary metabolites are largely accumulated in the pulp and include sugars, organic acids, and amino acids. Grape secondary metabolites predominantly include the phenylpropanoids, typically found in the berry skin, and comprise flavonoids, phenolic acids, stilbenes, and viniferins (Anesi et al., 2015). While sugars are largely imported to the fruit through the plant vascular tissues (Ollat et al., 2002), the overall levels and composition of organic acids, amino acids, and secondary metabolites at harvest are determined by the sum of complex metabolic processes occurring in the fruit during its lifetime (Conde et al., 2007; Deluc et al., 2007; Dai et al., 2013). Indeed, metabolic shifts caused by changes in climatic conditions in the immediate vicinity of the fruit (i.e., microclimate), such as light and temperature, are well-known to affect fruit composition (Jackson and Lombard, 1993; Downey et al., 2006).

Studies exploring the effect of microclimate on fruit primary metabolites largely agree that sunlight, and specifically an increase in the temperature load of the fruit, highly modulate malic acid levels, leading to a corresponding decrease in overall fruit acidity (Kliewer, 1965; Jackson and Lombard, 1993; Conde et al., 2007; Sweetman et al., 2014). However, results regarding the effect on the relatively stable tartaric acid are still controversial (Rienth et al., 2016; Young et al., 2016). Discrepancies also exist regarding the effect on amino acids, owing in part to the differential response of cultivated varieties. For example, filtering UV-B irradiance was found to increase the overall levels of amino acids in the juice of Riesling cv. (Schultz et al., 1998), but had no effect on the levels and composition of amino acids in Sauvignon Blanc cv. grapes at harvest (Gregan et al., 2012).

With regards to grape specialized metabolism, studies underline the flavonoids as a highly responsive group to light and temperature perturbations (Chalker-Scott, 1999; Winkel-Shirley, 2002). For instance, flavonol glucoside concentrations in the fruit positively correlated with increased SI on the cluster (Haselgrove et al., 2000; Spayd et al., 2002), and were found to be the most significant metabolites distinguishing the metabolic profile of shaded berries from that of exposed berries (Pereira et al., 2006). Fruit sunlight exposure was found to modulate the composition of anthocyanins, such as the proportion of acylated and coumarylated forms, the ratio between di-hydroxylated and tri-hydroxylated anthocyanins (Haselgrove et al., 2000; Spayd et al., 2002; Downey et al., 2004; Tarara et al., 2008; Chorti et al., 2010), and the proportion of ortho-di phenol anthocyanins (Rustioni et al., 2011). However, the overall accumulation of anthocyanins showed higher dependency on temperature conditions than on SI (Downey et al., 2006), with the effect of the interaction between incoming SI and fruit temperature on grape anthocyanin levels exhibiting a synergistic or antagonistic character, depending on the ranges of both factors (Tarara et al., 2008; Azuma et al., 2012).

While the overall environmental effect on grape has been investigated, few studies have addressed the within-cluster spatial profile of compound accumulation and variability. Pisciotta et al. (2013) characterized berry anthocyanin content considering its vertical position on the rachis and external and internal faces of the cluster (i.e., facing the inter-row or the canopy). However, no data regarding micrometeorological conditions were presented, hampering the interpretation of the results with respect to environmental conditions. Zhang et al. (2015) found greater polyphenol content in regions of the cluster characterized by lower berry surface temperature in Shiraz cv. More recently, by using model-generated microclimate data, Pieri et al. (2016) related the overall levels of berry flavonols, anthocyanins, and amino acids to cluster exposition and its impact on the estimated levels of SI and berry surface temperature (BST). Taken together, these works indicate that micro-scale environments play an active role in generating the spatial patterns of metabolite accumulation that affect fruit composition, as well as the uniformity of the crop. Optimizing crop uniformity is generally regarded as an important aspect in the overall improvement of wine quality (Keller, 2010), owing to its significant role in the composition of the harvested grape, and final wine (Kontoudakis et al., 2011; Liu et al., 2016). However, recent attempts to minimize field heterogeneity through the use of common vineyard management techniques, such as deficit irrigation and cluster thinning, were found inefficient (Calderon-Orellana et al., 2014). This stresses the importance of exploring the effectiveness of other potential techniques, including the modulation of clusterzone microclimate, to form more homogeneous conditions. As a key factor determining fruit intercepted irradiance and temperature (Smart and Sinclair, 1976), practical interventions to ameliorate fruit microclimate generally involve the regulation of sunlight exposure. Practices, such as basal leaf removal, or shading, whether by the use of artificial nets or by the use of

**Abbreviations:** GABA, gamma-Aminobutyric acid; Cyan-3-glu, Cyanidin-3- O-glucoside; Pet-3-glu, Petunidin-3-O-glucoside; Peo-3-glu, Peonidin-3-Oglucoside; Mal-3-glu, Malvidin-3-O-glucoside; Delph-3-glu, Delphindin-3-Oglucoside; Delph-3-acet, Delphinidin-3-O-(6′′-acetyl-glucoside); Cyan-3-acet, Cyanidin-3-O-(6′′-acetyl-glucoside); Pet-3-acet, Petundin-3-O-(6′′-acetylglucoside); Mal-3-acet, Malvindin-3-O-(6′′-acetyl-glucoside); Peo-3-acet, Peonidin-3-O-(6′′-acetyl-glucoside); Delph-3-coum, Delphinidin-3-O-(6′′ p-coumaroyl-glucoside); Mal-3-caffe, Malvidin-3-O-(6′′-caffeoyl-glucoside); Cyan-3-coum, Cyanidin-3-O-(6′′-p-coumaroyl-glucoside); Pet-3-coum, Petunidin-3-O-(6′′-p-coumaroyl-glucoside); Peo-3-coum, Peonidin-3-O-(6′′-pcoumaroyl-glucoside); Mal-3-coum, Malvidin-3-O-(6′′-p-coumaroyl-glucoside); Myr-3-glr, Myricetin-3-O-glucuronide; Rutin, Quercetin-3-O-rutinoside; Myr-3-glu, Myricetin-3-O-glucoside; Quer-3-glr, Quercetin-3-O-glucuronide; Quer-3-glu, Quercetin-3-O-glucoside; Kaemp-3-glr, Kaempferol-3-Oglucuronide; Kaemp-3-glu, Kaempferol-3-O-glucoside; Narin-chalc-glu, Naringenin-chalcone-4-O-glucoside; Hex., Hexoside.

extended canopies, are routinely used in the industry. However, integrated studies encompassing a gradient of sunlight exposure, profiling of metabolic data, and a quantitative characterization of microclimate conditions are lacking. Hence, our understanding of how sunlight regulation influences the spatial heterogeneity of fruit microclimate and its effect on fruit metabolic processes and coordination is currently limited.

In this study, we performed a spatial characterization of micrometeorological conditions and metabolic profiles within individual grape clusters subjected to differing levels and directions of sunlight exposure. The extended range of SI intensities allowed us to explore the responses of a large number of primary and secondary (phenylpropanoids) metabolites to a range of microclimate conditions typically found in the field. This study's insights will aid in defining strategies for sunlight regulation aimed at improving fruit composition and uniformity in challenging environments.

## MATERIALS AND METHODS

#### Site Description

The experiment was conducted during the 2014 and 2015 growing seasons, in a vineyard located in the heart of the Negev Desert, Israel (30◦ 36′ 55.22′′N, 34◦ 45′ 12.00′′E, 800 m altitude). This is an arid region with an average annual precipitation rate of 70 mm (Israel Meteorological Service), which, during the growing season, is characterized by stable meteorological conditions including high SI and elevated midday temperatures (Supplementary Figure 1). The vineyard was planted in 2007 with Vitis Vinifera L. cv. Cabernet Sauvignon grafted on 140 Ruggeri rootstock, irrigated using a covered drip-irrigation system as is common in the region. Rows are orientated north-south with a 30◦ angle to the north-east\south-west, and are trained in vertical shoot positioning (VSP). Three experimental rows were selected, with one border row between every two experimental rows. In each row, three groups of nine adjacent vines were marked as field repetitions, and the basal leaves in the vicinity of the clusters (up to 30 cm above the cordon) were completely removed at the onset of veraison. Each field repetition was further assigned one of three shading treatments: fully exposed, i.e., no net (Exposed), 30% shading net (30% shaded) and 60% shading net (60% shaded), using UV-stabilized woven mesh shading nets for agriculture (Ginegar, Israel) with the percentage representing the PAR filtering capacity as published by the manufacturer. The nets were placed using thin metal support wires, one next to each vine, in a manner that created a shading tunnel with a diameter of about 80 cm around the cluster zone in order to facilitate wind flow and prevent an increase in relative humidity (Supplementary Figure 2). Shading was placed from the onset of veraison (the day of basal leaf removal) until harvest date. Treatments were repeated once on each row in a way that represented all locations along the row to minimize effects of spatial differences both between and within the rows.

#### Meteorological Measurements

During the 2015 growing season, a detailed study of clusterscale micrometeorological conditions was performed. Incoming SI, air temperature, relative humidity (RH%), and wind speed and direction were continuously monitored in 15-min intervals from veraison to harvest by placing a multi-sensor (WS501- UMB, Lufft, Fellbach, Germany) connected to a data logger (CR200, Campbell Scientific, Utah, USA) 1 m above the canopy, positioned in the field to allow for a maximum fetch in the direction of the prevailing winds (northwest). The distance between the sensor and the most distant experimental vine was approximately 90 m.

Cluster-zone air temperature and RH% were measured continuously every 15 min from veraison to harvest by placing sensors equipped with an internal data logger (Hobo ProV2, Onset, MA, USA) at a shaded spot in the immediate vicinity of the clusters. Two sensors were placed at each treatment site to verify repeatability.

The SI intensity on the clusters was measured simultaneously at four horizontal axes: parallel and perpendicular to the vine row, and vertically, to the sky (Supplementary Figure 3). This was done by constructing a box positioning all five pyranometers (LI200R, Li-Core, NE, USA) and connecting them to a single data logger (21X, Campbell Scientific, UT, USA). The sensors were stabilized and leveled on a tripod and placed at the cluster zone, with representative locations and distances from adjacent clusters verified to simulate cluster conditions. This system was continuously rotated every 3 days between field treatments and normalized to the SI measured above the canopy to allow for a comparison between measurements made on different dates. Measurements were done on the east side of the canopy for the shaded treatments and on both sides of the canopy for the exposed treatment.

Berry surface temperature (BST) was measured on 28 July (post-veraison) and 19 August (pre-harvest) 2015, during two diurnal field campaigns from 6:00 to 20:00 using an infrared camera (T640, FLIR systems, OR, USA). Every hour, photos were taken from two representative clusters on each treatment, from each side of the canopy, for a total of 12 clusters. Each cluster was divided into three faces, north- and south-facing, and facing the inter-row (east for east-located clusters and west for westlocated clusters). Data analysis included a careful selection of berries located in the middle section of the cluster's vertical axis (i.e., middle height) in each photo and the exclusion of pixels representing non-grape background data.

# Berry Sampling

Sampling during both seasons was done several days prior to harvest date scheduled at 24◦Brix, to represent final berry composition and the potential of accumulated spatial differences. Samples were taken at pre-dawn in order to prevent any differences between sunlit and shaded berries at the time of sampling. Four vines per shading treatment were sampled, located in two out of the three field repetitions. Selected vines for sampling were at least four vines apart (**Figure 1A**). In each vine, four to six exterior-located clusters (furthest from the trunk toward the inter-row) were selected from each canopy side, to avoid possible shading by the canopy apart from midday hours. Each cluster was dissected into four horizontal orientations (**Figure 1B**), and samples were taken from the middle height

of each orientation. Berries from a single vine and canopy side, located on the same cluster plane, were pooled together and immediately frozen in liquid nitrogen. In the lab, skin, pulp and seeds were carefully separated while kept frozen on dry ice, placed in Eppendorf tubes, and stored at −80◦C until further processing. Analysis included grape skin tissues sampled during both seasons and pulp tissues sampled during the 2015 season.

# Analysis of Skin Phenylpropanoids

Grape skin samples were analyzed using Ultra Performance Liquid Chromatography coupled to a Quadrupole Time-of-Flight Mass-Spectrometer (UPLC QTOF-MS, Waters, MA, USA), following an extraction protocol for metabolite profiling as described in Weckwerth et al. (2004). Skin tissues were lyophilized and ground under liquid nitrogen using a RETSCH-mill (Retsch, Haan, Germany) with prechilled holders and grinding beads. The powder was weighed (40 mg), and metabolites were extracted in a 1-ml pre-chilled methanol:chloroform:water extraction solution (2.5:1:1 v/v). Internal standards, i.e., 0.2 mg/ml ribitol in water, 1 mg/ml ampicillin in water, 1 mg/ml corticosterone in methanol, were subsequently added. The mixture was then briefly vortexed, and 100 µl of methanol was added; the mixture was then placed on a horizontal shaker for 10 min at 1000 rpm. The samples were later

sonicated for 10 min (Elmasonic S30, Elma, Singen, Germany) and centrifuged for 10 min (20,817 × g, microcentrifuge 5417R, Eppendorf, Hamburg, Germany). The supernatant was decanted into new tubes, mixed with 300 µl of chloroform and 300 µl of MiliQ water (Millipore, MA, USA), vortexed for 10 s and then centrifuged at 20,817 × g for 5 min. Next, the water/methanol phase was separated, filtered (0.22 µm Millipore, MA, USA) and transferred to UPLC vials for analysis.

## LC/MS Conditions

UPLC-QTOF-MS conditions were exactly as described previously by Hochberg et al. (2013).

### LC/MS Annotation

MassLynxTM software (Waters) version 4.1 was used for system control and data acquisition. Metabolite annotation was validated using the standard libraries described in Arapitsas et al. (2012), based on retention time order, given in Degu et al. (2014). Metabolites were also annotated based on fragmentation patterns searched against the Chemspider metabolite database (http:// www.chemspider.com/), the consistency of their retention times with those of identified metabolites, and comparison with the data in the current scientific literature.

#### Analysis of Pulp Primary Metabolites

Pulp samples were manually crushed with a mortar and pestle while kept frozen with liquid nitrogen. Next, 100 mg of crushed, frozen powder was weighed and extracted by adding a methanol:chloroform:water solution (2.5:1:1), similar to the aforementioned phenylpropanoid extraction. Then, 70 µl of the extract were dried using a Concentrator Plus (Eppendorf, Hamburg, Germany) and derivatized exactly as described in Hochberg et al. (2015) with sorbitol as the internal standard. Glucose, fructose, tartaric and malic acids were quantified using a calibration curve of standards (Sigma-Aldrich, MO, USA) with 10 concentration points from 50 to 900 ng for glucose and fructose and 2.5 to 45 ng for malic and tartaric acids. A split ratio of 50:1 was used to correctly determine the levels of glucose, fructose, malic and tartaric acids, due to their relatively high abundance in the pulp. The GC-MS conditions were exactly as described previously by Hochberg et al. (2013).

The Xcalibur data system V2.0.7 was used for system control and data acquisition. Annotation was based on spectral searching supported by the National Institute of Standards and Technology (NIST, Gaithersburg, MA, USA) against RI libraries from the Max-Plank Institute for Plant Physiology (Golm, Germany).

#### Statistical Analysis

Statistical analysis was performed using R v3.3.1 in RStudio. A profile analysis, from the "profileR" package, was used in order to compare the spatial profiles of clusters from exposed and shaded treatments and both canopy sides. This analysis tested the multivariate spatial data with two separate null hypotheses: a) the multivariate profile between groups is parallel; and b) the multivariate levels between groups are equal. This was done separately for each of the 70 annotated metabolites. In addition, within-cluster differences were tested for significance using the aov() function for ANOVA and the post-hoc Tukey test using the "agricolae" package. The same method was used for comparing the within-cluster coefficient of variance for each metabolite between different shading treatments, to highlight significant differences in cluster homogeneity caused by shading. Differences in malate/tartaric acid levels were tested for significance by using the built-in t-test() function.

The Pearson correlation of the "corrplot" package was used in order to construct separate metabolite correlation matrices for fully exposed and 60% shaded clusters, based on the entire set of samples from each treatment.

A regression analysis of each metabolite with daily incoming SI levels was done using the entire set of samples obtained from all treatments and cluster orientation, for which irradiance data was available. Linear and logarithmic regressions were assessed by the built-in lm() function, using non-transformed and log-transformed irradiance data, respectively. The R 2 -values of metabolites found to have significant regressions were then calculated by using the mean of four biological replicates, in order to minimize the effect of within-group variability on this coefficient.

Heatmap figures were generated using TMeV v4.9, using the mean of four biological replicates. Hierarchical clustering of the heatmaps was based on the Pearson correlation.

### Network Analysis

Correlation networks were constructed based on the data obtained from Pearson correlation analyses performed in R (detailed above), separately for exposed and 60% shaded treatments, and for metabolites detected in pulp and skin tissues (i.e., primary and secondary metabolites). Visualization and computation of network properties were performed using the "MetScape" application and the NetworkAnalyzer tool, respectively, available in Cytoscape V3.4.0. Correlations were incorporated into the network if the r-value was r > 0.5 or r < (−0.5).

# RESULTS

### The Spatial Distribution of Micrometeorological Conditions across the Cluster Differed in Response to Shading and Position

Cluster-zone shading successfully filtered the incoming SI as shown in **Table 1**. Clusters shaded with 30% and 60% shading nets received 64% and 34%, on average, of the incoming SI measured for fully exposed clusters, respectively. This did not lead to measured differences in the cluster-zone air temperature and relative humidity between treatments (data not shown). Comparing the different orientations in the east-facing cluster, the east-south orientation (orientation 1) received the largest amount of daily SI, followed by east-north (orientation 2) and finally west-south (orientation 3). The higher SI recorded for east-north compared to west-south may be attributed to the sun's zenith relative to the row orientation at the time of direct sunlight exposure on the corresponding sensors.

Interestingly, a comparison between the exposed and the 30% shaded clusters, and the 30% shaded and the 60% shaded clusters revealed that incoming SI differences within a cluster exceeded the differences between treatments. Finally, while the percentage of SI intercepted by the different cluster orientations, compared to the reference of each treatment, remained unaffected, lowering the overall incoming SI levels to the cluster, by shading, effectively minimized the within-cluster heterogeneity in SI. Shading reduced the incoming energy differences between orientations 1 and 3 by 3-fold, from 6.8 to 2.3 MJ/m<sup>2</sup> /day for exposed and 60% shaded clusters, respectively.

As expected, filtering the incoming SI caused a reduction in BST (**Figure 2A**). As shown by the percentiles, during direct sun exposure hours (e.g., morning for east-facing clusters), shading reduced the maximum BST, while having no effect on the minimum values, which represent the temperature of the shaded cluster faces. Notably, during the afternoon, when the east-facing clusters were exposed only to diffused radiation, the minimum BST of the 60% shaded clusters increased, while it remained stable for the 30% shaded and fully exposed clusters. Shading clusters with 60% shading nets decreased their maximum temperature by up to 7.1◦C and daytime (6:00–20:00) mean BST by 0.7◦C (**Table 2**). Examining the distribution of BST within a cluster, presented in histograms (**Figure 2B**), revealed that shaded clusters were more homogeneous than exposed ones. While maximum differences in BST within an exposed cluster


*Sums of daily irradiance, measured within the cluster-zone to represent the different cluster orientations.Values marked by* \* *are estimated based on the mean filtering efficiency of each shading net. Relative sums of daily irradiance values represent the percentage compared to the daily irradiance in orientation 1 of each treatment separately. Values enclosed in parentheses represent the percentage compared to the daily irradiance of the exposed treatment for each orientation separately. Data were not collected for orientations 4 and 5.*


TABLE 2 | Berry surface temperature by cluster orientation in the three shading treatments.

*Berry surface temperature (BST) calculations based on hourly spatial measurements of representative clusters acquired on August 19, 2015. Data were not collected for orientations 4 and 5.*

reached 17◦C (33 to 50◦C at 11:00), they were 9.6◦C in 60% shaded clusters (34.3 to 43.9◦C at 12:00).

## Cluster Spatial Profile of Metabolite Level Was Affected by SI Intensity and Cluster Position

The spatial profile analysis, i.e., the levels and pattern of metabolite accumulation in response to the berry position/orientation, was conducted separately for the primary (**Figure 3**) and secondary metabolites (**Figure 4**), and summarized in Supplementary Table 1.

#### The Impact of Cluster Shading on the Spatial Pattern and Overall Levels of Fruit Metabolites

#### **Pulp primary metabolites**

Shading affected the spatial pattern (i.e., the pattern of accumulation in the different cluster orientations) of leucine, beta-alanine, and citrate in east-located clusters, increasing from internal to external orientations in exposed clusters, while shaded clusters had the opposite trend. Shading increased the levels of maleate and aspartate by 1.6- and 1.8-folds, respectively, and decreased the levels of tartaric acid, valine, and leucine by 1.3-, 2.2-, and 2.2-folds, respectively, in east-located clusters. In westlocated clusters, shading decreased the levels of phosphoric acid, erythritol, beta-alanine, valine, leucine, and GABA by 1.3-, 1.4-, 1.3-, 1.9-, 1.6-, and 2.1-folds, respectively.

#### **Skin phenylpropanoids**

Shading significantly affected the levels of 11 and 12 phenylpropanoids in the east- and west-located clusters, respectively, in 2015. Shading increased the levels of procyanidin B1 and epicatechin by 1.5- and 1.2-folds, respectively, and decreased the levels of the anthocyanins cyan-3-glu and cyan-3-acet (1.8- and 2-folds, respectively), phenylalanine and narin-chalc-glu (2.1- and 3-folds), and the flavonols myr-3-glu, myr-3-glr, quer-3-glu, kaemp-3-glu, and kaemp-3-glr by 1.7-, 1.4-, 2.1-, 4-, and 2.5-folds, respectively, in east-located clusters. In west-located clusters, shading significantly increased the levels of both mal-3-acet and mal-3-coum by 1.3-fold, and decreased the levels of cyan-3-glu, phenylalanine, narin-chalc-glu, and hydroxybenzoate hex (1.8-, 1.8-, 3-, and 1.5-folds), the stilbenes delta-viniferin and piceid (1.8- and 4.2-folds), and the flavonols myr-3-glu, quer-3-glu, kaemp-3-glu and kaemp-3-glr by 1.6-, 2.6-, 5.6-, and 2.8-folds, respectively.

#### The Impact of Cluster Canopy Side on the Spatial Pattern and Overall Levels of Fruit Metabolites **Pulp primary metabolites**

The canopy side of the cluster significantly affected the spatial pattern of malate, galactarate, maleate, threonine, glucose, and fructose in shaded clusters, while no significant effect was found in exposed clusters (**Figure 3B**). All mentioned metabolites followed a trend in which external cluster orientations had lower values than the internal ones, resulting in an opposite eastnorth-south-west pattern between the two sides of the canopy, corresponding with SI. In addition, canopy side significantly affected the overall levels of glucose and fructose in shaded clusters, and of lumichrome, beta-alanine, and GABA in exposed clusters. With the exception of lumichrome, all were found to be higher in clusters from the west side of the canopy than in those from the east.

#### **Skin phenylpropanoids**

Canopy side affected the spatial profile of 15 and 24 phenylpropanoid metabolites in exposed and shaded clusters, respectively, in the 2015 season (**Figure 4**). These included all the anthocyanins, with the exception of cyan-3-glu and vitisin A, the flavan-3-ols epigallocatechin and catechin, narin-chalcglu and phenylalanine, and the entire set of flavonols, with the exception of myr-3-glu. In the 2014 season, canopy side significantly affected the spatial profile of only four and five phenylpropanoid metabolites, in exposed and shaded clusters, respectively; of these, mal-3-glu and peo-3-coum, in exposed clusters, and the flavonols rutin, quer-3-glr, quer-3-glu, and kaemp-3-glu, in shaded clusters, repeated in both years. Canopy side had no significant effect on the levels of phenylpropanoids

FIGURE 3 | Cluster spatial level results for primary metabolites detected by GC-MS in berry pulp tissues sampled from different cluster orientations and shading treatments. (A) Illustrations of the daily irradiance levels in each cluster orientation including estimations based on diffusive light measurements and the measured mean filtering capacity of the shading nets. (B) Organic acids, amino acids, sugars, sugar alcohols, and other compounds found to be significantly affected by SI levels, cluster position and/or cluster orientation. Levels represent relative abundance based on ion count. Numbers on the X axis represent cluster orientations. Yellow bars represent fully exposed clusters, and dark gray bars represent 60% shaded clusters. Error bars are standard error (*n* = 4). Bars of the same cluster location and treatment, marked by different letters, represent significantly different values (α < 0.05). Information given in the boxes details the significant effects of treatment (upper box) for clusters located on the east (E) and the west (W) side of the canopy, as well as canopy side (lower box), on the spatial pattern and profile levels of metabolite accumulation in fully exposed (Ex) and 60% shaded (Sh) clusters. Asterisks in upper box indicate a significant effect of treatment on the spatial pattern of compound accumulation. Yellow and gray boxes indicate a significantly higher overall compound level in the exposed and 60% shaded treatments, respectively. Asterisks in lower boxes indicate a significant effect of canopy side on the spatial pattern and overall compound levels.

FIGURE 4 | Cluster spatial level results for phenylpropanoid metabolites detected by UPLC-QTOF-MS in berry skin tissues sampled from different cluster orientations and shading treatments. Included are pivotal phenylpropanoids and stilbenes, flavan-3-ols, flavonols, and anthocyanins found to be significantly affected by SI levels, cluster position and/or cluster orientation. Levels represent relative abundance based on ion count. Numbers on the X axis represent cluster orientations. Yellow bars represent fully exposed clusters, and dark gray bars represent 60% shaded clusters. Error bars are standard error (*n* = 4). Bars of the same cluster location and treatment, marked by different letters, represent significantly different values (α < 0.05). Information given in the boxes details the significant effects of treatment (upper box) for clusters located on the east (E) and the west (W) side of the canopy, as well as canopy side (lower box), on the spatial pattern and profile levels of metabolite accumulation in fully exposed (Ex) and 60% shaded (Sh) clusters. Asterisks in upper box indicate a significant effect of treatment on the spatial pattern of compound accumulation. Yellow and gray boxes indicate a significantly higher overall compound level in the exposed and 60% shaded treatments, respectively. Asterisks in lower boxes indicate a significant effect of canopy side on the spatial pattern and overall compound levels. Illustrations of the daily irradiance levels in the corresponding cluster orientation are given in Figure 3A.

in 2015, while in 2014, mal-3-glu was higher in the eastlocated clusters than in the west; the opposite was found for phenylalanine.

Overall, the proportion of metabolites showing significant responses to canopy side and shading were higher in the phenylpropanoids than in the primary metabolites, with sugars and hydroxy-cinnamates being the least affected chemical groups in pulp and skin tissues, respectively.

#### Common and Differential Responses of Metabolite Groups to SI

Hierarchical clustering of primary metabolites (**Figure 5A**) highlighted three major compound groups. The first group was composed mainly of nitrogenous compounds, including the amino acids leucine, valine, alanine, serine, and GABA, the biogenic amines putrescine and ethanolamine, and two sugar alcohols, galactinol, and erythritol. This group's compounds showed an increase from shaded to fully exposed clusters. In addition, in 30% shaded and fully exposed clusters, a generally higher abundance was found in external cluster orientations (i.e., facing the inter-row), especially for clusters located at the west side of the canopy, than in the internal orientations (i.e., facing the canopy), as seen for orientations 7 and 8 (external) compared to 4 and 5 (internal). The second group included metabolites associated with the TCA cycle, such as citrate, fumarate, malate, as well as the closely linked aspartate and maleate, in addition to glycolate, gluconate, and raffinose. These compounds showed a gradual decrease from the densely shaded to the fully exposed samples. In addition, their content was greater in the internal cluster orientations than in the external ones, irrespective of the treatment, as seen for orientations 4 and 5 compared to 1 and 8, respectively. The third group, comprising tartaric acid, the amino acids proline, beta-alanine and threonine, sucrose, glucose-6-phosphate, phosphoric acids, galactarate, malonate, transcaffeate, pyroglutamate, anhydro-gluopyranose, myo-inositol, and lumichrome, had a less pronounced pattern of change between treatments and cluster orientations.

Clustering the phenylpropanoids using the 2015 season data (**Figure 5B**) grouped together compounds according to their chemical properties and biochemical pathways, such as those belonging to the flavonols, flavan-3-ols and hydroxy-cinnamates. In general, the flavonols showed a strong increase from the densely shaded to the exposed clusters, while the flavan-3 ols and cinnamates had the opposite trend. In contrast to the mentioned groups, the anthocyanins showed metabolitespecific trends. Cyanidin (glycosylated and acetylated) and peonidin (glycosylated) increased from the densely shaded to the exposed clusters and were grouped with the flavonols, as well as phenylalanine and the stilbene piceid, exhibiting an opposite trend to that of the malvidin metabolites. The rest of the anthocyanins were grouped together and exhibited an optimum in the 30% shaded cluster samples.

The 2014 season data (Supplementary Figure 4), based on fully exposed and 60% shaded clusters, yielded similar trends. However, in contrast to the 2015 results, in the 2014 season, the coumaroylated forms of cyaniding, delphinidin, and petunidin clustered with the flavonols, while the glycosylated cyanidin and peonidin, together with the majority of the anthocyanins, formed a separate group.

# Within-Cluster/within-Vine Heterogeneity of Metabolite Abundance Was Affected by Exposure to SI

To study how the within-cluster heterogeneity of metabolite content was affected by shading, metabolite abundance values in samples taken from different orientations of an individual cluster were normalized to that specific cluster median. A hierarchical clustering of the samples (i.e., based on cluster orientations) from the 2015 season (Supplementary Figure 5) was used as a qualitative assessment of cluster heterogeneity where the magnitude of change from the cluster median is represented by the intensity of red and blue colors. Spatial patterns common to all treatments were evident, as samples from the internal orientations (i.e., samples 4 and 5) were clustered together and apart from the external ones (i.e., samples 1 and 8). Among the primary metabolites (Supplementary Figure 5A), organic and amino acids were the most heterogeneous compounds in all treatments, in contrast to the sugars glucose, fructose, and sucrose that were relatively uniform across cluster orientations. Finally, a subtle gradual increase in cluster heterogeneity was visible from the more uniform 60% shaded clusters to the more heterogeneous exposed clusters for the nitrogenous compounds putrescine, pyroglutamate, and ethanolamine, and the compounds trans-caffeate, lumichrome, and erythritol. Among the different phenylpropanoid groups (Supplementary Figure 5B), the flavonols were the most heterogeneous; furthermore, their within-cluster variability clearly increased with increasing levels of shading.

To further verify this trend, the coefficient of variance of each metabolite was calculated per vine (i.e., within all samples originating in a single vine). Values were compared between treatments, and a significant treatment effect on the coefficient of variance was found for five out of the total 70 metabolites (Supplementary Figure 6). Among the primary metabolites, putrescine had a significantly lower coefficient of variance in the 60% shaded clusters than in the 30% shaded and fully exposed clusters, showing an improvement in uniformity with shading. In the phenylpropanoid group, the glucuronide and glycosylated forms of the flavonols quercetin and kaempferol had a significantly higher coefficient of variance in the 60% shaded clusters than in the 30% shaded and fully exposed clusters, showing an improvement in cluster uniformity with an increasing degree of sunlight exposure.

### Modulation of Metabolite-Specific Abundance by Sunlight

The metabolite profiles of berries exposed to a gradient of SI intensity resulted in a significant linear regression between 33, out of 70, metabolites and the sum of daily SI. **Figure 6** shows the regression of 12 metabolites of interest that obtained R <sup>2</sup> > 0.5, highlighting major metabolic shifts in the fruit in response to SI. Among the primary metabolites (**Figure 6A**),

the two most abundant organic acids, malate and tartaric acid, were found to have contrasting responses to incoming SI levels, showing negative and positive regressions with incoming SI levels, respectively. This caused a significant increase in the ratio of tartaric acid/malate in fully exposed clusters compared to shaded ones (Supplementary Figure 7). As with malate, fumarate and aspartate showed similar negative responses. In contrast, a large number of nitrogenous compounds, including betaalanine, the branch-chained amino acids, valine and leucine, stress-related proline and GABA, and the biogenic amines,

putrescine and ethanolamine, had positive regressions with SI levels (**Figure 6A** and Supplementary Table 2).

Among the phenylpropanoids (**Figure 6B** and Supplementary Table 2), the flavan-3-ols catechin, epicatechin, epigallocatechin, and procyanidin B1 had negative regressions with SI levels. A similar negative trend was found for malvidin anthocyanins, and acetylated and coumaroylated peonidin, while cyanidinglucoside showed a contrasting response, verifying the preceding results shown in **Figure 5B**. As expected, the flavonols showed positive regressions with incoming SI. In addition, the logarithmic trend that best fitted the response of the glycosylated kaempferol and quercetin was in accordance with the higher variability of these metabolites found by within-cluster hierarchical clustering (Supplementary Figure 5), and calculated as an increase in the coefficient of variance, observed under low-light conditions (i.e., 60% shaded clusters) in Supplementary Figure 6.

#### Metabolic Coordination in Response to SI Differed between Exposed and Shaded Clusters

Correlation matrices were separately constructed for primary and phenylpropanoid metabolites based on samples obtained during the 2015 season, from all cluster orientations and canopy sides of the same treatment. This allowed a comparison of the metabolite coordination of exposed vs. 60% shaded clusters in response to changes in SI (**Figures 7A–D**). For both metabolite groups, exposed clusters had a higher number of negative correlations between metabolites and a lower number of positive correlations compared to the 60% shaded clusters. The same trend was found in the phenylpropanoid data obtained in 2014.

Primary metabolites (**Figures 7A,B**) had 136 negative correlations with r < −0.5 in exposed clusters compared to 36 in the 60% shaded clusters, and 270 positive correlations with r > 0.5 compared to 384, respectively. In the exposed clusters, the amino acids serine, alanine, leucine and valine, the biogenic amine putrescine, and galactinol were found to negatively correlate with glucose-6-phophate, glycolate, myo-inositol, glucopyranose, trans-caffeate, and gluconate, as well as with malate, fumarate, maleate, aspartate, and threonine. Fumarate, gluconate and aspartate were also negatively correlated with tartaric acid, glucose, fructose, GABA, beta-alanine, erythritol, and raffinose. Tartaric acid and galactarate were negatively correlated in addition to a strong negative correlation of proline and ethanolamine with malate. GABA, beta-alanine and erythritol were strongly positively correlated, as well as glucose and fructose, fumarate and gluconate and the group of glucose-6-phosphate, malate, glycolate, maleate, myoinositol, and threonine. In the 60% shaded clusters, alanine was negatively correlated with the associated citrate, malate and maleate, the amino acids GABA and proline and trans-caffeate. Pyroglutamate had a strong negative correlation with leucine, while gluconate was negatively correlated with raffinose. In contrast, a large group of metabolites, comprising serine, threonine, aspartate, GABA, myo-inositol, fumarate, phosphoric acid, maleate, glycolate, citrate, malonate, galactarate, sucrose, glucopyranose, and erythritol, were positively correlated, in addition to a strong positive correlation between tartaric acid, malate and glucose.

The phenylpropanoids (**Figures 7C,D**) shared 88 negative correlations with r < −0.5 in exposed clusters compared to 74 in the 60% shaded clusters and 506 positive correlations with r > 0.5 compared to 610, respectively. In the exposed clusters, phenylalanine was found to negatively correlate with a large number of metabolites, including all the annotated flavan-3-ols and anthocyanins, and to positively correlate with the flavonols of kaempferol, as well as quercetin conjugates (excluding quer-3-glr). The latter flavonols, in addition to hydroxy-benzoate, were negatively correlated with a large number of anthocyanins, including all malvidin metabolites and the acetylated forms of peonidin, petunidin and delphinidin, but not with their glycosylated and coumaroylated forms, nor with any of the cyanidin metabolites. In contrast, in the 60% shaded clusters, phenylalanine showed only few, weak negative correlations, and correlations between flavonols and anthocyanins were strongly positive. Instead, epicatechin and coumarate hex showed differing degrees of negative correlations with the entire set of annotated anthocyanins. Finally, correlations between the stilbene delta-viniferin and flavonols, which were slightly positive in the exposed clusters, were strongly negative in the shaded clusters.

Four networks were created based on the correlation matrices of the four datasets (Supplementary Figures 8A–D), two treatments (exposed and 60% shaded clusters), and two metabolic groups (primary and phenylpropanoid metabolites), across the different cluster locations and orientations. At r > 0.5 and r < −0.5, shading slightly increased the number of edges in the primary metabolite network from 157 to 169, the network density from 0.28 to 0.301, the clustering coefficient from 0.486 to 0.537 and the average node degree from 9.24 to 9.94. In the phenylpropanoid networks, shading increased the number of edges from 303 to 340, the network density from 0.481 to 0.54, the clustering coefficient from 0.71 to 0.77 and the average node degree from 16.83 to 18.89. Shading caused an at least 2-fold increase in the nodal degree of the primary metabolites glycolate, galactarate, phosphoric acid, maleate, malonate, beta-alanine, and fructose, and an at least 2-fold decrease in the nodal degree of the nitrogenous compounds valine, serine, leucine, proline, putrescine, ethanolamine and pyroglutamate, as well as glycerate and tartaric acid. In the phenylpropanoids, shading caused an at least 2-fold increase in the nodal degree of the flavonols quer-3-glu, rutin, kaemp-3-glu, and kaemp-3-glr, the flavan-3 ols catechin and Epicatechin, and the stilbene delta-viniferin, and an at least 2-fold decrease in the nodal degree of phenylalanine, the hydroxy-cinnamates p-coumarate and coutarate, hydroxybenzoate hex, procyanidin B1, vitisin A, and the stilbene piceid.

#### DISCUSSION

The accumulation levels of the major primary and secondary metabolites that determine a grape berry's sensorial profile vary as a function of the micrometeorological conditions in its immediate vicinity (Jackson and Lombard, 1993; Downey

et al., 2006). Nevertheless, few studies have investigated how these conditions vary within a vine and across a single grape cluster, and how their variation affects the spatial pattern of the fruit metabolic profile in the vineyard and crop uniformity at harvest. This study provided a detailed spatial characterization of metabolite abundance in clusters located on both canopy sides and subjected to different degrees of sun exposure. The coupling of this sampling layout with high resolution measurements of both SI and BST was used to assess the role of sun exposure as a determining factor in the spatial pattern and variability in grape clusters' chemical composition, and to expand current knowledge regarding compound-specific responses and plasticity to SI.

SI filtering was proven to be an efficient tool to minimize the within-cluster variability of both incoming irradiance and BST. Artificial shading resulted in decreased BST of sunlit berries and an increased BST of shaded berries. This phenomenon could not be explained by diffusive light or air temperature, since the diffusive light intensity was lower in the shaded treatment and air temperature within the cluster-zone showed no differences (data not shown). For both SI and BST, differences between the orientations of the same cluster exceeded those between treatments, stressing the extent of within-cluster variability and the range of conditions that can be investigated within a single grape cluster. Finally, due to the strong correlation between SI and BST, as previously suggested (Smart and Sinclair, 1976) and found here, and since SI plays an important role in both the direct triggering of fruit-located photoreceptors and in fruit thermal balance**,** SI was selected as the explanatory variable in this work.

Elevated fruit temperatures have previously been suggested to increase the metabolic flux toward the TCA cycle and to modify its regulation, utilizing malate to enhance its anaplerotic capacity in grapevine berries (Sweetman et al., 2014; Rienth et al., 2016). The former work related the flux increment to the enhanced biosynthesis of pyruvate (valine, leucine, serine and glycine), oxaloacetate (aspartate, threonine and isoleucine) and 2-oxoglutarate-driven compounds (GABA, proline and putrescine). Indeed, increased TCA flux and the accumulation of all or part of the mentioned metabolites have been described in response to a wide array of abiotic stresses in a number of plant species, as summarized by Krasensky and Jonak (2012) and Obata and Fernie (2012). Here, SI conditions (i.e., cluster location and sun exposure levels) triggered responses in malate and oxaloacetate-associated compounds (malate, fumarate and aspartate) that contrasted with those in pyruvate (valine, leucine and serine) and 2-oxoglutarate-derived (proline, GABA and putrescine) nitrogenous compounds. Increased SI levels involve a concomitant increase in tissue temperature, and their combination is expected to exert oxidative stress (Foyer et al., 1994). Indeed, redox homeostasis was recently suggested to be the link between the metabolic modulation of grapevine berries, found in different leaf-removal studies (Young et al., 2016). This may explain the increase in the levels of proline, GABA and the polyamine putrescine, which share a role in mitigating oxidative stress (Krasensky and Jonak, 2012). Furthermore, the decrease in TCA cycle metabolites and the increase in specific amino acids may also have resulted from an arrest in glycolytic activity coupled with protein degradation, generating amino acids in a non-biosynthetic manner (Araujo et al., 2011; Lehmann et al., 2012; Obata and Fernie, 2012). Unfortunately, current knowledge is lacking regarding the direct effect of SI on fruit primary metabolism. Hence, such metabolic shifts are currently attributed to the secondary effects of temperature, UV-B irradiance and combined stress responses. Future research on the photo-receptor-mediated modulation of primary metabolism can greatly contribute to our understanding.

When considering the results of the network analysis, a higher coordination of the metabolic processes in the central metabolism was observed in this study in shaded grapes. Compared to fully exposed grapes, glycolysis, the TCA cycle, and amino acid metabolism shared a greater number of correlations, and the relations were mainly positive. The fact that the acclimation of the exposed and shaded clusters to perturbations in SI levels resulted in opposite types of correlations between nitrogen and carbon metabolites points to a differential coordination, possibly resulting from a comparison between non-stress (shaded clusters) and stress-related (fully exposed) responses. As described above, and as shown previously in potted grapevines and grape cell cultures subjected to shortterm heat treatment (Sweetman et al., 2014; Ayenew et al., 2015), and berries of vines exposed to water deficit (Hochberg et al., 2013), a combined stress response could lead to the specific accumulation of nitrogen metabolites. In contrast, a positive coordination was maintained between glycolysis sugars and TCA cycle intermediates in both conditions, emphasizing the tight co-regulation existing between these two metabolic pathways.

The induction of the flavonoids biosynthesis by fruit-located photo-receptors was shown in a number of species, including grapevine, as reviewed in Zoratti et al. (2014). Summarizing a large number of studies, it can be concluded that while the R2R3 MYB transcription factors family includes both positive and negative regulators of the flavonoid biosynthesis, light seemed to strictly induce the expression of positive ones. However, a more complex interaction is evident, considering the consequent increase in fruit temperature. As found by a number of studies (Spayd et al., 2002; Mori et al., 2007; Tarara et al., 2008; Azuma et al., 2012), above a certain threshold, temperature causes a reduction in the anthocyanin levels, possibly through degradation, whether enzymatic or non-enzymatic (Vaknin et al., 2005; Mori et al., 2007; Chassy et al., 2015). This results in an expected antagonistic effect of elevated BST and strong SI conditions, as found in the exposed berries. Under these assumptions, the response of anthocyanins to increasing SI levels is expected to reflect the metabolite turnover between biosynthesis and degradation. The negative linear effect of SI on the levels of acylated peonidin and malvidin and mal-3 glu, measured in this study, might not be solely attributed to the consequent increase in temperature, as the proportion of acylated anthocyanins was found to increase with rising temperatures (Downey et al., 2004; Tarara et al., 2008), and malvidin metabolites were found to be exceptionally stable to high temperatures in Mori et al. (2007). Instead, considering the contrasting linear increase in the levels of cyanidin-glucoside, it is possible that the combined heat and light modulation of the biosynthesis-related genes' expression patterns, as found in Azuma et al. (2012), was involved in the preferential biosynthesis of upstream metabolites as was observed for cyanidin-glucoside. This may explain the overall positive net turnover of cyanidinglucoside, while downstream metabolites, such as acylated peonidin and malvidin, showed a clear negative result.

The induction of flavonol accumulation by increasing SI levels, measured here, is in accordance with a large number of studies regarding light-induced flavonol biosynthesis in grape berries (Czemmel et al., 2009; Matus et al., 2009; Carbonell-Bejerano et al., 2014; Liu et al., 2015) and overall accumulation (Haselgrove et al., 2000; Downey et al., 2004; Cortell and Kennedy, 2006; Pereira et al., 2006; Matus et al., 2009; Azuma et al., 2012)**,** yet the regression between SI and flavonol-glucoside accumulation in grape skins has been found so far to be purely linear (Haselgrove et al., 2000). The extensive gradient and relatively high SI intensity present in our study may be the cause for this discrepancy, yet whether this trend is due to a stagnation in biosynthesis or to an increase in temperaturedriven degradation remains unclear.

The accumulation of monomeric and condensed Flavan-3 ols in grape skin peaks around veraison, followed by a decrease during the final stages of ripening (Downey et al., 2004; Fujita et al., 2007; Cohen et al., 2012). Increased SI exposure was found to increase flavan-3-ols levels at veraison, yet, since sunlit berries showed a faster decline during ripening, differences were no longer evident at harvest (Downey et al., 2004; Fujita et al., 2007). In this study, conducted between veraison and harvest, the linear negative effect of SI on the levels of flavan-3-ols metabolites was in accordance with the cited literature. A possible explanation could be a delay in the typical post-veraison decrease in the expression of the biosynthesis-related genes LAR and ANR under shading, as found in Fujita et al. (2007).

The stilbenes, which play a role in antifungal activity (Langcake, 1981; Pont and Pezet, 1990; Pezet et al., 2004) and offer potential health promoting effects (Bradamante et al., 2004; Baur et al., 2006; Gresele et al., 2011), accumulate in grapevine tissues in response to biotic and abiotic stresses, such as mold development, wounding, and UV-C irradiation treatments (Fritzemeier and Kindl, 1981; Bais et al., 2000; Vannozzi et al., 2012). Attempts to understand the impact of environmental factors on the accumulation of stilbenes in grape skin tissues have yielded confounding results, possibly owing to the importance of genetic, developmental, and pedological factors (Versari et al., 2001; Bavaresco et al., 2007, 2008, 2012; Berli et al., 2008; Carbonell-Bejerano et al., 2014; Degu et al., 2016). In this study, no significant correlation was found between SI and the levels of the stilbenes piceid and delta-viniferin. However, their levels were significantly higher in the exposed, compared to the shaded, clusters that were positioned on the west side of the canopy but not the east, an opposite trend to that of the major anthocyanins Mal-3-acet and Mal-3-coum. Considering the UV-B induction of stilbene synthase (STS) expression and stilbene accumulation (Versari et al., 2001; Carbonell-Bejerano et al., 2014), and the fact that chalcone synthase and STS compete for the same substrate (Jeandet et al., 1995; Vannozzi et al., 2012), the measured increase in stilbene accumulation in the exposed, compared to the shaded, berries may have depended on a combination of increased STS expression and higher substrate availability, provided by a possible inhibition of anthocyanin biosynthesis.

The differences in metabolic coordination between the exposed and shaded clusters in phenylpropanoid metabolism are intriguing. Exposing clusters to high SI generated a strong negative association between the precursor of the polyphenol pathway, phenylalanine, and the anthocyanins, which was not evident in the shaded clusters. In addition, in the exposed clusters, anthocyanins were more strongly correlated with narinchalc-glu, but less with flavonols, and the correlation between flavonols and stilbenes shifted from highly negative (shaded) to partially positive (exposed). Anthocyanin accumulation in grape-tissue culture was recently shown to be decoupled from phenylalanine under conditions in which biosynthesis-related gene expression was down-regulated (Manela et al., 2015). As a result, the tissue accumulated higher levels of flavonols and stilbenes. It is possible that a similar phenomenon occurred in the fully exposed clusters, as biosynthetic gene expression may have been hampered under conditions that included elevated temperatures (Azuma et al., 2012). Taken together, these lines of evidence are in support of a repartitioning of carbon precursors of the polyphenol pathway from anthocyanin biosynthesis to that of stilbenes and flavonols. This hypothesis should be confirmed in future studies by implementing stable isotopes based flux analysis in detached berries.

Our results revealed that shifting the intensity and direction of solar irradiance (SI) significantly modulated the spatial patterns of the accumulation of organic and amino acids, the main sugars glucose and fructose, and the majority of skin phenylpropanoid metabolites, across the grape-cluster, while hydroxy-cinnamates were not affected. In addition, filtering the irradiance intensity significantly affected the levels of 24 metabolites across the spatial locations, including organic and amino acids, flavonols, flavan-3-ols, anthocyanins and stilbenes. The within-cluster spatial heterogeneity was characterized by large variations in the flavonol levels, found to be significantly affected by sunlight exposure. Overall, the effect of SI conditions on skin phenylpropanoids was comprehensive, while it was found to be more specific in the case of pulp primary metabolites. Together, these findings suggest that sunlight plays a major role in shaping the spatial accumulation of quality-related compounds within a single grape cluster.

The calculated fold changes in metabolite accumulation, across the spatial dataset presented here (Supplementary Table 3), summarizes the compound plasticity of 70 primary and secondary metabolites to light and temperature perturbations, revealing the potential modulating effect of sunlight regulation on fruit composition. Considering that a single layer of grapevine leaves absorbs at least 60–70% of the visible wavelengths (Schultz, 1996), the gradient of SI intensities and metabolite fold change values found in this study may represent, yet possibly underestimate, the range found within a single, non-defoliated, commercial vine.

# CONCLUSIONS

Grape berries' acclimation to their surrounding environment involves local metabolic shifts, which affect their chemical composition and quality at harvest. Sunlight exposure triggers a complex response through both irradiance-mediated signaling and accumulated heat. Here, we show that this metabolic acclimation to sunlight drives the spatial variability of chemical composition between berries on a single cluster, and that it involves the interaction and modulation of partitioning between several biochemical processes in both pulp and skin. These include the accumulation of pyruvate and 2-oxoglutarate-derived nitrogenous compounds, at the expense of malate, fumarate and aspartate in the pulp, and the accumulation of phenylalanine, narin-chalc-glu, cyan-3-glu and the flavonols, accompanied with a decrease in flavan-3-ols, hydroxy-cinnamates and malvidin anthocyanins, in the skin. **Figure 8** illustrates the main findings of the study.

clusters, while mesh trapezoids represent 60% shaded clusters.

This study characterized, yet did not isolate, the accompanying climatic components, such as temperature, that are known to influence fruit metabolic processes. Therefore, our findings may apply to a defined climate, namely warm and arid to semi-arid regions. The existing knowledge gap regarding the intriguing interaction between (SI) and berry temperature, as well as the contribution of temperature-driven processes to the overall fruit metabolic response, may hinder the extrapolation of our findings to systems experiencing different climatic conditions.

At the practical level, (SI) is the most easily and readily controlled climatic factor. In the context of climate change, these results will aid in designing a knowledgebased use of sunlight regulation as tool to improve grape composition, under the conditions that are expected to prevail in an ever-expanding number of commercial vineyards worldwide.

# AUTHOR CONTRIBUTIONS

NR, NA, and AF conceived and planned the study. NR and NW applied the viticultural treatments, and collected the meteorological data and berry samples in the field. NR and NW analyzed the meteorological data and prepared the berry samples for extraction. NW processed and analyzed the thermal images, and NR performed the sample extraction and analysis using the UPLC-QTOF-MS and GC-MS devices. NR integrated the data and performed the data analysis. NR wrote the body of the paper with AF and NA. All authors reviewed and approved the manuscript.

#### FUNDING

This work was partially funded by the Koshland Foundation for the Support of Interdisciplinary Research in Combating Desertification, and the Frances and Elias Margolin Trust.

#### ACKNOWLEDGMENTS

The authors would like to thank the owner of the Nana Farm, Eran Raz, for his professional maintenance of the experimental parcel, and Noga Sikron, Mariela Leiderman, Talya Samani and

#### REFERENCES


Biruk Ayenew for their technical guidance and support in the lab and field.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2017. 00070/full#supplementary-material

stable isotopes. Food Chem. 166, 448–455. doi: 10.1016/j.foodchem.2014. 06.002


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Reshef, Walbaum, Agam and Fait. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Transcriptome-Wide Identification of Novel UV-B- and Light Modulated Flavonol Pathway Genes Controlled by VviMYBF1

Stefan Czemmel1,2, Janine Höll<sup>2</sup> , Rodrigo Loyola3,4, Patricio Arce-Johnson<sup>3</sup> , José Antonio Alcalde<sup>4</sup> , José Tomás Matus<sup>5</sup> and Jochen Bogs2,6,7 \*

<sup>1</sup> Quantitative Biology Center, University of Tübingen, Tübingen, Germany, <sup>2</sup> Centre for Organismal Studies Heidelberg, Heidelberg, Germany, <sup>3</sup> Departamento de Genética Molecular y Microbiología, Pontificia Universidad Católica de Chile, Santiago, Chile, <sup>4</sup> Departamento de Fruticultura y Enología, Pontificia Universidad Católica de Chile, Santiago, Chile, <sup>5</sup> Centre for Research in Agricultural Genomics, CSIC-IRTA-UAB-UB, Barcelona, Spain, <sup>6</sup> Dienstleistungszentrum Ländlicher Raum Rheinpfalz, Viticulture and Enology Group, Neustadt/W, Germany, <sup>7</sup> Fachhochschule Bingen, Bingen am Rhein, Germany

#### Edited by:

Claudio Bonghi, University of Padua, Italy

#### Reviewed by:

Zhanwu Dai, INRA, France Ilaria Filippetti, Università di Bologna, Italy

> \*Correspondence: Jochen Bogs jochen.bogs@dlr.rlp.de

#### Specialty section:

This article was submitted to Crop Science and Horticulture, a section of the journal Frontiers in Plant Science

> Received: 02 March 2017 Accepted: 06 June 2017 Published: 22 June 2017

#### Citation:

Czemmel S, Höll J, Loyola R, Arce-Johnson P, Alcalde JA, Matus JT and Bogs J (2017) Transcriptome-Wide Identification of Novel UV-B- and Light Modulated Flavonol Pathway Genes Controlled by VviMYBF1. Front. Plant Sci. 8:1084. doi: 10.3389/fpls.2017.01084 Flavonols constitute a group of flavonoids with important photoprotective roles in plants. In addition, flavonol content and composition greatly influences fruit quality. We previously demonstrated that the grapevine R2R3-MYB transcription factor (TF) VviMYBF1 promotes flavonol accumulation by inducing the expression of flavonol synthase (VviFLS1/VviFLS4), a key step of the initial flavonol pathway. Despite this, gene networks underlying flavonol modification in grapevine including both structural and regulatory genes remain poorly understood. In order to identify flavonol modifying genes and TFs acting downstream of VviMYBF1 a microarray-based transcriptome analysis was performed on grapevine hairy roots ectopically expressing VviMYBF1 or a Green Fluorescent Protein as control. VviFLS1 was induced in VviMYBF1 transgenic roots and glycosylated flavonols accumulated significantly compared with control lines. Among the differentially expressed genes, potential flavonol-modifying enzymes with predicted rhamnosyltransferase (e.g., RhaT1) or glycosyltransferase (e.g., GT3) activities were identified. In addition, important TFs of the MYB and bZIP families such as the proanthocyanidin regulator VviMYBPA1 and the UV-B light responsive HY5 homolog VviHYH were significantly altered in their expression pattern by overexpression of VviMYBF1. Co-temporal expression analysis demonstrated positive correlation of VviMYBF1 with VviFLS1, VviGT3, and VviRhaT1 during berry development and in fruits ripened with different light and UV-B radiation conditions at field. These results show that VviMYBF1 overexpression led to the identification of novel genes of the flavonol pathway and that the flavonol modifying machinery can be influenced by agricultural practices to optimize flavonol composition in grapes.

Keywords: flavonols, MYB, gene regulation, HY5, UV-containing light, grapevine

# INTRODUCTION

fpls-08-01084 June 20, 2017 Time: 17:4 # 2

Flavonols are the most ubiquitous flavonoids found in dietary plant-based foods (Middleton et al., 2000) and provide the second most abundant group of flavonoids in grapevine (Vitis vinifera L.) fruits (Georgiev et al., 2014). Flavonols largely accumulate in grape berry skins and show a remarkable facet of chemical diversity. In skins of the cultivar (cv.) 'Shiraz', modification of flavonols were mainly identified as 3-, 7-, and 4<sup>0</sup> -O-glycosylations of the basic flavonol scaffold (Downey and Rochfort, 2008; **Figure 1**). Flavonol profiles strongly depend on grapevine cultivars (Zhu et al., 2012), but in general, the main representatives in red grapes are quercetin-3-O-glucosides followed by myricetin, whereas quercetin and kaempferol derivatives constitute the most prominent flavonol compounds in white grapes (Mattivi et al., 2006; Castillo-Muñoz et al., 2007). The quality of red wines can be influenced by flavonols as the color of the wines is positively influenced by the copigmentation phenomenon, which is due to molecular associations between anthocyanins and flavonols or other uncolored phenolic compounds in solutions (Boulton, 2001). In recent years, it has been shown that biomedical activities of flavonols are tightly linked to their chemical diversity. The efficiency of flavonols as antioxidant compounds greatly depends on their chemical structure demonstrating a decrease in antioxidant and antibacterial capacity of flavonol glycosides compared to flavonol aglycones (Mori et al., 1987; Vinson et al., 1999; Burda and Oleszek, 2001; Montoro et al., 2005).

Although flavonols are produced as secondary metabolites, the relationship between their biological function and modification in planta still remains elusive. In grapevine a boost in flavonol accumulation is often observed in response to any agronomical practice that favors the exposure of grape brunches to sunlight (Teixeira et al., 2013). Consequently, flavonol biosynthesis has been extensively studied in response to its induction by UV-containing light in grapes (Price et al., 1995; Czemmel et al., 2009; Matus et al., 2009). In Arabidopsis it was shown that the tt4 and tt5 mutants, which are defective in the gene CHALCONE SYNTHASE and CHALCONE ISOMERASE, respectively, are sensitive to high-irradiance UV-containing light (Li et al., 1993).

exclusively glycosylated in position 3 of the C ring.

the flavan skeleton are abbreviated as follows: Glc, glucose; Gal, galactose; Rha, rhamnose; Me, methyl group. Note that flavonols detected in grape so far are

The authors showed that this effect was correlated with low content or even absence of kaempferol derivatives in leaves (Li et al., 1993), suggesting an important role of flavonols as UV screens in planta. Outside its role in UV protection, modified flavonols were found to act as endogenous flavonol inhibitors of polar auxin transport in the model plant Arabidopsis (Yin et al., 2014). Conversely the expression or activity of flavonol glycosyltransferases have been shown to be influenced by auxin levels (Santelia et al., 2008). Moreover flavonol aglycones but not flavonol 3-O-glycosides were able to restore pollen fertility in conditionally male-fertile petunia pollen both in vivo and in vitro (Vogt and Taylor, 1995).

Induction of flavonol biosynthesis by solar radiation can be attributed to an upregulation of flavonol biosynthetic genes leading to an increase in glycosylated flavonols. This process is negatively affected in the grape berry skin in response to light depletion (Matus et al., 2009) or UV-B filtering (Carbonell-Bejerano et al., 2014). As seen from these studies and many others, visible light and UV-B strongly affect the expression of the R2R3-MYB transcription factor (TF) VviMYBF1 and its target, the first flavonol branch gene FLAVONOL SYNTHASE 1 (VviFLS1, also known as FLS4 by Fujita et al., 2006; **Figure 1**), whose relation was demonstrated in grape suspension cells treated with UV-B light (Czemmel et al., 2009). Besides this well-described regulatory mechanism, only two structural genes have been identified so far involved in flavonol modification, namely VviGT5 and VviGT6, which encode a UDP-glucuronic acid:flavonol-3-O-glucuronosyltransferase and a bifunctional UDP-glucose/UDP-galactose:flavonol-3-Oglucosyltransferase/galactosyltransferase, respectively (Ono et al., 2010). From these two genes, GT5 (in addition to MYBF1 and FLS1) has been directly associated with the UV-B signaling pathway in grapevine (Loyola et al., 2016).

In addition to the above-mentioned VviMYBF1, other R2R3-MYB TFs have been described as regulators of the phenylpropanoid pathway (reviewed by Czemmel et al., 2012; Matus, 2016). Similar to other plants, these TFs provide the common denominators in the regulation of structural genes of all flavonoid branches whereas co-factors encoding the basic helix–loop–helix (bHLH) domains (also referred to as MYC proteins) and conserved WD repeats (WDR) have been so far exclusively associated to the regulation of anthocyanin/PA accumulation and not to flavonol biosynthesis (Czemmel et al., 2012). The main players in grapevine are VviMYB5A and VviMYB5B proteins that are considered to be general regulators of the flavonoid biosynthetic pathway while VviMYBPA1, VviMYBPA2, and VviMYBPAR are PA biosynthesis regulators and VviMYBA1, VviMYBA2, VviMYBA6, and VviMYBA7 are specific for anthocyanin biosynthesis (Kobayashi et al., 2002; Bogs et al., 2006, 2007; Walker et al., 2007; Deluc et al., 2008; Terrier et al., 2009; Koyama et al., 2014; Matus et al., 2017).

In Arabidopsis the VviMYBF1 homologes AtMYB12, AtMYB11, and AtMYB111 regulate flavonol biosynthesis in a tissue-specific manner (Stracke et al., 2007). In both grape and Arabidopsis, flavonol related MYBs are regulated by the bZIP TF ELONGATED HYPOCOTYL 5 (AtHY5 and VviHY5) contributing to the establishment of UV-B tolerance in these species (Stracke et al., 2010; Loyola et al., 2016). This is in line with findings that the hy5 mutant shows flavonol-deficient roots and downregulation of AtMYB12 expression (Lee et al., 2007; Stracke et al., 2010).

Besides the well-established control mechanism of the initial flavonol pathway gene VviFLS1 by VviMYBF1, it is unclear how the immense biodiversity of flavonol compounds found in grapes is achieved. In order to elucidate the regulation in biosynthesis and modification of flavonols in grapevine, VviMYBF1 was ectopically expressed in grapevine hairy roots (HRs). Follow up experiments on these HRs allowed to identify novel structural and regulatory genes of the flavonol branch. The hypothesis that genes involved in flavonol modification were under similar transcriptional control by VviMYBF1 and environmental factors such as UV-B light was tested. Microarray analysis revealed several differentially regulated candidate genes, which are potentially involved in flavonol modification, transcriptional regulation and response to light. We selected two genes encoding for a putative flavonol rhamnosyltransferase (VviRhaT1) and a glycosyltransferase (VviGT3), which is closely related to the previously characterized flavonol modifying enzymes GT5 and GT6 (Ono et al., 2010), for downstream analysis. Expression of these candidates correlated with flavonol accumulation and VviMYBF1 and VviFLS1 expression in VviMYBF1 overexpressing grapevine HRs and also in response to UV-containing light in field and greenhouse experiments. These genes are promising VviMYBF1 target structural genes that can be further investigated in relation to their impact on flavonol biosynthesis in response to common agronomical practices in the vineyard.

# MATERIALS AND METHODS

#### Cloning of MYBF1 Construct

The cloning of VviMYBF1 from the cultivar (cv.) 'Shiraz' (GenBank locus accession: FJ948477) into the vector pART7 (Gleave, 1992) to give the construct pART7MYBF1 was described previously (Czemmel et al., 2009). For generation of HRs overexpressing VviMYBF1, pART7MYBF1 was NotI digested and the insert (VviMYBF1 under control of CaMV35S) was cloned into pART27 (Gleave, 1992) to give pART27MYBF1. The binary vector pART27 contains spectinomycin resistance for bacterial selection and kanamycin resistance for selection in planta.

# Plant Material Collection for Developmental Series

The collection of the samples for the grapevine developmental series of V. vinifera cv. 'Pinot Noir' was described before (Höll et al., 2013). In short, sampling was performed from the start of floral initiation until harvest with early flower and later berry samples collected from a commercial vineyard during the 2007 to 2008 season in Schriesheim near Heidelberg, Germany. The berry samples derived from ∼100 berries from 20 different plants and bunches and were sampled at weekly intervals, as described previously (Downey et al., 2003). All samples were frozen in

liquid nitrogen upon collection on the field and stored at −80◦C until analyzed.

# Hairy Root Culture and Transformation

Before HR transformation, pART27-MYBF1 or pKGWFS7 control plasmid expressing GFP-GUS were transferred by electroporation into the Agrobacterium rhizogenes strain ATCC 15834, carrying the Ri plasmid pRi15834 (neither the chromosome nor the plasmid ATCC 15834 of this A. rhizogenes strain carry an antibiotic marker).

Transformation and induction of transgenic VviMYBF1 or GFP-GUS expressing grapevine HRs derived from leaves from in vitro-grown cv. 'Chardonnay' plants was performed exactly as described in Höll et al. (2013). After transformation and induction, transgenic HRs were routinely kept on HR) medium (Höll et al., 2013) and transferred to fresh medium after 4 weeks. HR tissues used for RNA extraction and high performance liquid chromatography (HPLC) analysis were rapidly frozen in liquid nitrogen, ground to a fine powder using a Retchmill (MM20, Retch) and stored at −80◦C until further use.

# RNA Extraction and Quantitative PCR (qPCR) Analysis

Total RNA was extracted from grinded material from the developmental series of V. vinifera cv. 'Pinot Noir' as previously described in Höll et al. (2013). RNA of MYBF1 and GFP transgenic grapevine HRs was isolated following the protocol of the EURx GeneMATRIX Universal RNA purification kit (Roboklon). Prior to RT reactions, RNA samples were heated to 65◦C for 10 min and immediately placed on ice to destroy secondary structures. Synthesis of cDNA was performed with SuperScriptTM III First-Strand Biosynthesis System (Invitrogen, catalog no. 18080-051) according to the instructions in the manual for RT using oligo dT primers. Typically, 1–2 µg of RNA were reverse transcribed in a volume of 20 µl for 1 h at 50◦C followed by termination of the enzyme reaction at 85◦C for 5 min. After digest of the residual RNA strands using RNase H, 30 µl of RNase free H2O were added. This 50 µl solution served as cDNA stock for further application and was stored at −20◦C. PCR reactions were carried out for flower and berry skin samples from the cv. 'Pinot Noir' developmental series as well as HR samples by using 0.5 µl of 10 µM forward and 0.5 µl of 10 µM reverse primer, 5 µl cDNA (diluted 1:20), 7.5 µl of 2x ABsoluteTM QPCR SYBR <sup>R</sup> Green Fluorescein Mix (ABgene) and 1.5 µl of H2O in a final volume of 15 µl (for all set of qPCR primers used in this work, see Supplementary Table 1). Thermal cycling conditions were identical for all primer pairs: 96◦C for 15 min followed by 96◦C for 30 s, 58◦C for 30 s, and 72◦C for 30 s for 35 cycles, followed by a melt cycle from 50 to 96◦C. QPCR was carried out using SYBR green method for detection of double stranded PCR products on an iCyclerTM optical Module real-time cycler (Bio-Rad). UBIQUITIN1 (previous TIGR database id: TC32075, VIT\_16s0098g01190) was used for normalization of target gene levels in grapevine HRs and light experiments while the three reference genes GAPDH (VIT\_17s0000g10430), UBIQUITIN1 and EF1-alpha (VIT\_06s0004g03220) were used to normalize gene expression during cv. 'Pinot Noir' berry development as done previously (Höll et al., 2013). The primers MYBintF and MYBintR2 were used to detect the transcript level of VviMYBF1 in grapevine by amplification of a 214 bp PCR-fragment from the 3 0 region of the gene. The primers VviFLS1 and VviFLS2 were used to detect a 154 bp amplicon of the VviFLS1/FLS4 gene (for other primer pairs see Supplementary Table 1). The efficiency of the primers was tested in preliminary experiments with dilutions of plasmid or purified PCR products maintaining an r 2 value ≥ 0.96. With all cDNAs used, the primer set gave a single PCR product, which was verified by agarose gel electrophoresis and by determination of the melt curves for the product at the end of each run.

# Flavonol Staining

Fresh cross sections from 3 to 4 weeks old HRs were transferred in a freshly prepared solution of 0.25% (w/v) diphenylboric acid 2-amino-ethyl ester (DBPA, Naturstoffreagenz A, Roth) and 0.00375% (v/v) Triton X-100. Fluorescence was visualized with either an inverted epifluorescence microscope (Leica DM IRB, Leica) or a stereomicroscope (Leica MZ FL III, Leica) using a DAPI fluorescence filter with an excitation wavelength of 340–380 nm and an emission wavelength of 425 nm. Images were captured using the Leica Image Manager (IM) 50 software and handled with Adobe Photoshop version 8.0.1 without changing the color parameters.

## Microarray Analysis

Microarray raw data generation was produced at the Centro Nacional de Biotecnologia (CNB, Madrid) by using the Grapegen GeneChip, which is an Affymetrix custom made chip containing 23,096 probe sets corresponding to ∼15,800 different annotated genes (Pontin et al., 2010) and covering ∼52% of the genome. Quality assessment of microarray data and differential expression (DE) analysis were carried out in the statistical language R (version 3.2.1) mainly using the affy (version 1.48.0) and LIMMA packages (version 3.26.9) by the Quantitative Biology Center (QBiC) in Tübingen<sup>1</sup> . Based on PCR analysis for the overexpressed TF VviMYBF1 and its target VviFLS1, four VviMYBF1 (lines no. 32, 33, 34, and 62) and four GFP overexpressing lines (lines no. 77, 79, 83, and 97) out of 10 stable transformants were selected. During raw data control analysis on all eight lines (four GFP and four VviMYBF1 lines each) several sensitive measures to assess array quality such as Normalized Unscaled Standard Error (NUSE) and Relative Log Expression (RLE) confirmed low quality of array corresponding to the GFP control line 79. Since the array for line 79 was detected as an outlier, DE analysis has been performed on four VviMYBF1 and three GFP lines excluding GFP line 79. Raw data (CEL files) and meta-information (e.g., experimental design, sample names) of microarray data were deposited to NCBI's Gene Expression Omnibus (GEO) and can be accessed under the GEO identifier GSE95532.

<sup>1</sup>https://portal.qbic.uni-tuebingen.de/portal/

### Light and UV-B Radiation Treatments

Four different sunlight or UV-B radiation experiments were conducted in cv. 'Cabernet Sauvignon' commercial plants (either treated at field or uprooted and transferred to a greenhouse).

#### Experiment 1

Three sunlight depletion treatments were conducted at field in commercial cv. 'Cabernet Sauvignon' vines during the 2006–2007 ripening season, as described by Matus et al. (2009). In this study, different sunlight exposures were generated: (i) full shading of fruits by the plant's own canopy (0% exposure), (ii) full sunlight exposure from veraison (ripening onset) onward, generated by displacement of leaves around the cluster region (100% exposure), and (iii) 25% exposure, by shading from veraison until 6 weeks after which leaves were displaced for cluster illumination.

#### Experiments 2 and 3

High (∼0.3 Wm−<sup>2</sup> ) and Low (∼0.1 Wm−<sup>2</sup> ) UV-B exposure treatments were applied to clusters from 9 year old potted vines in a UV-free greenhouse during 2011–2012 and 2012–2013 growing seasons, respectively, as described by Loyola et al. (2016).

#### Experiment 4

A UV-B filtering radiation treatment (here called -UV-B) was applied in cv. 'Cabernet Sauvignon' plants from a commercial vineyard in Maipo Valley, Chile (33◦ 430 28<sup>00</sup> S, 70◦ 450 9.71<sup>00</sup> W) during 2011–2012 growing season. The filtering treatment consisted in blocking solar UV-B radiation by installing a 100 µm clear polyester film at the position of grape clusters. Grape clusters from both east and west side of the rows were treated, but only grape clusters from the east side of each experimental row (exposed to sunlight during the morning until midday) were sampled. The experimental design consisted in four blocks with five plants each (biological replicates n = 4). Three berries per cluster (randomly sampled) and four clusters per plant were used for each sample. The treatments started at 6 weeks before veraison. Weeks −3, 0, 3, and 6 after veraison (WAV) were considered for RNA extraction and gene expression quantification while weeks −2, 1, 4, and 7 after veraison were used for HPLC analysis. A total of 60 berries were sampled for each biological replicate per condition at each sampling date. Berries were immediately peeled and deseeded. Berry skins were frozen in liquid nitrogen and stored at −80◦C until RNA was extracted. Total flavonols were measured as quercetin equivalents and environmental parameters such as solar UV-B irradiance, temperature, solar radiation at cluster level and total solar radiation were measured around veraison from 08.00 to 19.45 h in the east side of the row.

### Separation of Hairy Root and Berry Skin Flavonols by HPLC

High performance liquid chromatography analysis was performed on methanolic extracts of HR lines using a reverse phase HPLC (Kontron Instruments 322 pump system/360 autosampler/335 HPLC detector; Kontron) with a Symmetry C18 column (3,5 µm, 4,6 mm × 150 mm, catalog no. WAT200632, Waters) protected by a guard column (Czemmel et al., 2009). Liquid nitrogen-frozen and subsequently homogenized sample aliquots of 20–50 mg were used to extract flavonoids by adding 200 µl of 50% (v/v) methanol (HPLC grade) in H2O. Samples were then sonificated for 20 min in an ice water bath and centrifuged for 10 min at 13000 rpm. 200 µl of the clear supernatant were loaded for HPLC analysis. Separation was carried out with a binary gradient of solvent A [10% formic acid (v/v) in H2O] to solvent B [100% (v/v) methanol, HPLC grade]. The gradient conditions were 0 min, 17% solvent B; 15 min, 35% solvent B; 40 min, 37% solvent B; 42 min, 100% solvent B; 50 min, 100% solvent B; 51 min, 17% solvent B; 58 min, 17% solvent B. The column was maintained at 40◦C and the flow rate was 1 mL/min. Data acquisition and processing was performed by Kroma System 2000 software (Kontron). Data are presented as HPLC chromatogram peak areas over time and expressed as milli-absorbance units (mAU) at 520 nm for anthocyanins and at 353 nm for flavonols. Concentrations were calculated from calibration curves prepared from commercial standards (Phytolab) and expressed as quercetin-3-O-glucoside equivalents for flavonols and malvidin-3-O-glucoside equivalents for anthocyanins. Acidic hydrolysis was performed to identify the compounds as flavonols by cleaving the glycosyl group for a shift to the aglycone form by the addition of a 3 N HCL solution, 50% (v/v) methanol in water and incubation for 3 h at 95◦C. All HPLC separation experiments were performed in three independent extractions of from the same biological material.

High performance liquid chromatography quantification of flavonols in berry skins from light experiments was different to flavonol measurements from HRs and berry skin and flower samples from the developmental series of cv. 'Pinot Noir.' For light experiments, flavonol measurements were conducted as described in Matus et al. (2009).

#### Statistics

For statistical evaluation of microarray data, the LIMMA package was used (Ritchie et al., 2015). First, its function lmFit() fitted linear models on the quality controlled and RMA normalized expression values of each gene across the seven samples with the factor genotype (with the two levels MYBF1 and GFP-GUS). Then the functions eBayes() and topTable() with the argument adjust.method = "BH" were used to compute moderated t-statistics and extract a table of the top-ranked genes from a linear model fit with p-values corrected for multiple testing by applying the false discovery rate (FDR) from Benjamini and Hochberg (1995), respectively. Genes were considered differentially expressed (DE) with a FDR corrected p < = 0.05. No log fold change (logFC) cutoff was applied to assess DE of a gene.

To statistically assess differences in expression of genes and flavonol accumulation between VviMYBF1 and GFP control HR lines, an unpaired two-tailed t-test was performed with assumption of equal standard deviation (SD) between both sample groups. Results were considered statistically different at p < 0.05.

For each of the four UV/light experiments, differential gene expression was assessed using a two-way analysis of variance

factors time (in WAV) and treatment (+UV-B, −UV-B) and an interaction term between those two factors. To analyze at what time points the treatment effects differed, a post hoc analysis was performed on the model using Tukey's test. Genes were considered statistically DE at this time point at p < 0.05.

Note that for accumulations of individual flavonol in response to light no statistical test was performed.

#### Accession Numbers

Microarray data were submitted in the Gene Expression Omnibus (GEO) under accession number GSE95532.

# RESULTS

#### Selection of Hairy Roots for Transcriptome Analysis

Previous studies indicate that the grapevine HR system provides a suitable tool to identify novel target genes regulated by an ectopically expressed MYB TF (Cutanda-Perez et al., 2009; Terrier et al., 2009; Khater et al., 2012; Höll et al., 2013). Transgenic HRs were generated by infiltration of V. vinifera cv. 'Chardonnay' leaves with A. rhizogenes carrying either VviMYBF1 or GFP (control) cDNA. Presence of VviMYBF1 transgene was verified by PCR, whereas control lines expressing GFP were selected by PCR and fluorescence microscopy. Four VviMYBF1 (lines no. 32, 33, 34, and 62) and four GFP overexpressing lines (lines no. 77, 79, 83, and 97) out of 10 stable transformants were analyzed for transcript amount of VviMYBF1 and its known target VviFLS1 (Czemmel et al., 2009). During microarray quality control analysis, line no. 79 was detected as an outlier (for details, see Materials and Methods) and downstream analysis has been performed on four VviMYBF1 and three GFP lines excluding GFP line 79. Additionally, HR flavonols were stained in vivo with diphenylboric acid-2-aminoethylester (DPBA, **Figure 2A**) and quantified using HPLC (**Figure 2B**). By using fluorescence microscopy, we observed orange like bodies (indicative of the accumulation of quercetin-derivatives) exclusively in VviMYBF1 overexpressing HRs (**Figure 2A**). These bodies were prominent in epidermal cell layers while endodermal cells and parenchymatic cells of the cortex accumulate only trace amounts or did not show any orange staining indicative for flavonol accumulation. Flavonols, mainly quercetin 3-glucoside, quercetin 3-galactoside and a so far unknown flavonol derivative accumulated significantly higher in VviMYBF1 overexpressing lines as compared to GFP control lines when measured quantitatively using HPLC (**Figure 2B**). To show that MYBF1 induces only the flavonol branch of the flavonoid pathway anthocyanins were measured in the HR lines. Neither in the VviMYBF1 nor in the GFP overexpressing lines anthocyanins were detected.

With the aim to decipher the mechanisms that underlie the phenotypic changes observed in HRs ectopically expressing VviMYBF1, qPCR analysis was deployed. Expression of VviMYBF1 in selected VviMYBF1 lines exceeded significantly its expression compared to GFP control lines (**Figure 2C**),

FIGURE 2 | Selection of transgenic hairy root (HR) lines. Lines for microarray analysis were selected under the following criteria: flavonol accumulation as visualized by fluorescence microscopy on in situ diphenylboric acid-2 aminoethylester (DPBA) stained root cross sections (A), flavonol content as shown by high performance liquid chromatography (HPLC) analysis (B) and gene expression analysis to verify transcript levels of the VviMYBF1 transgene and its well-characterized target VviFLS1 (C,D). DBPA staining analysis shown in (A) was conducted with sections of the VviMYBF1 line 33 (left) and control line 77 (right). HPLC results in (B) are the mean values of three independent extractions from the same biological material. Gene expression analysis in (C,D) is shown relative to the expression of UBIQUITIN1 gene (previous TIGR database id: TC32075, VIT\_16s0098g01190) with standard deviations derived from three qPCR runs with three technical replicates each. Stars indicate significant differences (p < 0.05) between MYBF1 and control HRs based on t-test.

names are indicated on the x-axis with MYBF1 samples on the left and control samples on the right. (B) Principal component analysis (PCA) of gene expression in hairy roots expressing MYBF1 (blue) or GFP control (red). The plot shows the first two principal components (PC1 and PC2) that account for 38% and 22% of the total variation of the data.

which resulted in a pronounced induction of its target, VviFLS1 (**Figure 2D**). Results from this homologous expression experiment clearly support the role of VviMYBF1 in regulation of flavonol biosynthesis by induction of VviFLS1 transcription. These results showed that selected VviMYBF1 and GFP HR lines were suitable candidates for a differential transcriptome analysis approach to identify novel target genes of VviMYBF1 involved in flavonol biosynthesis and its regulation.

# Transcriptome Analysis to Identify VviMYBF1 Target Genes

To identify novel target genes of VviMYBF1, total RNA was isolated from the four VviMYBF1 and the three GFP overexpressing HR lines to perform a Microarray-based transcriptome analysis. Principal component analysis (PCA) clearly separated both, VviMYBF1 and GFP HR lines in distinct groups (**Figure 3B**) indicating differences between underlying gene expression pattern in VviMYBF1 and GFP control groups which is also visible using a heatmap (**Figure 3A**). To test the a priori hypothesis that genes will have significantly different mean expression values between VviMYBF1 and GFP lines, pairwise comparison between the two groups VviMYBF1 and GFP was performed and moderated t-statistics and associated p-values were generated, which were further adjusted for multiple testing using Benjamini and Hochberg's method to control the FDR (Benjamini and Hochberg, 1995). In total, 548 probesets corresponding to 484 differentially expressed genes (DEGs; that is because several oligonucleotides hybridize to the same gene) were DE based on an adjusted p-value smaller than 0.05 (**Table 1** and Supplementary Table 1). These 548 probesets represent ∼2% of the probes on the chip (23,096 probesets in total). 279 probesets corresponding to 230 genes have been called positive DE, while the other 269 probes, corresponding to 254 genes, show significant negative regulation in VviMYBF1 compared to GFP HR lines. Amongst the DEGs, three candidate genes were found which could modify flavonol scaffold structures (**Table 1**): a putative UDP-sugar flavonoid/flavonol glycosyltransferase [VIT\_11s0052g01580, VviGT3 (Ono et al., 2010)], a putative UDP-rhamnose:rhamnosyltransferase (VIT\_00s0218g00170, VviRhaT1) and a putative flavonol modifying sulfotransferase (VIT\_17s0000g04930, VviST1). An investigation of the positive DEGs for TFs revealed VviHYH (VIT\_05s0020g01090) and VviMYB4A (VIT\_03s0038g02310), which showed 1.5 and a 1.6 fold inductions in gene expression in VviMYBF1 versus control HRs, respectively (Supplementary Table 1). Amongst the pool of significantly downregulated transcripts several genes of the PA biosynthetic pathway were identified: the PA regulatory TF gene VviMYBPA1 (VIT\_15s0046g00170) and genes under genetic control of VviMYBPA1 including two chalcone synthase isoforms (VviCHS3 and VviCHS1, VIT\_05s0136g00260, VIT\_14s0068g00930) and anthocyanidin reductase (VviANR, VIT\_00s0361g00040). Microarrays also indicate that transcript abundance of a yet uncharacterized FLAVONOL SYNTHASE isoform, VviFLS5 [(VIT\_18s0001g03430 (Fujita et al., 2006)] was negatively affected by overexpression of VviMYBF1 (**Table 1**).

With the aim to validate the microarray experiment, expression of the three candidates flavonol-modifying genes VviST1, VviRhaT1, and VviGT3 and the PA regulator VviMYBPA1 were analyzed by qPCR (for primers used see Supplementary Table 1). Quantitative analysis revealed that VviGT3, VviST1 and VviRhaT1 were indeed induced in VviMYBF1 lines compared to GFP controls (**Figures 4A–C**), whereas transcript abundance of VviMYBPA1 was negatively influenced by ectopic expression of the flavonol regulator VviMYBF1 (Supplementary Figure 1). Expression of the previously characterized flavonol modifying genes VviGT5 and VviGT6 were not checked in the HR samples, as probesets representing these genes were not present on the microarray chip design used here.



The probesets correspond to significantly upregulated and downregulated genes that passed the filter criteria of a multiple adjusted p-value of 0.05. Note that Table 1 is sorted from highest to lowest log fold change (logFC). The respective probesets correspond to the 12Xv1 genome accession. Note that the microarray chip used here does not contain a probe corresponding to the expressed transgene VviMYBF1. The probeset VVTU16103 for the known VviMYBF1 target gene VviFLS1/VviFLS4 was therefore used as a positive control for DE analysis.

## Candidate Gene Expression during Grape Berry Development

For correlative analysis of gene expression during grape berry ripening, the same V. vinifera cv. 'Pinot Noir' developmental series was used, in which gene expression related to stilbene biosynthesis was measured previously (Höll et al., 2013). Expression pattern correlation between VviMYBF1 and VviFLS1 transcripts was highest at eleven-to-ten weeks prior to beginning of berry ripening (veraison), correlating with previous results on flavonol accumulation and VviFLS1 and VviMYBF1 expression pattern (Downey et al., 2003; Czemmel et al., 2009). Monitoring expression pattern of the putative VviMYBF1 target genes in developing berries showed that VviST1 and VviGT3 are low expressed during berry ripening (**Figures 4D,E**) though showing correlation with VviFLS1 and VviMYBF1 expressions at some early developmental stages (Downey et al., 2003; Czemmel et al., 2009). There is correlation of VviRhaT1 expression with flavonol accumulation throughout all stages of development as VviRhaT1 transcript amounts gradually increase from early to late stages of development (**Figure 4F**), which is similar to flavonol accumulation in grapevine (Downey et al., 2003).

#### Candidate Gene Expression and Flavonol Analysis under Changes in UV-Containing Light Exposure

Fruit zone leaf removal is a common viticultural practice that exposes grape berry clusters to sunlight, reducing humidity and improving the accumulation of various secondary metabolites important for fruit quality (Reynolds et al., 1995; Teixeira et al., 2013). In order to evaluate whether gene expression of candidate flavonol-modifying genes is influenced by agronomical practices that alter the exposure of berries to UV-containing light, four experiments were conducted: Experiment 1: sunlight depletion generated by leaf displacement around clusters of fieldgrowing plants (**Figures 5A–D**), Experiments 2–3: control versus 'High UV-B' and 'Low UV-B' radiation in clusters of 9 yearold plants growing in a greenhouse (Supplementary Figure 2); and Experiment 4: UV-B filtering in clusters of field-growing plants (**Figures 5E–H** and Supplementary Figure 3A). For Experiment 4 not only gene expression levels were measured (**Figures 5E–H**), but also total (Supplementary Figure 3B) and individual flavonol levels determined (Supplementary Figure 4). Furthermore, environmental parameters such as solar UV-B irradiance, temperature, solar radiation at cluster level and total solar radiation were measured (Supplementary Figure 5). For further experimental details for these four light experiments, refer to Section "Light and UV-B Radiation Treatments".

The MYB TF VviMYBF1 was strongly upregulated in all developmental stages at 100% exposure in the sunlight field experiment compared with 0% exposure (**Figure 5A**). Its regulation was also higher in the +UV-B condition of the filtering experiment at 0 and +3 WAV (**Figure 5E**). In the greenhouse experiments, VviMYBF1 expression was more strongly induced by UV-B irradiation (+UV-B) at low intensities but not at high intensities (Supplementary Figures 2A,E). The induction at low intensities was prominent at early stages of development (−3 WAV and time point 0, Supplementary Figure 2B).

Consequently, VviFLS1, the target gene of VviMYBF1, was also upregulated by sunlight and UV-B, most prominently at all time points of 100% exposure in the sunlight experiment compared with 0% exposure (**Figure 5B**), and at 0 and +3 WAV in the UV-B filtering field experiment (**Figure 5F**). VviFLS1 was already induced at −3 WAV, although not statistically significant (**Figure 5F**). Expression of VviFLS1 was also affected in the greenhouse experiments under UV-B irradiance, especially at +3 WAV and +6 WAV (Supplementary Figures 2B,F).

VviRhaT1 expression was induced by +UV-B in the UV filtering experiment around veraison (**Figure 5G**) but showed no significant response to 25 or 100% sun exposure in the light field experiment except at +7 WAV (**Figure 5C**) or to low UV-B irradiance conditions in the greenhouse (Supplementary Figure 2G). In contrast to VviMYBF1 and VviFLS1, VviRhaT1 expression profile throughout development was similar in the 25 and 100% exposure treatments in the sunlight field experiment (**Figure 5C**).

Materials and Methods, D–F). Data points in (D–F) are given as weeks from the onset of ripening (time point 0). Note that from 6 weeks before veraison (-6), berry skin has been separated from seeds. Flowering occurred 8 weeks before veraison. Each sample of the developmental series from –6 on corresponds to a pool of >100 berries collected from ∼20 plants, as described in Höll et al. (2013). Values and standard deviations are derived from two PCR runs with triplicate PCR reactions each. Stars in (A–C) indicate significant differences (p < 0.05) between MYBF1 and control HRs based on t-test. No statistical analysis was performed for data in (D–F).

Expression of the three genes MYBF1 and FLS1 and RhaT1 are indicative of a rapid induction to late sunlight exposure (stage +7 WAV in the 25% exposure treatment), as this is the stage when leaves are replaced and clusters are exposed to light (see diagram at the bottom of **Figure 5** and see Materials and Methods) which is supported by a similar expression level of the respective gene at the same stage in the 100% exposure treatment (**Figures 5A–C**).

Regarding VviGT3, there is a consistent two-to-three-fold induction of its expression in response to UV-B at 3 weeks before veraison in the greenhouse UV-B irradiance or filtering

FIGURE 5 | Transcript profiles of known and putative flavonol biosynthetic genes in the berry skin in response to light and UV-B-modifying field conditions. Expression data results for VviMYBF1 (A,E), VviFLS1 (B,F), VviRhaT1 (C,G), and VviGT3 (D,H) are shown in two field radiation experiments: cluster sunlight-depletion generated by leaf displacement at field (A–D) and UV-B radiation filtering experiment at field (E–H, see Materials and Methods). Results from the greenhouse experiments are shown in Supplementary Figure 2. Data points in the UV-B filtering field experiments are given as weeks from the onset of ripening (veraison, time point 0). Sunlight percentages for each treatment refer to the range of time under sunlight exposure: 0% corresponds to full shading of fruits by plant leaves, 100% corresponds to full sunlight exposure from veraison onward, generated by movement of leaves around the cluster region, and 25% shading from veraison until 6 weeks after which leaves were displaced for cluster illumination. Each of the two experiments is exemplified with a diagram at the bottom of the Figure. The experimental design consisted in four blocks with five plants each (biological replicates n = 4). Three berries per cluster (randomly sampled) and four clusters per plant were used for each sample. Gene expression in cv. 'Cabernet Sauvignon' berries is shown relative to UBIQUITIN1 expression. Values and standard deviations derived from one PCR run with duplicate PCR reactions on each of the four biological replicates. Stars indicate significant differences at indicated time points across treatments (p < 0.05) based on a two-way ANOVA followed by Tukey's post hoc test. Note that the lines in (B) indicate that at each time point in 0% versus 100% sun exposure differences in gene expression of VviFLS1 were statistically significant.

experiments (Supplementary Figures 2D,H and **Figure 5H**). In contrast, VviGT3 gene expression was not detectable in post-veraison samples such as in the light field experiment (**Figure 5D**), which correlates well with its low expression at late stages of development (**Figure 4E**).

While VviST1 transcript levels were detectable in the HR lines overexpressing VviMYBF1, in none of the light experiments expression of VviST1 was detectable, also correlating to its low expression found in the cv. 'Pinot Noir' developmental series (**Figure 4D**).

In order to correlate gene expression levels with flavonol accumulation, flavonols were quantified in the UV-B filtering field experiment. Flavonols show a significant increase in response to UV-B at different time points (Supplementary Figure 3B). When analyzing for individual flavonol compounds it was found that many flavonols are 3-O-glycosylated, as reported before (Carbonell-Bejerano et al., 2014) and the mostly affected were 3-O-glucosides of quercetin, kaempferol, and isorhamnetin (Supplementary Figure 4).

#### DISCUSSION

### Selection of VviMYBF1 Transgenic Hairy Root Lines for Microarray Analysis

Several studies have shown that target gene specificity of TFs may differ when introduced into heterologous hosts (Bovy et al., 2002; Luo et al., 2008). Therefore microarray-based transcriptome analysis was performed on the homologous system V. vinifera cv. 'Chardonnay' HRs overexpressing VviMYBF1 and compared to GFP expressing control roots. For microarray analysis, suitable HR lines were pre-selected by qPCR to detect the presence of significant amounts of VviMYBF1 and its known target gene, VviFLS1 (Czemmel et al., 2009). In situ staining of transgenic VviMYBF1 HRs for flavonol compounds clearly supported qPCR analysis and demonstrated that flavonols accumulated in VviMYBF1 transgenic roots but not in GFP (control) lines. Flavonol compounds were visible in yellow vacuolar-like cytoplasmic inclusion bodies but not uniformly located in all HR tissues. Interestingly, similar structures named anthocyanin vacuolar inclusions (AVIs) were observed after overexpression of the anthocyanin regulator VviMYBA1 (Cutanda-Perez et al., 2009). The sectors in VviMYBF1 root cross sections accumulating highest levels of flavonols include skin layers (epidermis, hypodermis), endodermis and vascular bundles, whereas no flavonols could be observed in the cortical parenchyma cells. As flavonol regulation by VviMYBF1 is not dependent on the presence of bHLH proteins (Czemmel et al., 2012), a lack of flavonol accumulation in HR tissues such as cortical parenchyma cells might result from either the presence of inhibitors (e.g., transcriptional repressors) or the lack of WD-repeat proteins or other yet unknown activators in these tissues. Main known flavonols found in VviMYBF1 HR extracts were quercetin-3-O-glucoside and also quercetin-3-Ogalactoside. Both of these quercetin-derived compounds have been reported as abundant compounds found in grape leaves and berries (Castillo-Muñoz et al., 2007; Downey and Rochfort, 2008).

### Overexpression of VviMYBF1 in Hairy Roots Identifies Novel Putative Flavonol Pathway Genes Controlled by UV-Containing Light

It is well-known that flavonols, in a similar way as anthocyanins, mainly exist as glycosylated forms whereas corresponding aglycones rarely exist in planta (Mattivi et al., 2006). Conjugation with glucose leads to increased water solubility, bioavailability and reduced toxicity, as respective reaction products can be subsequently stored in the vacuole. For example, kaempferol-diglucoside was suggested as one substrate for the tonoplast localized MATE protein Flower Flavonoid Transporter (FFT) in Arabidopsis (Thompson et al., 2010). Sugar moieties at the flavonol core skeleton also influence antioxidant potential (Vinson et al., 1999; Burda and Oleszek, 2001; Montoro et al., 2005) and they are able to protect polyphenols from enzymatic oxidation by plant peroxidases (Regev-Shoshani et al., 2003). These flavonol modifications, also in addition to their content and composition, can be remarkably different among grape varieties, during berry development (Downey et al., 2003; Mattivi et al., 2006; Castillo-Muñoz et al., 2007) and in response to UV-containing light (Price et al., 1995; Czemmel et al., 2009; Matus et al., 2009). Still, major knowledge gaps exist in the regulatory network underlying flavonol biosynthesis under developmental and environmental conditions that determine visible and UV-B light qualities.

Microarray analysis of VviMYBF1 overexpressing HR lines was therefore centered toward the identification of novel structural and regulatory genes involved in flavonol biosynthesis. Promising candidates, namely two glycosyltransferases (VviGT3, VviRhaT1) and a putative flavonol-sulfotransferase (VviST1) were identified using high throughput DE profiling. qPCR analysis of transgenic HRs overexpressing VviMYBF1 confirmed the microarray results and showed that VviGT3, VviRhaT1, and VviST1 are significantly upregulated (**Figure 4**).

Results in this work clearly support the idea that VviGT3 is under the control of VviMYBF1 to produce glycosylated flavonols in young developing berries. Despite the many existences of glycosylated flavonols in planta and their high importance as sunscreen pigments, when comparing grapevine to the model plant Arabidopsis (Yonekura-Sakakibara et al., 2007, 2008), only a few genes involved in flavonol glycosylation have been characterized in vitro (Ono et al., 2010). The same holds true for the identification of only a few structural genes involved in glycosylation of other flavonoids (PAs, anthocyanins) in grapes (Ford et al., 1998; Hugueney et al., 2009; Khater et al., 2012). VviGT3 is located in the same orientation on chromosome 11 in close vicinity to VviGT5 and VviGT6 (Ono et al., 2010). This tandem organization suggests that VviGT3 arose by tandem duplication and might provide another flavonol modifying gene. Sequence similarity between VviGT3 and VviGT5 (64%) or VviGT6 (64%) is lower compared to the sequence identity between VviGT5 and VviGT6 (88%), which implies functional differences. Although the GT3 protein has not been functionally categorized yet, co-expression analysis reveals some spatial and potential functional differences compared to GT5/GT6. VviGT5 and VviGT6 were shown to be co-expressed with VviFLS1 in berries, leaves and petioles (Ono et al., 2010) with higher GT5 expressions after the onset of ripening (Loyola et al., 2016). Results from light experiments and berry development show that VviGT3 transcripts are more profoundly expressed and inducible at early stages of development (**Figure 4E**), which correlates well with peaks of VviMYBF1 and VviFLS1 expression in grapes (Downey et al., 2003; Czemmel et al., 2009). Furthermore,

VviGT3 is light inducible only at early stages of development (**Figure 5H**) where it is expressed (**Figure 4E**). The absence of induction in the light field experiment (**Figure 5D**) can be explained by the fact that all samples used there correspond to post-veraison stages where VviGT3 is low expressed, as it was also demonstrated in the cv. 'Pinot Noir' developmental series (**Figure 4E**). VviGT3 is therefore regulated by light in a similar way as VviGT5 and VviGT6 and the flavonol marker genes VviFLS1 and VviMYBF1 (Carbonell-Bejerano et al., 2014). Neither in the cv. 'Corvina' gene atlas (Fasoli et al., 2012) nor in the cv. 'Pinot Noir' series VviGT3 expression was observed in skin samples of berries toward ripening (**Figure 4E**). These results indicate that at very early stages of development VviMYBF1 could regulate a cascade of GT expression profiles including VviGT3, VviGT5 and VviGT6 to attach sugar moieties [mainly of the class of 3-O-galactosides and 3-O-glucosides (Supplementary Figure 4)] to flavonol skeletons produced by the co-induced VviFLS1 gene. Later in berry development, flavonol modifications attached by VviGT3 might not be needed in ripening berry skins, in which the transcriptional network of VviMYBF1 activates the expression of other flavonol structural genes except VviGT3. All UV radiation-increased flavonols found in this study were glycosylated, corroborating the results of Carbonell-Bejerano et al. (2014). The structural similarities between GT5, GT6, and GT3 and the pronounced accumulation of 3-O-galactosides and 3-O-glucosides in the field experiment render the possibility that GT3, besides GT5 and GT6, contribute to the glycosylation of flavonols under UV-light regimes to provide non-toxic UV screens to young berries before veraison.

In conjunction with its albeit modest but present induction by UV-containing light during late stages of development, the VviRhaT1 gene was induced in VviMYBF1 HRs when using microarray analysis and confirmative qPCR analysis. The Grape Gene Atlas data showed that cDNA of VviRhaT1 is present ubiquitously in 50 out of 54 measured samples (Fasoli et al., 2012). It is expressed in all tissues where VviGT3 is expressed but in addition in many berry and fruit related tissues and additionally in leaves where VviRhaT1 expression correlates with VviMYBF1 and VviFLS1 transcript abundances. This correlates well with the uniform expression of this gene across the cv. 'Pinot Noir' developmental series (**Figure 4F**). VviRhaT1 shows an expression pattern very similar to flavonol accumulation on a per berry basis (Downey et al., 2003). This indicates its involvement in flavonol modification throughout rather than at the beginning of berry ripening at which stage VviMYBF1 and VviFLS1 expression peaks (Czemmel et al., 2009). This also supported by a co-induction of quercetin-rhamnoside (+1 WAV) and VviRhaT1 (0 and +3 WAV) gene expression around or post-veraison in the UV-B filtering field experiment (**Figure 5G** and Supplementary Figure 4H) indicating a putative role of VviRhaT1 as flavonol modifying enzyme. This function is further corroborated by the high sequence similarity to At2g22590 (UGT91A1) from Arabidopsis, which presumably encodes a rhamnosyltransferase under transcriptional control of the flavonol regulators MYB12, MYB11, and MYB111 (Stracke et al., 2007). This regulatory link was identified by transcriptome analysis on a flavonol deficient mutant (myb11-myb12-myb111) showing strong downregulation of the transcripts encoding UGT91A1. Transient expression analysis in Arabidopsis protoplasts confirmed that the promoter of UGT91A1 was responsive to flavonol regulators (Stracke et al., 2007).

Whereas O-glycosylation of flavonoids, which may be influenced by VviGT3 and VviRhaT1, has been at least partially studied in grapes, almost nothing is known regarding flavonoid sulfurylation in developing grape berries. In plants, flavonol sulfotransferases (STs) have been initially described in Flaveria spp. (Varin and Ibrahim, 1989) and sulfurylated flavonols have often been associated with the nucleus (Grandmaison and Ibrahim, 1995; Naoumkina and Dixon, 2008). In addition, another study suggests that sulfurylated flavonols might play a positive role in the regulation of polar auxin transport by acting as antagonist to quercetin (Faulkner and Rubery, 1992). A detailed examination of grapevine chromosome 11 revealed that genes encoding GTs alternate with two putative flavonol sulfotransferase genes with one of them, VviST1, being significantly overexpressed in VviMYBF1 transgenic root lines but not induced by sunlight or UV-B. Expression data regarding flavonol STs in grapes are sparse. The Grape Gene Atlas shows that VviST1 is only present in 13 out of 54 tissue samples and only in berry pericarp and berry flesh when considering fruit samples (Fasoli et al., 2012). VviST1 is expressed in roots but not in flowers, which contrasts the expression of VviMYBF1 in these samples. The low expression, as reported in the Grape Gene Atlas, is supported here when studying expression pattern of ST1 in developing berries (**Figure 4D**) but thereby shows some correlation with early expression of VviMYBF1 in developing berries (Czemmel et al., 2009).

Taken together, these results confirm a regulative effect of VviMYBF1 on VviFLS1, VviRhaT1, and VviGT3 gene expression in VviMYBF1 HRs, during berry development and at particular time points in response to sunlight and UV-B radiation. In contrast, greenhouse and UV-B field experiments did not gain any insight on a light-dependent regulation of the putative flavonol sulfotransferase gene VviST1. These results extend our knowledge about flavonol glycosylation in response to light and leave open the question whether flavonol sulfurylation can be modified by agronomical practices influencing light.

# VviMYBF1 Is Part of a Regulatory Cascade Involving VviHY5/VviHYH and the PA Regulator VviMYBPA1

Microarray analysis on VviMYBF1 HRs identified the grape HY5 homolog (VviHYH) and VviMYBPA1 as target of VviMYBF1 (**Table 1** and Supplementary Table 1). While the bZIP factor VviHY5 could not be analyzed in this work because it is not represented on the Grapegen GeneChip, transient expression of VviHY5 in grapevine plantlets resulted in an increase of the expression of VviMYBF1 (Loyola et al., 2016). Furthermore, the presence of HY5 binding elements typical for bZIP TFs in the promoters of VviMYBF1 and VviFLS1 genes was proven by in silico approaches (Czemmel et al., 2009). These results suggest a positive feedback loop of VviMYBF1 and VviHYH/VviHY5 in grapevine. A similar relationship is found

between the VviMYBF1 and VviHYH homologes AtMYB12 and AtHY5 in Arabidopsis. Lee et al. (2007) identified HY5 binding sites (ACEs) in the promoters of numerous Arabidopsis genes including AtCHS, AtFLS and most prominently AtMYB12 (Lee et al., 2007). This information was used by Stracke et al. (2010) to demonstrate that in Arabidopsis HY5 is required for the transcriptional activation of the AtMYB12 and AtMYB111 genes under UV-B and visible light. This is consistent with the observation that the hy5 mutant shows flavonoid-deficient roots (Sibout et al., 2006) and downregulation of AtMYB12 transcripts (Stracke et al., 2010). Taken together, these results support the theory that the UV-B response machinery (including the TFs VviHY5 and VviMYBF1) exist in grapes to propel flavonol accumulation through the activation of the regulatory network consisting of both, MYB and bZIP TFs (Malacarne et al., 2016; Matus, 2016).

Our transcriptome study also demonstrated a downregulation of structural and regulatory genes of the proanthocyanidin (PA) pathway by VviMYBF1 and therefore indicated a competition between the flavonol and PA branch of the pathway (**Table 1**). Indeed, qPCR analysis on VviMYBF1 HRs showed that there is a negative influence of the flavonol regulator VviMYBF1 on the expression of the PA regulatory TF VviMYBPA1 (**Table 1** and Supplementary Figure 1). Concomitantly, genes which were shown to be under control of VviMYBPA1 such as two CHALCONE SYNTHASE isoforms, ANR and CHI (Terrier et al., 2009) were also downregulated. These results are in line with previous findings that also proposed a competition between flavonol and PA/anthocyanin biosynthesis (Czemmel et al., 2009). Interestingly, an inverse relationship – a differential regulation of VviMYBF1 expression by VviMYBPA1 – seems not to exist as the expression of the key target of VviMYBF1, VviFLS1 was not altered in VviMYBPA1 HRs (Terrier et al., 2009). In context with the contrasting expression profiles of VviMYBF1 and VviMYBPA1 in developing cv. 'Shiraz' berries (Bogs et al., 2007; Czemmel et al., 2009), the data presented here imply that VviMYBF1 is a positive regulator of flavonol biosynthesis at the expense of PA accumulation via downregulation of VviMYBPA1 transcription during initial berry ripening stages.

# CONCLUSION

In this study the grapevine flavonol regulator VviMYBF1 was overexpressed in the homologous model system of cv. 'Chardonnay' HRs. Microarrays, qPCR analysis and light/UV exclusion experiments identified promising novel flavonol biosynthetic genes, such as a flavonol glucosyltransferase (VviGT3) and a rhamnosyltransferase (VviRhaT1). These structural genes in context with regulatory genes such as VviHYH, both being under control by VviMYBF1, could play a role in the biosynthesis of flavonols, acting downstream of VviFLS1 and playing a role in UV light responses. Subsequent biochemical characterization of the substrate specificity of the structural candidate genes and the transcriptional potential of the identified VviMYBF1 targeted TFs will foster our understanding of flavonol biodiversity in fruit and thrive the development of molecular tools and agronomical practices to optimize flavonol biosynthesis in response to UV light. As flavonols are more and more explored as important quality determinants of fruit-derived products and have been suggested as quality markers for different grape varieties (Ritchey and Waterhouse, 1999; Hermosín-Gutiérrez et al., 2011), genes involved in flavonol biosynthesis could be implemented as molecular traits during marker-assisted breeding approaches. These techniques will provide tools to the wine industry to optimize fruit quality by adaptation of flavonol content and composition using viticultural practices such as optimization of light regimens.

# AUTHOR CONTRIBUTIONS

JB conceived and designed the study. SC wrote the manuscript, performed the experiments related to HRs except the qPCR work, analyzed all the data and run the bioinformatics workflow for microarray analysis at QBiC. JH did the qPCR analysis experiments in HRs and developmental series. JM, RL, PA-J, JA performed the sunlight and UV-B greenhouse and field experiments. JM revised the study critically for important intellectual content and together with JH and JB revised the manuscript.

# FUNDING

We acknowledge financial support for this research from the Bundesministerium für Bildung und Forschung and its initiative Genomanalyse im Biologischen System Pflanze, Comisión Nacional de Investigación Científica y Tecnológica, Chile (CONICYT; Ph.D. grant no. 21120255 to RL), Fondo Nacional de Desarrollo Científico y Tecnológico (FONDECYT grants no. 1150220, 1040551 and FONDECYT postdoctoral grant no. 3150578 to RL) and Programs ECOS-CONICYT (grant no. C11B01) and Millennium Nucleus for Plant Synthetic Biology and Systems Biology NC130030.

# ACKNOWLEDGMENT

We especially thank Cornelia Walter for excellent technical lab assistance and Pablo Carbonell-Bejerano and José Martínez-Zapater to obtain high quality microarray data with the Affymetrix custom-made Grapegen GeneChip.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2017.01084/ full#supplementary-material

# REFERENCES

fpls-08-01084 June 20, 2017 Time: 17:4 # 14



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Czemmel, Höll, Loyola, Arce-Johnson, Alcalde, Matus and Bogs. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Dissecting the Biochemical and Transcriptomic Effects of a Locally Applied Heat Treatment on Developing Cabernet Sauvignon Grape Berries

Fatma Lecourieux 1‡, Christian Kappel 2 †‡, Philippe Pieri <sup>2</sup> , Justine Charon<sup>2</sup> , Jérémy Pillet 2 † , Ghislaine Hilbert <sup>2</sup> , Christel Renaud<sup>2</sup> , Eric Gomès <sup>3</sup> , Serge Delrot <sup>3</sup> and David Lecourieux <sup>3</sup> \*

<sup>1</sup> Centre National de la Recherche Scientifique, Institut des Sciences de la Vigne et du Vin, UMR Ecophysiologie et Génomique Fonctionnelle de la Vigne, Villenave d'Ornon, France, <sup>2</sup> Institut National de la Recherche Agronomique (INRA), Institut des Sciences de la Vigne et du Vin, UMR Ecophysiologie et Génomique Fonctionnelle de la Vigne, Villenave d'Ornon, France, <sup>3</sup> Université de Bordeaux, Institut des Sciences de la Vigne et du Vin, UMR Ecophysiologie et Génomique Fonctionnelle de la Vigne, Villenave d'Ornon, France

Reproductive development of grapevine and berry composition are both strongly influenced by temperature. To date, the molecular mechanisms involved in grapevine berries response to high temperatures are poorly understood. Unlike recent data that addressed the effects on berry development of elevated temperatures applied at the whole plant level, the present work particularly focuses on the fruit responses triggered by direct exposure to heat treatment (HT). In the context of climate change, this work focusing on temperature effect at the microclimate level is of particular interest as it can help to better understand the consequences of leaf removal (a common viticultural practice) on berry development. HT (+ 8 ◦C) was locally applied to clusters from Cabernet Sauvignon fruiting cuttings at three different developmental stages (middle green, veraison and middle ripening). Samples were collected 1, 7, and 14 days after treatment and used for metabolic and transcriptomic analyses. The results showed dramatic and specific biochemical and transcriptomic changes in heat exposed berries, depending on the developmental stage and the stress duration. When applied at the herbaceous stage, HT delayed the onset of veraison. Heating also strongly altered the berry concentration of amino acids and organic acids (e.g., phenylalanine, γ-aminobutyric acid and malate) and decreased the anthocyanin content at maturity. These physiological alterations could be partly explained by the deep remodeling of transcriptome in heated berries. More than 7000 genes were deregulated in at least one of the nine experimental conditions. The most affected processes belong to the categories "stress responses," "protein metabolism" and "secondary metabolism," highlighting the intrinsic capacity of grape berries to perceive HT and to build adaptive responses. Additionally, important changes in processes related to "transport," "hormone" and "cell wall" might contribute

#### Edited by:

José Tomás Matus, Centre for Research in Agricultural Genomics (CRAG), Spain

#### Reviewed by:

Alessandro Vannozzi, University of Padua, Italy Claudio Pastenes, University of Chile, Chile

#### \*Correspondence:

David Lecourieux david.lecourieux@inra.fr

#### † Present Address:

Christian Kappel, Institut für Biochemie und Biologie, Universität Potsdam, Potsdam, Germany; Jérémy Pillet, Laboratorio de Bioquímica y Biotecnología Vegetal, Universidad de Málaga, Málaga, Spain ‡ These authors have contributed equally to this work.

#### Specialty section:

This article was submitted to Crop Science and Horticulture, a section of the journal Frontiers in Plant Science

Received: 27 October 2016 Accepted: 10 January 2017 Published: 31 January 2017

#### Citation:

Lecourieux F, Kappel C, Pieri P, Charon J, Pillet J, Hilbert G, Renaud C, Gomès E, Delrot S and Lecourieux D (2017) Dissecting the Biochemical and Transcriptomic Effects of a Locally Applied Heat Treatment on Developing Cabernet Sauvignon Grape Berries. Front. Plant Sci. 8:53. doi: 10.3389/fpls.2017.00053

**84**

to the postponing of veraison. Finally, opposite effects depending on heating duration were observed for genes encoding enzymes of the general phenylpropanoid pathway, suggesting that the HT-induced decrease in anthocyanin content may result from a combination of transcript abundance and product degradation.

Keywords: grapevine, berry development, microclimate, high temperature, microarrays, metabolomics/metabolite profiling, climate change

## INTRODUCTION

Grapevine is probably the fruit species whose economical value depends the most on climatic conditions. Besides the soil and local viticultural practices, which are components of the socalled "terroir" effect, grapevine and wine quality are determined by vintage that integrates the different climatic parameters experienced by the plant during growth and ripening. Among these environmental factors, temperature is a major regulator affecting both grapevine phenology and fruit composition (Schultz, 2000; Jones et al., 2005).

The Intergovernmental Panel on Climate Change (IPCC) predicts a significant increase of mean temperatures ranging between 1.8 and 4◦C by the end of the 21st century, depending on the scenario (IPCC, 2013). Given that grapevine varieties have been ranked according to their thermal preferences (Huglin, 1978), this may lead to change the varieties traditionally grown in AOC areas, or even threaten the sustainability of some Mediterranean viticultural areas (Hannah et al., 2013; Fraga et al., 2016).

However, mean temperature perceived by the plant may be altered by common viticultural practices that may be envisaged as a mean to cope with the effects of climate change (Van Leeuwen et al., 2013). For example, the extent of leaf removal directly affects light exposure and temperature of the berries. Berry temperatures depend on row orientation and significantly differ between east and west-exposed sides in the case of north-south row orientation. The temperature also interacts with the leaf/fruit (source/sink) relationships to determine yield (Sadras and Moran, 2013) and with the circadian clock to determine transcriptomic responses (Rienth et al., 2014). Moderate increases of berry temperatures obtained in open-top chambers uncouple the anthocyanin and sugar contents in red varieties (Sadras and Moran, 2012) and affect berry sensory traits (Sadras et al., 2013). Because of the direct irradiative effects, the temperature of reproductive organs may differ very significantly from that of the whole plant or surrounding air (Spayd et al., 2002; Pieri and Fermaud, 2005; Cola et al., 2009).

Secondary metabolites, which are among the most prominent discriminating compounds for berry quality, are synthesized in situ under the direct influence of the local micro-environment (mostly temperature and light conditions). In this context, it is important to investigate the effects of high temperatures on berry metabolism and composition. These effects depend on the imposed temperature gradient, and on the time and duration of the heat stress. During flowering, high temperatures inhibit berry set and reduce yield (Greer and Weston, 2010). After fruit set, high temperatures stimulate sugar accumulation at the expense of other qualitative compounds (Greer and Weston, 2010) such as organic acids (Champagnol, 1984), flavonols, anthocyanins, amino acids (Spayd et al., 2002; Pereira et al., 2006; Mori et al., 2007; Tarara et al., 2008; Cohen et al., 2012) and aromas (Schultz, 2000). Together with other parameters such as the carbohydrate source-sink ratio (Ollat and Gaudillère, 1998; Parker et al., 2014; de Toda and Balda, 2015), ambient air temperature is a wellknown factor that influences berry growth and veraison timing (Duchene et al., 2010), but the impact of heat treatment (HT) at the berry level on ripening onset has not been described extensively.

In the present work, we compare the effects of HT (+8 ◦C, 14 days) imposed on berry clusters at 3 different stages (green, veraison, ripening) of berry development. This experimental setup avoids the complexity of the responses involving both light and heat effects, and allows a detailed study of the relevance of developmental stages in the response to HT. Biochemical analysis and gene expression studies were conducted at 3 time points (1, 7, and 14 days) after the beginning of HT, which also allowed us to study the time course of the responses. Our results showed dramatic and specific biochemical and transcriptomic changes in heat exposed berries, depending on the developmental stage and the stress duration. Heating strongly altered the berry concentration of amino acids and organic acids and decreased the anthocyanin content at maturity. When applied at the herbaceous stage, HT delayed the onset of veraison. More than 7000 genes were deregulated in at least one of the nine experimental conditions contributing to the postponing of veraison. Finally, opposite effects depending on heating duration were observed for genes encoding enzymes of the general phenylpropanoid pathway, suggesting that the HT-induced decrease in anthocyanin content may result from a combination of transcript abundance and metabolite degradation.

# MATERIALS AND METHODS

#### Plant Material

Fruiting cuttings of Vitis vinifera L. cv Cabernet Sauvignon (Ollat and Gaudillère, 1998) were grown in a greenhouse, in 0.5 L pots containing a mixture of perlite, sand and vermiculite (1:1:1). A drip irrigation system supplied water and a complete nutrient solution to the roots five times a day all along the experiment, avoiding any water or nutrient shortage. All fruiting cuttings bore only one single cluster and lateral shoots were removed as soon as they appeared during growth. Before the experiment, the tip of each shoot was removed as soon as 16 leaves per plant were produced to maintain approximately the same leaf area in all plants and a high leaf to fruit ratio (Ollat and Gaudillère, 1998). Therefore, all bunches were assumed to experience neither water nor assimilate limitation. In addition, the fruit cuttings were selected on the basis of similar vegetative growth and vigor as well as size and compactness of bunches.

## Heat Treatment, Temperature Measurements, and Sampling

Both control and heat treatments were applied for 14 d periods, at 3 phenological stages, namely middle-green (30 days after fruit set), veraison (50% of berries turning to a visible red color) and middle-ripening (80 days after fruit set). Total soluble solids (TSS) content in berries was determined before, during and after each treatment using a digital refractometer (Atago, Tokyo). Before treatments, middle-green, veraison and middleripening berries displayed a TSS of 3.9 ± 0.3, 8.9 ± 1.4 and 14.1 ± 1.1◦Bx, respectively (mean of replicates ± SD). Three sets of 25 fruiting cuttings were used as controls [GC (Green Control), VC (Veraison Control) and RC (Ripening Control)]; the temperature of their bunch closely followed greenhouse ambient air temperature. Three sets of 25 bunches from other fruiting cuttings were submitted to heat treatment: GHT (Green Heat Treatment), VHT (Veraison Heat Treatment) and RHT (Ripening Heat Treatment). The clusters were submitted to an elevated temperature airflow produced by fan heaters (common domestic models, used at 1000 W). Only bunches were heated since shoots, leaves and roots were all protected from the heated airflow by extruded polystyrene foam deflectors. Air heating was applied repeatedly during 14 d, from 7:00 a.m. to 7:00 p.m. every day to mimic the usual diurnal temperature course of exposed berries. To avoid any differential effect linked to airflow and possible mechanical stress, simple fan blowers were used to create a continuous airflow around the control clusters during 14 d, from 7:00 a.m. to 7:00 p.m. every day. The flesh temperatures in control and heated berries (10 replicates for each treatment) were monitored continuously by copper-constantan thermocouples inserted into the berries, connected to a Campbell datalogger (Campbell Scientific). After treatments, each set of fruiting cuttings was replaced in control conditions until harvest.

The experiment was conducted over 3 years. The samples collected the first year were used for biochemical and transcriptomic analysis. Samples from years 2 and 3 were used for biochemical and RT-qPCR analysis.

#### Sampling

In order to analyse short- and long-term responses for each treatment, 3 control and 3 heat-stressed berries per cluster (25 clusters per condition) were sampled in the evening (7:00 p.m.) at 3 different time points after treatment (1, 7, and 14 d), immediately frozen in liquid nitrogen and stored at −80◦C. To obtain 3 experimental replicates, each of the 3 berries collected per cluster was used to constitute 3 independent groups of 25 berries for each condition. The pool of deseeded berries from each group was used as a biological replication and underwent independent RNA extractions.

For the biochemical analysis conducted around harvest, 3 biological replicates were prepared as described above, with berries collected from the same 25 fruiting cutting used for each treatment. Then, frozen berries (5 berries per replicate) were slightly thawed and separated quickly into skin, pulp, and seed. The skin and pulp were immediately ground into fine powder in liquid nitrogen using mortar and pestle.

# Metabolites Quantification

#### Sugars, Organic Acids, and Amino Acids Contents

An aliquot of 500 mg fine powder of pulp was extracted sequentially with ethanol (80 and 50%), dried in a Speed-Vac, and re-dissolved in 2.5 mL de-ionized water. Glucose and fructose content were measured enzymatically with an automated micro-plate reader (Elx800UV, Biotek Instruments Inc., Winooski, VT, USA) according to the method of Gomez et al. (2007). Tartaric acid content was assessed by using the colorimetric method based on ammonium vanadate reactions (Pereira et al., 2006). Malic acid was determined using an enzyme-coupled spectrophotometric method that measures the change in absorbance at 340 nm from the reduction of NAD+ to NADH (Pereira et al., 2006). The amino acid content was determined after derivatization (Cohen and Michaud, 1993) using a Waters 2695 HPLC system equipped with Waters 474 fluorescence detector (Waters, Milford, MA, USA). Twenty amino acids were identified and quantified as described by Pereira et al. (2006). The results were expressed in µmol.L−<sup>1</sup> juice.

#### Anthocyanins Quantification

An aliquot of 500 mg of berry skin powder was freeze-dried for 72 h and the dried powder (50 mg) was extracted in 1.0 mL methanol containing 0.1% HCl (v/v). Extracts were filtered through a 0.45 µm polypropylene syringe filter (Pall Gelman Corp., Ann Harbor, MI, USA) for HPLC analysis. Each individual anthocyanin was analyzed with HPLC as described in Soubeyrand et al. (2014). Quantification was carried out by peak area integration at 520 nm. The concentration of individual anthocyanins was calculated in milligrams per gram (mg. g−<sup>1</sup> ) of dry skin weight (DW) using malvidin 3-O-glucoside (Extrasynthese, Genay, France) as external standard.

# RNA Extraction and cDNA Production

Berries collected from Cabernet Sauvignon fruit cuttings were quickly frozen in liquid nitrogen, ground to a fine powder with a Dangoumau blender, and stored at −80◦C prior to use. Total RNA from deseeded berries was extracted according to Lecourieux et al. (2010). RNA isolation was followed by DNase I treatment. The purity and quantity of the RNA were determined using a Nanodrop 1000 spectrophotometer (Thermo Scientific). RNA integrity was determined using a Bioanalyzer 2100 (Agilent) with RNA 6000 Nano Kit I (Agilent). For each sample, reverse transcription was performed from 2 µg of purified RNA using the Moloney murine leukemia virus reverse transcriptase (Promega) according to the manufacturer's instructions. The cDNA obtained was diluted (1:10) in distilled water.

# Microarray Hybridization and Data Processing

Total RNAs extracted from 3 biological replicates per condition were hybridized with 60-mer oligoarrays bearing a set of probes for 29,582 unigenes (NimbleGen Gene Expression 12 × 135K Arrays). Labeling, hybridization and scanning were carried out at the GeT-transcriptomic platform (GenoToul-Toulouse, France). Microarray data were analyzed using the R and R/Bioconductor software (Gentleman et al., 2004; R Core Team, 2013). Quality control was done using the arrayQualityMetrics package (Kauffmann et al., 2009), one identified outlier (GHS14\_1) was excluded from further analyses. Data were normalized using the Robust Multi-array Average algorithm (RMA) (Irizarry et al., 2003). Principal component analysis was done using the prcomp function, visualizations were made using ggplot2 (Wickham, 2009). Differentially expressed genes between treatment and control samples for all stages and treatment durations were identified using Limma package (Smyth, 2005). Differentials with absolute fold changes above 2 and BH (Benjamini and Hochberg, 1995) corrected P-values below 0.05 were considered significant. Significantly affected gene categories based on the MapMan Ontology (Thimm et al., 2004; Usadel et al., 2005) were identified using a Chi-square test. MapMan mappings for the Cribi 12X grapevine genome were based on closest homologs regarding to the Arabidopsis thaliana genome. Empirical cumulative distribution function plots for selected categories were made using the latticeExtra package. Gene annotations were taken from Grimplet et al. (2012).

The raw microarray data were submitted to Gene Expression Omnibus (NCBI) and are accessible through GEO accession number GSE86551.

# RESULTS

### Heat Treatment and Berry Temperature Recording

HT was applied locally on clusters of Cabernet Sauvignon fruiting cuttings at 3 developmental stages, namely middle-green, veraison and middle-ripening (**Figure 1A**). Three different sets of 25 plants per condition (control or HT) were used for each developmental stage. HT was applied every day during 14 d from 7:00 a.m. to 7:00 p.m. to mimic usual daytime temperature course of sunlight-exposed berries. Under control conditions, average daytime berry flesh temperature measured were 26.5◦C ± 1.4 (control green stage, GC clusters), 25.8◦C ± 2.2 (control veraison stage, VC clusters), and 26.2◦C ± 2.5 (control ripe stage, RC clusters). HT increased the average daytime pulp temperature to 34.7 ± 1.4◦C, 33.7 ± 1.9◦C, and 35.2 ± 2.5◦C for GHT, VHT, and RHT clusters respectively (**Figure 1B**). Depending on the developmental stage, this experimental set-up led to average pulp temperature differences of 8.2◦C ± 1.3 (GHT vs. GC), 7.9◦C ± 1.1 (VHT vs. VC) and 9.0◦C ± 0.8 (RHT vs. RC) between heat-treated and control berries (**Figure 1B**).

The HT imposed during the herbaceous stage delayed the onset of anthocyanin accumulation by 2 to 3 weeks (GHT vs GC berries), (**Figure 2A**). Furthermore, for control clusters 4 weeks were needed to reach complete color turning stage (100% clusters (n = 25) while at least one additional week was required to reach the same stage in the heated clusters. However, even at the last stage, the percentage of uncolored berries per cluster (46%) remained much higher than in control clusters (8%) (**Figure 2C**). A similar delay in the increase in total soluble solids (TSS) content was also observed in stressed berries (**Figure 2B**) while the average berry weight was not affected by the treatments (data not shown).

# Biochemical Analysis of the Treated Berries

The biochemical content of berries collected at harvest from control fruiting cuttings was compared with that of berries exposed to a 2-week period HT at 3 different developmental stages (**Table 1**). The berry hexose content (glucose and fructose) at harvest was not affected whenever the fruit-localized HT took place. By contrast, applying HT at veraison and to a lesser extent at ripening stage decreased malate content while tartrate content was slightly higher in VHT clusters than in other samples. In the present study, 20 amino acids including the non-proteinogenic ones were quantified. The concentration of 7 amino acids (THR, ARG, TYR, PHE, CYS, LYS, GABA) was significantly increased by a HT applied at veraison or ripening stages. Conversely, the PRO amount decreased dramatically after mid-ripening treatment (**Table 1**).

The total anthocyanin concentration (TSA) in berry skin at harvest was reduced by about 50% in VHT and RHT when compared to control and GHT berries (**Table 1**). Interestingly, the GHT berries that colored later (**Figure 1**, Supplementary data) displayed a TSA comparable to control berries at harvest. Cabernet Sauvignon berries contained higher amounts of tri-hydroxylated anthocyanins than dihydroxylated ones in both control and heated conditions. Heat exposure preferentially decreased the proportion of dihydroxylated anthocyanins, regardless of the period of HT treatment (**Table 1**). Acylated anthocyanins represented a higher proportion of the total anthocyanin pool in HT berries (**Table 1**). The concentration of malvidin-3-O-glucoside, which is the most abundant anthocyanin in Cabernet Sauvignon berries (Dimitrovska et al., 2011; Lorrain et al., 2011) remained unaffected at harvest whereas the amount of one of its acylated forms (malvidin-3-O-(6′ -acetyl)-glucoside) was significantly increased in GHT clusters. By contrast, the concentrations of dephinidin, cyanidin, petunidin, peonidin and of their 3-acetylglucoside derivatives were severely reduced by heat irrespectively of the period of treatment in GHT, VHT and RHT berries. The inhibiting effect of HT on the corresponding 3-coumaroylglucoside derivatives was less pronounced.

#### Alteration of Global Berry Transcriptome in Response to Localized Heat Treatment

Principal component analysis of the whole normalized gene expression dataset showed that the 3 replicates of each experimental condition are well-grouped and therefore adequate for further analysis (**Figure 3**; Supplementary Figure 1). Principal

component 1 (PC1) and PC2 explained 26% of the total variance in gene expression and can be attributed to development. PC1 (18.9%) clearly separated the green stage from both veraison and ripening stages, whereas separation between veraison and the two other stages can be distinguished on PC2 (7.1%) (Supplementary Figure 1). PC3 explained 6.2% of the total variance and splits HT from control samples in a similar proportion whatever the developmental stage. Finally, the samples were clearly separated according to the treatment duration, potentially reflecting a developmental dependent response (**Figure 3**).

A total of 7518 transcripts were differentially expressed (fold change > 2, p-value adj. < 0.05) in at least one of the 9 conditions (Supplementary Table 1), corresponding to 25.4% of the unigenes represented on the microarray slide. The overlaps in DEGs under these nine conditions were depicted with 3-way Venn diagrams, according to developmental stage (**Figure 4A**) or stress duration (**Figure 4B**). The strongest HT effect was observed for the herbaceous stage with 5287 DEGs, compared to 4122 and 5061 DEGs identified after HT exposure at veraison and ripening stage, respectively (**Figure 4A**; Supplementary Table 2). A total number of 4141, 6612, and 3717 genes were deregulated after 1, 7, and 14 days of HT, respectively (**Figure 4B**; Supplementary Table 2). Except for one condition (V14D), the number of up-regulated genes was always higher than the number of

down-regulated genes. The strongest HT effect was observed at G7D with 4024 DEGs (2135 up-regulated and 1889 downregulated genes, 7 and 6% of the grapevine unigenes, respectively) whereas the smallest effect (577 DEGs) was observed in G1D. The comparison of all 9 conditions showed that only 36 genes were steadily induced under HT condition whereas continuously



Metabolite concentrations were determined around harvest. Means ± SE are shown (n = 3). Different letters indicate significant differences between control and stress conditions within a given compound at P < 0.05 accoring to a Tukey's test. CT, control; GHT, heat treatment applied during the herbaceous stage; VHT, heat treatment applied during veraison stage; RHT, heat treatment applied during the ripening stage. nd, not detected.

down-regulated transcripts were not found (Supplementary Table 3). These genes will be further considered in the discussion section below.

### Identification of Significantly Altered Functional Categories According to the Stage and the Treatment Duration

To gain insight into the functional categories impacted by HT, the 7518 DEGs were distributed into 35 MapMan functional categories, the so-called BINs (Usadel et al., 2005). The proportion of DEGs representing each category was determined for each of the nine experimental conditions (**Figure 5**). Functional enrichment analysis was performed to better identify the significantly altered functional categories according to the stage and the stress duration. HT triggered a wide range of effects on the berry transcriptome (**Figure 5**; Supplementary Table 4). Twenty-seven of the 35 MapMan BINs were significantly altered in HT berries in at least one condition, but only 3 functional categories, namely "Stress" (BIN 20), "Protein" (BIN 29) and "Secondary metabolism" (BIN 16), were profoundly affected in all 9 conditions. The 24 other categories were differentially affected according to the developmental stage and the stress duration, and particularly correspond to Photosynthesis (BIN 1), Cell wall (BIN 10), Hormone metabolism (BIN 17), RNA (BIN 27), DNA (BIN 28), Signaling (BIN 30), and Transport (BIN 34).

#### "Stress" -Associated BIN

Four hundred twenty seven DEGs were related to the functional category "Stress" and mainly belonged to the "abiotic/heat stress" cluster (Supplementary Table 5; 98 out of the 427 "Stress" DEGs). Most of the transcripts associated to "heat stress" category were predominantly up-regulated (Supplementary Figure 2A) and belong to the HSP family (Heat Shock Protein). Transcripts encoding universal stress proteins (USPs; VIT\_17s0000g04260, VIT\_04s0079g00610, VIT\_08s0032g00590) also accumulated in response to HT.

#### "Protein" -Associated BIN

HT deeply affects protein homeostasis. Indeed, numerous HSP and chaperones genes were up-regulated in HT berries (Supplementary Tables 5, 6). These proteins play an important role in protein-protein interactions (Kotak et al., 2007a; Bokszczanin and Fragkostefanakis, 2013). The abundance of several FK506-binding protein (FKBP) related transcripts was also increased by HT (VIT\_19s0015g01100, VIT\_07s0031g01150, VIT\_08s0007g04340, VIT\_13s0064g00580, VIT\_01s0011g00930, VIT\_00s0260g00070). These proteins belong to the large family of peptidyl prolyl cis–trans isomerases that can function as chaperones. Protein synthesis and degradation were also strongly affected in heated berries as suggested by the 60 DEGs linked to protein synthesis category and the 348 genes belonging to proteolysis (Supplementary Tables 5, 6).

Protein degradation through the ubiquitin-proteasome system (UPS) plays an essential role in diverse cellular pathways, cell-cycle progression, DNA repair, and degradation of damaged proteins as well as in signal transduction. Particularly, UPS components are major players in plant acclimation to abiotic stresses (Stone, 2014; Guerra et al., 2015). In the present study, 167 transcripts potentially linked to the ubiquitin machinery were deregulated after HT (102 up-, 56 down-, 9 both up- and down-regulated transcripts; Supplementary Table 6). Most of these DEGs encode different putative ubiquitin ligases (E3). Orthologs of each of the 3 major E3 classes (namely RING-type (Really Interesting New Gene); HECT-type (Homology to E6-Associated Carboxyl-Terminus), and U-box-type) were up-regulated in heatstressed berries. While some E3-like transcripts accumulated whatever the developmental stage (VIT\_09s0002g00220, VIT\_12s0034g01390, VIT\_05s0124g00230, VIT\_06s0009g03670, VIT\_08s0040g02600, VIT\_19s0027g00320), most of these were transiently or specifically deregulated at a particular stage. For instance, some E3-related transcripts were only HT upregulated during the herbaceous stage (VIT\_17s0000g09790, VIT\_18s0041g01090, VIT\_14s0068g02150, VIT\_14s0066g02580, VIT\_08s0056g01410) whereas others responded to HT during veraison or ripening stages (VIT\_19s0015g00660, VIT\_01s0011g02950, VIT\_01s0026g00300, VIT\_07s0005g01360, VIT\_08s0007g04790, VIT\_18s0001g02280, VIT\_00s0160g00270, VIT\_18s0001g06220). Furthermore, the CRL group (Cullin based Ring E3 ligases), corresponding to the largest class of ubiquitin ligases (Stone, 2014), and especially the CUL1 based E3s, [also referred to as Skp1-Cullin-F-box (SCF)] were strongly impacted by HT. Indeed, an ortholog of the Arabidopsis adaptor protein S-Phase kinase-associated protein (SKP; VIT\_03s0038g02480) and many F-box proteins are up-regulated at the transcriptional level upon HT.

#### "Secondary Metabolism" -Associated BIN

The secondary metabolism produces compounds of critical importance for berry quality and wine bitterness and astringency

(Lund and Bohlmann, 2006; Ali et al., 2010; Kuhn et al., 2014; Robinson et al., 2014). Three hundred twenty seven DEGs were related to "Secondary metabolism" and mainly belonged to the subcategories "isoprenoids," "phenylpropanoid-lignin," and "flavonoids" (Supplementary Table 7).

Numerous genes affected by HT correspond to aroma and aroma-precursor related gene. Terpenes (predominantly eucalyptol, β-caryophyllene, and α-humulene) are usually present at low levels in Cabernet Sauvignon grapes, accumulating during the preveraison stage whereas benzene derivatives (2 phenylethanol and 2-phenylethanal) appear at late ripening (Kalua and Boss, 2009; Robinson et al., 2014). The strong repression of genes encoding the 1-deoxy-D-xylulose-5 phosphate synthase (VIT\_05s0020g02130, VIT\_09s0002g02050, VIT\_11s0052g01730, VIT\_11s0052g01780) suggests that volatile terpenoids biosynthesis may be decreased by HT (Supplementary Table 7). The 1-deoxy-D-xylulose-5-phosphate synthase catalyzes the synthesis of isopentenylpyrophosphate (IPP), which is the precursor of all terpenes. The condensation of IPP and its isomer DMAPP (dimethylallylpyrophosphate) forms geranyl diphosphate (GPP) that is used by terpene synthase (TPS) to produce monoterpenes and derivatives. The Pinot Noir reference genome contains 89 putative TPS, among which half have been functionally characterized (Martin et al., 2010). Among the 55 probe sets giving reliable results and representing transcripts of functional, partial and pseudo TPS on the NimbleGen array (Cramer et al., 2014), 16 transcripts showed differential abundance in HT berries (Supplementary Table 8). Thirteen TPS were transiently repressed in GHT, VHT and or RHT clusters whereas only three TPS were upregulated by HT (VviTPS25, VviTPS26, VviTPS50). From the literature, most of these TPS accumulate at the late stages of ripening in vineyard conditions (Cramer et al., 2014). Finally, the transcript abundance of several terpene-related genes decreased in berries directly exposed to HT, in a similar way to that observed after exposing the whole vine to HT (Rienth et al., 2014). These repressed genes encode geraniol 10-hydroxylase, (-)-germacrene D synthase and linalool synthase (Supplementary Table 8).

Within the terpene family, carotenoids are a complex subgroup of isoprenoid pigments playing diverse roles in plants and providing nutritional value. Carotenoids also lead to C13-norisoprenoids which contribute the characteristic aromas of Vitis vinifera varieties (Mendes-Pinto, 2009). Among the 42 putative grape carotenoid metabolic genes (Young et al., 2012), 21 transcripts showed differential abundance after HT in at least one of the 9 conditions, most of these being down-regulated (Supplementary Tables 7, 9). The abundance of 2 out of 3 phytoene synthase transcripts (VviPSY, VIT\_12s0028g00960,

VIT\_06s0004g00820), encoding the enzyme that catalyzes the first step committed to carotenoid biosynthesis, decreased upon HT. This down-regulation was also observed for downstream genes, namely phytoene dehydrogenase (VIT\_04s0023g01790), carotene desaturase (VIT\_14s0030g01740), carotenoid isomerase (VIT\_08s0032g00800, VIT\_12s0035g01080), lycopene cyclase (VIT\_06s0080g00810, VIT\_11s0016g01880) and carotene hydroxylase (VIT\_04s0023g00080). The β-carotene hydroxylase 2 transcripts (VviBCH2, VIT\_16s0050g01090) were the only ones strongly accumulating under HT, and especially at the herbaceous stage.

Methoxypyrazines (MPs) are strongly odorant volatile molecules with vegetable-like fragrances that participate to the distinct herbaceous/ bell pepper characters of some wines such as Cabernet Sauvignon (Dunlevy et al., 2009; Darriet et al., 2012; Kuhn et al., 2014). Isobutyl methoxypyrazine (IBMP) is the predominant MP in Cabernet Sauvignon berries, accumulating throughout the pre-veraison stage before declining during the ripening phase. The last step of its biosynthesis was recently deciphered in grapevine, through the unambiguous identification of VviOMT3 (VIT\_03s0038g03090), an O-methyltransferase capable of converting the nonvolatile precursor 2-hydroxy-3-isobutylpyrazine (IBHP) into IBMP (Dunlevy et al., 2013; Guillaumie et al., 2013). By contrast to VviOMT4, VviOMT1 and VviOMT2 are active with a broad range of substrates but methylate IBHP in vitro, although with poor affinity (Dunlevy et al., 2010). Interestingly, VviOMT3 transcripts were strongly repressed in green berries exposed to HT (Supplementary Table 7). A similar decrease was also observed in GHT berries for VviOMT4 (VIT\_03s0038g03080) and VviOMT1 (VIT\_12s0059g01790) transcripts, whereas VviOMT2 (VIT\_12s0059g01750) remained unaffected.

Grape berry phenolics derived from the phenylpropanoid pathway participate to sensory properties, color and protection against environmental stress (Teixeira et al., 2013). Our biochemical analysis highlighted the dramatic effects of local warming on the onset of veraison and on the final anthocyanin contents in berries at harvest (**Table 1**; **Figure 2**). MapMan analysis (**Figure 6**) shows the contrasted heat responses of DEGs related to the general phenylpropanoid and to the flavonoid biosynthesis pathways (**Figure 6**; Supplementary Figure 3; Supplementary Table 10). The transcriptomic responses of GHT samples clearly differed from both VHT and RHT samples. During the herbaceous stage, genes involved in the general phenylpropanoid pathway (phenylalanine ammonialyase PAL, cinnamate 4-hydroxylase C4H, 4-coumarate-CoA ligase 4CL) were induced by HT regardless to the stress duration. By contrast, in VHT and RHT samples, the same genes were up-regulated after 1 day, but strongly repressed after 7 and 14 days of treatment. Many transcripts related

to the flavonoid biosynthesis pathway were repressed in GHT clusters. Two chalcone synthase transcripts (VviCHS1: VIT\_14s0068g00920, VviCHS2: VIT\_14s0068g00930) encoding the first committed enzyme in flavonoid biosynthesis (Parage et al., 2012) were repressed after 7 days. A significant repression by HT was also observed for many flavonoid transcripts of the late biosynthetic pathway, such as flavonoid 3 ′ -hydroxylase (VviF3′H, VIT\_11s0016g01020, VIT\_11s0016g01030, VIT\_09s0002g01090), flavonoid 3′ 5 ′ -hydroxylase (VviF3′ 5 ′H), dihydroflavonol 4-reductase (VviDFR, VIT\_16s0039g02350, VIT\_18s0001g1280) and leucoanthocyanidin dioxygenase (VviLDOX, VIT\_08s0105g00380), whereas the flavanone 3-hydroxylase gene (VviF3H, VIT\_16s0098g00860) was upregulated (**Figure 6**; Supplementary Figure 3; Supplementary Table 10). The HT led to contrasted effects on the same set of genes in VHT and RHT clusters when compared with GHT. VviF3H transcript was significantly less abundant whereas VviCHI (chalcone isomerase, VIT\_19s0014g00100) and VviF3′H genes were mostly up-regulated. Moreover, the repressive effect of HT on VviF3′ 5 ′H isoforms was not observed in VHT clusters and only transiently detected in RHT berries. In addition, the structural genes VviANR (anthocyanidin reductase, VIT\_00s0361g00040), and VviLAR (leucoanthocyanidin reductase, VIT\_01s0011g02960, VIT\_17s0000g04150) involved in proanthocyanidins (PA) synthesis were significantly repressed under HT, according to the stage and/or duration of the stress.

The anthocyanidin aglycones are further modified through glycosylation, methylation and acylation events, leading to the production of a wide variety of anthocyanin compounds. Glycosylation of flavonoids is catalyzed by enzymes from the large glycosyltranferase (GT) family, represented by 240 genes in the grape genome (Ono et al., 2010). Glycosylation enhances the structural diversity and modifies the functional properties of these secondary metabolites. For instance, the expression of UFGT [UDP-glucose:flavonoid 3-O-glycosyltransferase, renamed VviGT1, VIT\_16s0039g02230 (Parage et al., 2012)], catalyzing the 3-O-specific glycosylation of anthocyanidin is critical for the coloration of grape skin (Boss et al., 1996). Twenty-nine putative VviGT were deregulated in berries exposed to HT (**Figure 6**; Supplementary Figure 3; Supplementary Table 10). Despite the loss of anthocyanins upon HT, VviGT1 expression remained unaffected by HT. By contrast, VviGT5 (VIT\_11s0052g01600) and VviGT6 (VIT\_04s0023g01290) transcripts were strongly down-regulated in VHT and RHT clusters. VviGT5 and VviGT6 drive the glycosylation of flavonols, which increase their water solubility and their accumulation (Ono et al., 2010).

In grapevine, anthocyanins can also be modified through the action of O-methyltransferases (AOMTs) and acyltransferases (ACTs) before being transported into the vacuole (Fournier-Level et al., 2011; Rinaldo et al., 2015). These modifications modulate berry color by reducing anthocyanin reactivity and increasing their stability and solubility in water. None of the two grape AOMTs (VviAOMT1; VIT\_01s0010g03510, VviAOMT2; VIT\_01s0010g03490) that were previously described as effective anthocyanin 3′ - and 3′ ,5′ -O-methyltransferase (Hugueney et al., 2009; Lücker et al., 2010; Fournier-Level et al., 2011) was significantly deregulated by HT (Supplementary Table 10). However, other putative VviAOMTs transcripts (VIT\_11s0016g02610, VIT\_07s0031g00350, VIT\_03s0063g00140, VIT\_12s0028g03110), were transiently up- or down-regulated by HT depending on the ripening stage and stress duration. HT also modified the transcript amounts of various ACTs (Supplementary Table 10). Indeed, 5 putative ACTs were transiently repressed in GHT berries (VIT\_12s0134g00590, VIT\_12s0134g00630, VIT\_12s0134g00600, VIT\_12s0134g00650, VIT\_12s0134g0660), whereas 3 ACTs (VIT\_03s0017g00870, VIT\_12s0134g00590, VIT\_12s0134g00630) were up-regulated in both VHT and RHT berries. VIT\_03s0017g00870 corresponds to Vvi3AT, an enzyme recently associated to the production of the common grape berry acylated anthocyanins (Rinaldo et al., 2015).

Stilbenes and lignins represent branching points in the phenylpropanoid pathway. Stilbenes are a small family of phytoalexins synthesized by plants in response to biotic and abiotic stresses. Under normal growth conditions, berries stilbene content increases from veraison to ripening, with significant differences among V. vinifera varieties (Gatto et al., 2008). Fortyeight stilbene synthases (STSs) genes catalyzing the biosynthesis of the stilbene backbone were found in grapevine genome. This represents an unusual example of functional redundancy (Parage et al., 2012). Most of these (39 genes) were impacted in berries exposed to HT, displaying a quite similar and noticeable expression profile (**Figure 6**; Supplementary Figure 3; Supplementary Table 10). In GHT clusters, 18 STS transcripts accumulated after 1 and 7 d of treatment before being repressed at 14 d. In VHT and RHT fruits, the STSs were transiently induced at 1 d and then strongly repressed over the 2 weeks of experiment.

Lignification can also be induced as a response to various biotic and abiotic stresses, as shown in Citrus fruit after postharvest HT (Yun et al., 2013). Conversely, the cell wall lignin content was reduced in the skin of grape mature berries experiencing a water stress (Vannozzi et al., 2012; Fernandes et al., 2015). In the present study, numerous transcripts (39) potentially involved in the lignin biosynthetic pathway differentially accumulated in HT berries (**Figure 6**; Supplementary Figure 3; Supplementary Table 10). Two cinnamoyl-CoA reductase (VviCCR) gene isoforms were affected in an opposite way by HT, VIT\_14s0066g01150 being induced, and VIT\_09s0070g00240 being repressed. In GHT and VHT clusters, the transcript levels of 16 cinnamyl alcohol dehydrogenase (VviCAD) genes were mainly repressed whereas a few ones accumulated in RHT berries. The content of 3 ferulate 5-hydroxylase transcripts (VviF5H) was reduced in both VHT and RHT berries. Finally, HT led to contrasted effects on caffeic acid O-methyltransferase (VviCOMT) and peroxidase transcripts (catalyzing the polymerization of monolignols into lignins).

While many structural genes were transcriptionally affected in HT berries, the positive regulatory genes identified so far in grapevine were either weakly (VviMYBPA1, VviMYB14, VviMYB15, VviMYBF1) or not deregulated (VviMYBA1, VviMYBA2, VviMYBA3, VviMYC1, VviMYCA1, VviMYBPA2) in HT berries (Supplementary Table 11). Accordingly, HT did not affect the amount of VviGT1 transcript (VIT\_16s0039g02230) that encodes the anthocyanidin glycosyltransferase catalyzing the limiting step for anthocyanin accumulation and are a direct target of VviMybA1 transcription factor (TF; VIT\_02s0033g00410) (Cutanda-Perez et al., 2009). For comparison, no effect of HT was observed on VviMYBA1 transcripts in berries from fruiting cuttings (Carbonell-Bejerano et al., 2013), whereas a VviMYBA1 repression was reported in two others studies (Yamane et al., 2006; Rienth et al., 2014). It is also noteworthy that both VviMYBC2-L3 (VIT\_14s0006g01620) and VviMYB4b (VIT\_04s0023g0371) transcripts were transiently enhanced in VHT and RHT berries, respectively. These two TFs were recently described as negative regulators of the phenylpropanoid pathway in grape (Cavallini et al., 2015). While HT did not significantly impact the expression of genes known to act in the transport of flavonoids (reviewed by Zhao (2015), either directly (VviAM1, VviAM3, VviABCC1, VviMATE1, VviMATE2) or indirectly (VviGST1, VviGST4), the present work pointed out a significant deregulation of many putative ABC or MATE transporters and Glutathion S-transferase family members (Supplementary Tables 1, 11). It cannot be excluded that these proteins could be involved in the control of flavonoid transport in HT grape berries, as well.

In addition to synthesis, stabilization and vacuolar sequestration, the anthocyanin content can also be modulated through degradation that may involve enzymes such as laccases, polyphenol oxidases, class III peroxidases, and β-glucosidases (Oren-Shamir, 2009). Our experiments showed a strong HT effect on the expression of laccase genes, potentially impacting the polymerization rate of various phenolic compounds. Out of the 93 laccase-annotated genes in the grapevine genome, 33 transcripts were deregulated by HT (Supplementary Table 12). Interestingly, the laccase TT10 (Transparent Testa 10, VIT\_18s0075g00600) showed a maintained up-regulation by HT whatever the stage of development. In Arabidopsis, TT10 was proposed to participate in the oxidative polymerization of phenolic compounds (Pourcel et al., 2005). In a similar way, numerous genes encoding putative polyphenol oxidases (4), peroxidases (25) and β-glucosidases (16) were impacted by HT (Supplementary Table 12). In the present study, HT induced several peroxidase genes including class III type (VIT\_07s0130g00220, VIT\_18s0001g06850) whereas others were repressed (VIT\_18s0001g06840, VIT\_18s0001g06890). Recently, Movahed et al. (2016) showed that the expression of 3 of the 5 peroxidase genes (VIT\_14s0066g01850, VIT\_06s0004g07770, VIT\_07s0191g00050, VIT\_11s0016g05320 and VIT\_18s0072g00160) that were most strongly expressed in grape berry pericarp during ripening was influenced by temperature elevation. In agreement with these results, the present data also pinpoints the alteration of 4 of these genes (VIT\_14s0066g01850, VIT\_06s0004g07770, VIT\_11s0016g05320 and VIT\_18s0072g00160) in response to HT but with different kinetics and intensities.

# DISCUSSION

#### Locally Applied HT Delays the Onset of veraison and Alters Berry Composition at Maturity

The temperature differences used in this experiment (**Figure 1**) are within the range of values obtained in vineyards between sun-exposed and shaded berries. For instance, the temperature of exposed Merlot berry can exceed air temperature by more than 10◦C above ambient temperature, especially for clusters directly exposed to solar radiation (Pieri and Fermaud, 2005).

The present work highlights the strong inhibiting effects of HT (∼ 35◦C) locally imposed to clusters, on fruit-specific processes required for the onset of veraison. Conversely, damping diurnal berry temperature fluctuations (daytime cooling and night-time heating of clusters starting from fruit set) advanced the onset of ripening (Cohen et al., 2008; Tarara et al., 2008), as well as separated treatments (daytime cooling or night-time heating) even though the effects were less marked in the latter case (Cohen et al., 2012).

High temperatures promote the decrease of organic acid content observed after veraison, by exacerbating the malic acid breakdown (Ford, 2012; Rienth et al., 2016). Elevated temperatures accelerate the utilization of malate, enhancing the anaplerotic capacity of the TCA cycle for amino acid biosynthesis (Sweetman et al., 2014). By contrast to previous reports describing an increase in berry concentrations of various amino acids after exposure of whole grapevines to HT, the present data were obtained after berry-localized treatments. After a 14 d HT of whole grape fruiting cuttings, Carbonell-Bejerano et al. (2013) observed in fruits an increase in the content of 5 amino acid including TYR and PHE. Similarly, Sweetman et al. (2014) reported a boost in the berry concentration of 10 amino acids, among which THR, PRO and GABA, after applying at veraison an 11 d HT to potted Shiraz vines. The accumulation of amino acids is a widespread phenomenon of higher plants response to various abiotic stresses, including HT (Krasensky and Jonak, 2012; Bokszczanin and Fragkostefanakis, 2013). This increase may result from an up-regulation of amino acid synthesis, a decreased amino acid catabolism, and/or of an enhanced of stress-induced protein breakdown or to a change in amino acids imported from the plant. In grape berries exposed to HT, Sweetman et al. (2014) suggested that the amino acid accumulation was rather due to de-novo biosynthesis than to proteolysis. Among others, the increased levels of PRO and GABA may have a beneficial effect upon exposure to environmental cues (Krasensky and Jonak, 2012; Bokszczanin and Fragkostefanakis, 2013). PRO, which might act as a compatible osmolyte, a free radical scavenger and a protein chaperone, has been reported to have a protective role against abiotic stresses in many plant species. By contrast, tobacco and Arabidopsis plants do not accumulate PRO under HT and an excess of PRO reduces thermotolerance in Arabidopsis (Lv et al., 2011). PRO hyper-accumulated in Shiraz berries submitted to warming (Sweetman et al., 2014), but not in heat- treated Muscat Hambourg berries (Carbonell-Bejerano et al., 2013). PRO content was even strongly reduced in Cabernet Sauvignon fruits as shown in the present study (**Table 1**). These differences may be due to different experimental procedures (local vs whole plant treatments) and/or result from variety-dependent responses. Therefore the exact role of PRO in grape berry adaptation to HT remains to be determined.

The amount of the osmolyte GABA also increased in RHT berries (**Table 1**). This compound may act as a signaling molecule and affect different processes, including the control of the carbon–nitrogen balance and the protection against oxidative stress (Fait et al., 2007; Krasensky and Jonak, 2012). Recently, Sweetman et al. (2014) showed that a GABA shunt was up-regulated in warmed grape berries. This shunt is important for stress tolerance when the TCA cycle is down-regulated (Fait et al., 2007).

PHE and TYR whose concentration also increased in VHT and RHT berries (**Table 1**) are synthesized via the shikimate pathway followed by the branched aromatic amino acid metabolic pathway. Beside their role as building blocks of proteins, PHE and TYR serve as precursors for a wide range of secondary metabolites. Particularly, PHE is the substrate of PAL, the key enzyme of phenolic biosynthesis. The accumulation of PHE in HT berries may result from the strong repression of PAL genes observed during ripening, and to a larger extent, to the repression of genes encoding enzymes involved in the general phenylpropanoid pathway (**Figure 6**; Supplementary Table 7). Dai et al. (2014) reported that the up-regulation of the phenylpropanoid pathway genes and the increased accumulation of anthocyanins in response to high sugar supply is paralleled by a decrease in berry PHE content.

Heat exposure reduced the concentration of total anthocyanins and particularly the amount of dihydroxylated anthocyanins, and increased the proportion of acylated anthocyanins (**Table 1**). This result agrees with previous studies showing that application of elevated temperatures (>30◦C) to fruits results in anthocyanin degradation and inhibition of their accumulation. The intensity of this phenomenon depends on the type of anthocyanin derivative and on the grape variety (Kliewer and Torres, 1972; Spayd et al., 2002; Mori et al., 2007; Cohen et al., 2008; Tarara et al., 2008; Azuma et al., 2012).

Altogether, the experimental design set up with Cabernet Sauvignon fruiting cuttings to address the direct impact of elevated temperatures on berry development provided reliable biochemical results that are comparable to those observed in vineyards or using potted grapevines. Moreover, the alterations observed in heated berries directly result from the cluster exposure to HT. Finally, while veraison is delayed in GHT berries, their biochemical contents at harvest are comparable to that observed in control fruits.

## Locally Applied HT Deeply Affects Grape Berry Transcriptome and Triggers Adaptative Responses

Our results indicate that a local HT application deeply affected the berry transcriptome whatever the stage and the stress duration (**Figure 4**). The amplitude of the response is consistent with previous studies reporting that about 5% of the plant transcriptome is up-regulated 2-fold or more in response to HT (Mittler et al., 2012). The berry transcriptome varied with a similar amplitude when microvines were subjected to a short 2 h stress period during the day, whereas gene expression changes were more pronounced when this 2 h-HT was applied during the night (Rienth et al., 2014). The present work also revealed that only few genes were heatderegulated whatever the developmental stage and the stress duration. The comparison of all 9 conditions showed that only 36 genes were steadily induced under HT condition whereas continuously down-regulated transcripts were not found (Supplementary Table 3). Not surprisingly, 24 genes encoding various HSPs are listed among the permanently up-regulated genes. Other genes correspond to transcripts encoding putative transcription factors (TFs) from the MYB (VIT\_08s0007g06180) and the AP2/ERF (VIT\_15s0046g01390) families. These TFs contrast with many other TFs that are deregulated in at least one experimental condition (602 DEGS, Supplementary Table 13), and take part in the observed deep remodeling of the berry transcriptome under heat. Especially, the function of AP2/ERF members in plant abiotic stress responses is clearly established (Mizoi et al., 2012) and few recent works suggested a similar role for this TF family in grapevine (Carbonell-Bejerano et al., 2013; Zhu et al., 2013; Rocheta et al., 2014). Two others transcripts coding for a sterol O-acyltransferase (VIT\_00s2300g00010) and a truncated receptor-like kinase (VIT\_00s0437g00010) respectively, showed disrupted expression levels in all the nine conditions (Supplementary Table 3). The closest ortholog of VIT\_00s2300g00010 in Arabidopsis corresponds to AtASAT1 (Acyl-CoA Sterol Acyltransferase1), an enzyme catalyzing the phytosterol ester biosynthesis in seeds (Chen et al., 2007). The precise role of this enzyme in the context of heat-induced responses remains to be determined. Finally, VviGOLS1 (VIT\_07s0005g01970) encoding a galactinol synthase (GOLS) was consistently up-regulated by HT, confirming our previous work (Pillet et al., 2012).

#### Heat-Induced Adaptive Responses in Berries

Essential for cell growth and viability, HSPs function as molecular chaperones in maintaining protein quality and folding, and are required for the acquisition of the thermotolerance (Bokszczanin and Fragkostefanakis, 2013). A massive up-regulation of HSP genes is a highly conserved response in heat-exposed plants (Finka et al., 2011). This response is a result of the complex signaling cascade, whose final steps consist in the activation of heat shock TFs (HSFs) and their binding to the HSP promoters (Mittler et al., 2012). HSF activity is regulated at the transcriptional, post-transcriptional and post-translational levels (Mittler et al., 2012) and 21 Arabidopsis HSF transcripts are upregulated in response to various environmental stress conditions (Miller and Mittler, 2006). Data mining of the 12X version of the grape genome allowed the identification of 19 putative VviHSFs (Pillet et al., 2012; Scharf et al., 2012) among which 6 were up-regulated in HT berries (Supplementary Table 14). Among these, VIT\_04s0008g01110 corresponds to VviHSFA2 that was recently reported as involved in the regulation of heat responses in stressed berries (Pillet et al., 2012). HSFA2 is a key player for basal and acquired thermotolerance, extending the effect of heat acclimation in both tomato and Arabidopsis (Charng et al., 2007). AtHSFA2 which is the most highly HT-induced HSF (Busch et al., 2005), serves as a regulatory amplifier of a subset of genes (Schramm et al., 2008) and is required for the induction and maintenance of HT memory-related genes (Lämke et al., 2016). The transcriptional regulation of the heat-induced responses may also be due to VviMBF1c (Multiprotein bridging factor 1c; VIT\_11s0016g04080) that showed enhanced expression pattern in HT berries (Supplementary Table 14), in good agreement with two recent studies (Carbonell-Bejerano et al., 2013; Rienth et al., 2014). In Arabidopsis, AtMBF1c acts as a thermotolerance mediator through the heat-dependent regulation of 36 different transcripts (Suzuki et al., 2011). The existence of a similar MBF1c regulon was proposed in ripening berries of microvines submitted to HT (Rienth et al., 2014). In the present study, the relationship between VviMBF1c and putative members of this regulon was not so tight, probably due to a much longer heat exposure (14 days vs. few hours).

Several transcripts encoding putative USPs also accumulated in response to HT (Supplementary Table 5). The USP family plays an important role in stress resistance in bacteria (Kvint et al., 2003). Based on expression profiles, a similar role in abiotic stress tolerance was suggested for various plant USPs (Kerk et al., 2003; Maqbool et al., 2009; Isokpehi et al., 2011). Recently, the Arabidopsis AtUSP was shown to exhibit a redox-dependant chaperone function and to enhance plant tolerance to heat shock and oxidative stress (Young Jun et al., 2015). The role of USPs in grapevine remains to be elucidated.

#### Protein Homeostasis in Heated Berries

The maintenance of protein homeostasis, which includes the control of synthesis, intracellular sorting, folding, the function and degradation of proteins, is fundamental to ensure growth and development of plants under normal and stressful environmental conditions. In Arabidopsis, FKBP62 (ROF1) and FKBP65 (ROF2) are involved in acquired thermotolerance through the interaction with HSP90.1 and HSFA2 (Meiri and Breiman, 2009; Meiri et al., 2010). ROF1 contributes to the transcription activity of HSFA2 but ROF2, in the presence of ROF1, abolishes this activity. This suggests that ROF2 acts as a HT modulator through a negative feedback regulation of HSFA2 (Meiri et al., 2010). While expression of the grapevine VviROF1 ortholog (VIT\_00s0769g00010) was not affected in our experiments (Supplementary Table 6), VviROF2 transcripts (VIT\_00s0260g00070) accumulated upon HT at green and ripening stages (G7D, R1D and R7D). Half of the DEGs linked to the functional category "protein synthesis" encode ribosomal proteins known to prevent inhibition of protein synthesis under HT (Muñoz and Castellano, 2012). A massive heat-dependent deregulation of ribosomal protein genes was also described in wheat (Qin et al., 2008). Interestingly, one of the constantly and strongly up-regulated genes encodes a chloroplast ribosomal protein (RPS1; VIT\_13s0101g00050), a heat responsive protein involved in retrograde activation of heat responses in Arabidopsis (Yu et al., 2012). Particularly, RPS1 was proposed as a critical factor in the activation of HSFA2 and of its target genes. Many genes related to protein targeting (34 DEGs) were also affected in HT berries and mainly up-regulated during the green stage (Supplementary Table 6).

HT-induced protein degradation is a complex process involving a multitude of proteolytic pathways, with many DEGs encoding cysteine protease, serine protease, arginine protease, subtilase or metalloprotease activities. For instance, 24 transcripts encoding AAA/FtsH metalloproteases were deregulated in heated berries (Supplementary Table 6). This protease family is essential for the protein quality control of mitochondrial and chloroplastic membranes, and to prevent damages caused by stress conditions (Janska et al., 2013). For example, VviFtsH6 transcripts (VIT\_14s0108g00590) strongly accumulated in berries when the HT was applied during the herbaceous stage. In Arabidopsis, the FtsH6 protease is involved in the degradation of the light-harvesting complex II (LHC II) during high light acclimatization and dark-induced senescence (Zelisko et al., 2005). The involvement of FtsH proteases in the quality control of the photosystem II was also described under moderate HT (Yoshioka et al., 2006) and a similar role may be attributed to FtsH6 in green photosynthetic berries exposed to HT. Another metalloprotease encoding gene (VIT\_03s0088g00320) was highly up-regulated by HT at all stages. This transcript encodes an Arabidopsis EGY3 ortholog (At1g17870) belonging to the EGY family (ethylene-dependent gravitropism-deficient and yellow-greenlike 3). AtEGY3 is an AtHSFA2 target gene (Nishizawa et al., 2006). The involvement of this putative membrane and plastidial metalloprotease in heat responses was recently pinpointed (Laranjeira and coworkers, unpublished, PhD thesis; http:// hdl.handle.net/1822/18268). Both FtsH6 and EGY3 genes were also recently identified as heat responsive genes in Brassica napus siliques (Yu et al., 2014). The deregulation of numerous E3 ligase genes in HT berries may reflect an important role in triggering adaptive responses. Indeed, recent findings highlighted the role of various E3 ligases in mediating abiotic stress tolerance in Arabidopsis and in different crop species (Stone, 2014; Guerra et al., 2015). Most stress-related E3 ligases identified so far facilitate responses to environmental stimuli by modulating the abundance of key downstream stress-responsive TFs, thus affecting stress-related changes in gene expression. A non-proteolytic function of ubiquitin modification has also been reported in abiotic stress tolerance (Stone, 2014). For example, the rice E3 ligase OsHCI1 (Heat and Cold induced 1) monoubiquitinates some nuclear proteins in plants exposed to HT, which results in their translocation to the cytoplasm, and promotes HT tolerance (Lim et al., 2013). In apple, the E3 ligases MdCOP1s negatively regulate the peel anthocyanins content of fruits by modulating the degradation of the MdMYB1 protein (Li et al., 2012). A similar role can be envisaged in heat exposed berries to control the stability of flavonoid associated TFs. Many E2 conjugating enzyme genes are stressinducible (Stone, 2014; Guerra et al., 2015). Two E2 transcripts were up-regulated when HT was applied on ripening berries. The first one (VIT\_06s0004g08200) is an E2 enzyme ortholog of the Arabidopsis ubiquitin-conjugating enzyme 28 (UBC28, At1g64230). The second one (VIT\_18s0001g10100) encodes a SUMO Conjugating Enzyme (SCE1a) binding small ubiquitinlike modifiers (SUMOs) to a wide range of cellular proteins. SUMOylation of AtHSFA2 represses its transcriptional activity and SUMO overexpression decreases sHSPs accumulation after HT (Cohen-Peer et al., 2010).

#### Heat Treatment Potentially Decrease Berry Phenolic Quality and Aromatic Potential

Low temperatures are beneficial to aroma production in the cool climate white grape cultivars (Duchêne and Schneider, 2005). Conversely, the aromatic potential of berries exposed to HT may be reduced (Belancic et al., 1997; Falcão et al., 2007), possibly due to heat-induced transcriptional changes (Rienth et al., 2014). Cabernet Sauvignon berries contain various aromatic molecules such as terpenes and C13-norisoprenoids (Robinson et al., 2014) and are typified by specific volatile thiols and methoxypyrazines (Bouchilloux et al., 1998). Our transcriptomic data indicate that direct exposure of Cabernet Sauvignon berries to HT may decrease their aromatic potential through deregulation of numerous aroma and aroma precursor-related genes (Supplementary Tables 7, 9). Indeed, volatile terpenoids biosynthesis may be decreased by HT, as suggested by the heat repression of many key enzymes of the biosynthetic pathway (1-deoxy-D-xylulose-5-phosphate synthase, terpene synthase, geraniol 10-hydroxylase, (-)-germacrene D synthase and linalool synthase). Comparable results were obtained after exposure of the whole vine to HT (Rienth et al., 2014). Our data also revealed that most of 21 DEGs linked to carotenoid metabolism were down-regulated after HT, with the exception of VviBCH2. In Arabidopsis, overexpression of BCH2 improved tolerance to high light and HT by catalyzing the conversion of β-carotene to zeaxanthin and therefore preventing membranes from oxidative damage (Davison et al., 2002). While a constant decline in carotenoid abundance is generally observed after veraison in various cultivars including Cabernet Sauvignon (Deluc et al., 2009), our results suggest that HT may contribute to the decrease of carotenoid (except zeaxanthin) concentration before veraison, and/or accelerate its decrease after veraison. However, the possible consequences in term of aroma potential are not clear since only one type of CCD transcripts (VviCCD4a, VIT\_02s0087g00910) was decreased by HT. Another gene (VviCCD4b, VIT\_02s0087g00930) was transiently up-regulated by HT at veraison. The CCD enzymes convert their carotenoid substrates to C13-norisoprenoids, which encompass desirable flavor and aroma compounds in grapes and wine. In normal growth conditions, up-regulation of VviCCD4a and VviCCD4b was observed during ripening in Sauvignon Blanc and Pinotage (Young et al., 2012; Lashbrooke et al., 2013), whereas VviCCD4a expression pattern during berry development was dependent on the growth temperature regime applied to Cabernet Sauvignon fruit cuttings (Guillaumie et al., 2011). Carotenoid concentration in grape berries is influenced by microclimate through an effect of light exposure on the cluster (Kwasniewski et al., 2010; Young et al., 2015) and possibly through an impact of the temperature as suggested by our results. Finally, the present work highlights the HT repressive effects on 3 out of 4 VviOMTs (Supplementary Table 7), including VviOMT3, responsible for the synthesis of the predominant methoxypyrazine IBMP (Guillaumie et al., 2013). Natural (climate, soil) and viticultural factors impact on IBMP concentration in grapes and wines, and grapes ripened under HT produce wines with reduced IBMP contents (Darriet et al., 2012). Considering the strong correlation between VviOMT3 expression and berry MP content, our data implies that HT lead to a strong reduction in IBMP synthesis during the herbaceous stage, resulting from the repression of the key gene VviOMT3, thus drastically reducing IBMP content in ripe berries.

Grape berry phenolic compounds contribute to organoleptic properties, color and protection against environmental cues. Our data show that a local warming of developing berries strongly affects the phenylpropanoid metabolic pathway. The effects depend on the stress duration and on the developmental stage (**Figure 6**; Supplementary Figures 2B, 3). Interestingly, the repression of VviPAL genes that correlated well with the accumulation of its substrate PHE and the deregulation of genes involved in anthocyanin stabilization (VviAOMTs and VviACTs) may contribute to the significant decrease of anthocyanin contents in HT berries at harvest (**Table 1**). Therefore, the delayed onset of veraison observed for GHT clusters could be considered at the transcriptional level. Additionally, together with the repression of VviFLS1 (flavonol synthase, VIT\_18s0001g03470; also named FLS4, Fujita et al., 2006; Czemmel et al., 2009), the down-regulation of VviUFGT genes (VviGT5 and VviGT6) may contribute to the decrease of the flavonol content in heated berries. This result agrees with a recent work reporting the repression of VviGT5 in detached berries subjected to elevated temperatures (Loyola et al., 2016). Since previous work reported little or no effect on skin flavonol amounts (Spayd et al., 2002), the consequences of a locally applied HT on flavonol content still have to be determined. The transcriptional regulation of many flavonoid structural genes under HT suggests a control through the associated TFs. Since no or weak effect was observed on the expression of the positive TFs identified so far (Supplementary Table 11), several hypothesis can be proposed among which an up-regulation of negative regulators such as VviMYBC2-L3 and VviMYB4b (Supplementary Table 11; Cavallini et al., 2015), and a post-transcriptional control of the flavonoid associated genes as observed for the lightcontrolled stability of MYB1 protein in apple (Li et al., 2012).

A relationship between increased anthocyanin catabolism and elevated temperatures was proposed for grape berries (Mori et al., 2007). Although it is poorly known, anthocyanin degradation may involve enzymes such as laccases, polyphenol oxidases, class III peroxidases, and β-glucosidases (Oren-Shamir, 2009). In plants, the role of laccases remains largely unknown but their involvement in different steps of phenylpropanoid metabolism has been suggested. Fang et al. (2015) identified a vacuolar located ADE/LAC protein (anthocyanin degradation enzyme/laccase) responsible for the epicatechin-mediated anthocyanin degradation in litchi fruit during pericarp browning. Likewise, a gene encoding a putative rice laccase plays an important role in responses to abiotic stress (Cho et al., 2014). Our work strengthens the idea that the decrease in berry flavonoid content induced by HT involves an enzymaticmediated degradation of these molecules in the vacuolar compartment. Indeed, laccases, polyphenol oxidase, class III peroxidase, and β-glucosidase genes were deregulated in local HT conditions (Supplementary Table 12). Particularly, the role of vacuolar class III peroxidase in anthocyanin degradation has been highlighted in different plant species including grape (Calderon et al., 1992; Vaknin et al., 2005; Zipor and Oren-Shamir, 2013; Zipor et al., 2015). For instance, a transcriptional up-regulation of VviPrx31 (VIT\_14s0066g01850) was observed in berries exposed to HT (Movahed et al., 2016; and Supplementary Table 1). In Petunia, its overexpression reduced anthocyanin contents in petals exposed to heat.

Finally, genes involved in stilbenes and lignins biosynthesis were also specifically altered by HT during berry development. These alterations may modify berry texture and tolerance to the environment.

### Heat Deregulation of Green Berry Transcriptome Can Contribute to Delay veraison

During the herbaceous stage, heating significantly affected categories associated with stress and secondary metabolism (as discussed above), as well as categories associated with transport, cell wall, or hormones (**Figure 5**; Supplementary Table 15). The deregulation of these categories could explain, at least in part, the delayed veraison typified by the late accumulation of sugars and anthocyanins in GHT clusters (**Figure 2**).

#### Membrane Transporters

Many genes encoding proteins related to transport processes were enriched in GHT conditions (Supplementary Table 15). These genes were mainly repressed in GHT berries, in agreement with previous reports indicating that HT exposure led to a decrease in micronutrient transport. Genes involved in the calcium homeostasis were also downregulated by applying HT to green berries, and particularly CNGCs genes. CNGCs are nonspecific cation channels that are regulated by cyclic nucleotides such as cAMP or cyclic GMP (Ward et al., 2009) and that contribute to Ca2<sup>+</sup> signaling in the context of developmental processes, biotic and abiotic stress responses (Jha et al., 2016). Calcium may act through facilitating developmental and stress response signaling, stabilizing membranes, influencing water relations and modifying cell wall properties through crosslinking of de-esterified pectins (Hocking et al., 2016). Beside a sum of DEGs potentially linked to calcium homeostasis, the HT also impacted numerous genes related to signaling (protein kinase, transcription factors, Ca2+-binding proteins) (Supplementary Table 1). Potassium transporter gene expression was also altered after GHT exposure. As potassium is required for cell expansion, alteration in the expression of its transporters may affect berry growth. Additionally, as K<sup>+</sup> transporters and channels are known targets of ABA (that play an important role in berry ripening) (Davies et al., 2006), the alteration of their expression during the herbaceous stage may contribute to delay fruit ripening. MATE efflux transporters and ABC transporters were specifically affected by HT. For instance, while members of subfamily B were up-regulated at GHT, MATE, and ABCC and G were down-regulated. Interestingly, ABCG, the most affected subfamily, is involved in fruit maturation and exhibits a protective role (detoxification, vacuolar transport of ABA and glucosyl ester, anion transport) (Andolfo et al., 2015). It is the largest plant ABC transporter subfamily divided in two groups: WBCs and PDRs. The first group is involved in the extrusion of cuticular lipids and the second, in resistance to pathogens, antimicrobial terpenoids and auxinic herbicides, and in transport of signaling molecules or in secretion of volatile compounds (Kretzschmar et al., 2011; Andolfo et al., 2015). The repression of most of these transporters could potentially contribute to decrease anthocyanin accumulation.

The down-regulation of sugar transporter genes by HT fits well with earlier reports showing that HT results in the delay or arrest of the ripening process and of the accumulation of sugars (Greer and Weston, 2010; Greer and Weedon, 2013). However, in our experimental conditions, the postponing of the ripening process appears only when HT is directly applied on green clusters (GHT). At later stages, most of these sugar transporters were either down-regulated or not affected by HT (Supplementary Table 15). Altogether, these modifications in the amount of transporter transcripts may affect the berry physiology by modifying the pH and the nutrient content. As a consequence, amounts of certain primary and secondary metabolites were also affected in heated berries, resulting in the delayed onset of veraison.

#### Cell Wall Metabolism

Genes involved in cell wall composition were strongly altered in green berries during HT, highlighting the importance of cell wall adjustment (Supplementary Table 15). Heating effects on cell wall metabolism were also observed in VHT and RHT clusters. The fine-tuning of the cellulose-hemicellulose networks appears to be crucial for tolerance to heat (Tenhaken, 2014; Le Gall et al., 2015). The present transcriptomic data suggest differential cell wall synthesis and remodeling in heat-exposed berries, potentially affecting the overall fruit growth even if the exact consequences remain to be determined. In vineyard conditions, Dal Santo et al. (2013) found a correlation between the season climate and some differentially expressed genes encoding enzymes involved in cell wall structural modifications (especially CESs, PAEs, and XETs). XETs contribute to cell wall expansion and loosening, through their action on xyloglucan frame. After whole grape exposure to heat, an increased level of XETs transcripts was measured in warmed green berries. It was postulated that enhanced expression could be related to the adaptation of berry volume to temperature and the need for more flexible cell walls (Rienth et al., 2014). Our data reinforce these observations with 19 up-regulated XETs in GHT clusters. Cell wall loosening through increase level of expansin might help to maintain cellular functions during HT (Le Gall et al., 2015).

#### Hormone Homeostasis and Signaling

Fruit development and responses to environmental cues are controlled by plant hormones (Davies and Böttcher, 2009; Kuhn et al., 2014) and particularly by ABA. The present work revealed that berry heating significantly affects the expression of genes involved in ABA metabolism and signaling (Supplementary Table 9). The deregulation of ABA biosynthesis and signaling by applying HT during the herbaceous stage may contribute to the postponing of veraison and/or help to face HT. Interestingly, NCED expression and particularly VviNCED2 and VviNCED4 was impacted in GHT but also in VHT and RHT clusters. The down regulation of these 2 genes contrasts with data from Carbonell-Bejerano et al. (2013) reporting an HT-induction of these two genes after veraison and an increased ABA level in full ripe heated berries. The differences between both studies might be due to differences in the genotype and/or in the experimental conditions (temperature regime, microenvironment versus Lecourieux et al. Local Heating Impacts Berry Development

whole plant exposure, stress kinetic). Genes involved in ABA signaling were also affected. For instance, VviABI3 (VIT\_07s0005g05400) strongly accumulated in GHT berries (Supplementary Table 9). ABI3 is a B3-domain TF that is a part of the core ABA signaling network and the corresponding gene is up-regulated in Cabernet Sauvignon berries treated with ABA (Rattanakon et al., 2016). In normal growth conditions, VviABI3 transcripts accumulated during the lag phase prior veraison in Cabernet Sauvignon (Deluc et al., 2007). Thus, heating seems to accelerate the expression of VviABI3 during the herbaceous stage. In Arabidopsis, ABI3 is exclusively expressed in seeds, controlling gene expression programs that are essential to achieve seed maturation (Baud et al., 2016). Whereas the role of ABI3 in nonseed tissues remains unknown, a functional connection was demonstrated between ABI3 and HSFs in sunflowers embryos and during seed development in Arabidopsis (Rojas et al., 1999; Kotak et al., 2007b). Rattanakon et al. (2016) identified others TFs showing modified expression level after ABA treatment of grape organs, these TFs belonging to the AP2/ERF, NAC, bZIP/ABRE, and MYC/MYB families. In berries exposed to heat, more than 20 transcripts from this set of TFs were deregulated what can be potentially due to ABA action (Supplementary Table 15). The strongest deregulated TF of this list corresponds to the bZIP/ABRE Abscisic Acid Insensitive protein 5 (VviABI5; VIT\_06s0080g00340, VIT\_08s0007g03420) that significantly accumulated in GHT clusters whatever the treatment duration (Supplementary Table 15). ABI5 is depicted as a key regulator in the ABA signaling pathway, and a recent work highlighted its contribution in the ABA-dependent stimulation of plant thermotolerance (Lee et al., 2015). After interaction with the plant-specific RNA-binding protein FCA, ABI5 enhanced antioxidant activity under HT conditions. Particularly, ABI5 promotes the expression of genes encoding antioxidants, including 1-CYSTEINE PEROXIREDOXIN 1 (PER1). In agreement with these data, the heat-induced expression profiles of VviABI5 and VviPER1 (VIT\_05s0020g00600) were closely related in berries, suggesting a conserved mechanism in the fruit (Supplementary Table 1). Peroxiredoxins, which are thiol-based peroxidases, enhance plant tolerance to oxidative and heat stresses. More broadly, in this work, 70 DEGs were identified in heated berries as ROS (reactive oxygen species) scavenging/detoxifying enzymes and various antioxidants (Supplementary Table 1). This kind of adaptative response, described in different plants exposed to HT (de Pinto et al., 2015), may help the berries to manage potential damages due to oxidative burst. Indeed, ROS mainly attributable to NADPH oxidase activity accumulate under stress conditions including HT (Miller et al., 2009). Accordingly, two VviRBOH (respiratory burst oxidase homolog; VIT\_14s0060g02320, VIT\_01s0150g00440) genes were up-regulated in heat-exposed berries (Supplementary Table 1).

Besides ABA, our transcriptomic data (Supplementary Table 1) also suggest that HT affects the metabolism of auxin, ethylene and jasmonic acid. These hormones are involved in the control of grape berry development (Davies and Böttcher, 2009) and in the activation of key genes responsible for HT response (Bokszczanin and Fragkostefanakis, 2013; Sharma and Laxmi, 2015). For instance, indole-3-acetic acid (IAA) inhibits berry growth, sugar and anthocyanin accumulation (Kuhn et al., 2014) and prevents ripening. Indeed, the decrease in IAA content and the increase of its conjugated form are needed to induce ripening (Davies et al., 1997; Böttcher et al., 2010). Interestingly, our work showed that application of a HT to green berries resulted in the reduction of IAA-amido synthetase (VviGH3, VIT\_07s0005g00090) transcript abundance and in an increase in the expression of two transcripts encoding IAA-amino acid hydrolases (VIT\_11s0016g02700, VIT\_08s0007g02740), thus suggesting that heating is disrupting IAA conjugating process in GHT clusters and therefore postponing the onset of ripening. Further work is needed to address the respective role of each hormone and their interactions in the context of grape berry grown under HT.

# CONCLUSIONS

This work provides the first molecular data describing the effect of HT at the microclimate level and brings important information on the consequences of temperature elevation in the context of leaf removal practice. HT effects depend both on the developmental stage and on the stress duration. Heating delayed the onset of veraison and strongly altered the berry biochemical composition at harvest. These physiological modifications could be partly explained by the deep remodeling of heated berry transcriptome. The intrinsic capacity of grape berries to perceive heat stress and to build adaptive responses is suggested by the deregulation of categories such as "stress responses," "protein metabolism" and "secondary metabolism." Additionally, important changes in processes related to "transport," "hormone" and "cell wall" might contribute to the postponing of veraison. Furthermore, opposite effects depending on heat duration were observed for genes encoding enzymes of the general phenylpropanoid pathway, suggesting that the HT-induced decrease in anthocyanin content may result from a combination of transcript abundance and product degradation. However, one cannot exclude that this process could also be regulated at the protein level as HT strongly affects protein homeostasis related genes. Finally, the data reported here provide a rich transcriptomic resource for functional characterization of the genes that potentially control HT response and/or adaptation in grapevine. The functional characterization of some putative candidates is in progress.

# AUTHOR CONTRIBUTIONS

DL and PP designed the research; DL oversaw the research; JP, PP, and DL performed the greenhouse experiments; FL, JC, and DL carried out the RNA extraction for transcriptomic analysis and performed RT-QPCR; CK performed the bioinformatics analysis; JC, JP, GH, and CR did the metabolic analysis; FL, CK, and DL analyzed and interpreted the data; FL, CK, SD, and DL drafted the manuscript; PP and EG critically revised the manuscript. All authors read and approved the final manuscript.

#### FUNDING

This research received funding from the Agence Nationale de la Recherche for the project "DURAVITIS" (grant no. ANR-2010- GENM-004-01).

#### ACKNOWLEDGMENTS

The authors would like to thank Pr. Laurent Torregrosa, Dr. Charles Romieu, Dr. Markus Rienth, Dr. ZhanWu Dai and Pr. Grant Cramer for valuable discussions and scientific

#### REFERENCES


interactions. The authors also thank Dr. Pablo Carbonell Bejerano for critical reading of the manuscript. For the production of the fruiting cuttings and technical assistance during greenhouse experiments, we express our gratitude to Jean Pierre Petit and Guillaume Pacreau.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2017. 00053/full#supplementary-material


anthocyanin biosynthesis and red fruit coloration in apple. Plant Physiol. 160, 1011–1022. doi: 10.1104/pp.112.199703


in response to several types of environmental stress. Plant J. 48, 535–547. doi: 10.1111/j.1365-313X.2006.02889.x


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Lecourieux, Kappel, Pieri, Charon, Pillet, Hilbert, Renaud, Gomès, Delrot and Lecourieux. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Whole Plant Temperature Manipulation Affects Flavonoid Metabolism and the Transcriptome of Grapevine Berries

Chiara Pastore<sup>1</sup>† , Silvia Dal Santo<sup>2</sup>† , Sara Zenoni<sup>2</sup> , Nushin Movahed<sup>1</sup> , Gianluca Allegro<sup>1</sup> , Gabriele Valentini<sup>1</sup> , Ilaria Filippetti<sup>1</sup> \* and Giovanni Battista Tornielli<sup>2</sup>

<sup>1</sup> Department of Agricultural Sciences, University of Bologna, Bologna, Italy, <sup>2</sup> Department of Biotechnology, University of Verona, Verona, Italy

#### Edited by:

Ashraf El-kereamy, University of California, United States

#### Reviewed by:

Ramesh Katam, Florida A&M University, United States David Lecourieux, University of Bordeaux 1, France Maria Carmen Gomez-Jimenez, University of Extremadura, Spain

> \*Correspondence: Ilaria Filippetti ilaria.filippetti@unibo.it

†These authors have contributed equally to this work.

#### Specialty section:

This article was submitted to Technical Advances in Plant Science, a section of the journal Frontiers in Plant Science

> Received: 24 February 2017 Accepted: 17 May 2017 Published: 06 June 2017

#### Citation:

Pastore C, Dal Santo S, Zenoni S, Movahed N, Allegro G, Valentini G, Filippetti I and Tornielli GB (2017) Whole Plant Temperature Manipulation Affects Flavonoid Metabolism and the Transcriptome of Grapevine Berries. Front. Plant Sci. 8:929. doi: 10.3389/fpls.2017.00929 Among environmental factors, temperature is the one that poses serious threats to viticulture in the present and future scenarios of global climate change. In this work, we evaluated the effects on berry ripening of two thermal regimes, imposed from veraison to harvest. Potted vines were grown in two air-conditioned greenhouses with High Temperature (HT) and Low Temperature (LT) regimes characterized by 26 and 21◦C as average and 42 and 35◦C as maximum air daily temperature, respectively. We conducted analyses of the main berry compositional parameters, berry skin flavonoids and berry skin transcriptome on HT and LT berries sampled during ripening. The two thermal conditions strongly differentiated the berries. HT regime increased sugar accumulation at the beginning of ripening, but not at harvest, when HT treatment contributed to a slight total acidity reduction and pH increase. Conversely, growing temperatures greatly impacted on anthocyanin and flavonol concentrations, which resulted as strongly reduced, while no effects were found on skin tannins accumulation. Berry transcriptome was analyzed with several approaches in order to identify genes with different expression profile in berries ripened under HT or LT conditions. The analysis of whole transcriptome showed that the main differences emerging from this approach appeared to be more due to a shift in the ripening process, rather than to a strong rearrangement at transcriptional level, revealing that the LT temperature regime could delay berry ripening, at least in the early stages. Moreover, the results of the in-depth screening of genes differentially expressed in HT and LT did not highlight differences in the expression of transcripts involved in the biosynthesis of flavonoids (with the exception of PAL and STS) despite the enzymatic activities of PALs and UFGT being significantly higher in LT than HT. This suggests only a partial correlation between molecular and biochemical data in our conditions and the putative existence of post-transcriptional and post-translational mechanisms playing significant roles in the regulation of flavonoid metabolic pathways and in particular of anthocyanins.

Keywords: grape, ripening, flavonoid, transcriptome, temperature

# INTRODUCTION

fpls-08-00929 June 1, 2017 Time: 19:43 # 2

Agriculture, and in particular viticulture, is highly dependent upon climatic conditions during the growing season. The predicted climate change therefore presents a major challenge for wine production. Although Vitis vinifera shows large variations in terms of tolerance to abiotic summer stresses, i.e., high temperature (HT) and radiation and low water availability (Palliotti and Poni, 2016), the effects of a temperature increase on berry composition have been widely studied, in terms of both extreme heatwaves or mild-to-moderate increase during ripening. Since berry temperature and solar radiation often act synergistically and sun exposure of grape bunches can be modified by viticulture practices, several researches have focused on the effect of both parameters simultaneously (Bergqvist et al., 2001; Spayd et al., 2002; Downey et al., 2004; Cortell and Kennedy, 2006; Tarara et al., 2008; Azuma et al., 2012; Movahed et al., 2016) and only a few computed the precise role of temperature in these multi-factor studies (Mori et al., 2005, 2007; Yamane et al., 2006; Cohen et al., 2012; Sadras et al., 2013a,b; Rienth et al., 2016).

High sugar concentration at harvest is often associated with thermal increase, as indicated by the trend observed in the last decades (Petrie and Sadras, 2008; Mira de Orduña, 2010; Sadras and Petrie, 2011). However, several experiments showed that sugar accumulation is not or only slightly affected (Spayd et al., 2002; Mori et al., 2005; Mori et al., 2007; Sadras and Denison, 2009; Movahed et al., 2016), or sometimes even reduced (Greer and Weston, 2010; Greer et al., 2010; Carbonell-Bejerano et al., 2014; Rienth et al., 2016) by air temperature increase. The different results can probably be ascribed to variation in diurnal temperature levels, since temperatures over 30◦C may lead to the stopping of soluble solids transport from leaves to berry, but may sometimes indirectly cause a higher concentration by evaporative loss (Keller, 2010).

Temperature has been known for some time to have significant effects on berry acidity, accelerating the breakdown of malic acid (Rienth et al., 2016) and decreasing the titratable acidity the greater the heat summation (Tarara et al., 2008). Intriguingly, other studies suggested a cultivar-dependent thermal response of acidity and pH (Bergqvist et al., 2001; Sadras et al., 2013a,b; Movahed et al., 2016).

Of particular interest are the effects of temperature on the phenylpropanoid biosynthetic pathway, involved in the biosynthesis of flavonoids (anthocyanins, flavonols, and tannins) that play a crucial role in grape and wine composition with regards to color, bitterness and stability, and also in the biosynthesis of non-flavonoid compounds (i.e., stilbenes). A negative correlation between elevated temperature during the day (over 30◦C) and anthocyanin concentrations has recently been explored (Mori et al., 2007; Carbonell-Bejerano et al., 2013; Movahed et al., 2016). Some authors pointed out the effect of increasing temperatures on the reduction of the enzymatic activity of some key enzymes involved in flavonoid biosynthesis as phenylalanine ammonia-lyase (PAL), which presides the first step of general phenylpropanoid biosynthesis, and UDPglucose:flavonoid 3-O-glucosyltransferase (UFGT), which is involved in the last and specific step of anthocyanin biosynthesis (Mori et al., 2007; Movahed et al., 2016).

Flavonols are known to behave as UV-protectants and to play a role in co-pigmentation with anthocyanins. Flavonols in the berry can be affected by sunlight exposure, which usually promotes strong enhancement in concentrations and in the expression of flavonol biosynthesis-related genes (Spayd et al., 2002; Downey et al., 2004; Czemmel et al., 2009). On the contrary, temperature seems to have less effect than light in flavonol synthesis control and under thermal increase flavonols can be unaffected (Tarara et al., 2008) or slightly reduced (Azuma et al., 2012). Temperature appears to have little impact on tannins (Cohen et al., 2012) whose accumulation in skins and seeds occurs predominantly before veraison (Downey et al., 2004).

Grapevine transcriptomic analysis has provided a wealth of data concerning the mechanisms responsible for the temperature effects on berry composition, especially on sugars, acidity and anthocyanin concentrations (Carbonell-Bejerano et al., 2013; Pastore et al., 2013; Rienth et al., 2014, 2016; Lecourieux et al., 2017). Several authors reported that the loss of anthocyanin synthesis following HT is due to the reduced expression of anthocyanin biosynthetic genes (Yamane et al., 2006; Azuma et al., 2012; Lecourieux et al., 2017). Sometimes, however, despite a sharp reduction in terms of anthocyanin concentration, a concomitant reduction in anthocyanin biosynthetic genes expression was not found, as reported for Muscat Hamburg berries on fruiting cuttings (Carbonell-Bejerano et al., 2013) and in Cabernet Sauvignon and Sangiovese vines grown under increasing temperature (Mori et al., 2007; Movahed et al., 2016). In these cases, also an involvement of anthocyanin degradation, implying the action of peroxidases should be hypothesized (Mori et al., 2007; Movahed et al., 2016) as it was previously seen in other plant species, as Brunfelsia flower petals (Vaknin et al., 2005), litchi (Zhang et al., 2005), and strawberry fruits (Chisari et al., 2007).

Despite recent progress, the direct and indirect effects of temperature on the grape ripening process, and specifically on flavonoid composition, are far from being completely unraveled. In particular, there is growing interest in charting the impact of temperature in specific viticultural areas and different seasons on flavonoid composition. Here, we analyzed the grapevine cultivar Sangiovese, the most cultivated Italian variety, comparing the effects of two thermal regimes on the berry skin biochemical composition, flavonoid-related enzymatic activity, and whole transcriptome during ripening.

# MATERIALS AND METHODS

# Grapevine Plant Material and Growing Conditions

Experiments on grapevine berries were conducted in 2012 on 6-year-old uniformly potted plants (V. vinifera cv. Sangiovese). The vines, grafted on SO4 rootstocks, were grown in 30-liter pots containing a 1:1 mixture of sand and soil (27% sand, 46% silt and 27% clay, clay loam soil). The number of shoots was standardized to nine per vine. In addition, to achieve a uniform leaf area on all

the vines, the tip of each shoot was removed and 15 main leaves were maintained before the experiment started. At the beginning of bunch closure [BBCH 77, (Lorenz et al., 1995)], 10 vines were selected and bunch numbers were adjusted to 11–12 per vine.

The vines were assigned to two treatments: low temperature (LT) and HT. Five LT vines were placed from 1 week before veraison to harvest in a plastic greenhouse (20 m<sup>3</sup> ) where the air temperature was controlled by a cooler and a fan was used to homogenize environmental conditions in the greenhouse (Supplementary Figure 1). During the night, the tunnel was opened.

Five HT vines were placed in an identical plastic greenhouse, without fan, whose basal segment was open but all the canopies were covered to maintain similar illumination to the LT vines.

The average, maximum and minimum air temperatures were recorded using air temperature sensors (TL20, 3M, Milan, Italy) in both greenhouses during the ripening period (**Table 1**).

Both greenhouses were made of polyethylene film (MOP, Bologna, Italy) that did not alter the spectral composition of light. The incident light during the day, which outside ranged from 500 to 2000 µmol·m<sup>2</sup> ·s −1 , was reduced by the polyethylene film up to 12% within the visible range. The humidity recorded during the experiment was comparable between LT and HT greenhouses.

All vines were automatically watered daily and were well supplied with nutrients.

#### Berry Temperature Monitoring

Berry temperature was monitored in 10 bunches from each treatment using 10 T-type thermocouples (RS component, Milan, Italy) positioned in the sub-cuticular tissues of the berry skin. Each probe was then connected to a CR10X data logger (Campbell Scientific Ltd, Leicestershire, United Kingdom), registering temperature data every 20 min during the development period.

#### Berries Sampling

Berries were sampled before the treatment (T0, 1 week before veraison), at veraison (T1) and 10 (T2), 20 (T3), 32 (T4), and 45 (T5) days after veraison, corresponding to harvest. The berries were collected at the same time of day (9–10 am). At T0 five berries from each of the ten vines were sampled and pooled and this procedure was repeated 4 times to create four independent biological replicates of 50 berries each. Upon thermal treatment imposition, nine berries were randomly selected from each of the five treated vines and pooled. The same sampling procedure was repeated four times to create four independent pools of 45


Average, maximum, and minimum air temperature recorded during the experiment under the two different growing regimes. LT = low temperature, HT = high temperature.

berries per each sampling date/treatment combination. In total, the experiment entailed the collection and the analysis of 44 berry samples [4 control samples + (2 thermal regimes × 5 stages × 4 biological replicates)].

From each biological replicate about 20 berries were weighed and directly tested for the evaluation of soluble solids (◦Brix), titratable acidity and pH. The remaining berries were peeled and the skins were immediately frozen in liquid nitrogen and stored at −80◦C for subsequent metabolic analyses, enzyme activity and expression analysis.

## Soluble Solids, Titratable Acidity, and pH Measurements

The sampled berries were crushed and the must was sieved and used for soluble solids analysis with a temperature-compensating CR50 refractometer (Maselli Misure Spa, Parma, Italy). We then diluted 5 ml of the same must seven times with bi-distilled water for titration using a Crison Compact Titrator (Crison, Barcelona, Spain) with 1 N, 0.5 N or 0.25 N NaOH (Sigma-Aldrich, St. Louis, MO, United States), according to the stage of berry ripening to obtain pH and titratable acidity data (expressed as g L−<sup>1</sup> of tartaric acid equivalents).

#### Analysis of Grape Berry Anthocyanins and Flavonols

Total anthocyanins and flavonols were analyzed in all 44 samples by soaking 2–3 grams of peeled skins, depending on the berry phenological stage, per each sampling date/treatment combination in 50 mL methanol for 24 h (Mattivi et al., 2006), then storing the extracts at −20◦C.

To analyze the total concentrations of each flavonol aglycone, an aliquote of 5 ml of methanolic extract was completely dried under vacuum. To achieved the acid hydrolization of flavonol glucosides, the pellet was resuspended in 2.5 ml of methanol and 2.5 ml of 2M trifluoroacetic acid (Sigma–Aldric, Saint Louis, MO, United States) in milliQ water. The reaction was conducted at 100◦C in a boiling hot water bath, with a condenser, for 2 h. The reactions product was then completely dried under vacuum and the pellet obtained resuspended in 1 ml of methanol until HPLC analyses (Mattivi et al., 2006).

HPLC separation and quantifications of anthocyanins and flavonols (Mattivi et al., 2006) were performed on a Waters 1525 HPLC (Waters, Milford, MA, United States) equipped with a diode array detector (DAD) and a Phenomenex (Castel Maggiore, Bologna, Italy) reversed-phase column (RP18, 250 mm × 4 mm, 5 µM). Anthocyanins were quantified at 520 nm using an external calibration curve with malvidin-3-glucoside chloride as the standard (Sigma-Aldrich). Flavonols were quantified at 370 nm with the corresponding external standards (myricetin, quercetin, and kaempferol) purchased from Extrasynthese (Genay, France).

#### Analysis of Skin Berry Tannins

Skin tannins extraction was performed following the procedure proposed by Downey et al. (2003): About 100 mg of skins per each sampling date/treatment combination were ground to a fine powder separately, extracted with a solution containing 70%

acetone for 24 h in dark room and measured by HPLC using the same equipment used for anthocyanins analysis. After free monomers were removed, the tannin content was determined by acid-catalyzed cleavage in the presence of excess phloroglucinol as described by Kennedy and Jones (2001). Individual reversedphase HPLC separations were used to determine the abundance of free monomers and cleaved proanthocyanidins by measuring absorbance at 280 nm (Downey et al., 2003). The concentrations of free monomers and hydrolyzed terminal subunits were determined from standard curves prepared with commercial standards of catechin, epicatechin, epicatechin-gallate and epigallocatechin (Extrasynthese, France).

#### Enzymatic Activity Assays

Berry skins (0.2 gr) were ground with a mortar and pestle in liquid nitrogen to a fine powder. For PAL and UFGT activity assays, the protein extraction was performed according to the methods of Mori et al. (2005, 2007), respectively. Peroxidase activity was instead measured on berry skin after protein extractions as described by Ushimaru et al. (1997).

For PAL activity measurement, the reaction mixture consisted of 0.5 ml of phenylalanine and 0.5 ml of protein extract. The assay mixture was incubated at 37◦C for 60 min. The reaction was terminated by adding 0.5 ml of HCl acid (18%). The quantity of the product, trans-cinnamic acid, was calculated using its extinction coefficient of 9630 M−<sup>1</sup> cm−<sup>1</sup> at 290 nm. One unit (U) of PAL activity, expressed on berry skin fresh weight, was defined as the production of 1 mol of trans-cinnamic acid per minute.

The UFGT assay was performed on the protein extract using either cyanidin or delphinidin as substrate in 200 mM Tris-HCl (pH 7.5), containing 0.1 mM cyanidin or delphinidin and 10 mM UDP-glucose. After incubation for 5 min at 37◦C, the reaction was stopped by adding 150 µl 5% HCl. The concentration of cyanidin-3-glucoside and delphinidin-3-glucoside was calculated at 520 nm and pH 1, using extinction coefficients of 26,900 M−<sup>1</sup> cm−<sup>1</sup> and 26,000 M−<sup>1</sup> cm−<sup>1</sup> , respectively. One unit (U) of UFGT activity, expressed on berry skin fresh weight, was defined as the production of 1 mol of cyanidin-3-glucoside or delphinidin-3-glucoside per second.

Guaiacol peroxidase activity was determined in the ripening berry skin as described by Ushimaru et al. (1997), using pyrogallol as the electron donor. The reaction mixture comprised the protein extract in 50 mM sodium phosphate buffer (pH 7.0), 0.1 mM H2O<sup>2</sup> and 50 mM pyrogallol (H2O<sup>2</sup> and pyrogallol were freshly prepared just before use). The absorbance at 430 nm was recorded immediately after the addition of pyrogallol and after 7 min at room temperature, and was compared to a blank with no protein extract added. One unit (U) of peroxidase activity, expressed on berry skin fresh weight, was defined as the amount of enzyme that catalyzes the oxidation of 1 µmol of pyrogallol per minute.

#### Statistical Analyses

The experiment had a completely randomized design and the agronomic parameters and biochemical data were submitted to analysis of variance (ANOVA) using SAS statistical software (SAS Institute, Cary, NC, United States) with four replications for each treatment. The temperature data were analyzed with 10 replications for each treatment.

# RNA Extraction and Microarray Analyses

Total RNA was isolated from approximately 400 mg of pulverized berry skins from three biological replicates sampled at all dates except harvest, for a total of 27 berry samples [3 control samples + (2 thermal regimes × 4 stages × 3 biological replicates)], using the Spectrum Plant Total RNA kit (Sigma–Aldrich), with modifications as described in Dal Santo et al. (2016). RNA quality and quantity were determined using a Nanodrop 2000 spectrophotometer (Thermo Scientific, Wilmington, DE, United States) and a Bioanalyzer Chip RNA 7500 series II (Agilent, Santa Clara, CA, United States). We hybridized 5 µg of total RNA per sample to a NimbleGen microarray 090818\_Vitus\_exp\_HX12 chip (Roche, NimbleGen Inc., Madison, WI, United States), according to the manufacturer's instructions (Dal Santo et al., 2013). Statistical analysis of the microarray data was conducted using TMeV v4.8<sup>1</sup> . Statistical analysis of microarrays (SAM) was performed with a false discovery rate (FDR) of 0.1% and ANOVA using α = 0.05 and standard Bonferroni correction. Heat maps were created using log2-transformed expression values and then median-centered by transcript. Cluster analysis was performed by the k-means method (KMC) with Pearson's correlation distance. Principal component analysis (PCA) was conducted using SIMCA P+ v13 (Umetrics, United States). Gene Ontology (GO) annotation was applied using the BiNGO v2.3 plug-in tool in Cytoscape v2.6 with PlantGOslim categories, as described by Maere et al. (2005). Overrepresented PlantGOslim categories were identified using a hypergeometric test with a significance threshold of 0.05. STEM v1.3.8 was used for clustering, comparing and visualizing gene expression data (Ernst et al., 2005).

### Reverse Transcription (RT) and Real Time qPCR

One microgram of extracted RNA was treated with 2 units (U) of Turbo DNase (TURBO DNA-free kit—Ambion) according to the instructions provided with the commercial kit. DNase- treated RNA was then used for cDNA synthesis using the SuperScriptIII Reverse Transcriptase kit (Invitrogen) following the producer's indications. In order to assess if the cDNA had been properly produced, an amplification with primers designed on the 30UTR of an actin coding gene (VIT\_12s0178g0020, (Pastore et al., 2011) was performed. Real Time qPCR was performed using GoTaq <sup>R</sup> GreenMaster Mix kit (Promega) to amplify a specific region of target genes (UFGT, VvUFGT – VIT\_16s0039g02230; PAL – VIT\_00s2849g00010 and the peroxidase VvPrx31 – VIT\_14s0066g01850) with previously described primer pairs (Movahed et al., 2016). Primers and cDNA were mixed with the Power SYBR <sup>R</sup> Green PCR Master Mix (Applied Biosystems, Foster City, CA, United States) and the reaction was carried out on an ABI PRISM StepOne Sequence Detection System (Applied Biosystems, Foster City, CA, United States) using the following cycling conditions: 95◦C hold for 10 min followed by 45 cycles

<sup>1</sup>mev.tm4.org/

at 95◦C for 30 s, 55◦C for 30 s and 72◦C for 20 s. 95◦C hold for 2 min followed by 40 cycles at 95◦C for 15 s, 55◦C for 30 s, 60◦C for 30 s, and 95◦C for 15 s. Non-specific PCR products were identified by the dissociation curves. Amplification efficiency was calculated from raw data using LingReg PCR software (Ramakers et al., 2003). The mean normalized expression (MNE)-value was calculated for each sample referred to the ubiquitin expression according to the Simon equation (Simon, 2003). Standard error (SE) values were calculated according to Pfaffl et al. (2002).

#### Accession Numbers

Grape berry microarray expression data are available in the Gene Expression Omnibus under the series entry GSE92864<sup>2</sup> .

#### RESULTS

### Two Different Thermal Regimes Differently Affect Ripening Parameters

We set up an experimental design to impose two different thermal regimes in potted grapevine plants over the course of grape ripening. Five vines were placed in a plastic greenhouse where the air temperature was artificially cooled whereas other five vines were placed in an identical plastic greenhouse with only the basal segment open. These two conditions were named LT and HT, respectively, with HT representing the closest condition to the 2012 thermal regime. HT and LT berry temperature were strongly differentiated in the two greenhouses (**Table 2** and **Figure 1**). Indeed, the average berry temperature during the treatment period was ∼21.8◦C in the LT greenhouse and 26.5◦C in the HT greenhouse (**Table 2** and **Figure 1A**). Maximum temperatures were higher in HT during ripening, with several heatwaves, sometimes reaching values of around 40◦C in HT berries, corresponding to 7–8◦C higher than the maximum temperatures detected in LT berries (**Figure 1B**). Overall, HT berries accumulated an additional 238 Degree Days (DD) compared to LT berries (**Table 2**). The number of hours with berry temperature exceeding 30◦C was four-fold greater in the HT greenhouse, while berry temperature exceeding 35◦C was registered only in the HT greenhouse (**Table 2** and **Figure 1B**).

<sup>2</sup>http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE92864

TABLE 2 | Average Berry Temperature, accumulated Degree Days (DDs), and heatwaves hitting the berries (calculated as the number of hours with average berry temperature > 30◦C and > 35◦C) under the two different growing regimes.


Values represent means of ten replicates. LT = low temperature, HT = high temperature. Asterisks indicate significant differences between HT and LT using ANOVA (∗P < 0.05).

On the contrary, the minimum temperatures in HT and LT were aligned (**Figure 1B**).

The LT thermal regime slightly delayed technological ripening in berries, which showed a lower level of ◦Brix and pH and higher values of titratable acidity compared to HT. However, at harvest, the soluble solids reached comparable values in HT and LT berries (**Figure 2**).

Until 10 days after veraison, there was no detectable difference in total anthocyanin concentration between the two thermal regimes (**Figure 3A**). Starting from 10 days after veraison, total anthocyanin accumulation rate began to accelerate in the skin of LT berries compared to HT. This different behavior was maintained at harvest, when the concentration of total anthocyanins in the skin of LT grapes was almost doubled compared to HT (**Figure 3A**). The most abundant individual anthocyanin in Sangiovese berry skin was malvidin-3-glucoside, in both LT and HT treatments (**Figure 3A**). However, its percentage was significantly lower in HT than LT (**Figure 3A**). This decrease of malvidin-3-glucoside was counterbalanced in HT by an increase in the percentage of petunidin-3-glucoside, delfinidin-3-glucoside, and especially of cyanidin-3-glucoside, albeit non-significant.

The accumulation trend of flavonols was similar to that of anthocyanins in both HT and LT. At harvest, flavonol concentration in LT berry skin was thus three-times of that found in HT berry skin (**Figure 3B**). The analyses of the single flavonol compounds at harvest showed a higher quercetin and lower myricetin percentage in LT berries compared to HT (**Figure 3B**). Kaempferol was instead unaffected by different growing temperature regimes (**Figure 3B**).

Berry skin tannins decreased during ripening without significant differences between HT and LT in any developmental stage analyzed. Likewise, regardless of thermal regime, the percentage, measured at harvest, of the four individual flavan-3-ol monomers catechin, epicatechin, epicatechin-gallate and epigallocatechin did not differ between HT and LT berries, epicatechin being the tannin compound present in the greatest concentration in both treatments (**Figure 3C**).

Our results suggest a different effect of LT and HT thermal regimes, to a less extent, on technological berry ripening and, more strongly, on flavonoid increase, with the enhancement of anthocyanin and flavonol accumulations under LT conditions.

#### The Effect of Temperature on Berry Skins Whole Transcriptome

The transcriptome of Sangiovese berry skins under LT and HT growing temperature regimes was assessed at five sampling times (i.e., before the treatments (T0), at veraison (T1) and 10 (T2), 20 (T3) and 32 (T4) days after veraison). The dataset was initially screened by Significance Analysis of Microarrays (SAM, 9 groups, FDR = 0.1%) to select genes that were differentially modulated under our experimental conditions. Analysis of Variance (ANOVA, 9 groups, α = 0.05, standard Bonferroni correction) was applied to transcripts positive in the SAM in order to retrieve the most significantly modulated transcripts (6441 genes, Supplementary File 1).

To verify the uniformity of biological replicates and investigate the transcriptomes of LT and HT berries, we performed a PCA, obtaining a significant model (8 PCs, R2X = 0.941, Q2 (cum) = 0.848, **Figure 4A**). PC1 explained ∼60% of the total dataset variability and mostly reflected differences among the four sampling dates (**Figure 4A** and Supplementary Figure 2), suggesting a slight delay in LT berry ripening at stages 3 and 4 in comparison to berries ripened under HT conditions. PC2 accounted for 13.3% of the total variability and mainly described differences between first and last sampling times, and intermediate ones (Supplementary Figure 2). Notably, PC3, explaining ∼8% of total dataset variability reflected differences among T0, LT and HT samples (**Figure 4A**). Indeed, the expression profiles of the first and last percentile PC3 loadings showed more expression in LT and HT samples, respectively (**Figures 4B,C**). The GO enrichment analysis revealed that the positive PC3 loadings were significantly enriched in the functional categories Cellular process, Biosynthetic process and Secondary metabolic process (**Figure 4E**). In particular, the first percentile PC3 loadings, more expressed in LT, comprised several genes involved in the metabolism of phenylpropanoids (5 Stilbene Synthase, STS, and 4 PAL), in the plant response to abiotic and biotic stresses (several ascorbate oxidases and glutaredoxines as well as many R proteins), in carbohydrate metabolism (one alcohol dehydrogenase and one trehalose-6 phosphate phosphatase). The transcript of ethylene response factor VvERF075, which belongs to the AP2/ERF superfamily and is usually upregulated in berry skin during ripening (Licausi et al., 2010), was also found more abundant in LT. Furthermore, we found among the positive PC loadings the transcription factor VvNAC60 already described as a putative master regulator of the transition between unripe and ripe red berries (Palumbo et al., 2014).

No significant GO enrichment could be found in the negative loadings of the PC3, more expressed under

HT regime (**Figure 4D**). The last percentile of the PC3 loadings showed higher level of transcript under HT regime of several cell-wall related transcripts (including two cellulose synthases, a pectinacetylesterase and a xyloglucan endotransglucosylase/hydrolase), of regulatory and structural genes involved in cytokinin metabolism (two isopentenyltransferases) and in ethylene metabolism (the 1-aminocyclopropane-1-carboxylate oxidase), in protein degradation and biosynthesis (including several proteases and amino acid transporters) and in the metabolism of carbohydrates and lipids (including an aldose 1-epimerase and a desaturase). Two genes involved in flavonoid biosynthesis and regulation, the dihydroflavonol-4-reductase and the transcription factor VvMybPA1, previously reported as a regulator of proanthocyanidin biosynthesis (Bogs

LT at the same date using ANOVA (∗P < 0.05). ns, not significant.

et al., 2007) were also more expressed in HT compared to LT.

In summary, the comparison of berry skin transcriptomes highlighted the effect of the temperature on gene expression during fruit ripening. In particular, LT condition was characterized by a higher expression of transcripts associated with the metabolism of phenylpropanoids.

### Gene Expression Profiles Inspection Highlights a Strong Effect of Temperature on Early Phenylpropanoid Pathway

We focused on changes in expression profiles of genes scoring an absolute value of fold change |FC| ≥ 2, identifying 2,257

of genes positively (right) and negatively (left) correlated to the third principal component were selected within the first (positive) and last (negative) percentile of the third component loadings. Enriched GO terms for the genes negatively (D) and positively (E) correlated to the PC3. The network graphs show BiNGO visualizations of the overrepresented GO terms. Categories in GoSlimPlants (Maere et al., 2005) were used to simplify this analysis. Colored nodes represent GO terms that are significantly overrepresented (p < 0.05). T0 = plants before treatment, LT = low temperature, HT = high temperature. Sample names are composed by temperature treatment abbreviation followed by the indication of the developmental stage (1, 2, 3, or 4), and by the description of the biological replicate (A–C). Gray, blue, and purple indicate samples of control, LT-treated and HT-treated, respectively.

annotated genes (Supplementary File 2). Differences in the timing of activation/repression of the differentially expressed genes during ripening in HT and LT was investigated in more detail by applying the short time-series expression miner (STEM) clustering method (Ernst et al., 2005) to the 2,257 identified genes (Supplementary File 3). **Figure 5** reports the most significant results of the STEM approach. Some of the transcriptional trends changes were also confirmed by Real Time qPCR analysis (Supplementary Figure 3).

Notably, many PAL and STS genes resulted as much more expressed in LT berry skins after veraison (**Figure 5A**), whereas they showed a slower rate of activation in HT berry skins (HT-2 → LT-10 in STEM analysis – Supplementary File 3), corroborating the PCA results (**Figure 4A**). Interestingly, a similar trend of expression was also detected for a protondependent oligopeptide transporter (POT), which contains a PTR2 domain that characterizes both nitrate and peptide transporters (Widhalm et al., 2015).

We found transcripts showing a peak of expression 10 days after veraison in HT, while their expression generally increased during the whole ripening period in LT (**Figure 5B**; HT-23 → LT-22 in STEM analysis – Supplementary File 3). Some genes involved in anthocyanin biosynthesis and transport (VvUFGT and VvGST4) and other phenylpropanoid/flavonoid related genes (one PAL, VvCHS3 and VvF3H1), were found to belong to this group, together with transcripts involved in hormone metabolism, such as one jasmonate O-methyltransferase and one gibberellin 20 oxidase, and with one gene that showed high homology with a cold-induced wall associated kinase (Cao et al., 2009).

We instead found that in HT berry skins, starting from 10 days after veraison, there was an earlier activation (HT-22 → LT-14 in STEM analysis – Supplementary File 3) of genes involved in the biosynthesis of volatile aromas [the terpene synthases VvTPS25 and VvTPS26, (Martin et al., 2010)], in cell wall metabolism and in DNA metabolism (**Figure 5C**). The same trend was shared by one peroxidase transcript (peroxidase 50).

Several genes that were downregulated in both HT and LT berries during ripening showed a different expression profile, i.e., a rapid and progressive decrease of expression starting from veraison in HT berries, and a much slower decrease in LT berries (**Figure 5D**; HT-4 → LT-12 in STEM analysis – Supplementary File 3). Cell wall-related transcripts, including genes linked to chlorophyll degradation and response to biotic stresses showed this trend. VvPrx31, a gene coding for a peroxidase putatively associated with anthocyanin degradation (Movahed et al., 2016) was also found in this group.

In order to point out the genes differentially expressed between LT and HT, we decided to consider just genes with a fluorescence expression threshold value ≥ 100 and transcripts in which the |FC| between LT and HT was ≥ 2 in at least one stage of development. By this approach, we obtained 417 differentially expressed genes (Supplementary File 4) that were over-represented in the GO functional categories of Secondary metabolic process, Generation of precursor metabolites and energy, Response to biotic stimulus, and Biosynthetic process (Supplementary Figure 4). These genes were grouped into five different FC clusters by K-means clustering (KMC) analysis highlighting the times with the greatest difference in gene expression between LT and HT (**Figure 6**). At a glance, the genes in each cluster showed a peak of expression at a given stage in LT. In the other ripening stages, expression of the same genes may be higher in HT, however, the FC of the LT/HT ratio at the peaking stage was generally much higher compared to the HT/LT values in the other stages. The most represented cluster included genes with high FC between LT and HT at 10 days after veraison (**Figure 6B**). The remaining differentially expressed genes were almost equally divided in the other four clusters (**Figure 6**). In the first cluster (**Figure 6A**), collecting genes more promptly activated by the LT regime at veraison (stage 1), we found a glycerol-3-phosphate acyltransferase 3 (AtGPAT3) to be the most differentially expressed gene (Supplementary File 4). Furthermore, several ERF/AP2 transcription factors (Licausi et al., 2010) were among the most differentially expressed genes belonging to this cluster (**Figure 6A** and Supplementary File 4). STSs were the most differentially expressed genes in LT regime at 10 and 20 days after veraison (**Figures 6B,C**), confirming the strong involvement of such genes in association with LT regime, as previously observed with the STEM approach (**Figures 4C**, **5A**). In these clusters, we also found genes involved in the response to biotic stimuli, such as wound induced proteins, beta 1–3 glucanase, and pathogenesis-related proteins, indicating a general activation of defense mechanisms under LT conditions. The cluster in **Figure 6D** is characterized by high FC in LT/HT 20 days after veraison (Supplementary File 4). Interestingly, this cluster included three ABC transporters and three isoforms of LRR receptor kinase CLAVATA1 (CLV1), which seem to be able to confer resistance to various abiotic stresses (Grzeskowiak et al., 2013). Lastly, several genes involved in volatile compounds synthesis were present in the last cluster, which groups genes showing a peak of expression in LT at 32 days after veraison (**Figure 6E** and Supplementary File 4). Two galactinol synthases, were also included in this cluster.

Overall, using this FC clustering approach we were able to validate expression of genes related to phenylpropanoid biosynthesis identified by STEM analysis, but also to retrieve other genes involved in the metabolism of volatile compounds, lipids and hormones that showed high sensitivity to thermal changes.

#### Temperature Affects the Activity of Enzymes Involved in Anthocyanin Metabolism

In LT treatment, growing temperatures positively impacted on the anthocyanins accumulation compared to HT (**Figure 3A**). However, such an accumulation was not fully explained by differences in transcription of genes related to phenylpropanoid/flavonoid metabolism. We therefore evaluated the enzymatic activity of PAL, the key enzyme of phenylpropanoid biosynthesis pathway (Zhang and Liu,

FIGURE 5 | Growing temperature regime affects specific clusters of genes. Four selected significant profiles (<5% Bonferroni correction method) of the 2,257 genes modulated in different growing temperature, from among 25 profiles obtained by STEM analysis. For each cluster, the average gene expression trend (top panels) and heat map of all the genes' expression profiles (bottom panels) are depicted. See Supplementary File 3 for the complete comparison profile table and clusters numbering. (A) HT 2 → LT 10, (B) HT 23 → LT 22, (C) HT 22 → LT 14, (D) HT 4 → LT 12. T0 = plants before treatment, LT = low temperature, HT = high temperature. Sample names are composed by temperature treatment abbreviation followed by the indication of the developmental stage (1, 2, 3 or 4). Data are the average of the three biological replicates. Gray, blue, and purple indicate samples of control, LT-treated and HT-treated, respectively.

2015), of UFGT, which catalyzes the last step of anthocyanin biosynthesis and is considered the key enzyme for this pathway (Kobayashi et al., 2001), and of peroxidases, which are supposed to degrade anthocyanins under elevated temperature growing conditions (Mori et al., 2007; Movahed et al., 2016). PAL activity gradually decreased in both temperature regimes until 10 days after veraison and was steady thereafter in HT, whereas it abruptly increased in LT, especially between 10 and 20 days after veraison. Afterward, high levels of PAL activity still persisted in LT berries. Thus, from the end of veraison to harvest, PAL activity under HT conditions was significantly lower in berry skins in comparison to LT (**Figure 7A**).

In order to elucidate a possible effect of the two thermal regimes on the affinity of UFGT for delphinidin or cyanidin, the enzymatic assay was performed twice using both substrates (**Figure 7B** and Supplementary Figure 5). After an initial fluctuating trend, in both cases the activity of UFGT showed an increasing trend from 10 days after veraison to harvest in HT and LT berries (**Figure 7B** and Supplementary Figure 5) and a strong and significant reduction of UFGT activity was observed in HT compared to LT from 20 days after veraison to harvest. No differences were found in terms of higher or lower affinity of the UFGT enzyme for delphinidin or cyanidin substrates as the activity of the enzyme showed comparable values between the two assays after veraison (**Figure 7B** and Supplementary Figure 5). The peroxidase activity of HT and LT berry skin showed a similar trend during ripening, with a significantly higher activity in HT than LT berries from veraison to harvest (**Figure 7C**), supporting a role of peroxidases in anthocyanins degradation under HT conditions.

Overall, the enzymatic activity analyses indicated the presence of a different balance between anthocyanin biosynthesis and degradation under LT and HT thermal regimes due to an increased action of PAL and UFGT enzymes in LT and of peroxidases in HT.

# DISCUSSION

# Increased Temperature Accelerates Sugar Accumulation and Acidity Depletion during Ripening

The increasing temperature associated with climate change is expected to modify air and land temperatures in most vine growing regions, which will undergo a warming of 2 to 4◦C in the next decades (Hannah et al., 2013). Mild to moderate temperature increases were shown to cause phenological changes in grapevine, accelerating the vegetative development and fruit maturation, ultimately affecting the berry composition (Duchene and Schneider, 2005; Jones et al., 2005). The two regimes analyzed in this study (LT and HT) can be considered representative of the inter-seasonal thermal variability occurring in recent years in the typical Sangiovese growing area of north-central Italy (Filippetti et al., 2013; Teslic et al., 2016 ´ ), with a tendency toward the thermal level of HT. The experiment design aimed to affect temperature regimes, without modifying other environmental parameters. The main differences between HT and LT regimes during preveraison-to-harvest period were mainly associated to an increase in maximum temperature under HT conditions. There was a general acceleration of sugar accumulation and total acidity reduction in the HT thermal regime. The effect of elevated temperature on sugar accumulation may depend on the amount of temperature variation as it has been reported that HTs (≥40◦C) could impact on the photosynthetic supply of sugar to the berry, causing a significant reduction in sugar accumulation (Greer and Weston, 2010; Greer and Weedon, 2013). On the contrary, the milder conditions of our experiment, in which temperatures ≥ 40◦C were seldom recorded under HT regime, led to an increased sugars accumulation, compared to LT, in all developmental stages except at harvest time. Indeed, a decrease in acidity and increase in pH associated with HT have

been reported in other grape varieties (i.e., Shiraz, Chardonnay and Cabernet Franc) grown under warm conditions (Sadras and Moran, 2012; Bonada and Sadras, 2015). However, other studies registered no effect of temperature increase on either titratable acidity or pH, nor on both simultaneously (Sadras and Moran, 2012; Greer and Weedon, 2013; Movahed et al., 2016) corroborating a previous hypothesis that grape berry acidity and pH depend on the interaction between cultivar and the amount of temperature increase (Sadras and Moran, 2012).

# Increased Temperature Negatively Affects Anthocyanin and Flavonol Concentrations

The influence of temperature on the flavonoid concentration of grape berry has been extensively reviewed (Downey et al., 2006; Teixeira et al., 2013). The biosynthesis of tannin and flavonols is high at flowering and in the berry skin the accumulation increases from fruit set until 1 to 2 weeks after veraison (Kennedy and Jones, 2001; Downey et al., 2003; Bogs et al., 2005). Anthocyanin accumulation, on the contrary, starts from veraison and reaches its maximum in the latest phases of fruit maturation, when their synthesis ceases (Boss and Davies, 2001). In our study, no significant relationship could be verified between temperature increase and total skin tannin concentration in Sangiovese berries (**Figure 3**). These results, in line with the findings by Cohen et al. (2012), suggested that increasing temperature from veraison to harvest has little impact on tannin accumulation, probably because these compounds have already been synthesized. Furthermore, skin tannins are the most stable flavonoids under diverse growing conditions, due to their chemical structure which is widely variable in size, ranging from dimers to polymers with more than 40 units (Teixeira et al., 2013). This could cause less susceptibility to potential degradative processes induced by temperature.

The HT thermal regime induced a similar decrease in the concentration of flavonols and anthocyanins (**Figure 3**). Flavonols are very sensitive to changes in environmental conditions. For example, sunlight is known to enhance flavonols accumulation in berries (Downey et al., 2006), reflecting their role as UV protectants (Spayd et al., 2002; Pastore et al., 2013). Since we maintained the same light intensity and quality between the two greenhouses, our data highlighted the strong temperature effect on this class of compounds. The reduced accumulation of anthocyanins in Sangiovese berries during ripening under increased temperature has been already reported for various genotypes in different conditions (Spayd et al., 2002; Mori et al., 2005, 2007; Tarara et al., 2008; Movahed et al., 2016).

Anthocyanins and flavonols changed also their composition, in berry skins. In particular, HT berries had a lower percentage of the anthocyanin malvidin 3-G and the flavonol quercetin. In grapevine, growing temperatures have been associated with increased proportions of highly hydroxylated and methylated anthocyanins (Mori et al., 2007; Tarara et al., 2008; Cohen et al., 2012). Conversely, in a recent study an increase in the degradation of petunidin and malvidin glucoside at elevated temperatures was observed in Cabernet Sauvignon grapes exposed to labeled phenylalanine (Chassy et al., 2015). The relationship between berry temperature and flavonols profiles has been less extensively studied. Consistently with our results, a higher proportion of flavonols with di-hydroxylation, as quercetin, was detected in Merlot berries when temperature was reduced by approximately 8◦C in comparison with control temperature (Cohen et al., 2008).

Overall, this evidence supports the high susceptibility of anthocyanins and flavonols to air temperature, and the critical

role of experimental conditions in this kind of assessment. Accumulation of these classes of secondary compounds is the result of complex and interconnected processes such as synthesis, degradation, hydroxylation, methylation, acylation and transport, thus it is not always possible to determine general univocal relations.

## Transcriptomic Analysis Highlights Processes Affected by Temperature in Berry Skin

We analyzed the entire transcriptome of the berry skin in cv. Sangiovese during ripening exploiting a combination of complementary statistical approaches to retrieve those transcripts mostly associated with each of the two temperature regimes. In particular, the PCA analysis suggested that, beyond a slight transcriptional hastening of ripening in a few developmental stages of HT samples, a clear rearrangement in the skin transcriptome (∼8% total variability of the dataset) can be ascribed to the imposition of different thermal regimes, throughout the course of the experiment.

The inhibitory effects of HT on stilbene biosynthetic pathway have already been described by Rienth et al. (2014) and in our conditions many members of the STS and PAL gene families were induced under LT regime suggesting a clear activation of stilbene biosynthesis. A coordinated gene expression of PAL and STS was observed in grape berry, suggesting that several enzymatic steps in the stilbene biosynthetic pathway are co-regulated (Zenoni et al., 2016).

Interestingly, several ERF transcription factors resulted as promptly higher expressed under LT compared to HT regime, whereas few were less expressed under LT regime, suggesting that members of this family of transcription factors, known regulators of thermotolerance in plants (Carbonell-Bejerano et al., 2013) play a central role in the response to temperature variation in grapevine berry skin as reported for Arabidopsis (Cheng et al., 2013), Chickpea (Cicer arietinum) (Deokar et al., 2015) and tomato (Severo et al., 2015). Moreover, the presence of ACO1 and ACS among genes more expressed in LT suggests that the ethylene signaling is required in the regulation of the temperature-driven ripening processes of grapevine berries. Another response observed under LT regime was associated to the reaction toward environmental solicitations. This included the glycerol-3-phosphate acyltransferases (GPAT) that catalyze the acylation at sn-1 position of glycerol-3-phosphate to produce lysophosphatidic acid (LPA), an important intermediate for the formation of different types of acyl-lipids (Chen et al., 2011) and two galactinol synthases, whose involvement in multiple abiotic stresses responses (Zhuo et al., 2013) and in particular in heat stress responses (Pillet et al., 2012) has been reported.

Lastly, some genes involved in the biosynthesis of aromas, such as linalool synthase, delta-cadinene synthase, vetispiradiene synthase and a germacrene synthase, were found more expressed under LT conditions. These results well corroborate previous reports regarding the reduction in aromatic potential in grapevine berries exposed to HT (Rienth et al., 2014).

# Transcriptional and Post-transcriptional Regulation of Genes Belonging to the Phenylpropanoid/Flavonoid Pathway

The accumulation of flavonoids and anthocyanins in Sangiovese berries skin was strongly affected by the two thermal regimes. However, a clear correlation between the upregulation of genes involved in anthocyanin and flavonol biosynthesis and transport and the higher levels of these compounds under LT conditions was not revealed. The slight difference of the VvUFGT and VvGST4 expression profile in LT compared to HT seems insufficient to support the significant difference in anthocyanin accumulation. Similar results were obtained by other authors reporting that mRNA accumulation of anthocyanin biosynthetic genes was not reduced under HT growing conditions (Mori et al., 2007; Carbonell-Bejerano et al., 2013; Rienth et al., 2014). Moreover, the expression of VvMYBA1, the transcription factor that activates VvUFGT and VvGST4 in grape berry skin, was not significantly affected by the thermal regime, under our experimental conditions. A first hypothesis to explain the strong differences observed in terms of flavonoid accumulation between LT and HT, based on the expression profiles of other genes involved in such metabolism, suggests the involvement of the proanthocyanidin regulator VvMybPA1 in the activation of an alternative branch under HT regime, competing with anthocyanins pathway. Another hypothesis arose from the high expression of several PAL members in LT conditions, which could supply substrates to both stilbenes and, presumably, flavonoid accumulation. An additional possibility derives from the observation that the enzymatic activities of PALs and VvUFGT were significantly higher in LT thermal regime. This was consistent with the transcriptomic data for PALs, whereas clearly discordant with the VvUFGT expression profile. These enzymatic assays clearly support the higher levels of anthocyanins and flavonols observed in berry skins under LT conditions, and that a posttranscriptional mechanism may be crucial in the regulation of the late steps of anthocyanin biosynthesis. Recent researches have revealed that post-translational mechanisms may play significant roles in the regulation of phenylpropanoid/flavonoid metabolic pathways. In Arabidopsis and apple (Malus domestica), where MYB transcriptions factors are required for anthocyanin accumulation and for the expression of structural genes in the anthocyanin biosynthesis pathway, it has been demonstrated that the repression of anthocyanin accumulation in darkness requires an interaction between a proteasome protein complex and MYBs (Li et al., 2012; Maier et al., 2013). Furthermore, a post-translational regulation mediated by the ubiquitin/26S proteasome system of the anthocyanin-related TRANSPARENT TESTA 8 (TT8) transcription factor has been detected in Arabidopsis (Patra et al., 2013). The possibility of a posttranscriptional regulation of the enzymatic steps of the pathway has also been described, showing that a multiple-level control governs the enzymatic activity of PAL (Zhang et al., 2013; Zhang and Liu, 2015).

The analysis of peroxidase activity revealed a significant increase under HT conditions, supporting that this class of

enzymes may trigger the temperature-dependent degradation of anthocyanins (Mori et al., 2007). Our transcriptomic survey did not reveal a significant increase in the gene expression of the recently characterized VvPrx31 (Movahed et al., 2016) in HT conditions compared to LT, suggesting that new peroxidases isoforms (as peroxidase 50) may be involved in anthocyanins degradation following increasing temperature.

Overall, it is possible to hypothesize that different environmental conditions could influence anthocyanin biosynthesis and degradation in grapevine through posttranscriptional modifications of key structural genes of the pathway, such as UFGT and peroxidases.

#### CONCLUSION

Our results provide valuable insights into the understanding of the mechanisms that underlie Sangiovese berries response to changes in growing temperatures that could be useful to identify the most suitable areas for Sangiovese cultivation under current climate change, given the great sensitivity of this cultivar to the increasing temperature. Two thermal regimes characterized by difference of 5◦C in average and 7◦C in maximum air temperature, specifically affected berry ripening. The LT regime delayed ripening in the early phases and had great impact on grape phenolic composition and on the activity of some enzymes involved in flavonoid biosynthesis, enhancing the accumulation of anthocyanins and flavonols. Conversely, berries grown at HT showed an increase in peroxidase activity, which could concur to the reduced accumulation of flavonoids found in these conditions. Transcriptional analyses identified the existence of a strong effect of both thermal regimes on the whole transcriptome, but the partial correlation between biochemical and molecular data requires further research to elucidate the existence of posttranscriptional and post-translational mechanisms involved in

#### REFERENCES


the balance between biosynthesis and degradation of flavonoids and in particular of anthocyanins.

## AUTHOR CONTRIBUTIONS

CP performed the RNA extraction, contributed to the enzymatic and HPLC analyses and wrote the manuscript. SDS conducted the microarray experiments, interpreted the microarray data and wrote the manuscript. SZ and GT designed the microarray experiments and critically revised the manuscript. NM, GA, and GV sampled the material and contributed to the enzymatic and HPLC analysis. IF conceived and supervised the study, wrote and critically revised the manuscript.

# FUNDING

CP was granted by Alma Mater Studiorum – University of Bologna research fellowship. SDS was financially supported by the Italian Ministry of University and Research FIRB RBFR13GHC5 project "The epigenomic plasticity of grapevine in genotype per environment interactions."

### ACKNOWLEDGMENT

We thank Emilia Colucci for her help in setting up the greenhouses.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2017.00929/ full#supplementary-material

the Grapevine, ed. K. A. Roubelakis-Angelakis (Dordrecht: Kluwer Academic Publishers), 1–33.


different stress signals. Plant Physiol. 162, 1566–1582. doi: 10.1104/pp.113. 221911



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Pastore, Dal Santo, Zenoni, Movahed, Allegro, Valentini, Filippetti and Tornielli. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Constraint-Based Modeling Highlights Cell Energy, Redox Status and α-Ketoglutarate Availability as Metabolic Drivers for Anthocyanin Accumulation in Grape Cells Under Nitrogen Limitation

#### Edited by:

José Tomás Matus, Universitat Autònoma de Barcelona, Spain

#### Reviewed by:

Maria Pedreno, Universidad de Murcia, Spain Roque Bru-Martinez, University of Alicante, Spain

#### \*Correspondence:

Eric Gomès eric.gomes@inra.fr †These authors have contributed

#### Specialty section:

equally to this work.

This article was submitted to Plant Breeding, a section of the journal Frontiers in Plant Science

Received: 07 January 2018 Accepted: 16 March 2018 Published: 17 May 2018

#### Citation:

Soubeyrand E, Colombié S, Beauvoit B, Dai Z, Cluzet S, Hilbert G, Renaud C, Maneta-Peyret L, Dieuaide-Noubhani M, Mérillon J-M, Gibon Y, Delrot S and Gomès E (2018) Constraint-Based Modeling Highlights Cell Energy, Redox Status and α-Ketoglutarate Availability as Metabolic Drivers for Anthocyanin Accumulation in Grape Cells Under Nitrogen Limitation. Front. Plant Sci. 9:421. doi: 10.3389/fpls.2018.00421 Eric Soubeyrand<sup>1</sup>† , Sophie Colombié<sup>2</sup>† , Bertrand Beauvoit<sup>2</sup>† , Zhanwu Dai<sup>3</sup> , Stéphanie Cluzet<sup>4</sup> , Ghislaine Hilbert<sup>3</sup> , Christel Renaud<sup>3</sup> , Lilly Maneta-Peyret<sup>5</sup> , Martine Dieuaide-Noubhani<sup>6</sup> , Jean-Michel Mérillon<sup>4</sup> , Yves Gibon<sup>2</sup> , Serge Delrot<sup>1</sup> and Eric Gomès<sup>1</sup> \*

<sup>1</sup> UMR 1287 Ecophysiologie et Génomique Fonctionnelle de la Vigne, Université de Bordeaux, Institut des Sciences de la Vigne et du Vin, Bordeaux, France, <sup>2</sup> UMR 1332 Biologie du Fruit et Pathologie, INRA-Bordeaux, IBVM, Bordeaux, France, <sup>3</sup> UMR 1287 Ecophysiologie et Génomique Fonctionnelle de la Vigne, INRA-Bordeaux, Institut des Sciences de la Vigne et du Vin, Bordeaux, France, <sup>4</sup> EA 3675 GESVAB, Université de Bordeaux, Institut des Sciences de la Vigne et du Vin, Bordeaux, France, <sup>5</sup> UMR 5200 Laboratoire de Biogenèse Membranaire, Université de Bordeaux, Bordeaux, France, <sup>6</sup> UMR 1332 Biologie du Fruit et Pathologie, Université de Bordeaux, IBVM, Bordeaux, France

Anthocyanin biosynthesis is regulated by environmental factors (such as light, temperature, and water availability) and nutrient status (such as carbon, nitrogen, and phosphate nutrition). Previous reports show that low nitrogen availability strongly enhances anthocyanin accumulation in non carbon-limited plant organs or cell suspensions. It has been hypothesized that high carbon-to-nitrogen ratio would lead to an energy excess in plant cells, and that an increase in flavonoid pathway metabolic fluxes would act as an "energy escape valve," helping plant cells to cope with energy and carbon excess. However, this hypothesis has never been tested directly. To this end, we used the grapevine Vitis vinifera L. cultivar Gamay Teinturier (syn. Gamay Freaux or Freaux Tintorier, VIVC #4382) cell suspension line as a model system to study the regulation of anthocyanin accumulation in response to nitrogen supply. The cells were sub-cultured in the presence of either control (25 mM) or low (5 mM) nitrate concentration. Targeted metabolomics and enzyme activity determinations were used to parametrize a constraint-based model describing both the central carbon and nitrogen metabolisms and the flavonoid (phenylpropanoid) pathway connected by the energy (ATP) and reducing power equivalents (NADPH and NADH) cofactors. The flux analysis (2 flux maps generated, for control and low nitrogen in culture medium) clearly showed that in low nitrogen-fed cells all the metabolic fluxes of central metabolism were decreased, whereas fluxes that consume energy and reducing power, were either increased (upper part of glycolysis, shikimate, and flavonoid pathway) or maintained (pentose phosphate pathway). Also, fluxes of flavanone 3β-hydroxylase, flavonol synthase, and anthocyanidin synthase were strongly increased, advocating for

a regulation of the flavonoid pathway by alpha-ketoglutarate levels. These results strongly support the hypothesis of anthocyanin biosynthesis acting as an energy escape valve in plant cells, and they open new possibilities to manipulate flavonoid production in plant cells. They do not, however, support a role of anthocyanins as an effective mechanism for coping with carbon excess in high carbon to nitrogen ratio situations in grape cells. Instead, constraint-based modeling output and biomass analysis indicate that carbon excess is dealt with by vacuolar storage of soluble sugars.

Keywords: anthocyanins, grapevine, cell redox status, energy escape valve hypothesis, constraint-based modeling

#### INTRODUCTION

Flavonoids are naturally occurring secondary metabolites belonging to the group of polyphenols, which are ubiquitous in all land plants, with currently over 9,000 compounds identified (Buer et al., 2010). Among polyphenols, flavonoids encompass over 6000 distinct molecules, divided into aurones, flavones, flavonols, flavanols, anthocyanins, phlobaphenes, and isoflavonoids, the last two being almost exclusively synthesized in maize and leguminous plants (Hichri et al., 2011). They exhibit a large variety of biological roles in plants. They control pollen fertility in many species (Taylor and Jorgensen, 1992) and influence auxin transport (Peer and Murphy, 2007). Light absorbing pigments such as anthocyanins and aurones color flower petals and fruit epicarp, thus facilitating pollinator attraction and seed dispersal (Mol et al., 1998). With regard to human health, the consumption of grapes or grape-derived products, has been correlated with a reduced incidence of a number of chronic illnesses (Iriti and Faoro, 2009; Kozlowska and Szostak-Wegierek, 2014), and flavonoids have been proposed as major contributors of these health-promoting effects (Butelli et al., 2008; Qin et al., 2011).

Anthocyanins, which are key compounds for premium red wine making, are present in the skin (epicarp) of the red grape berries, and sometimes, in the case of the so-called "teinturier" cultivars also in the pulp (mesocarp) (Petrussa et al., 2011; Guan et al., 2016). Hence, in order to optimize anthocyanin content in the berries, it is important to understand the molecular regulation of the anthocyanin production by environmental factors and viticultural practices. Anthocyanins are synthesized through the phenylpropanoid and flavonoid pathways, starting with phenylalanine as a precursor, and splitting into two branches to produce the di- and tri-hydroxylated flavonoids (**Figure 1**; Tanaka et al., 2008; He et al., 2010). The accumulation and the proportion of these compounds in the berry skin depends on genetic, developmental, and environmental factors (Rio Senegade et al., 2008; He et al., 2010; Dai et al., 2011) as well as on viticultural practices (Downey et al., 2006). Light, temperature, irrigation, and nitrogen supply have been shown to impact grape berry anthocyanin content (Dai et al., 2011; Berdeja et al., 2015; Keller, 2015; Habran et al., 2016).

In several crops like tomato or grapevine and model plants such as Arabidopsis, nitrogen depletion increases the concentration of phenolics in general and of anthocyanins in particular (Keller and Hrazdina, 1998; Hilbert et al., 2003; Fritz et al., 2006; Feyissa et al., 2009; Løvdal et al., 2010; Soubeyrand et al., 2014; Habran et al., 2016). The transcriptional regulation of the phenylpropanoid and flavonoid pathways in response to low nitrogen availability has been extensively studied in Arabidopsis (Lillo et al., 2008), tomato (Løvdal et al., 2010), tobacco (Fritz et al., 2006), or grapevine (Soubeyrand et al., 2014). These studies shed light on the molecular mechanisms underlying the regulation of the flavonoid metabolism by nitrogen depletion, pointing out the responsiveness of the pathway's positive and negative regulators, i.e., R2-R3MYB and LBD transcription factors. They do not, however, address the pending question of the existence of a metabolic "driver" that would fuel the increase in flavonoid biosynthesis. According to Hernández and Van Breusegem (2010), a plausible candidate for such a metabolic driver could be the cell energetic status. High carbon-to-nitrogen ratio leads to an energy excess, both in terms of high ATP and high reducing power (NADH and NADPH); and an increase in flavonoid biosynthetic metabolic fluxes would act as an "energy escape valve," helping plant cells to cope with that energy excess. Indeed, the flavonoid pathway consumes ATP and NADPH reducing equivalents in several of its enzymatic steps particularly when the shikimate pathway, that links the central metabolism to the phenylpropanoid pathway, is also taken into account (**Figure 1**).

Constraint-based modeling can be used in order to test that hypothesis by comparing maps of metabolic fluxes in the two contrasted situations, i.e., in nitrogen limiting condition compared to the control condition. Mathematical modeling of metabolism is a particularly promising tool as it offers a systems approach to analyze the structure, dynamics, and behavior of complex metabolic networks. In plant research, the issue of modeling metabolism is increasingly gaining attention, and several mathematical modeling approaches applied to plant metabolism exist (for reviews, see Giersch, 2000; Morgan and Rhodes, 2002; Poolman et al., 2004; Rios-Estepa and Lange, 2007). Constraint-based modeling such as flux balance analysis (FBA, Orth et al., 2010) allows the prediction of metabolic fluxes at steady-state by applying mass balance constraints to a stoichiometric model. Mass-balance information, such as growth rate, biomass composition, and substrate consumption rate, are used to fix boundaries on the flux solutions space (Reed and Palsson, 2003) and an objective function is used to identify the optimal flux distribution among all possible steady-state flux

distributions. This modeling has the advantage of not requiring the knowledge of enzyme kinetic parameters.

The present work aims to investigate the metabolic flux reorganization that is involved in the response of anthocyanin accumulation to nitrogen supply, taking advantage of a grapevine red cell suspension system, the GT3 Vitis vinifera L., cv. Gamay Teinturier (syn. Freaux or Gamay Freaux Tintorier, Vitis International Variety Catalogue #4382, Decendit and Mérillon, 1996). To this end, the cells were cultivated in control or nitrogen limiting conditions. Then we generated and compared flux maps of plant cell metabolism by coupling the network of heterotrophic metabolism previously described (Colombié et al., 2015) with the overall reactions to phenolic compounds production (anthocyanins, flavonols, tannins, and stilbenes), paying special attention to energetic processes by balancing cofactors. The results are consistent with excessive ATP and reducing equivalent (NADPH mostly) as well as α-ketoglutarate availability acting as "pushers" that increase anthocyanin and more broadly polyphenol biosynthesis in nitrogen-depleted cells.

#### MATERIALS AND METHODS

## Grapevine Cell Culture Growth and Sampling

Vitis vinifera cv. "Gamay Fréaux" var. teinturier GT3 cell suspensions were sub-cultured on a modified Gamborg B5 medium, supplemented with 20 g L−<sup>1</sup> sucrose, 250 mg L−<sup>1</sup> casein hydrolysate, 0.1 mg mL−<sup>1</sup> 1-naphthalene acetic acid and 0.2 mg mL−<sup>1</sup> kinetin (Saigne-Soulard et al., 2006). Cells were routinely sub-cultured every 7 days in 250 mL Erlenmeyer flasks containing 50 mL of culture medium. For experimental purpose, 7-days old cells were inoculated, with a 1:6 (v/v) ratio in 200 mL of the same medium but containing either 5 mM (final concentration, low nitrogen, N−) or 25 mM (final concentration, control, N) KNO3, in 1 L Erlenmeyer flasks. The ammonium concentration was identical in both N<sup>−</sup> and N treatments (2 mM). For each sampling point, three replicate flasks of cell culture were harvested at 0 (or 1 for experiment 1), 4, 6, 8, and 11 days post-inoculation by vacuum filtration, quickly washed twice with

ice-cold distillated water, weighed and quick-frozen in liquid nitrogen. Frozen cells were then reduced to fine powder in a liquid nitrogen-cooled MM200 ball grinder (Retsch, Haan, Germany), and stored at −80◦C until further analysis.

#### Cell Biomass and Metabolites Content Analysis

#### Phenolic Compounds

Anthocyanins and flavonols were analyzed on powdered freezedried cells, which were extracted and analyzed according to Soubeyrand et al. (2014).

Tannins and stilbenes were extracted from 40 mg of freezedried cells with 4 mL of methanol (100%) overnight at +4 ◦C. The samples were centrifuged at 6,000 × g for 10 min. Two milliliters of supernatant were vacuum-dried using a SpeedVac SC 110 plus (Thermo Fisher Scientific, Saint-Herblain, France) for the analysis of stilbenes and 50 µL were used for the analysis of the total phenolic content. Then, 100 µL MeOH (100%) and 1 mL H2O were added to the vacuum-dried samples and filtered through an Ion Exchange Resin (Dowex 50 WX 4-400) to remove anthocyanins. Extracts were vacuum-dried using a SpeedVac SC 110 plus and the dry pellet was re-suspended in 800 µL of MeOH/H20 50/50 (v/v) for the HPLC analysis. Extracts were then filtered through a 0.45 µm polypropylene syringe filter (Pall Gelman Sciences Corp., Ann Arbor, MI, United States). Stilbenes analysis was performed with a Summit HPLC System consisting of P680 pump, ASI-100TTM autosampler and UVD 340U UV-Vis detector operating at 320 nm (Dionex Corporation, Sunnyvale, CA, United States) (Ramirez-Lopez and DeWitt, 2014). After injecting 20 µL, separation was achieved at ambient temperature on a reverse-phase Ultrasphere ODS column 25 cm × 4.6 mm, 5 µm particle size with an Ultrasphere ODS guard column 4.5 cm × 4.6 mm (Beckman Instruments Inc., Fullerton, CA, United States). All reagents were of analytical grade. Separation was performed according to Saigne-Soulard et al. (2006).

Total phenolic content was assessed by the Folin–Ciocalteu method (Singleton and Rossi, 1965). The assay mixture (3 mL) contained 50 µL of extract, 450 µL MeOH/H2O (50/50, v/v), 250 µL Folin–Ciocalteu reagent and 2.25 mL of ultrapure water. After 3 min, 2 mL Na2CO<sup>3</sup> (75g L−<sup>1</sup> ) were added and the samples were incubated at 50◦C for 5 min and absorbance was read at 760 nm. Calculation of phenolics was based on a standard curve prepared using gallic acid, and the results were expressed as mg gallic acid equivalents per liter (mg GAE L−<sup>1</sup> ).

#### Sugars and Amino Acids

Five hundred milligrams of cell powder (FW) were extracted from cell suspension samples using decreasing concentrations of ethanol: ethanol 80%, ethanol 50% (v/v) and ultrapure water. All three supernatants were pooled, vacuum-dried using a Speed Vac SC 110 plus (Thermo Fisher Scientific, Saint-Herblain, France). The dry pellet was re-suspended in 2 mL of ultrapure water and stored at −20◦C before further analysis.

Amino-acid content was analyzed by the method described by Cohen and Michaud (1993), modified according to Martínez-Lüscher et al. (2014). Briefly, after derivatization with 6 aminoquinolyl-N-hydroxysuccinimidyl-carbamate, amino acids were analyzed using a Waters 2695 HPLC system equipped with a Waters 474 fluorescence detector (Waters, Milford, MA, United States). Separation was performed on a Nova-Pack C18 AccQ-Tag column (Waters, Milford, MA, United States) at 37◦C with elution at 1 mL min−<sup>1</sup> with a 67 min linear gradient (eluent A, sodium acetate buffer, 140 mM at pH 5.7; eluent B, acetonitrile 60% in water (v/v)). Chromatograms corresponding to excitation at 250 nm and emission at 395 nm were recorded and quantified with chemical standards purchased from Sigma (St. Louis, MO, United States).

Soluble sugars (glucose, fructose, and sucrose) were measured enzymatically with a microplate reader (ELx800UV, BioTek Instruments Inc., Winooski, VT, United States) as described by Gomez et al. (2007).

#### Malic Acid, Total Starch, and Proteins

Malate, starch, and protein content were measured as described in Biais et al. (2014).

#### Cell Wall Total Polysaccharides

Total cell wall polysaccharides quantification was subcontracted to the BIBS platform of INRA-Nantes<sup>1</sup> , using 100 mg of freezedried cell powder, as described in Colombié et al. (2015).

#### Lipids

Five hundred milligrams of cell powder (FW) were extracted by 1 mL of MeOH:H2SO<sup>4</sup> (40:1, v/v), supplemented with 2 µg of heptadecanoic acid (internal standard) and incubated 60 min at 80◦C in screw-capped tubes. Then, 400 µL hexane and 1.5 mL of ultrapure water were added, vigorously mixed and centrifuged at 3,000 g for 5 min. The organic phase was collected and transferred to injection vial to analyze fatty acids by GC-FID (Gas Chromatography coupled to Flame Ionization Detection), as described by Maneta-Peyret et al. (2014).

#### Total Carbon and Nitrogen Content

Cell total carbon and nitrogen contents were determined by Dumas' combustion method, with a Flash EA 112 autoanalyzer (Thermo fisher, Courtaboeuf, France), following the manufacturer's instructions and using 8 mg of freeze-dried cell powder. In the case of culture medium analysis, 250 µL of freeze-dried medium were used instead.

#### Total Nucleic Acid

The total DNA content was measured using the deoxyribosespecific diphenylamine reaction, using 15 mg of freeze-dried cell powder as starting material, and salmon sperm DNA for calibration (Colombié et al., 2015).

#### Enzyme Capacity Determinations

Phenylalanine ammonia-lyase (PAL) activity was measured according to Gagné (2007). Approximately 250 mg of cell powder were extracted by vigorous shaking with 40 mg polyvinylpolypyrrolidone (PVPP) and 2.5 mL extraction buffer composed of 0.1 mM Tris–HCl (pH 8.8), 5 mM EDTA, 0.05% spermidine (w/v), 4 mM β-mercaptoethanol,

<sup>1</sup>http://www.bibs.inra.fr

and 1 mM phenylmethylsulfonyl fluoride (PMSF, added just prior extraction). The samples were centrifuged for 20 min at 16,000 × g at 4◦C. The protein extract was desalted on a PD-10 column (Sephadex resin G-25, PD-10 column, GE Healthcare) equilibrated with 25 mL of 0.1 mM Tris–HCl (pH 8.8). Aliquots of desalted proteins were frozen in liquid nitrogen and stored at −80◦C. Spectrophotometric assays contained 300 µL of protein extract in 30 mM L-phenylalanine in 0.1 M Tris–HCl (pH 8.8) and 150 µL of 30 mM L-phenylalanine in 0.1 M Tris–HCl (pH 8.8). Reactions were incubated for 15 to 180 min at 37◦C. The amount of trans-cinnamic acid formed in the assay was measured spectrophotometrically at 290 nm. PAL activity was expressed as µg of cinnamic acid formed per µg FW−<sup>1</sup> .

For all other enzyme measurements [glucose-6-phosphate dehydrogenase (G6PDH), phosphoglucomutase (PGM), fructose hexokinase (FK), glucose hexokinase (GK), shikimate dehydrogenase (SD), enolase (ENO), pyruvate kinase (PK) and glutamine synthetase (GS)], aliquots of 20 mg frozen FW cell powder were extracted by vigorous mixing with extraction buffer (Nunes-Nesi et al., 2007). FK, GK, G6PDH, PK, SD, and GS were assayed as described in Gibon et al. (2004). ENO was assayed as described by Biais et al. (2014), PGM was assayed according to Gibon et al. (2009).

#### Respiration Measurements

Oxygen consumption rates of cells were measured with Clark's electrode at 25◦C in a 1 mL thermostatically controlled chamber. Respiration assays of growing cells were performed in the GT3 cell suspension medium under stirring. Seven hundred and fifty microliters of cell suspension were centrifuged (1,500 × g for 5 min) and the resulting pellet gently re-suspended in 1 mL of cell culture medium. Respiration rates were initially expressed in nmol O<sup>2</sup> min−<sup>1</sup> g <sup>−</sup><sup>1</sup> FW.

#### Coenzyme Analysis

All extractions were performed at 4◦C with 200 mg of frozen powder cell. For the assays of NAD<sup>+</sup> and NADP+, aliquots of frozen cells were extracted with 500 µL of 0.2 N HCl then incubated for 5 min at 80◦C. Fifty microliters of 0.2 M NaH2PO<sup>4</sup> (pH 5.6) was added and the extract was neutralized to a final pH in the range from 5.5 to 6.5 with 0.2 M NaOH. To quantify NADH and NADPH, other aliquots of frozen cells were extracted as for NAD<sup>+</sup> and NADP<sup>+</sup> except that the extraction medium was 0.2 M NaOH and the heated sample was neutralized with 0.2 N HCl to a final pH in the range from 7.5 to 8.5.

Coenzyme content was quantified by adapting methods described by Wilhelm (2009). The reaction buffer was composed by Tris/KOH (pH 7.7) qsp 350 µL, 100 µL of 10 mM methylthiazolyldiphenyl-tetrazolium (MTT) and 50 µL of 4 mM phenazine ethosulfate (PES). For the NAD<sup>+</sup> and NADH assay, the reaction was started by adding 3.5 U of alcohol dehydrogenase (ADH, Roche, Melan, France) and 10 µL ethanol (99%). For the NADP<sup>+</sup> and NADPH assay, 1.6 U of glucose-6-phosphate dehydrogenase (Roche) and 0.5 M glucose-6-phosphate were added to the assay. Absorbance was read at 570 nm for 10 min, and the results were expressed in nmol g−<sup>1</sup> FW.

#### Modeling

Concentrations of accumulated metabolites and biomass components were converted from gram-to mole-basis and then multiplied by the specific growth rate calculated at day 4 and 6 in order to calculate the corresponding fluxes used as constraints in the flux balance model. Stoichiometric network reconstruction encompassing central and polyphenol metabolism (model in sbml format, Supplementary Presentation 2) and mathematical problems were implemented using MATLAB (Mathworks R2012b, Natick, MA, United States) and the optimization toolbox, solver quadprod with interior-point-convex algorithm for the minimization.

#### Statistical Analysis

Statistical analyses were done using the statistical package of the "R" software (R Development Core Team, 2010). A oneway analysis of variance (ANOVA) was used. Unless otherwise stated, the mean of the 3 biological replicate treatments was used in data analysis. Unless otherwise stated, comparisons of means were performed using HSD.r multiple comparisons function of Tukey's post hoc test at P < 0.05.

### RESULTS

#### Anthocyanin Accumulation in Cells Cultivated Under Control and Low Nitrogen Conditions

Dry biomass accumulation kinetics were nearly identical for cells cultured in either control or low nitrogen conditions, increasing from about 1.6 g DW−<sup>1</sup> of cells per L at culture initiation to about 10 g DW−<sup>1</sup> at the 8th day of culture, and starting to decline afterwards (**Figure 2A**). Conversely, total anthocyanin accumulation patterns were strikingly different in control and low nitrogen condition cultures (**Figure 2B**). In control cells (N, 25 mM NO<sup>3</sup> <sup>−</sup>), total anthocyanin content was fairly stable during the culture (around 2–3 mg g DW−<sup>1</sup> ), whereas in cells cultivated at low nitrogen (N−, 5 mM NO<sup>3</sup> <sup>−</sup>) total anthocyanin cell content strongly increased from the 4th day of culture to the 12th day, reaching a maximum of about 20 mg g DW−<sup>1</sup> at the end of the culture.

This result has been reproduced with three cell cultures (named exp. 1, exp. 2, and exp. 3), that were conducted in the same conditions. Similar growth profiles were observed with about 10 g DW−<sup>1</sup> of cells per L at the 11th day of culture (Supplementary Figure 1). Plotted in log-scale, the cell biomass concentration increased linearly up to the 6th day of culture. Maximal specific growth rate was reached at the 4th day of culture, in the exponential growth phase. Consequently, the steady state was assumed for modeling, i.e., the state when there was no accumulation of internal metabolites of the network, at day 4. Nevertheless, all data have also been collected and analyzed at the 6th day of culture to assess the evolution of the metabolic fluxes from day 4 to 6. For each experiment, the specific growth rate (µ in day−<sup>1</sup> ) was determined as the growth rate (g DW day−<sup>1</sup> ) related to the biomass (g DW) and the average

was: 0.288 ± 0.013 day−<sup>1</sup> and 0.160 ± 0.045 day−<sup>1</sup> at day 4 and 6 of the culture, respectively.

## Nitrogen and Carbon Consumption in Culture Medium

Both total nitrogen and total carbon have been determined in the medium (for experiments 2 and 3, Supplementary Figure 2). Only in the case of low nitrogen condition (N−), nitrogen was limiting and even fully depleted since the 6th day of culture. The total carbon concentration in the medium was stable since the beginning up to the 8th day of culture, implying that there was no carbon limitation.

#### Glucose and Fructose as Metabolized Sugars

Sucrose (20 g L−<sup>1</sup> ) was the carbon source supplied in the culture medium, but it is generally cleaved to form hexoses by cell wall invertase activity (Atanassova et al., 2003; Chen et al., 2013). Measurements of sucrose, glucose, and fructose in the medium during 2 cultures (exp. 2, Supplementary Figure 3; exp. 3, Supplementary Figure 4) showed that hexose concentrations were higher than sucrose concentration at the 4th of culture. Thus glucose and fructose were assumed to be the main sugars metabolized by the cells.

#### Metabolic Fluxes Modeling Flux-Balance Model

The flux-balance model was constructed by integrating biochemical and physiological knowledge about the stoichiometry of reactions and the boundary conditions, i.e., the definition of external compounds. The model describes one cell and assumes that the suspension is homogeneous. The model combines the central metabolism previously described (Beurton-Aimar et al., 2011; Colombié et al., 2015) dedicated to breakdown and transformation of extracellular nutriments to produce energy and metabolic precursors (amino acids, proteins, cell wall, . . .) and the secondary metabolic pathway to produce the main polyphenols (anthocyanins, flavonols, tannins, and stilbenes). This network of reactions (schematized in Supplementary Figure 5, and the list of the stoichiometric reactions in Supplementary Table 1) includes the glycolysis, the tricarboxylic acid cycle (TCA), the pentose phosphate pathway, starch metabolism, and sucrose metabolism. The carbon source was described through glucose and fructose uptake (Vglc-up, Vfru-up). The inorganic nitrogen source was nitrate (Vno3-up) involving enzymes of the nitrogen assimilation pathway [nitrate reductase (Vnr), glutamine synthetase (Vgs), and glutamate synthase (Vgogat)]. Ammonium, with a low concentration (2 mM), was neglected as nitrogen source. For the phenolic pathway, three reversible reactions involving naringenin, dihydroquercetin, and leucocyanidin (Vnar, Vdhq, and Vlcc) and two irreversible reactions involving cinnamate and coumaroyl coenzyme A (Vpal and Vcoum) were connected to central metabolism via phenylalanine. The fluxes directed toward the main phenolic compounds, i.e., anthocyanins, flavonols, tannins, and stilbenes were described by four overall reactions (Vanthoc, Vflav, Vtannins, and Vstilb, respectively). The main biosynthetic processes were described with overall reactions: (1) cell wall polysaccharides from UDP-glucose (Vcw), (2) protein synthesis (Vprotein) according to the measured amino acid composition of proteins (Supplementary Table 2), (3) fatty acids synthesis (diacyl glycerol, Vdag) from pyruvate and trioses phosphate according to total fatty acid biomass measurement (Supplementary Table 3), and (4) nucleotides synthesis (DNA and RNA, Vnucleotides) from ribose-5-phosphate by using plant metabolic pathway databases<sup>2</sup> . All other accumulated compounds were described as a simple accumulation: (1) malate (Vac-mal), (2) soluble sugars, i.e., glucose (Vac-glc), fructose (Vac-fru), and sucrose (Vac-suc), and (3) four groups of free amino acids, glutamate (Vac-Glu), aspartate (Vac-Asp), alanine (Vac-Ala), and phenylalanine (Vac-Phe). It has been checked that no metabolites were excreted in the medium (data not shown).

<sup>2</sup>http://www.plantcyc.org

Energy intermediates, both ATP and NAD(P)H, were explicitly taken into account. The cofactors NADP/NADPH were linked to biomass and the phenolic compounds production, and the cofactors NAD/NADH and FAD/FADH were linked to ATP synthesis via two essential reactions of oxidative phosphorylation (Vnrj1 and Vnrj2), which are associated to the mitochondrial respiration. Recycling of AMP by adenylate kinase is described by Vadk. The portion of synthesized ATP that is not used for growth has been balanced by the model as an ATP hydrolyzing reaction (Vnga-ATPm) that physiologically represents cellular maintenance (Amthor, 2000). Finally, all the cofactors were defined as internal metabolites, which means that they were balanced, thus constraining the metabolic network not only through the carbon and nitrogen balance but also through the redox and energy status.

In summary, the model of the metabolic network describes the main growth components of the cell through a set of n reactions involving m metabolites whose mint were internal metabolites. At steady-state, the mass balance equation is expressed by

$$\frac{dX\_{\text{int}}}{dt} = NV = 0\tag{1}$$

With Xint the vector of mint internal metabolites, V = (vi) t i=1...n the flux vector composed by the rates of n reactions of the network, and N = (nij) <sup>i</sup>=1...mint, <sup>j</sup>=1...<sup>n</sup> the stoichiometry matrix where nij is the stoichiometric coefficient of metabolite xint,<sup>i</sup> in reaction j. To solve the system, a lower and an upper bound constrained each flux.

#### Constraints Limiting the Flux Space and Resolution

The first type of constraints applied to limit the flux space to flux directions was inferred from thermodynamic properties of reversibility or irreversibility. Thus, among the internal reactions of the metabolic network, 33 were irreversible as indicated by unidirectional arrows on Supplementary Figure 5, which meant that their lower bounds were set to zero.

The second type of constraints was the maximal enzyme capacities. Experimentally determined activities of enzymes of central metabolism and flavonoid pathway, considered as maximal enzyme capacities (converted in mmol g DW−<sup>1</sup> day−<sup>1</sup> , Supplementary Table 4) were used to limit each corresponding flux in the metabolic network. The same values, but negative, were used as lower bounds of reversible enzymes. When the capacity of a given enzyme was not known, the bounds were set to infinity.

The third type of constraint concerned the respiration rate (see Section "Materials and Methods"). The sum of the two reactions of ATP synthesis by oxidative phosphorylation (Vnrj1 and Vnrj2) was constrained by the respiration measurements: 3.54 ± 0.18 and 2.86 ± 0.24 mmol g DW−<sup>1</sup> day−<sup>1</sup> in control and low nitrogen conditions at day 4 of culture, respectively.

Finally, the essential constraints required to set up the system were the external fluxes, also called exchange fluxes. Assuming steady state, these 16 fluxes (rates) were calculated from experimental data (Supplementary Table 5) and used as both lower and upper bounds. Also, respiration rates (Supplementary Table 6) were used to constrained ATP synthesis fluxes.

The mass balance of accumulated metabolites and biomass components covered an average of 81 and 91% of the dry biomass in control and nitrogen-limiting conditions, respectively (Supplementary Table 5). The accumulation of phenolic compounds in nitrogen-limiting condition was followed by an increase in sugar accumulation in cells at the expense of proteins synthesis and malate accumulation, especially at day 6 (**Figure 3**).

Flux minimization, which leads to a unique solution (Holzhutter, 2004), was used as the objective function to solve the system and generate flux maps in both N and N− conditions (**Figure 4**). Unsurprisingly, flux maps obtained with low nitrogen cultured cells compared to control condition showed higher fluxes in primary than in secondary metabolism both at 4 and 6 days of culture. Also a lower flux of ATP synthesis is pointed in low nitrogen condition.

More than the absolute values of the calculated fluxes, we were interested in the relative changes in the fluxes to look for cell metabolism reprograming under low nitrogen condition (Supplementary Table 7 and **Figure 5**). Concerning external fluxes (4th day) the main changes in nitrogen-limiting condition were the increase in accumulation of phenolic compounds (except flavonols), hexoses and starch (**Figure 5A**). Conversely protein synthesis, sucrose accumulation, and respiration were decreased. Consequently, the calculated fluxes in the main pathways (glycolysis, TCA, PPP. . .) were decreased of about 20 to 30%, except the fluxes of the phenolic pathway which were increased: Vmacl, Vshik, Vpal, and Vcoum by 38% and, Vnar, Vdhq, and Vlcc by 26%. More surprisingly, two internal fluxes, the PPi-dependent phosphofructokinase (Vpfp) and the pyruvate kinase (Vpk) were strongly increased (80 and 63%), and also glucose and fructose uptakes, were also slightly increased (7 and 6%) at day 4. Finally, fluxes corresponding to enzymatic steps of the flavonoid biosynthesis that use α-ketoglutarate (α-KG) as a reducing agent and convert it to succinate, namely Vdhq and Vanthoc were increased by 25 and 104%, also at day 4. Conversely, α-KG conversion to succinate the TCA cycle (Vkgdh) was decreased by 25% in low nitrogen culture conditions.

The global behavior of external fluxes was exacerbated at day 6 and resulted in similar observations than at day 4, i.e., a global diminution of all fluxes (40–50%) but here without significant change in Vpfp and Vpk and sugar uptake (**Figure 5B** and Supplementary Table 7). Changes in fluxes for Vdhq and Vanthoc were further enhanced by low nitrogen culture conditions, compared to control, with an increase of 129 and 518%, respectively. A third α-KG-dependent flavonoid biosynthetic flux was also strongly enhanced by 320%. TCA-linked metabolic flux that converts α-KG into succinate (Vkgdh) was reduced by 50%.

The internal metabolite concentrations were not accessible with the flux-balance model. Then complementary analyses have been done to determine the total contents of redox metabolism coenzymes (NAD+, NADH, NADP+, and NADPH). While NADH was slightly affected, NADPH significantly increased in low nitrogen condition compared to the control (Supplementary Table 8). Thus, the NADP+/NADPH ratio was significantly

lower in low nitrogen condition at day 4 (**Figure 6**). The same trend was observed at day 6, but was not deemed statistically significant according to Student's t-test (**Figure 6**). These results clearly showed an excess of NADPH, concomitant with the accumulation of anthocyanin (at day 4 and 6) and stilbene compounds (at day 4).

# DISCUSSION

#### Low Nitrogen Stimulates Anthocyanin Biosynthesis in GT3 Grapevine Cells

Under nitrogen limitation, an increase of the total anthocyanin content, especially peonidin and petunidin derivatives was noted. Moreover, the low phenylalanine concentration in cells cultivated in limiting nitrogen condition is in agreement with an increase in the phenylpropanoid catabolic flux, supported by the increase in PAL activity. Stimulation of anthocyanin biosynthesis in vineyard-grown grape berries by low nitrogen availability has been well documented in the literature, with an increase of around 30% in berry anthocyanin content (see for example Keller and Hrazdina, 1998; Hilbert et al., 2003; Soubeyrand et al., 2014). Similar studies using grapevine cell suspensions are much more scarce, however. In our experiments, GT3 grapevine cell suspension responded to low nitrogen in the culture media by a c.a. 700%, in average, increase total anthocyanin accumulation. This result is in the same order of magnitude as the results previously obtained on strawberry (Mori and Sakurai, 1994), or Gamay Fréaux grapevine (Do and Cormier, 1991), thus validating the GT3 cell suspension culture used in this work for acquiring the dataset that allowed us to perform FBA modeling.

intensity.

# Building a FBA Model That Links Central Primary Metabolism and the Polyphenol Secondary Metabolism

The metabolic model utilized in this work was sufficiently detailed to describe the global functioning of the cell. The originality of this work was to couple both primary and secondary metabolism, including the flavonoid biosynthetic pathway. As far as we know, in plant science only few models take into account secondary metabolic pathways. A genome-scale metabolic model of maize has already been reconstructed (Saha et al., 2011). Bekaert et al. (2012) described their updated mathematical model of Arabidopsis thaliana Columbia metabolism, which adds the glucosinolates, an important group of secondary metabolites, to the reactions of primary metabolism. In a recent review, Collakova et al. (2012) showed that metabolism can be modeled mathematically by using models and genomescale models (GEMs) predicting the combination of flux values of a defined metabolic network given the influence of internal and external signals. Nevertheless plant GEMs tend to be accurate in predicting only qualitative changes in selected aspects of central carbon metabolism, while secondary metabolism is largely neglected mainly due to the missing (unknown) genes and metabolites. As such, these models are suitable for exploring metabolism in simplified models such cell cultures in plants grown in favorable (controlled) conditions, but not in field-grown plants that have to cope with environmental changes in complex ecosystems (Collakova et al., 2012).

## Cell Energy and Reducing Power as a Driver for Anthocyanin Biosynthesis in Grape Cells

The question of the existence of a metabolic driver that would fuel the increase in anthocyanin (and more generally in flavonoid) biosynthesis in such situation remains open. One emerging property of the FBA-generated flux maps is the fact that in low nitrogen conditions (N−), several enzymatic steps that consume ATP and reducing power (NADPH or NADH) have their metabolic flux either maintained or increased (i.e., starch synthase, phosphofructokinase, enzymes of the pentosephosphate pathway, all the enzymes of the phenylpropanoid and flavonoid biosynthetic pathway, as well as the stilbene biosynthetic pathway). Conversely, most of the metabolic fluxes that lead to ATP, NADH, or NADPH formation were decreased by low nitrogen conditions (i.e., phosphoglycerate kinase and pyruvate kinase in the lower part of the glycolytic pathway, the

p < 0.05.

malic enzyme, most of the TCA cycle enzymes with noticeable exception of the malate dehydrogenase which has its metabolic flux slightly increased). This strongly advocates for a link between cell energy status (i.e., excess of ATP and reducing power) and secondary metabolism, confirming an hypothesis made by Hernández and Van Breusegem (2010). Recently, redox-dependent modulation of the anthocyanin pathway has been reported in Arabidopsis leaves during exposure to high light intensity (Page et al., 2011; Viola et al., 2016), or in Citrus callus (Cao et al., 2015). FBA modeling results strongly support that hypothesis, as well as actual NADP+/NADPH ratio measurements, pointing toward ATP and NADPH excess as a metabolic driver for flavonoid (and particularly anthocyanin) biosynthesis in grapevine GT3 cells. In the same review from Hernández and Van Breusegem (2010) also hypothesized that flavonoid biosynthesis could also constitute a carbon sink in situations of high carbon-to-nutrient ratio. Indeed in leaves from plants such as Rosemary or Tea trees, flux analysis suggests that up 20% of the fixed carbon would flow through the phenylpropanoid pathway, leading to a phenolic content accounting for up to 30% of dry matter, making it the main non-structural carbon sink of the plant, and thus an efficient mechanism to deal with carbon excess, without mobilizing any nitrogen (Hernández et al., 2004; Rippert et al., 2004; Yao et al., 2005). In the case of GT3 grape cells, however, model flux calculations and biomass composition analysis demonstrated that anthocyanins, and more broadly flavonoids, represent only a marginal storage sink for non-structural carbon (0.49 and 1.5% of total dry matter, at day 4 and 6, respectively, in N− condition), ruling out a role of anthocyanin (and more broadly flavonoids) biosynthesis as an effective mechanism for coping with carbon excess in high carbon to nitrogen ratio situations. Instead, FBA model output and biomass analysis indicate that carbon excess is dealt with by diverting embolic flux to vacuolar storage of soluble sugars (hexose and to a lesser extent sucrose) and malic acid. This discrepancy could be linked to fact that cell suspensions and whole organs such as leaves obviously differ in their behavior in term of carbon management. Leaves can act both as source and sinks for carbon, whether cultured cells only acts as carbon sink. The comparison is thus limited, but nevertheless points out two potentially different strategies for leaves and grape cells to cope with carbon excess.

#### α-Ketoglutarate Levels as a Potential Regulator of Anthocyanin Biosynthesis in Grape Cells

Besides the fact that low nitrogen culture conditions might lead to an altered cell energy status (i.e., an excess of ATP and NADPH), another output of the FBA-generated flux maps is that three fluxes of the flavonoid pathway that use α-KG were strongly up-regulated in low nitrogen cultured cells. α-KG has emerged in the past decade as a signal molecule in plants, linking TCA cycle to secondary metabolism, including the flavonoid pathway (Araújo et al., 2014). Indeed, three enzymatic steps of the flavonoid pathway use α-KG as reducing agent in their catalytic cycle: the flavanone 3β-hydroxylase, the flavonol synthase and the anthocyanidin synthase (Turnbull et al., 2004). Under low nitrogen culture conditions, consumption of α-KG by GOGAT for glutamate synthesis is bound to decreased. This is advocated by model output that predicts a diminution of 21 and 50% at day 4 and 6, respectively, potentially leading to an increase of cell α-KG level, which would be used to fuel anthocyanin and more generally flavonoid biosynthesis. Thus, α-KG availability would be part of the metabolic driver that lead to enhanced flavonoid biosynthesis high carbon-to-nitrogen ratio conditions. Actual α-KG level measurements would be required to further advocate this hypothesis.

#### CONCLUSION

Flux balance analysis modeling was used to investigate metabolic flux reprogramming in grapevine cells in response to low nitrogen culture conditions and revise the well-known upregulation of anthocyanin biosynthesis in response to low

nitrogen availability. Model outputs unambiguously point toward cell energy excess and increased α-KG availability as the metabolic drivers of anthocyanin synthesis (and more broadly flavonoid synthesis) under high carbon-tonitrogen ratio conditions. This work was conducted in a cell suspension culture, and the next obvious question is whether such a metabolic driver effect is also occurring in ripening berries of red grape varieties, which accumulate anthocyanins to high levels in their exocarp cells, a key feature for high quality red wine making. Further modeling and biochemical work is needed to address that question.

## AUTHOR CONTRIBUTIONS

ES, GH, CR, StC, LM-P, and BB performed the experiments and the analytical work. SoC performed model construction and calculations, participated to data analysis and manuscript writing. MD-N generated the flux maps. YG, J-MM, ZD, and SD discussed the results and performed manuscript critical reading. EG led the project and designed the experimental flowchart, discussed the results and coordinated the manuscript writing and critical reading.

#### REFERENCES


#### FUNDING

ES was supported by a Ph.D. grant from the Ministère de l'Éducation Nationale, de l'Enseignement Supérieur et de la Recherche and the work was partially funded by a grant from the FR BIE (Fédération de Recherche Biologie Intégrative et Ecologie) of University of Bordeaux.

#### ACKNOWLEDGMENTS

We thank the Metabolome Facility of Bordeaux Functional Genomics Centre. We are indebted to the BIBS Facility [IBISA/BioGenOuest Biopolymers, Interactions, Structural Biology platform (BIBS), UR 1268 BIA, INRA Angers-Nantes, F-44300 Nantes, France] for cell wall analyses.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2018.00421/ full#supplementary-material

flavonoid/anthocyanin accumulation in citrus. BMC Plant Biol. 15:27. doi: 10. 1186/s12870-015-0426-4




Yao, L., Caffin, N., D'arcy, B., Jiang, J., Shi, R., Singanusong, X., et al. (2005). Seasonal variations of phenolic compounds in Australia-grown Tea (Camellia sinensis). J. Agric. Food Chem. 53, 6477–6483. doi: 10.1021/jf050382y

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Soubeyrand, Colombié, Beauvoit, Dai, Cluzet, Hilbert, Renaud, Maneta-Peyret, Dieuaide-Noubhani, Mérillon, Gibon, Delrot and Gomès. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Dissecting the Variations of Ripening Progression and Flavonoid Metabolism in Grape Berries Grown under Double Cropping System

Wei-Kai Chen1,2, Xian-Jin Bai<sup>3</sup> , Mu-Ming Cao<sup>4</sup> , Guo Cheng<sup>4</sup> , Xiong-Jun Cao<sup>4</sup> , Rong-Rong Guo<sup>5</sup> , Yu Wang1,2, Lei He1,2, Xiao-Hui Yang1,2, Fei He1,2, Chang-Qing Duan1,2 and Jun Wang1,2 \*

<sup>1</sup> Center for Viticulture and Enology, College of Food Science and Nutritional Engineering, China Agricultural University, Beijing, China, <sup>2</sup> Key Laboratory of Viticulture and Enology, Ministry of Agriculture, Beijing, China, <sup>3</sup> Guangxi Academy of Agricultural Sciences, Nanning, China, <sup>4</sup> Grape and Wine Research Institute, Guangxi Academy of Agricultural Sciences, Nanning, China, <sup>5</sup> Guangxi Crop Genetic Improvement and Biotechnology Laboratory, Nanning, China

#### Edited by:

Soren K. Rasmussen, University of Copenhagen, Denmark

#### Reviewed by:

José Tomás Matus, Universitat Autònoma de Barcelona, Spain Chonghuai Liu, Zhengzhou Fruit Research Institute (CAAS), China

> \*Correspondence: Jun Wang jun\_wang@cau.edu.cn

#### Specialty section:

This article was submitted to Plant Breeding, a section of the journal Frontiers in Plant Science

Received: 05 June 2017 Accepted: 23 October 2017 Published: 10 November 2017

#### Citation:

Chen W-K, Bai X-J, Cao M-M, Cheng G, Cao X-J, Guo R-R, Wang Y, He L, Yang X-H, He F, Duan C-Q and Wang J (2017) Dissecting the Variations of Ripening Progression and Flavonoid Metabolism in Grape Berries Grown under Double Cropping System. Front. Plant Sci. 8:1912. doi: 10.3389/fpls.2017.01912 A double cropping system has been commercially adopted in southern China, where there is abundant sunshine and heat resources. In this viticulture system, the first growing season normally starts as a summer cropping cycle; then, the vine is pruned and forced, resulting in a second crop in winter. Due to climate differences between the summer and winter growing seasons, grape ripening progression and flavonoid metabolism vary greatly. Here, the metabolites and transcriptome of flavonoid pathways were analyzed in grapes grown under two growing seasons at different stages. Notably, the winter cropping cycle strongly increased flavonoid levels by several times in comparison to summer grapes, while the summer season took a major toll on anthocyanin and flavonol accumulation, since the winter cropping greatly triggered the expression of upstream genes in the flavonoid pathway in a coordinated expression pattern. Moreover, the ratio of VviF305 <sup>0</sup>Hs (flavonoid 305 0 -hydroxylase) to VviF30Hs (flavonoid 3<sup>0</sup> -hydroxylase) transcript levels correlated remarkably well with the ratio of 3 05 0 -substituted to 3<sup>0</sup> -substituted flavonoids, which was presumed to control the flux of intermediates into different flavonoid branches. On the other hand, the phenological phase also varied greatly in the two crops. Compared to summer cropping, winter growing season accelerated the duration from budburst to veraison, therefore advancing the onset of ripening, but also prolonging the duration of ripening progression due to the purposes to harvest high-quality grapes. The differential expression pattern of hormonerelated genes between the two cropping cycles might explain this phenomenon.

Keywords: transcriptome, secondary metabolism, subtropical climate, double cropping system, Vitis vinifera, flavonoid biosynthesis, ripening

# INTRODUCTION

The growing season of grapevine is the time of year during which local climatic conditions (i.e., temperature, sunlight, and rainfall) permit its normal growth. In most temperate regions, grapevines undergo dormancy from late fall to early spring, and a single pruning and harvest is the conventional grapevine practice. However, in southern China, which shows extremely high

**137**

temperature and concentrated rainfall during July and August, this traditional cultivation pattern does not adapt to local climate, and it takes a major toll on crop and grape quality, since these rainy months increase the occurrence of fungal diseases (Bai et al., 2008). In conventional practices, the compound buds remain in a stationary state during the current growing season, and they break dormancy in the following spring (Lavee and May, 1997). However, if these compound buds are forced out of dormancy early during the current season, a double cropping viticulture system can be achieved (Lin et al., 1985; Bai et al., 2008; Gu et al., 2012). Two crops of table grapes per year were achieved in Taiwan by a combination of pruning, defoliation, and chemical treatments (Lin et al., 1985). Similarly, in the Brazilian Southeast, the making of good wines was also attained by a double pruning approach (Favero et al., 2011). Therefore, the farmers in southern China introduced this strategy to overcome all detrimental environment under conventional practices.

With the benefit of high total yield and good quality, the double cropping system has great potential in subtropical viticulture regions (Lin et al., 1985). Recently, the winter grapes in Brazil showed physicochemical characteristics more suitable than those from the summer growing season for winemaking purposes, since the summer grapes were featured by higher cluster weight and titratable acidity while the winter crops were characterized by higher total soluble solids (TSS) content and pH value (Júnior et al., 2017). This viticulture system (**Figure 1**) has been commercially adopted in southern China, in which the first growing season normally starts in March to July; then, the vine is pruned and forced by a cyanamide solution in August, resulting in a second bud break in August, and a second crop until January of the following year (Lin et al., 1985; Bai et al., 2008). In some vineyards, the purpose of the first cropping is not to harvest grapes in summer, but to have the inflorescence primordia differentiated into latent buds (Dias et al., 2017). The second cropping resembles the extending of fruit ripening from summer to the autumn-winter of the growing season (Gu et al., 2012). This double cropping system could minimize the negative impact of a rainy and hot climate with the help of rain-shelter treatment, and it could maximize the use of sunshine and heat resources in subtropical regions.

Flavonoids are a group of natural compounds that share a common polyphenolic flavan skeleton. The biosynthetic pathways for anthocyanins, flavonols, and flavan-3-ols in plants share a common upstream step through phenylalanine ammonia-lyase (PAL) to flavanone 3-hydroxylase (F3H). Flavonol synthase (FLS) catalyzes dihydroflavonols to form their corresponding flavonol aglycones, and leads the flavonoid flux into the flavonol branch. On the other hand, leucoanthocyanidins and anthocyanidins can be converted into flavan-3-ols by leucoanthocyanidin reductase (LAR) and anthocyanidin reductase (ANR), respectively. As for anthocyanin synthesis, UDP-glucose: flavonoid 3-O-glucosyltransferase (UFGT) is considered as the key enzyme that determines the formation of anthocyanidin-3-O-glucosides. Due to internal and external differences between grapes, the flux through the flavonoid pathway toward downstream anthocyanins, flavonols, and flavan-3-ols varies significantly (Castellarin et al., 2007; Mouradov and Spangenberg, 2014). The climate of Nanning County in southern China is characterized by long growing seasons with higher temperature months from June to September. Many studies have shown that the flavonoid compounds of grape berries are greatly influenced by climate conditions (reviewed by Downey et al., 2006). High temperature often increases the degradation of flavonoid compounds, resulting in poor coloration (Yamane et al., 2006; Mori et al., 2007; Gu et al., 2012). Due to climate differences between the summer and winter growing seasons, especially the reverse temperature evolution pattern during grape berry development, flavonoid composition and content are often distinct between the two crops (Xu et al., 2011; Zhu et al., 2017). It is understandable that these variances in metabolites have much to do with the transcriptomic difference of grapes between the two growing seasons.

Since the first report of the draft genome sequence of the Pinot Noir grapevine (Jaillon et al., 2007), genomewide transcriptome research revealing the regulation of berry ripening and its associated metabolic processes has become a hot topic (Fabres et al., 2017; Wong and Matus, 2017). Transcriptomic analyses also provide a comprehensive approach to study the transcriptional responses of grapes to changing environment conditions (Sun et al., 2017). Water deficit activates the expression of phenylpropanoid pathway transcripts, which increases flavonoid content in wine grape (Grimplet et al., 2007; Deluc et al., 2009; Savoi et al., 2016). Solar ultraviolet radiation triggers regulatory responses through the UV-B radiation-specific signaling pathway, which also results in the activation of phenylpropanoid biosynthesis (Carbonell-Bejerano et al., 2014). The transcriptional expression patterns of flavonoid biosynthesis in wine grapes grown in two regions with distinct climates also showed regional differences (Li et al., 2014). Furthermore, several mRNA expression profiling studies have been reported that show a detailed analysis of gene expression during grape berry development (Deluc et al., 2007, 2009; Sweetman et al., 2012; Degu et al., 2014). However, to the best of our knowledge, the transcriptomic analysis of seasonal variation on grape flavonoid compounds has not been reported within a double cropping system.

Thus, in the present study, we performed transcriptome and metabolite analysis of flavonoid biosynthesis both in summerand winter-grapes of Vitis vinifera L. cv. "Cabernet Sauvignon" (CS) and V. vinifera cv. "Riesling" by RNA-sequencing and HPLC-ESI-MS/MS. Comparative analysis of the transcript and metabolite profiles revealed season- or cultivar-specific patterns of flavonoid biosynthesis. This paper provides insights into the mechanisms of growing season influence on flavonoid accumulation.

#### MATERIALS AND METHODS

### Experimental Vineyard and Double Cropping Viticulture Practices

The experiment was conducted in 2014 and 2015 on 7-year-old grapevines grafted onto SO4 rootstock in the vineyards of the

Guangxi Academy of Agricultural Sciences, located in southern China (22◦ 360 6400N, 108◦ 140 1300E, altitude 104 m), where it is typically a subtropical humid monsoon climate with abundant sunshine and heat resources. The vines were planted in east-west rows under a rain-shelter treatment with an inter- and intra-row vine spacing of 3.5 m × 1.5 m and were managed in a closing Y-shaped training system with 2 × 4/5 shoots per meter and 1.0 m cordon above ground. Two widely planted wine grape cultivars, "CS" and "Riesling," were selected for research.

The grape viticulture regime in the subtropical region is dominated by double cropping systems (Lin et al., 1985; Bai et al., 2008). In these cultivation practices, the vines are pruned twice and the grapes are harvested twice per year. To produce the first crop, the vines were pruned and enforced with 2.5–3.0% hydrogen cyanamide in mid-February when the temperature was maintained above 10◦C. The terminal bud was not sprayed to avoid apical dominance. In addition, the soil was kept moist to accelerate germination. Then, the grapevines were in full bloom around mid-April, while the veraison stage ranged from mid-June to early July, followed by the harvest stage (summer grape). This whole period was termed the summer cropping cycle. Then, the vines were pruned and forced again in August, leading to the second crop (winter grape) in early January of the following year. In detail, the grapevines were pruned on approximately 20th August, were manually defoliated, and an average of 5–10 buds were left for each cane. Hydrogen cyanamide was smeared only at the terminal bud and in the vicinity of the pruning-wound surface. Then, the bud burst 5–8 days later and initiated the winter cropping cycle.

Grape berries in three biological replicates were collected at four E-L stages (Coombe, 1995) according to berry color and TSS (◦Brix) for each crop as follows: pea-size berries (E-L 31), the onset of veraison (E-L 35), the end of veraison (E-L 36), and the harvest stage (E-L 38). In brief, 300 berries were selected from both the sunny and shady sides of at least 50 whole vine selections, among which 100-berry sub-samples were processed immediately to determine the physicochemical parameters. The remaining samples were frozen in liquid nitrogen and stored at −80◦C for subsequent RNA and flavonoid extraction. TSS of the juices was determined with digital pocket handheld refractometer (PAL-1, Atago, Japan), and titratable acidity was measured by acid–base titration.

# Extraction of Flavonoid Compounds

The fresh skins were peeled from the grapes and were immediately ground into powder in liquid nitrogen. Afterward, the skin powder was lyophilized at −50◦C and used for extraction of anthocyanin, flavan-3-ol, and flavonol. The anthocyanins and flavonols were simultaneously extracted in two analytical replicates according to a previous report (Downey et al., 2007). Three aliquots of grape skin powder (0.10 g) were immersed in 1.0 ml of 50% methanol in water, were ultrasound sonicated for 20 min, and were centrifuged. Then, the supernatant was collected and the residues were re-extracted again. All the supernatants were mixed, filtered through a 0.22-µm nylon membrane, and transferred to HPLC auto-sampler vials. The extraction of flavan-3-ol was also conducted twice for each sample according to method described by Liang et al. (2012). Grape skin powder (0.10 g) was mixed with 1 ml of phloroglucinol buffer (0.5% ascorbate, 300 mM HCl, and 50 g/l phloroglucinol in methanol), incubated at 50◦C for 20 min, neutralized with 1 ml of sodium acetate (200 mM, pH 7.5), and finally centrifuged at 8000 × g for 15 min. This procedure was repeated three times, and the supernatants were combined and filtered for HPLC analysis.

#### Analysis of Flavonoid Compounds

Analysis of flavonoid compounds was carried out using an Agilent 1200 Series HPLC–MSD trap VL equipped with a variable wavelength detector (for flavonol) or a diode array detector (for flavan-3-ol and anthocyanin). The mass spectrometric acquisition parameters were as follows: ESI interface, positive ion mode (for anthocyanin) or negative ion mode (for flavonol and flavan-3-ol), 35 psi nebulizer pressure, 10 ml/min drying N<sup>2</sup> flow rate, 350◦C drying N<sup>2</sup> temperature, capillary voltage 3000 V, and scans at m/z 100–1000.

Anthocyanin extract was injected onto a Kromasil C18 column (250 mm × 4.6 mm, 5 µm). The mobile phases A and B were aqueous 2% formic acid and acetonitrile containing 2% formic acid, respectively. The flow rate was 1.0 ml/min, and the solvent gradients were as follows: from 6 to 10% B over 4 min, from 10 to 25% B over 8 min, isocratic 25% B for 1 min, from 25 to 40% B over 7 min, from 40 to 60% B over 15 min, from 60 to 100% B over 5 min, and from 100 to 6% B over 5 min. Other conditions were as follows: injection volume, 30 µl; detection wavelength, 525 nm; and column temperature, 50◦C.

Flavan-3-ol compounds were separated on a reversed phase Zorbax SB-C18 column (250 mm × 4.6 mm, 5 µm) using mobile phase A of aqueous 0.2% acetic acid and mobile phase B of acetonitrile: 0.2% acetic acid (4:1) at a flow rate of 1 ml/min and were monitored at 280 nm at 25◦C. The elution gradients of solvent B were as follows: 0 min, 10% B; 20 min, 10% B; 30 min, 15% B; 40 min, 20% B; 50 min, 33% B; 55 min, 40% B; 58 min, 100% B; 63 min, 100% B; and 64 min, 10% B.

For flavonol separation, mobile phase A was a mixture of formic acid:acetonitrile:water (85:50:865) and B was formic acid:acetonitrile:water:methanol (85:250:215:450). The column selected was a Zorbax SB-C18 (4.6 mm × 250 mm, 5 µm), with the temperature maintained at 40◦C. The gradient conditions were as follows: 0% B over 7 min; 24.2 min, 14.2% B; 27 min, 15.7% B; 39 min, 23.5% B; 45 min, 26% B; 51.6 min, 32% B; 61.8 min, 40% B; 62.3 min, 60% B; 67.8 min, 100% B; and 78.4 min, 0% B. The flow rate was 0.63 ml/min, and the detector wavelength was 360 nm.

# Transcriptome Sequencing and Data Analysis

Three biological replicates for each sample were performed. A sub-sample of 50 berries were randomly selected from each biological replicate for RNA extraction. Total RNA was extracted from the frozen deseeded berries (whole pericarp) at three development stages (E-L 35, 36, and 38) using a SpectrumTM Plant Total RNA Kit (Sigma–Aldrich, Carlsbad, CA, United States) to conduct transcriptome analysis on the Illumina HiSeqTM 2000 platform with 50-bp single reads and were then aligned against the reference grapevine genome 12×V2, allowing no more than two mismatches. Transcriptome de novo assembly was conducted using the short reads assembling program Trinity with a fixed k-mer length of 25. To determine gene expression levels, the longest transcript was chosen to calculate the fragments per kilobases per million reads (FPKM) value when more than one transcript was obtained for a single gene (Mortazavi et al., 2008). For the functional annotation, unigene sequences were aligned to databases as described previously (Li et al., 2014). Differentially expressed genes (DEGs) between the samples were identified by the R package called "DESeq2." A false discovery rate ≤0.01 and a fold change ≥2 were set as the threshold to judge the significance of gene expression differences. Gene Ontology (GO) and KEGG enrichment analysis of DEGs was used to select candidate genes. The data have been deposited in the NCBI Gene Expression Omnibus (GEO) database and are accessible through GEO accession GSE103226.

# Statistical Analysis

Heatmap visualizations were performed using the R package "pheatmap." Principal component (PC) analysis was done using MetaboAnalyst 3.0<sup>1</sup> . A one-way analysis of variance (ANOVA) was used to measure the differences between flavonoid contents employing Duncan's multiple range tests at p < 0.05. The column plots were prepared using OriginPro 9.2 (OriginLab Corporation, Northampton, MA, United States).

# RESULTS

### Meteorological Data and Phenological Characteristics

The environmental condition in southern China corresponds to a typically subtropical humid monsoon climate, which is characterized by a hot and humid summer and a mild to cool winter (Supplementary Table S1). The climatic conditions displayed great differences between the two growing seasons (**Figure 2**). Basically, the daily temperature showed reverse

<sup>1</sup>http://www.metaboanalyst.ca/faces/home.xhtml

evolution patterns in the two growing seasons from budbreak to harvest. In the summer cropping cycle, the temperature was low at an early stage, then it increased gradually until the veraison and ripening stages. The mean temperature in the summer season approached 30◦C, and the daily maximum temperature frequently exceeded 35◦C, which has been considered as detrimental to plant growth and anthocyanin accumulation (Cohen et al., 2012; de Rosas et al., 2017). While in the second cropping cycle, the temperature dropped from the point of flowering until the ripening process, with a mean temperature of approximately 20◦C during the entire winter cropping. In addition, the extremely high temperature hours in the winter season during grape berry ripening was almost negligible, only

2 h versus 210 h in the summer. Another important difference in climate was rainfall. Southern China always experiences abundant and concentrated rainfall in the summer or, to a lesser extent, in the autumn. The heavy rainfall in June and July could greatly reduce grape quality, making the double cropping practice appealing, as this approach harvested double crops and shifted fruit ripening from the hot and rainy summer (July and August) to the mild and cool winter (January) in each growing season (Bai et al., 2008). However, the sunshine hours during berry development, ranging from fruit-set to harvest, were distinctly higher in summer cropping in both years, and this was also the case for the growing degrees days and photosynthetically active radiation. Diurnal temperature between day and night was

another important climatic index (Cohen et al., 2012), and it was 1–4◦C lower in winter cropping.

Due to climate difference in the two growing seasons, the phenological phase also varied greatly in the two crops (Supplementary Table S1). The entire period of berry development in winter cropping, from fruit-set to full-ripeness, was longer than or equal to that of the summer cycle, but the duration of the early phenological phase was hastened and the late progression was delayed. The winter cropping was characterized by shorter green and veraison stages, but a longer berry ripening stage. Compared to stage durations of cv. "CS" described in single cropped vineyards (Li et al., 2014), both crops showed shorter berry development periods partially due to higher temperature. The grapes of summer crop showed a longer veraison stage than that from single cropped vineyards, which was nearly equal to that in winter season. Comparing the ripening duration, the single cropped grapes ranged between the summer crop and winter crop herein (Li et al., 2014). With regard to whole duration from flowering to harvest, the years of 2014 and 2015 showed opposite results. The difference on the ripening duration in the winter cropping was behind this phenomenon, resulting from distinct differences in rainfall. It was noted that "CS" and "Riesling" demonstrated similar flowering, veraison, and harvest dates under the double cropping system.

To determine the impact of the cropping cycles on berry development and flavonoid metabolism, we collected berries at four development stages for each crop. The berries in summer cropping showed higher TSS at E-L 31, but showed an opposite result at E-L 38, and no significant difference was found at E-L 35 and E-L 36 (**Figure 3**). Berries sampled from both crops in both cultivars showed similar patterns of physicochemical characteristics during grape berry development (**Supplementary Figure S1**). Berry weight increased along with berry development, but the berries in the winter cropping were significantly smaller than those of the summer cropping at E-L 35, 36, and 38. The same phenomenon also occurred in berries from two separate years, with smaller berries in 2014 than that in 2015, which coincided with rainfall.

#### Impact of Cropping Season on Flavonoid Metabolites across Berry Development

The flavonoid compounds were analyzed to explore the effect of the double cropping system on grape berries in a subtropical region and to compare the metabolites with the transcriptome. A total of 14 and 8 flavonol glycosides were identified in "CS" and "Riesling," respectively, which corresponded to six types of free aglycone in "CS" and three types in "Riesling" (Supplementary Table S2). The content of total flavonols in both "CS" and "Riesling" was significantly higher in winter grapes than the content in summer grapes, except for a single point of E-L 31 of "Riesling" in 2015 (**Figure 4**), in agreement with the data reported by Zhu et al. (2017). The total flavonol content showed an increasing trend for "CS," but a declining trend for "Riesling." It has been suggested that kaempferol-type, quercetin-type, and isorhamnetin-type flavonols are generally present in both red and white grapes, while myricetin-type flavonol and its derivatives

are only accounted for in V. vinifera red grapes, which is in agreement with the present study (Castillo-Muñoz et al., 2007, 2010). For "CS" grapes, quercetin and myricetin were the most abundant flavonols, the proportion of which showed an inverse relationship during berry development. Quercetin accounted for more than 90% of the flavonol at stage E-L 31, while at E-L 38, its proportion reduced to 60–70% in summer grapes and 40–50% in winter grapes. Similarly, the proportion of glucoside and glucuronide also varied with the development stage and growing season. The green stage featured 3-O-glucuronides, while the harvest stage was characterized by 3-O-glucosides. In addition, the proportion of 3-O-glucosides at E-L 38 was 10% lower in winter grapes than in summer grapes. However, in "Riesling" grapes, the flavonol content was dominated by quercetin and 3- O-glucuronide, and their proportion showed no consistent trend versus the developmental stage.

The anthocyanins in "CS" displayed progressive accumulation in both crops across grape development. The content of total anthocyanins at E-L 38 in winter grapes, which was 2047.4 mg/kg FW in 2014 and 1682.3 mg/kg FW in 2015, was almost 10-fold higher than the content in summer grapes (**Figure 4**). In addition, the individual accumulation patterns of delphinidin-3- O-glucoside, cyanidin-3-O-glucoside, petunidin-3-O-glucoside, peonidin-3-O-glucoside, and malvidin-3-O-glucoside, and their acylated derivatives all showed similar trends (Supplementary Table S2). Malvidin-3-O-glucoside and its acylated derivatives accounted for 80% of the anthocyanins in summer grapes, while in winter grapes, the proportion of malvidin was significantly lower, dropping to 45% in 2014 and 60–70% in 2015. In addition, there was an increase in the percentages of delphinidin and petunidin in winter grapes in both years, but an increase in the proportion of cyanidin and peonidin occurred only

in 2014. Thus, the content of 3<sup>0</sup> 5 0 -substituted anthocyanins increased drastically in winter grapes relative to summer grapes, but its proportion sometimes decreased. Additionally, the ratios of methylated/non-methylated and acylated/non-acylated anthocyanins also showed significant differences between the two growing seasons, with higher methylation and acylation levels in summer grapes. In fact, the proportion of acylated anthocyanins was previously shown to greatly decrease with exposure to sunlight and high temperatures (Mori et al., 2007; Tarara et al., 2008).

Flavan-3-ols, the immediate competitors of the precursors for flavonol and anthocyanin synthesis, showed a mild reduction in total content as the stage progressed in the summer crop and maintained a relatively stable level in winter grapes, with the exception of "Riesling" in the 2015 winter crop, which displayed a transient peak at E-L 35, indicating the relatively limited effects of the cropping cycle on flavan-3-ols (**Figure 4**). The content of flavan-3-ols was higher in "CS" berries than "Riesling" berries, as well as flavonol. Comparing the total flavan-3-ol content between the two cropping cycles, the result in 2014 showed significant differences, but no significant difference was found in 2015. It seemed that the summer grapes had more flavanol at E-L 31, after which were reduced to lower levels than that in winter crops at later stages. Zhu et al. (2017) also found higher flavan-3-ol levels in winter grapes of cv. "Muscat Hamburg" at maturity, but no consistent trend was found in cv. "Khoyo." In regard to flavanol profiles, epicatechin, as expected, was the most abundant fraction in both varieties, and its proportion was slightly but significantly higher in winter grapes at E-L 38. Epicatechin-3-O-gallate was the second main constituent, and it accounted for approximately 20% of the total composition, but showed no detectable differences between the two growing seasons.

Dihydrokaempferol represents the branching node in the flavonoid pathway, and it converts to dihydroquercetin and dihydromyricetin with the catalysis of F30H (flavonoid 3 0 -hydroxylase) and F3<sup>0</sup> 5 <sup>0</sup>H (flavonoid 3<sup>0</sup> 5 0 -hydroxylase), respectively, giving rise to the 3<sup>0</sup> -substituted and 3<sup>0</sup> 5 0 -substituted flavonoid compounds (Degu et al., 2014). Since almost no 3 0 5 0 -substituted flavonoids existed in "Riesling" grapes, we only analyzed the proportion of 3<sup>0</sup> -hydroxylated and 3<sup>0</sup> 5 0 hydroxylated flavonoids in "CS" (**Figure 4**). At E-L 31, there was no significant difference among the four groups in the proportion of 3<sup>0</sup> -substituted flavonoids (approximately 85%). Then, their proportion gradually decreased and displayed significant differences at E-L 38, approaching 65–80% in summer grapes and 35–50% in winter grapes at E-L 38, due to an abundant accumulation of 3<sup>0</sup> 5 0 -substituted anthocyanins and flavonols.

# Transcriptomic Changes of Flavonoid Pathway under Double Cropping System

The genetic control of the flavonoid pathway is well known, but the mechanisms of seasonal influence in the double cropping system are poorly understood. Therefore, one of the goals in this study was to explore the molecular changes in the berries under different cropping cycles and to correlate these changes with

FIGURE 5 | Principal component analysis of the whole normalized gene expression dataset. Four different colors (one per each type of sample: CS-W, CS-S, R-W, R-S) and then for each color three different tones (35, 36, and 38 EL stages) were used to represent different samples. Ellipses encircle the three replicates of each sample subjected to the same stage. CS/R, "Cabernet Sauvignon"/"Riesling"; S/W, summer/winter crop.

metabolite accumulation. Berries of three selected developmental stages (E-L 35, 36, and 38) from summer and winter cropping in 2014 were chosen to perform transcriptome analyses by RNA-Seq. PC analysis for the whole normalized gene expression was performed and most three biological replicates of each sample were well-grouped (**Figure 5**). PC1 explained 25.4% of the total variance in gene expression and separated the ripening stage of winter crop from both green and veraison stages of "Riesling" sample. PC2 explained 15.7% of the total variance and clearly separated "CS" at ripening stage of winter crop from other "CS" samples, and separation between two varieties can also be distinguished on PC2 (**Figure 5**). The expression patterns of the transcripts involved in the phenylpropanoid and flavonoid pathways were depicted as FPKM values across three developmental stages in Supplementary Table S3. In addition, the influences of cropping season on the expression of flavonoid genes were expressed as the log2-fold change of the transcript abundance in winter cropping compared to summer cropping (**Figure 6**). A subset of DEGs participating in multiple branches of flavonoid metabolism was identified in two varieties.

The early phenylpropanoid pathway, acting upstream of the flavonoid pathway branch, provides the precursors of p-coumaroyl-CoA for subsequent phenolic biosynthesis. The winter cropping displayed a distinct and significant upregulation of most upstream genes during ripening, among which one unigene of Vvi4CL (4-coumarate: CoA ligase, VIT\_206s0061 g00450) was downregulated in both cultivars. Interestingly, a

flavonoid pathway transcripts are depicted in dotted red rectangle. The complete data set can be accessed in Supplementary Table S3.

recent study conducted on red blotch-infected berries found 25 suppressed genes in phenylpropanoid metabolism, while this unique Vvi4CL was induced by the disease during berry ripening (Blanco-Ulate et al., 2017). It is noteworthy that p-coumaroyl-CoA is a branch point toward flavonoid and stilbene synthesis. In the present study, the transcripts encoding stilbene synthase (STS) were increased in "CS" during the ripening process, while they decreased in "Riesling" across the three developmental stages (Supplementary Table S3). The VviSTS unigenes were significantly modulated by the growing season, and their transcripts were distinctly induced as the berry developed from veraison to post-veraison in both cultivars, which coincided with the accumulation of stilbenes at the onset of ripening (Deluc et al., 2011). In parallel, the cropping season greatly triggered

the expression of upstream genes in the flavonoid pathway, the transcripts of which showed coordinated expression patterns with genes in the phenylpropanoid pathway, ensuring a sufficient quantity of precursors for the subsequent synthesis of flavonoid compounds.

The flavonoid metabolic pathway shares a common upstream route, and then the flux is diverted to 3<sup>0</sup> -hydroxylated or 3<sup>0</sup> 5 0 hydroxylated flavonoids separately via the enzyme F30H or F3<sup>0</sup> 5 <sup>0</sup>H. Similar with upstream genes, including two VviF3H (VIT\_204s0023g03370, VIT\_218s0001g14310), two VviF30H (VIT\_203s0063g01690, VIT\_209s0002g01090), and six VviF3<sup>0</sup> 5 <sup>0</sup>H (VIT\_206s0009g02810, VIT\_206s0009g02830, VIT\_206s0009g02840, VIT\_206s0009g02970, VIT\_206s0009 g03000, VIT\_206s0009g03010), were significantly upregulated around the veraison stage in the winter berries versus the summer berries. Of particular interest were the transcripts encoding the F3<sup>0</sup> 5 <sup>0</sup>H enzyme, which were mainly expressed in the red grapes of "CS." The transcript levels of VviF3<sup>0</sup> 5 0H were extremely low in "Riesling," suggesting that a different transcriptional regulation mechanism for VviF3<sup>0</sup> 5 <sup>0</sup>H existed in white and red varieties (Matus et al., 2017). The simultaneous upregulation of both VviF30H and VviF3<sup>0</sup> 5 <sup>0</sup>H transcripts in winter berries compared to summer berries of "CS" also modulated the relative abundance of the different flavonoid forms. In the current study, the ratio of VviF3<sup>0</sup> 5 <sup>0</sup>Hs to VviF30Hs levels in "CS" was significantly and distinctly higher in winter cropping berries during ripening, in parallel with their higher ratio of 3<sup>0</sup> 5 0 -substituted to 3<sup>0</sup> -substituted flavonoids, which was presumed to control the flavonoid composition of grape berries. Similarly, a previous study also found that the temporal and variety-specific expression of VviF30H and VviF3<sup>0</sup> 5 <sup>0</sup>H in grapes occurred in coordination with the accumulation of the respective hydroxylated metabolites (Bogs et al., 2006; Sun et al., 2016).

The downstream flux of flavonoid metabolism in grape berries included multiple branches, and the related genes showed divergent expression patterns across the three developmental stages in the two cropping cycles. Contrary to the coordinated upregulation of upstream genes, a subset of transcripts involved in the late biosynthetic pathway was significantly repressed at some stages, such as VviDFR (dihydroflavonol 4-reductase, VIT\_215s0048g01010, VIT\_219 s0014g04980, VIT\_202s0025g01260), VviFLS (VIT\_218s0001g 03430), VviLDOX (leucoanthocyanin dioxygenase, VIT\_208 s0105g00380), and VviGT5 (UDP-glucuronic acid: flavonol 3-Oglucuronosyltransferase, VIT\_211s0052g01600), whereas other transcripts, such as VviDFR (VIT\_216s0039g02350, VIT\_218 s0001g12800), VviFLS (VIT\_211s0118g00390, VIT\_218s000 1g03470), VviLDOX (VIT\_202s0025g04720), and VviLAR (VIT\_201s0011g02960, VIT\_217s0000g04150), were significantly upregulated in many cases. Of the five known VviFLS genes, the enzyme of which commits to flavonol biosynthesis, only two transcripts of VviFLS4 (VIT\_218s0001g03470) and VviFLS5 (VIT\_218s0001g03430) were expressed, and they showed differential expression patterns in the two cropping cycles (Fujita et al., 2006). In this study, the transcript level of VviFLS4 was extremely low, and it was only upregulated by winter cropping

at E-L 38 in "CS" and at E-L 35 in "Riesling." Genes encoding two previously characterized flavonol glycosyltransferases, VviGT5 and VviGT6 (UDP-glucose/UDP-galactose: flavonol-3- O-glucosyltransferase, VIT\_211s0052g01630), were co-expressed with VviFLS4, which is in agreement with a previous report (Malacarne et al., 2015). Czemmel et al. (2017) also demonstrated positive correlation of VviMYBF1 with novel genes of the flavonol pathway, VviGT3 (flavonol glycosyltransferase, VIT\_211s0052g01580) and VviRhaT1 (flavonol rhamnosyltransferase, VIT\_200s0218g00170), during berry development. In the present study, the transcript of the former was detected at low levels (FPKM < 0.1), while the latter was expressed with a FPKM ranging from 5 to 15 (Supplementary Table S3). Additionally, VviRhaT1 was significantly upregulated at E-L 35 in "CS," which might also account for the higher flavonol in winter grapes. LAR and ANR are key regulators of flavan-3-ol and proanthocyanidin biosynthesis. The expression of VviANR (VIT\_200s0361g00040) was higher in winter grapes at E-L 35, but no significant difference was found. Winter cropping significantly upregulated VviLAR1 (VIT\_201s0011g02960) at E-L 36 and VviLAR2 (VIT\_217s0000g04150) at E-L 38 in "CS," while in "Riesling," only VviLAR2 was significantly affected, with an upregulation at E-L 36 and a downregulation at E-L 38. Comparing the expression patterns between the two seasons, VviLAR2 in "Riesling" kept a moderate and stable level in summer cropping, while the transcript in winter cropping increased to twofold higher at veraison, and then it dropped sharply at harvest. Hence, it may be speculated that the transcript of VviLAR2 leads to the peak flavan-3-ol content at E-L 35 in winter grapes. Furthermore, there is a range of anthocyanin biosynthetic enzymes relevant to the glycosylation, methylation, and acylation events of anthocyanin, such as UFGT (Boss et al., 1996), anthocyanin O-methyltransferase (AOMT; Fournier-Level et al., 2011), and anthocyanin acyltransferase (3AT; Rinaldo et al., 2015). The expression patterns of these genes involved in anthocyanin modification across three developmental stages were similar with the expression patterns of the upstream metabolic genes, the transcripts of which peaked at E-L 36, with higher levels in the winter cropping than in the summer cropping. The expression of VviUFGT, catalyzing the formation of anthocyanin-3-O-glucosides, is critical for berry coloration (Boss et al., 1996). The constantly higher expression of VviUFGT in winter cropping correlated remarkably well with the greater abundance of anthocyanins in winter grapes than summer grapes. The two AOMTs responsible for anthocyanin methylation (Fournier-Level et al., 2011) were also significantly upregulated in winter cropping, which failed to explain the methylation variation, since the proportion of malvidin was dramatically decreased in winter grapes. It was noticed that two transcripts, VIT\_205s0062g00300 and VIT\_205s0062g00310, exhibited similar patterns in the two cultivars, but they differed in their magnitude of abundances; these transcripts might encode a UDP-glucose: anthocyanidin 5,3-O-glucosyltransferase (53GT) with homology to a flavonol glucosyltransferase-like protein. The winter cropping significantly downregulated the expression of Vvi53GT in "Riesling" at E-L 31, and then upregulated it at E-L 38. In fact, the real role of VIT\_205s0062g00300 and

VIT\_205s0062g00310 in grapes still needs further research, and the reason for their high abundances in "Riesling" is unclear.

# Transcriptional Modulation of Flavonoid Biosynthesis

Two classes of genes are required for flavonoid biosynthesis in grapes, the first class is structural genes that encode enzymes in the metabolic pathway, and the second class includes regulatory genes that control the transcription of these biosynthetic genes. The flavonoid pathway genes are known to be coordinately controlled by the interactions of R2R3–MYB, basic helix-loophelix (bHLH), and WD40-repeat transcription factors (TFs) in response to developmental cues or external stress factors, with MYB being central to the transcriptional complexes (Heppel et al., 2013). The ternary complex of MYB/bHLH/WD40 binds to responsive elements in the promoters of biosynthesis genes, activating transcription of genes in the pathway. While many structural genes were significantly upregulated in winter cropping berries, many previously identified regulatory genes were either weakly affected or were not affected by cropping cycle (Supplementary Table S3).

VviMYBF1 regulates a narrow set of genes involved in flavonol biosynthesis, potentially comprising the genes VviCHS (chalcone synthase) and VviFLS4 (also named VviFLS1; Czemmel et al., 2009, 2017). The expression of VviMYBF1 was very low in both varieties, and its transcript abundance decreased across the three developmental stages, similar to a previously reported expression pattern in developing Shiraz berries (Czemmel et al., 2009). The present study showed an upregulation of VviMYBF1 in the winter cropping cycle at E-L 38/E-L 35 in "CS"/"Riesling" (**Figure 6**), that was not well correlated with the expression pattern of VviFLS4. With respect to the regulation of proanthocyanidin-specific biosynthesis, many regulators have been recently characterized, including VviMYBPA1, VviMYBPA2, and VviMybPAR (Bogs et al., 2007; Terrier et al., 2009; Koyama et al., 2014). In addition, VviMYB5a and VviMYB5b, as well as the negative repressors VviMYBC2-L1 and VviMYBC2-L3, regulate the genes involved in proanthocyanidin biosynthesis and several steps in the upstream pathway (Deluc et al., 2006, 2008; Cavallini et al., 2015). The genes encoding the above TFs were expressed at some stage during ripening, and some of them showed similar expression patterns between the two varieties. It has been reported that MYB5a and MYB5b tightly exert their regulation in a temporal way during berry development, with MYB5a predominantly acting in the early stages and MYB5b acting near the later ripening process (Deluc et al., 2008; Matus, 2016). Similarly, the transcript profiles of VviMYB5a and VviMYB5b were distinct during ripening, with peak levels of VviMYB5a at E-L 35 and peak levels of VviMYB5b at E-L 38. The transcript levels of VviMYBPA2, VviMYBPAR, and VviMYBC2-L2, on the other hand, remained low throughout berry development, and they were not differentially expressed between the two cropping cycles. The expression of VviMYBPA1 was significantly downregulated in winter cropping berries at E-L 35 and E-L 38 in both varieties, which correlated well with the expression of two VviDFR transcripts (VIT\_219s0014g04980, VIT\_202s0025g01260). Interestingly, the expression of two

negative regulators, VviMYBC2-L1 and VviMYBC2-L3, was significantly upregulated at E-L 36 by winter cropping in "CS"; VviMYB4a was significantly upregulated as well, and it might negatively regulate flavonoid or phenolic acid synthesis (Cavallini et al., 2015). MYB14 and MYB15 were two TFs demonstrated to specifically activate STS genes (Höll et al., 2013). The expressions of VviMYB14 and VviMYB15 matched well with STSs profiles in "CS" rather than in "Riesling," suggesting different transcriptional regulation of stilbene biosynthesis in two varieties.

In V. vinifera grapes, two VviMYBA genes in a single gene cluster are responsible for berry color variation and anthocyanin accumulation, among which are the VviMYBA1 and VviMYBA2 genes encoding putative regulators of anthocyanin biosynthesis in red grapes (but these genes are non-functional in white grapes), while VviMYBA3 is only statistically associated with berry color without functional validation (Walker et al., 2007; Fournier-Level et al., 2009). These three VviMYBAs showed similar expression patterns in "CS," namely, they were significantly upregulated by winter cropping, especially at E-L 35 and E-L 36, in parallel with the expression profiles of VviUFGT, VviAOMT, and Vvi3AT, and the evolution of anthocyanin levels. More recently, two bHLH proteins of VviMYC1 (VIT\_207s0104g00090) and VviMYCA1 (VIT\_215s0046g02560) were demonstrated to promote anthocyanin accumulation in cooperative interaction with VviMYBA1 (Matus et al., 2010; Hichri et al., 2011), the transcripts of which showed no significant variation between the two cropping cycles. VviWDR1 (VIT\_216s0098g00870), which contributed positively to the accumulation of anthocyanins, was downregulated by winter cropping at E-L 38, contrary to the expression change of VviMYBA1.

In fact, apart from the widely acknowledged MYB, bHLH, and WD40 TFs, some regulators that belong to the WRKY, AP2/ERF, MADS-box, GATA, and bZIP families were also positively or negatively involved in flavonoid metabolisms (reviewed by Hichri et al., 2011). A recently identified TF of VvibZIPC22 is involved in the regulation of flavonoid biosynthesis, the expression of which was induced by UV light, paralleled by accumulation of the VviFLS4 transcript and flavonol compounds (Malacarne et al., 2016). In the present research, VvibZIPC22 was significantly upregulated by winter cropping in both varieties, in good agreement with the accumulation of flavonols. However, no correlation was found between the expression profiles of VviFLS4 and VvibZIPC22. It is relevant to mention the photomorphogenic factors from the bZIP family, which play important roles in mediating light-dependent flavonoid regulation, especially the biosynthesis of flavonols (Loyola et al., 2016; Matus, 2016). Many components in the UV-B signaling pathway that specifically perceive UV-B radiation have been identified in V. vinifera, including UV-B RECEPTOR 1 (UVR1, VIT\_207s0031g02560), two ELONGATED HYPOCOTYL 5 grape homologs (HY5, VIT\_204s0008g05210; HYH, VIT\_205s0020g01090), and two CONSTITUTIVE PHOTOMORPHOGENIC 1 (COP1-1, VIT\_212s0059g01420; COP1-2, VIT\_210s0523g00030), which could mediate flavonol biosynthesis in grapes under UV-B exposure (Carbonell-Bejerano et al., 2014; Liu et al., 2015; Loyola et al., 2016). In southern China, UV-B radiation was shown to

be higher in the summer, due to stronger sunlight and a longer duration of sunlight, although we did not detect the precise daily dose of UV-B. Concomitantly, most genes involved in the UV-B response pathway were significantly repressed in the winter cropping, except for VviCOP1-1 (Supplementary Table S3), which aligns with the climatic conditions. Solar UV radiation was shown to enhance flavonol accumulation (Carbonell-Bejerano et al., 2014; Loyola et al., 2016). However, the changes in VviUVR1, VviHY5, and VviHYH expression could not match the flavonol variation between the two cropping cycles, which might be resulted from temperature-responsive changes in HY5 levels, since HY5 protein has been shown to degrade at high temperature (Kim et al., 2017; Park et al., 2017).

#### Seasonal Response of Plant Hormone-Related Genes

Based on many years of cultivation experiences, grape farmers found the progression of ripening was different between the two cropping cycles. Since plant hormones play important roles in the ripening process of fruit, the expression of ABA and ethylene associated genes was analyzed. It was thought that VviNCED (9-cis-epoxycarotenoid dioxygenase) encodes the key enzyme for the bulk of ABA biosynthesis, and its expression correlated well with ABA accumulation (Sun et al., 2010; Young et al., 2012). The transcript abundances of VviNCED2 (VIT\_210s0003g03750) and VviNCED3 (VIT\_219s0093g00550) peaked at E-L 35 in both varieties, and they decreased after veraison. It was clear that the winter cropping upregulated three VviNCEDs in "CS," especially VviNCED3, which showed massive fold-changes in all of the berry stages. In "Riesling," VviNCED3 was upregulated at E-L 35 while VviNCED6 was significantly promoted at E-L 36 (**Figure 7** and Supplementary Table S4). The relative expression of genes involved in ABA catabolism, such as ABA 8 0 -hydroxylase (VIT\_202s0087g00710), was also significantly increased in both varieties. ABA plays a crucial role in response to a variety of abiotic stresses, such as drought, salinity, and extreme temperature. Thus, the drought due to low rainfall in the winter cropping had a more pronounced influence on ABA induction. In the ABA signaling pathway, the transcript abundances of VviPYR1/PYL/RCAR (VIT\_205s0077g01550, VIT\_202s0012g01270, VIT\_215s0046g01050) were significantly increased in winter cropping berries, while the expressions of VviPP2Cs (protein phosphatase 2C) were also significantly upregulated, except for two transcripts (VIT\_201s0011g03910, VIT\_206s0004g06840), which showed significant declines after veraison. For VviSnRK2s (sucrose-non-fermenting1-related kinase 2), both upregulated and downregulated transcripts were identified between two cropping cycles. ABA responsive element binding factors play a crucial role in ABA-dependent gene activation. There were no significant changes in the gene expression of ABFs in "Riesling," while significant downregulation was observed in "CS," except for VviABF2 (VIT\_218s0001g10450), which was confirmed to modulate ABA-dependent berry ripening processes, such as the promotion of phenolic synthesis and cell wall softening (Nicolas et al., 2014).

Considering the important role of ethylene in the ripening process and its cross-talk with ABA (Sun et al., 2010; Cramer et al., 2014), the expression changes in the ethylene biosynthesis and signaling pathways were also analyzed (Supplementary Table S4). Similar to the role of NCED in ABA biosynthesis, 1-aminocyclopropane 1-carboxylate oxidase (ACO) is known to determine the production of ethylene. The transcript accumulation of VviACO1 was suggested to match the occurrence of the ethylene peak in "CS" clusters (Chervin et al., 2004). The study here also showed decreasing expression levels of VviACO1 (VIT\_200s2086g00010) as berry ripening progressed from veraison, except for a sudden rise in the summer grapes of "Riesling" at harvest. The expression of putative VviACO1 and three other candidate genes (VIT\_205s0049g00280, VIT\_205s0049g00390, VIT\_205s0049g00420) were significantly repressed in winter grapes in both varieties in comparison to their expression in the summer cropping. The VviACO2 gene was expressed at high levels throughout berry development, with a significant upregulation in the winter berries of "CS" at postveraison. Only one transcript of VviACS (1-aminocyclopropane 1-carboxylate synthase, VIT\_202s0025g00360) was differentially expressed between the two cropping cycles, which peaked at E-L 35 and had higher levels in "Riesling" berries. Interestingly, winter cropping repressed the ethylene biosynthesis pathway, while the ethylene signaling pathway was greatly induced. Particularly, many genes involved in ethylene signaling, such as the ethylene receptors ethylene response 2 (ETR2) and ethylene insensitive 4 (EIN4), the negative regulators constitutive triple response 1 (CTR1), EIN3-binding F-box 1 and 2 (EBF1/2), and ethylene response factor (ERF), were significantly upregulated in winter cropping at different stages in both varieties.

Auxin plays positive roles in plant growth and delays ripening-associated processes. In grape berries, auxin is produced through the combined action of tryptophan aminotransferase related (TAR) and YUCCA (YUC) proteins in a two-step biosynthesis pathway (Böttcher et al., 2013). As expected, many transcripts in the auxin biosynthesis and signaling pathways were expressed in developing berries, peaking at E-L 35. Gene expression of the putative auxin biosynthesis members from the TAR and YUC families was analyzed. Four members of VviTAR were downregulated in "CS," but only VviTAR1 and VviTAR2 were differentially expressed. Comparing the three developmental stages, the transcript of VviTAR1 peaked at E-L 36 in summer cropping, while it increased gradually until E-L 38 in winter cropping. Both varieties showed similar expression patterns, but the winter cropping downregulated VviTAR1 in "CS" while it upregulated the transcript in "Riesling." The auxin-influx transporter AUX1 mediates the uptake of auxin, and many transcripts encoding AUX1 were upregulated by winter cropping. Auxin/indole acetic acid (AUX/IAA) proteins are a family of transcriptional repressors that play a central role in auxin response. Most VviAUX/IAA genes were downregulated at E-L 36 in "CS" while in "Riesling," the effects mainly occurred at E-L 38. In addition, there were many transcripts that also showed upregulation at E-L 38 in "Riesling," including VviTAR1 and VviAUX1. Gretchen Hagen 3 (GH3) plays a role in the conjugation of indole-3-acetic acid to aspartate at the onset of grape berry ripening (Böttcher et al., 2011); one putative transcript of

FIGURE 7 | Expression profiles of DEGs related to hormone biosynthesis and signaling pathway during grape development in "CS" and "Riesling" in 2014. The log2-transformed FPKM values (Winter/Summer) are represented by the color map. Blue and red boxes indicate downregulated and upregulated transcripts, respectively, in winter berries versus large berries. The numbers indicate different components. The complete data set can be accessed in Supplementary Table S4.

VIT\_203s0091g00310 was upregulated at E-L 35 in winter versus summer grapes, and another of VIT\_219s0014g04690 was promoted at E-L 38. The SMALL AUXIN UP RNA (SAURs) act as positive or negative regulators of auxin synthesis, and their transcription depended on the level of active auxin (Hagen and Guilfoyle, 2002). Downregulation was found to dominate the differential expression of VviSAURs in winter versus summer berries. Among the 10 VviSAURs, two (VIT\_209s0002g00650, VIT\_208s0040g01390) showed consecutive upregulation along three stages in "CS," but in "Riesling," the upregulation only occurred at E-L 38.

#### DISCUSSION

The metabolism of flavonoids in grapevine is closely correlated with environmental stresses, variety and tissue, developmental cues, and phytohormone regulation. There have been many transcriptomic studies showing the dynamics of gene expression and flavonoid accumulation in response to internal and external stimuli (Degu et al., 2014; Carbonell-Bejerano et al., 2014; Li et al., 2014; Savoi et al., 2016; Blanco-Ulate et al., 2017; Lecourieux et al., 2017; Sun et al., 2017). Here, a comparative parallel analysis was conducted on grapes grown under a double cropping system in two cultivars. To the best of our knowledge, this is the first systemic research of double cropping berries from both transcriptomic and metabolic views during the course of berry development, which enables a comprehensive description of the cropping cycle controlling flavonoid synthesis in grape during ripening.

# Winter Cropping Promotes Phenylpropanoid–Flavonoid Metabolism during Berry Ripening

Considering that the environmental conditions vary greatly in the two growing seasons, we propose that the climate factors of temperature, light, and rainfall exert great contributions to the variation of flavonoid compounds between the two cropping cycles. Investigations into the influences of climate factors on flavonoid biosynthesis in a vineyard have various approaches at different stages (reviewed by Kuhn et al., 2014). These cultural practices or environmental factors resulted in a range of differences in the levels and profiles of flavonoid compounds in grape berries. The effects of the cropping cycle on flavonoid accumulation are the result of a combination of many climate factors across the whole life cycle in the same vineyard. Notably, these three classes of flavonoids: anthocyanins, flavonols, and flavan-3-ols, respond differently to the climatic stimulus.

In grapes, flavonols are important ultraviolet light protectants and play pivotal roles in fresh fruit and the resulting wine. Flavonol accumulated in berries across the entire berry development process, showing an increase from pre-veraison to harvest in cv. "CS" but remaining relatively constant during ripening in "Riesling," which resulted in a different content and profile of flavonol products between the two cropping cycles. It has been reported that flavonols are the flavonoids that are most drastically affected by different light levels, correlating with the expression pattern of VviFLS4 but not with that of VviFLS5 and coinciding with its putative transcriptional regulator, VviMYBF1 (Czemmel et al., 2009; Matus et al., 2009; Carbonell-Bejerano et al., 2014). Contrarily, the berries in the summer cropping, which had abundant sunshine hours and high illuminance, accumulated lower amounts of flavonols than the winter grapes in both varieties. What was more interesting was the expression profiles of VviFLS4 and VviMYBF1; the former was upregulated at the harvest stage of "CS" versus post-veraison of "Riesling," while the latter was upregulated in "CS" but downregulated in "Riesling," and both were expressed at low transcript levels. The VviFLS5 gene was expressed at a level several times higher than VviFLS4 and showed a downregulation in winter cropping in both cultivars. The fact that there was discrepancy in flavonol biosynthesis and the expression profiles of VviFLS and VviMYBF1 in the present study indicated the involvement of post-transcriptional control or other regulators, which require further research. Notably, some differences not correlating well could be due to the fact that both shaded and sunny berries were sampled together. Besides, the deseeded berries were used to extract RNA while the flavonols were isolated from the skin, which also had an effect. Several studies have shown that light radiation plays a profound effect on flavonol synthesis, while temperature has no effect or a weak effect on their content (Spayd et al., 2002; Cohen et al., 2012; Azuma et al., 2015). Thus, the induction of flavonols in response to climatic stresses is a complex process, and the promotion by some factors in the winter cropping was suggested to have a greater effect than the offset from lower light radiation. Downey et al. (2003) displayed the two distinct periods of flavonol synthesis; the first occurred around flowering and the second occurred during berry ripening, which all coincided with the expression of VviFLS4. Therefore, the green berries in the winter cropping accumulated more FLS around fruit-set in response to strong light, which might have also impacted the veraison stage, contributing to larger amounts of flavonol. Additionally, the active upstream pathway in winter cropping could provide a greater abundance of precursors for the production of flavonols in grape skin.

UV-B radiation is a key environmental signal, exerting a strong effect on flavonol synthesis in grape berries. Briefly, UV-B is specifically perceived by UVR8, which in turn, function together with a positive COP1 to activate a range of signaling cascades mediated by HY5 and HYH; then, they finally regulate transcription of target genes, resulting in downstream responses (Loyola et al., 2016; Matus, 2016). Flavonol biosynthesis in grapes is suggested to be predominantly stimulated by the UV-B response pathway through the activation of HY5 and HYH (Liu et al., 2015; Czemmel et al., 2017). In addition, VviMYBF1 could act on the UV-B signaling cascade by activating HYH (Czemmel et al., 2017), thus inducing the direct and indirect activation of VviFLS4 and several targets in the phenylpropanoid pathway, such as VviCHS3 and VviGT5 (Carbonell-Bejerano et al., 2014; Loyola et al., 2016). The expression pattern of the UV-B response machinery in the two cropping cycles was correlated with light radiation, but it failed

to explain the variations in flavonol content in the current study, since enhanced levels of flavonols normally coincide with the upregulation of genes in the light-signaling pathway (Loyola et al., 2016).

Flavan-3-ols are monomeric subunits of condensed tannins in grape skin and seeds, contributing greatly to the body and mouthfeel of wines being produced. However, relatively little is known about the mechanism of environmental impact on flavan-3-ol production in grape skins, despite the fact that it shares common upstream steps with flavonol and anthocyanin. The grape skin had high levels of flavan-3-ols compared with other flavonoids; they accumulated from fruitset until veraison and then declined afterward, in parallel with the expression patterns of VviANR and VviLAR (Bogs et al., 2005). In the present study, the content of total flavan-3 ol in skin was determined in two consecutive years, but it varied only in 2014. The evolution pattern of total flavan-3-ols in grape skins also varied in two cropping cycles. Since the massive production of flavan-3-ols occurred at the green stage, it was hard to correlate the gene expression profiles with its accumulation, as we only conducted RNA-Seq from the onset of ripening. However, a higher expression of VviLAR1 or VviLAR2 might explain the higher levels of flavan-3-ols in the winter cropping. To date, several regulatory TFs related to flavan-3-ol biosynthesis have been found in grape: the MYB positive regulators of VviMYBPAR and VviMYBPA1/PA2 and the MYB C2 repressors of VviMYBC2- L1/L3 (Bogs et al., 2007; Terrier et al., 2009; Koyama et al., 2014; Cavallini et al., 2015), many members of which also worked beyond the phenylpropanoid pathway in addition to VviMYB5a/5b (Deluc et al., 2006, 2008). According to the regulatory mechanisms of flavonoid biosynthesis in grapes, VviMYBPAR, VviMYBPA1, VviMYBPA2, and VviMYB5a are particularly involved in regulation before the onset of ripening, while VviMYB5b also controlled the general flavonoid pathway after veraison (Deluc et al., 2008; Matus, 2016). However, VviMYB5b is more related to anthocyanins in the ripening stage rather than proanthocyanidins (Cavallini et al., 2014). Thus, it was hard to correlate the flavanol evolution pattern with gene expression profiles during ripening. Additionally, Carbonell-Bejerano et al. (2013) also found that elevated temperature promoted the expression of VviMYBPA1, but repressed VviMYB5a, indicating that these regulators responded differently to external stimuli during grape development and that the regulatory network was tightly organized in a spatial and temporal way.

As specific metabolites of the flavonoid pathway, anthocyanins are synthesized only in red grapes under the strict control of multiple regulatory factors. A rapid accumulation of anthocyanins in red varieties begins from veraison until harvest, leading to more assimilated carbon flux into the production of anthocyanins rather than flavan-3-ols. The positive regulators VviMYBA1 and VviMYBA2 are specific to controlling the biosynthesis of anthocyanin in red V. vinifera grapes mediated through VviUFGT, while in white grapes, a retrotransposon of Gret1 inserted in the promoter region of the VviMYBA1 gene and two non-synonymous mutations that occurred in the VviMYBA2 coding region disrupt their regulatory function (Boss et al., 1996; Kobayashi et al., 2004; Walker et al., 2007). In the winter cropping of "CS," both VviMYBA1 (fold change > 1.5) and VviMYBA2 (fold change approximately 4.0) were significantly promoted at pre- and post-veraison, they then significantly upregulated the key gene of VviUFGT (fold change > 2.0) at post-veraison, and finally, they contributed to the drastic accumulation of anthocyanins (fold change approximately 10) during ripening. As far as we know, little research has been conducted on double cropping grapes in subtropical climates, except for two contrasting works. One also found higher flavan-3-ol and anthocyanin levels in winter berries at harvest (Xu et al., 2011), while the other one showed no difference between the two cropping cycles at maturity (Chou and Li, 2014). Differences in grape genotype, environmental conditions, and cultural practices might explain this conflicting result. Recent studies also confirmed that high temperature (35◦C) strongly reduced anthocyanin synthesis and enhanced its degradation (Carbonell-Bejerano et al., 2013; Lecourieux et al., 2017). Moreover, the repressors VviMYB4a and VviMYBC2- L1/L3 were significantly upregulated in the winter cropping at post-veraison, in accordance with the potential roles of balancing the inductive effects of activators (Cavallini et al., 2015).

Here, it is worth discussing the competition between F30H and F3<sup>0</sup> 5 <sup>0</sup>H in grapes, since the profiles of flavonoids varied significantly in the two cropping cycles, especially for anthocyanins and flavonols. In contrast to the high expression of VviF30H and VviF3<sup>0</sup> 5 <sup>0</sup>H in red grapes, the expression of the VviF3<sup>0</sup> 5 <sup>0</sup>H transcript in white grapes during ripening was extremely low, suggesting a different regulatory mechanism (Bogs et al., 2006). Thus, it was interesting to note an accumulation of the 3<sup>0</sup> 5 0 -substituted compounds of flavan-3 ol (i.e., epigallocatechin), but not that of flavonol (myricetin), in the white "Riesling" grapes. Considering the great "loss" of the F3<sup>0</sup> 5 <sup>0</sup>H pathway in "Riesling" during ripening, we only discussed the competitive relation of F30H and F3<sup>0</sup> 5 <sup>0</sup>H in "CS." The expression patterns of VviF30H and VviF3<sup>0</sup> 5 <sup>0</sup>H coordinated with the upstream pathway genes, peaking around post-veraison, but varying in magnitude. Concurrent with higher the expression ratio of VviF30H to VviF3<sup>0</sup> 5 <sup>0</sup>H in summer grapes, there was an increase in the proportion of 3<sup>0</sup> -substituted flavonoids in the summer cropping, which is similar to a previous result induced by water deficits (Castellarin et al., 2007). Clearly, the prevalence of VviF3<sup>0</sup> 5 <sup>0</sup>H over VviF30H would lead to more dihydromyricetin, the precursor of myricetin, epigallocatechin, and delphinidin, and in contrast, it would yield less dihydroquercetin, the precursor of quercetin, catechin, and cyanidin (Kuhn et al., 2014). The flavonol composition was almost all quercetintype compounds (95%) at the green stage, and this fraction decreased during ripening and dropped to a low level (70% in summer versus 40% in winter) at harvest, while the flavan-3-ols were almost exclusively catechin-type compounds (85%), and its proportion showed no difference regardless of the sample stage or cropping cycle (Supplementary Table S2). With respect to anthocyanin, the effects of the cropping cycle were somewhere in between. The anthocyanin profile showed no

consistent difference in 2015, whereas in 2014, the percentage of cyanidin-type compounds was significantly higher in the winter cropping cycle (15% in summer versus 25% in winter). Hence, the work here confirmed that F30H acted as an early flavonoid gene, while F3<sup>0</sup> 5 <sup>0</sup>H worked at later stages (Degu et al., 2014). The higher fraction of these 3<sup>0</sup> -substituted flavonoids at harvest in summer cropping resembled the metabolic profiles of the grapes under heat or sunlight exposure (Guan et al., 2014; Lecourieux et al., 2017). A recent work showed that VviMYBA1 could specifically induce VviF3<sup>0</sup> 5 <sup>0</sup>H and promote tri-hydroxylated fractions (Matus et al., 2017). Therefore, in winter cropping, the induced F3<sup>0</sup> 5 <sup>0</sup>H might divert more flux to the 3<sup>0</sup> 5 0 -substituted sub-branch and might provide more precursors to synthesize anthocyanins and flavonols. However, the production of 3<sup>0</sup> 5 0 -substituted flavan-3-ol reached a plateau in spite of a large amount of precursors, which is likely due to flavonol-specific control or the low substrate activity of LAR for leucodelphinidin.

## Winter Cropping Advances the Onset of Veraison and Accelerates Ripening Progression

In fruits, ripening is a complex event that involves major physiological and metabolic changes controlled by plant hormones, which are also signals in response to developmental and environmental cues (Kuhn et al., 2014). However, to date, the ripening mechanism of non-climacteric fruits is still poorly understood, especially with respect to the regulation of phytohormones. Since the climatic conditions varied greatly in the two cropping cycles, it inevitably affected the initiation of veraison and the progression of ripening, which was a good opportunity to learn about the ripening mechanism in grapevines.

Understanding how plants respond to environmental stimuli is crucial to improving yield and grape quality in the field. Berry ripening is affected by multiple climatic factors, of which water deficit (Castellarin et al., 2007), light exposure (Matus et al., 2009), and damping diurnal temperature range (Cohen et al., 2012) advance the onset of ripening, while high temperature (Lecourieux et al., 2017) and shading treatment (Matus et al., 2009) delay it. In the current study, the phenological phase varied greatly between the two growing seasons. Winter cropping accelerated the duration from budburst to post-veraison, therefore advancing but also prolonging the subsequent ripening process. The difference in ambient temperature accounted for most of this phenomenon. On the one hand, the optimum temperature range between 25 and 30◦C was associated with higher rates of plant growth (Buttrose, 1969; Cohen et al., 2008), and this was the case for processes from bud-break to flowering in the winter cycle. On the other hand, with respect to grape berries, low temperature and damping diurnal temperature fluctuation in the winter cropping hastened their development and ripening processes (Kliewer and Torres, 1972; Cohen et al., 2008), while the extremely high temperature in the summer might inhibit fruit growth and berry ripening. The photoperiod or the daylength is also considered a fundamental environmental signal that affects phenological development (Sreekantan et al., 2010). The phases of flowering, onset of ripening, bud dormancy, and leaf senescence are modulated by day length, together with other stress factors (Keller, 2010). This seasonal decrease in the photoperiod and temperature might also be associated with the temporal variation in ripening progression. Moreover, berry weight was negatively correlated with the onset of veraison, since the larger berry seemed to be a dilution of the smaller one, needing more time to achieve the accumulation of the primary and secondary metabolites. The reason for the advancement of veraison in wintering cropping might be, at least in part, due to the smaller berry size. More importantly, the berries in winter cropping showed a hastened increase in TSS, especially before or around veraison, and then the rate dropped to a similar level to summer berries during ripening. Another consequence of high temperature is a hastened decline of acidity in berries (Carbonell-Bejerano et al., 2013). The sugar:acid balance at harvest is an important trait of fruit quality, and a gradual fall in acidity is also coupled with a sugar increase. During the ripening process in winter, the temperature was extremely low (approximately 10◦C), so it took a long time to complete the degradation of acid, resulting in more sugar accumulation, and this was the reason why winter cropping prolonged the ripening duration. Thus, the longer duration of ripening stage in winter versus summer cropping was due to the purposes to harvest high-quality grapes.

The regulation of grape development and ripening in response to external or internal cues involves a dynamic interplay among hormones. Albeit unclearly, the functional reciprocity among plant hormones was suggested to control ripening transitions and progression (Sun et al., 2010; Böttcher et al., 2013; Pilati et al., 2017). The transition of ripening, termed as veraison, is accompanied by the modulation of hormones, concurrently with many physical, chemical, and physiological changes (Castellarin et al., 2016). Here, most genes encoding NCED/ACO/TAR, the key enzymes of ABA/ethylene/auxin biosynthesis, showed coordinated expression patterns, peaking around pre-veraison, in spite of their different expression levels. Particularly, during early berry development, the level of endogenous auxin was high and peaked around veraison, then its rapid decline occurred from pre-veraison, followed by sequential increases in ethylene and ABA content (Böttcher et al., 2013; Kuhn et al., 2014). Several studies confirmed the ethylene peak preceded the ABA peak at pre-veraison, and they suggested the trace ethylene could induce the expression of VviNCED and therefore the biosynthesis of ABA (Chervin et al., 2004; Sun et al., 2010). In turn, the exogenous ABA treatment on pre-veraison berries also triggered ethylene biosynthesis, suggesting a functional and positive interaction between ABA and ethylene (Sun et al., 2010; Pilati et al., 2017). In the present study, genes in ABA-related pathway were significantly upregulated in winter cropping cycle, which might explain the advance of ripening initiation.

The applications of ABA and ethylene before veraison seemed to accelerate the onset of berry ripening, whereas the synthetic auxins treatment delayed the ripening initiation (Sun et al., 2010; Ziliotto et al., 2012). Ziliotto et al. (2012) found an "antagonistic" effect between auxin and ethylene, and a "synergistic" effect between auxins and ABA. The genes involved

in ABA biosynthesis and perception were repressed by auxin spraying at pre-veraison, while the ethylene biosynthetic pathway was triggered (Ziliotto et al., 2012). Ziliotto et al. (2012) found a peak of ethylene biosynthesis genes coincided with high expression levels of auxin biosynthesis genes by a pre-veraison auxin analogs treatment, leading the berries back to the preveraison stage. Similarly, when the pre-veraison fruits were treated with ethylene-releasing compound, the transient increase of auxin specifically induced by ethylene would counteract the positive effect of excess auxin, thus delaying the initiation of ripening phase (Böttcher et al., 2013). In this context, VviACO1, the transcript that was confirmed to be consistent with the ethylene peak (Sun et al., 2010), was induced by summer cropping, together with VviTAR in "CS," probably resulting from the strong antagonism between auxin and ethylene.

As regards to auxin degradation, it has been shown that GH3 determined this process by conjugation of auxin with amino acids at pre-veraison, and it positively responded to the application of exogenous ABA and ethylene, representing a signal of the berry ripening (Böttcher et al., 2010). The decrease of auxin levels after veraison was probably due to the formation of IAA-Aspartate, which was closely correlated with the VviGH3 transcript levels (Böttcher et al., 2010; Corso et al., 2016). And GH3-1 was claimed to be responsible for the auxin homeostasis at pre-veraison (Böttcher et al., 2010). Here, the VviGH3-1 (VIT\_203s0091g00310) expression was peaked at E-L 35 and remained at high levels at E-L 36 and E-L 38, in agreement with previous report (Böttcher et al., 2010). The transcript of VviGH3-1 was upregulated at E-L 35 in winter grapes compared to summer grapes, with a 0.5-fold higher in "CS" and a 1.2-fold higher in "Riesling" (Supplementary Table S4). This behavior of VviGH3-1 might be associated to the earlier onset of veraison in winter cropping cycle, since the action of GH3 could potentially control of berry ripening rate (Böttcher et al., 2010, 2013; Corso et al., 2016).

Taken together, the differential expression pattern of hormone-related transcripts between the two cropping cycles explained the ripening variation at a hormone level. In general, ABA is acted as a stress-stimulated signal, and the water deficit and low temperature in winter cropping acted as a positive regulator of ABA production (Deluc et al., 2009). Furthermore, coloring by the synthesis of pigments is another major change during the onset of ripening (Castellarin et al., 2016). The deeper color with more anthocyanins in winter cropping also coincided with the transcript levels of VviNCED, reproducing the major effects of ABA on the rise of anthocyanins, together with a sugar increase (Wheeler et al., 2009).

#### Final Remarks

Climate conditions have far-reaching implications for grape cultivation in the field. The extremely high temperature and frequent, intense precipitation in the conventional grapevine growing cycle take a major toll on grape yield and quality in southern China. Thus, the development of a double cropping system per year is a great breakthrough, not only minimizing the impact of bad weather in a subtropical monsoon climate but also improving the quality and yield of out-of-season grapes. Our research here was mainly focused on the flavonoid metabolism, since the grapes grown under two cropping cycles showed distinctly different phenolic compounds in addition to their differences in skin coloration. In addition, we demonstrated that the winter cropping cycle promoted the biosynthesis of flavonoids by (i) avoiding many types of detrimental weather events and making good use of the abundant heat and light resources in southern China; (ii) prolonging the duration of ripening stage to give the berry more time to accumulate flavonoid compounds; (iii) altering the expression patterns of flavonoid-related TFs, particularly with the upregulation of VviMYBA1, VviMYBA2, VviMYBF1, and VviMYB5a and the downregulation of VviMYBPA1, which in turn, greatly induced the flavonoid biosynthetic genes; (iv) triggering the ABArelated ripening processes, which also positively coincided with anthocyanin accumulation; and (v) correlating with their smaller berries and higher sugars. Thus, the alterations in ripening regulatory networks and the flavonoid biosynthetic pathway probably mainly occurred at pre-veraison, leading to a great increase in metabolic gene expression around post-veraison and subsequent flavonoid accumulation.

# AUTHOR CONTRIBUTIONS

JW, X-JB, and C-QD designed the experiments on vineyard samples. M-MC, GC, and X-JC promoted the double cropping system and sampled the grapes. W-KC processed the samples for RNA isolation for RNA-seq analysis and flavonoid extraction for HPLC-MS/MS analysis, and drafted the manuscript. X-HY participated in the process of flavonoid extractions. R-RG, YW, and LH provided statistical and bioinformatics analysis. FH and JW revised the manuscript and provided suggestions. All authors contributed to discussion of the results and approved the final manuscript.

# FUNDING

This work was supported by China Agriculture Research System (CARS-29).

# ACKNOWLEDGMENTS

We thank the staff from the Grape and Wine Research Institute in the Guangxi Academy of Agricultural Sciences for sampling and pruning, particularly Li Chen, Guo-Pin Chen, and Shu-Yu Xie for providing technical support.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2017.01912/ full#supplementary-material

FIGURE S1 | The evolution of berry fresh weight of 'Cabernet Sauvignon' and 'Riesling' under double cropping system in 2014 and 2015.

# REFERENCES

fpls-08-01912 November 8, 2017 Time: 17:51 # 18


under double pruning management. Sci. Agric. 74, 134–141. doi: 10.1590/1678- 992x-2015-0384



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Chen, Bai, Cao, Cheng, Cao, Guo, Wang, He, Yang, He, Duan and Wang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

#### Edited by:

Ashraf El-kereamy, University of California, United States

#### Reviewed by:

M. Teresa Sanchez-Ballesta, Instituto de Ciencia y Tecnología de Alimentos y Nutrición (ICTAN-CSIC), Spain Jerome Grimplet, Instituto de Ciencias de la Vid y del Vino (ICVV), Spain

#### \*Correspondence:

Simone D. Castellarin simone.castellarin@ubc.ca

#### †Present address:

Stefania Savoi and Jose C. Herrera, Division of Viticulture and Pomology, Department of Crop Sciences, University of Natural Resources and Life Sciences Vienna, Tulln, Austria

‡These authors have contributed equally to this work.

#### Specialty section:

This article was submitted to Technical Advances in Plant Science, a section of the journal Frontiers in Plant Science

> Received: 04 April 2017 Accepted: 12 June 2017 Published: 10 July 2017

#### Citation:

Savoi S, Wong DCJ, Degu A, Herrera JC, Bucchetti B, Peterlunger E, Fait A, Mattivi F and Castellarin SD (2017) Multi-Omics and Integrated Network Analyses Reveal New Insights into the Systems Relationships between Metabolites, Structural Genes, and Transcriptional Regulators in Developing Grape Berries (Vitis vinifera L.) Exposed to Water Deficit. Front. Plant Sci. 8:1124. doi: 10.3389/fpls.2017.01124 Multi-Omics and Integrated Network Analyses Reveal New Insights into the Systems Relationships between Metabolites, Structural Genes, and Transcriptional Regulators in Developing Grape Berries (Vitis vinifera L.) Exposed to Water Deficit

Stefania Savoi1,2†‡, Darren C. J. Wong<sup>3</sup>‡ , Asfaw Degu<sup>4</sup> , Jose C. Herrera<sup>1</sup>† , Barbara Bucchetti<sup>1</sup> , Enrico Peterlunger<sup>1</sup> , Aaron Fait<sup>4</sup> , Fulvio Mattivi2,5 and Simone D. Castellarin<sup>3</sup> \*

<sup>1</sup> Department of Agricultural, Food, Environmental and Animal Sciences, University of Udine, Udine, Italy, <sup>2</sup> Department of Food Quality and Nutrition, Research and Innovation Centre, Fondazione Edmund Mach, San Michele all'Adige, Italy, <sup>3</sup> Wine Research Centre, The University of British Columbia, Vancouver, BC, Canada, <sup>4</sup> The Jacob Blaustein Institutes for Desert Research, Ben-Gurion University of the Negev, Sede Boqer, Israel, <sup>5</sup> Center Agriculture Food Environment, University of Trento, San Michele all'Adige, Italy

Grapes are one of the major fruit crops and they are cultivated in many dry environments. This study comprehensively characterizes the metabolic response of grape berries exposed to water deficit at different developmental stages. Increases of proline, branched-chain amino acids, phenylpropanoids, anthocyanins, and free volatile organic compounds have been previously observed in grape berries exposed to water deficit. Integrating RNA-sequencing analysis of the transcriptome with large-scale analysis of central and specialized metabolites, we reveal that these increases occur via a coordinated regulation of key structural pathway genes. Water deficit-induced up-regulation of flavonoid genes is also coordinated with the down-regulation of many stilbene synthases and a consistent decrease in stilbenoid concentration. Water deficit activated both ABA-dependent and ABA-independent signal transduction pathways by modulating the expression of several transcription factors. Gene-gene and genemetabolite network analyses showed that water deficit-responsive transcription factors such as bZIPs, AP2/ERFs, MYBs, and NACs are implicated in the regulation of stressresponsive metabolites. Enrichment of known and novel cis-regulatory elements in the promoters of several ripening-specific/water deficit-induced modules further affirms the involvement of a transcription factor cross-talk in the berry response to water deficit. Together, our integrated approaches show that water deficit-regulated gene modules are strongly linked to key fruit-quality metabolites and multiple signal transduction

**157**

pathways may be critical to achieve a balance between the regulation of the stressresponse and the berry ripening program. This study constitutes an invaluable resource for future discoveries and comparative studies, in grapes and other fruits, centered on reproductive tissue metabolism under abiotic stress.

Keywords: abiotic stress, central metabolism, drought, grapevine, fruit quality, ripening, RNA-sequencing, specialized metabolism

#### INTRODUCTION

fpls-08-01124 July 6, 2017 Time: 16:47 # 2

Drought is considered one of the major threats for crops in the predicted future climatic scenarios. In fruit crops, drought can impact the accumulation of metabolites that determine fruit quality (Ripoll et al., 2014). Generally classified as drought tolerant, grape is often not irrigated or minimally irrigated to improve the berry composition (Chaves et al., 2007). Previous research showed that water deficit causes large changes on the specialized metabolism of the grape berry, promoting the synthesis of specific volatile organic compounds (Bindon et al., 2007; Savoi et al., 2016), carotenoids (Deluc et al., 2009), and phenolics (Castellarin et al., 2007a,b; Hochberg et al., 2015). The grape berry central metabolism is also affected by water deficit. Deluc et al. (2009) reported an increase of proline concentration in Cabernet Sauvignon berries, and the increase paralleled the induction of key genes for the proline synthesis such as the one codifying for the pyrroline-5-carboxylate synthase. Hochberg et al. (2015) reported that also other central metabolites, including sucrose, several amino acids and organic acids, ascorbate, and raffinose can take part in the metabolic response of the grape berry to water deficit.

These results indicate that a complex regulation of several metabolic pathways, possibly determined by common or specific molecular signals, underlay the metabolic response of the grape berry to water deficit. In the model plant Arabidopsis thaliana, the transcriptional response to drought is modulated by both ABAdependent and ABA-independent signal transduction pathways (Yamaguchi-Shinozaki and Shinozaki, 2006; Singh and Laxmi, 2015). In the ABA dependent pathway, the accumulation of ABA is sensed by PYR/PYL/RCAR-PP2C receptor complex, and activates a class III SnRK2s that phosphorylates four transcription factors (TFs), ABA-responsive element (ABRE) binding protein 1 (AREB1), AREB2, ABRE binding factor 3 (ABF3), and ABF1. These TFs regulate several downstream genes by binding to the ABRE cis-regulatory element (CRE) present in the promoter region of these genes. Drought induced ABA also regulates the activity of MYB/MYC, NAC, WRKY, and NF-Y TFs. In the ABA-independent signaling pathway, the DREB2A, a protein that belongs to the AP2/ERF family, plays a pivotal role in the transcriptional response under drought stress (Yoshida et al., 2014). Recently, a cross-talk between ABA-dependent and ABA-independent pathways has been hypothesized under drought stress, with AREB/ABFs inducing DREB2A (Nakashima et al., 2014; Singh and Laxmi, 2015). In grape, ABA participates in the drought response of multiple organs and in the process of fruit ripening. In the xylem sap of shoots and roots, ABA levels negatively correlate with stem and root water potential, and increases of ABA concentration correspond to decreases of leaf transpiration (Rossdeutsch et al., 2016). ABA related genes were shown to be involved in the transcriptional response of the leaf to drought (Dal Santo et al., 2016; Hopper et al., 2016) and, in the grape berry, water deficit increased ABA concentration and ABA related transcripts, but not consistently between varieties (Deluc et al., 2009).

Among fruit crops, grapes are the most considered crop in studies on the impact of drought on fruit composition. The adoption of large-scale metabolite and transcript analyses (Deluc et al., 2009; Hochberg et al., 2015; Savoi et al., 2016) has strongly enhanced our understanding of the metabolic response of the grape berry to water deficit; however, systems approaches considering the analysis of primary and secondary metabolism, the molecular pathways, and the network relationships between genes and metabolites involved in the response to water deficit are still missing in grape, as well as in other fleshy fruits. Furthermore, the modulation of the volatile organic compounds (VOCs) under water deficit has been poorly investigated in wine grapes (Savoi et al., 2016), although they represent a key component of the final wine flavor.

Recent studies have successfully considered a multi-omics approach in several annual herbaceous crops, such as Medicago truncatula (Zhang et al., 2014), Oryza sativa (Maruyama et al., 2014), and Zea mays (Opitz et al., 2016). In these studies, RNA-sequencing and large-scale metabolite analyses were adopted and revealed several commonalities in the metabolic response between these crops and A. thaliana. Network-based analyses using multi-omics data provide a powerful tool for discovering links between and within the many layers of biological complexity that governs plant functions such as the coordinated regulation of genes and metabolic pathways. This approach has been performed in a few fruit crop studies to prioritize candidate genes involved the control of fruit development and composition (Mounet et al., 2009; Zamboni et al., 2010; Osorio et al., 2011, 2012; Savoi et al., 2016). We aimed to apply a similar approach to deeply characterize the molecular and metabolite response to water deficit in the grape berry, to uncover the relationships between water deficit-responsive genes and the metabolite accumulation during berry development, and to identify the key putative molecular regulators that underlay the metabolic response to water deficit. For that purpose, we conducted a two season experiment where two contrasting water regimes were applied and the levels of transcripts and metabolites were analyzed at several berry developmental stages. Transcripts

and metabolites that were modulated by water deficit were used for constructing molecular networks to investigate their relationships and to identify the major molecular pathways that underlay the response of grape berry metabolism to this abiotic stress.

# MATERIALS AND METHODS

## Field Experiment, Physiological Measurements, and Sample Preparation

Field experiments were conducted in 2011 and 2012 at the University of Udine experimental farm on 18/19 years old Vitis vinifera L. 'Merlot' (clone R3 onto SO4 rootstock) vines. The vines were planted with 2.5 m × 1.0 m spacing and trellised to a spur cordon system. To ensure a proper management of the water regime during the experimental trial and consistent treatments across seasons, four rows were covered with an ethylene-vinylacetate (EVA) film at the beginning of the seasons, as described in Herrera et al. (2015). Two irrigation treatments were imposed at approximately 25 days after anthesis (DAA): control (CT) vines where weekly irrigated maintaining the midday stem water potential (9Stem) above −0.6 MPa and water deficit (WD) vines were not irrigated from fruit set until 9Stem was lower than −1.4 MPa, whereupon irrigation was managed to maintain 9Stem between −1.0 and −1.4 MPa. Each treatment was replicated four times in plots of 10 vines each in a completely randomized design. Plant water status was monitored weekly by measuring midday 9Stem (Savoi et al., 2016).

Berries were randomly sampled for analyses seven times during each season: three times before the beginning of ripening (30, 44, and 60 DAA in 2011, and 26, 40, and 53 DAA in 2012), at the beginning of ripening (74 DAA in 2011 and 67 DAA in 2012), and three times during berry ripening (87, 100, and 115 DAA in 2011, and 81, 95, and 106 DAA in 2012). These developmental stages corresponded to E-L 31, 32, 33, 35, 36, 37, and 38 in the E-L system, respectively. The onset of ripening (veraison, defined as the day at which 50% of the berries had changed color from green to red) was recorded at 69 and 60 DAA in 2011 and 2012, respectively. In both seasons, the last sampling point coincided with the harvest date.

At each date, two sets of berries were randomly collected from each plot, for a total of four biological replicates per irrigation treatment. The first set of 60 berries was used for measuring berry weight, total soluble solids (TSS), titratable acidity (TA), and pH as described in Herrera et al. (2015). The second set of 40 berries was used for the metabolite and transcript analyses; samples were snap frozen with liquid nitrogen, and stored at −80◦C.

For metabolite and transcript analyses, whole berries, without pedicel, were grinded to a fine powder under liquid nitrogen using an analytical mill (IKA-Werke GMbH & Co.). The frozen powder was aliquoted for metabolite and RNA extraction as described below. Moreover, a quality control (QC) sample for metabolite analysis was prepared by pooling together aliquots of all the samples. Large-scale targeted metabolite analyses were undertaken in both seasons, while transcriptome analyses were performed only in 2012.

#### Metabolite Analyses

Primary metabolites were extracted from 100 mg of frozen powder, derivatized for GC-MS analysis and analyzed in a Trace GC Ultra gas chromatograph coupled to a DSQII quadrupole mass spectrometer (Thermo Scientific) as described in Degu et al. (2014). XCalibur software was used for the mass spectra identification using the NIST library (United States) and the RI libraries from the Max-Planck Institute for Plant Physiology (Germany). The QC sample was used for data normalization.

Metabolites were determined as described in Savoi et al. (2016) unless specified. Phenylpropanoids, stilbenoids, flavonols, flavan-3-ols, and proanthocyanidins chromatographic analysis was carried out using a Waters Acquity UPLC system (Milford) coupled to a Waters Xevo triple-quadrupole mass spectrometer detector (Milford). Compounds were identified with TargetLynx software based on their reference standard, retention time, and qualifier and quantifier ion, and were accurately quantified by their calibration curve and expressed as mg/Kg of grapes. Anthocyanins were analyzed as described in Sivilotti et al. (2016) using a HPLC (Shimadzu) equipped with a diode array detector. The concentration of individual anthocyanins was expressed in oenin chloride equivalents and expressed as mg/Kg of grapes. Carotenoids chromatographic analysis was performed in a 1290 Infinity Binary UPLC (Agilent) equipped with an RP C30 3 µm column Spectra components and elution profiles were determined with the R package 'alsace' 3.0. Compounds were quantified from linear calibration curves built with standard solutions and expressed as mg/Kg of grapes.

Free (non-glycosylated) VOCs analysis was performed with a Trace GC Ultra gas chromatograph (Thermo Scientific) coupled to a TSQ Quantum Tandem mass spectrometer. XCalibur software was used for the peaks identification. VOCs were identified by comparing the retention times of individual peaks with the retention times of their reference standards, and by identifying the mass spectra using the NIST library. The ratio of each VOC area to the d8-acetophenone internal standard area was considered to reduce technical variability among extractions and chromatographic runs and VOCs quantity were expressed as µg/Kg of grapes of d8-acetophenone equivalents.

Extractions and injections of the samples were performed in a random sequence and QC samples were injected at the beginning of the sequence and every six sample injections.

### RNA Extractions and RNA Sequencing Analysis

Transcriptome analyses were performed on the samples collected at 26, 53, 67, 81, and 106 DAA in 2012. Three out of the four biological replicates per treatment were considered. RNA extraction, RNA quality and quantity determination, library preparation and quantification, sequencing and QC analysis were performed as described in Savoi et al. (2016). Reads were aligned against the reference grape genome V1 PN40024 12X (Jaillon et al., 2007) using the software TopHat version 2.0.6 (Trapnell et al., 2012) with default parameters. Aligned reads were counted with htseq-count (version 0.6.0), in intersectionnon-empty mode for overlap resolution (Anders et al., 2015).

Differentially expressed (DE) genes [false discovery rate (FDR) less than 0.05] analysis was performed with the R package 'DESeq2' (Love et al., 2014). Annotation of gene functions was done according to Grimplet et al. (2012) and Naithani et al. (2014) and retrieved from recent literature. Gene ontology analyses for each sampling were carried out as described in Savoi et al. (2016).

## Quantitative Real-Time Polymerase Chain Reaction

The validation of RNA-Seq data was carried out on a set of DE genes using the quantitative real-time polymerase chain reaction (qPCR) technique. The reverse transcription of RNA samples was performed with the QuantiTect Reverse Transcription Kit (Qiagen); specific primers for 12 selected genes were designed with Primer3web version 4.0.0 (**Supplementary Table S1**). Real-Time PCR experiments were performed on a Bio-Rad CFX96TM using SsoFastTM EvaGreen <sup>R</sup> Supermix. qPCR run condition were as per instructions with annealing temperature of 58◦C. VviAP47 (VIT\_02s0012g00910) was used as a reference gene (Reid et al., 2006). In order to validate the technical and biological reliability of the transcriptome dataset, qPCR analysis was carried out on samples collected at different stages of development (starting from the sampling that preceded the onset of ripening) in 2011 and 2012.

#### Statistical and Network Analyses

A one-way ANOVA was performed using JMP 7 (SAS Institute Inc.) to detect significant differences (P < 0.05) between irrigation treatments at each sampling date for the several physiological and compositional parameters considered. Heatmaps representing the log<sup>2</sup> fold change (log2FC) of metabolite concentrations between treatments (WD/CT) were drawn with R software. Metabolites were clustered with Person correlation and a complete link. Principal component analyses (PCAs) on metabolite and transcriptome datasets were performed using the R software. Co-expression analysis was performed using weighted correlation gene co-expression network analysis (WGCNA) package in R (Langfelder and Horvath, 2008) using the log<sup>2</sup> fold change (WD/CT) [dataset1] and the variance stabilized transformed (VST) counts [dataset2] of DE genes at 26, 53, 67, 81, and 106 DAA, separately to identify highly correlated genes sharing similar water deficit co-response and development accumulation patterns, respectively. Empirical P-value of Pearson Correlation Coefficient (PCC) values (statistical significance) of dataset1 and dataset2 were estimated by 1,000 permutations using the 'rsgcc' package (Ma and Wang, 2012). Enrichment for CRE in the gene promoters (1 kb upstream of the 5<sup>0</sup> UTR or TSS) of each WGCNA co-response modules was conducted as described previously (Savoi et al., 2016; Wong et al., 2016). Known CREs of 6-, 7-, 8-mers (222 in total) were analyzed. Enrichment of CREs was validated with the hypergeometric test adjusted with FDR correction. Putative CREs were deemed significantly enriched under FDR < 0.01. Stress co-response and development-regulated submodules were created using PCC threshold > 0.8 with an empirical P-value < 0.01 and visualized using Cytoscape software (version 3.1.1) (Shannon et al., 2003).

#### Accession Number

All raw sequence reads have been deposited in NCBI Sequence Read Archive<sup>1</sup> . The BioProject accession is PRJNA348618.

# RESULTS

#### Impact of Water Deficit on Fruit Development

In 2011, differences in 9Stem between treatments started from 50 DAA and the lowest values in WD vines (<−1.4 MPa) were measured at late stages of fruit development (95 and 110 DAA). Whereas in 2012, differences between the treatments started 10 days earlier (39 DAA), and the lowest values in WD vines (<−1.4 MPa) were measured at 75 and 88 DAA (**Figure 1**). At this stage, the 9Stem in WD vines was −0.8 and −1.2 MPa in 2011 and in 2012, respectively. In both seasons, 9Stem of WD vines was consistently lower than −1.0 MPa for the entire ripening period.

Water deficit significantly reduced berry weight at 74, 100, and 115 DAA in 2011 (**Figure 1**) and at 67, 81, 95, and 106 DAA in 2012 (**Figure 1**). The reduction in berry weight was also reflected in lower vine productivity in both seasons (**Supplementary Table S2**).

Significant differences in TSS and TA were observed only in 2012 (**Figure 1**). In this season, water deficit increased TSS before (40 and 53 DAA) and after (95 and 106 DAA) the onset of fruit ripening, while TA was observed to be lower and higher in WD than in CT at 26 and 53 DAA, respectively.

# Impact of Water Deficit on Berry Metabolites

A total number of 101 compounds, belonging to the central (primary) and specialized (secondary) metabolism, were identified (standard annotation level 1) and quantified. A complete list of the compounds identified, with their machine readable chemical identifier, is reported in **Supplementary Table S3**, and the kinetics of the accumulation of these compounds in CT and WD berries during development in both vintages is reported in Supplementary Figure S1.

Principal component analyses were used to compare the metabolite profiles of CT and WD berries during development in both seasons considering the whole set of metabolites and the central and specialized metabolism separately (**Figure 2**). The first two principal components represent from 64.17 to 67.67% of the variance in the datasets. In all the PC1-PC2 score plots, clear separation among samples was observed based on the berry developmental stage. This separation was consistent between the two seasons; samples collected at similar developmental stages in the two seasons clustered together in the PC1-PC2 score plots. The PC2 and PC1 separated the developmental stages before and after the beginning of fruit ripening, respectively.

<sup>1</sup>http://www.ncbi.nlm.nih.gov/sra

Separation between irrigation treatments was observed from 100 DAA in 2011 and from 81 DAA in 2012. This large and coordinated change in the whole composition of the berry was more evident when the whole set of metabolites was included in the analysis than when central and specialized metabolism were analyzed independently (**Figure 2**). In both seasons, differences in the berry metabolite profiling between CT and WD berries were maximized at the last sampling stage.

Among the 34 central – amino acids, sugars, organic acids, polyols, and polyamines – metabolites analyzed, a total of 11 and 15 metabolites were significantly modulated by water deficit at one or more developmental stages in 2011 and 2012, respectively (**Figure 3** and Supplementary Figure S1). Most differences were observed during berry ripening, when in WD berries the level of leucine, valine, isoleucine, threonine, proline, and putrescine increased consistently in the two seasons. On the contrary, raffinose decreased before the beginning of ripening in both seasons as a result of water deficit, as well as malate decreased at late stages of ripening.

The specialized metabolites analyzed in this study were phenolics, carotenoids, and free VOCs. Among the 39 phenolics detected, 22 and 28 were significantly modulated by water deficit in 2011 and 2012, respectively (**Figure 4** and Supplementary Figure S1). Eighteen of them were consistently modulated between the two seasons. Water deficit increased the concentration of benzoic and cinnamic acids. Specifically, gallic acid was increased by water deficit at harvest in both seasons, while trans-caftaric, trans-coutaric, and trans-fertaric acids were increased only in 2012. Water deficit strongly increased the concentration of most anthocyanins detected. Stilbenoids, such as trans-resveratrol, piceatannol, and pallidol, were decreased by water deficit in both seasons at harvest, while cis- and trans-piceid were significantly decreased by water deficit only in 2011 even though they had a similar trend but showed no significance in 2012. Monomeric flavan-3-ols, such as catechin and epicatechin gallate, were decreased by water deficit in both seasons during berry ripening. Conversely, proanthocyanidins B1, B2+B4 were generally increased by water deficit. Flavonols were not consistently affected by water deficit; quercetin-3-O-rutinoside was significantly increased by water deficit, but only at 95 DAA in 2012. However, the method used for detecting and quantifying flavonols was unable to detect the glycosylated myricetin, one of the main flavonols produced in red grapes (Mattivi et al., 2006).

Carotenoids such as violaxanthin, neoxanthin, and lutein 5-6-epoxide, decreased in WD berries during ripening, while zeaxanthin increased (**Figure 5** and Supplementary Figure S1).

A strong increase of free VOCs was consistently observed between the two seasons at late developmental stages under water deficit. Remarkably, eight free VOCs were consistently modulated between seasons by water deficit (**Figure 5** and Supplementary Figure S1). C6 VOCs, such as (E)-2-hexenal and 3-hexenol, increased in WD berries at the onset of ripening, while the C6 VOC hexanol increased at 74 DAA in 2011 and at 40 DAA in 2012, but decreased at 26 DAA in 2012. Furthermore, C5, C7, C8, and C9 VOCs, such as 1-penten-3-ol, (E)-2-heptenal, (E)-2 octenal, 1-octen-3-ol, and nonanol increased in WD berries at late ripening stages.

### Impact of Water Deficit on Berry Transcriptome

Larger metabolic changes occurred in 2012, when water deficit started earlier and was overall more severe during the season (**Figure 1**), thus transcriptome analysis was undertaken on berry samples collected during this season at five berry developmental stages (before the onset of ripening, 26 and 53 DAA; at the onset of ripening, 67 DAA; and during ripening, 81 and 106 DAA).

The average number of unique reads that mapped the 12X V1 version of the grape genome (Jaillon et al., 2007) was 27.1 M (**Supplementary Table S4**). Among the 29,971 genes of the grape genome, 23,253 (77.6%) genes were expressed at 26 DAA, 23,220 (77.5%) at 53 DAA, 21,997 (73.4%) at 67 DAA, 22,453 (74.9%) at 81 DAA, and 22,162 (73.9%) at 106 DAA.

A PCA was performed to compare the transcriptome profiles of the 30 independent samples analyzed (2 treatments × 5 developmental stages × 3 biological replicates) (**Figure 6**). The first two principal components explain 61.9 and 15.8%, of the variance among samples, respectively. Berry transcriptome were clearly separated accordingly to the developmental stage; moreover, within developmental stages, the berry transcriptomes of WD berries grouped together and were separated from the transcriptomes of CT berries.

The total number of DE genes between CT and WD was 5,167 (**Supplementary Table S5**). Water deficit modulated the expression of 214 genes (175 up-regulated; 39 down-regulated) at 26 DAA, 90 genes (38 up-regulated; 52 down-regulated) at 53 DAA, 1,290 genes (662 up-regulated; 628 down-regulated) at 67 DAA, 2,900 genes (1,569 up-regulated; 1,331 downregulated) at 81 DAA, and 2,925 genes (1,431 up-regulated; 1,494 down-regulated) at 106 DAA (**Figure 6**). Several of them were differentially regulated at more than one developmental stage (**Figure 6**). Thirty GO categories (slim biological processes) were significantly overrepresented among the DE genes as presented in Supplementary Figure S2.

The expression of 12 selected DE genes was tested at several developmental stages in both seasons with a qPCR (Supplementary Figure S3). This analysis indicates that differences in the gene expression level between treatments remained consistent regardless the platform or the season considered.

#### Transcriptional Regulatory Networks of Berries under Water Deficit

Transcription factors are central in regulating many plant biological processes; including developmental processes and response to the environment. A total of 447 TFs out of 2,211 possible TFs encoded in the grape genome (Grimplet et al., 2012) were modulated in the berry in response to water deficit. In this study, emphasis was given on the ripening stages (67, 81, and 106 DAA), when large metabolite and transcriptome

responses to water deficit occurred. A large proportion of the water deficit-modulated TFs belongs to the MYB (33 genes), bHLH (33 genes), AP2-ERF (28 genes), C2H2 (27 genes), WRKY (26 genes), NAC (23 genes), C3H (20 genes), HB (18 genes), GRAS (16 genes), and bZIP (14 genes) families (**Supplementary Table S5**). A selection of these genes is reported in **Figure 7**.

Central components of the ABA-independent drought response pathway are the AP2/ERF TFs encoding dehydrationresponsive element binding proteins (DREBs) that bind the dehydration-responsive element/C-repeat (DRE/CRT) sequence (Mizoi et al., 2012). Several of these genes, were significantly modulated by water deficit. For example, VviERF1, an upstream component of the jasmonic acid and ethylene signaling pathway, induced by high salinity and drought stress in Arabidopsis (Cheng et al., 2013), was strongly up-regulated by water deficit during fruit ripening. Another AP2/ERF-DREB (VviRAP2.4) was up-regulated by water deficit at late ripening. In Arabidopsis this TF confers drought tolerance by activating drought-responsive genes (Lin et al., 2008).

Our data indicate that basic leucine zipper (bZIP) proteins – components of the ABA-dependent signaling pathway – might also be implicated in the regulation of the drought response in fruits.

MYB TFs are key modulators of plant metabolism and development, and have been shown to be involved in the drought response. Several MYB TFs were strongly modulated by water deficit. Among them, we noted VviMYB14 (VIT\_07s0005g03340) and VviMYB15 (VIT\_05s0049g01020), VviMYB5B (VIT\_06s0004g00570), VviMYBA1 (VIT\_02s0033g00410), VviMYBC2-L1 (VIT\_01s0011g04760), and VviMYBPA1 (VIT\_15s0046g00170) that are involved in regulating various branches of the phenylpropanoid, stilbenoid, and flavonoid metabolism.

In this study, most of the DE NACs were up-regulated by water deficit. Among the other TFs modulated by water deficit, two auxin response factors (VviARF3 – VIT\_10s0003g00420; VviARF2 – VIT\_17s0000g00320) – possibly implicated in fruit development and ripening (Kumar et al., 2014) – encoding orthologous genes for AtARF3/ETTIN and AtARF2, were down- and up-regulated during ripening, respectively. Example of grape ARFs that may integrate multiple signaling pathways includes, VviARF2 (VIT\_17s0000g00320), the homolog of tomato ARF2, and VviARF5 (MONOPTEROS,

indicates the class of metabolites.

VIT\_18s0001g13930). Arabidopsis ARF2 and ARF5 are master regulators of auxin hormone responses, largely targeting the genes implicated in hormone-mediated signaling pathway, growth, and tissues development, among others (O'Malley et al., 2016).

Two homeobox-leucine zipper proteins (VviHB7 – VIT\_15s0048g02870, and VviHB12 – VIT\_16s0098g01170), encoding for orthologous genes of AtHB7 and AtHB12 (Valdés et al., 2012), were significantly up-regulated under water deficit. Particularly, VviHB12 was up-regulated from 53 DAA onward, with the highest induction at 81 DAA when water deficit reached the highest severity. Also, two of the WRKYs (VviWRK18 – VIT\_04s0008g05760, and VviWRKY40 – VIT\_09s0018g00240) involved in the ABA signaling pathway (Geilen and Böhmer, 2015) were down-regulated under water deficit during ripening. Vice versa, one WRKY TF (VviWRKY71 – VIT\_12s0028g00270), that has been shown to be involved in the oxidative stress process, as well as in the salicylic acid and jasmonic acid signaling pathways (Guo and Qin, 2016), was up-regulated at 81 DAA by water deficit.

#### Modulation of Central and Specialized Pathway Genes under Water Deficit

Many genes DE under water deficit codify for enzyme involved in major central and specialized pathways. Detailed description of how these major pathways were modulated is presented and discussed in Supplementary Figures S4–S7.

Water deficit affected the expression of genes involved in 9 steps out of 10 of the glycolysis metabolic pathway (Supplementary Figure S4). Significant differences in the gene expression were observed from 67 DAA onward, and the majority of these genes were up-regulated by water deficit. Few genes of the TCA cycle were moderately modulated at 81 or 106 DAA. Moreover, several sugar transporters, possibly

involved in the monosaccharide, sucrose, and polyol and hexose intake into the berry cells were modulated. Hexoses are the major sugars accumulated in the grape berry during ripening and, interestingly, most of the hexose transporters were downregulated.

Water deficit increased the concentration of the short branched-chain amino acids leucine, valine, and isoleucine in both seasons (**Figure 3**) and several genes involved in valine and leucine biosynthesis were modulated during berry ripening (Supplementary Figure S5). Genes that underlay the synthesis of proline including glutamate dehydrogenases (GluDH), glutamate synthases (GluS), and a pyrroline-5-carboxylate synthase (P5CS) were strongly up-regulated during the final stages of ripening. Furthermore, genes that promote the decarboxylation of arginine or ornithine into polyamines were up-regulated.

Consistently with the differences observed in the phenylpropanoid, stilbenoid, and flavonoid accumulation (**Figure 4**), Water deficit strongly modulated most steps of the related biosynthetic pathways (Supplementary Figure S6); it promoted the expression of the branch of the flavonoid pathway which leads to the production of tri-hydroxylated anthocyanins (Castellarin et al., 2007b) and down-regulated 28 out of 45 stilbene synthases (STSs).

Water deficit also affected the expression of several genes of the carotenoid pathway (Supplementary Figure S7), mostly by up-regulating them. Carotenoids such as neoxanthin and violaxanthin can be cleaved by 9-cis-epoxycarotenoid dioxygenase (NCED) and further modified to produce the drought and ripening related hormone ABA. Three VviNCEDs were up-regulated in WD berries during ripening.

The molecular pathways that underlay the VOC production in fruits, as well as their modulation under water deficit remain largely unknown. Several VOCs detected in this study are produced from the peroxidation of free C18 polyunsaturated fatty acids, such as linolenic and linoleic acids, which lead to the production of C6, C9 (Kalua and Boss, 2009) and putatively C5 (Shen et al., 2014) VOCs. The fatty acid degradation involves lipoxygenases (LOX), hydroperoxide lyase (HPL), hexenal isomerases (HI), and alcohol dehydrogenases (ADH) (Schwab et al., 2008). We found that a 13-LOX, a HPL and five ADHs were up-regulated in WD berries during berry ripening. The induction of VviLOX (VIT\_06s0004g01510) in WD berries at 106 DAA, may explain the higher accumulation of C5, C8, and C9 VOCs observed. In addition, we identified two grape genes codifying for (Z)-3:(E)-2-hexenal isomerases, recently identified paprika and tomato (Kunishima et al., 2016), consistently up-regulated in WD berries from 67 DAA onward.

Finally, we observed a strong modulation of many transcripts involved in the reactive oxygen species (ROS) related pathways

(production, scavenging, and signaling) involved in the plant responses to abiotic stresses (**Supplementary Table S5**) (Dal Santo et al., 2016).

## Predicted Water Deficit-Regulated Modules Link Central Players in the Metabolic Response

To determine the correlation pattern among DE genes and analyze their regulation during berry development and in response to water deficit, WGCNA was performed (Langfelder and Horvath, 2008). Eleven co-response gene modules (clusters) of highly correlated genes based on the water deficit-modulation were identified, with each module containing up to six generalized development-based accumulation patterns (**Figure 8** and **Supplementary Table S6**). In seven modules (named WD1, 2, 5, 6, 8, 9, and 11) we observed a general upregulation of the gene expression under water deficit; on the contrary, in four modules (named WD3, 4, 7, and 10) we observed a general down-regulation of the gene expression. Based on the fact that water deficit was stronger on metabolite and transcript abundance during fruit ripening, we focused on the DE that decreased in expression during ripening (sub-module DEV1 and DEV4), the ones that peaked at the onset of ripening (sub-module DEV5), the ones which were generally highly expressed from the onset of ripening to harvest (sub-module DEV3), and the ones which showed a steady increase in transcripts during ripening before peaking at harvest (sub-module DEV2).

In order to gain more insights into the regulatory control of water deficit-induced genes, the promoters of module genes were evaluated for CRE enrichment (**Figure 8** and **Supplementary Table S6**). We evaluated 222 CREs, 6- to 8 mer in length, given that these lengths usually define the primary DNA element recognized by various plant TFs (Franco-Zorrilla et al., 2014). We have recently shown that, in grape, many of these CREs are indeed bona-fide CREs with large biological relevance (Wong et al., 2016, 2017). The ACGTGKC CRE was highly enriched in general water deficit-induced modules, such as in WD2, WD9, and WD11 modules. Similarly, CACGTG CRE was highly enriched in the WD2 and WD11 modules. Although the core DRE element (RCCGAC) was not enriched in the modules, the 7-mer DRE, RCCGACA, and other DRE-related CREs (e.g., RYCGAC and MACCGMCW) involved in drought responses were enriched in the WD9 and WD11 modules. Examples of genes containing ABRE and G-box CREs in promoters include all three water deficit upregulated VviNCEDs (VIT\_05s0051g00670, VIT\_10s0003g03750, and VIT\_19s0093g00550) – whose homologs in Arabidopsis (e.g., AtNCED3) regulate ABA biosynthesis during drought stress (Yamaguchi-Shinozaki and Shinozaki, 2006). Other CREs enrichment in these DE genes have been associated with droughtinduced transcriptional pathways, including NACR (CATGTG enriched in the WD11 module and TTRCGT in the WD2 and WD11 modules), GCC-box (GCCGCC enriched in the WD1 and WD11 modules), and MYB (CCGTTA enriched in the WD2 and WD11 modules), while others may be novel, such as AuxRE/ETT (TGTCGG enriched in the WD2 and

WD11 modules) and ZAT6 (ACACTA enriched in the WD8 module).

Module-metabolite co-response was first determined for each module satisfying a PCC > 0.8, as a preliminary step to understand the coordinated regulation of modules with target metabolites (**Supplementary Table S6**). For instance, the WD2 module was correlated with 11 of 15 anthocyanin compounds profiled, in agreement with the many flavonoid pathway genes (e.g., four flavonoid 3<sup>0</sup> 5 0 -hydroxylases –VviF3<sup>0</sup> 5 <sup>0</sup>Hs, the UDP-glucose:flavonoid-3-O-glucosyltransferase –VviUFGT, one leucoanthocyanidin dioxygenase –VviLDOX, and one flavanone 3-hydroxylases –VviF3H) that were localized in that module. To a lesser extent, the WD11 module was correlated with four different anthocyanin compounds coinciding with fewer flavonoid pathway genes (e.g., three VviF3<sup>0</sup> 5 <sup>0</sup>Hs, two chalcone synthases –VviCHSs, and one anthocyanin-acylglucoside transporter –VviAnthoMATE2). The WD9, WD2, and WD11 modules were correlated with various amino acids. For example, the WD11 module correlated to valine, threonine, and proline, and contains key structural genes involved in the formation of precursors for proline (such as VviGluDH, VviGluS, and VviP5CS) and threonine (e.g., threonine synthase – VviTS). Correlations between stilbenoid compounds with the modules considered were generally weak (PCC < 0.8), despite a large number of VviSTS genes being localized into WD3 (8 VviSTSs) and WD4 (19 VviSTSs) modules. As regards the correlation of modules with VOCs concentration, the WD2 module was correlated with 1-octen-3-ol and nonanol, and contained

FIGURE 8 | Functional overview of biological modules modulated by water deficit in the grape berry. (A) Violin plots represent the distribution of log2FC (WD/CT) of all DE genes in stress (WD) co-response gene modules. Box plots represent the distribution of variance stabilized transformed expression values of DE genes in development (DEV) co-regulated gene modules. (B) Overview of selected enriched cis-regulatory elements (CREs) in stress co-response genes modules. Colors of circles representative the TF family that recognize a designated CRE. Sizes or circles represent the number of genes containing a designated CRE in their promoter region. Opacity of circles (in color) represent the enrichment score, –log10(FDR), of CREs. Only enriched CREs (FDR < 0.01) in designated modules are depicted. Network representation of WD2 and WD11 modules, having ripening-associated expression patterns (DEV2/3/5), centered on significantly correlated TF and/or structural genes with (C) anthocyanin (pink, lavender blue, cyan, blue, and purple), and (D) amino acid compounds (orange, red, and green). Structural genes, TF, and metabolites are represented by circle, square, and triangle nodes, respectively. Thick edges in light red and gray edges represent associations between genes/TFs (PCC > 0.8; P < 0.01) for (C,D), respectively. Gene-metabolite (PCC > 0.8; P < 0.01) are depicted in thinner edges and the different colors denote different categories, as reported in (C,D).

VviHPL1. Interestingly, several transcripts involved in the degradation of storage lipids such as triacylglycerol lipase and phospholipase A2 were also found in this module. Finally, strong correlations were observed between the WD5 module and zeaxanthin; consistently, the module included several structural carotenoid genes such as phytoene desaturase and ζ-carotene desaturase.

In order to discover new links between and within the many layers of biological complexity that governs the metabolic response of the grape berry to water deficit, the coordination between metabolite and gene patterns and water deficit responses was further investigated. For the metabolitegene centered subnetwork, significant correlations between metabolites (anthocyanin and amino acids) and annotated structural pathway genes and TFs were investigated further for sub-modules DEV2, DEV3, and DEV5 in WD2 and WD11. The WD2 and WD11 modules were considered for anthocyanin (**Figure 8**), while only module WD11 was considered for amino acid metabolite-gene correlations (**Figure 8**). In addition, only regulatory genes from the bZIP, AP2/ERF, MYB, and NAC families were mainly considered given a consistent enrichment of the associated CREs in the promoters of both modules. Focusing on the anthocyanin subnetwork, we observe frequent correlations of bZIP (VviABF4 and VviGBF3), AP2/ERF (VviRAP2.1 and VviRAP2.4), and NAC (VIT\_02s0012g01040 and VIT\_10s0003g00350) TFs with various anthocyanin in the WD2 module. Consistently, frequent correlations of AP2/ERF (VviERF62) and NAC (VviNAC87 and VviRD26) TFs were also observed with anthocyanin in the WD11 module. As expected, frequent correlations of VviMYBA1 (WD11) to three VviF3<sup>0</sup> 5 <sup>0</sup>Hs, and to eight anthocyanins further corroborate its role in regulating various anthocyanin pathway genes during berry development (Walker et al., 2007; Rinaldo et al., 2015) and under water deficit (Castellarin et al., 2007b). Interestingly, the ABRE, DRE/CRT and GCC, and/or NACR were also found in the promoter of VviMYBA1and of various flavonoid genes in WD11 module (**Supplementary Table S6**); this might provide additional clues into the regulation of anthocyanin biosynthesis under water deficit. In the amino acid subnetwork, VviERF1 and VviNAC1/VviNAC33 (VIT\_19s0027g00230) were among the top five highly correlated TFs with proline in this module, and were also connected with VviP5CS in the network. ERF1 is involved in the regulation of P5CS during normal growth and abiotic stresses in Arabidopsis (Cheng et al., 2013). Reinforcing the observed gene and metabolite correlations, VviP5CS promoter also contains CRE signatures potentially relevant for VviERF1 binding, via the GCC-box (GCCGCC) and DRE (MACCGMCW) CRE (**Supplementary Table S6**) suggesting a conserved regulatory mechanism regulating proline via VviP5CS in the berries in response to water deficit stress. Other TF binding sites such as NAC (TTRCGT and TTACGTGT) (**Supplementary Table S6**) in P5CS promoter highlight its potential regulation via VviNAC1/VviNAC33. In addition, VviERF62 and VviRD26 were also correlated with VviGluS and CRE signatures related to DRE, GCC-box, and NACR are found within VviGluS promoter indicating potential regulation of these TFs on VviGluS.

# DISCUSSION

Our results represent the first multi-omics study of the metabolic changes induced by water deficit in grape berry and, to the best of our knowledge, in any fruit. We revealed that both the ABAdependent and ABA-independent signal transduction pathways are modulated by water deficit during fruit ripening, highlighting their central role in a reproductive organ such as the berry, in addition to their widely accepted roles in vegetative tissues of many plants (Yamaguchi-Shinozaki and Shinozaki, 2006; Shinozaki and Yamaguchi-Shinozaki, 2007). Gene-gene, genemetabolite network analyses, as well as gene promoter analysis shed light on the poorly understood systems relationships between regulators, structural genes, and metabolites in the fruit response to water deficit not only in grapes but also in other fruits. The integrated network analysis associated genes involved in amino acid, phenylpropanoid, and flavonoid pathways with the stress-responsive TFs (bZIPs, AP2/ERFs, MYBs, and NACs) that took part in the water deficit-stress signal into tightly interconnected modules. Enrichment of specific CREs (ABRE, DRE, NACR, GCC-box, MYB, AuxRE/ETT, and ZAT6) in DE genes of specific modules was consistent with the module membership of TFs that recognize these sites. This provides strength into the role of these TFs in modulating water deficit-responsive genes. The analysis of gene-metabolite co-response networks in ripening-associated sub-modules led us to propose several VviAP2/ERF and VviNAC members as putative regulators of the amino acid and anthocyanin accumulation in the grape berries under water deficit. Network and promoter analysis also revealed that the participation of VviARFs and AuxRE, involved in the auxin signaling pathway, may be an important component of the fruit response to water deficit.

The coordinated response of genes and metabolites (**Figure 8** and **Supplementary Table S6**) to water deficit in a fleshy fruit is similar to what observed in various tissues of A. thaliana (Harb et al., 2010), M. truncatula (Zhang et al., 2014), and Z. mays (Opitz et al., 2016). Several studies have shown that both ABA-dependent and ABA-independent pathways are indispensable for water deficit response in vegetative tissues (Yamaguchi-Shinozaki and Shinozaki, 2006; Shinozaki and Yamaguchi-Shinozaki, 2007). Here, we show that similar mechanisms apply in a reproductive organ such as the grape berry. In regards to the ABA-dependent signaling pathway, several VviAREB/ABFs and VvibZIP genes were modulated by water deficit during berry ripening when deficit reached its highest severity. In regards to the ABA-independent pathway, VviAP2/ERFs-DREBs TFs were highly induced by water deficit (Mizoi et al., 2012) as reported in Arabidopsis (Lin et al., 2008; Cheng et al., 2013). The relevance for the observed modulation of these pathways in the grape may be attributed with higher ABA sensitivity, reduced transpiration rate, and improved drought tolerance (Kang et al., 2002; Fujita et al., 2005).

In our dataset, one TF in particular (VviHB12) was upregulated by water deficit at 53, 67, 81, and 106 DAA. The induction of VviHB12 at 53 DAA, just before the onset of

ripening and when major effects of water deficit on the level of transcripts and metabolites were not observed yet, suggests that VviHB12 may be involved in one of the earliest responses to water deficit. The Arabidopsis AtHB12 is strongly induced by water deficit and ABA, and participates in the regulation of ABA signaling through the regulation of PP2C and ABA receptor gene expression (Olsson et al., 2004; Valdés et al., 2012). Similarly to what was found in Arabidopsis, the overexpression of this gene is also related with the higher expression of a VviPP2C highly expressed during fruit ripening (Supplementary Figure S3).

The involvement of multiple stress regulons (Nakashima et al., 2014) might be critical to orchestrate the balance between stress-responsive regulation and the berry ripening program. These examples suggest that increased osmotic stress signals induced by water deficit may further fine-tune the ripening program through regulating multiple interacting TFs, possibly accelerating ripening (Castellarin et al., 2007b; Herrera and Castellarin, 2016). Further supporting this, several TF 'switch' genes that may be master regulators of berry ripening, such as VviMYBA1-2, VviNAC1/VviNAC33- 47-71, and VviLBD15-18-38 (Palumbo et al., 2014), are induced under water deficit from the onset of ripening onward.

Furthermore, this study confirmed several metabolic reprogramming patterns previously reported in grapevine as well as in other plants. WD increased the accumulation of proline and branch chain amino acids such as leucine, valine, and isoleucine, confirming their role in drought response in grapevine (Deluc et al., 2009; Hochberg et al., 2015), A. thaliana (Nambara et al., 1998; Urano et al., 2009) and O. sativa (Maruyama et al., 2014). In the case of proline increase a parallel up-regulation of the key biosynthetic gene VviP5CS and of VviGluDH was observed. A study in tobacco and grape has shown that abiotic stress-induced ROS activate GluDH expression and enhances GluDH activity to produce glutamate for proline biosynthesis (Skopelitis et al., 2006). The coordinative induction of VviGluDH with VviP5CS under water deficit, observed in WD berries during ripening, may be a conserved mechanism necessary for maintaining a large amount of glutamate available for proline accumulation. NAC binding sites in VviP5CS promoter and the CRE signatures related to DRE, GCCbox, and NACR found within VviGluS promoter indicate potential regulation of these TFs on the proline accumulation. Direct implication of NAC1 on P5CS, or RD26 and ERF62 on GluS regulation has not been shown in plants, but some evidences show that other plant NAC and AP2/ERF members, such as JUNGBRUNNEN1/ANAC042 (Wu et al., 2012), OsNAC5 (Song et al., 2011), or GmERF3 (Zhang et al., 2009) are directly implicated in drought stress-mediated proline accumulation.

Several transcripts involved in phenylpropanoid and flavonoid biosynthesis, including VviMYB5b – a generic regulator of this pathway (Deluc et al., 2008; Cavallini et al., 2014), were enhanced by water deficit in parallel with a higher accumulation of related metabolites, such as benzoic and cinnamic acids, and anthocyanins. Previous studies have already reported a modulation of these pathways in grape berries exposed to water deficit (Deluc et al., 2009; Savoi et al., 2016). Nonetheless, our gene-metabolite network analysis identified correlations between specific structural and regulatory flavonoid genes (e.g., CHSs, LDOX, UFGT, AOMT, F30 5 <sup>0</sup>Hs, and MybAs) and anthocyanin modulated by water deficit.

The large demand for precursor for anthocyanins production possibly determines the observed impairment of stilbenoid production which decreased both in biosynthesis and concentration, indicating a redirection of phenylpropanoids to the flavonoid pathway instead to the stilbenoid one. However, the two MYBs, VviMYB14 and VviMYB15, that regulate stilbene biosynthesis in grapevine (Höll et al., 2013) do not correlate with the transcripts levels of VviSTSs modulated in WD berries, suggesting that other TFs might contribute to the stilbenoid regulation (Wong et al., 2016). One candidate could be VviMYBC2-L1, a negative regulator of flavonoid (anthocyanin and proanthocyanidin) and stilbenoid biosynthesis (Huang et al., 2014; Cavallini et al., 2015), that was up-regulated at 67 and 81 DAA, potentially repressing VviSTS transcripts. Stilbenoid production increased in Cabernet Sauvignon berries exposed to water deficit (Deluc et al., 2011) but were not significantly affected in Tocai Friulano (Savoi et al., 2016), indicating that the degree of water deficit and the genotype may be key factors for stilbenoid accumulation under drought events.

Analysis of gene-metabolite co-response networks in ripening-associated sub-modules, revealed a strong coordinated response between structural pathway genes and metabolite whilst identifying known regulators for grape anthocyanin biosynthesis (e.g., VviMYBA1-2). Gene-metabolite correlation networks has been successfully applied to prioritize candidate genes involved the control of fruit composition and development in tomato (Mounet et al., 2009) and in grapes (Zamboni et al., 2010; Savoi et al., 2016). In this study, we identified new candidate regulators for anthocyanin compounds, including several VviNACs that may play direct and/or indirect roles in regulating structural genes or specific pathway regulators, respectively. A recent study demonstrated that NAC TFs (PpBL, PpNAC1, and PpNAC2) can trans-activate PpMYB10.1 (homologs to VviMYBA1-2) promoter and that the silencing of PpBL inhibits anthocyanin pigmentation in peach fruits (Zhou et al., 2015). Little is known on the regulation of grape amino acid metabolism. The WD11 metabolite-gene subnetwork centered on proline, valine, and threonine correlated with the expected pathway genes such as VviP5CS, VviGluDH, VviGluS, VviTS, and other amino acid metabolism genes, and with TFs, such as VviERF1, VviNAC1, and VviERF62. Functional validation of these regulatory modules is necessary to confirm their role in the regulation of critical anthocyanin and amino acid biosynthetic genes in response to water deficit.

Many ABRE, DRE/CRT, and NACR CREs involved in stress-responsive transcription in vegetative tissues (Yamaguchi-Shinozaki and Shinozaki, 2006) were found in promoters

of DE genes and were also enriched in water deficitinduced modules. This highlights the conserved role of these stress-responsive CREs in modulating water deficit-responsive genes in reproductive tissues such as berries. Nonetheless, our analysis also show additional CRE pertaining to auxin responses may be an important component of the fruit response to water deficit, an observation/role that has not been implicated before in other fruit systems. Enrichment of the auxin response element (AuxRE/ETT – TGTCGG) in water deficit-induced modules (e.g., WD2 and WD11) suggests that ARFs that bind to these sites may play an important role in regulating water deficit-induced genes in berries, potentially via auxin signaling. ARFs have relevant function in drought-stress responses in plants like Glycine max (Ha et al., 2013) and regulate many aspects of fruit development and ripening (Kumar et al., 2014). Some of them [e.g., SlARF2, Hao et al. (2015)] are central components of fruit (tomato) development and ripening regulatory network. These observations reinforces that stress-responsive CREs may serve a critical role in ripening regulatory networks and fruit maturation. Many fruit ripening-associated TFs bind to these elements (Karlova et al., 2014; Kumar et al., 2014; Leng et al., 2014), of which some (e.g., VviABF2 and tomato SlNAC4 – homologs of Arabidopsis ATAF1 and VviATAF1) are known to regulate both abiotic stress responses and fruit ripening (Nicolas et al., 2014; Zhu et al., 2014). Finally, limited water availability affects VOCs production in several plant organs (Nowak et al., 2010; Griesser et al., 2015), including fleshy fruits, such as apple (Behboudian et al., 1998; Hooijdonk et al., 2007), and tomato (Veit-Köhler et al., 1999). We have recently reported that water deficit modulates the synthesis of monoterpenes in white grapes (Savoi et al., 2016) and, interestingly, at the end of ripening several volatiles such as 1-octen-3-one, (E)-2-heptenal, (E)-2-octenal, and nonanol were commonly up-regulated by water deficit in Merlot as well as in Tocai Friulano. The higher accumulation of VOCs compounds under water deficit may be the result of complex modulation of fatty acid degradation pathway genes, such as VviLOX and VviHPL1 observed here (**Supplementary Table S5**). Furthermore, consistent up-regulation of triacylglycerol lipase and phospholipase A2 transcripts suggests an additional role for storage lipid degradation on various VOC accumulation. In cucumber, phospholipase A2 has been demonstrated increase the susceptibility of lipid body membrane to lipolytic enzymes such as LOX and lipases via partial degradation of membrane proteins and associated phospholipid monolayer (Rudolph et al., 2011).

Indeed, silencing of TomloxC in tomatoes has been shown to increase C8 and C10 compounds (Orzaez et al., 2009) and decrease C5 and C6 ones in ripe fruits. This is probably due to higher precursor availability to other LOXs that may catalyze the production of C8 and C9 precursors (Senger et al., 2005). Shen et al. (2014) also demonstrated that the modulation of tomato HPL correlates with changes in C5 and C6 volatiles in fruits. Consistent induction of VviHPL1 from the onset of ripening onward may explain higher accumulation of VOC aldehydes under water deficit. Moreover, a role of LOXs and HPLs in the drought adaptation process has been already suggested in Arabidopsis (Grebner et al., 2013; Savchenko et al., 2014). While various chain length volatiles impart important flavor attributes (Baldwin et al., 2004; Ripoll et al., 2014), these compounds may also act as powerful signaling molecules activating abiotic stress response (Alméras et al., 2003; Yamauchi et al., 2015) and potentially hasten the developmental program in berries during water deficit. Both abiotic stress and ripening processes involves oxidative stress signals. Recent studies have demonstrated that (E)-2 hexenal, 2-butenal, and 3-hepten-2-one treatment in Arabidopsis seedlings can trigger large transcriptome changes involving abiotic stress genes and TFs, such as DREBs (Yamauchi et al., 2015). The possibility of VOCs in modulating maturationrelated transcriptome changes (e.g., senescence and ripening) in fruit tissues should not be discounted in light of these observations.

# CONCLUSION

Our results confirm several previously reported modulation of the primary and specialized metabolism and also provide new insight into the stilbenoid and volatile compounds response to water deficit. The integration with network analysis revealed major water deficit-regulated gene modules that are strongly linked to central and specialized metabolites as well as multiple signal transduction pathways (e.g., regulation of anthocyanin and amino acids via members of VviAP2/ERF and VviNAC TF families). Activation of both ABA-dependent and ABAindependent signaling pathway may also be critical to achieve a balance between the regulation of the stress response and the berry ripening program. Further functional analyses are needed to characterize the putative identified modulators of this metabolic response. This study represents a first step into understanding the transcriptional control and their downstream regulatory cascades in grapes or other fruits while providing an important resource for breeding opportunities, irrigation management, and comparative studies centered on reproductive tissue metabolism under abiotic stress in fruit crops.

# AUTHOR CONTRIBUTIONS

SS participated in the design of the study, carried out the specialized metabolite analyses, RNA extractions, part of the transcriptome data analysis, and drafted part of the manuscript; DCJW carried out part of the transcriptome data analysis, the network analysis, and drafted part of the manuscript; AD carried out the central metabolite analysis; JCH carried out the anthocyanin analysis; BB performed the field experiment; EP coordinated the field experiments; AF supervised the central metabolite analysis and critically revised the manuscript; FM participated in the design of the study, supervised the metabolite analysis, and critically revised the manuscript; SDC conceived the study, coordinated the experiments, supervised the field experiment, transcriptome analysis, and network analysis,

interpreted the results, and drafted part of the manuscript. All authors read and approved the final manuscript.

#### FUNDING

This study was funded by the European Territorial Cooperation program (Sustainable viticulture and improvement of the territorial resources of the grape and wine industry), the Fondazione Edmund Mach (GMPF Program), the COST Action FA1106 Quality Fruit, Genome British Columbia (10R21188), and the Natural Sciences and Engineering Research Council of Canada (10R23082).

#### ACKNOWLEDGMENTS

We would like to thank Panagiotis Arapitsas, Georg Weingart, Silvia Carlin, and Domenico Masuero for technical support in the metabolite analyses, and Federica Cattonaro and Mara Miculan of the Institute of Applied Genomics (Udine, Italy) for technical support in the RNA-sequencing analysis.

#### REFERENCES


#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2017.01124/ full#supplementary-material

TABLE S1 | List of genes and primer sequences assayed for expression by qPCR.

TABLE S2 | Impact of water deficit on grape productivity.

TABLE S3 | List of compounds identified in this study using GC-MS, UHPLC-MS/MS, HPLC-DAD, and HS-SPME-GC-MS platforms.

TABLE S4 | RNA sequencing analysis metrics.

TABLE S5 | Summary of differentially expressed genes and associated information at 26 (A), 53 (B), 67 (C), 81 (D), and 106 (E) days after anthesis. (F) TFs differentially expressed. (G) Information on TFs reported in Figure 7. (H) List of differentially expressed genes involved in ROS production, scavenging, and signaling pathways and associated information.

TABLE S6 | (A) Weighted correlation gene co-expression network analysis (WGCNA). (B) CRE sequence, length, and description, number of genes containing each CRE, total occurrence in the genome, enrichment value, and the list of genes containing each CRE. (C) Module-metabolite co-response matrix determined for each module.



to the production of the leaf aldehyde in plants. J. Biol. Chem. 291, 14023–14033. doi: 10.1074/jbc.M116.726687


rin) and ethylene receptor (Nr) mutants reveals novel regulatory interactions. Plant Physiol. 157, 405–425. doi: 10.1104/pp.111.175463


expression of anionic glutamate dehydrogenases to form glutamate for proline synthesis in tobacco and grapevine. Plant Cell 18, 2767–2781. doi: 10.1105/tpc. 105.038323


SlNAC4, functions as a positive regulator of fruit ripening and carotenoid accumulation. Plant Cell Physiol. 55, 119–135. doi: 10.1093/pcp/ pct162

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Savoi, Wong, Degu, Herrera, Bucchetti, Peterlunger, Fait, Mattivi and Castellarin. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

,

# Transcriptional Responses to Pre-flowering Leaf Defoliation in Grapevine Berry from Different Growing Sites, Years, and Genotypes

Sara Zenoni<sup>1</sup> \*, Silvia Dal Santo<sup>1</sup> , Giovanni B. Tornielli<sup>1</sup> , Erica D'Incà<sup>1</sup> , Ilaria Filippetti<sup>2</sup> Chiara Pastore<sup>2</sup> , Gianluca Allegro<sup>2</sup> , Oriana Silvestroni<sup>3</sup> , Vania Lanari<sup>3</sup> , Antonino Pisciotta<sup>4</sup> , Rosario Di Lorenzo<sup>4</sup> , Alberto Palliotti<sup>5</sup> , Sergio Tombesi5,6 , Matteo Gatti<sup>6</sup> and Stefano Poni<sup>6</sup>

<sup>1</sup> Department of Biotechnology, University of Verona, Verona, Italy, <sup>2</sup> Department of Agricultural Science, University of Bologna, Bologna, Italy, <sup>3</sup> Dipartimento di Scienze Agrarie, Alimentari e Ambientali, Università Politecnica delle Marche, Ancona, Italy, <sup>4</sup> Department of Agricultural and Forest sciences, University of Palermo, Palermo, Italy, <sup>5</sup> Dipartimento di Scienze Agrarie, Alimentari e Ambientali, Università di Perugia, Perugia, Italy, <sup>6</sup> Dipartimento di Scienze delle Produzioni Vegetali Sostenibili, Università Cattolica del Sacro Cuore, Piacenza, Italy

#### Edited by:

Giuseppe Ferrara, Università degli Studi di Bari Aldo Moro, Italy

#### Reviewed by:

Ana Margarida Fortes, Instituto de Biossistemas e Ciências Integrativas (BioISI), Portugal Massimiliano Corso, Université libre de Bruxelles, Belgium Christine Böttcher, Commonwealth Scientific and Industrial Research Organisation (CSIRO), Australia

> \*Correspondence: Sara Zenoni sara.zenoni@univr.it

#### Specialty section:

This article was submitted to Crop Science and Horticulture, a section of the journal Frontiers in Plant Science

> Received: 20 January 2017 Accepted: 07 April 2017 Published: 02 May 2017

#### Citation:

Zenoni S, Dal Santo S, Tornielli GB, D'Incà E, Filippetti I, Pastore C, Allegro G, Silvestroni O, Lanari V, Pisciotta A, Di Lorenzo R, Palliotti A, Tombesi S, Gatti M and Poni S (2017) Transcriptional Responses to Pre-flowering Leaf Defoliation in Grapevine Berry from Different Growing Sites, Years, and Genotypes. Front. Plant Sci. 8:630. doi: 10.3389/fpls.2017.00630 Leaf removal is a grapevine canopy management technique widely used to modify the source–sink balance and/or microclimate around berry clusters to optimize fruit composition. In general, the removal of basal leaves before flowering reduces fruit set, hence achieving looser clusters, and improves grape composition since yield is generally curtailed more than proportionally to leaf area itself. Albeit responses to this practice seem quite consistent, overall vine performance is affected by genotype, environmental conditions, and severity of treatment. The physiological responses of grape varieties to defoliation practices have been widely investigated, and just recently a whole genome transcriptomic approach was exploited showing an extensive transcriptome rearrangement in berries defoliated before flowering. Nevertheless, the extent to which these transcriptomic reactions could be manifested by different genotypes and growing environments is entirely unexplored. To highlight general responses to defoliation vs. different locations, we analyzed the transcriptome of cv. Sangiovese berries sampled at four development stages from pre-flowering defoliated vines in two different geographical areas of Italy. We obtained and validated five markers of the early defoliation treatment in Sangiovese, an ATP-binding cassette transporter, an auxin response factor, a cinnamyl alcohol dehydrogenase, a flavonoid 3-O-glucosyltransferase and an indole-3-acetate beta-glucosyltransferase. Candidate molecular markers were also obtained in another three grapevine genotypes (Nero d'Avola, Ortrugo, and Ciliegiolo), subjected to the same level of selective pre-flowering defoliation (PFD) over two consecutive years in their different areas of cultivation. The flavonol synthase was identified as a marker in the pre-veraison phase, the jasmonate methyltransferase during the transition phase and the abscisic acid receptor PYL4 in the ripening phase. The characterization of transcriptome changes in Sangiovese berry after PFD highlights, on one hand, the stronger effect of environment than treatment on the whole berry transcriptome rearrangement during development and, on the other, expands existing knowledge of the main molecular and biochemical modifications occurring in defoliated vines. Moreover, the identification of candidate genes associated with PFD in different genotypes and environments provides new insights into the applicability and repeatability of this crop practice, as well as its possible agricultural and qualitative outcomes across genetic and environmental variability.

Keywords: grapevine, pre-flowering defoliation, berry transcriptome, flavonoid, secondary metabolite

#### INTRODUCTION

fpls-08-00630 April 29, 2017 Time: 12:26 # 2

Viticulture is still strongly bound to the concept of terroir, relating the sensory attributes of wine to the environmental conditions in which the grapes are grown (Van Leeuwen and Seguin, 2006). Though a shared definition of terroir is still hard to find, there is general consensus that the main factors composing terroir are climate, soil, cultivar/rootstock and human practices and that these factors strongly interact (Renouf et al., 2010). Quantifying the relative importance of each factor influencing terroir is an extremely difficult task since the variability of all factors involved must be represented. A quite considerable effort was made by van Leeuwen et al. (2004) who concluded that climate, soil, and cultivar had a decreasing importance in influencing performance of the cultivars Merlot, Cabernet franc, and Cabernet Sauvignon grown in three different soil environments and observed over 5 years.

A first important consequence of the complex climate × soil × cultivar interactions is that the same cultivar grown in different environments can originate products of different composition and market value. The capacity of a genotype to modulate its phenotype under different environmental conditions is defined phenotypic plasticity, a phenomenon of considerable interest in plant physiology. Over the last decade, a number of studies exploring metabolomic and transcriptomic bases of phenotypic plasticity in Vitis vinifera have been conducted in local cultivars such as Corvina and Garganega (Dal Santo et al., 2013a, 2016b; Fernie and Tohge, 2013; Anesi et al., 2015). These works demonstrated the direct effect of growing conditions on gene expression during berry ripening, allowing several environmentally modulated genes to be identified, including many belonging to the phenylpropanoid/flavonoid pathway (Dal Santo et al., 2013a, 2016b). The existence of a terroir-specific effect on berry transcriptome and metabolome was also revealed, which persists over several vintages (Anesi et al., 2015), and specific plastic transcripts were associated with groups of vineyards sharing common viticulture practices (Dal Santo et al., 2013a).

If it is agreed that human practices are an important component of the terroir concept (Renouf et al., 2010), forecasting their effects on grape composition and wine can greatly benefit from associating omics tools to traditional agronomic assessment. Among the different operations pertaining to grapevine canopy management, pre-flowering leaf removal is likely the one that has received the greatest interest from the scientific community over the last decade. Starting with the original work (Poni et al., 2006), a number of subsequent studies, representing a broad array of cultivars and environments (Diago et al., 2012; Gatti et al., 2012, 2015; Palliotti et al., 2012; Lee and Skinkis, 2013; Pastore et al., 2013; Risco et al., 2014; Komm and Moyer, 2015; Sternad Lemut et al., 2015; Sivilotti et al., 2016), have confirmed the technique to be valuable and repeatable for: (i) reducing vine yield through a decrease in fruit-set and/or berry size; (ii) decreasing cluster compactness, hence susceptibility to rot diseases, and (iii) improving grape composition in terms of total soluble solids (TSSs), phenolic and aroma compounds. However, it was also observed that the outcome of pre-flowering defoliation (PFD) could be quite variable between consecutive years (Gatti et al., 2015) and that different cultivars or the same cultivar in different environments could be influenced differently by the treatment (Kuhn et al., 2014).

Although the practice is widely used in viticulture, very little molecular information is available and, as a consequence, the definition of common mechanisms linking the impact of leaf removal to berry physiological and metabolic responses, is far from complete.

The pioneer study conducted on genome-wide expression analysis in cv. Sangiovese vines subjected to either pre-flowering or late season (i.e., at veraison) defoliation revealed a general delay in transcriptional ripening following both treatments (Pastore et al., 2013). Moreover, a more extensive transcriptome rearrangement in berries subjected to PFD was observed, which reflects the uncoupling of metabolic processes, in particular anthocyanin and flavonol synthesis, from the general ripening program (Pastore et al., 2013). A very recent study performed on Sauvignon blanc shed more light on grapevine response to an altered microclimate due to early leaf removal (Young et al., 2016). When main and lateral leaves were removed from the cluster zone at fruit-set in order to induce and maintain berry light exposure, higher levels of carotenoids and volatile terpenoids were found in the berries, in two consecutive years. The study also clearly demonstrated that the main physiological responses occur in the early stages of berry development, when berries are still photosynthetically active (Palliotti and Cartechini, 2001), and that the key response is the change in pigment levels and metabolite pools that have photoprotective and/or antioxidant functions (Young et al., 2016). Overall, it is clear that early defoliation combined with environmental conditions affects berry composition through changes in gene expression.

The complexity involved in the reprogramming of berry transcriptome, proteome, and metabolome during development has been progressively described in different grapevine varieties (Deluc et al., 2007; Zamboni et al., 2010; Fasoli et al., 2012; Agudelo-Romero et al., 2013; Dal Santo et al., 2013a; Anesi et al., 2015), demonstrating that a large part of metabolic changes characterizing berry formation and ripening are under transcriptional control. It is also known that grape berry

development involves the integration of multiple hormonal signals, with some hormones acting as promoters and others as repressors. In particular, in non-climacteric fruits, such as grape, where no burst in ethylene production is observed during ripening, the abscisic acid (ABA) seems to play a stronger role during ripening and its crosstalk with other growth regulators has been proposed at different berry stages (Davies and Bottcher, 2009; McAtee et al., 2013; Fortes et al., 2015). Despite the amount of information already reported, hormonal control in grape ripening is still poorly understood (Fortes et al., 2015). In this context, the identification of molecular markers, addressing the question of how stable and replicable is the link of PFD to favorable physiological and metabolic changes in berry, represents a huge challenge.

In this work, a comparative study of the agronomic and molecular berry responses to PFD was performed in four genotypes grown in different areas of cultivation over two consecutive years, with the aim of identifying genes whose expression could be attributable to this viticulture practice, regardless of site, year, and genotype. Molecular responses in Sangiovese berries during development from defoliated and untreated control vines in two different growing sites were investigated by a genome wide expression analysis. The expression profiles of selected candidate genes were assessed by qPCR in all experimental conditions and integrated with agronomic and ripening parameters, to unveil developmental and metabolic processes commonly affected in berries after PFD.

## MATERIALS AND METHODS

#### Plant Material, Experimental Layout, and Berry Sampling for Gene Expression Analyses

Berry samples for subsequent transcriptomic analyses and real time qPCR analyses were taken from mature and healthy vineyards located in Emilia Romagna (cvs Sangiovese and Ortrugo), Umbria (cv. Ciliegiolo), Marche (cv. Sangiovese), and Sicily (cv. Nero d'Avola), Italy. Sangiovese plots in Emilia Romagna and Marche shared the same rootstock (S.O.4.), whereas clones were different: clone R24 and clone SG12T, respectively. All vineyards were standard either cane or spur pruned vertically shoot positioned (VSP) trellises. Single vine spacing within row varied between 0.8 and 1.5 m, whereas between-row spacing was between 2.5 and 3.3 m, resulting in a vine density varying from 2020 to 5000 vines/hectare. More details regarding trellis structure, bud load, soil characteristics, climate trends and canopy and vineyard management practices can be found in Filippetti et al. (2011), Alagna et al. (2014), Gatti et al. (2015), and Silvestroni et al. (2016).

In 2012 and 2013, in each site × cultivar combination, two treatments were compared consisting of PFD performed at the "separated closed flowers" stage (Baggiolini, 1952) by removing the six basal main leaves of all shoots on each test vine (varying from 6 to 12 according to site; Supplementary Figure 1A) while any lateral shoot emerging from the same basal nodes at the time of defoliation was retained. PFD was compared with a non-defoliated control treatment (C).

Using an identical sampling protocol, berry sampling at each site × cultivar combination was performed on both control and PFD treatments at four development stages as follows: 20 days after leaf removal (Stage 1); hard and green berries at veraison (i.e., 1–5% slightly colored berries in a cluster) (Baggiolini, 1952) (Stage 2); soft, yet still not colored berries at veraison (Stage 3), berries at a TSS concentration of about 18◦Brix (Stage 4). On each sampling date, a batch of 60 berries was collected. In detail, three independent pools of 20 berries each were collected from clusters of different vines in order to create three biological replicates that represent almost the entire variability of the experimental design. The sampling was performed by carefully cutting each berry at the pedicel with scissors in order to avoid any damage or juice loss. Berries were immediately frozen in liquid nitrogen and then shipped to labs at the University of Verona (northern Italy) for transcriptome processing and expression analyses. Given the typical asynchrony in individual berry ripening, berry sampling for Stages 2 and 3 were performed on the same date for both treatments.

In total for the transcriptomic analysis on Sangiovese, the experiment entailed the collection and analysis of 48 berry samples (4 stages × 2 treatments × 2 sites × 3 biological replicates). For the real time qPCR on all site × cultivar combinations, the experiment entailed the collection and analysis of 240 berry samples (4 stages × 2 treatments × 5 sites × 3 biological replicates × 2 years).

### Vegetative Growth and Yield Components

For every site × cultivar combination vegetative growth capacity was expressed by estimated total final leaf area per vine and measured pruning weight. Total leaf area per vine was estimated by node counts and surface area of fully expanded main and lateral leaves (Lopes and Pinto, 2015), whereas 1-year pruning weight per vine was taken soon after leaf shedding in fall was completed. At harvest, each year, total yield and cluster number per vine were recorded, and mean cluster weight calculated accordingly. Single berry weight was taken on the samples then processed for must analyses and total berry number calculated from mean cluster weight. For more details on sample size and sampling procedures, please refer to the papers cited above. Source-to-sink balance was expressed as leaf area-to-yield ratio.

### Must Composition and Phenolic Compounds Analyses

Four-to-six 100-berry samples were taken pre-harvest from each genotype by treatment combination in the different test sites. The 100-berry sample was composed by five berries taken for a total of 20 clusters; two berries were sampled from the top portion or wings, two from the middle and one from the tip of the cluster in order to account for within cluster variability in ripening.

TSS concentration, pH, and titratable acidity (TA) were determined on must samples according to standard methods described in Iland et al. (2011).

Total anthocyanin concentration (mg/kg of fresh berry mass) was determined according to Iland (1988).

Flavonol compounds were extracted from grape skins as reported by Downey and Rochfort (2008). In brief: 0.100 g of lyophilized grape skins were extracted in 1.0 mL of 50% (v/v) methanol in water for 20 min with sonication. The extracts were centrifuged (5 min at 10000 × g at 4 ◦C), filtered through a 0.22 µm polypropylene syringe for HPLC analysis and transferred to HPLC auto-sampler vials.

The chromatographic method was developed using an Agilent 1260 Infinity Quaternary LC (Agilent Technology, Santa Clara, CA, USA) consisting of a G1311B/C quaternary pump with inline degassing unit, G1329B autosampler, G1330B thermostat, G1316B thermostatted column compartment and a G4212B diode array detector fitted with a 10 mm path, 1 µL volume Max-Light cartridge flow cell. The instrument was controlled using Agilent Chemstation software version A.01.05. Separation was achieved on a reverse-phase C-18 Synergi Hydro RP 80A, 250 mm × 4.6 mm, 4 µm (Phenomenex, Torrance, CA, USA). The solvents used were 5% (v/v) formic acid (solvent A) and acetonitrile (solvent B). The flow rate was 0.5 mL/min, with a linear gradient profile consisting of solvent A with the following proportions (v/v) of solvent B: 0–10 min, 2–10% B; 10–25 min, 10–12% B; 25–35 min, 12–30% B; 35–43 min, 30% B; 43–48 min, 30–40% B; 48–52 min, 40–50% B; 52–55 min, 50–60% B; 55–58 min, 60–98% B; 58–63 min, 98% B; 63–66 min, 98–2% B; 66–72 min 98% B. The column temperature was maintained at 40 ± 0.1◦C. Five microliters of sample extract was injected. The elution was monitored at 200–700 nm, detection by UV-Vis absorption with DAD scanning between 280, 320, and 370 nm. Anthocyanins and flavonols were identified using authentic standards and by comparing the retention times. Quantification was based on peak areas and performed by external calibration with standards.

### Statistical Analyses

A completely randomized block design was used and the agronomic parameters and must composition were subjected to analysis of variance (ANOVA, SAS statistical software, SAS Institute, Cary, NC, USA) and mean separation performed by t-test. In other cases, variability across treatments was expressed as mean ± standard error (SE).

#### RNA Extraction

Total RNA was isolated from approximately 400 mg of berry pericarp tissue (i.e., entire berries without seeds) using the SpectrumTM Plant Total RNA kit (Sigma–Aldrich), with modifications as described in Dal Santo et al. (2016a). Seeds were manually removed from the 20 berries of each biological replicate before the liquid nitrogen grinding procedure. RNA quality and quantity were determined using a Nanodrop 2000 spectrophotometer (Thermo Scientific, Wilmington, DE, USA) and a Bioanalyzer Chip RNA 7500 series II (Agilent, Santa Clara, CA, USA).

# Microarray Analyses and Statistical Approaches

We hybridized 5 µg of total RNA per sample to a NimbleGen microarray 090818\_Vitus\_exp\_HX12 chip (Roche, NimbleGen Inc., Madison, WI, USA), representing 29,549 predicted genes on the basis of the 12X grapevine V1 gene prediction version. The hybridization was performed according to the manufacturer's instructions (Dal Santo et al., 2016a). Statistical analysis of the microarray data was conducted using TMeV v4.8 (mev.tm4.org/). Statistical analysis of microarrays (SAM) was conducted with a false discovery rate (FDR) of 0.1% and ANOVA using α = 0.01 and standard Bonferroni correction in order to skim off genes that showed a high variability among the three biological replicates. Correlation matrixes were prepared using R software and Pearson's correlation coefficient as statistical metric to compare the values of the whole transcriptome (29,549 genes) in all analyzed samples. Correlation values were converted into distance coefficients to define the height scale of the dendrogram. Principal component analysis (PCA) was conducted using SIMCA P+ v13 (Umetrics, USA) and applied to the significantly modulated transcripts dataset (18,771 genes). Differentially modulated genes at each developmental stage in both sites were retrieved by performing a between-subjects (C vs. PFD samples) t-test (α = 0.01), assuming equal variance among samples. Gene Ontology (GO) enrichment analysis was performed with the AgriGO online software<sup>1</sup> , using Singular Enrichment Analysis (SEA) tool and Fisher's as statistical test method (Du et al., 2010). Heat maps were created using log2 transformed expression values and then median-centered by transcript. Cluster analysis was conducted on transcript mediancentered fluorescent values by the k-means method (KMC) with Pearson's correlation distance. We used the Figure of Merit (FOM) statistic to determine the optimal number of clusters (n = 10).

## Reverse Transcription (RT) and Real-Time qPCR

One microgram of total RNA was treated with DNase I (Promega) according to the instructions provided with the commercial kit. DNase treated RNA was then used for cDNA synthesis using the Improm-II TMReverse Transcriptase (Promega) following the producer's indications. The transcriptional profile was analyzed by real-time RT-PCR as described by Zenoni et al. (2011), using the SYBR Green PCR master mix (Applied Biosystems) and a Mx3000P real-time PCR system (Stratagene). Each expression value, relative to VvUBIQUITIN1 (VIT\_16s0098g01190), widely used as a suitable and robust reference gene during berry development (Chen et al., 2013; Dal Santo et al., 2013b, 2016b; Cramer et al., 2014), was determined in triplicate. Non-specific PCR products were identified by the dissociation curves. Amplification efficiency was calculated from raw data using LingRegPCR software (Ramakers et al., 2003). The mean normalized expression (MNE)-value was calculated for each sample referred to the ubiquitin expression

<sup>1</sup>http://bioinfo.cau.edu.cn/agriGO

Zenoni et al. Grape Responses to Defoliation

according to the Simon equation (Simon, 2003). Standard error (SE)-values were calculated according to Pfaffl et al. (2002). The primer sequences used in qPCR analysis are listed in Supplementary Table 1.

# Analysis of Correlation between Microarray and qPCR

Correlation between the microarray and qPCR results was performed for the six putative molecular markers of the PFD treatment in Sangiovese for the year 2012, and the statistical significance of this correlation determined. For the NimbleGen microarray, the data input into the correlation analysis was the Log2 value of the average of the three biological replicates for each gene × site × treatment combination. For qPCR, we used the mean Log2 ratio value reported by qPCR from all replicate. Prior to performing correlation analyses, the data were tested for normality using the Shapiro–Wilk test, as indicated by Morey et al. (2006). Because the data were not normally distributed, Spearman's Rho, instead of Pearson's correlation, was computed using R software. The calculated correlation coefficient was 0.4617597 (Spearman's Rho, p = 2.185e-06, n = 96). By normal standards [n = (96 – 2) = 94], the correlation between the NimbleGen microarray data and qPCR data for the indicated six genes, would be considered statistically significant (p < 0.005).

#### Accession Numbers

Grape berry microarray expression data are available in the Gene Expression Omnibus under the series entry GSE92980<sup>2</sup> .

# RESULTS

#### Impact of Pre-flowering Defoliation on Agronomic Parameters and Berry Transcriptome of cv. Sangiovese under Two Growing Conditions

In order to provide a preliminary evaluation of site × early defoliation interaction, the cv. Sangiovese was subjected in 2012 to the pre-flowering defoliation treatment (PFD) using the same protocol in Ancona (AN) and Bologna (BO) and, within each site, a non-defoliated control treatment (C) was also included.

Daily maximum, minimum, and mean air temperatures (T) as well as daily rainfall for the two locations evaluated from the 1 April until 30 September, showed some common features for the two sites, with a quite cool spring and long summer (Supplementary Figure 1B).

Agronomic parameters showed that the two sites shared significant differences for total leaf area and yield per vine, cluster weight, and berries per cluster between PFD and C treatments, with PFD showing lower values than C vines in all cases (**Table 1**). The remaining parameters, including the source–sink balance expressed as leaf area-to-yield ratio were either unchanged or slightly enhanced in AN, as concerns TA and total anthocyanins concentration.

To investigate the molecular changes that take place after defoliation during berry development, Sangiovese berry transcriptomes of PFD and C vines were compared at the four berry developmental stages in both sites (**Figure 1A**). A dendrogram of the global transcriptomic data revealed enough uniformity among the three biological replicates in the C and PFD samples at each time point (**Figure 1B**). The main separation among samples is related to berry stage, with Stage 1 resulting as the most divergent. Indeed, Stage 2 showed more similarity at the transcriptional level to Stages 3 and 4, suggesting that, despite berries still being hard and green at Stage 2, many molecular processes related to ripening were already activated. Interestingly, the second variable that strongly influenced sample association was site. In fact, except for Stage 1, AN and BO samples were characterized by distinctive berry transcriptomes during ripening. This evidence highlights the strong effect of growing conditions on berry transcriptome in Sangiovese and suggests that this effect is more evident when the ripening program is initiated.

Concerning the effect of the defoliation treatment on berry transcriptome we found that PFD and C vines were distinguishable only at specific combinations of berry stage and site. In fact, the separation between PFD and C is evident at Stage 1 in AN, at Stage 2 in BO, at Stage 3 in AN and at Stage 4 in BO. These results suggest that, in Sangiovese, PFD has a weaker effect on berry transcriptome than growing conditions.

To retrieve genes differentially modulated under our experimental conditions, the berry transcriptome dataset was screened by significance analysis of microarrays (SAM, 16 groups, FDR = 0.1%). Analysis of variance (ANOVA, 16 groups, α = 0.01, standard Bonferroni correction) was applied to transcripts positive in the previous SAM experiment in order to skim off the most significantly modulated transcripts. We obtained a reduced dataset of 18,771 genes (Supplementary File 1), which was inspected by PCA analysis. The two principal components, explaining 56.5% of the total dataset variability, allowed berry samples to be clearly separated on the basis of their developmental stage (**Figure 1C**). Sample distribution confirmed that Stage 2 is more similar to Stage 3 at both sites; moreover, it can be observed that, at transcriptional level, the ripening process (Stages 3 and 4) at AN is slightly advanced in comparison to BO (**Figure 1C**). Interestingly, principal component 3 (PC3), explaining 11.8% of the total dataset variability, clearly separated AN from BO samples, in particular after veraison, again evidencing the strong effect of growing conditions on the berry transcriptome rearrangement during ripening in cv. Sangiovese (Supplementary Figure 2). At no stage nor in either site were principal components found that separated PFD samples from C ones.

Notwithstanding the small effect of PFD on berry transcriptome, we focused on the identification of differentially expressed genes (DEGs) after PFD, regardless of the growing site. We then compared the PFD and C berry transcriptomes at each time point using a t-test (between subjects t-test, α = 0.01) for both sites separately. The number of DEGs identified between PFD and C vines at each stage was different in the two sites

<sup>2</sup>https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE92980

#### TABLE 1 | Agronomic and ripening parameters 2012.


Vegetative growth, yield parameters, and grape composition recorded in 2012 on cv. Sangiovese grapevines grown in Ancona and Bologna and subjected, within each location, to a pre-flowering defoliation (PFD) or undefoliated (Control). <sup>∗</sup>P < 0.05 and ∗∗P < 0.01, respectively.

(**Figure 2A** and Supplementary File 2). In particular, at Stage 1 and 3 a higher number of DEGs characterized AN, whereas at Stage 2 and 4 a higher number of DEGs was found in BO. These differences well mirrored the behavior observed in the dendrogram analysis. A total of 1746 and 1041 DEGs in at least one stage between PFD and C vines, were found in BO and AN, respectively.

By comparing the list of DEGs from BO and AN, only 125 genes were identified as differentially expressed at both sites (**Figure 2B** and Supplementary File 3). The GO enrichment analysis performed on the three group of DEGs, i.e., BO-specific, AN-specific and common, revealed a significant overrepresentation of the "response to stimulus" functional category in all groups of DEGs. Genes belonging to this functional category may be involved in the detection and response to external and endogenous stimuli and also to many stresses, such as biotic and abiotic stress, redox state, and others. Regarding the common DEGs, another two functional categories resulted as significantly overrepresented, the "biosynthetic process," and "cellular amino acid and derivative metabolic process," represented by an aspartate aminotransferase, two glycine hydroxymethyltransferases, a glutamine synthetase, and a serine hydroxymethyltransferase.

#### Identification of Berry Molecular Markers Associated to Pre-flowering Defoliation in Sangiovese

In order to identify putative molecular markers associated to the PFD in cv. Sangiovese, independently of site, we focused on the 125 DEGs shared by BO and AN. We checked the expression profile of the 125 genes during berry development on C vines in the two sites, in order to identify genes whose expression was not or slightly influenced by the growing conditions. A KMC clustering analysis identified 10 expression clusters and revealed a very high expression variability of these genes in C vines growing in the two sites, with only 38 genes belonging to same clusters of expression during development in BO and AN (**Figure 3A** and Supplementary File 4). Subsequently, we evaluated the pattern of Fold change (FC) between PFD and C at each stage in both sites, and we found that only 11 genes were affected by the PFD in a similar manner throughout berry development in BO and AN (**Figure 3B**). Among these genes we selected six characterized by an upregulation at all stages after the PFD treatment or by an upregulation till Stage 3 and a downregulation at Stage 4 (**Figure 3C**). The selected genes were represented by the ATP-binding cassette (ABC) transporter VvPDR20/VvABCG50 (VIT\_06s0061g01490), auxin response factor (ARF) 10 (VIT\_13s0019g04380), cinnamyl alcohol dehydrogenase (CAD; VIT\_00s0615g00020), flavonoid 3-O-glucosyltransferase (G3T; VIT\_11s0052g01630), geraniol 10-hydroxylase (G10H; VIT\_02s0012g02820), and indole-3 acetate beta-glucosyltransferase (IND; VIT\_13s0019g03040). The expression of these genes were validated by qPCR on PFD and C berries in the two sites at the four berry developmental stages (**Figures 3D–I**). The VvPDR20/VvABCG50 showed a peak of expression at Stage 2 in C berries throughout development, whereas in PFD berries its expression significantly increased at Stage 3 in both sites (**Figure 3D**). Although very few functional studies have been performed on ABC transporters activity in grape, a role in the vacuolar localization and transport of glucosylated anthocyanidins was demonstrated for the member VvABCC1 (Francisco et al., 2013). The expression profiles obtained by our analysis suggest a role of VvPDR20/VvABCG50 at the onset of ripening and evidenced that PFD delays its expression during berry development. In the case of the ARF the increase in expression at Stage 4 in C condition is significantly hastened by the PFD treatment at Stage 3 in both sites (**Figure 3E**). The CAD gene is characterized by a significant increase of expression at Stage 3 in PFD condition instead of the Stage 4 observed in C vines in AN, and by a significant increase at Stage 2 in PFD in both sites, particularly in BO (**Figure 3F**). This

FIGURE 1 | Whole transcriptome analysis of Sangiovese berries subjected to PFD treatment in two different sites. (A) Schematic representation of the sampling design used. (B) Cluster dendrogram of the whole transcriptome dataset in all analyzed samples. Pearson's correlation values were converted into distance coefficients to define the height of the dendrogram. Samples are colored according to the developmental stage of sampling. (C) Score scatterplot (PC1 vs. PC2) of the PCA model (9 Principal Components, R 2 (cumulative) = 0.903, Q<sup>2</sup> (cumulative) = 0.848) applied to the significantly modulated transcripts dataset. Samples are colored according to the developmental stage of sampling. Different treatments are indicated by different symbols, "✩" = Control and "O" = Pre-flowering defoliation.

behavior well mirrored the trend observed in the dendrogram, showing that the separation between PFD and C is more evident at Stage 2 in BO, and at Stage 3 in AN. The G3T was characterized by an increase in expression at Stage 4 in C condition in both growing sites. For this gene the PFD treatment led to a significant increase of expression level throughout berry development, in particular at Stage 2 in BO and Stage 3 in AN, similarly to the CAD (**Figure 3G**). The G10H showed a peak of expression at Stage 3 in C vines in both growing sites. The PFD treatment enhances the expression level of this gene at Stage 3 but also significantly hastens its induction at Stage 2 in both BO and AN (**Figure 3H**). Concerning the IND, involved in the regulation of auxin levels by IAA conjugation (Bottcher et al., 2010; Fortes et al., 2015), a flat expression trend during the first stages of berry development with a slight downregulation at Stage 4 was found in BO C vines, whereas a high expression was observed at Stages 1

FIGURE 3 | PFD treatment molecular markers of Sangiovese cultivar selection and real-time qPCR validation in 2012. (A) Heat map representing the fluorescence intensity of C vines in the 125 commonly modulated genes. KMC analysis was used to determine the transcripts with unaltered expression between Bologna and Ancona sites (highlighted as same cluster) and those with different expression (highlighted as different cluster). (B) Schematic representation of the Fold Change (FC), calculated between C and PFD vines at each developmental stage in Ancona and Bologna, in the 38 same cluster transcripts found in (A). The black arrows indicate the 11 genes showing a similar trend of FC. (C) FC between C and PFD vines at each developmental stage in Ancona and Bologna in a selection of six transcripts. (D–I) Real-time qPCR validation of the (D) ABC transporter VvPDR20-VvABCG50 (ABC; VIT\_06s0061g01490), (E) Auxin response factor 10 (ARF; VIT\_13s0019g04380), (F) Cinnamoyl alcohol dehydrogenase (CAD; VIT\_00s0615g00020), (G) Flavonoid 3-O-glucosyltransferase (G3T; VIT\_11s0052g01630), (H) Geraniol 10-hydroxylase (G10H; VIT\_15s0048g01490) and (I) Indole-3-acetate beta-glucosyltransferase (IND; VIT\_13s0019g03040) expression profiles in PDF and C Sangiovese vines during berry development in 2012. The mean normalized expression (MNE)-value was calculated for each sample referred to the VvUBIQUITIN1 (VIT\_16s0098g01190) expression according to the Simon equation (Simon, 2003). Bars represent means ± SE of three biological replicates. The significant modulation (t-test, p < 0.05) of gene expression between C and PFD berries at each stage per each site is indicated by an asterisk, red for BO and green for AN.

and 2 in AN, followed by a decrease in expression after veraison. The PFD induced a significant increase of IND expression in both sites but at Stage 2 in BO and Stage 3 in AN (**Figure 3I**).

Overall real-time qPCR analysis confirmed the microarray expression profiles for all the selected genes, demonstrating that these genes could be considered putative molecular markers of PFD in berry throughout development in cv. Sangiovese in 2012.

In order to investigate the influence of year on the PFD effect, pre-flowering leaf removal was also applied at both sites in 2013 using the same protocol. Seasonal weather data as daily air temperature and rainfall are shown in Supplementary Figure 1B.

Leaf area and yield per vine, cluster weight, and berries per cluster were again significantly lower in PFD vines, whereas pruning weight per vine and leaf area-to-yield ratio were not modified (**Table 2**). A striking berry size reduction was recorded in BO, while TSS were notably higher in PFD in both locations. However, while increased TSS did not achieve a concurrent increase in anthocyanins at the AN site, total anthocyanins were significantly higher in the defoliated vines at BO (**Table 2**).

Berry samples were collected following the same protocol used in 2012. The expression of the six genes identified as markers of PFD was analyzed on C and PFD berries collected during 2013 by the qPCR approach. The ABC transporter gene (ABC) showed the same expression profile as that revealed in 2012 in C and PFD vines, with a with a significant induction at Stage 3 in both PFD vines and a very high expression during PFD berry development in AN (**Figure 4A**). The ARF expression was confirmed in 2013 in C and PFD berries in both sites (**Figure 4B**), as well as the effect of PFD on CAD expression (**Figure 4C**). The expression of CAD in C berries was instead slightly different from 2012 in BO, with a clear decrease of expression from Stage 1 throughout berry development not observed the year before. The effect on G3T expression in PFD berries also resulted as the same in the 2 years, again with the significant and stronger effect at Stage 2 in BO and Stage 3 in AN (**Figure 4D**). On the contrary, the expression of GH10 gene was strongly affected by year in C berries and not influenced by PFD treatment in either site (**Figure 4E**). Lastly, beside the different expression in C vines at Stage 2 in AN, IND gene showed the same trend of significant induction as in 2012 due to the PFD treatment in both sites (**Figure 4F**).

Overall, the five genes, ABC, ARF, CAD, G3T, and IND resulted as being putative molecular markers of PFD treatment in Sangiovese during berry development, independently of growing site and year.

#### Identification of Putative Molecular Markers of Pre-flowering Defoliation in Different Genotypes

In order to evaluate if the five putative Sangiovese marker genes could also represent molecular markers of PFD for other genotypes cultivated in different environments, we applied PFD using the same protocol adopted for Sangiovese, on Nero d'Avola (ND), Ortrugo (OR), and Ciliegiolo (CI) cultivars, cultivated during 2012 and 2013in three different Italian areas, Palermo (PA-Sicily), Perugia (PE-Umbria) and Piacenza (PI-Emilia Romagna), respectively.

The environmental parameters recorded during 2012 and 2013 at the three sites are reported in the Supplementary Figure 3. In general, for sites located in north and central Italy (PI and PE) yearly weather patterns and rainfall showed some common features while the Sicilian location (PA) had a quite different trend. As observed for BO and AN, 2012 was marked by a quite cool spring and long summer (mid-June till end of August) in PI and PE, with hot spells reaching 40◦C and very limited rainfall (Supplementary Figures 1, 3). In PA, both seasons showed a more progressive increase in air temperature peaking around 40◦C in 2012 with basically no rainfall. 2013 was slightly cooler with some rain falling at the end of the season.

In 2012, yield per vine was significantly reduced by PFD regardless of cultivar, although responsiveness of some yield components showed variability (i.e., unchanged berry weight in Ciliegiolo and unchanged cluster weight and berries per cluster in Ortrugo) (**Table 3**). TSS at harvest were always increased by PFD in all cultivars and the same response was seen for total anthocyanins in cvs Ciliegiolo and Nero d'Avola. Except for the white cv. Ortrugo, must pH and TA were less responsive overall, whereas final LA/yield ratio was in general slightly enhanced in PFD.

In 2013, yield per vine and cluster weight were reduced in PFD regardless of location (**Table 4**). In agreement with the 2012 response, TSS and total anthocyanins were higher in the defoliated vines across all cultivars. Must pH and TA confirmed their relatively low sensitivity to the applied treatment, whereas moderately higher LA/yield ratios were again found in the PFD vines.

Berry samples for gene expression analysis were collected from C and PFD vines at the same four phenological stages used for Sangiovese. The real-time qPCR of the five Sangiovese PFD molecular markers performed on berries collected during 2012 showed that the five genes were not differentially modulated nor showed different expression profiles during berry development in Nero d'Avola, Ortrugo, and Ciliegiolo subjected to PFD (data not shown). These results demonstrated that these genes could not be considered molecular markers of PFD for other genotypes and/or for other environmental conditions. The transcriptomic dataset of 18771 genes obtained for Sangiovese was therefore inspected again with the aim of finding genes consistently modulated by PFD at the same stage in both sites, without necessarily showing a similar expression profile throughout the entire berry development. By using a threshold of | FC | > 2 between C and PFD we found that 6 genes were consistently modulated by PFD at Stage 1, 28 genes at Stage 2, 39 at Stage 3, and 19 at Stage 4 in both sites (Supplementary File 5). Among genes modulated at Stage 1, 5 resulted as upregulated by PFD and only 1, an unknown protein, downregulated. The flavonol synthase (VIT\_18s0001g03470) resulted as the most induced by the treatment in both sites and was therefore selected for further transcriptional investigation in other terroirs, years, and genotypes (Supplementary File 5). At Stage 2, 16 genes resulted as commonly upregulated by PFD, including the already analyzed geraniol 10-hydroxylase (G10H), some genes involved in carbohydrate metabolism, the MADS-box AGL20 (VIT\_15s0048g01240) and jasmonate

#### TABLE 2 | Agronomic and ripening parameters 2013.

fpls-08-00630 April 29, 2017 Time: 12:26 # 11


Vegetative growth, yield parameters, and grape composition recorded in 2013 on cv. Sangiovese grapevines grown in Ancona and Bologna and subjected, within each location, to a PFD or undefoliated (Control). <sup>∗</sup>P < 0.05 and ∗∗P < 0.01, respectively.

O-methyltransferase (VIT\_18s0001g12890). Among the 12 genes downregulated by the treatment we found three chitinases, one pathogenesis-related protein and the indol-3-acetic acid amino synthetase (Supplementary File 5). We selected the MADS-box

PFD berries at each stage per each site is indicated by an asterisk, red for BO and green for AN.

AGL20 and jasmonate O-methyltransferase for further analysis. Interestingly, at Stage 3 all the 39 commonly modulated genes resulted as upregulated by PFD. Among these genes at least 10 terpene synthases were found, together with three multidrug

equation (Simon, 2003). Bars represent means ± SE of three biological replicates. The significant modulation (t-test, p < 0.05) of gene expression between C and

#### TABLE 3 | Agronomic and ripening parameters 2012.

fpls-08-00630 April 29, 2017 Time: 12:26 # 12


Vegetative growth, yield parameters, and grape composition recorded in 2012 on cv. Nero d'Avola, Ortrugo, and Ciliegiolo grapevines grown in Sicily, Emilia Romagna, and Umbria, respectively, and subjected, within each location, to a PFD or undefoliated (Control). <sup>∗</sup>P < 0.05 and ∗∗P < 0.01, respectively.

#### TABLE 4 | Agronomic and ripening parameters 2013.


Vegetative growth, yield parameters, and grape composition recorded in 2013 on cv. Nero d'Avola, Ortrugo, and Ciliegiolo grapevines grown in Sicily, Emilia Romagna, and Umbria, respectively, and subjected, within each location, to a PFD or undefoliated (Control). <sup>∗</sup>P < 0.05, ∗∗P < 0.01, and ∗∗∗P < 0.001, respectively.

resistance-associated proteins, three serine carboxypeptidases and two genes involved in jasmonate metabolism, the VvJAZ2 and jasmonate O-methyltransferase, upregulated also at Stage 2 (Supplementary File 4). For further investigation, we chose one multidrug transporter (ABC- VIT\_09s0020g05380), the linalool synthase VvTPS62 (VIT\_00s0572g00020) and jasmonate O-methyltransferase. Finally, the 19 genes commonly modulated by PFD at Stage 4 all resulted as downregulated in comparison to the C vine. Among these genes we found three kinases, two NAC transcription factors and the ABA receptor PYL4 (VIT\_08s0058g00470) that were the most downregulated (Supplementary File 5). The latter was chosen for the next investigation.

The expression of the seven selected genes was investigated by qPCR in C and PFD berries of Nero d'Avola, Ortrugo, and Ciliegiolo during 2012, only at the corresponding stage. We found that flavonol synthase at Stage 1, jasmonate O-methyltransferase at Stages 2 and 3 and the ABA receptor PYL4 at Stage 4 confirmed the modulation of expression obtained by microarray analysis in Sangiovese in BO and AN (**Figure 5**). Instead, the MADS-box AGL20, ABC transporter and VvTPS62 were not commonly modulated by PFD in all genotypes (data not shown). The expression of flavonol synthase, jasmonate O-methyltransferase and the ABA receptor PYL4 was then evaluated by qPCR in C and PFD berries of all genotypes during 2013. They all showed the same modulation after treatment in all genotypes and growing conditions (**Figure 6**), resulting as good candidates for molecular markers of the PDF treatment.

Concerning the common upregulation of flavonol synthase, we determined main berry flavonols at harvest in both 2012

TABLE 5 | Main flavonols concentration.


expression according to the Simon equation (Simon, 2003). Bars represent means ± SE of three biological replicates. All genes in all genotypes resulted significantly

Concentration of quercetin, myricetin, and kaempferol recorded at harvest in berries of cvs Ortrugo and Sangiovese (Bologna site) vines subjected to PFD or undefoliated (control) in 2012 and 2013. Data are given as mean ± standard error (SE). Mean separation within row and parameter by t-test, P < 0.05. Absence of letters means ns. <sup>1</sup>Data are given on a skin dry weight basis. <sup>2</sup>Data are given on a berry fresh weight basis.

and 2013 on cvs Ortrugo and Sangiovese, evidencing a clear pattern of a significant increase in PFD treatment in the two sites and genotypes (**Table 5**). This trend was overwhelming vs. year-to-year variability and especially marked for quercetin 3-O glucuronide + quercetin 3-O glucoside and kaempferol 3-O glucoside.

modulated (t-test; p < 0.05) between C and PFD berries. The † indicates no significance.

These data strongly support the involvement of flavonol synthase in the PFD response in berry independently of growing site, year, and genotype.

#### DISCUSSION

#### Influence of Growing Site, Year, and Genotype on the Defoliation Response

The testing of four different genotypes over two consecutive years and having also compared a cultivar (Sangiovese) in the same year under two growing conditions (BO and AN) permits a proper discussion about the interactive effects between the above factors on vine response to PFD applied in all

instances with the same protocol (i.e., six main basal leaves removed at the "separated closed flowers" stage with retention of any laterals). Growing conditions (BO vs. AN) exerted an overall moderate effect on vine response variability to early leaf removal. It is notable that, in 2012, although AN vines were clearly under-cropped as compared to BO, variations of significantly modified parameters were always the same, while the same also held true for unmodified parameters, with the exception of TA and total anthocyanins (**Table 1**). In 2013, the influence of growing conditions was greater, albeit essentially limited to berry weight that was greatly reduced at BO site, while it was unchanged at AN. Since the vine balance given as LA/yield did not change in the two locations, it is likely that the improved total anthocyanins concentration at harvest in BO grapes results from inherently smaller berry size, hence higher skin-to-pulp ratio. This hypothesis is well supported by results obtained in a previous study showing that berries from pre-flowering defoliated Sangiovese vines, characterized by a significant increase of anthocyanin concentration in comparison to the control, also demonstrated a significant increase of berry skin thickness (Pastore et al., 2013).

Variability in the response to PFD attributable to year was, in AN and BO sites, more pertinent to must composition than to vegetative growth and yield parameters. In particular, TSS was more responsive in 2013, showing a large increase in PFD vines, whereas TSS at both sites was unchanged in 2012. The reasons for this difference are not easy to distinguish; however, the quite high TSS reached in AN in 2012 likely reflects the high LA/yield (>1.9 m<sup>2</sup> /kg) placing no limitations on the sugar accumulation process; in confirmation of this, PFD was more effective in increasing TSS in 2013 at quite low LA/yield ratio, suggesting that, in both sites, PFD might have benefitted from higher foliage "quality" due to lower canopy age from veraison onward (Poni et al., 2006). In the other sites, response to PFD over the 2 years was more consistent for all the parameters considered.

Comparing four genotypes over 2 years yielded a total of 10 C vs. PFD comparisons. All genotypes showed high responsiveness to the technique and no "recalcitrant" varieties could be discriminated. In more detail, in all cases yield per vine was significantly reduced after the early defoliation; among the main yield components cluster weight was reduced in 9 out of 10, berry weight in 5 out of 10, and berries per cluster in 6 out of 10. This confirms previous studies that PFD is extremely effective and consistent at reducing yield through either lower fruit set or smaller berry size or a combination of both, in turn resulting in looser clusters. This is not a surprising finding since the physiological background on which the technique relies is quite robust; it is well-known from the literature (Coombe, 1962; Hardie and Considine, 1976) that a calibrated source limitation imposed pre-flowering constrains the carbohydrate pool available to support flowering and fruit-set which, inherently, become limited. In terms of grape composition, it is likewise confirmed that, with very few exceptions, all genotypes subjected to PFD show a significant increase in TSS at harvest (7 cases out of 10) and, for red cultivars, in total anthocyanins (6 out of 8). The physiological bases for such response are also quite solid: (i) in PFD, ripening benefits from non-limiting or even higher final LA/yield ratios since induced yield limitation is often greater than the amount of leaf area removed "per se" with defoliation; (ii) as reported in Poni et al. (2006), in PDF the amount of carbohydrate supply per unit of grape fresh mass is higher from veraison onward due to an overall younger, hence more efficient canopy, and compensation mechanisms in either leaf area development or maximum photosynthetic rates and (iii) TSS and total anthocyanins can also be enhanced due to smaller berry size. Another quite remarkable and consistent feature of the PDF practice was that, despite the large increase in TSS, TA was reduced only in the Ortrugo trial, in five out of eight cases it was unchanged and on one occasion (AN, 2012) it was increased. This suggests that the technique is quite effective at decoupling the sugar/acid ratio and if TSS are increased, TA is not necessarily concurrently decreased. This feature is of special interest within a global warming scenario (Palliotti et al., 2014), where maintenance of adequate acidity in warm environments is an increasing concern.

#### Common Molecular Responses to Pre-flowering Defoliation in Sangiovese Berries Involved Genes Related to Secondary and Hormone Metabolism

Berry molecular responses to PFD were initially investigated by global gene expression analysis performed on berries at four developmental stages in defoliated and control Sangiovese vines, grown in BO and AN in 2012. Statistical analysis clearly revealed that the environment has a stronger effect than defoliation treatment on the transcriptome rearrangement during berry development in Sangiovese. Indeed, the correlation dendrogram showed a very clear distinction between BO and AN berry transcriptomes at each developmental stage and a weak and variable separation between berries from PFD and C vines. In addition, a PCA analysis revealed that no principal components were able to distinguish, at transcriptional level, berries from PFD and C vines. PCA analysis also showed that the ripening process was slightly advanced in AN in comparison to BO, reflecting the higher level of TSS, lower values of TA and higher anthocyanin content observed in ripe berries grown in AN, independently of the PFD effect (**Table 1**).

It is very unlikely that inter-clonal variation within the same genotype might have interfered in the transcriptional responses. Albeit some slight differences in berry transcriptome during development can be attributable to small variations in the sampling procedure adopted in the two locations, the growing site seems to strongly affect berry gene expression during development in Sangiovese, evidencing that a large part of its transcriptional ripening program is plastic. Previous attempts to quantify the transcriptomic plasticity in grapevine berry were recently reported for the red cultivar Corvina and white cultivar Garganega (Dal Santo et al., 2013a, 2016b). These studies demonstrated a wide berry phenotypic plasticity in both cultivars, in particular affecting

the secondary metabolism, suggesting that this phenomenon, which allows the production of different wines from the same cultivar and the adaptation of the same cultivar to diverse growing regions, is still scarcely characterized in grapevine.

Sangiovese is the top red variety grown in Italy with about 70.000 hectares (ISTAT, 2015), and is cultivated in several regions (i.e., Tuscany, Emilia Romagna, Marche, Umbria). It is well known that Sangiovese wine made with grapes from Emilia Romagna usually reaches a price tag that is 5–10 times lower than any Brunello di Montalcino label, which is likewise made with 100% Sangiovese grapes. These differences might reflect consumer perception and expectations as well as marketing strategies. In this context our transcriptomic survey, reflecting the growing site effect of two distinct Italian regions (Emilia Romagna and Marche), should be further explored to unveil the genotype × environment interactions of this important Italian grapevine cultivar.

The impact of the environment on Sangiovese berry transcriptome was further highlighted by the analysis of DEGs in berries from C and PFD vines in the two sites. Among the 1746 and 1041 DEGs found in BO and AN, respectively, only 125 were commonly differentially expressed, strongly suggesting that the effect of PFD on berry gene expression is mainly affected by growing conditions.

The GO enrichment analysis performed on these commonly DEGs revealed that "response to stimulus" and "cellular amino acid and derivative metabolic process" functional categories were significantly over-represented. The same two categories were previously found overrepresented in the list of genes differentially expressed at the end of veraison in Sangiovese berries subjected to both pre-flowering and veraison defoliation treatment (Pastore et al., 2013). The role in stress resistance of many genes involved in amino acid metabolism, such as the aspartate aminotransferase, glutamine synthetase, and serine hydroxymethyltransferase, was previously described (Moreno et al., 2005; Singh and Ghosh, 2013; de la Torre et al., 2014; Wu et al., 2016), strongly suggesting that stress response induction is one of the principal effects of PFD on berry transcriptome, independently of the environment. Interestingly, other functional categories were found among the 125 genes, predominantly the "secondary metabolic process," mainly represented by genes belonging to the phenylpropanoid pathway, and "hormone stimulus," with several genes related to auxin, ethylene, and ABA metabolism. These results support previous observations regarding the direct effect on the expression of genes involved in berry ripening exerted by leaf removal (Pastore et al., 2013). Moreover, the modulation of phenylpropanoidrelated genes in defoliated berries could represent a stress response, as previously observed in grapevine upon various stresses (Matus et al., 2009; Rienth et al., 2014; Corso et al., 2015).

In order to identify putative berry molecular markers of the PFD treatment, we focused only on genes characterized by a low plasticity during berry development. Among these, six were selected as putative markers of the treatment, being similarly modulated throughout berry development by PFD in the two sites and for their possible role in berry formation and ripening.

The expression profiles of these candidates were analyzed by qPCR in berries from PFD and C vines in BO and AN in 2012, in order to validate microarray data, and in 2013 in order to assess the vintage effect on their expression modulation after treatment. Five out of six selected genes demonstrated a consistent modulation of expression induced by PFD throughout berry development in both years and in both sites, emerging as putative molecular markers for this treatment in the cultivar Sangiovese.

The ABC transporter VvPDR20/VvABCG50 showed a delay in its peak of expression at Stage 3, when berries are softening and start to accumulate pigments, instead of Stage 2, when berries are still green and firm. ABC transporters superfamily constitutes one of the largest families of transmembrane proteins that plays important roles in the vacuolar accumulation of secondary metabolites, such as flavonoids, in detoxification and heavy metal sequestration, in chlorophyll catabolite transport and ion channel regulation (Klein et al., 2006; Kang et al., 2010). In grapevine, 135 putative ABC proteins were identified and classified (Cakir and Kilickaya, 2013). The VvPDR20/VvABCG50 gene belongs to the PDR subfamily in V. vinifera, the largest ABC transporter subfamily. In grapevine, no PDR-related ORF has been cloned in its entirety and characterized (Cakir and Kilickaya, 2013). However, in other species, members of this family confer resistance to various biotic and abiotic stresses (Moons, 2003; Lee et al., 2005; Stukkens et al., 2005). Interestingly, it was recently shown in Arabidopsis that PDR12 is a plasma membrane ABA uptake transporter that mediates cellular uptake of the phytohormone ABA in guard cells (Kang et al., 2010). VvPDR20/VvABCG50 gene is an interesting molecular marker directly affected by the PFD treatment in Sangiovese, and is a good candidate for future functional studies.

The indole-3-acetate beta-glucosyltransferase and the auxin responsive factor 10, involved in auxin metabolism and signaling, respectively, were also identified as PFD molecular biomarkers. It is generally acknowledged that auxin plays a role in fruit growth. However, the change in auxin levels during berry development and the dynamics of auxin transport and signaling are still under debate (Fortes et al., 2015). It has been demonstrated that auxin treatment of pre-veraison grape berries delays ripening and alters the expression of developmentally regulated genes, suggesting that low auxin levels are required to trigger the onset of ripening (Davies et al., 1997; Bottcher et al., 2011; Ziliotto et al., 2012). However, it has been also hypothesized that high auxin levels at pre-veraison stages are required for the induction of genes involved in the ripening inception (Ziliotto et al., 2012; Corso et al., 2016). The regulation of auxin levels is associated with the conjugation of indole acetic acid (IAA) by the indol-3-acetate beta-glucosyltransferase, which allows an increase in the conjugated form of auxin after veraison. The decrease of expression after veraison of the indol-3 acetate beta-glucosyltransferase gene, obtained in C Sangiovese vines in the 2 years, corroborates previous results obtained in three Portuguese varieties (Agudelo-Romero et al., 2013) and supports the role of this gene in auxin level regulation

during ripening. Interestingly, an increase in expression level of this gene was found as a consequence of PFD treatment, which delays the gene downregulation, possibly affecting the IAA homeostasis and, hence, the regulation of the onset of ripening.

Auxin response factors regulate the auxin-mediated gene expression (Tiwari et al., 2003). In grapevine, 19 ARF genes were recently identified (Wan et al., 2014; Corso et al., 2016) but no functional studies have been performed to date. The ARF 10 exhibited a peak of expression at the ripening stage in C vines, whereas the peak of expression was hastened at the end of veraison in PFD vines (Stage 3), strongly suggesting that the treatment can interfere with the ripening progress by affecting the auxin level at the end of veraison. Biochemical analyses aimed to quantify the free and conjugate auxin form in berries from PFD and C vines will be necessary to thoroughly characterize the possible impact on hormonal regulation exerted by the PFD.

The last two putative PFD molecular markers identified in Sangiovese were a CAD gene, involved in the last step of the synthesis of the monomeric precursors of lignin, and a G3T, involved in the glycosylation of flavonols. Both genes showed an increase of expression in berries from PFD vines, more pronounced at Stage 2 in Bologna and at Stage 3 in Ancona. The increase in expression of CAD was already observed in Sangiovese berries from PFD vines (Pastore et al., 2013). It was previously proposed that the increase in sunlight exposure of berries on defoliated vines induced the expression of genes involved in cell wall metabolism that allow an increase in berry skin thickness, providing more epidermal layers for protection against sunburn and storage of anthocyanin compounds (Pastore et al., 2013).

The G3T coincides with the previously characterized VvGT6, contributing to flavonol glycosylation (Ono et al., 2010). The upregulation of this gene obtained in berries from PFD vines is consistent with the higher flavonol content that berries subjected to PFD manifested (Pastore et al., 2013; this work).

Among the six genes, only the expression of geraniol 10-hydroxylase, a cytochrome P450 monooxygenase involved in the biosynthesis of terpenoids (Collu et al., 2001) and phenylpropanoids (Sung et al., 2011), showed an inconsistent modulation of expression between the 2 years. Differences in expression trend of this gene and its responsiveness to the defoliation treatment could be linked to the differences in weather patterns between 2012, marked by a long summer with hot spells, and the slightly cooler 2013. It is well known that terpenoids metabolism contributes to plant adaptation to the environment, in particular solar exposure, UV-B radiation (Gil et al., 2013; Zhang et al., 2014) and drought (Deluc et al., 2009; Selmar and Kleinwachter, 2013; Savoi et al., 2016).

#### General Molecular Markers of the Pre-flowering Defoliation

Expression analysis of the PFD Sangiovese putative markers was performed on another three genotypes subjected to the same treatment in different environments in 2012. A stronger influence of genotype and/or growing site than PFD treatment on these genes' expression was clearly revealed during berry development, indicating that they should not be considered as PFD markers for genotypes other than Sangiovese. We therefore selected new candidates by focusing on DEGs at each berry developmental stage, without evaluating the entire expression profile.

The inclusion in our experimental plan of an early stage of berry development (Stage 1), allowed previous transcriptomic results, focused only on the PFD effect on berry ripening phase, to be greatly improved (Pastore et al., 2013). Indeed, several DEGs in PFD berries in comparison to C were found at Stage 1 in both sites. This strongly supports recent results obtained in defoliated Cabernet Sauvignon berries showing that earlier berry stages react to leaf removal distinctly from the later developmental stages (Young et al., 2016).

A flavonol synthase resulted as one of the most upregulated genes in PFD Sangiovese berries at Stage 1 in both sites and its upregulation was confirmed by qPCR in all genotypes, environments, and years. Interestingly, the upregulation of the same gene was previously observed in both pre-flowering and veraison defoliated Sangiovese berries at the end of veraison (Pastore et al., 2013) and the expression of two isoforms of flavonol synthase was affected in defoliated berries of Cabernet Sauvignon (Matus et al., 2009). Quantification of the main berry flavonols at harvest in Ortrugo and Sangiovese in both years evidenced a significant increase in berries from PFD vines, in the two sites, genotypes and years, strongly supporting transcriptomic data. In Sangiovese PFD treated berries, the concentration of quercetin, the main flavonol in red grapes (Mattivi et al., 2006), and kaempferol, was more than twice that in control, in both years, whereas the increase of myricetin at harvest was less intense, as previously observed (Pastore et al., 2013). This shift in flavonol composition was not observed in Ortrugo, in which the abundance of all flavonol compounds increased after PFD treatment in both years, as previously reported for Merlot (Spayd et al., 2002). It was demonstrated that the induction of flavonols synthesis is positively correlated to sunlight exposure, reflecting their role as UV protectants (Price et al., 1995; Haselgrove et al., 2000; Downey and Rochfort, 2008; Matus et al., 2009). Our data suggested that the PFD-induced expression of flavonol synthase gene at an early stage of berry development is due to an increase in cluster sunlight exposure, causing a significant accumulation of flavonols in berries at harvest, and that this effect is shared across different environments, years, and genotypes.

A jasmonate O-methyltransferase gene resulted as being another common molecular marker of the defoliation for both Stage 2 and Stage 3 of berry development, being positively modulated by PFD in all tested conditions. The role of methyl jasmonate (MeJA) in the response to biotic and abiotic stresses was widely discussed in the past (Cheong and Choi, 2003; Wasternack and Song, 2016). In non-climacteric fruits such as strawberry and grape, JA levels are reported to be high in early development and decreasing to lower values in riper

fruits, enabling the onset of ripening to occur (Kondo and Fukuda, 2001). In grapevine, the gene coding for jasmonate O-methyltransferase was found downregulated in ripe fruits of three grape varieties (Agudelo-Romero et al., 2013). Interestingly, the downregulation of jasmonate O-methyltransferase during berry development was revealed by transcriptomic analysis on Sangiovese berries from C vines in both Bologna and Ancona in 2012. As a consequence, the upregulation observed at Stages 2 and 3 in berries from PFD vines corresponded to a delay in its downregulation and not to a genuine induction.

It was recently demonstrated that JA plays an important role in grape berry coloring and softening by inducing the transcription of several ripening-related genes, such as phenylpropanoid genes, cell wall metabolism-related genes and genes involved in aroma accumulation (Jia et al., 2016). Consistently, we observed that at Stage 2 and Stage 3 several terpenoid synthase genes (e.g., one geraniol 10-hydroxylase and several different terpene synthases), involved in berry aromatic compounds accumulation, was found commonly upregulated by PFD in Bologna and Ancona in 2012, together with jasmonate O-methyltransferase. Intriguingly, at Stage 3 also a jasmonate ZIM-Domain VvJAZ2 gene, involved in JA signaling cascade (Pauwels and Goossens, 2011; Wasternack and Song, 2016), resulted as commonly modulated by PFD in Sangiovese berries in both sites. Although the expression of this gene was not assessed in all conditions, we could hypothesize that the PFD impacts on berry ripening, possibly affecting the JA metabolism at veraison, when the JA level has recently been proposed to play a crucial role in ripening regulation (Jia et al., 2016).

The last identified berry PFD marker was the ABA receptor PYL4, which resulted as being the most commonly downregulated gene at Stage 4 in Sangiovese in 2012. Interestingly, all genes commonly modulated in PFD Sangiovese berries at Stage 4 in that year resulted as downregulated, suggesting that the treatment could affect many metabolisms by hastening their normal shutdown. This hypothesis also holds true for the ABA receptor PYL4, showing a decreasing expression trend during ripening in C conditions with a peak of expression at veraison, and an accelerated downregulation in PFD berries. The ABA receptor PYL4 belongs to the PYR/PYL/RCAR receptors family that, together with the PP2Cs and SnRK2s kinases, constitutes the complex molecular machinery involved in the ABA-mediated signaling pathway (Boneh et al., 2012).

Many studies indicated that ABA, together with other phytohormones like brassinosteroids (BRs) and ethylene may play an important role in several ripening-associated processes of grape berry (Davies and Bottcher, 2009; Kuhn et al., 2014). It was observed that free ABA levels increase around veraison, concurrently with sugar accumulation, berry coloration, and softening, whereas during ripening ABA levels may be controlled mainly by conjugation to glucose. Nevertheless, it was found that a set of genes involved in the ABA-mediated signaling pathway, including the ABA receptors PYL8 RCAR3 and PYL9 RCAR1, were upregulated at the mid-ripening phase in three Portuguese varieties (Fortes et al., 2011; Agudelo-Romero et al., 2013). This expression trend was consistent with the expression of PYL4 found in C vines, suggesting that later in ripening, ABA synthesis is not induced, but ABA-regulated processes are instead activated. The lower expression of PYL4 observed in PFD berries at harvest, which corresponds to a faster downregulation in comparison to C, suggests that PFD treatment hastened the ABAmediated ripening signaling possibly by inducing abiotic stress response early during development. This faster downregulation of PYL4 is particularly marked in Ortrugo cultivar, in which the highly significant increase in ripening parameters in both years suggests that the ripening process in this cultivar started much earlier than in the other cultivars in the berries after treatment.

# CONCLUSION

A comparison was made of physiological and molecular responses to PFD in four grapevine cultivars grown in different Italian geographical areas and during two consecutive years to evaluate the interactive effects between these factors on vine response to PFD and to determine the common effect of defoliation in berry at transcriptional level. All genotypes were highly responsive to the technique, the yield per vine being significantly reduced in all conditions. In terms of grape composition, a significant increase in sugar content and total anthocyanins at harvest was obtained in most genotypes. Sangiovese resulted as being the cultivar with the stronger variability in must composition in the response to defoliation. Global gene expression analysis performed on Sangiovese berries from defoliated and untreated control vines grown in two different sites highlighted, on the one hand, the strong effect of environment on the berry transcriptional ripening program in this cultivar and, on the other, allowed genes commonly regulated by selective leaf removal to be identified. The differential expression of these putative marker genes, mainly related to secondary metabolism and hormone signaling, could link the defoliation treatment to physiological and metabolic changes found in treated berries. These new insights greatly improve previous knowledge about molecular mechanisms on the basis of the qualitative outcomes of an important and widely used management technique in viticulture, allowing physiological responses in berry to the selective PFD practice to be precisely defined across genetic and environmental variability.

# AUTHOR CONTRIBUTIONS

SZ designed the microarray experiments, interpreted the micro array data, and wrote the manuscript, SDS performed the statistical analyses, SDS and GBT interpreted the microarray data and helped in drafting the manuscript, ED performed the RNA extraction, microarray experiments, and real-time qPCR analysis, IF, CP, GA, OS, VL, APi, RDL, APa, ST and MG conducted defoliation experiments, sampled the material and processed the data, CP, GA and MG performed the HPLC analysis,

SP conceived and supervised the study, wrote and critically revised the manuscript. All authors contributed to the revision of the manuscript.

#### FUNDING

The research leading to these results received funding from the Italian Ministry of University and Research PRIN 2009P3B89K\_003 Project "Characterization of genes involved in berry ripening after pre-bloom defoliation" and by the European INNOVINE Project FP7-311775 "Combining innovation in vineyard management and genetic diversity for a sustainable European viticulture." SDS was financially supported by the Italian Ministry of University and Research FIRB RBFR13GHC5

#### REFERENCES


Project "The epigenomic plasticity of grapevine in genotype per environment interactions."

#### ACKNOWLEDGMENT

We thank Martina Albarello for technical assistance in real-time qPCR analyses.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2017.00630/ full#supplementary-material

auxin-related genes in cabernet sauvignon berries. Front. Plant Sci. 7:69. doi: 10.3389/fpls.2016.00069



responses in green and ripening grapevine (Vitis vinifera) fruit. BMC Plant Biol. 14:108. doi: 10.1186/1471-2229-14-108


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Zenoni, Dal Santo, Tornielli, D'Incà, Filippetti, Pastore, Allegro, Silvestroni, Lanari, Pisciotta, Di Lorenzo, Palliotti, Tombesi, Gatti and Poni. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Global DNA Methylation Patterns Can Play a Role in Defining Terroir in Grapevine (Vitis vinifera cv. Shiraz)

Huahan Xie1,2† , Moumouni Konate1,2† , Na Sai1,2,3, Kiflu G. Tesfamicael1,2 , Timothy Cavagnaro<sup>2</sup> , Matthew Gilliham2,3, James Breen4,5, Andrew Metcalfe<sup>6</sup> , John R. Stephen<sup>7</sup> , Roberta De Bei<sup>2</sup> , Cassandra Collins<sup>2</sup> and Carlos M. R. Lopez1,2 \*

<sup>1</sup> Environmental Epigenetics and Genetics Group, University of Adelaide, Adelaide, SA, Australia, <sup>2</sup> The Waite Research Institute and The School of Agriculture, Food and Wine, University of Adelaide, Adelaide, SA, Australia, <sup>3</sup> The ARC Centre of Excellence in Plant Energy Biology, University of Adelaide, Adelaide, SA, Australia, <sup>4</sup> Robinson Research Institute, University of Adelaide, Adelaide, SA, Australia, <sup>5</sup> Bioinformatics Hub, School of Biological Sciences, University of Adelaide, Adelaide, SA, Australia, <sup>6</sup> School of Mathematical Sciences, University of Adelaide, Adelaide, SA, Australia, <sup>7</sup> Plant Genomics Centre, Australian Genome Research Facility Ltd., Adelaide, SA, Australia

#### Edited by:

José Tomás Matus, Centro de Investigación en Agrigenómica, Universitat Autònoma de Barcelona, Spain

#### Reviewed by:

Philippe Gallusci, Université de Bordeaux, France Melane Alethea Vivier, Stellenbosch University, South Africa

#### \*Correspondence:

Carlos M. R. Lopez carlos.rodriguezlopez@adelaide.edu.au

> †These authors have contributed equally to this work.

#### Specialty section:

This article was submitted to Plant Breeding, a section of the journal Frontiers in Plant Science

Received: 18 April 2017 Accepted: 11 October 2017 Published: 30 October 2017

#### Citation:

Xie H, Konate M, Sai N, Tesfamicael KG, Cavagnaro T, Gilliham M, Breen J, Metcalfe A, Stephen JR, De Bei R, Collins C and Lopez CMR (2017) Global DNA Methylation Patterns Can Play a Role in Defining Terroir in Grapevine (Vitis vinifera cv. Shiraz). Front. Plant Sci. 8:1860. doi: 10.3389/fpls.2017.01860 Understanding how grapevines perceive and adapt to different environments will provide us with an insight into how to better manage crop quality. Mounting evidence suggests that epigenetic mechanisms are a key interface between the environment and the genotype that ultimately affect the plant's phenotype. Moreover, it is now widely accepted that epigenetic mechanisms are a source of useful variability during crop varietal selection that could affect crop performance. While the contribution of DNA methylation to plant performance has been extensively studied in other major crops, very little work has been done in grapevine. To study the genetic and epigenetic diversity across 22 vineyards planted with the cultivar Shiraz in six wine sub-regions of the Barossa, South Australia. Methylation sensitive amplified polymorphisms (MSAPs) were used to obtain global patterns of DNA methylation. The observed epigenetic profiles showed a high level of differentiation that grouped vineyards by their area of provenance despite the low genetic differentiation between vineyards and sub-regions. Pairwise epigenetic distances between vineyards indicate that the main contributor (23–24%) to the detected variability is associated to the distribution of the vineyards on the N–S axis. Analysis of the methylation profiles of vineyards pruned with the same system increased the positive correlation observed between geographic distance and epigenetic distance suggesting that pruning system affects inter-vineyard epigenetic differentiation. Finally, methylation sensitive genotyping by sequencing identified 3,598 differentially methylated genes in grapevine leaves that were assigned to 1,144 unique gene ontology terms of which 8.6% were associated with response to environmental stimulus. Our results suggest that DNA methylation differences between vineyards and sub-regions within The Barossa are influenced both by the geographic location and, to a lesser extent, by pruning system. Finally, we discuss how epigenetic variability can be used as a tool to understand and potentially modulate terroir in grapevine.

Keywords: environmental epigenetics, DNA methylation, terroir, MSAP, msGBS, Vitis vinifera, Shiraz, Barossa

#### INTRODUCTION

fpls-08-01860 October 27, 2017 Time: 16:50 # 2

The ability of plants to produce alternative phenotypes in response to changing environments is known as phenotypic plasticity (Pigliucci, 2005). Genotypes with this characteristic are able to produce a variety of phenotypes including improved growth and reproduction (Lacaze et al., 2009). Grapevine (Vitis vinifera L.) is a highly plastic crop that exhibits large differences in fruit composition from a given variety depending upon the environmental conditions of the vineyard of origin (Dal Santo et al., 2016). Fruit traits that affect wine quality are thought to be largely driven by the interaction of a grapevine's genetic characteristics with environmental factors (i.e., climate, soil, and topography) and vineyard management (Robinson et al., 2012). The grapevine cycle extends for two seasons, with buds formed in a specific year giving rise to shoots that will carry fruit the next year (Keller et al., 2010). Environmental cues over two seasons can impact on yield (fruit quantity) and fruit composition by influencing the formation of the inflorescence primordia (Buttrose, 1974), flowering and fruitset (Petrie and Clingeleffer, 2005). Temperature and water availability are also known to influence sugar concentration, acidity, pH, color, and other characteristics in the fruit (Adams, 2006; Downey et al., 2006). Moreover, climate change predictions of elevated CO<sup>2</sup> and rising temperature are also likely to have an effect on the grapevine reproductive cycle and on fruit composition (Parra et al., 2010). All these variables, in conjunction with the wine making process, give a wine its distinctive character. The impact of the environment on grape composition and subsequent wine excellence has given rise to the concept of 'terroir,' a French term referring to terre, "land" (Fanet and Brutton, 2004).

Terroir is defined as the interaction between the physical and biological environment and applied viticultural and oenological practices that lead to unique characteristics in a final wine (Seguin, 1986). Extensive studies have been published on terroir, but generally, these focus on a single parameter such as climatic factors, soil structure, or soil microbiology (Harrison, 2000; Tonietto and Carbonneau, 2004). However, studying only one environmental parameter does not provide an entire understanding of how wine quality is influenced by terroir (van Leeuwen et al., 2004). A significant amount of work has also been published on the genetic basis of fruit composition in grapevines (e.g., Doligez et al., 2002). Despite these insights, further research is required on the molecular changes that are involved in the vine interaction with its environment.

One of the molecular changes worth investigating relates to environmentally induced epigenetic modifications. In fact, phenotypic plasticity has been previously associated to epigenetic variation (Vogt, 2015). Interestingly, analysis of epigenetic diversity has been shown to be more effective in discriminating inter-clonal variability in grapevine than the use of purely genetic molecular markers such as simple sequence repeats (SSRs) or amplified fragment length polymorphisms (AFLPs) (Imazio et al., 2002; Schellenbaum et al., 2008; Ocaña et al., 2013). Epigenetic mechanisms refer to potentially heritable (via mitosis or meiosis) molecular changes that affect gene expression leading to differences in phenotype without changing the organism DNA sequence (Jaenisch and Bird, 2003; Haig, 2004). Such mechanisms are involved in the control of a range of plant processes, including developmental control (Daccord et al., 2017), genomic imprinting (Köhler et al., 2012), and response to stress (Yaish et al., 2011; Tricker et al., 2012). It is now also widely accepted that epigenetic mechanisms have been the source of useful variability during crop varietal selection (Amoah et al., 2012; Bloomfield et al., 2014; Rodríguez López and Wilkinson, 2015).

Multiple environmental cues have been shown to induce persistent changes in epigenetic modifications, resulting in an epigenetic priming that can act over multiple vegetative (Kumar et al., 2016) or sexual generations (Tricker et al., 2012). Although whether environmentally induced epialleles have any effect on the phenotypes of future generations remains a matter of debate, such priming is considered by some as an adaptive strategy by which plants use their memory of the environment to modify their phenotypes to adapt to subsequent conditions (Kelly et al., 2012; Tricker et al., 2013a,b; Vogt, 2015). It is commonly accepted that DNA methylation constitutes an adaptation strategy to the environment (YunLei et al., 2009), and that changes in DNA methylation can produce altered phenotypes (Zhang et al., 2006; Herrera and Bazaga, 2011; Iqbal et al., 2011). Moreover, epigenetic mechanisms are now considered as potential drivers of rapid adaptation to the environmental variability (Bräutigam et al., 2013). These processes facilitate adaptation by regulating the expression of genes controlling phenotypic plasticity (Richards, 2006; Bossdorf et al., 2008) early in adaptive walks (Kronholm and Collins, 2016) but also by releasing cryptic genetic variation and/or facilitating mutations in functional loci over longer-term timescales (O'Dea et al., 2016). To this extent, there have been extensive studies establishing a link between DNA methylation in plants and environmental conditions both in wild (Fonseca Lira-Medeiros et al., 2010; Herrera and Bazaga, 2010; Alonso et al., 2016) and cultivated species (Zheng et al., 2017).

All major epigenetic mechanisms, DNA methylation, histone modifications, and RNA interference, are present in plants (Pikaard and Mittelsten Scheid, 2014; Holoch and Moazed, 2015; Wendte and Pikaard, 2017). In plants, DNA methylation (5mC) occurs at different cytosine contexts (CpG, CpHpG, or CpHpH) (H = A, T or C) (Richards, 1997; Baulcombe and Dean, 2014; Niederhuth and Schmitz, 2017) and it is induced, maintained or removed by different classes of methyltransferase in conjunction with environmental and developmental cues (Law and Jacobsen, 2010; Baulcombe and Dean, 2014). The contribution of DNA

**Abbreviations:** AMOVA, analysis of molecular variance; BAM file, Binary Alignment/Map file; BIC, Bayesian Information Criterion; CPM, counts per million; DAPC, discriminant analysis of principal components; DMG, differentially methylated gene; DMM, differentially methylated marker; FDR, false discovery rate; gDNA, genomic DNA; GeoD, geographic distance; GO, gene ontology; kb, kilobase; log2FC, logarithm 2 of fold change; MSAPs, methylation sensitive amplified polymorphisms; msGBS, methylation sensitive genotyping by sequencing; MSL, methylation sensitive polymorphic loci; NML, non-methylated polymorphic loci; PCA, principal component analysis; PCoA, principal coordinate analysis; PhiPT, measurement of epi/genetic diversity among populations; SI, Shannon's diversity index; SNP, single nucleotide polymorphism; TES, transcription end site; TSS, transcription start site.

methylation to plant performance has been extensively studied in model organisms and some annual crops (Rodríguez López and Wilkinson, 2015). However, we are only beginning to understand how long-living plants, such as grapevines, use epigenetic mechanisms to adapt to changing environments (Fortes and Gallusci, 2017). Effects of environmental conditions on non-annual crops performance can be very difficult to evaluate since many environmental factors interact over the life of the plant to ultimately contribute toward the plant's phenotype (Fortes and Gallusci, 2017). Although epigenetic mechanisms have been shown to act as a memory of the organism's growing environment during mitotic division (Morao et al., 2016), even after vegetative propagation (Raj et al., 2011; Guarino et al., 2015), very few studies have focussed on DNA methylation changes in grapevine. The few known studies in this field used MSAPs (Reyna-López et al., 1997) for the detection of in vitro culture induced epigenetic somaclonal variability (Baránek et al., 2015), and the identification of commercial clones (Imazio et al., 2002; Schellenbaum et al., 2008; Ocaña et al., 2013).

In this study, we hypothesize that DNA methylation can play a role in defining terroir. To test this hypothesis, we investigated the effect of environmental and management conditions on DNA methylation variation in grapevine cultivar Shiraz across 22 vineyards representative of The Barossa wine zone (Australia) (Robinson and Sandercock, 2014) using MSAPs. Finally, we used msGBS to characterize the genomic context of the observed regional genetic and epigenetic variability.

# MATERIALS AND METHODS

#### Vineyard Selection and Plant Material

Vines from 22 commercial vineyards located in the iconic Barossa wine zone (The Barossa hereafter) (Australia) were included in this study. Vineyards were chosen based on the knowledge that they produce premium Shiraz wines that are representative of the climate, soil and management practices that are used in the Barossa sub-regions as described by the Barossa Grounds Project (Robinson and Sandercock, 2014) [i.e., Eden Valley (three vineyards) and Barossa Valley (19 vineyards) which included vineyards in the five distinctive sub-regions within the Barossa Valley Region: Northern Grounds (four vineyards), Central Grounds (four vineyards), Eastern Edge (four vineyards), Western Ridge (four vineyards), Southern Grounds (three vineyards)] (Supplementary Table S1). To simplify the nomenclature, the Eden Valley region, Northern, Central, Southern Grounds, Eastern Edge, and Western Ridge will be defined as sub-regions hereafter. All vineyards were planted with own-rooted vines of the cv. Shiraz. Ten vineyards were planted with clone SA 1654 (Whiting, 2003), one with clone BVRC30 (Whiting, 2003), one with clone PT15 Griffith (Farquhar, 2005) and 10 of 'unknown' clonal status (Supplementary Table S1).

Nine vines from three rows from each vineyard were selected and vines adjacent to missing vines, end of row vines and border rows were excluded, to prevent differences in competition effects between plants. Also, rows containing sampled plants were selected from each vineyard after discussion with vineyard managers to capture the variability in each vineyard. A total of 198 plants (nine plants per vineyard) were selected to capture the diversity from each vineyard. Leaf samples (first fully expanded leaf at bud burst, E-L 7) (Coombe, 1995) were collected from three nodes per plant and pooled into a single sample per plant. All samples were taken before dawn (between 10:00 pm and sunrise) to minimize variability associated with differences in plant water status (Williams and Araujo, 2002). Samples were immediately snap-frozen in liquid nitrogen in the vineyard and stored at −80◦C until DNA extraction.

#### DNA Isolation

Genomic DNA extractions from all 198 samples were performed using the three pooled leaves per plant powdered using an automatic mill grinder (Genogrinder). The obtained frozen powder was used for DNA extraction using the Oktopure automated DNA extraction platform (LGC Genomics GmbH) according to the manufacturer's instructions. Isolated DNA was quantified using the Nanodrop 2000 spectrophotometer (Thermo Fisher Scientific, Wilmington, DE, United States). DNA final concentrations were normalized to 20 ng/µl using nanopure water (Eppendorf, Germany).

### Analysis of Genetic/Epigenetic Variability Using MSAP

Methylation sensitive amplified polymorphism analysis was performed as described by Rodríguez López et al. (2012). In brief, gDNA from 88 plants (four plants per vineyard) was digested with a combination of the restriction enzymes EcoRI and one of two DNA methylation sensitive isoschizomers (HpaII or MspI). Double stranded DNA adapters (See Supplementary Table S2 for the sequence of all oligonucleotides used) containing co-adhesive ends complementary to those generated by EcoRI and HpaII/MspI were ligated to the digested gDNA and then used as a template for the first of two consecutive selective PCR amplifications in which the primers were complementary to the adaptors but possessed unique 3<sup>0</sup> overhangs. The second selective PCR amplification used primers containing 3<sup>0</sup> overhangs previously tested on grapevine (Baránek et al., 2015). HpaII/MspI selective primer was 5<sup>0</sup> end-labeled using a 6-FAM reporter molecule for fragment detection using capillary electrophoresis on a ABI PRISM 3130 (Applied Biosystems, Foster City, CA, United States) housed at the Australian Genome Research Facility Ltd., Adelaide, SA, Australia.

Generated electropherograms were visualized using GeneMapper Software v4 (Applied Biosystems, Foster City, CA, United States). A binary matrix containing presence (1) absence (0) epilocus information was generated for each enzyme combination (i.e., EcoRI/HpaII and EcoRI/MspI). MSAP fragment selection was limited to allelic sizes between 95 and 500 bp to reduce the potential impact of size co-migration during capillary electrophoresis (Caballero et al., 2008). Different levels of hierarchy were used to group the samples. Samples were first grouped according to vineyard of origin. Then, samples were divided into their sub-regions of origin. Finally, samples were further separated into groups according to clones and the

vineyard management systems (i.e., pruning system used in their vineyard of origin) (Supplementary Table S1).

HpaII and MspI binary matrices were then used to compute Shannon's Diversity Index implemented using msap R package (v. 1.1.8) (Pérez-Figueroa, 2013) and PCoA was estimated in all regions to determine and visualize the contribution to the observed molecular variability within regions of NML and of MSL (genetic and epigenetic variability, respectively) (Smouse et al., 2015).

GenAlex v 6.5 software (Peakall and Smouse, 2012) was used for PCoA in order to visualize the molecular differentiation between Barossa sub-regions detected using MSAP profiles generated after the restriction of gDNA with HpaII or MspI. We then used AMOVA to determine the structure of the observed variability using PCoA. Molecular differences between vineyards and regions was inferred as pairwise PhiPT distances (Michalakis and Excoffier, 1996).

Mantel test analysis (Hutchison and Templeton, 1999) was used to estimate the correlation between the calculated pairwise molecular distances with (1) the GeoD [i.e., Log(1 + GeoD (Km)] and (2) differences in environmental variables among vineyards (i.e., vineyard altitude, regional average annual rainfall, regional growing season rainfall, regional mean January temperature, regional growing season temperature, and growing degree days). Mantel test was implemented in Genalex v 6.5 as described by Róis et al. (2013) and significance was assigned by random permutations tests (based on 9,999 replicates).

### Characterization of Genetic/Epigenetic Variability Using msGBS

Methylation sensitive genotyping by sequencing was performed as described by Kitimu et al. (2015). In brief, 200 ng of gDNA from nine samples from Northern, Central, and Southern Grounds (vineyards 1–4, 5–8, and 13–15, respectively) were digested using 8 U of HF-EcoRI and 8 U of MspI (New England BioLabs Inc., Ipswich, MA, United States) in a volume of 20 µl containing 2 µl of NEB Smartcut buffer at 37◦C for 2 h followed by enzyme inactivation at 65◦C for 10 min. Sequencing adapters were ligated by adding 0.1 pmol of the MspI adapters (uniquely barcoded for each of the 198 samples) and 15 pmol of the common EcoRI Y adapter (See Supplementary Table S2 for the sequence of all oligonucleotides used), 200 U of T4 ligase and T4 ligase buffer (New England BioLabs Inc., Ipswich, MA, United States) in a total volume of 40 µl at 24◦C for 2 h followed by an enzyme inactivation step at 65◦C for 10 min. Excess adapters were removed from ligation products using Agencourt AMPure XP beads (Beckman Coulter, Australia) at the ratio of 0.85 and following manufacturer's instructions. Single sample msGBS libraries were then quantified using Qbit 3 (Thermo Fisher). A single library was generated by pooling 25 ng of DNA from each sample. Library was then amplified in eight separate PCR reactions (25 µl each) containing 10 µl of library DNA, 5 µl of 5x Q5 high fidelity buffer, 0.25 µl polymerase Q5 high fidelity, 1 µl of each forward and reverse common primers at 10 µM, 0.5 µl of 10 µM dNTP and 7.25 µl of pure sterile water. PCR amplification was performed in a Bio-Rad T100 thermocycler consisting of DNA denaturation at 98◦C (30 s) and 10 cycles of 98◦C (30 s), 62◦C (20 s), and 72◦C (30 s), followed by 72◦C for 5 min. PCR products were then re-pooled and DNA fragments ranging between 200 and 350 bp in size were captured using the AMPure XP beads following manufacturer's instructions. Libraries were sequenced using an Illumina NextSeq High Output 75 bp pair-end run (Illumina Inc., San Diego, CA, United States) at the Australian Genome Research Facility (AGRF, Adelaide, SA, Australia).

#### msGBS Data Analysis

Analysis of genetic diversity between regions was performed by SNP calling using TASSEL 3 (Bradbury et al., 2007) on msGBS sequencing results. Only SNPs present in at least 80% of the samples were considered for analysis. PCA was implemented on TASSEL 3 using the selected SNPs. To identify any possible geographical genetic structure, the optimal number of genetic clusters present in the three regions were computed using BIC as effected by DAPC using adegenet 2.0.0<sup>1</sup> .

Identification of significant DMMs between regions was then computed using the package msgbsR<sup>2</sup> (accessed on 26/08/2016). In brief, raw sequencing data was first demultiplexed using GBSX (Herten et al., 2015) and filtered to remove any reads that did not match the barcode sequence used for library construction. Following demultiplexing, paired-end reads were merged using bbmerge in bbtools package (Bushnell, 2016). Merged reads were next aligned to the 12X grapevine reference genome<sup>3</sup> . Alignment BAM files where then used to generate a read count matrix with marker sequence tags, and used as source data to perform subsequent analyses using msgbsR in the R environment (R Core Team, 2015). Finally, the presence of differential methylation between regions was inferred from the difference in the number of read counts from all sequenced MspI containing loci that had at least 1 CPM reads and present in at least 15 samples per region. Significance threshold was set at Bonferroni adjusted P-value (or FDR) < 0.01 for difference in read CPM. The logFC (logarithm 2 of fold change) was computed to evaluate the intensity and direction of the region specific DNA methylation polymorphism.

To determine how the observed changes in DNA methylation between sub-regions were associated to protein coding genes, the distribution of DMMs was assessed around such genomic features, as defined in Ensembl database<sup>4</sup> , by tallying the number of DMMs between the TSS and the TES and within five 1 kb windows before the TSS and after TES of all V. Vinifera genes, using bedtools (Quinlan and Hall, 2010).

Genes within 5 kb of a DMM were referred to as DMGs. DMGs in each pairwise regional comparison were grouped into those showing hypermethylation or hypomethylation, and were next used separately for GO terms enrichment, using the R packages: GO.db (Carlson, 2016) and annotate (Gentleman, 2016). Significant GO terms were selected based on Bonferroni adjusted P-values at significance threshold of 0.05. Finally, GO

<sup>1</sup>http://adegenet.r-forge.r-project.org/

<sup>2</sup>https://github.com/BenjaminAdelaide/msgbsR

<sup>3</sup>http://plants.ensembl.org/Vitis\_vinifera/

<sup>4</sup>http://plants.ensembl.org/biomart/martview/

terms containing DMGs in all three pairwise comparisons were visualized using REViGO (Supek et al., 2011).

# RESULTS

Analysis of MSAP profiles obtained from DNA extractions of the first fully expanded leaf of 88 individual vines selected from 22 commercial vineyards within the six Barossa sub-regions (**Figure 1** and Supplementary Table S1) yielded 215 fragments comprising 189 from MspI and 211 from HpaII, of which 80 and 84%, respectively, were polymorphic (i.e., not present in all the analyzed samples/replicates when restricted with one of the isoschizomers). Comparison of the HpaII and MspI banding patterns showed that in average, 42.1% of analyzed bands represented fully methylated or SNP containing loci, 22.3% represented hemimethylated loci, 19.6% represented unmethylated loci, and 18.1% represented loci containing internal cytosine methylation (Supplementary Table S3).

## Analysis of Genome/Methylome Differences within Wine Sub-regions in the Barossa

Principal coordinate analysis of the MSAP profiles generated from NML (genetic variability) and by MSL (epigenetic variability) (Pérez-Figueroa, 2013) revealed a higher separation between vineyards when using epigenetic information than when using genetic (**Supplementary Figure S1**). The capacity of both types of variability to differentiate between vineyards was more evident on vineyards in the Southern Grounds (**Supplementary Figures S2G,H**). Both PCoA analysis and Shannon's diversity index showed significantly higher epigenetic than genetic diversity for all sub-regions (**Supplementary Figure S2** and **Table 1**). Among sub-regions, Southern Grounds had the highest epigenetic diversity (0.581 ± 0.124) and Western Ridge the lowest (0.536 ± 0.143). Genetic diversity showed the highest value in the Southern Grounds (0.374 ± 0.143) and the lowest in the Northern Grounds (0.240 ± 0.030).

## Analysis of Genome/Methylome Differences between Wine Sub-regions in the Barossa

We used AMOVA (**Table 2**) to obtain an overview of the molecular variability between all the studied sub-regions. Overall, MSAP profiles generated using restriction enzyme MspI achieved better separation between sub-regions than those generated using HpaII. Of all 30 calculated molecular pairwise distances between sub-regions (PhiPTs), 25 were significant (P < 0.05) (**Table 2**). Calculated PhiPT values ranged from 0.115 (PhiPT of Northern Grounds vs Southern Grounds calculated using MspI) and 0.012 (PhiPT of Central Grounds vs Eastern Edge calculated using HpaII).

Analysis of molecular variance on MSAP profiles indicates that the majority of the observed variability is explained by differences within vineyards (81% using profiles generated with MspI and 91% with HpaII). A significant proportion of the

total variability detected was associated to differences between vineyards (17% with MspI and 8% with HpaII) and 2 and 1% was due to differences between sub-regions (MspI and HpaII, respectively).

# Effect of Vineyard Location on Methylome Differentiation

To determine if environmental differences between vineyards influenced the observed epigenetic differences we studied the vineyards' pairwise geographic and molecular distances correlation. Vineyards located on the North–South axis of the Barossa Valley [i.e., vineyards 1, 2, 3, and 4 (Northern Grounds), 5, 6, 7, and 8 (Central Grounds), and 13, 14, and 15 (Southern Grounds)] (**Figure 2A**) were selected as Northern and Southern Grounds showed the greatest epigenetic

differentiation (**Table 2**). PCoA analysis showed that Central Grounds samples occupied an intermediate Eigen space between Northern and Southern Grounds samples with coordinate 1 (24% of the observed variability) representing the North– South axis (**Figure 2B**). Moreover, Mantel test showed a significant (P = 0.0003) positive correlation (R <sup>2</sup> = 0.3066) between pairwise vineyard epigenetic and GeoDs (**Figure 2C**). Then, Mantel test analysis was implemented to compare the observed molecular differences against environmental variables. Differences in vineyard altitude showed a small but significant positive correlations (R <sup>2</sup> = 0.1615, P = 0.013) with PhiPT values between vineyards (**Supplementary Figure S3**). We then investigated if clone and vineyard management systems could be contributing to this correlation, by comparing the epigenetic/GeoDs correlation of 10 vineyards planted with clone 1654 [vineyards 1 and 4 (Northern Grounds), 7 (Central Grounds), 9 and 12 (Eastern Ridge), 15 (Southern Grounds) 16, 17, 18, and 19 (Western Ridge) (**Figure 3A**)] and of six vineyards



Columns %Polym MSL and %Polym NML show the percentage of polymorphic loci of each class per region. Shannon diversity indices are reported as mean (± SD). Wilcoxon rank test provides statistical support for all Shannon diversity indices (P < 0.0001).


TABLE 2 | Molecular distances (PhiPT) between Barossa Valley wine producing sub-regions.

PhiPT values were calculated using MSAP profiles generated from 88 grapevine plants (four individual plants per vineyard) using restriction enzyme combinations MspI/EcoRI (above diagonal) and HpaII/EcoRI (below diagonal). P-values (shown in parenthesis) were calculated based on 9,999 permutations. Pairwise regional comparisons showing not significantly different PhiPT values are highlighted in bold. A total of 22 vineyards were included in the analysis: Northern Grounds (4), Central Grounds (4), Southern Grounds (3), four vineyards in Eastern Edge (4), four vineyards in Western Ridge (4) and Eden valley (3).

planted with the same clone (1654) and trained using the same pruning system (i.e., spur pruning) [vineyards 1 (Northern Grounds), 7 (Central Grounds), 9 (Eastern Ridge), 15 (Southern Grounds), 16 and 19 (Western Ridge) (**Figure 4A**)]. Again, PCoA shows that the main contributor (23–24%) to the detected variability is associated to the distribution of the vineyards on the N–S axis. Mantel test showed a positive correlation for both epigenetic/GeoD comparisons, however, although both correlations were significant (P < 0.05), the correlation among vineyards pruned using the same system (**Figures 4B,C**) was higher than that observed when all pruning systems were incorporated in the analysis (**Figures 3B,C**).

#### msGBS Analysis of Genome/Methylome Differentiation between Northern, Central, and Southern Grounds

TASSEL 3 was then implemented on msGBS data for SNP calling from 99 samples collected in 11 vineyards in the Northern, Central, and Southern Grounds sub-regions. This generated a total of 8,139 SNPs of which 4,893 were present in at least 80% of the sequenced samples. PCA analysis using filtered SNPs showed very low level of genetic structure, with only five plants from vineyard 3 (Northern Grounds) separating from the rest (**Supplementary Figure S4A**). However, this clustering was

FIGURE 3 | Analysis of the correlation between molecular differentiation and geographic distance (Km) of vineyards planted with clone 1654 in the Barossa region. (A) Location of the selected Barossa Valley vineyards from the three sub-regions distributed along the Barossa Valley North–South axis Northern Grounds (blue), Central Grounds (green), Eastern Edge (red), Southern Grounds (yellow), and Western Ridge (purple). Arrow indicates the direction of geographic North. (B) PCoA representing genetic/epigenetic differences between leaf samples collected from four plants/vineyard. Percentage of the variability captured by each PC is shown in parenthesis. (C) Correlation between pairwise genetic/epigenetic distance (MspI PhiPT) and geographical distance [Log(1 + GeoD) (Km)] between vineyards. Shown equations are the correlation's R <sup>2</sup> and the Mantel test significance (P-value was estimated over 9,999 random permutations tests). PCoA and PhiPT for Mantel test were based on presence/absence of 215 loci obtained from MSAP profiles generated using MspI.

not supported by DAPC (i.e., the optimal clustering solution should correspond to the lowest BIC) which indicated the optimal number of clusters for this data set is 1 (**Supplementary Figure S4B**) suggesting a lack of genetic structure in the vineyards/regions analyzed.

Principal components-linear discriminant analysis (PC-LDA) was then used to visualize differences in DNA methylation detected using msGBS. DNA methylation profiles clustered samples by their sub-region of origin, with Northern and Central Grounds being separated by differential factor (DF1) from Southern Grounds while DF2 separated Northern from Central Grounds (**Figure 5**). These results were supported by the higher number of DMMs found when comparing samples from Southern to samples from Central or Northern Grounds than when comparing Northern to Central (**Table 3**).

We next investigated the association of the detected DMMs to annotated protein-coding genes in the grapevine genome by surveying their location and density within and flanking such genomic features. A total of 3,598 genes were deemed differentially methylated (i.e., presented one or more DMMs within 5 kb of the TSS or the TSE) or within genes (**Table 3**). Quantification of such DNA methylation changes showed that, in average, methylation levels are higher in the northern most region in each comparison (i.e., NG > CG > SG) (**Figure 6A**). The majority of detected DMMs associated to a gene were present in the body of the gene and the number of DMMs decreased symmetrically with distance from the TSS and the TES (**Figure 6B** and Supplementary Tables S4–S6). Finally, as observed with all DMMs, the comparison between Northern and Central Grounds samples showed the lowest number of DMGs (**Table 3**, **Figure 6C**, and Supplementary Tables S4–S6).

To gain further insight into the functional implications of the DNA methylation differences detected between sub-regions, we used GO.db (Carlson, 2016) and annotate (Gentleman, 2016) to assign 1,144 unique GO terms to the observed DMGs (adjusted P-value < 0.05). As observed with DMMs and DMGs the comparison between Northern and Central Grounds samples showed the lowest number of GO terms containing DMGOs (**Table 3**, **Figure 6C**, and Supplementary Tables S4–S6). REViGO semantic analysis of GO terms shared by all three pairwise regional comparisons (**Figure 7**) showed an increase of gene enrichment (i.e., a decrease in adjusted P-values) with GeoD (e.g., see **Figure 7** for comparisons between Northern Grounds and Southern Grounds (A,B) and Central Grounds and Northern Grounds (C,D). Three hundred and eleven DMGs (8.6% of the total) were allocated in GO terms associated to

response to environmental stimulus (161 and 150 abiotic and biotic challenges, respectively) (**Figure 7** and Supplementary Tables S7, S8), which included GO terms in the semantic space of plant response to light, temperature, osmotic/salt stress and defense to biotic stimulus.

# DISCUSSION

In this study, we analyzed the effect of growing region on the methylation profiles of Shiraz plants using MSAP and msGBS. Both techniques use methylation sensitive enzymes to discover DNA methylation polymorphisms between samples. Although the use of restriction enzymes has the obvious limitation of being capable of detecting such polymorphisms only on the context of their recognition sequence, the technology has been extensively validated over the last 20 years and is considered highly reliable (Yaish et al., 2014; Li et al., 2015; Guevara et al., 2017).

# Grapevine DNA Methylation Patterns Are Region Specific

Analysis of HpaII and MspI generated MSAP profiles showed that the methylation profiles of the six different sub-regions were significantly different (P < 0.05) in 25 of the 30 possible pairwise comparisons (**Table 2**). Variability among vineyards and sub-regions was higher in MspI generated profiles (17 and 2%) than in HpaII profiles (8 and 1%), indicating that the

TABLE 3 | Identification of DMMs, DMGs, and GO terms (DMGOs) between sub-regions in Barossa Shiraz.


Cells contain the number of DMMs, DMGs, or DMGOs detected in each pairwise comparison. Differential methylation between sub-regions was calculated using msGBS data from nine Shiraz plants per vineyard [Northern Grounds (NG): four vineyards, Central Grounds (CG): four vineyards and Southern Grounds (SG); three vineyards]. Directionality of methylation (i.e., hyper/hypomethylation) indicates an increase or decrease in DNA methylation in the second region compared to the first region in each pairwise comparison.

FIGURE 6 | Analysis of DMGs and GO terms (DMGOs) among three wine sub-regions in Barossa Shiraz. Genes were considered differentially methylated if located within 5 kb of at least one DMM (FDR < 0.01). DMMs were generated using msGBS on nine plants per vineyard (Northern Grounds: four vineyards, Central Grounds: four vineyards, and Southern Grounds; three vineyards). (A) Directionality of methylation differences between regions. Boxplots show the distribution of the intensity of changes in DNA methylation level between regions, represented here as the fold-change (2 power log2FC) in read counts for a given msGBS markers between two regions. Median shows the direction of the methylation flux at a whole genome level in each region comparison (i.e., positive medians indicate a global increase in DNA methylation (hypermethylation) while negative medians indicate a global decrease in DNA methylation (hypomethylation) in the second region in the (Continued)

#### FIGURE 6 | Continued

comparison (e.g., Northern Grounds is hypermethylated compared to Southern Grounds). (B) Distribution of 3598 region specific DMMs around genes. Columns –5 to –1 and 1 to 5 represents the number of DMMs per kb around V. vinifera genes. Column 0 indicates the number of DMMs within the coding sequence (i.e., between the transcription start and end sites) of V. vinifera genes. (C) Shared DMGs and DMGOs between regional comparisons. Venn diagrams show the number of unique and shared DMGs and DMGOs between each regional pairwise comparison (i.e., Blue: hyper/hypomethylated genes and GOs in Northern Grounds compared to Southern Grounds; Yellow: in Central Grounds compared to Southern Grounds; and Green: in Central Grounds compared to Northern Grounds).

detected regional epigenetic differences are, at least partially, sequence context specific (Tricker et al., 2012; Meyer, 2015). Calculated PhiPT values showed low levels of molecular differentiation between sub-regions, even when those differences were statistically significant (**Table 2**). This could be explained by the high proportion of the total variability associated to differences between individual plants (81–91%) compared to 1–2% associated to differences between sub-regions. Such high levels of molecular differentiation between individuals could be due to the random accumulation of somatic variation with age, which can be genetic or epigenetic in nature. A specific limitation of MSAPs is its inability to distinguish between a fully methylated site from a site containing a genetic mutation (Yaish et al., 2014). PCA of genetic polymorphisms detected using msGBS results showed a high level of genetic variability between plants (**Supplementary Figure S4A**) which is characteristic of long living plants in general (Baali-Cherif and Besnard, 2005) and in grapevine in particular (Torregrosa et al., 2011). However, DAPC did not detect any sample clustering associated to their sub-region of origin (**Supplementary Figure S4B**) indicating that genetic diversity is not structured in a geographic manner. Although both genetic and epigenetic somatic variation can be random (Vogt, 2015), different growing conditions will differentially affect the DNA methylation profiles of otherwise genetically identical individuals (Consuegra and Rodríguez López, 2016) as previously shown on clonally propagated Populus alba (Guarino et al., 2015). It is, therefore, not surprising to find that epigenetic profiling was a better predictor of sample origin than genetic profiling alone both using MSAP data (**Table 2** and **Supplementary Figure S1**) or msGBS data (**Figures 5**–**7** and **Supplementary Figure S4**). This suggests that although genetic differences between regions or vineyards can partly contribute to the observed molecular differentiation between vineyards/subregions, epigenetic differences are the major driver of such differentiation.

Previous studies have shown that in some instances clonal variability in grapevine is better explain by epigenetic than genetic differences (Imazio et al., 2002; Schellenbaum et al., 2008; Ocaña et al., 2013). It is therefore possible, that the epigenetic differences observed here are not associated to regional environmental differences but that they were present since the time of planting or due to environmental variations that may have occurred at the time of the sampling. For this reason, further research including information from more than one season is

needed to validate the DNA methylation differences between

of differentiation when inter-vineyard variability was analyzed (**Supplementary Figures S2G,H**), suggesting a major contributor to the observed molecular variability between vines in the Southern Grounds is linked to the vineyard of origin. Taken collectively, these results suggest that the specific growing conditions from each subregion impose DNA methylation patterns on grapevine plants specific for each region as previously shown both in cultivated (Guarino et al., 2015) and wild plant populations (Fonseca Lira-Medeiros et al., 2010). Not surprisingly, and contrary to what has been shown in natural plant populations (Fonseca Lira-Medeiros et al., 2010; Róis et al., 2013), no clear negative correlation between genetic and epigenetic diversity was observed in the studied vineyards. This is most probably due to the intensive phenotypic selection to which grapevine cultivars have been under since domestication and the relative low levels of environmental disparity to which vines growing in the same vineyard are exposed to.

## Differences in Altitude and Pruning System Correlate with Vineyard Epigenetic Differentiation

Principal coordinate and Mantel test analysis showed that the correlation between epigenetic and GeoD between vineyards on the North–South axis of the Barossa Valley (**Figure 2A**) was significant (P = 0.0003) (**Figure 2C**) and that the main contributor to the observed epigenetic differences was the position of the studied vineyards along the N–S axis (**Figure 2B**). This suggests that environmental differences between locations could be contributing to the observed molecular differences between sub-regions or vineyards (**Figure 3**). Moreover, the correlation (R <sup>2</sup> = 0.3066) between epigenetic and GeoD among vineyards planted with clone 1654 on the N–S axis (**Figure 2**) supports the Shannon diversity analysis that indicate that the different genetic backgrounds used in this study do not greatly affect the epigenetic differences observed between regions (**Table 1**). Conversely, differences in vineyard altitude appear to be a contributor to the detected epigenetic differentiation between vineyards (**Supplementary Figure S3**). Previous work

FIGURE 7 | REViGO semantic analysis of differentially methylated GO terms shared by all three regional pairwise comparisons. Functional enrichment of GO-terms was carried out for the genes deemed differentially methylated (DMGs) hypermethylated (185) (A,C) or hypomethylated (211) (B,D) in Northern Grounds compared to Southern Grounds (A,B) and Central Grounds compared to Northern Grounds (C,D) using GO.db and annotate and summarized using REViGO. Bubble color indicates the p-value for the FDRs (the first 10 terms are labeled with legends in black. A detailed list of all GO terms containing DMGs has been supplied as a Supplementary Tables S7 and S8); circle size indicates the frequency of the GO term in the underlying GO database (bubbles of more general terms are larger).

fpls-08-01860 October 27, 2017 Time: 16:50 # 11

has shown that sun exposure can have significant effects both in berry metabolomic profiles (Son et al., 2009; Tarr et al., 2013) and on the epigenetic profiles of plants growing in different environments (Guarino et al., 2015). Although altitude does not necessarily affect sun exposure, it can have a profound effect on the UV levels experienced by plants (approximately 1% increase every 70 m gain in altitude). Our results suggest that, although DNA methylation in and around genes changes in both directions (hyper- and hypo-methylation), on average, it increases with altitude (i.e., NG > CG > SG; vineyard average altitude 301, 277, and 236 m, respectively) (**Figure 6A**).

Due to the nature of the msGBS approach used here, all sequenced DMMs are in the CHG context. Global methylation levels on this context varies widely between plant species (9.3% in Eutrema salsugineum to 81.2% in Beta vulgaris) (Niederhuth et al., 2016). In V. vinifera, genome-wide weighted CHG methylation level is 20.4%, more than double than that found in A. thaliana (Niederhuth et al., 2016). Although the analysis of DNA methylation has traditionally focus on the CG context, CHG differential methylation has been reported to be more prominent than CG differential methylation in other perennial crops (i.e., Apple, Malus domestica) (Daccord et al., 2017). In this study, the majority of detected DMMs associated to a gene were present in the body of the gene (**Figure 6B**). The function of gene body methylation (GbM) is not yet well understood (Zilberman, 2017) and recent studies have shown that GbM can be lost over evolutionary time with no deleterious consequences, suggesting that it might not be required for plant viability (Bewick et al., 2016). However, plant accessions with higher average GbM have been shown to have higher average expression of gene body methylated genes (Dubin et al., 2015; Wang et al., 2015). Moreover, GbM has also been proposed as a regulator of alternative splicing (Wang et al., 2016) and suppressor of intragenic cryptic promoters and transposable element (Maunakea et al., 2010). In particular CHG GbM has been associated to the silencing of genes lacking CG GbM (Niederhuth and Schmitz, 2017) and to the repression of splicing in maize (Regulski et al., 2013). Remarkably, global methylation levels of the CG, CHG, and CHH contexts has been proposed to be an adaptive trait to environmental variables such as latitude, aridity and photosynthetically active radiation temperature, respectively (Dubin et al., 2015). It is, therefore, tempting to speculate that the differences in GbM observed between regions reflect plant adaptation to their local environments that could be affecting alternative splicing, which has been itself been proposed as an adaptive mechanism (Ast, 2004).

Functional analysis of the DMGs between sub-regions generated GO terms associated to plant response to light stimulus (Supplementary Table S8). More importantly, the number of genes associated to such GO terms was higher in comparison between regions with bigger differences in altitude [74 and 46 genes in comparison SG vs. NG (65 m difference) and SG vs. CG (41 m), respectively] than in the pairwise comparison with lower difference in altitude (6 genes NG vs. CG (24 m)]. Although this positive polynomial grade 2 correlation (R <sup>2</sup> = 1) was generated using only three data points, it is tempting to speculate that differences in light incidence due to differences in altitude are triggering the observed changes in DNA methylation in response to light stimulus genes. Especially when previous work has shown that, in grapevine leaves, increased UV levels trigger the synthesis of non-flavonoid phenolics such as resveratrol (Sbaghi et al., 1995; Teixeira et al., 2013). Interestingly, DNA methylation has been previously linked to the regulation of the gene VaSTS10, which controls the synthesis of resveratrol in Vitis amurensis (Kiselev et al., 2013; Tyunin et al., 2013).

The correlation between epigenetic and GeoDs observed between vineyards planted with clone 1654 and pruned with the same method (spur pruning) (**Figure 4**) was reduced when all vineyards planted with clone 1654 were considered irrespectively of the pruning system used (**Figure 3**). This suggests that differences in pruning system, in conjunction with environmental conditions, might be contributing to the epigenetic differences observed between vineyards and subregions in this study. However, further research on the effect of the observed change in DNA methylation with vineyard altitude and pruning on gene expression are needed to validate the hypothesis that such changes might be regulating plant adaptation to such environmental cues.

# CONCLUSION

Vintage, geographic location, and vineyard management have been shown to influence both vegetative growth (Jackson and Lombard, 1993) and fruit composition in grapevine (Roullier-Gall et al., 2014). In light of the results shown here, we propose that epigenetic processes in general and DNA methylation in particular, could constitute an important set of molecular mechanisms implicated in the effect that provenance and vintage has, not only on plant vegetative growth, but also on fruit and wine quality. It is important to stress that since global patterns of DNA methylation are tissue/organ specific (Rodríguez López et al., 2010), the observed differences in DNA methylation profiles between plants growing in different regions can only be taken as indicative of those occurring in leaves. However, the effect of the environment on the epigenetic profiles of different tissues in plants reflects their mode of development. That is, unlike mammals, plants growth and organ formation occurs from stem cell populations in the meristems (Pikaard and Mittelsten Scheid, 2014). Previous studies (e.g., Verhoeven et al., 2010; Tricker et al., 2013a) have shown that environmentally induced markers detected in leaf tissue can be found on subsequent generations. This indicates that the DNA methylation markers observed in leaves were also present in the meristematic tissue that ultimately produced the reproductive organs. For this reason, it is plausible to expect that region-specific markers detected in grapevine leaves, could also be present in other organs such as berries since these are originated from the same shoot apical meristems as leaves. Although preliminary, our results open the door to speculate that epigenetic priming (Tricker et al., 2013b) could act as a form of epigenetic memory of the vineyard's environment that would ultimately contribute, at least partially, to the uniqueness of wines produced in different regions. Testing this hypothesis will require the integrative analysis of fruit DNA methylation, gene expression, and metabolite composition data from multiple seasons to account for the effect of inter-annual climatic variations on fruit composition (Fabres et al., 2017).

## AUTHOR CONTRIBUTIONS

fpls-08-01860 October 27, 2017 Time: 16:50 # 13

HX and MK carried out the experiments and contributed to data analysis. NS performed gene ontology analysis on msGBS data. KT performed TASSEL analysis on msGBS data. TC, MG, JB, AM, and JS contributed to the design of the research project. RDB and CC contributed to the design of the research project, site selection and collection of material. CL contributed to the design of the research project, data analysis and drafted the manuscript. All authors read and contributed to the final manuscript.

#### FUNDING

This study was funded through a Pilot Program in Genomic Applications in Agriculture and Environment Sectors jointly supported by the University of Adelaide and the Australian Genome Research Facility Ltd. MK was supported by the Australian Agency for International Development (AusAID) Ph.D. scholarship. CL is supported by a University of Adelaide Research Fellowship. MG is supported by the Australian Research Council through Centre of Excellence (CE1400008) and Future Fellowship (FT130100709) funding.

#### ACKNOWLEDGMENTS

The authors would like to gratefully acknowledge the Barossa Grounds Project and in particular the growers that allowed us to sample material from their properties and supplied information about their vineyards and management strategies. Dr. Kendall R. Corbin performed DNA extractions from all samples used in this study. Personnel in the viticulture group, Dr. Sandra Milena Mantilla, Annette James, and Valentin Olek contributed to collection of material.

# REFERENCES


# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2017.01860/ full#supplementary-material

FIGURE S1 | Analysis of genetic and epigenetic differences within Barossa Valley sub-regions. PCoAs represent variability of NML (genetic variability) (A,C,E,G,I,K) and of MSL (epigenetic variability) (B,D,F,H,J,L) as classified by the msap (v. 1.1.8) software (Pérez-Figueroa, 2013) of leaf samples in vineyards from Northern Grounds (A,B), and the Barossa Valley's Western Ridge Central Grounds (C,D), Eastern Edge (E,F), and Southern Grounds (G,H), Western Ridge (I,J), and Eden valley (K,L). Coordinates 1 and 2 are shown with the percentage of variance explained by them. Points represent individuals from each vineyard. Vineyard code (NG, CG, EE, SG, WR, and EV) is shown as the centroid for each vineyard. Ellipses represent the average dispersion of those poinst around their center. The long axis of the ellipse shows the direction of maximum dispersion and the short axis, the direction of minimum dispersion.

FIGURE S2 | Analysis of regional genetic and epigenetic diversity. Red symbols indicate samples analyzed using genetic information only, black symbols represent samples analyzed using epigenetic information only according to the R package for statistical analysis of MSAP data "msap." PCoAs were calculated using MSAP profiles generated from gDNA extracted from E-L 7-stage leaves (Coombe, 1995) of 88 grapevine plants grown in vineyards located in the six Barossa Valley wine sub-regions (A, Northern Grounds; B, Central Grounds; C, Southern Grounds; D, Eastern Edge; E, Western Ridge; F, Eden valley) (four individual plants per vineyard) using restriction enzyme combinations MspI/EcoRI and HpaII/EcoRI.

FIGURE S3 | Analysis of the correlation between epigenetic differentiation and environmental differences among vineyards planted along the Barossa Valley North–South axis: Mantel test analysis of the correlation between pairwise epigenetic distance (MspI PhiPT) and differences in altitude between vineyards. Shown equations are the correlation's R <sup>2</sup> and the Mantel test significance (P-value was estimated over 9,999 random permutations tests). PhiPT values were based on presence/absence of 215 loci obtained from MSAP profiles generated using MspI.

FIGURE S4 | Analysis of the grapevine genetic diversity in vineyards planted along the Barossa Valley North–South axis. (A) PCA representing genetic structure calculated using 4,893 high quality SNPs (i.e., present in at least 80% of the samples) in genomic DNA collected from 11 vineyards [Northern Grounds (blue) four vineyards, Central Grounds (green) four vineyards, and Southern Grounds (yellow) three vineyards (nine plants/vineyard)]. (B) Identification of the optimal number of genetic clusters present within the three sub-regions compared using BIC as implemented by DAPC using adegenet 2.0.0 (i.e., the optimal clustering solution should correspond to the lowest BIC).


complex traits in diverse samples. Bioinformatics 23, 2633–2635. doi: 10.1093/ bioinformatics/btm308


fpls-08-01860 October 27, 2017 Time: 16:50 # 14



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Xie, Konate, Sai, Tesfamicael, Cavagnaro, Gilliham, Breen, Metcalfe, Stephen, De Bei, Collins and Lopez. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Prospect on Ionomic Signatures for the Classification of Grapevine Berries According to Their Geographical Origin

Youry Pii<sup>1</sup> \*, Anita Zamboni<sup>2</sup> , Silvia Dal Santo<sup>2</sup> , Mario Pezzotti<sup>2</sup> , Zeno Varanini<sup>2</sup> and Tiziana Pandolfini<sup>2</sup>

<sup>1</sup> Faculty of Science and Technology, Free University of Bozen-Bolzano, Bolzano, Italy, <sup>2</sup> Department of Biotechnology, University of Verona, Verona, Italy

The determination of food geographical origin has been an important subject of study over the past decade, with an increasing number of analytical techniques being developed to determine the provenance of agricultural products. Agricultural soils can differ for the composition and the relative quantities of mineral nutrients and trace elements whose bioavailability depends on soil properties. Therefore, the ionome of fruits, vegetables and derived products can reflect the mineral composition of the growth substrate. Multi-elemental analysis has been successfully applied to trace the provenance of wines from different countries or different wine-producing regions. However, winemaking process and environmental and cultural conditions may affect a geographical fingerprint. In this article, we discuss the possibility of applying ionomics in wines classification on a local scale and also by exploiting grape berry analyses. In this regard, we present the ionomic profile of grapevine berries grown within an area of approximately 300 km<sup>2</sup> and the subsequent application of chemometric methods for the assignment of their geographical origin. The best discrimination was obtained by using a dataset composed only of rare earth elements. Considering the experiences reported in the literature and our results, we concluded that sample representativeness and the application of a preliminary Principal Component Analysis, as pattern recognition techniques, might represent two necessary starting points for the geographical determination of the geographical origin of grape berries; therefore, on the basis of these observations we also include some recommendations to be considered for future application of these techniques for grape and wines classification.

Keywords: grape, wine, ionomic profile, traceability, rare earth elements, ICP-MS

#### INTRODUCTION

The geographical origin and the authenticity of food products are often related to the overall perception that consumers have in terms of quality, thus having a strong impact on the commercial value of the goods. In the last decades, fingerprinting techniques based on the chemical analyses of agricultural products followed by multivariate statistical approaches have been developed, aiming at identifying and classifying products according to their geographical origin (Versari et al., 2014). These methods assume that the chemical composition of the food product under study (e.g., mineral elements, stable isotopes ratios, and metabolites) is depending on the provenance

#### Edited by:

José Tomás Matus, Centre for Research in Agricultural Genomics, Spain

#### Reviewed by:

Jose Antonio Alcalde, Pontifical Catholic University of Chile, Chile Andreas Zitek, University of Natural Resources and Life Sciences, Vienna, Austria

> \*Correspondence: Youry Pii youry.pii@unibz.it

#### Specialty section:

This article was submitted to Technical Advances in Plant Science, a section of the journal Frontiers in Plant Science

> Received: 16 January 2017 Accepted: 10 April 2017 Published: 24 April 2017

#### Citation:

Pii Y, Zamboni A, Dal Santo S, Pezzotti M, Varanini Z and Pandolfini T (2017) Prospect on Ionomic Signatures for the Classification of Grapevine Berries According to Their Geographical Origin. Front. Plant Sci. 8:640. doi: 10.3389/fpls.2017.00640

**213**

environment (Versari et al., 2014). The fingerprinting methods based on mineral element composition of food stuff have been largely adopted in the last years to trace the geographical origin of wine, olive oil, honey, cheese, coffee, vegetable, fruits, and spices (Danezis et al., 2016). One of the most popular techniques adopted for these analyses is the inductively-coupled plasma mass spectrometry (ICP-MS), which can be used for the determination of both the ionomic profile and the isotope ratios (Baxter et al., 1997; Rebolo et al., 2000; Wieser et al., 2001; Castiñeira et al., 2004; Coetzee et al., 2005; Šelih et al., 2014; Capici et al., 2015; Mimmo et al., 2015; Popescu et al., 2015; Scampicchio et al., 2016).

In the particular case of agricultural products, it is postulated that the presence and the concentration of the mineral elements might reflect their geographical provenance (Almeida and Vasconcelos, 2003). Considering that the natural diffusion of mineral elements follows a pathway starting from the rocks, going through the soil and, finally, reaching the plant, it is thus conceivable that ionomic profile of plant organs and tissues is dependent on the geochemistry of the soil on which crops are cultivated (Geana et al., 2013). In addition, also anthropogenic activities, including the soil management, the use of fertilizers and phytochemicals, might determine alterations in the ionomic signature of agricultural products (Pepi et al., 2016a).

#### TRACEABILITY OF WINES

Among a wide variety of experiences dedicated to the geographical tracing of food products, a large number of studies, aiming at finding out reliable fingerprinting methods, have been carried out on wines traceability, most probably due to their relatively high commercial value. For these reasons, the elemental composition of different type of wines have been investigated with the aim of correlating them to the provenance soil for geographical tracing purposes (for an extensive review see Versari et al., 2014) (**Table 1**). However, the critical reading of the scientific literature published in this field of research demonstrates that the determination of the chemical descriptors for the origin of wines are strongly dependent on a plethora of factors, as for instance the number of samples used in the analyses, the type of wine (i.e., white, red, or rosè), the pattern recognition technique applied for the statistical analysis [e.g., Discriminant Analysis, Principal Component Analysis (PCA), Cluster Analysis, Stepwise Linear Discriminant Analysis and similar] and, most importantly, the geographical origin (Baxter et al., 1997; Díaz et al., 2003; Marengo and Aceto, 2003; Castiñeira et al., 2004; Jos et al., 2004; Thiel et al., 2004; Coetzee et al., 2005, 2014; Angus et al., 2006; Capron et al., 2007; Galgano et al., 2008; Serapinas et al., 2008; Forina et al., 2009; Fabani et al., 2010; Catarino et al., 2011; Rodrigues et al., 2011; Martin et al., 2012; Zou et al., 2012; Azcarate et al., 2013; Geana et al., 2013; Šelih et al., 2014). As also shown in **Table 1**, the majority of geographical tracing studies explores the analytical dataset by means of unsupervised pattern recognition analyses (e.g., PCA) and, once the most discriminant variables have


been found, ad hoc statistical analyses, specifically supervised methods, are run in order to exacerbate the clusterization and to extract further information from the dataset. In addition, it could also be inferred that the power of the technique exploited for the chemical analysis (i.e., AAS, ICP-OES, ICP-AES, and ICP-MS) might be determinant for the tracing purposes.

Therefore, the use of multi-elemental profile of wines as a fingerprinting technique requires the careful identification of suitable elements that generally reflect the characteristic features of the provenance soil. Thus, the correlation between chemical composition of wine and provenance soil is usually considered an important prerequisite for classification of wines according to geographical origin. In wine analyses, the choice of the indicator elements should also take into account possible distortions due to agricultural practices, environmental conditions and winemaking process.

Winemaking is a complex process that involves multiple stages, as for instance blending, fermentation, rectification, and clarification, which have shown to influence element concentration in the final product. Early studies have shown that the concentration of elements may either increase (e.g., Al, Cd, Cr, Fe, Pb, and V) (Kristl et al., 2002; Almeida and Vasconcelos, 2003) or decrease (i.e., Al, Cd, Co, Cr, Fe, Pb, and V) (Eschnauer et al., 1989) in the processes of must fermentation and wine fining. More recently, Aceto et al. (2013) carried out a geographical tracing study on Moscato wines and demonstrated that the concentration of lanthanides, used as chemical markers, is conserved from soil to must, whilst the fingerprinting was affected by the treatments with bentonites. These observations led to the conclusion that wine traceability could be only pursued if the fining treatments were alternative to the bentonites ones (Aceto et al., 2013).

#### GEOGRAPHICAL ORIGIN OF GRAPE BERRIES USING RARE EARTH ELEMENTS AS CHEMICAL DESCRIPTORS: THE VERONA REGION CASE

Some of the limitations in wine fingerprinting, as discussed above, may be overcome by analyzing the chemical composition of berries. In particular, this would circumvent the problems associated with the chemical changes caused by winemaking, in particular for Rare Earth Elements (REEs).

It is widely accepted that the composition in terms of REEs in the rocks is reflected also in the soil and in the plant tissues, even though a certain degree of variability is observed depending on the plant species (Wang et al., 1997; Wyttenbach et al., 1998; Zhang et al., 2002; Oddone et al., 2009). In the case of Vitis vinifera, the distribution of REEs within the berries has been studied in different cultivars (e.g., Chardonnay, Cabernet Sauvignon, Italian Riesling) by ICP-MS techniques (Bertoldi et al., 2009; Yang et al., 2010). In particular, Bertoldi et al. (2009) were able to show that Europium was accumulated in grape berries seed. Collectively, the results obtained within these studies prompted other authors to exploit these features (i.e., REEs) to study the geographical origin of wines, also considering the recent evidence demonstrating that different rootstocks do not significantly affect the REE content in the grape berries (Pisciotta et al., 2017).

For instance, the afore-mentioned Moscato tracing work represents a comprehensive study in which the wine production chain has been investigated for geographical discrimination purposes and the correlation between the soil composition in terms of REEs and their concentration of berries and musts has been examined (Aceto et al., 2013). The unsupervised pattern recognition analyses carried out on musts did not highlight any difference between samples collected in the Moscato DOCG geographical region (Aceto et al., 2013). On one hand, these results indeed highlighted the power of REEs to assess the belonging of Moscato samples to the DOCG area; on the other, no striking distinctions between the different vineyards were found. Further insight in the use of REEs for the determination of geographical provenance was obtained with a study considering the REEs profile in berries of the "Glera" cultivar sampled in five different vineyards in the Veneto region, Italy (Pepi et al., 2016b). The authors established a correlation between the REEs concentration in the berries and REEs available fraction in the soil; this indeed allowed the discrimination of the provenance (Pepi et al., 2016b), provided that the geological origin of the soil in the vineyard considered across the Veneto region was fairly diverse.

Few examples also indicated the possibility to use the trace element composition for classification of wines produced in wine growing regions located in small geographical area (Coetzee et al., 2005; Šelih et al., 2014). In this context, we decided to investigate whether the ICP-MS multi-elemental analysis followed by multivariate statistical analyses could be effective in the distinction of grape samples originating from neighboring vineyards (within an area of 300 km<sup>2</sup> ). We harvested berries of V. vinifera cv. Corvina (clone 48) at full ripening stage (Brix degree ranging between 18 and 24) from eleven vineyards located in the three most important wine production macroareas of the Verona region, namely Bardolino, Valpolicella, and Soave. The sampling rationale and procedures, as well as the sampling sites, were previously described by Dal Santo et al. (2013) and Anesi et al. (2015). The samples were homogenized, mineralized, and the concentration of 34 mineral elements in grapevine berries was determined by ICP-MS. These data were used as chemical descriptors to establish, through chemometric methods, criteria for assigning their geographical origin.

In order to obtain a comprehensive view of the whole dataset, the concentration of mineral elements was used to build a heat map, in which each value has been calculated as the log2 of the ratio between the element concentration in the sample and the average concentration of that element in all the samples (**Figure 1A**). Within the heat map, it was

used for the scatter plot accounts for 59.50% of the total variance. (C) The PCA analysis was carried out considering only micro-, macro- and beneficial elements generating a model composed of five components, which described 96.97% of the total variance. The scatter plot was obtained by combining the first two components, accounting for 74.92% of the variance. (D) The PCA analysis was carried out considering only REEs generating a model composed of two components, which describe 98.65% of the total variance. (E,F) Scatter plot obtained by PCA of the sub-data set encompassing only micro-, macro- and beneficial elements. The age of the vineyards (E) and the rootstock genotype (F) are highlighted.

possible to differentiate two big groups of elements: a first group, encompassing mostly plant macro- and micronutrients (except for Ga and Rb), which showed strong variations, both positive and negative, and a second group that was formed mostly by REEs, which showed milder fluctuations between samples (**Figure 1A**). Micro- and macronutrients, as well as beneficial elements (Marschner, 2011), are actively taken up, accumulated, and differentially allocated in tissues and organs, and their homeostasis is tightly regulated in order to avoid nutritional imbalances (Williams and Salt, 2009). Therefore, substantial variations in the micro- and macronutrient concentrations are to be expected, depending on the soil type and on the rootstock genotype. On the other hand, REEs are not essential to plants; still they can be absorbed following the route of Ca, with which they share a similar ionic radius (Pickard, 1970; Hu et al., 2004). This is supported by the observations that Ca can be replaced by REEs in several biochemical and physiological functions (Pickard, 1970; Hu et al., 2004; Liu and Hasenstein, 2005; Babula et al., 2008; Xiaoqing et al., 2009; Carpenter et al., 2015; Yang et al., 2015). Besides the natural variations of element concentrations due to soil characteristics and origin, it is noteworthy that Cu, in 5 samples out of 11, was more abundant as compared to the global average value (**Figure 1A**). This behavior, also documented by Geana et al. (2013) in Romanian wine samples, might be due to Cu accumulation in soil, following the agronomical practice of using Cu-based fungicides for the protection of grapevine plants against downy mildew.

Pattern recognition analyses were carried out on the whole dataset in order to highlight possible differences and similarities among the samples considered, finally aiming at the geographical origin discrimination. The PCA generated a six-component model, accounting for a total variance of 97.66%. The first two components, which together explained about 59.50% of the total variance, have been used to graphically represent the model (**Figure 1B**). The validity of the PCA models were assessed by the cross-validation approach previously described Bro et al. (2008) and Pii et al. (2015). Despite accounting for more than half of the total variance, the model obtained failed in describing the geographical provenance of samples, since they resulted randomly scattered across the diagram, except for three samples belonging to the Bardolino area that closely clustered together in the same quarter (**Figure 1B**).

According to differences in element behavior displayed in the concentration heat map (**Figure 1A**), the whole dataset was split into two sub-datasets, the first encompassing micro-, macro- and beneficial elements, and the second comprising REEs, and they were subjected to PCA (**Figures 1C,D**).

The multivariate analysis of the first sub-dataset (i.e., micro-, macro- and beneficial elements) generated a five components model, accounting on the whole for 96.96% of the total variance. The scatter plot obtained combining the first two components, which represented 74.92% of the variance, showed neither the separation of samples according with the geographical origin nor any other clear clustering (**Figure 1C**). Indeed, the distribution along the first component was mainly driven by Cu and, to a lower extent, by Fe, Na and B (Data not shown). As previously discussed, the differential accumulation of Cu could be due to agronomical practices (Geana et al., 2013); nonetheless, it has also been observed that the element composition of berries is also dependent on the rootstock genotype (Ga¸stoł and Domagała-Switkiewicz, ´ 2013). For this reason, both the age of the vineyard and the rootstock genotype have been highlighted within the PCA model discussed above (**Figures 1C,E,F**). In spite of this, any clear clustering regarding the classification of the samples considered (i.e., vineyard age and rootstock genotype) was obtained (**Figures 1E,F**).

On the other hand, when only the REEs were considered for PCA, a two-components model accounting for 98.65% of the total variance has been obtained (**Figure 1D**). The scatterplot showed the separation of samples into two distinct clusters along the first component of the model, one encompassing samples from Bardolino and Valpolicella vineyards and the other comprising Soave vineyards. Nevertheless, two outliers from the other vineyards clustered with Soave samples (**Figure 1D**). According to the loading plot, the separation along the first component was mainly driven by Lu, whereas the other REEs contributed to the separation of samples along the second component (data not shown). This behavior might be due to the fact that Lu showed the strongest variation in concentration among the REEs group (**Figure 1A**). To the best of our knowledge, Lu has not emerged as discriminant element yet.

## CONCLUSION AND RECOMMENDATIONS FOR FUTURE STUDIES

In conclusion, the new data presented here showed that the whole ionomic signature of the grape berries did not fully allow the discrimination of their geographical origin, most likely due to the heterogeneity in the characteristics (i.e., vineyards age, rootstock genotypes and agricultural practices) of the vineyards and the limited number of samples analyzed. Nonetheless, our data confirm that the multielemental analyses based on REEs of agricultural products might be a powerful technique to trace the geographical origin of foodstuff, and, in this specific case, of grapevine berries and musts. Furthermore, our data indicate that the ionomic signature can be suitable even for agricultural products originating from neighboring regions.

On the bases of this experience and the pieces of research published in the literature, we suggest making the following recommendations, which may be considered in the experimental design, aiming at improving the efficacy and the resolution of the predictive tool:


The commercial value of wines greatly depends on the authentication of their geographical origin, which represents a benefit for both consumers and wine producers. The ionomic signature appears as a powerful and flexible method to trace wine provenance even at the level of wine-producing sub-regions. Its

flexibility relies on the availability of multiple elemental markers, different types of samples (wine, must, grape) for chemical analysis and numerous analytical and statistical methods. The optimization of these parameters, as well as the application of a sufficiently large number of variables, may allow tailoring the experimental set up for each wine-making area.

#### AUTHOR CONTRIBUTIONS

YP, AZ, ZV, and TP: Designed the experiments. YP and AZ: Samples and Data Analyses. SD and MP: Provided the samples

#### REFERENCES


and information about the vineyards. YP, AZ, ZV, and TP: Critical discussion of the data. YP and TP: Paper preparation. TP: Research coordination.

#### FUNDING

This project was financed by "Fondo Sociale Europeo nel Veneto" Project Number 1695/1/1/1103/2010, Project Title "Sviluppo di metodi analitici per la tracciabilità e l'autenticazione dei vini" with the partnership of Unione Italiana Vini.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Pii, Zamboni, Dal Santo, Pezzotti, Varanini and Pandolfini. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Induction of Noble Rot (*Botrytis cinerea*) Infection during Postharvest Withering Changes the Metabolome of Grapevine Berries (*Vitis vinifera* L., cv. Garganega)

Stefano Negri 1 †, Arianna Lovato1 †, Filippo Boscaini <sup>1</sup> , Elisa Salvetti <sup>1</sup> , Sandra Torriani <sup>1</sup> , Mauro Commisso<sup>1</sup> , Roberta Danzi <sup>2</sup> , Maurizio Ugliano<sup>1</sup> , Annalisa Polverari <sup>1</sup> , Giovanni B. Tornielli <sup>1</sup> \* and Flavia Guzzo<sup>1</sup> \*

*<sup>1</sup> Biotechnology Department, University of Verona, Verona, Italy, <sup>2</sup> Unione Italiana Vini Soc. coop, Verona, Italy*

#### *Edited by:*

*Alma Balestrazzi, University of Pavia, Italy*

#### *Reviewed by:*

*Michaela Griesser, University of Natural Resources and Life Sciences, Austria Claudio Pastenes, Universidad de Chile, Chile*

#### *\*Correspondence:*

*Giovanni B. Tornielli giovannibattista.tornielli@univr.it Flavia Guzzo flavia.guzzo@univr.it*

*† These authors have contributed equally to this work.*

#### *Specialty section:*

*This article was submitted to Crop Science and Horticulture, a section of the journal Frontiers in Plant Science*

> *Received: 06 March 2017 Accepted: 26 May 2017 Published: 21 June 2017*

#### *Citation:*

*Negri S, Lovato A, Boscaini F, Salvetti E, Torriani S, Commisso M, Danzi R, Ugliano M, Polverari A, Tornielli GB and Guzzo F (2017) The Induction of Noble Rot (Botrytis cinerea) Infection during Postharvest Withering Changes the Metabolome of Grapevine Berries (Vitis vinifera L., cv. Garganega). Front. Plant Sci. 8:1002. doi: 10.3389/fpls.2017.01002* The natural or induced development of noble rot caused by the fungus *Botrytis cinerea* during the late stages of grapevine (*Vitis vinifera* L.) berry ripening is used in some traditional viticulture areas to produce high-quality wines such as Sauternes and Tokaji. In this research, we wanted to verify if by changing the environmental conditions during post-harvest withering we could induce the noble rot development on harvested berries in order to positively change the wine produced from withered Garganega berries. Therefore, we exposed the berries to postharvest withering under normal or artificially humid conditions, the latter to induce noble rot. The presence of noble rot symptoms was associated with the development of *B. cinerea* in the berries maintained under humid conditions. The composition of infected and non-infected berries was investigated by untargeted metabolomics using liquid chromatography/mass spectrometry. We also explored the effects of the two withering methods on the abundance of volatile organic compounds in wine by yeast-inoculated micro-fermentation followed by targeted gas chromatography/mass spectrometry. These experiments revealed significant metabolic differences between berries withered under normal and humid conditions, indicating that noble rot affects berry metabolism and composition. As well as well-known botrytization markers, we detected two novel lipids that have not been observed before in berries infected with noble rot. Unraveling the specific metabolic profile of berries infected with noble rot may help to determine the compounds responsible for the organoleptic quality traits of botrytized Garganega wines.

Keywords: postharvest withering, Garganega grapes, noble rot induction, metabolomics, VOCs

### INTRODUCTION

The necrotrophic ascomycete Botrytis cinerea has been described as a 'Jekyll and Hyde' fungus because it causes devastating gray mold disease in grapevine plants but is also responsible for noble rot in ripe and overripe berries, which allows the production of high-quality sweet wines such as Sauternes and Tokaji (Fournier et al., 2013). Gray mold caused by B. cinerea is one of the

**220**

most severe grapevine diseases, reducing both the quality and quantity of berries. The resulting wines are poor because the infected berries have an unfavorable composition and the pathogen also produces toxic compounds that affect yeast and thus inhibit the fermentation process (Bocquet et al., 1995; Hong et al., 2011; Agudelo-Romero et al., 2015). However, the development of B. cinerea as noble rot (botrytization) is a favorable process lasting 10–20 days and is typical of particular wine productions. The berry is transformed by the penetration of fungi through stomata, wounds or microfissures on the fruit surface (the pourri plein stage), the permeabilization of the fruit skin encouraging water loss and sugar concentration, and finally enzymatic maceration (the pourri rôti stage) (Ribéreau-Gayon et al., 2006). At the end of this process, further fungal development is arrested by the high sugar concentration and, if still on the plants, the botrytized berries can be harvested individually. Noble rot confers a berry composition which is distinct from that of berries with gray rot and uninfected berries and is potentially associated with desirable aroma characters of the resulting wine. In addition to directly producing potent odorants such as phenylacetaldehyde, lactones and vanillin (Lopez Pinar et al., 2016), noble rot infection can indeed stimulate production, in the berry, of cysteine and glutathione conjugates which can be transformed by the yeast into the powerful aroma compound 3-mercaptohexanol (Thibon et al., 2009, 2011). The developmental transition between gray rot and noble rot is influenced by environmental conditions and soil characteristics. Moist nights, foggy mornings and dry, sunny days promote the slow infection that results in noble rot, whereas strong rainfall and high humidity facilitate the more aggressive gray mold (Ribéreau-Gayon et al., 1980; Gubler et al., 2013).

Recioto di Soave is an Italian passito wine (i.e., a wine produced from dehydrated grapes) made from the white-skinned berries of the cultivar Garganega. Grape dehydration (known as withering) takes place after harvest in a dedicated room known as the fruttaio. Slow postharvest withering can favor noble rot development, induced by particular environmental conditions and/or artificial B. cinerea inoculation, thus allowing botrytization to be implemented in regions where natural noble rot is uncommon (Lorenzini et al., 2012; Tosi et al., 2013).

The metabolomic and transcriptomic changes that occur in black-skinned grape berries of various cultivars during traditional postharvest withering in fruttaio have recently been described (Zenoni et al., 2016). The behavior of withering berries is strongly cultivar dependent. The berries of slow-withering cultivars are more metabolically active during the process, showing both de novo synthesis of various metabolites (especially stilbenes) and a higher number of differentially expressed genes. Transcriptomics and metabolomics have also been used to investigate the response of white-skinned Sémillon berries during noble rot infections (Blanco-Ulate et al., 2015). The fruits of this cultivar respond to B. cinerea infection by upregulating genes involved in the response to pathogens and stress, fruit ripening, and hormone metabolism, and by accumulating certain secondary metabolites such as phenylpropanoids and terpenes.

Here we used untargeted metabolomics based on liquid chromatography/mass spectrometry (LC-MS) to characterize the metabolites of Garganega berries during postharvest withering in fruttaio, under standard conditions and with artificial humidification used to induce B. cinerea colonization. We also used targeted gas chromatography/mass spectrometry (GC-MS) to characterize the volatile organic compounds (VOCs) in wines produced from berries exposed to the two postharvest withering treatments.

# MATERIALS AND METHODS

#### Withering Methods and Berry Sampling

Approximately 170 kg of Garganega berries was harvested at the commercial ripening stage (soluble solids content = 18.5 ± 0.25◦Brix) in Monteforte d'Alpone (Verona, Italy) at the beginning of October and transported to the Pasqua Vigneti e Cantine winery. The berries were placed in perforated plastic boxes (plateaux, ∼5 kg in each) in a ventilated withering facility under natural conditions (17–20◦C, 78–82% relative humidity) and sampled (T0).

Brix degrees were measured weekly in three randomly selected replicates using a DBR35 digital refractometer (Giorgio Bormac, Carpi, Italy). Three boxes were also weighed weekly using a CH50K50 electronic balance (Kern, Balingen, Germany) in order to determine the weight loss of berry bunches during withering. After 29 days, when the weight loss was ∼30% of the initial weight, grape clusters were sampled (T1) and half of the plateaux were covered with plastic film at 15–17◦C. Water-filled trays were placed inside to increase the relative humidity (88–94%) and encourage B. cinerea development. The remaining plateaux were left under normal withering conditions (15–17◦C, 68–75% relative humidity). The two different environmental conditions were imposed for 32 days before the final samples were taken obtaining T2-n (normal withering, ventilated) and T2-i (induced noble rot) samples. The total duration of dehydration was 61 days. The relative humidity and temperature inside and outside the covering were monitored using Hobo Pro v2 sensors connected to data loggers (Onset Computer Corporation, Bourne, MA, USA).

Three independent pools of 3 kg of berries each were collected for each treatment and used in the following analysis. For T0, T1 and T2-n grape berries were randomly sampled whereas for T2-i berries were visually selected for noble rot symptoms. Each sample was used to determine the average berry weight, the number of B. cinerea colony forming units (CFUs) and for LC-MS analysis. The remaining of T0, T2-n, and T2-i berry samples (about 2 Kg for each biological replicates) were also pressed and the resulting musts were micro-fermented followed by GC-MS analysis of the wine. A simplified experimental workflow is reported in **Figure 1**.

# Enumeration of *B. cinerea* CFUs

We randomly selected 100 g of berries from each biological replicate and crushed them independently. The juices were serially diluted in 25% Ringer's solution (Oxoid, Basingstoke, UK) and 100-µl aliquots were spread on triplicate plates of Botrytis Selective Medium (Edwards and Seddon, 2001). The B. cinerea

CFUs were counted on the plates after 5–7 days of incubation at 20◦C in the dark.

## Extraction of Metabolites and LC-MS Analysis

250 g of frozen berries for each pool was ground in liquid nitrogen and 300 mg of berry powder was extracted in three volumes of cold LC-MS-grade methanol. After mixing, the samples were sonicated for 15 min at 4◦C and centrifuged (16 000 rcf, 10 min, 4◦C). The supernatants were diluted 1:2 in LC-MS-grade water, passed through 0.2µm Minisart RC4 filters (Sartorius-Stedim Biotech, Göttingen, Germany) and analyzed by reversed-phase high-performance liquid chromatography (RP-HPLC) using a Gold 127 HPLC System (Beckman Coulter, Brea, CA, USA) equipped with a C18 guard column (7.5 × 2.1 mm, 5µm particle size) and an Alltech (Nicholasville, KT, USA) RP C18 column (150 × 2.1 mm, 3µm particle size). A gradient between solvent A (0.5% formic acid and 5% acetonitrile in water) and solvent B (100% acetonitrile) was set as follows: 0–10% B in 2 min, 10–20% B in 10 min, 20–25% B in 2 min, 25–70% B in 7 min, isocratic for 5 min, 70–90% B in 1 min, isocratic for 14 min, 90–0% B in 1 min, and 20 min equilibration. For each sample, 20µl was injected at a flow rate of 0.2 ml min−<sup>1</sup> .

The HPLC instrument was coupled on-line to an Esquire 6000 ion trap mass spectrometer equipped with an electrospray ionization (ESI) source (Bruker Daltonik, Bremen, Germany). Mass spectra were recorded in alternating positive and negative ionization mode within the range 50–1,500 m/z with a target mass of 400 m/z. Nitrogen was used as the nebulizing gas (50 psi, 350◦C) and drying gas (10 L min−<sup>1</sup> ) and the vacuum pressure was 1.4 × 10−<sup>8</sup> bar. For fragmentation analysis, MS/MS and MS<sup>3</sup> spectra were recorded in positive and negative ionization modes in the range 50–1,500 m/z. Helium was used for collision induced dissociation (amplitude = 1 V). MS data were collected using Esquire Control v5.2 software and processed using Esquire Data Analysis v3.2 software (both provided by Bruker Daltonik).

Metabolites were identified by comparing retention times, m/z values and fragmentation patterns with those of commercial standards in our in-house library. When no authentic standard compounds were available, identification relied on the fragmentation patterns in online databases such as MassBank (www.massbank.jp) or reported in the literature. Neutral losses of 132, 146, and 162 Da were considered diagnostic of the loss of pentose, deoxyhexose, and hexose sugars, respectively.

## Lipid Extraction and LC-MS Analysis

Botrytis cinerea strain B05.10 (Amselem et al., 2011) was inoculated into flasks containing 125 ml potato dextrose broth (Formedium, Hunstanton, UK) at a concentration of 7 × 10<sup>6</sup> conidia/ml, and incubated for 7 days at 22◦C, shaking at 120 rpm. B. cinerea mycelia were recovered by filtration and ground in liquid nitrogen. We resuspended 300 mg of frozen berry powder or B. cinerea mycelia in 300µl LC-MS-grade water and then mixed the suspension with 3 ml glacial chloroform/methanol (2:1). The samples were vortexed for 30 s, stored on ice for 1 h, sonicated for 15 min and centrifuged (25 min, 4,500 rcf, 4 ◦C). The chloroform phases (∼2 ml) were recovered, placed in 2-ml plastic tubes and centrifuged again (10 min, 16,000 rcf, 4 ◦C). The supernatants were recovered, partially evaporated in a Heto Holton Maxi-Dry Plus Vacuum (Thermo Fisher Scientific, Waltham, MA, USA) and tubes containing the same extract were pooled. Finally, the solvent was completely evaporated, the residue was resuspended in three volumes (w/v) of LC-MS-grade methanol and sonicated for 3 min. One sample, arbitrary selected as Quality Control 1 (QC1), and a methanolic solution including 1µg/µl palmitic acid as QC2 (Sigma-Aldrich, St Louis, MO, USA) were analyzed at the beginning and end of the experiment, respectively. Finally, the solutions were passed through a Minisart RC4 0.2-µm filter and 20µl was injected into the abovementioned LC-MS system. The solvents were 0.5% (v/v) formic acid in LC-MS-grade water (A) and 100% acetonitrile (B). A gradient was established from 50 to 100% B in 10 min, followed by 65 min under isocratic conditions and then from 100 to 50% B in 1 min. The column was finally equilibrated for 15 min. MS analysis was carried out by equipping the mass spectrometer with an atmospheric pressure chemical ionization (APCI) source and using the same parameters described above for medium-polar metabolites.

#### Micro-Vinification

Fermentation trials were carried out using musts from T0, T2-n, and T2-i berry samples. The musts were separated from the pomace and mixed with 0.3 g/l activating agent (Apapiù Mix, Tebaldi, Colognola ai Colli, Italy) and 15 mg/l sodium metabisulfite (Sigma-Aldrich) before transferring 500 ml of each must carefully into sterile bottles. After measuring the Brix degrees, the musts were inoculated with 10<sup>6</sup> CFU/ml of the commercial wine strain Saccharomyces cerevisiae Mycoferm CRU 69, previously activated by following the manufacturer's instructions (Ever, Pramaggiore, Italy). Each fermentation experiment was performed in triplicate at a controlled temperature of 18◦C. We monitored the fermentation kinetics for 14 days by gravimetric analysis to determine the loss of weight due to the production of CO2.

### GC-MS Analysis of Micro-Vinificated Wines

Volatiles were analyzed by gas chromatography–mass spectrometry (GC-MS) after solid–phase extraction (SPE). SPE was performed using ENV<sup>+</sup> cartridge (1 g, 40–140µm; Isolute, IST Ltd., Mid Glamorgan, UK) and an Aspec XL Sample Processor for SPE (Gilson Inc. Middleton, WI, USA). The cartridges were sequentially conditioned with methanol (9.5 ml) and distilled water (19 ml). A total of 38 ml of wine sample diluted 1:2 with distilled water, and 1–heptanol added as internal standard (500µg/l) was loaded onto the cartridge. The residue was washed with 19 ml of distilled water. The free aroma compounds were eluted with 9 ml of dichloromethane. The solution was dried with Na2SO<sup>4</sup> and concentrated to 0.4 ml by nitrogen flow stream. GC–MS analysis was performed with 6980N Network GC System coupled with a 5975 XL EI/CI MSD (Agilent Technologies, Santa Clara, CA, USA), equipped with DB–WAX Bonded PEG fused silica capillary column (60 m × 320µm i.d. × 0.25µm film thickness; Agilent Technologies). Instrumental conditions were: electron impact (EI) mode 70 eV; injector temperature 200◦C; He carrier flow 1.5 ml/min; column temperature 50◦C for 4 min, rising to 240◦C at 4◦C/min, then 20 min at 240◦C; and injection volume 2.0µL in splitless mode. The analyses were performed in SCAN mode. NIST data bank and co-injection of pure reference standards were used to identify the compounds.

# Statistical Analysis of Samples

Statistical significance between samples analyzed for percentage of weight loss, soluble solid content, average berry weight, B. cinerea enumeration and fermentation kinetics was evaluated by t-student. For LC-MS and GC-MS data, raw chromatograms were converted to netCDF files for peak alignment and area extraction using MZmine software (http://mzmine.sourceforge.net/) and multivariate statistical analysis was applied to the resulting dataset using SIMCA v.13.0 (UmetrixAB, Umeå, Sweden). Pareto scaling was applied to all analytical methods. Unsupervised principal component analysis (PCA) was used to identify the major clusters defined by the samples, and two supervised methods, namely partial least squares discriminant analysis (PLS-DA) and orthogonal projections to latent structures discriminant analysis (OPLS-DA/O2PLS-DA), were used to compare classes in order to identify metabolites that characterize different withering stages. PLS-DA models were validated by a permutation test (200 permutations) and the corresponding OPLS-DA/O2PLS-DA models were cross-validated by analysis of variance (ANOVA) with a threshold of p < 0.01.

## RESULTS

# Artificial Humidification of Berries Increases the Prevalence of *B. cinerea* Colonization

The berries maintained under standard conditions showed a linear weight loss during the first 40 days of withering followed by slower weight loss toward the end of the process, whereas the berries maintained under artificially high humidity lost weight more slowly (p < 0.01, **Figure 2A**). The Brix degrees increased during withering due to the loss of water and consequently the concentration of sugars, hence the profile was complementary to the loss of weight, increasing more slowly in the covered berries (**Figure 2B**).

At the end of the withering process, only the covered berries showed the typical symptoms of noble rot, with roughly 70% of the berries visually appearing as chocolate-brown colored, more shriveled and dehydrated compared to control berries (**Figures 2C,D**). These berries were visually selected for further analysis. The comparison between the average berry weight of visually selected T2-n and T2-i berries revealed that T2-i weight was slightly lower than the control berries (p < 0.05, **Figure 2E**). This was also associated with a higher prevalence of B. cinerea colonization as determined by counting the number of CFUs on selective medium, confirming that the conditions used for induction supported B. cinerea growth in its latent form (p < 0.05, **Figure 2F**). No gray mold symptoms were observed among the covered berries.

# Untargeted Metabolomics Reveals Fungal Metabolites, Plant Phytoalexin Accumulation, and Plant Metabolite Depletion

Berries were sampled for metabolomic comparison at the beginning of the experiment (T0), 29 days later just before the two different withering conditions were applied (T1), and at the end of the experiment, separately for the berries exposed to conventional withering (T2-n) and the high humidity conditions (T2-i). Representative LC-MS chromatograms obtained in negative and positive ionization modes are shown in **Figure 3**. The blue zones highlight metabolites that increased in abundance during dehydration but were largely consumed in T2-i berries in comparison with T2-n berries. These metabolites included caffeoyl tartaric acid (caftaric acid), the amino acids leucine/isoleucine, phenylalanine, and tryptophan (together with its caffeic acid derivative), flavan-3-ols and flavonols. The red zones indicate metabolites that became more abundant or appeared de novo in the artificially humidified berries (T2-i) in comparison with T2-n berries. These included uridine 5'-diphospho-N-acetylglucosamine, a resveratrol tetramer and 13-keto-9Z,11E-octadecadienoic acid/13-oxo-9Z,11Eoctadecadienoic acid (13-KODE). The metabolites present in all samples are summarized in **Table 1**.

The chromatograms were used to build two data matrices (negative and positive ionization mode, respectively). The negative ionization data matrix contained 257 m/z features, 74 of

(B) soluble solids (◦Bx) during normal withering (T2-n) and induced noble rot (T2-i). (C) Appearance of sampled berries withered under natural conditions or (D) under higher humidity to induce noble rot. (E) Average berry weight in the T2-n and T2-i samples at the end of the withering process. (F) Enumeration of *Botrytis cinerea* colony forming units (CFUs) in samples T0, T1, T2-n, and T2-i. Vertical bars represented standard deviations (SD) of means (*n* = 3). Asterisks refer to *t*-student *p*-values obtained from T2-n and T2-i comparison (\**p* < 0.05, \*\**p* < 0.01).

which were tentatively identified. They included 52 metabolites plus adducts and fragments (Supplementary File 1, Datasheet 1). The positive ionization data matrix contained 356 m/z features, 33 of which were identified, corresponding to 22 different metabolites (Supplementary File 1, Datasheet 2).

The two data matrices were explored by multivariate analysis (O2PLS-DA). The results obtained for the negative ionization data matrix are shown in Supplementary Figures 1A,B. The O2PLS-DA loading plot was expressed as a pq(corr) value, representing the correlation between the p part of the model (the class of samples) and the q part of the model (the metabolites). The spatial closeness among the metabolites (black triangles) and the samples (blue squares) reflects their relationships, revealing the concentration effect from T0 to T1 and T2, but also specific effects in the botrytized samples that differ substantially from the non-botrytized controls (Supplementary Figure 1B). To overcome concentration effects and focus on concentrationindependent effects, the data matrices were normalized for weight loss, and the signal intensities were expressed relative to the weight at the beginning of the experiment (Supplementary File 1, Datasheets 3, 4). The results following data normalization are shown for the negative data matrix (**Figures 4A,B**). The enrichment of metabolites previously observed in the T2-n samples is now effectively shared among the T0, T1, and T2 n samples (light blue circle) confirming these metabolites are characteristic of the entire traditional withering process. Despite this normalization effect, a few metabolites were typical of the T0 and T2-n control samples (green and orange circles, respectively). Interestingly, the botrytized berries (brown circles) were strongly characterized by a group of specific metabolites which correlate negatively with the traditional withering process. The metabolites are listed in Supplementary File 2, Datasheet 1.

OPLS-DA analysis was also applied to both the negative and positive ionization mode data matrices for the normal withering (T2-n) and botrytized (T2-i) berry samples (**Figures 4C,D**) better highlighting differences between T2-n and T2-i. The set of strongly characteristic metabolites identified by this analysis (Supplementary File 2, Datasheet 2) reflects the depletion and de novo production of metabolites in T2-i as already highlighted in **Figure 3** and reported in **Table 1** (e.g., resveratrol derivatives, 13-KODE and uridine 5′ -diphospho-N-acetylglucosamine).

The depleted compounds represent diverse metabolite classes (e.g., sugars, amino acids, flavonoids and some stilbenes). Other metabolites accumulate rapidly in the botrytized berries, including pantothenic acid, some stilbenes (dimers, trimers and tetramers but not the monomers), glucose-6-phosphate, uridine 5 ′ -diphospho-N-acetylglucosamine, a lipid putatively annotated as 13-KODE, and other unidentified metabolites. The presence of the N-acetylglucosamine donor uridine 5′ -diphospho-Nacetylglucosamine indicates an active fungal metabolism because this sugar is a precursor of the chitin found in the fungal cell wall. Some of the molecules accumulating in botrytized berries were also clearly detectable in the chromatograms as major signals (**Figure 3**), including one resveratrol tetramer (**Figure 3**) and two metabolites with retention times of 30 and 32 min respectively (not shown). These two metabolites showed similar behavior, suggesting similar chemical properties. Both showed the chloride and formic adducts as main signals in negative ionization mode,

and the molecular ions in positive ionization mode, and both generated fragments at m/z 355, 337, and 206. Although we were unable to identify these molecules, the higher retention time suggested they are lipids. We therefore re-extracted the methanol extracts with chloroform, and the resulting lipid fractions were analyzed by LC-APCI-MS using a method optimized for lipid analysis. The lipid profiles of T2-n and botrytized (T2-i) samples are shown in Supplementary Figure 2. This analysis confirmed the lipid-like nature of the two unidentified T2-i metabolites (highlighted in the figure) and showed that the general lipid profile is otherwise similar between the T2-n and T2-i samples. The same peaks could not be detected using the same LC-APCI-MS approach following the extraction of lipids from B. cinerea strain B05.10 grown in vitro, indicating that the two unidentified lipids might not be typical constituents of the fungus (data not shown). On the other side, we cannot exclude that the wild type B. cinerea strains developed in this experiments have different composition than the used reference B05.10 strain (Amselem et al., 2011).

## The Wines Produced by Botrytized and Conventionally Withered Berries Show Different Aromatic Profiles

Micro-scale vinification was performed on fresh berries (T0), withered botrytized berries (T2-i) and conventionally withered berries (T2-n). The ◦Brix values of the musts from the T0, T2-i and T2-n berries were 18.67 ± 0.25, 31.13 ± 0.11, and 32.87 ± 0.06, respectively. The must fermentation rate (calculated as grams of CO2/100 ml of must) was generally higher in the musts from withered grapes (**Figure 5A**). Moreover, musts from the conventionally withered grapes showed a more vigorous fermentation compared to the musts from botrytized berries (p < 0.05).

The aromatic profile of the three wines was analyzed by GC-MS 14 days after the beginning of the vinification. The O2PLS-DA model of the entire GC-MS data matrix revealed that the three wines showed distinct aromatic compositions (**Figures 5B,C**). The compounds characterizing the three wines are listed in **Table 2**, and their pq(corr) values are shown in Supplementary File 3. Wines from T2-n grapes were mainly characterized by fruity aromas, whereas botrytized wines were characterized by spicy aromas (**Figure 5C**).

When the two withering processes were compared, wine from the naturally withered berries was strongly characterized by the presence of ethyl-4-hydroxy butanoate, as well as benzyl alcohol, eugenol, guaiacol, homovanillic alcohol, homovanillic acid, trans-3-hexenol, β-damascenone, and methyl vanillate. In contrast, the botrytized wine was strongly characterized by the presence of N-(3-methylbutyl)acetamide, as well as sherry lactone 1, benzaldehyde, 1-octen-3-ol, trans-8-dihydroxylinalool,


*Peak numbers refer to the chromatogram profiles in Figure 3. Rt, retention time (min); Nf, not fragmented. Asterisks refer to those metabolites which average peak areas was significantly different between T2-n and T2-i as assessed by t-student test.*

\**p* < *0.05,* \*\**p* < *0.01.*

ethyl vanillate, ethyl isoamyl succinate, diethyl succinate, p-cresol, ho-diendiol, 4-terpineol, γ-nonalactone, and ethyl phenylacetate (**Figure 5D** and Supplementary File 3). When considering only the more abundant VOCs (more than 500 ppb) showing at least a two-fold difference in abundance between the two samples, the wine produced from naturally-withered berries was characterized by isovalerianic acid, isoamylacetate, decanoic acid, and homovanillic acid, whereas the botrytized wine was characterized by N-(3-methylbutyl)-acetamide, sherry lactones 1 and 2, benzaldehyde and 4-terpineol.

#### DISCUSSION

#### The Natural Development of Noble Rot Can Be Strongly Induced in Garganega Berries Undergoing Postharvest Dehydration

The postharvest induction of noble rot could be used for the production of botrytized wines in regions with climates unsuitable for natural botrytization and also in those with suitable climates, to overcome the unpredictability of natural botrytization. However, a controlled widespread noble rot development on dehydrating grapes is not easy to achieve because the natural or forced ventilation of cases to accelerate dehydration makes the berries less susceptible to infection with B. cinerea (Barbanti et al., 2008). Fedrizzi et al. (2011) investigated natural botrytization in Corvina berries during withering, but in this case it was necessary to discard rotten berries developing gray mold and to manually separate the botrytized and non-botrytized fruit. Lorenzini et al. (2012) showed that noble rot can be induced under postharvest laboratory conditions by inoculating Garganega and Corvina berries with the fungus. The ability to achieve widespread noble rot development during natural withering has been reported anecdotally (Ferrarini et al., 2009; Vannini and Chilosi, 2013).

Here we demonstrated the ability to induce noble rot development in Garganega berries without the concomitant development of gray mold by implementing a special management strategy during postharvest withering, comprising an initial period of normal withering to allow partial berry dehydration (which prevents the development of gray mold by ensuring the adequate concentration of sugars) followed by a

The metabolites which strongly characterize each sample are highlighted with colored circles and are listed in Supplementary File 2 (Datasheet 1). The light blue circle comprises all metabolites that characterize the natural withering process and negatively correlate with berries infected with noble rot (T2-i). Correlation loading plots for the OPLS-DA models of negative (C) and positive (D) data matrices show the distribution of metabolites between T2-n and T2-i berries. All metabolites with pq(corr) values > 0.7 or < −0.7 are considered highly characteristic of T2-n (highlighted in yellow) or T2-i (highlighted in brown) berries and are listed in the Supplementary File 2 (Datasheet 2).

period of increased humidity achieved by covering the berries in the presence of water-filled trays. This simple procedure increased the humidity without affecting the temperature, slightly reduced the rate of berry dehydration, and encouraged the development of B. cinerea infection without the need for artificial inoculation because the fungus is commonly present in the vineyard and in cellars as an environmental contaminant. Noble rot induction was confirmed by berry characteristics and the enumeration of B. cinerea CFUs in selective medium. We characterized the changes in the metabolite profile of grapes and wines attributable to the proliferation of the fungus. However, the possibility that the modified air humidity could be the cause of part of the differences between T2-n and T2-i cannot be completely ruled out.

# *B. cinerea* Growth and Plant Defence Can Be Monitored by Untargeted Metabolomics

Untargeted metabolomics based on LC-MS revealed metabolites associated with the infection of berries by B. cinerea. Some of these metabolites were biochemical markers of the fungus, including structural components and products of fungal metabolism, while others were derived from the berries and represent the onset of plant defense mechanisms during withering.

The LC-MS data matrix revealed many imprints of fungal metabolism, including the presence of the N-acetylglucosamine donor uridine 5′ -diphospho-N-acetylglucosamine, which is utilized by fungi including B. cinerea as a substrate for the enzyme chitin synthase (Causier et al., 1994). The declining levels of many grape metabolites in botrytized fruits suggests they were degraded by fungal metabolism, including sucrose, hydroxycinnamic acids (coutaric, caftaric, and fertaric acids), amino acids, lignans, and many flavonoids (including flavan-3 ols and flavonols). The loss of polyphenols has been reported in other white-berry cultivars infected with noble rot, including Chenin Blanc (Carbajal-Ida et al., 2016) and Chardonnay (Hong et al., 2012) although there was a specific increase in the abundance of flavan-3-ols in Chenin Blanc, in contrast to other polyphenols (Carbajal-Ida et al., 2016). However, cultivars such as Sémillon accumulated high levels of phenylpropanoids following the onset of noble rot (Blanco-Ulate et al., 2015). Therefore, the impact of B. cinerea on phenylpropanoid metabolism appears to be cultivar dependent.

The stilbenes are phytoalexins that are known to accumulate during botrytization (Landrault et al., 2002; Blanco-Ulate et al., 2015). The observed decline in the abundance of stilbene monomers (resveratrol and resveratrol glucoside) could reflect the consumption of these metabolites by the fungus, but the concomitant increase in the levels of stilbene dimers, trimers and tetramers suggests that botrytization causes the aggregation of stilbene monomers into oligomers. The accumulation of the oxylipin 13-KODE could also represent a plant defense response because this metabolite is induced as a defense molecule in soybean (Glycine max) in response to fungi such as Aspergillus niger, A. oryzae, Rhizopus oligosporus, and A. niger wry (Feng et al., 2007). To the best of our knowledge, this is the first

report describing the induction of 13-KODE in grapevine berries in response to noble rot. However, octadecadienoic acids, the precursors of KODE oxylipins, have been proposed as potential positive metabolic markers of gray mold (Agudelo-Romero et al., 2015). The botrytized fruits also accumulated large amounts of pantothenic acid, D-glucose-6-phosphate and two unannotated lipids.

### The Postharvest Induction of Noble Rot Influenced the Accumulation of Wine Aroma Compounds

The proliferation of B. cinerea induced remarkable changes in the accumulation of VOCs, affecting several aroma compounds that may contribute to the sensory characters of white wines. From a quantitative perspective, N-(3-methylbutyl)acetamide was the strongest marker of botrytized wine in agreement with previous studies of botrytization in Recioto di Soave (Azzolini et al., 2013; Tosi et al., 2013), Amarone (Fedrizzi et al., 2011), and Fiano (Genovese et al., 2007) wines. From a qualitative perspective, several VOCs detected at lower concentrations but with a potentially higher impact on aroma (Francis and Newton, 2005) were also influenced by noble rot. The sherry lactone isomers and γ-nonalactone were detected at higher concentrations in botrytized wines, in agreement with previous reports (Genovese et al., 2007; Sarrazin et al., 2007; Azzolini et al., 2013; Tosi et al., 2013). Although lactones do not contribute directly to the aroma of botrytized wines, they are involved in perceptive interaction phenomena resulting in an enhanced sensory contribution, e.g., synergy between γ-nonalactone and eugenol can enhance the overripe orange aroma notes typical of noble rot wines (Stamatopoulos et al., 2014). Eugenol is mostly derived from contact between the wine and oak wood, so the storage in oak barrels of Garganega wines from berries infected with noble rot could enhance these overripe orange aromas. Terpenes such as citronellol, ho-diendiol, hydroxylated linalool derivatives and 4-terpineol became more abundant during dehydration in the berries with noble rot, especially in the case of 4-terpineol. Likewise, the norisoprenoid 3-oxo-α-ionol (which gives rise to the tobacco aroma compound megastigmatrienone) accumulated to higher levels during withering. Terpenes and norisoprenoids are two important groups of aroma compounds that contribute the floral, fruity and tobacco-like attributes of wines. They accumulate in the berries as free molecules and as glycosylated precursors, which can be revealed by the action of yeast during fermentation or by acid hydrolysis during wine aging (Ugliano

#### TABLE 2 | Aroma compounds highlighted in Figure 5C characterizing T0, T2-n, and T2-i musts.


*The corresponding pq(corr) values are reported in Supplementary File 3. Items marked with an asterisk have to be multiplied by 10<sup>3</sup> .*

et al., 2006). Dehydration can favor their accumulation, and the presence of B. cinerea can facilitate their release from precursors by means of the pool of enzymes released into the must (Donèche, 1993).

Several volatile benzenoids, generally characterized by sweet/spicy aroma notes, were shown to increase in response to noble rot and/or simple dehydration, including benzaldehyde, vanillin, cresols, guaiacols and eugenol. Although benzaldehyde is often associated with the development of B. cinerea (Genovese et al., 2007; Fedrizzi et al., 2011), the behavior of volatile benzenoids has not been investigated in detail. However, Genovese et al. (2007) also observed the accumulation of eugenol and vinyl guaiacol in white wines prepared from berries infected with noble rot. The mushroom-like aroma compound 1-octen-3-ol is found at significantly higher concentrations in wines produced from botrytized berries. B. cinerea and other pathogens such as Uncinula necator (powdery mildew) produce 1-octen-3-ol, which at high concentrations can reduce the quality of berries and introduce mushroom off-odors in the finished wine (Darriet et al., 2002).

Interestingly, several aroma compounds arising from yeast metabolism, in particular the powerful fruit-smelling esters isoamyl acetate, ethyl butanoate, ethyl hexanoate and ethyl octanoate, were present at lower concentrations in the botrytized wines compared to wines from either the fresh or dehydrated berries. When comparing fresh and dehydrated berries, it was clear that dehydration without noble rot infection favored the accumulation of these metabolites during fermentation, probably due to the higher concentrations of nitrogen available to the yeast (Ugliano et al., 2006; Vilanova et al., 2007). The lower concentrations of esters in the botrytized wines could therefore reflect the depletion of nitrogen or the release of esterases by B. cinerea.

In conclusion, the analysis of the volatile fraction of wines and evaluation of the potential odor contribution of different volatiles indicated that wines from dehydrated berries were generally characterized by higher content of fresh fruit-smelling compounds (esters), whereas noble rot induced the accumulation of several spicy aroma compounds such as lactones, combined with compounds with floral attributes such as 4-terpineol and the mushroom smelling compound 1-octen-3-ol.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

GT: designed the experiments; SN and FB: performed the LC-MS-based untargeted metabolomics; SN and RD: performed the GC-MS based metabolomics; MC: performed the lipidomic experiment; ST and ES: did the bacterial counts and analyses; AP and AL: projected and executed the sampling and the agronomical analyses; SN and FG: analyzed the metabolomics data; FG: wrote the manuscript; SN, AL, MU, and GT: contributed to the draft writing; MU, AP, and ST: critically revised the manuscript; all authors read and approved the final manuscript.

#### FUNDING

This work was supported by Regione Veneto - POR - Fondo Sociale Europeo 2007–2013 – Ob. Competitività Regionale e Occupazione - Reg. 1081/2006. Asse IV "Capitale Umano." This work benefited from the networking activities coordinated within the EU-funded COST ACTION FA1106 "An integrated systems approach to determine the developmental mechanisms controlling fleshy fruit quality in tomato and grapevine."

#### ACKNOWLEDGMENTS

We are grateful to the winery "Pasqua Vigneti e Cantine SpA" for providing Garganega samples and for allowing us to set up the experiment in their withering facility.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2017. 01002/full#supplementary-material


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Negri, Lovato, Boscaini, Salvetti, Torriani, Commisso, Danzi, Ugliano, Polverari, Tornielli and Guzzo. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Metabolite Profiling Reveals Developmental Inequalities in Pinot Noir Berry Tissues Late in Ripening

Amanda M. Vondras<sup>1</sup> , Mauro Commisso<sup>2</sup> , Flavia Guzzo<sup>2</sup> and Laurent G. Deluc<sup>1</sup> \*

<sup>1</sup> Deluc Laboratory, Department of Horticulture, Oregon State University, Corvallis, OR, United States, <sup>2</sup> Guzzo Laboratory, Department of Biotechnology, University of Verona, Verona, Italy

Uneven ripening in Vitis vinifera is increasingly recognized as a phenomenon of interest, with substantial implications for fruit and wine composition and quality. This study sought to determine whether variation late in ripening (∼Modified Eichhorn-Lorenz stage 39) was associated with developmental differences that were observable as fruits within a cluster initiated ripening (véraison). Four developmentally distinct ripening classes of berries were tagged at cluster véraison, sampled at three times late in ripening, and subjected to untargeted HPLC-MS to measure variation in amino acids, sugars, organic acids, and phenolic metabolites in skin, pulp, and seed tissues separately. Variability was described using predominantly two strategies. In the first, multivariate analysis (Orthogonal Projections to Latent Structures-Discriminant Analysis, OPLS-DA) was used to determine whether fruits were still distinguishable per their developmental position at véraison and to identify which metabolites accounted for these distinctions. The same technique was used to assess changes in each tissue over time. In a second strategy and for each annotated metabolite, the variance across the ripening classes at each time point was measured to show whether intra-cluster variance (ICV) was growing, shrinking, or constant over the period observed. Indeed, berries could be segregated by OPLS-DA late in ripening based on their developmental position at véraison, though the four ripening classes were aggregated into two larger ripening groups. Further, not all tissues were dynamic over the period examined. Although pulp tissues could be segregated by time sampled, this was not true for seed and only moderately so for skin. Ripening group differences in seed and skin, rather than the time fruit was sampled, were better able to define berries. Metabolites also experienced significant reductions in ICV between single pairs of time points, but never across the entire experiment. Metabolites often exhibited a combination of ICV expansion, contraction and persistence. Finally, we observed significant differences in the abundance of some metabolites between ripening classes that suggest the berries that initiated ripening first remained developmentally ahead of the lagging fruit even late in the ripening phase. This presents a challenge to producers who would seek to harvest at uniformity or at a predefined level of variation.

#### Edited by:

Simone Diego Castellarin, University of British Columbia, Canada

#### Reviewed by:

Amnon Lichter, The Volcani Center, Israel Dario Cantu, University of California, Davis, United States

\*Correspondence:

Laurent G. Deluc laurent.deluc@oregonstate.edu

#### Specialty section:

This article was submitted to Crop Science and Horticulture, a section of the journal Frontiers in Plant Science

> Received: 03 April 2017 Accepted: 08 June 2017 Published: 30 June 2017

#### Citation:

Vondras AM, Commisso M, Guzzo F and Deluc LG (2017) Metabolite Profiling Reveals Developmental Inequalities in Pinot Noir Berry Tissues Late in Ripening. Front. Plant Sci. 8:1108. doi: 10.3389/fpls.2017.01108

Keywords: uneven ripening, crop heterogeneity, metabolomics, HPLC-MS, fruit composition, Vitis vinifera

# INTRODUCTION

fpls-08-01108 June 28, 2017 Time: 18:39 # 2

That producers seek to define and pursue optimal levels of enologically important metabolites in grapes is understood. However, intra-cluster variation is an important consideration as well, given the link between fruit uniformity and crop quality (Carroll et al., 1978; Selvaraj et al., 1995; Barbagallo et al., 2011; Kontoudakis et al., 2011; Liu et al., 2016). This contrasts the prevalent paradigm wherein the mean amount of a metabolite for a population of berries, rather than the variability inherent to that population, influences harvesting decisions. Whether optimal levels of traditional markers that influence harvest decisions (sugars, pigments, tannins, and organic acids) coincide with desirable levels of heterogeneity is largely unexplored.

Several studies have examined the ways in which fruits within a cluster vary, why fruits may initiate ripening unevenly, and means of managing heterogeneity (Cawthon and Morris, 1982a,b; Coombe, 1992; Fernandez et al., 2006; Friend et al., 2009; Gray and Coombe, 2009; Pagay and Cheng, 2010; Calderon-Orellana et al., 2014b; Gouthu and Deluc, 2015). The uneven onset of ripening in a cluster (véraison) has been attributed to fruits' seed content, weakly to flowering time, and the interplay of hormones (Böttcher et al., 2010; Gouthu and Deluc, 2015; Vondras et al., 2016). Then, between véraison and harvest, intracluster variance (ICV) is reduced in terms of gene expression, ◦Brix, color index, and size (Gray and Coombe, 2009; Pagay and Cheng, 2010; Gouthu et al., 2014). However, differences at harvest are still observed and not without consequences. Although one study found no significant relationship between crop price and crop heterogeneity (Calderon-Orellana et al., 2014a), Carroll et al. (1978) showed that wines from fruits belonging to the least and most advanced berries had the lowest sensory scores. They observed differences in sugar, pH, titratable acidity, wine tannins and color between different classes of berries. In Syrah, larger berries at commercial harvest had lower quality characteristics and a yellow–green color indicative of incomplete maturity and possibly higher seed catechin extractability (Barbagallo et al., 2011). In recognizing that substantial variation at harvest limits accurate determination of phenolic maturity, Kontoudakis et al. (2011) also showed that wines from higher density (high sugar) berries were associated with higher ethanol content, pH, color intensity, total phenolic indexes, anthocyanins, and polymerization of proanthocyanidins and lower titratable acidity and bitterness; the resulting wines were higher quality and better balanced. In another study, less-dense grapes contributed fewer anthocyanins and more seed tannins than skin tannins, detrimentally affecting wine composition, while denser berries had the highest total phenolic content (Liu et al., 2016). Whether or not this variability is predominantly due to developmental differences is unexplored, though previous reports have demonstrated variation associated with other factors, like fruit position within clusters (Kasimatis et al., 1975; Tarter and Keuter, 2005; Pagay and Cheng, 2010; Pisciotta et al., 2013).

If fruits are developmentally equals, then dynamic tissues should undergo key developmental transitions, like véraison and dehydration, uniformly. Therefore, perhaps the most appropriate time to make such assessments is as those transitions occur. Late ripening, which we define here as the period of extended ripening immediately following ripeness (Coombe, 1995), is a period during which fruits dehydrate. Distinctive wines are produced using both on- (Rolle et al., 2009; Bowen and Reynolds, 2015; Khairallah et al., 2016; Lukic et al., 2016 ´ ) and off-vine (Bellincontro et al., 2004; Costantini et al., 2006; Moreno et al., 2008; Toffali et al., 2011; Zenoni et al., 2016) dehydration strategies (Figueiredo-González et al., 2013). Both practices have similar effects on sugars, secondary metabolism, and cell integrity (Zamboni et al., 2008) and desiccation can produce responses analogous to those of water stress (Bellincontro et al., 2009). This developmental window is not only important for winemakers because of the dramatic metabolic changes that occur, but also because it might be used to better appreciate developmental inequality within a cluster.

The purpose of this study was to determine whether intracluster variation late in ripening was linked to differences in developmental progress that are observable as fruits unevenly being ripening at véraison. Toward this objective, fruits were tagged as members of qualitative developmental categories or "ripening classes" based on their color at véraison and collected as fruits passed what would be considered commercial harvest into a stage that could be described as on-the-vine withering. If developmental differences persisted between fruits in a cluster, they might be best captured (1) as fruits transition into this stage and (2) in "dynamic" tissues (tissues that demonstrate they are changing within the window observed). A multivariate technique called Orthogonal Projections to Latent Structures-Discriminant Analysis (OPLS-DA) was used to clarify the extent to which different berry tissues remained dynamic in the late ripening period and to determine if and due to which metabolites berries late in ripening could be segregated based on their developmental category at véraison. It was concluded that (1) overall, fruits that were developmentally distinct at véraison remain distinguishable late in ripening, (2) skin, pulp, and seeds were not equally dynamic in the late ripening period, and (3) this period was marked by both reductions and expansions in variation for many metabolites, though most annotated metabolites showed no significant changes in ICV over the period.

## MATERIALS AND METHODS

#### Experimental Design

This study was conducted in 2011 at the Oregon State University Woodhall experimental vineyard in Alpine, Oregon. Pommard grapevines, clones of Vitis vinifera L. cv. Pinot noir, grown on 101-14 rootstock, and trained in a double Guyot system with vertical shoot positioning were used. The five vines used for this study were managed using standard viticultural techniques. On each plant, six primary clusters were chosen on both the east (three clusters) and west (three clusters) side of plants.

A non-invasive tagging technique was used to label four qualitatively distinct ripening classes of fruits at véraison (Lund et al., 2008; Gouthu et al., 2014) on September 10th. Here, véraison is defined as when ∼50% of the cluster remains green,

while ∼50% has visibly initiated ripening. Among these fruits, Green Hard (GH), Green Soft (GS), Pink (PS), and Red (RS) fruits were randomly selected and tagged throughout each of the selected clusters. GH and GS were completely green with no evidence of color-change. GH and GS were distinguished by touch, with GH having no perceptible deformation. PS often exhibited green and pink marbling or were light pink in color. RS berries were dark pink or red. Within each cluster, representatives of each ripening class were tagged at véraison using different colored strings. Then, six berries from each ripening class were sampled from each plant 34, 41, and 48 days after véraison: October 14th (t1), 21st (t2), and 28th (t3). In this study, one biological replicate is equal to six berries of a particular ripening class and from one of the plants used. Sampled berries were immediately frozen on dry ice and then stored at −80◦C. These sampling dates corresponded approximately to stage 39 (overripe) in the modified Eichhorn-Lorenz system for classifying grapevine growth stages (Coombe, 1995).

#### Berry Measurements

Total soluble solids and color were measured per berry (n = 5). A SPER Scientific digital refractometer (Scottsdale, AZ, United States) was used to measure total soluble solids in units of degrees Brix (◦Bx) and a Konica Minolta CR-300 chroma meter (Minolta Corp, Osaka, Japan) was used to quantitatively measure color [lightness (L), hue angle (h), and chroma (C)]. The color index of each berry was calculated as previously described (Carreño et al., 1995) and computed as (180 − h)/(L + C).

#### Metabolite Extraction

Approximately 40 mg of lyophilized material were weighed and extracted with 20 and 40 volumes (w/v) of cold, 90% methanol for pulp and seed, respectively. Skin tissues were subjected to 40 volumes (w/v) of cold 89.9% methanol acidified with 0.1% (v/v) of formic acid (Toffali et al., 2011). The extracts were vortexed, sonicated in an ice-filled ultrasonic bath (Falc Instruments, Bergamo, Italy) for 20 min at 40 kHz, kept in darkness for 2h at 4◦C and finally centrifuged at 13000 rpm for 10 min at 4◦C. Supernatants were collected and stored at −20◦C. The 200 µL of each extract were diluted 1:2 with LC-MS-grade water and filtered with Minisart RC 4 membrane filters (0.2 µm diameter pores, Sartorius) prior to injection into the HPLC-MS system.

#### Metabolite Separation, Detection, and Annotation

Twenty microliters of each diluted sample were drawn through a 508 Autosampler (Beckman Coulter, Fullerton, CA, United States) system and injected to a Beckman Coulter Gold 127 HPLC system (Beckman Coulter, Fullerton, CA, United States) equipped with a C18 guard column (7.5 mm × 2.1 mm) in front of an Alltima HP C18 column (150 mm× 2.1 mm, particle size 3 µm; Alltech Associates Inc, Derfield, IL, United States). Samples were analyzed randomly and in technical duplicate. The chromatographic solvents, conditions, and gradient are described in Anesi et al. (2015).

Metabolite detection was carried out with a Bruker ion trap Esquire 6000 (Bruker Daltonics GmbH, Bremen, Germany) equipped with an ESI ion source with the following specifications: 10 L/min for N<sup>2</sup> drying gas and 50 psi for the N<sup>2</sup> nebulizing gas heated at 350◦C. The analyses were performed in negative and positive alternate modality, setting a target mass of 400 m/z and a scan range of 50–3000 m/z. Metabolite fragmentation was performed up to MS<sup>3</sup> by using Helium gas and setting the fragmentation amplitude at 1 V. Chromatographic data were recorded up to 55 min with Esquire Control v5.2 software and the.d generated files were processed with the proprietary Data Analysis v3.2 (Bruker Daltonics), converted in net.cdf files and analyzed with open-source MZmine 2.10<sup>1</sup> software to create a data matrix reporting feature peak areas. After peak deconvolution, alignment and gap filling performed by MZmine can result in few missing values in the data matrix; such values were considered "missing" and not as zero (Commisso et al., 2017). The subsequent multivariate statistical analyses were carried out with SIMCA v13.0 (Umetrix AB, Umea, Sweden).

Metabolite annotation were made by comparing the m/z, retention time and fragmentation pattern (MS/MS and MS<sup>3</sup> ) of the detected signals with an in-house library of authentic commercial standards or, in their absence, with data reported in literature or online databases<sup>2</sup>,<sup>3</sup> . The confidence of each metabolite annotation was classified as prescribed by Sumner et al. (2007) and is defined in **Data Sheet 1**.

#### Statistical Analyses

Several statistical methods were used to explore intra-cluster differences (differences among the ripening classes) and changes within the cluster over time.

Pareto scaling was applied to all analytical methods (van den Berg et al., 2006; Toffali et al., 2011). PCA was used to identify and remove 20 probable outliers from the 180 samples. The data from the remaining 160 samples (n = 3–5) were analyzed by PCA and OPLS-DA using SIMCA 13.0 (Umetrix AB, Umea, Sweden) to identify metabolites or biomarkers that accounted for differences between ripening groups and distinguished time points. OPLS-DA models were cross validated by ANOVA (p-value < 0.05) and equivalent PLS-DA models were fit and tested by permutation (200 permutations) to avoid overfitting (Triba et al., 2014). In addition to meeting these criteria, only models with high, crossvalidated predictability (Q2 > 0.50) were considered as high confidence (Triba et al., 2014; Anesi et al., 2015). All features were used in SIMCA 13 analyses; only annotated metabolites are shown in figures and discussed.

ICV was estimated by averaging the five biological replicates within a ripening class and calculating the variance across the four ripening classes at a single time point for each annotated metabolite. Unidentified features/metabolites were not included in this analysis. An F-test for variance was used to test for significant changes in ICV between t1, t2, and t3 for each metabolite. In addition, a Tukey-test was used to test for

<sup>1</sup>http://mzmine.github.io/

<sup>2</sup>http://www.massbank.jp

<sup>3</sup>http://www.hmdb.ca

significant differences in relative metabolite amount between ripening classes.

#### Interactive Figures

Figures were constructed using Prism (GraphPad Software Inc., San Diego, CA, United States), SIMCA 13.0, and plot.ly, an online graphing resource. Though still images are presented in this manuscript, the interactive equivalents of plot.ly-generated figures are provided as supplemental HTML files for an enriched exploration of the results (**Presentation 1**). Given the interactive files, readers can include or exclude groups of metabolites by clicking on them in the figure legends, zoom in and out of specified regions of the plots, rotate three-dimensional figures, and identify individual data points which have not been labeled in the still images provided herein. These figures are helpful for visualizing the described trends in the data.

#### RESULTS

#### Developmental and Metabolic Inequality

We sought to determine whether intra-cluster variation late in ripening was associated with developmental inequalities that were apparent at véraison. This was enabled by tagging developmentally distinct "ripening classes" of fruits at véraison (GH, GS, PS, RS) and then sampling them late in ripening.

In terms of total soluble solids (**Figure 1A**), the four ripening classes trended toward uniformity at t2, but this uniformity was short-lived. Green fruits (GH and GS) had significantly lower ◦Brix than more advanced fruits (PS and RS) at t3. No significant differences were observed among the ripening classes at t1 or t2 (**Figure 1A**). With respect to berry color, GH showed significantly lower color index than PS and RS at every collection time and lower color index than GS at t2 and t3 (**Figure 1B**). The color index of GS was similar to PS at t1 and indistinguishable from PS and RS at t2 and t3. Together, **Figures 1A,B** examined separately present seemingly distinct narratives concerning fruit development and changes in ICV. ICV as defined by total soluble solids (◦Brix) describes clusters trending toward uniformity from t1 to t2, with inequalities reappearing after t2. Alternatively, ICV as defined by color index suggests that at and after harvest, differences between the berries that were the most and least advanced at véraison persist without an obvious point of relatively high uniformity. These initial observations indicate that (1) perception of cluster uniformity depends on the metabolites being measured, (2) developmental inequalities persist, though they may be temporarily masked, and (3) metabolite uniformity does not necessarily suggest developmental uniformity. This also suggests limits to how much developmental inequality observed at véraison is actually mitigated by harvest. Importantly, **Figure 1A** shows that fruits that were relatively advanced at véraison also initiated the dehydration stage first, as by this time increases in sugars are linked to dehydration rather than import.

Untargeted HPLC-MS was used to further assess ICV of metabolites separately in berry seeds, skin, and pulp. Following data acquisition, 139 metabolites were annotated using an inhouse library (**Data Sheet 1**). Including all features, annotated and unknown, Orthogonal Projections to Latent Structures – Discriminant Analysis (OPLS-DA) was used to determine whether the ripening classes were distinguishable late in ripening, whether fruits from t1, t2, and t3 were distinguishable overall and if so, which metabolites account for the segregation of different groups.

Prior to this, however, Principal Component Analysis (PCA) of all samples (including all annotated metabolites and unidentified features) revealed that the three berry tissues were remarkably distinct in their metabolite profiles (**Supplementary Figure S1**), with anthocyanins and stilbenes highest in and positively correlated with skin and with proanthocyanidins and flavanols in seed. Most flavonols and other flavonoids were associated with skin tissues. The distinct metabolic profiles of each tissue warranted analyzing each tissue separately to resolve any differences among the groups of interest. Interestingly, a high degree of similarity among the GH and GS berries and among the PS and RS berries was observed such that we were unable to model their differences with high predictability (Q2). This might indicate that the ripening classes are more similar to one another late in ripening than they were at véraison, but this is impossible to say conclusively without equivalent measures at véraison. The four ripening classes were aggregated into two

(B–D) are provided as in Presentation 1.

groups which could be reliably well-modeled by OPLS-DA in each tissue (**Figure 2**)—Lagging (GH + GS) and Advanced (PS + RS). Still, the original ripening classes are colored in **Figure 2**.

For each tissue, OPLS-DA was used to identify metabolites that define the ripening groups (**Figure 2** and **Supplementary Figure S2**) and define the intra-cluster metabolic changes during some of the latest stages of ripening (**Figure 3**). Model parameters are summarized in **Data Sheet 2**. Score plots (**Figure 2A** and **Supplementary Figure S2**) were used to visualize segregation among the samples, with the predictive component describing between-group differences and the orthogonal component describing within-group differences. For each tissue, the metabolic profiles of Lagging and Advanced berries were distinct, with no clear trends in within-group variance that could clearly be attributed to the original ripening classes (GH vs. GS and RS vs. PS). Next, S-plots with VIP integration were used to identify metabolites that best explain the segregation of ripening groups in each tissue (**Figures 2B–D**). S-plots show the covariance and correlation structure between the metabolites and predictive score. In other words, they show the reliability and influence of the metabolites on group segregation. The VIP score, also considered, is an additional metric that describes the extent to which any metabolite drives group distinctions. Metabolites with high VIP (>2) and relatively high |p (corr)| and |p| are putative biomarkers that define berries

late in ripening that were Lagging or Advanced at véraison (**Figure 2**).

The seeds of Lagging and Advanced berries were defined by their high levels of proanthocyanidins and sugars versus leucine/isoleucine, respectively (**Figure 2B**). The skins of Lagging and Advanced berries were distinguished by high anthocyanins and sugars, respectively (**Figure 2C**), and the pulps of Lagging berries were high in and defined by leucine/isoleucine (**Figure 2D**), in contrast to the high levels of leucine/isoleucine found in Advanced berries' seeds. Unsurprisingly, Advanced berry pulp was distinguishable by high levels of sugars (**Figure 2D**).

The score plots in **Figures 3A,C** also visualize segregation among the samples, but per their collection date and irrespective of their ripening class or group. No model that passed all acceptable thresholds upon cross-validation could be established to describe metabolic differences in seed or skin over time (seed, Q2 = 0.25, CV-ANOVA p > 0.05; skin, Q2 = 0.27, CV-ANOVA p = 0.022). Because the OPLS-DA skin-by-time model was valid, we have included it here despite low Q2. Metabolites with high scores in the corresponding S-plot (**Figures 3B,D**) indicate why the late ripening stages observed were distinctive. The skins of t1 and t3 berries were distinguished by their high levels of flavonols and other flavonoids versus phenylalanine and sugars, respectively (**Figure 3B**), and the pulp of t1 and t2/t3 berries

importance (VIP > 2) are indicated with a cross symbol. Putative biomarkers are labeled. Interactive versions of B and D are provided in Presentation 1.

could be segregated on the basis of high phenolic acids versus amino acids and sugars, respectively (**Figure 3D**).

Overall, biomarkers that define differences between ripening groups and were shared across all three tissues were exclusively sugars, specifically sucrose species and a di-hexose derivative. Leucine/isoleucine was an in-common biomarker between seed and pulp that distinguished Lagging from Advanced fruits (**Figure 4**).

Taken together, these results indicate that (1) variability late in ripening is associated with the developmental inequalities present at the ripening onset, (2) the metabolome remains dynamic post-harvest for pulp (and less for skin), with markers that define points during this late ripening period, and (3) that for seeds, the differences associated with ripening group at véraison exceeded those associated with change over the period observed.

### Trends in Intra-cluster Variance during Late Ripening

Next, trends in ICV late in ripening and the amount of ICV for metabolites with constant ICV were measured. The log<sup>10</sup> fold-change in variance was plotted for each metabolite between pairs of sequential time points. This allows the visualization of ICV patterns for each annotated metabolite in the data (**Figure 5**). The ICV patterns characteristic of each region are summarized in **Figure 5A**. How metabolite variance behaved as fruits enter this late phase should provide evidence regarding whether the ripening classes were developmentally uniform or not. Increasing ICV might suggest developmental inequality (Quadrants 2, 3, and 4), whereas decreasing ICV suggests migration toward developmental uniformity (+y/+x-axis and Quadrant 1). Metabolites with constant ICV would require further examination (Center).

None of the annotated metabolites, in any tissue, significantly and exclusively increased or decreased in variance (**Figures 5B–D**). Though, several metabolites did demonstrate a significant change in ICV between a single pair of time points in each tissue. Further, some non-significant but observable trends appear upon examining classes of metabolites.

In seed (**Figure 5B**), amino acids, anthocyanins, flavonols, and organic acids predominantly localized on the right-hand side of the plot. Most metabolites that showed significant changes in ICV fell in quadrants 2 or 4. Several, however, experienced significant reductions in variation, characteristic of the +y-axis and +x-axis regions including several pigments, a P2-type proanthocyanidin, and quercetin-o-glucuronide. In skin (**Figure 5C**), leucine showed significant changes in ICV characteristic of quadrant 2 over the time-course and most amino acids fell in quadrants 2 and 3. Anthocyanins and sugars localized predominantly in quadrant 2, and phenolic acids, flavonols and other flavonoids in quadrants 1 and 2. Stilbenes mostly fell in quadrants 1 and 4, and proanthocyanidins into 2 and 3. Like seed, most metabolites in skin that showed significant changes in ICV fell in quadrants 2 or 4, and few metabolites showed no increase in ICV. In skin, this included quercetin aglycone, citric acid, and a glucoside of cis-resveratrol. For pulp (**Figure 5D**), stilbenes and amino acids occurred in quadrants 2 and 3, anthocyanins and other flavonoids predominantly in quadrant 2, phenolic acids in 1 and 2, and flavanols in 1 and 4. Like seed and skin, most metabolites in pulp that showed significant changes in ICV over the time course fell in quadrants 2 and 4, and several exhibited reductions in ICV either between t1 and t2 or t2 and t3. In pulp, these included sucrose and a hexose fragment, several proanthocyanidins, flavanols, and flavonols.

Overall, though, most metabolites did not show significant changes in ICV over the time-course (**Figures 5B–D**). Taking these metabolites, the magnitude of variance that persisted in the cluster was explored (**Figure 6**). This group could include metabolites with persistently high or low variance across the ripening classes throughout this study; in other words, the amount of difference between ripening classes did not significantly change over time for these metabolites, and that difference could be either large or small. The magnitude of ICV also provides insight into the developmental uniformity of berries in the cluster. High, constant ICV might suggest persistent developmental differences, whereas low, constant ICV suggests developmental uniformity.

Classes of metabolites tended to exhibit similar levels of ICV, so ICV was summarized per metabolite class in **Table 1**. In all tissues, amino acids had comparatively high levels of persistent ICV. Some of the most contextually important metabolites for each tissue tended to be among the most persistently variable– proanthocyanidins and flavanols in seed (**Figure 6A**), anthocyanins in skin (**Figure 6B**), and sugars in pulp (**Figure 6C**). In addition, flavonols and phenolic acids in seed (**Figure 6A**) and

flavanols and proanthocyanidins in skin (**Figure 6B**) were among the least variable groups of metabolites.

These results indicate that the tendency toward reduced ICV is rare late in ripening, that variability will remain constant over this period for most metabolites (at least those considered here), and that high variability was often observed for the most spatiocontextually relevant metabolites.

#### DISCUSSION

This study predominantly examined the extent and behavior of intra-cluster variation late in the ripening phase and, similar to others who observed variation among berries, we observed variability between ripening classes (Carroll et al., 1978; Barbagallo et al., 2011; Kontoudakis et al., 2011; Rolle et al., 2011; Liu et al., 2016). In contrast to earlier studies which used berry density, color or weight classes to classify fruits and assess metabolite differences, we directly assessed variation associated with uneven ripening onset and, therefore, developmental inequality. If fruit uniformity at harvest is desirable, then understanding how metabolites accumulate in a developmentally diverse cluster of fruits, particularly late in ripening, should aid the identification of biomarkers to improve harvest decisions. We propose an approach to identify markers in the future and, given our data, the features that make this

change between both pairs of time points, cross. Significance threshold, p-value < 0.05. Interactive versions of (B–D) are provided in Presentation 1.

challenging. First, our inability to model four distinct ripening classes (instead, modeling Lagging vs. Advanced), which at véraison were distinguishable, does support some reduction in developmental variation across the classes as was observed in terms of gene expression and berry size by others (Gray and Coombe, 2009; Gouthu et al., 2014). However, developmentassociated differences among the ripening groups over the lateripening period were still identifiable, and there were diverse patterns in ICV except a significant, continual reduction in variation.

Fruits within a cluster are strong, competitive sinks during ripening (Coombe, 1988; Davies et al., 1999). However, changes in sugar concentration are more associated with dehydration than sugar import into berries near commercial harvest (Coombe and McCarthy, 2000). This change delineates the majority of the ripening phase from that observed in this study. Fruits may indeed undergo a reduction of intra-cluster variation during ripening (Gray and Coombe, 2009; Gouthu et al., 2014), but if fruits were truly developmentally uniform or were approaching uniformity, spatio-contextually relevant metabolites would have low ICV or only have exhibited reductions in ICV as fruits enter this late stage. Instead, the re-divergence of the ripening classes in terms of ◦Brix and other metabolites, as well as persistently high variance in others indicates that the fruits remain developmentally distinct. Metabolic uniformity does not necessarily imply developmental uniformity. This contrasts the conclusions of Gray and Coombe (2009), wherein fruits must developmentally synchronize to proceed into subsequent growth stages. In the interest of improving harvest decisions, trends in ICV may be a worthy consideration, given that if one waits longer to harvest, for instance, there is no guarantee that variation will continually reduce.

In each of the tissues studied, amino acids were among the most variable overall. Most amino acids showed no significant changes in ICV over this experiment, except for leucine/isoleucine hexose in seed and skin. In addition, leucine/isoleucine was able to distinguish ripening groups in seed and pulp. Significant differences between ripening classes were observed at one or more time points for arginine, phenylalanine, leucine/isoleucine, leucine/isoleucine hexose, and tryptophan; these amino acids and proline were also able to distinguish t1 from t3 berry pulp (**Supplementary Figure S3**). Together, proline and arginine constitute 90% of the Nitrogen content in grape juice and influence the perception of acidity in wine (Gerós et al., 2012). Arginine and phenylalanine are both sources of Yeast Assimilable Nitrogen (YAN), and phenylalanine specifically is the precursor for the phenylpropanoid pathway giving rise to flavonoids and stilbenes. As the most abundant yeast-assimilable, N-containing metabolite in juice, arginine content is one factor in the production of fruity and floral wine aromas (Gutiérrez et al., 2015). Lagging berries had significantly higher levels of arginine and phenylalanine than pink and red berries at one or more times in and pulp tissues (**Supplementary Figure S3**). skin More specifically, in skin, the level of arginine in GH berries was significantly higher than in PS and RS berries at times 1 and 3; arginine was significantly higher in GS than PS and RS at t2. Likewise, the level of phenylalanine in skin was

TABLE 1 | Tukey HSD tests comparing variances of metabolites classes within individual sampling dates.


In each metabolite class, only metabolites with persistent variance (as in Figure 6) were included in comparisons. Groups that do not share letters in common are significantly different, p-value < 0.05.

significantly higher in GS than in PS and RS at t2. In pulp, the level of phenylalanine was significantly higher for GH berries than the other classes at t2 and RS at t3. Similarly, the pulp levels of arginine in GH berries were higher than PS berries at t2 and RS berries at t3. Developmental differences in either arginine or phenylalanine could be important determinants of the differences we observed among downstream secondary metabolites. Leucine/isoleucine was among the metabolites best able to distinguish the seeds and pulp of less from more advanced berries. Phenylalanine and leucine participate in the production of higher alcohols during fermentation, namely 2-phenylethanol and isoamyl alcohol, the most abundant higher alcohols found in wine; these higher alcohols affect the aromas of wine and model solutions (Yoshizawa et al., 1961; Äyräpää, 1967; Vilanova et al., 2013; Noguerol-Pato et al., 2014; Vidal et al., 2014; Cameleyre et al., 2015). Previous work has implicated 2-phenylethanol in Pinot noir aroma, the variety used in this study (Miranda-Lopez et al., 1992; Girard et al., 2001). Finally, the potential implications of developmental inequalities in tryptophan are also interesting. Tryptophan is a precursor of auxin, a major regulator of fruit development and suspected precursor of 2-aminoacetophenone (AAP), an off-aroma described in white wines and the production of which can vary with harvest time (Hoenicke et al., 2001, 2002; Maeda and Dudareva, 2012; Schneider, 2014). Future studies could shed more light on how amino acid inequalities between individual berries originate and possibly propagate other metabolite inequalities, for better or worse.

Although ripening groups were distinguishable in each of the tissues examined, t1 vs. t3 fruits were only reliably differentiable in pulp and per their sugar and amino acid content. These observations add to previous reports which also demonstrate pulp continues to undergo metabolism and transcriptomic

changes during dehydration on and off the vine (Bellincontro et al., 2004; Costantini et al., 2006; Moreno et al., 2008; Rolle et al., 2009; Toffali et al., 2011; Bowen and Reynolds, 2015; Khairallah et al., 2016; Lukic et al., 2016 ´ ; Zenoni et al., 2016). For seed, the OPLS-DA and ICV patterns present seemingly contrasting results. OPLS-DA was unable to define seeds by their collection date, though was able to segregate ripening groups, and significant changes in ICV were observed for some metabolites between pairs of time points in seed. This suggests that the change in ICV over time in seed was not sufficiently large so as to define one time point versus another even though the difference between ripening groups may have expanded or contracted for some metabolites. Véraison for an individual berry marks the onset of ripening and coincides with the initiation of seed maturation, tannin oxidation, a cessation of seed growth, and seed dehydration (Ristic and Iland, 2005). The lack of or inability to observe changes is not entirely unexpected, given that the seed has matured by this period (Ristic and Iland, 2005) and changes that occur over time in seed might only be observed over longer time scales than in this study. Perhaps underlying factors that contributed to differences in seeds early in their development also influence seed composition after seeds have completed maturation such that they are distinguishable past the largest phases of seed development.

Tartaric and malic acids compose ∼90% of total berry acidity. Both acids are considered in harvesting decisions and impact final wine composition and perception. Typically, levels increase in the berry up to 4 weeks after anthesis and decline during ripening (Kliewer et al., 1967; Lamikanra et al., 1995). Consistent with this expectation, tartaric acid was consistently highest in and able to distinguish t1 versus t3 fruits in pulp. Furthermore, tartaric acid was significantly higher in GH fruits' pulp than in GS, PS, and RS (**Supplementary Figure S4**); tartaric acid was probably an unsuitable marker of Lagging fruits (by OPLS-DA) because GS was indistinguishable from the red berries over the time course (**Supplementary Figure S4**). Although OPLS-DA analyses indicate that changes in organic acids late in ripening among berries have more to do with a dynamic pulp over time, this observation upon close inspection suggests that developmental differences do persist such that GH is perpetually laggard if defined by its tartaric acid content. Further examination could reveal how these inequalities in primary metabolites occur in the first place, and the extent to which they, plus environmental and physiological factors, contribute to wine quality and ICV among secondary metabolites.

Among the secondary metabolites are berry phenolics, which contribute to the color, astringency, and bitterness of wines. Because of their importance, there is substantial interest in characterizing phenolic composition in a way to better inform harvest times and anticipate wine composition (Cagnasso et al., 2008; Tian et al., 2009; Kontoudakis et al., 2011). Persistently variable ripening classes would include, then, fruits at different levels of phenolic maturity. The color of red wines is influenced by anthocyanins and other phenolics (copigmentation). Anthocyanins were highest in Lagging fruits' skin, were able to distinguish Lagging from Advanced berries, and were among the most variable group of metabolites in skin tissues. Interestingly, developmental differences among the ripening classes were well-described by anthocyanins; being developmentally delayed and possibly having passed into the period of pigment decline (Hilbert et al., 2015), respectively, GH and RS fruits generally had lower anthocyanin levels, whereas the intermediate classes (GS and PS) tended to exhibit higher pigment levels (**Supplementary Figure S5**). Bautista-Ortín et al. (2006) showed anthocyanin extraction was improved in laterharvested fruits, even though they have lower pigment levels overall, to such a degree that slight over-maturation of fruit is perhaps desirable for determining harvest times (Bautista-Ortín et al., 2006); the developmental distance between the ripening classes may not only impact the abundance of anthocyanins, but their extractability as well.

Proanthocyanidins make significant contributions to wine bitterness and astringency in addition to their co-pigmentation activities. Kontoudakis et al. (2011) showed that underripe berries will differ from high density counterparts in their contribution of seed versus skin tannins, degree of polymerization, and resultant wine quality. Proanthocyanidins and flavanols were among the most variable metabolites in seeds, but only proanthocyanins were highest in and distinguished Lagging seeds (**Figure 2B** and **Supplementary Figure S6**). Per their seeds' proanthocyanidin content, pink and red berries could be considered developmentally ahead of green fruits. Again, however, changes in the seeds over time were not sufficient to define time points over this period. This result is consistent with Ristic and Iland (2005), who showed tannin accumulation closely tied to fruit development and ripening, peaking at véraison and declining in seeds as they dry, mature, and brown. Curiously, catechin accumulation also peaks at véraison in seeds and skin and declines during ripening (Kennedy et al., 2000; Downey et al., 2003a; Ristic and Iland, 2005), though flavanols did not define ripening groups in seeds as did the proanthocyanidins. Furthermore, proanthocyanidins and flavanols were the least persistently variable in skin tissues; the same trends in abundance (green > red) were not observed in skin even though they are contextually relevant within that tissue.

Unlike anthocyanins and proanthocyanidins, hydroxycinnamic acids were neither distinctive features of a ripening group, nor at remarkable levels of ICV. Upon closer examination, no significant differences were observed among the ripening classes for caftaric or coutaric acids in pulp, though their abundance did change over time (**Supplementary Figure S4**). However, significant differences in gallic acid (a hydroxybenzoic acid derivative) were observed among the ripening classes' seeds, with higher concentrations observed in pink and red fruits than in green fruits' seeds. This result is somewhat consistent with Tian et al. (2009), who showed higher levels of gallic acid in must for later harvested fruits and is additional evidence that RS and PS remained "ahead" of green fruits late in ripening.

Unlike other metabolite classes which provide evidence that fruits are developmentally distinct, stilbenes did not define either of the ripening groups, though many stilbenes showed relatively high levels of ICV in skin and changed in ICV between at least

one pair of time points in several tissues. High levels of stilbenes are found in berry skin, with variability between cultivars (Sun et al., 2006), can be induced to resist pathogens, and increase from véraison to harvest in Pinot noir (Gatto et al., 2008). Sources of variation besides developmental differences may contribute to the high ICV observed for stilbenes. Similarly, flavonols were not effective markers of developmental differences, but did distinguish t1 from t3 fruit overall, and were highly variable in a contextually relevant tissue—skin (Downey et al., 2003b).

If either uniformity or minimal heterogeneity is optimal, then when metabolites of interest are at a desired level and uniformity across the cluster would be pertinent to harvest decisions and identifying measurable biomarkers for this purpose would be valuable. This, of course, is complicated by developmental inequalities within the cluster that persist. An ideal marker would (1) have persistently low ICV such that any berry, regardless of developmental stage, could be used as a representative individual and (2) change over time so that a particular level of the marker might be associated with low ICV for a set of metabolites of interest. Although sugars are important in harvest decisions, those annotated here would be unsuitable markers by this definition. The sugars in our metabolite library were the best indicators for segregating ripening groups and several were indicative of their diversity, not of their uniformity. Identifying such a marker would at least require a longer time scale, more frequent sampling intervals than used in this study, thorough field testing, and would benefit from expanding grape-specific metabolite libraries (Aretz and Meierhofer, 2016). This study contributes to the growing body of work and interest in uneven ripening, its implications, and is pertinent to an ever-evolving paradigm in harvest decision-making. To better understand heterogeneity within the cluster and its impacts, future studies should continue to segregate variation associated with inherently unequal "starting positions" as well as other factors, such as microclimate and position in and of the cluster (Reshef et al., 2017).

#### AUTHOR CONTRIBUTIONS

All authors assisted in critical revision of this manuscript, approve of the version submitted for publication, and have agreed to be accountable for all aspects of the work, including ensuring that all questions regarding the accuracy or integrity of the work are appropriately investigated and resolved. This work was conceived by AV, FG, and LD. The experimental design and metabolite extraction were performed by AV and MC. The LC-MS instrumentation and metabolite identification were performed by MC and FG. The description of the LC-MS methodology was provided by MC. The statistical analyses were performed and manuscript was composed by AV.

# FUNDING

This work was funded by the Grape Research Coordination Network (NSF 0741876).

## ACKNOWLEDGMENTS

The authors are thankful for financial support for this project from the Grape Research Coordination Network and ongoing support from Oregon State University's College of Agriculture, the Oregon Wine Research Institute, and University of Verona's Department of Biotechnology.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2017.01108/ full#supplementary-material

FIGURE S1 | PCA bi-plot of all samples [p (corr)] and associated loadings [metabolites, t (corr)], plotted using components 1 and 2.

FIGURE S2 | Score plots from OPLS-DA analysis, by ripening group, of skin (A), and pulp (B) which show separation of samples. Hotelling's T2 Ellipse (95%) is not shown, but samples outside the ellipse are denoted with an asterisk. Samples are distributed along a predictive component (x-axis) and orthogonal component (y-axis) and are colored per their ripening class: GH, green; GS, light green; PS, pink; RS, red.

FIGURE S3 | Selected amino acid dynamics over the time course. Tukey HSD tests were used to compare ripening classes at individual time points. Significant differences between the ripening classes at single time points are denoted with different letters, p-value < 0.05. For visual ease, the points may be shifted slightly to the right. However, there are only three time points (t1, t2, and t3) at which measurements were taken.

FIGURE S4 | Selected organic and phenolic acid dynamics over the time course. Tukey HSD tests were used to compare ripening classes at individual time points. Significant differences between the ripening classes at single time points are denoted with different letters, p-value < 0.05. For visual ease, the points may be shifted slightly to the right. However, there are only three time points (t1, t2, and t3) at which measurements were taken.

FIGURE S5 | Selected anthocyanins over the time course. Tukey HSD tests were used to compare ripening classes at individual time points. Significant differences between the ripening classes at single time points are denoted with different letters, p-value < 0.05. For visual ease, the points may be shifted slightly to the right. However, there are only three time points (t1, t2, and t3) at which measurements were taken.

FIGURE S6 | Selected proanthocyanidins over the time course. Tukey HSD tests were used to compare ripening classes at individual time points. Significant differences between the ripening classes at single time points are denoted with different letters, p-value < 0.05. For visual ease, the points may be shifted slightly to the right. However, there are only three time points (t1, t2, and t3) at which measurements were taken.

DATA SHEET 1 | Metabolite annotation.

DATA SHEET 2 | SIMCA OPLS-DA model parameters and score plot axes details.

PRESENTATION 1 | Interactive figures.

characteristics of Vitis vinifera L. cv. Syrah. S. Afr. J. Enol. Vitic 32, 129–136. Bautista-Ortín, A. B., Fernández-Fernández, J. I., López-Roca, J. M., and Gómez-Plaza, E. (2006). The effect of grape ripening stage on red wine color. J. Int. Sci. Vigne Vin. 40, 15–24. doi: 10.20870/oeno-one.2006.40.1.879

Anesi, A., Stocchero, M., Santo, S. D., Commisso, M., Zenoni, S., Ceoldo, S., et al. (2015). Towards a scientific interpretation of the terroir concept: plasticity of the grape berry metabolome. BMC Plant Biol. 15, 191. doi: 10.1186/s12870-015-

Aretz, I., and Meierhofer, D. (2016). Advantages and pitfalls of mass spectrometry based metabolome profiling in systems biology. Int. J. Mol. Sci. 17, 632.

Äyräpää, T. (1967). Formation of higher alcohols from 14C-labelled valine and leucine. J. Inst. Brew. 73, 17–30. doi: 10.1002/j.2050-0416.1967.tb03012.x Barbagallo, M. G., Guidoni, S., and Hunter, J. J. (2011). Berry size and qualitative


0584-4

REFERENCES

doi: 10.3390/ijms17050632

fpls-08-01108 June 28, 2017 Time: 18:39 # 13


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Vondras, Commisso, Guzzo and Deluc. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# System-Level and Granger Network Analysis of Integrated Proteomic and Metabolomic Dynamics Identifies Key Points of Grape Berry Development at the Interface of Primary and Secondary Metabolism

#### Lei Wang<sup>1</sup> , Xiaoliang Sun<sup>1</sup> , Jakob Weiszmann1, 2 and Wolfram Weckwerth1, 2 \*

*<sup>1</sup> Department of Ecogenomics and Systems Biology, University of Vienna, Vienna, Austria, <sup>2</sup> Vienna Metabolomics Center, University of Vienna, Vienna, Austria*

#### Edited by:

*Simone Diego Castellarin, University of British Columbia, Canada*

#### Reviewed by:

*Darren Wong, Australian National University, Australia Flavia Guzzo, University of Verona, Italy*

> \*Correspondence: *Wolfram Weckwerth wolfram.weckwerth@univie.ac.at*

#### Specialty section:

*This article was submitted to Crop Science and Horticulture, a section of the journal Frontiers in Plant Science*

Received: *23 November 2016* Accepted: *02 June 2017* Published: *30 June 2017*

#### Citation:

*Wang L, Sun X, Weiszmann J and Weckwerth W (2017) System-Level and Granger Network Analysis of Integrated Proteomic and Metabolomic Dynamics Identifies Key Points of Grape Berry Development at the Interface of Primary and Secondary Metabolism. Front. Plant Sci. 8:1066. doi: 10.3389/fpls.2017.01066* Grapevine is a fruit crop with worldwide economic importance. The grape berry undergoes complex biochemical changes from fruit set until ripening. This ripening process and production processes define the wine quality. Thus, a thorough understanding of berry ripening is crucial for the prediction of wine quality. For a systemic analysis of grape berry development we applied mass spectrometry based platforms to analyse the metabolome and proteome of Early Campbell at 12 stages covering major developmental phases. Primary metabolites involved in central carbon metabolism, such as sugars, organic acids and amino acids together with various bioactive secondary metabolites like flavonols, flavan-3-ols and anthocyanins were annotated and quantified. At the same time, the proteomic analysis revealed the protein dynamics of the developing grape berries. Multivariate statistical analysis of the integrated metabolomic and proteomic dataset revealed the growth trajectory and corresponding metabolites and proteins contributing most to the specific developmental process. K-means clustering analysis revealed 12 highly specific clusters of coregulated metabolites and proteins. Granger causality network analysis allowed for the identification of time-shift correlations between metabolite-metabolite, protein- protein and protein-metabolite pairs which is especially interesting for the understanding of developmental processes. The integration of metabolite and protein dynamics with their corresponding biochemical pathways revealed an energy-linked metabolism before veraison with high abundances of amino acids and accumulation of organic acids, followed by protein and secondary metabolite synthesis. Anthocyanins were strongly accumulated after veraison whereas other flavonoids were in higher abundance at early developmental stages and decreased during the grape berry developmental processes. A comparison of the anthocyanin profile of Early Campbell to other cultivars revealed similarities to Concord grape and indicates the strong effect of genetic background on metabolic partitioning in primary and secondary metabolism.

Keywords: Vitis vinifera, berry development, mass spectrometry, primary metabolism, secondary metabolism, flavonoids, systems biology, data integration

# INTRODUCTION

Grapevine (Vitis vinifera L.) is one of the most important and widely cultivated economic crops. Grape berries are consumed either as fresh fruit or processed to raisins, juice and wine. Besides its enormous economical and nutritional values, grapes and grape products possess a wide variety of health benefits, such as antioxidation (Sánchez-Moreno et al., 1999; Doshi et al., 2006; Sánchez-Alonso et al., 2007; Sáyago-Ayerdi et al., 2009; Anastasiadi et al., 2010), cardiovascular protection (Tebib et al., 1994; Adisakwattana et al., 2010; Razavi et al., 2013), neuroprotection (Feng et al., 2007), anti-obesity (Kim et al., 2013; Zhang et al., 2013), etc.

The grape berry is a non-climacteric fruit. From fruit set to ripening, grape berries undergo three main developmental phases including two sigmoidal growth phases with an intermediate lag phase (Kennedy, 2002). The performance of grape berry development is characterized by dramatic changes in both physiology and biochemistry, including increases in volume and weight, changes in texture, color, aroma, acidity, sugar contents, susceptibility to disease, etc. The first growth phase (phase I) is characterized by fruit formation and enlargement due to the active cell division and expansion. In this phase, a notable accumulation of organic acids, especially malic and tartaric acid has been observed (Conde et al., 2007). Phase II, which is defined as a lag phase features a slow enlargement of berry volume caused by a stop in cell division. The grape berry is still green and hard at this phase. Organic acids continuously accumulate until veraison, which marks the beginning of phase III. During the last phase, the grape berries undergo a second sigmoidal growth accompanied by a decrease in acidity and increase in sugar content (Conde et al., 2007; Deluc et al., 2007; Fortes et al., 2011; Liang et al., 2011; Dai et al., 2013; Degu et al., 2014; Fraige et al., 2015; Cuadros-Inostroza et al., 2016). The peel of red varieties colors as a result of the accumulation of anthocyanins (Boss et al., 1996; Ali M. B. et al., 2011; Degu et al., 2014; Fraige et al., 2015). The grape berry becomes soft in the final phase and is ready to be harvested. Another generally adopted descriptive system is the E-L system which was proposed firstly by Eichhorn and Lorenz (1978) with a more detailed description of grape berry development stages.

Fruit development is an intricate process, featuring complex regulation and fine-tuned changes in metabolism. Its analysis requires the use of sensitive methods, which allow high sample throughput to cope with the amount of samples necessary to examine a time continuous process.

Since the release of the grapevine genome sequence in 2007 (Jaillon et al., 2007; Velasco et al., 2007), studies of developing grape berry based on transcriptomic (Deluc et al., 2007; Palumbo et al., 2014), proteomic (Giribaldi et al., 2007; Negri et al., 2008; Martinez-Esteso et al., 2011; Fraige et al., 2015) and metabolomic (Ali K. et al., 2011; Dai et al., 2013; Degu et al., 2015) techniques contributed extensively to our understanding of berry growing and ripening process. These studies not only enhanced and supplemented the morphological and physiological descriptions but also promoted the work to molecular level. Exploring the developmental process basing on a single level data results in a partial view of the progress. Several studies described the developmental process by combining transcriptomic and metabolomic profiles (Fortes et al., 2011; Agudelo-Romero et al., 2013; Degu et al., 2014). Considering that the proteome is the active part of the metabolic phenotype, integration and complex statistical correlation network analysis of those data will provide crucial information for the understanding of the metabolic and physiological changes (Weckwerth et al., 2004b; Morgenthal et al., 2005; Wienkoop et al., 2008; Valledor et al., 2013, 2014; Nukarinen et al., 2016; Wang et al., 2016a,c). Nonetheless, systematic analysis of integrated metabolome and proteome profiles of developing grape berries is still less covered. It is also problematic to schematize the metabolic dynamics of developing grape berry by summarizing or comparing those studies due to the coverage limitation of either developmental stages or metabolism branches. For instance, some studies only focus on primary metabolism (Dai et al., 2013) whereas others target flavonoid accumulation during grape berry ripening (Ali M. B. et al., 2011). Zamboni and coworkers integrated the transcriptomic, proteomic and secondary metabolite data of four developmental and three postharvest time points of Corvina grape berry into a complex statistical correlation network analysis for the identification of putative, stage-specific biomarkers (Zamboni et al., 2010). In addition, some studies worked on individual parts of grape berries, such as skin (Negri et al., 2008; Ali M. B. et al., 2011; Degu et al., 2014, 2015; Wu et al., 2014) or berries depleted of seed or peel (Martinez-Esteso et al., 2011; Fang et al., 2013; Fraige et al., 2015).

In this study, we harvested samples according to the modified E-L system (Coombe, 1995) from fruit set to ripening at 12 time points. Mass spectrometry based high-throughput platforms were applied for the metabolomic and proteomic analysis of both primary and secondary metabolism dynamics of developing grape berries. Multivariate statistical analysis of the dynamics of metabolites and proteins involved in primary metabolism i.e., glycolysis, tricarboxylic acid (TCA) cycle, amino acid metabolism as well as secondary metabolism i.e., flavonol, flavan-3-ols, anthocyanins and lignin unveiled metabolism interactions during the berry growing period.

# MATERIALS AND METHODS

#### Sample Collection

Berries at 12 developmental stages corresponding to EL 27, 29, 30, 31, 32, 33, 34, 35, 36, 37, 37.5 (to distinguish with the samples at early EL 37 stage), 38 (**Figure 1**) were harvested according to the modified Eichhorn-Lorenz system (E-L system) (Eichhorn and Lorenz, 1978; Coombe, 1995) from V. vinifera (Early Campbell) growing in the plant garden of University of Vienna (48◦ 13′ 50.2′′N 16◦ 21′ 28.2′′E) during the 2014 growing year. The plants did not receive any specific training system. Three biological replicates each containing 5 to 10 berries were collected for each developmental phase. The harvested berries were frozen in liquid nitrogen immediately and stored at −80◦C.

#### Metabolite and Protein Extraction

An integrative extraction of metabolites and proteins was performed according to a universal extraction protocol

(Weckwerth et al., 2004b) with some modifications. The grape berries were ground to fine powder in liquid nitrogen using mortar and pestle. 50 to 100 mg of material was extracted with 750µl of extraction solution (methanol: water: formic acid = 70:28:2) and 250µl of hexane. The mixture was homogenized by vigorous vortexing and incubated 30 min on ice. Then the mixture was centrifugated at 20,000 g for 8 min to separate the lipophilic and hydrophilic phases which were subsequently transferred into new tubes, respectively. The extraction procedure was repeated once with the lipophilic and hydrophilic phases pooled together with those from the first extraction, respectively. The extracts were dried under vacuum. Proteins were extracted from the residue pellets according to a previous protocol (Noah et al., 2013).

#### Metabolite Measurement, Identification and Quantification

The dried hydrophilic phases were re-dissolved in 400 µl of 50% methanol. For the primary metabolite analysis, 25µl of this re-dissolved hydrophilic phase was dried under vacuum and subsequently derivatized according to a modified protocol (Weckwerth et al., 2004b; Mari et al., 2013). Agilent <sup>R</sup> 6890 gas chromatograph coupled to a LECO Pegasus <sup>R</sup> 4D GC × GC-TOF spectrometer was used for the primary metabolite measurement. Instrument parameters were set as described previously (Doerfler et al., 2013). GC separation was performed at a constant flow 1 mL min−<sup>1</sup> helium. Initial oven temperature was set to 70◦C and hold for 1 min, followed by a ramp to 76◦C at 1◦C min−<sup>1</sup> and a second ramp at 6◦C min−<sup>1</sup> to 350◦C hold for 1 min. Transfer line temperature was set to 340◦C and post run temperature to 325◦C for 10 min. The metabolite identification and quantification was performed with LECO Chroma TOF <sup>R</sup> . Retention times (RTs) of the peaks were converted to retention indices (RIs) according to the RTs of spiked alkanes (C12-C40). Metabolites were annotated by comparing their RIs and mass spectra to those of standards in the GMD Golm database (Kopka et al., 2005) with a minimum match factor set to 700. The peak areas of the annotated metabolites corresponding to specific masses were extracted and used for relative quantification. Mixtures of standard compounds were measured under the same conditions at different concentrations to calculate the standard curves for absolute quantification.

For the secondary metabolite analysis, 10µl of the redissolved hydrophilic phase was mixed with 2.5µl of reserpine (5 mg l−<sup>1</sup> ) as an internal standard, 10µl of 1.0% formic acid (FA) solution and 77.5µl of water. After centrifugation at 20,000 g for 8 min, 5µl of the supernatant were loaded on Waters ACQUITY UPLC HSS T3 nanoACQUITY Column (particle size 1.8µm, dimension 100µm × 100 mm) via a HTC PAL Autosampler device coupled to an Eksigent nano LC pump and eluted with a non-linear gradient (Mari et al., 2013) at a constant flow rate of 500 nl min−<sup>1</sup> . The LC conditions were 5% B during 0– 3 min, a linear increase from 5 to 20% B during 3–25 min, from 20 to 40% B during 25–40 min and from 40 to 50% B during 40–55 min, finally from 50 to 95% B during 55–63 min followed by 15 min of maintenance with a flow rate of 500 nl min−<sup>1</sup> . Ionization was performed by a nano ESI source (Thermo Scientific, USA) in positive mode with the masses analyzed by a LTQ Orbitrap XLTM mass spectrometer (Thermo, Germany). Each full scan was followed by one MS/MS scan with the most abundant precursor ion fragmented by collision induced dissociation (CID) under 35% of the normalized collision energy during 90 ms activation time. The minimum signal threshold was set to 50,000. Before measurement, the machine was calibrated and standards were measured to check the condition of measurement. We also ensured linearity of the spiked internal standard in different concentrations. The combination of a very low flow rate (500 nL/min) and a gradient that minimized co-elution was chosen to minimize matrix effects. For secondary metabolite identification, accurate precursor masses, sum formula RTs together with mass accuracy were exported from Xcalibur (Thermo Xcalibur 2.2 SP1.48) and compared with the information from literature or standard compounds. The annotation levels were marked according to a standard proposed by the Metabolomics Standards Initiative (Sumner et al., 2007). LCquan (Thermo, v2.6.6.1128) was used for peak extraction and peak area integration.

#### Protein Digestion and Analysis

Protein concentration was determined by the Bradford method (Bradford, 1976) with a BSA standard curve. 100µg of protein were firstly reduced with dithiothreitol (DTT, 5 mM, 37◦C, 45 min); then alkylated with iodoacetamide (IAA, 10 mM, 23◦C, dark, 60 min) and finally 5 mM of DTT was added (23◦C, dark, 15 min). Endoproteinase LysC and trypsin were applied for digestion based on a previous protocol (Hoehenwarter et al., 2008). After digestion, samples were desalted with C18-SPEC-96 well plate (15 mg, Agilent, USA) according to the manufacturer's instruction. The eluted peptides were dried under vacuum and dissolved in 500µl of start gradient solution (4.5% acetonitrile, 0.1% FA). 1µg of the digested protein was loaded on an Ascentis Peptide ES-C18 column (particle size 2.7µm, dimension 15 cm × 100µm, Sigma-Aldrich, USA) and eluted with a 90 min linear gradient from 5 to 40% of mobile phase B (90% acetonitrile, 0.1% FA; phase A, 0.1% FA in water) at a constant flow rate of 400 nl min−<sup>1</sup> . The same ESI-LTQ-Obitrap equipment used for metabolite analysis was applied for peptide measurement. Each full scan was followed by 10 MS/MS scans in which the 10 most abundant ions were selected and fragmented by CID with 35% of the normalized collision energy during a 30 ms activation time. Minimum signal threshold was set to 10,000.

The obtained raw files containing peptide information were searched against a grape fasta file including 65,448 protein sequences from UniProt with the SEQUEST algorithm in Proteome Discoverer (v 1.3, Thermo Scientific). Searching parameters were set as below: maximum two missed cleavage sites, acetylation for N-terminal modification, oxidation of methionine for dynamic modification and carbamidomethylation of cysteine for static modification were allowed. Mass tolerance for precursors was set to 10 ppm and for fragment masses to 0.8 Da. False discovery rate (FDR) was set to 0.01. Protein candidates were defined by at least two peptides with high confidence. The obtained raw files and sequence information of the identified proteins were submitted to the public repository ProteomeXchange (Vizcaino et al., 2014) with the dataset identifier PXD003769 (http://www.proteomexchange. org/) as well as to the PROMEX database (http://promex.pph. univie.ac.at/promex/). Normalized spectral abundance factors (NSAFs) were calculated (Zybailov et al., 2006) for relative quantification. The protein candidates that are present in all the three biological replicates of at least one stage were considered for the statistical analysis. For the functional analysis, the identified protein sequences were blasted against a protein database of Arabidopsis thaliana (from PLAZA with 27,407 protein sequences) and Theobroma cacao (from PLAZA with 44,404 protein sequences) with the BLASTP function in NCBI (v 2.2.31, ftp://ftp.ncbi.nlm.nih.gov/blast/executables/blast+/LATEST/). A cacao mapping file from GoMapMan ("tca\_Phytozome9.1\_ transcript\_2015-01-09\_mapping.xlsx," http://www.gomapman. org/export/current/mapman) was applied for the functional analysis.

#### Statistical Analysis

The obtained metabolite data were normalized to fresh weight and dodecane (C12 alkane, GC-MS data) or total ion intensity (LC-MS data). Analysis of variance (ANOVA) and k-means clustering were performed within Matlab <sup>R</sup> (V8.4.0 R2014b; http://www.mathworks.com). The significant levels of the candidates were presented with lower case letters according to results of the Duncan's test (Duncan, 1955). K-means clustering analysis was repeated 100 times and finally the result with minimal total distance was selected. Principal component analysis (PCA), hierarchical clustering analysis and Granger causality analysis were performed with COVAIN under Matlab environment (Sun and Weckwerth, 2012). Granger causality analysis was performed on all the identified metabolites and proteins as well as the clusters after k-means clustering analysis with their time lag was set to 1, 2, 3, respectively. The correlations with p-values less than 0.05 were recorded. The network of Granger result was visualized in Cytoscape (http:// www.cytoscape.org/). The Venn diagram was drawn with Venny 2.0.2 (http://bioinfogp.cnb.csic.es/tools/venny/).

# RESULTS

### Metabolomic Profiles of Developing Grape Berry

The GC-TOF-MS platform allowed the annotation of 87 candidates including sugars, amino acids, organic acids together with simple amine and phenolic compounds, according to their RIs and mass spectra. Additionally, 49 flavonoids were annotated from LC-Orbitrap-MS data according to the accurate precursor masses, sum formula and their fragmentation patterns (**Table 1**). The detailed information (RI and RTs, quantification masses or



*(Continued)*


#### TABLE 1 | Continued

*Api, Apigenin; Caff, Caffeoyl; Cou, Coumaryl; Cy, Cyanidin; DHQ, Dihydroquercetin (taxifolin); Dp, Delphinidin; Glc, Glucose; Glu, Glucuronide; Kae, Kaempferol; Lar, Laricitrin; Lut, Luteolin; MDHQ, methyldihydroquercetin; Mv, Malvidin; Myr, Myricetin; Pg, Pelargonidin; Pn, Peonidin; Pt, Petunidin; Que, Quercetin; Rha, Rhamnoside; Tri, Tricetin.*

MS2 fragments and the integrated peak area, one way ANOVA results) of the annotated candidates is listed in **Table S1**. The dynamic patterns of the annotated metabolites were visualized by a hierarchical bi-clustering color map (**Figure 2A**). Sugars (including sugar alcohols, sugar acids), amino acids, organic acids and flavonoids were further shown in **Figures 2B–E**.

#### Sugars

The main sugars in grape berries are fructose, glucose and sucrose. In the present study, fructose constantly accumulated during development with significant increases before veraison and during ripening (**Figure 2B**). Glucose also significantly accumulated around veraison but declined afterwards (**Figure 2B**). The content of sucrose fluctuated during grape development (**Figure 2B**) with four inflection points at EL 30, 32, 34, and 36 respectively. Other sugars, sugar alcohols and sugar acids either decreased during development (ribose, xylose, myo-inositol, arabinose, rhamnose, galactaric acid) or showed the highest level at EL 32 (threonic acid, galactonic acid, gluconic acid, erythrirol) (**Figure 2B**).

#### Amino Acids

Amino acids showed distinct dynamics during grape berry development (**Figure 2C**). Arginine and asparagine were the most abundant amino acids in young berries, alanine and glutamine in mature berries (**Table S2**). Lysine, tyrosine, arginine, ornithine and phenylalanine increased significantly from EL 27 to EL 29, however, decreased dramatically until the end of lag phase (EL 34) and remained at a relatively low level during the second sigmoidal growth period (**Figure 2C**). Asparagine was in high abundance at the first two developing stages followed by a dramatic decline from EL 29 to EL 30 then stayed in low level until the end. Other amino acids fluctuated during grape berry developing and all showed a turning point at EL 34 which is the end of the lag phase and the beginning of the veraison (**Figure 2C**).

#### Organic Acids

The predominant organic acids detected in grape berry were malic acid, tartaric acid and citric acid which increased before veraison (EL 35) and decreased afterwards (**Figure 2D**). Other organic acids showed similar changing pattern except pyruvic acid, gallic acid and caffeic acid which were highest in the young berries and then decreased throughout the developmental process (**Figure 2D**).

#### Flavonoids

Grape and its products are rich in polyphenolics. These secondary metabolites, especially flavonoids, play multiple roles in grape and attract more and more attentions due to their health benefits (Anastasiadi et al., 2010; Kim et al., 2013; Zhang et al., 2013). During grape berry development, the detected flavonoids presented two distinct changing patterns (**Figure 2E**). All the anthocyanins accumulated during ripening whereas most of the candidates in the other subfamilies like proanthocyanins, flavan-3-ol, flavonol, flavanonol and their glycosides were abundant

in young berries and decreased during the time course of development (**Figure 2E**). The synthesis of anthocyanins splits into three branches, i.e., the monohydroxylated (pelargonidin, Pg), the dihydroxylated (cyanidin, Cy), and the trihydroxylated (delphinidin, Dp) branch. Cy and peonidin (Pn) glycosides which belong to the dihydroxylated branch were detected from stage EL 37 or even EL 36 whereas derivatives of the other two branches started to appear one stage later. Furthermore, in mature berries (EL 38), the relative abundance of Cy- and Dp- derivatives were higher than the corresponding derivatives of other aglycones (**Figure S1**). For instance, Cy-Cou-diGlc (**30**) and Dp-Cou-diGlc (**25**) were more abundant than petunidin- (Pt, **29**), malvidin- (Mv, **32**) and Pg- (**33**) coumaroyl-diglucoside; Cy-Glc (**2**) and Dp-Glc (**8**) were in higher level than glucoside of other aglycones (**Figure S1**).

#### Protein Profiles of Developing Grape Berry

In total, 1313 proteins were identified from all the samples (for sequences information see in **Table S3**). 848 candidates prevalent in all replicates of at least one stage were used for further statistical analysis. NSAFs and the ANOVA analysis result can be found in **Table S4**. The functions of all the protein candidates were annotated by blasting against protein sequences of A. thaliana and T. cacao. Blast results were summarized in **Table S4**. The matching with the T. cacao database yielded a higher amount of hits with a better blast quality and was therefore used for functional analysis. Subsequently, the identified protein candidates were assigned to corresponding functional bins according to the cacao mapping file from GoMapMan (**Table S4**).

Hierarchical bi-clustering analysis was applied to visualize the dynamic proteome profiles of developing grape berry (**Figure S2**). Samples of 12 developing stages were clustered into 3 groups indicated with color blue, red and green (**Figure S2**). Samples from stage EL 27, 29, 30 and 31 were assigned to group 1; EL 32, 33, 34, 35 group 2 and EL 36, 37, 37.5, 38 group 3. The Venn diagram (**Figure 3A**) shows 394 proteins were common to all groups and 157, 43 and 80 protein candidates are specific to group 1, 2, and 3, respectively. The functional distribution of these group specific proteins was summarized with pie charts (**Figure 3A**). There were 18 amino acid metabolism related proteins detected accounting for 5.83% of all the group 1 specific proteins whereas only 1 and 2 proteins were accounting for 1.61 and 1.43% respectively of group 2 and 3 (**Figure 3A**) indicating active amino acid metabolism at phase I on proteome level. There were 8 transport related proteins accounting for 2.59% of group 1 specific proteins and 5 accounting 3.57% of group 3 specific proteins whereas there was no transport related protein exclusive to group 2 (**Figure 3A**). Another notable point is that more proteins associated with secondary metabolism existed exclusively in group 1 (12, 3.88%) and 3 (12, 8.57%) than in group 2 (2, 3.23%) (**Figure 3A**) indicating the synthesis of secondary metabolites was more active in the beginning and the end of berry developmental stages than in the middle.

The annotated protein candidates were assigned to 50 functional bins (**Table S4**). The majority of functional categories include candidates involved in protein synthesis (11.41%), protein degradation (7.11%), RNA regulation of transcription (6.09%), signaling (4.38%), and abiotic stress response (4.30%) (**Figure 3B**). The changing patterns of the proteins in these functional categories were summarized into four groups by hierarchical clustering analysis (**Figure 3C**) with their summarized changing patterns shown on the right side. Proteins in 30 out of 50 functional bins were in lowest content around veraison (**Figure 3C** in red and green). In contrast, proteins involved in 7 functional groups were with highest abundance just before veraison (EL 33) (**Figure 3C** in blue). The functional bins in purple group (**Figure 3C**) involve proteins constantly accumulating during grape berry development.

Proteins related to abiotic and biotic stresses (4.30 and 2.03%, respectively) showed high abundance at early developmental stages or/and during ripening (**Table S3**, **Figure 3C** in red) indicating high resistance ability of grape berry against environmental and developmental stresses during these developmental phases. A larger amount of oxidative stress responsive proteins in young, green berries as well as increasing expression of pathogen responsive proteins after veraison have been previously reported (Giribaldi et al., 2007). Additionally, another study reported a parallel transcript profile of stress/pathogens responsive gene strongly expressed in ripening berries (Davies and Robinson, 2000).

Photosynthesis (6.1%) is another major functional category including candidates involved in light reaction (2.97%), photorespiration (0.86%), and calvin cycle (2.27%) (**Figure 3B**). Proteins involved in light reaction were in higher abundance in the earliest stage and around veraison (**Figure 3C** in blue) whereas those involved in photorespiration and calvin cycle were more abundant in young green berries. The levels of proteins in all of these three subgroups declined after veraison. The decrease in abundance of photosynthesis related proteins throughout grape berry development especially after veraison was consistent with previous proteomic studies (Martinez-Esteso et al., 2011; Fraige et al., 2015) and the physiological situation (Pandey and Farmahan, 1977) of developing grape berries.

Proteins associated with lipid metabolism were observed with high frequency and (3.1%, **Figure 3B**) showed increasing expression after veraison (**Figure 3C**). Proteins associated with secondary metabolism showed distinct changing patterns. Those involved in phenylpropanoid synthesis were in high abundance in young berries and then decreased during development (**Figure 3C** in red) whereas those related with later steps of flavonoid and isoprenoid biosynthesis were strongly accumulated after veraison (**Figure 3C** in purple). Proteins associated with synthesis of N and S containing metabolites were highly expressed around veraison. The distinct arrangement of protein expression reflected the metabolic adjustment during grape berry development.

#### Metabolome and Proteome Data Integration

#### Multivariate Statistical Analyses Reveal the Trajectory of Grape Berry Development

The PCA plot (**Figure 4A**) of the integrated metabolomics and proteomic dataset revealed a continuous trajectory during grape berry development. The separation of various developmental stages indicated a distinction of metabolism on metabolite and

protein levels. The first principal component (PC 1) accounting for 47.44% (**Figure 4A**) of the total variance characterized metabolic and proteomic specificities of grape berries at different developmental stages. Candidates with high absolute loading scores (**Table S5**) included metabolites, especially flavonoids, caffeic acid, gallic acid, lysine, asparagine, arginine and methionine together with protein candidates involved in development, lipid metabolism, cell wall construction, TCA cycle and protein degradation which accounted for most of the separation among developmental stages.

Further, k-means clustering analysis was applied to group candidates according to their changing patterns. **Figure S3** presents the 12 clusters with a bold red line indicating the averaged pattern of all the candidates in each cluster. Cluster 2 and 6 with 86 and 124 candidates, respectively, present candidates with higher abundance at early developmental stages (EL 27 to 31) (**Figure 4B**, **Figure S3**). Candidates in these two clusters include amino acids (methionine, phenylalanine, asparagine, ornithine, arginine, lysine, tyrosine), organic acids (caffeic acid, pyruvic acid, gallic acid), sugars and sugar alcohols (arabinose, rhamnose, myo-inositol), (dihydro)flavonol derivatives (epicatechin, 3, 20, 36, 49, 12, 31, 38, 39, 43, 45, compound number see in **Table 1**) and proteins involved in amino acid metabolism (**Table S6**). The candidates in cluster

between clusters that resulted from k-means clustering analysis with time lag set as 1, 2, and 3, respectively.

10 and 9 are with highest abundance at EL 32 and veraison (EL 34 and 35), respectively. The most abundant organic acids, i.e., malic acid and citric acid, together with some proteins involved in photosynthesis in cluster 5 were in a relatively higher level during the lag phase and veraison than during the two sigmoidal growth phases. In contrast, those candidates in cluster 3 and 4 showed opposite dynamic patterns including organic acids (shikimic acid, ascorbic acid, fumaric acid, maleic acid), amino acids (valine, leucine, proline isoleucine) and proteins involved in stress response (**Figure S3**, **Table S6**). All the annotated anthocyanins and four flavonol derivatives (5, 10, 21, 26, compound number see in **Table 1**), together with a high abundance of proteins involved in secondary metabolism and lipid metabolism in cluster 1, 8, and 12 mainly accumulated after veraison (EL 36 to 38) (**Figure S3**, **Table S6**). The candidates in cluster 7 show a continuous elevation in their abundance whereas those in cluster 11 decreased (**Figure S3**, **Table S6**).

#### Granger Causation Analysis of an Integrated Metabolomic/Proteomic Network during Grape Berry Development

Correlation network analysis has been widely applied in omics studies to investigate molecular correlations and connections in metabolism (Steuer et al., 2003; Weckwerth, 2003; Weckwerth et al., 2004a; Sun and Weckwerth, 2012). Those biomolecules with a higher node degree have more connections with other molecules and are therefore regarded as essential connection points in a metabolic network (Weckwerth, 2003; Weckwerth et al., 2004a). The connection degree of nodes might also vary in grape berries at different developmental stages, different cultivars (Cuadros-Inostroza et al., 2016) or under distinct growth conditions (Hochberg et al., 2013; Savoi et al., 2016). Thus, correlation network analysis is an appropriate method for searching important biomolecules involved in specific metabolism processes. Time-lagged correlation analysis by Granger causation represents an advanced level of correlation network analysis (Doerfler et al., 2013). Granger causality analysis was initially introduced by Granger (1969) to predict events based on time series data and time-lagged correlations in economics. It was also applied to some biological studies to interpret directed and nonlinear correlations between metabolites, transcripts and proteins (Walther et al., 2010; Doerfler et al., 2013; Valledor et al., 2014). To further extend the understanding of the dynamic correlations of all the identified metabolites and proteins during the developmental time course, Granger causality analysis was applied to all the candidates and clusters discussed above. The results indicated significant metabolite-metabolite, metabolite-protein and protein-protein correlations. **Figure 4C** shows the directed correlation of phenylalanine and phloretin (p-value = 0.02187) which indicated a strong effect of phenylalanine concentration on phloretin synthesis. Not only precursors but also enzymes showed significant correlation to their product synthesis. One example is the accumulation of anthocyanin synthase (ANS, A2ICC9) prior to anthocyanins (**Figure 4D**, **Table S7**). **Figure 4E** presents a significant granger correlation (p-value = 0.00538) between protein E0CSB6 (malate dehydrogenase) and D7T300 (ATPase) indicating the close relationship between TCA cycle and ATP production. Other time series correlations with pvalues less than 0.05 were summarized in **Table S7**. The Granger causation network (**Figure S4**) includes 674 nodes (**Table S7**) with 21 neighbors in average. Many amino acids (ornithine, arginine, phenylalanine, lysine, tyrosine, asparagine), organic acids (acid\_like2, shikimic acid) and their metabolism related proteins (**Table S7**) show highest node degrees revealing them as potential biochemical hub during grape berry development. The application of Granger analysis to the 12 clusters obtained from the k-means clustering analysis revealed significant time lagged correlations between these clusters. Significant directed connections among these clusters are shown in **Figures 4F–H** with the time lag set as 1, 2, and 3, respectively (one, two or three time points shifted). The Granger causality correlations from clusters 2 to 6, 6 to 10, 10 to 9, 9 to 12, and 12 to 8 (**Figure 4G** indicated with red arrows) were visualized as a line chart (**Figure 4B**) clearly showing their dynamic shift over the developmental time course, especially a significant correlation between cluster 9 and 12. Such combined Granger causality analysis with clustering analysis indicated a general metabolic shift from the metabolism of amino acids, sugars and some flavonoids to organic acid accumulation and finally to lipid and anthocyanin synthesis during grape berry development.

PCA and k-means clustering analysis presented the systemic dynamics of the metabolites and proteins during grape berry development. To understand the metabolism progress of developing grape berry in a biochemical context, we mapped the sugars, amino acids, organic acids, flavonoids and the related proteins on their corresponding biosynthetic pathways (**Figure 5**). The integration of dynamics of metabolites as well as proteins involved in both primary and secondary metabolisms presents metabolic checkpoints during grape berry development. This is further discussed below.

# DISCUSSION

# Primary Metabolism Dynamics

Primary metabolism plays an essential role in grape berry development. The products from primary metabolism pathways are not only crucial for grape survival but also endow grape berry specific characters which are further decisive of its market value.

#### Sugar Metabolism

Sugars, especially fructose, glucose and sucrose determine the sweetness of grapes, moreover, the alcohol concentration of wine. In grape berries, sucrose is mainly imported via phloem from source organs. Subsequently, the imported sucrose is either hydrolyzed to glucose and fructose by invertase or converted to glycolysis substrates via sucrose synthase (Susy) and UDP-glucose pyrophosphorylase (UGPase). The fluctuation of sucrose content during grape berry development might be caused by a disproportionate ratio of the import to the consumption. Furthermore, synthesis of sucrose from malate via the gluconeogenic pathway (Ruffner et al., 1975; Dai et al., 2013) might also contribute to the fluctuation of sucrose concentration, especially after veraison. The accumulation of glucose and fructose during grape berry development was reported previously (Wu et al., 2011; Dai et al., 2013). In the present study, fructose accumulated throughout the developmental process whereas glucose concentration did not continue to rise after veraison. Similar glucose dynamics were also observed in some table grape varieties i.e., "Thompson Seedless," "Crimson Seedless," and "Red Globe" (Muñoz-Robredo et al., 2011). In contrast to these findings, both glucose and fructose concentration constantly increased during grape berry development in some grape varieties and cultivars (Wu et al., 2011; Dai et al., 2013). The discrepancy in glucose accumulation patterns could be explained by the differences in the ripening process among varieties. The distinct expression pattern of invertase and Susy might explain the unequal accumulation of glucose and fructose. The decline

FIGURE 5 | Visualization of metabolite and protein dynamics on their biosynthetic pathways. Metabolites are written in black letters with blue line charts indicating their changing patterns whereas proteins are written in red letters with red line charts. Relative abundance of metabolites and proteins were averaged over three biological replicates. Bars represent standard errors. Susy, sucrose synthase; UGPase, UDP-glucose pyrophosphorylase; PFP, pyrophosphate-fructose 6-phosphate 1-phosphotransferase; FBPase, fructose 1, 6-bisphosphatase; PGK, phosphoglycerate kinase; PGM, phosphoglycerate mutase; PK, pyruvate kinase; PDC, pyruvate *(Continued)*

#### FIGURE 5 | Continued

dehydrogenase complex; IDH, isocitrate dehydrogenase; OGDC, oxoglutarate dehydrogenase complex; SCS, succinyl coenzyme A synthetase; SDH, succinate dehydrogenase; MDH, malate dehydrogenase; AspAT, aspartate aminotransferase; AS, asparagine synthetase; ASADH, aspartate-semialdehyde dehydrogenase; MetH, methionine synthase; MAT, methionine adenosyltransferase; PHGDH, phosphoglycerate dehydrogenase; PSAT, Phosphoserine transaminase; SHMT, serine hydroxymethyltransferase; OASTL, O-acetylserine (thiol)-lyase; GLDH, Glutamate dehydrogenase; GS, Glutamine synthetase; ASS, argininosuccinate synthase; ALAT, alanine aminotransferase; AGT, alanine-glyoxylate transaminase; CS, chorismate synthase; PAL, phenylalanine ammonia lyase; C4H, cinnamate 4-hydroxylase; 4-coumarate-CoA ligase; CCR, cinnamoyl-CoA reductase; CAD, cinnamyl-alcohol dehydrogenase; CHS, chalcone synthase; CHI, chalcone isomerase; F3H, flavanone 3-hydroxylase; F3′H, flavonoid 3′ -hydroxylase; F3′5 ′H, flavanoid 3′ ,5′ -hydroxylase; FLS, flavonol synthase; DFR, dihydroflavanol 4-reductase; ANS, anthocyanidin synthase; ANR, anthocyanidin reductase; 3-GT, anthocyanidin 3-O-glucosyltransferase.

in abundance of invertase since EL 31 caused a decrease in the production of glucose and fructose whereas the increasing expression of Susy ensured the continuous accumulation of fructose.

#### Glycolysis

The substances generated from sugar metabolism are subsequently incorporated into glycolysis. Metabolism along this process generates energy (ATP), reducing equivalents (NADH) as well as intermediates for amino acid biosynthesis, lipids and secondary metabolite production. The abundance of most glycolytic enzymes increased through the development and ripening process (**Figure 5**). Phosphoglycerate kinase (PGK) was the only glycolytic enzyme, which declined in abundance (**Figure 5**). The concentration of pyrophosphate-fructose 6 phosphate 1-phosphotransferase (PFP), enolase and pyruvate kinase (PK) in the cytosol strongly increased after veraison (**Figure 5**). Phosphoglycerate mutase (PGM) was increased during the first sigmoidal growing phase (EL 27 to EL 32) and maintained a relatively constant level afterward. The increase in abundance of glycolytic proteins was consistent with some former reports that studied the proteomic profile of grape skins and berry tissue without seeds (Negri et al., 2008; Kambiranda et al., 2014). However, some studies reported a decrease in abundance of glycolytic enzymes (Davies and Robinson, 2000; Giribaldi et al., 2007; Martinez-Esteso et al., 2011) or glycolytic intermediates (Dai et al., 2013) during berry ripening. Variety and differences in growth conditions might explain these different observations. Additionally, isoforms of enzymes may play different roles at a particular developmental stage (Chaturvedi et al., 2013; Ischebeck et al., 2014; Wang et al., 2016a,b), thus displaying varying dynamics during development. For instance, Fraige et al reported three isoforms of UDPase, of which two candidates decreased in abundance after veraison whereas one increased (Fraige et al., 2015).

#### TCA Cycle

The tricarboxylic acid cycle (TCA cycle) generates energy, reducing power and carbon skeletons, which makes it a central hub in metabolism. The Pyruvate dehydrogenase complex (PDC) converts pyruvate to acetyl-CoA which serves as fuel to the TCA cycle. During the grape berry development, PDC was in high abundance during the first growing phase (EL 27-32), followed by a significant decline at the lag phase (EL 32-34) and with a subsequent increase (**Figure 5**). Similar to PDC, isocitrate dehydrogenase (IDH), oxoglutarate dehydrogenase complex (OGDC), and succinyl coenzyme A synthetase (SCS) were in high abundance during the sigmoidal growth phases whereas in low abundance during the lag phase (**Figure 5**). The high abundance of these enzymes in the young and ripening berries was consistent with the great demand for energy and building blocks at these two phases. Aconitase and succinate dehydrogenase (SDH) were strongly expressed after veraison. In a former report, a sharp expression of aconitase was observed in ripening grape skin (Negri et al., 2008). Fumarase is the only enzyme whose expression gradually declined throughout the berry development (**Figure 5**). Malate dehydrogenase (MDH), catalyzing a reversible reaction between oxaloacetate and malate, was concentrated at a lower level at phase I whereas it showed a progressive increase in expression during ripening which was in agreement with previous reports (Martinez-Esteso et al., 2011; Kambiranda et al., 2014). The transcript levels of MDH and malic enzyme were reported to increase during grape berry ripening which might contribute to the decline of malate concentration after veraison (Deluc et al., 2007).

The contents of the intermediates, citrate, succinate, fumarate and malate gradually increased in early stages of development, up to a peak in concentration at EL 31 (fumarate and succinate), EL 33 (malate) or around veraison (citrate) with a subsequent decline. Similar dynamic patterns were observed in developing Cabernet Sauvignon berries with a peak in accumulation of most TCA cycle intermediates at veraison (Dai et al., 2013). The accumulation of these organic acids before veraison was parallel to the high abundance of enzymes at phase I. However, the gradually increasing expression of TCA cycle associated enzymes was accompanied by a decrease in the intermediates during grape berry ripening (phase III). The discrepancy between the increase in the abundance of the enzymes and the decrease in the content of the intermediates during grape berry ripening indicates a high metabolic flux through this pathway with an efficient incorporation of the intermediates in the synthesis of amino acids, lipids and secondary metabolites. In grape, organic acids are responsible for the titratable acidity which is an index for fruit quality. High amounts of organic acids endow young berries a sour taste for defense against herbivores. The organic acids in mature berries are essential for wine production as they protect the fermentation process from bacterial contamination. In wine they are responsible for the sour part of the taste. They are also essential for the color of wine by contributing to the stabilization of anthocyanins (Clemente and Galli, 2011). The changing patterns of those dominant organic acids i.e., malic acid, tartaric acid and citric acid, were consistent with previous reports (Deluc et al., 2007; Ali K. et al., 2011; Muñoz-Robredo et al., 2011; Fraige et al., 2015).

#### Amino Acids Metabolism

Amino acids are major transportable nitrogenous compounds in grape. In source organs, intermediates from glycolysis and TCA cycle can be utilized as precursors for the synthesis of amino acids. For instance, phosphoenolpyruvate is the precursor of aromatic amino acids that derive from the shikimate pathway; α-ketoglutarate and oxaloacetate are the precursors of glutamate and aspartate family amino acids, respectively. Asparagine and glutamine were the major amino acids in young and mature grape berries respectively. Both of them carry an extra amide group making them efficient nitrogen-carriers. They play important roles in the nitrogen assimilation, transportation and storage in plants. Asparagine and glutamine can be converted to Asp and Glu to serve as precursors for biosynthesis of many other amino acids, e.g., proline, arginine (from glutamate); methionine, threonine and lysine (from aspartate). Several enzymes involved in amino acid metabolism were detected and mapped on **Figure 5**. Aspartate aminotransferase (AspAT) catalyzes the reversible transfer of an amino group between aspartate and glutamate thus plays an important role in nitrogen distribution. In concordance with a previous report (Martinez-Esteso et al., 2011), AspAT abundancy gradually decreased during grape berry development. Methionine synthase (MetH), catalyzing the synthesis of methionine, is another essential amino acid of the aspartate family. The abundance of MetH increased until veraison, then stayed relatively constant during ripening in the present study. However, it decreased during green developmental stages and increased during ripening in the study of Martinez-Esteso et al. (2011). Glutamine synthetase (GS) is another crucial enzyme involved in nitrogen assimilation. GS, which catalyzes the condensation of glutamate and ammonia to generate glutamine, gradually declined in abundance from EL 27 to EL 36 and slightly increased afterwards. The transcript level of GS was shown to be significantly higher in phase I in a previous study (Deluc et al., 2007). Marinez-Esteso et al reported a decline in the level of GS before veraison (Martinez-Esteso et al., 2011) which is consistent with our result. However, the changing trend of GS after veraison was absent in their study.

#### Secondary Metabolism Dynamics

Grape is rich in polyphenolic compounds that derive from the phenylpropanoid pathway. These bioactive secondary metabolites play essential roles in protecting grape berry against abiotic stresses such as UV radiation (Pontin et al., 2010), high light and high temperature stresses (Ayenew et al., 2015), drought stress (Król et al., 2014) as well as biotic stresses (Gutha et al., 2010; Wallis and Chen, 2012). In addition, they also contribute to organoleptic features of grape berry and wine (Schmidtke et al., 2010; Gutiérrez-Capitán et al., 2014).

#### Lignin

Simple phenolic compounds that are synthesized from phenylalanine can be polymerized to lignin which is an essential component of the cell wall. In the present study, three enzymes, i.e., cinnamoyl-CoA reductase (CCR), cinnamylalcohol dehydrogenase (CAD) and ferulate 5-hydroxylase (F5H, Q9M4H8) involved in lignin synthesis were annotated. CCR and CAD, which catalyze the last two steps of monolignol synthesis not only impact lignification but also plant development. Absence of CCR and CAD resulted in dwarfism and sterility in Arabidopsis (Thevenin et al., 2011). One protein candidate was annotated as CCR (A5AXM6) and detected after veraison. Four candidates were annotated as CAD. The averaged expression pattern of CAD showed high levels in both young and ripening berries and low level around veraison (**Figure 5**). Aharoni et al. (2002) reported a comparable pattern of the transcription level of CAD in developing strawberries which are also non-climacteric fruits and undergo color turning phases. In their study, CAD expression level was high in green strawberries followed by a decreasing during the white and turning stages and finally increased again in red strawberries (Aharoni et al., 2002).

#### Flavonoids

Flavonoids are another class of secondary metabolites derived from the phenylpropanoid pathway and share common precursors with lignin. The detected flavonols and flavan-3-ols showed distinct changing patterns with anthocyanins. Parallel phenomena were observed before in other varieties, i.e., Cabernet Sauvigon (Ali M. B. et al., 2011; Degu et al., 2014), Shiraz (Degu et al., 2014) and Norton (V. aestivalis) (Ali M. B. et al., 2011) and supposed to be caused by the competition of precursors between anthocyanins and other subfamily members of flavonoids (Ali K. et al., 2011; Degu et al., 2014). The anthocyanin profile obtained from the present data was consistent with that of "Concord" (Liang et al., 2011) and "Pink Sultana" (Boss et al., 1996) in which Cy and Dp derivatives were the dominant anthocyanins. In contrast, Mv derivatives were the most abundant anthocyanins in the other varieties in these two studies (Boss et al., 1996; Liang et al., 2011) and other reports (an overview of all varieties is provided in **Table S8**) (Mazza et al., 1999; He et al., 2010; Papini et al., 2010; Ali M. B. et al., 2011; Degu et al., 2014, 2015). Noticeably, Early Campbell is a hybrid of Vitis vinifera and Vitis labrusca by crossing Moore Early with (Belvidere × Muscat of Hamburg). Both Moore Early and Belvidere are seedlings of Concord. The fruit taste and disease resistance of Early Campbell are similar to Concord (Robinson et al., 2012). The similar anthocyanin profile of Early Campbell with that of Concord is probably also due to its genetic background. The annotated proteins that were involved in the flavonoid pathway include chalcone synthase (CHS), chalcone isomerase (CHI), flavanone 3-hydroxylase (F3H), dihydroflavanol 4-reductase (DFR) and anthocyanidin synthase (ANS). CHS catalyzes the condensation of p-coumaroyl-CoA with malonyl-CoA to generate chalcone which is further isomerized to naringenin by CHI. Two proteins, annotated as CHS (A2ICC5, G4XGW2), were detected after veraison and increased in abundance during ripening (**Table S4**). In a previous study, three copies of Chss were found in the grape genome. The mRNA levels of Chs2 and Chs3 significantly coincided with anthocyanin and Chs1 and Chs2 with flavonol biosynthesis (Jeong et al., 2008). The expression of A2ICC5 and G4XGW2 were in accordance with anthocyanin accumulation indicating the involvement of these two CHSs in the coordination of anthocyanin synthesis. The expression level of CHI (A5ANT9) gradually increased before veraison then underwent a sharp decline at veraison (EL 35) with a subsequent recovery to the level before veraison. F3H catalyzes the hydroxylation of flavanones at 3-position to form dihydroflavonols. Both protein candidates annotated as F3H (A2ICC8, A2ICC8) were detected from EL 32 and showed increasing levels during grape berry development. The transcription of F3hs appeared to be induced for the biosynthesis of flavonols and anthocyanins (Jeong et al., 2008). Dihydroflavonols are further converted to either flavonols via flavonol synthase (FLS) or to leucoanthocyanidins via DFR. The competition between FLS and DFR influences the contents of flavonols and anthocyanins (Tian et al., 2015) which further affects the color (Lou et al., 2014) and the abilities of plants to cope with stress (Hua et al., 2013; Wang et al., 2013). DFR is a crucial enzyme in the flavonoid pathway involved in the synthesis of anthocyanins, proanthocyanins and tannins (Moyano et al., 1998; Zhang et al., 2008; Hua et al., 2013; Wang et al., 2016c). In the present study, two protein candidates (A5BGJ0, A5BIY8) were annotated as DFR. Their combined expression pattern indicated a progressive increase in abundance of DFR before veraison and a slight decrease afterwards (**Figure 5**). ANS catalyzing the conversion of colorless leucoanthocyanins to colored anthocyanidins was detected in ripening berries. The abundance of ANS (A2ICC9) significantly increased after veraison (**Figure 5**) in accordance with the anthocyanin accumulation. MYBrelated transcription factors (TFs) are involved in regulation of flavonoid synthesis (Czemmel et al., 2009). It was further reported that VvMYB5b was highly expressed after veraison and the anthocyanin synthesis was enhanced in transgenic tobacco due to ectopic expression of VvMYB5b (Deluc et al., 2008). In our study, four proteins i.e., A5ADL7, A5AHA8, F6GTT4, F6I581 were annotated as "MYB-related transcription factor". Their summarized content was lowest around veraison whereas the highest level was observed before veraison and during ripening (**Table S4** sheet 2). This pattern was neither directly correlated to the amount of anthocyanins in the developing grape berries nor to the protein levels of ANS (A2ICC9). Further we detected 7 bZIP family members (A5B427, A5BZF5, D7SUP9, D7TNE5, F6GTA6, F6GUN1, F6HBQ3). bZIP family member are thought to be involved in the regulation of flavonoid biosynthesis (Malacarne et al., 2016). In a recent study Loyola et al. propose that HY5 and HYH are involved in UV-Bdependent flavonol accumulation in grapevine (Loyola et al., 2016). The concentrations of the bZIP proteins in our study showed an increase toward veraison and a decrease afterwards. Because there is often not a direct dependency between transcriptional and translational/posttranslational control (Nukarinen et al., 2016) it is difficult to compare gene expression levels from other studies with protein levels from our study. Furthermore, the analysis of TFs requires in most cases a specific enrichment step before proteomic analysis. Future investigations will focus more on the discussed transcription factors and their control on developmental processes and flavonoid biosynthesis.

#### Stilbenes

Stilbenes are also derived from the phenylpropanoid pathway and were reported to be enriched in grape. However, we did not detect any stilbene or related enzymes in this study. This might be due to different growth and stress conditions or a different genetic background of the variety we studied. Two publications reported stilbene content in the peel of Early Campbell at veraison stage (Islam et al., 2014; Ahn et al., 2015). They also found that hairy vetch and ryegrass extracts and red and blue LED light induced stilbene accumulation as well as the expression of genes involved in stilbene synthesis. We used the whole berry as the study object which is different from using isolated peel. Furthermore there is evidence that support a competition between the synthesis of stilbenes and flavonoids. One evidence is the negative correlation between resveratrol and anthocyanin accumulation in 5 Vitis species at different developmental species (Jeandet et al., 1995). The other evidence is observed in transgenic strawberries. Hanhineva et al transformed a stilbene synthase gene to strawberries (35S:NS-Vitis3 line). While the STS gene was highly expressed in the transgenic strawberry line, CHS expression was down regulated (Hanhineva et al., 2009). These are two examples indicating the competition relationship between flavonoid and stilbene synthesis pathway. It is thus of interest to compare the flavonoid and stilbene content of this variety under different growth conditions and with other grape varieties that produce stilbenes to investigate the competition of stilbene and flavonoid biosynthesis.

# CONCLUSION

In summary, the analysis of grape berry development from fruit set to mature fruit by mass spectrometry based platforms revealed intimate correlations between the metabolome and the proteome at the interface of primary and secondary metabolism. The broad coverage of developmental stages included in the present study enabled a dense correlation network analysis of these dynamic processes covering central carbon metabolism such as sugar metabolism, glycolysis, TCA cycle, amino acid metabolism as well as secondary metabolism, especially the flavonoid pathway. Multivariate statistical analysis such as PCA, clustering analysis and Granger causality analysis provides a convenient data mining approach for the interpretation of the integrated metabolome and proteome dataset and revealed the systemic associations between metabolites and proteins during grape berry development. The application of Granger causality analysis is helpful in revealing time-lagged correlations between metabolites and proteins which is especially important for understanding the molecular timeshifts during developmental processes of grape berries. Together with other studies this work provides a reference point for future investigations of grape berry development in a variety of different genotypes.

#### AUTHOR CONTRIBUTIONS

LW and WW conceived and designed the experiments. LW performed the experiments. LW, XS, JW, and WW analyzed the data. WW provided the reagents, materials and analytical tools. LW wrote the manuscript. WW and JW revised the manuscript. All the authors approved the final manuscript.

#### FUNDING

LW was supported by a Ph.D. scholarship provided by China Scholarship Council (CSC) (Grant number: 201206220134).

#### ACKNOWLEDGMENTS

We would like to thank the gardeners for their great maintenance of the grapevine. We would like to thank all the MoSys members for fruitful discussions. We would like to thank Reinhard Turetschek for suggestions in protein blast analysis.

#### REFERENCES


#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2017. 01066/full#supplementary-material

Table S1 | Metabolite information.

Table S2 | Absolute quantification of 29 metabolite candidates.

Table S3 | Protein sequencing information.

Table S4 | NSAFs, ANOVA, BLAST results of protein candidates.

Table S5 | Loading scores.

Table S6 | K-means clustering analysis (12 clusters).

Table S7 | Granger causality analysis of the integrated metabolomic and proteomic data sets.

Table S8 | Anthocyanin profiles of grape berries of different cultivars.

Figure S1 | Relative abundance of anthocyanins in mature berries (EL 38).

Figure S2 | Hierachical bi-clustering analysis of protein candidates.

Figure S3 | K-means clustering analysis of the integrated dataset.

Figure S4 | Granger causality based network.

sports differing in berry skin colour. Aust. J. Grape Wine R. 2, 163–170. doi: 10.1111/j.1755-0238.1996.tb00104.x


Conde, C., Silva, P., Fontes, N., Dias, A. C. P., Tavares, R. M., Sousa, M. J., et al. (2007). Biochemical changes throughout grape berry development and fruit and wine quality. Food 1, 1–22.


pathway in grapevine and mediate flavonol accumulation in response to the environment. J. Exp. Bot. 67, 5429–5445. doi: 10.1093/jxb/erw307


cultivars (Vitis vinifera) in which anthocyanin synthesis is sunlight-dependent or -independent. PLoS ONE 9:e105959. doi: 10.1371/journal.pone.0105959


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Wang, Sun, Weiszmann and Weckwerth. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Drawing Links from Transcriptome to Metabolites: The Evolution of Aroma in the Ripening Berry of Moscato Bianco (Vitis vinifera L.)

Laura Costantini <sup>1</sup> \*, Christian D. Kappel <sup>2</sup> , Massimiliano Trenti <sup>1</sup> , Juri Battilana<sup>1</sup> , Francesco Emanuelli <sup>1</sup> , Maddalena Sordo<sup>1</sup> , Marco Moretto<sup>3</sup> , Céline Camps <sup>2</sup> , Roberto Larcher <sup>4</sup> , Serge Delrot <sup>2</sup> and Maria S. Grando1, 5

<sup>1</sup> Grapevine Genetics and Breeding Unit, Genomics and Biology of Fruit Crop Department, Research and Innovation Centre, Fondazione Edmund Mach, San Michele all'Adige, Italy, <sup>2</sup> UMR Ecophysiology and Grape Functional Genomics, Institut des Sciences de la Vigne et du Vin, University of Bordeaux, Villenave d'Ornon, France, <sup>3</sup> Computational Biology Platform, Research and Innovation Centre, Fondazione Edmund Mach, San Michele all'Adige, Italy, <sup>4</sup> Experiment and Technological Services Department, Technology Transfer Centre, Fondazione Edmund Mach, San Michele all'Adige, Italy, <sup>5</sup> Center Agriculture Food Environment, University of Trento, San Michele all'Adige, Italy

#### Edited by:

José Tomás Matus, Centre for Research in Agricultural Genomics, Spain

#### Reviewed by:

Philip Richard Young, Institute for Wine Biotechnology, South Africa Patricio Hinrichsen, Instituto de Investigaciones Agropecuarias, Chile

> \*Correspondence: Laura Costantini laura.costantini@fmach.it

#### Specialty section:

This article was submitted to Crop Science and Horticulture, a section of the journal Frontiers in Plant Science

Received: 10 February 2017 Accepted: 25 April 2017 Published: 16 May 2017

#### Citation:

Costantini L, Kappel CD, Trenti M, Battilana J, Emanuelli F, Sordo M, Moretto M, Camps C, Larcher R, Delrot S and Grando MS (2017) Drawing Links from Transcriptome to Metabolites: The Evolution of Aroma in the Ripening Berry of Moscato Bianco (Vitis vinifera L.). Front. Plant Sci. 8:780. doi: 10.3389/fpls.2017.00780 Monoterpenes confer typical floral notes to "Muscat" grapevine varieties and, to a lesser extent, to other aromatic non-Muscat varieties. Previous studies have led to the identification and functional characterization of some enzymes and genes in this pathway. However, the underlying genetic map is still far from being complete. For example, the specific steps of monoterpene metabolism and its regulation are largely unknown. With the aim of identifying new candidates for the missing links, we applied an integrative functional genomics approach based on the targeted metabolic and genome-wide transcript profiling of Moscato Bianco ripening berries. In particular, gas chromatography-mass spectrometry analysis of free and bound terpenoid compounds was combined with microarray analysis in the skins of berries collected at five developmental stages from pre-veraison to over-ripening. Differentially expressed metabolites and probes were identified in the pairwise comparison between time points by using the early stage as a reference. Metabolic and transcriptomic data were integrated through pairwise correlation and clustering approaches to discover genes linked with particular metabolites or groups of metabolites. These candidate transcripts were further checked for co-localization with quantitative trait loci (QTLs) affecting aromatic compounds. Our findings provide insights into the biological networks of grapevine secondary metabolism, both at the catalytic and regulatory levels. Examples include a nudix hydrolase as component of a terpene synthase-independent pathway for monoterpene biosynthesis, genes potentially involved in monoterpene metabolism (cytochrome P450 hydroxylases, epoxide hydrolases, glucosyltransferases), transport (vesicle-associated proteins, ABCG transporters, glutathione S-transferases, amino acid permeases), and transcriptional control (transcription factors of the ERF, MYB and NAC families, intermediates in light- and circadian cycle-mediated regulation with supporting evidence from the literature and additional regulatory genes with a previously unreported association to monoterpene accumulation).

Keywords: grapevine, Muscat, monoterpene, development, berry skin, metabolic and transcript profiling, integration, candidate gene

# INTRODUCTION

A great deal of the consumer interest in wine derives from its aroma characteristics. The major aroma-impact compounds in grape and wine are terpenoids (monoterpenes, sesquiterpenes, and in a wider acception also C13-norisoprenoids), phenylpropanoids/benzenoids, fatty acid derivatives, sulfur compounds, and methoxypyrazines (Dunlevy et al., 2009; Ebeler and Thorngate, 2009; Panighel and Flamini, 2014; Robinson et al., 2014; Black et al., 2015). The typical floral and citrus attributes of Muscat varieties are primarily determined by a combination of linalool, geraniol and nerol (Ribéreau-Gayon et al., 2000). The same monoterpenes contribute to the varietal aroma of Riesling in association with the linalool oxides, hydroxy-linalool, α-terpineol, citronellol, terpendiol I and hydroxy-trienol (Rapp, 1998). Likewise rose oxide, which is highly correlated with Muscat score in grapes (Ruiz-García et al., 2014), is also a potent odorant in Scheurebe and Gewürztraminer (Guth, 1997; Ong and Acree, 1999; Luan et al., 2005).

The terpene biosynthetic pathway is generally well known (Dudareva et al., 2013), even though a number of alternative non-canonical reactions may occur (Sun et al., 2016). Of the two systems responsible for the production of plant isopentenyl diphosphate (IPP) and dimethylallyl diphosphate (DMAPP), the primarily cytosolic mevalonic acid (MVA) and the plastidial methylerythritol phosphate (MEP) pathway, the latter has been suggested as the dominant route for monoterpene biosynthesis in grape berries (Luan and Wüst, 2002). Several lines of evidence (Battilana et al., 2009, 2011; Duchêne et al., 2009; Emanuelli et al., 2010; Martin et al., 2012; Wen et al., 2015) support the existence of at least two rate-limiting enzymes in the grapevine MEP pathway, namely the first (1-deoxy-Dxylulose 5-phosphate synthase, VvDXS1) and the last (4-hydroxy-3-methylbut-2-enyl diphosphate reductase, VvHDR). Both IPP and DMAPP are substrates for short-chain prenyltransferases, which produce prenyl diphosphate precursors for the large family of terpene synthases (TPSs). To date around 40 fulllength VvTPSs out of 53–89 predicted functional enzymes have been biochemically characterized (Martin et al., 2010) and some major players in grape Muscat aroma have been identified, like the α-terpineol synthase VvTer, the linalool synthase Lis, the linalool/nerolidol synthase VvPNLinNer1 and the geraniol synthase VvPNGer (Ebang-Oke et al., 2003; Martin and Bohlmann, 2004; Martin et al., 2012; Matarese et al., 2013; Zhu et al., 2014; Wen et al., 2015). Once a terpenoid alcohol skeleton has been produced, extensive modifications determine the final monoterpene composition of grapes and wines (Ribéreau-Gayon et al., 1975; Williams et al., 1989; Luan et al., 2004, 2005, 2006a,b; Mathieu et al., 2009). These secondary transformations are (at least in part) catalyzed by enzymes (Luan et al., 2006a; D'Onofrio et al., 2016) that in most cases have not been identified. The only exceptions are the three grape monoterpenol β-D-glucosyltransferases VvGT7, VvGT14 and VvGT15 and the cytochrome P450 CYP76F14 (Bönisch et al., 2014a,b; Ilc et al., 2017). The main reason for this gap is that such enzymes belong to large families with broad substrate tolerance and overlapping activities (Schwab, 2003; Nelson et al., 2008; Schwab and Wüst, 2015). A better knowledge of the missing enzymes might allow us to manipulate the formation of grape aroma compounds. For example, limiting the reactions responsible for the depletion of key odorants (e.g., through the selection of genotypes with low monoterpene glycosyltransferase or oxygenase activities in breeding programs) could be an alternative approach for the improvement of grape/wine flavor (Bönisch et al., 2014a; Hjelmeland and Ebeler, 2015).

The grapevine terpenoid pathway is intricately regulated by endogenous and environmental factors that enable spatially and temporally controlled metabolite production (Ebeler and Thorngate, 2009; Robinson et al., 2014). In other plant species a network of transcription factors (TFs) is involved in the regulation of this pathway, including members of the AP2, AP2/ERF, bZIP, MYB, MYC, NAC, WRKY, and YABBY families (De Geyter et al., 2012; Patra et al., 2013; Nieuwenhuizen et al., 2015; Wang et al., 2016). A tight regulation of terpene biosynthesis is additionally exerted at the post-transcriptional level involving both structural enzymes and transcription factors (Vom Endt et al., 2002; Hemmerlin, 2013; Rodríguez-Concepción and Boronat, 2015), as observed also in Vitis vinifera (Bönisch et al., 2014a; Matarese et al., 2014). A number of transcription factors that might control terpene synthesis have been recently predicted in grapevine through gene co-expression network analysis (Wen et al., 2015), though none of them has been yet demonstrated to regulate the expression of relevant terpene pathway genes. Similarly, the reasons of the differential accumulation of the main monoterpenes in grape berry tissues across development (Günata et al., 1985; Wilson et al., 1986; Park et al., 1991; Luan and Wüst, 2002), which is reflected in the identification of specific QTLs for linalool and geraniol/nerol (Doligez et al., 2006; Battilana et al., 2009), are still unknown.

This work aims at a better understanding of aroma determination in grapevine and at the identification of candidate genes for further functional analysis. To this purpose, we integrated gas chromatography/mass spectrometrybased quantitative analysis of selected metabolites with microarray-based transcriptomic analysis in Moscato Bianco (Vitis vinifera L.) ripening berries. According to the observed associations between metabolite and transcript profiles, we report several genes that may control the accumulation of free and glycosidically bound monoterpenes and additional aroma-related compounds.

# MATERIALS AND METHODS

#### Plant Material For Metabolic and Microarray Analyses

Berries of the cultivar Moscato Bianco (Vitis vinifera L.) were collected from pre-veraison to over-ripening in 2005, 2006, and 2007 (**Figure 1A** and **Table 1**). At each sampling date, ten bunches were taken from ten adult plants out of the ∼ 250 grown on Kober 5BB rootstocks in the experimental fields of FEM (Fondazione Edmund Mach, San Michele all'Adige, Italy). Care was taken to sample from different vines and positions within each vine. In the lab, berries were pooled in order

FIGURE 1 | Acidity, sugars (A) and monoterpene content (B–F) of the Moscato Bianco samples collected during berry development in 2005, 2006, and 2007. The five stages assayed by microarray analysis in 2006 are highlighted with red in (A). Exemplar monoterpenes with a major contribution to the total free monoterpene profile (B) during PV, R1, and R2 are shown in (D–F), respectively. A single biological replicate was considered at each stage in each season; bars in (D–F) correspond to the standard error calculated from six technical replicates, as described in Supplementary Data1: Method S1. The metabolites were quantified by using solid SPE-HRGC-MS and referring to the internal standard 1-heptanol. The lines connecting data points were smoothed through the specific option provided by Excel. E-L stage, growth stage according to the modified Eichhorn-Lorenz scheme (Coombe, 1995); PV, pre-veraison; V, veraison; R1, ripening (till technological maturity or stage E-L 38); R2, over-ripening (after technological maturity). The decimal E-L stages were arbitrarily assigned by the authors of the present study to facilitate the alignment of the sampling dates from the three different seasons.



The five stages used for microarray analysis are in boldface. E-L stage, growth stage according to the modified Eichhorn-Lorenz scheme (Coombe, 1995). The decimal E-L stages were arbitrarily assigned by the authors of the present study to facilitate the alignment of sampling dates from three different seasons (Figure 1). The weeks from veraison were established with a maximum tolerance of 2 days from the exact date.

to minimize environmental effects and then divided into two (2005 and 2007) or three (2006) batches. Berries from the first batch were homogenized to juice (80 mL) and analyzed for titratable acidity and soluble solids content by FT-IR (Fourier Transform InfraRed) spectroscopy with a FOSS instrument (FOSS NIRSystems, Oatley, Australia). Berries from the second batch were stored at −80◦C till metabolic analysis. Berries from

the third batch were hand-peeled, the skins were immediately frozen in liquid nitrogen and stored at −80◦C pending RNA extraction.

#### Metabolic Analysis

Thirty-two aroma-active compounds were quantified in their free and glycosidically bound form by using solid phase extraction (SPE) and high-resolution gas chromatography-mass spectrometry (HRGC-MS; Supplementary Data1: Method S1 and Supplementary Table S1) in the growing seasons 2005, 2006, and 2007.

Network analysis for 2006 metabolic data included pairwise correlation, hierarchical clustering with bootstrapping (Pvclust with 10,000 resamplings, see Suzuki and Shimodaira, 2006) and principal component analysis (PCA) and was applied to different data sets (free and glycosidically bound metabolites, absolute quantities and differentials, 5 and 13 time points).

#### Microarray Analysis

Based on monoterpene accumulation during berry development in 2006 (**Figure 2**), five time points were chosen along this season (**Figure 1** and **Table 1**). Total RNA was extracted from grape skins using the SpectrumTM Plant Total RNA Kit (Sigma-Aldrich, St. Louis, Missouri, USA). RNA quantity and quality were evaluated with a NanoDrop ND-8000 spectrophotometer (NanoDrop Technologies, Wilmington, Delaware, USA) and an Agilent 2100 Bioanalyzer (Agilent Technologies, Mississauga, Ontario, Canada).

Microarray experiments were carried out with a 70-mer oligoarray containing all 14,562 probes from the Array-Ready Oligo SetTM (AROS) for the Grape (Vitis vinifera) Genome version 1.0 (Operon Biotechnologies, Huntsville, Alabama, USA). At the time this platform represented a good compromise between genome coverage, cost and computational effort required for data analysis. RNA from points 2 to 5 was hybridized competitively with RNA from point 1 (pre-veraison), following the dye-swap experiment design (Churchill, 2002). A total of sixteen slides were used (four comparisons: 2 vs. 1, 3 vs. 1, 4 vs. 1, 5 vs. 1; two biological and two technical replicates). The biological and technical replicates corresponded to two subgroups from the unique pool of berries (third batch) and to the dye swaps, respectively. Details for probe synthesis,

FIGURE 2 | Evolution of monoterpenoids in their free (solid red line) and glycosidically bound (dashed blue line) form during Moscato Bianco berry ripening in 2006. A single biological replicate was considered at each stage; bars correspond to the standard error calculated from six technical replicates, as described in Supplementary Data1: Method S1 (technical replication is not available for 7-hydroxy-nerol, 7-hydroxy-citronellol, 4-terpineol, rose oxide I and II). The metabolites were quantified by using solid SPE-HRGC-MS and referring to the internal standard 1-heptanol. The lines connecting data points were smoothed through the specific option provided by Excel.

hybridization and scanning are described in Supplementary Data1: Method S2.

Spot intensities were quantified with the software MAIA 2.75 (Novikov and Barillot, 2007). After excluding poor quality spots due to bad spotting (e.g., spots with irregular shapes or highly unequal intensity distributions), median intensity gene expression data without background subtraction were normalized by a global lowess method followed by a print-tip median method with a modified version of the Goulphar script version 1.1.2 (Lemoine et al., 2006). Differentially expressed probes (DEPs) with a false discovery rate (FDR) < 1% and a cutoff of 2-fold change (FC) were identified with the R/Bioconductor Limma package using linear models (Smyth, 2004) and taking into account biological and technical replicates by doing a twofactor analysis. The earliest sample was used as the reference to whom all the other samples were compared. A multiple testing correction (Benjamini and Hochberg, 1995) was applied to adjust the FDR. The full raw expression dataset is available at the Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo/) under the accession number GSE76834.

#### Probe Functional Annotation

The 70-mer probes spotted on the Grape AROS V1.0 array represent 14,562 transcripts from The Institute for Genomic Research (TIGR) Grape Gene Index (VvGI), release 3 (August 13, 2003). The corresponding annotation is based upon a match between each oligo and the gene set of the 12X version of the grape genome at CRIBI (http://genomes.cribi.unipd.it/grape/) and is publicly available at GEO (http://www.ncbi.nlm.nih.gov/geo/) under the accession number GPL15453. Since this annotation provides every oligo with a text description but doesn't associate it to any gene prediction identifier, we independently achieved this information by blastN alignment against the grape gene sets at CRIBI (http://genomes.cribi.unipd.it/grape/, 12X version of the genome, V1 gene prediction, annotation from Grimplet et al., 2012) and IASMA (Velasco et al., 2007), as fully detailed in Supplementary Table S2. For the alignment against the CRIBI gene set, the following parameters were used: sequence identity ≥90%, minimum alignment length of 95%, maximum number of mismatches of 5 and maximum number of gaps of 5. The aligned 70-mers were found to correspond to 7,162 and 8,260 unique gene predictions at CRIBI and IASMA, respectively. The 14,562 probes were also grouped into main functional categories according to the Mapman BIN structure (Rotter et al., 2009; Supplementary Table S2).

#### Microarray Validation via Real-Time PCR

Real-time quantitative reverse transcription-PCR (qRT-PCR) was used to validate the microarray results. Since RNA from berries collected in 2006 was no more available, new samples (with three biological replicates from pooled berries) as closest as possible to those analyzed in 2006 were obtained in 2016 by adopting the same sampling procedure and the same protocol for RNA extraction.

Primers for the amplification of unique PCR products from 70 to 250 bp were designed on 15 Vitis vinifera gene predictions perfectly matching with the microarray 70-mers by using Primer3 (Untergasser et al., 2012), as reported in Supplementary Table S3. Details for the amplification reaction and expression analysis are described in Supplementary Data1: Method S3. The relationship between microarray and qRT-PCR data was established through Pearson correlation.

#### Discovery of a Link between Transcriptome and Metabolome in Aroma Development Integration of 2006 Transcriptomic and Metabolic Data

Different approaches were tested in order to discover transcripts linked with the accumulation of one or more metabolites. In a first identification step of candidates, the most stable expression changes were preferred to the biggest ones, hence the microarray probes with adjusted p-values < 0.05 in all comparisons were considered, irrespective of their fold change (4,450). Working at probe level instead of gene level was chosen for two main reasons: (1) different probes supposedly matching to the same gene (especially long genes) often show different expression values, which might be an indicator of alternative transcription and (2) the sequence specificity to CRIBI 12X gene predictions is not optimal for a number of probes spotted on the AROS array (this is especially true for probes related to secondary metabolism); for a detailed assessment of probe specificity, Moscato Bianco genome and transcriptome assembly would be required, which is out of the scope of this work.

#### **Pairwise correlation**

Pearson pairwise correlation was calculated between transcripts and metabolites across all the time points (log2-transformed differentials in the pairwise comparisons 2 vs. 1, 3 vs. 1, 4 vs. 1, and 5 vs. 1). With the goal of identifying aroma regulatory genes we also tested a two-step strategy, which was based on (1) search for candidate metabolism and transport genes by direct correlation with metabolites and (2) expression pairwise correlations between these enzyme/transporter-coding genes and any regulatory gene within the microarray. For this aim, Pearson correlations were computed both between differential gene expression ratios (n = 4) and microarray channel intensities (n = 32, when considering technical replicates separately).

#### **Correlation biclustering**

Based on the assumption that a gene might regulate the accumulation of a metabolite only at specific stages during ripening, correlation biclustering between transcripts and metabolites was achieved with QUBIC (Li et al., 2009) (log2 transformed data, Pearson correlation). Compared to the traditional clustering methods, biclustering algorithms discover local co-expression patterns (groups of genes/metabolites that show similar patterns under a specific subset of the experimental conditions) (Madeira and Oliveira, 2004). We manually inspected our biclustering data only in a few exemplar cases.

#### **Soft clustering**

Soft clustering of the metabolite and transcript differentials was performed by using the R/Bioconductor Mfuzz package (Kumar and Futschik, 2007) with the default value 1.25 for the fuzzy parameter m. A membership value in the range of 0–1 was assigned to each metabolite and probe. Soft clustering offers several advantages with respect to hard clustering; in particular, it has been suggested to be more suitable for time course microarray data in which expression patterns are often not well separated (Futschik and Carlisle, 2005; Kumar and Futschik, 2007). The biological significance of the clusters was analyzed by enrichment analysis of the MapMan functional categories assigned to the probes in each cluster. Specifically, Chi square and Fisher statistical tests were employed to search for significant differences (p-value < 0.05) between the observed number of probes within each MapMan functional category per cluster and the expected number of probes in that category based on the overall AROS genome array expression distribution.

#### **Selection of candidate genes**

From the whole set of transcripts with a potential association to monoterpenes (based on their correlation with metabolites and membership to soft clusters/biclusters harboring metabolites) we selected a subset of genes with significant expression changes and/or supporting evidence from the literature, like a relevant function in other plant species, co-localization with QTLs for monoterpene content and coexpression with genes involved in the terpene pathway. In particular, the QTL co-localization was stated when the V1 gene predictions fell into the 1-LOD confidence intervals of the QTLs for linalool, geraniol and nerol reported by Doligez et al. (2006) and Battilana et al. (2009) based on the analysis of different segregating progenies in 2– 3 seasons (depending on the progeny). The genomic region corresponding to each QTL confidence interval was determined from the physical position of the two neighboring markers, while the V1 gene prediction physical position was retrieved from Grimplet et al. (2012).

#### Integration of Transcriptomic and Metabolic Data Over Multiple Seasons to Verify a Subset of Candidate Genes

For the candidate genes assessed by both microarray and real-time PCR analyses (in 2006 and 2016, respectively), the association between expression and metabolic profiles was further tested by employing a general monoterpene quantification that considers the three seasons (2005, 2006, and 2007) as replicates. To this purpose, the average concentration among these seasons was computed for each metabolite at each developmental stage. Pearson and Spearman correlations were calculated between the transcriptional and metabolic data expressed as log<sup>2</sup> fold changes at the stages 2-5 (E-L 34.5, 36, 38, and 39) with respect to the first stage (E-L 31).

#### RESULTS AND DISCUSSION

Our study gives an example of the systems biology approach. Systems biology has been successfully applied to the discovery of regulatory and biosynthetic genes involved in the control of metabolite production (Yuan et al., 2008; Liberman et al., 2012), including examples from grape (Zamboni et al., 2010; Fortes et al., 2011; Agudelo-Romero et al., 2013; Costantini et al., 2015; Malacarne et al., 2015; Suzuki et al., 2015; Wen et al., 2015; Savoi et al., 2016).

#### Metabolic Analysis

The present work provides a temporal profiling of aromatic compounds in the Moscato Bianco ripening berry. The protocol used for the chemical analysis was optimized for molecules belonging to the monoterpenoid class, however it allowed the simultaneous quantification of additional metabolites. In particular, the content of 21 monoterpenoids, 3 C13 norisoprenoids, 5 phenylpropanoids/benzenoids and 3 C<sup>6</sup> aliphatic compounds was quantified from pre-veraison to overripening in 2005, 2006, and 2007. For several compounds a coherent accumulation trend was observed in the different years (**Figure 1** and Supplementary Figure S1A). The most significant correlation between seasons was observed for free linalool, nerol, α-terpineol, hydroxy-diendiol I + hydroxy-trienol, hydroxydiendiol II, hexanol, cis-3-hexen-1-ol, bound linalool, geraniol, nerol, trans-furan linalool oxide and benzyl alcohol. Other metabolites appeared instead to be more sensitive to seasonal effects, like rainfall and temperature (Supplementary Figure S1B). Hereafter, we will refer to 2006, which is the year assayed by microarray analysis.

The most abundant metabolites were monoterpenes (hydroxy-diendiol I, trans-geranic acid, linalool, geraniol and nerol), with concentrations higher than 600 µg/kg of berries (**Figure 2** and Supplementary Figure S2). The majority of monoterpenoids reached the highest amount in their glycosidically bound form. The main exceptions are represented by high oxidation state monoterpenes, like the two pyran linalool oxides, the two diendiols and rose oxide I. A clear prevalence of the free form was also observed for the C<sup>6</sup> aliphatic compounds, while the most abundant C13-norisoprenoids and phenylpropanoids/benzenoids were glycosidically bound, in agreement with previous analyses (Sánchez Palomo et al., 2006; D'Onofrio et al., 2016).

The quantity of many metabolites was significantly (at least 2-fold) altered during ripening. The compounds that changed most with respect to pre-veraison were linalool, geraniol, nerol, cis/trans-8-hydroxy-linalool, hydroxy-diendiol I and II in both forms; cis-furan linalool oxide, trans-geranic acid, 7 hydroxy-geraniol, 7-hydroxy-citronellol, hydroxy-trienol in their free form; trans-furan linalool oxide, 3-oxo-α-ionol, methyl salicylate, hexanol and cis-3-hexen-1-ol in their bound form (Supplementary Figures S3A,D, Supplementary Table S4).

The pattern of accumulation along berry development varied with the metabolite (**Figure 2** and Supplementary Figure S2). The concentration of the three compounds mainly contributing to Muscat aroma (linalool, geraniol, and nerol) was from low to moderate before veraison (August 17 or stage E-L 34.5 in this work) and then increased during ripening. Free linalool reached its maximum on September 13 (technological maturity or stage E-L 38) and decreased during over-ripening. A similar behavior was observed in 2005 and 2007, even though the peak corresponded to slightly earlier stages (**Figure 1E**), proving that technological and aroma ripening might not occur at the same time (Vilanova et al., 2012). Otherwise, free geraniol and nerol as well as the three bound forms showed a steady increase in their content during the sampling period. These results confirm previous findings (Günata et al., 1985; Ebang-Oke et al., 2003; Piazzolla et al., 2016). Several additional patterns were observed. For example, the four linalool oxides could be detected at berry onset; their concentration reached a minimum between July 31 (stage E-L 31.5 in this work) and August 8 (stage E-L 33) and then increased in at least one of the two forms to peak on September 13 (stage E-L 38) in their free form. While the glycosidically bound forms of the two diendiols showed a similar pattern of accumulation, free hydroxy-diendiol I and hydroxytrienol were highly concentrated before veraison and decreased over the course of berry ripening, with a trend opposite to that of free hydroxy-diendiol II (**Figures 1**, **2**). The high content of free hydroxy-diendiol I and hydroxy-trienol at berry onset, when free linalool was not yet produced, may indicate that their accumulation is regulated independently from that of their precursor.

In the attempt of simplification, metabolite network analysis was performed on a total of 52 (26 free and 26 glycosidically bound) compounds. Metabolite grouping was obtained through hierarchical clustering and principal component analysis by using different metabolic data sets (**Figure 3** and Supplementary Figures S3, S4). It is clearly evident that most monoterpenes are tightly correlated, which is indicative of their common metabolic origin and in agreement with previous findings (Ilc et al., 2016b). In particular, when considering the absolute amount of free metabolites at 13 time points, three main clusters (AU > 95%) could be identified: (1) cis-pyran linalool oxide (OxD), transpyran linalool oxide (OxC), trans-furan linalool oxide (OxA) and cis-furan linalool oxide (OxB); (2) geraniol, nerol, cis-8 hydroxy-linalool, benzyl alcohol, 2-phenylethanol, trans-geranic acid, citronellol, hydroxy-diendiol II, trans-8-hydroxy-linalool, linalool, hexanol, trans-3-hexen-1-ol, rose oxide I, rose oxide II and α-terpineol; (3) 4-terpineol and hydroxy-diendiol I + hydroxy-trienol. Additionally, within the second cluster a clear separation could be noticed between linalool on one side, geraniol and nerol on the other side (**Figure 3A** and Supplementary Figure S4A). Oppositely, when considering the absolute amount of bound metabolites at 13 time points a single significant cluster was obtained, which included most of the analyzed compounds. It can be easily noticed that bound cis-furan linalool oxide (OxB) has a peculiar behavior with respect to the other three linalool oxides (**Figure 3B** and Supplementary Figure S4C).

#### Microarray Analysis

Gene expression in Moscato Bianco berry skin at stages 2–5 was compared to stage 1 (pre-veraison). The two biological replicates assayed at each stage were confirmed to perfectly cluster together (Supplementary Figure S5). The total number of differentially expressed probes (DEPs) in at least one comparison was 2,228, which corresponds to 15% of the chip probes (Supplementary Table S5). As expected, the highest number of DEPs was observed in stage 5 vs. stage 1 (616 up-regulated and 1,132 down-regulated probes), whereas the lowest number was recorded in stage 2 vs. stage 1 (452 up-regulated and 506 down-regulated probes). A number of DEPs were common among comparisons (21, 19, and 28% of common DEPs among 2, 3, and 4 comparisons, respectively), whereas 32% of the DEPs were regulated at only one time point (data calculated from Supplementary Table S5).

### Microarray Validation via Real-Time PCR

Specific primers were designed for 15 candidate genes and the change in their expression during berry development was analyzed in skin tissues by qRT-PCR to validate the microarray dataset (**Figure 4**). A strong relationship was found between the microarray and qPCR fold changes in the expression levels of the 15 genes (overall Pearson correlation coefficient r = 0.84, with individual values ranging from 0.47 to 1), indicating the reliability of the whole transcriptome assay (**Figure 4** and Supplementary Table S3).

#### Discovery of a Link between Transcriptome and Metabolome in Aroma Development

Potential links between transcripts and metabolites were established based on correlation and clustering approaches, though they do not necessarily imply causation. To this purpose, 4,450 probes with adjusted p-values < 0.05 in all comparisons were considered, which included 1,906 out of the 2,228 DEPs with a cut-off of 2 fold-change and adjusted p-value < 0.01.

#### Integration of 2006 Transcriptomic and Metabolic Data

#### **Pairwise correlation**

Significant (at the 0.05 or 0.01 level) Pearson correlations could be established only in the absence of Benjamini and Hochberg (1995) correction for multiple testing (Supplementary Table S6). Consequently, this result was employed as criterium in the following candidate gene selection only in combination with additional supporting evidence. In the two-step strategy, positive pairwise expression correlations were discovered between 33 enzyme/transporter-coding genes correlated to metabolites and several regulatory genes within the microarray (Supplementary Table S6).

#### **Correlation biclustering**

Correlation clustering between transcripts and metabolites resulted in the identification of 419 biclusters, that are groups of probes with a common behavior toward a group of metabolites. The clustered probes and metabolites were found to belong to a number of biclusters ranging from 1 to 10 and from 3 to 182, respectively (Supplementary Table S7).

#### **Soft clustering**

Based on their expression profile across stages 1–5, the selected 4,450 probes and 52 metabolites were clustered into nine distinct Mfuzz groups (**Figure 5** and Supplementary Table S8). The distribution of probes per cluster within each MapMan functional category is shown in Supplementary Figure S6 and the enriched categories within each cluster are indicated in **Figure 5**. Probes annotated with the Mapman functional category "Secondary metabolism" were not found to be significantly over-represented in any cluster. Free geraniol and nerol were attributed to a distinct

cluster (cluster 6) with respect to free linalool (cluster 9), which reflects their Pvclust clustering (**Figure 3A** and Supplementary Figure S4A). This separation is mainly due to the decrease of free linalool, but not geraniol and nerol, from technological maturity onwards (**Figure 2**). The highly similar accumulation trend of geraniol and nerol likely reflects a common chemical origin (nerol is a geometrical isomer of geraniol), while their relationship with linalool is less clear. Oppositely, the bound forms of the three monoterpenoids accumulated to a similar extent (cluster 4 in **Figure 5**), suggesting dynamic changes in the distribution and concentration of these compounds.

#### Selection of Candidate Genes

Several genes with a potential association to aroma-related compounds were identified from the probes correlated and clustered with those metabolites (Supplementary Table S9, Supplementary Discussion in Supplementary Data1 and Supplementary Table S10). In particular, the contrasting behavior of free linalool and free geraniol/nerol encouraged us to search for genes specifically related to one or the other profile. The existence of linalool-specific metabolic pathways is even more intriguing if we consider that the prevalence of the linalool class on the geraniol one clearly distinguishes Moscato Bianco from other aromatic varieties (D'Onofrio et al., 2016).

From this broad gene set, the most promising candidates for monoterpene biosynthesis and its regulation were further selected (**Table 2**) based on supporting evidence from the literature, e.g., a relevant role for the homolog gene in other plant species, the co-localization with QTLs for monoterpene content (with a special attention to the linalool-specific QTLs on chromosomes 2 and 10, which were also detected in the Moscato Bianco genetic background by Battilana et al., 2009), or the coexpression from public transcriptomic databases with genes involved in the metabolic pahways under study, which may indicate functional association according to the "guilt-byassociation" principle. Hereafter, we discuss the most interesting findings from the present work; obvioulsy, we can't exclude that additional genes not included in this microarray platform may participate in monoterpene biosynthesis, as well as we can't know a priori whether our findings will be reproduced in other Muscat varieties with a genetic background different from Moscato Bianco.

#### **Monoterpene skeleton biosynthesis**

Early terpenoid pathway genes. The role of VvDXS isoforms in the development of aroma was previously investigated by realtime PCR on the same samples of Moscato Bianco analyzed here (Battilana, 2009), for which reason we did not repeat the analysis. In that study a significant up-regulation of VvDXS1 was found to precede the peak of linalool, geraniol and nerol resulting in a positive correlation between VvDXS1 expression profile and monoterpenoid accumulation. On the Grape AROS V1.0 array no probe could be found for VvDXS1, whereas four probes corresponding to other DXS isoforms (VIT\_04s0008g04970 and VIT\_00s0218g00110) were not differentially expressed during Moscato Bianco berry ripening.

Several pieces of evidence from different plant species suggest that flux control in the MEP pathway does not converge on a single rate-limiting enzyme, such as DXS, but may involve other enzymes like DXR (1-deoxy-D-xylulose 5-phosphate reductoisomerase) and HDR (Vranová et al., 2012; Hemmerlin, 2013). The lack of significant modulation and the decreasing

correlation between the fold changes (log2) in the expression levels of the 15 genes obtained by microarray and qRT-PCR analyses is shown in the last chart.

trend during berry ripening observed for VvDXR in our study (VIT\_17s0000g08390 in **Figure 4**) do not support a regulatory role, in agreement with Rodríguez-Concepción et al. (2001) and Mendoza-Poudereux et al. (2014). Oppositely, the expression of VvHDR (VIT\_03s0063g02030 in **Figure 4**) is consistent with the veraison-initiated accumulation of monoterpenes, as reported by Martin et al. (2012) and Wen et al. (2015) (**Table 2** and Supplementary Table S9).

Middle and late terpenoid pathway genes. In other plant species GPPS works as a heterodimeric complex; in particular, the levels of GPPS small subunit, but not GPPS large subunit, might play a key role in regulating monoterpene biosynthesis (Tholl et al., 2004). Consistently, the AROS probes for GPPS large subunit genes (VIT\_04s0023g01210 and VIT\_18s0001g12000) were neither differentially expressed during Moscato Bianco berry ripening nor correlated to any monoterpene. No probe could be identified for the GPPS small subunit.

Only three probes for terpene synthases are present on the Grape AROS V.1 array, which are not specific to any single gene prediction. One of them, showing the best match to the sesquiterpene synthases VIT\_18s0001g04280 and VIT\_18s0001g04530, was up-regulated during Moscato Bianco berry ripening (**Table 2** and Supplementary Table S9). It is worth noting that the same genes were reported to correlate with linalool and α-terpineol (Savoi et al., 2016).

An interesting candidate gene for the biosynthesis of monoterpenes is a nudix hydrolase (VIT\_10s0003g00880), whose expression increases along berry development (**Figure 4**). The corresponding probe belongs to cluster 4, which also harbors several monoterpenes (**Table 2**, Supplementary Tables S8, S9). Recently, a rose nudix hydrolase was reported to convert geranyl diphosphate to geranyl monophosphate, which in turn is hydrolyzed to geraniol by a phosphatase activity (Magnard et al., 2015). This alternative and completely new terpene synthase-independent route for monoterpene production might play a role also in other plants, including grapevine.

#### **Secondary monoterpene transformations**

Extensive oxidative monoterpene metabolism has been reported in grapes and wine, with a percentage of linalool oxygenation ranging from 52 to 97% (Ilc et al., 2016b). The main linalool oxidation products are trans/cis-8-hydroxy-linalool (by hydroxylation), trans/cis pyranoid/furanoid linalool oxides and polyhydroxylated derivatives or polyols like the odorless hydroxy-diendiol I and II (by epoxidation and hydrolysis). Similarly, C-8 oxygenated geraniol and citronellol derivatives

can be formed through hydroxylation, whereas the oxidation to geranial and neral (altogether named citral) is supposedly mediated by alcohol dehydrogenases. Geranic acid is another oxidation product of geraniol. Rose oxide is generated from citronellol by allylic hydroxylation and acid-catalyzed cyclization. Citronellol in turn arises from the reduction of geraniol and nerol (hydrogenation).

Members of the cytochrome P450 (CYP) 71 and 76 families were recently shown to metabolize linalool in Arabidopsis thaliana (Ginglinger et al., 2013; Höfer et al., 2014; Boachon et al., 2015). Interestingly, the CYP76 gene family has encountered an evident expansion in the grape genome (Nelson et al., 2008). In order to identify genes potentially implicated in grape monoterpenoid metabolism we looked


(Continued)

Frontiers in Plant Science | www.frontiersin.org

TABLE 2 | Summary

 of the most significant

 candidate

 genes for

monoterpene

biosynthesis.


TABLE 2 | Continued


(Continued)

TABLE 2 | Continued


Supplementary genes involved in the terpene pahway), profile consistency between microarray and qRT-PCR analyses (the genes assayed by both techniques are marked with an asterisk). The symbols # and § indicate references for grapevine and other plant species, respectively. For details, including the correspondence between microarray probes and V1 gene predictions, see Supplementary Table S9.

in VTCdb database (http://vtcdb.adelaide.edu.au/Home.aspx) for CYP genes coexpressed with linalool synthases, as in Ginglinger et al. (2013). This information was then added to our transcriptomic and metabolic integrated datasets. On this base, we propose some genes (VIT\_15s0048g01490, VIT\_15s0048g01590, VIT\_18s0001g13790, VIT\_00s0389g00030, VIT\_00s0389g00040) and eventually additional ones as potential CYPs involved in linalool metabolism (**Table 2** and Supplementary Table S9). Most of these candidates have been never reported elsewhere, and thus deserve further attention. Conversely, VIT\_15s0048g01490 and VIT\_18s0001g13790 were recently characterized by Ilc et al. (2017) but their biochemical activity was only tested on a limited number of compounds. Our findings suggest instead that these genes might play a role in the production of a broader set of hydroxylated and/or epoxidized products as in other species (Meesters et al., 2007; Ginglinger et al., 2013; Höfer et al., 2013, 2014; Boachon et al., 2015) and, even if the need for further oxidoreductases can not be excluded (Ilc et al., 2016a), they encourage to check this hypothesis by analyzing additional substrates (geraniol, nerol, citronellol) and products (e. g. pyranoid/furanoid linalool oxides, hydroxydiendiols, geranic acid, rose oxides) in CYP enzymatic assays. We also propose an epoxide hydrolase (VIT\_04s0023g02610) to be assessed for involvement in monoterpene oxidative metabolism (**Table 2** and Supplementary Table S9).

Based on their sequence similarity with terpenoid glucosyltransferases from different plant species and on their membership in biclusters harboring some glucosylated monoterpenes, we propose that VIT\_03s0180g00200, VIT\_03s0180g00320 (**Table 2**) and eventually other genes reported in Supplementary Table S9 (VIT\_03s0091g00040, VIT\_03s0180g00280, VIT\_05s0062g00430, VIT\_05s0062g00520, VIT\_05s0062g00630, VIT\_05s0062g00640) might code for enzymes that glucosylate monoterpenes along with additional metabolites. Most of these genes have been investigated in previous works but they were not considered as candidates for monoterpene glucosylation in view of their decreasing expression during berry development (Khater et al., 2011; Bönisch et al., 2014a,b). However, they might be involved in the production of glucosylated monoterpenes with a similar trend, like the high oxidation state monoterpenoids sharing the same biclusters (**Table 2** and Supplementary Table S9), which were not quantified in those papers. This hypothesis is not contradicted by the lack of gene annotation referring to the "Monoterpenoid biosynthesis" pathway and of positive correlation between transcript and monoterpenyl glucoside accumulation, as the same holds for the biochemically characterized monoterpenol glucosyltransferase VvGT7 (**Table 2**) and may be explained by the broad substrate tolerance and overlapping enzymatic activities of the large GT family. Monoterpenyl glucosides are only intermediates within the glycosylation pathway and post-transcriptional control is additionally involved (Bönisch et al., 2014a).

#### **Monoterpene transport**

Terpene transport within the cell and into the apoplast is an almost unexplored field. It may engage multiple pathways, e.g., (1) insertion of the hydrophobic terpenes into vesicle membranes

TABLE

2


Continued

followed by transport and fusion to the plasma membrane, (2) carrier proteins (like GSTs, glutathione S-transferases, and ABC, ATP-binding cassette transporters) that conduct these molecules to the (plasma) membrane, and (3) direct diffusion between the endoplasmic reticulum and/or plastidial (stromule) membranes and the plasma membrane (Ting et al., 2015). The fusion of vesicles with target membranes is mediated by a group of proteins called SNAREs (soluble NSF attachment protein receptors). Surprisingly and still without a clear underlying mechanism, both sesquiterpenes and monoterpenes were boosted when vesicle fusion was inhibited in Nicotiana benthamiana (Ting et al., 2015). Moreover, two Arabidopsis linalool synthases were detected in vesicular structures associated with the plastids (Ginglinger et al., 2013). Based on these findings, we included among our candidates a gene coding for a SNARE associated Golgi protein (VIT\_02s0012g01630). Plant ABCG transporters play a role in the flux of secondary metabolites, particularly of terpenoid origin (Kang et al., 2011). Interestingly, we found an ABCG gene (VIT\_16s0039g00010) that shows a profile consistent with monoterpene accumulation and is coexpressed with several monoterpene synthases in VTCdb. We also selected a glutathione S-transferase (VIT\_08s0040g03040) and two amino acid permeases (VIT\_06s0009g01140 and VIT\_08s0007g05210), which are coexpressed with monoterpene biosynthetic genes in VTCdb and positively correlated to several monoterpenes in the present study (**Table 2** and Supplementary Table S9).

#### **Monoterpene biosynthesis transcriptional regulation**

Recent works (Cramer et al., 2014; Wen et al., 2015) suggested that a group of ERF6-type transcription factors clustered on chromosome 16 are involved in aroma accumulation, based on the correlation of their transcript abundance and the transcript abundance of several terpenoid pathway genes. For some of these regulatory genes, e.g., the orthologs of CrORCA2, CrORCA3, and AaERF1 (De Geyter et al., 2012), no probe was found among the 4,450 probes used for our integrative analysis. Other ERF genes reported in the mentioned papers (VIT\_16s0013g00950, VIT\_16s0013g00980, VIT\_16s0013g00990, VIT\_16s0013g01030, VIT\_16s0013g01050, VIT\_16s0013g01060, not listed in Supplementary Table S9) belonged to clusters 1, 2, 7 and did not show any relevant positive correlation with monoterpenes. However, some of the AROS probes had only a partial match with these genes, as a consequence they might correspond instead to ERF gene isoforms not involved in flavor determination. Conversely, the genes VIT\_16s0100g00400 and VIT\_18s0001g05250 showed an expression profile consistent with the accumulation of monoterpenes in Moscato Bianco ripening berry (**Table 2** and Supplementary Table S9).

We also observed an interesting behavior (**Figure 4**, **Table 2**, and Supplementary Table S9) for TFs of the MYB (VIT\_14s0066g01090) and NAC (VIT\_19s0014g03300) families that promote mono- and sequiterpene production in other plant species (Reeves et al., 2012; Nieuwenhuizen et al., 2015). In particular, VIT\_14s0066g01090 (MYB24) has been proposed as a candidate transcriptional regulator of (mono)terpene biosynthesis also in grapevine (Matus, 2016; Savoi et al., 2016), for which reason it deserves further attention. Finally, based on the negative effect of GBF1 (G-box binding factor 1) and ZCT (zinc-finger Catharanthus transcription factor) proteins on the expression of the TIA (terpenoid indole alkaloid) biosynthetic genes Str (strictosidine synthase) and Tdc (tryptophan decarboxylase) (Sibéril et al., 2001; Pauw et al., 2004), we selected two genes (VIT\_15s0046g01440 and VIT\_18s0001g09230) negatively correlated with monoterpene accumulation during Moscato Bianco berry ripening (**Table 2** and Supplementary Table S9).

One of the signals dramatically impacting isoprenoid biosynthesis in higher plants is light, which activates the MEP pathway at the transcriptional and post-transcriptional level (Rodríguez-Concepción, 2006; Cordoba et al., 2009; Vranová et al., 2012; Mannen et al., 2014). Sunlight exclusion limits the synthesis and accumulation of terpenes also in grape berries (linalool and the bound forms being the most responsive) by especially affecting DXS and TPS genes (Zhang et al., 2014; Friedel et al., 2016; Joubert et al., 2016; Matus, 2016; Sasaki et al., 2016). Our findings (**Figure 4**, **Table 2**, and Supplementary Table S9) are consistent with a role, among others, for HY5 (LONG HYPOCOTYL5, VIT\_04s0008g05210) in the regulation of light-induced terpenoid biosynthesis in grapes, in agreement with other evidences (Carbonell-Bejerano et al., 2014a,b; Zhou et al., 2015; Loyola et al., 2016).

The isoprenoid pathway has also been reported to be under the circadian clock control. In particular, the emission of volatile terpenoids follows a diurnal rhythm and genes encoding enzymes involved in IPP biosynthesis (especially those from the MEP pathway) and downstream pathways are coexpressed with circadian clock genes and show typical circadian expression profiles (Cordoba et al., 2009; Vranová et al., 2012; Pokhilko et al., 2015). Some probes on the AROS array correspond to a gene of the circadian oscillator (VIT\_15s0048g02410) and fall into clusters harboring several monoterpenes (**Table 2** and Supplementary Table S9).

The expression profile of a number of additional transcription factors (including master regulators) and genes potentially involved in the post-transcriptional regulation (Hemmerlin, 2013) overlaps monoterpene accumulation during Moscato Bianco berry ripening, which supports a role in the control of monoterpene biosynthesis for VIT\_01s0026g01970, VIT\_02s0012g01040, VIT\_02s0012g01240, VIT\_02s0234g00100, VIT\_03s0038g02500, VIT\_04s0023g00130, VIT\_04s0023g01250, VIT\_04s0023g02950, VIT\_06s0004g07550, VIT\_07s0031g01320, VIT\_07s0031g01930, VIT\_07s0104g01050, VIT\_08s0007g05880, VIT\_09s0054g01780, VIT\_10s0003g03190, VIT\_12s0028g03860, VIT\_00s0214g00090, VIT\_00s0463g00020 (**Figure 4**, **Table 2**, Supplementary Table S9). To our knowledge, these genes represent new regulatory candidates for the production of several (cluster 4) or specific metabolites, like linalool (cluster 9) and geraniol/nerol (cluster 6), as suggested by their co-localization with QTLs and their correlation with enzyme/transporter genes correlated to metabolites.


(Continued)


TABLE 3 | Continued



(Continued)


correlation; A, microarray analysis; R, qRT-PCR analysis.

 was

 gray

 at

#### Integration of Transcriptomic and Metabolic Data Over Multiple Seasons to Verify a Subset of Candidate Genes

In order to confirm the above links between transcriptome and metabolome in aroma development, the 15 genes assessed by both microarray and real-time analyses were also tested for correlation with the metabolic profile over three seasons, which were considered as three biological replicates (**Table 3** and Supplementary Figure S1A). Significant correlations were found for all the genes except VvDXR (confirming the results from 2006 data) and VvHDR, which probably precedes monoterpene accumulation. Several compounds were affected, especially in their glycosidically bound form. Unsurprisingly, most of the metabolites with no correlation showed an inconsistent profile among seasons (e.g., free OxA and citronellol, bound α-terpineol) or a decreasing trend along berry ripening (e.g., free HOdiendiol I + HO-trienol and bound OxB). Though not ensuring a punctual conformity to the observations from a single year (**Table 2**), the findings from multiple seasons (**Table 3**) prove the general consistency of the outcomes of different techniques and years and argue for the reliability of the whole set of results based on the integration of 2006 transcriptomic and metabolic data.

#### CONCLUSION

Understanding the origin of grape aromatic compounds is essential in the breeding of new varieties and in the management of high-quality crops in a changing climate. In this work, previously undescribed gene-to-metabolite networks with a possible association to grape flavor were deduced by integrating the expression profiles of 4,450 gene tags and the accumulation profiles of 52 metabolites. Pairwise correlation and clustering methods pointed to several structural and regulatory genes potentially involved in the biosynthesis of monoterpenes, which

#### REFERENCES


paves the way for locating candidates for at least some of the missing links in the underlying pathway. Our collective findings contribute toward understanding the regulation of secondary metabolism in Muscat-type grape cultivars through the formulation of testable hypotheses regarding the function of specific genes.

#### AUTHOR CONTRIBUTIONS

LC, CK, JB, FE, SD, and MG contributed to the project design; LC, MT, JB, FE, MS, and CC took part in the experimental work; RL provided the metabolic analysis; LC, CK, and MM performed the statistical and bioinformatic analyses; LC and CK were involved in data interpretation; LC wrote the manuscript. All the authors approved the final version of this text.

#### FUNDING

This research was sustained by a Short Term Scientific Mission grant awarded to LC by the Institute of Vine and Wine Sciences (Bordeaux, France) and with the financial support provided by the Autonomous Province of Trento (Accordo di Programma).

#### ACKNOWLEDGMENTS

We thank Silvia Lorenzi for sample collection, Sergio Moser for metabolic analysis and Pietro Franceschi for statistical support.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2017. 00780/full#supplementary-material


information regarding the dynamics of grape ripening. BMC Plant Biol. 11:149. doi: 10.1186/1471-2229-11-149


(Syrah×Pinot Noir) mature grapes: integration of transcriptional profiling and metabolic quantitative trait locus analyses. J. Exp. Bot. 66, 4441–4453. doi: 10.1093/jxb/erv243


pathway in white grapes (Vitis vinifera L.). BMC Plant Biol. 16:67. doi: 10.1186/s12870-016-0760-1


synthase gene QH6 in modulating its rhythmic expression. Front. Plant Sci. 6:304. doi: 10.3389/fpls.2015.00304

Zhu, B. Q., Cai, J., Wang, Z. Q., Xu, X. Q., Duan, C. Q., and Pan, Q. H. (2014). Identification of a plastid-localized bifunctional nerolidol/linalool synthase in relation to linalool biosynthesis in young grape berries. Int. J. Mol. Sci. 15, 21992–22010. doi: 10.3390/ijms1512 21992

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Costantini, Kappel, Trenti, Battilana, Emanuelli, Sordo, Moretto, Camps, Larcher, Delrot and Grando. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Constructing Integrated Networks for Identifying New Secondary Metabolic Pathway Regulators in Grapevine: Recent Applications and Future Opportunities

#### Darren C. J. Wong<sup>1</sup> and José Tomás Matus <sup>2</sup> \*

<sup>1</sup> Ecology and Evolution, Research School of Biology, Australian National University, Acton, ACT, Australia, <sup>2</sup> Centre for Research in Agricultural Genomics, CSIC-IRTA-UAB-UB, Barcelona, Spain

#### Edited by:

Ashraf El-kereamy, University of California, USA

#### Reviewed by:

Sara Zenoni, University of Verona, Italy Jose A. Casaretto, University of Guelph, Canada

\*Correspondence: José Tomás Matus tomas.matus@cragenomica.es

#### Specialty section:

This article was submitted to Technical Advances in Plant Science, a section of the journal Frontiers in Plant Science

> Received: 16 February 2017 Accepted: 22 March 2017 Published: 12 April 2017

#### Citation:

Wong DCJ and Matus JT (2017) Constructing Integrated Networks for Identifying New Secondary Metabolic Pathway Regulators in Grapevine: Recent Applications and Future Opportunities. Front. Plant Sci. 8:505. doi: 10.3389/fpls.2017.00505 Representing large biological data as networks is becoming increasingly adopted for predicting gene function while elucidating the multifaceted organization of life processes. In grapevine (Vitis vinifera L.), network analyses have been mostly adopted to contribute to the understanding of the regulatory mechanisms that control berry composition. Whereas, some studies have used gene co-expression networks to find common pathways and putative targets for transcription factors related to development and metabolism, others have defined networks of primary and secondary metabolites for characterizing the main metabolic differences between cultivars throughout fruit ripening. Lately, proteomic-related networks and those integrating genome-wide analyses of promoter regulatory elements have also been generated. The integration of all these data in multilayered networks allows building complex maps of molecular regulation and interaction. This perspective article describes the currently available network data and related resources for grapevine. With the aim of illustrating data integration approaches into network construction and analysis in grapevine, we searched for berry-specific regulators of the phenylpropanoid pathway. We generated a composite network consisting of overlaying maps of co-expression between structural and transcription factor genes, integrated with the presence of promoter cis-binding elements, microRNAs, and long non-coding RNAs (lncRNA). This approach revealed new uncharacterized transcription factors together with several microRNAs potentially regulating different steps of the phenylpropanoid pathway, and one particular lncRNA compromising the expression of nine stilbene synthase (STS) genes located in chromosome 10. Application of network-based approaches into multi-omics data will continue providing supplementary resources to address important questions regarding grapevine fruit quality and composition.

Keywords: Vitis vinifera, stilbenes, flavonoids, ripening, systems biology, genome-wide, multi-omics, data integration

## NETWORK-BASED APPROACHES INTO OMICS DATA

Complex biological processes can be studied from a "multiomics" perspective thanks to the recent improvements in genome-wide techniques and systems biology approaches. Each omics data type is particularly useful in elucidating the constituents and function of a particular cellular domain. Together, they constitute layers of biological complexity. Genomic data generated from genome sequencing projects are commonly used to ascribe molecular function and biological processes based on sequence similarity, while transcriptomics and metabolomics data typically provide a global "snapshot" of gene expression and metabolite dynamics in various biological contexts.

For many omics data, interactions/associations between molecules can be represented as networks, where nodes (genes, proteins, metabolites) are connected by edges. These edges denote an association often inferred from correlational and informational theoretic measures such as Pearson correlation coefficient (PCC) and mutual information (MI), respectively. In the case of gene co-expression networks (GCNs), edges represent similar gene expression behaviors. Based on the "guilt by association" principle, genes involved in related processes share similar gene expression dynamics across a wide range of experiments (Wolfe et al., 2005). However, as functional information could be delimited to a reduced number of interactions within a gene network (Gillis and Pavlidis, 2012), subsequent targeted gene characterizations are needed to prove these relationships. Whether the function of a network is dependent or not on specific interactions, GCN analysis have proven to be a powerful tool for inferring gene function and coordinated biological processes related to plant metabolism (Persson et al., 2005; Itkin et al., 2013).

Other forms of networks constructed from omics datasets do not necessarily rely on abundance or expression levels to establish node relationships. For example, protein-protein interaction networks describe physically interacting protein pairs identified from high-throughput yeast two-hybrid screens (e.g., Arabidopsis Interactome Mapping Consortium, 2011). Also, genome-wide location studies (i.e., by using ChIP-Seq) allow determining regulatory networks for transcription factors (TF) and other DNA-binding proteins. These TF-binding networks have led to the identification of novel components and of new connections that alter the network diagrams originally drawn by genetic and molecular analyses (reviewed by Ferrier et al., 2011).

#### RECENT APPLICATION OF METABOLITE NETWORK-BASED APPROACHES IN GRAPES

Studies utilizing networks constructed from omics data profiled in the berry are continuously increasing (**Table 1**). Network analyses involving metabolite datasets (primary and/or secondary metabolites) are by far the most reported. These studies have included networks inferred from single contexts such as berry development and ripening (Zamboni et al., 2010; Dai et al., 2013; Wang et al., in press), or in combination with other factors including environmental influence (Guan et al., 2016; Savoi et al., 2016; Reshef et al., 2017), and/or cultivar differences (Degu et al., 2014; Cuadros-Inostroza et al., 2016). Network topology has also been investigated in detail to reveal critical metabolites and their regulation. For instance, Cuadros-Inostroza et al. (2016) showed that an increase in network connectedness and density (especially regarding primary metabolites) became prevalent at specific berry developmental stages such as fruit set and veraison (i.e., the onset of ripening). The same study, in concordance with Degu et al. (2014), highlighted that berry-metabolite networks from different cultivars could possess contrasting network topologies, albeit with overall network connections generally maintained. Metabolite networks from the cultivars cv. "Merlot" (Cuadros-Inostroza et al., 2016) and cv. "Shiraz" (Degu et al., 2014) were consistently denser to that off cv. "Cabernet Sauvignon."

Rewiring of berry metabolite networks under different environmental conditions or perturbations such as drought (Savoi et al., 2016) and sunlight exposure (Reshef et al., 2017) have also been reported. These studies have shown that higher network connectivity is commonly observed in perturbed networks. Such property could be associated to a tighter metabolic control of the metabolic pathways under investigation. Such is seen for phenylpropanoid and volatile organic compounds (VOC) in berries under prolonged drought compared to non-stress berries (Savoi et al., 2016). Similarly, primary metabolite networks encompassing compounds related to glycolysis, the TCA cycle, and amino acid metabolism showed higher network connectivity in shaded berries compared to fully exposed berries (Reshef et al., 2017).

Some metabolic-network studies have shown that certain metabolites (or classes) could act as important switches in the developmental regulation of metabolism during berry growth and ripening, given their high centrality (number of connections) or degree scores in their network. Dai et al. (2013) showed that trehalose-6-phosphate appeared to be the most connected compound in the primary metabolite network of cv. "Cabernet Sauvignon" grapes, with significant partial correlations to sugar metabolism, glycolysis, and TCA cycle intermediates. Altogether, these compounds may be implicated in coordinating metabolite dynamics during berry development. One recent study highlighted fucose as critical for coordinating metabolic regulation in a stage-specific manner, thus deprioritizing the importance of sugars such as glucose, fructose, and sucrose as a function of network centrality measure (Cuadros-Inostroza et al., 2016). These findings further demonstrate the complexity of berry metabolic regulation during development and ripening.

# GENE CO-EXPRESSION NETWORKS TO STUDY GRAPE BERRY RIPENING

The increased ease of transcriptome profiling, combined with availability of datasets shared by the grapevine research community in public repositories, has led to increased attention

#### TABLE 1 | Studies of grape berry development and composition involving molecular networks approaches.


G, genes; M, metabolite; Prot, protein; Prom, promoter; + R, includes resource/database; I, integrated (G, M, Prot, Prom, miRNA inclusive).

in the use of gene co-expression networks (GCNs) in the study of berry development and metabolism. GCNs can be classified into "condition-dependent" and "condition-independent" categories (Usadel et al., 2009). In grapes, several studies have focused on condition-independent GCNs (encompassing different cultivars, tissues, developmental stages, stress and vineyard management treatments) as it provides a more convenient and representative (albeit "static") relationship overview (**Table 1**). This approach has been useful for ascribing the most representative biological functions of the 134 grapevine R2R3-MYB transcription factors based on their top 100 co-expressed genes (Wong et al., 2016), where VviMYB13 (close homolog of VviMYB14 and VviMYB15) was identified as an additional STILBENE SYNTHASE regulator acting in a tissue- and/or stress-dynamic manner.

Platforms such as the ViTis Co-expression DataBase (VTCdb; Wong et al., 2013) and VESPUCCI (Moretto et al., 2016) have been successfully exploited to study the extent of transcription factor regulatory networks, providing support for targeted functional studies. Such is the case for the bZIP TF VvibZIPC22, which is involved in the regulation of flavonoid biosynthesis in grapes and may be also implicated in carbohydrate and amino acid metabolism, as inferred from VESPUCCI (Malacarne et al., 2016). Two other bZIP TFs (VviHY5 and VviHYH) were shown to co-operatively mediate flavonol accumulation in grapes in response to sunlight and ultraviolet radiation exposure (Loyola et al., 2016). As inferred from VTCdb and GCN analysis, these regulators were potentially implicated in carbohydrate and isoprenoid metabolism in addition to the control of the flavonoid pathway. Similarly, the involvement of the grapevine VviWRKY26 in the regulation of vacuolar transport and flavonoid biosynthesis was demonstrated using a combination of transcriptomic approaches including GCNs (Amato et al., 2017).

Condition-dependent GCNs have been constructed from tissue- or stress-specific datasets, including berry (Zamboni et al., 2010; Palumbo et al., 2014) or abiotic and biotic stresses (Wong et al., 2017). These GCNs provide several advantages over condition-independent networks as inferring gene function is largely simplified, providing a more "dynamic" overview of gene relationships that otherwise could be enhanced or lost in certain conditions (Obayashi et al., 2011). One example of a condition-specific GCN involves the study of the transcriptomes of five black-skinned cultivars across four berry phenological stages (Palumbo et al., 2014). The authors identified "fight-club" nodes and "switch" genes, having the latter unique expression profiles and network topological properties, such as a marked negative correlation connectivity to both neighboring genes and genes grouped to other modules in the network. Genes associated with transcription factor activity; cell wall modification and carbohydrate and secondary metabolism were found as candidate master regulators, potentially inducing large-scale transcriptome reprogramming during berry development (Palumbo et al., 2014).

Finally, miRNA and siRNA-mediated gene regulatory networks have also been constructed from high-throughput small RNA and degradome sequencing and computational target prediction methods (Zhang et al., 2012; Belli Kullan et al., 2015). These networks (not relying in abundance or expression levels) revealed novel modules such as miR156/miR172 regulatory circuits and VviTAS3/4 regulatory cascades, which are implicated in regulating plant growth and development and in the control of flavonoid biosynthesis, respectively.

# TOWARD THE INTEGRATION OF MULTI-OMICS DATA IN GRAPES

Although individual omic network methods have been widely used, a shift toward multi-omics data and integration is increasingly being adopted in plant biology (Proost and Mutwil, 2016), including grapevine (**Table 1**). Integration approaches allow building complex maps of molecular regulation and interaction. By these means, complex traits from these networks can be assessed (e.g., plasticity and evolution).

The first systems level study in grapes leveraged transcriptomic, metabolomic, and proteomic technologies to understand berry development and the postharvest withering process (i.e., controlled dehydration) in cv. "Corvina" grapes (Zamboni et al., 2010). Using a combination of hypothesis-free and -driven integration approaches, the authors were able to tease out putative berry stage-specific functional networks. As an outcome, a fully integrated network related to the withering process revealed key phenylpropanoid and stressresponsive genes (i.e., biotic, osmotic, and oxidative), together with proteins involved in oxidative- and osmotic-stress, and secondary metabolites such as acylated anthocyanins and stilbenes. Recently, integration of berry metabolome (primary and secondary) and proteome networks encompassing 12 developmental stages revealed a greater propensity of an energylinked metabolism in berries prior to veraison (Wang et al., in press). These observations corroborated earlier studies (Dai et al., 2013; Cuadros-Inostroza et al., 2016), demonstrating that pronounced changes in the berry occurs before veraison, characterized by a reduction of many early accumulating primary metabolites. Interestingly, the integrated network also revealed several modules with high node degree for many metabolites (amino acids and organic acids) and corresponding enzymes catalyzing their synthesis (Wang et al., in press).

Characterizing genes that regulate the accumulation of secondary metabolites throughout fruit ripening is key for improving quality traits and for predicting plant behavior in response to the environment. In this sense, transcript-metabolite associations have been used to prioritize candidate genes important for determining berry quality parameters under adverse environmental conditions (Savoi et al., 2016). Integrated transcript-metabolite networks encompassing monoterpenes that are both ripening-related and drought-modulated (e.g., linalool, nerol, α-terpineol) revealed many highly co-regulated transcripts to be involved in terpene and lipid metabolism. The authors further highlighted VviMYB24 as a promising regulatory candidate for monoterpene biosynthesis given consistent correlations with all three monoterpenes in their study.

Cis-regulatory element-driven networks have been recently constructed using integrated information of promoter CRE structure and network connectivity (Wong et al., 2017). Numerous CRE-driven modules inferred using conditiondependent GCNs (development-dynamic and stress-specific) highlighted roles in stress response (e.g., to drought and pathogens) and developmental processes (e.g., fruit ripening). For example, GCC-core sub-modules contained many genes that were highly induced in berries and leaves infected with fungi (Wong et al., 2017).

Cis-regulatory element enrichment maps or transcript information for miRNA target enrichment analysis can be easily integrated into plant GCNs. This approach has been used to prioritize target genes of the entire grape R2R3-MYB family (Wong et al., 2016) and also to explain the expression responses of module genes under prolonged drought stress in berries (Savoi et al., 2016). Enrichment for miRNA targets within GCNs has suggested a pivotal role of these molecules in regulating the expression of "switch" genes in a stage-specific manner (Palumbo et al., 2014). Finally, aggregating several networks into a community network can also be advantageous to effectively reveal discrepancies between individual networks while highlighting associations common across individual networks (Proost and Mutwil, 2016). This approach has been used by Loyola et al. (2016) to identify a set of high confidence targets of HY5 and HYH given by the combination of microarray and RNA-Seq data with genome-wide promoter inspections. It is noteworthy that "condition-independent" and "condition-dependent" approaches are still useful for providing a preliminary insight into co-expression relationships in grapes.

## AN ILLUSTRATION FOR THE INTEGRATION OF MULTILAYERED NETWORKS FOR DISSECTING THE COMPLEXITY OF THE BERRY'S PHENYLPROPANOID COMPOSITION

In grapes, phenylpropanoids influence their organoleptic properties and beneficial attributes to human health, highlighting the importance of their study. Several reports have demonstrated the complex nature of secondary metabolism in grapevine, both at the level of chemical composition and genetic regulation (Dal Santo et al., 2013; Costantini et al., 2015; Malacarne et al., 2015). Among the many phenylpropanoid compounds that influence the quality of grapes and wines, some of the most important are flavonoids (anthocyanins, flavonols and tannins) and stilbenes. These compounds accumulate in a temporal and compartmentalized manner and numerous regulators of their accumulation have been characterized to date (Reviewed by Kuhn et al., 2014; Matus, 2016). One strikingly relevant feature of the grapevine genome is that wine quality-related gene families are expanded in gene number (Martin et al., 2010; Vannozzi et al., 2012), including those related with transcription factor activity (Matus et al., 2008; Wong et al., 2016). Genomics and transcriptomics data originated from these and others studies suggest that the regulation of secondary metabolism in grape is a much more complex trait compared to plant model species. As large-scale omics data are periodically accumulating; there is an enormous potential for gene discovery in relation to grape secondary metabolic pathways.

To demonstrate how various biological networks can be integrated to study berry's phenylpropanoid composition, we gathered networks generated from gene co-expression analyses, predicted miRNA-gene and long non-coding RNA (lncRNA) gene interactions. First, we re-analyzed a comprehensive berry ripening RNA-Seq transcriptome dataset (five black-skinned cultivars sampled at four developmental stages; Palumbo et al., 2014) and constructed a ripening-specific gene co-expression network (PCC > |0.8|). This ripening-specific GCN was then used as a basis for lncRNA-gene network, which consisted of predicted lncRNAs (Vitulo et al., 2014) that showed strong correlation with a putative "interacting" gene (PCC > | 0.8 |) that was co-located within 100 kb flanking the lncRNA position. Using a comprehensive catalog of grapevine miRNAs (Belli Kullan et al., 2015; Pulvirenti et al., 2015), we also reanalyzed potential miRNA-mRNA interactions using psRNATarget with default parameters (Dai and Zhao, 2011). As the interpretation of each network at a global scale is out of the scope of this perspective, we focused our attention on the early phenylpropanoid and flavonoid (ePP and Fla) pathways and on the potential regulatory genes and their interactions (among genes, miRNAs, and lncRNAs). The resulting network is composed of 112 ePP/Fla pathway genes (differentially expressed during berry development and ripening) together with five miRNA and 14 predicted grapevine lncRNAs (**Figure 1**). GCN analysis revealed a strong co-regulation within early phenylpropanoid and flavonoid pathway genes maintaining few connections between both sub-pathway genes during the course of berry development and ripening.

Three clusters (I, II and III) were observed for Fla pathway genes sharing many positive correlations within each group (**Figure 1**). Cluster I contained genes mainly involved in the regulation of anthocyanin accumulation such as five flavonoid-3′ ,5′ -hydroxylases (F3′ 5 ′H), two anthocyanin-o-methyltransferases (AOMT1-2), the UDP-GLUCOSE:FLAVONOID 3-O-GLUCOSYLTRANSFERASE (UFGT) and ANTHOCYANIN-3-O-GLUCOSIDE-6′′-O-ACYLTRANSFERASE (3AT). Cluster II consisted of genes encoding proanthocyanidin biosynthesis genes including three predicted galloyl glucosyltransferases, ANTHOCYANIDIN REDUCTASE (ANR), and LEUCOANTHOCYANIDIN REDUCTASE (LAR), as well as upstream flavonoid pathway genes such as CHALCONE SYNTHASE (CHS) and CHALCONE ISOMERASE (CHI). One predicted antisense lncRNA (VIT\_203s0180n00020) collocated (within 50 kB) and positively correlated with one galloyl glucosyltransferase gene (VIT\_03s0180g00200). This cluster also contained genes encoding one 4-coumarate-Co-A ligase (4CL), two flavonol synthases (FLS4-5), one flavanone-3-hydroxylase (F3H), and one caffeic acid 3-o-methyltransferase (COMT), all of which were negatively correlated with genes from cluster I. Furthermore, grapevine miRNAs miR169r/t and grape-m0534 were predicted to target 4CL and FLS4, respectively. Several genes belonging to cluster II and I shared negative correlations (light blue solid edges, **Figure 1**). This separation is evident whereby the majority

interaction and lncRNA-gene co-location (within 100 kb). Clusters I, II and III connect genes belonging to the early phenylpropanoid and flavonoid (ePP and Fla) sub-pathways. Cluster IV shows a dense group containing predominantly PHENYLALANINE AMMONIA-LYASE (PAL) and STILBENE SYNTHASE (STS) genes that are largely co-expressed with positive co-expression correlations. Purple edges represent positive co-expression correlations between TFs and early phenylpropanoid pathway genes. Pie chart colors represent the presence of selected TF-binding sites (based on cis-regulatory element enrichment analysis) in promoter regions of the corresponding enzyme-coding genes. Light blue border edges depict STS genes located in chromosome 10.

of genes from cluster I are ripening-specific (i.e., upregulated from veraison onwards), while many genes from cluster II are mostly expressed during the early-to-mid stages of berry development (and subsequently downregulated as ripening progresses).

As there is much less evidence in the regulation of the early phenylpropanoid and stilbene sub-pathways compared to the regulation of flavonoid biosynthesis, we focused our attention on a fourth, highly connected cluster (IV) holding strong positive correlations within and between the two large PHENYLALANINE AMMONIA-LYASE (PAL) and STILBENE SYNTHASE (STS) gene families (**Figure 1**). Two cinnamate-4-hydroxylases (C4H) also shared many strong positive correlations with PAL genes and one 4CL was positively correlated with many STS encoding genes. Gene expressions within this cluster were mainly late-ripening specific, with many of them peaking at harvest. Promoters from cluster IV were highly enriched for cis-regulatory elements including those for R2R3-MYB, AP2/ERF, WRKY, bHLH, and bZIP TF binding. In particular, the MYB binding site CCWACC was present in one CCoAMT, two C4H, 10 PAL, and 27 STS genes. The potential regulation of these genes by MYB transcription factors is supported by recent studies showing that several grapevine MYBs may have regulatory roles controlling the levels of small weight phenylpropanoids and stilbenes (Höll et al., 2013; Cavallini et al., 2015). Our approach is novel in suggesting the regulatory roles by other TF families such as WRKY and AP2/ERF. For example, strong co-regulation of nine WRKY TF to 11 PAL and 44 STS genes coincided with the presence of WRKY cis-regulatory elements in many PAL and STS genes. Interestingly, one of the four predicted intergenic lncRNAs (VIT\_210s0042n00100) was co-located and strongly co-regulated with all nine STS positioned on chromosome 10. Recent evidence from several functionally characterized lncRNAs in animals and plants suggest that lncRNAs could operate as decoys, guides, signals, and scaffolds, acting as single molecules or complexes regulating pre- and post-transcriptional processes (Wang and Chang, 2011). As such, our observation raises the plausibility of a large-scale regulatory function between this lncRNA and STS genes. This STS-associated lncRNA may fulfill combinatorial roles for the fine-regulation of multiple STS, as signals for transcription activity in a stage-specific way or as guides for chromatin modifiers to the cluster of tandem-positioned STS of chromosome 10, potentially modulating DNA accessibility.

#### CONCLUSION

Multi-omics studies incorporating systems biology approaches in grapevine have facilitated the identification of new grape secondary metabolism regulators and have helped in the characterization of genome-wide responses to environmental factors. These studies have brought knowledge and new tools to understand how to modify and improve grape's quality.

#### REFERENCES


Additional efforts will still be needed to map protein-DNA and protein-protein landscapes at a large scale. Also, DNAse I hypersensitivity mapping could be useful to identify pioneering transcription factors controlling grape and wine quality traits.

# AUTHOR CONTRIBUTIONS

JTM conceived the article and planned its structure. DW and JTM searched and discussed the literature and wrote the manuscript. DW generated new network data. All authors have read and approved the manuscript.

#### ACKNOWLEDGMENTS

The authors wish to acknowledge Dr. Jason Argyris (Centre for Research in Agricultural Genomics, CRAG) for critically reviewing this work.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Wong and Matus. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Cultivar Diversity of Grape Skin Polyphenol Composition and Changes in Response to Drought Investigated by LC-MS Based Metabolomics

#### Edited by:

Simone Diego Castellarin, University of British Columbia, Canada

#### Reviewed by:

Jose Carlos Herrera, University of Natural Resources and Life Sciences, Vienna, Austria Markus Keller, Washington State University, United States

> \*Correspondence: Véronique Cheynier veronique.cheynier@inra.fr

#### Specialty section:

This article was submitted to Technical Advances in Plant Science, a section of the journal Frontiers in Plant Science

> Received: 23 May 2017 Accepted: 10 October 2017 Published: 27 October 2017

#### Citation:

Pinasseau L, Vallverdú-Queralt A, Verbaere A, Roques M, Meudec E, Le Cunff L, Péros J-P, Ageorges A, Sommerer N, Boulet J-C, Terrier N and Cheynier V (2017) Cultivar Diversity of Grape Skin Polyphenol Composition and Changes in Response to Drought Investigated by LC-MS Based Metabolomics. Front. Plant Sci. 8:1826. doi: 10.3389/fpls.2017.01826 Lucie Pinasseau<sup>1</sup> , Anna Vallverdú-Queralt <sup>1</sup> , Arnaud Verbaere<sup>1</sup> , Maryline Roques 1, 2, 3 , Emmanuelle Meudec<sup>1</sup> , Loïc Le Cunff <sup>3</sup> , Jean-Pierre Péros <sup>4</sup> , Agnès Ageorges <sup>2</sup> , Nicolas Sommerer <sup>1</sup> , Jean-Claude Boulet <sup>1</sup> , Nancy Terrier <sup>2</sup> and Véronique Cheynier 1, 2 \*

<sup>1</sup> Plateforme Polyphénols SPO, INRA, Montpellier SupAgro, Université de Montpellier, Montpellier, France, <sup>2</sup> SPO, INRA, Montpellier SupAgro, Université de Montpellier, Montpellier, France, <sup>3</sup> IFV Pôle national matériel végétal, UMT Génovigne, Montpellier, France, <sup>4</sup> AGAP, INRA, CIRAD, Montpellier SupAgro, Université de Montpellier, Montpellier, France

Phenolic compounds represent a large family of plant secondary metabolites, essential for the quality of grape and wine and playing a major role in plant defense against biotic and abiotic stresses. Phenolic composition is genetically driven and greatly affected by environmental factors, including water stress. A major challenge for breeding of grapevine cultivars adapted to climate change and with high potential for wine-making is to dissect the complex plant metabolic response involved in adaptation mechanisms. A targeted metabolomics approach based on ultra high-performance liquid chromatography coupled to triple quadrupole mass spectrometry (UHPLC-QqQ-MS) analysis in the Multiple Reaction Monitoring (MRM) mode has been developed for high throughput profiling of the phenolic composition of grape skins. This method enables rapid, selective, and sensitive quantification of 96 phenolic compounds (anthocyanins, phenolic acids, stilbenoids, flavonols, dihydroflavonols, flavan-3-ol monomers, and oligomers…), and of the constitutive units of proanthocyanidins (i.e., condensed tannins), giving access to detailed polyphenol composition. It was applied on the skins of mature grape berries from a core-collection of 279 Vitis vinifera cultivars grown with or without watering to assess the genetic variation for polyphenol composition and its modulation by irrigation, in two successive vintages (2014–2015). Distribution of berry weights and δ <sup>13</sup>C values showed that non irrigated vines were subjected to a marked water stress in 2014 and to a very limited one in 2015. Metabolomics analysis of the polyphenol composition and chemometrics analysis of this data demonstrated an influence of water stress on the biosynthesis of different polyphenol classes and cultivar differences in metabolic response to water deficit. Correlation networks gave insight on the relationships between the different polyphenol metabolites and related biosynthetic pathways. They also established patterns of polyphenol response to drought, with different molecular families affected either positively or negatively in the different cultivars, with potential impact on grape and wine quality.

Keywords: grape berry, Vitis vinifera, phenolic compounds, UHPLC-QqQ-MS, metabolomics, water deficit, large-scale studies

## INTRODUCTION

In the context of climate change, it is of prime importance to anticipate and predict the response of the different biota to the changes in environmental conditions, especially for plants, that are devoid of motility. Climate change is expected to affect plant composition and consequently, in the case of crop species such as grapevine, the quality of plant derived products. Among plant metabolites, secondary metabolites, including phenolic compounds, have been recognized as playing multiple roles in plant response to a wide range of biotic and abiotic stresses and in particular to water stress (Baker and Orlandi, 1995; Dixon and Paiva, 1995; Caldwell et al., 2003). They are also essential components of plant derived foods and beverages, responsible for major organoleptic properties such as color and taste and contributing health benefit (Manach et al., 2004).

Grape phenolic compounds comprise several families, divided between non flavonoids (hydroxybenzoic acids, hydroxycinnamic acids, and stilbenes) and flavonoids, based on the same C6-C3-C6 skeleton (flavonols, dihydroflavonols, flavan-3-ols, and anthocyanins). Each family is represented by several compounds differing by their hydroxylation level and by substitution of the hydroxy groups (methylation, glycosylation, acylation). For example, anthocyanins, the red grape pigments, are based on six aglycones which can be mono- or di-glucosylated and further acylated with acetic, p-coumaric, and caffeic acid, giving rise to a large number of compounds (Favretto and Flamini, 2000; Heier et al., 2002; Vidal et al., 2004a). Moreover, various anthocyanin derivatives such as anthocyanin dimers and flavan-3-ol anthocyanin adducts have been detected in grape skin extracts (Vidal et al., 2004b). Grape flavan-3-ols also show high diversity. They include several monomers (catechin, epicatechin, gallocatechin, epigallocatechin, and epicatechin 3-gallate) that are the constitutive units of oligomers and polymers (proanthocyanidins or condensed tannins), with degrees of polymerization ranging from 2 to over 100 in grape skin (Souquet et al., 1996).

The impact of water stress on grape berry composition has already been investigated (reviewed in Downey et al., 2006; Teixeira et al., 2013). However, those studies were performed on a few elite cultivars (e.g., Cabernet-Sauvignon, Chardonnay, Syrah, Merlot. . . ) analyzed for a limited number of phenolic metabolites, most often anthocyanins. In addition, results are hardly comparable between studies since differences in water regime were not applied at the same developmental stage and with the same intensity, and amounts of phenolic compounds were not expressed in the same units. Since water stress induces a decrease of berry size, and given that most of phenolic compounds are stored in external cell layers of the cells, an increase of phenolic concentration expressed as mg/g of fresh weight can be measured without any increase of content expressed in mg/berry (Bucchetti et al., 2011). As a general trend, water stress was shown to induce an increase of anthocyanin content and a qualitative modification of the anthocyanin pool, when fine analysis was performed (Castellarin et al., 2007; Bucchetti et al., 2011; Ollé et al., 2011; Hochberg et al., 2015). In contrast, conflicting results were obtained on other classes of phenolic compounds. For example, no (Kennedy et al., 2002; Ollé et al., 2011) or slight (Ojeda et al., 2002) modifications in flavan-3-ol composition and a reduction (Hochberg et al., 2015; Savoi et al., 2017) or increase (Deluc et al., 2011; Herrera et al., 2017) of stilbene accumulation have been observed in response to water deficit. Cultivar specificity of these responses has been reported by comparing cv. Chardonnay (Deluc et al., 2009) or cv. Syrah (Hochberg et al., 2015) to cv. Cabernet Sauvignon. This may be related to hydraulic behavior or to differences in phenological stages (Hochberg et al., 2015) as early and late water deficit affect phenolic composition in different ways (Ojeda et al., 2002; Ollé et al., 2011; Casassa et al., 2015).

Nevertheless, a major challenge for breeding of grapevine cultivars adapted to climate change and with high potential for wine-making is to describe and dissect the complex global phenolic response involved in adaptation mechanisms on a wide range of genotypes. The aim of the present study was to investigate the polyphenol composition and its modification in response to water deficit on a large panel of cultivars reflecting the genetic diversity of grapevines.

#### MATERIALS AND METHODS

#### Plant Material and Experimental Design

The diversity panel (DP) of 279 V. vinifera cultivars described by Nicolas et al. (2016) was used for this study. It is composed of three subgroups of 93 cultivars representing the three main genetic pools, which differ in use and geographical origin: wine West (WW), wine East (WE), table East (TE).

Each cultivar was over-grafted in 2009 on 6-years old vines of cultivar Marselan in a complete randomized block design with five blocks and one plant of each cultivar per block. The trial was located at the Domaine du Chapitre of Montpellier Supagro (Villeneuve-les-Maguelonne, France), maintained under classical local training system (double cordon, 4,000 plants/ha). A drip irrigation was installed in two blocks in order to create a water contrast with the other three blocks. In 2014 and 2015, irrigation was applied 2 days per week from the last third of June to the end of the berry sampling period (October 6th and October 16th, in 2014 and 2015, respectively). The quantity of supplied water was approximately of 10 mm per 10-day period. Data on total rainfall per 10-day period were obtained for the nearest climatic station.

## Sampling

Grape berries were collected at ripeness when sugar concentration reached 20◦Brix. To determine this sampling stage, regular measurements (three times a week from week 30) were performed with an optical refractometer using a few berries per cultivar/treatment. Three clusters were sampled per cultivar/treatment, their end parts were discarded and 100 berries randomly sampled to estimate mean berry weight. Thirty berries were then randomly selected and their skins isolated, frozen in liquid nitrogen, and stored at −80◦C until extraction and analysis. The remaining berries were crushed and the juice was filtered. An aliquot of 1 mL was prepared for the analysis of the 13C/12C ratio (δ13C).

#### δ <sup>13</sup>C Analysis

δ <sup>13</sup>C or carbon isotope discrimination is expressed compared to a standard and ranges at maturity stage from −27 p. 1000 (no water deficit) to −20 p. 1000 (severe water deficit stress, Van Leeuwen et al., 2001). Its measurement was subcontracted. Samples were freeze-dried, pre-weighed, encapsulated, and then sent to OEA Laboratories Limited (Cornwall, UK). They were analyzed by a Sercon 20-20 dual turbo pumped Continuous Flow Isotope Ratio Mass Spectrometer (CF/IRMS) linked to a Thermo EA1110 Elemental Analyzer (EA) NC dual tube configured fitted with a high performance Carbosieve G separation column. Samples and references were weight optimized for δ13C analysis according to elemental composition. IRMS calibration was scale normalized using isotope references USGS-40 and USGS-41a as lower and upper scale anchors with random QC sample checks within sample sequences. Absolute weights of carbon in samples were determined from the IRMS total beam values relative to the elemental composition of the references. References were weighed from bulk material to 6 decimal places using a Mettler UMX5 microbalance. Standard deviations for isotope reference materials was typically better than 0.15 for carbon.

#### Extraction and Sample Preparation for Polyphenol Analysis Extraction

The extraction procedure was adapted from that of Mané et al. (2007), as described by Pinasseau et al. (2016). Briefly, frozen skins were ground with liquid nitrogen with a Mortar Grinder Pulverisette 2 (Fritsch, Idar-Oberstein, Germany). One hundred milliliters of of powder were weighed and 500 µL of methanol was immediately added. Then 3.5 mL of acetone/H2O 70/30 (v/v) 0.05% trifluoroacetic acid were added. The mixture was crushed with Precellys (Bertin Technologies, Montigny-le-Bretonneux, France) during three cycles (3 × 40 s each). 3.5 mL were centrifuged with a Heraeus Multifuge X3R Centrifuge (ThermoFischer Scientific, Waltham, USA) (21,320 g, 5 min, 4 ◦C). Aliquots (1 mL) of the supernatant were dried with Genevac (SP Scientific, Warminster, PA, USA).

#### Sample Preparation for Determination of Polyphenol Composition

Five hundred microliters of methanol/H2O 50/50 (v/v) 1% formic acid were added on the solid obtained after evaporation with Genevac (SP Scientific, Warminster, PA, USA). After solubilization using an Ultrasonic Cleaner (VWR, Fontenay-sous-Bois, France) (30 min), the solution obtained was centrifuged with Hettich Mikro 220R (Hettich Lab Technology, Tuttlingen, Germany) (15,000 rpm, 15 min, 4◦C). Dilutions 1/20 were prepared. Pure and diluted samples were injected in triplicate for UHPLC-QqQ-MS analysis.

The phloroglucinolysis reaction was carried out on the solid obtained after evaporation with Genevac (SP Scientific, Warminster, PA, USA), following the procedure described in Pinasseau et al. (2016).

#### Instrumentation

Analyses were carried out using an Acquity UPLC system (Waters, Saint-Quentin-en-Yvelines, France) hyphenated to a triple quadrupole (QqQ) TQD mass spectrometer (Waters, Saint-Quentin-en-Yvelines, France). The UPLC system included a binary pump, a cooled autosampler maintained at 7◦C and equipped with a 5-µL sample loop, a 100-µL syringe and a 30 µL needle, and a diode array detection (DAD). The DAD spectra were recorded in the range of 210–600 nm (resolution 1.2 nm). MassLynx software was used to control the instruments and to acquire the data which were then processed with the TargetLynx software.

#### Chromatographic Conditions

The column used for chromatographic separation was a reversedphase Acquity HSS T3 1.8µm 1.0 × 100 mm (Waters, Saint-Quentin-en-Yvelines, France) protected by a 0.2µm in-line filter and maintained at 40◦C. The mobile phase consisted of 1% (v/v) formic acid in deionized water (solvent A) and 1% (v/v) formic acid in methanol (solvent B). The flow rate was 0.170 mL/min. Samples were injected into the column by using the Partial Loop with Needle Overfill injection mode with an injection volume of 1 µL.

#### **UPLC analysis of polyphenol composition**

Isocratic 1%B from 0.0 to 2.0 min, linear 1–5%B from 2.0 to 2.1 min, linear 5–10%B from 2.1 to 8.0 min, linear 10–28%B from 8.0 to 12.0 min, isocratic 28%B from 12.0 to 18.0 min, linear 28–45%B from 18.0 to 22.0 min, linear 45–99%B from 22.0 to 23.5 min, isocratic 99%B from 23.5 to 26.5 min. At the end of this sequence, the column was brought back to initial conditions with linear 99–1%B from 26.5 to 27.0 min, then re-equilibrated with isocratic 1%B from 27.0 to 30.0 min.

#### **UPLC Analysis of tannin Units after phloroglucinolysis**

Isocratic 2%B from 0.0 to 1.5 min, linear 2–7%B from 1.5 to 3.0 min, linear 7–40%B from 3.0 to 5.0 min, linear 40–99%B from 5.0 to 6.0 min, isocratic 99%B from 6.0 to 6.5 min. As the end of this sequence, the column was brought back to initial conditions with linear 99–2%B from 6.5 to 7.0 min, then re-equilibrated with isocratic 1%B from 7.0 to 10.0 min.

#### Mass Spectrometry Conditions

The mass spectrometer was operated in MRM mode with electrospray ionization (ESI) either in positive or negative ionization mode. The source and desolvation temperatures were respectively set at 120 and 450◦C. Nitrogen was used as desolvation (500 L/h) and cone (50 L/h) gas. Argon was used as collision gas at a flow rate of 0.16 mL/min. Capillary voltage was set at 3.5 kV in positive mode and 2.8 kV in negative mode.

### Polyphenol Composition Data

Lower molecular weight phenolic compounds including phenolic acids, stilbenes, anthocyanins, flavonols, dihydroflavonols, flavan3-ol monomers, dimers and trimers, and derived pigments and tannins, were analyzed by UHPLC-QqQ-MS in the MRM mode, using a method adapted from that described by Lambert et al. (2015). A few additional phenolic compounds detected in the grape extracts were identified and included in the method as detailed below. Glutathione in its reduced and oxidized forms was analyzed by UHPLC-QqQ-MS in the MRM mode as described by Vallverdú-Queralt et al. (2015). Flavan-3 ol units released after phloroglucinolysis were analyzed by UHPLC-QqQ-MS in the MRM mode (Lambert et al., 2015).

MRM transitions parameters of added target compounds that are commercially available were optimized by using the Intellistart tool of the Masslynx software which consists in automatically detecting the major fragments and optimizing cone voltages and collision energies. 1-galloyl-β-D-glucose (glucogallin) was characterized by the loss of glucose (−162Th). The main fragment (m/z 139Th) of (-)-epigallocatechin was the result of a Retro-Diels-Alder (RDA) fragmentation. Piceatannol was characterized by the loss of a diphenol (−110Th). These three molecules and quercetin-3-O-glucuronide were included in the calibration standards.

For new target analytes that are not commercially available, MRM parameters were optimized directly in grape extracts and compared to data reported in the literature. Pelargonidin 3-glucoside was characterized by the loss of glucose (−162Th) (Arapitsas et al., 2012) while pelargonidin 3-acetylglucoside and pelargonidin 3-coumaroylglucoside were characterized by the loss of the acetylglucose (−204Th) and coumaroylglucose (−308Th), respectively. These fragmentation patterns were specifically targeted in accordance with those of the other anthocyanins described by Lambert et al. (2015). Transitions of (epi)gallocatechin-malvidin 3-glucoside and (epi)gallocatechinpeonidin 3-glucoside were specifically targeted in accordance with their non galloylated equivalents described in Lambert et al. (2015). They are characterized by the loss of glucose (−162Th). Analysis of anthocyanin-tannin (A-T) bicyclic A-type adducts, namely peonidin 3-glucoside-(epi)catechin (m/z 753Th), petunidin 3-glucoside-(epi)catechin (m/z 769Th), malvidin 3-glucoside-(epi) catechin (m/z 783Th), malvidin 3-glucoside- (epi)gallocatechin (m/z 799Th), was optimized in the same way. The main fragments detected at m/z 313, 329, and 343Th, respectively for peonidin, petunidin, and malvidin derived A-T adducts result from a retro Diels-Alder (RDA) fragmentation (−168Th), the loss of the anthocyanin A-ring (−126Th) and that of the glucose substituent (−162Th) (Remy-Tanneau et al., 2003). Malvidin 3-glucoside dimer and malvidin 3-glucosidepeonidin 3-glucoside were characterized by the loss of the two glucose moieties (−324Th) (Vidal et al., 2004b). Glucosylated flavonols such as isorhamnetin 3-glucoside, kaempferol 3 glucoside, and syringetin 3-glucoside were qualified by the loss of the glucose (−162Th) (Vrhovsek et al., 2012). Fragmentation of laricitrin 3-glucoside (−162Th) was specifically targeted in accordance with fragmentation patterns of the other glucosylated flavonols (Lambert et al., 2015). Kaempferol 3-glucuronide was qualified by the loss of the glucuronide (−176Th) (Vrhovsek et al., 2012). Fragmentations of isorhamnetin 3-glucuronide, laricitrin 3-glucuronide, and syringetin 3-glucuronide (loss of the glucuronide −176Th) were optimized in accordance with the fragmentation pattern of the other glucuronidated flavonols (Lambert et al., 2015). Piceatannol 3-glucoside was characterized by the loss of the glucose (−162Th) (Vrhovsek et al., 2012). Fragmentation of (+)-gallocatechin (fragment at m/z 139Th after a RDA fragmentation) was optimized in accordance with (–)-epigallocatechin which is commercially available. Anthocyanins were expressed as equivalent malvidin 3-O-glucoside. Flavonol glucosides and flavonol glucuronides were expressed as equivalent quercetin 3-glucoside and quercetin 3-glucuronide, respectively. Piceatannol glucoside was expressed as equivalent piceid.

Quantitative data on 105 compounds was thus obtained. In addition, 17 variables have been calculated, including quantitative variables, namely total concentrations of native anthocyanins (s\_AN\_n), flavonols (s\_FO), stilbenes (s\_ST), hydroxycinnamic acid derivatives (s\_HC), hydroxybenzoic acid derivatives (s\_HB), flavan-3-ols (i.e., sum of tannin units released after phloroglucinolysis, s\_FA), and qualitative variables, %acylated anthocyanins (p\_AN\_acyl), %B-ring trihydroxylated anthocyanins (p\_AN\_tri), %B-ring methylated anthocyanins (p\_AN\_met), %B-ring monohydroxylated flavonols (p\_FO\_mono), %B-ring dihydroxylated flavonols (p\_FO\_di), %B-ring trihydroxylated flavonols (p\_FO\_tri), %Bring methylated flavonols (p\_FO\_met), %flavonol glucuronides (p\_FO-glucur), %B-ring trihydroxylated flavan-3-ol units (p\_FA\_tri) %galloylated flavan-3-ol units (p\_FA\_gall), mean degree of polymerization (dp\_FA), calculated as the molar ratio of total released units to total terminal units. The list of variables is given in **Table 1**, along with their codes and abbreviations.

#### Chemometrics

For the 2 years of sampling (2014 and 2015), chemometrics treatments were performed on the MRM data for the 105 compounds, sorted by families (same order in 2014 and 2015) anthocyanins, flavanols, stilbenes, etc. For each observation, the 105 compounds were associated to the 17 calculated parameters, and the three parameters from the vineyard: δ <sup>13</sup>C, refractive index, berry weight. Only cultivars for which both irrigated and non-irrigated observations were available were considered in each vintage. Samples with missing berry weight values were also eliminated. For the 105 MRM parameters, values below the quantification threshold were automatically replaced with a value corresponding to half of the threshold value.

#### TABLE 1 | list of variables, variable codes, and abbreviations.




(Continued)

(Continued)

#### TABLE 1 | Continued



PPt,Pn,Mv = AN4+AN5+AN6+AN9+AN10+AN11+AN15+AN16+AN17+AN21 +AN22+AN23+AN26+AN27+AN28

Pkaempf = FO9+FO10

Pquerc + Isorham = FO3+FO4+FO11+FO12

Pmyric+laric+syring = FO1+FO2+FO5+FO6+FO7+FO8 Pisorham+laric+syring = FO1+FO2+FO7+FO8+FO11+FO12

Pglucuronides = FO1+FO3+FO3+FO7+FO9+FO11.

One-way analysis of variance and principal component analysis were performed using the Fact toolbox of the Scilab software. Correlation networks were processed using Cytoscape. Hierarchical clustering of phenolic compounds and genotypes was performed using EXPANDER V6 (Sharan et al., 2003). The distance measurement used in the algorithm is (1-Pearson Correlation)/2, with complete linkage.

#### Reagents and Chemicals

Formic acid, HPLC grade methanol, acetone, hydrochloric acid, trifluoroacetic acid, ammonium formiate, L-ascorbic acid, and phloroglucinol were purchased from Sigma Aldrich (St Louis, MO, USA). Deionized water was obtained from a Milli-Q purification system (Millipore, Molsheim, France). Standards of (+)-catechin, (–)-epicatechin, (–)-epicatechin 3-O-gallate, reduced L-glutathione, oxidized L-glutathione, piceatannol, p-coumaric acid, protocatechuic acid, syringic acid, trans-caftaric acid, and trans-resveratrol were purchased from Sigma-Aldrich (St Louis, MO, USA). Standards of (–)-epigallocatechin, gallic acid, hydroxytyrosol, malvidin 3-O-glucoside chloride, malvidin

code formula Full name

(Continued)

3,5-di-O-glucoside chloride, procyanidin B2, quercetin 3-Oglucuronide, and taxifolin were purchased from Extrasynthese (Geney, France). Standards of caffeic acid, ferulic acid, and vanillic acid were purchased from Fluka (Buchs, Switzerland). Standards of 1-O-Galloyl-β-D-glucose and quercetin 3-Oglucoside were purchased from PlantMetaChem, Transmit GmbH (Gießen, Germany). Standard of trans-piceid was purchased from Selleckchem (Houston, TX, USA).

## RESULTS

## Genetic Diversity of Polyphenol Composition

After elimination of cultivars for which both samples were not available and/or essential data such as berry weight was missing, complete data was obtained for 208 cultivars in 2014, for 161 cultivars in 2015, and for 147 cultivars in both years. The list of samples collected in 2014 and 2015 and their harvest dates is provided in **Table S1**. Data for all cultivars in both vintages are available in Pinasseau et al. (2017).

Large differences in the phenolic composition were observed between cultivars. Tannins were very abundant in all cultivars with concentrations ranging from 0.4 to 7.5 mg berry−<sup>1</sup> in 2014, and over 12 mg berry−<sup>1</sup> in 2015. Anthocyanin contents ranged from less than 1µg berry−<sup>1</sup> in white cultivars to 8.5 and 14.7 mg berry−<sup>1</sup> , respectively in 2014 and 2015. Flavonols, and especially quercetin derivatives (quercetin 3-glucoside and quercetin 3-glucuronide), were also abundant, with concentrations ranging from 0.04 to over 6 mg berry−<sup>1</sup> in 2014 and from 0.06 to over 5 mg berry−<sup>1</sup> in 2015. Other polyphenol classes were hydroxycinnamic acids (8–2,000µg berry−<sup>1</sup> ) mostly represented by caftaric and coutaric acids, stilbenes (1–745µg berry−<sup>1</sup> ), among which cis- and trans- piceid and trans-resveratrol were the most abundant, dihydroflavonols (trace amounts to 196µg berry−<sup>1</sup> ), and hydroxybenzoic acids (trace amounts to 25µg berry−<sup>1</sup> ). A number of anthocyanin derivatives were also detected. Most of them (i.e., carboxypyranoanthocyanins; e.g., carboxypyranomalvidin 3-glucoside, called vitisin A, caftaric anthocyanin adducts, and series of flavanolanthocyanin, anthocyanin-flavanol, anthocyanin-ethyl-flavanol, and flavanol-ethyl flavanol adducts), were present in low amounts, except pyranoanthocyanins resulting from reaction of acetaldehyde with anthocyanins, especially pyranomalvidin 3-glucoside (vitisin B), detected at concentrations up to 400µg berry−<sup>1</sup> .

**Table 2** shows the correlation coefficients between irrigated and not irrigated populations, in 2014 and 2015, and between vintages for irrigated and not irrigated samples, for each of the 17 calculated polyphenol composition variables and for berry weight. Berry weight was highly correlated across all four conditions, as expected. The contents (per berry) and concentrations (per g of berry) of all polyphenol families, except flavonols and stilbenes, in irrigated and not-irrigated berries were highly correlated in 2014 but not in 2015. Correlations between years were low under both conditions. In contrast, for all qualitative variables, correlations between irrigated and not irrigated conditions were very high and correlations between years were only slightly lower.

Correlation networks established from the phenolic composition data showed several clusters. Correlations >0.8 are presented in **Figure 1**. The content of malvidin 3 glucoside was correlated on one hand with those of delphinidin 3-glucoside, petunidin-3-glucoside, and of their coumaroyl and caffeoyl derivatives and, on the other hand, with those of some anthocyanin derivatives [pyranomalvidin 3-glucoside, carboxypyranomalvidin 3-glucoside, (epi)gallocatechinmalvidin 3-glucoside, (epi)catechin-malvidin 3-glucoside, and (epi)catechin-petunidin 3-glucoside] (**Figure 1**, **A**). Pelargonidin, cyaniding, and peonidin 3-glucosides were correlated together and with peonidin derivatives, namely pyranopeonidin 3-glucoside and (epi)catechin–peonidin 3 glucoside (**B**) while their caffeoyl and p-coumaroyl esters formed another group (**C**). All acetylated anthocyanins were correlated together in a separate cluster (**D**). Other types of anthocyanin pigments, namely anthocyanin 3,5-di-O-glucosides (**E**), anthocyanin dimers (**F**), phenylpyranoanthocyanins correlated between them and with caftaric-anthocyanin adducts (**G**) and the different isomers of (epi)catechin-ethyl-peonidin−3 glucoside and (epi)catechin-ethyl-malvidin 3-glucoside (**H**) formed additional groups. Flavonols clustered in three different groups consisting of kaempferol and quercetin 3-glucosides (**I**), myricetin, laricitrin, and syringetin 3-glucosides (**J**), and laricitrin and syringetin 3-glucuronides (**K**), respectively. Stilbene glucosides (cis- and trans- piceids and piceatannol glucoside) formed another correlation network (**L**). Flavan-3-ol variables formed three clusters: (epi)catechin monomers and terminal units (**M**), (epi)gallocatechin terminal units (**N**), and (epi)catechin phloroglucinol derivatives (extension and upper units in the tannin structures) (**O**).

# Vine Water Status in 2014 and 2015

Information from the rain and irrigation data and from the measures of δ <sup>13</sup>C and berry weight was combined to characterize the vine water status during the vegetative seasons 2014 and 2015. Bar plots showing water quantities supplied by rainfall and irrigation are provided in **Figure 2**, showing that the total quantity of rainfall received within the plot trial the preceding winter and spring was very different After including data from 2013 (data not shown), the total rainfall received from November to the second third of June (before irrigation started) was 187.5 and 460.5 mm for 2014 and 2015, respectively. Another notable difference between the two vegetative seasons was the earlier occurrence of summer rainfall in 2015 as compared to 2014 (**Figure 2**).

#### Cultivar Response to Water Stress in Vintages 2014 and 2015

A first round of statistical analysis was performed with oneway ANOVA analysis on the four data sets (irrigated and notirrigated, 2014 and 2015) available for 147 cultivars (Table S2). The absence of significant differences (at p = 0.05) in refractive index values between conditions in both years confirmed that


TABLE 2 | Stability of polyphenol composition data and berry weight; correlations between irrigated and not irrigated berries in 2014 (2014 I/NI) and 2015 (2015 I/NI), and between 2014 and 2015 berries, under irrigated (I\_2014/2015) and not irrigated (NI\_2014/2015) conditions.

\*Calculated with colored (black, red, and pink) cultivars only.

berries were actually collected at the same developmental stages, while differences between years indicated a slight vintage effect. However, large phenotypic diversity was observed on berry weight (**Figure 3**). Water deficit induced a slight shift toward smaller berry size in 2014, with the major class below 1.5 g and between 1.5 and 2.5 g per berry, respectively, in non-irrigated and irrigated berries. Distribution of berry sizes was not impacted by irrigation in 2015. Large variations were also observed for δ <sup>13</sup>C values within the collection (**Figure 4**). Irrigation induced larger shifts in 2014 than in 2015 and the whole population showed much lower values in 2015 than in 2014, regardless of the irrigation regime. Berry weight was significantly lower in not-irrigated berries in 2014 but not in 2015. Irrigation induced significant differences on the δ <sup>13</sup>C values in both vintages, but water stress was much lower in 2015with δ <sup>13</sup>C values significantly higher than in 2014. Taken together, these data indicate that irrigation induced a marked contrast in 2014 but a very limited one in 2015.

ANOVA analysis performed on the 105 polyphenol variables expressed in mg per g of berry (Table S2) showed that most tannins and flavonols and of their total concentrations were significantly reduced by irrigation in 2014 but not in 2015. In 2015, the concentrations of cis-resveratrol and piceatannol were significantly increased by irrigation and that of glucogallin was significantly reduced. Significant vintage effect was also found on over 50 compounds, with significantly higher levels in 2015 for the majority of them, except gallocatechin and epigallocatechin which were more abundant in 2014.

When the analysis was performed on the data expressed per berry (Table S2), no significant difference was found in the levels of phenolic compounds between irrigated and not irrigated conditions in 2015 whereas seven compounds from the flavan-3-ol family and oxidized glutathione were significantly increased by irrigation in 2014. Numerous compounds, distributed within all polyphenol families, were significantly higher in 2015 than in 2014, as well as total flavan-3-ols, flavonols, hydroxybenzoic acids, and hydroxycinnamic acids.

One way ANOVA was also performed separately on the complete 2014 and 2015 data sets (**Table 3**). There was no statistically significant difference between irrigated and not-irrigated conditions (at p = 0.05) in 2015 on polyphenol composition. In contrast, in 2014, irrigation induced significant changes in the content (per berry) of 16 polyphenols and in the concentration (per g of berry) of 47 compounds.

Taken together, these results indicate that berries were probably not exposed to any sufficient water stress regime in 2015 to induce changes in their phenolic composition. Consequently, data from 2015 were not further explored in this study.

# Impact of Water Stress on Polyphenol Composition

Principal component analysis (PCA) was performed on the phenolic composition data of all berry skin samples collected in 2014, expressed in mg per g of fresh berry. Projection of the samples on the first two principal components, accounting together for 37% of the variance, showed large cultivar differences, as well as a strong impact of irrigation (**Figure 5A**). White and red cultivars were separated along the first axis which was negatively correlated with the concentrations of most phenolic compounds, including anthocyanins, especially delphinidin, petunidin, and malvidin 3-glucosides, myricetin, laricitrin, and syringetin glycosides, hydroxybenzoic acids, especially gallic and syringic acids, and epigallocatechin, both in the free form and as terminal units of proanthocyanidins (**Figure 5B**). Non-irrigated samples generally appeared shifted negatively along the first axis, indicating that they contained higher levels of these molecules.

ANOVA analysis of variance performed on the polyphenol composition data set expressed per g berry (**Table 3**) indicated that berries from irrigated vines contained significantly lower concentrations of the cis isomers of resveratrol and piceid, of all tannin units determined after phloroglucinolysis, and of most benzoic acids, hydroxycinnamic acids, and flavonols. The concentrations of some anthocyanins, namely 3-glucosides of pelargonidin, delphinidin, petunidin and malvidin, cyanidin 3,5 diglucoside and petunidin 3,5-diglucoside were also significantly decreased, as well as those of some anthocyanin derivatives, namely pyranoanthocyanins, tannin-anthocyanin adducts, and caftaric anthocyanin adducts. Other variables such as the concentrations of flavan-3-ol monomers were not significantly modified.

When PCA was performed on the phenolic composition data expressed per berry (**Figure 6A**), most samples appeared shifted along the first and/or second axis, but in different directions. Again, white cultivars were separated from red cultivars along the first axis, which was negatively associated with the same phenolic compounds as in the previous PCA (**Figure 6B**).

When the data was expressed per berry, 16 compounds were significantly increased in berries from irrigated samples (**Table 3**). Thus, water stress induced a significant decrease of the biosynthesis of catechin and epicatechin, both as flavan-3-ol monomers and as constitutive units of proanthocyanidins, total flavan-3-ols, phenyl- and catechyl-pyranoanthocyanins, caftaricanthocyanin adducts, (epi)catechin-ethyl-malvidin-3- glucoside, caffeic acid, and piceatannol. Moreover, some qualitative flavan-3-ol variables, namely tannin mDP, and % trihydroxylated tannin units were significantly reduced by irrigation.

Unsupervised hierarchical clustering of metabolites and cultivars affected by drought was performed on the response of polyphenol composition to water status, with data expressed as

log (irrigated/non-irrigated), with polyphenol contents expressed per berry. The resulting plot (**Figure 7**) shows different response patterns for different cultivars and for the different groups of analytical variables. Groups of compounds whose content varies in the same direction in response to irrigation can be distinguished. Cluster **a** contained mainly mono- and di-hydroxylated flavonols and dihydroflavonols (astilbin and engeletin, respectively mono- and dihydroxylated on the Bring). The most abundant flavanol subunits (and also the sum of tannins) were grouped in cluster **b**, and linked with cluster **c** containing anthocyanin-flavanol derivatives linked with an ethyl-bridge. Cluster **d** grouped hydroxycinnamates and several of their reaction products with anthocyanins (pyranoanthocyanins and caftaric-anthocyanins). Most of the flavanol monomers and terminal units are clustered in the close **e1** and **e2**. Clusters**f1** and **f2** contained respectively mono-and dihydroxylated anthocyanins along with some of their derivatives and trihydroxylated anthocyanins. The latter encompassed **g1** and **g2,** containing trihydroxylated flavonols. It is also noticeable that β-glucogallin was included in **f2**. All stilbenes shared the same response to irrigation and were clustered in cluster **h**.

The same data (log (irrigated/non-irrigated), calculated from polyphenol concentrations expressed in mg berry−<sup>1</sup> ) was used to establish the correlation network shown in **Figure 8**. Only the correlations >0.8 are presented. Major clusters corresponded to stilbenes (**A**), native anthocyanins derived from delphinidin, petunidin, and malvidin (**B**), from peonidin (**C**), and from pelargonidin (**D**), caftaric and coutaric acids (**E**), kaempferol and quercetin 3-glucosides (**F**), catechin and gallocatechin monomers (**G**), (epi)catechin units of tannins (**H**), (epi)gallocatechin units of tannins (**I**), and anthocyanin derivatives, especially phenylpyranoanthocyanins and caftaric-anthocyanin adducts (**J**). Two additional clusters consisted of pyranopeonidin 3 glucoside, cyanidin 3-acetylglucoside and cluster D (**K**) and pyranomalvidin 3-glucoside with the 3-acetylglucosides of delphinidin and petunidin (**L**).

Berry anthocyanin, flavonol, hydroxycinnamic acid, and stilbene contents were increased or decreased under irrigated conditions in some cultivars. Groups of cultivars whose composition varies similarly in response to irrigation are clustered together (**Figure 7**). For example, irrigation resulted in increased and decreased stilbene levels in most cultivars of group

1 and group 2, respectively. The opposite pattern was observed for flavan-3-ols. The distribution of colored (i.e., black, red, and pink) cultivars and white cultivars, and that of the three genetic groups (WW, WE, TE) in some of the subgroups defined by unsupervised hierarchical clustering has been compared to that of the whole population (**Figure 9**). Chi2 tests performed on each subgroup showed that colored cultivars are overrepresented in subgroups 1-2-1 and 2-1-1 and underrepresented in subgroup 2-2-2 and cultivars from WW origin are overrepresented in subgroup 2-1-2 (Table S3). Although both contain mostly colored cultivars, subgroups 1-2-1 and 2-1-1 show different response to irrigation, with decreased anthocyanins (**Figure 7**, **f1,f2**), Bring trihydroxylated flavonols (**g1, g2**) and stilbenes (**h**) in the latter, and reduced tannins (**b, e1, e2**) and increased stilbenes (**h**) in the former. Distributions of the harvest dates for each group under irrigated and not-irrigated conditions were also examined (**Figure 10**)**.** Chi-2 tests (Table S3) performed on the entire population showed that the distribution of harvest dates was similar for all subgroups under irrigated conditions but significantly different under not irrigated conditions. Chi-2 test values calculated for each subgroup indicated that the distribution of harvest dates in some of them was significantly different from that of the whole population, although the difference was significant at p = 0.05 only for 2-2-1. Thus, not irrigated cultivars of subgroups 1-1-2 and 1-2-1 and cultivars of subgroups 2-1-1, 2-2-1, and 2-2-2 were harvested earlier and later, respectively, and cultivars of subgroup 1-2-1 were shifted toward later harvest dates under irrigated conditions.

# DISCUSSION

## Cultivar Differences in the Polyphenol Composition of Grape Berry Skins

Major polyphenol families detected in berry skin samples were flavan-3-ols, including monomers and proanthocyanidins, anthocyanins, flavonols, hydroxycinnamic acids, and stilbenes, along with lower amounts of dihydroflavonols and benzoic acids, as classically reported. All families showed wide ranges of concentrations across the diversity panel. Anthocyanin contents enable distinction between white, pink, and red cultivars (Castellarin and Di Gaspero, 2007; Pelsy, 2010) although white grape berries also contain trace amounts of anthocyanin pigments (Arapitsas et al., 2015). Red cultivars were also characterized by the presence of flavonols with trihydroxylated B-rings, i.e., derived from myricetin, laricitrin, and syringetin glycosides, which are known to be specific of red cultivars (Mattivi et al., 2006). Cultivar differences in hydroxycinnamic acid contents have also been reported and related to differences in cultivar susceptibility to enzymatic browning (Cheynier et al., 1990). Genetic determinism of flavonols has been studied through QTL analysis (Malacarne et al., 2015), but this information is still lacking for stilbenes. However, their concentrations are also believed to highly depend on environmental factors as they are involved in plant defense against UV exposure and fungal attacks (Teixeira et al., 2013). It is noteworthy that some of the samples contained very high levels of flavonols and especially of quercetin and kaempferol

derivatives compared to values reported earlier (Mattivi et al., 2006). This may be related to environmental conditions as berry concentration of quercetin glycosides have been shown to increase dramatically following sunlight exposure (Price et al., 1995; Spayd et al., 2002; Downey et al., 2004). Similarly, the lack of correlation between the flavan-3-ol contents of berries collected from a population of 141 grapevines cultivars over 2 successive years indicated that tannin accumulation is mostly driven by environmental factors rather than genetically determined (Huang et al., 2012b). In contrast, qualitative profiles within the different polyphenol groups are known to be cultivar characteristics. Thus, chemotaxonomic approaches based on grape anthocyanin profiles (Roggero et al., 1988; Mazza, 1995; Fournier-Level et al., 2009) or hydroxycinnamic acid profiles (Boursiquot et al., 1986) have been proposed. Skin flavan-3 ol composition also appeared highly conserved between years, meaning that it is mostly linked to genetic factors (Huang et al., 2012b). Our data, showing high correlations between years and between irrigation regimes for qualitative polyphenol variables and low correlations as well as strong vintage effect for quantitative variables (**Table 2**), confirm that the polyphenol profiles depend on cultivar while contents are affected by environmental factors, as reported in the above cited literature.

In addition to the expected native anthocyanins, several anthocyanin derivatives were detected. Among them, caftaric anthocyanin adducts were present only in trace amounts. As these adducts result from enzymatic oxidation catalyzed by grape polyphenoloxidase (Sarni-Manchado et al., 1997), this indicates that no enzymatic oxidation took place during sample preparation. Pyranoanthocyanins and carboxypyranoanthocyanins resulting from reaction of anthocyanins respectively with acetaldehyde (Cheynier et al., 1997) and pyruvic acid (Fulcrand et al., 1998) have been reported in grape (Arapitsas et al., 2015). Anthocyanin dimers have also been isolated from grape skins (Vidal et al., 2004b). Strong correlations between the levels of malvidin-3-glucoside and peonidin 3 glucoside and those of vitisin B and pyranopeonidin 3-glucoside, respectively (**Figure 1**), substantiate the hypothesis that these compounds are formed in vivo. Moreover, the high level of vitisin B detected in some cultivars indicates that acetaldehyde is present in subcellular compartments in rather large amounts together with anthocyanins. Two major groups of tannin-anthocyanin reaction products were also detected. Flavanol-anthocyanin adducts resulting from cleavage of tannins followed by addition with anthocyanins have been detected in wine (Salas et al., 2004) and in various fruits including grapes (Gonzalez-Paramas et al., 2006). Flavanol-ethyl-anthocyanins resulting from condensation of anthocyanins and flavanols with acetaldehyde are well known to occur in wine (Timberlake and Bridle, 1976; Arapitsas et al., 2012) and have been detected in cranberry extracts (Tarascou et al., 2011). Reactions of (epi)catechin, anthocyanins, and acetaldehyde yield complex mixtures of TABLE 3 | Results of the ANOVA performed on the data of irrigated (I) and non-irrigated (NI) vines separately on 2014 and 2015; polyphenol composition data in and microgram per berry and microgram per g of berry; variable codes are provided in Table 1.


(Continued)

#### TABLE 3 | Continued


(Continued)

#### TABLE 3 | Continued


<sup>a</sup>Variable codes as in Table 1.

\*SNK: results of the Student-Newman-Keuls grouping (p < 0.05).

<sup>b</sup>A few missing values have been removed from the calculation.

products, including pyranoanthocyanin, (epi)catechin-ethylanthocyanin, and (epi)catechin-ethyl-(epi)catechin derivatives (Vallverdú-Queralt et al., 2017a,b). Molecules of the last group have been detected in wine (Cheynier et al., 1997) but this is the first report of their presence in grape. Flavanol-anthocyanins correlated with their anthocyanin precursors (**Figures 1**, **A** and **B**) while flavanol-ethyl-anthocyanins formed a specific cluster (**Figure 1**, **H**). Although these molecules could also form during sample preparation, the levels reported here and the relatively long reaction rates compared to the duration of our extraction procedure suggest that they were present in planta.

Other correlations networks established for the 2014 data only (not shown) showed additional relationships between malvidin-3-glucoside and anthocyanin dimers (malvidin 3-glucoside and malvidin 3-glucoside–peonidin 3-glucoside), and syringic acid that arises from degradation of malvidin (Furtado et al., 1993; Vallverdú-Queralt et al., 2016) and between peonidin 3-glucoside and vanillic acid. Formation of syringic and vanillic acids respectively from malvidin 3-glucoside–peonidin 3-glucoside can be promoted by light and heat exposure (Furtado et al., 1993).

Correlations between variables (**Figure 1**) can be interpreted in terms of biosynthetic pathways. Indeed, anthocyanin and

FIGURE 5 | PCA of the MRM phenolic composition data of berry skin samples collected in 2014 (mg g−<sup>1</sup> ); (A), projection of the samples on PC1 and PC2; red and white cultivars are represented in red and in green, respectively; IR, irrigated, NI, not-irrigated. (B), loadings of the variables (coded as in Table 1) on PC1. AN, native anthocyanins+dimers; AP, pyrano anthocyanins; AF, anthocyanin-flavanol adducts; AC, caftaric-anthocyanin adducts; HF, dihydroflavonols; FO, flavonols; ST, stilbenes; FA, flavanols (tannins); HB, hydroxybenzoic acids; HC, hydroxycinnamic acids; OT, others.

FIGURE 6 | PCA of the MRM phenolic composition data of berry skin samples collected in 2014 (mg berry−<sup>1</sup> ); (A), projection of the samples on PC1 and PC2; red and white cultivars are represented in red and in green, respectively; IR, irrigated; NI, not-irrigated. (B), loadings of the variables (coded as in Table 1) on PC1. AN, native anthocyanins+dimers; AP, pyrano anthocyanins; AF, anthocyanin-flavanol adducts; AC, caftaric-anthocyanin adducts; HF, dihydroflavonols; FO, flavonols; ST, stilbenes; FA, flavanols (tannins); HB, hydroxybenzoic acids; HC, hydroxycinnamic acids; OT, others.

flavonol variables clustered together according to their B-ring substitution (trihydroxylated/dihydroxylated) and/or acylation pattern. Clustering of those compounds according to their B-ring hydroxylation pattern is consistent with the ability of F3'H and F3'5'H to use both anthocyanin and flavonols as substrate (Bogs et al., 2006). Flavonols clustered in three correlation networks corresponding to B-ring trihydroxylated compounds (myricetin, laricitrin, and syringetin derivatives) and other (mono or dihydroxylated) flavonols and substitution by glucose or glucuronic acid. This probably reflects the high sugar specificity of the already described Vitis flavonol glycosyltransferases: when VvGT5 is quite exclusively a glucuronyl donor, VvGT6 catalyzes both flavonol glucosylation and galactosylation (Ono et al., 2010). Flavan-3-ols also clustered following their B-ring hydroxylation pattern, (epi)-catechin flavanol units, and (epi)gallocatechin units forming different groups. Strong correlations between terminal units and the corresponding monomers likely reflect the analytical method

as monomers contribute to terminal units. Moreover, upper units, detected as the corresponding phloroglucinol adducts, were separated from terminal units, suggesting that both types of units have different precursors, as already suspected (Stafford et al., 1982; Huang et al., 2012a). Acetylated anthocyanin derivatives were correlated, regardless of the anthocyanin B-ring substitution while glucosylated, coumaroylated, and caffeoylated anthocyanins derived from malvidin, petunidin, and delphinidin (trihydroxylated) and from other anthocyanidins formed distinct groups. Although the already characterized acyltransferase Vv3AT is able to use both aliphatic and aromatic acyl-CoA as substrate (Rinaldo et al., 2015), this suggests that anthocyanin acylation with acetic acid and with hydroxycinnamic acids could involve alternative biosynthetic mechanisms (Bontpart et al., 2015).

Finally, correlations of anthocyanins with their derivatives and degradation products (**Figure 1**, clusters **A** and **B**), and clustering of molecules such as anthocyanin dimers (**Figure 1**, **F**), flavanol-ethyl anthocyanins (**Figure 1**, **H**) and hydroxyphenyland catechyl-pyranoanthocyanins, resulting from anthocyanin reactions with p-coumaric and caffeic acid (**Figure 1**, **G**) reflect their formation from the same precursors and/or through identical reaction mechanisms.

### Impact of Water Deficit on the Polyphenol Composition of Grape Berry Skins

Not-irrigated vines suffered water stress in 2014 but not in 2015. Indeed, water stress classically induces a decrease in berry weight (Roby et al., 2004; Bucchetti et al., 2011). In 2015, irrigation had no significant impact on berry weight, which indicates that berries were not exposed to any sufficient water stress regime to induce phenotypic changes. Corroborating this hypothesis, in 2015, berry weights were higher and δ <sup>13</sup>C values were lower than those of berries from irrigated vines in 2014. Accordingly, none of the polyphenol variables showed significant differences between berries from irrigated and not irrigated vines (**Table 3**). In contrast, in 2014, water stress induced significant loss of berry weight as well as significant differences on the concentration of several polyphenols. This confirms that polyphenols are part of the chemical arsenal allowing adaptive response to abiotic stress, being protective molecules against oxidative damages by

scavenging Reactive Oxygen Species (ROS) produced during stress (Rontein et al., 2002).

Not irrigated berries contained higher levels of most phenolic compounds when expressed in mg g−<sup>1</sup> fresh berry weight (**Table 3**, **Figure 5)**. This concentration effect can be attributed to reduced berry size under water stress, as observed earlier (Roby et al., 2004; Bucchetti et al., 2011). However, shifts between irrigated and non-irrigated samples on the PCA performed on the phenolic composition data expressed in mg per berry (**Figure 6**) showed that water status affected polyphenol biosynthesis. Water deficiency is known to impact berry development, decrease berry weight, and modulate accumulation of secondary metabolites including polyphenols (Kennedy et al., 2002; Roby et al., 2004). Data available on a limited number of genotypes suggest that the response to moderate water stress differs depending on the level of irrigation and/or water stress, on the berry development stage when water deficit occurs and on the cultivar (Ojeda et al., 2002; Teixeira et al., 2013). Thus, several studies have shown an increase in the accumulation of stilbenes, flavonols, and anthocyanins and enhanced transcription of genes involved in these pathways following moderate water deficiency while other studies failed to observe these effects or even observed a decrease. In Syrah, water deficit applied before or after veraison resulted in an increase of total anthocyanin contents and differences in the anthocyanin profiles (Ollé et al., 2011). Data on the effect of environment on tannin biosynthesis is still scarce: water deficiency in Cabernet Sauvignon (Kennedy et al., 2002; Castellarin et al., 2007) or in Syrah (Ollé et al., 2011), or thermic variation in Merlot (Cohen et al., 2012) did not affect tannin accumulation. In another study, a decrease or increase in tannin accumulation was reported in Syrah exposed respectively to early (between anthesis and veraison) or late (after veraison) water stress (Ojeda et al., 2002). However, tannin accumulation might also be related to biotic stress exposure (Dixon et al., 2005). As well, the significant increase of cis-resveratrol and piceatannol concentrations observed in 2015 under irrigated conditions may be due to plant response to increased fungal pressure as stilbenes are known to be involved in defense against fungi (Jeandet et al., 2002).

ANOVA performed on the 2014 samples (**Table 3**) showed that the levels of 16 MRM variables expressed per berry were significantly higher in irrigated berries. Tannins were the major family affected by irrigation, with both quantitative (increase of catechin and epicatechin, detected as monomers, and as terminal and upper units of tannin chains, and of total flavan-3-ol levels) and qualitative (decrease of % B-ring trihydroxylated units and mean DP) variations. A decrease of tannin DP in irrigated vines has been reported earlier (Ojeda et al., 2002) and the proportion of B-ring trihydroxylated units was reduced in shaded berries (Cortell and Kennedy, 2006). Other affected compounds included piceatannol, caffeic acid, and pigments resulting from reactions of anthocyanins with hydroxycinnamic acids, i.e., caftaric-anthocyanin adducts, phenylpyranoanthocyanins and catechyl-pyranoanthocyanins.

Several groups of variables were affected by irrigation in the same way and formed clusters on the correlation networks established from the response of MRM variables to irrigation (log; irrigated/non-irrigated of the concentrations expressed in mg berry−<sup>1</sup> , **Figure 8)**. Thus, clustering of phenylpyranoanthocyanins, catechylpyranoanthocyanins, and caftaric-anthocyanin adducts indicated that they were simultaneously increased upon irrigation. Stilbenes formed another cluster, indicating that they were not only closely related, as shown by clustering of their concentrations (**Figure 1**, **L**) but also impacted in the same way by water deficit (**Figures 7**, **h**, **8**, **A**). Transcription of genes involved in the biosynthesis of stilbene precursors has been shown to increase and decrease in response to water stress in Cabernet Sauvignon and Chardonnay, respectively (Deluc et al., 2011).

Other clusters grouped together members of the different flavonoid families (anthocyanins, flavan-3-ols, and flavonols) which were further sorted according to their B-ring substitution pattern. Thus, malvidin, delphinidin, and petunidin derivatives (trisubstituted on the B-ring) and peonidin derivatives (disubstituted on the B-ring) formed different groups (**Figures 7**, f1 and **f2**, **8**, **B** and **C**). Water deficit has been shown to enhance expression of flavonoid 3′ ,5′ -hydroxylase (F3′ 5 ′H), involved in B-ring trihydroxylation, relative to that of flavonoid 3 ′ -hydroxylase (F3′H), involved in B-ring dihydroxylation, and, consequently, to increase the proportion of B-ring trihydroxylated anthocyanins (Castellarin et al., 2007). An increase of O-methyltransferase expression also correlated with accumulation of malvidin and peonidin derivatives (Castellarin et al., 2007). However, in another study, water deficit applied before and after veraison affected anthocyanin composition differently, enhancement of malvidin accumulation being observed only with post-veraison stress (Ollé et al., 2011). Among flavonols, drought responses of kaempferol and quercetin glucosides (respectively mono and dihydroxylated on the B-ring) were correlated (**Figures 7**, **a**, **8**, **F**). Quercetin glycosides have been shown to accumulate following UV exposure of the berry (Price et al., 1995) and are believed to play a role in UV protection. Interestingly, expression of F3′ 5 ′H and biosynthesis of B-ring trihydroxylated anthocyanins were reduced in tissues protected from light exposure by shading of the berries or accumulation of phenolic compounds acting as UV screens in external tissues (Guan et al., 2014). The presence of β-glucogallin in the same response group to water status as trihydroxylated flavonols and anthocyanins is unexpected. This compound is suspected to be an intermediate in the flavanol galloylation pathway (Bontpart et al., 2015). However, glucose ester of hydroxybenzoic acid was already described as glucose donor for flavonoid glucosylation (Nishizaki et al., 2013). In the flavanol family, (epi)catechin based upper tannin units clustered with terminal catechin (**Figure 8**, **H**) and (epi)gallocatechin tannin units formed a different cluster (**Figure 8**, **I**), again indicating different responses of tannins based on trihydroxylated and dihydroxylated B-rings. However, catechin and gallocatechin monomers formed another group (**Figure 8**, **G**), suggesting that biosynthesis of flavan-3-ol monomers and of proanthocyanidins are differently regulated. The last two clusters (**L** and **K**) were similarly based on dihydroxylated and trihydroxylated anthocyanin B-rings, respectively. However, unexpected correlations of these molecules with pyranoanthocyanins derived from reaction of acetaldehyde with other anthocyanins require further investigation.

# Genetic Diversity of Grapevine Response to Drought Shown by Metabolomics

Grouping of polyphenols according to their drought response across the diversity panel provided confirmation of the results of variance analysis (e.g., for flavan-3-ols). Additional correlation networks between molecules that had not been detected in the global ANOVA treatment suggest different responses of the molecular clusters in different cultivars. Unsupervised hierarchical clustering of cultivars and metabolites (including MRM data and calculated variables) affected by drought (**Figure 7**) was performed to explore this hypothesis. Molecules clustered by family and, within some families, by B-ring substitution pattern, confirming the impact of irrigation on some specific branches of the biosynthetic pathway shown by the correlation networks. Clustering of cultivars confirmed cultivar differences in the molecular response to drought. Some of these differences may be related to differences in polyphenol metabolism as some cultivars accumulate specific polyphenol classes (e.g., colored vs. white cultivars). Three cultivar subgroups comprised an excess of colored or white cultivars, compared to the whole population (**Figure 9**). However, color did not fully explain the clustering based on polyphenol response to drought. Genetic groups did not appear as a major factor although cultivars from genetic group WW were overrepresented in subgroup 2-1-2. Moreover, cultivars from group 1 were generally harvested earlier under not irrigated conditions than those of group 2 (**Figure 10**). Differences in precocity may also induce different polyphenol response to irrigation as some compounds such as flavan-3-ols and hydroxycinnamic acids are biosynthesized at early development stages while anthocyanins and flavonols accumulate after veraison. Finally, some of the observed responses may be indirect responses, for instance due to differences in cultivar responses to other types of stresses, such as UV-stress since water regime also impacts canopy (for instance for flavonols), or biotic stress (for stilbenes).

This preliminary study is based on only two vintages, with no treatment replicate for individual cultivars. However, some cultivars with contrasted responses to water stress have been identified and could be used in future more detailed studies. In particular, it will be of interest to also characterize the plant physiological status, to determine if these contrasted behaviors are related to the near iso/anisohydric phenomenon, analyze their stability and explore potential interferences with phenological stages and environmental factors.

### Large Scale Metabolomics Studies Shedding Light on Polyphenol Composition

The MRM method used in this study was targeted on a large number of polyphenolic compounds, including 96 molecules analyzed directly and 9 additional compounds released after acid-catalyzed depolymerization of proanthocyanidins in the presence of pholoroglucinol. Such experiments had never been performed at a large scale, in terms of number of targeted molecules and of number of studied cultivars. The large data collected made it possible to establish correlation networks that confirmed previous knowledge and provided new information on grape polyphenol metabolism. Conversely, the patterns established validate interpretation of mass spectrometry data for most of the compounds analyzed. However, a few compounds appeared as outliers in some of the clusters, raising questions on their attribution. For example, the signal attributed to petunidin-3,5-diglucoside clustered with pigments derived from reactions of anthocyanins with phenolic acids, suggesting confusion or contamination with another molecule of this group. This will be explored further, potentially leading to the discovery of new compounds. Similarly, clustering of β-glucogallin with B-ring trihydroxylated anthocyanins and flavonols is surprising. This may be related to a role of glucogallin in their biosynthesis, as proposed above. However, formation of β-glucogallin may also reflect degradation of delphinidin and/or myricetin, followed by glucosylation of the resulting gallic acid. In this case, other anthocyanins are expected to follow the same catabolic process. For example, degradation of malvidin or syringetin and of peonidin or isorhamnetin should similarly yield syringoyl- and vanilloyl-glucose which have not been included in the molecular targets of the MRM method. Detection of these new molecules would help validate this hypothesis.

# CONCLUSIONS

A large scale experiment involving cultivation of an association panel of 279 V. vinifera cultivars designed to represent the genetic and phenotypic variation encountered in cultivated grapevine and metabolomics analysis targeted to a large number of polyphenolic compounds (polyphenomics) was performed in 2014 and 2015. Chemometrics analysis of the data showed large differences in polyphenol composition related to genetic factors, environmental factors (i.e., water stress), and genetic x environment interactions. Correlation networks shed light on the relationships between the different polyphenol metabolites and related biosynthetic pathways. In addition, detailed polyphenomics analysis confirmed that polyphenol reactions described in wine take place in the berries. Finally, this paper reports the first large scale study demonstrating an influence of water stress on the different classes of polyphenols but also cultivar differences in the types and extents of drought responses, with different molecules affected either positively or negatively and different impacts on grape and wine quality. This work will be the foundation for identifying the genetic basis of the drought differential response of the cultivars in term of polyphenol composition, through Genome-Wide Association Study.

# AUTHOR CONTRIBUTIONS

LP and AV developed the MRM methods and performed the analyses, MR developed and applied the extraction protocols, EM and AVQ interpreted the mass spectrometry data, NS supervised the metabolomics analysis, JB and NT performed the chemometrics analysis, LL, JP, AA, NT, NS, and VC conceived the designed research and interpreted the results. All authors contributed to drafting and/or critical revision of the work and approved the manuscript.

#### FUNDING

The research leading to these results has received funding from the European Community's Seventh Framework Program (FP7/2007-2013) under the grant agreement no FP7-311775, Project Innovine. Financial support from GIS IBiSA (Infrastructures en Biologie Santé et Agronomie), Région Languedoc Roussillon, and INRA CNOC for funding of the UPLC–MS equipment and funding from Alfonso Martín Escudero Foundation for the postdoctoral fellowship of AVQ are also acknowledged.

#### REFERENCES


#### ACKNOWLEDGMENTS

The authors gratefully acknowledge Gilles Berger, Yves Bertrand, Jean-Luc Guiraud, Thérèse Marlin, and Léa Ollier for technical assistance.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2017. 01826/full#supplementary-material

Table S1 | List of cultivars collected in 2014 and/or 2015, with their codes, genetic groups, colors, and harvest dates under irrigated (I) and not irrigated (NI) conditions.


affected by interseason weather variability. J. Agric. Food Chem. 65, 5868–5878. doi: 10.1021/acs.jafc.7b01466


in grape skins. J. Agric. Food Chem. 52, 7144–7151. doi: 10.1021/jf0 48939h

Vrhovsek, U.,Masuero, D., Gasperotti, M., Franceschi, P., Caputi, L., Viola, R., et al. (2012). Versatile targeted metabolomics method for the rapid quantification of multiple classes of phenolics in fruits and beverages. J. Agric. Food Chem. 60, 8831–8840. doi: 10.1021/jf2051569

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Pinasseau, Vallverdú-Queralt, Verbaere, Roques, Meudec, Le Cunff, Péros, Ageorges, Sommerer, Boulet, Terrier and Cheynier. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Differences in Flower Transcriptome between Grapevine Clones Are Related to Their Cluster Compactness, Fruitfulness, and Berry Size

#### Jérôme Grimplet\*, Javier Tello † , Natalia Laguna and Javier Ibáñez

Departamento de Viticultura, Instituto de Ciencias de la Vid y del Vino (Consejo Superior de Investigaciones Científicas, Universidad de La Rioja, Gobierno de La Rioja), Logroño, Spain

#### Edited by:

Giovanni Battista Tornielli, University of Verona, Italy

#### Reviewed by:

Gregory Alan Gambetta, Ecole Nationale Supérieure des Sciences Agronomiques de Bordeaux-Aquitaine, France Silvia Dal Santo, University of Verona, Italy

#### \*Correspondence: Jérôme Grimplet

jerome.grimplet@icvv.es

#### † Present Address:

Javier Tello, Division of Viticulture and Pomology, Department of Crop Sciences, University of Natural Resources and Life Sciences Vienna (BOKU), Tulln, Austria

#### Specialty section:

This article was submitted to Crop Science and Horticulture, a section of the journal Frontiers in Plant Science

Received: 16 January 2017 Accepted: 07 April 2017 Published: 27 April 2017

#### Citation:

Grimplet J, Tello J, Laguna N and Ibáñez J (2017) Differences in Flower Transcriptome between Grapevine Clones Are Related to Their Cluster Compactness, Fruitfulness, and Berry Size. Front. Plant Sci. 8:632. doi: 10.3389/fpls.2017.00632 Grapevine cluster compactness has a clear impact on fruit quality and health status, as clusters with greater compactness are more susceptible to pests and diseases and ripen more asynchronously. Different parameters related to inflorescence and cluster architecture (length, width, branching, etc.), fruitfulness (number of berries, number of seeds) and berry size (length, width) contribute to the final level of compactness. From a collection of 501 clones of cultivar Garnacha Tinta, two compact and two loose clones with stable differences for cluster compactness-related traits were selected and phenotyped. Key organs and developmental stages were selected for sampling and transcriptomic analyses. Comparison of global gene expression patterns in flowers at the end of bloom allowed identification of potential gene networks with a role in determining the final berry number, berry size and ultimately cluster compactness. A large portion of the differentially expressed genes were found in networks related to cell division (carbohydrates uptake, cell wall metabolism, cell cycle, nucleic acids metabolism, cell division, DNA repair). Their greater expression level in flowers of compact clones indicated that the number of berries and the berry size at ripening appear related to the rate of cell replication in flowers during the early growth stages after pollination. In addition, fluctuations in auxin and gibberellin signaling and transport related gene expression support that they play a central role in fruit set and impact berry number and size. Other hormones, such as ethylene and jasmonate may differentially regulate indirect effects, such as defense mechanisms activation or polyphenols production. This is the first transcriptomic based analysis focused on the discovery of the underlying gene networks involved in grapevine traits of grapevine cluster compactness, berry number and berry size.

#### Keywords: Vitis vinifera, cluster architecture, phenotyping, transcriptomics, somatic variation

# INTRODUCTION

Grapevine (Vitis vinifera L.) is one of the most valuable horticultural crops in the world, with a total grape production of 77 million ton (2013, http://faostat3.fao.org). The value of any table grape, grape juice, or wine product relies fundamentally on disease-free and high quality fruits. Cluster compactness, an issue specific to grapevine, directly impacts fruit quality and disease susceptibility:

**321**

Berries in compact clusters tend to ripe more asynchronously, impacting quality at harvest and compact cluster are also more susceptible to diseases, such as Botrytis cinerea (Molitor et al., 2012b).

Cluster compactness is a complex trait, resulting from the interaction of parameters related to cluster architecture and berry morphology, each contributing differently within a cultivar. Shavrukov et al. (2004) indicated the internode length of inflorescence rachis is the major trait responsible for inflorescence openness in four grape cultivars. However, a smaller berry size is responsible for loose cluster in Albariño (Alonso-Villaverde et al., 2008), while in other study, cluster density is correlated with the number of seeds per berry in the progeny of two wine grape cultivars (Bayo-Canha et al., 2012). More recently, our group has dissected the cluster compactness trait on a large set of table and wine cultivars (Tello et al., 2015). This exhaustive survey indicates that the berry number and the length of the rachis main axes (cluster architecture) are the most critical parameters for cluster compactness, followed by berry size. Each of these cluster compactness features is specific to different development stages. (i) Architecture related parameters are defined early. At the end of the first season summer, the primary latent bud contains a compressed shoot with inflorescence meristems, tendril and leaf primordia. In the second season, during initial stages of bud swelling, the inflorescence branch meristems can additionally ramify to form further inflorescence branch meristems that divide into a group of flower meristems (normally three). At that point, the inflorescence/cluster architecture is essentially set, as rachis elongation is limited after flowering (Coombe, 1995; Shavrukov et al., 2004). (ii) Final berry number in the cluster depends on the initial number of flowers and the fruit set rate that occurs after anthesis, although a compensation effect does exist (May, 2004). The initial number of flowers in the inflorescence is determined early in the second season, before bud burst, and it is noted that high temperatures at this stage decrease the number of flowers eventually formed (Ezzili, 1993). The availability of carbohydrate reserves in the trunk and roots (from the previous season) may also be a limiting factor (Bennett et al., 2002). Fruit set rate depends on the success of the pollination and fertilization processes, and also on the competition with other sink organs, mainly growing shoots. (iii) Two main factors are responsible for the size of the ripe berry at harvest: the cell number and their volume. Cell division is particularly active before anthesis and stops when the berry reach the lag phase, at the beginning of ripening (véraison). From that point only growth by cell enlargement occurs (Harris et al., 1968; Dokoozlian, 2000).

Little is known about the molecular basis or genetic factors responsible for differences in cluster compactness among grapevine cultivars and clones. Experimental treatments to reduce cluster compactness involve enlarging inflorescence main axes, reducing fruit set, and/or reducing berry size. Plant hormones control grapevine reproductive development and flowering timing through the gibberellin:cytokinin balance. Gibberellins mediate the formation of the inflorescence axis, while cytokinins regulate the differentiation into flowers and are specifically needed for the growth of pistil (Pool, 1975). ABA concentration is high before anthesis, and auxin transport is needed to avoid abscission and promote fruit set (Kühn et al., 2014). The application of the gibberellins inhibitor prohexadione-Ca causes a loosening effect by reducing berry size and/or number of berries, likely through disturbing pollination and cell division processes (Molitor et al., 2011; Schildberger et al., 2011). The application of gibberellic acid pre-bloom promotes the growth of the inflorescence (Hed et al., 2011; Molitor et al., 2012a), while gibberellin treatments during bloom reduce fruit set and increase berry size (Ben-Tal, 1990).

The availability of the grapevine genome sequence (Jaillon et al., 2007; Velasco et al., 2007) allowed high throughput studies of the grapevine that are leading to an increased knowledge of the molecular events occurring behind physiological processes. In this work we performed transcriptomic analyses of Garnacha Tinta clones, with stable differences in specific compactnessrelated parameters (berry number, berry size), to identify genes and gene networks involved in cluster compactness characteristics. From this transcriptomics study, 183 candidate genes were selected for an association analysis in a collection of grapevine varieties (Tello et al., 2016).

# METHODS

#### Plant Material

In the early 2000s, Gobierno de La Rioja prospected the entire Rioja region and collected hundreds of grapevine (Vitis vinifera L.) plants of different cultivars, usually old plants and/or plants with particular characteristics. Each of these plants was multiplied by cuttings and grafted on Richter 110 rootstock. Five clonal grafted vines per original plant were planted together in a single plot at the experimental vineyard of La Grajera (Logroño, La Rioja). This clone collection includes 501 clones from Garnacha Tinta, which were screened for cluster compactness, in sequential steps. First, the compactness of all the clones was visually assessed. Then, nine clones were selected for phenotyping during the next season and six of these were also phenotyped during a second and third season. Finally, four of these clones, two with compact clusters ("compact clones") and two with loose clusters ("loose clones") were selected for transcriptomic analysis.

#### Phenotyping

In three successive seasons, six selected Garnacha Tinta clones were phenotyped for several variables related to cluster compactness using five clusters per clone as described by Tello and Ibáñez (2014). All the clones were subjected to pair-wise comparisons for phenotypic variables grouped in four categories: plant (e.g., fertility), cluster architecture (e.g., cluster length), fruitfulness (berry number and seed number) and berry size (**Supplementary Table 1**). Clone pairs differing only in one category were favored, but the most selective criterion was consistency over the seasons for the observed significant pair-wise differences, and some clone comparisons with non-consistent differences were discarded. Finally, four clones (368, 906, 1134, and 1154) were used for transcriptome analysis (**Supplementary Table 2**).

# Experimental Design and Sampling for Transcriptome Analysis

The experimental design was determined in accordance with the significant pair-wise differences consistently observed between the selected clones of Garnacha Tinta over the three seasons (**Supplementary Table 1**, **Figure 1**). Organs and stages were sampled based on specific differential parameters: berry number, seed number and berry size (**Table 1**). For berry number, flowers were sampled at the end of flowering (E-L 26, Coombe, 1995) before possible abscission or set (**Table 1**; comparisons G1-26, G2-26, G3-26, and G4-26). Seed number is also determined at that step since it depends on the success of pollination. Spring buds at budburst were sampled in two clones to study the initial number of flowers (E-L 3: comparison G4-03), when flowers start to differentiate (Pouget, 1981; Dunn and Martin, 2000).

Berry size is determined by cell division and cell expansion. So, analyses for berry size were carried out on flowers at the end of flowering, when cell division is active (E-L 26; comparison G1-26), and on green berries at the beginning of véraison (E-L 34), when cell division is complete and berry enlargement by cell expansion begins (Dokoozlian, 2000; comparison G1-34). As berries are in different developmental stages within the same cluster at a given time, sampled berries were classified according to their density by flotation on NaCl solutions (Carbonell-Bejerano et al., 2016). Green berries floating in a solution of 80 g/l NaCl and sinking in a solution of 60 g/l NaCl were selected.

Three replicate samples were collected from different vines. After collecting, samples were immediately frozen in liquid nitrogen, and then kept in the laboratory at −80◦C until RNA extraction.

# RNA Extraction and Microarray Hybridization

Total RNA was extracted from samples using the Spectrum plant total RNA kit (Sigma, www.sigmaaldrich.com) as recommended by manufacturer. DNase I digestion was carried out with the RNase-Free DNase Set (QIAGEN). RNA integrity and quantity were assessed with a Nanodrop 2000 spectrophotometer (Thermo Scientific) and an Agilent's Bioanalyzer 2100. Microarray hybridizations were performed at the Genomics Unit of the National Centre for Biotechnology (CNB-CSIC, Madrid).

Synthesis of cDNA, labeling, hybridization, and washing steps were performed according to the NimbleGen arrays user's guide. Each sample was hybridized to a NimbleGen microarray 090818 Vitis exp HX12 (Roche, NimbleGen), which contains probes targeted to 29,549 predicted grapevine genes and 19,091 random probes as negative controls. Images were analyzed using NimbleScan v2.6 software (Roche), which produces.xys files containing the raw signal intensity data for each.


TABLE 1 | Experimental design for each of the comparisons performed between Garnacha Tinta clones.

Organs and sampling stages for transcriptomics analyses were chosen based on the stable phenotypic differences found in three seasons. C, Compact clone; L, Loose clone.

#### Microarray Data Processing

The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus (Edgar et al., 2002) and are accessible through GEO Series accession number GSE67708 (www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE67708).

Raw intensity values were processed using the R package oligo (Carvalho and Irizarry, 2010). Individual probes raw expression values were computed from.xys files and the in house pd info builder package pd.vitus.exp.vitnames designed to fit the 12Xv1 annotation nomenclature. Normalization was performed with Robust Multi-Array Average (RMA; Irizarry et al., 2003). Resulting RMA expression values were log2 transformed. Distributions of expression values processed via RMA of all arrays were very similar with no apparent outlying arrays.

#### Microarray Data Analysis

Each condition (clone × stage/organ) was performed in three biological replicates. Differential expression analyses of the comparisons presented in **Table 1** were performed with the ebayes (Smyth, 2004) method from the package limma in R. The cutoff of differentially expressed genes was set to a p < 0.05 after Benjamini-Hochberg correction with at least a 2-fold ratio difference of expression. Principal component analysis (PCA) was performed in R using the pca package with the ppca method. Hierarchical clustering was performed using MultiExperiment Viewer (Saeed et al., 2003) based on Pearson's correlation and using the average linkage option and optimal gene ordering. The stringent set was obtained by clustering genes with a distance threshold <0.05. The tolerant set was obtained by clustering genes with a distance threshold <1.5.

To identify the biological functions over-represented within selected probe sets, functional enrichment analyses were performed using FatiGO (Medina et al., 2010; P < 0.05). Functional categories were based on manual annotation of 12Xv1 grape genome assembly, described in Grimplet et al. (2012).

#### Cytoscape/VitisNet Analysis

Expression data were uploaded in Cytoscape version 3 (Shannon et al., 2003) and analyzed with VitisNet (Grimplet et al., 2012). According to FatigoGO analysis, networks related to enriched categories were selected for manual inspection. The visual style in the figures was designed to best represent changes in flower by including notifications of the genes over-expressed in compact or loose clones. A color gradient was used depending on the presence of the differentially expressed gene (DEG) in 2 or 3 (light color) to 4 (dark color) comparisons, to have a visual representation of the DEG degree of recurrence in the comparisons. Network ID corresponds to the VitisNet ID (Grimplet et al., 2009).

# RESULTS AND DISCUSSION Phenotyping and Comparison of the

# Clones

In a multi-cultivar framework, our group identified the major morphological factors influencing the cluster compactness trait (Tello and Ibáñez, 2014; Tello et al., 2015). Different variables, classified within four major groups (plant, cluster architecture, fruitfulness (berry and seed number) and berry size), were phenotyped in a large set of diverse cultivars, and it was concluded that the length of the cluster main axes and berry number were the main discriminant variables for cluster compactness, followed by the berry size. In the present work, a similar set of variables was used to study Garnacha Tinta clones and only clone pairs consistently differing in selected variables were used for analyses. Cluster compactness remained consistent through the seasons, but some of the significant differences observed the first season were not stable over the three seasons. Therefore, some clone comparisons were discarded. Finally, four clones of Garnacha Tinta, two loose (368 and 1154) and two compact clones (906 and 1134) were used for transcriptome analysis (**Table 1**, **Supplementary Table 2**).

Similar to cluster compactness, the berry number showed very consistent differences through the seasons in the four clone comparisons (**Table 1**, **Figure 1**, **Supplementary Table 1**). The compact clones produced significantly tighter clusters than loose clones and carried a significantly greater number of berries in all the comparisons studied during the three seasons.

Berry number was the only differential variable in comparison G3, but in the remaining comparisons additional seasonally stable differences appeared in other traits. Thus, comparison G1 was selected to examine the transcriptional changes observed between a loose clone and a compact clone with significant differences in berry number and berry size. The compact clone (1134) produced more and larger berries than the loose clone (368) (**Figure 1**, **Supplementary Table 1**).


TABLE 2 | Number of differentially expressed genes (DEG) at every time point for every comparison Compact vs. Loose clone.

Finally, comparisons G2 and G4 were selected to examine differences in global gene expression related to the two variables included in the fruitfulness category: berry number and seed number. In these comparisons, the compact clones always had more berries per cluster and more seeds per berry than the loose clones (**Figure 1**, **Supplementary Table 1**). This was expected, as both the number of seeds and fruit set are related to pollination and fertilization, and flower fate (abscission or berry set) greatly depends on the existence of at least one fertilized ovule in the flower (Kassemeyer and Staudt, 1982).

#### Global Gene Expression Data

The greatest number of differentially expressed genes (DEG) between clones was in flower at the end of flowering (E-L 26), while very few differences could be seen in spring buds (G4-03) (**Table 2**). **Figure 2** represents the first two axes of a PCA of the expression data obtained for the four studied clones at the end of bloom, E-L 26. The first component of the PCA represented 73% of the total variation and seemed related to compactness. Component 2 accounted for 9% of the total variability, separating genotypes.

The replicates from compact and loose clones were clearly separated, however clones of the same compactness presented a large variation. There were differences in the sampling dates to match physiological state but they did not seem to be related to the variation, since more variability could be observed between some replicates sampled the same day (data not shown). It is known that, within the same inflorescence, there are flowers in different stages of development, including those with already fertilized ovules, others with fertilization in progress, and others that have not been fertilized and probably will drop (Kühn et al., 2014). These flower stages are not visually distinguishable during sampling, but their transcriptomic profiles are probably different, because there are evidences in grapevine indicating that pollination rapidly modifies gene expression (Kühn and Arce-Johnson, 2012). The proportion of flowers in each of those stages would vary differently between clones, partially explaining the consistent differences in the number of berries observed in the four comparisons. This is probably the major cause for the gene expression differences observed in the four pair-wise comparisons at the end of flowering (**Table 2**, E-L 26).

At that stage, a greater number of DEG was observed in comparisons involving the compact clone 1134 (G1-26 and G2- 26, 5320 and 4446 genes, **Table 2**) than in comparisons with the other compact clone, 906 (2413 and 665 genes). That variation may be the result of an asynchronous floral development in clone 1134, which would lead to the sampling of slightly different flower stages in the compared clones. This would be supported by the high number of DEG (1607 genes) observed at E-L 26 between

compact clone 906. Red dot: Garnacha Tinta compact clone 1134.

clones 1134 and 906 (**Table 2**). This difference is reduced later, as illustrated in comparison G1-34, where the two clones involved (1134 and 368) reached similar transcriptomes, with a minimal number of differentially expressed genes between them at E-L 34.

In comparison G1, in addition to a different number of berries, a consistent difference in berry size was observed, unlike in G2. So, the differentially expressed genes found in G1-34 and (partly) in G1-26 could be related to the fact that clone 368 showed a smaller berry size than 1134 during the three studied years.

The number of significant DEG obtained for the stage end of flowering (E-L 26) was much lower in loose clones (368 + 1154) than in compact clones (1134 + 906) when all comparisons are considered. Only 70 gene transcripts were more abundant in the loose clones and 400 in the compact clones (**Table 2**). Many genes, however, were differentially expressed between the compact clone 1134 and any loose clone (2051 genes showed a greater expression in loose clones, 1683 in the compact clone).

# Functional Categories Analysis

Functional categories enrichment analysis was performed in order to identify the main mechanisms impacted in cluster compactness and their related traits. Since PCA showed greatest differences in expression pattern in clones 1134 and 368, analyses were performed considering several situations: group 1 includes the genes differentially expressed in all comparisons at E-L 26 (**Table 3**) that are specifically related to compactness independent of the clone (400 over-expressed genes in compact clones, 70 in loose clones); group 2 contains the genes specifically regulated in the most extreme compact clone 1134, i.e., differentially expressed in G1 and G2: genes expressed in clone 1134 (2051 over-expressed genes) vs. all the loose (1703 over-expressed genes) (**Table 4**); and group 3 comprises the genes specifically regulated in the most extreme loose clone 368, i.e., differentially expressed in G1 and G3 comparisons: over-expressed genes in

#### TABLE 3 | Over-represented functional categories in all E-L 26 comparisons with P < 0.05.


Values are expressed as log2 ratio group/genome.

clone 368 (1400) vs. all the compact (560) (**Table 5**). If a category is enriched in both the compact and loose clones, this means that there are distinct genes from that category represented in a larger proportion in both sets than in the whole transcriptome. As indicated in 4.2, for group 1 few genes were differentially expressed in the loose clusters considering all comparisons at E-L 26; therefore, only two categories were enriched in the loose clustered type (**Table 3**). Overall, several patterns of expression emerged from the three enrichment analyses. Within the functional categories related to the metabolism, several functional categories indicate a dramatic shift of expression of genes involved in the metabolism. The category related to cell growth and death was over-represented in all the clones with compact clusters (**Tables 3**–**5**). Cytoskeleton, chromosome organization and biogenesis and DNA metabolism were also over-abundant in the compact clones (**Tables 3**–**5**), indicating a possible greater cellular replication activity in the compact clones. Categories related to cell wall showed clear specificity of transcript expression in either compact or loose clones. Pectinrelated categories were only over-represented in the compact clones and cellulose biosynthesis in the loose clones (**Tables 4, 5**). Phenylpropanoids-related categories showed dramatic changes in gene expression, the phenylpropanoid metabolism category was over-represented in both the compact and loose clones (**Tables 4, 5**). The lignin biosynthesis category seems more abundant in the compact clone (1134) when compared with both loose clones (**Table 4**) and the loose clone 368 when compared with both compact clones (**Table 5**). Terpenoids and alkaloids categories also seemed to be over-represented in the loose clones (**Tables 4, 5**). In addition the plant-pathogen interaction category was also over-represented in the loose clones vs. 1134 (**Table 4**). Several categories related to hormone signaling were also over-represented in the loose clones, such as Auxin, brassinosteroids, cytokinins, jasmonate, and ethylene signaling (**Tables 4, 5**). Transporters showed a balanced pattern; however, oxygen transport was more abundant in the loose clones (**Tables 4, 5**). Ion transport-related categories were also over-represented in both types of clones.

#### VitisNet Analysis Indicates Metabolic Pathways Related to Cluster Compactness

Networks were manually inspected to find those that presented relevant changes. These analyses allowed us to identify key networks and possible causes for cluster compactness as well as important information on early fruit development that will be discussed along this section (**Figures 3**–**6**, **Supplementary Images 1**–**10**). We observed changes in gene expression between compact and loose clones in flowers, and the clone with more and bigger berries (clone 1134) showed more differences with the loose clones than the other compact clone. It was however difficult to clearly distinguish if differences in cell replication or timing impacted fruit set (and thus berry number), the number of cells (berry size), or both. We identified four main categories of genes showing differential expression, related to: cellular activity, pathogens interaction, hormonal response and phenylpropanoids biosynthesis.

#### Loose and Compact Clones Show Great Difference in Flower Transcriptome Indicating a Distinct Cell Division Rate and/or Asynchronous Development

Comparison between flowers of clones producing tight clusters and clones producing loose clusters indicated a distinct cell division rate and/or asynchronous development. Most noticeably genes related to a greater activity in production of cellular material were more abundant in the compact clones. Evidences were specifically gathered at the level of carbohydrate and nucleic acid metabolism as well as the regulation of cell cycle and cell division.

#### **Carbohydrate metabolism. Cell wall**

The composition and size of the fruits as they grow are very dependent of the efficiency of the flower as a nutrient sink (Bihmidine et al., 2013) and significant differences were

#### Grimplet et al. Cluster Compactness Transcriptomics in Grapevine

#### TABLE 4 | Over-represented functional categories in G1 and G2 comparisons with P < 0.05.


(Continued)

#### TABLE 4 | Continued


Values are expressed as log2 ratio group/genome.

observed between compact and loose clones in the carbohydrate metabolism in flower.

Important regulators of the sucrose metabolism (**Figure 3**) were seen to have isogenes specifically expressed in flower. Most noticeably, cell wall invertase (VIT\_04s0008g01140) had greater expression in the loose clones than in clone 1134 and a vacuolar form was more expressed in the compact clones than in clone 368 (VIT\_16s0022g00670). The cell-wall forms have been associated with rapidly growing tissues (Eschrich, 1980), they were induced by wounding and pathogenic attack (Sturm and Chrispeels, 1990), and have been implicated in phloem unloading and source/sink regulation (Eschrich, 1980; Roitsch et al., 1995). Gene expression in flower also indicated that starch seems to be preferentially catabolized into dextrin and maltodextrin with the increase of expression of several isogenes of alpha-(7 isoforms) and beta-amylases (2 isoforms) in the loose clones with respect to clone 1134. Higher expression of starch synthase (VIT\_00s1488g00020) might indicate greater starch production in loose clones. Additionally a possible regulator of amylases (Liu and Thornburg, 2012), a transcript homologous to Myb305, was more abundant in the loose clone 368 vs. compact clones (VIT\_14s0083g01060, **Figure 3**). However, the change of carbohydrate and cell osmolarity might be reminiscent of the

#### TABLE 5 | Over-represented functional categories in G1 and G3 comparisons with P < 0.05.


(Continued)

#### TABLE 5 | Continued


Values are expressed as log2 ratio group/ genome.

flower opening mechanism (van Doorn and Van Meeteren, 2003) thus it would maintain turgor in the flowers of the loose clone, indicating a slight difference in the timing (delay) in loose against compact clones. As mentioned above, this difference could not be phenotyped since the samples were in an equivalent external stage: flowers were sampled at the end of flowering, with fallen stamen.

The next step was to identify the potential fate of the carbohydrates that would be produced from the DEG in the compact clones. In plants, most of the carbon fixed by photosynthesis is incorporated into cell wall carbohydrates. Compact clones showed an increase of expression of several transcripts involved in the biosynthesis of compounds that might be related to an increase of cell wall material. Starting from the fructose, all the enzymes that are involved in the biosynthetic pathway of both D-mannose and GDP mannose (**Figure 3**) presented at least an isoform over-expressed in flowers of the compact

FIGURE 4 | Adapted Cytoscape networks including transcripts differentially expressed in flowers between loose and compact clones related to nucleic acid metabolism. Genes over-expressed in compact clones in all comparisons are in dark red. Genes over-expressed in compact clones in 2 or 3 comparisons are in red. Genes over-expressed in loose clones in all comparisons are in dark green. Genes over-expressed in loose clones in 2 or 3 comparisons are in green. Figure is adapted from networks 10230 and 10240 from Grimplet et al. (2009).

clones. These included fructokinases (VIT\_18s0089g01230, VIT\_05s0102g00710), hexokinases (VIT\_18s0001g14230), mannose-6-phosphate\_isomerase (VIT\_01s0011g03750), phosphomannomutase (VIT\_01s0011g03750, only G1). Other genes that might be involved in the biosynthesis of the predominant cell wall components arabinose or UDP-xylose (Seifert, 2004) (**Figure 3**), such as UDPglucuronate 4-epimerase (VIT\_15s0048g00330, only G1), UDP-glucuronate decarboxylase (VIT\_05s0077g02300) and alpha-N-arabinofuranosidase (VIT\_08s0032g00890, VIT\_12s0055g01180) were also more abundant in compact clones. In cell wall (**Supplementary Image 1**) many differences between the expression levels of isogenes were observed. There were a few families that seem to be specific to one or the other cluster type. One of them, the pectinacetylesterase, which is involved in the regulation of pectin acetylation, had three isogenes over-expressed in flowers of the compact clones vs. clone 368, as well as four isogenes of the fasciclin-like arabinogalactan proteins involved in cell adhesion (Johnson et al., 2003). They might also be involved in cell expansion, since a mutant was observed causing swelling in roots (Shi et al., 2003). To complete the picture related to cell wall, several major families of cell wall related proteins showed differential expression between isogenes in a large amount but were evenly

represented between the loose and compact clones, amongst them, the pectin methylesterase inhibitors, the pectinesterases, the pectate lyases or the xyloglucan endotransglycosylases. These centrally important aspects of expansion are also mediated by auxin, which is critical for skin strength in the earliest stages in flowers (Reeves et al., 2012). Overall while the gene expression is contrasted between clones, no routes leading to specific cell wall metabolites emerged as specific in any cluster type probably because substrate specificity of isogenes is not yet well-described.

In summary, there are differences between loose and compact clones in the expression of genes related to carbohydrate and cell wall metabolism. It can be hypothesized that cells in the compact clones were dividing more actively, triggering a large cascade of events that would explain the high number of differentially expressed genes but will likely complicate the identification of the primary genetic factors initiating the events.

#### **Purine and pyrimidine biosynthesis**

Transcripts involved in the metabolism of nucleic acid components were another indicator of differences in cellular activity between compact and loose clones. Several genes related to purine metabolism (**Figure 4**) tended to be upregulated in flowers of the compact clones. Genes coding for enzymes involved in the next part of the pathway were

more clearly over-expressed in the flowers of the compact clones indicating greater production of deoxynucleotides. The ribonucleoside-diphosphate reductase presented three isogenes (VIT\_07s0031g01990, VIT\_14s0068g02000, VIT\_07s0031g02000) over-expressed in the compact clones. The nucleoside-triphosphatases (VIT\_19s0015g01800, VIT\_10s0003g01720 over-expressed in compact clone 1134) were involved in the conversion of ATP and GTP into ADP and GDP for RNA biosynthesis. Adenine phosphoribosyltransferase (VIT\_00s1847g00010), which is involved in the purine salvaging, was also over-expressed in flowers of the compact clones.

As observed for the purine metabolism pathway, an increase of the gene expression in compact clones was observed in the pyrimidine pathway (**Figure 4**) with few changes affecting the earlier biosynthesis steps. The changes specifically affected transcripts for the interconversion of nucleotides triphosphate toward the production of deoxynucleotide diphosphates. Activity related to DNA repair was observed by the increase of dUDP pyrophosphatase (Dubois et al., 2011) and then production of dUMP. These observations might be linked to processes in the replication and repair related networks. In addition, transcripts of genes coding for enzymes involved in the pyrimidine salvaging pathways were also more abundant in compact clones, such as cytidine deaminase (VIT\_17s0000g01370) and thymidine kinase (VIT\_12s0057g00500).

The VitisNet analysis indicates significant differences in the nucleotide metabolism pathway between flowers of compact and loose clones. It suggests that these differences were not affecting the de novo biosynthesis of the nucleotides but may be related to interconversion and salvaging. These results shall be put in perspective with observations in the networks related to genetic regulation, more specifically DNA repair.

#### **Regulation of DNA replication and repair mechanisms**

The rate of DNA repair is dependent on many factors, including cell type, cell age and extracellular environment. In the studied clones at E-L 26 we observed a greater activity of this pathway in compact clones. Most of the DEG involved in DNA replication (twenty; **Figure 5**) were more abundant in the compact clones, like other pathways related to replication and repair. One of the six identified MCM genes related to DNA replication has previously been shown to be up-regulated after fertilization (Dresselhaus et al., 2006). MCM6 is essential for both vegetative and reproductive growth and development in plants (Dresselhaus et al., 2006).

Base excision repair (**Figure 5**) is the predominant DNA damage repair pathway for the processing of small base lesions. A large portion of the genes (10 over 30 genes) belonging to this network was more abundant in the compact clones. The differentially expressed transcripts were found either in the network common branches or in the mechanism of reparation of segments of 2–13 nucleotides.

DNA mismatch repair (**Figure 5**) is a highly conserved biological pathway that plays a key role in maintaining genomic stability (Li, 2008). Several of the over-expressed genes in the compact clones also belonged to the DNA polymerase complex. Homologous recombination (**Figure 5**) is essential for the accurate repair of DNA double-strand breaks (potentially lethal lesions), and acts before the cell enters mitosis. Once again, over-expression of genes related to DNA replication and repair in flowers of compact clones is another marker of a more active cell division compared to loose clones.

#### **Cell cycle and regulation of actin cytoskeleton**

Transcripts of about a third of the genes (97/322) involved in the cell cycle network (**Supplementary Image 2**) were more abundant in compact clones. The observation of a greater expression of these genes, combined with transcript abundance related to DNA processing mechanisms was another indicator of a more active division state in the compact clones. In addition, the expression of the genes involved in the network of regulation of actin cytoskeleton (**Supplementary Image 3**) was more abundant in flowers of the compact clones (82 of 343 genes).

Overall, differences were observed between compact and loose clones of Garnacha Tinta in the expression of cell divisionrelated genes (carbohydrate and nucleic acid metabolism as well as regulation of cell cycle and cell division). This could be due to a difference in the rate of fecundated flowers in the two clone types, resulting in the comparison of a pool with a greater ratio of fecundated flowers (in the compact clones) vs. a pool with a lower ratio (in the loose ones) that would be in different cell division states. At this point of the development, cell division is in an exponential phase (Harris et al., 1968; Ojeda et al., 1999) and slight differences could significantly be reflected in the transcriptome. This would eventually lead to the greater number of berries observed in the compact clones. In addition, the differential activity in terms of cell replication could lead to a differential final cell number in the berries, and ultimately to a different berry size. In comparison G1-26, this could explain the differences in berry size between clone 1134 and clone 368 (**Supplementary Table 1**). Since few differences were observed in networks related to cell expansion (the other cell growth mechanism in berries), it would be the differential number of cells what affect final berry size in this case.

The differences observed in the cell division-related gene expression could also be due to a slight delay in the development progress in the loose clone with respect to the compact one. Given that the berries derived from flowers that opened first ("first berries") have less probability to abscise than the later opening flowers (Kühn et al., 2014), a delay in the development could produce a greater berry abscission rate, thus affecting berry number. Unfortunately it was not possible to determine if the transcriptome differences were due to only one or the two possible causes proposed.

#### Plant Pathogen Interaction and Relation to the Jasmonate/Methyl Jasmonate Interconversion

DEG involved in the mRNA surveillance pathway were predominantly expressed in the loose clones. They belonged more specifically to pre-mRNA 3′ -end processing machinery and non-sense-mediated decay (NMD). Some of them, such as SMG7 (VIT\_00s0527g00010, VIT\_00s0640g00020) appear to regulate the expression of the genes involved in pathogen response in Arabidopsis (Rayson et al., 2012). Therefore, the expression of genes related to the plant-pathogen interaction was further examined.

Salicylic acid (SA) is a signal molecule involved in interactions between plants and pathogens. Enzymes potentially involved in its biosynthesis pathway did not exhibit differential expression of their corresponding transcripts. However, some genes involved in SA signaling were differentially expressed in flowers and that are known to be involved in pathogen response. Homologous to EDS1 (4 adjacent isogenes on the genomic sequence) were over-expressed in flowers of the loose clones vs. clone 1134. These genes were known to be involved in R proteinmediated signaling (Dempsey et al., 2011). Twenty R proteins (**Supplementary Image 4**) presented more abundant transcripts in flowers of the loose clones. In the plant-pathogen interaction network (**Supplementary Image 5**), several isoforms of BAK1 (8 over 19) and EIX1/2 (7/20) genes were more abundant in flowers of the loose clones vs. clone 1134. These genes are known to act together in the plant defense against pathogens induced by ethylene (Bar et al., 2010). Differences in expression of genes potentially regulated by ethylene (**Supplementary Image 6**) were observed in both compact and loose clones but members of Ethylene Response Factor subfamily were clearly more abundant in the loose clones. These genes corresponded to the subfamily IX or B-3 according to Nakano et al. (2006). The genes in group IX have often been linked in defensive gene expression in response to pathogen infection (Berrocal-Lobo et al., 2002) and this group contains PTI genes (Gu et al., 2002) that were known to be regulated by EDS1 (Dempsey et al., 2011). The WRKY transcription factors can also play a role in the defense mechanism (Rushton et al., 2010) and many of them were overexpressed in loose clones (20 genes).

Transcript level-related evidence of differential accumulation of jasmonic acid was unclear since expression of different transcripts coding for proteins involved in its biosynthesis in the alpha-linolenic acid metabolic pathway was increased in compact or loose clones. The isoforms of jasmonate O methyltransferase/VIT\_04s0023t03810 VIT\_04s0023t03800 VIT\_04s0023t03790) were over-expressed in the flowers of the compact clones and the methyl jasmonate esterases (VIT\_00s0253t00170, VIT\_00s0253t00160 VIT\_00s0253t00150) were preferentially expressed in the loose clones. The first enzyme catalyzes the conversion of jasmonate to methyl-jasmonate (MEJA) and the esterase catalyzes the demethylation of methyljasmonate. Jasmonate needs to be in the demethylated form to trigger defense response to herbivores (Wu et al., 2008), while MEJA is most likely involved in plant morphology.

There is no obvious reason explaining the greater expression of genes potentially involved in pathogen interaction in loose clusters, but both (pathogen-related gene expression and cluster loosening) could be consequences of the flower abscission process. The activation of different defense responses at flower abscission zones was described in tomato (Meir et al., 2011). Grapevine inflorescences treated to increase flower abscission showed up-regulation of pathogenesis-related genes (Domingos et al., 2016).

#### Gibberellin and Auxin Biosynthesis and Signaling Were Likely to Play a Role in Compact Clones **Gibberellins**

Application of gibberellins (GA) on the clusters is widely used in the table grape industry to control fruit set, elongate rachis or increase berry size (Coombe, 1960). It has different effects depending on the treatment concentration and timing. When applied at bloom, gibberellins affect fruit set and berry size (Dokoozlian and Peacock, 2001). We hypothesize that differences in the gibberellins metabolism or signaling would be observed at flowering between compact and loose clones in flowers of clones differing in berry number (and berry size in G1). Several transcripts coding for enzymes involved in GA biosynthesis (diterpenoid biosynthesis, **Figure 6**) were more abundant in flowers of the compact clones in the comparison G1-26, such as copalyl diphosphate synthase (VIT\_07s0151g01070 loose clones vs. 1134), ent-kaurene synthase (VIT\_07s0151g01040 loose clones vs. 1134), gibberellin-20 oxidase (VIT\_04s0044g01650 loose clones vs. 1134, VIT\_04s0044g02010) and the regulator BME3 (VIT\_13s0019g04390 only G1). Moreover, flowers of the loose clones showed higher expression of transcript coding for the enzyme converting active GAs (GA1, GA3, GA4, and GA7) to inactive GAs (GA34, GA8): GA2-oxidase (VIT\_10s0116g00410, VIT\_19s0140g00120 loose clones vs. 1134). These findings are in agreement with Giacomelli et al. (2013) proposing that the pool of bioactive GAs in grapevine flowers during flowering and fruit set is controlled by a fine regulation of the abundance and localization of GA oxidase transcripts.

Genes involved in GA signaling (**Figure 6**) did not show differential expression between compact and loose clones in flowers but several genes known to be regulated by GA showed greater expression in the compact clones. Interestingly, several GASA-like transcripts showed preferential expression in the compact clones (VIT\_08s0007g05860, VIT\_18s0072g01110, VIT\_14s0066g01790, VIT\_03s0091g00390, and VIT\_14s0108g00740). GASA proteins are involved in diverse processes, and GASA4 in Arabidopsis is present in flower and involved in the seed development and yield (Roxrud et al., 2007). One of the transcripts (VIT\_03s0091g00390) corresponds to the SNAKIN subfamily which is known to be an antimicrobial (Segura et al., 1999) but more recently its role in the cell division was described (Nahirñak et al., 2012).

#### **Auxin**

In grapevine, auxin is a growth factor required for fruit growth. No significant observation could be made on auxin biosynthesis related transcripts to identify a possible greater production in compact or loose clones. More significantly, transcripts involved in the auxin transport (**Figure 6**) were more abundant in flowers of the compact clones, such as PINOID (VIT\_11s0016g04910, VIT\_13s0074g00730) and the auxin efflux carriers PIN3 (VIT\_08s0040g01230, VIT\_17s0000g02420) PIN6 (VIT\_18s0001g15420) PIN5 (VIT\_04s0023g00320) PIN10 (VIT\_08s0040g01220), and AUX1 (VIT\_03s0038g02140). As mentioned above, it has been recently shown that berries derived from flowers that open first have less probability to abscise than the flowers that open later, and that this ability requires decreased ethylene-related gene expression dependent on polar auxin transport (Kühn et al., 2014). Later, Kühn et al. (2016) found that polar auxin transport and transcripts of four putative PIN genes decreased in conjunction with increased abscission, and the inhibition of polar auxin transport resulted in fruit drop. In this context, over-expression of auxin transporter genes could be related to a greater final number of berries in the cluster by contributing to lower the number of abscised flowers or fruitlets.

In the auxin regulation pathway (**Figure 6**), transcripts coding for proteins related to the early response to auxin were upregulated in flowers of the compact clones, including six transcripts for AUX/IAA and seven transcripts for SAUR. Quantitatively, ARF6 was one of the most differentially expressed genes in the G4 comparison (G4-03 and G4-26). ARF6 is known to be present in the flower and embryo, and in Arabidopsis it was specifically localized in the lower tier of the embryo and suspensors (Rademacher et al., 2011). Recently, Su et al. (2016) found that ARF6 and ARF8 are required in Arabidopsis for gradient auxin response and can mediate auxin-induced gene activation in somatic embryogenesis induction. In tomato, down-regulation of ARF6 and ARF8 by microRNA 167 led to floral development defects and female sterility (Liu et al., 2014). ARF4 was the second Auxin Response Factor over-expressed in the compact clone 1134 (in G1-26). It has been characterized in tomato fruit (Sagar et al., 2013), where lowers chloroplast production and starch and is down-regulated by presence of sugars. The expression of ARF4 in tomato increases between anthesis and 4 days post-anthesis and might be involved in fruit set (Zouine et al., 2014). In grapevine ARF4 is more abundant in high seed content berries at ripening (Gouthu and Deluc, 2015).

Cross-talk between GAs and auxins has proven to play an important role during fruit set in tomato via the activation of GA biosynthtetic enzyme GA20 oxydase by auxin (de Jong et al., 2009) two transcripts coding for GA20ox are over-expressed in compact clones (**Figure 6**). In grapevine crosstalk beween these two hormones is also critical in flower set initiation and parthenocarpy (Jung et al., 2014; Lu et al., 2016).

#### Genes Involved in Phenylpropanoids Biosynthesis Show That Important Secondary Metabolites Might Be Specifically Expressed within Clones

A significant number of genes involved in the biosynthesis of phenylpropanoids, flavonoids and anthocyanins were differentially expressed between cluster types, although most of them showed isogenes preferentially expressed in either compact or loose clones. There were also differences in the transcript abundance of genes affecting the production of several important secondary metabolites. All but three of the 46 stilbene synthase genes were preferentially expressed in flowers of the loose clones vs. clone 1134 (**Supplementary Image 7**). It was shown that over-expression of grapevine stilbene synthase VIT\_16s0100g00910 can induce parthenocarpy in tomato (Ingrosso et al., 2011) and thus this gene might be related to the control of berry number. These authors also observed that greater amounts of stilbene were related to pollen sterility.

Several transcripts coding for enzymes potentially involved in the anthocyanin biosynthesis showed preferential expression in flowers of compact clones (**Supplementary Image 9**), including three Anthocyanidine rhamnosyltransferase (VIT\_00s0820g00020, VIT\_15s0046g01950, VIT\_00s0218g00140) and three Anthocyanidin 3-Oglucoside-6"-O-malonyltransferase (VIT\_12s0134g00660, VIT\_12s0134g00620, VIT\_12s0134g00640). The latter two were up-regulated in clusters with small berries. The earlier steps in the phenylalanine biosynthesis (**Supplementary Image 10**) also showed a greater gene expression in the flowers of the compact clones, including shikimate dehydrogenase (VIT\_14s0030g00650, VIT\_14s0030g00660), shikimate kinase (VIT\_18s0001g01730), and prephenate dehydratase (VIT\_10s0116g01670).

### CONCLUSIONS

The characterization of the differential expression in clones of Garnacha Tinta presenting phenotypic differences in traits related to cluster compactness allowed us to identify networks and candidate genes potentially involved in those traits. The flowers at the end of bloom seem to be an organ and developmental stage of crucial importance for the traits studied, while much less differences were observed in spring buds and young berries. Our study focused on the end of flowering which is a particularly active period of rapid changes but other stages could also play important role in compactness and a fine monitoring of the flowering stages would improve our knowledge. In the case of the analysis on berry, the microclimate caused by different compactness levels may also influence the genes expression and make more difficult the discrimination between genetics and environmental factors. All the stable differential traits considered (berry number, seed number and berry size), are potentially affected by the magnitude of cell division rate, and many related gene networks showed different expression levels, indicating a greater division rate in compact clones with more berries (and eventually more seeds or larger berries). Differential expression of transcripts involved in hormone signaling and transport support that auxin and gibberellins play a central role in fruit set, and some identified key genes have been noted. Other hormones, such as ethylene and jasmonate may differentially regulate potential indirect effects, such as the activation of some defense mechanism or polyphenols production.

# AUTHOR CONTRIBUTIONS

JG performed the gene expression analysis and interpretation. JT and NL performed phenotyping analysis. JI designed the study. JG and JI drafted the manuscript. All authors read and approved the final manuscript.

## FUNDING

This work was financially supported by the projects AGL2014- 59171R (co-funded by FEDER) and AGL2010-15694 and the Ramon y Cajal grant RYC-2011-07791, all from the Spanish MINECO. JT was the recipient of a predoctoral fellowship from MINECO (Grant: BES-2011-047041).

### ACKNOWLEDGMENTS

The authors acknowledges R. Aguirrezábal, S. Hernáiz, B. Larreina, M. I. Montemayor, and E. Vaquero for their technical assistance. We acknowledge Gobierno de La Rioja for the collection and maintenance of the clones. The authors would also like to thanks Anne Fenell for critical review of the manuscript.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2017. 00632/full#supplementary-material

Supplementary Image 1 | Adapted Cytoscape networks including transcript differentially expressed in flowers between loose and compact clones related to cell wall metabolism. Genes over-expressed in compact clones in all comparisons are in dark red. Genes over-expressed in compact clones in 2 or 3 comparisons are in red. Genes over-expressed in loose clones in all comparisons are in dark green. Genes over-expressed in loose clones in 2 or 3 comparisons are in green. Figure is adapted from networks 40006 from Grimplet et al. (2009).

Supplementary Image 2 | Adapted Cytoscape networks including transcript differentially expressed in flowers between loose and compact clones related to cell cycle. Genes over-expressed in compact clones in all comparisons are in dark red. Genes over-expressed in compact clones in 2 or 3 comparisons are in red. Genes over-expressed in loose clones in all comparisons are in dark green. Genes over-expressed in loose clones in 2 or 3 comparisons are in green. Figure is adapted from networks 44110 from Grimplet et al. (2009).

Supplementary Image 3 | Adapted Cytoscape networks including transcript differentially expressed in flowers between loose and compact clones related to regulation of actin cytoskeleton. Genes over-expressed in compact clones in all comparisons are in dark red. Genes over-expressed in compact clones in 2 or 3 comparisons are in red. Genes over-expressed in loose clones in all comparisons are in dark green. Genes over-expressed in loose clones in 2 or 3 comparisons are in green. Figure is adapted from networks 44810 from Grimplet et al. (2009).

Supplementary Image 4 | Adapted Cytoscape networks including transcript differentially expressed in flowers between loose and compact clones related to R proteins from plant–pathogen interaction. Genes over-expressed in compact clones in all comparisons are in dark red. Genes over-expressed in compact clones in 2 or 3 comparisons are in red. Genes over-expressed in loose clones in all comparisons are in dark green. Genes over-expressed in loose clones in 2 or 3 comparisons are in green. Figure is adapted from networks 34627 from Grimplet et al. (2009).

Supplementary Image 5 | Adapted Cytoscape networks including transcript differentially expressed in flowers between loose and compact clones related to plant–pathogen interaction. Genes over-expressed in compact clones in all comparisons are in dark red. Genes over-expressed in compact clones in 2 or 3 comparisons are in red. Genes over-expressed in loose clones in all comparisons are in dark green. Genes over-expressed in loose clones in 2 or 3 comparisons are in green. Figure is adapted from networks 34626 from Grimplet et al. (2009).

Supplementary Image 6 | Adapted Cytoscape networks including transcript differentially expressed in flowers between loose and compact clones related to ethylene signaling. Genes over-expressed in compact clones in all comparisons are in dark red. Genes over-expressed in compact clones in 2 or 3 comparisons are in red. Genes over-expressed in loose clones in all comparisons are in dark green. Genes over-expressed in loose clones in 2 or 3 comparisons are in green. Figure is adapted from networks 30008 from Grimplet et al. (2009).

Supplementary Image 7 | Adapted Cytoscape networks including transcript differentially expressed in flowers between loose and compact clones related to phenylpropanoids biosynthesis. Genes over-expressed in compact clones in all comparisons are in dark red. Genes over-expressed in compact clones in 2 or 3 comparisons are in red. Genes over-expressed in loose clones in all comparisons are in dark green. Genes over-expressed in loose clones in 2 or 3 comparisons are in green. Figure is adapted from networks 10940 from Grimplet et al. (2009).

Supplementary Image 8 | Adapted Cytoscape networks including transcript differentially expressed in flowers between loose and compact clones related to flavonoids biosynthesis. Genes over-expressed in compact clones in all comparisons are in dark red. Genes over-expressed in compact

#### REFERENCES


clones in 2 or 3 comparisons are in red. Genes over-expressed in loose clones in all comparisons are in dark green. Genes over-expressed in loose clones in 2 or 3 comparisons are in green. Figure is adapted from networks 10941 from Grimplet et al. (2009).

Supplementary Image 9 | Adapted Cytoscape networks including transcript differentially expressed in flowers between loose and compact clones related to anthocyanin biosynthesis. Genes over-expressed in compact clones in all comparisons are in dark red. Genes over-expressed in compact clones in 2 or 3 comparisons are in red. Genes over-expressed in loose clones in all comparisons are in dark green. Genes over-expressed in loose clones in 2 or 3 comparisons are in green. Figure is adapted from networks 10942 from Grimplet et al. (2009).

Supplementary Image 10 | Adapted Cytoscape networks including transcript differentially expressed in flowers between loose and compact clones related to ethylene signaling. Genes over-expressed in compact clones in all comparisons are in dark red. Genes over-expressed in compact clones in 2 or 3 comparisons are in red. Genes over-expressed in loose clones in all comparisons are in dark green. Genes over-expressed in loose clones in 2 or 3 comparisons are in green. Figure is adapted from networks 10400 from Grimplet et al. (2009).

#### Supplementary Table 1 | Phenotypic data and pair-wise t-tests of the selected clones during three seasons.

Supplementary Table 2 | Gene expression values data. Sheet 1: normalized data for each samples, value are expressed as log2 of intensity. Sheet 2: functional annotation. Sheet 3: significance of the differential expression, ratio of transcripts expression in each comparison, genes meeting the cutoff of fold change >2 and P < 0.05 are reported as 1 for compact clones and −1 for loose clones.


for grapevine bunch compactness. Aust. J. Grape Wine Res. 21, 277–289. doi: 10.1111/ajgw.12121


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer SDS and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review.

Copyright © 2017 Grimplet, Tello, Laguna and Ibáñez. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Abscisic Acid Is a Major Regulator of Grape Berry Ripening Onset: New Insights into ABA Signaling Network

Stefania Pilati<sup>1</sup> \*, Giorgia Bagagli<sup>1</sup> , Paolo Sonego<sup>1</sup> , Marco Moretto<sup>1</sup> , Daniele Brazzale<sup>1</sup> , Giulia Castorina<sup>2</sup>† , Laura Simoni<sup>2</sup> , Chiara Tonelli<sup>2</sup> , Graziano Guella3,4, Kristof Engelen<sup>1</sup> , Massimo Galbiati<sup>2</sup> and Claudio Moser<sup>1</sup>

#### Edited by:

José Tomás Matus, Centre for Research in Agricultural Genomics (CRAG), Spain

#### Reviewed by:

Fatma Lecourieux, Centre National de la Recherche Scientifique (CNRS), France Javier Ibáñez, Instituto de Ciencias de la Vid y del Vino, Spain Alonso Gastón Pérez-Donoso, Pontificia Universidad Católica de Chile, Chile

\*Correspondence:

Stefania Pilati stefania.pilati@fmach.it

#### †Present address:

Giulia Castorina, Dipartimento di Scienze Agrarie e Ambientali – Produzione, Territorio, Agroenergia, Università degli Studi di Milano, Milan, Italy

#### Specialty section:

This article was submitted to Crop Science and Horticulture, a section of the journal Frontiers in Plant Science

> Received: 03 March 2017 Accepted: 06 June 2017 Published: 21 June 2017

#### Citation:

Pilati S, Bagagli G, Sonego P, Moretto M, Brazzale D, Castorina G, Simoni L, Tonelli C, Guella G, Engelen K, Galbiati M and Moser C (2017) Abscisic Acid Is a Major Regulator of Grape Berry Ripening Onset: New Insights into ABA Signaling Network. Front. Plant Sci. 8:1093. doi: 10.3389/fpls.2017.01093 <sup>1</sup> Research and Innovation Centre, Fondazione Edmund Mach, San Michele all0Adige, Italy, <sup>2</sup> Dipartimento di Bioscienze, Università degli Studi di Milano, Milan, Italy, <sup>3</sup> Department of Physics, Bioorganic Chemistry Lab, University of Trento, Trento, Italy, <sup>4</sup> Istituto di Biofisica, Consiglio Nazionale delle Ricerche, Trento, Italy

Grapevine is a world-wide cultivated economically relevant crop. The process of berry ripening is non-climacteric and does not rely on the sole ethylene signal. Abscisic acid (ABA) is recognized as an important hormone of ripening inception and color development in ripening berries. In order to elucidate the effect of this signal at the molecular level, pre-véraison berries were treated ex vivo for 20 h with 0.2 mM ABA and berry skin transcriptional modulation was studied by RNA-seq after the treatment and 24 h later, in the absence of exogenous ABA. This study highlighted that a small amount of ABA triggered its own biosynthesis and had a transcriptome-wide effect (1893 modulated genes) characterized by the amplification of the transcriptional response over time. By comparing this dataset with the many studies on ripening collected within the grapevine transcriptomic compendium Vespucci, an extended overlap between ABAand ripening modulated gene sets was observed (71% of the genes), underpinning the role of this hormone in the regulation of berry ripening. The signaling network of ABA, encompassing ABA metabolism, transport and signaling cascade, has been analyzed in detail and expanded based on knowledge from other species in order to provide an integrated molecular description of this pathway at berry ripening onset. Expression data analysis was combined with in silico promoter analysis to identify candidate target genes of ABA responsive element binding protein 2 (VvABF2), a key upstream transcription factor of the ABA signaling cascade which is up-regulated at véraison and also by ABA treatments. Two transcription factors, VvMYB143 and VvNAC17, and two genes involved in protein degradation, Armadillo-like and Xerico-like genes, were selected for in vivo validation by VvABF2-mediated promoter trans-activation in tobacco. VvNAC17 and Armadillo-like promoters were induced by ABA via VvABF2, while VvMYB143 responded to ABA in a VvABF2-independent manner. This knowledge of the ABA cascade in berry skin contributes not only to the understanding of berry ripening regulation but might be useful to other areas of viticultural interest, such as bud dormancy regulation and drought stress tolerance.

Keywords: Abscisic acid (ABA), grapevine (Vitis vinifera), berry ripening, RNA sequencing, promoter analysis, AREB/ABF

# INTRODUCTION

fpls-08-01093 June 21, 2017 Time: 11:34 # 2

Grape is a traditional world-wide cultivated crop, whose fruit is consumed fresh or dried as table grapes, fermented to produce wines, spirits and vinegar or transformed into pharmaceutical health-promoting products. The process of fruit development has been intensively studied initially to improve quality of production and more recently to maintain high quality under changing climatic conditions. Grape berry development can be divided into two phases of berry growth: an initial phase from fruit set until green hard berries, characterized by embryo maturation in the seeds, pericarp intense cell duplication, malic and tartaric acid accumulation and proanthocyanidin synthesis, and a final phase of ripening, characterized by fruit softening, mesocarp cell enlargement and sugar and aroma accumulation and skin coloring. The onset of ripening, that is the transition from the first to the second phase, implies an extensive reprogramming of the berry transcriptome, as observed in several "-omic" studies (Deluc et al., 2007; Pilati et al., 2007; Fasoli et al., 2012; Lijavetzky et al., 2012). An integrated network analysis recently highlighted that this transition occurs via an extensive gene down-regulation driving the suppression of the vegetative growth metabolism and the activation of maturation-specific pathways (Palumbo et al., 2014).

Such transcriptional and metabolic reprogramming is orchestrated by numerous signals, such as hormones, in particular ABA, ethylene and brassinosteroids, reviewed in (Kuhn et al., 2014), physiological modifications, such as cell turgor and elasticity (Castellarin et al., 2011), and metabolic factors, such as sugar and reactive oxygen species accumulation (Gambetta et al., 2010; Pilati et al., 2014). However, their reciprocal influence on ripening inception has not been disentangled so far, due to the complexity of the system. Recently, Castellarin et al. (2015) proposed a timeline of events leading to the onset of ripening: an initial fall of elasticity and turgor pressure in the berry is followed by ABA and sugar accumulation and then skin coloring. Yet, an ABA sharp increase at ripening onset has been reported in numerous studies (Deluc et al., 2007; Wheeler et al., 2009; Gambetta et al., 2010; Sun et al., 2010). The fast accumulation of ABA in the cells is due the its typical positive feedback loop, triggered by a small amount of hormone coming possibly from the leaves (Antolìn et al., 2003) or by diffusion from the mature, dormancy-acquiring seeds (Kondo and Kawai, 1998). This ripening-specific accumulation is reported also in peach, sweet cherry and tomato (Zhang et al., 2009; Sun et al., 2012; Tijero et al., 2016).

Abscisic acid regulates a variety of plant developmental processes, such as leaf senescence, seed maturation and dormancy, bud dormancy and adaptive responses to abiotic and biotic stresses, in particular drought and salinity, by means of stomata closure, osmotic potential regulation and/or wax deposition (Nambara and Kuchitsu, 2011). During plant evolution, ABA conserved its ancestral role in cellular responses modulation to stimuli affecting the cell water status, but acquired new functions in the regulation of different processes, sometimes also in a species specific way (Takezawa et al., 2011; Wanke, 2011). In fleshy fruit ripening, the relationship between ABA and sugar accumulation and turgor pressure, which together determine water uptake and cell enlargement, could represent an example of acquired functionality of ABA in angiosperms (Wheeler et al., 2009; Gambetta et al., 2010). In tomato, both a transgenic line blocked in ABA synthesis and one overexpressing a transcription factor activating ABA response demonstrated the effect of ABA on tomato texture and firmness and primary metabolites accumulation (Sun et al., 2012; Bastías et al., 2014). A similar decrease in fruit firmness was observed transforming tomato with the Vitis homolog of this transcription factor (Nicolas et al., 2014). ABA effect on berry skin coloring was highlighted in studies focused on seedless varieties of table grapes, where the absence of seeds correlated with low amounts of ABA and low color development, recovered by ABA treatments (Kondo and Kawai, 1998; Ferrara et al., 2013). Nonetheless, studies focused on the effects of adverse external conditions highlighted that anthocyanin accumulation in condition of water stress required not only ABA, but also sugar accumulation and possibly other stimuli, suggesting that ABA is not the only necessary signal for color development, or it can exert an indirect effect (Castellarin et al., 2007; Wheeler et al., 2009; Gambetta et al., 2010; Ferrandino and Lovisolo, 2014).

Abscisic acid signaling network encompasses genes involved in its biosynthesis, degradation, conjugation and transport, whose reciprocal transcription and enzymatic activities determine ABA cellular content, and genes involved in its perception and signaling cascade. The knowledge of this network has burst recently, taking advantage of the combination of molecular, biochemical, forward and reverse genetic studies in Arabidopsis, reviewed in (Finkelstein, 2013) and the availability of the genome sequence of Vitis vinifera (Jaillon et al., 2007; Velasco et al., 2007), which accelerated the transfer of knowledge from the model plant to this crop. The early steps of ABA biosynthesis take place in the plastid as part of the MEP pathway leading to carotenoids production. NCED catalyzes the first committed step in ABA biosynthesis, and is rate-limiting, reviewed in (Nambara and Marion-Poll, 2005). NCEDs are encoded by multigene families and the expression of the specific family members is tightly regulated in response to stress or developmental signals contributing to ABA synthesis in different contexts. In addition to ABA synthesis, catabolism is a major mechanism for regulating ABA levels: ABA can be irreversibly hydroxylated at the 8<sup>0</sup> position by P-450 type monooxygenases to give an unstable intermediate (8<sup>0</sup> -OH-ABA) isomerized to phaseic acid; or reversibly esterified to ABA-glucose ester (ABA-GE), which can accumulate in vacuoles or apoplast as storage. Transporters of the G subfamily of the ATP-binding cassette (ABC) family mediate the import and export of ABA through the plasmalemma (Jarzyniak and Jasinski, 2014).

**Abbreviations :** ABA, abscisic acid; ABRE, ABA-responsive element; AREB/ABF, ABRE-binding proteins/ABRE-binding factors; CIPK, CBL-interacting protein kinase; CPK, calcium-dependent protein kinase; NCED, 9-cis-epoxycarotenoid dioxygenase; PP2C, protein phosphatase 2C; PYR/PYL/RCAR, pyrabactin resistance1/PYR1-like/regulatory components of ABA receptors; RBOH, respiratory burst oxidase homolog; SnRK2, sucrose-non-fermenting1-related kinase 2.

Abscisic acid perception and signaling in grapevine has been recently elucidated in root and leaf (Boneh et al., 2011, 2012) and in fruit (Gambetta et al., 2010), by identifying and partially characterizing ABA receptors, PP2Cs and SnRK2 kinases. The best characterized ABA receptors in Arabidopsis are soluble proteins of the family PYR (pyrabactin resistant), PYL (PYR-like) or RCAR (regulatory component of ABA receptor). Eight RCARs, four of which were induced by ABA in leaf, were identified in grapevine (Boneh et al., 2011). ABA binds to PYR/PYL/RCAR proteins, resulting in a conformational change that enhances stability of a complex with clade A PP2Cs, which in Arabidopsis are all induced by ABA and include the ABA insensitive mutants abi1 and abi2 (Merlot et al., 2001). In grapevine, nine PP2Cs have been identified (Gambetta et al., 2010; Boneh et al., 2011): six are induced by ABA in leaf, and five are induced at véraison in the berry. In the absence of ABA, the PP2Cs keep SNF1 related kinases (SnRKs) inactive through physical interaction and dephosphorylation. When ABA binds to its receptors, they recruit PP2C, thus releasing the inhibition of SnRKs which become active by autophosphorylation and activate more than 50 target proteins, which include transcription factors as well as other targets. In grapevine, seven SnRKs were identified and they appeared differently modulated in leaf, root and fruit upon abiotic stresses and development. PP2Cs may also dephosphorylate other classes of kinases, e.g., the ABA-stimulated calcium-dependent protein kinase (ACDK), linking ABA to calcium signaling (Yu et al., 2006) and widening the cascade. Among the transcription factors activated by ABA, a subgroup of the bZIP family, called AREB/ABF (Liu et al., 2014), are directly activated by SnRKs (Yoshida et al., 2010): in grapevine 11 putative ABAresponsive bZIPs have been identified by sequence homology and one of them, VvABF2, has been recently characterized (Nicolas et al., 2014).

The present work analyzes the early transcriptional events occurring in green hard, pre-véraison berry skin treated with exogenous ABA showing the dramatic effect of this hormone on ripening onset and identifying candidates targets of VvABF2, thus expanding our knowledge on ABA network in the fruit.

# MATERIALS AND METHODS

# Plant Material, Biochemical Analyses, ABA Treatment

During 2011, two clusters of V. vinifera cv. Pinot Noir ENTAV115 were collected almost daily between 9 am and 10 am at 6–7 weeks post-flowering (wpf), corresponding to EL-33 and EL-34, at the FEM study site (San Michele all0Adige-TN Italy). Each cluster represented one biological replicate. Each cluster was divided in smaller bunches and then half of them, at random, were pressed for must analysis by means of Fourier transform infrared spectroscopy (FTIR) using the instrument WineScanTM Type 77310 (Foss Electric, Denmark) while the remaining small bunches were rapidly frozen. Frozen berries were peeled with a scalpel and the skins were ground to obtain a fine powder. Skin anthocyanin concentration was measured after methanol extraction (1 g berry skin powder in 10 mL methanol) according to the double pH differential method (Cheng and Breen, 1991). Lipid extraction and analysis were performed as described in Pilati et al. (2014). ABA detection and quantification was carried by LC-UV-MS technique. In particular we used as stationary phase a column Kinetex C18 (5 µm, 150 mm × 4.6 mm, flow 0.8 mL/min) and as mobile phase an A:B gradient elution (A = H2O + 0.5% formic acid; B = MeOH + 0.5% formic Acid) with B changing from 40 to 55% in 10 min. 10 µL of pure ABA solution or raw skin berry extracts (solution prepared in MeOH/CHCl<sup>3</sup> 9:1) were injected in every chromatographic run. The retention time of ABA was 8.9 min. ABA was detected both by ion-positive mode ESI-MS analysis (characteristic ions at m/z 287, 265, 247, 229, 201, 187, and 173) and by photodiode array detector. ABA was quantified by interpolation on a working curve (absorbance vs. concentration) built on three ABA solutions at concentration 1.9, 19, and 190 ng/µL and the corresponding absorbance measured at fixed wavelength of 262 nm (R <sup>2</sup> = 0.998). During 2013, three clusters of V. vinifera cv. Pinot Noir ENTAV115 were collected between 9 and 10 am at 7 wpf, corresponding to EL-33, at the FEM study site (San Michele all0Adige-TN Italy). Each cluster represented one biological replicate. Berries were detached from each cluster by cutting the petiole, at 2–3 mm distance from the berry. Berries were washed in 0.1% Plant Preservative Mixture (PPM, Duchefa) water solution for 30 min. 20 berries per cluster were put in a 100 mL water solution containing 0.2 mM ABA (Sigma) and 0.5% methanol (used to dissolve ABA) or 0.5% methanol as control. After 20 h mild shaking, the berries were extensively rinsed and then eight berries were frozen in liquid nitrogen. The remaining berries, both treated and not treated with ABA, were plated on solid medium in Petri dishes (0.9% agar, 10% sucrose, 0.1% PPM) for 24 h, and then eight were frozen as described above. ABA and sucrose concentrations were taken from Gambetta et al. (2010).

# RNA Extraction and Expression Analysis by qPCR

Total RNA was extracted from the skin powder samples using Spectrum Total Plant RNA kit (Sigma) and was quantified using a Nanodrop 8000 (Thermo Scientific). The integrity was checked using Bioanalyzer 2100 (Agilent) and RNA Nano Chips. For Real-time PCR analyses, first strand cDNA was synthesized from 2 µg RNA using the SuperScript VILO cDNA Synthesis Kit (Invitrogen) according to the manufacturer's instructions. The cDNAs were mixed with Fast SYBR Green Master Mix (Applied Biosystems) and amplified on a ViiA 7 Real Time PCR System (Applied Biosystems) using an initial heating step at 95◦C for 20 min, followed by 40 cycles of 95◦C for 1 min and 60◦C for 20 s, using the primers reported in Supplementary Table S1. Raw fluorescence data were extracted using Viia 7 Software v1.0. Ct and reaction efficiency were calculated using LinRegPCR (Ruijter et al., 2009). Relative expression was calculated according to (Pfaffl, 2001) by centering expression values for each gene on the mean value. Three reference genes (Actin, SAND and GAPDH) were used for normalization with geNorm (Vandesompele et al., 2002). For RNA sequencing, 2 µg RNA for each sample were shipped in dry ice to Genomicx4life (Salerno, Italy).

# RNA-Seq Analysis and Identification of Differentially Expressed Genes

Sequencing has been performed on Illumina Hi-seq 1500, producing 100 nt directional single-end reads. Raw reads were pre-processed for quality using fastqc v.0.11.2<sup>1</sup> and cleaned with cutadapt v.1.12 (Martin, 2011). The resulting reads were aligned to the grape (12x v1<sup>2</sup> ) genome using the Subread aligner (Liao et al., 2013). Raw read counts were extracted from the Subread alignments using the featureCount read summarization program (Liao et al., 2014). The summarized read count data was used to identify DEGs among various treatments by using the voom method (Law et al., 2014), which estimates the mean-variance relationship of the log-counts, generating a precision weight for each observation that is fed into the limma empirical Bayes analysis pipeline (Smyth, 2004). A Volcano Plot generated using the ShinyVolcanoPlot Web App<sup>3</sup> was used to select sets of DEGs for each comparison based on both p-value and expression fold change. In the present work, a maximum p-value of 0.05 and a minimum absolute fold change of 2 were imposed. Raw sequences were deposited at the Sequence Read Archive of the National Center for Biotechnology<sup>4</sup> under BioProject accession number PRJNA369777.

## Gene Annotation and Promoter Analysis

Vitisnet gene annotation has been used (Grimplet et al., 2009), except for the transcription factors gene families already characterized in grapevine, for which the specific published annotation has been used (Licausi et al., 2010; Wang N. et al., 2013; Liu et al., 2014; Wang et al., 2014; Wong et al., 2016). Functional class enrichment was performed on GO (Gene Ontology) terms (annotation<sup>5</sup> ) using the TopGO Bioconductor package (Alexa and Rahnenfuhrer, 2016) and on Vitisnet gene annotation (Grimplet et al., 2009) taking advantage of the VESPUCCI grape compendium (Moretto et al., 2016). Promoter analysis has been performed on the 1-kb promoters of the genes modulated at 20 and 44 h using DREME software (Bailey, 2011). Statistically enriched motifs were annotated using the "DAP motifs" database for Arabidopsis (O'Malley et al., 2016). The enriched motifs were searched and counted in the dataset using Patmatch software (Yan et al., 2005).

#### Plasmid Constructs

Two type of constructs were prepared for transient expression, using the Gateway system (Karimi et al., 2002). The coding sequence of ABF2 (VIT\_18s0001g10450) was amplified from Pinot Noir berry cDNA using Phusion DNA polymerase (Finnzymes) and the primers ABF2Fw and ABF2rev and cloned into pENTR-D-TOPO vector (Invitrogen), sequenced and transferred into pK7WG2, under the control of 35S promoter. The 1-kb promoters of VvNAC17 (VIT\_19s0014g03290), Armadillo-like (VIT\_17s0000g08080), Xerico-like (VIT\_01s0137g00780), VvMYB143 (VIT\_00s0203g00070) and HB5 (VIT\_04s0023g01330) were amplified from PN40024 genomic DNA using Phusion DNA polymerase and the pairs of primers indicated in Supplementary Table S1. These DNA fragments were cloned in pENTR-D-TOPO, sequenced and transferred into PHGWFS7, upstream of EGFP and GUS reporter genes.

#### Transient Expression Assay in N. benthamiana

Promoter activation assays were performed in 5-week-old Nicotiana benthamiana plants agroinfiltrated as described in Li (2011). Three leaves from four tobacco plants, representing four biological replicates, were co-infiltrated with the activating plasmid pK7WG2:CaMV35S:ABF2 and individual pHGWFS7:promoter:GUS target constructs. Leaves co-infiltrated with the pK7WG2 empty vector and each of the pHGWFS7:promoter:GUS plasmids were used as a control for the trans-activation assay. 48 h after the first infiltration, leaves were infiltrated with 50 µM ABA dissolved in 10 mM MgCl<sup>2</sup> and 0.07% EtOH, or with a mock solution (0.07% EtOH in 10 mM MgCl2). Leaf samples were collected at 15 min, 1 and 3 h after the beginning of the ABA treatment. qPCR analysis of GUS expression was performed as previously described for the grape samples, using the primers qPCR\_GUSF1 and qPCR\_GUSR1 (Supplementary Table S1). GUS expression was normalized using the Elongation factor 1α (EF-1α) gene (AF120093), amplified with primers pEFfw and pEFrev (Schmidt and Delaney, 2010).

# RESULTS

#### Treatment of Pre-véraison Berries with ABA

In order to identify the exact moment preceding the onset of ripening, we focused our attention on the 2 weeks preceding color break in Pinot Noir berries. Samples were collected daily during season 2011 and analyzed for biochemical and molecular parameters which are known from both literature and our experience to change dramatically at ripening inception (Pilati et al., 2014). These include biochemical profiles, such as total acidity, sugar and anthocyanins content (**Figure 1A**), galactolipid peroxidation state and ABA content in berry skins (**Figure 1B**) and gene expression profiles, such as those of lipoxygenase A (LOXA, VIT\_06g0004s01510) – responsible of the enzymatic galactolipid peroxidation – and of NCED1 (VIT\_19s0093g00550) – first committed enzyme in ABA biosynthesis (**Figure 1C**). This preliminary analysis showed that in the 48 h preceding anthocyanins accumulation (on July 14th), all these parameters undergo a transition, which marks the beginning of a distinct developmental phase, i.e., the ripening. This discontinuity implies an extensive regulation occurring within the cells to trigger all the metabolic pathways characterizing the biochemical changes of fruit ripening. To study the role of ABA as a trigger of ripening, in 2013 we collected berries at the hard green pre-véraison stage (E-L 33)

<sup>1</sup>https://www.bioinformatics.babraham.ac.uk/projects/fastqc/

<sup>2</sup>http://genomes.cribi.unipd.it/

<sup>3</sup>https://github.com/onertipaday/ShinyVolcanoPlot

<sup>4</sup>www.ncbi.nlm.nih.gov/sra

<sup>5</sup>http://genomes.cribi.unipd.it/DATA/V1/ANNOTATION/

and treated them with exogenous ABA. The treatment was performed on detached berries with short petioles in an aqueous medium containing 0.2 mM ABA for 20 h, which allowed for both homogenous ABA diffusion into the berry skin trough functional stomata and accurate experimental reproducibility. After 20 h, eight berries were collected, rinsed with water and frozen while the remaining were rinsed and plated for additional 24 h on solid medium containing 10% sucrose, in the absence of ABA, and then frozen (**Supplementary Figure S1**).

expression in the skin of berries treated with 0.2 mM ABA for 20 h and then cultured on solid 10% sucrose medium without ABA for 24 h (light gray) and controls (dark gray). Normalized relative quantities were obtained by RT-PCR analysis using the two best reference genes. Data represent the mean of three biological replicates.

Sucrose was added for avoiding osmotic stress according to (Gambetta et al., 2010) in order to mimic berry sugar content at véraison (6–7◦Brix).Treatment efficacy has been verified by two independent approaches. Firstly, ABA was quantified in skin samples of control and ABA treated berries by HPLC-UV-MS showing an average accumulation of 1.62 micrograms ABA/gr FW skin powder at 20 h and 0.55 µg/gr FW at 44 h in treated samples (**Figure 2A**). The latter value was similar to physiological values measured at véraison in other studies, such as 300 ng/gr FW in Wheeler et al. (2009) and 200 ng/gr FW in Sun et al. (2010), whereas higher ABA level at 20 h could be explained by ABA direct uptake. The very low ABA measured in control samples is consistent with berries at the green hard pre-véraison stage (**Figure 1B**). Secondly, the expression of the two genes known to be up-regulated at véraison, LOXA and NCED1, has been measured by qPCR (**Figure 2B**). NCED1 was up-regulated 66

times by ABA treatment at 20 h and 198 times at 44 h; LOXA showed a similar behavior, as it was up-regulated 4 and 13 times at 20 and 44 h, respectively. These profiles are consistent with those observed in 2011 in the transition from E-L 33 to E-L 34.

Moreover, the induction of NCED1 can explain the intracellular ABA level measured at 44 h.

# ABA Extensively Modulates the Berry Skin Transcriptome

Transcriptomes of treated and control samples were analyzed by RNA-sequencing (Supplementary Table S2). Principal component analysis (PCA, **Figure 3**) shows that the four conditions are well-separated, while the biological replicates are grouped together. Major variance (41.5%) distinguished the two time-points (20 vs. 44 h), while treated vs. untreated berries were neatly separated along the second principal component, which explained 19.5% of the variance. It appeared that ABA treatment extensively impacted on pre-véraison berry skin transcriptome.

Treated vs. untreated samples within each time-point were statistically compared to extract the lists of significantly modulated genes and a further restriction on fold change (greater than 2) was applied. 871 genes resulted modulated at 20 h by ABA and 1512 at 44 h; 490 genes were modulated at both time-points, with a coherent trend (**Figure 4**; and Supplementary Table S3). Interestingly, ABA-induced transcriptional modulation increased over time, regardless of the absence of the external stimulus. This amplification in the response suggests that ABA likely acted as a primer of a broad cellular program. According to RNA-seq analysis, NCED1 was induced 32 and 64 times at 20 and 44 h, respectively, whereas LOXA was induced 3 and 12 times, thus confirming previous qPCR analysis. Genes have been functionally annotated using Vitisnet (Grimplet et al., 2009) integrated with manual curation. For functional class enrichment analyses both Vitisnet and Gene Ontology were used (Supplementary Table S4). Genes modulated exclusively at 20 h, the less abundant group, were enriched in classes related to stress, cell wall modification, photosynthesis, respiration and translation. They could represent a stress response due to the excess or sudden delivery of ABA in the treatment, which induced a high turn-over of proteins involved in basic energy metabolism and cell wall. The set of ABA-positively modulated genes at both 20 and 44 h was enriched in genes involved in cell regulation: ABA and ethylene networks were over-represented, along with transcription factors related to these hormones, such as members of the large bZIP, APETALA2 and MYB families. At 44 h post-treatment, the number of genes positively modulated equaled that of the negatively modulated and the functional categories related to the metabolic pathways typical of the ripening process were enriched. Lipid and carbohydrate metabolism, cell wall modification, and flavonoid metabolism were over-represented among the up-regulated genes, supporting the role ABA plays in regulation of sugar metabolism and accumulation, cell enlargement and softening and color development. Ethylene, Auxin and ABArelated categories remained over-represented, suggesting that these hormones regulated not only the onset but also the process of grape berry ripening. Photosynthesis was over-represented among the down-regulated genes, suggesting that ABA triggers the switching off of this basal metabolism inducing the transition to a specialized sink organ, such as the ripe berry.

## Most of ABA Responsive Genes Are Involved in Ripening

In order to outline the role of ABA at ripening onset, a metaanalysis using the grapevine expression data compendium Vespucci (Moretto et al., 2016) was performed. Seven experiments in which berry ripening transcriptome had been analyzed were selected to visualize how the sets of genes modulated in the skin by ABA at the two time-points were modulated during physiological ripening. These experiments were performed in different berry tissues (seed, pulp, skin, and pericarp) in six different cultivars: Cabernet Sauvignon (GSE11406), Sauvignon Blanc (GSE34634), Corvina (GSE36128), Pinot Noir (GSE49569 and GSE31674), Muscat Hamburg (GSE41206) and Norton (GSE24561). Heatmaps representing the comparisons are shown in **Figure 4**, while the tables which generated the heatmaps, downloaded from Vespucci website, are available as Supplementary Table S5. Within each comparison, a percentage of genes ranging between 49 and 87% were modulated coherently by ABA and during ripening in all tissues and cultivars (**Figure 4**). However, the heatmaps showed also variable profiles, likely related to the tissue and/or cultivar specific modulation of these transcripts. While the smallest overlap occurred with genes up-modulated exclusively at 20 h, the sets of genes modulated at both 20 and 44 h and those modulated only after 44 h showed a more extended overlap with ripening, between 64 and 87%, higher when considering the down-regulated genes. Functional categories enrichment analyses were repeated on the restricted sets of genes modulated both by ABA and during ripening (**Figure 4** and Supplementary Table S4). In general, the analysis reproduced the results described in the previous section, with small refinements, such as for genes up-modulated at 20 and 44 h in which more specific categories related to carbohydrate metabolism, such as aminosugars, galactose and glycerolipids, appeared enriched or changes, as occurred for genes up-modulated at 44 h, in which the two categories of cell wall and flavonoid biosynthesis were lost while that of fruit ripening and abscission were gained. On a broad scale, we show that green hard berries treated with ABA are not simply responding to a stimulus, rather activating genes that are typical of the ripening program.

Palumbo and collaborators highlighted that the transition from immature-to-mature stage in the berry is characterized by an extensive transcriptomic down-regulation, anti-correlated to a small group of 190 so called "switch genes" (Palumbo et al., 2014). Interestingly, we identified in our set of positively modulated genes 80 of such "switch genes" (Supplementary Table S6). This finding is consistent with ABA playing an important role in ripening regulation, partly by switching off typical vegetative pathways, such as those related to photosynthesis. Besides, 13 out of these 80 candidate "ABA-responsive switch genes" were predicted to be regulated post-transcriptionally by miRNA (Palumbo et al., 2014), suggesting that ABA modulation

p-value < 0.05) and twofold change threshold. Green bars on the left are proportional to the number of repressed genes, while red bars on the right to the number of induced genes. On left and right sides, heatmaps of the down- and up-ABA modulated genes obtained in Vespucci, selecting 7 experiments on berry ripening performed in six Vitis vinifera cv.: Cabernet Sauvignon (GSE11406), Sauvignon Blanc (GSE34634), Corvina (GSE36128), Pinot Noir (GSE49569 and GSE31674), Muscat Hamburg (GSE41206) and Norton (GSE24561). Genes are clustered according to their expression profiles. The number of the transcripts coherently modulated in ABA treatment and berry ripening is indicated beside the heatmaps by green/red bars. Outer left and right columns, enriched functional categories of each subset of genes coherently modulated by ABA and during ripening are reported. For the complete output of the GO and Vitisnet enrichment analyses, refer to Supplementary Table S4.

can occur both via direct targets activation and via posttranscriptional mechanisms.

Finally, we compared our results with the list of genes modulated in a grapevine cell culture by a 1-h treatment with 20 µM ABA (Nicolas et al., 2014). Only 55 genes were modulated by ABA in both experiments and they were mostly up-regulated (Supplementary Table S7). They are related to ABA network and cell response to abiotic stresses, such as drought, dehydration, osmotic stress and potentially represent a basal ABA signaling core conserved in any type of cell.

#### ABA Network in Berry Skin at the Onset of Ripening

Abscisic acid treatment of berries at pre-véraison stage significantly modulated several genes involved in ABA metabolism, perception and signaling, which are summarized in **Figure 5**. All the gene families involved in ABA metabolism appeared transcriptionally affected by ABA: NCED family, involved in ABA biosynthesis, ABA 8<sup>0</sup> hydroxylase, in its degradation, ABA glucosidase, in ABA conjugation with sugar moieties and ABC transporters of the G subfamily, involved in ABA transport. The participation of the different NCED isoforms in berry ripening has been widely reported (Lund et al., 2008; Wheeler et al., 2009; Sun et al., 2010; Young et al., 2012). Using Vespucci, we could visualize the expression of the five NCEDs present in grapevine genome in at least three cultivars during berry development and post-harvest withering (**Supplementary Figure S2**). By combining this information with our results, we can state that VIT\_19s0093g00550, called NCED3 in Young et al. (2012) and NCED1 in Sun et al. (2010), is very rapidly induced in the skin by ABA treatment and its upregulation specifically occurs in the pulp and skin tissues at véraison, while the gene is not modulated later in ripening (**Supplementary Figure S2**). In our results, there are two other NCEDs which are modulated though to a lesser extent and only at 44 h: VIT\_02s0087g00910, which is also modulated during post-harvest withering in Corvina, and VIT\_10s0003g03750, which does not seem related to the process of berry ripening (**Supplementary Figure S2**). To understand their function, further investigations are needed. Instead, from Vespucci analysis, the isoform VIT\_02s0087g00930 seems induced during the whole berry ripening in all tissues, slightly more in the pulp.

Concerning ABA perception and signaling, both ABA receptors of the PYL/PYR/RCAR family and PP2C phosphatases were affected by the presence of exogenous ABA. The two receptors (VIT\_08s0058g00470 and VIT\_02s0012g01270), previously identified as RCAR5 and 7 by Boneh et al. (2011) and as PYL3 and PYL1 by Li et al. (2012), were down-regulated in our experiment. Three PP2C phosphatases were strongly induced at 20 and 44 h: VIT\_11s0016g03180 was identified as AtABI1 homolog by phylogenetic analysis (named PP2C-2 in Gambetta et al., 2010), and was characterized in leaf and root (named PP2C4) by Boneh et al. (2011). VIT\_06s0004g05460 and VIT\_13s0019g02200, corresponding to PP2C-6 and PP2C-3 in Gambetta et al. (2010) and PP2C9 and PP2C8 in Boneh et al. (2011) were found induced at ripening onset in developing berries, anticipated in deficit irrigation conditions and induced by ABA in the skin of in vitro cultured berries. Out of the five identified Snf1-related kinases involved in ABA response, one (VIT\_02s0236g00130) corresponding to SnRK2.1 in Boneh et al. (2012) was up-regulated at 20 and 44 h. In **Figure 5**, other kinases belonging to the calcium dependent protein kinases family (CPK/CDPK/CDKs) or calcineurin B-like interacting protein kinases (CIPKs) and the three tiers of mitogen-activated protein kinase cascades (MAPKs, MAPKKs, and MAPKKKs) were included, based on Arabidopsis literature reporting their potential involvement in ABA signaling (reviewed in Finkelstein, 2013). In grapevine, some CPK and CDPK have been studied due to their up-regulation at véraison and relationship to ABA or drought stress (Yu et al., 2006; Cuéllar et al., 2013). Interestingly, also a LRK10 receptor kinase was found induced at 20 and 44 h and was included, as it is induced also in ABA-treated cell culture and in Arabidopsis one isoform, AtLRK10L1.2, is implicated in ABA response and drought resistance (Lim et al., 2015).

Known direct targets of SnRK2 phosphorylation are NADPH oxidases of the Respiratory Burst Oxidase family (Rboh), leading to the production of hydrogen peroxide, plasma membrane anion and K+-channels, regulating ion transport and stomata opening, and basic domain/leucine zipper (bZIP) transcription factors of the ABA responsive element binding factor subgroup (AREB/ABF), mediating ABA-responsive genes transcription (Wang P. et al., 2013). In our experiment, representatives for all these functional categories were modulated (**Figure 5**, top). Interestingly, only one member of the grapevine AREB/ABF predicted subgroup (Liu et al., 2014) was modulated, namely the VvAREB/ABF2 (VvbZIP045, VIT\_18s0001g10450), recently characterized by Nicolas et al. (2014). VvAREB/ABF2 likely represents the isoform phosphorylated by SnRK2 in berry skin cells at ripening onset and activating down-stream genes of the ABA signaling cascade.

### Extending the Regulatory Circuit of ABA Signaling

With the aim of identifying candidate target genes of VvABF2 activity, promoter regions of the genes modulated at 20 and 44 h were analyzed to find significantly enriched cis-acting motifs. The three most enriched motifs that were found were recognized by bZIP transcription factors, NAC and ABF subgroup of the bZIP family, respectively (**Figure 6**). Calculating the frequency of refined consensus motifs highlighted that the ABRE motif, "CACGTGT/GC," was about threefold more represented in our ABA-modulated gene set compared to the whole genome set of promoters. The recurrence of ABREs within each promoter was calculated: it ranged from 0 to 4 and is represented in **Figure 5**. ABREs are present in genes involved in ABA network, like NCEDs, PP2Cs and PYL/RCAR receptors but also in some TFs of the NAC, MYB, HB, bZIP and ERF families. In particular, VvMYB143 and VvNAC17 were the most enriched in ABREs and were also modulated by ABA in cell cultures (Nicolas et al., 2014). VvMYB143 belongs to the subgroup S11 of R2R3-MYB TFs, whose function has not been characterized

the genes chosen for in vivo promoter activation assay by VvABF2.

promoter analysis. Underlined genes are modulated in the same way in grape berry cell cultures treated with ABA, reported in Nicolas et al. (2014). Arrows indicate


FIGURE 6 | In silico promoter analysis of the genes modulated at both 20 and 44 h. The 1 kb promoter regions have been analyzed in DREME (Bailey, 2011) and motif enrichment has been calculated using the whole genome promoters as reference. The three most enriched motifs and the transcription factor families they are likely recognized by, are reported. A refined search of the most enriched exact sequence motifs has been performed using Patmatch (Yan et al., 2005) and the corresponding frequencies are reported in the last two columns.

yet (Wong et al., 2016). VvNAC17 belongs to subgroup III (Wang N. et al., 2013) and has been recently characterized for VvABF2 activation in tobacco protoplasts (Nicolas et al., 2014). Other two genes, VIT\_17s0000g08080 and VIT\_01s0137g00780, were highly modulated by ABA in our experiment and also in grapevine cell culture (Nicolas et al., 2014) and possess 4 and 2 ABREs in their promoters, respectively. The first gene is annotated as Armadillo b-catenin, because it contains the 3D motif Armadillo, involved in binding of large molecules such as proteins or DNA but also the U-box domain, for recruiting E2-adenylated ubiquitin in the protein degradation pathway (Coates, 2003). The gene VIT\_01s0137g00780 was annotated as "unknown." RNA-seq data have been used to refine this gene prediction, revealing an incorrect splicing site (Supplementary Table S8). The improved transcript prediction based on reads mapping was used for homology search by BLAST algorithm in other species, highlighting the presence of a RING Zinc finger domain, as in Xerico (Ko et al., 2006). Interestingly, in Arabidopsis, genes containing these domains are reported to be involved in ABA response modulation and in drought resistance (Ko et al., 2006; Moody et al., 2016). These two genes will be hereafter referred to as Armadillo-like and Xericolike, for brevity. Vespucci analysis showed that VvNAC17 and Armadillo-like were part the genes modulated by ABA at 20 and 44 h and during ripening, while VvMYB143 and Xericolike were characterized by more variable profiles in the different cultivars and tissues and would need more specific investigation to precisely describe their modulation.

These four genes were further investigated in a transient transactivation assay in N. benthamiana to verify their dependence on VvABF2 activity. The transcription factor HB5, activated by ABA but lacking ABREs in its promoter, was included as a negative control. Tobacco leaves were co-infiltrated with the target promoters fused to the reporter GUS along with either the VvABF2 activator or the control empty vector. The constitutive over-expression of ABF2 on its own is insufficient to induce expression of the downstream target genes, as ABF2 activity involves the ABA-dependent phosphorylation of its N-terminal domain, as demonstrated in Arabidopsis for AtABF2 (Fujita et al., 2005). Consequently, 2 days after the agro-infiltration, leaves were re-infiltrated either with 50 µM ABA or with a mock solution. GUS gene expression was analyzed by qPCR at 15 min, 1, and 3 h after the ABA treatment (**Figures 7A–E**). The promoters of VvNAC17 and Armadillo-like showed a remarkable activation by VvABF2 following ABA treatment, in agreement with the presence of four ABREs in their sequences (**Figure 7F**). Expression of VvMYB143, which contains two ABRE motifs in its promoter, was activated by ABA, irrespectively of the presence of VvABF2. Conversely, despite the presence of two ABF binding domains, the Xerico-like promoter did not show any activation in all the conditions, as observed for the negative control gene HB5.

# DISCUSSION

This work outlines the importance of ABA in the initial phases of berry skin ripening and describes in detail the molecular components of its network, discriminating ripening-specific isoforms and identifying new elements of the signaling cascade.

The positive correlation between the rate of ABA accumulation and the ripening rate of the berry has been firstly reported in 1973 (Coombe and Hale, 1973). Since then, numerous studies highlighted the reciprocal influence and importance of ABA, ethylene, auxins and brassinosteroids at the onset of ripening, based on the observation that variations in the level of an hormone affects the relative concentration of the others and consequently the timing of ripening (Coombe and Hale, 1973; Davies et al., 1997; Symons et al., 2006; Sun

fpls-08-01093 June 21, 2017 Time: 11:34 # 10

et al., 2010; Su et al., 2015). A previous work investigated at the transcriptomic level the effect of ABA on berry ripening, by means of the Affymetrix Vitis Genechip (Koyama et al., 2010). More recently two additional works, performed using NimbleGen whole genome arrays, addressed the transcriptional response to ABA in grapevine cell cultures and in different grapevine organs (Nicolas et al., 2014; Rattanakon et al., 2016). These studies highlighted that ABA is a ubiquitous signal raising specific responses according to the cell and tissue type and to the precise developmental stage. In the present work, great attention has been paid to the experimental setting, in the attempt to simulate the molecular events occurring in the berry just before ripening starts. Therefore, the precise developmental stage of green hard berry has been identified by means of a preliminary study in 2011, based on daily sampling during the 2 weeks before color break. This study suggested three molecular markers useful to define the pre-véraison stage, characterized by very low content of ABA and very low expression levels of the two genes LOXA and NCED, and the transition to the ripening stage, characterized by a significant increase in all these values.

In order to capture the early events of ABA response in the context of berry skin ripening onset regulation, green hard berries were collected in 2013 and treated with ABA for 20 h in liquid and then cultured for additional 24 h in the absence of exogenous ABA, to assay its role as a trigger. ABA uptake and its effect on berry development have been initially assayed by measuring the values of the three markers, which suggested the occurrence of the transition (**Figure 2**). Then berry skin transcriptomes have been analyzed by RNA-seq. PCA analysis highlighted that ABA treatment extensively impacted on berry

skin cells transcriptome (**Figure 3**). In fact, 871 genes were modulated by ABA compared to mock treated samples after 20 h, and 1512 after 44 h, indicating that the response to ABA amplified over time, as a signaling cascade (**Figure 4**). Even though this cascade has been triggered by a small amount of exogenous ABA, we actually know from the up-regulation of the enzyme NCED and the measured intracellular ABA level at 44 h (**Figure 2**) that ABA triggered its own biosynthesis through a positive feedback loop. As our focus was on the role of ABA in ripening onset, we narrowed our attention on those genes modulated by ABA during physiological ripening. This comparison has been performed using the grapevine expression data compendium Vespucci (**Figure 4**). Remarkably, 1346 (71%) ABA-skin responsive genes appeared modulated also during ripening in the berry tissues and cultivars considered, strongly supporting the regulatory role this hormone has in this non-climacteric plant. Nonetheless, this analysis highlighted genes characterized by a more variable profile, related to tissue and/or cultivar specific modulation, which will deserve further characterization. Based on functional enrichment analyses, the categories related to signaling were over-represented within the genes induced at 20 and 44 h (**Figure 4** and Supplementary Table S4), namely the whole ABA network, from biosynthesis and perception to bZIP transcription factors and the ethylene signaling cascade mediated by AP2/EREBP transcription factors. It is interesting to note that some ethylene responsive factors were rapidly modulated by ABA before ethylene biosynthesis was stimulated, suggesting that they might actually represent points of convergence between the two hormones signaling cascades, explaining the tight interconnection between these two hormones at the onset of ripening (Sun et al., 2010). Even though not statistically over-represented, many genes involved in auxin response (IAA, SAUR, ARF), metabolism (GH3) and transport and five genes involved in brassinosteroids biosynthesis and signaling were modulated by ABA, supporting the previously proposed model of a complex integrated signaling network (Kuhn et al., 2014).

Conversely, functional classes related to ripening-specific metabolisms were found over-represented at the 44 h timepoint. In particular, the up-regulated set was enriched in starch and sucrose metabolism, ethylene signaling, including AP2/EREBP transcription factors, whereas the down-modulated genes were mainly related to photosynthesis. The occurrence of this modulation at 44 h might reflect an indirect effect of ABA on these pathways, possibly mediated by other signals, such as ethylene and/or sugars. Besides, the modulation of many genes related to ions and water transport and cell wall modification supports the role of ABA in the fine tuning of sugar accumulation and water uptake to control cell osmosis which, in concert with cell wall structure modifications, drives the process of cell distension in the skin at the onset of berry ripening. The role of ABA in stimulating color accumulation through the activation of regulatory and structural genes of the anthocyanin pathway is still elusive. Treatments with this hormone after véraison were effective in stimulating berry coloring (Peppi et al., 2008; Ferrara et al., 2013; Katayama-Ikegami et al., 2016). Other studies showed that sugars were effective in promoting anthocyanins accumulation both in grapevine cell cultures (reviewed in Lecourieux et al., 2014) and in vitro berry cultures (Dai et al., 2014). In our analysis, we observed the up-regulation of the TF VvMYBA2 and of some structural genes such as flavonoid 3<sup>0</sup> 5 <sup>0</sup> hydroxylases and UDPglucose: anthocyanidin 5,3-O-glucosyltransferases, at 44 h, even though the whole pathway was not statistically enriched and we did not observe berry coloring. This delay was probably due to the fact that other cofactors from the MYB-bHLH-WD40 complex (such as MYC1 or MYCA1) were still not induced, suggesting that other signals beside ABA were required. Interestingly, VvMYBA7, recently characterized as a regulator of the anthocyanins synthesis in vegetative organs, was up-regulated by ABA but only at the 20 h time-point, suggesting its direct induction by ABA, but not its participation to the ripening program (Matus et al., 2017). The role of ABA in downregulating photosynthesis under stressful conditions, such as drought, salinity or low temperature, and during developmental processes, such as senescence (Lee et al., 2004; Yang et al., 2011; Gao et al., 2016), has been extensively described. It is not surprising then that at 44 h the functional enriched categories within the down-regulated genes involve different aspects of the photosynthetic metabolism. The switching off of this metabolism, central in green vegetative tissues, requires a highly regulated and concerted ensemble of reactions in order to avoid dangerous accumulation of reactive oxygen species or metabolic unbalances. The comparison with a previous meta-analysis focused on the genes involved in the transition from immature-to-mature stage in the berries highlighted that a relevant proportion (42%) of these genes were up-regulated by ABA (Palumbo et al., 2014). This observation supports the importance of ABA in triggering the transition to ripening and the down-regulation of the photosynthetic metabolism (Supplementary Table S5). Moreover, this comparison indicated that ABA regulates gene expression not only at the transcriptional level, by means of transcription factors modulation, but also post-transcriptionally, modulating miRNAs transcription.

As mentioned above, ABA is an ancient and ubiquitous signaling molecule, evolutionary linked to plant adaptation to dry terrestrial land (Takezawa et al., 2011; Wanke, 2011). It controls transpiration, dehydration tolerance and other water status-associated processes such as seed and bud maturation and dormancy, root growth, leaf morphogenesis and senescence, thus making this compound a key player in integrating plants growth and development with environmental conditions (Nambara and Kuchitsu, 2011). The specificity of cellular ABA response must thus rely on specific isoforms of upstream perception components and on the wide variety of downstream signaling cascades. In this study, such ripening specific isoforms have been identified (**Figure 5**) and they represent strong candidates for experimental validation. Concerning the ABA biosynthetic enzyme NCED, a véraison specific isoform, VIT\_19s0093g00550 was identified (**Supplementary Figure S2**). Two ABC transporters of the G subfamily, VIT\_18s0072g01220 and VIT\_01s0011g02730, could be involved in ABA import/export, the former being induced also in ABAtreated grapevine cell cultures (Nicolas et al., 2014). Three PP2C

genes out of the nine previously characterized (Boneh et al., 2011), were up-regulated in the berry by ABA and during ripening; one of these, ABI1, is induced also in the study by (Nicolas et al., 2014). Many kinases of the calcium- or calcineurin-dependent families, beside the better characterized SnRK2, are induced by ABA and likely to be related to this signaling cascade. Known transcription factors directly activated by SnRK2 belong to the bZIP family and are named ABA responsive element binding (ABRE/ABF). In grapevine, VvABF2 (VIT\_18s0001g10450, VvbZIP45), initially named GRIP55 due to its induction during ripening (Davies and Robinson, 2000), has been exhaustively characterized, showing its induction by ABA and its prevalent transcription in berry skin and seeds (Nicolas et al., 2014). As we also found VvABF2 induced by ABA in berry skins in our study, we tried to identify its possible targets. By an in silico promoter analysis of the genes modulated at 20 and 44 h, a motif very similar to the ABA responsive elements (ABREs) present in the Arabidopsis database was found as significantly enriched (**Figure 6**). We focused our attention on four genes that were strongly up-regulated by ABA at both time-points and in cell cultures and enriched in ABREs: two TFs, VvMYB143 and VvNAC17, and two uncharacterized genes possibly involved in proteasome-dependent protein degradation, Armadillo-like and Xerico-like. These genes are modulated during berry ripening, even if MYB143 and Xerico-like need further investigation as they show a tissue and/or cultivar specific behavior. Interestingly enough, all these genes are reported in literature to be related to osmotic stress (Denekamp and Smeekens, 2003), drought stress (Ko et al., 2006) and/or ABA accumulation (Kong et al., 2015). In particular, the transcriptional modulation of Armadillo-like and Xerico-like genes and their involvement in ABA response via protein stability indicates a third level of ABA response regulation, in addition to the transcriptional and miRNAmediated post-transcriptional ones. This translational regulation level could affect either the perception/activation mechanism of ABA signaling, as suggested by Kong et al. (2015) or proteins down-stream in the cascade, as suggested by Liu et al. (2011) and Seo et al. (2012) in Arabidopsis.

The trans-activation assay performed in tobacco leaves showed that VvNAC17 and Armadillo-like were strongly activated by VvABF2, consistently with the presence of four ABREs in their promoters (**Figure 7**). Instead, VvMYB143 was activated by ABA irrespectively of the presence of VvABF2, suggesting that endogenous tobacco transcription factors can mediate its ABA dependent expression. Finally, Xerico-like neither showed significant activation by ABA nor by VvABF2. This is in apparent contrast with the observation that Xericolike expression is activated in ABA-treated grapevine cell cultures constitutively over-expressing VvABF2 (Nicolas et al., 2014). One possible explanation is that other possibly grapespecific factors are required to prime Xerico-like expression in addition to VvABF2 (e.g., additional transcription factors or VvABF2-interacting proteins). Interestingly, an in silico analysis identified Xerico-like among the co-expressed VvMYB143 genes and enriched in MYB-core type I binding motif, suggesting that VvMYB143 might be the regulator of Xerico-like (Wong et al., 2016).

# CONCLUSION

We proved the importance of ABA signaling to trigger the onset of ripening in the skin of green hard berries, occurring via an extensive gene modulation. Many molecular components of the ABA network, encompassing metabolism, perception and signaling, have been identified and many have been proposed as candidates to be experimentally validated. Four genes have been experimentally characterized showing different behaviors in response to ABA. As these genes are related to ABA/drought tolerance in other species, it will be of interest to functionally characterize them not only at ripening onset, but also under abiotic stress conditions.

# AUTHOR CONTRIBUTIONS

SP designed the study, performed time-course study in 2011 and ABA treatment in 2013 together with DB, supervised all the analyses and drafted the manuscript, GB performed the RNAseq experiment and worked on data analysis, KE, PS, and MM performed the bioinformatic analyses (RNAseq raw data analysis, gene co-expression and promoter analyses), GC, LS, and MG performed promoter activation assays in tobacco leaves, GG did ABA and lipid peroxidation analyses, CT revised the manuscript, CM contributed to the design of the study and interpretation of the results, supervised all the study and revised the manuscript. All authors revised and approved the final manuscript.

# FUNDING

The research was supported by the Autonomous Province of Trento (PAT-ADP 2013-2016). It also benefited from the networking activities coordinated under the EU-funded COST ACTION FA1106 "An integrated systems approach to determine the developmental mechanisms controlling fleshy fruit quality in tomato and grapevine." LS was supported by a fellowship from Fondazione Umberto Veronesi per il Progresso delle Scienze, Milan, Italy.

# ACKNOWLEDGMENT

We would like to thank Mickael Malnoy for providing the vectors for promoter activation assays in tobacco and Lisa Giacomelli for technical support.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2017.01093/ full#supplementary-material

FIGURE S1 | Description of the ABA treatment.

FIGURE S2 | Visualization of NCED gene family during berry ripening in three Vitis vinifera cultivars.

# REFERENCES

fpls-08-01093 June 21, 2017 Time: 11:34 # 14



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Pilati, Bagagli, Sonego, Moretto, Brazzale, Castorina, Simoni, Tonelli, Guella, Engelen, Galbiati and Moser. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Insights into the Role of the Berry-Specific Ethylene Responsive Factor VviERF045

Carmen Leida<sup>1</sup> \*, Antonio Dal Rì<sup>1</sup> , Lorenza Dalla Costa<sup>1</sup> , Maria D. Gómez<sup>2</sup> , Valerio Pompili<sup>1</sup> , Paolo Sonego<sup>3</sup> , Kristof Engelen<sup>3</sup> , Domenico Masuero<sup>4</sup> , Gabino Ríos<sup>5</sup> and Claudio Moser<sup>1</sup>

<sup>1</sup> Genomics and Biology of Fruit Crops Department, Research and Innovation Center, Fondazione Edmund Mach, San Michele all'Adige, Italy, <sup>2</sup> Instituto de Biología Molecular y Celular de Plantas, Universidad Politécnica de Valencia-Consejo Superior de Investigaciones Científicas, Valencia, Spain, <sup>3</sup> Computational Biology Department, Research and Innovation Center, Fondazione Edmund Mach, Trento, Italy, <sup>4</sup> Food Quality and Nutrition Department, Research and Innovation Centre, Fondazione Edmund Mach, Trento, Italy, <sup>5</sup> Fruit Tree Breeding Department, Instituto Valenciano de Investigaciones Agrarias, Moncada, Spain

During grape ripening, numerous transcriptional and metabolic changes are required in order to obtain colored, sweet, and flavored berries. There is evidence that ethylene, together with other signals, plays an important role in triggering the onset of ripening. Here, we report the functional characterization of a berry-specific Ethylene Responsive Factor (ERF), VviERF045, which is induced just before véraison and peaks at ripening. Phylogenetic analysis revealed it is close to the SHINE clade of ERFs, factors involved in the regulation of wax biosynthesis and cuticle morphology. Transgenic grapevines lines overexpressing VviERF045 were obtained, in vitro propagated, phenotypically characterized, and analyzed for the content of specific classes of metabolites. The effect of VviERF045 was correlated with the level of transgene expression, with highexpressing lines showing stunted growth, discolored and smaller leaves, and a lower level of chlorophylls and carotenoids. One line with intermediate expression, L15, was characterized at the transcriptomic level and showed 573 differentially expressed genes compared to wild type plants. Microscopy and gene expression analyses point toward a major role of VviERF045 in epidermis patterning by acting on waxes and cuticle. They also indicate that VviERF045 affects phenolic secondary metabolism and induces a reaction resembling a plant immune response with modulation of receptor likekinases and pathogen related genes. These results suggest also a possible role of this transcription factor in berry ripening, likely related to changes in epidermis and cuticle of the berry, cell expansion, a decrease in photosynthetic capacity, and the activation of several defense related genes as well as from the phenylpropanoid metabolism. All these processes occur in the berry during ripening.

Keywords: ERF, RNA-seq, over-expressing transgenic lines, VOCs, wax, Vitis vinifera

# INTRODUCTION

Fruit ripening is a developmental process whereby mature seed-bearing organs undergo physiological and metabolic changes that promote seed dispersal. These changes affect the nutritional value of fruit and are thus of key relevance for human and animal diet, but it also makes the fruits more susceptible to pathogen attacks, reasons for which the process attracts considerable attention from the scientific community (Giovannoni, 2004).

#### Edited by:

José Tomás Matus, Centre for Research in Agricultural Genomics, Spain

#### Reviewed by:

Ana Margarida Fortes, University of Lisbon, Portugal Kazuya Koyama, National Research Institute of Brewing, Japan

> \*Correspondence: Carmen Leida carmen.leida@gmail.com

#### Specialty section:

This article was submitted to Crop Science and Horticulture, a section of the journal Frontiers in Plant Science

Received: 17 September 2016 Accepted: 15 November 2016 Published: 09 December 2016

#### Citation:

Leida C, Dal Rì A, Dalla Costa L, Gómez MD, Pompili V, Sonego P, Engelen K, Masuero D, Ríos G and Moser C (2016) Insights into the Role of the Berry-Specific Ethylene Responsive Factor VviERF045. Front. Plant Sci. 7:1793. doi: 10.3389/fpls.2016.01793

Grapevine is one of the most important cultivated crops in the world; the fruit is used as a source of fresh fruit, or once fermented, for production of wine and distilled beverages. The beginning of grape ripening, called véraison, coincides with a dramatic metabolic re-arrangement, affecting the accumulation of sugars, metabolism of acids, berry softening and coloring, and fruit growth. Ripening control in non-climacteric fruits, such as grapes, was originally thought to be ethylene independent, but recent evidence demonstrates a common genetic regulatory mechanism between climacteric and non-climacteric fruits (Lin et al., 2009). For example, a small amount of ethylene was measured in non-climacteric strawberries and this production was correlated to the expression of an ACC oxidase 1 gene (Trainotti et al., 2005). Other evidence includes the observation that climacteric, such as tomato, and non-climacteric species, such as grapevine, share common ripening regulators like members of the MADS-box, Zn-fingers, and bZIP transcription factor (TF) families (Fei et al., 2004).

There are hints suggesting that ethylene is also affecting grape ripening. The application on grapes of the ethylene releasing compound 2-chloroethylphosphonic acid (CEPA) 3–6 weeks before véraison causes a delay of the ripening process, while treatments 2 weeks before véraison accelerate the start of grape ripening (Coombe and Hale, 1973). Application of the inhibitor of the ethylene receptor 1-methylciclopropene (1-MCP) before véraison delays berry growth, acid degradation, sucrose production, and coloring (Chervin et al., 2004). A peak of endogenous ethylene has also been detected in grapevine berries, although at much lower concentrations than in climacteric fruits, 1 week before véraison (Chervin et al., 2004). The potential role of ethylene in the ripening of non-climacteric fruits is likely to occur via cross-talk with other hormones such as abscisic acid, auxin and brassinosteroids, all of which are known to play a part in grapevine berry ripening (Hale et al., 1970; Coombe and Hale, 1973; Davies et al., 1997; Jeong et al., 2004; Symons et al., 2006).

A key step in ethylene signal transduction is the activation of ethylene responsive factors (ERFs) that belong to the large superfamily of AP2/ERF TFs, specific to plants (Nakano et al., 2006). These factors are characterized by the presence of one or more AP2/ERF domains, consisting of 58–59 amino acids folded in one α-helix and a β-sheet, that recognizes the GCC box (5<sup>0</sup> -AGCCGCC-3<sup>0</sup> ) cis-element in the promoter of the target genes (Fujimoto et al., 2000). Based on the number of AP2/ERF domains and presence of other conserved domains, this superfamily can be divided into three families called AP2, ERF and RAV. The ERF family is characterized by one single AP2/ERF domain and it comprises the CBF/DREB and ERF sensu stricto subfamilies (Sakuma et al., 2002). ERF members have been discovered in many plant species due to the high degree of conservation of AP2/ERF domain (Nakano et al., 2006; Zhang et al., 2008; Zhuang et al., 2008), including grapevine, where 132 and 149 AP2/ERF genes have been predicted (Zhuang et al., 2009; Licausi et al., 2010). ERF and DREB factors are often involved in fruit ripening control, and plant response to stress (Nakano et al., 2006). Among ERF proteins involved in fruit ripening are factors characterized in plum, apple and tomato. Seven ERFs have been proposed to regulate plum fruit development and ripening, based on their gene expression patterns (El-Sharkawy et al., 2009). MdERF1 and MdERF2 are regulated by ethylene in apple as suggested by exogenous MCP treatment and varietal studies (Wang et al., 2007). Overexpression and silencing of the tomato LeERF1 gene has revealed an important role in plant development, fruit ripening and softening (Li et al., 2007), and tolerance to drought (Lu et al., 2010). Members of the SHINE clade of ERF factors (Aharoni et al., 2004) are involved in the regulation of lipid biosynthesis and the accumulation of cuticular waxes in tomato, leading to drought tolerance and recovery from water deficit (Shi et al., 2013).

In this study we focus on VviERF045, a factor phylogenetically related to the SHINE clade of ERF genes from Arabidopsis (Aharoni et al., 2004) which is specifically induced after véraison in grapevine fruit, and thought to play a role in the ripening process (Pilati et al., 2007; Fasoli et al., 2012; Lijavetzky et al., 2012; Palumbo et al., 2014). Five transgenic lines overexpressing VviERF045 were obtained and used for functional characterization through phenotypic observation and metabolic and transcriptomic analyses.

#### MATERIALS AND METHODS

#### Plant Material, 1-MCP and Etephon Treatments

Fruits were harvested from 'Pinot Noir' grapevine 10-years old plants cultivated in open field at Fondazione Edmund Mach (FEM) in San Michele all'Adige (Italy), following standard cultural practices and disease management. During 2006, three independent clusters were collected weekly starting from 4 to 10 weeks after anthesis (WAA) and at 14 WAA. Seeds, buds, tendrils, adult and young leaves, roots and flowers were also collected. The fruit (10 WAA) was dissected into pulp, skin and seed.

1-MCP and etephon treatments (both at 5 ppm) were performed at 7, 8, 9 WAA for 24 h, in a polyethylene bag wrapped around the cluster. Véraison (berry color change) occurred at 7 WAA. Mock treatments were applied to the control samples. Plant material was immediately frozen at −80◦C and stored until analysis.

#### Phylogenetic Analysis

The protein sequences of VviERF045, 7 ERFs from Prunus salicina (El-Sharkawy et al., 2009) and the three best blastx matches to VviERF045 from Solanum lycopersicum and Arabidopsis were aligned with MUSCLE (Edgar, 2004). In order to assess the real orthologs, a reciprocal best hit approach was used. Genebank accession numbers are listed in **Figure 1F**. A distance matrix was constructed according to the PAM model and clustered with the Neighbor-Joining method, using the EMBL-EBI bioinformatic tools framework (Li et al., 2015). The reliability of the phylogenetic grouping was assessed by bootstrapping (1000 replicates).

FIGURE 1 | VviERF045 expression pattern and protein sequence comparative analysis. Left panel: real time RT-qPCR analysis of VviERF045 expression profile in different tissues (A): B, Bud; S, Shoot; ML, Mature Leaf; YL, Young Leaf; F, Flower; R, Root; T, Tendril; 5WAA, berry at 5 weeks after anthesis (WAA); 10 WAA, berry at 10 WAA. ML and YL were used as reference samples. (B) VviERF045 expression in different fruit parts: pulp, skin and seed at 10 WAA. Pulp is taken as calibrator (C) VviERF045 expression at different developmental stages, berry developmental stages are indicated as WAA, (v) indicates véraison. (D) VviERF045 expression after 1-MCP and ethephon treatment: points represent 1-MCP treatment, points and lines represent ethephon treatment and continuous line represents control. Error bars represent SD and are based on three biological and two technical replicates. Data were normalized using ubiquitin and tubulin as reference genes. Different letters in the figure mean significant difference (p < 0.05) according to Tuckey's post hoc test. Right panel: (E) Phylogenetic tree of the ERF amino acid sequences from Prunus salicina [PsERF1a (FJ026009), PsERF1b (FJ026008), PsERF12 (FJ026003), PsERF3a (FJ026005), PsERF3b (FJ026004), PsERF2a (FJ026007), PsERF2b (FJ026006)], Arabidopsis thaliana [AT5G25190 (NP\_197901.1), AT1G15360-SHN1 (NP\_172988.1), AT5G25390-SHN3 (NP\_197921.1), AT5G11190-SHN2 (NP\_196680.1)], Solanum lycopersicum [SlSHN1 (XP\_004235965.1), SlSHN3 (XP\_004240977.1), SlERF1 (NP\_001234848.1)] and Vitis vinifera ERF from clade V [VviERF045, VviERF042, VviERF051, VviERF044, VviERF043, VviERF048, VviERF049, VviERF041, VviERF050, VviERF047, VviERF046]. The aa sequences were selected based on these criteria: (i) Grapevine ERFs belonging to the same clade of VviERF045, (ii) Prunus salicina ERF sequences related to fruit ripening, (iii) Best blastx matches to VviERF045 from A. thaliana and S. lycopersicum. Numbers on the branch represent the percentage for bootstrap value n = 1000. (F) Alignment of the amino acid sequences of clade V of the phylogenetic tree. 'AP2,''mm' and 'cm' conserved domains are represented as blue rectangles.

#### Production of Transgenic Lines

The complete coding region of VviERF045 (GenBank accession number KX179904) was amplified with Pfu Ultra Hotstart DNA polymerase (Stratagene, San Diego, CA, USA), starting from cDNA from mature berry. The purified PCR product was cloned into pENTR-D TOPO cloning vector (Invitrogen, Carlsbad, CA, USA), sequenced and transferred to pK7WG2 binary vector (Karimi et al., 2002) downstream of the 35SCaMV

promoter, by using the Gateway technology (Invitrogen). The Agrobacterium strain EHA105 containing the VviERF045 binary vector and the pCH32 virulence helper plasmid were used for grape transformation. Gene transfer experiments were performed as described in Dalla Costa et al. (2016) on embyogenic calli of Vitis vinifera cv. 'Brachetto'. Transgenic and wild type plants were grown and propagated in vitro.

# Expression by Quantitative Real-Time PCR (RT-qPCR) Analysis

Each sample was composed of a pool of leaves (first five leaves from the apical meristem) from five different in vitro plants. Total RNA was extracted from 100 mg of leaf powder by using SpectrumTM Plant Total RNA kit (Sigma–Aldrich, St Louis, MO, USA), adding 1% PVP40 in the extraction buffer. Total RNA was quantified with Nanodrop8000 Spectrophotometer (Thermo Scientific, Waltham, MA, USA). RNA integrity was checked by agarose gel electrophoresis. Total RNA (1 µg) was treated with Ambion <sup>R</sup> DNA-free DNase Treatment in order to remove contaminating DNA (Life technologies, Carlsbad CA, USA), and subsequently reverse transcribed with SuperScript <sup>R</sup> VILOTM cDNA Synthesis Kit (Invitrogen) in a final volume of 20 µL, according to manufacturer's instructions. One microliter of a 10X diluted first strand cDNA was used for each amplification reaction in a final volume of 20 µL. RT-qPCR was performed in a ViiATM 7 Real-Time PCR System (Applied Biosystems, Foster City, CA, USA), using the KAPA SYBR Fast qPCR Master Mix (Kapa biosystems, Wilmington, MA, USA). Reaction composition and conditions followed manufacturer's instructions. The cycling protocol consisted of 10 min at 95◦C, then 40 cycles of 30 s at 95◦C and 60 s at 60◦C. Specificity of the PCR was assessed by the presence of a single peak in the dissociation curve after the amplification and through size estimation of the amplified product. The relative standard curve method was used to quantify relative expression genes in case of efficiency less than 90%. Otherwise the 1Ct method was used as described in Applied Biosystems user's manual. Results were calculated as the average of three independent biological replicates for each line, repeated twice, using tubulin and ubiquitin as reference genes (Supplementary Table S2). For the amplification of VviERF045, we used two different primer pairs, namely "VviERF045" and "VviERF045endog" (Supplementary Table S2). Both primers of the first pair anneal on the coding sequence and they measure the total expression of the endogenous and exogenous (transgene) VviERF045 copies. Unlike, the reverse primer of the second pair anneals on the 3 <sup>0</sup>UTR region of the transcript which is present only in the endogenous gene copy but not in the transgenic one. The "VviERF045endog" primers were thus used to amplify specifically endogenous gene expression both in the transgenic lines and in the different grapevine tissues (**Figure 1**).

### RNA-Seq Analysis and Identification of Differentially Expressed Genes (DEGs)

Total RNA was extracted from three independent pools of leaves (1 g) from in vitro grown plants as described above. RNA-Seq for transgenic line L15 and control were performed using an Illumina HiSeq2000 sequencing service (Illumina, Inc., San Diego, CA, USA). Samples were sequenced twice in separated lanes. Pairedend (2 × 100 bp) and raw reads were pre-processed for quality using fastqc 0.11.2<sup>1</sup> and adapter trimming with qtrim v0.94<sup>2</sup> . The resulting pre-processed reads were aligned to the reference transcriptome of Vitis vinifera (V1 grapevine annotation)<sup>3</sup> using the bowtie2 aligner v2.2.3 (Langmead and Salzberg, 2012) and deposited in Gene Expression Omnibus<sup>4</sup> series entry GSE77240. The summarized read count data was used to identify DEGs among various treatments by using the voom method (Law et al., 2014), which estimates the mean-variance relationship of the logcounts, generating a precision weight for each observation that is fed into the limma empirical Bayes analysis pipeline (Smyth et al., 2008). DEGs were identified between OE\_ERF and WT using a P-value of 0.05 and a log2-fold change greater than 1.5 and lower than −1.5 (**Figure 4B**; Supplementary Table S4; **Supplementary Figure S1**).

# Functional Analysis

Differentially expressed genes were analyzed by BLAST2GOv 3.0.9 (Conesa et al., 2005) and TopGO (Alexa et al., 2006). The analysis with TopGO was done by comparing three statistical methods (Fisher's, weight, Kolmogorov–Smirnov), and selecting the best 10 GO terms.

# Phenolic Metabolites Determination

Leaves from transgenic lines and control were sampled as described above (three biological replicates). Approximately 100 mg of powder from each sample was extracted in sealed glass vials using of a mixture of water/methanol/chlorophorm (20:40:40). Phenolics were extracted following Vrhovsek et al. (2012) method and UPLC chromatography was performed by injecting 2 µL of each sample. The same extract was used to measure anthocyanins by UPLC (Arapitsas et al., 2012).

# Lipid Profile Analysis

The lipid profile in leaves was determined following Della Corte et al. (2015) protocol, starting from 100 mg of powdered leaves and injecting 5 µL of lipid extracted solution into the LC-MS/MS system.

# Chlorophylls and Carotenoids Quantification

Leaves from in vitro cultivated transgenic lines and control were collected and powdered with liquid nitrogen (three biological replicates). Total carotenoids and chlorophylls were extracted from 100 mg samples using acetone 80% and read with a spectrophotometer at the wavelengths 470, 646.8, and 663.2 nm. Chlorophylls and carotenoids were determined following Lichtenthaler (1987) method.

<sup>1</sup>http://www.bioinformatics.bbsrc.ac.uk/projects/fastqc

<sup>2</sup>https://ccb.jhu.edu/software/fqtrim/index.shtml

<sup>3</sup>http://genomes.cribi.unipd.it/grape/

<sup>4</sup>http://www.ncbi.nlm.nih.gov/geo/

### Leaf Area Measurement

fpls-07-01793 December 9, 2016 Time: 15:1 # 5

Leaf area was measured with Iris Electronic Eye Analyzer VA300 (iBiosys Solutions Ltd, UK) and calculated with AlphaSoft 12.44 (Alpha MOS, France) using a fixed area as reference. In case of folded leaves the doubled part was cut and pasted aside with GIMP 2.6.12 image manipulation program (GNU GPL) in order to measure the whole leaf area. A non-parametric test was preferred for statistical analysis given the non-normality of data. We used the ggplot2 R package to graphically present these data in form of boxplots, using the geom\_boxplot function (Wickham, 2009).

### PCA Analysis and Heatmaps

Principal component analysis (PCA) of the metabolites (**Supplementary Figure S4**) was obtained with R after scaling and centering the data. Heatmap representation of secondary metabolite content in the transgenic lines (**Figures 3**, **5**, and **6**) was calculated for each metabolite. Values were scaled by subtracting the mean value of WT and dividing by the standard deviation. Significance was assessed by ANOVA test followed by Tuckey's test using R<sup>5</sup> .

#### Optical Microscopy and CryoSEM

For lipid staining, a Sudan IV (Sigma–Aldrich) stock solution (0.1% w/v in isopropyl alcohol) was diluted 1:1 with glycerol, mixed well, and allowed to sit at room temperature for 30 min and syringe filtered to remove precipitates. The fourth or fifth leaves, starting from apex, were taken and cut in little squares of 4–5 mm. Leaves were included in agarose 5%, sectioned in 30 µm slices with a vibratome, stained for 30 min, mounted in distilled water with a cover slip and viewed immediately. Images were captured using a microscope Eclipse E600 (Nikon, Melville, NY, USA).

For Cryo-SEM, leaves were harvested, mounted on SEM stubs attached to a CT-1000C Cryo-transfer system (Oxford Instruments, Oxford, UK) and frozen in liquid N2. The frozen leaves were transferred to the cryo-stage of a JEOL JSM-5410 scanning electron microscope (SEM). The samples were then fractured, sublimated by controlled warming to −90◦C, and sputter coated with a thin film of gold. Finally, leaves were viewed at an accelerating voltage of 15 keV and captured at 1000x and 2000x magnification.

# RESULTS AND DISCUSSION

#### VviERF045 Is a Berry-Specific Transcription Factor Induced at Ripening and Closely Related to the ERFs from the SHINE-Clade

Expression analysis via RT-qPCR showed that VviERF045 is highly expressed in fruit at 10 WAA, while its expression is much lower in other tissues such as root, stem, leaf, bud, flower and green berry (**Figure 1A**). During berry development VviERF045

<sup>5</sup>www.R-project.org/

expression raises starting from 7 WAA and peaks 2–3 weeks later, about at the end of the véraison period, (**Figure 1B**) at which time, VviERF045 expression is more pronounced in skin and pulp rather than seed (**Figure 1C**). These observations suggest that VviERF045 might play a regulatory role in the berry ripening process. Although several members of the ethylene response factor family are ethylene inducible (Pirrello et al., 2012), berries treated around véraison (7, 8, and 9 WAA) with 1-MCP or etephon showed no significant differences in the expression of VviERF045 compared with the control (**Figure 1D**). However, our study could not exclude that treatments done in a prevéraison stage could have led to the same results.

Former phylogenetic analysis assigned VviERF045 to clade V of the ten clades identified for the 122 grapevine members of the ERF superfamily (Licausi et al., 2010). In this study we have made a more focused analysis comparing VviERF045 and other VviERFs from cladeV to 15 highly similar and previously characterized ERF protein sequences from Prunus salicina, Arabidopsis thaliana, and Solanum lycopersicum (**Figure 1E**). Our phylogenetic tree highlights that the seven ERFs related to fruit development and ripening in Prunus salicina (El-Sharkawy et al., 2009) cluster apart from the other ERF sequences. On the other side, VviERF045 and its homologs from tomato and Arabidopsis fall into a single clade together with VviERF046 and VviERF047. Interestingly, one subgroup into this clade contains exclusively wax biosynthesis genes whose overexpression results in a glossy leaf phenotype and increased drought tolerance: AT1G15360 (SHN1), AT5G11190 (SHN2), and AT5G25390 (SHN3) from Arabidopsis, and SlSHN1 and SlSHN3 from tomato (Tournier et al., 2003; Aharoni et al., 2004; Shi et al., 2013). They share a high degree of similarity since in addition to the conserved AP2 domain they display two other conserved motifs located in the middle and the C-terminus of the protein sequence ('mm' and 'cm'; Aharoni et al., 2004). The remaining three sequences, which comprise VviERF045, SlERF1, and the Arabidopsis AT5G25190, form a distinct subgroup within clade V that distinguishes itself most notably by a deletion of six and one amino acid(s) in the 'mm' and 'cm' domains, respectively (**Figure 1F**).

AT5G25190 was reported to be induced by 1 aminocyclopropane-1-carboxylic acid (ACC) and salt (Zhang et al., 2011) as well as by drought (Huang et al., 2008), but it was shown that its overexpression does not lead to a typical leaf shine phenotype (Aharoni et al., 2004). SlERF1 overexpression leads to several phenotypic effects including ethylene triple response on etiolated seedling, leaf development, enhanced fruit ripening and softening (Li et al., 2007) and improved tolerance to drought stress (Lu et al., 2010).

# Phenotypic Characterization of VviERF045 Transgenic Lines

Fourteen transgenic lines overexpressing VviERF045 from the Vitis vinifera cv. 'Brachetto' were generated (see Materials and Methods). In five of them, the expression of VviERF045 (due to transcription of exogenous and endogenous gene copies) was much higher than in the wild type and they were used for further functional characterization. In lines L6 and L7 the expression of

FIGURE 2 | Characterization of VviERF045 overexpressing lines. (A) Relative expression level of the endogenous + exogenous gene (blue bars), and endogenous gene (orange bars) in the transgenic lines. Error bars are based on data from three biological and two technical replicates. Data were normalized using ubiquitin and actin as reference genes. Relative expressions with the same letter are not significantly different (p < 0.05) according to Tuckey's post hoc test. The ANOVA test refers to the red bars since for blue bars no significant difference was observed. (B) In vitro phenotype of transgenic and WT plants of the same age.

VviERF045 was about 100-fold increased; in the other three lines it was increased around 25–30 times (**Figure 2A**).

The overall phenotype of the transgenic lines seemed directly related to the level of expression of the transgene, affecting not only leaf morphology and color, but also root biomass and architecture (**Figure 2B**) and this was particularly evident in case of L6. Several leaf features correlated strongly with the expression level of VviERF045, such as the leaf blade insertion angle on the petiole (**Figure 2B**), the leaf area and the leaf margins (**Figures 3A,B**). In general, L6 leaves displayed an acute insertion angle, a globular and chlorotic surface (almost yellow) and leaf margins curved toward the abaxial surface of the leaf, resembling somehow an 'epinastic' phenotype (Barry et al., 2001). Unlike L6, WT plants carried leaves with an insertion angle ranging between 180◦ and 140◦ , with a plane and bright green surface and an evident dentate margin. L15-19-22 showed an intermediate phenotype with insertion angles of 90–120◦ , a WTlike dentate leaf margin and a light green color, while L7 leaves were more similar to L6 in form and color. The analysis of pigment contents confirmed these phenotypic observations, with L6 showing lower values of chlorophyll a and b and carotenoids (**Figure 3C**). The transgenic lines showed also a general reduction in leaf number and leaf area. The total leaf area in L6 was about 10 times lower than in WT. L7, 15, and 19 showed comparable

leaf areas, about half of the WT, whereas L22 was more similar to WT (**Figures 3A,B**). The VviERF045 overexpressing lines had a smaller root system with short and thick roots (**Figure 2B**). This could be due to a defective auxin gradient, which plays a key role in root development (Overvoorde et al., 2010).

# Effect of VviERF045 Overexpression on the Transcriptome

Three pools of leaves harvested from in vitro plants of the transgenic line L15 and from WT plants were used to compare the two transcriptomes by a RNA-Seq approach. L15 was selected since it showed a high level of VviERF045 expression, while growing sufficiently well in vitro and in vivo. Between 29 and 79 million of paired-end reads of 100 nucleotides were obtained for each replicate, and on average 79% of them were properly aligned in both senses (Supplementary Table S1).

A multidimensional scaling approach to the analysis of the expression data highlighted that the three replicates of L15 were well separated from those of WT (**Figure 4A**). Using a volcano plot, 573 DEGs between L15 and WT were identified in the region with absolute values of log2 fold change greater than 1.5 and a p-value <0.05 (**Figure 4B**; Supplementary Table S4).

To identify over-represented gene categories within the DEGs, we ran an enrichment analysis with both Blast2GO (Conesa et al., 2005) and TopGO (Alexa et al., 2006). Blast2GO found 35 categories (p-value < 0.05) from the 'biological function' subtree enriched with respect to the reference transcriptome (**Supplementary Figure S3A**). By grouping these GO categories into broader functional categories, phenylpropanoid metabolism, signaling and amino acid metabolism were over-represented. The analysis of GO terms using Blast2GO showed a higher percentage of transferase, protein phosphorylation, protein kinase and receptor activities, which suggests the participation of VviERF045 in complex regulatory pathways. In addition, the presence of genes related to secondary metabolite pathways, such as trihydroxystilbene synthase, naringerin-chalcone synthase, flavonoid biosynthesis, and flavonoid metabolic process suggests the involvement of VviERF045 in the synthesis and metabolism of phenolic compounds (**Supplementary Figure S3A**).

Nine major GO categories came out enriched by applying weight count versus classic count in Fisher's enrichment test by using TopGO (Alexa et al., 2006): protein phosphorylation (Supplementary Table S4), wax biosynthetic process, response to endogenous stimulus, cotyledonal vascular tissue/pattern formation, drug transmembrane transport, jasmonic acid biosynthesis, response to salt stress (Supplementary Table S4), defense response to bacterium (Supplementary Table S4), and negative regulation of endopeptidase activity (**Supplementary Figure S3B**).

#### Analysis of Metabolites

We measured 56 phenolic secondary metabolites (**Supplementary Figure S4**), including chlorophyll a and b and total carotenoids (**Figure 3C**), hydroxycinnamic acids, benzoic acids, stilbenes, flavonols, flavan-3-ols and anthocyanins (**Figure 5A**) and lipids (**Figure 6A**) for the all five transgenic lines and WT (Supplementary Table S3). To have a general idea of the dispersion of the metabolic data, a PCA was computed for all the metabolites (**Supplementary Figure S4**). In this analysis the WT and the L6 plants were the most separated groups, whereas the other lines showed an intermediate position. Thus, the metabolomic study confirmed the extreme behavior of L6 line observed in the phenotypic characterization.

### VviERF045 Is Involved in Plant Growth and Development

Among the DEGs we found a significant number of genes whose Arabidopsis putative orthologs are involved in growth and development, and more specifically are associated to the development of anatomical structures, the formation of cotyledonal vascular pattern, procambial histogenesis and multidimensional cell growth (Supplementary Table S4).

Noteworthy was VIT\_04s0008g01970, coding for the putative ortholog of the ERECTA (ER) gene, which appeared strongly down-regulated in L15. ER codes for a leucine-rich repeat receptor-like Ser/Thr kinase that is a major transcriptional regulator with pleiotropic effects on development and plant physiology. It controls plant transpiration efficiency, modulating stomatal opening and CO<sup>2</sup> fixation (Masle et al., 2005), stomatal density and patterning (Lampard et al., 2008), abaxial-adaxial identity (Qi et al., 2004), petal shape and size (Abraham et al., 2013), ethylene induced hyponastic growth and leaf petiole angle (Van Zanten et al., 2010), leaf area and plant biomass during shade avoidance syndrome (SAS) (Kasulin et al., 2013), and resistance against specific pathogens such as fungi (Häffner et al., 2014). Indeed er loss of function mutants show reduced plant size, rounder and shorter leaves, shorter petioles and compact inflorescences in Arabidopsis. These features closely resemble those we observed in the lines overexpressing VviERF045, namely reduced leaf biomass, leaves with globular surface and different leaf margins and changes in leaf-petiole angle (**Figure 2B**). In L15, VIT\_18s0001g10160, coding for the putative WUSCHEL-RELATED HOMEOBOX4 (WOX4) grapevine ortholog, was induced twofold with respect to WT plants. WOX4 is expressed in the pro-cambium and plays an important role in vascular meristem organization. Recent evidence indicated that ER participates also in vascular development, acting upstream to WOX4 (Uchida and Tasaka, 2013), and our results suggest a similar interplay between these genes in grapevine.

# VviERF045 Regulates Lipid Metabolism As Well As Cuticle and Waxes Synthesis

Our results show that VviERF045 is functionally linked to lipid metabolism, specifically to the synthesis of cuticle and cuticular waxes. Optical images revealed a different pattern of the lipid distribution on the surface of L6 leaves compared to WT (**Figure 7A**). In the latter, the reddish color was evenly distributed along the cuticular layer that covers the epidermal cells, while in L6 the stain was observed in groups of intracellular droplets, similar to lipid bodies, in the epidermal layers. Scanning Electron Microscopy (Cryo-SEM) confirmed a striking difference in the structure of the epicuticular waxes between the two (**Figure 7B**):

FIGURE 4 | RNA-Seq analysis of L15 line. (A) Classical multidimensional scaling showing the percentage of variance explicated by PC1 and PC2. (B) Volcano plot used for selection of differentially expressed genes (DEGs) between L15 and WT lines. Spots represent genes. The red line represents the significance threshold of p-value (0.05) and the blue lines represent log2 fold change values of 1.5 and −1.5. Genes located above the red line and on the right-hand side or the left-hand side of blue lines were selected as DEGs.

and highly significant differences (p < 0.01). A legend with the color scale is reported on the top left part. B.A.D., Benzoic acid derivative. (B) Transcripts involved in their biosynthesis. 4.3.1.25 Phenylalanine ammonium lyase (PAL) (VIT\_16s0039g01170, VIT\_16s0039g01240, VIT\_16s0039g01280, VIT\_16s0039g01300, VIT\_16s0039g01360), 1.1.1.195 cinnamyl-alcohol dehydrogenase (CAD), (VIT\_03s0110g00310, VIT\_13s0064g00270), ferulate 5-hydroxylase (F5H) (VIT\_17s0000g03930), 2.3.1.74 Chalcone reductase (CHR) (VIT\_01s0011g06440), 21.1.104 Caffeoyl-CoA O-methyltransferase 1 (VIT\_07s0031g00350), 1.11.1.7 peroxidase (VIT\_12s0028g01830, VIT\_14s0068g01920, VIT\_10s0116g01780, VIT\_18s0001g06890, VIT\_16s0098g00820), 2.3.1.95 STS (stilbene synthase) (VviSTS2, VviSTS3, VviSTS5, VviSTS6, VviSTS7, VviSTS10, VviSTS13, VviSTS15, VviSTS17, VviSTS18, VviSTS19, VviSTST20, VviSTST21, VviSTST25, VviSTS25, VviSTS28, VviSTS29, VviSTS30, VviSTS31, VviSTST37, VviSTST38, VviSTST39, VviSTST42, VviSTS46, VviSTS47 (Vannozzi et al., 2012). 2.3.1.116 Quercetin 3-O-glucoside-6<sup>00</sup> -O-malonyltransferase (VIT\_12s0134g00630). 2.4.1.115 anthocyanidin 3-O-glucosyltransferase (VIT\_03s0017g02110, VIT\_16s0022g01970), 2.4.1- UDP-glucose (VIT\_18s0041g00830, VIT\_18s0041g00840, VIT\_18s0041g00930, VIT\_18s0041g01010, VIT\_16s0050g01680). Green color means down-regulated gene, red color up-regulation.

the WT cuticle appeared heavily decorated with wax aggregates, while the L6 leaf surface was smooth. The wax decoration in the other transgenic lines was reduced compared to WT.

The lipid composition of leaf samples was analyzed to further understand the observed alterations at the cuticular level. Six lipid compounds belonging to the classes of fatty acids, sterols, glycerolipids, glycerophospholipids and sphingolipids appeared significantly modulated (**Figure 6A**). The steroid-like lanosterol was at a higher concentration in WT and diminished in transgenic lines proportionally to VviERF045 expression (**Figure 6A**). Lanosterol is known to be a component of the tomato cuticular waxes. In the lcer6 mutant, an increase of lanosterol together with other sterols and triterpenic cycles, was accompanied by a proportional decrease in long chain alkanes (Leide et al., 2007). This result is in line with the down-regulation in L15 of a squalene monoxigenase (VIT\_00s0441g00020), involved in the oxidation of squalene to squalene epoxide, the precursor for lanosterol biosynthesis (**Figure 6B**).

Cuticular waxes are made of very-long-chain (VLC) fatty acids (FAs), synthesized starting from plastidial C16-C18 FAs, elongated into VLCFAs in the endoplasmic reticulum membrane, and subsequently modified into primary alcohols and wax esthers (**Figure 6A**, Yeats and Rose, 2013). Several genes involved in FA elongation and wax biosynthesis (VIT\_00s0179g00380, VIT\_01s0011g03490, VIT\_16s0050g00830) were expressed at lower level in the L15 transgenic line (Supplementary Table S4), implying VviERF045 overexpression reduces long FA and wax content. Down-regulation was observed for the putative orthologs of the Arabidopsis CER1 and CER2 genes, the grapevine genes VIT\_15s0021g00050 and VIT\_05s0029g00480. The cer1 mutant of Arabidopsis is blocked in the conversion of stem wax C30 aldehydes to C29 alkanes, leading to a lack of secondary alcohols and ketones. This biochemical impairment results in a reduced wax load on the leaf surface of the cer1 mutants (Bourdenx et al., 2011), resembling the cuticular phenotype we observed in the transgenic lines in this study. The closest sequence to VIT\_05s0029g00480 is CER26, the homologue of CER2, which encodes for an acyl-transferase enzyme involved in the elongation process of C28 FAs (Pascal et al., 2013). VIT\_08s0007g00390, similar to the Arabidopsis PROTEOLYSIS 6 (PRT6), was less expressed in the L15 line. prt6 mutants are impaired in lipid degradation and retain oil bodies in the cells, similar to the ones observed in the external layers of the L6 leaves in our analysis (**Supplementary Figure S6**) (Holman et al., 2009). The cytochrome P450 genes VIT\_02s0025g03320 and VIT\_07s0031g01680 were found down-regulated in VviERF045 overexpressing lines. VIT\_02s0025g03320 belongs to the CYP86A subfamily, known to have ω-hydroxylase activity on midchain FAs (Yeats and Rose, 2013). The similar Arabidopsis gene LACERATA was reported to be involved in cutin biosynthesis (Wellesen et al., 2001). The most similar sequence to VIT\_07s0031g01680 in Arabidopsis belongs to the CYP96A subfamily, which includes MIDCHAIN ALKANE HYDROXYLASE 1 (MAH1, CYP96A15), an enzyme of the decarbonylation pathway catalyzing the synthesis of cuticular wax secondary alcohols and ketones from VLC alkanes (Greer et al., 2007). L15 plants also displayed lower expression of three lipases belonging to the GDSL family (VIT\_13s0106g00350, VIT\_18s0041g02160, VIT\_18s0086g00220). Members of this large gene family appear to have a broad range of activities in the regulation of plant development, morphogenesis, synthesis of secondary metabolites, and defense response (Chepyshko et al., 2012). Recently, specific members within the family have been shown to play a role in cutin synthesis by catalyzing the formation of cutin ester oligomers (Yeats et al., 2014).

The only two genes of the wax biosynthetic pathway which resulted up-regulated in L15 were VIT\_02s0012g02500

and VIT\_15s0046g00490, encoding for a putative stearoyl-acyl carrier protein-desaturase (S-ACP-DES) and a putative wax synthase/diacylglycerol acyltransferase 1 (WSD1), respectively (**Figure 6A**). In plants, S-ACP-DESs tune the ratio of saturated to monounsaturated FAs (Kachroo et al., 2007). In Arabidopsis, WSD1 is responsible for the esterification of VLC primary alcohols to long chain wax esthers using C16 FAs as substrates (Li et al., 2008).

CR, Cuticular Ridge. Lower part: images of the adaxial leaf surface (1000X). WA, Wax Aggregates.

As our results pointed toward a role of VviERF045 in regulating cuticle biosynthesis, we compared the effects of its overexpression with those described for the major known regulators, namely the SHINE gene family and some specific MYB TFs. WAX INDUCER1/SHINE1 (WAX1/SHN1) was the first TF identified (Aharoni et al., 2004). It is an ERF sequence of clade V, whose overexpression gives rise to dwarf plants with curved and glossy leaves, lower stomata density, thicker cutin and higher

wax density. The cauline leaves of the gain of function mutant shine display cuticular ridges similar to those here reported on the L7 leaves (Aharoni et al., 2004; Kannangara et al., 2007). When SHN1 and the other two closely related members SHN2 and SHN3 were silenced, Arabidopsis plants exhibited, among other phenotypic traits, a decrease in cutin load (Shi et al., 2011), and changes in cuticle structure and lipid composition have also been demonstrated in tomato (Shi et al., 2013). SHINE regulators exert their function by acting on several cuticle- and epidermisassociated genes, including CYTOCHROME P450s, GSDL-type LIPASES, ACYLTRANSFERASES, LONG CHAIN ACYL\_CoA SYNTHASES, CER1 and CER2 (Kannangara et al., 2007; Shi et al., 2011, 2013). Genes with similar functions were down-regulated in L15 transgenic plants, as reported above.

AP2-containing TFs can be either activators or repressors depending on the effect on transcription of specific target genes. Transcriptional repressors are further classified as active or passive repressors: active repressors contain a repression domain (RD), which allows these proteins to actively prevent transcription of a target gene; passive repressors do not have an RD and suppress transcription by competing with transcriptional activators for binding to the target sequence (Licausi et al., 2013). VviERF045 can not be classified as an active repressor because its sequence does not display a RD. The cuticular phenotype of the lines, as revealed by microscopical investigation, as well as the negative regulation of the cuticle- and waxrelated genes in L15, are similar to those reported for SHINE silenced lines of Arabidopsis and tomato, indicating VviERF045 as a potential passive repressor. In particular VviERF045 might negatively regulate VviERF042 encoded by VIT\_09s0002g06750 and VviERF044 encoded by VIT\_04s0008g05440, which are down-regulated in L15 (Supplementary Table S4). Interestingly, VviERF042 and VviERF044 are the putative horthologues of the Arabidopsis SHINE1and SHINE3 genes, whose silencing leads to a decrease in cutin load and to changes in cell wall structure (Shi et al., 2011) similar to the ones observed in the transgenic lines of this study.

The FA chain α-linolenic acid is also the precursor of the phytohormone Methyl Jasmonate (MeJA) via the action of a lipoxygenase and a jasmonate O-methyltransferase. In L15 we found up-regulated the genes encoding for these two enzymes (VIT\_06s0004g01470 and VIT\_14s0006g02170), as well as for a MeJA esterase (VIT\_00s0253g00090), catalyzing the inverse reaction from MeJA to JA (**Supplementary Figure S6**). MeJA and JA are considered to be defense-related hormones and they do not seem to play a major role during berry ripening. It is not clear if the observed induction of the MeJA biosynthetic genes in L15 is related to the stress induced by the expression of the transgene or to direct regulatory effect of VviERF045.

### VviERF045 Modulates Genes Involved in Secondary Metabolic Processes

Phenolics are a large and complex group of secondary metabolites with chemical properties that contribute to pigmentation and defense against several biotic and abiotic stresses in grapes (Ali et al., 2010). Their biosynthesis starts from the amino acid phenylalanine which is converted into a vast array of molecules belonging to the major classes of the phenylpropanoids (hydroxycinnamic acids, stilbenes and lignins) and of the flavonoids (flavonols, proanthocyanidins and anthocyanins) (**Figure 5B**).

Several DEGs belonging to the phenylpropanoid and flavonoid biosynthetic pathway (Supplementary Table S4) encode for enzymes often positioned at the branching point of the pathway. In agreement with the overall increases in phenolic compound concentration in the transgenic lines, the majority of related DEGs also were up-regulated (Supplementary Table S4, **Figure 5B**). This includes the induction of five PHENYLALANINE AMMONIA-LYASE (PAL) encoding genes (VIT\_16s0039g01170, VIT\_16s0039g01240, VIT\_16s0039g01280, VIT\_16s0039g01300 and VIT\_16s0039g01360), which catalyze the conversion of L-phenylalanine to trans-cinnamic acid and ammonia. Among the DEGs there are also genes known to affect lignin amount and composition, suggesting that this metabolic class was likely induced as indicated by the high level of vanillin in L6 (Vanholme et al., 2010). In particular, different genes coding for cinnamyl-alcohol dehydrogenases (CADs), ferulate 5 hydroxylase (F5H), caffeoyl-CoA o-methyltransferase (COMT1) and several peroxidases were induced in L15 (**Figure 5B**). It is interesting to note that SlSHN3 silenced tomato lines showed a thicker cell wall of the epidermal cells, and that Ambavaram et al. (2011) reported that AtSHN2 controls secondary cell wall biosynthesis (lignin and cellulose) acting on CAD genes (Ambavaram et al., 2011), observations that support VviERF045 acting as a SHINE factor.

Twenty-three stilbene synthase (STS) genes appeared positively associated to VviERF045 over-expression. STSs form a rather expanded gene family in grapevine, including at least 33 members (Vannozzi et al., 2012), and they produce the basic stilbene structure, trans-resveratrol, from one p-coumaroyl-CoA and three malonyl-CoA molecules. Trans-resveratrol can then be modified by hydroxylation, methylation, glycosylation, or condensation of more units to form the ample class of stilbenoids, which represent the major antimicrobial phenolic compounds in grapevine (Jeandet et al., 2002; Malacarne et al., 2011). These compounds are also produced upon abiotic stresses such as UV-light, salinity stress (Ismail et al., 2012), and during leaf senescence and fruit ripening (Gatto et al., 2008).

The only highly accumulated stilbene common to all transgenic lines was the glucoside derivative of t-resveratrol, trans-piceide, but in L15 and L19 also the monomers cis-piceide, isorhapontin, astringin and the dimers pallidol and ampelopsin D exhibited higher levels than in WT plants. Since polymeric forms of resveratrol are usually produced during fungal attacks (Malacarne et al., 2011), this might indicates that VviSTS upregulation in L15 was mainly driven by a more general stress (Cuendet et al., 2000). In grapes, flavan 3-ols are mainly present in skin and seed tissues, where they accumulate before véraison. In vegetative organs, their content constantly increases during leaf development, but their synthesis decreases in old leaves (Bogs et al., 2005). They are found as monomers, namely catechin, epicatechin and epicatechin 3-O-gallate, as well as oligomers, and polymers called proanthocyanidins (PA), also known as condensed tannins. In our transgenic lines, compounds of this class, either in monomeric or dimeric form (procyanidin B), or condensed to caffeic acid, were clearly found at higher concentration than in WT (**Figure 5A**). As flavan 3-ols appear to function in resistance against various biotic and abiotic stresses, including UV irradiation by decreasing oxidative stress (Hammerbacher et al., 2014), it is likely that the transgenic lines face a more stressful situation than WT plants, due, for example, to cuticle impairment and to reduced photosynthetic capacity.

In our experiment, a significant higher content of peonidin p-coumaryl3glu and to a lesser extent of the glycosylated forms of cyanidin, delphinidin and malvidin, were observed in most transgenic lines (**Figure 5A**). The glycosylated forms of the flavonols quercetin and isorhamnetin displayed a similar behavior. Where the main role of anthocyanins in grapes is the red berry pigmentation to attract animals for seed dispersal, the main function of flavonols is UV-protection. Both classes are antioxidant molecules induced during different stresses, which might be the main reason of their increase in the transgenic overexpression lines. In case of the anthocyanins, the expression data were congruent with the metabolic data for two anthocyanidin 3 o-glucosyltransferases up-regulated in L15 (VIT\_03s0017g02110, VIT\_16s0022g01970, Supplementary Table S4), but less coherent for five MATE genes (VIT\_11s0052g01560, VIT\_11s0052g01540, VIT\_07s0031g00750, VIT\_00s0225g00080, VIT\_11s0052g01500 Supplementary Table S4), which were down-regulated. This grapevine protein family plays a role in the H+-dependent transport of acylated anthocyanins into the vacuole (Gomez et al., 2009), and the observed down-regulation possibly indicates a problem with the vacuolar storage of these molecules.

Another important class of secondary metabolites affected in the transgenic over-expressing lines was the photosynthetic pigments, namely chlorophylls and carotenoids. As expected from the pale leaf color, the analysis of chlorophylls and carotenoids confirmed a much lower concentration in the transgenic lines, with a minimum in L6 (**Figure 3C**). During fruit ripening the photosynthetic apparatus is dismantled (Lijavetzky et al., 2012), and our results suggest that VviERF045 might play such a role in the berries. In Arabidopsis COP1-INTERACTING PROTEIN 7 (CIP7) is involved in light-dependent anthocyanin and chlorophyll accumulation (Yamamoto et al., 1998). The putative CIP7 gene of grapevine (VIT\_00s1306g00010) was down-regulated in our study, as confirmed by RT-qPCR (**Supplementary Figure S2**). This gene was reported to be down-regulated at véraison in five red Italian varieties (Palumbo et al., 2014), as well as during post harvest withering (Fasoli et al., 2012). Other L15 repressed genes related to chlorophyll metabolism are FERRITINS (VIT\_08s0058g00410, VIT\_08s0058g00430, VIT\_08s0058g00440), iron-storage proteins involved in the regulation of free iron levels in the cells, whose impairment cause rapid natural senescence with leaf yellowing accompanied by accelerated decrease of maximal photochemical efficiency and chlorophyll degradation (Murgia et al., 2007).

In the transgenic lines, we observed the up-regulation of sesquiterpene synthase genes encoding for delta-cadinene synthase, alpha-farnesene synthase and valencene synthase (Lücker et al., 2004) (**Supplementary Figure S5**). Sesquiterpenes are a class of volatile terpenoids enriched in the epicuticular wax layer of the berry fruit. They act as antimicrobial volatile compounds (Petronilho et al., 2014) and they are induced by pathogenic fungi as well as by elicitors and MeJA (Hampel et al., 2005), but they contribute to the typical flavor of aromatic grape varieties too.

### VviERF045 in Fruit Ripening

fpls-07-01793 December 9, 2016 Time: 15:1 # 13

Berry ripening is a complex physiological process under tight regulation, which begins about 8 WAA and proceeds for about 5–6 weeks. From ripening onset, the berry undergoes chlorophyll degradation, accumulation of color, sugar and aroma compounds, organic acid catabolism, and an increase in berry size and elasticity (Coombe and McCARTHY, 2000).

Among the ERF regulatory factors possibly linked to the berry ripening process, identified previously in a microarray experiment on Pinot Noir berries at three developmental stages (Pilati et al., 2007), we selected VviERF045 for further characterization, since this factor displays a fruit ripening specific expression (**Figure 1**). An important role for VviERF045, as major switch in berry ripening, was recently also proposed by Palumbo et al. (2014).

Although our study was not conducted on berries, but in leaves from in vitro plants, implying that the results cannot be transferred straightforwardly to the fruit system, we have observed the modulation of several processes in the transgenic overexpressing lines, which are also typical of grape ripening: changes in the epidermis and in the cuticle, a decrease in photosynthetic capacity, and the activation of several defense related genes.

In this study, we collected clear evidence that VviERF045 regulates wax biosynthesis and the morphology of the cuticle and probably of the cell wall in the epidermal cells by modulating a set of specific genes. The phylogenetic proximity of VviERF045 to the SHYNE clade (**Figure 1E**) of ERFs, known to function in cuticle and epidermis patterning, further corroborates this conclusion.

At ripening onset, three processes take place, all of which imply a modification of the outer structures of the epidermal cells and thus possibly the intervention of VviERF045: berry softening, berry expansion (Coombe, 1992), and a reduction in the thickness of cuticular waxes (Rogiers et al., 2004). In the overexpressing transgenic line L15, genes known to be involved in these berry processes, such as an endo-1,4-betaglucanase (VIT\_04s0008g02010) involved in cell wall disassembly (Libertini et al., 2004), three expansins (VIT\_06s0004g04860, VIT\_06s0004g07970, VIT\_12s0059g00190), a polygalacturonase PG1 (VIT\_07s0005g01550), and a pectinesterase (VIT\_11s0016g00330) related to berry expansion and skin softening (Deytieux-Belleau et al., 2008), are down-regulated compared to the WT plants. These same genes are induced in the berry, at ripening onset. This might suggest that VviERF045 down-regulates these enzymes to counterbalance an excessive cell wall disassembling. The post-véraison development of an amorphous layer of cuticular waxes and the observation that deposition of epicuticular wax ceases at véraison as reported in Shiraz berries (Rogiers et al., 2004), is in line with our microscopic analyses (**Figure 7**) and the repression of cuticle and wax biosynthetic genes in L15 (**Figure 6**). With the beginning of berry ripening, the photosynthetic apparatus is dismantled and consequently the photosynthetic capacity of the berry drops dramatically (Pandey and Farmahan, 1977). VviERF045 could contribute to this switch-off in virtue of its effect on chlorophylls and carotenoids content (**Figure 3**), and the down-regulation of genes important for chlorophyll accumulation, like CIP7 and FERRITINs.

Many pathogen-resistance genes appear modulated by VviERF045 (Supplementary Table S4), suggesting its action also increases plant defense via activation of the basic immune defense system. Among the proteins that change their levels of expression during berry ripening, there are many pathogenesis-related proteins (PRs). PRs are highly abundant at ripening and generally lowly expressed or absent in unripe berries. The presence of this class of proteins in healthy fruit suggests that they may play a role in fruit development, or that they are part of a pre-emptive defense when softening and sugar accumulation make fruit attractive targets for pathogens (Davies and Robinson, 2000).

To further corroborate the importance of the obtained results in understanding berry ripening regulation, we ran in silico analyses taking advantage of the grapevine gene expression compendium VESPUCCI (Moretto et al., 2016). We looked whether the 563 DEGs modulated in the L15 to WT plants comparison, were expressed in the berry during ripening, in order to gain insights about their role in the process. Five hundred and forty five DEGs (18 genes were not unique in the database) were analyzed in 389 condition contrasts (Supplementary Table S5) mostly derived from samples of berries at different phenological stages, between EL 27 and EL 41. Interestingly, a large fraction (70%) of the DE genes appeared either up- (153 genes) or down-regulated (231 genes) (Supplementary Table S4), indicating that these genes are indeed modulated during ripening. The two groups were also enriched in functional classes characteristic of berry ripening like starch and sucrose metabolism, auxin biosynthesis, ethylene signaling and phenylpropanoid biosynthesis in the case of the up-regulated genes, cell wall and HomeoBox TFs in the case of the down-regulated ones. An important interaction between ethylene and auxin in the control of berry ripening has been recently elucidated (Böttcher et al., 2013). Within the DEGs, we found 7 ERF encoding genes: two SHINE putative horthologues (VviERF042 and VviERF044) that were downregulated, and other five ERFs that were strongly up-regulated. In this last group with the exception of VviERF045, there were four ERF genes (VviERF093, VviERF111, VviERF118, VviERF120), from clade IX or X, previously shown to be induced in the transition from véraison to ripe berries either in skin or in flesh (Licausi et al., 2010). These evidences strongly suggest an involvement of these ERF TFs in the control of berry ripening.

# CONCLUSION

We have functionally characterized VviERF045 by overexpressing the encoding gene in in vitro grown grapevine plants and

by phenotyping them at morphological and molecular level. VviERF045 seems to regulate, in coordination with other ERF factors, including the putative horthologues of the Arabidopsis SHINE1 and 3 genes, different processes such as the structuring of the epidermis and cuticle of the berry, cell expansion, photosynthesis, phenylpropanoid metabolism and the activation of several defense related genes. If this functional role will be confirmed by follow-up studies on the fruits of the transgenic lines, we can predict that having the possibility to adjust the expression of VviERF045 by well-timed viticultural practices (e.g. water stress, hormonal treatments) or by breeding, might allow to improve grape quality and plant resilience. The expression of VviERF045 can be used as an expression marker of the plant resilience status.

# AUTHOR CONTRIBUTIONS

CL, AD, VP, and DM did the experimental work, CL, LD, MG, and GR assessed the best way to prove the gene function, CL did the phylogenetic trees, PS and KE elaborated RNA-seq data and were involved in data interpretation, CL, GR, and CM substantially contributed to the design of the work. All the authors revised it critically for important intellectual content and approved the final version of this manuscript.

### FUNDING

CL was supported by the Marie Curie FP7-PEOPLE-2011-CIG action program-[Graperipe project n. 303907]. Network activities have been supported by COST1106 action.

#### ACKNOWLEDGMENTS

Authors thank Pietro Franceschi for suggestions in the PCA statistical analysis, Valentino Poletti and Susanna Micheli for helping in in vitro micropropagation, Luca Zulini for chlorophyll and carotenoids measure method, Marisol Gascón Irún and Manuel Josep Planes Insausti for the excellent service with the optic and electronic microscope, Michele Perazzolli for preliminary work on VviERF045 expression analysis and Ivana Gribaudo for 'Brachetto' embryogenic callus.

#### REFERENCES


# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2016.01793/ full#supplementary-material

#### FIGURE S1 | Library size of each RNA-seq sample replicate.

FIGURE S2 | Comparison of RNA-seq and real time RT-qPCR analysis.

Expression profiles of VviERF045 VIT\_04s0008g06000, RPT2 VIT\_00s0665g00050, PIR VIT\_00s1306g00010, NAC VIT\_01s0026g02710, ERECTA VIT\_04s0008g01970, SBP VIT\_04s0210g00170, N-acetyltransferase VIT\_05s0020g03680, HB1 VIT\_12s0059g01190, SA VIT\_14s0006g02170, YABBY VIT\_15s0048g00550, Peroxidase VIT\_16s0098g00820, HB1 VIT\_18s0001g10160. Lines represent expression levels (log Fold Change) by RNA-seq analyses in WT and L15, reported as means and standard errors of three independent biological replicates. Histograms represent the relative expression levels (logFC) to the expression of the WT, as assessed by real time RT-qPCR and reported as means and standard errors of three biological and two technical replicates for each plant line. Different letters show significant differences among samples with p < 0.05 and Tuckey's significance test. In case of no significance no letters are reported in the figure.

FIGURE S3 | GO term enrichment in selected DEGs. (A) Blast2GO Fisher's enrichment test analysis for GOterms. Blue bars indicate Test Set (DEGs from L15 vs. WT comparison) while red line indicate the Reference Set (entire reference transcriptome). On the X-axis is reported the percentage of sequences for each GO category (B) Best 9 GOterms by comparing classic with weight method from Fisher's test elaborated with TopGO (Alexa et al., 2006).

FIGURE S4 | PCA score and loading plots of metabolites in the transgenic lines. Different color points represents different samples: black (L15), red (L19), green (L22), blue (L6), light blue (L7), pink (WT). Distribution of the average values in (A) all the analyzed metabolites, (B) phenolic compounds, (C) anthocyanins (D) lipids (E) chlorophylls (Ca and Cb coincide) and total carotenoids (Cxc). The most weighted loadings are represented in each plot.

FIGURE S5 | Transcripts involved in Terpene biosynthesis. 4.2.3.46 alpha-farnesene synthase (AFS1) (VIT\_00s0361g00060, VIT\_00s0392g00030, VIT\_00s0392g00060), 1.14.99.7 (VIT\_00s0441g00020) squalene monoxygenase, 4.2.3.13 (VIT\_18s0001g04710) (+)-delta-cadinene synthase, 4.2.3.75 (-)-germacrene D synthase (VIT\_18s0001g04990, VIT\_18s0001g05240), 4.2.3.119 (VIT\_08s0007g06860) pinene synthase, 1.14.-.- CYP82C4 (VIT\_18s0001g11480), 1.3.3.9 CYP72A1 secologanin synthase (VIT\_19s0135g00150), 1.14.13.72 C-4 sterol methyl oxidase (VIT\_00s2125g00010), CYP724B1 (VIT\_14s0066g00170), CYP90B1 Steroid 22-alpha-hydroxylase (VIT\_04s0023g01630, VIT\_04s0023g01640, VIT\_12s0057g01460). Green color means down-regulated gene, red color means up-regulated gene.

FIGURE S6 | Transcripts involved in alpha-linolenic metabolism. 1.13.11.12 LOX1 (VIT\_06s0004g01470) lipoxygenase 1, 21.1.141 Jasmonate O-methyltransferase (VIT\_14s0006g02170), MJAE MeJA esterase (VIT\_00s0253g00090). Red color means up-regulated gene.




cooperatively regulated by ethylene and jasmonate in Arabidopsis thaliana. J. Plant Res. 119, 407–413. doi: 10.1007/s10265-006-0287-x


growth in Arabidopsis thaliana is controlled by ERECTA. Plant J. 61, 83–95. doi: 10.1111/j.1365-313X.2009.04035.x


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Leida, Dal Rì, Dalla Costa, Gómez, Pompili, Sonego, Engelen, Masuero, Ríos and Moser. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Sequence Polymorphisms and Structural Variations among Four Grapevine (Vitis vinifera L.) Cultivars Representing Sardinian Agriculture

Luca Mercenaro<sup>1</sup> , Giovanni Nieddu<sup>1</sup> , Andrea Porceddu<sup>1</sup> , Mario Pezzotti<sup>2</sup> and Salvatore Camiolo<sup>1</sup> \*

<sup>1</sup> Dipartimento di Agraria, Università degli Studi di Sassari, Sassari, Italy, <sup>2</sup> Dipartimento di Biotecnologie, Università degli Studi di Verona, Verona, Italy

The genetic diversity among grapevine (Vitis vinifera L.) cultivars that underlies differences in agronomic performance and wine quality reflects the accumulation of single nucleotide polymorphisms (SNPs) and small indels as well as larger genomic variations. A combination of high throughput sequencing and mapping against the grapevine reference genome allows the creation of comprehensive sequence variation maps. We used next generation sequencing and bioinformatics to generate an inventory of SNPs and small indels in four widely cultivated Sardinian grape cultivars (Bovale sardo, Cannonau, Carignano and Vermentino). More than 3,200,000 SNPs were identified with high statistical confidence. Some of the SNPs caused the appearance of premature stop codons and thus identified putative pseudogenes. The analysis of SNP distribution along chromosomes led to the identification of large genomic regions with uninterrupted series of homozygous SNPs. We used a digital comparative genomic hybridization approach to identify 6526 genomic regions with significant differences in copy number among the four cultivars compared to the reference sequence, including 81 regions shared between all four cultivars and 4953 specific to single cultivars (representing 1.2 and 75.9% of total copy number variation, respectively). Reads mapping at a distance that was not compatible with the insert size were used to identify a dataset of putative large deletions with cultivar Cannonau revealing the highest number. The analysis of genes mapping to these regions provided a list of candidates that may explain some of the phenotypic differences among the Bovale sardo, Cannonau, Carignano and Vermentino cultivars.

Keywords: Vitis vinifera, next generation sequencing, structural variation, SNP, CNV, Run of homozygosity

# INTRODUCTION

Grapevine berries (Vitis spp.) are marketed worldwide as wine, fresh and dried fruits, and as ingredients for cosmetics and nutraceuticals<sup>1</sup> . These diverse applications are possible due to the broad genetic basis of cultivated grapevine germplasm (Laucou et al., 2011; Emanuelli et al., 2013; Maul et al., 2015), which has been propagated independently by many civilizations throughout history (Imazio et al., 2006; This et al., 2006). There are now thousands of cultivated varieties, many

<sup>1</sup>http://www.oiv.int/

#### Edited by:

José Tomás Matus, Centre for Research in Agricultural Genomics (CRAG), Spain

#### Reviewed by:

Rosa Arroyo-Garcia, Instituto Nacional de Investigación y Tecnología Agraria y Alimentaria, Spain Sergio Lanteri, University of Turin, Italy

> \*Correspondence: Salvatore Camiolo scamiolo@uniss.it

#### Specialty section:

This article was submitted to Crop Science and Horticulture, a section of the journal Frontiers in Plant Science

> Received: 06 March 2017 Accepted: 06 July 2017 Published: 20 July 2017

> > Citation:

#### Mercenaro L, Nieddu G, Porceddu A, Pezzotti M and Camiolo S (2017) Sequence Polymorphisms and Structural Variations among Four Grapevine (Vitis vinifera L.) Cultivars Representing Sardinian Agriculture. Front. Plant Sci. 8:1279. doi: 10.3389/fpls.2017.01279

grown in the traditional wine-producing countries of Europe, which have arisen by spontaneous mutation, hybridization, self-fertilization, and interactions with viruses (Arroyo-Garcìa et al., 2006). There is significant evidence of introgression from wild vine (Vitis vinifera europaea subsp. sylvestris) in current commercial cultivars (Sefc et al., 2003; Crespan, 2004; Myles et al., 2011).

The domesticated grapevine is thought to have originated in the Mediterranean (Zohary, 1995; Imazio et al., 2006) although a secondary center of domestication has been proposed in the western Mediterranean (Grassi et al., 2003; Chessa and Nieddu, 2005; Arroyo-Garcìa et al., 2006). The high diversity of local germplasm reflects the domestication of the wild relatives and has been conditioned by vegetative propagation and the repeated introduction of cuttings and plants (Myles et al., 2011). One of the proposed secondary origins is the island of Sardinia, the second largest island in the Mediterranean Sea (Grassi et al., 2003; Chessa and Nieddu, 2005). This ancient civilization was among the first in the western part of the basin to cultivate grapevine and process the berries (Ucchesu et al., 2015, 2016). Wild vines still grow near hundreds of Sardinian rivers, wetlands and commercial vineyards, and the berries are often used to make wine. The extant Sardinian grapevine germplasm includes hundreds of cultivars differing in agronomic performance, berry chemical composition and oenological potential (Castia et al., 1992; Calb et al., 2006; Vacca et al., 2009; Nieddu, 2011). Even so, only 26 traditional cultivars are recorded in the national grape varietal register, although dozens more can be found in older Sardinian vineyards. Despite the large number of cultivars present in Sardinia, few are routinely used for wine production. The amount of land dedicated to viticulture in Sardinia is 26,000 ha, 85% of which is represented by six cultivars. These include the three major red berry varieties Cannonau, Carignano and Bovale sardo, and the major white berry variety Vermentino (Nieddu, 2011).

Cannonau is the most important red berry cultivar grown in Sardinia (8000 hectares, 29% of the total) and is considered synonymous with Garnacha tinta cultivated in Spain and Grenache noir cultivated in France (di Rovasenda, 1877; Molon, 1906; Viala and Vermorel, 1991). This relationship has been confirmed by ampelographic analysis and the comparison of isoenzymes and molecular markers (Calb et al., 2006; De Mattia et al., 2007; Cipriani et al., 2010). The Grenache family is cultivated on 185,000 ha of land worldwide (Anderson and Aryal, 2013) and the existence of differentiated locally adapted genotypes has been proposed (Ortiz et al., 2004; Mercenaro et al., 2016b), with the sequence diversity of numerous accessions sampled in France, Spain and Italy clustering according to the sampling area (Meneghetti et al., 2011). A comparison between cultivars form Armenia and Georgia (the first areas of grapevine domestication) with European cultivars showed that Cannonau was more closely related to the transcaucasian varieties than to other Italian cultivars (Crespan, 2010). Carignano is grown mostly ungrafted on 2,000 ha of land in south-western Sardinia characterized by extremely sandy soil (Mercenaro et al., 2014). It is genetically similar to the Spanish varieties Carinena and Mazuelo, and the French variety Carignan noir (Mercenaro et al., 2014). These synonymous varieties are cultivated on 80,000 ha of land worldwide (Anderson and Aryal, 2013). Bovale sardo is cultivated on ∼800 ha in central Sardinia and is characterized by extensive intra-varietal differences (Nieddu, 2011). It probably has a local origin (Grassi et al., 2003). Finally, Vermentino is one of the most promising white wine cultivars (ranking fifth by volume of wine sold in Italy). It is traditionally cultivated in the west Mediterranean region and was recently introduced into Australia, South Africa and Argentina. Sardinia and France each cultivate Vermentino on ∼4,200 ha, with further vineyards in Ligury and West Tuscany. Sardinian Vermentino has been awarded DOCG status (Controlled and Guaranteed Denomination of Origin).

The combination of high throughput sequencing technologies and the grapevine reference genome (Jaillon et al., 2007) has facilitated comprehensive sequence analysis in diverse grapevine germplasms. Cultivars with different agronomic and oenological characteristics have been re-sequenced to identify genetic differences underlying the distinct phenotypes (Da Silva et al., 2014; Di Genova et al., 2014; Cardone et al., 2016) and comprehensive sequence variation maps are under construction to link these differences with transcriptomic and metabolomic data, as well as information about grapevine breeding practices (Ray and Satya, 2014). In this regard, availability of NGS data boosted, particularly in the last decade, the identification of candidate genes involved in response to stress (Xu et al., 2014), in the production of essential secondary metabolites (Kim and Buell, 2015) in both model and non-model plants (Unamba et al., 2015). Indeed, Giannuzzi et al. (2011) identified duplications in grapevine cultivar Pinot Noir hosting genes responsible for adaptation and response to environmental changes. The analysis of genomic features in different V. vinifera cultivars will expand our knowledge of the evolution of the grapevine genome and will facilitate breeding programs.

Here we report a thorough characterization of genomic sequence variation in four Sardinian cultivars compared to the PN40024 reference genome to determine the genomic characteristics underlying the phenotypic differences among these varieties. SNPs and indels for the four Sardinian cultivars were compared to data from three additional cultivars (Gewurztraminer, Sultanina and Tannat) that are not typical of this island agriculture. The present study aims to characterize the reported cultivars in terms of SNPs/indels, complex structural variations and degree of homozigosity, in order to speculate those features possibly underlying their phenotypic peculiarities.

# MATERIALS AND METHODS

#### Reference Sequence and Annotation

The grapevine reference genome with corresponding annotations and associated gene ontology terms (V. vinifera, cv. Pinot noir, PN40024 12× assembly V1 prediction) was downloaded from the Grapevine Genome CRIBI Biotech Center website<sup>2</sup> . We used gff2sequence (Camiolo and Porceddu, 2013) to generate coding sequences (CDSs) and the 5<sup>0</sup> and 3<sup>0</sup> untranslated regions (UTRs).

<sup>2</sup>http://genomes.cribi.unipd.it/grape/

#### DNA Resequencing

fpls-08-01279 July 18, 2017 Time: 17:22 # 3

High molecular weight DNA was extracted from nuclei starting form 3–5 g of young leaves of V. vinifera cv. Cannonau, Bovale, Carignano and Vermentino, using the procedure described in Zhang et al. (1995) without embedding the nuclei in agarose plugs, but directly performing the lysis of nuclear walls with detergent and proteinase K. Resequencing with an Illumina HiSeq 2000 instrument at the Istituto di Genomica Applicata (IGA, Udine, Italy) produced paired-end short reads of variable length and number (Supplementary Table S1). The produced reads have been deposited in the SRA database with the accession numbers SRR5803837, SRR5803836, SRR5803839 and SRR5803838 for Bovale, Cannonau, Carignano and Vermentino, respectively. Sequence read datasets were quality filtered using the NGS QC toolkit (Patel and Jain, 2012) with default parameters prior to downstream analysis. Quality filtering together with the entire downstream analysis pipeline described below was also carried out on three additional grapevine varieties (Gewurztraminer, Tannat and Sultanina) for comparison. The resequencing data corresponding to these latter three cultivars were retrieved from the Sequence Read Archive (Gewurztraminer, ERR514999; Sultanina, SRR924200; and Tannat, SRR863595 and SRR863618).

#### Alignment of Reads and Single Nucleotide Variation Detection

Filtered reads were aligned to the reference genome using Altools (Camiolo et al., 2016) (edit distance = 5% of the read length, base quality cutoff = 10). The embedded Altools Pileup Analyzer module was used to create a pileup formatted file reporting only essential data such as coverage and presence/absence of SNPs/indels at each genomic position. Base information was retrieved only when the corresponding position was covered by at least three reads, whereas SNPs and indels were considered only at positions featuring at least the average coverage and with the polymorphism supported by at least half of this value. The statistical significance of the called SNPs and indels was estimated using VarScan (Koboldt et al., 2012) with default parameters and applying a p-value cutoff of 0.05. In order to avoid possible bias in the detection of polymorphisms we only retained unambiguously mapped reads at this stage. Finally a threshold of 0.2 in the allele frequency was applied to minimize the detection of somatic mutation. In this condition only 0.1% of the called polymorphisms proved to feature an excessive depth of coverage (e.g., 6 times the coverage standard deviation) possibly underlining alignment artifacts.

#### Genome-Wide Data Visualization

The Altools Sliding Analysis module was used to visualize the alignment statistics along the genome (e.g., coverage, SNPs and indels). Briefly, each chromosome was split into a series of adjacent windows (windows size and step = 20,000 bp) that were investigated in terms of average coverage and polymorphism density. Only positions that were covered by reads were used at this stage. The resulting data were used to plot a genome circular representation using Circos (Krzywinski et al., 2009), with all the reported values normalized to the genome average.

#### Depth of Coverage Analysis

The Altools Coverage Analyzer tool was used to detect copy number variation (CNV) and presence/absence variation (PAV) in the grapevine cultivar genomes. False CNV due to known genomic repeats was avoided by comparison with the reference genome. Therefore, we first generated a simulated Illumina paired-end reads dataset for the PN40024 reference genome (average coverage 40×) using DWGSIM<sup>3</sup> and repeated the alignment and pileup procedures. Simulated reads were used in place of the real resequencing data to reduce the effect of non-homogeneous coverage and hemizygous loci in the reference genome (although we are aware that the usage of real reference genome resequencing data may take into account possible technical artifacts in the reads generation step). We then computed the coverage ratio between the target (all paired reads properly mapping were considered at this stage) and reference genomes in 500-bp adjacent windows. Significant fluctuations in the ratio identified using the DNAcopy algorithm (Seshan and Olshen, 2010) were deemed to explain the CNV. PAV was called if coverage was detected in the reference genome but not in the target. Only structural variations longer than 1000 bp were retained for downstream analyses. We decided not to use the paired end information to detect smaller structural variations, since the used DNA library featured a short insert. Hereafter we define gains as those genomic regions featuring a significant higher copy number in the target genome compared to the reference. Similarly we define losses as those regions featuring a lower copy number in the target genome (including no copies for zero-coverage areas). It is important to note that such definitions are not intended as statements of phylogenesis because the absence of an outgroup makes it impossible to establish which genome has lost or gained DNA during evolution. For the same reason, no effort was made to infer the copy number of these regions. The Altools Genic Extractor tool was used retrieve the annotated genes within the CNV and PAV regions.

#### Large Deletions Analysis

Reads mapping onto the reference produced a sam formatted alignment file (Li et al., 2009) that was used to investigate the occurrence of large deletions. We first filtered the initial dataset by removing all reads mapping at multiple positions (only alignments featuring the XT:A:U tag were retained). This step was necessary to exclude the interpretation of genome duplication events within the same chromosome as large deletions. Paired-end reads mapping at a distance between 10,000 and 1,000,000 bp, e.g., incompatible with the estimated insert size, were then considered to reflect a large deletion event. Only structural variations confirmed by at least three paired-end read mates were used for downstream analysis. Such a task was performed by using the Large deletion finder software within the Altools suite. The Genic Extractor tool was used to retrieve the annotated genes within the large deletions. Large deletions

<sup>3</sup>https://github.com/nh13/DWGSIM

together with CNV were not analyzed for the three outgroup cultivars for the sake of clarity and to keep the focus on the varieties typical of the Sardinian agriculture (although they will be considered for future studies).

#### Genic Polymorphism Analysis

Single nucleotide polymorphisms and indels were mapped to the genic portion of each genome using the Altools module Polymorphism Analyzer. This estimated the number of events with the potential to modify polypeptide structures, e.g., non-synonymous substitutions responsible for amino acid replacements or premature stop codons, or indels creating a frameshift in the CDS. Transcripts featuring more than five SNPs were aligned to the corresponding reference gene and dn/ds was calculated by using scripts incorporating the Biopython (Cock et al., 2009) library cal\_dn\_ds and using the Maximum Likelihood estimation method. Significance of the selection signal was tested by a Fisher's exact test.

### Regions Characterized by Extensive Homozygosity

Regions of homozygosity (ROH) were identified using plink (Purcell et al., 2007) with a sliding window of 500 SNPs and a minimal ROH size of 50 kb with one heterozygous or missing SNP allowed for each window. Because ROHs may arise due to hemizygosity, we excluded all ROHs that overlapped regions identified as losses.

## Gene Ontology Studies

The R package topGO was used to carry out single-gene enrichment analysis and to determine ontology codes for biological processes and molecular functions.

# RESULTS

#### Alignment Statistics

Paired-end genomic reads representing cultivars Bovale, Cannonau, Carignano and Vermentino were mapped to the PN40024 reference genome and reads representing cultivars Gewurztraminer, Sultanina and Tannat were downloaded from public databases for comparison. As shown in **Table 1**, there was significant diversity among the cultivars in terms of several sequence diversity parameters. The SNP density (number of SNPs per Mbp of covered reference genome sequence) varied from a minimum of 5508.0 for Bovale to a maximum of 8522.1 for Vermentino. The indel density (number of indels per Mbp of covered reference genome sequence) ranged from 213.4 for Gewurztraminer to 728.4 for Vermentino. The ratio of the total number of heterozygous/homozygous SNPs varied from 0.6 for Bovale to more than 2 for Cannonau, Gewurztraminer and Tannat. On the other hand, the ratio of the total number of homozygous/heterozygous indels was ∼2 in most cultivars, although Bovale was exceptional with a ratio of 0.7. Among the varieties cultivated in Sardinia, the highest sequence diversity compared to the reference genome was observed for Vermentino, as confirmed by the lowest number of aligned sites. It is important to note that both the total number of reference bases covered by reads and the total number of reads (e.g., the depth of re-sequencing) differed widely for each cultivar. However, it is unlikely that these factors influenced our diversity estimations significantly because we used a conservative approach in which variant calling was restricted to genomic regions covered by a number of reads at least equal to the average genome coverage. Mild tendencies toward a compositional shift emerged from compositional analysis of the polymorphic sites: the average GC content of the polymorphic sites was lower in the reference genome than in the resequenced cultivars, particularly in the case of Cannonau, Gewurztraminer and Tannat (**Table 1**).

To gain insight into the level of sequence diversity at regions presumably subjected to purifying selection, we extrapolated the sequence polymorphisms within the transcripts (CDS and UTRs). As expected, the polymorphism density was much lower in these regions, particularly in the CDS, where sequences are under greater selective pressure due to their role in protein synthesis (**Table 2**). The density of indels in the CDS was even lower, presumably due to their ability to cause disruptive frameshifts (**Table 2**). In some cultivars, the sequence variation in genic regions was dissimilar to the variation observed at the whole-genome level. Cannonau showed the least genomic variation but the highest SNP density in transcripts, although this trend was not uniform throughout the transcript. Indeed, Cannonau UTRs (but not CDSs) proved to be more polymorphic than the other cultivars with the exception of Gewurztraminer (**Table 2**). Bovale, the most similar to the Pinot noir reference genome at the genic level, ranked second in terms of CDS diversity. Indel density in the CDS was uniform in all cultivars with the exception of the two white berry varieties Vermentino and Sultanina, which showed a remarkably higher number of such polymorphisms in CDSs.

Many SNPs caused the loss or gain of stop codons (**Table 2**). Again, Cannonau was distinguished from the other cultivars with the highest number of both premature and new stop codons in both homozygous and heterozygous genomic regions. In contrast, the two white berry varieties Vermentino and Sultanina showed the lowest number of premature stop codons in both homozygous and heterozygous genomic regions. Most of these genes can be considered as pseudogenes because plant transcripts with premature stop codons are usually targeted for degradation via the nonsense mediated decay pathway. We found 1296 putative pseudogenes among the four Sardinian cultivars, 118 of which contained two or more premature stop codons in at least one cultivar. Among these pseudogenes, 75.6% were specific for one cultivar and only 1.3% were shared by all cultivars.

#### Homozygosity Islands

We next investigated the allelic variability of SNPs along chromosomes, seeking ROHs (chromosome regions featuring uninterrupted runs of consecutive homozygous SNPs) which are common features of many resequenced genomes (Ku et al., 2011; Metzger et al., 2015). We set a minimal ROH size of 50 kb with a sliding window of 500 homozygous SNPs, allowing for one missing or heterozygous SNP per window.

TABLE 1 | Polymorphisms statistics for 7 grapevine cultivars (four grown in Sardinia + 3 outgroups).


TABLE 2 | Polymorphisms statistics for 7 grapevine cultivars (four grown in Sardinia + 3 outgroups).


(a) SNPs and Indels percentage are calculated by dividing the number of polymorphism occurrences by the total length of the polymorphic regions.

The cultivars could be assigned to two groups, the first with many ROHs (Bovale, Vermentino and Carignano) and the other with few ROHs (Cannonau Sultanina and Tannat) with Gewurztraminer showing intermediate behavior (**Table 3**). As expected, the proportion of the genome included in ROHs was associated with the ROH number. A large proportion of the genome was found within ROHs in the first group: 17.3% in Bovale (34,847,149 bp), 8.6% in Vermentino (24,538,350 bp) and 5.9% in Carignano (18,274,355 bp). A much smaller proportion was found in the second group: 1.2% in Cannonau (4,674,100 bp), 0.9% in Sultanina (3,816,666) and 0.8% in Tannat (3,593,251 bp). The intermediate cultivar Gewurztraminer showed an intermediate proportion of 3.2% (11,412,631 bp). However, there were only minor differences between the two groups in the frequency of ROH distribution. The cultivars with many ROHs tended also to have larger ROHs, whereas those with fewer ROHs tended to have smaller ROHs (**Figure 1A**). Interestingly the frequency distribution of SNP density within ROHs distinguished the two groups more clearly: Carignano, Bovale and Vermentino contained more ROHs with densely clustered SNPs, whereas Cannoanu, Sultanina and Tannat contained more ROHs with sparse SNPs (**Figure 1B**). The ROHs were distributed along all 19 chromosomes, although in a nonuniform manner (**Figure 1C**). Only 16,402 bp of the ROH sequence was common to all cultivars, and this contained 31 protein-coding genes (Supplementary Table S2). An average of 62.5% ROH sequence in each cultivar was private, i.e., restricted to that variety.

#### Structural Variation

Copy number variation, PAV and large deletions are complex structural variations that can be inferred by the analysis of coverage variation along chromosomes. The Altools module Sliding Analysis was used to visualize these variations, and


TABLE 3 | Regions of homozygosity (ROH) statistics for the 7 analyzed grapevine cultivars.

For the four varieties cultivated in Sardinia, private ROHs are calculated within this group.

as reported for other resequenced cultivars (Da Silva et al., 2014; Di Genova et al., 2014; Cardone et al., 2016) we found that the coverage was not homogeneous along chromosomes (**Supplementary Figure S1**). We therefore used a digital comparative genome hybridization approach to identify duplicated/deleted genomic regions in the Sardinian cultivars. These were identified as regions with a copy number significantly higher (gains) or lower (losses) than the corresponding regions in the reference genome. However, we did not determine the actual copy number of these regions in the reference genome so the terms gain and loss are not intended to indicate the direction of mutational events during evolution. We identified 6526 genomic regions with significant differences in copy number among the four cultivars compared to the reference sequence with 81 regions being shared between all four cultivars and 4953 specific to single cultivars. On average, we found that 4.3% of the reference genome was duplicated and 1.4% was deleted in the Sardinian cultivars. Furthermore, 81 of the CNVs (49 gains and 32 losses) corresponding to 316,000 bp (131,000 bp in gains and 185,000 bp in losses) were common to all Sardinian cultivars, whereas 619.1 CNVs were unique to individual cultivars (**Table 4**). The common CNVs encompassed 12 protein-coding genes (Supplementary Table S2). The Cannonau genome contained ninefold more duplicated regions than the Vermentino genome and ∼2.5-fold more than the Carignano and Bovale genomes. In contrast, the Bovale genome showed the highest number of low-copy-number regions followed by Vermentino, Carignano and finally Cannonau (**Table 4**). The length distribution of gains and losses in the Sardinian cultivars is shown in **Supplementary Figure S2**, and **Figure 2** presents a circular genomic map of the distribution of gains and losses along each chromosome. Most of chromosomes 1 and 17 together with the whole of chromosome 10 did not show any gains in any of the cultivars. Vermentino showed the lowest number of chromosomes involved in gain events with chromosomes 14, 3 and 4 featuring only a few such variations. Several common patterns also emerged from the distribution of losses. For example, we observed a common high density of loss events in chromosome 16, but a very low number in chromosome 17. The absence of gains/losses within extended genomic portions must be considered in the light of the DNAcopy algorithm high stringency (e.g., segmentation default p-value < 0.01), which can result in a poorer sensitivity. In this regards, applying a higher p-value during the genome segmentation step actually resulted in the emergence of previously undetected CNVs events (results not shown). Notably, this phenomenon only apparently affects more the detection of gain compared to loss events (**Figure 2**) due to the losses datasets being enriched also in zero-coverage genomic portions.

Transposable elements (TE) are known to play a primary role in shaping the genomic architecture of plants (Carrier et al., 2012; Bai et al., 2016) and may contribute to the occurrence of CNVs. Indeed a relevant, although variable, proportion of the detected CNV proved to overlap annotated TE for all the analyzed cultivars (Supplementary Table S3). We found that, on average, TE overlap 23.2 and 6.7% of the detected gains and losses, respectively. Notably, CNVs proved to host a higher percentage of TE in cultivar Cannonau with a relative abundance almost double than those observed for the other varieties. A more detailed analysis of TE types and distribution within CNV regions was beyond the scope of this manuscript and will be reported elsewhere.

The distribution of large deletions clearly differentiated the four Sardinian cultivars, with Cannonau featuring the highest number (1990), Vermentino the lowest (50) and Bovale and Carignano featuring intermediate numbers of 419 and 529, respectively (**Figure 2** and **Table 5**). Approximately 1,100,000 bp included in the large deletions was common to the four Sardinian varieties and this encompassed 44 protein-coding genes (Supplementary Table S2). The proportion of private large deletions ranged from 16.9% in Vermentino to 54.5% in Cannonau (**Table 5**).

#### Functional Diversity

We next investigated whether the sequence and structural polymorphisms within genes provided insight into the adaptive and/or artificial selection traits of the cultivars. Gene ontology enrichment analysis was applied to the putative pseudogenes, revealing that several biological process categories such as "defense response" and "apoptotic process" were significantly overrepresented in the Sardinian cultivars (Supplementary Table S4).

We calculated the rate of non-synonymous (dn) and synonymous (ds) substitutions at loci featuring more than

five SNPs and used the dn/ds ratio to identify the genes under either purifying (dn/ds < 1) or diversifying selection (dn/ds > 1). Gene ontology single-gene enrichment analysis revealed several common features among the Sardinian cultivars. Several genes involved in methionine biosynthesis appeared subject to purifying selection in all the red berry Sardinian varieties. Similarly, a number of genes involved in the regulation of auxin response factor (ARF) signal transduction appeared subject to purifying selection in all the Sardinian varieties with the exception of Carignano (Supplementary Table S5a). Finally, genes involved in apoptosis and other defense processes appeared subject to positive selection in all the cultivars (Supplementary Table S5b).

TABLE 4 | Copy number variations statistics for the four analyzed Sardinian cultivars.


Gene ontology enrichment analysis of the genes within ROHs indicated the predominance of primary metabolism, stress response and secondary metabolism categories (Supplementary Table S6). However, within these wide classes each cultivar featured specific biological process or molecular functions. ROHs in Bovale were enriched for genes involved in defense responses and the biosynthesis of salicylic and jasmonic acids. ROHs in Cannonau were enriched for genes involved in solute transport across cellular membranes and responses to biotic and abiotic stress, such as cold, wounding and fungi. Stress response genes were also significantly enriched in the Carignano ROHs, together with genes encoding strictosidine synthetases and those



involved in cytoskeletal organization. The ROHs in Vermentino were enriched for genes involved in embryo sac development, trehalose biosynthesis and oxidation/reduction.

The ontologies of genes in CNV regions depended on whether the regions were gains or losses. The gained regions were enriched for genes involved in flavonoid synthesis and other secondary metabolic processes, especially in Cannonau (Supplementary Table S7). In contrast, the lost regions were enriched for stressresponse genes (Supplementary Table S8). As stated above, the gain or loss of regions was relative to the reference genome, so a significant enrichment should be interpreted as evidence that mutation (either deletion or duplication) has affected regions hosting specific gene functions rather than enrichment of the function with respect to the gene copy number in the reference sequence. Gene ontology enrichment analysis focusing on genes within large deletions also revealed the prevalence of genes that respond to biotic/abiotic stress. Notably, several ontologies were shared among the Sardinian cultivars, with 11 common genes involved in cycloartenol biosynthesis lost in three of the varieties (Supplementary Table S9).

#### DISCUSSION

Viticulture and wine-making play a primary role in the Sardinian economy. Indeed, almost 26,000 ha of the island is devoted to grapevine cultivation yielding ∼500,000 hectoliters of wine every year (Nieddu, 2011). Cannonau, Bovale and Carignano are among the most widespread red berry cultivars, and Vermentino is by far the most widely cultivated white berry cultivar. These varieties were resequenced to investigate genomic characteristics potentially associated with their distinct phenotypes.

#### Genetic Diversity and Distribution of Sequence Polymorphisms

Sequence reads from the Sardinian varieties were aligned to the Pinot noir PN40024 reference genome, allowing the identification of several forms of sequence polymorphism, such as SNPs and indels, as well as structural variations such as CNVs, PAVs and large deletions. The cultivars showed wide variation in several sequence diversity parameters, and 2,421,176 SNPs were discovered by comparing the Sardinian cultivars with three varieties not grown in Sardinia.

Cannonau was most similar to the reference genome in terms of the percentage of homozygous SNPs and reference bases covered by reads (**Table 1**) whereas Vermentino showed the greatest divergence from the Pinot noir genome due to the greater number of genomic positions not covered by reads, and the frequency of SNPs/indels at both the genomic and genic levels. This may reflect the original selection of this cultivar for the production of table grapes (Nieddu, 2011), in accordance with previous studies highlighting marked genomic differences between wine and table varieties (Myles et al., 2011). Indeed, several alignment statistics were common between Vermentino and Sultanina, a well-known table variety, such as the high number of indels within the transcripts and the lower number of mutations producing stop codons (**Table 2**). The ratio of heterozygous to homozygous SNPs differed substantially among the cultivars, suggesting their breeding histories were also distinct. Bovale showed the lowest ratio of heterozygous to homozygous SNPs, and historical data suggest this cultivar originated by local breeding with the selection of several clones. Based on simple sequence repeat (SSR) polymorphism, several closely related clones have been identified that can be assigned to a cluster of Bovale-like genotypes (Meneghetti et al., 2013). The breeding of these clones may have been characterized by intercrossing and the selection of Bovale-related materials, including Bovale muristellu and Bovale murru (Grassi et al., 2003).

Uninterrupted arrays of homozygous SNPs, defined as ROHs, are often considered as signatures of inbreeding. Several ROH parameters are reliable predictors of the breeding histories of carriers, including their size, SNP density and distribution. Using a conservative approach, we identified the extent of ROHs in all the cultivars. Only a small proportion of the total ROH sequence was shared among all the cultivars. Notably, the length distributions of ROHs in each cultivar were similar, with most belonging to the smallest length classes. ROHs have been associated to inbreeding events in several systems and the length distributions of ROHs has been taken as a marker of the timing and extent of inbreeding: large ROHs are associated with recent inbreeding whereas smaller ones are older and thus usually diagnostic of germplasm origin. Following these considerations the rather homogenous ROH distributions we observed may reflect the limited number of sexual reproduction events typical of grape breeding. In Cannonau, the lower number of ROHs together with the higher percentage of heterozygous polymorphisms may suggests a more complex breeding history than the other cultivars grown in Sardinia. This is supported by evidence that Cannonau clusters more closely to varieties cultivated in the near East than with other Italian varieties (Crespan, 2010). However, we urge caution in interpreting these results only in terms of inbreeding because, in species vegetatively propagated, regions with reduced heterozygosity (and thus with high level of homozigosity) may be coincident with mosaic structural variations. Application of dedicated software together with resequencing experiments featuring higher depth of coverage will be needed to discriminate ROH origin in the analyzed cultivars (Marroni et al., 2017).

#### Signatures of Selection

Selection for desirable traits may have driven the emergence of unique genomic features in each of the cultivars so we searched

for genes under purifying and positive selection by calculating the dn/ds ratio at each polymorphic locus. Our data indicated that the selected traits play key roles in the plant life cycle. For example, six genes involved in methionine biosynthesis were found to be under purifying selection in all the Sardinian red berry varieties (Supplementary Table S5a). Methionine metabolism appears to be involved in the ripening of berries given that the derivative S-adenosylmethionine is required for the production of ethylene during maturation (Agudelo-Romero et al., 2013) and methionine precursors differ widely in abundance from veraison onward in diverse grapevine varieties (Giribaldi et al., 2010). Several ARF genes were also found to be under purifying selection, and this family of regulators is also implicated in grapevine berry ripening (Wan et al., 2014). Finally, an enrichment in biological processes involved in the cell shape regulation (e.g., "regulation of cell shape," "microtubule-based movement," "actin filament-based movement") emerged when analyzing genes under purifying selection. In this regards the cytoskeleton of plant cells is believed to play a role in the response to several external stimuli such as heat or cold that are sensed as a mechanical load upon the membrane (Nick, 2013).

In contrast, stress-response genes (particularly those involved in apoptosis) were found to be under positive selection, which may provide the genetic variation needed to deal with a wider range of local conditions (Supplementary Table S5b). The plasticity of stress-response genes was also confirmed by the gene ontology enrichment analysis of transcripts featuring premature stop codons in all the cultivars. In this regard, previous reports highlighted a diverse response to abiotic stresses (e.g., water depletion) for cultivars Cannonau, Carignano and Bovale (Mercenaro et al., 2016a) with such a trend being also confirmed when Vermentino was compared to the international cv Chardonnay (Mercenaro et al., 2012). A wide variety of genes proved to be involved in the response to several abiotic stresses (e.g., high light, high heat and drought) also in other grapevine varieties. Interestingly, transcriptomic analyses revealed that the number and type of differentially expressed stress related genes may largely vary when comparing different cultivars resulting in candidate gene sets that are poorly overlapping (Rocheta et al., 2016).

# Structural Variation

Complex structural variations such as CNVs, PAVs and large deletions contribute to both intraspecies and interspecies genetic variation. CNV polymorphisms are widely studied in humans because they are associated with many severe diseases (Buchanan and Scherer, 2008). CNV has only recently been investigated in plants and CNVs may be more abundant in intergenic regions, although CNVs involving genes have also been reported (Zmie ˙ nko et al., 2013 ´ ).

Copy number variation proved to be non-homogeneously distributed along the chromosomes of the analyzed cultivars. Interestingly, the occurrence of gains (and, at a lesser extent, losses) was not detected in extended portions of chromosomes 1, 10 and 17. Although we cannot exclude that technical reasons may have contributed to such a phenomenon (see Result), other causes should be taken into consideration. The presented varieties may share a higher homology with the reference PN40024 cultivar in the highlighted genomic portions. Indeed, chromosomes 1 and 17 featured SNPs frequency values below the average for all the analyzed cultivars (data not shown). Additionally, the unequal distribution of TE (whose presence is highly correlated with the occurrence of CNVs) may contribute to the observed trend (i.e., chromosome 17 contains 4.7 repetitive sequences per Mb, that is the lowest value among the V. vinifera chromosomes).

Having identified CNVs in the genomes of the four Sardinian cultivars, we extracted the associated genes and used a single-gene enrichment analysis to investigate their ontologies. We found that biological processes and molecular functions related to stress responses were the most overrepresented categories among these genes in all four cultivars (Supplementary Tables S7, S8). However, each cultivar was also characterized by unique ontologies. For example, among the molecular functions specifically overrepresented in the Cannonau gained genomic regions we found 13 genes involved in the synthesis of naringenin and resveratrol, and 6 of the 12 known genes involved in the synthesis of jasmonate, which enhances the production of resveratrol. This observation seems to be in line with previous reports highlighting a high content of this longevity-linked (Bhullar and Hubbard, 2015) secondary metabolite in Cannonau (Franco et al., 2000; Corder et al., 2006). Genes involved in the synthesis of resveratrol and naringenin-chalcone were also overrepresented in the gained regions of Bovale together with genes involved in the metabolism of hydrogen peroxide, e.g., a molecule whose accumulation proved to vary during the V. vinifera plant cycle (Qsaib et al., 2014). Notably, genes in the gained regions of Carignano and Vermentino shared several biological processes and molecular functions related to redox activity and electron transport.

The main processes represented by genes in the lost regions of all four cultivars were related to stress responses, thus confirming the widespread genomic plasticity of this class of genes. In Cannonau, the molecular function "chitinase activity" was also overrepresented in the lost regions, and this is associated with resistance to fungal pathogens (Busam et al., 1997). In Bovale, the molecular function "strictosidine synthase activity" was overrepresented in the lost regions, concurring that the absence of these enzymes in grapevine has no impact on fitness (Zhang et al., 2014).

Finally, several genes were lost in all the red berry varieties due to large deletions events. In particular, 11 genes with cycloartenol synthase activity were potentially lost in at least one allele of Cannonau, Bovale and Carignano. Cycloartenol synthase converts 2,3-oxidosqualene to cycloartenol, which is the first step in the biosynthesis of sterols. Arabidopsis thaliana plants with a mutation in this gene failed to produce progeny suggesting a role in male gametophyte function (Babiychuk et al., 2008). Because the grapevine cultivars we investigated have been bred by vegetative propagation for several centuries it is likely that some gene required for pollen development may be lost in large deletions without this phenomenon being counter selected. Notably a significantly higher number of ontologies associated with the synthesis of resveratrol emerged for genes within large

deletions in Vermentino. This finding was confirmed in other white grape wines (Gewurztraminer and Sultanina), providing a genetic explanation for the lower resveratrol content of white wines compared to reds (Bavaresco, 2003).

#### CONCLUSION

We produced a list of CNV, SNP and indels which could be of functional significance and thus contribute to explain agronomic differences among cultivars. Although the reported polymorphisms rely on a mere in silico investigation, the high stringency of the method together with an extensive quality check of our pipeline (see Data Sheet 1) allowed to produce reliable inferences. The integration of such data with transcriptomic and metabolomic analyses under different stress conditions will allow to narrow the number of candidate regions under investigations and construct hypothesis breeding strategies to improve V. vinifera resilience.

### AUTHOR CONTRIBUTIONS

SC and AP contributed to the design and conception of the work together to the drafting of the manuscript. LM, GN, and MP contributed to the interpretation of the generated data and to

#### REFERENCES


the revision of the manuscript. All the authors approved the final draft of the submitted manuscript.

#### FUNDING

We would like to thank the R.A.S. for funding as a part of the project "Effetto dello stress idrico sulle risposte fisio-metaboliche e genetiche della vite in Sardegna" (L.R. 7 - CRP 7900).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2017.01279/ full#supplementary-material

FIGURE S1 | Coverage distribution along 19 chromosomes of Vitis vinifera for the 4 cultivars grown in Sardinia. All possible mapping locations were considered at this stage in order to highlight common patterns due to highly repetitive regions. From outward to inward: 19 Vitis vinifera chromosomes ideogram, Bovale, Cannonau, Carignano, Vermentino. Colors range ascendingly from green (low coverage) to red (high coverage) and represents coverage values that were normalized on the individual average genomic coverage (9 color classes were used, e.g., green, dgreen, vdgreen, vvdgreen, black, vvdred, vdred, dred, red, with color prefixes v, very and d, dark).

FIGURE S2 | Length distribution of gained regions, lost regions and large deletions, in the four analyzed Sardinian grape cultivars.



grapevine by means of different PCR-derived marker systems. Mol. Biotechnol. 48, 244–254. doi: 10.1007/s12033-010-9365-3


Molon, G. (1906). Ampelografia. Milan: U. Hoepli.


Viala, P., and Vermorel, V. (1991). Ampélographie. Paris: Masson.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Mercenaro, Nieddu, Porceddu, Pezzotti and Camiolo. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Omics Approaches for Understanding Grapevine Berry Development: Regulatory Networks Associated with Endogenous Processes and Environmental Responses

Alejandra Serrano<sup>1</sup> , Carmen Espinoza<sup>1</sup> , Grace Armijo<sup>1</sup> , Claudio Inostroza-Blancheteau<sup>2</sup> , Evelyn Poblete<sup>1</sup> , Carlos Meyer-Regueiro<sup>1</sup> , Anibal Arce<sup>1</sup> , Francisca Parada<sup>1</sup> , Claudia Santibáñez1,3 and Patricio Arce-Johnson<sup>1</sup> \*

<sup>1</sup> Laboratorio de Biología Molecular y Biotecnología Vegetal, Departamento de Genética Molecular y Microbiología, Pontificia Universidad Católica de Chile, Santiago, Chile, <sup>2</sup> Núcleo de Investigación en Producción Alimentaría, Facultad de Recursos Naturales, Escuela de Agronomía, Universidad Católica de Temuco, Temuco, Chile, <sup>3</sup> Ecophysiology and Functional Genomic of Grapevine, Institut des Sciences de la Vigne et du Vin, Institut National de la Recherche Agronomique, Université de Bordeaux, Bordeaux, France

#### Edited by:

Simone Diego Castellarin, University of British Columbia, Canada

#### Reviewed by:

Grant Cramer, University of Nevada, Reno, United States Gregory Alan Gambetta, Ecole Nationale Supérieure des Sciences Agronomiques de Bordeaux-Aquitaine, France

> \*Correspondence: Patricio Arce-Johnson parce@bio.puc.cl

#### Specialty section:

This article was submitted to Crop Science and Horticulture, a section of the journal Frontiers in Plant Science

Received: 27 March 2017 Accepted: 10 August 2017 Published: 07 September 2017

#### Citation:

Serrano A, Espinoza C, Armijo G, Inostroza-Blancheteau C, Poblete E, Meyer-Regueiro C, Arce A, Parada F, Santibáñez C and Arce-Johnson P (2017) Omics Approaches for Understanding Grapevine Berry Development: Regulatory Networks Associated with Endogenous Processes and Environmental Responses. Front. Plant Sci. 8:1486. doi: 10.3389/fpls.2017.01486 Grapevine fruit development is a dynamic process that can be divided into three stages: formation (I), lag (II), and ripening (III), in which physiological and biochemical changes occur, leading to cell differentiation and accumulation of different solutes. These stages can be positively or negatively affected by multiple environmental factors. During the last decade, efforts have been made to understand berry development from a global perspective. Special attention has been paid to transcriptional and metabolic networks associated with the control of grape berry development, and how external factors affect the ripening process. In this review, we focus on the integration of global approaches, including proteomics, metabolomics, and especially transcriptomics, to understand grape berry development. Several aspects will be considered, including seed development and the production of seedless fruits; veraison, at which anthocyanin accumulation begins in the berry skin of colored varieties; and hormonal regulation of berry development and signaling throughout ripening, focusing on the transcriptional regulation of hormone receptors, protein kinases, and genes related to secondary messenger sensing. Finally, berry responses to different environmental factors, including abiotic (temperature, water-related stress and UV-B radiation) and biotic (fungi and viruses) stresses, and how they can significantly modify both, development and composition of vine fruit, will be discussed. Until now, advances have been made due to the application of Omics tools at different molecular levels. However, the potential of these technologies should not be limited to the study of single-level questions; instead, data obtained by these platforms should be integrated to unravel the molecular aspects of grapevine development. Therefore, the current challenge is the generation of new tools that integrate large-scale data to assess new questions in this field, and to support agronomical practices.

Keywords: grapevine fruit development, seed development, biotic and abiotic stresses, transcriptomics, metabolomics

## INTRODUCTION

fpls-08-01486 September 6, 2017 Time: 16:34 # 2

The grapevine (Vitis vinifera), one of the most important fruit crops worldwide, provides berries that can be used as fresh fruit, raisins, and for wine making and distillation of liquors. The grapevine has fleshy berries derived from the ovary of the flower, whose development is a complex process that can be divided into three stages with distinctive physiological and biochemical characteristics (Coombe and McCarthy, 2000). During the first stage (stage I) there is an exponential increase in berry size due to rapid cell division and growth, leading to the establishment of the final number of cells (Coombe and Hale, 1973). Some of the principal compounds that are present in the berry at stage I are tartaric and malic acids, which accumulate mainly in skin and flesh and confer acidity to fruits and wine (Sweetman et al., 2009, 2012). The second stage (stage II) is a lag phase in which important physiological and biochemical changes occur, such as softening and coloring. Within this stage, veraison takes place, characterized by the beginning of the synthesis of anthocyanins, soluble flavonoids compounds that provide color to red varieties (**Figure 1**) (Boss et al., 1996). Sucrose, originating from leaves, reaches the fruits through the phloem, and is then hydrolyzed forming glucose and fructose (Robinson and Davies, 2000; Kennedy, 2002; Vignault et al., 2005; Deluc et al., 2007; Hayes et al., 2007; Fontes et al., 2011). Stage II is thus a transition between an unripe fruit and the third stage of development (stage III or ripening). The latter involves important morphological and physiological changes, like color development (Boss et al., 1996), turgor reduction and berry enlargement (Chervin et al., 2008), and decreased acidity (Costenaro-da-Silva et al., 2010), among others. In addition, hormonal changes that occur throughout development positively or negatively regulate ripening (**Figure 1**; Gerós et al., 2012). Therefore, during ripening, a large number of complex transcriptional and/or post-transcriptional regulatory processes are triggered. In this review, we focus on the integration of global approaches, including proteomics, metabolomics, and especially transcriptomics, to understand grape berry development and the influence of environmental factors on this process. Thus, we will cover initial fruit development, with emphasis on seed formation; the importance of coloration and hormonal changes during development, especially on ripening; and finally, the effect of environmental factors on this process will be discussed.

# GRAPE BERRY DEVELOPMENT FROM AN OMICS PERSPECTIVE

#### Seed Development and Seedless Fruits

Grape berry development begins after fertilization, when in a process known as fruit set, the ovary changes from a stationary state and experiences an abrupt increase in size that occurs due to cell division and enlargement, leading to rapid pericarp growth. Throughout this period, seed development is an important process, mainly because seeds produce auxins, gibberellins (GAs) and cytokinins, which play multiple roles in grape berry development (Keller, 2010). Seed and berry development are coordinated, and the changes that seeds undergo have an impact on fruit ontogeny. The first stage of berry development is characterized by a rapid increase in seed size, during which embryogenesis and endosperm growth occur. At the second stage, about 10 and 15 days prior to veraison, seeds reach their final size and maximum fresh weight, and at the beginning of the third stage, embryo growth ceases and the endosperm accumulates reserves until the seeds turn dormant (**Figure 1**; Keller, 2010).

Ripe berries usually contain up to four seeds derived from four ovules (Dokoozlian, 2000; Kennedy, 2002). However, seedless grape varieties have been spontaneously found in nature and have been preserved over the years through asexual propagation. Seedless berries develop naturally via two different mechanisms, parthenocarpy and stenospermocarpy, which generate berries without or with rudimentary seeds, respectively (Varoquaux et al., 2000). In order to understand the main differences between parthenocarpy and stenospermocarpy, we will discuss the few available Omics analyses of early stages of berry development and seed formation.

#### Parthenocarpy

In parthenocarpic fruits, the stimulus of pollination is sufficient to trigger fruit set (Dokoozlian, 2000). Since the ovary is able to enlarge and form a berry without ovule fertilization, there is no seed in the fruit (Varoquaux et al., 2000). Until now, few parthenocarpic grape cultivars have been described. Of these, cv. Corinto Bianco (CB), a somatic variant of the seeded cv. Pedro Ximenez (PX), constitutes a good model to study seed formation (Vargas et al., 2007). To understand the molecular differences between CB and PX genotypes, flowers at 1 and 10 days pre-anthesis were transcriptionally compared using microarray (Royo et al., 2016). The analyses allowed the identification of 1958 differentially expressed genes between CB and PX. Interestingly, several genes that are specifically expressed in reproductive organs were down-regulated in CB. Processes such as cell wall biosynthesis, cytoskeleton biogenesis, vesicular transport, signaling through G proteins or phosphatidylinositol, among others, were enriched. Also, 14 single-nucleotide polymorphisms (SNP) were identified between both genotypes, which could explain the parthenocarpy phenotype (Royo et al., 2016). Considering that microarrays deliver limited information, a suitable approach to analyze the different stages of development in more detail, would be using RNA-seq technologies, in order to

**Abbreviations:** ABA, abscisic acid; ABF, ABA-response element binding factor; ARF, auxin response factor; ASR, ABA stress responsive element; BR, Brassinosteroids; CAA, carbonic anhydrase; CAB, chlorophyll a/b binding protein; CB, Corinto Bianco; ERF, ethylene response factor; GAs, gibberellins; GLD, grapevine leafroll disease; GLRaV, Grapevine leafroll-associated virus; GVA, Grapevine virus A; HXK, hexokinase; IAA, indole-3-acetic acid; JA, jasmonic acid; LOX, chloroplast lipoxygenase; NCED, 9-cis-epoxycarotenoid dioxygenase; PX, Pedro Ximenez; QTL, quantitative trait locus; ROS, reactive oxygen species; RSPaV, Rupestris Stem Pitting virus; SDI, seed development inhibitor; SNP, singlenucleotide polymorphism; SnRK1, sucrose-non-fermentative related kinase 1; SSH, suppression subtractive hybridization; T6P, trehalose-6-phosphate; UV, ultraviolet.

FIGURE 1 | Fruit development and environmental effects. Scheme of the most important changes that berries and seeds undergo during development, and the main environmental factors affecting this process. (A) Boxes indicate the phase where each stress condition (temperature, water-related stress and UV-B radiation, Botrytis cinerea and viruses) affect grape berry and its development. (B) Changes in size, color, brix degree, and pH during berry ripening and (C) variations in hormonal content during grape berry development. (D) Seed development showing the time in which parthenocarpy and stenospermocarpy can take place. The main events are indicated by open triangles. Bars represent the described changes throughout development, in which gray and white represent the higher and lower estimated referential values for each parameter, respectively.

gain further insights into the understanding of seed development and to generate new parthenocarpic varieties.

#### Stenospermocarpy

fpls-08-01486 September 6, 2017 Time: 16:34 # 4

During stenospermocarpy, pollination and fertilization occur normally, but a few weeks later, the embryo and/or the endosperm abort and the berries that have been generated possess just traces of seed (Varoquaux et al., 2000). It has been demonstrated that stenospermocarpy occurs in several seedless varieties, and is stable and unaffected by environmental factors (Zhang et al., 2013). However, little is known about the molecular mechanisms that underlie stenospermocarpy in grapes. The most accepted hypothesis proposes the existence of a dominant regulator gene called Seed Development Inhibitor (SDI), which could control three other recessive genes (Bouquet and Danglot, 1996). Different studies based on quantitative trait locus (QTL) analysis have reported a main QTL in linkage group 18 (LG18) (Doligez et al., 2002; Cabezas et al., 2006; Mejía et al., 2007; Costantini et al., 2008), which could explain between 50 and 70% of the seedlessness phenotype in grapes; LG18 could be considered as the SDI locus trait. In this context, VvAGL11 (MADs BOX transcription factor) was in silico mapped to the SDI locus and it has been proposed as the main functional candidate gene for seedlessness (Mejia et al., 2011). In fact, it was demonstrated that the silencing of a VvAGL11 homologous gene in tomato (Solanum lycopersicum L. cv Micro-Tom) generates fruits with few or rudimentary seeds (Ocarez and Mejia, 2016). Based on genome sequencing data, it is known that in the stenospermocarpic variety cv. Thompson Seedless, the VvAGL11 gene has an insertion of 15 bp in the 50UTR, which could be the cause of the seedless phenotype (Di Genova et al., 2014). In addition, in cv. Sultanine Monococco, which is also a seeded variety of Thompson, the VvAGL11 transcript level is higher in comparison with the seedless variety (Ocarez and Mejia, 2016), supporting the hypothesis that this gene is one of the main regulators of seed formation in grapes.

On the other hand, due to abnormal ovules may be formed during flower development before or after meiosis, through either the abnormal development of the nucellus or ovule integuments, or the degeneration of the egg in the embryo sac (Ebadi et al., 1996). In order to determine the molecular bases of this phenotype, flowers from cv. Thompson Seedless and cv. Thompson Seeded, a spontaneous mutant with seeded berries, were compared through suppression subtractive hybridization (SSH) (Hanania et al., 2007, 2009). The results demonstrated that ch-Cpn21, a gene that encodes for a chloroplastic chaperonin, is repressed in developing flowers of cv. Thompson Seedless. Likewise, the silencing of this gene in tobacco plants (Nicotiana benthamiana) and tomato induces seed abortion (Hanania et al., 2007). The use of somatic variants in combination with current transcriptomic technologies, would be very promising in the study of stenospermocarpy, helping to discover new genes playing important roles in seed abortion.

Based on the studies of Costenaro-da-Silva et al. (2010) and Nwafor et al. (2014), several genes have been associated with early stages of grape berry development. These include VvUBP1, a heterogeneous-nuclear ribonucleoprotein, VvFS41, a putative S1-like ribosomal protein involved in mRNA processing and synthesis of proteins related with cell division during the first days of berry development, VvERF1 and VvERF9, which encode for transcription factors related to several developmental processes, VvDOF1, possibly related to seed development, and VvRIP1 and ABI3, which have been related to hormone signaling, among others. Many of these genes have pleiotropic effects, so it is difficult to estimate their specific molecular contribution to the stenospermocarpy phenotype. Some of them could be involved in this process, but their functional characterization is needed to test this hypothesis.

#### Non-characterized Mechanisms of Seedlessness

A demonstrated way to produce seedlessness is the exogenous application of GAs before bloom or during anthesis. It is believed that GAs promote seedless grapes by inhibiting pollen germination, allowing unfertilized ovules to enlarge and form fruits, as occurs in parthenocarpy (Kimura et al., 1996; Cheng et al., 2013). However, another study suggests that exogenous GAs interfere with seed development, as described in stenospermocarpy (Cheng et al., 2013). So, the mechanism involved in this response is not clear. A transcriptional analysis by RNA-seq was performed in GA3-treated flowers of the seeded cv. Kyoho and a comparison with non-treated flowers was carried out (Cheng et al., 2015). This study demonstrated that GA<sup>3</sup> application modifies the expression profile of genes related to developmental processes, such as cellular metabolism, biosynthesis of different metabolites, stress response, transport, etc. Also, changes in the expression of genes related to flowering, fruit, and embryonic development were found. Within the genes possibly related to seedlessness, the Pelo gene, whose mouse homolog has a role in meiosis and causes embryonic lethality (Adham et al., 2003), was repressed after GA<sup>3</sup> treatment, and was correlated with seedlessness in grapes (Cheng et al., 2015). The Pelo gene probably has conserved roles across several species. However, deeper functional studies are needed to corroborate this information in plants and to determine if this gene does indeed fulfill a role in seed development, and more studies are needed to correlate any transcriptional changes with particular phenotypes. Recent studies have demonstrated that reactive oxygen species (ROS) are present throughout the entire seed's life cycle (Jeevan Kumar et al., 2015). In fact, the oxidative damage induced by an imbalance in plant redox homeostasis can affect normal seed development, leading to abortion (Cheng et al., 2013). Pathways related to ROS scavenging and detoxification are significantly affected after GA<sup>3</sup> treatment (Cheng et al., 2015). So, probably, exogenous GA application generates physiological changes that could induce seedless fruits through a ROS-related mechanism, but further research is needed to understand the role of ROS regarding the presence or absence of seeds. Naturally occurring seedlessness could be the result of a series of coordinated transcriptomic switches that cause a global reprogramming of the cell. To date, little is known about the seedless phenotype in grapevines, presenting a great challenge for researchers. The best model for understanding seedlessness is to compare somatic variants (seeded versus seedless) through global approaches, since they

have the same genetic background and could be used to discover new genes involved in this phenotype. Even though somatic variants are rare in nature, it is clear that these comparisons are much more informative than the use of two different varieties.

### First Stage of Grape Berry Development

The first stage of grape berry development (stage I) is initiated with fruit set. During the first 2 weeks, berry size increases markedly as auxin and GAs directly promote cell division and enlargement (Ojeda et al., 1999; Bottcher et al., 2010; Fortes et al., 2015). Tartaric, malic, and other organic acids, along with different phenolic precursors such as tannins and hydroxycinnamates, are synthesized, modifying the organoleptic properties of the berries (Deluc et al., 2007). Besides, minerals, micronutrients, and aroma-related compounds are present. Transcriptomic analysis of young berries in cv. Shiraz revealed an enrichment of hormone signaling responsive transcripts, suggesting that hormone-controlled metabolic pathways are highly active in early stages of development (Sweetman et al., 2012). During this stage, GAs are the key regulators of fruit set, cell division and cell expansion (Fortes et al., 2015). RNAseq analysis of cv. Centennial Seedless berries treated with GA<sup>3</sup> (12 days after flowering), revealed a repression of an abscisic acid (ABA)-response element binding factor (ABF) and ethylene response factors (ERFs) (Chai et al., 2014). Showing the occurrence of both GA3–ABA and GA3–ethylene crosstalk. The role of jasmonic acid (JA) in grapes remains unclear, but, as has been demonstrated in potato (Solanum tuberosum) leaves, it might stimulate cell division (Ravnikar et al., 1992). In grape, high levels of JA are present during the first stage, which then decrease in mature berries (Fortes et al., 2015). A proteomic study in cv. Muscat Hamburg has reported abundant levels of chloroplast lipoxygenases (LOX), enzymes that provide intermediates for JA biosynthesis during green berry development, followed by a decrease when berries reach a size of 15 mm (Martinez-Esteso et al., 2011). In the case of auxins, it has been proposed that they have a role in fruit growth delaying ripening (Fortes et al., 2015). In berry flesh of cv. Kyoho, a high concentration of auxins has been reported, in particular of indole-3-acetic acid (IAA) during the beginning of stage I, with a rapid decrease at the end of this stage and throughout stage II, consistent with the high rate of cell division observed in the first stage (Zhang et al., 2003). Considering that the final number of cells in the grape berries is defined in the first stage of development (Dokoozlian, 2000), the interaction between hormones regulating cell division is key to cluster progress, and might be an interesting target in studies aimed at improving yield.

# Second Stage of Grape Berry Development

#### Main Changes in Metabolites during Stage II

The second stage of grape berry development (stage II) is a lag phase, where the rate of increase in both fresh and dry weight is very low. At the end of this stage, veraison occurs, which is the transition from the second to the third stage of berry development, and is considered the onset of ripening. Different physiological and biochemical changes take place during veraison, of which anthocyanin synthesis and sugar accumulation are the most characteristic and important processes. In fact, anthocyanins are one of the main pigments present in colored grape berry skins (Souquet et al., 1996), while sugar content is widely considered one of the most important properties that define ripening (Guelfat-Reich and Safran, 1971; Jayasena and Cameron, 2008).

Depending on the cultivar, five types of anthocyanins are frequently found in V. vinifera, which are associated with organoleptic properties such as color (in the case of red wine), bitterness, astringency and also as antioxidant molecules with beneficial effects on human health (Dixon et al., 2005). Anthocyanin biosynthesis occurs through the phenylpropanoid pathway, in which two types of genes are involved: structural genes, encoding for biosynthetic enzymes, and regulatory genes, which are associated with temporal and spatial regulation of the structural genes (Deluc et al., 2007). Both, structural and regulatory genes are present in colorless and colored grapevine cultivars, but in the case of white cultivars, color is not expressed due to multiallelic mutations in MybA1 and MybA2, that prevent the transcription of these two important positive regulators of the phenylpropanoid pathway (Kobayashi et al., 2004). In the case of red cultivars, MYBA1 and MYBA2 transcription factors control anthocyanidin glycosylation through the regulation of flavonoid 3-O-glucosyltransferase (UFGT) expression (Ford et al., 1998). The anthocyanin biosynthesis pathway is not only regulated by MYB transcription factors, as it is also controlled by the critical transcriptional R2R3-MYB/bHLH/WD40 (MBW) complex in grapevine (Wu et al., 2014).

Flavonoid 3<sup>0</sup> -hydroxylase (F30H) and flavonoid 3<sup>0</sup> ,50 hydroxylase (F3<sup>0</sup> 5 <sup>0</sup>H) genes seem to be an important regulatory points in anthocyanin biosynthesis (Castellarin and Di Gaspero, 2007; Matus et al., 2016). Their proteins belong to the cytochrome P450 protein family and compete for a common precursor for the biosynthesis of red and blue anthocyanins, respectively (Bogs et al., 2006). Metabolic and transcriptomic analyses determined that in cv. Cabernet Sauvignon and cv. Shiraz, the F3<sup>0</sup> 5 <sup>0</sup>H gene is up-regulated, whilst that of F30H is down-regulated (Degu et al., 2014). However, further studies are necessary to understand the fine regulation of the phenylpropanoid pathway, focusing on the anthocyanin branch. In this case, the use of varieties with different berry skin colors would be informative. None of the aforementioned studies consider pink varieties, which have an intermediate color between red and white varieties, and could be used to complete the overview of anthocyanin biosynthesis in a fuller range of colors.

In a recent analysis using Omics approaches to analyze ripe berry skins of five cultivars, Cabernet Sauvignon, Merlot and Pinot Noir (red cultivars), and Chardonnay and Semillon (white cultivars) (Ghan et al., 2015), several transcripts and metabolites were mapped to the phenylpropanoid pathway. A higher transcript abundance for enzymes involved in anthocyanin biosynthesis, such as phenylalanine ammonialyase (PAL), chalcone synthase (CHS), flavanone 3-dioxygenase

(F3H), leucoanthocyanidin dioxygenase (LDOX), and UFGT was observed only in red cultivars. Shikimate was the most abundant metabolite in cv. Cabernet Sauvignon, which acts as a precursor for aromatic amino acid biosynthesis within the shikimate pathway (Maeda and Dudareva, 2012). This intermediary is important because it allows the transfer of the carbon skeleton into anthocyanin structures, and could become a crucial point in the study of anthocyanins in different varieties.

Sugar accumulation (mainly of glucose and fructose) is another important process that begins in veraison and continues throughout ripening. Sugar sensing mechanisms may play important roles during grape berry ripening, as they do in other aspects of plant development (Smeekens et al., 2010; Wind et al., 2010). Thus, the role of sugar is covered in the context of the third stage of berry development (see Third Stage of Grape Berry Development).

#### Hormonal Control during Stage II

ABA levels are high in young berries and then fall until veraison. A microarray analysis carried out over seven sequential points of berry development in cv. Cabernet Sauvignon, revealed that the transcript abundance of 9-cis-epoxycarotenoid dioxygenase (NCED1), the enzyme that conducts the limiting step in ABA synthesis (Tattersall et al., 2007), increases during the lag phase and peaks at veraison (Deluc et al., 2007). A similar expression pattern was shown for a gene encoding for the ABA signaling transduction protein phosphatase 2C ABI1, while the gene encoding a transcription factor of the same pathway, ABI3/VP1 (Abscisic acid Insensitive 3/Viviparous 1), showed the highest transcript abundance during lag phase (Deluc et al., 2007). Several studies have highlighted the control that ABA exerts over the biosynthesis of anthocyanins; at the transcriptional level by upregulation of biosynthetic genes, and at the metabolic level by increasing anthocyanin content (Wheeler et al., 2009; Giribaldi et al., 2010; Cramer et al., 2014). In this context, A 2- DE proteomic approach in cv. Cabernet Sauvignon showed that ABA treatment before veraison increases three proteins required for flavonoid biosynthesis: chalcone isomerase, dihydroflavonol-4-reductase, and anthocyanidin reductase (Giribaldi et al., 2010).

In non-climacteric fruits, such as grape, the role of ethylene is not fully understood due to the low levels of this hormone during development and the technical difficulties associated with its quantification (Symons et al., 2012). Nevertheless, several reports indicate a possible role of this hormone in grape berry ripening, mainly supported by the consistent presence of a small peak about 2 weeks after veraison (Chervin et al., 2004). These data are consistent with the findings of Pilati et al. (2007), who showed that the expression of the ACC synthase gene, involved in ethylene synthesis, increases just prior to veraison and decreases afterward, together with a peak in expression of ACC oxidase around veraison, which encodes for the enzyme responsible for the last step of ethylene biosynthesis.

Brassinosteroids (BR), on the other hand, are steroid hormones that have been implicated in the ripening of nonclimacteric fruits (Symons et al., 2006; Chai et al., 2013). The transcript abundance of the brassinosteroid receptor 1 gene (BRI1) peaks in the entry to lag phase and declines thereafter (Deluc et al., 2007). The expression profile of VvBR6OX1, which encodes for the enzyme that converts 6-deoxocastasterone to castasterone (the bioactive BR in grapes) shows a peak of induction just prior to veraison (Pilati et al., 2007). This evidence is consistent with an increase in BR levels at veraison and the high content observed during ripening in cv. Cabernet Sauvignon berries (Symons et al., 2006). Interestingly, exogenous application of BR increases anthocyanin content leading to premature grape berry coloration, similar to the effect of ABA. The connection between the molecular pathways of BR and ABA that regulate initial events of ripening stages has yet to be clarified. Based on transcriptomic analysis of cv. Merlot berries, it has been hypothesized that BR might be an early signal for ripening, modulating ethylene content (Ziliotto et al., 2012). In this model, the small peak of ethylene could upregulate genes associated with ABA biosynthesis and then initiate all ripening-associated ABA-induced metabolic changes (Ziliotto et al., 2012).

It has been well documented that auxin has a negative role during grape berry ripening. In fact, IAA levels (the active form) remain low from veraison throughout ripening, and auxin treatments during pre-veraison inhibit ripening (Davies et al., 1997; Bottcher et al., 2010, 2011; Ziliotto et al., 2012). Two auxin carriers (an AUX1-like and a PIN1-like) are expressed before veraison, while two auxin response factors (ARFs), ARF5 and ARF18, and an auxin receptor of the ABP family are expressed at pre-veraison and are then repressed during ripening (Pilati et al., 2007; Fortes et al., 2011). It has been suggested that ethylene represses auxin biosynthesis and thus regulates the balance between auxin and ABA to initiate ripening (Ziliotto et al., 2012). Probably, a network coordinated by ABA, BR, ethylene, and auxin levels are regulating the ripening stage, however, the master regulators that connect all these pathways are still unknown.

#### Third Stage of Grape Berry Development Main Changes in Metabolites during Stage III

During the third stage of development (Stage III), berries approximately double in size and there is a marked decrease in organic acid concentration and a dramatic accumulation of glucose and fructose (∼1 M each) in the vacuole of flesh cells (Fontes et al., 2011; Dai et al., 2013). The scientific community has gained understanding about the complexity and diversity of sugar-sensing systems, including hexokinase (HXK), protein kinases, as well as, novel molecular regulators, such as trehalose-6-phosphate (T6P) (Li and Sheen, 2016). It has been shown that the HXK enzyme, responsible for the 6-phosphorylation of glucose and fructose, plays a dual-function with both catalytic and regulatory activities and therefore, links gene expression and metabolism in plants (Moore et al., 2003). HXK-dependent signaling represses photosynthetic related-genes in the presence of hexoses, forming a repressive complex that is directly associated with the promoter regions of several genes including those that encode for chlorophyll a/b binding protein (CAB) and carbonic anhydrase (CAA) (Cho et al., 2006).

In grapevine, a genome wide analysis using the completely sequenced V. vinifera genotype PN40024 (cv. Pinot Noir) led

to the identification of six members of the HXK family (Çakir, 2014). Four genes that encode for HXKs in cv. Cabernet Sauvignon were analyzed (Gambetta et al., 2010). The authors showed that these genes are highly regulated at the transcriptional level during berry development. Specifically, HXK-1, HXK-2, and HXK-3, which were induced during ripening, while HXK-4 was repressed (Gambetta et al., 2010). Interestingly, under water deficit conditions, HXK-4 was induced during the third stage of berry development compared to the control under well irrigated conditions, indicating that there is both genetic and environmental control of the sugar sensing mechanisms during ripening (Gambetta et al., 2010).

Protein kinases are the major components of intracellular signaling and are responsible for rapid responses to changes in the environment. VviSK1, a protein kinases with sugar signaling function during berry development, whose transcript was shown to be accumulated after sucrose treatments in cv. Cabernet Sauvignon suspension cells (Lecourieux et al., 2010), positively affects sugar accumulation in grape cells and controls glucose transport through the regulation of four genes that encode the hexose transporters VvHT3, VvHT4, VvHT5, and VvHT6. Moreover, during berry development, VviSK1 transcripts decrease after the green stage and increase again after veraison, when sugar is accumulated (Lecourieux et al., 2010). Another protein kinase that may participate in sugar signaling during ripening is SnRK1 (Sucrose-non-fermentative Related kinase 1). In plants, SnRK1 receives inputs from hormones, as well as, sugar phosphates, and has been linked to several developmental processes and the control of primary and secondary metabolism, including photosynthesis and anthocyanin biosynthesis (Baena-Gonzalez et al., 2007; Nunes et al., 2013; Tsai and Gazzarrini, 2014). In grapevine, SnRK1 transcripts accumulate continuously in cv. Cabernet Sauvignon berries from the green stage until ripening (Gambetta et al., 2010). Nonetheless, to our knowledge, the abundance and activity of this protein kinase has not been measured in grapevine berries.

Phosphate sugars are other members of the sugar signaling landscape. Among them, T6P, which is generated by primary metabolism (Lunn et al., 2014) has been recently uncovered as a signal molecule with major implications in plant growth, development, and metabolism (Van Houtte et al., 2013; Wahl et al., 2013). Transcriptomic studies had uncovered that several genes that control T6P abundance are regulated during berry development. In one of the first gene expression profile analyses using AFLP in berry samples from cv. Corvina, the authors reported a T6P-phosphatase as one of the most upregulated genes in postharvest (Zamboni et al., 2008). Moreover, Deluc et al. (2007), using the first commercially available grapevine Affymetrix, identified different profiles for genes encoding T6Psynthase, which was overexpressed in the early days before veraison, and T6P-phosphatase, which was overexpressed at postharvest. Suggesting that the abundance of T6P is highly controlled during grape berry development.

In plants, T6P is linked to sugar signaling and the control of SnRK1, which is linked to developmental processes and control of metabolic pathways, including repression of anthocyanin biosynthesis (Baena-Gonzalez et al., 2007). In grapevine, several transcriptomic studies have shown that orthologs genes of SnRK1 and of the enzymes that control T6P homeostasis are highly regulated during berry development (Deluc et al., 2007; Gambetta et al., 2010). Therefore, the SnRK1/T6P pathway may be an important component of sugar signaling during berry development, and in this context, it remains to be studied whether the activity of SnRK1 protein kinase is actually inhibited by T6P, as found in other plant tissues. The Omics studies shown so far, using mainly transcriptomic and metabolomic approaches, have been useful for the identification of several sugar-signaling components, leading to the proposal that new mechanisms or candidate genes are involved in berry ripening. Nonetheless, specifically in the field of signaling through protein kinases, it is known that transcript accumulation is not the only, or main mechanism that influences their role and activity. In this perspective, it may be necessary to perform more protein-oriented Omics studies such as proteomics or phosphoproteomics. These are powerful technologies and could help to elucidate the importance of protein kinase signaling during berry development.

#### Hormonal Control during Stage III

Microarray and RNA-seq analyses have uncovered transcriptional reprogramming during ripening (Fasoli et al., 2012). At the onset of ripening in cv. Cabernet Sauvignon, low levels of IAA are required, while the auxin conjugate to aspartate (inactive form) concentration is high (Bottcher et al., 2010). In the case of IAA conjugate formation, the up-regulation of a gene coding for GH3.1 was found at veraison (Bottcher et al., 2010), in contrast to a decrease in GH3.3 expression (Pilati et al., 2007). On the other hand, genes coding for AUX–IAA proteins, transcriptional repressors of auxin-responsive genes, are down-regulated during ripening, while genes coding for IAA19 and IAA16 are up-regulated around veraison. Likewise, a gene homologous to Arabidopsis amidase AtAMI1, that in vitro synthesizes IAA from indole-3-acetamide, decreases its expression during ripening (Pilati et al., 2007) indicating a complex regulation for the maintenance of low levels of active auxin during ripening.

Related to the ethylene metabolism, the role of this hormone during ripening has not been clearly established (Chervin et al., 2004). Nevertheless, it is known that the transcript abundance of genes coding for ACC synthase decrease at veraison, while several genes coding for ACC oxidase are down-regulated and only one is up-regulated during ripening (Terrier et al., 2005). The intricate regulation of the ethylene signaling pathway during ripening seems to be more consistent and clearer during the later stage of this process. Cramer et al. (2014) assessed the transcriptome of Cabernet Sauvignon berries in the late stages of ripening using whole-genome microarrays. They reported that several positive regulators of the ethylene pathway are upregulated, including three different ethylene receptors (VviETR1, VviETR2, and VviEIN4) and several members of the ERF family of transcription factors. Moreover, the negative regulator of ethylene signaling, VviCTR1 is downregulated at the transcript level during late ripening in both pulp and flesh (Cramer et al., 2014). Supporting the idea of an active ethylene signaling role during berry ripening.

Regarding BRs, it has been shown that exogenous application to grape berries significantly promotes ripening, whilst endogenous BR levels dramatically increase at the onset of ripening and then decrease (Symons et al., 2006). These results coincide with the transcript accumulation of the VvBR6OX1 gene observed by Pilati et al. (2007), responsible for the synthesis of the bioactive BR, castasterone. In addition, a gene coding for an enzyme putatively involved in castasterone catabolism (castasterone 26-hydroxylase), leading to the inactivation of this BR, is down-regulated at ripening (Fortes et al., 2011). On the other hand, a gene related to BR biosynthesis that codes for steroid 5-alpha-reductase (DET2), was less expressed around veraison (Fortes et al., 2011). Diminished expression of biosynthetic genes could be associated with negative feedback regulation by increasing levels of BRs.

The ABA concentration increases dramatically during berry ripening (Coombe and Hale, 1973). Several reports suggest that this hormone plays a major role controlling color development (Koyama et al., 2010) and softness (Gambetta et al., 2010). ABA levels are directly related to changes in NCED activity (Wheeler et al., 2009), and indeed, NCED1 transcripts peak around veraison and decrease at advanced ripening (Deluc et al., 2007). A proteomic analysis in berry skins of cv. Barbera at different stages throughout ripening revealed that the most abundant proteins belong to the ABA stress responsive elements (ASR) family, representing nearly 13% of the total protein spot volume in early ripening (Negri et al., 2008). In Arabidopsis, the proposed model of ABA signaling involves the protein kinases SnRK2, which act as positive elements in signaling downstream of ABA. SnRK2s interact with the negative regulators PP2C protein phosphatases that inhibit the activity of SnRK2s. Recently, Liu et al. (2016) identified eight VviSnRK2 genes in the grapevine genome and generated a detailed co-expression network of the ABA signaling components, including transcription factors from the ABF family. They found a high co-expression coefficient of both VviSnRK2.8 and VviSnRK2.11 with VviABF2, which is an important transcriptional regulator of ABA-dependent signaling during grape berry ripening (Nicolas et al., 2014). VviABF2 expression rises from veraison until ripening, and transcriptomic analysis of VviABF2-overexpressing grapevine cells allowed the identification of several co-overexpressed genes regulated by ABA (Wong et al., 2013; Nicolas et al., 2014). The regulation of the ABA signaling pathway is complex; the cellular, physiological and transcriptomic responses to this hormone change dramatically in a tissue-specific manner (Rattanakon et al., 2016), and are also cultivar-dependent (Rossdeutsch et al., 2016). Moreover, as has been mentioned, the gene families that act positively, or negatively downstream of ABA are composed of several genes, and it is plausible that subspecialization of members in these families exists. The complexity mentioned above is a significant challenge for researchers when attempting to extract mechanisms related to ABA signaling. An important issue to address in the coming years is that of understanding the differential sensitivity to ABA across cultivars, and the crosstalk of ABA with other signals that seem to be important in berry development, such as sugars and other hormones.

# THE EFFECT OF THE ENVIRONMENT ON GRAPE BERRY DEVELOPMENT

Grape berries are constantly exposed to several biotic and abiotic factors that, to some extent, can affect their normal development and trigger positive or negative changes. In most cases, these factors negatively impact grape cultivation at different stages of plant and berry development during pre- and post-harvest (Armijo et al., 2016a). In this review, some of the most relevant grapevine abiotic and biotic stresses are discussed.

## Abiotic Stress

Climate change has caused significant warming in most grapegrowing areas, increasing some important abiotic stresses like heat, drought and UV radiation (Teixeira et al., 2013; Keller, 2015). These stresses mainly affect phenolic metabolism and, at the same time, berry composition and development (**Figure 1**).

Changes in temperatures during vegetative grape development are associated with changes in berry harvest date (Meier et al., 2007). Studies of transcripts, metabolites and proteins also show that sugar accumulation and other parameters related to color and aroma could be affected. Moderate warmer temperatures (∼25◦C) lead to higher berry sugar content (Coombe, 1987), while higher temperatures (>30◦C) negatively affect photosynthesis, with consequent reductions in sugar, anthocyanin, and malic acid accumulation, followed by a decrease in berry size and weight (Sadras and Moran, 2012; Teixeira et al., 2013; Rienth et al., 2016; de Rosas et al., 2017). Sugar and organic acid metabolism are desynchronized in ripening grapevine fruits at high temperatures, and secondary metabolism is diminished due to the transcriptional repression of their respective genes (Rienth et al., 2016; de Rosas et al., 2017). Thus, high temperatures are a negative regulator of berry development at ripening, but the mechanism behind this is still not clear. Integrated global analyses are required to identify the possible genes associated with the changes in the corresponding physiological traits.

In general, V. vinifera is considered as a salt and drought tolerant species (Tattersall et al., 2007). However, stress caused by water availability is having progressively more impact, due to can generate significant effects on grapevine cultivation. In response to salinity and drought, plants intensify the synthesis of ABA, which is transported to the aerial organs, inducing changes in the expression of genes related to their acclimatization (Tattersall et al., 2007; Cramer et al., 2011). Few studies have addressed salinity and drought stress in berries. For instance, it has been shown that water stress can increase berry flavonol content and affect the expression of genes involved in biosynthesis of stilbene precursors (Teixeira et al., 2013). All these analyses suggest a differential response to water limiting abiotic stresses that is cultivar dependent. Likewise, the optimal growth temperature for grapevines may vary between cultivars, and the activation of ABA and ethylene signaling pathways can differ according to their sensitivity or tolerance to drought. These responses have consequences in grapevine berries, since a common mechanism in response to stress in these organs, is

the induction of the anthocyanins accumulation, which act as protective molecules.

#### The Effect of UV Radiation in Grape Berry Development

Vitis vinifera is often cultivated in Mediterranean climates with varied UV-B radiation dosages (Martinez-Luscher et al., 2013), and it is considered as well adapted to solar radiation due to a variety of physiological responses, mainly based on antioxidant enzyme activities and secondary metabolites. The UV-B spectrum (280–315 nm) can provoke potential damage in macromolecules, including DNA, induce ROS, and disrupt several cellular processes in all living organisms (Frohnmeyer and Staiger, 2003; Jenkins, 2009). In grapevine, several studies have been performed to discover the processes associated with the UV radiation response during berry development. In this context, VviHY5 (ELONGATED HYPOCOTYL 5), VviHYH (VviHY5 HOMOLOGUE), and VviUVR1 (the photomorphogenic factor UV-B RECEPTOR 1) genes were characterized (Loyola et al., 2016). In this work, the authors described that the expression of VviHY5 and VviHYH differs during grape berry and inflorescence development upon exposure to low or high UV-B radiation, while VviUVR1 expression was not regulated by UV-B. Studies performed by Carbonell-Bejerano et al. (2014) indicated that grapevine berries respond to UV-B through the activation of the phenylpropanoid pathway and the production of photoprotective compounds. The accumulation of polyphenolic compounds in the berry involves specific UV-responsive genes that induce the expression of phenylpropanoid pathway related genes and several MYB transcription factors that regulate this pathway (Matus et al., 2009; Berli and Bottini, 2013). Matus et al. (2009) demonstrated that different light conditions increase the accumulation of flavonoid compounds in grape berries, while Loyola et al. (2016) shown that high and low UV-B radiation induce flavonol accumulation in this organ. Carbonell-Bejerano et al. (2014) suggest that UV-B radiation triggers flavonol accumulation in grape berry skin of cv. Tempranillo and induces the expression of VvFLS1 and VvGT5, two flavonol biosynthetic genes. Furthermore, several flavonol biosynthetic genes are regulated by the R2R3-MYB transcription factor VvMYBF1, which triggers flavonol and anthocyanin production in grape berries exposed to solar UV radiation (Czemmel et al., 2009, 2017; Matus et al., 2009; Matus, 2016). Genome-wide microarray studies performed in grape berry skins of cv. Pinot Noir exposed to UV-C light (100–280 nm), showed 238 upregulated genes (more than fivefold), including several genes encoding for proteins related to stilbene synthesis (Suzuki et al., 2015). These authors also reported that UV-C light increases levels of phenolic compounds like resveratrol and its analogs. Similar results were observed in berries of cv. Tempranillo exposed to solar UV radiation (Carbonell-Bejerano et al., 2014). In general, several volatile compounds accumulate in grape berries during ripening, but the amount of these compounds depends on specific irradiance levels and the type of radiation (Joubert et al., 2016).

Summarizing, there are numerous studies demonstrating that light can affect anthocyanin accumulation in berry skins. Which can be explained by changes in the expression of structural genes related to the phenylpropanoid pathway, as well as regulatory genes such as those of the MYB, bHLH, and WD40 families (Wu et al., 2014).

#### Biotic Stress

In addition to abiotic stress, grape berry development can be influenced by biotic factors, such as pathogens, of which fungal and viral diseases are the most common and harmful, negatively affecting fruit quality.

#### Fungal Infections: Botrytis cinerea and Its Dual Effect on Berry Development

The most important fungal disease affecting grape berry development is gray mould, caused by B. cinerea. Grape berries are resistant to the infection until veraison, but are highly susceptible at the onset of ripening and harvest (Kelloniemi et al., 2015). As a necrotrophic pathogen, B. cinerea secretes lytic enzymes and phytotoxins in order to promote cell degradation (Armijo et al., 2016b). Most of the agronomically relevant grapevine cultivars are susceptible to this pathogen, leading to significant losses worldwide.

Different large-scale approaches have been carried out in order to understand the regulatory networks and processes involved in the grape berry–B. cinerea interaction, and to characterize how berry development is affected. Transcriptomic and metabolic analysis of cv. Marselan, comparing B. cinerea berries at veraison with ripe berries, revealed that the former activates an early burst of ROS, together with multiple defense responses, including a salicylate-dependent pathway, resveratrol synthesis and cell-wall strengthening. In contrast, ripe berries activate the JA-dependent pathway against the fungus (Kelloniemi et al., 2015). As a common response, both developmental stages displayed an upregulation of genes encoding WRKY transcription factors, pathogenesis-related proteins, glutathione S-transferase (involved in cellular detoxification), stilbene synthase and PAL (involved in phenylpropanoid biosynthesis), and production of anthocyanins and phytoalexins. Global metabolic changes in cv. Marselan induced by B. cinerea infection correlate the greater resistance of veraison berries with an accumulation of resveratrol and caffeic, ferulic, and chlorogenic acids (Kelloniemi et al., 2015). Also, significantly higher levels of proline, glutamate, arginine, and alanine were detected in B. cinerea-infected ripe berries of cv. Chardonnay, as well as, an accumulation of glycerol, gluconic acid, and succinate, mainly in the berry skin (Hong et al., 2012). A reprogramming of carbohydrate and lipid metabolism toward an increased synthesis of secondary metabolites with antioxidant properties, such as trans-resveratrol and gallic acid, was also observed by Agudelo-Romero et al. (2015) in cv. Trincadeira. During later stages of infection, energy metabolism (photosystem I supercomplex) and secondary metabolism (phenylpropanoid and stilbenoid biosynthesis) also seemed to be downregulated (Agudelo-Romero et al., 2015).

Contrary to the effects caused by gray mould on grape berries, some particular cases of B. cinerea infection can generate favorable effects on wine grapes, in an interaction known as noble

rot. Botrytized wines are produced from grapes that have been affected by this fungus under specific environmental conditions, which are typically hot and dry. The infection produces berry dehydration, altering metabolic processes and the saprophytic microbiota (Magyar, 2011). The berry–fungus interaction promotes the accumulation of secondary metabolites that enhance wine grape composition in ripe berries. Transcriptomic and metabolic analyses of noble rot in cv. Semillon determined that anthocyanin biosynthesis is the most consistent hallmark of noble rot. In addition, the biosynthesis of terpenes and fatty acid aroma precursors increase during the infection (Blanco-Ulate et al., 2015). APETALA2/ETHYLENE RESPONSIVE FACTOR (AP2-ERF), and NON APICAL MERISTEM/ARABIDOPSIS TRANSCRIPTION FACTOR/CUPSHAPED COTYLEDON (NAC) transcription factors, were up-regulated during noble rot (Blanco-Ulate et al., 2015). Early products of the phenylpropanoid pathway are accumulated in noblerotted berries, such as rosmarinic acid (a cinnamic acid derivative with antioxidant and aromatic properties). Also, a significant accumulation of several flavonoid glycosides and flavanones was detected, along with build ups of cyanidin-3 rutinoside, delphinidin-3-rutinoside, cyanidin-3-gentiobioside, and delphinidin-3-gentiobioside, anthocyanins that are normally scarce in white-skinned grape berries. Other aromatic compounds such as acetophenones, benzoic acid derivatives, methoxyphenols, and phenolic glycosides showed increased abundance in noble rot, together with gallic acid, a precursor of tannin biosynthesis (Blanco-Ulate et al., 2015).

In conclusion, reprogramming of secondary metabolites and hormonal pathways are common features in B. cinerea-infected grape berries. Additionally, it has been shown that during grape berry development, the fruit undergoes changes that facilitate fungal infection, such as fruit softening, organic acid and sugar level modifications, loss of the preformed defenses and decreased stilbene production, among others. On the other hand, noble rot also alters berry metabolism by inducing stress responses and accelerating ripening to enhance the colonization process. B. cinerea infections can affect the color and sugar concentration, improving wine grape composition. This effect is caused by an imbalance of hormone synthesis and perception, which in turn activates several ripening-associated pathways. However, the mechanism behind this acceleration is still under study.

#### Viral Diseases and Their Effect on Grape Berry Development

Viral diseases are also common in grapevine plantations. Infections caused by these pathogens are highly complex, due to the large number of viral agents described and the occurrence of multiple infections (Prosser et al., 2007; Martelli, 2014; Jooste et al., 2015; Naidu et al., 2015). Grapevines show no resistance against virus; instead, viruses and host plants establish compatible interactions, where pathogens spread throughout all plant tissues, unimpeded by the resistance responses, generating global cellular stress and developmental defects. However, in compatible interactions, hosts are not passive against the pathogen, and molecular, cellular and physiological responses can be observed (O'Donnell et al., 2003; Ehrenfeld et al., 2005). In general, grapevine viruses infect vegetative organs, but infections also have consequences for berry development, causing a reduction in berry setting, and delayed berry ripening (Martelli, 1993, 2014). Molecular changes during berry ripening in virus-infected grapevine plants have been less characterized than leaf symptomatology. For instance, characterization of the Grapevine leafroll-associated virus 3 (GLRaV-3) infection in the red cv. Cabernet Sauvignon revealed the presence of viral particles in berry tissues together with massive transcriptional changes, which were more pronounced during ripening (Aquea et al., 2011; Vega et al., 2011). Since this virus is restricted to the phloem (Martelli, 2014), GLRaV-3 infection could physically modify sugar accumulation, altering source–sink relationships.

Transcript profiling analyses performed in cv. Cabernet Sauvignon berries at veraison and ripening, using the V. vinifera Affymetrix GeneChip, revealed numerous changes in transcripts related either to viral infection or to berry development (Vega et al., 2011). About 400 genes showed differential expression between veraison and ripening, in uninfected tissues. However, only half of these exhibited such differences when the two stages were compared in infected berries. Thus, viral disease greatly modifies the transcript abundance profile during berry development. The number of differentially expressed genes in infected berries was higher during ripening (146 up- and 86 down-regulated genes) than at veraison (41 up- and 14 downregulated genes), suggesting that the former stage could be more dramatically affected by virus infection. Among the transcripts that change in infected berries, is a group related to sugar transport and metabolism, including ATOCT2, a carbohydrate transmembrane transporter; ATSPS4F, a putative sucrose-phosphate synthase; a short-chain dehydrogenase/reductase (SDR) family protein involved in sugar metabolism; SUS2, sucrose synthase 2; and BFRUCT3, a betafructosidase. In agreement with this, glucose and fructose levels also decreased during ripening in infected berries. Several genes from the phenylpropanoid pathway were repressed by viral infection during ripening, such as CHS2 and UFGT, as well as genes that encode for transcription factors MYBPA1 and MYBA (anthocyanin biosynthesis), and FLS1, related to the flavonol biosynthetic pathway. These results were further supported by a decrease in total anthocyanin content and flavonol concentration during ripening in infected berries (Vega et al., 2011).

A characterization of berries of the Italian cv. Nebbiolo harboring a mixed infection of GLRaV-1, Grapevine virus A (GVA) and Rupestris Stem Pitting virus (RSPaV), showed significant differences in bud burst index, berry weight, titratable acidity, and resveratrol content when compared with uninfected berries (Giribaldi et al., 2011). In that study, a proteomic analysis revealed that mixed viral infection affects proteins related to cell structure metabolism in pulp, such as pectin methylesterase, N-acetyl-gamma-glutamyl-phosphate reductase, plastid movement impaired 1, phosphoglycerate kinase, polyphenol oxidase and alpha-tubulin, among others (Giribaldi et al., 2011). A thorough study carried out over three seasons, on the effects of grapevine leafroll disease (GLD) on cv. Merlot, showed that infection impacted greatly on yield, as well as on fruit quality (Alabi et al., 2016). For instance, the

authors consistently found a lower fruit yield over the seasons evaluated, supporting previous conclusions that GLD negatively affects vine performance. In virus-infected cv. Merlot plants, developing green berries showed minor compositional changes in comparison to uninfected plants. However, after veraison, dramatic variations were observed as a consequence of viral disease, suggesting that the virus can affect ripening-related processes occurring from veraison onward, as previously shown for cv. Cabernet Sauvignon (Vega et al., 2011).

In general, transcriptomic and metabolic data support the observation that viral diseases delay grape berry ripening, altering several characteristic parameters associated with this stage, such as sugar accumulation and color, among others. However, more studies should be carried out in order to establish how viruses alter grapevine berry ripening, how cultivars and environmental factors interact to produce the complete symptomatology, and how these multiple cues modify berry ripening.

#### CONCLUSION

Significant progress has been made toward understanding grape berry development, and how environmental factors can positively or negatively regulate this process. In this field, Omics platforms have been an important tool in the elucidation of the mechanisms underlying these interactions. Due to the lack of transgenic lines and suitable technologies for reverse genetics in grapes, Omics analyses have allowed us to make progress in unraveling the complex mechanisms that take place during berry development. Of the future challenges, the establishment of a robust model to assess biological questions is key. In grapes, the availability of mutant varieties and related cultivars with contrasting phenotypes is an advantage, but differences between cultivars could be more complex than expected. Therefore, global analysis should be carried out. In this context, as Omics provide numerous tools that generate huge data sets from an overall perspective, the integration of this information is the next challenge which needs to be addressed in order to understand the different processes underlying grape berry development. Systems biology deals with the integration of these data sets, advancing the way in which biological processes are studied from gene-by-gene

#### REFERENCES


studies toward a global perspective, where the different processes are depicted in regulatory networks. Those networks are useful in the prediction of gene function, while providing new insights into the regulatory mechanisms at a global level. The generation of robust networks to identify new regulators and genome-wide responses to environmental factors requires a vast number of data sets and the integration of multi-omics studies. In grape berries, most of the Omics studies are based on transcriptomic and metabolomic profiles; more-integrated networks are hindered by the lack of proteomic and phosphoproteomic studies. Another challenge in the bioinformatic field is the standardization and centralization of the stored data in order to facilitate the access to, and analysis of, Omics studies. Currently in grape, due to the multiple sources of data and gene annotation, there is a lack of consensus in the integrative tools available. For instance, annotation version 1 (V1) and version 2 (V2) differ in the number of annotated genes, with V2 having around 2000 new genes and 3000 putative long non-coding RNAs (lncRNA). The integration of the different annotations is a task that remains unresolved by the scientific community studying grape. Therefore, to improve our current knowledge, further Omics studies are undoubtedly necessary, yet this new data must be integrated with systems biology tools in order to comprehensively depict the associated regulatory networks.

### AUTHOR CONTRIBUTIONS

The manuscript was written by AS, CE, GA, CI-B, EP, CM-R, CS, FP, AA, and PA-J. AS was involved in revising the manuscript critically for important intellectual content. Manuscript editing was conducted by AS and CE. Figure design was conducted by CM-R. All authors contributed, read and approved the final manuscript. PA-J made the final approval.

# FUNDING

This work was supported by FONDECYT postdoctoral research 3150608 (AS), Millennium Nucleus of Plant Systems and Synthetic Biology NC130030 and FONDECYT 1150220.





and the promotion of Vitis vinifera L. berry ripening by abscisic acid. Aust. J. Grape Wine Res. 15, 195–204. doi: 10.1111/j.1755-0238.2008.00045.x


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Serrano, Espinoza, Armijo, Inostroza-Blancheteau, Poblete, Meyer-Regueiro, Arce, Parada, Santibáñez and Arce-Johnson. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# How Single Molecule Real-Time Sequencing and Haplotype Phasing Have Enabled Reference-Grade Diploid Genome Assembly of Wine Grapes

Andrea Minio<sup>1</sup> , Jerry Lin<sup>1</sup> , Brandon S. Gaut <sup>2</sup> and Dario Cantu<sup>1</sup> \*

<sup>1</sup> Department of Viticulture and Enology, University of California, Davis, Davis, CA, United States, <sup>2</sup> Department of Ecology and Evolutionary Biology, University of California, Irvine, Irvine, CA, United States

Keywords: heterozygosity, inbreeding depression, cabernet sauvignon, comparative genomics, grape

#### Edited by:

pan-genome

José Tomás Matus, Centre for Research in Agricultural Genomics, Spain

#### Reviewed by:

Jordi Garcia-Mas, Institute for Research and Technology in Food and Agriculture, Spain Michela Troggio, Fondazione Edmund Mach, Italy

#### \*Correspondence:

Dario Cantu dacantu@ucdavis.edu

#### Specialty section:

This article was submitted to Technical Advances in Plant Science, a section of the journal Frontiers in Plant Science

> Received: 04 April 2017 Accepted: 02 May 2017 Published: 17 May 2017

#### Citation:

Minio A, Lin J, Gaut BS and Cantu D (2017) How Single Molecule Real-Time Sequencing and Haplotype Phasing Have Enabled Reference-Grade Diploid Genome Assembly of Wine Grapes. Front. Plant Sci. 8:826. doi: 10.3389/fpls.2017.00826

## HIGH HETEROZYGOSITY IS A CHALLENGE FOR GRAPE GENOME ASSEMBLY

Domesticated grapevines (Vitis vinifera) have relatively small genomes of about 500 Mb (Lodhi and Reisch, 1995; Jaillon et al., 2007; Velasco et al., 2007), which is similar to other small-genomes species like rice (430 Mb; Goff et al., 2002), medicago (500 Mb; Tang et al., 2014), and poplar (465 Mb; Tuskan et al., 2006). Despite their small genome size, the sequencing and assembling of grapevine genomes is difficult because of high levels of heterozygosity. The high heterozygosity in domesticated grapes may be due, in part, to their domestication from an obligately outcrossing, dioecious wild progenitor. Domesticated grapes can be selfed, in theory, because their mating system transitioned to hermaphroditic, self-fertile flowers during domestication. In practice, however, selfed progeny tend to be non-viable, presumably due to a high deleterious recessive load and resulting inbreeding depression. As a consequence of these fitness effects, most grape cultivars are crosses between distantly related parents (Strefeler et al., 1992; Ohmi et al., 1993; Bowers and Meredith, 1997; Sefc et al., 1998; Lopes et al., 1999; Di Gaspero et al., 2005; Tapia et al., 2007; Ibáñez et al., 2009; Cipriani et al., 2010; Myles et al., 2011; Lacombe et al., 2013).

One such cultivar is Cabernet Sauvignon, one of the most widely cultivated wine grape cultivars. Cabernet Sauvignon was produced from a cross between Sauvignon Blanc and Cabernet Franc sometime before the seventeenth century in the Aquitaine region of France (Bowers and Meredith, 1997). Whether a spontaneous hybrid or a product of human breeding, all of the Cabernet Sauvignon grown around the world is thought to have resulted from this single hybridization event. Just as the parents of Cabernet Sauvignon have been identified, the genetic origin of many other important wine grape cultivars is known, and they often originate from the direct crossing of common, distantly-related cultivars (Strefeler et al., 1992; Ohmi et al., 1993; Qu et al., 1996; Bowers and Meredith, 1997; Sefc et al., 1998; Lopes et al., 1999; Crespan and Milani, 2001; Vouillamoz et al., 2003, 2004; Di Gaspero et al., 2005; Vouillamoz and Grando, 2006; Lacombe et al., 2007, 2013; Tapia et al., 2007; Boursiquot et al., 2009; Ibáñez et al., 2009; Cipriani et al., 2010; Myles et al., 2011; García-Muñoz et al., 2012). Due to this intraspecific hybridization process, levels of heterozygosity in grape cultivars can easily exceed 11% (Jaillon et al., 2007).

High heterozygosity is challenging for genome assembly, because heterozygous genomes typically produce more fragmented sequences than haploid or homozygous genomes of similar size and complexity (Yu et al., 2005; Argout et al., 2011; The Tomato Genome Consortium, 2012). The goal of standard assembly approaches is to collapse homologous regions with sufficient similarity into haploid consensus sequences, but divergent haplotypes in heterozygous regions typically result in multiple, difficult to resolve assembly paths which must then be assembled separately. Additionally, the boundaries between haploid consensus contigs and heterozygous regions cannot be resolved with a unique path; as a result they are left unlinked, which breaks assembly contiguity (**Figure 1A**). Altogether, elevated heterozygosity increases fragmentation and inflates the size of the total assembly, potentially doubling the genome size if the majority of the two homologous genomes are assembled separately (Huang et al., 2012; Li et al., 2012; Safonova et al., 2015). Fragmentation and retention of redundant regions can also lead to inaccurate gene models, apparent paralogous genes and duplicated blocks, incorrect gene copy number, and synteny breaks.

#### INITIAL ATTEMPTS TO SEQUENCE THE GRAPE GENOME

Despite the challenges in assembling heterozygous genomes, the commercial and cultural importance of the grapevine has led to several sequencing attempts. Two genome reference drafts for the common grapevine were released in 2007 (Jaillon et al., 2007; Velasco et al., 2007). Remarkably, these were the first genomes of any fruiting crop to be sequenced and only the fourth for flowering plants. These reference genomes, both of which utilized the Pinot Noir cultivar, were assembled using different approaches to address heterozygosity. The first genome by Jaillon et al. reduced heterozygosity by inbreeding a line of Pinot Noir (var. PN40024) to ∼7% heterozygosity (Jaillon et al., 2007). To produce the second genome, Velasco et al. sequenced a Pinot Noir clone (ENTAV115) directly then assembled contigs that represented separate homologous chromosomes (Velasco et al., 2007). Unsurprisingly, these early efforts are poor by current standards. The PN40024 genome had ∼8.4-fold coverage and was assembled into 19,577 contigs with a contig N<sup>50</sup> of only 65.9 kbp. Later sequencing increased coverage to up to 12x and the contig N<sup>50</sup> of the PN40024 genome to 102.7 kb (**Figure 1B**). The ENTAV115 genome used both Sanger pairedreads and 454 sequencing to achieve a total coverage of ∼4.2x. Although riddled with gaps and potentially omitting large regions of repetitive sequences where genes could be located, the two genomes provided valuable insights into grape genomes. Together they revealed that the Pinot Noir genome features: (i) ∼30,000 protein-coding genes, comparable to Arabidopsis but about 75% of rice and poplar; (ii) a high proportion of repetitive elements comprising an estimated ∼40% of the genome; (iii) complex patterns of gene duplications consistent with one or more paleopolypoidy events; (iv) expansion of gene families that influence the organoleptic properties of the berry; (v) a typical number (∼200) of NBS-LRR genes, which often function in disease resistance, and (vi) a standard complement of genes involved in disease signaling pathways. Despite its limitations, the PN40024 genome assembly has proven to be invaluable to the grape research community. Cited in over 2,000 articles, it has served as a reference in more than 3,000 genome-wide transcriptional analyses.

Following the publication of the PN40024 genome in 2007, no genome reference of equivalent or greater quality has been released for V. vinifera. Only a handful of studies have attempted to use bona fide genome-wide approaches to measure diversity within the species (Giannuzzi et al., 2011; Da Silva et al., 2013; Di Genova et al., 2014; Cardone et al., 2016). With the advent of second generation short read sequencing, attempts were made to perform de novo assembly and reference based resequencing of grape cultivars. These attempts failed to provide a high quality representation of the sequenced grape genotypes. A de novo approach was adopted to assemble the genome sequence of Thompson Seedless, a ubiquitous multipurpose cultivar. Despite an enormous sequencing depth (327x), the short fragment size did not permit resolution of repetitive regions, resulting in an extremely fragmented assembly (Di Genova et al., 2014; **Figure 1B**). For the wine grape cultivar Tannat (Da Silva et al., 2013), the authors applied a reference based assembly approach, which had proved to be effective in assembling multiple Arabidopsis genotypes (Gan et al., 2011). However, reference-based assembly failed to reconstruct genotype specific sequences with Tannat data, demonstrating that large scale resequencing initiatives like the 1,000 Human Genome project (Auton et al., 2015) and the 1,001 Arabidopsis Genomes project (Alonso-Blanco et al., 2016) would not succeed for Vitis. In fact, while the approach supported variant calling with de novo assembly to resolve regions highly divergent in sequence between Tannat and PN40024, it was unable to recover regions absent in the reference but present in Tannat. Consequently, over 10% of the gene space was not represented in the assembly, illustrating that the genomic sequence of one cultivar is insufficient for representing the total variability of the species. To improve representation of the V. vinifera pan-genome and encompass the variability of the species, we need the complete de novo assembled genomes of additional genotypes. Moreover, as grape cultivars are intraspecific hybrids of different genotypes, assembly of each genome should include a diploid representation of the genome to preserve information about the characteristics of each haplotype.

# RECENT DEVELOPMENTS IN GRAPE GENOME SEQUENCING

Single Molecule Real Time (SMRT) DNA sequencing (Pacific Biosciences) has emerged as a leading technology for characterizing complex structural variations, supporting and refining the assembly of complex genomes in hybrid fashion or alone for reconstructing highly continuous assemblies of both

#### FIGURE 1 | Continued

sequences from the Cabernet Sauvignon primary contigs were aligned using GMAP (Wu and Watanabe, 2005) to the Cabernet Sauvignon haplotigs and the PN40024 chromosomes to identify the shared part of the represented gene space. Only alignments with identity ≥80% and coverage ≥66% were considered. In similar fashion, coding sequences from the Cabernet Sauvignon haplotigs were aligned against the primary contigs and the PN40024 chromosomes, and coding sequences from PN40024 were aligned against both primary contigs and haplotigs of Cabernet Sauvignon.

small and highly repetitive genomes (Chin et al., 2013; Doi et al., 2014; Huddleston et al., 2014, 2016; Gordon et al., 2016; Ricker et al., 2016; Seo et al., 2016; Vij et al., 2016). The advantage of SMRT technology arises from the delivery of long reads, currently averaging over 30 kbp and potentially approaching 100 kbp. In addition to facilitating assembly of more contiguous genomes, long reads carry the necessary information to phase haplotypes over multiple kilobase distances. The open-source software, FALCON-unzip (Chin et al., 2016), was developed specifically to utilize the long reads generated using SMRT sequencing technology and assemble diploid genomes into highly contiguous and correctly phased diploid genomes. The algorithm first constructs a string graph composed of "haploid consensus" contigs together with bubbles representing structural variant sites between homologous loci. Sequenced reads are then phased and separated for each haplotype on the basis of heterozygous positions. Phased reads are finally used to assemble the backbone sequence (primary contigs) and the alternative haplotype sequences (haplotigs) (**Figure 1A**). The combination of primary contigs and haplotigs constitute the final diploid assembly with phased single-nucleotide polymorphisms and structural variants between the two haplotypes.

We have recently reported the assembly using SMRT technology and FALCON-unzip of the highly heterozygous diploid genome of Cabernet Sauvignon (Chin et al., 2016), one of the most widely cultivated wine grape cultivars. As it is the progeny of Cabernet Franc and Sauvignon Blanc, two cultivars with extremely divergent phenotypical traits, reconstructing the diploid structure of Cabernet Sauvignon is necessary for identifying the alleles inherited from the parent cultivars. We sequenced the Cabernet Sauvignon genome with a coverage depth of ∼140x using SMRT sequencing technology. Sequencing reads were then assembled using FALCON-unzip into a highly contiguous genome that integrated phased haplotype information. FALCON-unzip generated a set of primary contigs (591.4 Mbp in 718 contigs with N<sup>50</sup> = 2.17 Mbp, **Figure 1B**) that covers one of the two haplotypes, and a set of correlated haplotigs (367.8 Mbp in 2,037 contigs with N<sup>50</sup> = 0.80 Mbp). The assembled sequences exceed PN40024 contigs and Thompson Seedless scaffolds by nearly two orders of magnitude in size (**Figure 1B**), ranking this assembly not only as the best V. vinifera genome assembly but also among the highest quality plant genomes published to date, including other genomes sequenced with SMRT technology (Sakai et al., 2015; VanBuren et al., 2015; Jiao et al., 2016; The UC Davis Coffee Genome Project, 2017). Symptomatic of the extreme divergence in allele sequences in Vitis, the length of the primary assembly was inflated with respect to the expected genome size, illustrating one of the challenges of sequencing highly heterozygous genomes (Chin et al., 2016). After manual removal of un-phased haplotigs, the primary assembly is an ideal candidate for scaffolding or hybrid assembly with optical maps to produce a genome assembly of even higher quality.

Preliminary gene model prediction identified over 34,000 protein coding sequences on the primary assembly of the Cabernet Sauvignon genome and nearly 24,000 on the haplotigs (Chin et al., 2016). Just a few hundred of PN40024 annotated coding genes did not find any suitable alignment on the Cabernet Sauvignon assembly (411 genes; identity ≥80% and coverage ≥66%), but nearly 4,900 Cabernet Sauvignon loci could not be found on the PN40024 genome (**Figure 1D**). These results are in accordance with other studies that reported presence/absence polymorphisms of gene models between wine grape cultivars (Da Silva et al., 2013; Venturini et al., 2013; Jiao et al., 2015), but the high number of genes not found in PN40024 likely reflects its incompleteness. Moreover, nearly 2,100 coding sequences identified in the Cabernet Sauvignon haplotigs were not found on the primary assembly (**Figure 1D**). While limited by the preliminary status of the annotation, these observations point to a high degree of structural variation between homologous chromosomes. Moreover, these structural variations are likely to have functional consequences since they encompass coding sequences. The variability between haplotypes may also impact and potentially confound the analysis of RNAseq data. In the worst case, the expression of haplotype-specific loci that are not represented on the reference genome would be assigned to the most similar genomic region of the reference, which is likely to generate expression mismeasurement artifacts. As shown in **Figure 1C**, in the presence of a diploid reference (primary contigs plus haplotigs), about 10% more RNAseq reads map at ≥99% identity. This observation suggests that when both alleles are represented in the reference reads align to their respective haplotype; RNAseq can therefore be used to determine allelic specific gene expression.

#### CONCLUSIONS

Genome resequencing projects of both prokaryotic and eukaryotic organisms have clearly shown that one genome sequence is insufficient to properly describe the genetic characteristics of a species (Tettelin et al., 2005; Donati et al., 2010). In order to grasp comprehensive genetic variability and complete gene pools in outcrossing species, such as grape, we also need to go beyond the generation of haploid consensus sequences and focus our efforts to begin assembling diploid genome sequences with phased haplotypes. As discussed in this article, long read sequences and bioinformatic tools that take advantage of them have solved a critical bottleneck in grape genomics. As long-range scaffolding technologies, such as those based on proximity ligation–based methods like Hi-C (Putnam et al., 2016) or optical maps (Hastie et al., 2013; Yoon et al., 2016) are optimized for highly heterozygous plant genomes, we expect that reference-grade genome references will quickly become available for many grape species and cultivars of interest. This genomic information will allow us to identify core sequences that are common to all cultivars, as well as dispensable sequences comprising partially shared and non-shared genes that contribute to inter-cultivar phenotypic variation. This genomic information will also enable the identification of the genetic bases of economically important traits to accelerate the breeding of new cultivars and rootstocks.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

DC and AM conceived the article. Figure was prepared by AM and DC. AM, JL, BG, and DC wrote the first draft of the manuscript. DC revised and finalized.

### ACKNOWLEDGMENTS

The genome sequencing of Cabernet Sauvignon in the Cantu lab is supported by J. Lohr Vineyards and Wines and by E. & J. Gallo Winery. Part of this work is carried out in collaboration with UC Davis Chile and funded by the Chilean Economic Development Agency (CORFO).


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Minio, Lin, Gaut and Cantu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# A Concise Review on Multi-Omics Data Integration for Terroir Analysis in Vitis vinifera

Pastor Jullian Fabres<sup>1</sup> , Cassandra Collins<sup>2</sup> , Timothy R. Cavagnaro<sup>2</sup> and Carlos M. Rodríguez López<sup>1</sup> \*

<sup>1</sup> Environmental Epigenetics and Genetics Group, Plant Research Centre, School of Agriculture, Food and Wine, University of Adelaide, Glen Osmond, SA, Australia, <sup>2</sup> The Waite Research Institute, The School of Agriculture, Food and Wine, The University of Adelaide, Glen Osmond, SA, Australia

Vitis vinifera (grapevine) is one of the most important fruit crops, both for fresh consumption and wine and spirit production. The term terroir is frequently used in viticulture and the wine industry to relate wine sensory attributes to its geographic origin. Although, it can be cultivated in a wide range of environments, differences in growing conditions have a significant impact on fruit traits that ultimately affect wine quality. Understanding how fruit quality and yield are controlled at a molecular level in grapevine in response to environmental cues has been a major driver of research. Advances in the area of genomics, epigenomics, transcriptomics, proteomics and metabolomics, have significantly increased our knowledge on the abiotic regulation of yield and quality in many crop species, including V. vinifera. The integrated analysis of multiple 'omics' can give us the opportunity to better understand how plants modulate their response to different environments. However, 'omics' technologies provide a large amount of biological data and its interpretation is not always straightforward, especially when different 'omic' results are combined. Here we examine the current strategies used to integrate multi-omics, and how these have been used in V. vinifera. In addition, we also discuss the importance of including epigenomics data when integrating omics data as epigenetic mechanisms could play a major role as an intermediary between the environment and the genome.

#### Edited by:

Giovanni Battista Tornielli, University of Verona, Italy

#### Reviewed by:

Pablo Carbonell-Bejerano, Instituto de las Ciencias de la Vid y del Vino (ICVV), CSIC, Spain Stefania Pilati, Fondazione Edmund Mach, Italy

\*Correspondence:

Carlos M. Rodríguez López carlos.rodriguezlopez@adelaide.edu.au

#### Specialty section:

This article was submitted to Crop Science and Horticulture, a section of the journal Frontiers in Plant Science

Received: 08 February 2017 Accepted: 02 June 2017 Published: 20 June 2017

#### Citation:

Fabres PJ, Collins C, Cavagnaro TR and Rodríguez López CM (2017) A Concise Review on Multi-Omics Data Integration for Terroir Analysis in Vitis vinifera. Front. Plant Sci. 8:1065. doi: 10.3389/fpls.2017.01065 Keywords: multi-omics, environment, Vitis vinifera, data integration, epigenetics, transcriptomics, metabolomics

# INTRODUCTION

Grapevine is one of the most economically important fruit crops, and it is largely used for wine production (FAO, 2012). Most the chemical compounds that give its unique characteristics to wine are synthesized during berry development (Conde et al., 2007). However, fruit/wine composition is strongly influenced by the interactions between the plant's genome and the local growing conditions (including the vine management system), and the oenological practices of each winery (**Figure 1**), which could explain why it is so difficult to replicate a wine from a region outside that area.

Terroir is defined as the interactions between the plants, the environment and human factors (Gladstones, 2011) and it is frequently used to relate wine sensory attributes to its geographic origin (Van Leeuwen and Seguin, 2006). Although the relevance of terroir is still under debate (Anesi et al., 2015), a better understanding of how the environment affects grape berry composition can have a

**405**

significant effect on viticulture. To achieve such an understanding, it is necessary to identify the elements that drive terroir and analyze the interaction between them and the grapevine.

#### DECODING Terroir

from the copyright holder).

Terroir has been long studied, through the characterization of the different environmental factors affecting berry composition and wine quality, and climate exerts the strongest effect on berry composition (Robinson et al., 2012). Soil physicochemical properties as well have been identified as an influential factor defining the uniqueness of berry composition by vines grown in a specific climate (Cheng et al., 2014; Zerihun et al., 2015). Grapevine microbiome community may play an important role determining wine quality (Burns et al., 2015; Bokulich et al., 2016). Efforts have been made to study the grapevine microbiome landscape in relation to the vegetative growth cycle of the plant (Pinto et al., 2014), the post-harvest treatment of berries (Salvetti et al., 2016) and provenance (Bokulich et al., 2016) (For a review on microbiome analysis see Ibrahim and Kumar, 2017). Less work has been done to elucidate the molecular mechanisms involved in the plant response to terroir. A strategy to better understand the genome and environment interaction is to use 'omics' technologies. Omics refers to high throughput technologies that generate a large amount of data for each sample, allowing a deeper insight of the mechanisms regulating biological systems.

## ANALYSIS OF Terroir EFFECT ON GRAPE COMPOSITION USING TRANSCRIPTOMICS

Using transcriptomics is possible to study the grape's complete set of RNA transcripts encoded by the genome using high throughput methods (Hale et al., 2005). Dal Santo et al. (2013) performed gene expression analysis in a single Corvina clone cultivated in 11 different vineyards for three consecutive years.

Samples strongly clustered by season, known as a vintage effect, rather than common environmental conditions. However, the genes that showed more variation in expression between years were those involved in secondary metabolism, (mainly the biosynthesis of phenylpropanoids). Only when samples from a single vintage (i.e., 2008) were analyzed, it was observed that 5% of the studied annotated coding genes were differentially regulated under different growing conditions and agronomical practices. Anesi et al. (2015) complemented this study by analyzing the transcriptome and metabolome of the same cultivar. They identified metabolites that could describe a terroir signature for each vineyard. Moreover, it was possible to correlate terroir-sensitive metabolites with changes in the transcript level of genes involved in the biosynthesis of these metabolites. Similar results were obtained by Dal Santo et al. (2016) as they identified a clear correlation between gene expression and accumulation of phenylpropanoids and flavonoids in the variety Garganega grown at four different vineyards. Small RNA profiles have been analyzed to understand the interaction between genotype and environment in the varieties Sangiovese and Cabernet Sauvignon. In silico analysis suggests that microRNAs may be involved in berry development and the accumulation of secondary metabolites (Paim Pinto et al., 2016). Transcriptional analysis of berries from different regions has also shown that transcripts from the abscisic acid (ABA) biosynthesis pathway are among the most terroir sensitive genes (Sun et al., 2015). ABA is a plant hormone that regulates important steps in plant growth and development as well as play a key role in plant biotic and abiotic stress response (Cutler et al., 2010). ABA concentrations affect anthocyanin and flavonol accumulation (Koyama et al., 2010), suggesting a possible mechanism through which the environment affects grape berry composition and wine flavor and aroma.

## ANALYSIS OF Terroir EFFECT ON GRAPE COMPOSITION USING METABOLOMICS

Metabolomics is defined as the identification and quantification of metabolites using high-throughput techniques (Cevallos-Cevallos et al., 2009). This technology can screen higher numbers of products than more traditional approaches (Pereira et al., 2006; Atanassov et al., 2009; Hong, 2011), while the use of non-targeted metabolomics approaches allows the identification of un characterized metabolites (Panighel et al., 2015). Terroir can be explored by analyzing berry metabolite composition through different analytical methods (For a review in grape and wine metabolomics see Cozzolino, 2016). Son et al. (2009) identified that differences in berry metabolomes associated to environmental regional differences (radiation and rainfall) could explain the observed differences in wine composition. Similar results were obtained by Tarr et al. (2013) who distinguished the metabolic signatures of different grapevine varieties. Metabolomic analysis has also been performed to identify chemical compounds that can be associated to regional wine quality traits (Gambetta et al., 2016, 2017). Roullier-Gall et al. (2014) assessed the metabolomics profiles of two different terroirs, which were just 2 km apart, over three vintages. Although vintage had the greatest effect in the berry's metabolite composition, differences in fruit chemical composition associated to nearby terroirs could be detected when vintages were individually analyzed. This suggests that subtle geographical differences have a significant effect on grape/wine composition even when variability within vineyards can be relatively high (Mulas et al., 2011).

# MULTI-OMICS INTEGRATION

The aim of integrating multi-omic data is to reduce the gap between data generation and the ability to analyze and understand the biological mechanisms behind an organism's response to environmental cues. The objective of multi-omic data integration is to combine different types of data to construct a model that can be used to predict complex traits and phenotypes (**Figure 2**). This approach also allows the identification of biomarkers and of previously unknown relationships between the datasets (Rajasundaram and Selbig, 2016). Through the integration of environmental information with genomic, epigenomic, transcriptomic, and metabolomic data, we hypothesize that it will be possible to better understand the effect of terroir at a molecular level. The use of a multi-omic approach will also help reduce the incidence of false positives generated from single source data sets (Aho, 2013; Ritchie et al., 2015). However, integration of multi-omics data is not a trivial task, because the diversity of characteristics of the data generated from the different high throughput technologies (machine sensitivity, error rate, data structure) makes its combination challenging.

# APPROACHES TO DATASETS INTEGRATION

Analysis of large data sets from different origins has been done using two main approaches: network models (NMs) and pathway analysis (PA). Both share the basic idea of storing the data in a clear and meaningful way. NMs use concepts from mathematical graph theory, to represent biological components (e.g., genes) as nodes and their interactions (physical, genetic or functional) as their links (For a review on NM applied to plant biology see Fukushima et al., 2014). NMs are classified as homogeneous or heterogeneous depending on the number of different levels of information integrated (Gligorijevic and ´ Pržulj, 2015). Homogenous approaches integrate datasets with the same type of nodes and therefore cannot analyze the connectivity between multiple datasets simultaneously. However, complex biological questions such as the molecular regulation of fruit composition in grapevine are increasingly being addressed through the integration of multiple layers of cellular information (Wong and Matus, 2017), including but not restricted to genomics, transcriptomics, proteomics

and metabolomics, using heterogeneous methods. Bayesian networks (BNs) and Kernel-based methods (KBMs) are heterogeneous approaches commonly used for data integration (Zhang, 2009; Gligorijevic and Pržulj, 2015 ´ ). BNs are efficient detecting relationships hidden in noisy datasets but they are computationally demanding (Gligorijevic and Pržulj, 2015 ´ ) and are therefore, better suited for the interrogation of small datasets in hypothesis driven questions (Gligorijevic and Pržulj, ´ 2015) (i.e., the analysis of terroir on defined pathways such as those leading to the biosynthesis of metabolites related to fruit quality). KBMs are not as computationally demanding and so can integrate large molecular, structural and phenotypic datasets (Mizrachi et al., 2017), making them ideal for data driven terroir exploratory studies, biomarker discovery or for the reclassification of previously identified drivers of quality (Qi et al., 2008).

On the other hand, pathway analysis requires well documented biochemical pathways where omics data is combined to seek overrepresented groups (Wanichthanarak et al., 2015). For example, multiple co-inertia analysis (MCIA) can detect explanatory omic features even when they are not present in all datasets (Meng et al., 2014), which makes it attractive for the integration of terroir data from different studies. Random Forest implemented for pathway analysis (Pang et al., 2006), can be used to predict fruit/wine quality traits associated to terroir integrating multi-omic and phenotypic data as shown recently for potato (Acharjee et al., 2016).

Most of these multi-omics analysis approaches are pipelines that perform task sequences which share statistical methods (Bersanelli et al., 2016). Correlation analyses are the most common approaches performed to find relationships between the omics data. Simple correlation analyses, like Pearson or Spearman correlation, are widely used for multi-omics data integration (Rajasundaram et al., 2014; Rajasundaram and Selbig, 2016). Partial least square/projections to latent structures (PLS) and its extension, orthogonal partial least square (OPLS) (Tobias, 1995) have also been used for data integration from multiomics results. Even though their predictive power is similar, OPLS results are much easier to interpret and outliers are quickly detected. OPLS can be used as a discriminate analysis (OPLS-DA), to identify differences between the overall data properties while removing systematic variation (Kirwan et al., 2012). However, these methods provided little insight when they are used in complex biological systems (highly multicollinear systems) (Wanichthanarak et al., 2015).

Modifications to these methods have been implemented to facilitate the interpretation of the data, for example, sparse PLS (sPLS) (Chun and Keles˛, 2010) can better predict phenotypes through multi-omics data integration than previous methods (Rajasundaram et al., 2014). Orthogonal 2PLS (O2PLS), capable of dealing with unrelated systematic variation between datasets (Bouhaddani et al., 2016), has been successfully used for data integration of transcriptomics and metabolomics results from aspen under different light treatments (Bylesjö et al., 2007). Srivastava et al. (2013) used orthogonal projections to latent structures (OnPLS), an extension of O2PLS, to integrate transcriptomics, proteomics, and metabolomics data to construct a model that could identify biological relevant events in the oxidative stress response in poplar.

#### DATA INTEGRATION IN V. vinifera

In plant science, most of data integration of omics results comes from model plants; however, there is an increase in publications on multi-omics data integration in V. vinifera.

One of the first publications in multi-omics data integration in V. vinifera was the work of Zamboni et al. (2010). Integrating transcriptome, proteome and metabolome data, they identified stage specific biomarkers for berry development. Data integration was performed using two strategies, one hypothesis driven (i.e., a hypothesis was tested) and the other hypothesis free (i.e., discovery driven), in both cases principal component analysis (PCA), O2PLS and O2PLS-DA were used.

Using five different omics technologies and correlation analysis (PCA and Pearson correlation) together with biochemical pathway analysis (KEGG, PlantCyC and VitisCyC), Ghan et al. (2015) could differentiate biochemical characteristics from five different cultivars. Moreover, Anesi et al. (2015) studied the terroir effect in V. vinifera cultivar Corvina in seven different sites over a 3 years period using metabolome and transcriptome data. Using correlation analyses (PCA, PLS-DA and O2PLS-DA) they could identify a terroir signature in the berry metabolome composition for each growing site. Network analyses have been recently adopted to integrate grapevine multi omics results (Wong and Matus, 2017). For example, Palumbo et al. (2014) using network-based methods, identified "fight-club" nodes (genes with negatively correlated profiles) that may be relevant for the control of berry transition between development and ripening.

There are also online resources available that can help analyze omic data from V. vinifera. For example, VitisNet (Grimplet et al., 2009, 2012) offers manually annotated molecular networks (16,000 genes and 247 networks) where omics data can be loaded to visualize changes in the transcriptome, proteome and metabolome for a given experiment. VTCdb (Wong et al., 2013) is a gene co-expression database for V. vinifera that allows exploring transcription regulation. With more than 29,000 genes (95% of the predicted grapevine transcriptome) to query co-expression networks, VTCdb offers the possibility to analyze the transcriptional network of grapevine development, metabolism and stress response. VitisCyc (Naithani et al., 2014) is a grapevine metabolic pathway database that also allows omics data to be uploaded (transcriptome, proteome and metabolome) and to analyze changes in metabolic networks in each experiment. VESPUCCI (Moretto et al., 2016) is a manually annotated gene expression compendium exploratory tool that can be used to investigate grapevine's gene expression patterns.

#### PHENOTYPIC PLASTICITY THROUGH EPIGENETIC MODIFICATIONS

Epigenetics is the study of heritable phenotypes that occur through modifications that alter DNA activity without modifying its basic nucleotide structure (Feil and Fraga, 2012). Many epigenetic mechanisms, acting in an interactive and redundant fashion (Grant-Downton and Dickinson, 2005; Berger et al., 2009), have been described to date, with DNA methylation probably being the best-studied of all (Rapp and Wendel, 2005). DNA methylation affects chromatin condensation in a rapid and reversible manner (Grativol et al., 2012). In turn, the regional level of chromatin condensation affects the transcriptional state of nearby genomic features such as genes and transposable elements (Zhang et al., 2006). Global changes in DNA methylation associated to local environments can be analyze using a myriad of methods (Kurdyukov and Bullock, 2016). Bisulfite modification of genomic DNA combined with whole genome sequencing (BS-Seq) is the gold standard for methylation analysis because it can assess an entire methylome with single base resolution (Krueger et al., 2012). However, due to their lower cost, other approaches such as next generation sequencing following the capture of the methylated fraction of the genome or its fragmentation using methylation sensitive restriction enzymes (Bock et al., 2010; Li et al., 2010; Kitimu et al., 2015) are better suited to study large number of samples. Both generate quantitative and qualitative information of the methylation status of a reduced but significant representation of the total genome.

Environmental signals are one of the elements that can have a major effect in modifying the DNA methylation patterns leading to gene expression changes that ultimately affect the plant phenotype (Feil and Fraga, 2012). The idea that the environment could modify the epigenetic status, and these modifications passed to the offspring (Tricker et al., 2013) or maintained as epigenetic memory on long lived organisms (Latzel et al., 2016), has attracted attention from scientists studying mechanisms involved in adaptation to local environments (Consuegra and Rodríguez López, 2016) and how these could be used to enhance crop performance (Rodríguez López and Wilkinson, 2015). There are many reported examples of how the environment affects the epigenome in natural environments and how epigenetic variations in plant populations could help to overcome the lack of genetic diversity (Fonseca Lira-Medeiros et al., 2010; Verhoeven et al., 2010).

One of the most well-known examples in which the environment affects the phenotype through epigenetic modifications is vernalization (Feil and Fraga, 2012). Through this process, plants in temperate regions mitigate the deleterious effects of low winter temperatures on flower and fruit development by breaking dormancy only after having been exposed to a cold period (Kumar et al., 2016). Unusual environmental conditions during dormancy such as high winter temperatures have been shown to exert a negative effect on fruit quality and yield on perennial crops requiring a vernalization period (Sugiura et al., 2012). Recent work in apple shows how methylation and expression levels of key genes involved in flowering and fruit set are modified by the level of chill received during bud dormancy (Kumar et al., 2016), indicating that the environmentally induced changes observed in fruit quality could be regulated by DNA methylation.

Together these studies suggest that the environment can have a long lasting phenotypic effect in plants through epigenetic changes without the need for genetic variation, and that epigenetic mechanisms could be working as intermediaries between environmental variation and the plant genome, and in this way, potentially contributing to plant phenotypic plasticity. Moreover, this mechanism could give plant populations a way of adapting to the local growing conditions (Platt et al., 2015; González et al., 2016). However, to our knowledge, almost

all epigenetic studies done in V. vinifera have focused on the identification of commercial clones (Imazio et al., 2002; Schellenbaum et al., 2008; Ocaña et al., 2013) and on the assessment of in vitro culture on the epigenome (Baránek et al., 2015), there is, therefore, a lack of information of how the environment affects a grapevine's epigenome and to what extent this interaction affects fruit quality. Until now, there are no studies looking at the epigenome to understand the control of gene expression in V. vinifera and how environmental signals can change the regulation of metabolic pathways through epigenetic modifications. In our view, the inclusion of epigenomic data on the analysis of the terroir effect will not only increase the resolution of analysis but will also help us to understand the regulatory mechanisms behind the observed differences.

#### CONCLUSION

There is no doubt that the elements affecting grapevine growth and fruit composition are complex and multifarious. While the concept of terroir is widely discussed, the underlying mechanisms remain somewhat enigmatic. However, with the recent parallel development of omics technologies and of statistical approaches for their integration, we are reaching a point where it may be possible to overcome this challenge. The geographic delimitation of a terroir is the first challenge to overcome before its molecular characterization. This delimitation could be achieved 1. Empirically, based in the number of significantly different environmental subregions present in the study or/and 2. based on the traditionally defined wine regions. Moreover, the masking effect that environmental inter-annual variations can have over single year measurements demands the incorporation of data from multiple seasons to be able to determine terroirs with enough confidence. Ideally such seasons should be, from a weather perspective, variable within the range characteristic for the region of study to be able to capture its "normal terroir."

Understanding how the genome, environment and viticulture practices interact to affect fruit quality will allow us the

#### REFERENCES


opportunity to implement agricultural practices aimed to obtain the desired fruit characteristics for every climate/cultivar combination (Jones and Davis, 2000), leading to more efficient use of resources and better management of vineyards. In addition, grape growers can maximize the terroir effect on the grapevine to highlight the uniqueness of their vineyards ultimately increasing their industrial competitiveness. We propose that the integration of multi-omic and environmental datasets will contribute to a better understanding of the drivers of the terroir effect in grapevine. Moreover, multiple dataset data integration will increase our understanding of the molecular mechanisms involved in the regulation of multifactorial genome by environment interactions. Finally, it is increasingly recognized that plants are involved in complex interactions their soil and epiphytic microbiomes, which can affect their phenotype (Mueller and Sachs, 2015). The 'omics' era gives us the ability to explore the nature and consequences of biotic/abiotic interactions and so, a future challenge will be to bring the concept of the holobiont (the plant host plus its microbiomes) into the analysis of terroir and its effect on grapevine growth and fruit composition.

#### AUTHOR CONTRIBUTIONS

PF, TC, CC, and CR designed and wrote the manuscript.

#### ACKNOWLEDGMENTS

PF was supported by Australian Grape and Wine Authority (AGWA) and University of Adelaide Graduate Research Scholarships. CR is supported by a University of Adelaide Research Fellowship. This study was funded through a Pilot Program in Genomic applications in Agriculture and Environment Sectors jointly supported by the University of Adelaide and the Australian Genome Research Facility Ltd.




**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Fabres, Collins, Cavagnaro and Rodríguez López. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Plant Stress Responses and Phenotypic Plasticity in the Epigenomics Era: Perspectives on the Grapevine Scenario, a Model for Perennial Crop Plants

Ana M. Fortes<sup>1</sup> \* and Philippe Gallusci<sup>2</sup> \*

<sup>1</sup> Faculdade de Ciências, Instituto de Biossistemas e Ciências Integrativas, Universidade de Lisboa, Lisboa, Portugal, <sup>2</sup> UMR EGFV, Université de Bordeaux, Institut national de la recherche agronomique, Institut des Sciences de la Vigne et du Vin, Villenave-d'Ornon, France

#### Edited by:

José Tomás Matus, Centre for Research in Agricultural Genomics, Spain

#### Reviewed by:

Carlos Marcelino Rodriguez Lopez, University of Adelaide, Australia Patricio Arce, Pontifical Catholic University of Chile, Chile

#### \*Correspondence:

Ana M. Fortes amfortes@fc.ul.pt Philippe Gallusci philippe.gallusci@inra.fr

#### Specialty section:

This article was submitted to Crop Science and Horticulture, a section of the journal Frontiers in Plant Science

Received: 02 November 2016 Accepted: 16 January 2017 Published: 06 February 2017

#### Citation:

Fortes AM and Gallusci P (2017) Plant Stress Responses and Phenotypic Plasticity in the Epigenomics Era: Perspectives on the Grapevine Scenario, a Model for Perennial Crop Plants. Front. Plant Sci. 8:82. doi: 10.3389/fpls.2017.00082 Epigenetic marks include Histone Post-Translational Modifications and DNA methylation which are known to participate in the programming of gene expression in plants and animals. These epigenetic marks may be subjected to dynamic changes in response to endogenous and/or external stimuli and can have an impact on phenotypic plasticity. Studying how plant genomes can be epigenetically shaped under stressed conditions has become an essential issue in order to better understand the molecular mechanisms underlying plant stress responses and enabling epigenetic in addition to genetic factors to be considered when breeding crop plants. In this perspective, we discuss the contribution of epigenetic mechanisms to our understanding of plant responses to biotic and abiotic stresses. This regulation of gene expression in response to environment raises important biological questions for perennial species such as grapevine which is asexually propagated and grown worldwide in contrasting terroirs and environmental conditions. However, most species used for epigenomic studies are annual herbaceous plants, and epigenome dynamics has been poorly investigated in perennial woody plants, including grapevine. In this context, we propose grape as an essential model for epigenetic and epigenomic studies in perennial woody plants of agricultural importance.

Keywords: DNA methylation, epigenomics, grape, Histone Post-Translational Modifications, small RNAs, Vitis vinifera

#### INTRODUCTION

Epigenetic mechanisms regulate chromatin structure, gene expression, transposon mobility and DNA recombination (He et al., 2011; Pikaard and Scheid, 2015). They generally refer to modifications of gene expression that can be inherited through mitosis or meiosis yet without changes in the underlying DNA sequences (Eichten et al., 2014) and also include chromatin modifications that may lead to stable alteration of the transcriptional programming of non-dividing cells even after removal of the triggering signals (Avramova, 2015).

Epigenetic regulation is mediated by a complex interplay among different molecular actors. These include the DNA methylation/demethylation machinery, enzymes mediating histone post-translational modifications (PTMs), the remodeling of chromatin organization and specific

classes of small RNAs and long non-coding RNAs (Lauria and Rossi, 2011; Pikaard and Scheid, 2015; Gallusci et al., 2016). Briefly, in plants 5 methyl-cytosine (m5C) is found in all sequence context, including the CG and CHG (H = A, T, or C) symmetrical motives and the non-symmetrical CHH motif (reviewed in Gehring, 2013). DNA methylation is maintained in a post-replicative way by three classes of DNA methyltransferases: DNA METHYLTRANSFERASE 1 (MET1) and CHROMOMETHYLASE 3 (CMT3) for CG and CHG contexts, respectively, and by the DOMAIN REARRANGED METHYLTRANSFERASE 2 (DRM2), which requires an siRNA guide and reestablishment after each cycle of DNA replication or by CMT2 for the asymmetric CHH context (Du et al., 2012; Matzke and Mosher, 2014). Finally, DNA methylation can be lost after replication when maintenance of DNA methylation is not functional or actively reversed by DNA Glycosylase-Lyases (Piccolo and Fisher, 2014).

Histone PTMs are also essential epigenetic signals that can occur at the N-terminal tail of core histones (H2A, H2B, H3, H4) through acetylation, methylation, phosphorylation and ubiquitination (Berr et al., 2011). Histone acetylation and methylation at lysine residues are established by histone acetyltransferases (HATs) and histone lysine methyltransferases (HKMTs), respectively, which are encoded by complex multigenic families. These epigenetic marks can be removed by histone deacetylases (HDACs) and histone demethylases (HDMs), respectively (Berr et al., 2011; He et al., 2011; Pikaard and Scheid, 2015).

The recent development of epigenome profiling has boosted our understanding of the dynamics and function of epigenetic marks in plants. Several approaches have been recently developed (Schmitz and Zhang, 2011; Lee and Kim, 2014<sup>1</sup> ). So far, histone PTM analysis relies on Chromatin Immunoprecipitation (ChIP) using specific antibodies followed by hybridization to tilling arrays (ChIP- chip, Makarevitch et al., 2015) or by high throughput sequencing (ChIP-Seq, Wang et al., 2009). DNA methylation landscape can be studied by making use of methyl sensitive restriction enzyme to enrich DNA in methylated or un-methylated sequences that are subsequently hybridized to tilling arrays or sequenced (Kim et al., 2014). Alternatively, methylated regions can be selected using m5C specific antibodies (MeDIP), and analyzed with tilling arrays (MeDip-ChIP) or by Next Generation Sequencing (Medip Seq). Both approaches were used for methylome analysis for example in Arabidopsis, or poplar (Zhang et al., 2006; Zilberman et al., 2006; Kim et al., 2014). In particular, Medip-Seq was used to analyze the changes in methylation patterns during in vitro culture of cassava (Kitimu et al., 2015). But the golden standard for methylome analysis is the combination of bisulfite conversion of DNA to high throughput sequencing that allows analyzing the methylation landscape at a single base resolution (Whole Genome Bisulfite sequencing: WGBS). The methylomes of Arabidopsis (Cokus et al., 2008; Lister et al., 2008; Stroud et al., 2013), rice (Li et al., 2012; Garg et al., 2015), maize (Eichten et al., 2013), tomato (Zhong et al., 2013), Brassica (Chalhoub et al., 2014) and many others (Niederhuth et al., 2016) have now been described using this approach.

In this perspective, we will firstly focus on the analysis of the genome wide distribution of epigenetic marks in plants under stresses. However, most species used for epigenomic studies are annual herbaceous plants and little is known about epigenomes in perennial woody plants. Indeed, omics' approaches have been initiated in grape to understand environmental effects on plant and fruit development (Fortes et al., 2011; Agudelo-Romero et al., 2015). In addition, a few studies have indicated that epigenetic mechanisms might be involved in various aspects of grape development (Aquea et al., 2011). However, knowledge of grape epigenomes and of their variation has remained very limited until now (Niederhuth et al., 2016). Yet, grapevine presents several features that make it a relevant model for the study of epigenetic mechanisms due to the fact that is a perennial woody plant and the fruit maturation is subjected to nonclimacteric molecular and hormonal regulation (Fortes et al., 2015). Grapevine varieties are preserved in their distinct genetic backgrounds through clonal propagation. However, phenotypic diversity exists within clones (Pelsy, 2010) that is unlikely to be solely driven by differences in DNA sequence. These facts contribute to the relevance of grape as a model for epigenetic and epigenomic studies in perennial woody plants of agricultural importance.

#### EPIGENETIC REPROGRAMMING DURING ABIOTIC STRESS RESPONSES

Recent studies have shown the differential regulation of genes encoding epigenetic regulators (Fang et al., 2014; Li et al., 2014; Su et al., 2015) as well as local chromatin and DNA methylation changes in response to a variety of abiotic stresses including cold, salinity, drought, osmolality, or mineral nutrition, thereby highlighting the relevance of epigenetic regulations in these contexts (Chen et al., 2010; Luo et al., 2012; González et al., 2013; Bocchini et al., 2015; Kim et al., 2015; Liu et al., 2015). Consistent with these results, genome wide analyses of histone PTMs and DNA methylation distribution have revealed global epigenomic reprogramming in plants under abiotic stresses. In a recent study, trimethylation at lysine 4 on histone 3 (H3K4me3), a mark normally associated with gene expression, was analyzed in Arabidopsis plants under drought stress using ChIP-seq and showed to be highly dynamic and positively correlated with the transcription level of drought induced genes in response to stress (Dijk et al., 2010). Similar results were found in rice (Zong et al., 2013) and in moss (Widiez et al., 2014). Osmotic stress also causes an increase in phosphorylated histone H3 threonine 3 (H3T3ph) located at pericentromeric regions where it is thought to help maintaining the heterochromatin structure (Wang et al., 2015). Interestingly, H3T3ph is also present in active genes where it seemed to antagonize H3K4me3, suggesting that H3T3ph may have a repressive function on gene expression during osmotic stress (Wang et al., 2015) a role also suggested for histone deacetylase HDA9 (Zheng et al., 2016). In addition, priming effects in Arabidopsis were

<sup>1</sup>https://www.plant-epigenome.org/

shown to be partly mediated by remodeling of the epigenomic landscape, and involves the repressive mark H3K27me3 (Sani et al., 2013).

Recently, a specialized histone H1 variant was shown to be required for a substantial part of DNA methylation associated with environmental stress in Arabidopsis (Rutowicz et al., 2015) and two DEAD-box RNA helicases were suggested to be involved in epigenetic silencing of gene expression leading to suppression of Arabidopsis stress response (Khan et al., 2014).

In addition, DNA methylation is also critical for the responses of plant to abiotic stresses. This was initially shown by the demonstration that Arabidopsis mutants deficient in various steps of the RdDM pathway or in CHG maintenance methylation are affected in their capacity to modulate the stomatal index under low relative humidity (Tricker et al., 2012), present an hypersensitivity to heat exposure (Popova et al., 2013) or an enhanced sensitivity to phosphate starvation (Yong-Villalobos et al., 2015). These results are consistent with an important function of the DNA methylation dynamics in the regulation of abiotic stress–responsive genes. Indeed drought stress, but also nutrient deprivation cause extensive remodeling of DNA methylation patterns in Arabidopsis (Colaneri and Jones, 2013; Yong-Villalobos et al., 2015; Wibowo et al., 2016), barley (Chwialkowska et al., 2016) or Populus (Liang et al., 2014). In this latter case, modulation of DNA methylation at repetitive elements appeared essential for the control of adjacent gene expression (Liang et al., 2014) a function also suggested in maize where TEs could be used as local enhancers for stress responsive genes (Makarevitch et al., 2015). Similarly Pi deficiency in rice modulates DNA methylation at TEs located close to genes highly induced under this stress (Secco et al., 2015). In this case, however, TEs were hyper-methylated an event that occurred after gene induction most likely to prevent potentially deleterious activity of TEs located in the vicinity of highly induced stress responsive genes.

As a conclusion, the results discussed above are consistent with the idea that abiotic stresses cause significant reprogramming of chromatin not only related to gene expression, but also to the control of chromosome organization. In addition, evidence of transgenerational inheritance of plant responses to stress has been provided (Tricker et al., 2013; Migicovsky et al., 2014); although this process appears limited to and mainly mediated by the female gamete (Wibowo et al., 2016).

#### EPIGENETIC REPROGRAMMING DURING PLANT BIOTIC STRESS RESPONSES

Regarding histone Post-Translational Modifications and DNA methylation occurring upon biotic stress there is lesser information available than for abiotic stress. However, recent findings indicate that chromatin modifications contribute to plant immunity against both necrotrophic and biotrophic pathogens (reviewed by Ding and Wang, 2015). In fact, the expression of R genes which are central regulators of plant immunity was shown to be regulated by Arabidopsis E3 ubiquitin ligase genes HISTONE MONOUBIQUITINATION1 (HUB1) and HUB2 (Zou et al., 2014). Histone monoubiquitination at the R gene locus had an impact on immune responses. The loss of- function mutant bon1-1 has enhanced disease resistance to the virulent pathogen Pst DC3000 and both HUB1 and HUB2 mediate its autoimmune responses. In another case, HDA19, an Arabidopsis histone deacetylase, was shown to play a negative role in basal defense mediated by the SA-dependent signaling pathway. Loss of HDA19 causes increased expression of SA biosynthetic genes and defense genes and promotes resistance to the virulent Pst DC3000 (Choi et al., 2012). Dimethylated or trimethylated histone H3 Lys 27 (H3K27me2/3) marks silent or repressed genes involved in stress responses in plants. Li et al. (2013) showed that the rice Jumonji C protein gene JMJ705 encodes a histone Lys demethylase that specifically reverses this mark. An increase in JMJ705 expression in transgenic plants removes H3K27me3 from defense-related genes, induces their expression with involvement of jasmonic acid, and enhances plant resistance to biotic stress. Interestingly, Soyer et al. (2014) showed that chromatin-based transcriptional regulation can also act on effector gene expression in fungi during plant infection. Pathogen infection has been also reported to change histone modifications in some defense response genes (De-La-Pena et al., ´ 2012).

The profiling of the DNA methylomes of plants exposed to bacterial pathogen, avirulent bacteria, or salicylic acid revealed numerous stress-induced differentially methylated regions (DMRs) often coupled to differential gene expression (Dowen et al., 2012). Mutant plants globally defective in maintenance of CG methylation (met1-3) or non-CG methylation (ddc, drm1-2 drm2-2 cmt3-11) were markedly resistant to bacterial colonization.

DNA demethylation likely primes transposable elements as well as defense gene induction through the concomitant activation of their transactivators and/or the interference with other chromatin marks (Yu et al., 2013). Some immuneresponse genes, containing repeats in their promoter regions, are negatively regulated by DNA methylation. These defense gene loci may lose DNA methylation so that they are more easily activated at the transcriptional level (Yu et al., 2013). This is corroborated by the study of Le et al. (2014); the DNA methylases ROS1, DML2, and DML3 were shown to play a role in fungal disease resistance in Arabidopsis since a triple mutant rdd (ros1 dml2 dml3), presents down-regulation of stress response genes and increased susceptibility to a fungal pathogen. Furthermore, these authors showed that DNA demethylases target promoter transposable elements in stress responsive genes to positively regulate them.

#### NATURAL AND INDUCED EPIGENOMIC VARIATION, PHENOTYPIC PLASTICITY AND BREEDING

Natural epigenomic variation occurs during species evolution (Hirsch et al., 2013) and together with genetic variation is

likely involved in the phenotypic diversity and plasticity of plants. Epigenetic variation is sensitive to environmental inputs; epialleles induced by the environment or experimentally may be formed at a higher rate than alleles generated from genetic variation and may also be inherited leading to better adaptation to the environment (**Figure 1**; Hirsch et al., 2013).

Experimentally induced epialleles have been produced in Arabidopsis by generating Epigenetic Recombinant Inbred Line (EpiRILs) populations derived from decrease in DNA methylation 1-2 (ddm1-2) or the met1 parents (Johannes et al., 2009; Reinders et al., 2009). EpiRILs were subsequently used to identify epiQTL corresponding to DMRs that determine two complex traits, flowering time and primary root length (Cortijo et al., 2014). Interestingly, these EpiRILs present variation in growth capacity (Hu et al., 2015) and are more sensitive to salinity stress than the Col0 parent line suggesting that ddm1 derived epigenotypes limit the ability to adapt to this stress (Kooke et al., 2015). As an alternative approach, a stochastically hypomethylated population was generated by selfing Brassica rapa plants previously treated with the demethylating agent 5-Azacytidine (Amoah et al., 2012). This population was used for forward screening of agronomic traits such as flowering time, seed protein content and fatty acid components. These results suggest that a portion of QTLs that have been used by breeders so far may be due to epigenetic, rather than genetic variation (Springer, 2013).

DNA methylation may also have an important role in the long term adaptation of plants (**Figure 1**; Garg et al., 2015). Two rice cultivars with contrasting sensitivity to drought stress and salinity showed clearly different methylation landscapes; part of the DMRs between cultivars were associated with genes involved in stress responses (Garg et al., 2015).

Indeed variation in methylation patterns have been also observed in natural populations and might be associated with specific environmental traits. In a recent study, Dubin et al. (2015) showed by analyzing Arabidopsis accessions from Northern and Southern Sweden that CHH methylation at transposons increases with temperature and this was associated with major genetic variants at the CMT2 locus. In the same study, Gene Body Methylation which was not modified by temperature was shown to be correlated with the latitude of origin; Southern accessions being less methylated than Northern one. This was associated with a lower expression of the targeted genes in Southern accessions consistent with local adaptation of the accessions.

Epialleles impacting plant traits have now been identified in many plants (Rodríguez López and Wilkinson, 2015) since the initial characterization of the cycloidea and Cnr epimutations in snapdragon and tomato, respectively (Cubas et al., 1999; Manning et al., 2006; Poole et al., 2006). For example, Vitamin E in tomato is determined by epigenetic variations linked to a SINE retrotransposon located in the promoter region of a gene involved in the vitamin synthesis. This work showed that naturally occurring epialleles may be responsible for regulation of nutritionally important metabolic QTLs and determination


TABLE 1 | Genes involved in epigenetic mechanisms differentially modulated in Trincadeira grapes infected with the fungus Botrytis cinerea at green hard stage (EL33) and véraison stage (EL35).

Details on microarray analysis available in Agudelo-Romero et al. (2015).

of agronomic traits (Quadrana et al., 2014). In another study, the complex trait of Energy use efficiency was shown to possess an epigenetic component that is stably inherited, allowing the creation of distinct isogenic sublines that can be used in breeding (Hauben et al., 2009). Thus, induced or natural epigenetic diversity may represent an unexplored resource of phenotypical variations that could be used in plant breeding programs, as recently discussed in Rodríguez López and Wilkinson (2015).

## GRAPEVINE EPIGENOMICS AND EPIGENETICS: A MODEL PLANT FOR PERENNIAL CROP PLANT

Studies on Arabidopsis revealed functional aspects of epigenetic regulation of gene expression but present limitations since Arabidopsis has only 5% of methylated cytosine in the genome whereas many crops contain more than 20% (Lee and Kim, 2014). In fact, mutations in epigenetic regulators seem to have a higher impact in crops than in Arabidopsis (Mirouze and Vitte, 2014; Gallusci et al., 2016). In addition, Arabidopsis contains very few transposable elements comparing to crops (reviewed by Lee and Kim, 2014). Polymorphisms in transposon insertions and repeats can originate natural epigenetic variation. Furthermore, while the distribution of the genes along the chromosomes of Arabidopsis is fairly homogeneous, this situation may differ in crops. For example, Vitis vinifera genome is characterized by alternation of large regions with high and low gene density (Jaillon et al., 2007).

Several studies have already emerged in crops, in particular, recent analyzes carried out in tomato fruits (Zhong et al., 2013; Liu et al., 2015) constitute a relevant background for studies in grape. It is not yet known whether the epigenetic control of ripening is similar in all fleshy fruits or is limited to the tomato and related wild species (reviewed in Gallusci et al., 2016). Nevertheless, the expression patterns of several genes involved in DNA methylation and histones modifications indicate that epigenetic factors are involved in the onset of véraison in grape and a global decrease in DNA methylation may eventually occur during grape ripening (Fortes et al., 2011) as reported for tomato (Zhong et al., 2013; Liu et al., 2015). In this context, the lack of available mutants in grape constitutes a limitation comparing to tomato. However, studies addressing the methylation status of promoters of genes involved in easily identified traits can shed light on epigenetic regulation of gene expression in grape (**Figure 1**).

Chemical treatments that affect DNA methylation patterns could also be utilized to generate epimutations (Amoah et al., 2012) though they may not be as stable as genetic mutants. Several examples of epimutations in crops are mentioned in the review by Zhang and Hsieh (2013). Epimutagenesis may allow the opportunity to explore allelic variation and novel combinations of alleles without relying upon recombination (Springer, 2013).

Analysis of the distribution of epi-marks and DNA methylation in grape in relation with gene expression profiles and fruit quality traits would likely identify epialleles that could be used as important new targets for plant breeding (**Figure 1**). DNA methylation may generate multiple epialleles with various expression levels, thereby leading to continuous quantitative variation of a trait (Zeng and Cheng, 2014). On the other hand, Kitimu et al. (2015) identified candidate epimarks that distinguish between field cuttings and meristem culture cassava samples. Specific methylation signatures may be used in the future for the diagnosis of somaclonal variants and clonal stocks in grapevine.

Grape combines several specific features that could make it an appealing model to study epigenetic regulations in woody perennial plants. It is used as one of the main models for nonclimacteric fruits and also flower development is programmed 1 year in advance; the impact of environmental conditions on flower and subsequently fruit development seems to be in part determined by the environmental conditions the year before. Grape also has specific requirements such as grafting, and clonal propagation. In this context, epigenetic variability could add to the genetic diversity of grape to shape the phenotypic variations observed in this plant. Consistent with this view, clonal diversity within V. vinifera varieties has been distinguished using the methylation-sensitive amplified polymorphism technique, highlighting the usefulness of using epigenetic markers in intravarietal diversity studies (Ocaña et al., 2013). Grafting could also impact the epigenetic state of both rootstocks and shoots (scions), **Figure 1**. Recently, Lewsey et al. (2015) showed that mobile sRNAs regulate the DNA methylation landscape genome wide, and may be an important mechanism of genome defense in crops. They showed that site-specific transmission of epiallelic states from one accession to another can be achieved by grafting and by de novo methylation of unmethylated DNA, consistent with the idea that some effects of grafting are due to the movement of small RNAs. In grapevine, grafting with rootstocks induced the up-regulation of genes associated with DNA methylation and chromatin modification in the shoot apical meristem (Cookson and Ollat, 2013). Clarifying these mechanisms may open doors to innovative applications to enhance grapevine tolerance to stresses and grape quality.

In line with these ideas, the recent analysis of the transcriptomic changes associated with grape infection with the necrotrophic pathogen Botrytis cinerea suggested that epigenetic mechanisms are involved in the reprogramming of fruit defense (Agudelo-Romero et al., 2015). Genes coding for histones, DNA (cytosine-5)-methyltransferase, helicases, DICER and ARGONAUTE proteins were modulated during the infection, whereas those associated with TEs mobility were down-regulated (**Table 1**).

Base-resolution methylomes and high-throughput sRNA profilings are already available in more than 34 species (Niederhuth et al., 2016) including V. vinifera. Comparing the epigenomes of wild and cultivated Vitis species with and without biotic and non-biotic stresses will bring insights on the epigenetic basis of grapevine resistance to adverse conditions with potential impact in breeding strategies. Moreover, epigenetic marks may participate in the priming mechanisms to better withstand biotic and abiotic stresses (Crisp et al., 2016), another topic that deserves attention in order to moderate stress susceptibility and increase climate change resilience in grapevine. Interestingly, these epimarks can also be used in the future for distinguishing agronomic practices and terroir certification of wines.

Previously, transgenerational systemic acquired resistance, was demonstrated to be a prominent defense mechanism toward downy mildew pathogen and involves DNA methylation (Luna and Ton, 2012). In grapevine, a further layer of complexity can be added since memory in perennial plants is affected every year in meristems committed to flowering. Furthermore, the reason why epigenetic regulation in response to stress can be transient or transgenerational are not clear (Tricker, 2015). It is also not known the contribution of pathogen-responsive siRNAs in transgenerational immune priming and how they drive the selection of new phenotypes especially in perennial plants.

A deeper understanding of the molecular mechanisms involving tissue-specific epigenetic changes underlying genotype × environment interactions may be beneficial for long-term improvement of grapevine performance in less predictable climates with new sources of diseases.

In a near future, epigenetic marker-assisted breeding strategies will be applied to select for agronomical desirable epigenetic quantitative traits (**Figure 1**). Crop improvement via locusspecific epigenetic manipulation has become increasingly feasible with TALE- or CRISPR-based genome editing technologies (Mendenhall et al., 2013; Zhang and Hsieh, 2013). Such technologies can be expected to play an important role in grapevine improvement once transgenesis' protocols are optimized for different cultivars.

# AUTHOR CONTRIBUTIONS

AF and PG designed the perspective and wrote the manuscript.

# FUNDING

Funding was provided by the Portuguese Foundation for Science and Technology (SFRH/BPD/100928/2014, FCT Investigator FCT050, PEst-OE/BIA/UI4046/2014).

# ACKNOWLEDGMENTS

The authors would like to thank Prof. Graham Seymour from the University of Nottingham for carefully reading the manuscript and the COST (European Cooperation in Science and Technology) Action FA1106 "Quality fruit."

# REFERENCES

fpls-08-00082 February 2, 2017 Time: 16:19 # 7



DNA methylation and suppression of genes controlling stomatal development. J. Exp. Bot. 63, 3799–3814. doi: 10.1093/jxb/ers076


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Fortes and Gallusci. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.