# METABOLOMICS IN CROP RESEARCH – CURRENT AND EMERGING METHODOLOGIES

EDITED BY : Marta Sousa Silva, Andreia Figueiredo, Ute Roessner and Carlos Cordeiro 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 978-2-88963-158-2 DOI 10.3389/978-2-88963-158-2

#### 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

# METABOLOMICS IN CROP RESEARCH – CURRENT AND EMERGING METHODOLOGIES

Topic Editors:

Marta Sousa Silva, Faculdade de Ciências da Universidade de Lisboa, Portugal Andreia Figueiredo, Faculdade de Ciências da Universidade de Lisboa, Portugal Ute Roessner, The University of Melbourne, Australia Carlos Cordeiro, Faculdade de Ciências da Universidade de Lisboa, Portugal

Image created by Marta Sousa Silva

The plant metabolome is highly complex, being composed of over 200,000 metabolites. The characterization of these small molecules has been crucial to study plant growth and development as well as their response to environmental changes. The potential of metabolomics in plant research, particularly if applied to crop plants, is also extremely valuable in the discovery of biomarkers and in the improvement of crop yield and quality. This Frontiers Research Topic addresses many applications of metabolomics to crop research, based on different analytical platforms, including mass spectrometry, and nuclear magnetic resonance. It comprises 13 articles from 109 authors that show the importance and the contribution of metabolomics in the analysis of crop's traceability and genetic variation, in the study of fruit development, and in the understanding of the plant's response to the environment and to different biotic and abiotic stresses.

Citation: Sousa Silva, M., Figueiredo, A., Roessner, U., Cordeiro, C., eds. (2019). Metabolomics in Crop Research – Current and Emerging Methodologies. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-158-2

# Table of Contents

#### *06 Editorial: Metabolomics in Crop Research—Current and Emerging Methodologies*

Marta Sousa Silva, Carlos Cordeiro, Ute Roessner and Andreia Figueiredo

#### METABOLOMICS IN THE STUDY OF ENVIRONMENTAL EFFECTS ON CROPS


Sonia Demasi, Matteo Caser, Michele Lonati, Pier L. Cioni, Luisa Pistelli, Basma Najar and Valentina Scariot


Yoshio Tamura, Tetsuya Mori, Ryo Nakabayashi, Makoto Kobayashi, Kazuki Saito, Seiichi Okazaki, Ning Wang and Miyako Kusano

*59 Effect of Solid Biological Waste Compost on the Metabolite Profile of*  Brassica Rapa *ssp.* Chinensis

Susanne Neugart, Melanie Wiesner-Reinhold, Katja Frede, Elisabeth Jander, Thomas Homann, Harshadrai M. Rawel, Monika Schreiner and Susanne Baldermann

#### METABOLOMICS IN THE STUDY OF FRUIT DEVELOPMENT

*72 Dynamic Labeling Reveals Temporal Changes in Carbon Re-Allocation Within the Central Metabolism of Developing Apple Fruit* Wasiye F. Beshir, Victor B. M. Mbong, Maarten L. A. T. M. Hertog, Annemie H. Geeraerd, Wim Van den Ende and Bart M. Nicolaï

*88 A Systems Analysis With "Simplified Source-Sink Model" Reveals Metabolic Reprogramming in a Pair of Source-to-Sink Organs During Early Fruit Development in Tomato by LED Light Treatments* Atsushi Fukushima, Shoko Hikosaka, Makoto Kobayashi, Tomoko Nishizawa, Kazuki Saito, Eiji Goto and Miyako Kusano

#### METABOLOMICS IN THE STUDY OF BIOTIC STRESSES

*105 A Non-targeted Metabolomics Approach Unravels the VOCs Associated With the Tomato Immune Response Against* Pseudomonas syringae

María Pilar López-Gresa, Purificación Lisón, Laura Campos, Ismael Rodrigo, José Luis Rambla, Antonio Granell, Vicente Conejero and José María Bellés

*120 Untargeted Metabolomics Approach Reveals Differences in Host Plant Chemistry Before and After Infestation With Different Pea Aphid Host Races*

Carlos Sanchez-Arcos, Marco Kai, Aleš Svatoš, Jonathan Gershenzon and Grit Kunert

*133 Polyphenolic Composition of Lentil Roots in Response to Infection by*  Aphanomyces euteiches

Navid Bazghaleh, Pratibha Prashar, Randy W. Purves and Albert Vandenberg

*145 Phenolic Profile and Susceptibility to* Fusarium *Infection of Pigmented Maize Cultivars*

Jamila Bernardi, Lorenzo Stagnati, Luigi Lucini, Gabriele Rocchetti, Alessandra Lanubile, Carolina Cortellini, Giovanni De Poli, Matteo Busconi and Adriano Marocco

*158 Identification of Biomarkers for Defense Response to* Plasmopara viticola *in a Resistant Grape Variety*

Giulia Chitarrini, Evelyn Soini, Samantha Riccadonna, Pietro Franceschi, Luca Zulini, Domenico Masuero, Antonella Vecchione, Marco Stefanini, Gabriele Di Gaspero, Fulvio Mattivi and Urska Vrhovsek

*169 Identification of Lipid Markers of* Plasmopara viticola *Infection in Grapevine Using a Non-targeted Metabolomic Approach* Lise Negrel, David Halter, Sabine Wiedemann-Merdinoglu, Camille Rustenholz,

Didier Merdinoglu, Philippe Hugueney and Raymonde Baltenweck

# Editorial: Metabolomics in Crop Research—Current and Emerging Methodologies

Marta Sousa Silva1,2 \*, Carlos Cordeiro1,2, Ute Roessner <sup>3</sup> and Andreia Figueiredo<sup>4</sup> \*

<sup>1</sup> Laboratório de FTICR e Espectrometria de Massa Estrutural, Faculdade de Ciências, Universidade de Lisboa, Lisbon, Portugal, <sup>2</sup> Centro de Química e Bioquímica, Faculdade de Ciências, Universidade de Lisboa, Lisbon, Portugal, <sup>3</sup> School of BioSciences, The University of Melbourne, Melbourne, VIC, Australia, <sup>4</sup> Biosystems and Integrative Sciences Institute, Faculdade de Ciências, Universidade de Lisboa, Lisbon, Portugal

Keywords: metabolomics, mass spectrometry, nuclear magnetic resonance, metabolic profiling, crop development

**Editorial on the Research Topic**

#### **Metabolomics in Crop Research—Current and Emerging Methodologies**

#### Edited and reviewed by:

Kazuki Saito, RIKEN Center for Sustainable Resource Science (CSRS), Japan

#### \*Correspondence:

Marta Sousa Silva mfsilva@fc.ul.pt Andreia Figueiredo aafigueiredo@fc.ul.pt

#### Specialty section:

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

> Received: 17 July 2019 Accepted: 19 July 2019 Published: 02 August 2019

#### Citation:

Sousa Silva M, Cordeiro C, Roessner U and Figueiredo A (2019) Editorial: Metabolomics in Crop Research—Current and Emerging Methodologies. Front. Plant Sci. 10:1013. doi: 10.3389/fpls.2019.01013 The plant metabolome is highly complex, being composed of over 200,000 metabolites (Fiehn, 2002). The characterization of these small molecules has been crucial to study plant growth and development as well as their response to environmental changes. The potential of metabolomics in plant research, particularly if applied to crop plants, is immense. Besides the aforementioned applications, it is extremely valuable in the discovery of biomarkers and in the improvement of crop yield and quality (Alseekh et al., 2018). This Frontiers Research Topic addresses many applications of metabolomics to crop research, based on different analytical platforms, including mass spectrometry, and nuclear magnetic resonance (NMR). It comprises 13 articles from 109 authors that show the importance and the contribution of metabolomics in the analysis of crop's traceability and genetic variation, in the study of fruit development, and in the understanding of the plant's response to the environment and to different biotic and abiotic stresses.

Numerous agricultural and food products are recognized by their qualities that result mainly from their geographical origin (Vandecandelaere et al., 2009). An unequivocal traceability of a crop guarantees, not only its origin, but also its quality and safety. Busconi et al. followed a targeted metabolomics approach to specifically analyse the phenolic profile of Vanilla × tahitensis and its relation to plant traceability. The phenolic compound profile of different Vanilla × tahitensis clearly discriminated the Papua New Guinea samples from the Tahitian ones from French Polynesia. Additionally, it was possible to separate the Vanilla cultivars that comprised the two Tahitian samples: one exclusively from "Haapape" cultivar and the other from a mixture of both "Haapape" and "Tahiti" cultivars. Within the same region, latitude, and altitude also influence the metabolic composition of a plant. Demasi et al. analyzed different populations of Lavandula angustifolia (lavender) grown in nine peripheral alpine regions for future selection of higher quality flowers and essential oils. A targeted strategy based on the analysis of volatile compounds and essential oils revealed that latitude significatively influenced phytochemical composition, while altitude didn't affect the phytochemical profile of L. angustifolia. These results can be explored for the future ranking of lavender cultivation sites, thus promoting quality and value linked to geographical origin.

In the search for quality traits within crops, untargeted metabolomics approaches have been also very useful. Ellis et al. performed untargeted metabolic profiling of Pisum sativum (pea) mature seeds from genetically marked lines to identify which genotypes were enriched in specific compounds related to enduse quality traits. Indeed, there were sets of compounds in mature seeds associated with their genetic variation, and this information can be used to assist future breeding programmes. Seed quality can be compromised by the occurrence of pests and diseases in the crop. Pisum sativum plants can be affected by the pea aphid Acyrthosiphon pisum, a phloem-feeding insect comprised of different biotypes (or host races), each specialized on a specific crop legume species, including Medicago sativa (alfalfa) and Trifolium pratense (red clover), besides P. sativum (Peccoud et al., 2009). Interestingly, all the host races of this insect can develop on Vicia faba (faba bean). Sanchez-Arcos et al. studied these four plant-herbivorous insect systems using an untargeted metabolomics strategy to identify the metabolites involved in the specificity of pea aphid interaction with the different host plants. All these crop legumes are also affected by other pathogens, including the Aphanomyces euteiches, a soil-borne oomycete that causes Aphanomyces root rot (ARR) (Gaulin et al., 2007). Bazghaleh et al. followed a targeted metabolomics approach focused on polyphenolic profiling, to study the resistance of several lentil genotypes (Lens sp.) to A. euteiches infection. Several differences were found within the root polyphenolic profiles from the wild and the different cultivars of lentils, and between healthy and infected roots. The observed relationship between polyphenol composition and tolerance to A. euteiches will be valuable for future selection of resistant plants. Similar results were obtained by Bernardi et al. while studying the susceptibility of Zea mays (maize) cultivars to the fungus Fusarium verticillioides. Again, a targeted strategy was chosen, focusing on the screening of phenolic compounds from pigmented and non-pigmented maize cultivars. The maize cultivar with highest phenolic content showed highest resistance to Fusarium infection, a promising result toward the selection of more resilient maize plants.

Studying another plant-pathogen system, Chitarrini et al.'s work also showed a great potential for future application for the development of resistant varieties. Using different analytical methods, they identified biomarkers in a resistant grapevine associated with the defense against the biotrophic oomycete Plasmopara viticola, the causative agent of downy mildew. This study contributes for a better understanding of the mechanisms of grapevine interaction and resistance to downy mildew. Negrel et al. took a further step in elucidating this grapevine-P. viticola interaction and characterized P. viticola's metabolome using an untargeted metabolomics approach. These pathogen biomarkers can be used to in the development of a monitoring assay for the early detection of P. viticola in grapevine.

While analyzing both compatible and incompatible interactions between tomato (Solanum lycopersicum) and Pseudomonas syringae, López-Gresa et al. characterized the profile of volatile organic compounds associated with the tomato immune response to this bacteria. These results can be used in the future development of resistant tomato plants, thus preventing this agricultural problem and contributing to a more sustainable production. Another concern regarding stable fruit development is the availability of an optimal light environment. This has been a problem in several countries with fewer daylight hours. For tomato, a supplementary light system is often used in plants grown in greenhouses. However, an adequate plant growth and fruit development depends on the correct implementation of lightning strategies. Fukushima et al. performed an integrative omics strategy to investigate the metabolic changes in early fruit development of single-leaf tomato plants, with only one fruit truss, exposed to different intensities of red LED (lightemitting diode) light. The compounds that responded to the LED treatment and most contributed to the increase of fruit size of tomato plants were metabolites mainly involved in carbohydrate metabolism and the biosynthesis of several amino acids. This is quite relevant given the importance of carbon allocation for fruits during their development, for which a balanced source-sink relationship is essential to ensure adequate fruit nutritional quality and yield (Smith et al., 2018). Following this line of research, Beshir et al. used isotopically labeled substrates and metabolomics to investigate carbon re-allocation changes throughout the development of apple fruit. For the first time it was possible to create a thorough understanding of the metabolic changes' dynamics occurring during the different growth stages of fruit development using dynamic isotope labeling experiments.

Productivity increase in a crop and the efficient use of resources has been achieved through controlled growth in plant factories. Although not a natural environment, these closed production systems, with controlled lightning strategies and reduced environmental pollutants, became more sustainable and attractive to the food industry (Kosai, 2013). Tamura et al. analyzed the metabolite profiles of lettuce leaves grown under hydroponic conditions or fertilized soil, to investigate how cultivation conditions affected leaf metabolic composition. The results showed that the metabolic profile of both lettuce cultivars analyzed was greatly influenced by the cultivation method. Among the affected metabolites are the ones responsible for taste and functional ingredients, like amino acids, and phenolic compounds. Neugart et al. also analyzed the effect of soil fertilization with biological waste compost in the metabolic composition of Brassica rapa ssp. Chinensis (pak choi) sprouts. Indeed, the addition of biological waste from food production (coffee, aronia, and hop) strongly affected sprout metabolic profile, by increasing the concentration of carotenoids and decreasing the glucosinolates and phenolic compounds. The studies from Tamura et al., Fukushima et al., and Neugart et al. show that a detailed assessment of the effect of alternative cultivation systems, like greenhouses, and plant factories, is crucial in the quality and nutritional value of crop products, particularly the evaluation of the light, soil, and fertilization conditions.

Metabolomics has a huge potential in crop plant research. This Research Topic presents current and emerging approaches in metabolomics applied to crop research that we believe will be a future reference in the field and contribute to improve plant productivity and quality. The currently available analytical platforms provide different strategies for metabolomics studies, from targeted, quantitative metabolite profiling, to global untargeted metabolic fingerprinting. In the field of quantitative metabolomics analysis, NMR and mass spectrometry coupled to liquid chromatography (LC-MS) or gas chromatography (GC-MS) are the techniques of choice (Lei et al., 2011). However, currently, Fourier-Transform Ion-Cyclotron-Resonance mass spectrometry (FT-ICR-MS) offers the greatest potential for untargeted metabolomics, being able to simultaneously detect and identify thousands of metabolites in very high-throughput assays, providing extreme-resolution and ultra-high-mass accuracy, and successfully used in crop metabolomics (Aliferis and Jabaji, 2012; Adrian et al., 2017; Maia et al., 2019).

## AUTHOR CONTRIBUTIONS

This manuscript was written by MS and AF. It was revised and edited by all authors prior to submission.

#### REFERENCES


#### FUNDING

This work was supported by projects UID/MULTI/00612/2013, PEst-OE/QUI/UI0612/2013, PEst-OE/BIA/UI4046/2014, PTDC/BAA-MOL/28675/2017, and by IF 00819/2015 to AF and CEECIND/02246/2017 to MS, all from Fundação para a Ciência e Tecnologia (Portugal). We also acknowledge the support from the Portuguese Mass Spectrometry Network (LISBOA-01-0145- FEDER-022125) and the Project EU\_FT-ICR\_MS, funded by the Europe and Union's Horizon 2020 research and innovation programme under grant agreement no. 731077.

#### ACKNOWLEDGMENTS

As Editors of this topic, we wish to thank all the colleagues, authors, reviewers, and support staff who contributed to the success of this Frontiers Research topic. We also acknowledge the Frontiers Editorial Office and the Chief Editors for their technical support.

source of bioactive nutraceutical compounds. Food Funct. 10, 3822–3827. doi: 10.1039/C8FO02328J


**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 © 2019 Sousa Silva, Cordeiro, Roessner and Figueiredo. 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) and the copyright owner(s) 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.

# Phenolic Profiling for Traceability of Vanilla ×tahitensis

Matteo Busconi <sup>1</sup> \* † , Luigi Lucini 2†, Giovanna Soffritti <sup>1</sup> , Jamila Bernardi <sup>1</sup> , Letizia Bernardo<sup>2</sup> , Christel Brunschwig3, 4, Sandra Lepers-Andrzejewski <sup>4</sup> , Phila Raharivelomanana<sup>3</sup> \* and Jose A. Fernandez <sup>5</sup>

<sup>1</sup> Department of Sustainable Crop Production, Università Cattolica del Sacro Cuore, Piacenza, Italy, <sup>2</sup> Institute of Environmental and Agricultural Chemistry, Università Cattolica del Sacro Cuore, Piacenza, Italy, <sup>3</sup> Equipe EIMS (Etude Intégrée des Métabolites Secondaires), UMR 241 EIO Université de la Polynésie Française, Tahiti, French Polynesia, <sup>4</sup> Département Recherche et Développement, Etablissement Vanille de Tahiti, Raiatea, French Polynesia, <sup>5</sup> IDR- Laboratorio de Biotecnología y Recursos Naturales, Universidad de Castilla-La Mancha, Albacete, Spain

#### Edited by:

Andreia Figueiredo, Universidade de Lisboa, Portugal

#### Reviewed by:

Ibrahim M. Abu-Reidah, An-Najah National University, Palestine Gonçalo Graça, Imperial College London, United Kingdom

#### \*Correspondence:

Matteo Busconi matteo.busconi@unicatt.it Phila Raharivelomanana phila.raharivelomanana@upf.pf

† These authors have contributed equally to this work.

#### Specialty section:

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

> Received: 17 July 2017 Accepted: 25 September 2017 Published: 12 October 2017

#### Citation:

Busconi M, Lucini L, Soffritti G, Bernardi J, Bernardo L, Brunschwig C, Lepers-Andrzejewski S, Raharivelomanana P and Fernandez JA (2017) Phenolic Profiling for Traceability of Vanilla ×tahitensis. Front. Plant Sci. 8:1746. doi: 10.3389/fpls.2017.01746 Vanilla is a flavoring recovered from the cured beans of the orchid genus Vanilla. Vanilla ×tahitensis is traditionally cultivated on the islands of French Polynesia, where vanilla vines were first introduced during the nineteenth century and, since the 1960s, have been introduced to other Pacific countries such as Papua New Guinea (PNG), cultivated and sold as "Tahitian vanilla," although both sensory properties and aspect are different. From an economic point of view, it is important to ensure V. ×tahitensis traceability and to guarantee that the marketed product is part of the future protected designation of the origin "Tahitian vanilla" (PDO), currently in progress in French Polynesia. The application of metabolomics, allowing the detection and simultaneous analysis of hundreds or thousands of metabolites from different matrices, has recently gained high interest in food traceability. Here, metabolomics analysis of phenolic compounds profiles was successfully applied for the first time to V. ×tahitensis to deepen our knowledge of vanilla metabolome, focusing on phenolics compounds, for traceability purposes. Phenolics were screened through a quadrupole-time-of-flight mass spectrometer coupled to a UHPLC liquid chromatography system, and 260 different compounds were clearly evidenced and subjected to different statistical analysis in order to enable the discrimination of the samples based on their origin. Eighty-eight and twenty three compounds, with a prevalence of flavonoids, resulted to be highly discriminant through ANOVA and Orthogonal Projections to Latent Structures Discriminant Analysis (OPLS-DA) respectively. Volcano plot analysis and pairwise comparisons were carried out to determine those compounds, mainly responsible for the differences among samples as a consequence of either origin or cultivar. The samples from PNG were clearly different from the Tahitian samples that were further divided in two different groups based on the different phenolic patterns. Among the 260 compounds, metabolomics analysis enabled the detection of previously unreported phenolics in vanilla (such as flavonoids, lignans, stilbenes and other polyphenols).

Keywords: Vanilla ×tahitensis, food metabolomics, phenolics, traceability, authenticity

# INTRODUCTION

Vanilla is a flavoring traditionally recovered from the cured beans of the orchid genus Vanilla. It is also one of the three most expensive spices in the world, along with saffron and cardamom (Hondrogiannis et al., 2013). Other than as a fragrance, in traditional Mexican medicine vanilla was considered as a medicinal plant with multiple positive effects on men's health (Rain and Lubinsky, 2011). In eighteenth and nineteenth centuries, vanilla was included in the European and American pharmacopeia for its medicinal uses (King et al., 1898; Bythrow, 2005). Anti-inflammatory, antiviral, analgesic, antiseptic and aesthetic properties of vanilla have been recently reported (Duke et al., 2003). Antioxidant properties of vanilla and vanilla constituents (essential oil or extract) have been reported by different authors (Kumar et al., 2002; Teuscher et al., 2005; Maurya et al., 2007). The genus Vanilla is indigenous of Central America, and in particular of Mexico, and comprises over 100 different species (Soto Arenas, 2003; Soto Arenas and Dressler, 2010), of which only two are currently cultivated for commercial purposes, Vanilla planifolia Jacks. ex Andrews and Vanilla ×tahitensis (previously V. tahitensis Moore). Vanilla plants are grown in hot-humid tropical climates and have certain agro-ecological requirements in terms of temperature (20–32◦C), precipitation (average from 2,000 to 3,000 mm per year), altitude (from sea level to 600 m), shade (50–70%), welldrained soil rich in humus, support trees (as they are hemiepiphytic orchids). Optimal flowering, and, consequently, pod production, requires specific climatic conditions of a dry and cool season of at least 2 months (Hernandez and Lubinsky, 2011). The hybrid nature of Tahitian vanilla was recently determined (Lubinsky et al., 2008). Analysis of cpDNA and nuclear ITS sequences provided evidence that Tahitian vanilla is a hybrid between V. planifolia and V. odorata C. Presl, with V. planifolia being the female parent. Vanilla ×tahitensis is traditionally cultivated on French Polynesian islands, where vanilla vines were first introduced during the nineteenth century (Costantin and Bois, 1915; Bouriquet, 1954; Florence and Guérin, 1996; Lepers-Andrzejewski et al., 2012), currently, vanilla production is mainly carried out in the Leeward Islands (high islands: Raiatea, Tahaa, Huahine), Society archipelago (Lepers-Andrzejewski and Dron, 2010). Subsequently, within a short period of time, diversification resulted in the origination of about 14 cultivars identified over the years by local producers (Lepers-Andrzejewski et al., 2010). Among the isolated cultivars, two of them, "Haapape" and "Tahiti," which are morphologically and genetically differentiated (Lepers-Andrzejewski et al., 2011), became the most widespread and most commercialized cultivars in French Polynesia. Since the 1960s, V. ×tahitensis has been introduced into other Pacific countries such as Papua New Guinea (PNG), cultivated and sold as "Tahitian vanilla," although both sensory properties and aspect are different. These differences depend more on factors such as genetic traits, environment and technology (curing method and storage conditions). Actually, V. ×tahitensis, being less restrictive to humidity and supporting better rainfall, was found to be more suitable to the climate of PNG than V. planifolia and the Sepik area (well-drained alluvial plains) is the main vanilla production region (Mac Gregor, 2005). The volatile composition of vanilla in general and of V. ×tahitensis in particular has recently been investigated and the results were reported in the literature (Pérez-Silva et al., 2006; Brunschwig et al., 2009; Lepers-Andrzejewski et al., 2010; Brunschwig et al., 2012; Takahashi et al., 2013; Brunschwig et al., 2016).

According to official data (http://www.fao.org/faostat), in 2014, global vanilla production was over 7,000 t, with French Polynesia ranking ninth among the vanilla producers in the world. Considering the high economic value of vanilla, in the last years, several publications have dealt with the development of reliable methods to trace vanilla production according to the species or the different geographic origin. Tracing the genetic and/or the geographic origin of vanilla is crucial because the species, the variety, the environment (in particular climatic conditions) and the production method (curing methods and storage) imply characteristic different flavors that could be reflected in the quality and the price of the product. From an economic point of view, it is important to ensure V. ×tahitensis traceability and to guarantee that the product which is marketed is part of the future protected designation of the origin "Tahitian vanilla" (PDO) (Journal Officiel de la Polynésie Française, 2014, 2016). Vanilla ×tahitensis traceability has already been carried out via different analytical techniques, such as gas chromatography–flame ionization detection (GC-FID) and gas chromatography–mass spectrometry (GC-MS), to analyse the volatile compounds for quality control (Brunschwig et al., 2016); analysis of the stable isotopes of carbon and hydrogen evidencing that V. ×tahitensis has more heavy carbon than V. planifolia and that isotopes can be used to discriminate the geographic origin of the samples (Sølvbjerg Hansen et al., 2014); wavelength dispersive X-ray fluorescence to identify the elemental composition and the geographic origin of vanilla samples (Hondrogiannis et al., 2013). The use of metabolomics, allowing the detection and simultaneous analysis of hundreds or thousands of metabolites from different matrices, has recently gained high interest in food traceability (Oms-Oliu et al., 2013). Metabolomic techniques have been applied for the analysis of raw food material (cultivar identification, study of different metabolites that accumulate during plant growth, ripening and postharvest) and processed plant-derived food (food classification, authenticity assessment, food control) (Oms-Oliu et al., 2013). Up to now, the study of vanilla metabolome was applied to V. planifolia, albeit not for traceability purposes. Some studies concerning the metabolome of V. planifolia green pods from La Réunion (Palama et al., 2011) and the changes in metabolome composition in leaves (Palama et al., 2010) and pods (Palama et al., 2009) in different developmental stages have recently been published. Recently, Gu et al. (2017) carried out a comparative metabolomics analysis by using high-performance liquid chromatography– mass spectrometry (LC–MS) to analyse vanilla metabolome before and after curing to study the biosynthesis of vanillin during the curing process of vanilla. They evidenced the presence of at least seven different putative pathways of vanillin biosynthesis some of them possibly correlated with microbial activity.

In other species, metabolomics analysis was applied successfully to assess, among others, the authenticity of processed plant-derived food such as fruit juices (Vardin et al., 2008), coffee (Oliveira et al., 2009), vegetable oils (Rohman and Man, 2012; Ruiz-Samblás et al., 2012), and tomatoes (Lucini et al., 2017).

In the present work, we applied, for the first time, the comprehensive profile of the phenolic compounds for traceability purposes in V. ×tahitensis. Samples from two different areas, French Polynesia (FP) and PNG, were surveyed and the profile of the phenolic compounds investigated by UHPLC-ESI/QTOF-MS.

## MATERIALS AND METHODS

# Vanilla Samples

Two commercial samples of Vanilla ×tahitensis pods were obtained from two commercial providers ("Pacific Natural Product" and "Tahiti vanille" Brand); both of these samples were grown on the Leeward Islands and harvested in 2013. The pods have been cured following the traditional Polynesian method (Lepers-Andrzejewski et al., 2010). A sample of V. ×tahitensis pods (250 g) from PNG was provided by the NARI organization (PNG's National Agricultural Research Institute, donation from Dr. Sergie Bang) from the East Sepik region and collected from the 2013 harvest. It was used as reference for a comparison with V. ×tahitensis from French Polynesia. The pods have been cured according to the methods currently used in PNG (see discussion). At the end, three batches were available: batch1, pods of Tahitian vanilla, belonging to the "Haapape" cultivar, Vanille de Tahiti Brand from Pacific Natural Product, grown on the Leeward Islands (French Polynesia FP); batch2, mixture of pods of Tahitian vanilla, cultivars "Haapape" and "Tahiti," "Tahiti vanille" Brand, from the Leeward Islands (FP); batch3, pods of V.×tahitensis from PNG.

#### Profiling of Phenolic Compounds

Ten independent pods per treatment were analyzed as individual samples. The phenolic compounds were screened through a quadrupole-time-of-flight mass spectrometer coupled to a UHPLC liquid chromatography system (UHPLC-ESI/QTOF-MS), on the basis of the approach described by Lucini et al. (2015). Samples were extracted in 10 volumes of 50 mM HCOOH in 80% methanol, using an IKA T10 Ultra-Turrax to comminute samples (3 min at 30,000 rpm). The extracts were then centrifuged at +4 ◦C and filtered through a 0.22 µm cellulose membrane, diluted five times in 50% methanol and transferred to an amber vial for LC-ESI/Q-TOF-MS analysis. A 1290 UHPLC liquid chromatograph, equipped with a binary pump and coupled to a G6550 iFunnel QTOF mass spectrometer through a Dual Electrospray JetStream ionization system (all from Agilent Technologies Santa Clara, CA, USA), was used to profile phenolic compounds. The mass spectrometer was operated in positive MS-only (SCAN) mode to acquire spectra in the range 50–1,000 m/z. Extracts were injected (6 µL) and chromatographed under a water-methanol gradient elution (from 6% methanol to 92% methanol in 35 min), using an Agilent Zorbax Eclipse Plus C18 column (50 × 2.1 mm, 1.8µm). Lock masses and source conditions were optimized for phenolic compounds in previous experiments (Lucini et al., 2017). Briefly, nitrogen was used as drying gas (8 L min−<sup>1</sup> and 330◦C), nebulizer pressure was 60 psig and capillary voltage was 3,500 V. Blanks were analyzed between samples and lock masses (m/z 121.0509 and 922.0098) were continuously infused during chromatographic runs to achieve higher accuracies.

Raw data were processed via the Agilent Profinder B.0700 software using the "find-by-formula" algorithm. With this purpose, features mass and retention time were aligned and then the whole isotopic profile (isotopic spacing and isotopic ratio) was used for compounds' annotation, together with the monoisotopic accurate mass, against the database exported from Phenol-Explorer 3.6 (Rothwell et al., 2013). In addition, recursive analysis (using retention time as mandatory in the second ID step, with a tolerance of 0.1 min) and frequency filter were applied (only those compounds being in at least 80% of replications within at least one condition were retained). Therefore, based on the strategy applied, identification was carried out according to Level 2 (putatively annotated compounds) as set out by the COSMOS Metabolomics Standards Initiative (http://cosmos-fp7.eu/ msi).

## Statistical Analysis

The abundance value for each compound in the dataset was log2-transformed, normalized at 75th percentile and baselined to the median in the dataset. One-way analysis of variance (ANOVA) (p < 0.05, Benjamini-Hochberg multiple testing correction) has been carried out on the starting data set of metabolites. Unpaired t-test (p < 0.05, Benjamini-Hochberg FDR multiple testing correction) and fold-change analysis (cut-off = 2) were combined into Volcano plot analysis. Subsequently, unsupervised hierarchical cluster analysis (Euclidean similarity measure and Wards linkage rule) was generated on the basis of fold-change heatmaps.

Finally, the raw dataset was exported in SIMCA 14 (Umetrics, Malmo, Sweden), Pareto scaled (to reduce the relative importance of larger values and partially preserve data structure) and elaborated for orthogonal partial least squares discriminant analysis (OPLS-DA) prediction modeling. Herein, the variation between the three batches was separated into predictive and orthogonal (technical and biological variation) components. The presence of outliers was investigated according to Hotelling's T2 (i.e., the distance from the origin in the model plane), using 95 and 99% confidence limits for suspect and strong outliers, respectively. Method validity was next tested using CV-ANOVA (P < 0.01) and permutation testing after inspecting model parameters (goodness-of-fit R2Y and goodness-of-prediction Q2Y). Regarding Q2Y prediction ability, a value > 0.5 indicates good model quality (Rombouts et al., 2017). Variable importance in projection (VIP analysis) was used to evaluate the importance of metabolites and select the most discriminant ones (VIP score > 1).

# RESULTS

Ten single cured pods were recovered from each of the three vanilla batches and analyzed independently from the others. The use of an informative approach, such as UHPLC-ESI/QTOF-MS together with a comprehensive database (Phenol-Explorer), allowed the annotation of 260 phenolic compounds. All these phenolics belonged to a small number of main classes: flavonoids (120 compounds), phenolic acids (53 compounds), lignans (18 compounds), stilbenes (6 compounds) and one last group labeled as "other polyphenols" (63 compounds) (**Supplementary Table 1**). To the best of our knowledge, we report for the first time the occurrence of phenolic compounds from V. ×tahitensis, which do not belong to flavoring components (mainly composed by volatile components), such as flavonoids, lignans, stilbenes and other classes of polyphenols (curcuminoids). Based on the fold-change analysis, a heat map was developed and subsequently, an unsupervised hierarchical cluster analysis was carried out (**Figure 1**). It is important to underline that with the applied method we did not obtain an absolute quantification of the compounds but an indication of the relative abundance, based on the area of the peaks, of the different phenolics in the samples under comparison. Three main clusters could be identified: (1) all the pods from PNG batch3 (brown cluster); (2) eight pods from FP batch2 (blue cluster); (3) all the pods from FP batch1 (red cluster) plus two pods from FP batch2. Samples from PNG presented a characteristic phenolic profile, differing from that of the Tahitian ones, thus indicating that samples could be discriminated in terms of their geographic origin. Within the Tahitian pods, the metabolic profiles were quite similar, but still sufficiently different to allow two further sub-clusters to separate the Tahitian sample sets.

Two approaches were then applied to investigate the most differential compounds within the sample set. Initially, the data set was analyzed by using a one-way ANOVA (P < 0.05) that resulted in 88 differential compounds out of the 260 phenolics identified (**Supplementary Table 2**). Flavonoids were the most represented compounds (39 compounds), followed by "other polyphenols" and phenolic acids (21 and 20 compounds, respectively), stilbenes and lignans (four compounds per class). Among the phenolic acids, hydroxycinnamic acids were the most represented subclass (13 compounds) while within the flavonoids, flavonols (nine compounds), flavones (eight compounds) and anthocyanins (eight compounds) were more represented. Vanillin (4-Hydroxy-3-methoxybenzaldehyde) content was significantly different between the three samples according to one-way ANOVA, but the fold change was not high enough to be recognized as discriminant with the Volcano plot analysis in the pairwise comparison (see later). Among the 260 compounds, only one anisyl derivative (anisaldehyde) was detected, but not selected among the most discriminant compounds with ANOVA nor with the subsequent analyses (OPLS-DA and Volcano plot). This highlights that such metabolomics analysis was more powerful to detect, in V. × tahitensis, compounds even more discriminant than the characteristic odor-active anisyl compounds (Brunschwig et al., 2012, 2016).

The results of OPLS-DA agreed with the unsupervised cluster analysis being able to separate the samples according to their origin and, possibly, the cultivar of origin (**Figure 2**). Indeed, the characteristics of the model were excellent: R2Y = 0.969 and Q2Y = 0.877. No outlier samples could be observed by Hotelling's T2, whereas both CV-ANOVA (P = 4.3 10−<sup>14</sup> for regression) and permutation test (Supplementary Figure 2) showed a more than adequate degree of validation.

Overall, all these results evidenced that differences among the three samples were included in the compounds data set, thus driving the need for a more detailed assessment of which compounds these differences could be ascribed too. The VIP analysis from OPLS-DA allowed to select a small number of discriminant compounds, reducing the data set to only 23 highly discriminant compounds. Individual VIP scores are reported in **Table 1**, together with the discriminant compounds grouped in phenolic classes. Flavonoids (12 different compounds), other polyphenols and phenolic acids (four different compounds each one) were the most represented subclasses of phenolics (**Table 1**). Among the most discriminant compounds, two stilbenes (resveratrol and pterostilbene) have been detected.

Volcano plot analysis (unpaired t-test; p ≤ 0.01; fold-change cut off = 2) was carried out to determine those compounds mainly responsible for the differences among samples as a consequence of either origin or cultivar. Three different pairwise comparisons were carried out: (1) between the two Tahitian batch samples; (2) between FP batch2 and PNG; (3) between FP batch1 and PNG. The discriminant compounds for all the three different comparisons have been reported in **Supplementary Table 3**, in three active sheets, along with the corresponding fold change (an estimate of the relative abundance of the compounds in the samples under pairwise comparison) and regulation. The volcano plot graphic output of the comparisons FP batch1 and PNG and FP batch2 and PNG is reported in Supplementary Figure 1.

Twenty-one compounds (**Table 2**) were discriminant between the samples from the same geographic location (French Polynesia): flavonoids (10 compounds, highly represented by flavonols) and phenolic acids (five compounds, almost exclusively hydroxycinnamic acids) were the most frequent classes of phenolics. Discriminant hydroxycinnamic acids were derivatives of coumaric and caffeic acid. Six compounds were grouped under a class named, in phenol-explorer (Rothwell et al., 2013), as "other polyphenols." Twelve compounds were upregulated, while nine were down-regulated in FP batch1 with respect to the mixture (batch2). Fifty percent of flavonoids were respectively up- and down-regulated. Among other classes of phenolics, the up-regulated compounds in FP batch1 prevailed. Those compounds identified as discriminant through both the t-test and OPLS-DA are marked in **Table 2** with an asterisk.

Contrary to what was observed between the Tahitian samples, the other comparisons evidenced a higher number of differential compounds, respectively, 82 (FP batch2 vs. PNG) and 57 (FP batch1 vs. PNG). Checking these differential compounds through Venn analysis (Supplementary Figure 1), 46 metabolites were found in both the comparisons, while 36 and 11, respectively, were exclusive of the single comparisons for the two Tahitian batches when compared to

FIGURE 1 | Results of a not averaged, unsupervised hierarchical cluster analysis on the phenolic profile in the vanilla pods analyzed. The intensity of the compounds was used to build the heat maps on which the clusters were generated. Samples from PNG (brown) are clearly different with respect to Tahitian samples (red, batch1, and blue, batch2) in regards to the phenolic composition. Among the Tahitian samples, two patterns can be distinguished.

**13**

PNG (**Tables 3**, **4**). The common metabolites showed always the same kind of regulation in Tahitian samples, up or down, respectively, with respect to PNG samples; no compounds showing different regulations in the two comparisons could be observed (**Table 3**). The p-values of the common metabolites, as resulted from the two different pairwise comparisons, have been reported in **Table 3**. As an example, focusing on stilbenes, pterostilbene and resveratrol 3-O-glucoside were highly up-regulated, while piceatannol was down-regulated, both in batch2 vs. PNG and in batch1 vs. PNG. Among these compounds responsible for discrimination between the two different locations, flavonoids were the most represented ones (20 compounds), followed by phenolic acids (nine compounds). Within flavonoids, anthocyanins and flavonols were the most abundant compounds and glycosylated forms (e.g., malvidin 3-O-galactoside, petunidin 3-O-galactoside, petunidin TABLE 1 | Compounds discriminating between the three vanilla samples, as obtained from VIP analysis from Orthogonal Projections to Latent Structures Discriminant Analysis (OPLS-DA).


3-O-rutinoside) were highly represented. Globally, 31 and 15 compounds, respectively, were significantly up- and downregulated in Tahitian samples. In more detail, flavonoids (17 out of 20) and phenolic acids (6 out of 9) were mainly up-regulated, while within other polyphenols, more or less the same number of compounds were up- or down-regulated (5 and 6 out of 11, respectively). Eight of the compounds reported in **Table 3** were also selected by OPLS-DA as differential molecules able to discriminate samples based on both the geographic origin and genotype (five out of eight were up-regulated in Tahitian samples).

Focusing on the exclusive metabolites (**Table 4**), the number of differential compounds was lower between FP batch1 (11) vs. PNG than between FP batch2 vs. PNG (36). Furthermore, up-regulated compounds were more abundant than the downaccumulated ones, the corresponding p-values are reported in **Table 4**. Among the exclusive compounds of batch1 and of batch2, flavonoids were predominant: 7 out of 11 for batch1 and 19 out of 36 for batch2 (flavones and flavonols were particularly abundant). Exclusive compounds of batch2 were also phenolic acids (9 out of 36) and, despite the small number, stilbenes (**Table 4**). Among the stilbenes, both pinosylvin and delta-viniferin were up-regulated in batch2 with respect to PNG. Overall, out of five differential stilbenes (**Tables 3**, **4**), four were up-regulated and just one was down-regulated in Tahitian samples. Nine compounds among the exclusive metabolites were also recognized by OPLS-DA analysis; seven out of nine were up-regulated in Tahitian samples.

#### DISCUSSION

While traceability is usually most relevant when it concerns public health, the possibility to discover the origin of products, ingredients and their attributes from the farm throughout the whole food chain to the consumer is gaining more importance. Regarding plant-derived products, being able to trace the origin is extremely important. This is because the characteristics of these products are strongly influenced, other than by the genetic constitution of the cultivated varieties, by environmental conditions and by local traditional curing methods, both able to influence the final metabolic profile of the product (Brunschwig et al., 2016). Genetic bases being equal, the origin of the products, considered as the sum of environment and production technologies, implies characteristic flavors, subsequently reflected in premium prices that can represent a target for falsifications and frauds. This is particularly true in the case of V. ×tahitensis. Therefore, the development


TABLE 2 | Discriminant metabolites differentiating pods of batch1 from those of batch2.

The compounds were grouped in their chemical classes and sub-classes. The up- and down-regulation and the corresponding p values, as a result of the Volcano analysis (p ≤ 0.05, fold-change cut-off = 2.0), are reported.

\*Compounds also recognized as discriminant via OPLS-DA analysis.

of control procedures to protect the high-quality Tahitian production is of great interest and scientifically supported by the results of recent papers (Brunschwig et al., 2009, 2016; Takahashi et al., 2013). Among them, Brunschwig et al. (2016) have analyzed the volatile composition and the sensory properties of V. ×tahitensis from different geographic origins (French Polynesia and PNG), evidencing a clear difference in the composition of these compounds between the two different geographic locations. V. ×tahitensis from PNG was clearly different from V. ×tahitensis from French Polynesia. According to the authors, these differences were mainly a consequence of the curing technology. Our study could be considered as a prosecution of Brunschwig et al. (2016), although a different analytical approach was applied, to increase the knowledge on vanilla chemical composition toward the development of a traceability procedure for this product. Along with the analysis of chemical composition, the development of traceability methods based on DNA might be a topic of great interest. On this basis, we also tried to recover DNA from the cured pods. However, the techniques employed for DNA isolation from pods, both commercial kits and customized protocols, were not efficient and no DNA, or no PCR grade DNA, could be recovered (data not shown). Although further research is required to implement DNA-based approaches, these results strongly drive the analysis of secondary metabolites for traceability purposes. In the present study, we applied for the first time metabolomics, focused on the phenolic compounds, to V. ×tahitensis traceability. Ten pods for each one of the three samples (batch1 – cultivar "Haapape" FP; batch2 – "Haapape + Tahiti" FP; batch3 – PNG) were analyzed evidencing a quali-quantitative phenolic profile involving 260 compounds. The decision to focus our attention on phenolics has been driven by the fact that: (1) these secondary metabolites are highly associated to different environmental conditions, curing methods and genotypes, thus showing high discriminant capacity (Klockmann et al., 2016); (2) such an in-depth analysis of these chemically diverse compounds was not previously carried out for this species; (3) with respect to volatiles, phenolics are expected to be more stable during storage.

Pods from FP were softer than pods from PNG and this is very likely a direct consequence of the two different curing methods adopted: in the traditional FP method, the pods are harvested when fully mature, exposed in the shade for a natural browning (no high-temperature scalding step to stop maturation), alternatively dried in the sun and wrapped in cotton material overnight; they are then finally air-dried to stabilize the flavor and keep the water content at about 50%; in the PNG method it is included a high-temperature scalding step to stop maturation and drying to about 40% water content. Using ten individual pods for each samples increases the number of independent replicates strongly supporting the reliability of the results. Flavonoids were the most abundant phenolics, followed

TABLE 3 | Common metabolites, differentiating Tahitian and PNG samples, evidenced by matching the results of the two Volcano analyses: batch2 vs. PNG (b2-PNG) and batch1 vs. PNG (b1-PNG).


Metabolites are reported according to the chemical class and sub-class of the compounds. The up- and down-regulation with the corresponding p values, as a result of the Volcano analysis (p ≤ 0.05, fold-change cut-off = 2.0), are reported. Each compound reported in the Table had the same regulation (up or down) in Tahitian samples in both comparisons. The fold change of up and down regulation and other statistics have been reported in Supplementary Table 3. \*Compounds also recognized as discriminant via OPLS-DA analysis.

#### TABLE 4 | Exclusive metabolites differentiating Tahitian and PNG samples.


By matching the differential metabolites from the two Volcano analyses, these metabolites were present only in the pods of batch1 or only in batch2. Metabolites are reported according to the chemical class and sub-class of the compounds. The up- and down-regulation and the corresponding p values, as a result of the Volcano analysis (p ≤ 0.05, fold-change cut-off = 2.0), are reported. The fold change of up and down regulation and other statistics have been reported in Supplementary Table 3.

\*Compounds also recognized as discriminant via OPLS-DA analysis.

by phenolic acids and by a high number of compounds classified as other polyphenols. Small, albeit significant was the presence of stilbenes, such as resveratrol, molecules that have recently been widely studied, mainly in grapevine, for their role in protecting plants against diseases (Bavaresco et al., 2016) and for their health benefits because of a potent antioxidant activity (Marques et al., 2009). It was noted that out of five differential stilbenes, four were highly up-regulated in Tahitian samples.

Unsupervised cluster analysis (**Figure 1**) clearly evidenced that the whole phenolic profile was able to separate the three batches in different clusters based on the origin: PNG vanilla was clearly different from FP vanilla. Likely, the differences between Tahitian and PNG phenolic profiles were a consequence of the combination of the environment and the adopted curing technologies between FP and PNG. Indeed, being the exact composition of the V. ×tahitensis sample from PNG unknown, in principle, we cannot completely exclude also a possible effect of genotype. According to us, the genetic effect, in this case, should be less probable considering that cultivars "Haapape" and "Tahiti," the two mainly produced cultivars in FP (being "Haapape" the first and "Tahiti" the second most frequently grown cultivars) are also the most widespread cultivars out of FP and in particular in PNG. Further, two phenolic patterns could be distinguished, within the Tahitian main cluster, basically separating the pods from FP batches 1 and 2, and in this case the differences could be mainly a consequence of the different genotypes (Haapape and Tahiti) being the curing method the same and the environmental conditions more similar and uniform. Considering the whole data set, two pods of batch2 were placed within the batch1 cluster. These results confirmed that the pods of FP batch1 have a uniform phenolic profile strongly supporting their belonging to the single cultivar "Haapape." On the other hand, the FP batch2 was confirmed as a mixture of cultivars and, considering that the 10 pods representing batch2 were randomly selected for the analysis, the most likely scenario is that eight out of ten analyzed pods belong to "Tahiti" and two out of ten belong to "Haapape."

The OPLS-DA analysis confirmed the results of the cluster analysis, clearly separating the three batches according to their origin (**Figure 2**), and evidenced that 23 differential metabolites (mainly flavonoids, 12 out of 23 compounds, **Table 1** and **Figure 2**) can provide the same separation power as the cluster analysis. Taken together, these results support the utility of metabolomics in: separating pods based on a different origin and eventually on the different cultivars, even if vines were cultivated in the same area and the pods were processed using the same method; finding the most discriminant metabolites on which eventually base the development of traceability procedures.

Pairwise comparisons were carried out through Volcano plot analysis in order to identify the up- and down-accumulated compounds that can be more interesting in discriminating the different samples. A first comparison was carried out between the two Tahitian samples (**Table 2**). In this case, 21 compounds were detected to be highly up- or down-accumulated between the two groups (the fold change and the relative abundance of the compounds in the samples is reported in **Supplementary Table 3**). About half of these compounds were up-regulated in "Haapape" with respect to the mixture. The main differences were among flavonoids (particularly flavonols) and phenolic acids (mainly hydroxycinnamic acids). Amongst flavonols, three compounds were respectively up- and downregulated in "Haapape"; among cinnamic acids, three and one compounds were respectively up- and down-regulated in the same treatment. These differences can be a consequence of genetic differences between the two cultivars. Indeed, the "Haapape" and "Tahiti" genotypes share common genetic markers, but they differ in their ploidy level, with "Tahiti" being diploid and "Haapape" tetraploid (Lepers-Andrzejewski et al., 2011). This difference can be reflected in a different metabolomics composition of the pods.

By matching the two pairwise comparisons, FP batch1 vs. PNG and FP batch2 vs. PNG, the differential metabolites could be separated in two categories: compounds common between the two comparisons (46 metabolites, **Table 3**) and compounds exclusive of the single comparisons (7 and 36 metabolites, respectively, **Table 4**). Among the metabolites reported in **Table 3**, generally, flavonoids and phenolic acids were the most abundant classes of compounds. Being shared in both pairwise comparisons, we can postulate that these metabolites were independent of the genotype and mainly related to the different origin. This hypothesis is strengthened by the fact that, in both comparisons, the trend of accumulation was always the same for all the compounds reported in **Table 3**. Furthermore, most of the discriminant metabolites were up-regulated in "Tahitian" batches as compared to the PNG group.

As reported before, the particular cultivar(s) of the PNG sample was/were unknown; nevertheless between batch1 and PNG and between batch2 vs. PNG there were, respectively, 7 and 36 differential molecules. Considering this, the phenolic profile of the PNG sample was more similar to that of batch1 (cultivar "Haapape"), supporting a hypothesis that PNG pods can belong to the cultivar "Haapape" and that the main differences between the two batches are mainly a consequence of the different origin.

Regardless of the cultivar(s) considered, a significantly higher number of differential metabolites was pointed out when geographical origin was adopted as classification criterion under all statistics applied. Similarly, unsupervised hierarchical clustering evidenced that origin was the principal classification parameter. Although discrimination of origin via metabolomics was effective, specific markers could not be pointed out. Indeed, the discrimination potential is related to the actual profile of a wide variety of phenolic compounds, both in terms of their identity and abundance, with flavonoids and hydroxycinnamic acids playing a major role.

Considering our phenolic profiling approach has not been described previously, it becomes difficult to compare our results with markers previously reported in the literature. Nonetheless, most of the markers are aroma-related compounds that are likely more informative regarding vanilla quality rather than its origin. Vanillin is a phenolic aldehyde and one of the most important compounds in the primary extracts from vanilla beans as well as the principal flavor and aroma compound in vanilla. Brunschwig et al. (2016) have reported a significant difference in vanillin content between V. ×tahitensis grown and processed in PNG and in Tahiti, with a higher concentration in samples from PNG. In this case, vanillin values were significantly different among the three batches by ANOVA, but vanillin was not identified as one of the most differential compounds evidenced by Volcano analysis. This meant that, while a significantly different content of vanillin was present, the fold change value was lower than 2.0 (cut-off value adopted in the Volcano analysis) and that vanillin was not among the compounds mainly influenced by the origin under the present condition of analysis. This difference with respect to Brunschwig et al. (2016) can be a consequence of the different extraction method and of the different analytical technique. On the other hand, it is important to stress out that having several markers is expected to strengthen the discrimination capability of the analytical approach. With this regard, metabolomics followed by multivariate chemometrics is among the approaches gaining most of the popularity in traceability. Important characteristic constituents of V. ×tahitensis anisyl derivatives (anisyl alcohol, anisaldehyde, methyl anisate, anisyl formate, and anisyl acetate), which were previously detected by HPLC analysis (Brunschwig et al., 2009), were not found as discriminant in this study. However, this can be easily ascribed to their volatility that hampered their ionization efficiency at the electrospray interface. On the other hand, although these compounds are known to play an important role in the vanilla aroma, the whole phenolic profile was wide and differed under both a qualitative and quantitative point of view. With this regard, the latter appeared to be more informative for traceability purposes.

As a conclusion, in the present study, metabolomics focused on the phenolic profile was successfully applied for the first time to V. ×tahitensis in order to increase our knowledge of vanilla metabolome for traceability purposes. Among the results, the most significant were: (1) the discrimination of the samples based on their origin: PNG samples were clearly different from Tahitian samples; this variation could be mainly related to the different origin (i.e., a combination of pedo-climatic conditions and curing methods adopted in the two countries); (2) the grouping of the Tahitian samples based on the two patterns, which could be explained considering that the first sample corresponded to only one cultivar ("Haapape") while the second one was a mix of 2 cultivars ("Haapape" and "Tahiti"). These findings evidenced the utility of the metabolomics analysis to detect a high number of discriminating compounds; this possibility, in

#### REFERENCES

Bavaresco, L., Lucini, L., Busconi, M., Flamini, R., and De Rosso, M. (2016). Wine resveratrol: from the ground up. Nutrients 8:222. doi: 10.3390/nu8040222


combination with robust multivariate chemometrics, might open to the possibility to develop a standard procedure for traceability and authentication based on the metabolic profile of the pods. Of course, further studies are necessary to deepen our vision on vanilla metabolome and on the metabolic variations correlated with the cultivars and the origins, in order to detect and then validate selected highly discriminant compounds suitable for authentication. Finally, our deep profiling approach allowed the annotation of unexpected phenolic components, not yet reported, whose presence could be linked to the medicinal properties of vanilla. Polyphenols, such as those identified in this study, significantly benefit human health (Del Rio et al., 2013) and they can deserve further perspective assessments for an up-dated pharmaceutical valuation of vanilla.

#### AUTHOR CONTRIBUTIONS

MB designed the study, in cooperation with PR and JF. CB, SL, LB, GS, and JB carried out the plant experiments, contributed to interpretation of data and drafted the manuscript. LL developed the mass spectrometric method, performed statistics and helped to draft the manuscript. MB, PR, and JF draft and critically revise the manuscript. All authors read and approved the final manuscript.

#### ACKNOWLEDGMENTS

We are grateful to Dr. Sergie Bang (NARI) for providing V. ×tahitensis pods from Papua New Guinea.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2017. 01746/full#supplementary-material

Supplementary Table 1 | A full list of the 260 phenolic compounds is provided.

Supplementary Table 2 | The compounds, significantly different among the three samples (one-way ANOVA, multiple testing correction, p-value 0.05), are reported with the corresponding P-value.

Supplementary Table 3 | The results of the volcano plot analysis reporting, for each pairwise comparison, the compounds characterized by a fold change higher than 2.


**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 Busconi, Lucini, Soffritti, Bernardi, Bernardo, Brunschwig, Lepers-Andrzejewski, Raharivelomanana and Fernandez. 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.

# Latitude and Altitude Influence Secondary Metabolite Production in Peripheral Alpine Populations of the Mediterranean Species Lavandula angustifolia Mill.

Sonia Demasi <sup>1</sup> , Matteo Caser <sup>1</sup> , Michele Lonati <sup>1</sup> , Pier L. Cioni <sup>2</sup> , Luisa Pistelli <sup>2</sup> , Basma Najar <sup>2</sup> and Valentina Scariot <sup>1</sup> \*

*<sup>1</sup> Department of Agricultural, Forest and Food Sciences, University of Torino, Grugliasco, Italy, <sup>2</sup> Department of Pharmacy, University of Pisa, Pisa, Italy*

#### Edited by:

*Marta Sousa Silva, Universidade de Lisboa, Portugal*

#### Reviewed by:

*Ligia Salgueiro, University of Coimbra, Portugal Inger Martinussen, Norwegian Institute of Bioeconomy Research (NIBIO), Norway*

> \*Correspondence: *Valentina Scariot valentina.scariot@unito.it*

#### Specialty section:

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

> Received: *20 April 2018* Accepted: *18 June 2018* Published: *05 July 2018*

#### Citation:

*Demasi S, Caser M, Lonati M, Cioni PL, Pistelli L, Najar B and Scariot V (2018) Latitude and Altitude Influence Secondary Metabolite Production in Peripheral Alpine Populations of the Mediterranean Species Lavandula angustifolia Mill. Front. Plant Sci. 9:983. doi: 10.3389/fpls.2018.00983* *Lavandula angustifolia* Mill. has a great economic importance in perfumery, cosmetics, food manufacturing, aromatherapy, and pharmaceutical industry. This species finds its phytosociological optimum in the sub-Mediterranean region. Latitudinal and altitudinal gradients are expected to affect species diversification in peripheral alpine populations. In this study, phenotypic traits including morphometric parameters, volatile organic compounds (VOCs) and essential oils (EOs) were analyzed in lavender peripheral populations selected in order to explore different ecological conditions. Plants were cultivated under uniform conditions to observe variations due to the genetic adaptation to native environments and to exclude the short-term response to environmental factors. Results showed qualitatively and quantitatively intra-specific variations in secondary metabolites, mainly along the latitudinal gradient, while minor effect was attributable to the altitude. This latter affected more the morphometric parameters. As the latitude augmented, VOCs showed lower content of monoterpene hydrocarbon (mh) and higher content of oxygenated monoterpenes (om); whereas EOs showed higher content of mh and non-terpene derivatives (nt) and lower content of sesquiterpene hydrocarbons (sh). Lavender aroma and EO composition varied in every population, for a total of 88 and 104 compounds identified, respectively. Eleven and 13 compounds were responsible for 95% of the dissimilarity, with linalool, linalyl acetate and 1,8-cineole as major contributors. As the latitude augmented, linalool decreased and 1,8-cineole increased while linalyl acetate content was unaffected. These results are discussed with regards to the potential adoption of the lavender peripheral alpine populations for the improvement of quality and productivity of lavender cultivations, especially in mountainous areas.

Keywords: lavender, Lamiaceae, volatile organic compounds, essential oils, ecological gradient

# INTRODUCTION

The genus Lavandula (Lamiaceae family) comprises approximately 39 species, with a natural occurrence in the Mediterranean region, to the Arabian Peninsula, South West Asia and India (Lis-Balchin, 2003). This genus is constituted by small evergreen shrubs, with aromatic foliage and flowers, and due to its great economic importance this genus was the subject of several studies. Indeed, 1,000 tons of lavandin essential oil (L. × intermedia Emeric ex Loisel., EO) are produced worldwide every year, while only 200 tons of EO from lavender (L. angustifolia Mill.) together with 200 tons of EO from spike lavender (L. latifolia Medik.) are produced (Lesage-Meessen et al., 2015). Lavender dried flowers and EOs are commonly employed in perfumery and cosmetics, food manufacturing and aromatherapy (Cavanagh and Wilkinson, 2002; Gonçalves and Romano, 2013; Prusinowska and Smigielski, 2014). Biological and antioxidant properties of lavender EO have been extensively assessed (Cavanagh and Wilkinson, 2002; Raut and Karuppayil, 2014; Shahdadi et al., 2017). Numerous therapeutic activities have been reported, such as convulsion relief, anxiety and depression improvement, along with a positive effect to treat several neurological disorders (Cavanagh and Wilkinson, 2002; Caputo et al., 2016; López et al., 2017; Rahmati et al., 2017). The highest quality lavender oil comes from L. angustifolia (syn. L. spica L., L. officinalis Chaix., L. vera DC.) inflorescences, which contain high levels of linalyl acetate and linalool and low amount of camphor, considered positive features by both cosmetic and pharmaceutical industry (Cavanagh and Wilkinson, 2002; Kim and Lee, 2002; Shellie et al., 2002). Although the genus Lavandula has been widely studied in wild and controlled environment (Da Porto and Decorti, 2008; Da Porto et al., 2009; González-Coloma et al., 2011; Gonçalves and Romano, 2013; Pistelli et al., 2013, 2017; Hassiotis et al., 2014; Lesage-Meessen et al., 2015; Caputo et al., 2016; Kirimer et al., 2017; López et al., 2017; Rahmati et al., 2017), lavender aroma profile and its EO composition are continuously subjected to many investigations (Prusinowska and Smigielski, 2014).

Secondary metabolites from plants are involved in several functions such as attraction for pollinators, signaling between plants, defense and protection against biotic and abiotic stresses (Knudsen et al., 2006; Bakkali et al., 2008; Dudareva et al., 2013; Raut and Karuppayil, 2014; Nogués et al., 2015; Caser et al., 2016, 2017). The plant-environment interaction largely contributes to the phytochemicals expression (Kesselmeier and Staudt, 1999; Binns et al., 2002; Dudareva et al., 2013; Holopainen et al., 2013; Selmar and Kleinwächter, 2013; Loreto et al., 2014), and has already been proved to affect the chemical composition of lavender (Prusinowska and Smigielski, 2014). However, the variability in the emission of volatile organic compounds (VOCs) and in the EO biosynthesis is also evidently regulated by genetics, even though the mechanisms involved still need clarifications (Kesselmeier and Staudt, 1999; Sangwan et al., 2001; Brahmkshatriya and Brahmkshatriya, 2013). The intraspecific variation in secondary metabolite production in plants has been reviewed and explained by genetic drift, relaxed selective pressure, introgression of traits through hybridization, gene pleiotropic effects, and phenotypic plasticity (Eckert et al., 2008; Raguso, 2008; Dicke and Loreto, 2010). For instance, qualitatively and quantitatively differences in secondary metabolites have been reported not only between wild and cultivated plants (L. luisieri, González-Coloma et al., 2011), but also among plants of the same wild population, as described in lavandin of northwest Italy (Maffei and Peracino, 1993; Peracino et al., 1994) and in L. latifolia (Muñoz-Bertomeu et al., 2007).

The populations of plants that are geographically situated at the peripheral limit of a species distribution range are usually expected to exhibit lower abundance and intra-genetic diversity, according to the widely accepted "abundant center model." Moreover, a higher genetic differentiation than populations at the center is likely to occur (Eckert et al., 2008). Altitudinal gradient was also proved to affect species diversification in peripheral populations of L. latifolia (Herrera and Bazaga, 2008). Unfortunately, intra-specific studies on secondary metabolites production performed in plants in different environmental conditions do not inform whether the observed variations represent the genetic adaptation to specific environments, or the short-term response to environmental factors (Spitaler et al., 2006). Thus, trials under uniform cultivation practices are preferable to limit environmental influence on plant secondary metabolite production.

This study aimed to investigate if the environment differences, such as latitude and altitude, affected peripheral population diversification of L. angustifolia. Morphology, VOCs, and EOs were analyzed in lavender plants collected along both latitudinal and altitudinal gradients in West Alps (Italy), and grown in pots under uniform conditions, looking for traits of productive interest.

#### MATERIALS AND METHODS

## Native Environment

Lavender (L. angustifolia) is widespread throughout Italy (Pignatti, 1982), where it is a common component of lowgrowing shrub vegetation on calcareous soils. Lavandula angustifolia finds its phytosociological optimum in the sub-Mediterranean region within the order Ononidetalia, that describes meso-xerophile, basophile, supra-oromediterranean vegetation communities (Theurillat et al., 1995). It is the only species naturally occurring in the West Italian Alps (Piedmont region). Nine peripheral populations (**Figure 1**) were selected according to the natural range of distribution of the species (Pignatti, 1982), in order to explore different ecological conditions (**Table 1**, Supplementary Figure 1) in terms of latitude (Susa Valley in the northern range, Stura Valley in the core areas, and Tanaro Valley in the southern part of the Piedmont region) and altitude (low, medium and high altitude). Susa Valley represents the highest latitudinal range of distribution of the

**Abbreviations:** AC, Apocarotenoids; EO, Essential Oils; GC-MS, Gas Cromatography-Mass Spectrometry; GI, Growth Index; MH, Monoterpene Hydrocarbons; NT, Non Terpene Derivatives; OM, Oxygenated Monoterpenes; OS, Oxygenated Sesquiterpenes; SH, Sesquiterpene Hydrocarbons; SPME, Solid Phase Micro Extraction; VOCs, Volatile Organic Compounds.

species in the Alps (Aeschimann et al., 2004). As the latitude decreases (from Susa to Tanaro Valley), the annual rainfalls decrease and yearly temperatures increase, thus there is a clear ecological gradient moving toward the typical Mediterranean conditions (Fick and Hijmans, 2017).

## Plant Material and Pot Cultivation

Portions of lavender branches were collected from wild plants during the flowering season of 2014, from August to September. In each of the nine selected sites, the main area of lavender distribution was identified as a central point and from there, an area with a radius of 10 m was plotted. Inside that area, 10 individuals were randomly chosen and twenty cuttings were immediately prepared from each individual. Specimens are currently available and cultivated in the DISAFA facilities (45◦ 03′ 58.5′′N; 7◦ 35′ 29.1′′E, WGS84 System). Cuttings were allowed to root in peat substrate under plastic tunnels in an organic nursery, specialized in medicinal and aromatic plants (Frat.lli Gramaglia, Collegno, Italy; 45◦ 05′ 22.4′′N, 7◦ 34′ 26.4′′E, 302 m. a.s.l.). In spring 2015, the percentage of successful rooting was calculated and the rooted plants were transplanted in pots, with a mixture of peat and green compost (70-30, % v/v) and transferred outdoors. Irrigation was provided when needed and manual weed control was performed. Plants were fertigated with NPK fertilizer (1:2:2) three times during spring and once during autumn. The cultivation cycle lasted two years and during the first one (Summer 2015), non-destructive VOC analysis was performed in vivo on flowered pot plants to characterize lavender aroma and evaluate their potential differences. As ecological characteristics of the sampling site did have an influence on lavender performance, during the second year (Summer 2016), biometric parameters and EOs were analyzed.

# Plant Performance Evaluation

Biometric and morphological characteristics of 123 rooted flowered plants were evaluated once a week from the beginning of blooming, according to and implementing the guidelines of the International Union for the Protection of New Varieties of Plants (UPOV) proposed for lavenders. The selected parameters were: number of spikes per plant (n), spike length (cm), flowering time (% of flowered plants in each population per week), together with plant height (cm) and width (cm), used to calculate the Growth Index (i.e., the plant volume; cm<sup>3</sup> ) according to Demasi et al. (2017). The survival percentage was evaluated at the end of the second cultivation cycle.

# VOCs Detection and Analysis

Emitted volatiles were analyzed using a Supelco (Bellofonte, USA) SPME device coated with polydimethylsiloxane (PDMS, 100µm) in order to sample the headspace of each living/flowering plant. Plants of StuM population did not bloom, thus data on VOCs concerned eight populations, for a total of 33 plants. Sample was introduced individually into a 30 ml glass conical flask and allowed to equilibrate for 30 min. After the equilibration time, the fiber was exposed to the headspace for 15 min at room temperature; once sampling was finished, the fiber was withdrawn into the needle and transferred to the injector of the GC system, where the fiber was desorbed. GC-FID and GC-MS analyses were performed according to Pistelli et al. (2017) using a Varian CP-3800 apparatus (Paolo Alto, California, USA) equipped with a DB-5 capillary column (30 m × 0.25 mm i.d., film thickness 0.25µm) and a Varian Saturn 2000 ion-trap mass detector. The oven temperature was programmed rising from 60◦C to 240◦C at 3◦C/min; injector temperature, 220◦C; transfer-line temperature, 240◦C; carrier gas, He (1 ml/min).

# Essential Oil Extraction and Analysis

Essential oil of lavender plants was obtained by hydro distillation of dried aerial parts of each sample using a Clevenger-type apparatus according to the Italian Pharmacopoeia (Helrich, 1990). Flowered plants of StuM population did not produce enough material to obtain EO, thus data on EOs concerned eight populations, for a total of 122 plants. The obtained EOs were immediately injected in GC-FID and GC-MS according to the method described in Pistelli et al. (2017). Identification of the constituents was based on the comparison of the retention times with those of authentic samples in comparing their linear retention indices (LRI) relative to a series of n-hydrocarbons, and on computer matching against commercial (Adams, 2007; NIST/EPA/NIH Mass Spectral Library, 2014) and home-made mass spectra library built up from pure substances and components of known EOs as well as MS literature data.

# Statistical Analyses

Data were tested for the homogeneity of variance, then one-way ANOVA was performed to analyze latitude and altitude effect on biometric parameters and means were separated according to Ryan-Einot-Gabriel-Welsch-F post-hoc test (REGW-F). Statistically significant differences induced by latitude on chemical classes and compounds of VOCs and EO were assessed with the one-way ANOSIM with Euclidean-based similarity. The percentage contribution of each compound to the observed dissimilarity was assessed through the similarity percentage analysis (SIMPER, Euclidean distance). For each


TABLE 1 | Geographical characteristics and climatic conditions of *L. angustifolia* sampling sites in Piedmont region (West Italian Alps).

compound the difference between valleys were tested with Mann-Whitney pairwise test. Besides, VOC and EO chemical classes were subjected to the regression analysis to investigate their eventual link to altitude or latitude of native environment. The value for statistical significance was P < 0.05. The ANOVA analysis was performed with SPSS software (version 22), while all other analyses were performed with Past software (version 3).

#### RESULTS AND DISCUSSION

#### Morphology and Performance of Cultivated Lavender

Lavenders of West Alps differently performed under cultivation, and influence of sea distance (latitude) or altitude of the native environment was observed in some biometric parameters (**Table 2**). Rooting percentage after 3 months in seed cells was generally below 40% and ranged among populations (SusH 23.9%, SusM 17.0%, SusL 38.0%; StuH 15.8%, StuM 9.6%, StuL 35.3%; TanH 37.5%, TanM 23.0%, TanL 39.5%). Rooting ability was therefore generally low, especially considering that selected lavender cultivars often reach 95% of successful rooting (Lis-Balchin, 2003). However, the propagation of wild adult plants by cutting is needed to produce genetically homogeneous individuals and the rooting percentage achieved in this study on L. angustifolia is considered of interest by local nurseries. Significant effects were observed due to both altitude and latitude of native environment, with a more successful rooting in plants evolved closer to the Mediterranean conditions (Tanaro) and from lower altitudes. The survival rate of rooted cuttings after 2 years of cultivation ranged between 54% (SusL) and 83% (StuL), but was independent from the geographical origin of the plants (**Table 2**). During the first cultivation cycle, 9% of pot lavenders bloomed, while in the second year, all plants bloomed (**Figure 2**); the first flowered plants were recorded on June 6th 2016 and the last on August 19th 2016. Interestingly, almost all lavenders of high populations flowered simultaneously in 1 week (SusH 100%, StuH 100%, and TanH 96%) regardless the latitude, followed by medium and low altitude populations, with the latter group defining a more gradual trend during summer. Knowledge of flowering times is essential to plan nursery activities (Lis-Balchin, 2003) or harvesting time for the production of EOs or dried flowers. Indeed, the nurseries can exploit lavenders with a flower production concentrated in a short period, since higher number of plants are available per sale. Furthermore, producers can optimize flower harvesting in the field when a simultaneous blooming occurs. Nonetheless, a gradual blooming can ensure flowered plants availability in the nursery and gardens for a longer period.

Altitude affected also morphological parameters and flower yield (**Table 2**). Contrariwise to low altitude groups, plants from higher elevation developed a more compact shrub (GI = 2,003 cm<sup>3</sup> ), with higher number (11.26) and longer spikes (37.8 cm) during cultivation, but producing lower yield of inflorescences (1.16 gDW/plant), attributable to a less number of flowers per spike. Altitudinal gradient effect has been previously observed in other Lamiaceae species, displaying lower height at high altitudes, to avoid wind damages and improve photosynthetic conditions (Kofidis and Bosabalidis, 2008 and references in it). However, compact plants can facilitate field operations, especially harvesting. Growth and number of flower spikes were influenced also by the population distance from the sea. Plants from Susa Valley showed a compact growth (GI = 1,886 cm<sup>3</sup> ) and a low number of flower spikes (5.4), contrary to what observed in Stura plants (GI = 3,706 cm<sup>3</sup> and 10.6). Tanaro plants were compact (GI = 2,470 cm<sup>3</sup> ) and produced a high number of spikes (8.8), showing an interesting habit for ornamental and field cultivation purposes. A latitudinal gradient in spike length and flower yield were not detected in the second cultivation cycle. These data could be helpful in the selection of alpine peripheral L. angustifolia populations with interesting traits for nurseries both for ornamental and production purposes.

#### Influence of Native Environment on Volatile Emission

The aroma profile represents the volatiles spontaneous emitted from the living plants and is a highly complex component of floral phenotype (Raguso, 2008). High intra and interspecific variations in secondary metabolite production and aroma composition were reported in Lavandula genus, including L. latifolia and L. luisieri (Sanz et al., 2004; Muñoz-Bertomeu


TABLE 2 | Differences in rooting percentage, survival rate, growth index (GI), spike number and spike length recorded in the cultivated lavenders, according to altitude (High, Medium, and Low) and latitude (Susa, Stura, and Tanaro) of the native environment.

§*Means followed by the same letter in the same column denote no significant differences according to REGW-F test (P* < *0.05). ns,*\**,*\*\* *or* \*\*\* *indicates non-significant differences or significant at P* ≤ *0.05, 0.01, or 0.001, respectively.*

et al., 2007; González-Coloma et al., 2011) and other ornamental plants, such as rose, magnolia, and butterfly bush (Azuma et al., 2001; Raguso, 2008; Gong et al., 2014).

altitudes; dotted lines, lower altitudes.

The volatile chemical classes of lavender fragrance were mainly represented by terpenes, with low amount of non-terpene derivatives (NT, 1.1–6.1%) and apocarotenoids (AC, 0.1–0.5%) (**Figure 3**, Supplementary Table 1). Oxygenated monoterpenes were the most abundant VOCs in all the populations (OM, 61.2–78.1%), followed by monoterpene hydrocarbons (MH, 8.9–22.4%), sesquiterpene hydrocarbons (SH, 4.8–20.5%), and oxygenated sesquiterpenes (OS, 0.1–1.9%). Interestingly, SH were generally present in higher amounts in the aroma of the lowest-altitude populations (SusL 15.6%, StuL 18.3%, TanL 20.5%), and AC were not detected in SusM, SusL, and in TanM populations. One-way ANOSIM performed on chemical classes showed that populations far from the sea (Susa) significantly differed from the closest ones (Tanaro) in their phytochemical composition (p = 0.038), indicating a clear distinction of aroma profiles. Moreover, a linear negative regression fitted between latitude and MH and a positive with OM (**Figures 4A,B**), whereas a negative regression model fitted between SH and altitude (**Figure 4C**). Monoterpenes are high volatile terpenoids emitted with warm temperature (>20◦C). Their amount in flower fragrance is variable and often exceeds 50% of the total flower emission (Holopainen et al., 2013), as found in our study. Sesquiterpenes are the most diverse class of terpenoids (Holopainen et al., 2013); they are more abundant in plant

volatiles than oils and are effectively used in aromatherapy (Cavanagh and Wilkinson, 2002), giving a sweet aromatic note (Prusinowska and Smigielski, 2014). Therefore, altitude influence

on sesquiterpenes content found in this study could be an interesting starting point to explore the selection of fragrance characteristics.

Considering single compounds, 88 VOCs were identified in lavender aroma profile (Supplementary Table 1). The identified compounds accounted for more than 97% of the total composition. The number of compounds in each population varied from 42 (TanM) to 62 (StuL). Interestingly, every population had a peculiar aroma composition and all populations shared only 27 compounds. Among these compounds, the sum of linalyl acetate, linalool and 1,8-cineole were present more than 50% in SusM and SusL populations and more than 40% in the others; 1,8-cineole was less abundant in TanM and TanL; 4-terpineol and (E)-β-ocimene occurred in high percentages only in TanM lavender (7.7% and 7.5%, respectively), which showed also higher amount of lavandulyl acetate (4.5%). Linalool (22.9%), β-caryophyllene (13.9%), as well as borneol (6.7%) evidenced the highest amount in TanL. Da Porto and Decorti (2008) found a different arrangement in VOCs of lavender cultivated in Northeast Italy, in particular with 1,8-cineole, linalool, camphor, linalyl acetate, and β-caryophyllene being the 70% of the total aroma. Moreover, (E)-β-ocimene was present in smaller amounts (0.2–0.7%) compared to our study (1.6–7.5%).

The one-way ANOSIM performed using VOCs composition showed a similar trend to what observed in chemical classes, with a clear differentiation between Susa and Tanaro populations (p = 0.018), respectively the farthest and the closest sites to Mediterranean conditions. According to the SIMPER analysis, 11 compounds were responsible for the 95% of this dissimilarity in lavender fragrance (**Table 3**), with 1,8-cineole, linalyl acetate and linalool as major contributors (50.65, 16.33, and 13.16% of the contribution, respectively). The effect of sea distance was particularly seen on five out of 11 compounds. In detail, linalool, (E)-β-ocimene and 4-terpineol increased in the populations closer to the Mediterranean Sea, whereas 1,8-cineole and bornyl acetate had an opposite behavior. These latter constituents usually adversely affect lavender fragrance (Prusinowska and Smigielski, 2014), while linalool is responsible for the fresh and floral smell (Chizzola, 2013; Prusinowska and Smigielski, 2014). It is one of the best-examined and most often reported monoterpene in flowers (Buchbauer and Ilic, 2013; Holopainen et al., 2013). Moreover, there are evidences that linalool acts synergistically with linalyl-acetate and both compounds are required for the anxiolytic effect of the inhaled EOs (Buchbauer and Ilic, 2013). Thereof peripheral populations adapted to Mediterranean ecological conditions (Tanaro) appear more interesting for the selection of fragrance genotypes.

Both isomers of β-ocimene, together with linalool, are frequently found in species pollinated by butterflies, bees, and moths (Andersson, 2003; Dötterl and Schäffler, 2007; Gong et al., 2014). TanM population emitted the highest quantity of both isomers [(Z)-β-ocimene, 6.4% and (E)-β-ocimene, 7.5%] and could have therefore differentiated its VOC emission to attract its peculiar type of pollinators (Dudareva et al., 2013).

Qualitative and quantitative variation was found in this study in the fragrance composition of L. angustifolia species according to its distance from the sea, while Stark et al. (2008) found only qualitative differences concerning other secondary metabolites (flavonoids and tannins). Several phenological events and environmental parameters may drive intra-specific floral scent variation, as observed also between and within Phlox cultivars (Majetic et al., 2014), highlighting that species contain a great potential of variation. Nonetheless, the uniform cultivation conditions adopted in our trial pointed out that variation in lavender fragrance composition could be partly attributed to its genetic adaptation to ecological conditions of peripheral alpine sites.

## Influence of Native Environment on EO Composition

Generally, EO yields were consistent with previous studies on L. angustifolia (Da Porto et al., 2009; Prusinowska and Smigielski, 2014) which are very low in SusH and TanM to reach 0.69% in SusL (Supplementary Table 2). Plants from the lowest altitudes


TABLE 3 | List of VOCs responsible for dissimilarity induced by latitude in lavender aroma according to the SIMPER analysis.

§*Data followed by the same letter in the same line are not significantly different according to Mann-Whitney test. ns,* \**,* \*\* *or* \*\*\* *indicates non-significant differences or significant at P* ≤ *0.05, 0.01 or 0.001, respectively.*

showed higher yields in Susa and Tanaro valley (0.69 and 0.37%, respectively). Previous studies highlighted the same behavior in other aromatic plants such as Origanum vulgare (Vokou et al., 1993; Giuliani et al., 2013), Coriandrum sativum (Shams et al., 2016), and Rosmarinus officinalis (Tuttolomondo et al., 2015). Giuliani et al. (2013) found a correlation between higher yields at decreasing altitudes and, but not limited to, an increased presence of glandular peltate trichomes, the structures responsible for the accumulation of EOs. Considering distance from Mediterranean conditions, Susa lavenders were the most productive in terms of EO, followed by Tanaro and Stura plants.

The mean of the relative abundance of chemical classes was distributed as follows (**Figure 5**, Supplementary Table 2): OM (71.2%), OS (13.7%), SH (6.1%), NT (4.9%), MH (3.2%), and AC (0.3%). A different distribution was recorded by Pistelli et al. (2017) in L. angustifolia 'Maillette', where the amount of OS and SH were lower than that observed in this study (0.5 and 1.8%, respectively). Commonly, monoterpenes represent the major class in the oil of Lavandula species (Lis-Balchin, 2003) and our data are consistent with values reviewed by Prusinowska and Smigielski (2014). In this study, significant positive linear regression fitted between MH and NT concentration with latitude (**Figures 6A,B**) and AC content with altitude (**Figure 6D**); instead, a negative linear regression fitted between SH and latitude (**Figure 6C**), an opposite behavior than that observed in Teucrium polium EO by Sadeghi et al. (2014).

One-hundred-four compounds were identified in lavender EO (Supplementary Table 2), accounting for more than 99% of the total EO composition. The number of compounds in each population varied from 63 (SusH) to 83 (StuL). These numbers were higher than those previously reported in L. angustifolia species and cultivars cultivated in the Tyrrhenian coast (Venskutonis et al., 1997; Pistelli et al., 2017). Interestingly, each EO was a mixture of different compounds, but all populations shared at least 40 constituents. The EO composition of L. angustifolia is reported to be the most variable inside the genus (Lis-Balchin, 2003). However, the characteristic compounds found in Lavandula oil usually are

FIGURE 5 | Chemical classes composition (%) of the *L. angustifolia* essential oils after two cultivation cycle in uniform conditions. H, high altitude; M, medium altitude; L, low altitude; MH, monoterpene hydrocarbons; OM, oxygenated monoterpenes; SH, sesquiterpene hydrocarbons; OS, oxygenated sesquiterpenes; NT, non terpene derivatives; AC, apocarotenoids.

linalool, linalyl acetate, 1,8-cineole, β-ocimene, 4-terpineol, and camphor (Cavanagh and Wilkinson, 2002). Accordingly, in the present study linalool and linalyl acetate were the most abundant in all the EOs analyzed, with the highest amount recorded in TanL (linalool 35.0%) and StuH (linalyl acetate 26.3%). However, linalool content was present in lower percentage than in previous studies on lavender (Venskutonis et al., 1997; Pistelli et al., 2013, 2017). In fact, linalool production in L. angustifolia varies widely (Prusinowska and Smigielski, 2014) depending on temperature, flower development, and rainfall (Hassiotis et al., 2014). Linalyl acetate content registered in this work is in accordance with that reported for lavender 'Maillette'



§*Data followed by the same letter in the same line are not significantly different according to Mann-Whitney test. ns,*\**,*\*\* *or*\*\*\* *indicates non-significant differences or significant at P* ≤ *0.05, 0.01 or 0.001, respectively.*

(Pistelli et al., 2017), but lower than in lavender species grow in Lithuania (Venskutonis et al., 1997). According to the percentage content of the characteristic compounds of lavender essential oil reported in the European Pharmacopoeia 6.0 (Ph. Eur.) none of the analyzed EOs completely fulfill the expected limits. StuH and TanL were closely related to the established ranges except for linalool (less than 20%) in StuH and linalyl acetate (less than 25%) in TanL. The percentages of 4-terpineol, 3 octanone, limonene, and lavandulyl acetate were in the range fixed by Ph. Eur. for all the tested EOs. Among the compounds reported in ISO 3515:2002 for lavender EO<sup>1</sup> only 3-octanone, limonene and lavandulyl acetate respected the fixed limits in all the analyzed EOs. The analysis of flower fragrance has been suggested as a good indicator of the EO composition (An et al., 2001). However, EO composition frequently differed from the aroma profiles (see section Influence of Native Environment on Volatile Emission), as already highlighted in several studies on lavenders (Cavanagh and Wilkinson, 2002). These differences were caused by processes such as hydrolysis, thermal degradation, and molecular rearrangements that are promoted during the hydro distillation method (Da Porto and Decorti, 2008).

The one-way ANOSIM performed on EO compounds revealed significant differences among populations at different latitudes (p = 0.009), likewise to what observed in lavender aroma (see section Influence of Native Environment on Volatile Emission), with a clear separation between populations with more (Tanaro) and less (Susa) Mediterranean ecological conditions. According to the SIMPER analysis, thirteen compounds were responsible for 95% of this difference (**Table 4**), with linalool being the major contributor (45.98%). The effect of sea distance was seen on nine out of the 13 compounds. In particular, linalool, 4-terpinenol, β-caryophyllene, thujapsan-2 α-ol, and (Z)-γ-bisabolene were higher in populations closer to the Mediterranean Sea (Tanaro), while 1,8-cineole, τ -cadinol, 1-octen-3-yl acetate and α-terpineol where higher in the furthest populations (Susa). Linalool and terpineol are of great interest for medicinal industry due to their positive effects on the central nervous system (Prusinowska and Smigielski, 2014). β-caryophyllene, with its woody and spicy aroma, is commonly used in the fragrance and cosmetic industry. However, it has also antibiotic, anesthetic, anti-inflammatory, antioxidant, and anti-spasmodic activity (Buchbauer and Ilic, 2013). Therefore,

<sup>1</sup> ISO 3515:2002 Oil of Lavender (Lavandula angustifolia Mill.).

similarly to results on fragrance, the populations closer to Mediterranean conditions (Tanaro) could be considered valuable germplasm for selection and breeding purposes, aimed to the production of EO, despite the low yields obtained. Indeed it is needed to consider that the cultivation occurred in pot. In proper open field conditions higher biomass and EO yields are expected. No significant differences were detected in EO compounds according to altitude, likewise to VOCs composition (see Section Influence of Native Environment on Volatile Emission), despite there are evidence that altitude significantly affected chemical composition of EO in other Lamiaceae plants (Giuliani et al., 2013; Sadeghi et al., 2015). Chograni et al. (2010) observed also a phytochemical variation in the leaf EO of wild L. multifida belonging to the same bioclimatic zone, possibly explained by genetic factors. Similarly, in the present study, the ecological conditions of peripheral sites may have induced different adaptation mechanisms in lavender, leading to different phytochemical compositions. Since the EO obtained in this work did not respect the limit established by Ph. Eur., they can be used for other industrial application, or for cosmetic purposes.

#### CONCLUSIONS

We hereby provide the first phytochemical description of secondary metabolites in L. angustifolia peripheral populations of the West Italian Alps. The influence of sea distance (latitude) and altitude on VOC and EO composition was analyzed in plants grown under uniform cultivation conditions. Interestingly, results showed that the ecological gradient created germplasm heterogeneity. Compared to other studies, altitude seemed to affect mainly biometric parameters, and to a low extent the phytochemical composition of L. angustifolia, while sea distance had an influence on both morphological and phytochemical traits. Lavenders of West Italian Alps disclosed a great potential for the development of a valuable local product, where the cultivation of the thermophilous species L. × intermedia cannot be applied due to the lower average temperature of the studied alpine areas. Among the studied alpine peripheral populations, plants of Tanaro Valley, which have evolved in almost

#### REFERENCES


Mediterranean ecological conditions, generally performed better in both morphological and phytochemical characteristics. To obtain uniform productions, the study of the phytochemical variation and the selection and propagation of interesting genotypes must be promoted, together with the improvement of cultivation protocols.

#### AUTHOR CONTRIBUTIONS

SD and MC equally contributed to this work in the acquisition of data, analysis and interpretation of data, and drafting of manuscript. BN and PC carried out phytochemical analyses. ML contributed to lavender population selection and data analysis. LP and BN contributed to data interpretation and discussion. VS conceived, designed and coordinated the work. LP and VS critically revised and approved the manuscript.

#### FUNDING

This work was partially supported by Fondazione Cassa di Risparmio di Torino (Project Le colture officinali: un approccio di filiera per aumentare la competitività delle aziende piemontesi e valorizzare il territorio, grant number 2014.0976) and by the program Interreg V-A Francia Italia Alcotra Attività innovative per lo sviluppo della filiera transfrontaliera del fiore edule–Antea n. 1139.

#### ACKNOWLEDGMENTS

The authors thank Walter Gaino and Paolo Lo Turco for their contribution in the collection and propagation of lavender plants; and Frat.lli Gramaglia nursery for hosting cultivation.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2018. 00983/full#supplementary-material


phytochemical parameters in Helichrysum petiolare Hilliard and BL Burtt. Ind. Crops Prod. 83, 680–692. doi: 10.1016/j.indcrop.2015.12.053


Pignatti, A. (1982). Flora d'Italia. Bologna: Edagricole.


in flowering heads of Arnica montana cv. ARBO Phytochemistry 67, 409–417. doi: 10.1016/j.phytochem.2005.11.018


**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 Demasi, Caser, Lonati, Cioni, Pistelli, Najar and Scariot. 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(s) 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.

# NMR Metabolomics Defining Genetic Variation in Pea Seed Metabolites

Noel Ellis 1,2,3, Chie Hattori <sup>1</sup> , Jitender Cheema<sup>1</sup> , James Donarski <sup>4</sup> , Adrian Charlton<sup>4</sup> , Michael Dickinson<sup>4</sup> , Giampaolo Venditti <sup>4</sup> , Péter Kaló<sup>5</sup> , Zoltán Szabó<sup>5</sup> , György B. Kiss <sup>6</sup> and Claire Domoney <sup>1</sup> \*

<sup>1</sup> John Innes Centre, Norwich, United Kingdom, <sup>2</sup> IBERS, Aberystwyth University, Aberystwyth, United Kingdom, <sup>3</sup> Faculty of Science, School of Biological Sciences, University of Auckland, Auckland, New Zealand, <sup>4</sup> Fera Science Ltd., York, United Kingdom, <sup>5</sup> National Agricultural Research and Innovation Centre, Agricultural Biotechnology Institute, Gödöllo,˝ Hungary, <sup>6</sup> AMBIS Biotechnology Ltd., Budapest, Hungary

Nuclear magnetic resonance (NMR) spectroscopy profiling was used to provide an unbiased assessment of changes to the metabolite composition of seeds and to define genetic variation for a range of pea seed metabolites. Mature seeds from recombinant inbred lines, derived from three mapping populations for which there is substantial genetic marker linkage information, were grown in two environments/years and analyzed by non-targeted NMR. Adaptive binning of the NMR metabolite data, followed by analysis of quantitative variation among lines for individual bins, identified the main genomic regions determining this metabolic variability and the variability for selected compounds was investigated. Analysis by t-tests identified a set of bins with highly significant associations to genetic map regions, based on probability (p) values that were appreciably lower than those determined for randomized data. The correlation between bins showing high mean absolute deviation and those showing low p-values for marker association provided an indication of the extent to which the genetics of bin variation might be explained by one or a few loci. Variation in compounds related to aromatic amino acids, branched-chain amino acids, sucrose-derived metabolites, secondary metabolites and some unidentified compounds was associated with one or more genetic loci. The combined analysis shows that there are multiple loci throughout the genome that together impact on the abundance of many compounds through a network of interactions, where individual loci may affect more than one compound and vice versa. This work therefore provides a framework for the genetic analysis of the seed metabolome, and the use of genetic marker data in the breeding and selection of seeds for specific seed quality traits and compounds that have high commercial value.

Keywords: genetic map, genetic variation, pea, seed, metabolite, nuclear magnetic resonance

# INTRODUCTION

Metabolite profiling, based on chemical fingerprints provided by nuclear magnetic resonance (NMR) spectroscopy, provides an approach for the unbiased assessment of changes in the content of small molecules in response to genetic and/or environmental factors. Such profiles provide a useful and rapid method for assessing the changes that occur in the metabolome as a consequence

#### Edited by:

Marta Sousa Silva, Universidade de Lisboa, Portugal

#### Reviewed by:

Fabio Sciubba, Sapienza Università di Roma, Italy Matthew Paul, Rothamsted Research (BBSRC), United Kingdom

> \*Correspondence: Claire Domoney claire.domoney@jic.ac.uk

#### Specialty section:

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

> Received: 27 February 2018 Accepted: 25 June 2018 Published: 17 July 2018

#### Citation:

Ellis N, Hattori C, Cheema J, Donarski J, Charlton A, Dickinson M, Venditti G, Kaló P, Szabó Z, Kiss GB and Domoney C (2018) NMR Metabolomics Defining Genetic Variation in Pea Seed Metabolites. Front. Plant Sci. 9:1022. doi: 10.3389/fpls.2018.01022

**33**

of plant genotype and/or the interaction between genotype and environment (Messerli et al., 2007). The use of NMR spectroscopy for holistic studies of plant metabolism predates the term "metabolomics" (Fiehn et al., 2000) by some years (Moore et al., 1983; Belton and Ratcliffe, 1985; Ratcliffe, 1987; Fan, 1996).

NMR spectroscopy provides a method of choice to facilitate the efficient analysis of the large number of samples that is necessary to deal with the expected intrinsic variability of plant, or equivalent, biological materials particularly where these need to be grown or cultured under field or similar "near-natural" conditions. Such has been the case for the study of "substantial equivalence" in genetically modified plants, where NMR has been used in the analysis of field samples of wheat (Baker et al., 2006). Higher amounts of maltose and/or sucrose and differences in free amino acids were apparent in a transgenic line, and these observations were followed by more detailed studies of the amino acid composition using gas chromatography-mass spectrometry (GC-MS). NMR has also been employed to evaluate the effects of genetic modification and assess the effect of drought-stress on the Pisum sativum L. (pea) leaf metabolome (Charlton et al., 2004, 2008). Significant changes in resonances under droughtstress conditions were attributed to a range of compounds, both primary and secondary metabolites, including proline, valine, threonine, homoserine, myoinositol, γ -aminobutyrate (GABA) and trigonelline (nicotinic acid betaine). Some of these changes translated to alterations in the seed metabolome in the same experiments (unpublished data).

It has been shown, using GC/MS analyses of Arabidopsis developing seeds, that the seed desiccation period is associated with a major increase in the levels of free metabolites; these include the nitrogen-rich amino acids (asparagine, lysine, and arginine), the aromatic amino acids (tryptophan, phenylalanine, tyrosine), serine, alanine, the non-proteinogenic amino acid GABA, TCA-cycle intermediates, fumarate and succinate, and the levels of sucrose, galactose, arabinose, trehalose, sorbitol, galactitol, gluconate-6-phosphate and glycerate (Fait et al., 2006). Few studies have been carried out to investigate the effects of genetic variation on the metabolite composition of seeds. For the seeds of many crops, quality traits may be defined in terms of the synthesis of a number of key metabolites, for example the concentration of 2-acetyl-1-pyrroline (2AP) in rice linked to fragrance quality (Shi et al., 2008). An alternative to the expensive and time-consuming GC/MS method for assaying 2AP content in breeding programmes is offered by the demonstration that the metabolite is controlled by a gene, betaine aldehyde dehydrogenase, for which allelic variation has been described (Shi et al., 2008).

In pea, the molecular basis for many seed quality traits is largely unknown. An exception to this is the understanding of the control of sucrose content at a gross level, where naturally occurring mutants with defects in starch biosynthesis have elevated sucrose contents in their seeds. Mutations at two genetic loci (r and rb) have been exploited in the development of some of the varied food uses of pea seeds (Wang et al., 1998). Studies have shown the many pleiotropic effects that mutations at r and rb exert on seed metabolism overall; these include changes to nitrogen/protein accumulation, water content and seed shape when compared with wild-type lines (Perez et al., 1993; Casey et al., 1998; Lyall et al., 2003). These alterations to seed composition can be mimicked to similar or greater extents in mutants induced either through chemical mutagenesis or transgenesis, where additional genes of starch biosynthesis have been targeted (Wang and Hedley, 1991; Wang et al., 1998; Weigelt et al., 2009). While the r and rb loci are determined by mutations in a starch-branching enzyme and the large subunit of ADP-glucose pyrophosphorylase (AGPase), respectively, (Bhattacharyya et al., 1990; Hylton and Smith, 1992), transgenic lines of pea expressing RNAi constructs targeting the small subunit(s) of AGPase have shown a very similar phenotype, when compared with wild-type lines (Weigelt et al., 2009).

In many legume species, oligosaccharides derived from galactinol and sucrose are synthesized in seeds. In pea, these comprise the raffinose oligosaccharide (RFO) group of compounds, which include stachyose and verbascose in addition to raffinose. Quantitative and qualitative variation for these compounds has been described for pea, lentil and Medicago (Frias et al., 1994, 1999; Karner et al., 2004; Vandecasteele et al., 2011). Although RFOs are generally regarded as anti-nutrients in seeds, research in Medicago suggests that these compounds are related to seed vigor (Vandecasteele et al., 2011), while additional studies highlight their role in plant stress responses (Nakabayashi and Kazuki, 2015).

The aim of this study was to determine the extent to which variation in the metabolome of mature seeds was under genetic control and to investigate the main types of compounds involved in such regulation. This information could be used further to identify genotypes that are enriched in particular seed components, some of which may be associated positively or negatively with quality and/or health-promoting traits. Given the knowledge of the impact of the allelic state at the r and rb loci (above) and the variation that exists within these genotypes with respect to seed maturation, we sought to assess the extent of metabolome variation within r and rb genotypes of pea. In this paper, we define a metabolite phenotype for seeds from genetically marked recombinant inbred r and rb mutant lines. We describe variation within the metabolome of mature seeds from the recombinant inbred lines, for which we provide substantial genetic marker information and a framework for the analysis of metabolite data in relation to genetic loci and markers. Furthermore, for some of the identified metabolites, candidate genes have been identified for the control of metabolite content.

# MATERIALS AND METHODS

#### Plant Materials

A selection of recombinant inbred lines (RILs) from three mapping populations (JI 281 × JI 399, 32 lines; JI 15 × JI 399, 38 lines; JI 15 × JI 1194, 26 lines) and their parent lines (all available from the JIC Pisum germplasm collection; https://www.seedstor. ac.uk/search-browseaccessions.php?idCollection=6) were grown in microplots (1 m<sup>2</sup> ) at two locations, John Innes Centre, Norwich (JIC) and at the Processors & Growers Research Organization, Peterborough (PGRO), over two consecutive seasons (Year 1, 2011 and Year 2, 2012). The lines comprise 100 variant vining seed genotypes (either r or rb mutants), derived from crosses that have integrated genetic maps and are densely populated with genetic markers (Supplementary File S1). The JI 1194 parent is r, JI 399 is rb, and JI 281 and JI 15 are wild type for both r and rb.

Seeds were treated with Wakil seed treatment and sown directly into plots in bird-proof cages in the spring (March). Plants were irrigated and sprayed for protection against aphids as necessary. Mature (senesced) plants and their seeds were harvested together in July. Seeds were threshed and hand-picked to remove any foreign objects, while phenotype checks ensured the identity, integrity and purity of the genetic stock. From these, seed aliquots (approximately 6g) were prepared for NMR metabolite analysis.

#### NMR Analysis

The NMR profiles of Year 1 and Year 2 samples (mature seeds from r and rb RILs grown at two sites) were analyzed by <sup>1</sup>H high resolution NMR spectroscopy. Dried pea samples were ground into a fine powder and extracted with 1:1 methanol: water (150 mg per 1.5 mL). Samples were vortexed for 30 min before centrifugation (20,817g for 10 min). Methanol was removed from 900 µL of every supernatant by passing a stream of nitrogen over the sample for approximately 1 h. Samples were lyophilized overnight and then re-constituted in 700 µL NMR sample buffer (250 mM sodium phosphate, pH 7.0; 0.5 mM trimethylsilyl propanoic acid, TSP, dissolved in D2O), centrifuged (20,917g for 10 min) and 540 µL transferred to a labeled NMR tube. Sodium azide (60 µL aliquot of 10 mM, dissolved in D2O) was added to every sample to prevent microbial growth before NMR analysis. Extracts were also produced from the seed material using deuterated chloroform to ensure that metabolites which were not soluble in deuterated phosphate buffer solution were analyzed.

All spectra were acquired using an 11.7 T Bruker 500 MHz NMR spectrometer equipped with a 5 mm TCI cryoprobe. Acquisition and processing of the raw data were performed by using Topspin 2.13 patch level 6 (Bruker, Germany). NMR parameters and the magnetic field homogeneity were optimized using a control pea seed extract. The magnetic field was locked on the deuterium signal of the D2O and the homogeneity was optimized. The free induction decay (FID) was recorded using a 30◦ <sup>1</sup>H flip angle determined from a 90◦ pulse length of 11.25 µs. A relaxation delay of 3 s was inserted into the pulse sequence to ensure that quantitative data were acquired. Repetitions (256) of 65,536 complex points were collected over a spectral width of 7002.8 Hz, with the center of the spectrum at 500.1323546 MHz. The NMR probe head was maintained at a temperature of 300 K and the sample remained static during data collection. These parameters resulted in a total experiment time of approximately 45 min per sample.

#### NMR Data Processing

The data were Fourier transformed and an interactive phase correction applied to the spectrum. A baseline correction was applied and the spectrum referenced to the TSP peak at 0 ppm, the area of which was set to unity for all processed spectra using FELIX software (Accelrys, San Diego, CA, USA). Spectral binning of the resonances was performed using bespoke software, Metabolab, a graphical user interface developed using the Matlab platform. Adaptive binning was applied to the data by using the undecimated wavelet transform at a predefined level to reduce the number of variables and limit the effect of the variation of chemical shifts (Davis et al., 2007). Using this approach on the data acquired for different experiments (2 years) resulted in a difference in the total bin number determined for the two data sets. However, the bin identities could be compared, based on their defined limits.

# NMR Compound Identification

The identification of metabolites was performed by comparing resonances in the bins with the resonances of spectral data available either from a list of standards present in an internal database or from literature. As a literature source, the following on-line NMR databases were used:


To assign the binned data, the profiles of all acquired spectra were superimposed to determine the range of chemical shifts of all resonances included in the binned area. Following identification of chemical shift values, the listed databases were interrogated and a list of the most likely candidate metabolites was formed. The spectrum of the candidate metabolite was compared with the spectra acquired from the samples either directly, by using the spectra of the compounds in the internal database, or indirectly, using the on-line NMR databases (1–3 above). One-dimensional <sup>1</sup>H and two-dimensional homonuclear and heteronuclear correlation NMR experiments (1H –1H TOCSY and <sup>13</sup>C – <sup>1</sup>H HSQC) were also used to aid the assignment. The acquisition parameters for the TOCSY and HSQC experiments are given in Supplementary Table 1. A set of resonances was attributed to aglycone derivatives of anthocyanins, based on the study of Kirby et al. (2013).

# Normalization of NMR Bin Data and Determination of the Mean of Absolute Deviation (MAD)

The analysis of variation within any one bin across RILs was carried out following normalization of the bin values to a mean of 100 and standard deviation of 1, which resulted in all values being positive. Binned NMR data were normalized according to the formula below to facilitate further data processing.

A'<sup>L</sup> is the normalized bin area for the line L calculated as follows:

a<sup>L</sup> = NMR peak area

mu<sup>a</sup> = mean of the peak area for the RILs and the parents of the mapping population

SD<sup>a</sup> = standard deviation of the peak area for the RILs and the parents of the mapping population

A'<sup>L</sup> = 100 + ((a<sup>L</sup> - mua)/ SDa)

As a result, each bin has a mean of 100 and a standard deviation of 1.

Due to the number of data points to be analyzed (968 NMR bins for year 1, 990 NMR bins for year 2), an initial prioritization of bins for mapping analysis was achieved, using Mean of Absolute Deviation from the mean (MAD) values as a measure of the variation within any given bin. MAD values provide a measure of the absolute deviations of a set of data about the data's mean, that is, it is the average distance of the data set from its mean. Although high MAD values indicate bins with high variation across the population, this variation in phenotype does not necessarily indicate genetic variation. However, analysis of bins which have higher MAD values increases the possibility of detecting those bins for which quantitative trait loci (QTL) could be mapped. Due to the normalization carried out (as above), MAD values for the normalized data ranged between 0 and 1. Heat maps were generated to visualize MAD value distributions along the NMR spectrum and to compare relative MAD values among datasets (Supplementary Figure 1).

Linkage map analysis (see below) was performed for bins of interest, whereby quantitative variation within NMR bins was shown to be associated with genetic loci, if the two groups of lines carrying one or the other parent marker at that locus showed significant difference in NMR signal strength. Analysis by t-tests identified a set of bins with highly significant associations to genetic map regions. The correlation between MAD value and probability of genetic association for each bin was examined, using Pearson's correlation coefficient, in order to validate the usefulness of MAD values as a method of prioritization.

# Genetic Analysis of Quantitative Data Derived for NMR Bins

The genetic marker data and associated genetic maps for seven linkage groups (LG) of the three recombinant inbred pea populations (JI 281 × JI 399; JI 15 × JI 399; JI 15 × JI 1194) which formed the basis for this study are available as Supplementary Data (Supplementary File S1, with the genetic map data available as Supplementary File S1–Figures M1–M9). Briefly, the genetic markers determined were based on genespecific polymorphisms, as well as sequence-specific amplified polymorphic markers based on the retroelement PDR1 (Knox et al., 2009). Linkage analysis was carried out for three sets of RILs, and genetic maps obtained by ordering all available markers and determining their relative positions using THREaD MAPPER, a web-based software developed at JIC (Cheema et al., 2010), which can be accessed at http://threadmapper.org/ threadmapper. The linkage maps generated were used to draw genetic map charts for assessing genetic loci associated with quantitative variation in NMR signals, which can be visualized across the NMR spectrum for all datasets as movies at http:// www.threadmapper.org/qdips. Additionally, genetic markers associated with groups of NMR signals were analyzed in Excel, based on p-values as described below.

For high-throughput genetic analysis of quantitative trait data, single marker analyses were performed for 12 data sets (RIL population, year and location), using the linkage maps generated and the NMR bin quantitative signals as phenotypic data (968 and 990 bins for year 1 and year 2, respectively). A programme (available on request) was developed in house to generate p-values from Student's t-tests between RIL and maternal/paternal alleles. For most subsequent analyses, p-values were transformed to -log<sup>10</sup> to enable map charts to be plotted for visual investigation.

Significance thresholds were determined for all datasets. To minimize the impact of false positive signals generated by multiple t-tests, the significance threshold for each dataset was determined individually. First, the frequency distribution of all p-values associated with markers for all bins within a dataset was plotted within the range of 0 < p ≤ 0.05 with intervals of 0.0001 (**Figure 1**, blue line) and the average of ten successive intervals generated to remove noise from raw data (**Figure 1**, black line). The data were subjected to randomization (100 times resampled) and the mean p-value frequency derived from the randomized data (**Figure 1**, red line) plotted, with error bars of ± 3 standard deviations (**Figure 1**, yellow shadow). A significance threshold for p-value was determined where the plot of the experimental dataset crossed over the upper error bar of the resampled (randomized) dataset as indicated (**Figure 1**, green shadow).

# RESULTS

#### Metabolite Analysis

NMR spectra were acquired and peaks identified by adaptive binning (see section Materials and Methods, Davis et al., 2007) which allowed clear separation of peaks, and therefore calculation of peak areas, but with the consequence that the peaks did not coincide exactly between years. Some resonance frequencies included within a peak were known and thus bonds and compounds that contributed to the peak area could be identified. A list of correspondences and potential contributing compounds are given in Supplementary Tables 2, 3.

Peak areas were imported to Microsoft Excel Worksheets and, for a given data set (year and RIL population), the data were normalized so that each peak had an area of 100 and a standard deviation of 1. This meant that all peak areas were positive and that statistical analyses did not unduly emphasize variation in intense peaks and thus global analyses could be applied to the whole data set.

For a given population of RILs, the peak areas for each individual were available and the difference between the mean score for the lines with contrasting alleles was calculated. The expectation for each genotype is that the mean is 100 and the standard deviation is 1. The expected value of the difference between the two means is therefore 0 and given the number of individuals of each genotype a Student's t-test statistic can be generated. A related test calculated the mean absolute difference (Mean of Absolute Deviation from the mean, MAD) of the peak areas; this is greatest if there are two data subsets, one greater than 100 and the other less than 100. Example heat maps of MAD values in relation to the NMR spectrum are shown in Supplementary Figure 1, where regions of the spectrum showing

off point of p-values above the significance threshold. Data were determined for JI 15 × JI 1194 RILs, mature seeds, year 1, PGRO location.

consistently high variation are apparent. MAD values and the t statistic were well-correlated (Supplementary Figure 1).

The t-tests performed provided a probability value (p) for the two means being different from each other, but this is seriously confounded by multiple testing (ca. 1,000 bins and 790 markers). We therefore examined the frequency distribution of p as compared to randomized data in order to identify a threshold significance value for p (**Figure 1**).

### Genetic Map Based Analysis

The determination of cut-off p-values generated large numbers of "significant" associations (**Table 1**). This suggested that, for most NMR bins, some genetic marker(s) could explain a component of their variation. While this is of theoretical interest, it does not focus attention on specific marker/metabolite associations. An alternative approach was taken where the minimum p-value for each marker was plotted against the genetic map of each RIL population (**Figure 2**). This identified those regions of the genetic map with the most significant effect on the metabolite profile and, once these had been identified, the NMR bin most affected by that marker could be identified.

The plot identifies regions of the genetic map that have important effects on the metabolite profile. It should be realized that these are regions of relative importance because the low pvalues are included in the estimation of the standard deviations. If one of the most extreme peaks was missing, then the standard deviations used to estimate significance would be of a lower value. Those associations that are consistent between years are symmetrically reflected about the genetic map and equivalent TABLE 1 | Summary of pea recombinant inbred populations for which metabolite data were collected for mature seeds in 2 years and two locations.


The significance threshold (see Figure 1) determined for every dataset [population identity, year (Y1 or Y2) and location of plant growth (JIC or PGRO)] is listed, together with the number of markers for which significant quantitative variation in NMR bin signal intensities was determined.

positions on the genetic maps can be seen. The seven regions identified as peaks on the map (**Figure 2**) are discussed below.

#### Peak 1 and Peak 6

These peaks are associated with a segment at the top of linkage group (LG) I characterized by the microsatellite marker PC20 and the gene encoding a small subunit of AGPase (Aubert

p-value [–log<sup>10</sup> (p)] for each marker is plotted (in blue, below the map for year 1 data, and −1 times this value in brown above the map for year 2 data). The red lines correspond to +3 and +5 standard deviations of all p-values for that population and location. Solid red arrows indicate marker-NMR resonance associations that are

consistent between years, open arrows indicate those that reach the threshold value in 1 year, all numbered for reference in the text.

et al., 2006). The classical locus D also maps in this region (Ellis and Poyser, 2002) and this locus is known to regulate the pattern of anthocyanin deposition. The resolution of the map is insufficient to identify a single causative allelic difference. However, although the peak assignments for this region (peaks 1 and 6) in the three populations at the two sites do not coincide exactly (**Figure 2**), the region includes some common genetic markers; the associated compounds are listed in **Table 2**. Note that these markers identify only the most significant associations, so a lack of replication does not mean that a shared association does not occur. One peak, and compound, implicated more than once is the flavanone glycoside hesperidin. The p-value data for the additional peaks identified for hesperidin, as shown in Supplementary Figure 2 for the JI 281 × JI 399 population (JIC, years 1 and 2), illustrate the consistency of this association. All but one of the bins that include a resonance assigned to hesperidin behave co-ordinately across the genetic map. This is consistent with the signals being derived from variation in the abundance of hesperidin or a closely related compound. The lack of significance for one bin (and an equivalent resonant range in both years) could be explained by a contribution from an additional signal in that bin from a compound that does not co-vary in abundance with hesperidin. Here, additional resonances are associated with aromatic compounds, tryptophan and its catabolite tyramine. Taken together, these signals suggest that there is allelic variation in this LG I region that alters the regulation of compounds closely related to the anthocyanin pathway, and the D locus may therefore be implicated.

#### Peak 2

This is a broad peak on LG II (**Figure 2**) and has its highest significance value in the population JI 281 × JI 399 associated with the classical gene A, which regulates anthocyanin biosynthesis and corresponds to a gene encoding a bHLH transcription factor (Hellens et al., 2010). The peak resonances are listed in **Table 3**; most of the bins are in the aromatic region of the spectrum, two in the sugar range and one in the expected range for aliphatic amino acids. The distribution of year 1 and year 2 p-values for the bin tentatively assigned as the flavonoid naringin in the JI 281 × JI 399 RILs (JIC location) are shown in Supplementary Figure 3 and show no significant p-value associated with A. Surprisingly although A and a segregate in all three populations analyzed, no signal for an anthocyanin was detected in this analysis. Kirby et al. (2013) have undertaken an NMR analysis of anthocyanins in Rhus typhina, identifying profiles with multiple resonances, and so it was expected that signals from pea anthocyanins might be similarly scattered throughout the NMR spectrum. The resonances identified by these authors as corresponding to aglycones can be aligned with the bins we defined in this group. These results are shown in Supplementary Figure 4, and suggest that it is likely that some A-regulated anthocyanins are detected by this analysis of mature seeds.

#### Peak 3

Peak 3 corresponded to the top region of LG III in two of the three populations analyzed (**Figure 2**). This peak, close to the rb locus, may be considered an artefact due to there being very few RbRb genotypes within these two populations, where the vining genotypes selected for analysis were rb mutant lines (see section Materials and Methods). The rb mutation is a consequence of a nine-base pair deletion in the gene encoding the large subunit of AGPase (Rayner et al., 2017),



Summary provides the population identity, year and location of plant growth (JIC or PGRO), NMR bin number and range, and compound information. In bold are the bin numbers and/or compound data that were identified consistently.

which maps close to the top of LG III. It is noteworthy that this peak is missing from the JI 15 × JI 1194 population where all individuals are RbRb. Nevertheless, it is of interest that variation in metabolite profiles reflected the status of the rb locus.

#### Peak 4

A peak in the JI 281 × JI 399 population in the middle of LG IV for plants grown at PGRO is seen for both years (**Figure 2**). The peak of this value corresponding to bin 681 in JI 281 × JI 399 (Y1, PGRO) was not assigned to a known compound.

#### Peak 5

For both the JI 281 × JI 399 and JI 15 × JI 399 populations a peak can be seen in LG VI (**Figure 2**). The significant signals are for bin 490 in year 2 in JI 281 × JI 399 (JIC) and in JI 15 × JI 399 (PGRO), which was not assigned to a known compound but there is a resonance noted at 5.1052 ppm.

#### Peak 7

This corresponds to bin 879 in year 1 in the JI 15 × JI 1194 population and corresponds to an unidentified compound with a resonance at 1.1974 ppm.

#### Compound Based Analyses

Using the genetic map as a way of identifying interesting compound/marker associations identified those regions of the map which had the most profound effect on the metabolite pool. However, this approach was limited because the association between NMR bins and known compounds within those bins was poorly established. It did suggest, however, that there are regions of the genome that have a major impact on the seed metabolome and which require further characterization, for example using 2D NMR or complementary analytical methods. The extent of this genetically controlled variation was revealed to be greater than initially expected, compared to analysis of leaf metabolomes (Charlton et al., 2008). A complementary approach was to examine variation associated with priority compounds, or classes of compounds, for which NMR resonances have been established.


TABLE 3 | Peak 2 resonances and associated NMR bin data.

Summary provides the population identity, year (Y1 or Y2), location of plant growth (JIC or PGRO), NMR bin number and range, and compound information. In bold is the bin number identified more than once.

Here we need to consider two problems. The first is that any bin contains more than one resonance frequency and so the signal intensity may reflect the abundance of more than one compound. The second problem is that any particular resonance may derive from more than one molecule, for example if the molecules differ at remote positions. One way to overcome this is to consider the behavior of the signals from molecules with many assigned resonances which might be expected to behave co-ordinately. Several amino acids fulfil these criteria and are discussed below.

#### Isoleucine

There are 23 bins that report the intensity of resonances from isoleucine in year 1 and year 2 data (see Supplementary Table 3). If these all report variation in the abundance of the same compound, then the p-values for each marker should be strongly correlated. Comparing the year 1 and year 2 data for JI 281 × JI 399 grown at JIC (Supplementary Table 4A), this is clearly not the case. The highest correlation coefficient is 0.344 (year 1 bins 795 and 904 vs. year 2 bin 977), and the lowest is −0.377 (year 1 bin 848 vs. year 2 bin 915). In contrast, comparison of the pvalues for bins assigned to isoleucine within either year 1 or year 2 were highly correlated, with correlation coefficients reaching 0.97 in year 1 and 0.99 in year 2 (Supplementary Table 4B). This suggests that different bins are in fact reporting on co-varying determinants of the NMR signal, consistent with reporting on the same compound (or set of compounds). However, not all correlations were high even within an assignment class (defined in terms of the source of the NMR resonance in Supplementary Table 3). This is consistent with some bins reporting on resonance due to similar bonds in related (but different) compounds, or interference from resonances generated from different compounds that fall within the same bin.

The corresponding bins have different ppm ranges in different years; this may mean that the partitioning of signals is different between the data sets from the 2 years which may be more important than environmental effects on the biological samples. Indeed, the average of the correlation coefficients comparing overlapping bins between years is −0.024 ± 0.140 (n = 21), while the average of the correlation coefficient comparing non-overlapping bins between years is −0.027 ± 0.110 (n = 209), which would be consistent with the interpretation that overlapping bins are no more closely related than different bins that contain a resonance from the same compound; in other words, there are confounding signals within a bin. Partitioning these signals along the genetic map is therefore a useful way of dissecting out commonalities across the bins as shown in **Figure 3**. Although some peaks are consistent between **Figures 3A,B**, no peaks are consistent between years in **Figure 3B**, consistent with the suggestion that the bins are not comparable between years.

#### Leucine

Three bins were assigned to leucine in the year 1 and year 2 data, with details provided in Supplementary Table 5 and Supplementary Figure 5. These bins are close to one another and illustrate the way bins correspond between years. The data also highlight how adaptive binning can result in differences in the distribution of resonances among bins between the year 1 and year 2 datasets. (For example, the year 1 bin 903 includes the leucine resonance at 0.9494 ppm and the isoleucine resonance at 0.9584 ppm, but these are in separate bins in the year 2 data).

The distribution of –log<sup>10</sup> (p) values for leucine is shown on the genetic linkage map in **Figure 4**. Here the reflection of the pattern above and below the map shows the consistency of the data between years and sites, which is particularly noticeable

top for each isoleucine signal. The –log10 (p)-values for each marker are plotted above the genetic map for NMR resonances associated with different parts of the isoleucine molecule. Different bins are plotted with a slightly different color, those from year 1 in red and those in year 2 in blue. Overall, the identified peaks (labeled a–h) (Continued)

FIGURE 3 | are different between years, with the exception of peak h. The multiple bins assigned to isoleucine multiplet 7 are deconvoluted in (B). Mapping variation attributed to isoleucine multiplet 7. LG I–VII of the genetic map of the JI 281 × JI 399 RIL population are displayed horizontally from bottom to top. The –log<sup>10</sup> (p)-values for each marker are plotted above the genetic map for year 2 data and below the genetic map for year 1 data. The color coding for the various bins is indicated adjacent to the relevant map. The bin groupings are according to the correlations given in Supplementary Table 4. The peaks (labeled a–h) correspond to those identified in (A) for isoleucine multiplet 7. The red lines above the map correspond to 3 SD units for the variation in p-values.

on LG II in the region of the classical genetic marker A. The correlations among leucine associated bins for two years are given in Supplementary Table 6. The correlations between sites and years for LG II are given in **Table 4**, where the most different site/year combination is JIC in year 2. Remarkably, the strongest and most consistent signal is coincident with the A locus. The direction of this effect shows that the allele a is associated with an increase in signal intensity (Supplementary Figure 6), implying a role for this locus in regulating compounds beyond anthocyanins.

TABLE 4 | Linkage group II correlation coefficients of p-values among sites and years for leucine related NMR signals in the JI 281 × JI 399 RILs (Year 1, Year 2) across two sites (JIC, PGRO).


Shading intensity is proportional to the value of the correlation coefficient.

# The JI 281 × JI 399 Recombinant Inbred Population

Focussing on a single RIL population limits the analysis to a pair of alternative alleles, and we have shown above that the least correlated pair is the year 1 and year 2 data for the population JI 281 × JI 399 grown at JIC. We therefore examined these data sets and filtered according to the −log<sup>10</sup> (p) values, selecting only marker/bin associations having a −log<sup>10</sup> (p) value greater than 5 standard deviations from the mean of all values. For both years, this is a more stringent selection than using the pvalues obtained using data randomization (see above). These data are summarized in **Figure 5**. Where there is correspondence in the identification of a marker/bin association at this level of stringency, the two types of symbol are coincident, whereas regions unique to a given year are indicated by the presence of a single symbol type (**Figure 5**). The resonance signals for all the bins where a compound has been identified are listed in Supplementary Table 7 and the compounds affected are listed in **Table 5**.

#### Raffinose and Related Oligosaccharides in the JI 281 × JI 399 Population

The raffinose family of oligosaccharides (RFOs) are among the list of compounds in **Table 5**. These three (raffinose, stachyose, verbascose) are related in terms of their biosynthesis (Peterbauer et al., 2002). One of the enzymes involved (raffinose synthase, Rfs) shows genetic variation that maps approximately centrally on LG III (close to PSAB124, PSAA491 and PSAC18 markers in the JI 281 × JI 399 population; close to agpS1\_SNP3 on LG III in Iglesias-García et al., 2015). The gene encoding a second enzyme of this pathway, stachyose synthase (Sts), has been mapped to LG V in another cross (cv. Princess × JI 185, not used in this study). It is therefore of interest to describe how allelic variation for those bins, which contribute to the set displayed in **Figure 5** and are associated with only one of these compounds, is distributed on the genetic map. This is illustrated in **Figure 6**.

The graph (**Figure 6**) includes an association between one of the verbascose-related bins (566) and the location of stachyose synthase on linkage group V. Overall, the correlation between the two verbascose bins (566, 544) is low, likely due to additional resonance signals. However, the correlations of bin 566 with the others assigned uniquely to RFOs suggests that this set of bins is reporting on related compounds. The slight elevation of –log<sup>10</sup> (p) values near the location of the raffinose synthase gene does not reach the threshold level. This is notable, as there are eight amino acid differences between the deduced raffinose synthases of JI 281 and JI 399, four of which are predicted to lie within the mature protein (Q216K, R253W, G329V, and M379V for JI 281 and JI 399, respectively). The most significant associations for this group of compounds are with regions of LG II (stachyose, verbascose) and LG IV (raffinose, stachyose) (**Figure 6**).

Genetic loci on different linkage groups are associated with effects on raffinose concentration (**Figure 6**). The three raffinose bins (591, 589, and 587) are generally well correlated for the whole map, where the lowest correlation is for bins 589 and 587 with r = 0.729. However, the analysis for individual linkage groups has a range of correlation values, as would be expected if there is some interference from additional resonances that are under distinct genetic control. The most variable pair is 589 and 587 and their minimum correlation (for LG II) is 0.298 and maximum is 0.980 (for LG IV). Nevertheless, the strong correlation among these three bins is consistent with them reporting reasonably well on the same or a related compound. The most striking feature of these correlations according to linkage group is the contrast between LG II and LG IV. For LG II most correlations are positive; three negative pairwise combinations involve bin 587, with the other raffinose related bins having a positive correlation to both stachyose and verbascose bins. Linkage group II is the least differentiated in terms of these bins (measured as the mean average deviation of the non-self-correlations. For LG I–VII, these are: I, 0.331; II, 0.254; III, 0.351; IV, 0.616; V, 0.307; VI, 0.301; VII, 0.420, and overall 0.299). In contrast, LG IV is the most differentiated, with raffinose and stachyose positively correlated, but these are negatively correlated with verbascose, consistent with an allelic difference in the final step of the pathway. This is also seen on the –log<sup>10</sup> (p) plot toward the right-hand side of LG IV (**Figure 6**), where the color symbols are well separated, suggesting a difference in control of the early and late steps in the RFO pathway. Within this group of compounds, the most intense NMR signal was from stachyose (bin 585) and this showed the largest actual difference in signal intensity between the contrasting allelic states (higher with the JI 399 allele); the greatest percentage difference between the allelic classes was for bin 544 (verbascose) which was higher when associated with the JI 281 allele.

# DISCUSSION

In this paper, we investigate the genetic control of significant metabolites in pea seeds and provide a framework for their analysis in association with genetic maker data. Two approaches were adopted to examine the extent to which genetic, rather than environmental, control was important in determining the metabolome of seeds derived from three mapping populations: a map-based and a compound-based analysis. Despite the difficulties in associating NMR bin resonances exclusively with specific compounds, the screens have identified classes of compounds that should be investigated further as well as regions of the genetic map that warrant further investigation in relation to the compounds that are affected. We conclude that:

the NMR spectrum is shown vertically to indicate relative signal intensity. The blue spots represent the upper and lower bounds of a selected year 1 bin and the red crosses are the upper and lower bounds of selected year 2 bins (both JIC location). Bins were selected that had –log10 (p) values greater than the mean plus 5 SD units.

TABLE 5 | Summary of compounds identified which differed in JI 281 × JI 399 RILs with high significance.



An association between anthocyanin/phenylpropanoid derivatives and the nature of the allele at the A locus on LG II provides an example where the associated gene is a strong candidate for the observed effect, based on knowledge of flower color and seed trait differences associated with A/a. However, the highly significant differences in the branched chain amino acids (leucine, isoleucine) also associated with this locus suggests a wider impact on amino acid metabolism. This may be explained by considering that the anthocyanins

bottom to top. The –log10 (p) values for each oligosaccharide bin/marker are plotted above the genetic map (year 1, JIC). The threshold value for p determined from the randomization and t-test is indicated by the red horizontal line. The positions of raffinose synthase and stachyose synthase genes on linkage groups III and V are indicated by black triangles. The three bins assigned to raffinose (591, 589, and 587) are indicated in shades of red, stachyose (585) in green, and the two verbascose bins (566 and 544) in blue. On the right of the genetic map, a correlation analysis of the p-values for each bin within each linkage group is indicated, together with a color scale for the correlation range −1 to +1. At the bottom of the correlation scores, the correlation between the bins for the whole linkage map is given.

are derived from phenylalanine/ phosphoenolpyruvate, while leucine/isoleucine are derived from pyruvate. Therefore, a reduced flux from phosphoenolpyruvate to phenylpropanoids in a mutants may generate a higher flux from pyruvate and hence more leucine/ isoleucine. This hypothesis is in agreement with the directional change in these amino acids (Supplementary Figure 6). Although the pool of free amino acids is relatively small in mature seeds, in comparison with proteinderived amino acids, it is likely to represent a component of the seed metabolome which is significant to seed storage and early germination. Fait et al. (2006) showed that the metabolic preparation for germination and efficient seedling establishment is initiated during seed desiccation. Understanding the genetic control of such variation is therefore of academic as well as economic interest. Other work has highlighted the impact of single gene changes on the seed metabolome; metabolomic profiling of pea lines down-regulated for AGPase has demonstrated the widespread consequences for metabolism of changes to this single gene (Weigelt et al., 2009). Significant variation in relative amounts of amino acids and in polyamine metabolism was reported in a study of seeds from wild type and mutant pea lines, differing by the presence or absence of pea albumin 2 genes, normally expressed in seeds (Vigeolas et al., 2008).

The genetic loci associated with variation in RFOs are equally of interest, with some genetic control possibly attributed to genes encoding the major synthetic enzymes of the pathway (LG III and LG V), but a much higher level of significance implicating control by genetic loci on LG II and IV. The below threshold variation associated with the different Rfs genes in JI 281 and JI 399 is in agreement with the different Rfs alleles encoding proteins that do not differ greatly in functionality. Certainly, none of the variant Rfs regions are predicted to be of high relevance to protein function (using CODDLE and PARSESNP programmes). The association of verbascose variation with LG V and the Sts gene may be consistent with the JI 281 Sts allele progressing the galactosylation of RFOs further than its JI 399 counterpart. Transfer of a further galactinol residue to stachyose gives verbascose, a reaction which is probably catalyzed by a bifunctional stachyose synthase (Peterbauer et al., 2001). In combination, these loci (**Figure 6**) may be important for determining seedling vigor. In Medicago truncatula, seven of the 12 QTL for germination rate or post-germinative growth parameters co-located with sucrose/RFO QTL (Vandecasteele et al., 2011). A significant negative correlation was also found between seed vigor traits and sucrose: RFO ratio and, in addition, 80% of the variation in the stachyose: verbascose ratio co-located with a stachyose synthase gene. The genetic control of RFOs is of additional interest, given their involvement more generally in abiotic and biotic stress responses (Cao et al., 2013; Nakabayashi and Kazuki, 2015).

Further development of the framework presented here for association of NMR resonances and genetic variation could include two-dimensional NMR on the contrasting genotypes, focussing on the resonances identified as being significant. Additionally, HPLC and/or GC-MS could supplement these analyses. The identification of candidate genes implicated in the genetic regions highlighted by this work could be accelerated by using the fast neutron mutant population, which has been developed for pea in one of the genetic backgrounds studied here (rb mutant) and where large genomic regions have been shown to be deleted (Domoney et al., 2013). Deletions could be positioned with respect to the genetic map and (when available) the genome sequence of pea to identify a subset of fast neutron mutants in which the NMR signals could be compared. Mutants affected in the relevant signal would presumably carry a deletion in the gene of interest and therefore it could be identified. These approaches would be complementary to those presented by Luo (2015) for metabolite-based genome-wide association studies in plants.

#### REFERENCES

Aubert, G., Morin, J., Jacquin, F., Loridon, K., Quillet, M. C., Petit, A., et al. (2006). Functional mapping in pea, as an aid to the candidate gene selection and for investigating synteny with the model legume Medicago truncatula. Theor. Appl. Genet. 112, 1024–1041. doi: 10.1007/s00122-005- 0205-y

#### CONCLUSION

NMR analysis of genetically marked lines of pea has revealed genetic variation associated with sets of metabolites present in mature seeds. Some of this variation may be explained by few genetic loci, including variation in compounds related to aromatic amino acids, branched-chain amino acids, sucrose-derived metabolites, secondary metabolites and some unidentified compounds. Overall there is extensive variation within r or rb genotypes that has major implications for seed quality traits and may impact nutritional and/or organoleptic parameters. This variation is under the control of multiple loci distributed throughout the genome, presenting an array of possibilities for breeders. Our approach shows how the major genetic determinants of such variation can be identified and therefore managed within a breeding programme. The combined analysis thus presented provides a framework for the genetic analysis of the seed metabolome. The genetic marker datasets provided may be used in the further analysis of seed components that relate directly to seed storage and end-use quality traits.

#### AUTHOR CONTRIBUTIONS

NE, CH, and CD conceptualized the research. NE, PK, ZS, GK, and CH performed genetic mapping and analyzed genetic map data. JD, MD, AC, and GV carried out the NMR and data analysis. CH, JC, and NE devised and performed the bioinformatic analysis of the NMR and genetic data; NE and CD drafted and finalized the paper.

#### ACKNOWLEDGMENTS

This work was supported by: Biotechnology and Biological Sciences Research Council (BBSRC) (BB/J004561/1, BB/P012523/1) and the John Innes Foundation, The Department for Environment, Food and Rural Affairs (Defra) (CH0103, Pulse Crop Genetic Improvement Network), a BBSRC/Defra LINK project (BBSRC: BB/H009787/1; Defra: LK09126; AHDB: FV 351), and the European Union (Grain Legumes Integrated Project, GLIP, a Framework Programme 6 project, FOOD– CT−2004–506223). We thank Carol Moreau, Lorelei Bilham, and Catherine Chinoy (John Innes Centre) for technical assistance with parts of this work.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2018. 01022/full#supplementary-material


information content and allele discrimination. Plant Meths. 5:10. doi: 10.1186/1746-4811-5-10


**Conflict of Interest Statement:** JD, AC, MD, and GV were employed by The Food & Environment Research Agency, a government agency (now the company Fera Science Ltd.). GK is employed by the company AMBIS Biotechnology Ltd.

The remaining 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 Ellis, Hattori, Cheema, Donarski, Charlton, Dickinson, Venditti, Kaló, Szabó, Kiss and Domoney. 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(s) 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.

# Metabolomic Evaluation of the Quality of Leaf Lettuce Grown in Practical Plant Factory to Capture Metabolite Signature

Yoshio Tamura1,2, Tetsuya Mori<sup>3</sup> , Ryo Nakabayashi<sup>3</sup> , Makoto Kobayashi<sup>3</sup> , Kazuki Saito3,4 , Seiichi Okazaki5,6, Ning Wang<sup>1</sup> and Miyako Kusano1,3 \*

<sup>1</sup> Graduate School of Life and Environmental Sciences, University of Tsukuba, Tsukuba, Japan, <sup>2</sup> Central Research Institute for Feed and Livestock, National Federation of Agricultural Co-operative Associations, Tsukuba, Japan, <sup>3</sup> RIKEN Center for Sustainable Resource Science, Yokohama, Japan, <sup>4</sup> Graduate School of Pharmaceutical Sciences, Chiba University, Chiba, Japan, <sup>5</sup> Keystone Technology, Yokohama, Japan, <sup>6</sup> Graduate School of Environment and Information Sciences, Yokohama National University, Yokohama, Japan

#### Edited by:

Andreia Figueiredo, Universidade de Lisboa, Portugal

#### Reviewed by:

Francisco A. Tomas-Barberan, Consejo Superior de Investigaciones Científicas (CSIC), Spain Josep Maria Bayona, Consejo Superior de Investigaciones Científicas (CSIC), Spain Luigi Lucini, Università Cattolica del Sacro Cuore, Italy

\*Correspondence:

Miyako Kusano kusano.miyako.fp@u.tsukuba.ac.jp

#### Specialty section:

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

Received: 15 February 2018 Accepted: 30 April 2018 Published: 27 June 2018

#### Citation:

Tamura Y, Mori T, Nakabayashi R, Kobayashi M, Saito K, Okazaki S, Wang N and Kusano M (2018) Metabolomic Evaluation of the Quality of Leaf Lettuce Grown in Practical Plant Factory to Capture Metabolite Signature. Front. Plant Sci. 9:665. doi: 10.3389/fpls.2018.00665 Vegetables produce metabolites that affect their taste and nutritional value and compounds that contribute to human health. The quality of vegetables grown in plant factories under hydroponic cultivation, e.g., their sweetness and softness, can be improved by controlling growth factors including the temperature, humidity, light source, and fertilizer. However, soil is cheaper than hydroponic cultivation and the visual phenotype of vegetables grown under the two conditions is different. As it is not clear whether their metabolite composition is also different, we studied leaf lettuce raised under the hydroponic condition in practical plant factory and strictly controlled soil condition. We chose two representative cultivars, "black rose" (BR) and "red fire" (RF) because they are of high economic value. Metabolite profiling by comprehensive gas chromatography-mass spectrometry (GC-MS) and liquid chromatography-mass spectrometry (LC-MS) resulted in the annotation of 101 metabolites from 223 peaks detected by GC-MS; LC-MS yielded 95 peaks. The principal component analysis (PCA) scatter plot showed that the most distinct separation patterns on the first principal component (PC1) coincided with differences in the cultivation methods. There were no clear separations related to cultivar differences in the plot. PC1 loading revealed the discriminant metabolites for each cultivation method. The level of amino acids such as lysine, phenylalanine, tryptophan, and valine was significantly increased in hydroponically grown leaf lettuce, while soil-cultivation derived leaf lettuce samples contained significantly higher levels of fatty-acid derived alcohols (tetracosanol and hexacosanol) and lettuce-specific sesquiterpene lactones (lactucopicrin-15-oxalate and 15-deoxylactucin-8-sulfate). These findings suggest that the metabolite composition of leaf lettuce is primarily affected by its cultivation condition. As the discriminant metabolites reveal important factors that contribute to the nutritional value and taste characteristics of leaf lettuce, we performed comprehensive metabolite profiling to identify metabolite compositions, i.e., metabolite signature, that directly improve its quality and value.

Keywords: leaf lettuce, plant factory, hydroponic system, soil cultivation, metabolomics

# INTRODUCTION

fpls-09-00665 June 22, 2018 Time: 14:49 # 2

For the stable supply of vegetables, plant factories use hydroponic cultivation that controls vegetable growth and development under closed environments by regulating important factors for plant growth such as the temperature, humidity, light, growing medium, and plant nutrition (Kozai, 2013). Many use artificial lighting for the cultivation of leafy vegetables and herbs. Because light-emitting diodes (LEDs) suppress heat, their energy requirement is lower than of other light sources such as fluorescent lamps. As plant factories involve higher initial financial investments and running costs than soil cultivation (Tokimasa and Nishiura, 2015), their cultivation conditions must be optimized to yield economically viable horticultural outputs.

Leaf lettuce (Lactuca sativa L. var crispa) is a popular leafy vegetable grown in plant factories. Optimization of hydroponic cultivation conditions has focused on increasing its yield (Touliatos et al., 2016) and on the accumulation of functional nutrients (Miyagi et al., 2017). Different light intensities influence the composition of important phytochemicals such as anthocyanins, carotenoids, chlorophylls, phenolics, and ascorbic acid in baby leaf lettuce (Li and Kubota, 2009; Samuoliene et al., 2013). Moreover, the soilless cultivation can be more efficient in term of nutrient requirements for vegetables growth than that of soil culture (Rouphael et al., 2004; Palermo et al., 2012). Although the metabolite composition of leaf lettuce grown under hydroponic cultivation has been investigated, a comparison of the primary and secondary metabolite profiles in lettuce grown under hydroponic- and soil cultivation conditions is still needed.

Metabolomics has been used to identify the effect of different cultivation environments on plants and for the quality evaluation of agricultural products (Kusano et al., 2011b; Zeng et al., 2014; Sung et al., 2015; Garcia et al., 2016; Ya-Qin et al., 2017). According to Lisec et al. (2006), plants can produce approximately 200,000 metabolites and specific metabolites are produced in different plant species (Kusano et al., 2011b, 2014a,b), even in cultivars and ecotypes (Sulpice et al., 2009). As mass spectrometry (MS)-based metabolomics yields highly sensitive results and facilitates high-throughput data acquisition, it has been combined with various types of analytical separation techniques, including gas chromatography (GC), liquid chromatography (LC), and capillary electrophoresis (CE) (Gowda and Djukovic, 2014). Since primary metabolites contribute to taste, integrated comprehensive GC-MS analysis and sensory evaluation were applied to determine the taste characteristics of green tea grown under various artificial light conditions, and bidimensional GC could provide extended metabolic phenotyping of natural variants in rice (Kusano et al., 2007; Miyauchi et al., 2017). LC-MS-based metabolite profiling aimed at detecting secondary metabolites including functional ingredients in foods and plants (Xie et al., 2008; Ma et al., 2013) revealed that environmental factors affected the accumulation of flavonoids in tea (Zhang et al., 2014). Comprehensive LC-MS analysis combined with ultra-high-performance LC (UHPLC) and quadrupole high-resolution time-of-flight MS (qTOF-MS) was used for secondary metabolite profiling to evaluate the quality of three artichoke cultivars and their commercial products (Farag et al., 2013). Furthermore, CE-MS-based metabolomics to analyze charged metabolites, i.e., amino acids and organic acids, has been utilized for quality control of agricultural products such as edamame beans, pork meat, and sake (rice spirit) (Sugimoto et al., 2010, 2012; Muroya et al., 2014). Additionally, the use of nuclear magnetic resonance (NMR) spectroscopy in plant metabolomics study also complements analytical technique choices in providing robust and reproducible metabolic data (Sobolev et al., 2005; Pereira et al., 2014; Sekiyama et al., 2017).

Metabolite profiling is a promising analytical method for the elucidation of taste characteristics and for the functional evaluation of agricultural products because unlike targeted analysis, it returns a variety of component profiles for many samples. Moreover, multi-MS-based metabolite profiling yields wider coverage for the detection of a large variety of metabolites than single chromatographic techniques combined with MS (Arbona et al., 2009; Lee et al., 2014; Sung et al., 2015). It not only detects variations in taste-related metabolites elicited by differences in cultivation conditions, but also contributes to the identification of quantitative changes in specialized metabolites in vegetables. Taken together, collection of metabolite alterations data of leaf lettuce from different cultivation systems is very important for clarifying their quality.

We performed GC-MS- and LC-MS analysis for the comprehensive capture of cultivation-specific- and taste-related metabolites in leaf lettuce (**Figure 1**). To differentiate the metabolite composition, including primary and secondary metabolites, of two leaf lettuce cultivars, black rose (BR) and red fire (RF), grown under hydroponic or fertilized soil conditions, including light sources and nutrient, we performed integrated MS-based metabolite profiling to capture metabolite signature for BR and RF, and thus to establish their quality indexes.

# MATERIALS AND METHODS

# Plant Materials and Growth Condition

Two cultivars of L. sativa L. cv. BR and RF were grown for 33 days under hydroponic or soil conditions. The soil was treated with optimized concentrations of liquid fertilizer. The BR and RF seeds were purchased from Kaneko Seed Co. (Japan) and Takii Seed Co., Ltd. (Japan), respectively.

For hydroponic cultivation of BR and RF lettuce we applied the methods used by Keystone Technology Inc. (Japan). The conditions were: temperature, 22◦C; 16 h light/8 h dark cycle (16 l/8 days); photosynthetic photon flux density (PPFD), 150 µmol m−<sup>2</sup> s −1 from light-emitting diode (LED) light (peak wavelength: blue = 460 nm, green = 525 nm, red = 660 nm). For soil cultivation the growth-chamber (LPH-350S, Nippon Medical and Chemical Instruments Co., Ltd., Japan) conditions were: temperature, 24◦C; 16 l/8 days; PPFD, 140 µmol m−<sup>2</sup> s −1 from white LED light, 2–3 times per week watering (Tamoi et al., 2017). The leaves of each plant were counted starting with the smallest leaf and the leaf growth stages were recorded as "young," "middle" and "old" (Kusano et al., 2011a). The collected leaf samples were as follows: hydroponic cultivation; 2 cultivars × 2 sampling positions × 3 biological replicates, soil cultivation;

2 cultivars × 2 sampling positions × 4 biological replicates. In total, there were 28 leaf samples (**Figure 1**). Leaf samples were collected in bulk from 4 sites on leaves in the middle- and old stage using an 8-mm leaf punch disk (Fujiwara Scientific Company Co., Ltd., Japan) and immediately frozen in liquid nitrogen. Samples were stored at −80◦C until analysis.

#### Metabolite Profiling by GC-TOF-MS

For the extraction of 25-mg leaf samples (fresh weight) we used 1 ml methanol/chloroform/H2O (3:1:1, v/v/v) containing 10 stable isotope references. The extraction was performed for 10 min at 15 Hz and 4◦C using a mixer mill (Retsch MM301, Retsch GmbH, Germany) (Kusano et al., 2007). The mixture was then centrifuged, the supernatant was placed in a glass insert vial, and evaporated to dryness in a vacuum concentrator (Savant SPD2010 SpeedVac, Thermo Fisher Scientific Inc., United States). The dry extracts were derivatized using 30 µl of methoxylamine hydrochloride-HCl (20 mg/ml in pyridine) for 23 h at room temperature, then for 1 h in 30 µl of N-methyl-Ntrimethylsilyltrifluoroacetamide (MSTFA) at 37◦C with shaking; heptane (30 µl) was added to the extract mixture.

The extract (1 µl) was injected in into an Agilent 6890N GC instrument (Agilent Technologies, United States) via a CTC CombiPAL autosampler (CTC Analytics, Switzerland) for primary metabolite profiling. For separation we used an Rxi-5 Sil MS column [RESTEK, United States, inner diameter (ID), 30 m × 0.25 mm; film thickness, 0.25 µm]. Helium was the carrier gas delivered at a constant flow rate of 1 ml/min. The initial GC oven temperature was set at 80◦C for 2 min, raised to 320◦C at a rate of 30◦C/min, and then held constant for 3.5 min. Data acquisition was performed on a Pegasus IV TOF MS instrument (LECO Corp., United States); the acquisition rate and range were 30 spectra/s and m/z 60–800, respectively. Alkane standard mixtures (C8–C20 and C21–C40, Sigma-Aldrich, Japan) were used for calculating the retention index (RI).

## Metabolite Profiling by LC-TOF-MS

The dried samples were extracted (7 min, 18 Hz, 4◦C) with 150 µl of 80% MeOH and zirconia beads containing 2.5 µM lidocaine and 10-Camphorsulfonic acid per mg dried weight using a mixer mill. Zirconia beads were used to increase extraction efficiency in particularly for cationic compounds, e.g., anthocyanins. After centrifugation, the supernatant was filtered using an Oasis HLB µElution plate (Waters Co., United States) (Nakabayashi et al., 2014). Extract (1 µl) was injected in into a Waters Acquity UPLC instrument coupled with a Waters Xevo G2 QTOF-MS instrument for metabolite profiling. The analytical conditions were LC column: Acquity-bridged ethyl hybrid (BEH) C18 (ID, 100 mm × 2.1 mm; 1.7 µm particle diameter; Waters); solvent system: solvent A (water including 0.1% formic acid); solvent B (acetonitrile including 0.1% formic acid); gradient program: 99.5% A/0.5% B, 0 min; 99.5% A/0.5% B, 0.1 min; 20% A/80% B, 10 min; 0.5% A/99.5% B, 10.1 min; 0.5% A/99.5% B, 12.0 min; 99.5% A/0.5% B, 12.1 min; 99.5% A/0.5% B, 15.0 min; flow rate: 0.3 ml/min at 0 min, 0.3 ml/min at 10 min, 0.4 ml/min at 10.1 min, 0.4 ml/min at 14.4 min, and 0.3 ml/min at 14.5 min; column temperature: 40◦C; MS detection: capillary voltage, +3.0 keV; cone voltage, 25.0 V; source temperature, 120◦C; desolvation temperature, 450◦C; cone gas flow, 50 l/h; desolvation gas flow, 800 l/h; collision energy: 6 V; mass range: m/z 50–1500; scan duration: 0.1 s; interscan delay: 0.014 s; data acquisition: centroid mode; polarity: positive/negative; Lockspray (leucine enkephalin); scan duration: 1.0 s; interscan delay: 0.1 s. MS/MS data were acquired in ramp mode under the following analytical conditions: (1) MS: mass range, m/z 50–1500; scan duration, 0.1 s; inter-scan delay, 0.014 s; data acquisition, centroid mode; polarity, positive/negative; and (2) MS/MS: mass range, m/z 50–1500; scan duration, 0.02 s; inter-scan delay, 0.014 s; data acquisition, centroid mode; polarity, positive/negative; collision energy, ramped from 10–50 V. In this mode, MS/MS spectra of the top 10 ions (>1000 counts) in an MS scan were automatically obtained. If the ion intensity was below 1000, we did not perform MS/MS data acquisition but moved on to the next top 10 ions. Data acquisition was with Progenesis CoMet (Nonlinear Dynamics). Peak normalization was with lidocaine (positive mode) and 10-camphorsulfonic acid (negative mode).

#### Data Pre-processing

Non-processed data (NetCDF format) from GC-TOF-MS were exported to MATLAB (Mathworks, United States); custom scripts were used for data normalization, baseline correction, and subsequent analysis. Lastly, processed data obtained from hyphenated data analysis (HDA) were identified or annotated using an in-house metabolite library in PRIMe (Platform for RIKEN Metabolomics<sup>1</sup> ), and the library in the Golm Metabolome

<sup>1</sup>http://prime.psc.riken.jp

Tamura et al. Cultivation Methods Influenced Lettuce Metabolites

Database (GMD) (Jonsson et al., 2005, 2006). Peaks were normalized with the cross-contribution compensating multiple standard normalization (CCMN) method (Redestig et al., 2009). Chemical assignment of data acquired from LC-TOF-MS was performed using reported MS- or MS/MS data (Wu and Prior, 2005; Abu-Reidah et al., 2013) and the KNApSAcK database (keyword: Lactuca sativa). The m/z values were set as monoisotopic masses [(M+H)<sup>+</sup> or (M-H)−] under a tolerance match limit of 0.01 Da.

#### Statistical Analysis

fpls-09-00665 June 22, 2018 Time: 14:49 # 4

To determine the effect of the different cultivation conditions on the metabolite profile of leaf lettuce we performed multivariate statistical analysis including principal component analysis (PCA) and orthogonal partial least square-discriminant analysis (OPLS-DA) of the unit variance scaled and log-10-transformed data matrix obtained from GC- and UPLC**-**TOF-MS using SIMCA-P 14 (Umetrics, Sweden). Significant metabolites in leaf lettuce were evaluated based on their differing variable importance in projection (VIP) values calculated with OPLS-DA and Q-values adjusted via the false discovery rate (FDR) approach (Benjamini and Hochberg, 1995) using R package samr<sup>2</sup> . The effect of the different cultivation methods on the resulting common metabolites in the cultivars was visualized on the Venn diagram (VENNY ver.2.1<sup>3</sup> ).

#### RESULTS

#### Experimental Design

To investigate the effect of different cultivation conditions, cultivars, and leaf positions on the metabolite composition of leaf lettuce, we studied BR and RF cultivars raised for 33 days under the strictly controlled soil and hydroponic conditions (**Table 1** and **Figure 1**). As metabolite profiles are largely affected the extent of visual phenotypes (Fiehn et al., 2000), we optimized soil condition for minimizing differences of visual phenotypes when compared to lettuce phenotypes harvested from the practical plat factory. The same type of liquid fertilizer was applied under both conditions and the light intensity was almost the same. Different

<sup>2</sup>https://cran.r-project.org/web/packages/samr/index.html

<sup>3</sup>http://bioinfogp.cnb.csic.es/tools/venny/

TABLE 1 | Soil- and hydroponic-cultivation conditions.


<sup>∗</sup>PPFD, photosynthetic photon flux density (µmol m−<sup>2</sup> s −1 ).

concentration of the fertilizer was used for each cultivation because appearance and growth was different when we used the similar concentration for both cultivations (data not shown). Comparison of their phenotypes showed that the BR and RF lettuce leaves exposed to soil and hydroponic conditions were distinct (**Figure 2**). Soil-based cultivation yielded smaller colored leaves than hydroponic farming. Soil-grown BR and RF leaves were of the typical green color with dark- (BR) and bright-red pigments (RF) at the tip. For GC- and LC-MS, tissue samples from leaves in the old- and middle growth stage were collected from the same plant using an 8-mm leaf disk. The metabolites obtained by GC- and LC-MS were identified or annotated, and the metabolite profile data were then integrated for data interpretation by multivariate statistical analysis.

### Metabolite Profiling of Leaf Lettuce Grown Under Different Cultivation Conditions

To investigate metabolite changes, including primary and secondary metabolites, in BR and RF lettuce grown under different cultivation conditions, we harvested 28 leaf samples from each cultivar for comprehensive GC- and LC-MS analyses. GC-MS detected 223 peaks; 101 were identified- or provisionally identified primary metabolites. Positive- and negative-ionization mode-LC-MS followed by an MS/MS library search made it possible to annotate 2 of 30- and 30 of 65-detected compounds, respectively (Supplementary Table 1). Unsupervised multivariate statistical analysis by PCA was then performed to visualize the extent of metabolite changes elicited in the cultivars, under the different cultivation conditions, and in the leaf position.

In detail, "cultivation conditions" could differentiate on the first principal component, PC1 (22.9%), in which soil-based cultivation was recorded in the positive quadrant and hydroponic in the negative (**Figure 3**). Except for samples obtained from oldlayer BR leaves grown hydroponically (H-BR), the plots of the other leaves were not clearly separated. This observation suggests that there is little difference in the metabolite composition of the cultivars and of leaves sampled at different growth stages.

We next identified the discriminant metabolites in leaf lettuce by OPLS-DA. The score scatter plot in **Figure 4A** shows that the samples were clearly separated by the cultivation condition [PredComp1 = 20.5%, overall predictive performance of the model = R2 (Y): 0.99%, Q2 (cum) = 0.92%]. Therefore, the cultivation conditions had more impact on metabolite changes than the cultivar-type or the leaf growth stage.

## Discriminant Metabolites in Soil- and Hydroponic-Cultivated Leaf Lettuce

The discriminant metabolites for our quality index of leaf lettuce were screened from all 318 metabolites based on the selection criteria of VIP>1.0 and FDR<0.05 (**Figure 4B** and Supplementary Table 2). We then compared changes in the metabolite profiles of hydroponically- and soil-grown lettuce. Leaves harvested after soil cultivation contained more typical lettuce sesquiterpene lactones and fatty acid-derived alcohols [log2-fold change (FC) ranging from -3.37 to -1.45] than hydroponically grown leaf lettuce which contained more amino acids (log2-FC ranging from 0.84 to 4.25).

In both cultivars we looked for metabolites that differed when they were grown under the two different conditions. We extracted discriminant metabolites that showed FDR<0.05 and |log2-FC| >1; our findings are presented in a Venn diagram (**Figure 5** and Supplementary Tables 3, 4). When GC-MS and LC-MS detected the same metabolite, we selected the metabolite peak with the higher VIP value. The Venn diagram of metabolites that accumulated under hydroponic conditions (log2-FC>1) shows that common metabolites in BR and RF were amino acids (lysine, phenylalanine, pyroglutamate, tryptophan, and tyrosine) and phosphoric acid (**Figure 5A**). Soil-specific metabolites (log2-FC<-1) were sugars, organic acids, sesquiterpenes, and fatty acid-derived alcohols. Common metabolites in soil-grown BR and RF were arabinose, sucrose, myo-inositol, β-sitosterol, 1-hexacosanol, 1-tetracosanol, cystine, galactonic acid, galacturonic acid, and hydroxybenzoic acid (**Figure 5B**). The cultivar-specific metabolites in BR leaves grown hydroponically were amino acids (isoleucine, leucine and valine), caffeoylmalic acid, and coniferoside; in RF leaves grown hydroponically they were 2-propenoic acid and glutamate (**Figure 5A**). BR-specific metabolites obtained from plants grown in soil were sugars (erythritol and raffinose), organic acids (2-oxoglutaric, glutaric, and shikimic acids), and sesquiterpenes (lactucopicrin-15-oxalate and 15-deoxylactucin-8-sulfate); RFspecific metabolites included organic acids (glyceric and suberic acids) and phenolics (caffeoyltartaric-p-coumaroyl and p-coumaroylquinic acids) (**Figure 5B**). The different cultivation systems did not significantly alter the content of pigment and flavonol compounds in leaf lettuce, while these compounds tended to be accumulated in BR leaves (**Figure 6**). The application of the two cultivation methods could significantly differentiate (p < 0.05) taste-related compounds that might influence the sensory acceptance of lettuce, including glutamate (umami) (Halpern, 2000; Hounsome et al., 2008; Kurihara, 2009), sucrose (sweetness), and lactucopicrin-15-oxalate (bitterness) (**Figure 6**). Thoroughly, RF leaves from hydroponic cultivation had significantly more glutamate content than that of soilcultivated RF leaves. Then again, both BR and RF leaves grown hydroponically had significantly less sucrose and lactucopicrin-15-oxalate levels than leaves from soil cultivation.

#### DISCUSSION

## Effect of Cultivation Conditions on Leaf Lettuce Growth

The leaves of BR and RF lettuce grown for 33 days in soil or under hydroponic conditions exhibited different morphological

characteristics, i.e., their color and shape, but similar size (**Figure 2**). It has been reported that metabolite profiles largely affect from differences that is considered to be derived from difference of growth stages as well as visible phenotypes (Kusano et al., 2011a; Clevenger et al., 2015; Li et al., 2016). Soilcultivated leaves had more reddish appearance than leaves grown hydroponically. Under both growth conditions the BR leaves were reddish and more darkly green than RF leaves, suggesting leaf pigmentation is affected by cultivar characteristics (Caldwell and Britz, 2006). According to Stutte et al. (2009) and Son and Oh (2013), the combination of red and blue LED light yields red leaf lettuce while the sole exposure to red LED light failed to yield pigmented leaves. These differences in appearance, including the leaf morphology and color, may influence sensorial preferences and the acceptance RF and BR lettuce (McWatters et al., 2002). As plant pigmentation reflects the level of chlorophyll and other secondary metabolites such as carotenoid and anthocyanin, we hypothesized that environmental- and cultivar-related factors impact the primary and secondary metabolite composition associated with metabolic pathways (Sass-Kiss et al., 2005; Caldwell and Britz, 2006). Therefore, we used comprehensive GC- and LC-MS-based metabolic profiling to identify specific metabolites related with the BR and RF lettuce phenotype except for chlorophylls and carotenoids. Because the MS-based metabolomics platform cannot detect these compounds.

# Metabolomic Evaluation of the Quality of Leaf Lettuce Grown Under Different Cultivation Conditions

We demonstrate that the cultivation condition exerts greater influence on the metabolite composition of leaf lettuce than

the cultivar type or the leaf position. Environmental factors of cultivation systems can contribute to the elicitation of taste- and nutrition-related compounds in various plants (Ohashi-Kaneko et al., 2007; Zhang et al., 2014). Okazaki et al. (2008), who examined the influence of the primary metabolite composition and of nitrogen availability in two spinach cultivars grown under different inorganic nitrogen concentrations, found that the efficiency of their nitrogen use was not the same.

We varied the light source intensity and the liquid fertilizer concentration in our study of RF and BR lettuce grown

hydroponically and in soil. Miyagi et al. (2017) reported that red LED radiation combined with high concentrations of carbon dioxide and liquid fertilizer increased the biomass and the level of some amino acids in head lettuce grown hydroponically. Nitrogen, the main component of liquid fertilizer, impacts the plant metabolism. We found higher concentrations of amino acids (leucine, isoleucine, tryptophan, tyrosine, and phenylalanine) in the leaves of hydroponically- than soil-grown BR and RF lettuce. This phenomenon might also be affected by the applied nutritional values in both cultivation systems wherein more nitrogen nutrition was provided for hydroponic condition than in soil cultivation (**Table 1**). Due to the elicitation of higher amino acid levels, hydroponic cultivation may result in stronger "umami" perception than soil cultivation (Morris et al., 2007; Wang et al., 2016). On the other hand, the accumulation of sugars, fatty acid-derived alcohols, and β-sitosterol was lower in hydroponically- than soil-grown lettuce leaves. Our findings agree with earlier reports that the accumulation of sugars may be affected by environmental factors or stress such as light- and nutrient-conditions (Okazaki et al., 2008; Rosa et al., 2009; Miyagi et al., 2017). The level of leaf epidermal wax compounds such fatty acid-derived alcohols and β-sitosterol might be decreased

in lettuce leaves grown in a hydroponic culture system (Baker et al., 1979). Also, the three-times shorter shelf life of hydroponically grown leaf lettuce may be attributable to lower dehydration (Manzocco et al., 2011), suggesting that the decreased level of fatty acids and sterols we observed in hydroponically grown lettuce contributed to the change in its leaf epidermal composition. Additionally, increased levels of fatty acids were considered to reflect stress of lettuce in soil (Yang and Ohlrogge, 2009).

To evaluate the effect of environmental growth conditions on RF and BR lettuces, we compared their metabolite profiles and extract metabolite signature for developing a quality index. Compared with hydroponically grown lettuce, the accumulation of sesquiterpene lactones and phenolics was markedly greater in BR- and RF-lettuce leaves harvested after soil cultivation. Seo et al. (2009) reported that the correlation between mineral nutrients intake and producing of sesquiterpene lactones in leaf lettuce. Exposure to short-wavelength light (blue LED and UV**-**B treatments) elicits an increase in phenolics, flavonoids, and the pigment content in lettuce leaves (Ouzounis et al., 2015) and in Arabidopsis thaliana (Kusano et al., 2011c) and our light conditions may have exerted a similar effect. Caldwell and Britz (2006) suggested that differences in the photosensitivity trait among cultivars changes the metabolite composition in leafy vegetables. The higher level of phenolics in soil- than hydroponically grown RF lettuce may increase its antioxidant and antimutagenic effects but render it more bitter-tasting and more astringent (Ren et al., 2001; Altunkaya et al., 2016; Filippo D'Antuono et al., 2016). BR-specific metabolites such as sesquiterpene lactones found in soil-grown plants may also enhance its bitter taste (Chadwick et al., 2016; Filippo D'Antuono et al., 2016). On the other hand, the sugar components in these soil-grown lettuces may mask their bitter taste and enrich their taste complexity (Beck et al., 2014; Chadwick et al., 2016).

### CONCLUSION

Our study shows that the two cultivation methods greatly affect the metabolite profile of RF and BR leaf lettuce, including the metabolites responsible for their taste and their functional ingredients (e.g., amino acids and phenolic compounds). These metabolites may contribute to the overall quality and sensorial perception of leaf lettuce grown under hydroponic and soil cultivation methods (Manzocco et al., 2011; Mampholo et al., 2016). Metabolomic elucidations for lettuce have been conducted using various analytical techniques, including GC-MS, LC-MS,

#### REFERENCES

Abu-Reidah, I. M., Contreras, M. M., Arráez-Román, D., Segura-Carretero, A., and Fernández-Gutiérrez, A. (2013). Reversed-phase ultra-high-performance liquid chromatography coupled to electrospray ionization-quadrupole-time-of-flight mass spectrometry as a powerful tool for metabolic profiling of vegetables: Lactuca sativa as an example of its application. J. Chromatogr. A 1313, 212–227. doi: 10.1016/j.chroma.2013. 07.020

CE-MS, and NMR analyses (Pereira et al., 2014; Garcia et al., 2016; Zhao et al., 2016; Hurtado et al., 2017; Miyagi et al., 2017; Kitazaki et al., 2018). These techniques enabled to detect most of nutritional and pigment metabolites in the evaluated lettuce. Thus, metabolomic data can be used as an important basis for further detailed assessment of cultivation systems and horticulture products not only for lettuce, but for other vegetables. As the next step, it will be needed to evaluate which growth factor(s) are essential to affect phytochemical accumulation as well as yield in plant factories. We document that integrated metabolic profiling is a powerful tool for the comprehensively evaluation of the quality of leaf lettuce. Studies are underway to examine the effect of the light properties and of the liquid fertilizer concentration on the quality and nutritional value of leaf lettuce.

## AUTHOR CONTRIBUTIONS

MK initiated the study conception. MK, KS, and SO designed the analysis. YT performed the statistical analysis and drafted the manuscript. TM, RN, and MKo worked on data acquisition and analysis. MK supervised the study. SO and NW created and provided the samples. YT and MK wrote the manuscript and interpreted biological meaning of the results.

# FUNDING

This research was partly supported by the "Sustainable Food Security Research Project" in the form of an operational grant from the National University Corporation.

#### ACKNOWLEDGMENTS

We thank Drs. Thomas Moritz, Pär Jonsson, and Hans Stenlund (Umeå Plant Science Centre) for their assistance with data analysis and Dr. Yonathan Asikin (University of Tsukuba) for helpful comments on the manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2018.00665/ full#supplementary-material




unpasteurized Japanese sake with storage. J. Agric. Food Chem. 60, 2586–2593. doi: 10.1021/jf2048993


**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 Tamura, Mori, Nakabayashi, Kobayashi, Saito, Okazaki, Wang and Kusano. 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.

# Effect of Solid Biological Waste Compost on the Metabolite Profile of Brassica rapa ssp. chinensis

Susanne Neugart 1†, Melanie Wiesner-Reinhold1†, Katja Frede1,2, Elisabeth Jander <sup>2</sup> , Thomas Homann<sup>2</sup> , Harshadrai M. Rawel <sup>2</sup> , Monika Schreiner <sup>1</sup> and Susanne Baldermann1,2 \*

*<sup>1</sup> Leibniz Institute of Vegetable and Ornamental Crops, Großbeeren, Germany, <sup>2</sup> Department of Food Chemistry, Institute of Nutritional Science, University of Potsdam, Nuthetal, Germany*

Large quantities of biological waste are generated at various steps within the food production chain and a great utilization potential for this solid biological waste exists apart from the current main usage for the feedstuff sector. It remains unclear how the usage of biological waste as compost modulates plant metabolites. We investigated the effect of biological waste of the processing of coffee, aronia, and hop added to soil on the plant metabolite profile by means of liquid chromatography in pak choi sprouts. Here we demonstrate that the solid biological waste composts induced specific changes in the metabolite profiles and the changes are depending on the type of the organic residues and its concentration in soil. The targeted analysis of selected plant metabolites, associated with health beneficial properties of the Brassicaceae family, revealed increased concentrations of carotenoids (up to 3.2-fold) and decreased amounts of glucosinolates (up to 4.7-fold) as well as phenolic compounds (up to 1.5-fold).

#### Edited by:

*Marta Sousa Silva, Universidade de Lisboa, Portugal*

#### Reviewed by:

*Lachlan James Palmer, Flinders University, Australia Bartosz Adamczyk, University of Helsinki, Finland*

#### \*Correspondence:

*Susanne Baldermann baldermann@igzev.de*

*† These authors have contributed equally to this work.*

#### Specialty section:

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

> Received: *20 November 2017* Accepted: *22 February 2018* Published: *16 March 2018*

#### Citation:

*Neugart S, Wiesner-Reinhold M, Frede K, Jander E, Homann T, Rawel HM, Schreiner M and Baldermann S (2018) Effect of Solid Biological Waste Compost on the Metabolite Profile of Brassica rapa ssp. chinensis. Front. Plant Sci. 9:305. doi: 10.3389/fpls.2018.00305* Keywords: metabolite profiling, LC/MS, pak choi, carotenoids, phenolic compounds, glucosinolates

# INTRODUCTION

Rising world population, globally changing climate, intensification and competition of biomass usage as source of bioenergy and urban horticultural practice are only a few examples which enhance people's concern about the efficient, multiple use of bio-mass and recycling of residues. Biological waste is generated at various steps within the food production chain, e.g., production of dairy products, oil industry, beverage producers including breweries, processing of fruits and vegetables, and coffee producers (e.g., Mahro et al., 2015). These industries generate large quantities of biological waste and 13 million tons are produced in Germany per annum (Mahro et al., 2015). Much of these wastes are either burnt or treated by oxygen-driven biological methods, such as composting, which serves a dual purpose, i.e., valorization via material value as well as decreasing the pollution potential (Murthy and Naidu, 2012; Shemekite et al., 2014). On the other hand biological food waste is a reservoir of complex carbohydrates, proteins, lipids, and nutraceuticals and can form the raw materials for commercially important metabolites. Different modes of pretreatments, followed by recovery procedures have been applied to attain value-added products (natural antioxidants, vitamins, enzymes, cellulose, starch, lipids, proteins, and pigments) of spent coffee of high significance to pharmaceutical, cosmetic, and food industries (Murthy and Naidu, 2012). The valorization of the solid coffee by-products in the non-food sector is also gaining more importance (Murthy and Naidu, 2012). During the technological processing of fruits and berries, large amounts of solids remain containing pulp, skins, seeds etc. In Germany, industrial processing

**59**

produces around 7 million tons and in Europe ca. 30 million tons of such wastes. Aronia and hop wastes represent two of these potential wastes. In case of berries, such wastes have also been termed as "pomace." The major part of these residues is either used for animal feed or is more generally disposed–the disposal being costly. Aronia pomace and spent hops have been utilized to extract bioactive antioxidant phenolics (Baranowski et al., 2009; Yui et al., 2013; D'Alessandro et al., 2014; Brazdauskas et al., 2016; Takahashi and Osada, 2017; Xu et al., 2017). Furthermore, compost can enhance the organic content of soil. In this context, the use of organic soil amendments can influence soil fertility, microbial soil communities as well as plant growth and health. The use of compost is positively associated with higher plant available water holding capacity and lower bulk density and can foster beneficial microorganism populations (Grassi et al., 2015; Pan et al., 2016). Studies report the potential of using diverse cover crop green manure in field vegetable production; however the potential of organic waste as compost arising within the food production chain remains to be investigated. Green manures contribute to nitrogen, phosphorus, and potassium contents of soil (Thönnissen Michel, 1996; Diniz et al., 2007; De Neve, 2017; Verlinden et al., 2017). Moreover, eliminating synthetic fertilizers in agriculture is an issue of global concern, due to the associated decrease in soil fertility after its use, pollution, and contamination of produced horticultural products; all with possible relations to human health. The effect of organic production on the secondary metabolite profiles has been Investigated (Soltoft et al., 2011), however it remains unclear how the usage of residues of the food production chain modulates secondary plant metabolite profiles.

Apart from sustainable production and processing in recent years demands concerning the quality of foods is changing and consumers expect safe, high-quality plant foods. Quality is determined by a variety of aspects, including appearance, aroma, and taste, but increasingly also of value-giving plant metabolites. Secondary plant metabolites are defined as bioactive non-nutrients in fruits, vegetables and other plant foods and have been linked to reducing the risk of chronic diseases, i.e., cardiovascular or aging eye diseases (Hartley et al., 2013; Larsson et al., 2013). Secondary plant metabolites include carotenoids, glucosinolates, and phenolic compounds.

Here we report the effect of biological waste used as compost on the content of secondary plant metabolites in Brassica rapa ssp. chinensis (pak choi). Pak choi can be cultivated under a variety of agro-climatic conditions and is a fast growing green leafy vegetable, which is very frequently consumed in Asian countries and in steadily rising quantities in Europe. As with other species, the amount of bioactive substances is known to depend on several parameters, including genotype, growing conditions comprising also soil characteristics, and developmental stage.

Pak choi is rich in secondary plant metabolites and contains considerable amounts of β-carotene and lutein with average concentrations of around 4.5 and 8 mg/100 g edible fresh weight (FW), respectively (Reif et al., 2013). The concentration of βcarotene in other vegetables such as broccoli (0.8–2 mg/100 g FW), cauliflower (white, 0.01 mg/100 g FW), or Chinese cabbage (0.4–3 mg/100 g FW) is substantially lower (Harbaum et al., 2007).

As Brassica species pak choi is rich in glucosinolates, too. The concentrations of these valuable secondary metabolites are ranging from 100 to 300 mg/100 g FW (Wiesner et al., 2013b). The concentrations are comparable to other vegetables such as cabbage (90–200 mg/100 FW), kohlrabi (109 mg/100 g FW) and more than in cauliflower (60 mg/100 g FW) or broccoli (80 mg/100 g FW) (Sones et al., 1984; Kushad et al., 1999).

Moreover, Brassica vegetables are a rich source for phenolic compounds. Diverse phenolic compounds have been described in vegetables of the Brassicaceae family. These include the flavonoids kaempferol, quercetin, and isorhamnetin derivatives (Schmidt et al., 2010). As in other plants these core structures are glycosylated with sugar moieties and can be additionally acylated (Schmidt et al., 2010). Former studies presented a variety of kaempferol, quercetin, and isorhamnetin derivatives as well as various hydroxycinnamic acids in pak choi (Harbaum et al., 2007). The concentration of flavonoid derivatives in the leaf ranged from 4.68 to 16.67 mg/g and the concentration of hydroxycinnamic acid derivatives ranged from 1.48 mg/g of to 5.83 mg/g of dry matter (Rochfort et al., 2006).

We used solid biological waste as compost of the food production of coffee, aronia and hop and were particularly interested if these can modulate the phytochemical concentrations and profiles. We applied LC-ToF-MS to investigate the effect of the organic residues on metabolite profiles in pak choi. In addition, targeted analyses of carotenoids and chlorophylls, glucosinolates, and phenolic compounds were performed.

#### MATERIALS AND METHODS

#### Plant Material

Sprouts of pak choi (Brassica rapa ssp. chinenesis) cv. Black Behi were grown in green houses in Grossbeeren, Germany in summer 2015. The sprouts were grown on soil and mixtures of soil and compost. The soil used consisted of 35% vulcan clay, 50% turf, and 15% bark humus (Fruhstorfer Erde, Vechta, Germany, for further characterization see **Table 1).** Five or 10% of the soil was replaced by compost (percent by weight) since these quantities did not impact the plants' development. The biological waste used as compost was dried and grounded to a powder prior mixing with the soil. Spent coffee grounds were obtained from DEK Deutsche Extrakt Kaffee GmbH Berlin (Berlin, Germany) and the spent aronia organic material from Aronia Original Naturprodukte GmbH (Dresden, Germany). Hop production residue was provided by Hopsteiner HHV Hallertauer Hopfenveredlungsgesellschaft m.b.H. (Mainburg, Germany). The soil or soil compost mixtures were filled into aluminum foil trays (33 × 10 cm) and wetted with tap water if needed. Three grams of seeds (= about 1,100 seeds) was sown per replicate. The total aerial tissue of the sprouts with fully expanded cotyledons were harvested after 12 days above the surface, ∼2– 3 cm long) and immediately frozen in liquid nitrogen and stored at −80◦C until lyophylisation. Prior analysis the samples were



\**double lactate extract.*

ground to a homogeneous, fine powder with a Retsch ball mill. All experiments were performed in triplicate.

## Non-targeted Analysis of Non-volatile Metabolites

To assess differences in the composition of metabolites of sprouts grown on pure soil and compost soil mixtures non-volatile metabolites were analyzed by UHPLC-ToF-MS according to Errard et al. (2015, 2016). Ten milligrams of finely powdered tissue was extracted five times with 1.5 mL of 70% methanol acidified with 0.1% formic acid by shaking for 5 min. The volume of the combined supernatants was adjusted to a final volume of 10 mL. The extract was filtered through a 0.2µm PTFE membrane prior to UPLC-ToF-MS (Agilent Technologies 6230 TOF LC/MS) analysis. Samples (5 µL) were injected into an Agilent Zorbax Extend—C18 Rapid resolution HT (50 × 2.1 mm, 1, 1.8µm) column. Sample and column temperatures were maintained at 4◦ and 30◦C, respectively. The metabolites were eluted at a flow rate of 0.4 ml min−<sup>1</sup> using a chromatographic gradient [A: 0–3 min 98% (isocratic), 3–15 min 98–15%, 18 min 15–0%)] of the two mobile phases A; 0.01% aqueous formic acid and B; 0.01% formic acid in acetonitrile. The capillary voltage was set to 3.5 kV in the electrospray source. The source temperature was set to 320◦C. The nebulizer gas flow was set to 8 L min−<sup>1</sup> at 35 psi. Spectra were collected in negative or positive ionization mode at over the m/z range 70–1,200.

#### Analysis of Carotenoids

The carotenoids were extracted three times from 5 mg of freeze dried material using 0.5 ml methanol: tetrahydrofuran solution (1:1, v/v) according to Errard et al. (2015) and Mageney et al. (2016). The extracts were shaken at 1,000 rpm for 5 min at room temperature and centrifuged at 4,500 rpm for 5 min at 20◦C. The combined extracts were evaporated in a stream of nitrogen. The extracts were dissolved in 0.02 ml dichloromethane and 0.18 ml of isopropyl alcohol. Prior analysis the solutions were filtered through a 0.2µM PTFE membrane and kept at 4 ◦C in the auto sampler during the analysis. The separation was performed on a C30-column (YMC Co. Ltd., Japan, YMC C30, 100 × 2.1 mm, 3µm) on an Agilent Technologies 1290 Infinity UHPLC. Mixtures of methanol, methyl-tert-butyl-ether and water in different volume ratios (solvent A: 81/15/4 and solvent B: 6/90/4) were used as mobile phases at a flow rate of 0.2 ml min−<sup>1</sup> . To enhance the ionization 20 mM ammonium acetate was added to the mobile phases. The pigments were analyzed on an Agilent Technologies 6230 TOF LC/MS equipped with an APCI ion source in positive ionization mode. The gas flow rate was set to 8 L min−<sup>1</sup> and the temperature to 325◦C, the vaporizer temperature to 350◦C and the nebulizer pressure was set to 35 psi. The voltage was set to 3,500 V and a fragmentor voltage of 175 V was applied at a corona current of 6.5 µA. Stock solutions of the authentic standards were prepared individually and the concentrations were determined spectro-photometrically using the substance specific wavelengths and extinction coefficients. Identification was achieved by cochromatography with references substances. External standard calibration curves were used for quantification by dose–response curves.

#### Analysis of Glucosinolates

Desulfo-glucosinolates were analyzed after a method reported by Witzel et al. (2015). Twenty milligrams of lyophilized sample material were extracted by shaking with 750 µL 70% hot methanol (70◦C) for 10 min at 1,400 rpm. As internal standard 100 µL 2-propenyl glucosinolate (sinigrin) (1 mM) were added to the extraction mixture. The samples are centrifuged at 4,500 rpm at room temperature and the supernatant transferred to sample tube. The samples are re-extracted twice with 500 µl of hot methanol. The combined supernatants were applied to an acetic acid-activated DEAE Sephadex A-25 (0.5 mL bed volume) and desulfolated by 75 µL ß-glucuronidase/arylsulfatase over night at room temperature. Desulfo-glucosinolates were obtained by elution with 0.5 mL ultrapure water (2x) and the samples filtered through a SpinX (0.22 µMSpinX Costar Cellulose acetate, 3 min 3000 rpm) prior analysis by UHPLC-DAD (Agilent Technologies 1290 Infinity). The separation was performed on a Poroshell 120 EC-C18 (100 × 2.1 mm, 2.7µm) column in gradient mode using water and 40% acetonitrile as mobile phases at a flow rate of 0.4 mL min−<sup>1</sup> . The glucosinolate concentrations were calculated using 2-propenyl glucosinolate as an internal standard at a detection wavelength of 229 nm.

#### Analysis of Phenolic Compounds

The analyses were performed based on a modified method by Schmidt et al. (2010). Twenty milligrams of lyophilized sample material were extracted with 600 µL 60% methanol by ultra-sonication followed by 20 min shaking at 1,400 rpm. The sample was pelleted at 12,000 rpm at 20◦C and reextracted by 400 and 200 µL. The combined supernatant was evaporated and re-dissolved in 500 µL of 20% methanol. Prior analysis the samples were filtered through a SpinX (0.22µM SpinX Costar Cellulose Acetate, 3 min 3,000 rpm). The phenolics were analyzed on an Agilent Technologies 6230 TOF LC/MS equipped with an ESI ion source in positive ionization mode. The gas temperature was set to 350◦C at a flow rate of 10 L min−<sup>1</sup> , the vaporizer to 320◦C and the nebulizer pressure was set to 35 psi. The voltage was set to 3,500 V and a fragmentor voltage of 350 V was applied. The separation was performed on a Prodigy ODS1 100 A (150 × 2.1 mm, 5µm) column in gradient mode using 0.5% acetic acid and acetonitrile as mobile phases at a flow rate of 0.4 mL min−<sup>1</sup> . Stock solutions of the authentic standards were prepared individually and identification was achieved by cochromatography with references substances. External standard calibration curves were used for quantification by dose–response curves.

#### Analysis of Selected Soil Properties

The analyses have been performed by validated analytical methods of LUFA (Hoffmann, 1991). The analyses of the soil-pH (A 5.1.1), the EC value, salt content (A 10.1.1) as well as NO<sup>−</sup> 3 and NH<sup>+</sup> 4 (A 6.1.4.1) have been described by Seck-Mbengue et al. (2017). The total C, N and S concentrations were determined by Dumas combustion using the Elementar Analyser Vario EL (Elementar, Elementar Analysensysteme GmbH, Langenselbold, Germany) according to LUFA A 2.2.5. Minerals have been determined after LUFA (A 6.2.4.1/A 6.2.4.2) using an ICP-OES iCAP 7400 instrument (Thermo Scientific GmbH, Dreieich, Germany) at the following detection wavelengths: P 213.618 nm, K 766.490 nm, and Mg 285.213 nm.

#### Data Pre-treatment and Data Analysis

The raw data from the UPLC-ToF-MS analysis were transformed and processed by Mass Profiler Professional (MPP; Version 12.1). The raw data were extracted by using the molecular feature extraction of Mass Hunter B.06.00 using the following settings: small molecules, min. 500 counts, ion species [M+H]+, [M-H]−, [M+Na]+, [M+NH4] <sup>+</sup>, H2O as neutral loss, peak spacing tolerance 0.025 ppm and a quality score based on mass accuracy, isotope abundance, and isotope spacing of 80%. Raw data files were imported to MPP for recursive workflow. Statistical analysis was carried out by MPP and included oneway-ANOVA followed by post-hoc Tukey's HSD test (p ≤ 0.05). Compounds subjected to a minimum fold change of 2 were considered for identification. Afterwards, formulas were generated using the above mentioned ions and using a match tolerance of 10 ppm. Putative identification was obtained by using the Metlin database (Vers. 4.0, 24768 compounds). The statistical analysis of the results obtained from the targeted analysis of carotenoids, glucosinolates, and phenolic compounds was performed by Statistica 12. Shapiro–Wilk test has been used to test normal distribution. The means were compared by one-way-ANOVA followed by post-hoc Tukey's HSD test (p ≤ 0.05).

# RESULTS

# Characteristics of Soil and Soil Compost Mixtures

Changes induced by addition of compost to the soil have been investigated and are summarized in **Table 1**. The pH of the soil compost mixtures was not affected by the addition of coffee, but increased from 5.7 to 5.9 using aronia and to 6.5 using hop compost. The EC value increased after addition of hop compost and was lower using aronia compost and coffee compost. Major changes have been detected after addition of hop compost. Mg, K, P, NO<sup>−</sup> 3 , and NH<sup>+</sup> 4 increased in the hop compost soil mixtures as well as the total salt content. In aronia compost soil mixtures the total salt content was much lower compared to the common soil, however cannot be explained by the changes in Mg, K, P, or NO<sup>−</sup> 3 and NH<sup>+</sup> 4 and further analyses are required to explain the changed salt contents. The total N was the highest in the hop compost soil mixtures which also yield in a changed C/N ratio compared to the other soils. The carbon contents were in general higher after addition of biological waste as compost, whereas the impact on the total S content was minimal.

# Effects of Composts on Metabolite Profiles

LC-ToF-metabolomics was used to investigate the effect of biological waste as compost on non-volatile metabolites in pak choi. Therefore, metabolomic differences were combined with statistical analysis to determine the degree of metabolic differences caused by changing soil compositions. To evaluate the effect of organic residues on the metabolome, visualization of the altered metabolite profiles was done using principal component analysis (**Figure 1**). The first three principal components accounted for 28.6, 25.2, and 14.8% of the variance in pak choi samples, respectively. The samples could be clearly classified by soil conditions. We observed clusters for sprouts grown solely on soil and those grown on different soil-compost mixtures. There is a slightly different response in the sprouts depending on the amount of biological waste inside the soil. The overall composition was reflected by metabolic differences in the primary and secondary metabolite metabolism. We observed changes in the amino acid, energy, hormone, phenylpropanoid, and sugar metabolism (**Table 2**). For instance the amino acids isoleucine, leucine, methionine, and proline were all increased in sprouts grown on the different composts compared to the control group. Changes in other metabolites of the amino acid metabolism were related to the type and concentration of compost, e.g., threonine increased in sprouts grown on coffee or hop compost or 2-oxoglutarate and oxaloacetate in sprouts grown on coffee or aronia compost. Pyruvate was detected in higher concentration in sprouts grown on coffee compost and 4-hydroxyphenylpyruvate in those grown on hop compost. Glycerinaldehyde-3-phosphate, a metabolite of the sugar metabolism, decreased in any sample grown on compost, whereas galactose, galactose-1-phosphate, and glucose were found in higher concentrations in sprouts grown on coffee compost. Two metabolites of the energy metabolism have been tentatively identified. AMP strongly increased in sprouts grown on coffee or aronia compost, whereas dAMP decreased in

samples grown on hop or aronia compost. Different pathways of the hormone metabolism were involved in the response to the usage of composts, such as salicylate, abscisic acid, and jasmonic acid. Salicylate and an epi-jasmonic acid derivative were reduced compared to the control, except for the sprouts grown on 10% coffee compost and cis-abscisate strongly induced by coffee and hop compost. An induction of the ascorbate levels have been found in samples grown on coffee or hop compost and for dehydroascorbate in those grown on coffee or aronia compost. Malate was detected in higher concentrations in sprouts grown on coffee or hop compost compared to the control group. Compounds of the phenylpropanoid metabolism were reduced compared to sprouts solely grown on soil, except for a tentatively identified isorhamnetin derivative which was accumulated in sprouts grown on coffee or hop compost.

## Effect of Compost on Carotenoids, Glucosinolates, and Phenolic Compounds

On the basis of the putatively identified compounds we expected changes in valuable secondary plant metabolites and carried out a targeted analysis of carotenoids, glucosinolates, and phenolic compounds in the lyophilized sample material.

The factorial analysis of major carotenes, xanthophylls, glucosinolates, and phenolic compounds revealed that the different soil mixtures had noticeable impacts on the secondary plant metabolite profile (**Figure 2**).

The control group (sprouts grown on pure soil) and sprouts grown on soil mixtures with coffee, hop, or aronia composts show altered secondary plant metabolite profiles. Whereas slight differences were observed for the profiles of sprouts grown on coffee compost (5 and 10%) or aronia compost (5 and 10%), we observed higher variations among sprouts grown on hop compost (5 and 10%) (**Figure 2A**). Carotenoids contribute importantly to changes in sprouts grown on coffee and hop compost (5%), and glucosinolates in samples grown on aronia (5 and 10%) and hop compost (10%). The phenolic compounds show a more structure-specific response and contribute to the different metabolite profiles in sprouts grown on hop, aronia, or coffee compost (**Figure 2B**). More in detail, we estimated single carotenoid, glucosinolate, and phenolic compounds variability in sprouts of pak choi grown on pure soil (control) and compost-soil mixtures (**Table 3**).

#### Effect of Compost on Carotenoids

Highest changes were observed for the xanthophyll lutein (up to 3.2-fold), which is also the major carotenoid in pak choi sprouts. The concentration in the control group was 236 ± 46.6µg/g and with all treatments showing a significant increase and the largest increase found in sprouts grown on 10% hop compost. The lutein concentration in this sample group was 966.9 ± 48.3µg/g. In the treatments, the concentration of neoxanthin was found to increase by up to 3-fold. Major changes compared to the TABLE 2 | Metabolites affected significantly by organic composts (coffee, hop, aronia) in soil (n.d. not detected; – reduced compared to the control; + increased compared to the control; level of change: weak (+)/(−) (0–1), moderate +/− (1–5), strong ++ (5–20), very strong + + + >100).


control group (51.6 ± 7.5µg/g) were detected again for sprouts grown on 10% hop compost (186.9 ± 9.3µg/g). The changes in the xanthophyll zeaxanthin were less pronounced with max 1.7-fold change in case of sprouts grown on 10% hop compost (22.4 ± 1.1µg/g control compared to 37.9 ± 1.9µg/g 10% hop compost), too. The alterations in the β-carotene contents were minor compared to the variations observed for the xanthophylls. Slightly higher concentrations have been detected in all sprouts grown on compost, but significant effects were only found in the samples grown on hop compost or 5% aronia compost.

#### Effect of Compost on Glucosinolates

The total amount of the glucosinolates decreased in all sprouts except in the sample where 5% of the soil was replaced by compost derived from hop residue. In case of compost from coffee the reduction was from 13.62 ± 0.014µg/g in the control to 12.19 ± 0.153µg/g and 12.17 ± 0.118µg/g in sprouts grown on 5 and 10% coffee compost, respectively. Higher losses have been found after addition of aronia (5.66 ± 0.105µg/g 5% aronia and 4.36 ± 0.079µg/g 10% aronia) or hop compost (2.94 ± 0.006µg/g 10% hop) (**Table 3**). In all samples grown on soil-compost mixtures the aliphatic glucosinolates were reduced. Highest losses were found in those samples grown on aronia compost (5 and 10%) and on hop compost (10%). Using aronia compost the concentrations dropped by 0.59 (5% aronia compost) and 0.68 (10% aronia compost), from 12.29 ± 0.11µg/g to 5.10 ± 0.11µg/g and 3.94 ± 0.04µg/g. In the pak choi sprouts grown on 10% hop compost the loss of aliphatic glucosinolates was 0.79 corresponding to 2.5 ± 0.01µg/g. Interestingly, in sprouts grown on coffee compost, just the concentration of the aliphatic 4-methylthiobutyl glucosinolate was significantly reduced. The indole glucosinolates were significantly affected using coffee (5 and 10%), aronia (5 and 10%) or hop compost (10%). Noteworthy, in sprouts grown on the coffee compost the concentration of the methoxylated indole glucosinolates increased significantly. The 4-methoxyindole-3 ylmethyl glucosinolate was significantly affected by compost containing 5% coffee (0.30 ± 0.014µg/g) residues compared to the control (0.14 ± 0.003µg/g), whereas the 10% coffee compost affected more the 1-methoxyindole-3-ylmethyl glucosinolate (0.16 ± 0.068 – 0.74 ± 0.040µg/g). The precursor of both, the indole-3-ylmethyl glucosinolates, kept almost unaffected compared to the concentration in sprouts grown on pure soil. In contrast, soils mixed with aronia (5 and 10%) and hop (10%) reduced significantly the concentration of all indole glucosinolates compared to the control. The aromatic 2-phenylethyl glucosinolate was significantly reduced in all samples grown on soil-compost mixtures compared to pure soil (**Table 3**).

#### Effect of Compost on Phenolic Compounds

The concentration in the phenolic compounds (2154.43 ± 93.70µg/g) did not change (5 and 10% coffee compost, and 5% aronia compost) or were lower (5 and 10% hop compost, and 10% aronia compost compared to pure soil). In pak choi sprouts grown on 5% hop compost the concentration was reduced by 0.25 (1665.94 ± 76.36µg/g), on 10% hop compost by 0.31 (1525.24 ± 33.01µg/g, and on 10% aronia compost by 0.23 (1719.27 ± 7.23µg/g).

More specifically, single phenolic compounds showed a particular response in respect of the various soil composts (**Table 3**); e.g., caffeoylmalate and kaempferol-3-O-caffeoyldiglucoside-7-O-glucoside were reduced from control values of 28.39 ± 0.74µg/g by up 100-fold and 18.23 ± 1.48µg/g by 2-fold,

respectively. The concentrations of isorhamnetin-3-O-glycoside-7-O-glycoside and kaempferol-3-O-sinapoyl-diglucoside-7-Oglucoside were higher in samples grown on coffee compost (7.63 ± 0.25µg/g and 748.00 ± 41.73, respectively) and aronia compost (16.16 ± 0.33µg/g and 664.7 ± 0.60µg/g, respectively). To summarize, the addition of hop compost (5 and 10%) to the soil resulted in a decrease of malates and all kaempferol glycosides except for kaempferol-3-O-glucoside, whereas the use of aronia compost (5 and 10%) showed a decrease of malates but an increase of most kaempferol glycosides depending on the compost percentage added to the soil. However, the coffee composts (5 and 10%) were presented by decrease of caffeoylmalate to a minimum of 0.28 ± 0.13µg/g, but only slight or no changes in the kampferol glycosides showing an increase of kaempferol-3- O-sinapoyl-diglucoside-7-O-glucoside to 787.18 ± 21.89µg/g, kaempferol-3-O-glucoside to 7.83 ± 0.58µg/g, and kaempferol-3-O-hydroxyferuloyl-diglucoside-7-O-glucoside to 188.85 ± 7.45µg/g.

# DISCUSSION

#### Effects of Composts on Metabolite Profiles

In this study profiling methods and targeted analyses have been used to explore the overall impact of biological waste of food chain production used as composts on the metabolite profile of pak choi sprouts. Metabolomic approaches for studying changes induced by biotic (Errard et al., 2015) and abiotic factors (Jorge et al., 2016) became an important analytical tool in plant science (Witzel et al., 2015). The metabolite profile differed when comparing sprouts grown composts and control plants grown solely on soil. The overall composition was reflected by metabolic differences in the primary and secondary metabolism.

Interestingly, changes in diverse hormone signaling pathways have been observed. It is well-known that abscisic acid (Taylor et al., 2000), salicylic acid (Janda et al., 2007), and jasmonic acid (Creelman and Mullet, 1995) act as signals in plants to response to biotic and abiotic challenges alone or in combination. These changes lead to alternation in primary and secondary metabolites of which the latter one will be discussed specifically in the following subchapters. Whereas, many amino acids were reduced in response to biotic stress (Errard et al., 2016), in the present study the application of composts increased their concentrations. This accounts also for some sugars. The changes in ascorbate might be also correlated with changes in the xanthophyll cycle, where one of the key enzymes, the violaxanthin de-epoxidase, requires ascorbate as co-substrate and the enzymatic activity is pH-depended (Arnoux et al., 2009). The glucosinolate levels also depend on ascorbate, because ascorbate influences the myrosinase activity—the enzyme responsible for the hydrolyzation of glucosinolates to the bioactive breakdown products (Hasapis and Macleod, 1982). But, in our experimental setup the myrosinase and glucosinolates kept separated from each other and further research is needed to evaluate the possible interaction between ascorbate and glucosinolate concentrations.

Responses to abiotic stress are dynamic and complex and to get a comprehensive understanding further systematic studies are required to understand the specific changes induced by the different composts.

#### Effect of Compost on Carotenoids, Glucosinolates, and Phenolic Compounds

We focused our recent study on the secondary metabolites because they are associated with health beneficial properties. Carotenoids are well-known antioxidants and they can lower the risk of chronic diseases, including cardiovascular diseases (Gammone et al., 2017) and type-2 diabetes mellitus (Akbaraly et al., 2008), and several types of cancer (Bolhassani, 2015). Lutein and zeaxanthin can prevent age-related eye diseases TABLE 3 | Carotenoids, glucosinolates, and phenolic compounds in pak choi grown on soil and soil-organic compost mixtures.


*Means [*µ*g/g DW] were compared by one-way-ANOVA followed by post-hoc Tukey's HSD test (p* ≤ *0.05). Letters indicate significant differences within a group (in alphabetical order from the lowest to highest) and bold numbers show significant difference compared to the control group (post-hoc Dunnet's test, p* ≤ *0.05). Each value represents the mean of three samples* ± *S<sup>E</sup> .*

*Carotene: Car,* β*-Carotene; Xanthophylls: Lut, Lutein; Neo, neoxanthin; Zea, Zeaxanthin; aliphatic Glucosinolates: (GS): 4MTB, 4-Methylthio-butyl GS; 4MSOB, 4-Methyltsulfinly-butyl GS; 5MSOP, 5-Methylsufinyl-pentyl GS; 3But, 3-Butenyl GS; 4Pent, 4-Pentenyl GS, (R)-2OH3But, (R)-2-Hydroxy-3-butenyl GS; (S)-2OH3But, (S)-2-Hydroxy-3-butenyl GS; 2OH4Pent, 2- Hydroxy-4-pentenyl GS; indolic GS: I3M, Indol-3-ylmethyl GS; 4OHI3M, 4-Hydroxyindol-3-ylmethyl GS; 4MOI3M, 4-Methoxyindol-3-ylmethyl GS; 1MOI3M, 1-Methoxyindol-3-ylmethyl GS; aromatic GS: 2PE, 2-Phenylethyl GS; Hydroxybenzoeic acids: CM, Caffeoylmalate; CQ, Caffeoylquinicacid; SM, Sinapoylmalate; Flavonoids: I3G, Isorhamnetin-3-O-glucoside; I3G7G, Isorhamnetin-3-O-glycoside-7-O-glycoside; K3G, Kaempferol-3-O-glucoside; K3CdG7G, Kaempferol-3-O-caffeoyl-diglucoside-7-O-glucoside; K3dG7G, Kaempferol-3-Odiglucoside-7-O-glucoside; K3FdG7G, Kaempferol-3-O-feruloyl-diglucoside-7-O-glucoside; K3SdG7G, Kaempferol-3-O-sinapoyl-diglucoside-7-O-glucoside; K3hFdG7G, Kaempferol-3-O-hydroxyferuloyl-diglucoside-7-O-glucoside; K3pCdG7G, Kaempferol-3-O-p-coumaroyl-diglucoside-7-O-glucoside.*

such as macular degeneration (Mares, 2016; Frede et al., 2017). Crops have been bred or engineered to increase β-carotene as precursor of vitamin A (Bai et al., 2011). Glucosinolates, as characteristically secondary metabolites in Brassica vegetables, are of interest due to the wide range of health-promoting properties of their breakdown products, e.g., for anti-carcinogenic (Traka and Mithen, 2009; Wu et al., 2012), anti-diabetic (Waterman et al., 2015; Guzman-Perez et al., 2016), and anti-inflammatory effects (Bentley-Hewitt et al., 2014; Herz et al., 2016). For human health phenolic compounds are of special interest due to their antioxidant activity (Zietz et al., 2010), as well as anti-inflammatory and anti-cancerogenic effects (Pan et al., 2010; Chen and Chen, 2013).

Structure-specific responses of carotenoids, glucosinolates, and flavonoid glycosides, in Brassica species are also found in response to other eco-physiological conditions such as light or temperature (Schonhof et al., 2007; Neugart et al., 2012; Mageney et al., 2016). However, the sprouts were grown under the same environmental conditions and therefore interacting climatic effects are not expected.

#### Effect of Compost on Carotenoids

The structure-specific response of phenolic compounds as well as the variation in carotenoids and glucosinolates might be partly due to the various nitrogen contents of these composts. We measured the total nitrogen in the organic composts. The highest total nitrogen content was found in the hop compost and slightly higher values in the coffee and aronia compost soil mixtures. Plants response differently to changes of nitrogen concentration and form in soil. For instance, reduction in nitrogen fertilization yield in a reduction of carotenoids in lettuce (Becker et al., 2015), but carotenoid pigments increased in response to increasing nitrogen concentrations in kale (Kopsell et al., 2007) as they did in this study. In other studies, the carotenoid concentrations were not affected such as in leafy Brassica species (e.g., Fallovo et al., 2011). Under organic farming practices, higher concentrations of carotenoids were detected in cauliflower (Lo Scalzo et al., 2013). The content of carotenoids in carrot roots and human diets was not significantly affected by the agricultural production system (organic vs. conventional) or year, despite differences in fertilization strategy and levels (Soltoft et al., 2011), whereas βcarotene was found to be higher in organically grown Brassica species (Kapusta-Duch et al., 2014). This is in accordance to this study, where higher carotenoid concentrations were detected in sprouts grown on compost.

Carotenoids also react in response to stresses. In our study, neoxanthin increased in sprouts grown on soil containing composts from coffee and hop. Neoxanthin positively influences ABA accumulation in response to dehydration (North et al., 2007). However, ABA is only one plant hormone influenced by growing conditions and the metabolic profiling revealed that changes not only occurred in the abscisic acid signaling, but also in the salicylic acid, and jasmonate pathways (**Table 2**). Salicylic acid concentrations have been positively associated with carotenoid concentrations in maize and Indian mustard (Khodary, 2004; Thakur and Sohal, 2014), whereas methyl jasmonate either reduced the degradation of carotenoids in post-harvest lettuce (Kim et al., 2006) or increased β-carotene as well as mannitol in broccoli sprouts (Natella et al., 2016). The exogenous application of ABA, JA, and SA was positively correlated with photosynthetic pigment concentrations in turnip (Thiruvengadam et al., 2016). Therefore, the changes in plant hormones are one possible approach to explain the higher carotenoid concentrations in the present study. An additional factor might be the regulation of the violaxanthin de-epoxidase by ascorbate contributing to the accumulation of xanthophylls (Arnoux et al., 2009).

#### Effect of Compost on Glucosinolates

Glucosinolates are nitrogen containing compounds suggesting that nitrogen supply strongly affect glucosinolate concentration. The effect of nitrogen fertilization on changes in glucosinolates in kale has been reported recently. Increasing nitrogen concentrations in the soil results in reduction of aliphatic glucosinolates and an activation of the glucosinolate catabolic process was observed under ammonium nutrition (Groenbaek et al., 2014; Marino et al., 2016). However, increasing glucosinolate contents were found in broccoli and red cabbage grown on organic soil (Meyer and Adam, 2008). But, the effect of organic soil seems to be genotype-dependent. Picchi et al. (2012) found that the total glucosinolate content as well content of the major glucosinolates decreased in cauliflower grown on organic fields compared to cauliflower grown under conventional conditions. But, these effects were shown just for one cultivar; the glucosinolates in the cultivar Magnifico showed the opposite effect and were increased.

The factor analysis revealed that the aronia compost in both concentrations and hop compost (10%) had the strongest effect on the glucosinolates (**Figure 2**). In sprouts grown on these compost mixtures all individual glucosinolates decreased. This compost induced glucosinolate reduction could be due to a unbalanced ratio of nitrogen and sulfur in the soil: an optimal or high nitrogen supply combined with an insufficient sulfur supply leads to a decrease of aliphatic and increase of indole glucosinolates (Li et al., 2007). But, the optimal ratio of N/S is not only speciesspecific, but structure-specific, too. High N/S ratio promote the formation of aliphatic and indolic glucosinolates in broccoli florets (Schonhof et al., 2007), but in turnips a lower ratio of N/S lead to an increase of aromatic glucosinolates (Li et al., 2007). However, in our experiments the N/S ratio does not explain the decreases or increases of the individual glucosinolates. Furthermore, the ratio of alkenyl to hydroxyalkenyl glucosinolates is changed from about 1.3 to 2.1 (5%) and 3.5 (10%) in sprouts grown on aronia compost. It might be that aronia residues in the soil influence the conversion of alkenyl to hydroxyalkenyl glucosinolates due to the induction of the 2-oxo acid dependent dioxygenase (BrGSL-OH) as higher concentrations of 2-oxo acids have been detected in sprouts grown on aronia compost in this study. In contrast, in the coffee compost samples the aliphatic glucosinolates concentrations are slightly decreased or unchanged. Only the methoxylated indole glucosinolates (4-methoxyindole-3-ylmethyl and 1-methoxyindole-3-ylmethyl glucosinolate) increased significantly in pak choi sprouts grown on coffee compost mixtures, which might be due to the ammonium nutrition (Marino et al., 2016). The authors determined different expression levels depending on the nitrogen source: higher expression levels of CYP79B2/B3 in the indole glucosinolate pathway under nitrate nutrition, and a higher expression of genes involved in the aliphatic glucosinolate pathway under ammonium nutrition. The paradox is that under ammonium nutrition the transcript levels for the genes of the indolic glucosinolate biosynthesis pathway were reduced, but at metabolite level the indolic glucosinolates were increased in Arabidopsis and broccoli (Marino et al., 2016). Furthermore, the gene expression of CYP81F1/F2/F3/F4, involved in the methoxylation of the indole-3-ylmethyl glucosinolate to the methoxylated indole glucosinolates, might be induced by ammonium. Further investigations are needed to shed light on links between nitrogen and glucosinolate metabolism.

Previous experiments showed a regulation of the glucosinolate metabolism by the salicylic acid or jasmonate pathways. In pak choi sprouts an induction of glucosinolates was observed by methyl jasmonate treatment, but no changes were observed after treatment with methyl salicylate (Wiesner et al., 2013a). Other researchers observed an increase of aliphatic and indole glucosinolates after salicylic acid treatment (Kiddle et al., 1994). The interaction of signaling molecules is more complex and regulated by different positive and negative back loops. Also it could be demonstrated in that a drought-induced accumulation of aliphatic glucosinolates is related to ABA formation B. juncea (Tong et al., 2014).

#### Effect of Compost on Phenolic Compounds

The structure-specific response of phenolic compounds might be partly due to the various nitrogen contents of these composts. A negative correlation between specific flavonoid aglycones (kaempferol and quercetin) and nitrogen fertilization in Brassica juncea and between flavonoid glycosides and nitrogen fertilization in lettuce has been proven previously (Fallovo et al., 2011; Becker et al., 2015). Also in kale a species-specific response on changes in nitrogen content has been reported for phenolic compounds (Groenbaek et al., 2014). In the kale cv. Reflex the total flavonoid glucosides were significantly reduced, which was in this species genotype mainly related to the changes in quercetin glucosides while kaempferol and isorhamnetin glycosides were less susceptible to higher nitrogen fertilization. However, the main compounds in pak choi and kale are monoacylated kaempferol and quercetin tri- and tetraglycosides which were reduced in kale with higher dose of nitrogen fertilization (Groenbaek et al., 2014). For the pak choi sprouts in this study the monoacylated kaempferol glycosides were reduced with soil-hop compost mixtures assuming a negative correlation of nitrogen and flavonoid glycosides. However, the results were partly different for the three soil compost mixtures tested which leads to the conclusion that nitrogen is not the only influencing factor and the effect of soil coffee and soil aronia compost cannot be explained like this. Organically grown plants are considered to protect themselves better and form phytoalexins like flavonoids (Asami et al., 2003). Asami et al. (2003) found in organically grown plants (marionberry, strawberry and maize) higher total phenolic contents than in conventionally produced plants. In tomatoes quercetin, kaempferol, and narigenin were higher in organically grown plants compared to conventional grown plants in a 10 year trial (Mitchell et al., 2007). Biologically food waste can be considered as organically fertilization and the flavonoid glycosides in pak choi are mainly decreased. There is the need to study other underlying mechanisms that triggers the production of flavonoids. Apart from nitrogen supply, hormones, and other signaling compounds released as roots exudates could contribute to the modulated secondary plant metabolite profiles (Broeckling et al., 2008). Interestingly, we found specific reactions for sprouts grown on soil-aronia composts mixtures, e.g., kaempferol-3-O-feruloyl-diglucoside-7- O-glucoside increased significantly. It might be speculated that this is the result of higher concentrations in the root environment and lower active release rates by the roots. Further investigations should also include the analysis of modulated soil microorganism communities.

Moreover, jasmonate, salicylate, and abscisic acid are generally known to enhance flavonoids and hydroxycinnamic acids (Chen et al., 2006; Sandhu et al., 2011), but this cannot directly be linked to the present results.

## CONCLUSION

Interestingly, addition of biological waste of the food production chain as compost had strong effects on plant metabolites profiles in sprouts of pak choi. Different composts of the food production chain impact differently the plant metabolite profile of pak choi sprouts. The usage of coffee, aronia or hop composts incorporated in soil increased the concentration of carotenoids and decreased the amount of glucosinolates and phenolic compounds being associated with health beneficial properties of vegetables. The usage of compost might be besides the application of chemical and physical elicitors another possibility to target the accumulation of specific metabolites e.g., for the production of carotenoids to prevent vitamin A deficiency or age-related macular degenerative diseases. However, taking into account the intrinsic quality and health effects of vegetables that is in large measure driven by the secondary plant metabolites unintentional changes such as observed for glucosinolates and phenolic compounds need further investigations. Further studies are necessary and would shed light on the metabolome dynamics and underlying mechanism responsible for the changes induced by composts.

#### AUTHOR CONTRIBUTIONS

SB designed the study in cooperation with MW-R, HR, and KF. MW-R, SN, and SB carried out the plant experiments, contributed to interpretation of data and drafted the manuscript. SB developed the mass spectrometric method for the non-targeted analyses and performed statistics. EJ contributed to the non-targeted analysis. KF analyzed the carotenoids and chlorophylls, SN analyzed the phenolic compounds and MW-R and MS the glucosinolates. TH, HR, and MS critically revised the manuscript. All authors read and approved the final manuscript.

#### ACKNOWLEDGMENTS

We thank E. Büsch, A. Jankowsky, A. Platalla, and M. Skoruppa, K. Schmidt for their assistance.

# REFERENCES


rapa L. ssp. chinensis L. (Hanelt.)] by HPLC-DAD and LC-MS/MS. J. Agric. Food. Chem. 54, 4855–4860. doi: 10.1021/jf060154j


monocyte chemotactic protein-1 levels in OLETF rats. J. Oleo Sci. 62, 283–292. doi: 10.5650/jos.62.283

Zietz, M., Weckmüller, A., Schmidt, S., Rohn, S., Schreiner, M., Krumbein, A., et al. (2010). Genotypic and climatic influence on the antioxidant activity of flavonoids in kale (Brassica oleracea var. sabellica). J. Agric. Food. Chem. 58, 2123–2130. doi: 10.1021/jf9033909

**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 Neugart, Wiesner-Reinhold, Frede, Jander, Homann, Rawel, Schreiner and Baldermann. 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.

# Dynamic Labeling Reveals Temporal Changes in Carbon Re-Allocation within the Central Metabolism of Developing Apple Fruit

Wasiye F. Beshir <sup>1</sup> , Victor B. M. Mbong<sup>1</sup> , Maarten L. A. T. M. Hertog<sup>1</sup> , Annemie H. Geeraerd<sup>1</sup> , Wim Van den Ende<sup>2</sup> and Bart M. Nicolaï 1, 3 \*

<sup>1</sup> Division of Mechatronics, Biostatistics and Sensors, Department of Biosystems, KU Leuven, Leuven, Belgium, <sup>2</sup> Laboratory of Molecular Plant Biology, Department of Biology, KU Leuven, Leuven, Belgium, <sup>3</sup> Flanders Centre of Postharvest Technology, Leuven, Belgium

#### Edited by:

Ute Roessner, University of Melbourne, Australia

#### Reviewed by:

John A. Morgan, Purdue University, United States Shyam K. Masakapalli, Indian Institute of Technology Mandi, India

> \*Correspondence: Bart M. Nicolaï bart.nicolai@kuleuven.be

#### Specialty section:

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

> Received: 12 July 2017 Accepted: 02 October 2017 Published: 18 October 2017

#### Citation:

Beshir WF, Mbong VBM, Hertog MLATM, Geeraerd AH, Van den Ende W and Nicolaï BM (2017) Dynamic Labeling Reveals Temporal Changes in Carbon Re-Allocation within the Central Metabolism of Developing Apple Fruit. Front. Plant Sci. 8:1785. doi: 10.3389/fpls.2017.01785 In recent years, the application of isotopically labeled substrates has received extensive attention in plant physiology. Measuring the propagation of the label through metabolic networks may provide information on carbon allocation in sink fruit during fruit development. In this research, gas chromatography coupled to mass spectrometry based metabolite profiling was used to characterize the changing metabolic pool sizes in developing apple fruit at five growth stages (30, 58, 93, 121, and 149 days after full bloom) using <sup>13</sup>C-isotope feeding experiments on hypanthium tissue discs. Following the feeding of [U-13C]glucose, the <sup>13</sup>C-label was incorporated into the various metabolites to different degrees depending on incubation time, metabolic pathway activity, and growth stage. Evidence is presented that early in fruit development the utilization of the imported sugars was faster than in later developmental stages, likely to supply the energy and carbon skeletons required for cell division and fruit growth. The declined <sup>13</sup>C-incorporation into various metabolites during growth and maturation can be associated with the reduced metabolic activity, as mirrored by the respiratory rate. Moreover, the concentration of fructose and sucrose increased during fruit development, whereas concentrations of most amino and organic acids and polyphenols declined. In general, this study showed that the imported compounds play a central role not only in carbohydrate metabolism, but also in the biosynthesis of amino acid and related protein synthesis and secondary metabolites at the early stage of fruit development.

Keywords: Malus domestica Borkh, Braeburn, fruit growth, GC-MS, metabolomics, <sup>13</sup>C-label accumulation

# INTRODUCTION

Apple (Malus domestica Borkh.) is a member of the Rosaceae family that includes many important fruit trees like pear, peach, cherry, apricot, and prune. Apple is the most important fruit in the world market followed by pear and peach (Brown, 2012). It is consumed widely for its flavor, health, and nutritional value. The composition of the mature apple fruit is the resultant of carbohydrate import and the metabolic processes occurring during fruit development.

In plants, carbon is usually exchanged between source and sink tissues as simple sugars, typically sucrose (White et al., 2015). However, in Rosaceae, sorbitol and sucrose are the two main photosynthates, comprising, for instance, respectively 70 and 30% of the components collected from phloem exudate of apple fruit stalks (Klages et al., 2001). Phloem unloading of soluble sugars in developing apple and pear fruit follows an apoplastic route (Zhang et al., 2004, 2014) with sorbitol and sugar transporters being responsible for the uptake by the cells (Watari et al., 2004; Fan et al., 2009; Peng et al., 2011). Sorbitol is converted into fructose by different types of sorbitol dehydrogenase (SDH) enzymes (Loescher et al., 1982) showing different subcellular localizations in the different tissue types of apple (Wang et al., 2009), perhaps indicating its distinct role in the different tissues. Sucrose is either directly transported into parenchyma cells by sucrose transporters (SUT), or is converted into fructose and glucose in the apoplast by cell wall bound invertase (CWI) before being transported into the cells by hexose transporters (HT) (Büttner and Sauer, 2000; Williams et al., 2000). Subsequently, the imported compounds enter the fruit's respiration pathway to generate the energy to fuel metabolic processes (White et al., 2015) and to contribute to sucrose, fructose, glucose, malate, and starch pools (Berüter et al., 1997; Li et al., 2012) (see **Figure 1**).

In the field of systems biology there is a growing interest in using "omics" technologies, mainly genomics, transcriptomics, proteomics, and metabolomics for better understanding growth and ripening related changes of apple fruit (Janssen et al., 2008; Zhang et al., 2010; Henry-Kirk et al., 2012; Li et al., 2012). Metabolomics plays a central role in systems biology focusing on identification and quantification of low molecular weight metabolites, which are end products of cellular regulation, especially when encountering various stress conditions (Fiehn, 2002; Fernie et al., 2004b; Roessner and Beckles, 2009). Metabolomics studies are commonly used to characterize complex physiological and biochemical changes occurring during fruit development (Zhang et al., 2010; Li et al., 2013). Nonetheless, knowledge of metabolite levels by itself is insufficient to unravel intracellular fluxes related with activity levels in different pathway and regulatory mechanisms. For example, fluxes through a pathway can change without a significant change in the levels of intermediate metabolites (Fernie et al., 2005). The application of isotopically labeled substrates to reveal the in vivo carbon flow levels through metabolic networks in responses to physiological stimuli or genetic modification has, lately, received extensive attention (Schwender et al., 2004; Sauer, 2006; Ampofo-Asiama et al., 2014; Buescher et al., 2015; Heux et al., 2017; Mbong et al., 2017a,b). Previous isotope labeling studies in developing apple fruit were focused at a specific growth stage (Berüter et al., 1997; Berüter, 2004) rather than covering the journey from flowering to fully mature fruit. This has triggered the question of how the metabolite levels and metabolic pathway activity changed with the various stages of apple fruit development, covering cell division, cell expansion, and maturation.

Therefore, the aim of this study was to create a comprehensive understanding of the dynamics of metabolic changes occurring throughout apple fruit development by studying the changing uptake and distribution of <sup>13</sup>C-label through feeding experiments on hypanthium tissue discs taken at distinct stages of fruit development. It is the first time that dynamic isotope labeling experiments have been performed at various stages of apple fruit development to study changes in carbon re-allocation during fruit growth. A wide range of polar metabolites were analyzed using gas chromatography coupled to mass spectrometry (GC-MS) characterizing the metabolic pool sizes and the <sup>13</sup>Clabel distribution in growing apple fruit at distinct stages of development covering all major events occurring during fruit development.

# MATERIALS AND METHODS

#### Plant Materials and Chemicals

Apples (Malus domestica Borkh., cv. "Braeburn") collected from seven designated 2-year-old trees grown at the KU Leuven Research orchard at Rillaar, Belgium (50◦ 57′ 48.8′′N, 4◦ 52′ 56.5′′E) were used for this study. During the 2015–2016 growing season fruit samples were harvested at five growth stages (30, 58, 93, 121, and 149 days after full bloom, i.e., after the flowering is fully completed, see Figure S1). Fruits were harvested in the morning and immediately transported to the MeBioS lab, KU Leuven, where the experiments were carried out. At each stage, at least 45 replicate fruits were collected from the seven trees.

Previously published physiological and morphological data from apple fruit development (Janssen et al., 2008; Li et al., 2012) were used to select stages matching five major events occurring during fruit growth and development, i.e., 30 days (cell division), 58 days (peak rate of cell expansion and starch accumulation), 93 days (decline of cell expansion rate), 121 days (decline of starch levels), and 149 days (ripening) after full bloom.

Analytical grade chemicals were purchased from Sigma-Aldrich (Belgium) and stored according to the manufacturer's instructions.

## Respiration, Ethylene Production, and Osmolality Measurements

The results used to characterize the physiological parameters of developing apple fruit were all expressed on a fresh weight basis. The respiration and ethylene production rates of developing apple fruit were measured based on the methods described in Bulens et al. (2011). With respect to the five growth stages, one or two fruits were placed in 1-liter air tight glass jars. The jars were flushed with regular air. After 2 h of flushing, the gas flow was stopped and the initial composition of the gases in the headspace was measured using a compact-GC (Interscience, Louvain La Neuve, Belgium). The final reading was taken after 24 h. Respiration and ethylene production rate were calculated from these data with respiration rate being expressed in terms of CO<sup>2</sup> production rate (nmol kg−<sup>1</sup> s −1 ).

The osmolality of the fruit was determined based on the freezing point depression of pressed juice extract of 10 fruits, as described previously (Berüter, 2004). After the water activity of the extracted juice was measured using aw-Kryometer (AWK-40; Nagy, Germany), the osmolality of the fruit was derived from a calibration curve prepared using NaCl<sup>2</sup> solutions of known osmolality.

in the cell wall space between sieve element-companion cell complexes (SE-CC) and parenchyma cells. Sorbitol is transported from the cell wall space into cells by sorbitol transporters (SOT). Sucrose is either directly transported into cells by sucrose transporters (SUT), or is converted into fructose and glucose in the cell wall space by cell wall bound invertase (CWI) before being transported into the cells by hexose transporters (HT). Metabolic pathways are described considering carbon metabolism in developing apple fruit and using the labeling of intermediates described in the text and information in the literature. Metabolites presented in gray text are those whose <sup>13</sup>C-labeling could not be reliably quantified in this study. αKG, α-ketoglutarate; 3PG, 3-phosphoglycerate; F6P, fructose 6-phosphate; Fru, fructose; G1P, Glucose 1-phosphate; G6P, Glucose 6-phosphate; GABA, γ-aminobutyrate; Glc, glucose; GPT, G6P/phosphate transporter; NI, neutral invertase; Oxa, oxaloacetate; PEP, phosphoenolpyruvate; PPP, pentose phosphate pathway; PPT, PEP/phosphate transporter; S6P, sucrose 6-phosphate; TCA cycle, tricarboxylic acid cycle; UDP-glc, uridine diphosphate-glucose; VI, vacuolar invertase; XPT, triose-phosphate transporter.

# <sup>13</sup>C-Isotope Labeling Experiments

In vivo <sup>13</sup>C labeling experiments were conducted using apple tissue discs cut from the harvested apple fruit, submerged in a liquid medium supplemented with 20 mM [U-13C]glucose (Figure S1). The choice of [U-13C]glucose was based on the assumption that glucose is transported into the parenchyma cells after sucrose being converted into glucose and fructose by apoplastic CWI. The fruit was sliced along the equatorial axis and tissue discs (∼10 mm diameter and ∼1 mm thickness) were collected from the hypanthium tissue using a cork borer. Excised tissue discs were washed three times in an isotonic solution containing 50 mM HEPES/KOH buffer (pH 7.0), 2 mM CaCl2, 2 mM MgCl2, 2 mM DTT (1,4-dithiothreitol) using betaine as an osmoticum, and while shaking at 90 rpm, to remove damaged cells. The osmotic strength of the solution was adjusted to the osmolality of the different growth stages to preserve the integrity of tissue discs submerged in liquid medium. As a result, the betaine concentration in the medium was increased from 160 mM at 30 days to 730 mM at 149 days. After washing thoroughly about 2.5 g of tissue discs was placed in a 50 ml flask containing 5 ml of buffer solution. This solution was identical to the one used to prepare the tissue discs, yet supplemented with 20 mM unlabeled glucose. After pre-incubation in unlabeled solution for 12 h the tissue discs were transferred to a solution containing [U-13C]glucose (≥99% enrichment). The solution was continuously aerated with humidified air at 20◦C in a controlled environment. Tissue discs were incubated for various time intervals (1, 2, 4, 6, 8, 10, or 24 h) after label introduction. Subsequently, samples were washed three times using 100 ml of a hypotonic solution of 50 mM HEPES/KOH buffer (pH 7.0) to remove the salt and excess of substrate from the tissue. After washing, excess medium was removed from the discs using paper tissue. The tissue samples were rapidly frozen in liquid N<sup>2</sup> and subsequently stored at −80◦C prior to GC-MS analysis. Each <sup>13</sup>C labeling experiment was performed three times starting from independent biological plant material.

#### Primary Metabolite and Starch Analysis

Extraction and derivatization of polar metabolites were performed following the protocols described by Bekele et al. (2014). The frozen hypanthium tissue samples were homogenized using Mixer Mill (Retsch, MM 200, Haan, Germany) at a frequency of 30 Hz for 1 min. 200 mg of fresh weight of powdered frozen apple tissue material was extracted using 1 ml of methanol and incubated in a thermomixer (Eppendorf AG, Hamburg, Germany) at 70◦C for 15 min, shaking at 1,400 rpm. The methanol extract was centrifuged at 22,000 g for 20 min at 4◦C to separate the aliquot from the pellet. Next, the supernatant aliquot was dried at 50◦C on a heating block (Stuart, sample concentrator (SBH CON/1), Bibby Scientific Limited Stone, and Staffordshire, UK) under a stream of nitrogen gas. Subsequently, metabolites were derivatized by methoxymation followed by silylation. Firstly, 120 µl of methoxyamine hydrochloride (20 mg methoxyamine hydrochloride (MEOX) in 1 ml pyridine) was added to each sample and incubated in the thermomixer for 60 min at 30◦C. Secondly, 120 µl of BSTFA (N,O-Bis(trimethylsilyl(TMS))trifluoroacetamide) was added to each sample and incubated in the thermomixer for 120 min at 45◦C. Finally, 100 µl of the derivatized sample was transferred into glass vials containing deactivated glass inserts. In addition, standard compounds were injected alongside the samples in order to allow for the calculation of absolute metabolite concentrations.

GC-MS analysis was performed on a GC 7890A coupled with 5975C MS (Agilent Technologies, Palo Alto, CA). Initially, the sample was volatilized at 230◦C, inside the deactivated glass liner (SGE Analytical Science, Victoria, Australia). The chromatographic separation was performed on HP-5 ms column (30 m × 250µm ID, 0.25µm film thickness, Supelco, Bellefonte, CA) with programmed temperature ramp. Helium was used as a carrier gas applying a constant flow of 1 ml min−<sup>1</sup> . Each sample was injected in two different split modes. One with split ratio of 5:1 injection, which was optimized for less abundant metabolites. For this injection, the oven temperature was set to 50◦C for 2 min, ramped at 10◦C min−<sup>1</sup> to 325◦C and held for 5 min at 325◦C (34.5 min run time). Second, a high split ratio of 500:1 was used for most abundant metabolites corresponding to malate, asparagine, quinate, fructose, glucose, sorbitol, and sucrose. In 500:1 split mode, the oven temperature was set to 90◦C for 2 min and ramped to 325◦C switching between 50 and 10◦C min−<sup>1</sup> and held at 325◦C for a further 5 min resulting in a run time of 17.3 min. The mass selective detector (MSD) was operated in the electron ionization (70 eV) mode with quadrupole and MS ion source temperatures maintained at 150 and 230◦C, respectively. The detector was activated to record throughout the mass spectra range 35–500 m/z.

Starch extraction and determination was carried out using the method described by Hendriks et al. (2003). The pellet obtained after the extraction of polar metabolites was further extracted with 80% ethanol at 80◦C for 15 min. The pellets were dried and homogenized in 0.2 mM KOH and incubated for 1 h by heating them at 95◦C. After acidification to pH 4.9 with 1 M acetic acid/sodium-acetate buffer, the suspension was digested overnight with a mixture of amyloglucosidase and α-amylase. The glucose content of the supernatant was then used to assess the starch content of the sample.

# Metabolite Annotation and <sup>13</sup>C-Label Enrichments

Metabolite annotation was performed using Agilent MSD Chemstation Software (Agilent Technologies, Santa Clara, USA) by comparing the acquired spectra with an in-house built library, with the Agilent Fiehn Metabolomics Library, and with the NIST2011 Library (National Institute of Standards and Technology, Gaithersburg, MD, USA) (Table S1). MS Correction Tool was used to correct isotopomer fractions of a particular metabolite fragment for the natural stable isotopes (Wahl et al., 2004). The percentage <sup>13</sup>C enrichment was calculated from the total abundance of <sup>12</sup>C and <sup>13</sup>C ions in a particular metabolite pool (Araújo et al., 2014).

#### Data Analysis

To reveal the correlation structure of the data principal component analysis (PCA) was conducted using the Unscrambler <sup>R</sup> X software (version 10.3, CAMO A/S, Trondheim, Norway). Heat maps were generated to compare the concentration of metabolites during fruit development (30, 58, 93, 121, and 149 days after full bloom) using the MultiExperiment Viewer software (MeV v4.9.0, http://www. tm4.org/, Saeed et al., 2003). A paired t-test analysis was used to compare the mean difference between the glucosyl and fructosyl moieties of sucrose with a significance level of p-value = 0.05 using JMP software, version 13.0 (SAS Institute Inc., Cary, NC, USA).

#### RESULTS

# Experimental Setup and Physiological Parameters of Developing Apple Fruit

The results reported in this article were carried out at five stages of fruit development (30, 58, 93, 121, and 149 days after full bloom) throughout the 2015–2016 growing season. Preliminary <sup>13</sup>C labeling experiments were performed to test the uptake capacity of [U-13C]glucose by tissue discs cut from developing apple fruit at three distinct growth stages (from the earlier growing season, 2014–2015) including fully ripe apple fruit. The latter showed little to no <sup>13</sup>C-label incorporation (Figure S2). In contrast, an appreciable amount of <sup>13</sup>C-label was incorporated into tissue from developing apple fruit with the percentage labeling varying with progressing fruit development (Figure S2). The percentage labeling of glucose increased with fruit development while fully ripe apple fruit showed the lowest uptake rate. In addition, the incorporation of <sup>13</sup>C-label into the various metabolites increased linearly with the concentration of [U-13C]glucose increasing from 5 to 20 mM (Figure S3).

Physiological parameters of apple fruit were measured during fruit development. The changes in fruit osmolality, mass, diameter, respiration and ethylene production rate, and starch content during fruit growth are shown in **Figure 2**. The osmolality of the fruit increased from 30 days all the way to fruit maturation. Simultaneously, fruit mass and diameter increased approximately linearly from 30 to 149 days. During fruit growth, the respiration rate decreased from 460 nmol kg−<sup>1</sup> s −1 in young fruit to 70 nmol kg−<sup>1</sup> s −1 at 121 days followed by a slight increase at 149 days. In contrast, ethylene production rate remained relatively low until 121 days, showing a slight increase at 149 days. Starch content reached its peak level within 93 days and decreased throughout the later stages of fruit growth.

# Metabolite Changes throughout Fruit Development

The metabolite changes of the developing apple fruit were analyzed using a GC-MS-based metabolite profiling approach. Typical chromatograms are shown in **Figure 3**. Two injections were realized in this study, one with a split ratio of 5:1 (**Figure 3A**), which was optimized for less abundant metabolites (34.5 min run time) and one with a split ratio of 500:1 (**Figure 3B**) used for the most abundant metabolites (17.3 min run time). To verify the identity of the metabolites individual standard compounds were injected. Detailed peak information for the metabolites identified are presented in Table S1.

**Figure 4** shows the PCA bi-plot that reveals associations among sugars, sugar alcohols, organic and amino acids, and polyphenols observed in the <sup>13</sup>C labeling experiments executed at each of the five growth stages. The measured metabolites are represented by the open circles whereas the individual tissue samples are represented by their scores (closed symbols) and colored according to the five growth stages. The amount of variation covered by PC1 and PC2 was 48 and 13%, respectively,

covering 61% of the total variance. The scores showed a clear order according to growth stage. The largest variation was observed between the five growth stages (30, 58, 93, 121, and 149 days) whereas the effect of changes in incubation time was less pronounced (as indicated by the size of the closed symbols). Based on the PCA bi-plot some of the sugars, such as fructose and sucrose were positively associated with progressing growth stage. In contrast, most of the organic and amino acids and polyphenols were negatively correlated with growth stage.

To provide a more detailed representation of the results, heat maps were generated to show the metabolite changes during fruit growth (**Figure 5**). Fructose content increased significantly from 30 to 93 days and varied slightly later in development. The increase of fructose content was accompanied by a decrease of sorbitol level and starch degradation (**Figure 2F**) in the later growth stages. In contrast to the behavior observed in sorbitol, sucrose content increased from 30 to 149 days. Glucose reached its maximum level within 58 days and decreased throughout the later stages of fruit growth. G6P and F6P were high at 30 days and decreased gradually throughout fruit growth. Organic acids such as fumarate, succinate, pyruvate, quinate, glycerate, α-ketoglutarate, threonate, shikimate, and dehydroascorbate contents were high at 30 days and decreased substantially throughout fruit growth. In contrast, malate content increased up to 58 days and decreased considerably toward 149 days. Like most organic acids, the levels of most amino acids were very high in the early development and decreased gradually throughout fruit development. Most importantly, the major phenolic compounds present in apple (epicatechin, catechin, and chlorogenate) were very high at 30 days but strongly declined throughout fruit development (**Figure 5**, Figure S6C).

#### Validation of Isotopomers Analysis

Fragments used for isotopomer analysis were selected by comparing the experimental mass isotopomer of tissue discs incubated with unlabeled substrate with the theoretically expected values (Table S2). The extent of incorporation of <sup>13</sup>C-label into a certain metabolite was calculated using the molecular ion whenever available, otherwise, the average labeling of multiple fragments containing the carbon backbone of the molecular ion was considered. In some cases, when no suitable fragments could be found to represent the entire carbon skeleton, a single fragment was selected to represent the labeling of the metabolite in question (Roessner-Tunali et al., 2004). For example, m/z 218 of valine contains C1-C2, whereas m/z 144 contains C2-C5. However, m/z 144 seems to be superimposed with a minor peak from m/z 147, introducing about 9.3% error in m3, making its use for isotopomer analysis ambiguous (see Figure S4A). As a consequence, fragment m/z 218 was selected to represent the percentage labeling in valine. Similarly, fragment m/z 319 (C3-C6) of glucose was selected instead of taking into account m/z 160 that contains the first two carbons of glucose

(C1-C2) because m/z 160 contains around 5% error (Figure S4B). In addition, as shown in **Figure 3C**, m/z 361 and 451 corresponding to the mass spectra of sucrose were selected to represent the glucosyl and fructosyl moieties of sucrose, respectively, using fragmentation simulation available in the NIST library. After the incubation of the tissue discs for 24 h with 20 mM [U-13C]glucose medium, a clear change in the mass isotopomer distributions of sucrose moieties were observed (**Figure 3D**). Koubaa et al. (2012) reported that the fructosyl moiety of sucrose contributes about 40% to the intensity of m/z 361, which, by the fragmentation simulation, is considered solely as the glucosyl moiety of sucrose. In contrast, the glucosyl moiety of sucrose contributes only a small amount (near 5%) to the intensity of m/z 451 (Koubaa et al., 2012), which implies that any dilution effect linked to the two moieties of sucrose could lead a different degree of labeling for each fragment. It is important to note that the fragments of sucrose at m/z 361, 437, and 451 gave the same isotopic enrichment (Table S2), as reported in previous studies (Alonso et al., 2005; Füzfai et al., 2008). In summary, we were able to estimate the natural stable istope abundance pattern in a range of fragments for each metabolite by selecting one or more representative fragments.

# Analysis of the <sup>13</sup>C-Label Accumulation

<sup>13</sup>C-isotope feeding experiments were performed to get a better insight into the metabolite changes and metabolic pathway activity during apple fruit development at five selected growth stages. Following the feeding of [U-13C]glucose, the <sup>13</sup>C-label was incorporated into the various metabolites to different degrees of labeling depending on incubation time, metabolic pathway activity, and growth stage (**Figure 6**). The <sup>13</sup>C-label was incorporated considerably into glycolysis intermediates 1 h after the addition of exogenous [U-13C]glucose. The labeling dynamics of metabolites in the glycolysis pathway was very fast and reached more or less a plateau within a few hours as indicated, for instance, by a nearly constant labeling of G6P (**Figure 6A**). Moreover, a higher percentage labeling of the majority of metabolites was observed at 30 days as compared to the later growth stages. In contrast, sorbitol, epicatechin, and catechin remained unlabeled during the 24 h incubation period, irrespective of the growth stages.

At 30 days, the enrichment of intracellular <sup>13</sup>C-glucose reached 2.2% within 6 h and increased to 4.8% after 24 h while G6P and pyruvate ramped to 54.5 and 24.1% within 6 h, respectively. Between 6 and 24 h, the labeling of G6P slightly decreased while a continuous increase or nearly constant labeling was observed in the labeling dynamics of downstream metabolites. The labeling of pyruvate, citrate, succinate, malate, GABA, isoleucine, phenylalanine, proline, and valine increased over the time course of the feeding experiment. The labeling of α-ketoglutarate, alanine, aspartate, glutamate, and serine appeared to be approaching isotopic steady state between 10 and 24 h of the incubation. The incorporation of <sup>13</sup>C-label into fructose and malate (the most abundant sugar and organic acid metabolite in apple, respectively) were quite similar and continuously increased during the time course of the experiment, except for fructose exhibiting a longer lag phase in <sup>13</sup>C-label accumulation than malate. As compared to fructose, a much larger proportion of label was directed to sucrose that strongly increased throughout the feeding experiment, reaching over 35% within 24 h. Moreover, there was no considerable difference between the labeling pattern of glucosyl and fructosyl moieties of sucrose (**Table 1**).

In the later growth stages (58–149 days), there was a much larger increase in the labeling of glucose in comparison with 30 days. The opposite was true for most of other metabolites, showing a decrease in percentage labeling throughout fruit development. It should be noted that GABA behaves differently. A decent amount of labeling of sucrose also took place at 58 days, exhibiting a gradual decrease in percentage labeling, whereas no considerable labeling of fructose could be found in the later growth stages.

#### Net <sup>13</sup>C-Incorporation in Contrast with Unlabeled Pools

**Figure 7** shows the total <sup>13</sup>C-label incorporated into selected metabolites in comparison to the unlabeled pools of tissue discs retrieved from five growth stages during the feeding experiment with 20 mM [U-13C]glucose. Interestingly, the net <sup>13</sup>C-incorporation into sucrose at 30 days reached a nearly constant value within a few hours (**Figure 7**) whilst the percentage <sup>13</sup>C enrichment strongly increased, reaching over 35% within 24 h (**Figure 6A**) as the pool size of sucrose was reduced (from 8.4 µmol g−<sup>1</sup> at time zero to 4.02 µmol g−<sup>1</sup> at 24 h). The percentage <sup>13</sup>C labeling of sucrose at 58 days was 11.9% at 24 h,

however, the net <sup>13</sup>C-incorporation (2.64 µmol g−<sup>1</sup> ) showed a 2 fold increase relative to the early growth stage (1.4 µmol g−<sup>1</sup> ; **Table 2**). This is because the total sucrose pool was smaller at 30 days (4.02 µmol g−<sup>1</sup> ) than 58 days (22 µmol g−<sup>1</sup> ).

Unlike most other metabolites, the total isotope accumulation in intracellular glucose was much more prominent in the later growth stage. For instance, after 24 h of incubation the absolute amount of the total isotope incorporated at 30 days in glucose reached 6.31 µmol g−<sup>1</sup> whereas the value was considerably higher in the later growth stage (149 days), reaching 11.2 µmol g −1 although the total concentration of glucose was decreased at the later growth stage (**Figure 7**). Fructose labeling was high in young fruit and decreased over the growth stage. In addition, more label was incorporated into malate than in all other organic acids combined (**Table 2**).

# DISCUSSION

# Rationale for Developing <sup>13</sup>C-Labeling Experimental Setup for Developing Apple Fruit

In developing apple fruit, sorbitol and sucrose enter the fruit respiratory metabolism (Berüter et al., 1997) predominantly through the apoplastic pathway (Zhang et al., 2004), essentially after being converted into free fructose and glucose (Loescher et al., 1982; Büttner and Sauer, 2000; Williams et al., 2000). Several studies have been published using labeling with [U-<sup>13</sup>C]glucose to study the central metabolic fluxes in many plant species (apple, Berüter, 2004; potato tubers, Roessner-Tunali et al., 2004; maize root tips, Alonso et al., 2007). To establish incorporation of <sup>13</sup>C labeled sugars within the central metabolism, we allowed tissue discs cut from growing fruit to take up uniformly <sup>13</sup>C labeled extracellular glucose from the medium. We therefore carried out an additional set of experiments to establish the rate of glucose uptake and the labeling kinetics of the various metabolite pools. The preliminary experiment was performed at three distinct growth stages: i.e., 30 days (cell division), 90 days (cell expansion), and 150 days (maturation) throughout the 2014–2015 growing season. These preliminary results indicated a fast labeling of the upstream metabolites resulting in saturating of the labeling within a few hours of incubation. The <sup>13</sup>C-label was incorporated into the various metabolites in just 1 h after the addition of labeled substrate. This rate of incorporation is similar to what was observed for the accumulation of [14C]sorbitol in apple tissue discs (Zhang et al., 2004) but very fast when compared with the uptake of carboxyfluorescein (a common fluorescent marker of phloem transport) supplied to the intact apple fruit pedicel, which takes up to 4 h to reach the fruit flesh (Zhang et al., 2004). It must be



The glucosyl (fragment m/z 361) and fructosyl (fragment m/z 451) moieties were used for the calculation of <sup>13</sup>C-label enrichment, the average labeling reflects the labeling state of sucrose. Based on a paired t-test identifying the difference in labeling between the two sucrose moieties, the overall difference of the mean and the corresponding p-value was calculated.

pointed out that in the current experiments [U-13C]glucose was supplied directly to tissue discs being a simplified model system whereas Zhang et al. (2004) supplied the carboxyfluorescein to the intact apple fruit pedicel so that it covered a longer transport distance to the fruit flesh.

An additional experiment showed that the incorporation of the <sup>13</sup>C-label into the various metabolites increased linearly with the concentration of [U-13C]glucose in the range of 5–20 mM (Figure S3). This is in good agreement with the relationship between the source leave and sink fruit reported previously (Paul and Foyer, 2001; Morandi et al., 2008; Dash et al., 2013; White et al., 2015). These authors indicated that growth may be controlled by source and sink strength. For example, increasing carbohydrate availability through photosynthesis can increase growth to a certain extent. Also for sorbitol it has been demonstrated that its uptake by isolated tissue discs and in intact apple fruit is linear with respect to concentration (Berüter and Studer-Feusi, 1995). The final experiments were conducted using 20 mM [U-13C]glucose to ensure excess amount of label in the medium to allow incorporation of decent amounts of label in a broad range of metabolites (see "Materials and Methods"). In addition, the osmotic strength of the medium was adjusted to the osmolality of the different growth stages in the range of 228–790 mOsmol kg−<sup>1</sup> by using betaine (**Figure 2A**) – thus to preserve the integrity of tissue discs submerged in liquid medium – whilst the glucose concentration was kept the same for the different growth stages.

# Dynamic Labeling Revealed Metabolite Features of Developing Apple Fruit

#### Reduction in Respiration Rate during Fruit Growth Can Be Related to Changing Label Incorporation

Like pear fruit (Zhang et al., 2005), apple is characterized by high respiration rate at the early growth stage which then generally decreases throughout the season until prior to initiation of fruit ripening (Bepete and Lakso, 1997). During the early stage (30 days) when cell division was most active (Janssen et al., 2008), fruit showed the highest respiration rate (**Figure 2D**), driving utilization of the imported labeled glucose into the wider metabolism (**Figure 6**). Given the expected high FK/HK (fructose- and hexose-kinase) enzyme activities at the early growth stage of the fruit (Li et al., 2012), rapid utilization of the imported sugars for cell division and fruit growth is facilitated (Berüter, 1985; Dash et al., 2013). With fruit development progressing respiration rate gradually declined and most of the labeled glucose taken up was no longer transferred to the other metabolites (**Figures 6**, **7**). This reduction might be associated with the decrease in HK activity during fruit growth (Li et al., 2012). Similarly, Beauvoit et al. (2014) reported a higher glycolytic flux during cell division and a sharp decline during cell expansion of tomato fruit.

In general, carbon accumulation in a sink fruit depends on sink characteristics such as uptake rate (sink size × sink activity), photosynthesis and respiration rate of the fruit, as well as acquisition of carbon via the peduncle (Henton et al., 1999; Paul and Foyer, 2001; White et al., 2015). As reported for tomato (Greve and Labavitch, 1991), the <sup>13</sup>C-label incorporation into actively growing tissue is much higher than into fully grown yet ripening tomato pericarp. Therefore, this data confirms the tight relationship between the fruit's respiration rate and its glycolytic capacity, both changing during fruit development (Figure S5). Moreover, the declining net isotope accumulation into various metabolites observed at 58 and 93 days growth stages showed a slightly higher accumulation in some of TCA cycle metabolites such as fumarate, succinate, and malate at the later growth stage of 149 days (**Table 2**), which was in parallel with a slight increase in respiration and ethylene production rate (**Figures 2D,E**). In addition, the <sup>13</sup>C-label incorporated into GABA was high at 149 days (1.5 µmol/g) as well as at 30 days (3 µmol/g), but lower at the intermediate growth stages (**Table 2** and **Figure 6C**). Thus, the observed increase in label accumulation in TCA cycle metabolites, as well as the slight increase in the respiration and ethylene measurements at final growth stage reflects the onset of the climacteric rise (Dilley, 1981). This suggested that around 149 days after full bloom the "Braeburn" apple fruit were sufficiently mature for commercial harvest, at least for the concerning orchard and growth season.

The percentage <sup>13</sup>C-label incorporated into glucose, especially in the early growth stages, was lower than the labeling of some other downstream intermediate metabolites such as G6P and pyruvate. This leads to the hypothesis that, early during development, tissue favors extracellular sugars over cellular sugars already stored in the vacuole as their main source of carbon, thus maximizing the fruit's net carbon loading. The fast glycolytic turnover of the imported labeled glucose only represents a small fraction against a large background of unlabeled glucose stored in the vacuole resulting in a low overall percentage of labeled glucose. This suggests a high carbon conversion efficiency in young fruit. The interfering effect of existing unlabeled pools of metabolites in interpreting labeling data has been recognized before (Alonso et al., 2005, 2007).

The pool of the glycolysis intermediate G6P was high at 30 days and decreased gradually throughout fruit development. The higher concentration of G6P early in development can be related to the higher respiratory flux in young fruits, resulting in a high incorporation of label into G6P, reaching over 54% (**Figure 6A**). Likewise, facilitated entry of radioactive glucose into the hexose phosphate pools were reported for tissue discs retrieved from 50% fully grown apple fruit (Berüter et al., 1997). G6P could have been incorporated into sucrose, starch and cell wall polysaccharides (Alonso et al., 2005), as well as into the glycolytic pathway as indicated by the massive <sup>13</sup>C-label incorporation in the current experiments. G6P can also be used as a substrate for inositol synthesis in the cytosol, which can serve as a precursor for phytate synthesis (Mitsuhashi et al., 2008). During the cell expansion phase at 93 days, a slightly higher percentage labeling of myo-inositol was observed in the current study (**Figure 6A**) although it was evident that the net <sup>13</sup>C-incorporation showed a higher value at 30 days relative to the later growth stage. The decreased G6P labeling between 6 and 24 h in young fruit is quite similar to that previously observed in potato tuber discs (Roessner-Tunali et al., 2004). The decrease could be related to either the turnover of unlabeled starch or the cycling of unlabeled sucrose. However, starch degradation mainly occurs in mature fruit (Janssen et al., 2008; Li et al., 2012) thus, the most likely explanation is the regeneration of G6P through sucrose cycling (Berüter et al., 1997; Beauvoit et al., 2014). In agreement with our hypothesis, a decreasing sucrose level was observed during the 24 h incubation at 30 days (**Figure 7**).

#### Sucrose Cycles in Growing Fruit

As reported for potato tuber discs (Roessner-Tunali et al., 2004), a substantial labeling of sucrose took place during fruit development (**Figure 6A**). This suggests that in addition to the large flux of sucrose translocated from the leaves (Klages et al., 2001), the fruit is actively sustaining its sucrose synthesis apparatus, confirming that sucrose is the major product of sugar metabolism. Sucrose may be synthesized via conversion of F6P and uridine diphosphate-glucose (UDP-Glc) catalyzed by sucrose phosphate synthase (SPS) or from UDP-Glc and fructose as catalyzed by sucrose synthase (SuSy) which has higher activity in young fruit (Li et al., 2012) (see **Figure 1**). Given there was no significant difference between the labeling pattern of glucosyl and fructosyl moieties (**Table 1**), sucrose synthesis is most likely mediated by SPS with F6P inheriting the labeling directly from G6P through isomerization, suggesting that the hexose-phosphate isomerase reaction is at equilibrium.


TABLE 2 | Total isotope accumulating into selected metabolites of apple tissue discs retrieved from five growth stages and incubated in 20 mM [U-13C]glucose for 24 h, at which most of the metabolites are reaching isotopic steady state.

Values are means ± SE (n = 3). nd, not detected, due to very low metabolite concentrations it was not possible to calculate reliably the relative distribution of mass isotopomers both in unlabeled and labeled ions.

Alonso et al. (2005) demonstrated that the labeling state of the glucosyl and fructosyl moieties in isotopic steady state conditions reflects the labeling state of G6P and F6P, respectively. At the same time, fructose labeling (**Figure 6A**) exhibited a longer lag phase as compared to G6P and sucrose. If sucrose was coming from fructose and UDP-Glc via the SuSy mediated reaction, the fructose moiety of sucrose would have been diluted by unlabeled fructose, at least in the first few hours of incubation when fructose labeling was not initiated. This means that sucrose synthesis via SuSy would lead to a higher labeling of the glucosyl moiety of sucrose, when labeled glucose was supplied (Geigenberger and Stitt, 1993). Altogether, equal labeling of the two moieties of sucrose can only be achieved through SPS and therefore, sucrose synthesis was most likely mediated by SPS.

In young fruit at 30 days, the overall sucrose content decreased by twofold (from 8.4 to 4.01 µmol g−<sup>1</sup> ) during 24 h of incubation. Its increased percentage labeling and its overall low and decreasing levels indicated sucrose was actively metabolized during early development in line with the high metabolic demand related to cell division and growth. Evidence for the rapid turnover of sucrose during the early growth stage of tomato fruit—after being transported into the vacuole, due to a higher vacuolar invertase (VI) activity—is reported by Beauvoit et al. (2014). Similarly, a high VI activity during the early growth stage of rapid cell division of apple and peach fruit has already been published (Li et al., 2012; Zhang et al., 2013). This implies that VI and SPS enzymes are recognized as the main players of sucrose cycling in many sink tissues.

At 58 days, where cell expansion rate reached its peak (Janssen et al., 2008; Li et al., 2012), there was a significant increase in total isotope accumulation in sucrose as compared to the early and the later growth stages (**Table 2** and **Figure 7**). The two fold increase in net <sup>13</sup>C-label incorporation into sucrose implies that <sup>13</sup>C allocation into reserves was high during cell expansion with limited conversion to intermediate metabolites. It underlines the importance of sucrose cycling in controlling the glucose level in the cell, regulating the glycolytic capacity of plant cells. This increased labeling of sucrose can be related to the peaking glucose phosphorylation activity and the higher glucose content during this period of apple fruit development as observed by Zhao et al. (2016). In line with their report, a higher glucose content was observed at 58 days (**Figure 5**, Figure S6).

SPS is known to play an important role in this as well but shows various behaviors for the different fruits. In developing peach fruit SPS activity was reported to increase during the most rapid sucrose accumulation phase (Lowell et al., 1989; Zhang et al., 2013), which is consistent with the higher sucrose labeling observed at 58 days. In apple fruit, SPS activity was shown to increase slightly with progressing fruit growth followed by a rapid increase during ripening, mirroring the reduction in starch level (Berüter and Studer-Feusi, 1997; Li et al., 2012), which is inconsistent with the currently decline in labeling of sucrose toward maturation. Similarly, an increase in SPS activity during kiwifruit ripening was reported (Nardozza et al., 2013). This concurrent high SPS activity and low labeling of G6P and sucrose currently observed in mature fruit (**Figure 6A**) leads to the hypothesis that G6P is more likely to originate from starch degradation in the plastid than from exogenous [U-13C]glucose. Nardozza et al. (2013) observed a positive correlation between starch accumulation in high starch kiwifruit genotypes and SPS activity which leads to sucrose accumulation during ripening.

Furthermore, the incorporation of label into fructose can be explained from sucrose hydrolysis into fructose catalyzed by invertase and/or SuSy; particularly VI and SuSy are known to be more active in the early growth stage of some Rosaceae fruit such as peach (Lo Bianco et al., 1999) and apple (Li et al., 2012). The observed slower and lower labeling of fructose at 30 days can be explained from the much larger pool of unlabeled fructose (**Figure 7**). Such slow labeling of fructose was already found for Arabidopsis indicating that the hydrolysis of sucrose into glucose and fructose is evidently a slow process (Szecowka et al., 2013). Moreover, the absence of fructose labeling in young fruit during the first few hours of incubation suggests the glucose-fructose isomerase activity might be negligible with fructose being mainly formed from sucrose. Berüter (2004) reported a lower labeling of fructose as compared to the labeling of sucrose after addition of <sup>14</sup>C-glucose. However, considering the absolute pool size of labeled carbon accumulated in fructose after 24 h (8.4 µmol g−<sup>1</sup> , **Figure 7**), which was higher than the pool of labeled sucrose, shows that the actual flux into fructose was very high. The decline in fructose labeling with fruit development (**Figures 6A**, **7**) can be explained from the decreased activities of invertase and SuSy observed for apple (Li et al., 2012). In general, the labeling of fructose supports the hypothesis of sucrose cycling in apple (Berüter et al., 1997; Beauvoit et al., 2014). In contrast to sucrose and fructose, there was no <sup>13</sup>C-label incorporated into sorbitol at any growth stage (**Figure 6A**) suggesting the conversion of sorbitol into fructose is effectively irreversible in developing apple fruit (Berüter et al., 1997).

Besides the gradual decline in percentage labeling, fructose and sucrose both increased in their concentration with fruit growth, becoming the most abundant soluble sugars in mature fruit (Figure S6). Fructose accumulation can be explained coming from either sorbitol or sucrose. Both sorbitol and sucrose are being transported from the leaves to the sink tissue and either are stored as such or converted into fructose (Loescher et al., 1982; Büttner and Sauer, 2000; Williams et al., 2000; Klages et al., 2001). The decrease of sorbitol levels during fruit growth can be related to the observed increase of fructose levels, which for sucrose was not the case. The increasing sucrose content with progressing fruit development can be associated with a decrease in sucrose-cleaving enzymes, invertase and SuSy in addition to an increase in sucrose synthesizing enzyme, SPS (Berüter and Studer-Feusi, 1997; Li et al., 2012). Another comprehensive example in addition to sucrose synthesis via SPS is starch degradation sustaining the energy used to fuel metabolic processes, hence, accumulation of incoming sucrose (Berüter, 2004; Nardozza et al., 2013; Mesa et al., 2016; **Figure 2F**).

#### Young Fruits Are Characterized by High Levels of Organic and Amino Acids and Polyphenols

Extensive metabolic changes occurred during fruit growth (**Figure 5,** Figure S6). Organic acids, free amino acids, and phenolic compounds exhibited a significant higher concentration in the early growth stage. These large pools of organic and amino acid metabolites can serve as building blocks for cell growth at the early growth stage. In addition to the larger pools of these metabolites, the higher accumulation of large amount of label in organic acids (**Figure 6B**) and amino acids (**Figure 6C**) observed in the early growth stage can be associated with a higher rate of glycolysis/TCA cycle metabolism and subsequent contribution to protein synthesis (Fernie et al., 2004a; Li et al., 2012) related to cell division. This allows the fruit to grow to a bigger extend (Farinati et al., 2017). The decrease in level and labeling with growth stage can be explained from the shift in processes from mainly cell division toward mainly cell expansion and maturation. Similarly, Ishihara et al. (2015) reported a higher rate of protein synthesis in young leaves characterized by a gradual decline with leave maturation.

The amount of label incorporated into malate was considerably higher than the total amount of label in the other TCA cycle intermediates (**Table 2**) which can be understood from the fact that malate is one of the major storage compounds in apple (Berüter et al., 1997). Interestingly, it should be noted that in all of the growth stages, fumarate showed a substantial increase in its concentration during the feeding experiments (see Figure S6B). Pyruvate and alanine showed a consistent pattern in their dynamics of label enrichment. The labeling of alanine was slightly higher than its precursor pyruvate. However, MS measurements may have been biased as consequence of very low pyruvate concentrations, as well as the high turnover rate of pyruvate and its localization in multiple subcellular compartments, with unlabeled pyruvate present in the various cellular compartments diluting the labeled pyruvate mainly present in the cytosol (Buescher et al., 2015). Isoleucine and valine, which are synthesized from pyruvate were strongly labeled. Aspartate and glutamate, which are formed by transamination reaction from oxaloacetate and α-ketoglutarate (Szecowka et al., 2013), also showed a more rapid <sup>13</sup>C-label accumulation at 30 days.

The absence of label in epicatechin and catechin (**Figure 6C**) metabolites can be explained by their metabolic remoteness from the supplied <sup>13</sup>C substrate although the label, to a certain level, did reach phenylalanine, which serves as a bridge between the plant primary metabolism and the polyphenolic pathway. Like most organic acids in the TCA cycle, chlorogenate, catechin, and epicatechin were synthesized, to a large extent, early during development (30 days) and gradually decreased during fruit growth (**Figure 5**). However, the expression level of anthocyanidin reductase and leucoanthocyanidin reductase, which are responsible for the synthesis of catechin and epicatechin in apple, are known to increase toward maturity (Henry-Kirk et al., 2012). Previous studies have suggested that the decline in the content of phenolic compounds is associated with a dilution effect linked to fruit growth (vacuole expansion in mature apple fruit) (Renard et al., 2007). In general, the changes of metabolite levels were consistent with previous studies of apple fruit growth (Zhang et al., 2010; Li et al., 2013).

In conclusion, the novelty of the present work is in the dynamic labeling experiments performed at various stages of fruit growth to study carbon re-allocation metabolism during apple fruit development. Interestingly, short time isotope feeding experiments showed a wide range of label distributions between the different growth stages, depending on the proximity of each metabolite to the substrate. Isotopic steady state labeling was achieved in the majority of metabolites within few hours of exogenous [U-13C]glucose addition. It is important to remark that young fruit is characterized by a greater degree of label accumulation, related to the higher metabolite demand during cell division and fruit growth. Due to the reduced metabolic activity, as mirrored by respiratory rate, <sup>13</sup>C reallocation into various metabolites gradually declined as the requirement for cell growth and carbon skeletons decrease with fruit development. The work presented here can serve as a platform for further studies to understand developmental changes associated with fruit growth. Positional isotope feeding experiments and metabolic modeling can be considered to furthermore quantify fluxes through the glycolysis and the pentose phosphate pathway.

#### AUTHOR CONTRIBUTIONS

WB, MH, AG, and BN designed the experiments. WB carried out the experiments, data analysis and prepared the figures. MH,

#### REFERENCES


AG, and BN guided the experimental work. WB interpreted the results and wrote the manuscript with contributions from all the authors (VM, MH, AG, WVdE, and BN).

#### FUNDING

This work was funded by the Research Council of the KU Leuven (OT 12/055 and C16/16/002) for financial support. This research was carried out in the context of the European COST Action FA1106 ("QualityFruit").

#### ACKNOWLEDGMENTS

The authors wish to thank the KU Leuven Research orchard staff (Johan Verheyen) for providing the apple fruit, Prof. John Lunn and Prof. Mark Stitt for useful discussions.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2017. 01785/full#supplementary-material


and carbohydrate changes in peach. Acta Physiol. Plant. 35, 589–602. doi: 10.1007/s11738-012-1100-1


Zhao, J., Sun, M., Hu, D., and Hao, Y. (2016). Molecular cloning and expression analysis of a hexokinase gene, MdHXK1 in Apple. Hortic. Plant J. 2, 67–74. doi: 10.1016/j.hpj.2016.06.005

**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 Beshir, Mbong, Hertog, Geeraerd, Van den Ende and Nicolaï. 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 Systems Analysis With "Simplified Source-Sink Model" Reveals Metabolic Reprogramming in a Pair of Source-to-Sink Organs During Early Fruit Development in Tomato by LED Light Treatments

Atsushi Fukushima<sup>1</sup> , Shoko Hikosaka<sup>2</sup> , Makoto Kobayashi<sup>1</sup> , Tomoko Nishizawa<sup>1</sup> , Kazuki Saito1,3, Eiji Goto<sup>2</sup> and Miyako Kusano1,4 \*

<sup>1</sup> RIKEN Center for Sustainable Resource Science, Yokohama, Japan, <sup>2</sup> Graduate School of Horticulture, Chiba University, Chiba, Japan, <sup>3</sup> Graduate School of Pharmaceutical Sciences, Chiba University, Chiba, Japan, <sup>4</sup> Graduate School of Life and Environmental Sciences, University of Tsukuba, Tsukuba, Japan

#### Edited by:

Ute Roessner, The University of Melbourne, Australia

#### Reviewed by:

Adriano Nunes-Nesi, Universidade Federal de Viçosa, Brazil Maria Valeria Lara, Universidad Nacional de Rosario, Argentina

\*Correspondence: Miyako Kusano kusano.miyako.fp@u.tsukuba.ac.jp

#### Specialty section:

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

Received: 31 January 2018 Accepted: 10 September 2018 Published: 09 October 2018

#### Citation:

Fukushima A, Hikosaka S, Kobayashi M, Nishizawa T, Saito K, Goto E and Kusano M (2018) A Systems Analysis With "Simplified Source-Sink Model" Reveals Metabolic Reprogramming in a Pair of Source-to-Sink Organs During Early Fruit Development in Tomato by LED Light Treatments. Front. Plant Sci. 9:1439. doi: 10.3389/fpls.2018.01439 Tomato (Solanum lycopersicum) is a model crop for studying development regulation and ripening in flesh fruits and vegetables. Supplementary light to maintain the optimal light environment can lead to the stable growth of tomatoes in greenhouses and areas without sufficient daily light integral. Technological advances in genome-wide molecular phenotyping have dramatically enhanced our understanding of metabolic shifts in the plant metabolism across tomato fruit development. However, comprehensive metabolic and transcriptional behaviors along the developmental process under supplementary light provided by light-emitting diodes (LEDs) remain to be fully elucidated. We present integrative omic approaches to identify the impact on the metabolism of a single tomato plant leaf exposed to monochromatic red LEDs of different intensities during the fruit development stage. Our special light delivery system, the "simplified sourcesink model," involves the exposure of a single leaf below the second truss to red LED light of different intensities. We evaluated fruit-size- and fruit-shape variations elicited by different light intensities. Our findings suggest that more than high-light treatment (500 µmol m−<sup>2</sup> s −1 ) with the red LED light is required to accelerate fruit growth for 2 weeks after anthesis. To investigate transcriptomic and metabolomic changes in leafand fruit samples we used microarray-, RNA sequencing-, and gas chromatographymass spectrometry techniques. We found that metabolic shifts in the carbohydrate metabolism and in several key pathways contributed to fruit development, including ripening and cell-wall modification. Our findings suggest that the proposed workflow aids in the identification of key metabolites in the central metabolism that respond to monochromatic red-LED treatment and contribute to increase the fruit size of tomato plants. This study expands our understanding of systems-level responses mediated by low-, appropriate-, and high levels of red light irradiation in the fruit growth of tomato plants.

Keywords: Solanum lycopersicum, light emitting diode, metabolite profiling, RNA sequencing, light stress, fruit development

# INTRODUCTION

fpls-09-01439 October 5, 2018 Time: 17:17 # 2

Tomato (Solanum lycopersicum), a member of the Solanaceae family, is the leading vegetable crop. Supplementary lighting [e.g., fluorescent- and high-pressure sodium lamps, and lightemitting diodes (LEDs)] is used for tomato production in Northern Europe and Canada (for example, see Heuvelink et al., 2006). It can compensate for low rates of photosynthesis and increases both the growth and yield of tomato plants when compared to natural light (Gosselin et al., 1996; Gunnlaugsson and Adalsteinsson, 2006). Most greenhouses and areas without sufficient daily light integral (DLI) require such supplementary lights to maintain the optimal light environment for the stable growth of tomato plants. A seasonal effect of supplementary light was observed throughout the year (except from June to August); it resulted in increases in the tomato yield (Heuvelink et al., 2006). Others documented that supplementary lighting had no- or negative effects (Gunnlaugsson and Adalsteinsson, 2006; Trouwborst et al., 2010). These observations suggest that DLI from natural and supplemental lighting per plant, the light source, and/or the cultivar play an important role in determining fruit growth rates and yield. Also, depending on the crop species and several growth factors (e.g., temperature, CO2, and air humidity), the light intensity [photosynthetic photon flux (PPF in µmol m−<sup>2</sup> s −1 )] should be optimized to provide sufficient supplementary lighting without eliciting leaf stress and associated leaf disorders (Moe et al., 2006; Darko et al., 2014).

Several tomato fruit characteristics, mainly the result of dramatic metabolic shifts during development and ripening, result in a complex system (Carrari and Fernie, 2006; Bovy et al., 2007). Increasing the fruit yield per plant is important but challenging as the molecular mechanism of the sourceto-sink balance, a key step toward fruit development, remains largely unclear. The translocation of carbohydrates like sucrose and other nutrients from source to sink is a major determinant of plant growth (Nguyen-Quoc and Foyer, 2001; Paul et al., 2001; Ruan, 2014). Plants strictly regulate the production of photoassimilates and the source-to-sink response to changing environments (Lemoine et al., 2013; Osorio et al., 2014). Of these, sucrose contributes to translocation as a main carbon source in phloem. Tomato plants overexpressing sucrose phosphate synthase (SPS), a key enzyme in the sucrose metabolism, exhibited substantially altered carbon allocation in photosynthetic leaves (Galtier et al., 1993, 1995; Micallef et al., 1995). A reduction in the activity of sucrose synthase (SuSy), which catalyzes the sucrose cleavage in tomato fruit, considerably reduced its sucrose unloading capacity (D'Aoust et al., 1999). A comprehensive and quantitative molecular understanding of the tightly coupled coordination of photosynthesis and sink capacity is important. With respect to the quality of tomato fruit, these systems are closely associated with the phloem loading of sucrose in the source and with unloading in sink tissues via the central carbon metabolism, although generally, photosynthesis in fruit is not essential (Kahlau and Bock, 2008; Lytovchenko et al., 2011).

Transcriptome analysis with microarrays and RNAsequencing (RNA-Seq) revealed important key factors involved in fruit ripening (Lin et al., 2008; Chung et al., 2010; Nakano et al., 2012; Fujisawa et al., 2014; Nguyen et al., 2014). The integration of transcriptomic and metabolomic approaches demonstrated that the detected primary metabolites, cell wallrelated metabolites, and pigments were not strongly correlated with known key genes involved in ripening, but implied a causal relationship between tricarboxylic acid (TCA) cycle intermediates and fruit ripening (Alba et al., 2005; Carrari et al., 2006; Mounet et al., 2009; Osorio et al., 2011). Despite the agricultural importance of the developmental process under supplementary lighting, the comprehensive metabolic and transcriptional behaviors along the developmental period remain to be fully elucidated. Also, their role under artificial supplementary lighting with LEDs (Goto, 2003; Darko et al., 2014) in the regulation of flowering and early fruit development (rather than fruit ripening) remains to be identified quantitatively and systematically.

We present integrative omics approaches to elucidate the metabolomic impact of red LED light of different intensities on single leaves during the early fruit development of tomato plants. We set up a special light-irradiation system, our "simplified source-sink model," which involves a single tomato leaf, a fruit truss, and monochromatic red-LED light delivered during early fruit development; red light is widely used for supplemental lighting. Our findings suggest that the proposed workflow promises to aid in the discovery of key pathways that contribute to increasing the fruit size of tomatoes.

#### MATERIALS AND METHODS

#### Plant Material and Growth Conditions

Seeds from tomato (Solanum lycopersicum 'Reiyo') were sown in 72-cell trays (Takii Seed, Kyoto, Japan) and grown in a soil mix (Napura Soil Mixes, Yanmar, Osaka, Japan) for 2 weeks in a growth chamber (MKV DREAM, Tokyo, Japan) at 25◦C/20◦C (light/dark, Japan) and 900 µL L−<sup>1</sup> CO<sup>2</sup> concentration. The light/dark cycle was 16 h/8 h for 2 weeks. Then the seedlings were transferred to 2.4 l pots and grown in a growth chamber (Asahi Kogyosha, Tokyo, Japan); the PPF level was adjusted to 450–500 µmol m−<sup>2</sup> s <sup>−</sup><sup>1</sup> when measurements were at the meristem of each tomato plant (light source: ceramic metal halide lamps). S. lycopersicum cv. 'Moneymaker' was also used and exposed to the same conditions of cv. 'Reiyo' for fruit measurements. The experiments were performed at Chiba University, Japan.

For LED irradiation, we exposed single leaves for 4 weeks to a red LED panel (23 cm × 12 cm, 18 W, Shibasaki, Saitama, Japan); the peak wavelength was 660 nm (Showa Denko K. K., Tokyo, Japan). To remove the effects of supplemental light from other factors, we removed all leaves and trusses except for the flowers on the second truss, the leaf just below the second truss, and the apical portions of the main shoot at the anthesis stage of the second truss (Hikosaka et al., 2013) (**Supplementary Figure S1**). Each plant was trimmed to bear a single leaf and a truss with three flowers (**Supplementary Figure S1B**). We used four light intensities at PPF 0-, 200-, 500-, and 1,000 µmol m−<sup>2</sup> s −1 (P0, P200, P500, P1000). Different PPFs were applied to post-anthesis tomato plants for 2 weeks after anthesis (WAA), corresponding to 14 days after anthesis (DAA). Leaf and fruit samples were harvested 0-, 1-, and 2 WAA at the same hour 1600 (JST, Japan Standard Time), corresponding to midday in the growth chamber.

#### Fruit Measurements

fpls-09-01439 October 5, 2018 Time: 17:17 # 3

It is known that the fresh weight (FW, in g) of tomato fruit can be estimated from the fruit length, diameter, and height (for example, see Mutschler et al., 1986). Therefore, fruit measurements were taken every week with a digital caliper and recorded as the estimated FW of each fruit. Biological replicates, n = 3.

#### RNA Isolation

Total RNA was isolated using the RNeasy Plant Mini kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions. The concentration, integrity, and extent of contamination by ribosomal RNA were monitored using an ND-1000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, United States) and a Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA, United States).

# cDNA Library Construction and RNA-Sequencing

Beads with oligo(dT) were used to isolate poly(A) mRNA after total RNA was collected from tomato tissues, leaves, and fruits. Fragmentation buffer was added to cut mRNA into short fragments to serve as templates; random hexamer primer was used to synthesize first-strand cDNA. Second-strand cDNA was synthesized using buffer, dNTPs, RNaseH, and DNA polymerase I. Short fragments were purified with the QiaQuick PCR extraction kit and resolved with EB buffer for end repair and for adding poly(A). The short fragments were connected with sequencing adapters. After agarose gel electrophoresis, suitable fragments were selected as templates for polymerase chain reaction (PCR) amplification. Lastly, the library was sequenced using Illumina HiSeqTM 2000. We complementary analyzed same samples used for microarray analysis per condition (n = 1 each).

#### Sequence Processing, Mapping Reads to a Reference, and Differential Expressions

After Illumina reads were quality-checked, demultiplexed and trimmed, they were clustered per library using RobiNA (Lohse et al., 2012). The remaining short reads used for assembly were aligned to the CDS sequences with Bowtie (Langmead et al., 2009) to identify rRNA contamination; two mismatches were allowed. The ribosomal filtered reads were then aligned against tomato genome sequence SL2.40 (ITAG2.3) (Tomato Genome, 2012). Differentially expressed genes (DEGs) were identified using the DESeq (Anders and Huber, 2010) with default parameters. The level of significance was set at a false discovery rate (FDR) < 0.05 (Benjamini and Hochberg, 1995). We used BiNGO (Maere et al., 2005) to analyze significantly over-represented gene ontology (GO) categories in the DEGs (FDR < 0.05).

#### Metabolite Profiling

Metabolite profiling by gas chromatography-time-of-flight mass spectrometry (GC-TOF-MS) was performed essentially as described (Kusano et al., 2007a,b, 2011a) but with tomato-specific modifications [see our meta-data (accession no. MTBLS699) in MetaboLights (Kale et al., 2016)]. Briefly, all raw data in netCDF format were pre-processed by hyphenated data analysis (HDA) (Jonsson et al., 2005, 2006) and the obtained data matrix was normalized and summarized using the cross-contribution compensating multiple standard normalization (CCMN) method (Redestig et al., 2009). For metabolite identification, we crossreferenced the obtained mass spectra with gas chromatography with electron impact mass spectrometry (GC-EI-MS) and retention index libraries (Schauer et al., 2005) in the Golm Metabolome database (Kopka et al., 2005) and our own inhouse libraries. According to the recommendation (Fernie et al., 2011), detailed information on metabolite identification was shown in **Supplementary Tables S2A,B**. The metabolite profile data (processed data) with our experimental design (phenodata) are also included in **Supplementary Tables S2G,H**. We compared the metabolite responses: (1) treatment comparison, i.e., highlight vs. lowlight treatment and (2) developmental comparison, e.g., 2 WAA vs. 1 WAA under LED irradiation at P1000. The control condition of comparison (1) was P200 red light, whereas 1 WAA was used as control condition in the case of (2). Each sample point was analyzed with six biological replicates.

# Statistical Data Analyses for Transcript Profiling by Microarrays and Metabolite Profiling

We used same microarray data that were analyzed in our previous study (Fukushima et al., 2012) [accession#: GSE35020 in NCBI GEO (Barrett et al., 2013)]. We re-analyzed the total of 18 samples (12 leaf- and 6 fruit samples); three biological replicates per sample were used. Data normalization, visualization, and correlation analysis based on Pearson's correlation were performed using R<sup>1</sup> and Bioconductor (Gentleman et al., 2004). DEGs and differentially accumulated metabolites were identified using the LIMMA method, which is based on linear model fitting (Smyth, 2004). The level of significance was set at FDR < 0.05 (Benjamini and Hochberg, 1995). Principal component analysis (PCA) was performed using the pcaMethods package (Stacklies et al., 2007), with log-transformation and unit variance scaling. To visualize the global transcript responses of gene regulatory networks and metabolic pathways of fruit- and leaf samples, we used MapMan software (v3.5.1R2) (Usadel et al., 2005). Genes were classified into different functional categories based on MapMan BIN from the ITAG2.4 annotation. We used BiNGO (Maere et al., 2005) to analyze significantly over-represented GO terms in the DEGs. The level of significance was set at FDR < 0.05 (Benjamini and Hochberg, 1995).

<sup>1</sup>http://www.r-project.org

# RESULTS

# Scope of the Study and Its Systematic Experimental Design

In preliminary experiments, we studied the fruit weight and leaf area of whole Solanum lycopersicum L., 'Reiyo' plants exposed or not exposed to red LED irradiation. We first recorded the fruit weight along the developmental stages 1-, 2-, 3-, and 4 WAA of plants grown without supplemental LED lighting (**Supplementary Figure S2A**). Under normal light (average P500, metal halide lamp without supplemental LED light), we observed a remarkable increase in the fruit weight between 1- and 2 WAA, suggesting that the period was critical for early fruit development and the time of cell expansion. When we recorded the fruit weight and leaf area of whole plants grown under supplemental red LED light (P1000), we detected no effect on the fruit biomass at 2 WAA (**Supplementary Figure S2B**).

Based on these preliminary findings we focused on 2 WAA and developed a custom LED light system to gain insights into molecular regulation governing early tomato fruit development and biologically relevant changes in the storage pattern and translocation under different light conditions. Our "simplified source-sink model" (Hikosaka et al., 2013) is comprised of a single tomato leaf and fruit truss (**Supplementary Figure S1**) and can be used to deliver red or other color light irradiation in greenhouses or under closed growth conditions, e.g., in climate chambers. Although we delivered 100% of red LED light to the plants, leaves exhibited few stress signs under the P200 condition. However, after P500 and P1000 high light (HL) exposure, they manifested stress signs and accompanying disorders, including leaf curling and senescence (**Supplementary Figure S3**), due mainly to high light intensity and seemingly enhancement of translocation.

## Enhanced Light Intensity Strongly Affects Leaf and Fruit Growth in Tomato

Using our simplified source-sink model we next assessed variations in the fruit size and shape due to different light intensities. **Figure 1A** shows a representative fruit shape developed under red LED irradiation (P200, P500, and P1000) in 2 WAA. Although the tomato plants were grown simultaneously in a controlled growth chamber under artificial conditions, there were variations in the fruit morphology due to uncontrollable factors affecting fruit set (**Supplementary Table S1**). For example, the developmental stage at P1000 irradiation in 4 WAA can correspond to breaker (in some case, it corresponds to red ripe). In the case of P500 in 4 WAA, the stage corresponds to mature green. The mean weight, height, and width were statistically greater after HL- than P200 treatment (p < 0.05, Welch's t-test) (**Figure 1B**). We also evaluated variations in the fruit size and shape obtained under the same conditions in a different year (i.e., independent S. lycopersicum L. 'Reiyo' experiments) (**Supplementary Figure S4**). S. lycopersicum cv., 'Moneymaker,' exposed to the same conditions also exhibited this tendency (**Supplementary Figure S5**). Our findings suggest that treatment with red LED light exceeding P500 is sufficient for fruit growth in tomato plants grown under our artificial conditions.

# Overview of Metabolite and Transcript Responses to High Light Irradiation

To study small-molecule metabolites and gene expressions during early fruit development under red LED light with different light intensities, we performed global metabolite and transcript profiling using the experimental designs of an established GC-MS method (Kusano et al., 2007a,b), Illumina-based RNA-Seq, and previously reported microarrays (Fukushima et al., 2012) (**Supplementary Figures S6**–**S8**). To visualize the extent of metabolomic and transcriptomic changes elicited by different light intensities, we performed PCA and applied the data matrices of the metabolite- and transcript profiles separately. The PCA score scatter plot revealed that the strong impact on metabolite levels in accordance with observance of the presence or absence of light along with PC1 (**Figures 2A,B**). RNA-Seq data (**Supplementary Figure S9**) also showed different clustering groups based on tissue-dependent differences in the PC1 axis, while growth stages with PC2 (**Figure 2C**). When we focused on light intensity-dependent metabolomic changes in fruits, we found that samples exposed to HL conditions were clustered, while samples with light treatments were clearly separated from dark samples (P0). These observations suggest that while our HL condition strongly affected the metabolite accumulation in tomato plants, its effect on developing fruit was not as large.

# Comparative Analyses of Metabolite Profiling of Tomato Fruit- and Leaf Samples Under Different Light Intensities

Our broad-range metabolite analysis identified HL-responsive metabolites and revealed changes in metabolite levels throughout the early fruit-development stage (**Figure 3**). We first focused on metabolites that exhibited a statistically significant difference when the plants were grown under HL and under control conditions (**Table 1** and **Supplementary Table S2**). When we compared between P1000 and P200 light intensities, we found that at P1000, sugars including glucose, fructose, and trehalose and cell wall related metabolites like xylose and mannose markedly increased in 1 WAA fruits while aromatic amino acids such as phenylalanine, tryptophan, and tyrosine, were decreased. There were fewer HL-responsive metabolites in 2 WAA than 1 WAA fruits, resulting in 10 significantly changed metabolites. Inverse changes were observed in the metabolite levels, for example, phenylalanine and tyrosine were increased in response to HL.

Except at 2 WAA, the trend observed in P500 and P1000 fruits was similar. Fruits examined in 2 WAA did not show a significant increase in metabolites; mainly polyol and some hydroxyl acids were decreased. Sucrose, glucose, and xylose increased in response to HL. The significantly increased organic acids in 1- and 2 WAA leaves exposed to HL were citrate, malate, succinate, quinate, and glycerate. Methionine levels in young leaves dramatically decreased in response to HL. In 1 WAA leaves, P500 HL largely affected the metabolome; the number of

FIGURE 1 | Fruit size and shape variations elicited by different light intensities. (A) A representative fruit shape developed under red LED panels (P200, P500, and P1000) in 2 WAA. Scale bar: 5 cm. (B) Measured tomato fruit sizes. Statistically significant differences between fruits exposed to light intensities at P200, P500, and P1000. We used a box and whisker plot, a graphical summary of a distribution. This plot can visualize the minimum, lower and upper quartiles (25% and 75%), median, and maximum of data. Regarding extreme values, outliers may be displayed as open circles. Data show the mean of the total weight, height, and width calculated with the Welch t-test. Differences of <sup>∗</sup>p < 0.05 were considered statistically significant. The samples were used for metabolite profiling. The results indicate that treatment with higher than P500 red LEDs is sufficient for fruit growth under our artificial conditions. Biological replicates, n = 3. WAA, week after anthesis.

Illumina-based RNA-Seq (C). We observed changes in the metabolite composition of fruits and leaves in the presence or absence of light (PC1). RNA-Seq data also revealed different cluster groups according to tissue-dependent differences in the PC1 axis. The recorded changes in the growth stages were observed under PC2. Biological replicates, n = 5–6 for metabolite profiling and n = 1 for transcript profiling obtained by RNA-Seq analysis.

significantly accumulated metabolites was higher than at P1000 HL although we observed a similar tendency at P500 and P1000.

Comparison of 1- and 2 WAA fruits showed that sucrose decreased under HL while glucose and fructose increased in an essentially linear manner during fruit development and in response to HL (**Figure 3** and **Supplementary Table S2E**). Cellwall related metabolites like xylose, mannose, and arabinose increased during fruit development. Irrespective of the light intensity, the level of most sugars, sugar phosphates, and some organic acids like citrate and aconitate was higher in fruits than leaves (**Supplementary Table S2F**). Most highly accumulated amino acids in fruits were γ-amino butyrate (GABA), glutamine, asparagine, branched chain amino acids (valine, leucine, and isoleucine), beta-alanine, and methionine.

### Genome-Wide Transcript Profiling Revealed a Wide Range of Variations in Gene Expression and Reflected Changes in Regulatory Networks Under High Light Treatment

We assessed the comprehensive transcript abundance using microarrays and RNA-Seq (**Supplementary Tables S3**, **S4**). To mitigate issues associated with the coverage of gene annotation in microarrays we also performed RNA-Seq with the Illuminabased platform (**Supplementary Figure S9**). Using both datasets we identified DEGs. At P1000 and P200, microarray-based approaches detected 137 up- and 252 down-regulated genes, respectively, in fruits (**Figure 4A** and **Supplementary Table S5**). GO enrichment analysis in "Biological Process" showed that the 137 up-regulated genes were significantly enriched in "ripening (FDR = 3.2E-3)" and "cell wall modification (FDR = 3.2E-3)," whereas the 252 down-regulated genes were related to, for example, "cell division (FDR = 1.8E-2)" and "microtubulebased process (FDR = 1.4E-2)." Ripening-related genes encoded pectin methylesterase PME2.1 (probeset ID = Les.3630.1.S1\_at), expansin 1 (probeset ID = Les.191.1.S1\_at), and PME1.9 (3 probeset IDs: Les.3122.1.S1\_a\_at, Les.3122.2.A1\_at, and Les.3122.2.A1\_a\_at). The RNA-Seq-based approach identified 60 up- and 340 down-regulated genes between P1000 and P200 treatments in tomato fruit (**Figure 4B** and **Supplementary Table S5B**). GO enrichment analysis using DEGs obtained by RNA-Seq indicated that the 60 up-regulated genes were involved in the "cell wall macromolecule catabolic and metabolic process (FDR = 5.1E-4)" and in lipid localization/transport (FDR = 0.047) (**Figure 4B**). The 340 down-regulates genes were related to "proteolysis (FDR = 1.1E-11)" and "negative regulation of molecular function (FDR = 9.3E-7)."

RNA-Seq of 1 WAA leaves also revealed that DEGs that were up-regulated at P1000 and were related to biological processes like "regulation of transcription (FDR = 1.2E-4)" included genes that encode MYB-related transcription factor, WRKY-like MYBrelated transcription factor, and heat-shock factor protein. Our microarray data supported this observation (**Supplementary Table S5**). The number of down-regulated DEGs was larger than of up-regulated DEGs in P1000 2 WAA leaves; we only observed eight up-regulated DEGs. Our analysis for down-regulated genes significantly over-represented "glycerol metabolic process (FDR = 7.2E-4)," "alditol metabolic process (FDR = 7.2E-4)," and "polyol metabolic process (FDR = 1.3E-2)." Together, the results of our global transcript analysis suggest the presence of highly complex transcription dynamics in tomato fruits and leaves exposed to P1000 and P200 and examined at different

upper quartiles (25% and 75%), median, and maximum of each metabolite data. Regarding extreme values, outliers may be displayed as open circles. In each boxplot, the investigated time-points are 1 WAA and 2 WAA (X-axis). The relative abundance on the Y-axis shows the normalized responses of the metabolite peaks obtained by GC-TOF-MS. Typical metabolite classes such as sugars, sugar alcohols, and amino- and organic acids are shown. Magenta- and green boxes represent samples from fruit and leaves, respectively. Biological replicates, n = 5–6. WAA, week after anthesis.

developmental stages. These findings are reflected as the systemslevel response to HL under artificial condition using red LEDs.

## Tomato Leaf and Fruit Samples Exhibited Inverse Changes in the Expression Patterns Involved in Light Reactions, Secondary Metabolism, and Cell-Wall Biosynthesis

For a comprehensive study of transcript level changes in metabolic pathways, we visualized our RNA-Seq data using MAPMAN (Thimm et al., 2004; Usadel et al., 2005). **Figure 5** presents an overview of the transcript profiles of 2 WAA fruitand leaf samples exposed to P1000- or P200 light treatment (**Supplementary Figure S10**). MapMan analysis demonstrated that, as a whole, changes in the expression patterns involved in light reactions, the secondary metabolism, and cell-wall biosynthesis exhibited an inverse tendency in fruit- and leaf samples. The marked up-regulation in the transcript level of P1000 fruits was associated with light reactions; in leaves those genes were down-regulated in response to HL. In leaves, genes involved in cell-wall biosynthesis and secondary metabolism were up-regulated, in fruits they were down-regulated.

To shift our focus onto dissecting the metabolic balance in fruit- and leaf samples as a whole plant system, we used Spearman's correlation (p < 0.05) to identify metabolites that



The listed metabolites exhibited a statistically significant difference under HL and control conditions at 1- and 2 WAA [LIMMA (Smyth, 2004); |log2FC| ≥ 1, false discovery rate (FDR) < 0.05]. Log2FC means logarithmically transformed values of fold-changes (e.g., P1000/P200). See also Supplementary Table S1. HL, high light; FC, fold-change.

exhibited a significant correlation in fruit and leaf samples across time. We found that fumarate showed a negative correlation between fruit and leaf samples, while 2-oxoglutarate exhibited a positive correlation (**Figure 6**).

#### DISCUSSION

To enhance the growth/yield and to improve the fruit quality of tomato plants, a physiological understanding of their metabolic and transcriptional responses during fruit development under artificial supplementary LED light is necessary. In our "simplified source-sink model" (Hikosaka et al., 2013) (**Supplementary Figure S1**), each tomato plant is pruned to have a single leaf and one fruit truss. To gain insights into the storage patterns and translocation in developing tomato fruits in response to environmental perturbation by HL irradiation (**Figure 1**), we performed sophisticated molecular phenotyping of samples exposed to red-LED lighting (660 nm peak). Metabolite and transcript analysis using GC-MS and RNA-Seq/microarrays enhanced our understanding of the cellular response of the fruit storage metabolism associated with primed fruit ripening and cell-wall biosynthesis, the stress response, and photosynthesis in response to the light wavelength and to HL stress imposed by our artificial irradiation system (**Figure 2**). Our findings emphasize the tightly coupled coordination of photosynthesis and sink capacity and provide an important list of the candidate metabolites, transcripts, and key pathways that contribute to the cellular metabolic shift in the course of early fruit development.

Our "simplified source-sink model" is appropriable because it removes as many unwanted variations due to unstable greenhouse conditions as possible. We used this experimental system in our earlier co-expression network analysis to infer candidate functional genes in tomato plants (Fukushima et al., 2012). Applying the experimental system also enabled to compare the extent of changes of light intensity, the photosynthesis rate, and fruit growth in tomato plants grown under two types of supplementary LED lighting methods (Hikosaka et al., 2013). First, they assessed the effects of LED light intensity on the fruit set, dry weight, dry mass ratio of a tomato fruit, and the net photosynthetic rate of a center leaflet, i.e., by applying the same method presented in the study. The second experiments in the study assessed whether number of leaves irradiated by supplemental lighting made effects on the photosynthetic rate of a whole tomato plant. As the light irradiation per leaf could increase photosynthetic rate in both experiments, the present study performed comprehensive molecular phenotyping of samples collected under different light intensities by applying the simplified experimental design.

An artificial environment can cause plant developmental and morphological differences, and their responses can mask essential traits. The plant response to a combination of multiple abiotic stresses in the field condition cannot be directly extrapolated from that to each stress exposed individually (Mittler, 2006; Mishra et al., 2012). In addition, we notice an emerging area, so-called 'Field Omics' (Alexandersson et al., 2014), trying to monitor and analyze different molecular behavior of samples harvested from crop field trials. The current field-omics approaches face a difficult and fundamental problem causing from high variance influenced by temporal and spatial differences in field trials. Such intrinsic heterogeneity in each field overwhelms effect of the experimental perturbations. At least partially, our data can be used as one of references to learn about the differences in molecular mechanisms between field- and laboratory tests, although our system greatly differs from the field trials.

To increase and improve light distribution, supplementary lighting is widely used as it can promote a good photosynthetic response and fruit growth in the lower plant layers. Others (Gosselin et al., 1996; Hovi et al., 2004; Gunnlaugsson and Adalsteinsson, 2006; Hovi-Pekkanen and Tahvonen, 2008; Pettersen et al., 2010) reported the positive effects of supplemental lighting on fruit growth and yield. Based on our evaluation of the tomato fruit size and of shape variations, exposure to around P500 red LED light is sufficient for fruit

FIGURE 4 | Gene ontology term enrichment analysis for HL-responsive genes in tomato fruits (2 WAA, P1000 vs. P200). We used microarray (A) and RNA-Seq data (B). BH FDR < 0.05, |log2FC| ≥ 1. WAA, week after anthesis.

growth not only with respect to S. lycopersicum L. 'Reiyo' but also S. lycopersicum cv. 'Moneymaker' (**Supplementary Figures S3**–**S5**). To confirm whether this observation is cultivardependent or independent, a larger number of samples and cultivars should be needed in a future study. As we used only monochromatic red LED panel in this study, future studies will also focus on other light quality treatments (e.g., blue and mixture of red and blue).

Our study highlighted metabolic shifts in the carbohydrate metabolism and in several key pathways that may contribute to early fruit development under HL condition (**Figures 2–4**). A wide range of plant metabolic responses to various stresses has been studied by metabolomic- and transcriptomic approaches (Shulaev et al., 2008; Urano et al., 2010; Obata and Fernie, 2012; Nakabayashi and Saito, 2015; Noctor et al., 2015). While plants need light for photosynthesis and their healthy growth, it can damage plant cells; strong light irradiance is an abiotic stress factor. Genome-wide analyses can be used to characterize plant stress responses to high (excess) light stress and it can contribute to enhancing our understanding of stress-signaling pathways and the central metabolism including glycolysis, the TCA cycle, and photorespiratory pathways (Obata and Fernie, 2012). Besides identifying HL stress-responsive metabolites like sucrose, inositol, and GABA in tomato leaves, we found that HL stress (both P1000 and P500) led to dramatic decreases in aromatic amino acids in fruits at the early developmental stage (**Table 1** and **Supplementary Table S1**). We also showed that the expression patterns associated with light reactions, the secondary metabolism, and cell-wall biosynthesis exhibited inverse changes when we compared fruit- and leaf samples (**Figure 5**). The coordinated expressions associated with light reactions indicate functional photosynthesis in immature green fruit of tomato plants, which is consistent with early reports (Piechulla et al., 1987; Wanner and Gruissem, 1991; Schaffer and Petreikov, 1997; Alba et al., 2004; Kahlau and Bock, 2008; Lytovchenko et al., 2011). In fruits, down-regulated genes were involved in cell wall degradation. For example, there were down-regulated genes encoding a polygalacturonase and a pectate lyase. Both genes are known to be up-regulated and their activities become dominant during tomato fruit ripening (Cheng and Huber, 1997). In leaves, the coordinated expressions involved in phenylpropanoid- and flavonoid biosynthesis included up-regulated genes encoding chalcone synthase (CHS) and dihydroflavonol 4-reductase (DFR). This implies that the both early flavonoid biosynthetic pathway- and the more anthocyanin-specific genes response to mitigate high light stress (Wulff-Zottele et al., 2010; Caldana et al., 2011; Kusano et al., 2011b).

We observed positive correlation relationships of 2 oxoglutarate between fruit and leaf samples (**Figure 6**). Recent works based on metabolite flux analysis and metabolic network models demonstrated that metabolite provision via TCA cycle has been operated in response to demand of physiological status in the cell (Sweetlove et al., 2010). Among the intermediates in TCA cycle, 2-oxoglutarate is a key compound relating to carbon-nitrogen metabolism (Hodges, 2002; Foyer et al., 2011; Araujo et al., 2014). Antisense of 2-oxoglutarate dehydrogenase complex, involving in enzyme reaction of 2-oxoglutarate as a substrate, exhibited reduction of tomato fruit biomass (Araujo et al., 2012). After inhibiting 2-oxoglutarate dehydrogenase complex in potato tuber, there was the significant decrease in the level of 2-oxoglutarate, while fumarate level was unchanged (Araujo et al., 2008). These observations imply that at least 2-oxoglutalate level and biomass of reproductive/storage organs are likely to be positively coordinated in Solanaceae such as tomato and potato.

profiling performed with microarrays and RNA-Seq showed that the expression patterns involved in light reactions, the secondary metabolism, and in cell-wall biosynthesis exhibited changes that tended to be inverse between fruit- and leaf samples. Metabolite profiling revealed that sugars and some organic acids, citrate, aconitate, and malate increased along the developmental stages in fruits. Amino acids like Gln, Val, Leu, Ile, and beta-alanine were abundant in fruits and tended to increase in the course of fruit development.

Attention has long been focused on striking shifts in cell-wall composition and pigments (Rose et al., 2004), the strict control of climacteric fruit ripening by phytohormones (Alba et al., 2005; Barry et al., 2005), source-sink regulation (Do et al., 2010), and the physiological transition during the parallel differentiation of photosynthetically active chloroplasts into chromoplasts, (for example, see Klee and Giovannoni, 2011; Pesaresi et al., 2014) that occur during development and ripening of tomato plants. Comprehensive molecular phenotyping using transcript and metabolite profiling showed that Aux/IAA and ARF genes play an important role in triggering the fruit set program (Wang et al., 2005; Rohrmann et al., 2011). Critical aspects of metabolic regulatory mechanisms, especially the central metabolism that controls fruit development in tomato plants, have been studied (Carrari et al., 2006; Osorio et al., 2011). Steinhauser et al. (2010) used a near-isogenic line population derived from a cross between S. lycopersicum 'M82' and S. pennellii to compare changes in the enzyme activity levels that can affect the plant metabolism during fruit development. The studies stressed that the plant metabolism and source-sink interaction can be strongly affected by genetic and environmental perturbations and their interactions.

Broad metabolite profiling that combines the use of multiple analytical platforms and our proposed system is required for assessing the plant secondary or specialized metabolism because, in response to artificial LED, changes in the protectants, antioxidants, and other pigments/nutrients like lycopene are largely unclear. A significant difference between our- and earlier studies is the use of LEDs to inspect and capture the precise responses to HL treatment of tomato leaves and

fruits. Specific wavelengths and bandwidths generated by our red LEDs yield the specific red spectrum more efficiently than red filters combined with other light sources and elicit specific plant growth. The light spectrum strongly affects plant growth and development (Goto, 2003; Darko et al., 2014; Kitazaki et al., 2018) and the blue light spectrum near UV may increase the level of polyphenols such as anthocyanin as protectants and/or antioxidants (Seyoum et al., 2006). Massa (2008) suggested that certain light wavelengths may help to protect plants from attacks by insects and pathogens that elicit plant diseases.

Finally, under strictly controlled systems and LEDs, tomato plants exhibited system-level dynamic behaviors in the metabolism (**Figure 7**). This was a precise plant response to the supplemental light source, i.e., red LED light, we used, and yielded new insights that differed from findings made when conventional filters were applied to broad-spectrum light. Our approaches avoid direct, unwanted large-scale effects resulting from unstable greenhouse conditions and different light intensities. Our strategy helps to deepen our understanding of systems-level responses during the growth of tomato fruit and provides fundamental resources for further studies to investigate the molecular basis of the high plasticity of the plant metabolism.

#### DATA AVAILABILITY

Transcriptome datasets generated in this study are downloadable from the NCBI Sequence Read Archive (SRA) with the accession number DRA001843. Microarray GeneChip data are available at the NCBI GEO (GSE35020) as described in our previous study (Fukushima et al., 2012). All metabolite data (<sup>∗</sup> .netCDF format) are also downloadable from MetaboLights (Kale et al., 2016) (accession no. MTBLS699).

# AUTHOR CONTRIBUTIONS

AF, SH, and MiK designed and performed the statistical data analyses and interpreted the data, and wrote the manuscript with contributions from the other co-authors. MiK, MaK, TN, and KS performed the transcript and metabolite profiling. SH and EG conducted and analyzed the measurement of physiological data. All authors read and approved the final manuscript.

#### FUNDING

This work was supported by National Bioscience Database Center (NBDC) , Japan Science and Technology Agency (JST) (Project ID 18062862), Elucidation of biological mechanisms of photoresponse and development of advanced technologies using light sponsored by Ministry of Agriculture, Forestry and Fisheries (MAFF) and by the "Sustainable Food Security Research Project" in the form of an operational grant from the National University Corporation.

# ACKNOWLEDGMENTS

We thank Drs. Thomas Moritz, Pär Jonsson, and Hans Stenlund from the Umeå Plant Science Centre (Umeå, Sweden) and Umeå University (Umeå, Sweden) for their assistance with the HDA and the raw-data analysis (RDA), Mr. Koji Takano and Ms. Miho Misawa (RIKEN Center for Sustainable Resource Science) for preparing the standard compounds, Ms. Ursula Petralia for editorial assistance, Ms. Mariko Hayakumo for experimental assistance, Dr. Tetsuya Sakurai, Mr. Yutaka Yamada, and Ms. Mikiko Takahashi (RIKEN Center for Sustainable Resource Science) for their computational assistance. Dr. Masahito Shikata helped to provide tomato cv. 'MoneyMaker' seeds from the National BioResource Project (NBRP), Ministry of Education, Culture, Sports, Science and Technology (MEXT), Japan.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2018.01439/ full#supplementary-material

FIGURE S1 | Simplified source-sink model and overall experimental design. (A) Schematic representation of red LED lighting in this study, named "simplified source-sink model." We removed all the leaves and trusses except for the flowers of the second truss, the leaf just below the second truss, and the apical portions of the main shoot at the anthesis stage of the second truss. (B) Tomato plant cultivation with our systems under three types of light intensities (P200, P500, and P1000).

FIGURE S2 | Fruit weight and leaf area under LED light irradiation in Solanum lycopersicum L., 'Reiyo' – a whole plant. (A) Fruit weight comparison along developmental stages under normal light condition (average PPF 500, metal halide lamp as 'Control') without supplemental LED lighting. Biological replicates, n = 3. (B) Measurement of fruit weight and leaf area under LED lighting (PPF 1000, red LED) at 14 days. A whole plant and a red LED panel were set in a semi-closed assimilation chamber. The leaf just below the second truss was irradiated by LED. The bar graph indicates mean and standard deviation. Biological replicates, n = 6. WAA, week after anthesis.

FIGURE S3 | Tomato plants and leaf morphology on exposure to different light intensities of red LED at 2 WAA. (A) Leaf and fruit growth of tomato plants. (B) Leaf morphology. Few stress signs in the leaf were visible under P200 condition. The leaves indicated stress signs and accompanying disorders after P500 and P1000 high light treatments. Scale bars represent 10 cm.

FIGURE S4 | Fruit size and shape variation under different light intensities in Solanum lycopersicum L., 'Reiyo' using "simplified source-sink model." Measurement of fruit sizes (A) and leaf area (B). WAA, week after anthesis. The bar graph indicates mean and standard deviation. Biological replicates, n = 1–6 for fruits and n = 3–6 for leaves.

FIGURE S5 | Fruit size under different light intensities in Solanum lycopersicum L., 'Moneymaker' using "simplified source-sink model." Measurement of fruit fresh weight (A) and dry weight (B). Three fruits per plant/treatment. We used a box and whisker plot, a graphical summary of a distribution. This plot can visualize the minimum, lower and upper quartiles (25% and 75%), median, and maximum of data. Regarding extreme values, outliers may be displayed as open circles. Biological replicates, n = 3.

FIGURE S6 | Experimental design for metabolite profiling. WAA, weeks after anthesis.

FIGURE S7 | Experimental design for RNA-Seq analysis. WAA, weeks after anthesis.

FIGURE S8 | Experimental design for microarray analysis. See in details, (Fukushima et al., 2012). WAA, weeks after anthesis.

FIGURE S9 | (A,B) Statistics for sequenced short reads and mapping results of Illumina-based RNA-Seq. WAA, weeks after anthesis.

FIGURE S10 | Overview of transcript profile based on Illumina-based RNA-Seq. The RNA-Seq data from leaf-samples (1 WAA) was described using MAPMAN software (http://mapman.gabipd.org/web/guest/mapman) (Thimm et al., 2004; Usadel et al., 2005). The fold change is visualized by pseudo-color: red, up-regulated by P1000 treatment; blue, down-regulated by P1000 treatment. The results show that expression patterns involved in light reactions, secondary metabolism, and the biosynthesis of cell wall exhibit opposite alteration tendency between fruit- and leaf samples (green rectangles). WAA, weeks after anthesis.

TABLE S1 | Fruit developmental stages used for the study.

TABLE S2 | Reporting metabolite data by GC-TOF-MS in this study. (A) Metabolite reporting checklist. (B) Summary of values for detected metabolites. This table is based on reporting suggestions (Fernie et al., 2011).

#### REFERENCES


Significantly changed metabolites, processed data matrix, and sample information are as follows. (C) Tomato fruits. Metabolite responses during the developmental stages. (D) Leaves. Metabolite responses to high light treatment. (E) Tomato fruits. Metabolite responses during the developmental stages. (F) Leaves. Metabolite responses during the developmental stages. (G) Summary of metabolome data matrix, and statistical data analysis. (H) Sample information.

TABLE S3 | Summarized transcriptome data matrix and significantly changed transcripts obtained by microarrays. FC, fold change; FDR, false discovery rate.

TABLE S4 | Summarized transcriptome data matrix and significantly changed transcripts obtained by RNA-Seq. FC, fold change; FDR, false discovery rate.

TABLE S5 | Results of enrichment analysis. The top five gene ontology (GO) terms (full) that were significantly enriched (hypergeometric test with Benjamini and Hochberg FDR correction, FDR < 0.05) among differentially expressed genes (DEGs) whose expression differed significantly in response to HL (LIMMA, FDR < 0.05 and |log2FC| ≥ 1) based on microarray (A) and RNA-Seq (B) analysis. FDR, false discovery rate; NS, not significant.

highlight regulatory aspects of metabolic network behavior. Plant Physiol. 142, 1380–1396. doi: 10.1104/pp.106.088534




**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 Fukushima, Hikosaka, Kobayashi, Nishizawa, Saito, Goto and Kusano. 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(s) 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 Non-targeted Metabolomics Approach Unravels the VOCs Associated with the Tomato Immune Response against Pseudomonas syringae

María Pilar López-Gresa† , Purificación Lisón\* † , Laura Campos, Ismael Rodrigo, José Luis Rambla, Antonio Granell, Vicente Conejero and José María Bellés

Instituto de Biología Molecular y Celular de Plantas, Universitat Politècnica de València – Consejo Superior de Investigaciones Científicas, Ciudad Politécnica de la Innovación, Valencia, Spain

#### Edited by:

Marta Sousa Silva, Universidade de Lisboa, Portugal

#### Reviewed by:

Xavier Daire, Institut National de la Recherche Agronomique, France Yonathan Asikin, University of Tsukuba, Japan

#### \*Correspondence:

Purificación Lisón plison@ibmcp.upv.es †These authors have contributed equally to this work.

#### Specialty section:

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

Received: 05 May 2017 Accepted: 21 June 2017 Published: 04 July 2017

#### Citation:

López-Gresa MP, Lisón P, Campos L, Rodrigo I, Rambla JL, Granell A, Conejero V and Bellés JM (2017) A Non-targeted Metabolomics Approach Unravels the VOCs Associated with the Tomato Immune Response against Pseudomonas syringae. Front. Plant Sci. 8:1188. doi: 10.3389/fpls.2017.01188 Volatile organic compounds (VOCs) emitted by plants are secondary metabolites that mediate the plant interaction with pathogens and herbivores. These compounds may perform direct defensive functions, i.e., acting as antioxidant, antibacterial, or antifungal agents, or indirectly by signaling the activation of the plant's defensive responses. Using a non-targeted GC-MS metabolomics approach, we identified the profile of the VOCs associated with the differential immune response of the Rio Grande tomato leaves infected with either virulent or avirulent strains of Pseudomonas syringae DC3000 pv. tomato. The VOC profile of the tomato leaves infected with avirulent bacteria is characterized by esters of (Z)-3-hexenol with acetic, propionic, isobutyric or butyric acids, and several hydroxylated monoterpenes, e.g., linalool, α-terpineol, and 4-terpineol, which defines the profile of an immunized plant response. In contrast, the same tomato cultivar infected with the virulent bacteria strain produced a VOC profile characterized by monoterpenes and SA derivatives. Interestingly, the differential VOCs emission correlated statistically with the induction of the genes involved in their biosynthetic pathway. Our results extend plant defense system knowledge and suggest the possibility for generating plants engineered to over-produce these VOCs as a complementary strategy for resistance.

Keywords: metabolomics, tomato, bacteria, VOCs, defense

# INTRODUCTION

Plants have developed multiple defense mechanisms to protect themselves against biotic and abiotic stresses. Accumulation of secondary metabolites, which display numerous biological properties, constitutes a major component of stress responses (Dixon, 2001; Pusztahelyi et al., 2015; Qian et al., 2015; Kalaivani et al., 2016). Among them, volatile organic compounds (VOCs) are a relevant group involved in plant protection against pathogens and herbivores, and also in attracting pollinators and seed dispersers (Dudareva et al., 2013). Plant VOCs include a wide range

of chemical structures, such as terpenoids and phenylpropanoids-benzenoids, as well as fatty and amino acid derivatives (Holopainen and Gershenzon, 2010; Granell and Rambla, 2013). They all have a low molecular weight and polarity, high vapor pressures, and possess the ability to both cross membranes freely and be released into the surrounding atmosphere (Dudareva et al., 2006).

The bacterial speck of tomato, caused by Pseudomonas syringae pv. tomato (Pst), is a major problem in the agricultural industry (Willis and Kinscherf, 2009). Tomato cultivar Rio Grande, which contains the Pto gene (RG-Pto), is resistant to Pst, which expresses effectors or avirulence genes avrPto/avrPtoB. Such "gene-for-gene" recognition (Pto-avrPto/avrPtoB) elicits Effector-Triggered Immunity (ETI) establishment in plants, which allows the control of bacterial spread, and results in an incompatible interaction (Jia et al., 1997; Dangl and Jones, 2001; Lin and Martin, 2005). In contrast, Pst, which bears a deletion of avrPto genes (1avrPto/1avrPtoB), becomes virulent to RG-Pto plants by causing disease in plants and the development of a compatible interaction (Salmeron et al., 1994). Therefore, this tomato pathosystem represents an excellent model to study both kinds of plant–pathogen interactions.

Different signal molecules, such as salicylic acid (SA), gentisic acid (GA), ethylene (ET), or jasmonic acid (JA), have been described to accumulate upon pathogen attacks in tomato plants. SA accumulation has been associated with avirulent infections (Block et al., 2005), while high levels of GA and ET have been found in compatible interactions in tomato (Bellés et al., 1999; Zacarés et al., 2007). JA has been associated with plant responses to herbivores or necrotrophic pathogens (Zhang et al., 2017). However, the role of these defensive molecules has not been characterized in tomato Rio Grande plants infected with virulent or avirulent Pst strains.

More and more studies are being conducted to understand the participation of small metabolites in plant–pathogen interactions (Allwood et al., 2008, 2012). However, the interest in VOCs has focused mainly on the plant response to herbivores and fruit quality, and studies of VOCs in the plant response to pathogens are scarce (Niinemets et al., 2013). Specifically, differential volatile emission has been described for tobacco and pepper leaves in response to both avirulent and virulent strains of Pseudomonas (Huang et al., 2003, 2005) and Xanthomonas (Cardoza and Tumlinson, 2006), respectively, but the biological meaning of this phenomenon is not well understood.

In this paper, we applied an untargeted GC-MS metabolomics approach to analyze the volatiles differentially emitted in RG-Pto tomato plants infected with either avirulent strain Pst DC3000 or virulent strain Pst DC3000 1avrPto/1avrPtoB. Besides, levels of classical defense hormones, such as SA, GA, ET, and JA, were characterized in both tomato interactions. Finally, the activation of a set of genes involved in the corresponding VOC biosynthesis pathways was also studied. Our results will unravel the VOC network that underlies the tomato immune response against Pseudomonas syringae.

# MATERIALS AND METHODS

# Bacterial Strains, Growth Conditions, and Inoculum Preparation

The bacterial strains used in this study were Pseudomonas syringae pv. tomato DC3000 (Pst DC3000), and Pst DC3000 that contains deletions in genes avrPto and avrPtoB (Pst DC3000 1avrPto/1avrPtoB) (Ntoukakis et al., 2009). Bacteria were grown overnight at 28◦C in 20 mL Petri dishes with King's B agar medium supplemented with different antibiotic doses: rifampicin (10 mg/mL) and kanamicin (0.5 mg/mL) for Pst DC3000, and rifampicin (10 mg/mL), kanamycin (0.25 mg/mL) and spectinomycin (2.5 mg/mL) for Pst DC3000 1avrPto/1avrPtoB. Then bacterial colonies were transferred to 15 mL of King's B medium supplemented with rifampicin (10 mg/mL), and were grown overnight at 28◦C with stirring. Bacteria were then pelleted by centrifugation and resuspended in 10 mM of MgCl2, which contained 0.05% (v/v) Silwet L-77, to an optical density of 0.1 at 600 nm. Dilution plating was used to determine the final inoculum concentration, which averaged at 1 × 10<sup>7</sup> CFU/mL.

#### Plant Material and Bacterial Inoculation

Tomato seeds from the cultivar Rio Grande that contained resistance (R) gene Pto (RG-Pto) were grown under greenhouse conditions with a 16/8-h (26/30◦C) light/dark photoperiod (300 µmol/m<sup>2</sup> /s) and 65% relative humidity in 12 cm-diameter pots that contained a 1:1 mixture of peat and vermiculite.

Inoculations with compatible and incompatible bacteria were produced by immersing 28-day-old RG-Pto plants in Pst DC3000 or Pst DC3000 1avrPto/1avrPtoB suspension, respectively, as previously described (Martin et al., 1993). For mock inoculations, plants were dipped in 10 mM of MgCl<sup>2</sup> solution that contained Silwet L-77 (0.05%) without the bacterial inoculum. The third and fourth leaves, from bottom to top, were harvested and frozen in liquid nitrogen at the indicated times. The fifth leaf was placed freshly in a 10-mL screw cap vial and was kept for 5 h to take ET measurements. Six biological replicates were analyzed for each time and tomato–bacteria interaction.

## Extraction and the HPLC Analysis of Salicylic and Gentisic Acids

Extraction of free and total SA and GA from tomato leaflets was performed according to our previously published protocol (Bellés et al., 2008). Aliquots of 30 µL were injected through a Waters 717 autosampler into a reverse-phase Sun Fire 5-µm C18 column (4.6 mm ×150 mm) equilibrated in 1% (v/v) acetic acid at room temperature. A 20-min linear gradient of 1% acetic acid to 100% methanol was applied using a 1525 Waters Binary HPLC pump at a flow rate of 1 mL/min. SA and GA were detected with a 2475 Waters Multi-λ Fluorescence detector (λ excitation 313 nm; λ emission 405 nm), and were quantified with the Waters Empower Pro software using authentic standard compounds (SA sodium salt and GA, Sigma–Aldrich, Madrid, Spain). Standard curves were performed for each compound using similar concentration ranges to those detected in the samples. Data were corrected for losses in the extraction

procedure, and recovery of metabolites ranged between 50 and 80%.

#### Jasmonic Acid and Ethylene Measurements

Ethylene production was measured as described by Lieberherr et al. (2003) with some modifications. Approximately 0.5 g of fresh tissue from the fifth tomato leaf, harvested at the indicated time points after bacterial inoculation, was quickly enclosed in gas-tight 10-mL glass vials fitted with a septum. After 5 h, 400 µL of the gas phase from the vial were analyzed by gas chromatography in a 4890A Hewlett Packard gas chromatograph fitted with a flame ionization detector (FID) using a Teknokroma capillary column (2 m × 1/6<sup>00</sup> OD × 1 mm ID, Alumina F1 80/100). The carrier gas was helium, used at a pre-column pressure of 140 kPa. The injector and detector temperatures were set at 200◦C, while the oven temperature was 80◦C. The retention time of the ET peak under these conditions was 2.5 min. For each time point, six replicates were analyzed and the amount of ET was calculated from the data recorded and analyzed with the MassLynx Waters software by constructing a standard ET curve.

For JA quantification, 250 mg of frozen tissue from the third and fourth tomato leaves were added to 80% methanol–1% acetic acid that contained the internal standard dihydrojasmonate (OlChemIm, Czechia), and were mixed by shaking for 1 h at 4 ◦C. The extract was kept at −20◦C overnight and was then centrifuged. The supernatant was dried in a vacuum evaporator. The dry residue was dissolved in 1% acetic acid and passed through a reverse phase Oasis HLB column (Seo et al., 2011). The dried eluate was dissolved in 5% acetonitrile–1% acetic acid, and the hormone was identified using a reverse phase Ultra Performance Liquid Chromatography (UPLC) system coupled to a Q-Exactive mass spectrometer (Orbitrap detector; Thermo Fisher Scientific) by targeted Selected Ion Monitoring (SIM). Separation was performed in a 2.6 µm Accucore RP-MS column, 50 mm long × 2.1 mm i.d.; (Thermo Fisher Scientific) using a 5–50% acetonitrile gradient that contained 0.05% acetic acid as the solvent system at 400 µL/min for 14 min. The JA concentration in the extracts was determined using embedded calibration curves with the authentic standard (OlChemIm, Czechia) and the Xcalibur 2.2 SP1 build 48 and TraceFinder software.

#### HS-SPME Extraction and the GC-MS Analysis of Volatile Compounds

For the volatile analysis, 100 mg of frozen tomato leaf powder were weighed in a 10-mL headspace screw-cap vial. One milliliter of a saturated CaCl<sup>2</sup> solution and 100 µL of 750 mM EDTA (adjusted to 7.5 pH with NaOH) were added, mixed gently and sonicated for 5 min. Extraction of volatile compounds was performed by headspace solidphase micro-extraction (HS-SPME) (Rambla et al., 2015). The pre-incubation and extraction periods, both at 50◦C, were 10 and 20 min, respectively. Adsorption was performed by means of a 65 µm DVB/PDMS fiber (Supelco, Bellefonte, PA, United States). Desorption was done in the injection port of the gas chromatograph for 1 min at 250◦C in the splitless mode. Volatile extraction and injection were performed automatically with a CombiPAL autosampler (CTC Analytics, Zwingen, Switzerland).

The chromatographic separation of compounds was performed in an Agilent 6890N gas chromatograph (Santa Clara, CA, United States) equipped with a DB-5 ms fused silica capillary column (60 m long, 0.25 mm i.d., 1 µm film thickness). The oven temperature conditions were 40◦C for 2 min, 5◦C/min ramp until 250◦C and then held isothermally at 250◦C for 5 min. Helium was used as the carrier gas at 1.2 mL/min at a constant flow. Detection was performed in an Agilent 5975B mass spectrometer (Santa Clara, CA, United States), which operated in the EI mode (ionization energy, 70 eV; source temperature 230◦C). Data acquisition was performed in the scan mode (m/z range 35–250; six scans per second). Chromatograms and mass spectra were recorded and processed by the Enhanced ChemStation software (Agilent).

The unequivocal compound identification of the 70 volatile compounds was carried out by comparing both mass spectra and retention times with those of pure standards. All the commercial standards were purchased from Sigma–Aldrich (Madrid, Spain). Three other compounds were tentatively identified by comparing their mass spectra with those in the NIST 05 Mass Spectral library. Such tentatively identified compounds are marked with an asterisk.

## RNA Extraction and the Quantitative RT-PCR Analysis

The total RNA of tomato leaves was extracted using TRIzol reagent (Invitrogen, Carlsbad, CA, United States), following the manufacturer's protocol. RNA was then precipitated by adding one volume of 6 M LiCl and keeping it on ice for 4 h. Afterward the pellet was washed using 3 M LiCl and was dissolved in RNase-free water. Finally, in order to remove any contaminating genomic DNA, 2 U of TURBO DNase (Ambion, Austin, TX, United States) were added per microliter of RNA.

For the quantitative RT-PCR (qRT-PCR) analysis, one microgram of total RNA was employed to obtain the corresponding cDNA target sequences using an oligo(dT)<sup>18</sup> primer and the PrimeScript RT reagent kit (Perfect Real Time, Takara Bio Inc., Otsu, Shiga, Japan), following the manufacturer's directions. Quantitative PCR was carried out as previously described (Campos et al., 2014). A housekeeping gene transcript, Elongation Factor 1 alpha (eEF1α), was used as the endogenous reference. The PCR primers were designed using the pcr Efficiency software (Mallona et al., 2011) and are listed in Supplementary Table S1. The primers used to amplify TomLOXF have been previously described (Shen et al., 2014).

#### Statistical Analysis

The statistical analyses of the signal compounds levels (GA, SA, ET, and JA) and the qRT-PCR of the selected genes were done by an analysis of variance (multifactor ANOVA) using Statgraphics Centurion XVI.

For the untargeted analysis of the volatile profile, the GC-MS data were processed with the MetAlign software (Wageningen,

Netherlands) for the alignment of the chromatograms and the quantitation of each MS feature. The resulting dataset was submitted to a Partial Least Square (PLS) study by the SIMCA-P software (v. 11.0, Umetrics, Umeå, Sweden) using unit variance (UV) scaling.

For the Hierarchical Cluster Analysis (HCA), the ratios for each VOC were calculated and log2-transformed for normalization. HCA was performed with the Acuity 4.0 software (Axon Instruments) with the distance metrics based on the Pearson correlation. The normalized data were represented as a heat map using the same software. The Pearson correlations between gene expression and the VOCs concentrations were performed with the SPSS 16.0 software by considering data at 10 and 18 hpi (hour post-inoculation).

#### RESULTS

### RG-Pto Tomato Plants Infected Either with the Avirulent or Virulent Bacterial Strain Displayed Noticeable Symptom Differences

Symptoms development of the tomato plants infected with either avirulent strain Pst DC3000 or virulent Pst DC3000 1avrPto/1avrPtoB is shown in **Figure 1**. These symptoms ranged on a 0–4 scale as follows: symptomless (0), weak (1), moderate (2), severe (3), and very severe (4). Inoculation of the RG-Pto tomato plants with the virulent strain resulted in chlorotic lesions appearing by 18 hpi, which displayed a symptom degree from (0) to (1). These initial lesions increased in intensity and size, and caused significant leaf damage that ranged from (2) to (3) at 24 and 36 hpi, respectively. By 48 hpi, necrotic lesions extended to the total area, and leaves lost their firmness and completely collapsed, which was the maximum level of symptomatology (4). Strong epinasty was observed in the leaves of these symptomatic tomato plants. In contrast, no symptoms (0) were observed at any time on the RG-Pto tomato plants inoculated with avirulent strain Pst DC3000 due to Pto-avrPto/avrPtoB recognition and ETI establishment. Therefore, these immunized tomato leaves were similar to the mock-inoculated plants.

#### Levels of Salicylic Acid, Gentisic Acid, and Ethylene Were Enhanced in the RG-Pto Tomato Plants Infected with the Virulent Bacterial Strain, While Jasmonic Acid Drastically Lowered

The levels of the signaling defense molecules, SA, GA, ET, and JA were analyzed in the plants infected with both virulent and avirulent strains in a time-course study. **Figure 2** shows the significant GA induction that occurred in all the inoculated plants that bore both interactions. The increased amount of GA was already evident in tomato plants at 10 hpi. Leaf GA accumulation increased as bacterial infection progressed, and the highest levels peaked at 48 hpi (45 nmol/g FW) in the symptomatic tomato plant leaves. By 24 hpi, a significantly higher GA value was obtained in the compatible interaction compared to the incompatible infection. No remarkable increments in SA were observed in the Pseudomonas-infected tomato plants at any time point, although differences in the SA levels started to become significant in the compatible interaction by 24 hpi.

The evolution of ET from the leaves of both the compatible and incompatible tomato interactions was also measured. A dramatic increase in the production of this stress hormone was detected in the compatible interaction, which paralleled GA accumulation and appearance of symptoms (**Figure 3A**). Maximum ET production (188 nL/gFW/h) was reached at 18 hpi, which coincided with the onset of symptom development. These high levels of ET (up to 10-fold) could explain the strong epinasty observed in the tomato plants infected with the virulent bacteria. Remarkably during this compatible infection, ET biosynthesis was higher than that observed in the incompatible interaction, and was significantly elevated at any time point. Regarding infection with the avirulent bacteria, the differences in ET emission between the infected and mock-inoculated plants was only significant at 24, 36, and 48 hpi.

In contrast, the JA levels drastically lowered during virulent infection, and almost non-detectable values were displayed at 10 hpi. Regarding avirulent infection, JA production remained unaltered compared to the mock-inoculated tomato plants. A similar tendency was observed at 24 hpi (**Figure 3B**).

#### The Volatile Profile of Tomato Leaves Altered Differentially upon Infection with Both Pseudomonas syringae Strains

In order to examine the VOCs involved in the tomato–pathogen interactions, changes in the levels of these metabolites in the RG-Pto tomato plants, which were either mock-inoculated or infected with avirulent strain Pst DC3000 or virulent Pst DC3000 1avrPto/1avrPtoB, were analyzed by GC-MS in the leaves infected between 10 and 48 hpi (see Materials and Methods). **Table 1** lists the VOCs detected on our HS-SPME/GC-MS platform for the mock and infected RG-Pto tomato plants. In all, 73 compounds were identified: 11 esters (9 aliphatic and 2 aromatic), 20 aldehydes (16 aliphatic, 3 aromatic and 1 norcarotenoid), 13 alcohols (6 aliphatic, 1 norcarotenoid, 5 monoterpenic, and 1 sesquiterpenic), 9 monoterpene hydrocarbons, 8 ketones (5 aliphatic, 1 aromatic, and 2 norcarotenoid), 3 sesquiterpene hydrocarbons, 2 furans, 2 nitriles, 4 aliphatic acids, and 1 aromatic hydrocarbon. They were all unequivocally confirmed by using pure standards, except for three of them, which were tentatively identified based on their mass spectra similarity (match > 900).

To manage the large amount of mass data, a multivariate data analysis was performed that consisted in a PLS analysis, where compound abundance was assigned to the X variable, and harvesting time (10, 18, 24, 36, and 48 hpi) and type of infection (mock, compatible, and incompatible interaction) were defined as stepwise Y variables. The PLS analysis (**Figure 4**) showed that the first component (PC1) explained changes in the chemical composition during the experiment (harvesting time),

1avrPto/1avrPtoB at 10, 18, 24, 36, and 48 hpi for ethylene production (A), and at 10 and 24 hpi for jasmonic acid (JA) accumulation (B). Different letters indicate the statistical significances with a p-value < 0.05 compared with the mock-inoculated plants.

while the metabolic alteration due to bacterial infection was clearly characterized by the second component (PC2). **Figure 4** also displays the over-emitted metabolites in both infected plants, which were identified by the loading plot analysis. As expected, the VOCs from the non-infected plants were chemically similar within 48 h of the experiment. However, the plants infected with both bacterial strains showed an evident variation in their metabolic profile compared to the mock plants. This indicated that VOCs emission was independent of symptomatology. However, no clear separation between the virulent and avirulent infections was observed in this PLS score plot.

In order to distinguish the defense metabolites associated with each plant–pathogen interaction, the two infections were independently analyzed. The volatile content of the mock-inoculated RG-Pto tomato plants was compared with that of the RG-Pto-infected one with either avirulent strain Pst DC3000 (Supplementary Figure S1) or virulent strain Pst DC3000 1avrPto/1avrPtoB (Supplementary Figure S2). In these PLS analyses, a marked separation between the infected and mock plants was clearly observed by PC2. In both cases, the loading plot revealed a specific set of VOCs that strongly contributed to the separation of samples according to the specific plant–bacterial infection. **Figure 5** shows a hierarchically clustered heat map, including the most discriminative compounds by comparing the volatile profile of the tomato plants infected with the avirulent or virulent Pst strains with the control plants (column 1 and column 2, respectively).

These defense compounds, induced by both infection types, derive from three important plant metabolic pathways: fatty acids, terpenoids and benzenoids. Among them, the most prominent VOCs produced upon both bacterial infections (red in columns 1 and 2) were some esters of (Z)-3-hexenol, such as (Z)-3-hexenyl acetate, (Z)-3-hexenyl propionate, (Z)-3-hexenyl isobutyrate, (Z)-3-hexenyl butyrate, some hydroxylated monoterpenes (HMT), such as linalool, α-terpineol, both (Z)- and (E)- isomers of linalool oxide, and an unidentified sesquiterpene. The statistical analyses showed that their differential induction was significant (Supplementary Table S2). Regarding benzenoids emission, an increase in the production of methyl salicylate (MeSA), salicylaldehyde, and ethyl salicylate was also observed in both infections. The accumulation of the VOCs that derived from salicylate was consistent with the SA accumulation detected in the compatible interaction since levels of these phenolic derivatives were also higher in this virulent infection (**Figure 2**).

#### Comparison of the VOCs Profiles of Tomato Leaves upon Infection with Virulent and Avirulent Pseudomonas syringae Strains Unraveled a Specific Volatile Response for ETI

In order to identify whether a specific set of volatile metabolites was involved in the establishment of effective defense such as ETI, another PLS analysis was performed by comparing the VOCs emitted from the tomato plants infected with avirulent strain Pst DC3000 and with virulent strain Pst DC3000 1avrPto/1avrPtoB (**Figure 6**). Once again, the second component clearly showed the different set of volatile compounds emitted by the plant that underwent either a compatible or an incompatible interaction during the 48 hpi period. This metabolomic approach allowed us to identify the differentially induced VOCs in each infection type, by using the loading plot analysis. The hierarchically clustered heat map shows the VOCs that were differentially emitted by the tomato plants infected with virulent Pst strains compared with the avirulent infection (**Figure 5**, column 3). Most VOCs, which were over-emitted by tomato plants during the establishment of the ETI triggered by the avirulent strain (green-colored), showed significant differences compared with the symptomatic infection (Supplementary Table S2).

#### TABLE 1 | List of the VOCs identified in tomato leaves by GC-MS.


(Continued)



Family Code: Alc, alcohol; Ald, aldehyde; Ar, aromatic hydrocarbon; Est, ester; Fur, furane; Ket, ketone; Mt hd, monoterpene hydrocarbon; Nit, nitrile; Sqt, sesquiterpene. <sup>∗</sup>Tentative identification based on mass spectrum. <sup>a</sup>Monoterpenederived compound. <sup>b</sup> In addition to the alcohol group, it has a tetrahydrofuran group. <sup>c</sup>Norcarotenoid compound. <sup>d</sup>Sesquiterpene-derived compound.

The specific VOCs differentially released from the symptomatic tomato plants (virulent/avirulent ratio > 1; Supplementary Table S2) were an unidentified sesquiterpene, SA derivatives methyl salicylate and salicylaldehyde, monoterpenes α-pinene, α-phellandrene, β-phellandrene and limonene, as well as two isoprenoid chlorides. The strong induction of the SA derivatives was previously observed when comparing the volatile profiles of these susceptible plants to their corresponding mock-inoculated plants (Supplementary Figure S2). This study also revealed enhanced emission of monoterpenes after infection with virulent strain Pst DC3000 1avrPto/1avrPtoB.

Interestingly, we identified a set of volatiles that were significantly over-emitted by tomato plants when effectively resisting disease (green-colored in column 3 of **Figure 5**). Among them, several HMT, such as linalool, α-terpineol, 4-terpineol, (Z) and (E)-linalool oxides, HMT-1, HMT-2, HMT-3, and HMT-4, as well as the esters (Z)-3-hexenyl propionate, (Z)-3-hexenyl butyrate and (Z)-3-hexenyl isobutyrate, were found. The induction of these compounds is mentioned above (when comparing the volatile profiles of the resistant plants with their corresponding mock-inoculated plants; Supplementary Figure S1). The specific over-emission of these green leaf volatiles (GLVs) esters and HMTs during ETI suggests that these VOCs could participate in the defense response.

As a result of these untargeted metabolomic analyses, we conclude that the infected tomato plants emitted quantitatively

different volatiles depending on each type of bacterial strain used in this experiment. Monoterpenes and SA derivatives were released at higher rates by the symptomatic plants upon successful bacterial infection, while HMT and hexenyl esters were differentially over-emitted during ETI establishment, which led to resistance.

### Bacterial Infection Induces the Specific Expression of the Genes Involved in VOCs Biosynthesis

To study whether differential volatile production was due to transcriptional activation, we analyzed the expression levels of several key genes involved in the VOC biosynthesis by qRT-PCR. The results of the mock-inoculated RG-Pto tomato plants and the plants infected with Pst DC3000 or Pst DC3000 1avrPto/1avrPtoB at 10, 18, 24, 36, and 48 hpi are shown in **Figure 7**. We used the induction of the tomato Pathogenesis-Related PR1 gene as a positive control of bacterial infection, and observed a correlation of the expression of this gene and symptom development (**Figure 7A**). GLVs esters are known to be synthesized by 13-lipoxygenases (13-LOX) via 13-hydroperoxides, which are later cleaved by 13-hydroperoxide lyases (13-HPL) into (Z)-3-hexenal. This last compound is reduced by alcohol dehydrogenase (ADH). Finally, ester formation is catalyzed by alcohol acyl transferases (AAT) (Scala et al., 2013a).

In tomato, six genes that encode various types of lipoxygenases (TomloxA-F) have been described (Mariutto et al., 2011). TomloxC, TomloxD, and TomloxF encode 13-LOX lipoxygenases, and are involved in the synthesis of oxylipins, which play an important role in the response to biotic stress. Tomlox D lipoxygenase participates in the synthesis of JA, while Tomlox C and Tomlox F are involved in the biosynthesis of GLVs. As **Figure 7B** shows, a significant induction of TomloxF was detected upon bacterial infection with both strains at all the studied time points, and this induction was greater when ETI had been established. Regarding alcohol acyltransferases, five AAT genes (SlAAT1 - 5) have been identified in tomato (Goulet et al., 2015). We observed that the induction of AAT1 followed a similar pattern to that of TomloxF (**Figure 7C**). These induction patterns statistically correlated with the emission of the GLV esters (Supplementary Table S3A) in both infections, and became higher during ETI establishment. Therefore, these data are consistent with the VOCs metabolomic analysis, and suggest a possible role of the biosynthesis of GLV esters in plant defense against bacteria.

Terpene synthases (TPS) catalyze the synthesis of mono-, sesqui- and diterpenes, and are responsible for the diversity of the isoprene compounds found in nature (Degenhardt et al., 2009). There are 44 TPS genes in Solanum lycopersicum, 29 of which are potentially functional (Falara et al., 2011). By qRT-PCR, we observed a significant induction of TPS5, also known as MTS1

(**Figure 7D**), in the immunized tomato plants, which peaked at 18 hpi. This gene induction correlated with the production of several HMTs, such as linalool or α-terpineol (Supplementary Table S3B).

#### DISCUSSION

Tomato VOCs have been associated mainly with either improved fruit quality (Rambla et al., 2014) or the response against herbivores (Wei et al., 2013). Here, we extend this knowledge to the volatile differential emission of tomato leaves infected with virulent or avirulent bacteria. We were particularly interested in studying the contribution of VOCs to the resistance mechanisms presented by tomato leaves, which display no symptoms upon bacterial infection (**Figure 1**).

To better characterize both types of compatible and incompatible interactions, the levels of the different signal molecules, e.g., SA, GA, ET, or JA, were measured at several time points (**Figures 2**, **3**). Simple natural phenolics SA and GA are fundamental components of the signal transduction pathway, which triggers defense responses against different invading pathogens in many plant species. However, the biological role of these signals depends on the plant–pathogen system (Métraux and Raskin, 1993; Bellés et al., 1999; Dempsey et al., 1999). SA accumulation is associated with incompatible interactions (Métraux et al., 1990; Rasmussen et al., 1991; Malamy et al., 1992; Silverman et al., 1993; Uknes et al., 1993; Shirasu et al., 1997), while GA is associated mainly with compatible plant–pathogen systems (Bellés et al., 2006; López-Gresa et al., 2010). GA could also constitute a signal molecule that is complementary to SA since exogenous treatments with this compound are able to induce different defense responses to those triggered by SA (Bellés et al., 1999; Campos et al., 2014). Plants also produce ET in response to most biotic and abiotic stresses (Abeles et al., 1992; Merchante et al., 2013). ET has been particularly involved

in the response of Rutgers tomato plants to bacterial pathogen Pseudomonas syringae when infiltrated into leaves (Bellés et al., 1989; Zacarés et al., 2007). Finally, lipid-derived hormone JA has been described as being associated with wounding response and necrotrophic infections, and produces an antagonistic effect on SA-mediated signaling (Zhang et al., 2017). Indeed JA-insensitive tomato jai1 mutants are more resistant to virulent Pseudomonas syringae DC3000 (Zhao et al., 2003). However, jasmonate-deficient def1 tomato mutants have been described as being more susceptible to Pseudomonas syringae and Xanthomonas campestris (Thaler et al., 2004), which thus indicates that the role of JA in tomato bacterial infections is still unclear.

We generally observed that SA, GA, and ET accumulated at higher levels in the tomato plants infected with the virulent bacterial strain compared to those infected with avirulent bacteria. These higher levels correlated with symptom severity. SA accumulation has been described to occur at 4 or 10 dpi in tomato plants inoculated with either avirulent or virulent Xanthomonas, respectively (O'Donnell et al., 2001; Block et al., 2005). Besides in these tomato interactions, an earlier increase in ET has been described in avirulent infection, while delayed ET synthesis happens in the tomato plants infected with virulent Xanthomonas (Ciardi et al., 2000; Block et al., 2005). These results contrast with those observed herein, where the highest ET and SA levels were detected in symptomatic bacterial infection. Several factors could be behind the different SA and ET levels described in each pathosystem, such as timing during infection progress, pathogen dose, greenhouse conditions, or the plant growth stage. Yet in all the cases, the accumulation of either SA or ET correlated with PR1 induction (**Figure 7A**), a classical marker gene that is useful for assessing disease development. The higher level of both ET and GA, detected in the RG tomato plants infected with the virulent Pst strain, was associated mainly with symptom development, and agrees with those previously described (Lund et al., 1998; Bellés et al., 2006). Regarding ET, the mutants and tomato genotypes impaired in ET perception or ET synthesis exhibit a significant reduction in disease symptoms versus the wild type upon infection with different pathogens (Lund et al., 1998). These previous results present compelling evidence that both biosynthesis and ET perception are critical for symptoms development in tomato leaves.

Unlike the accumulations of SA, GA, and ET, we observed that JA levels drastically lowered in the virulent infection in accordance with the well-known SA–JA antagonism (Zhang et al., 2017). However, in the plants infected with avirulent bacteria, JA accumulation remained comparable to the mockinoculated plants, similarly to the detected SA levels. Very few reported studies have monitored JA levels in bacteria-infected

plants. In Arabidopsis plants, no differences in JA levels between mock-inoculated plants and those infected with a virulent bacteria have been described at 2 and 24 hpi (Scala et al., 2013b). Unlike our dipping infection method, these authors performed inoculation with a syringe, which thus caused mechanical damage to both mock and infected leaves. This difference in the inoculation procedure could explain the divergence of our results with those previously published.

All these data indicate that the different accumulation patterns of these four signal molecules depend on the diversity of pathogens with a range of lifestyles. To our knowledge, this is the first study in which the levels of all these signal molecules have been measured in RG tomato plants infected with both Pst strains.

Interestingly, we observed that changes in VOCs emission were mostly independent of macroscopic symptoms since the tomato leaves infected with the avirulent strain overproduced some specific VOCs. We identified 73 emitted VOCs by the GC-MS technology. Although more than 300 VOCs have been reported in tomato fruit (Tikunov et al., 2005), very little information is available on the detailed volatile profile in tomato leaves (Buttery et al., 1987; Wang et al., 2001; Thelen et al., 2005; Zhang et al., 2008; Zhang and Chen, 2009; Proffit et al., 2011), thus our data contribute to this knowledge. The comparison of the VOCs profiles of tomato leaves upon infection with virulent and avirulent Pseudomonas syringae strains allowed us to identify the differentially emitted compounds associated with each interaction. The VOCs emission of a diseased leaf is enriched in monoterpenes and SA derivatives, while that of a resisting leaf is characterized mainly by esters of hexenyl GLVs and HMTs.

For the VOCs emitted by symptomatic tomato leaves, the induction of monoterpenes in tomato plants upon Botrytis infection has been described, where α-phellandrene and β-phellandrene, 2-carene, limonene, and α-pinene contributed to more than 95% of volatile emissions (Jansen et al., 2009). Nevertheless, the specific release of monoterpenes has been described in pepper plants exposed to the incompatible Xanthomonas pathogen, but not in compatible interaction (Cardoza and Tumlinson, 2006). MeSA is generally induced upon pathogen infection (Loake and Grant, 2007). In pepper plants infected with avirulent and virulent Xanthomonas strains, MeSA is also emitted at higher levels in the compatible interaction (Cardoza and Tumlinson, 2006). However, in tobacco plants infected with several P. syringae strains, MeSA levels become higher upon avirulent inoculation (Huang et al., 2003).

Biotic stresses have been described to trigger emissions of volatile molecules, which are products of the lipoxygenase (LOX) pathway, such as C<sup>6</sup> aldehydes, alcohols and derivatives, generally referred to as GLVs (Niinemets et al., 2013). For example, Pst infection provokes the emission of (Z)-3-hexenol and (E)-2-hexenal in bean and tobacco leaves, respectively (Croft et al., 1993; Heiden et al., 2003). GLVs are also emitted after fungal and virus infections (Scala et al., 2013a). In tomato plants, (Z)-3-hexenol, (Z)-3-hexenal, and (Z)-3-hexenyl acetate are the dominant LOX products in the volatile emission after Botrytis cinerea inoculation (Jansen et al., 2009). Volatile esters not only contribute to the aroma of many fruits and flowers, but are also related to plant defense and plant-to-plant signaling (Goulet et al., 2015). The ester (Z)-3-hexenyl acetate is one of the most abundant volatiles to be emitted from mechanically or herbivore-damaged Arabidopsis thaliana plants (D'Auria et al., 2007), and can prime a defense response in nearby plants (Engelberth et al., 2004; Frost et al., 2008). The emission of other (Z)-3-hexenyl esters has also been described in pepper plants upon Xanthomonas infection (Cardoza and Tumlinson, 2006). The induction of (Z)-3-hexenol and some of its derived esters upon both bacterial infection types in the tomato plants reported herein extend GLVs emission to other plant–pathogen interactions. Our results, together with those in which (Z)-3-hexenol induces defense genes in Arabidopsis and maize plants (Bate and Rothstein, 1998; Farag et al., 2005), suggest that this alcohol can act as a signaling molecule involved in plant response. Another short-chain alcohol, 3-pentanol, has been found to trigger induced resistance in Arabidopsis against Pseudomonas syringae (Song et al., 2015), and also in pepper against Xanthomonas axonopodis pv. vesicatoria (Choi et al., 2014), by priming the SA and JA signaling pathways. Besides, several reports have shown the antifungal (Vaughn et al., 1993) and antibacterial properties (Croft et al., 1993; Deng et al., 1993) of GLVs, which thus reinforces the role of these VOCs in plant defense. Our results also reveal the possible defensive role of GLV esterification since these GLV esters were overproduced during ETI establishment.

Some of these GLVs, including hexanal, (Z)-3-hexenal or (E)-2-hexenal, have also been described as major compounds emitted by tomato fruit (Rambla et al., 2014), where Pseudomonas syringae can cause damage (Xin and He, 2013). Although esterase activity in fruit is enhanced in the red-fruited species of the tomato clade (Goulet et al., 2012), it would be interesting to study whether infection with virulent or avirulent bacteria could produce the esterification of these aldehydes in fruit, which would thus extend the defensive role of these GLVs esters to other organs.

Other VOCs involved in the defense response of plants are terpenoids, which are emitted after wounding or egg deposition by insects. Terpenoids induced by herbivores act in plant defense by attracting insect predators, and by acting as repellents or toxic compounds (Turlings and Tumlinson, 1992; Wegener et al., 2001). Besides, they have been associated to resistance against downy mildew in grapevine (Alarcón et al., 2015). However, the role of terpenoids in plant-bacteria interactions is not well studied. Here, we detected different HMTs, such as (Z)- and (E)-linalool oxides, linalool, α-terpineol, 4-terpineol, and six putative HMTs emitted from both infected plant types, where induction was greater in avirulent infection. The release of linalool and β-ocimene has also been described in tobacco plants infected by an avirulent strain of Pseudomonas (Huang et al., 2003). Since monoterpenes are emitted mainly by symptomatic plants, and HMTs are differentially released by those displaying the immune response, terpene hydroxylation appears to be a key process in the plant defense response.

In order to study whether the increase in VOC was associated with the induction of the VOC biosynthesis machinery, the expression of several genes, e.g., Tomlox and AAT, involved in the biosynthetic pathway of the esters of GLVs, and TPS, implicated in the biosynthesis of terpenoids, was studied by qRT-PCR. We observed a positive correlation in the induction of TomloxF, AAT1, and MTS1 with the emission of the corresponding VOCs, which were differentially released in the tomato plants that displayed ETI.

Among the six described Tomlox isoforms, the induction of the TomloxF gene has also been described to result from the infection caused by Pseudomonas putida (Mariutto et al., 2011). Our results reinforce the defensive role of this tomato LOX isoform and validate the metabolomic analysis. Regarding the different isoforms of the tomato alcohol acyltransferases, AAT1 has been correlated to the production of GLV esters in tomato fruits (Goulet et al., 2015). Accordingly, we observed the induction of the AAT1 gene upon bacterial infection, and the corresponding GLV esters accumulation, being this induction greater in the tomato plants that exhibited ETI.

Volatile isoprenoids represent the most abundant group of volatile compounds in plants, and are common components of both their aroma and defensive response induced by herbivores and pathogens (Aharoni et al., 2005). The observed induction of MTS1 upon bacterial infection correlates with the detected emission of monoterpenes and HMTs, and agrees with the previously described role of TPS in plant defense. Transgenic plants of Arabidopsis thaliana, which overexpress the TPS (E) β-caryophyllene synthase gene, emit larger amounts of this sesquiterpene and are more resistant to bacterial infection, which confirms the role of this gene in defense (Huang et al., 2012). MTS1 expression in tomato leaves has been described to be induced by spider mite-infestation, wounding and JA treatment (van Schie et al., 2007).

Terpene synthases reaction products are subsequently modified by hydroxylation, methylation, acylation, reduction, oxidation, isomerization, or glycosylation, which gives rise to more complex terpene compounds. Hydroxylation of monoterpenes, which would lead to compounds such as α-terpineol, is performed by Cytochrome P450 enzymes (CYP450) (Höfer et al., 2014). There are approximately 250 CYPs in tomato<sup>1</sup> , which means that clarifying which CYP450 might be responsible for terpene hydroxylation and is, therefore, involved in defense through this signaling pathway, is a complex task. However, approximately 25 CYP proteins have been reported to

<sup>1</sup>www.solgenomics.net

be involved in the immune response of tomato plants to bacteria Pst, which helps limit the search of the CYP450 responsible for the hydroxylation of terpenes and confirms the importance of hydroxylation in plant defense response (Pombo et al., 2014).

Our results suggested that the esters of GLVs and HMTs could play a defensive role in the tomato plant response. Unlike SA accumulation, which is the classical signal molecule in incompatible interactions that accumulates at higher levels in virulent infections, these VOCs were differentially emitted at higher levels when plants efficiently resisted bacterial infection, which indicates that they could play a defensive role. Further studies, such as pharmacological or genetic approaches, could be conducted to test this possibility. These volatile compounds could also display interesting biological properties, such as antioxidant, antimicrobial, insecticide or resistance inducers, and could be good candidates for agrochemical and pharmaceutical industries. Besides, the generation of tomato transgenic plants over-expressing enzymes involved in the biosynthesis of these volatiles could result in a new biotechnological strategy to obtain resistance.

#### AUTHOR CONTRIBUTIONS

MPL-G and JMB conceived and designed the study. MPL-G, PL, and LC carried out the experiments. IR prepared the figures. MPL-G, JLR, and AG performed the GC-MS-based

#### REFERENCES


metabolomics approach. MPL-G and PL did the data processing and statistical analysis. MPL-G, PL, IR, VC, and JMB interpreted the results. MPL-G and PL wrote the manuscript. JMB handled the literature.

# FUNDING

This work was funded by Grant BIO2012-33419 from the Spanish Ministry of Economy and Competitiveness.

#### ACKNOWLEDGMENTS

We would like to thank Dr. Selena Giménez-Sánchez (Centro Nacional de Biotecnología, Madrid) for kindly providing us with both strains of Pseudomonas syringae pv. tomato DC3000. We thank Dr. Isabel López-Díaz and Dr. Esther Carrera for the hormone quantification carried out at the Plant Hormone Quantification Service, IBMCP, Valencia, Spain. We are also very grateful to Teresa Caballero and Ana Ruiz for their excellent technical support.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2017.01188/ full#supplementary-material

Citrus Exocortis Viroid infection in tomato plants. J. Phytopathol. 125, 198–208. doi: 10.1111/j.1439-0434.1989.tb01061.x



bacterial pathogen. New Phytol. 193, 997–1008. doi: 10.1111/j.1469-8137.2011. 04001.x


tomato leaf volatiles. J. Chem. Ecol. 37, 565–574. doi: 10.1007/s10886-011- 9961-0


**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 López-Gresa, Lisón, Campos, Rodrigo, Rambla, Granell, Conejero and Bellés. 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.

# Untargeted Metabolomics Approach Reveals Differences in Host Plant Chemistry Before and After Infestation With Different Pea Aphid Host Races

Carlos Sanchez-Arcos<sup>1</sup>† , Marco Kai<sup>2</sup>† , Aleš Svatoš<sup>2</sup> , Jonathan Gershenzon<sup>1</sup> and Grit Kunert<sup>1</sup> \*

<sup>1</sup>Department of Biochemistry, Max-Planck Institute for Chemical Ecology, Jena, Germany, <sup>2</sup>Research Group Mass Spectrometry/Proteomics, Max-Planck Institute for Chemical Ecology, Jena, Germany

#### Edited by:

Andreia Figueiredo, Universidade de Lisboa, Portugal

#### Reviewed by:

Anna-Maria Botha-Oberholster, Stellenbosch University, South Africa Javier Sanchez, University of Zurich, Switzerland

> \*Correspondence: Grit Kunert gkunert@ice.mpg.de

#### †Present address:

Carlos Sanchez-Arcos, Institute for Inorganic and Analytical Chemistry, Friedrich Schiller University Jena, Jena, Germany Marco Kai, Department of Biochemistry, University Rostock, Rostock, Germany

#### Specialty section:

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

Received: 15 March 2018 Accepted: 05 February 2019 Published: 28 February 2019

#### Citation:

Sanchez-Arcos C, Kai M, Svatoš A, Gershenzon J and Kunert G (2019) Untargeted Metabolomics Approach Reveals Differences in Host Plant Chemistry Before and After Infestation With Different Pea Aphid Host Races. Front. Plant Sci. 10:188. doi: 10.3389/fpls.2019.00188 The pea aphid (Acyrthosiphon pisum), a phloem-sucking insect, has undergone a rapid radiation together with the domestication and anthropogenic range expansion of several of its legume host plants. This insect species is a complex of at least 15 genetically different host races that can all develop on the universal host plant Vicia faba. However, each host race is specialized on a particular plant species, such as Medicago sativa, Trifolium pratense, or Pisum sativum, which makes it an attractive model insect to study ecological speciation. Previous work revealed that pea aphid host plants produce a specific phytohormone profile depending on the host plant – host race combination. Native aphid races induce lower defense hormone levels in their host plant than non-native pea aphid races. Whether these changes in hormone levels also lead to changes in other metabolites is still unknown. We used a mass spectrometrybased untargeted metabolomic approach to identify plant chemical compounds that vary among different host plant-host race combinations and might therefore, be involved in pea aphid host race specialization. We found significant differences among the metabolic fingerprints of the four legume species studied prior to aphid infestation, which correlated with aphid performance. After infestation, the metabolic profiles of M. sativa and T. pratense plants infested with their respective native aphid host race were consistently different from profiles after infestation with non-native host races and from uninfested control plants. The metabolic profiles of P. sativum plants infested with their native aphid host race were also different from plants infested with non-native host races, but not different from uninfested control plants. The compounds responsible for these differences were putatively identified as flavonoids, saponins, non-proteinogenic amino acids and peptides among others. As members of these compound classes are known for their activity against insects and aphids in particular, they may be responsible for the differential performance of host races on native vs. non-native host plants. We conclude that the untargeted metabolomic approach is suitable to identify candidate compounds involved in the specificity of pea aphid – host plant interactions.

Keywords: Acyrthosiphon pisum, pea aphid host races, legume metabolome, saponins, flavonoids, Medicago sativa, Trifolium pratense, Pisum sativum

# INTRODUCTION

fpls-10-00188 February 26, 2019 Time: 15:6 # 2

Insects are the most diverse group of eukaryotic species on earth (Stork, 1993), and herbivorous species constitute a major group of insects (Strong et al., 1984; Mitter et al., 1988). Many herbivorous insect species are specialized on certain plant species (Bernays and Graham, 1988). Even within a single insect species, specialization can occur with different populations feeding preferentially on different plant taxa (Drès and Mallet, 2002). Adaptation to multiple plant species can lead to the formation of host races or biotypes within a species of an insect herbivore and might therefore play an important role in insect speciation. Nevertheless, the mechanisms behind these adaptations are hardly understood so far. Why is a certain host race able to feed on one plant but not another one? To some extent, differences in the constitutive chemical profiles of the plant species might be crucial (Hegnauer, 2001). However, various plants can also react differently to a specific insect herbivore (Arimura et al., 2005; Wu and Baldwin, 2010; Hogenhout and Bos, 2011; Sanchez-Arcos et al., 2016). Such differential plant responses can become apparent in changes in i.e., the plant metabolome (Jansen et al., 2009; Tzin et al., 2015).

Secondary metabolites used by plants as defenses against herbivores can act directly as feeding deterrents or toxins that decrease food intake or food-utilization efficiency (Gabrys and Tjallingii, 2002; Kim et al., 2008), decrease survival (Behmer et al., 2011), or indirectly as attractants for natural enemies of herbivores (reviewed in Unsicker et al., 2009). Chemical compounds involved in direct and indirect plant defense include terpenoids, phenolics, cyanogenic glycosides, glucosinolates, and alkaloids (Dreyer et al., 1985; Avé et al., 1987; Zagrobelny et al., 2004; Unsicker et al., 2009; Mithöfer and Boland, 2012; Maag et al., 2015; Züst and Agrawal, 2016). Specialized insect herbivores have to somehow cope with the presence of defensive secondary metabolites in their host plants, and have evolved specific adaptations to enable feeding (Ehrlich and Raven, 1964; Berenbaum and Zangerl, 1998).

To evaluate the potential defensive effects of plant metabolites on insect herbivores, the identity and concentration of these substances must be measured. Targeted analyses can be used for such studies if previous knowledge suggests that a specific metabolite or metabolite class like flavonoids or glucosinolates might be important. However, in some cases previous knowledge about the chemical composition of the plants involved or compounds relevant to herbivores is not available. In these cases the untargeted investigation of the whole metabolome by the use of metabolomics, might be a good tool to reveal candidate chemicals that are the cause of different plant-insect interactions (Fiehn, 2002; Krastanov, 2010; Nakabayashi and Saito, 2013).

One especially interesting plant-insect herbivore system to study the role of plant metabolites for speciation is the interaction between the pea aphid complex (Acyrthosiphon pisum Harris, Homoptera, Aphididae) and its legume (Fabaceae) host plants. The pea aphid underwent a rapid diversification about 6500–9500 years ago (Peccoud et al., 2009b) that led to the development of at least 15 different sympatric host races or biotypes each specialized on one or a few legume host plants (called native host plants) on which they perform well and prefer for feeding (Peccoud et al., 2009a, 2015). Such host plant preferences lead to assortative mating and reproductive isolation among populations (Caillaud and Via, 2000; Peccoud et al., 2009a). However, all pea aphid host races can perform well on Vicia faba, which is considered to be the universal host plant for all pea aphid host races characterized to date. The existence of the different host races and the fact that the pea aphid genome is entirely sequenced (The International Aphid Genomics Consortium, 2010) has made the pea aphid a model for studying ecological speciation (Brisson and Stern, 2006; Peccoud and Simon, 2010) and provides the opportunity to investigate the mechanisms underlying host plant adaptation.

Pea aphids like all aphids feed on phloem sap. With their sucking mouthparts called stylets they navigate through the epidermis and mesophyll to reach the phloem. During this penetration process they pierce many plant cells, salivate into the cells and also suck tiny amounts of cell contents (Tjallingii and Esch, 1993; Martin et al., 1997). It is assumed that aphid salivary proteins are involved in the adaptation of pea aphid host races to host plants (Jaquiery et al., 2012). Recent studies with other aphid species have revealed a close relation between proteins secreted with the saliva into the plant and host plant reactions (Rodriguez and Bos, 2013; Kaloshian and Walling, 2016). While several aphid proteins were found to facilitate aphid feeding (Will et al., 2007; Bos et al., 2010; Atamian et al., 2013; Elzinga et al., 2014; Naessens et al., 2015); other aphid proteins induce defense reactions in the plant and could lead to an incompatible aphid – plant interaction (Chaudhary et al., 2014; Elzinga et al., 2014). Concerning the pea aphid host races and their adaptation to their native hosts, despite the efforts made to investigate the role of candidate saliva proteins on host plant adaptation (Jaquiery et al., 2012; Guy et al., 2016; Boulain et al., 2018; Nouhaud et al., 2018) saliva proteins important for host plant specialization are not yet known. However, it is known that legumes differ in their production of defense hormones depending on whether native or non-native pea aphid host races are feeding on plants (Sanchez-Arcos et al., 2016). These hormone differences might lead to changes in plant metabolomes, especially for compounds having a deterrent or toxic impact on aphids.

Several studies show that plant secondary compounds have detrimental effects on pea aphids (Züst and Agrawal, 2016). For example, higher levels of saponins and phenolic compounds led to a reduction of aphid population growth (Golawska et al., 2006; Golawska and Lukasik, 2009) and increased mortality (De Geyter et al., 2012). Diverse flavonoid glycosides reduced aphid fecundity (Golawska and Lukasik, 2012; Golawska et al., 2012a), while nitrogen-containing compounds caused a rejection of a potential host plant by pea aphids (Kordan et al., 2012). Although these targeted approaches revealed that some secondary metabolites affect pea aphid performance, in most of these studies just one plant species was used and often the pea aphid host race was unknown. Thus, the contribution of plant compounds to the maintenance and performance of pea aphid host races on legume plants is still largely unknown. Only one study (Hopkins et al., 2017) combined a metabolomic profiling approach with behavioral tests to understand the chemical signatures that underlie host preferences by A. pisum. This study investigated

the metabolome of uninfested plants of different plant species belonging to the genera Medicago and Trifolium, but was limited to constitutive defense compounds or other compounds important for initial acceptance.

To pave the way for later targeted analyses in which the contribution of plant metabolites to the maintenance and performance of pea aphid host races on different legume plants could be analyzed, we applied an untargeted mass spectrometrybased metabolomic approach. Polar and semi-polar fractions of three native host plants of the pea aphid, Medicago sativa, Pisum sativum, Trifolium pratense, and the universal host Vicia faba, each infested with their native or one of two non-native aphid host races were analyzed and compared to fractions of uninfested control plants. These data made it possible to evaluate the use of metabolomics in identifying plant metabolites potentially involved in determining host race-host plant interactions in pea aphids.

## MATERIALS AND METHODS

#### Plant Material

Four legume plant species, Medicago sativa (alfalfa) cultivar (cv.) "Giulia," Trifolium pratense (red clover) cv. "Dajana," Pisum sativum (pea) cv. "Baccara," and Vicia faba (broad bean) cv. "The Sutton," were grown in 10 cm diameter plastic pots with a standardized soil mixture (7:20 mixture of Klasmann Tonsubstrat and Klasmann Kultursubstrat TS1, Klasmann-Deilmann GmbH, Geeste, Germany), in climate chambers at 20◦C, 70 ± 10% relative humidity, and under a 16 h light/8 h dark photoperiod. The plants were watered twice a week. To have a sufficient amount of plant material for the extraction of metabolites, M. sativa and T. pratense plants were used 4 weeks after sowing, while P. sativum and V. faba were used 3 weeks after sowing.

#### Aphids

Three pea aphid (Acyrthosiphon pisum Harris) clones, each representing one pea aphid host race, were used in the experiments: the clone L84 representing the Medicago race (here called MR), the clone T3-8V1 representing the Trifolium race (TR), and the clone Colmar representing the Pisum race (PR). Aphids were initially collected from their native host plants T. pratense, M. sativa, and P. sativum, respectively, and genotypically assigned to their respective host race (for detailed information see Table S1 in Peccoud et al., 2009b). All aphids were reared on 4 week old broad bean plants. To synchronize the age of the aphids for the experiments, five apterous female adults were placed on a broad bean plant and were allowed to reproduce for 48 h and were then removed from the plants. Nymphs were kept on the plants for 9 days until they reached adulthood. Then they were transferred to new plants were they reproduced. This procedure was repeated until enough synchronized young adult aphids were available for the experiment. To avoid escape of aphids, all aphid-containing plants were covered with air permeable cellophane bags (18.8 × 39 cm, Armin Zeller, Nachf. Schütz & Co, Langenthal, Switzerland), and placed in a climate chamber under the same conditions described for the plant material.

# Experimental Design

Five adult apterous female aphids of each host race were placed in magnetic clip-cages (Ø 3.5 cm), on leaves of each plant species (two leaves for M. sativa and T. pratense, one leaf for P. sativum and V. faba plants). Leaves from all four plant species enclosed in magnetic clip cages but without aphids served as controls (**Supplementary Figure 1**). Ten replicates of each combination were employed. All the infested and control plants were placed in climate chambers at 20◦C, 70 ± 10% relative humidity, and under a 16 h light/8 h dark photoperiod. Plant material was sampled after 48 h, a period which allowed the aphids to settle and the plant to react to the aphid infestation (Sanchez-Arcos et al., 2016).

# Plant Material Sampling and Metabolite Extraction

For plant material sampling, the clip cages were carefully opened, and aphids were removed using a paintbrush. Control plants without aphids were brushed in the same way as aphid-infested plants to control for possible induction of metabolic changes due to contact with the paintbrush. Leaves enclosed in the clip cages were harvested and rapidly frozen in liquid nitrogen. Frozen samples were stored overnight in Eppendorf tubes (2 ml) at −80◦C and then freeze-dried for 48 h. Dried plant material was homogenized into a fine powder by adding three stainless steel beads (3 mm Ø) in each tube and vigorous shaking for 4 min on a paint shaker (Skandex shaker SO-10m, Fast & Fluid Management, Sassenheim, The Netherlands). Portions of 10 mg dried plant material were extracted with 1 ml ice-cold extraction solution containing 80% methanol acidified with 0.1% formic acid and 0.1 µg/ml of L-(+)-α-phenylglycine (as a lock mass internal standard). Samples were immediately vortexed for 10 s and continuously sonicated in a water bath at room temperature (20◦C) for 15 min at a maximum frequency of 35 kHz. After centrifugation (10 min at 4,500 g and −10◦C), supernatants were filtered using 0.45 µm PTFE AcroPrepTM 96-well filtration plates (Pall Corporation, Port Washington, NY, United States) and a vacuum filtration unit. All filtered plant extracts were stored at −80◦C until LC-Orbitrap-MS analysis.

#### Plant Extract Analysis

From each plant extract 10 µl were analyzed using a UHPLC system of the Ultimate 3000 series RSLC (Dionex, Sunnyvale, CA, United States) connected to an LTQ-Orbitrap XL mass spectrometer (Thermo Fisher Scientific, Bremen, Germany). UHPLC was performed on an AcclaimTM C18 column (150 × 2.1 mm, 2.2 µm, Dionex) pre-fitted with a C-18, 3.5 µm guard column (2.1 × 10 mm, Waters, Dublin, Ireland). Separation was accomplished using a gradient of 0.1% (v/v) formic acid in water (solvent A) and 0.1% formic acid in acetonitrile (solvent B) as follows: 0–5 min isocratic 100% (v/v) A, 5–32 min gradient phase to 100% B, 32–42 min isocratic 100% B, 42–42.1 min gradient phase to 100% (v/v) A, 42.1– 47 min isocratic 100% A. The flow rate was set to 300 µl min−<sup>1</sup> .

The electrospray ionization (ESI) source parameters were set to 4.5 kV spray voltage, and 35 V capillary transfer voltage at a capillary temperature of 275◦C. The samples were measured in the negative (NI) and positive (PI) ionization modes in separate runs using 30,000 m/1m resolving power (mass range of m/z 150–2000) in the Orbitrap mass analyzer. XcaliburTM software (Thermo Fisher ScientificTM, Waltham, MA, United States) was used for data acquisition and visualization.

UHPLC-MS<sup>2</sup> analysis of selected compounds was carried out by injecting l µl of each extract into a UHPLC system (Dionex UltiMate 3000, Thermo Fisher Scientific, Dreieich, Germany) coupled to a Q-Exactive Plus mass spectrometer (Thermo Fisher Scientific). Separation was performed on a AccucoreTM C18 column (2.1 × 100 mm, 2.6 µm, Thermo Fisher Scientific), using a gradient of 0.1% (v/v) formic acid in water (solvent A) and 0.1% formic acid in acetonitrile (solvent B) as follows: 0–0.2 min isocratic 100% (v/v) A, 0.2–8 min gradient phase to 100% B, 8–11 min isocratic 100% B, 11.1–12 min isocratic 100% (v/v) A. A constant flow rate of 400 µl min−<sup>1</sup> was set. For the parallel reaction monitoring, selective ion scanning in the negative ionization mode was used with the following parameters: target ions [M-H]<sup>−</sup> (m/z 1085.55 and m/z 605.19); resolution: 17,500; AGC target: 2 × 10<sup>5</sup> ; maximum IT: 100 ms; sheath gas flow rate: 60; aux gas flow rate: 20; sweep gas flow rate: 5; spray voltage: 3.3 kV; capillary temperature: 360◦C; S-lens RF level: 50; aux gas heater temperature: 400◦C; acquisition time frame: 3.75–5.88 min.

#### Data Preprocessing

Raw data files were converted to mzXML format files using the MSconvert tool (ProteoWizard 3.0x software) and uploaded to the interactive XCMS online platform (Tautenhahn et al., 2012). Parameter settings for XCMS data processing were as follows: A multigroup analysis was run in the centWave mode for feature detection (1 m/z = 2.5 ppm, minimum peak width = 10 s, and maximum peak width = 60 s); correction of the retention time was performed with an obiwarp method (profStep = 1); and for chromatogram alignment: minfrac = 0.5, bw = 5, mzwid = 0.015, max = 100, minsamp = 1. Tables with the intensities of the detected features were obtained as output. Features that were missing in 3 or more out of the 10 samples for each treatment combination were classified as sporadic features and were discarded from the data set. This resulted in 5735 features in the negative ionization mode and 9057 features in the positive ionization mode.

#### Data Analysis

For the initial colonization of plants by aphids, constitutive compounds that are uniquely characteristic of a plant species might play a crucial role. A feature was classified as unique to a certain plant species when it fulfilled the following criteria: It appeared in at least eight of ten samples of the plant species of interest, and was absent in the other plant species or appeared in not more than two out of ten samples per plant species. In the same way features were selected that appeared in two or more plant species. Thus 3132 features (NI) and 5323 features (PI) were analyzed. Unique features (1081 NI; 1013 PI) were filtered again

to remove adducts or isotopes which resulted in 280 (NI) and 289 (PI) possible metabolites.

To determine how the metabolic profiles differed between uninfested control plants, principal component analyses (PCAs) were performed with the web-based tool MetaboAnalyst (Xia et al., 2009; Xia and Wishart, 2011). Due to technical reasons (limited number of features which can be processed in the program) 40% of the features that were near constant throughout the plant species based on the interquartile range were filtered out prior to analyses. Thus 3441 features out of 5735 features (NI) and 5434 features out of 9057 features (PI) were analyzed in the PCA. Features were then mean-centered and divided by the standard deviation of each feature (equivalent to auto scaling) to make them comparable. In order to support the results of the PCA and to check which compound classes contribute most to the separation of the different host plants, partial least squares discriminant analyses (PLS-DA) with unique features only were performed (**Supplementary Figure 3**).

To determine how metabolic profiles changed within a given host plant species after pea aphid infestation, several PCAs were performed. Since it was assumed that most features would not differ among plants of the same species infested with different aphid host races and uninfested control plants, PCA analyses were carried out on the 5% of metabolomic features that changed most in each species. To identify the 5% most differently regulated host race features, all features were compared by a non-parametric one-way ANOVA on ranks and sorted by their false discovery rate (FDR). Before PCA these 5% most differently regulated features were normalized by log-transformation and scaled by mean centering and division by the standard deviation of each feature (auto-scaling) to make them comparable.

To detect chemical compounds that might be involved in plant defense, we looked for features that were down regulated in plants when a native aphid race was feeding but up regulated when non-native aphid races were applied using the pattern hunter tool in the MetabolAnalyst tool. A Spearman rank correlation analysis was performed against given patterns. A pattern was specified as a series of numbers, where each number corresponded to the concentration levels of the features in the corresponding group. For instance, the pattern "2-1-2- 2" corresponding to the groups "uninfested control plants – plants infested with the native aphid race – plants infested with non-native aphid race A – plants infested with nonadapted aphid race B" searched for features down regulated (positive correlation) or up regulated (negative correlation) exclusively by the native aphid race. To test whether intensities of selected features differed among the four treatments, oneway analyses of variance (one-way ANOVA) were performed. In case of significant differences, Tukey HSD tests were executed to reveal which groups were different from each other. These univariate analyses were conducted using R software version 3.2.0 (R Development Core Team, 2015).

Selected features were assigned to chemical groups through putative identifications by performing library mass searches allowing a mass deviation in all databases of 5 ppm, and checked for spectrum matches in METLIN, Human Metabolome Database (HMDB), MetFrag, MassBank, KEGG, and LipidMaps.

UHPLC-Orbitrap mass spectrometry of aqueous methanol were analyzed in (A) negative and (B) positive ionization modes. Colored circles represent the metabolic profiles of individual plants. Dotted ellipses represent the 95% confidence regions for each group.

Features not found in these libraries or databases were considered as "unknown."

#### RESULTS

## Pea Aphid Host Plant Species Differ in Their Constitutive Metabolic Profiles

To evaluate how the metabolic profiles of the four uninfested plant species, M. sativa, T. pratense, P. sativum and V. faba, differed, PCA were conducted. PCA plots showed a clear separation of all four plant species in both ionization modes (**Figure 1**). Biological replicates of each plant species always grouped together with small confidence intervals. The first principal components (PC1) explained 26.5 and 24.8% of the total variability for negative ionization mode (NI) and positive ionization mode (PI) datasets, respectively, whereas the second principal components (PC2) accounted for 20.8 and 20.9% (for NI and PI modes, respectively) of the total variability of the data set. For both ionization modes, the metabolic profiles of M. sativa, T. pratense, and V. faba were separated mainly along PC1. P. sativum metabolic profiles were separated from those of the other plant species along PC2.

To visualize the characteristic features shared among plant species, Venn diagrams were used (**Figure 2**). In both ionization modes, most features were shared between all four plants species (1676 out of 3132 features or 53.5% and 3865 out of 5323 features or 72.6% of all features for NI and PI, respectively), while features unique to a certain plant species were much less common. Only 34.5% (1081 features; NI) and 19% (1013

features; PI) of all features were assigned to only one plant species. M. sativa and T. pratense plants possessed a higher number of unique features in comparison to P. sativum and V. faba plants, while V. faba displayed the lowest number of unique features. Additionally, M. sativa and T. pratense plants shared more common features, with 149 and 109 features for NI and PI modes, respectively, than any other pair of plant species (**Figure 2**).

Metabolites unique to a pea aphid host plant could serve as host identification cues responsible for the acceptance or rejection of a potential host by the different host races, and might even function as defenses against non-native races. To obtain an overview of the unique compounds of each plant

species, we assigned putative identifications to the unique features and organized them by chemical classes (**Table 1** and **Supplementary Table 1**). M. sativa displayed the highest number of unique metabolites with 107 and 86 in NI and PI mode, respectively, followed by T. pratense with 103 and 83 metabolites, P. sativum with 57 (NI) and 53 (PI) metabolites, and V. faba

TABLE 1 | Numbers of unique metabolites in uninfested M. sativa, T. pratense, P. sativum and V. faba plants and their putative chemical classification.


Retention times, m/z values and putative chemical class of the compounds can be found in Supplementary Table 1.

with 13 (NI) and 67 (PI) metabolites. From all these compounds 55% (156 out of 280 compounds; NI) and 40% (115 out of 289 compounds; PI) were putatively identified. The most common class of putatively identified unique compounds among all plant species was the flavonoids. Steroidal and triterpene saponins were not only specific but also the most abundant classes in M. sativa (**Table 1**, **Supplementary Table 1** and **Supplementary Figure 3**).

#### Metabolic Profiles of Host Plant Species Are Modified Differently by the Various Pea Aphid Host Races

To find out how plant metabolic profiles were modified after infestation by the various pea aphid host races, PCAs were performed based on the 5% of metabolomic features that changed most among plants infested with different pea aphid host races and uninfested control plants. In general, the metabolic profiles of uninfested control plants were separated from the profiles of aphid-infested plants, and thus plant metabolomes changed significantly upon aphid infestation. The degree of separation among profiles of plants infested with different host races depended on the plant species. Whereas the metabolic profiles of V. faba and M. sativa changed substantially depending on the attacking host race, profiles of T. pratense and P. sativum changed to a smaller extent (**Figure 3**).

The metabolic profiles of aphid-infested M. sativa plants were separated from the profiles of uninfested control plants along the first principal component (**Figures 3A,B**). The first PC explained 29% (NI) and 23.4% (PI) of the total variability. Among the infested plants, the metabolic profiles of those infested with the native MR host race were separated from those infested with the non-native TR and PR host races mainly along the second principal component. This explained 17.5% (NI) and 12.3% (PI) of the total variability.

The metabolic profiles of aphid-infested T. pratense plants also separated from those of uninfested control plants, but only in the positive ionization mode (**Figure 3D**) and not the negative ionization mode (**Figure 3C**). Furthermore, those plants infested with the native TR host race were grouped apart from those of the non-native MR and PR host races, especially in the positive ionization mode (**Figure 3D**). Of the total variability in the metabolic profiles of T. pratense, 62.5% (NI) and 49.8% (PI), was explained by the first two principal components.

The proportion of variability in the metabolic profiles of P. sativum that could be explained by the first two principal components was slightly lower than in the other plant species, 39.5% (NI) and 39.1% (PI). In contrast to the other three plant species where metabolic profiles of uninfested plants separated well from profiles of aphidinfested plants, in P. sativum the metabolic profiles of uninfested plants overlapped to some extent with those of plants infested with the native PR host race (**Figures 3E,F**). However, in the negative ionization mode, the metabolic profiles of plants infested with the non-native MR and TR host races were separated from those of uninfested control plants along the first principal component (**Figure 3E**). In

the positive ionization mode, the metabolic profiles of plants infested with non-native TR race separated from the metabolic profiles of the other treatments along the second principal component (**Figure 3F**).

A large proportion of the variability in metabolic profiles of the universal host plant V. faba could be explained by the first two principal components (82.9% for NI and 47.6% for PI). Metabolic profiles changed drastically between the differently treated plants, especially in the negative ionization mode. There was a clear separation between infested and uninfested plants in both ionization modes along the first principal component (**Figures 3G,H**). The second principal component separated MR host race-infested and uninfested plants from PR and TR host race-infested plants in the negative ionization mode (**Figure 3G**).

#### Some Metabolites Are Reduced by Native Pea Aphid Host Races, but Induced by Non-native Races

Pea aphid host races perform much better on their native host plants than on other species (Ferrari et al., 2006, 2008; Peccoud et al., 2009a, 2015; Schwarzkopf et al., 2013). This difference may be a consequence of the ability of each race to suppress defense signaling processes on its native plant and reduce the levels of defenses (Sanchez-Arcos et al., 2016). When feeding on a non-native host plant, on the other hand, aphid feeding may trigger signaling that leads to the induction of defenses like toxic or deterrent compounds. To test these ideas, we searched for metabolites that showed (1) significantly reduced levels only after infestation with native aphid host races, or (2) significantly increased levels only after infestation with non-native aphid host races (**Table 2** and **Supplementary Table 2**).

M. sativa plants contained the highest number of metabolites that fit with these patterns, with 4 and 13 metabolites detected in the NI and PI modes, respectively, followed by T. pratense with 4 (NI) and 10 (PI) metabolites and P. sativum with 2 (NI) and 3 (PI) metabolites. A number of compounds showing these patterns could be assigned to chemical classes including several flavonoids, a triterpene saponin, a sterol and a jasmonate derivate in M. sativa, a flavonoid and a non-proteinogenic amino acid in T. pratense and a flavonoid in P. sativum. The abundance of two of the M. sativa compounds in different pea aphid treatments is illustrated in **Figure 4**. The levels of a putatively identified triterpene saponin (MS: with [M-H]<sup>−</sup> ion at m/z 1085.55, [M+FA (formic acid)-H]<sup>−</sup> adduct ion at m/z 1131.56, and [M-shikimic acid-H]<sup>−</sup> ion at m/z 911.46; **Supplementary Figures 2A,B**) were decreased by aphid infestation only with the native MR host race. On the other hand, the amount of a compound, putatively classified as a glycosylated flavonoid (MS: with [M-H]<sup>−</sup> ion at m/z 605.19, MS<sup>2</sup> : fragment ions at m/z: 577.16 and 561.20, **Supplementary Figures 2C,D**), was significantly increased upon infestation with the non-native

significantly increased only after infestation with non-native aphid host races (B). (A) Metabolite with [M-H] <sup>−</sup> m/z 1085.55, was decreased only by the native MR host race. (B) Metabolite with [M-H]<sup>−</sup> m/z 605.19 was induced only by the non-native PR and TR host races. Both metabolites were detected in negative ionization mode. Bars represent means ± SE. MR, Medicago race; TR, Trifolium race; PR, Pisum race; Uninfested, uninfested control plants. Different letters indicate significant differences (P ≤ 0.05) based on One-way ANOVA followed by Tukey HSD test. MS and MS<sup>2</sup> spectra of these two compounds can be seen in Supplementary Figure 2.

TR and PR host races, but remained at similar levels as in uninfested control plants when the native MR host race infested the plant (**Figure 4B**).

TABLE 2 | Numbers of compounds down-regulated only after infestation with native, but not non-native aphid host races or up-regulated only after infestation with non-native, but not native host races, and their putative chemical classification.


Retention times, m/z values and putative formulas of the compounds can be found in Supplementary Table 2.

# DISCUSSION

## Chemical Differences Among Pea Aphid Host Plants Prior to Infestation Are Correlated With Subsequent Aphid Performance

Untargeted mass spectrometry-based analysis of the metabolites of four legume host plants of the pea aphid showed significant differences among these plants prior to aphid feeding (**Figure 1**). These differences might help explain the differential performance of the various pea aphid host races on specific host plants since individual metabolites could stimulate or deter feeding in disparate ways. Interestingly, the chemical complexity of the four plant species (the number of unique features found in our mass spectrometry analysis) is correlated with the performance of pea aphid host races on the various plants. Medicago sativa, which contained the highest number of unique chemical features (**Figure 2**), was previously found to be completely unsuitable for feeding by all other host races except the Medicago host race (Peccoud et al., 2009a; Schwarzkopf et al., 2013; Sanchez-Arcos et al., 2016). Trifolium pratense, which contained the second highest number of unique chemical features, was found to be unsuitable for other host races (Peccoud et al., 2009a), but did support some growth and reproduction of the Medicago host race (Peccoud et al., 2009a; Sanchez-Arcos et al., 2016).

The exceptional behavior of the Medicago race on T. pratense might be ascribed to the many chemical features shared between M. sativa and T. pratense (**Figure 2**), which also parallels their close phylogenetic relationship as members of the Tribe Trifolieae of the Fabaceae (Wojciechowski et al., 2004; The Legume Phylogeny Working Group et al., 2013). On the other hand, V. faba and P. sativum, both belonging to the Tribe Vicieae (Wojciechowski et al., 2004), had a much lower number of unique features, and can therefore be considered less chemically complex than M. sativa and T. pratense. This might be the reason for the good performance of all pea aphid host races on V. faba and their

intermediate performance on P. sativum (Peccoud et al., 2009a; Schwarzkopf et al., 2013; Sanchez-Arcos et al., 2016).

The negative correlations between host plant chemical complexity and pea aphid performance suggest that the unique metabolites of each host might play a role in limiting aphid feeding. The most abundant class of unique chemical compounds found in M. sativa, and also responsible for the separation of M. sativa from other plant species were putatively identified as steroidal and triterpene saponins (**Supplementary Figure 3**). Saponins originate from the mevalonate pathway (**Supplementary Figure 4**), and although saponins are distributed in many plant families, they are frequently found in legumes (Vincken et al., 2007), especially the genus Medicago, which is well known for containing a complex mixture of triterpene saponins with a broad spectrum of biological properties (Carelli et al., 2015). Saponins are reported to be defenses against aphids (De Geyter et al., 2012; Golawska et al., 2012b, 2014) as well as other insect herbivores (De Geyter et al., 2007; Chaieb, 2010), pathogens (Oleszek et al., 1990; Abbruscato et al., 2014), and nematodes (D'Addabbo et al., 2011). Saponins exhibit strong detergent properties, and can interact with membranes disturbing permeability and leading to cell death and necrosis (Wink, 2008).

The most frequent class of unique chemical compounds putatively identified in the other pea aphid host plants studied was the flavonoids. They are synthesized by the phenylpropanoid pathway (**Supplementary Figure 4**) and are widely distributed in all plants (Iwashina, 2000), but flavonoids, particularly isoflavonoids, are especially abundant in legumes (Velázques et al., 2010; Wink, 2013). Flavonoids can be deterrent or toxic for insects (Brignolas et al., 1998; Widstrom and Snook, 2001; Haribal and Feeny, 2003) including phloem feeders such as aphids (Montgomery and Arn, 1974; Lattanzio et al., 2000), and also have activity against microbes (Grayer and Harborne, 1994; Rauha et al., 2000). Interestingly, in a previous metabolomics investigation of host plant choice in pea aphids, L-phenylalanine and L-tyrosine were found to be associated with differential acceptability of species of Medicago and Trifolium by the Medicago and Trifolium host races (Hopkins et al., 2017). Since L-phenylalanine is one of the major precursors of flavonoids, its correlation with pea aphid host choice may be due to the flavonoids formed from this amino acid. Thus untargeted chemical analysis of pea aphid host plants has revealed several candidate groups of compounds that may play a role in host race specificity. Future studies on these compounds are warranted to determine how they are involved in defining pea aphid-host plant interactions.

# Host Plant Compounds Induced or Suppressed by Pea Aphid Host Races May Influence Their Performance

Pea aphids may be affected not only by the constitutive chemistry of their host plants, but also by the changes induced upon aphid feeding. Thus in this study we compared the metabolic composition of host plants after feeding by the different host races with the composition of uninfested plants. Our results showed that host plant metabolomes were indeed modified by aphid feeding with modification occurring in a host race-dependent manner. The metabolic profiles of M. sativa and T. pratense plants after aphid infestation were consistently different from those of their respective uninfested control plants, and the metabolic profiles measured after feeding by the Medicago and Trifolium native host races were different from those after feeding by non-native host races (**Figures 3A–D**). In contrast, the metabolic profile of P. sativum after infestation by the native Pisum host race was not different from that of the control plants, but infestation with the non-native host races generated distinct changes in metabolic profiles. These patterns of metabolic change correspond well with changes in defense hormone levels found in a previous study after pea aphid feeding (Sanchez-Arcos et al., 2016). The salicylic and jasmonate levels of M. sativa and T. pratense were substantially altered by aphid infestation. Native host races caused slight induction or reduction of hormone levels, and non-native host races caused significant increases. In P. sativum, the native host race did not modify defense hormone levels compared to those found in uninfested control plants.

The metabolic changes measured after aphid feeding may result from a series of aphid behaviors that occur during the establishment of phloem feeding. En route to the phloem, aphids pierce many plant cells and secrete watery saliva into these cells (Martin et al., 1997; Powell, 2005). They also secrete gelling saliva into the apoplast to form a sheath around the stylet. Interestingly, even in incompatible interactions (such as those resulting from non-native host races) when aphids are not able to feed on the sieve elements, they insert their stylets and salivate into the plant (Schwarzkopf et al., 2013). The saliva contains proteins some of which may function as herbivore associated molecular patterns that bind to plant recognition receptors leading to the activation of plant defense responses (Hogenhout and Bos, 2011). For instance, a proteinaceous elicitor from M. persicae saliva induced resistance against this aphid in Arabidopsis, and pre-treatment of the host plant with aphid saliva decreased aphid fecundity (De Vos and Jander, 2009). Some of the metabolic changes observed after pea aphid infestation in this study might be a direct consequence of the defense responses induced by salivary proteins.

Aphids that are able to feed on a plant, such as native host races, may prevent the induction of defense responses by secreting effector proteins into the plant (Hogenhout and Bos, 2011; Elzinga and Jander, 2013; Van Bel and Will, 2016). For example, the effector molecule C002 from the pea aphid was shown to be essential for feeding on its universal host plant V. faba, and silencing of the C002 gene reduced aphid fecundity (Mutti et al., 2008). However, plants have developed mechanisms to recognize such effector molecules and consequently activate their defense. Thus the chemical changes occurring after pea aphid infestation may reflect the outcome of the plantaphid interaction.

The specificity of host race performance on different host plants may well be due to chemical differences among plants induced by infestation. Plants commonly synthesize new chemical compounds after herbivore attack, but this phenomenon has been studied for phloem feeding insects in only a few cases (Züst and Agrawal, 2016). Thus, in this study we sought compounds that were either induced by non-native host races but not native races, or suppressed by native host races but not non-native races. Such substances might be critical in facilitating the good performance of native host races and poor performance on non-native races. Interestingly, the number of compounds that were up or down regulated in this manner was again higher in M. sativa and T. pratense than in P. sativum (**Table 2**).

One of the compounds with these characteristics, putatively classed as a triterpene saponin, was unique to M. sativa and down regulated only by the native MR host race (**Figure 4A**). Saponins in plants have been reported to play a widespread role in resistance against insect herbivores (Applebaum et al., 1969; Fields et al., 2010; Nielsen et al., 2010), including in M. sativa (Nozzolillo et al., 1997; Adel et al., 2000), and in resistance against aphids in particular (Soulé et al., 2000; Golawska et al., 2012b). The adverse effects of saponins against the pea aphid have been observed in several studies (Golawska, 2007; De Geyter et al., 2012). These results suggest that the down regulation of the M. sativa saponin by the native Medicago host race, as described above, could be responsible for the increased performance on its native host plant. At the same time, the inability of non-native host races to down regulate this saponin may contribute to their poor performance due to the potential deterrent or toxic effects of this compound.

Another compound whose levels varied in a host racespecific manner was putatively classed as a sterol. Plant sterols are essential dietary requirements for insects, which cannot produce their own de novo, but need sterols as structural components of cell membranes and precursors in the formation of ecdysteroids (molting and sex hormones) (Clark and Bloch, 1959; Behmer and Nes, 2003; Behmer, 2017). However, sterols can also have negative effects on insects (Behmer, 2017). For instance the sterol stigmasterol suppressed the lifetime reproductive output of pea aphids when added to an artificial diet (Bouvaine et al., 2012). The aphid M. persicae had a reduced reproduction and a higher mortality when feeding on tobacco plants with high levels of atypical steroids (Behmer et al., 2011). Thus the sterol identified in this study may also represent a good candidate to be tested further with different pea aphid host races to deduce its role in determining patterns of host plant specificity.

Another compound that was induced in M. sativa after infestation with non-native, but not native host races was putatively classified as a flavonoid (**Figure 4B**). Increased levels of flavonoids in legumes after pea aphid infestation were previously reported to cause detrimental effects on the pea aphid (Golawska and Lukasik, 2012). Host race-specific regulation of flavonoids also occurred in T. pratense. Thus flavonoids are also good candidates to explain the differential performance of pea aphid host races.

# CONCLUSION

An untargeted metabolomics approach allowed the detection of a number of candidate compounds that could be responsible for the disparate performance of various pea aphid host races on leguminous plant species. Some of these candidates are constitutive substances that are unique to one of the host plants. Other candidates were differentially induced or suppressed following infestation by the various host races. The most abundant chemical classes represented were saponins and flavonoids. Based on these results, untargeted metabolomics should be more widely considered as a valuable technique to discover plant metabolites of potential importance in plant interactions with herbivores and pathogens. When related plant species differ in their susceptibility or resistance, metabolomic comparison before and after enemy attack may reveal the chemical compounds responsible.

# AUTHOR CONTRIBUTIONS

CS-A and GK conceived and designed the experiments. CS-A performed the experiments. MK measured the metabolic profiles. CS-A and GK analyzed the data. CS-A, GK, and JG interpreted the results and wrote the manuscript. All authors critically revised and consented to the final version of the manuscript.

# FUNDING

This project was supported by the Max Planck Society, Germany, and the German Academic Exchange Service (DAAD) within the framework of the PPP – Procope German-French collaboration program (Project Code 57049694).

# ACKNOWLEDGMENTS

We thank Jean-Christophe Simon (INRA, Le Rheu, France) for providing the aphid clones, Kerstin Zipfel, Christoph Scheck, Lindsey Roark, and Kristina Schädel for experimental help, Andreas Weber and Elke Goschala for plant cultivation, Tim Baumeister for his help with the visualization of PLS-DA results, and Ilka Vosteen, Elizabeth Pringle, and Maria Paulmann for helpful comments on previous versions of the manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2019.00188/ full#supplementary-material

# REFERENCES

fpls-10-00188 February 26, 2019 Time: 15:6 # 11




**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 © 2019 Sanchez-Arcos, Kai, Svatoš, Gershenzon and Kunert. 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(s) 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.

# Polyphenolic Composition of Lentil Roots in Response to Infection by Aphanomyces euteiches

#### Navid Bazghaleh\*, Pratibha Prashar, Randy W. Purves and Albert Vandenberg

Department of Plant Sciences, University of Saskatchewan, Saskatoon, SK, Canada

Polyphenols comprise the largest group of plant secondary metabolites and have critical roles in plant physiology and response to the biotic and abiotic environment. Changes in the content of polyphenols in the root extracts and root tissues of wild (Lens ervoides) and cultivated (Lens culinaris) lentil genotypes were examined in response to infection by Aphanomyces euteiches using liquid chromatography mass spectrometry (LC-MS). Genotype, infection and their interaction determined the composition of polyphenols in lentil roots. The levels of several polyphenols were lower in the root extract of the low-tannin genotype L. culinaris ZT-4 compared to L. ervoides L01-827A. Kaempferol derivatives including kaempferol dirutinoside and kaempferol 3-robinoside 7-rhamnoside were more concentrated in the healthy root tissues of L. ervoides L01-827A than in L. culinaris genotypes. Infection increased the concentration of kaempferol, apigenin, and naringenin in the root tissues of all genotypes, but had no effect on some polyphenols in the low-tannin genotype L. culinaris ZT-4. The concentrations of apigenin, naringenin, apigenin 4-glucoside, naringenin7-rutinoside, diosmetin, and hesperetin 7-rutinoside were higher in the infected root tissues of L. ervoides L01-827A compared with the L. culinaris genotypes. Organic acids including coumaric acid, vanillic acid, 4-aminosalicylic acid, 4-hydroxybenzoic acid, and 3,4-dihydroxybenzoic acid effectively suppressed the in-vitro hyphal growth of A. euteiches. Some of these bioactive polyphenols were more concentrated in roots of L. ervoides L01-827A but were low to undetectable in ZT-4. This study shows that genotypic differences exist in the composition of root polyphenols in lentil, and is related to the response to infection caused by A. euteiches. Polyphenols, particularly the organic acid content could be useful for selection and breeding of lentil genotypes that are resistant to Aphanomyces root rot (ARR) disease.

Keywords: lentil, Lens ervoides, Lens culinaris, polyphenols, Aphanomyces euteiches, root rot, liquid chromatography, mass spectrometry

# INTRODUCTION

Lentil (Lens culinaris Medik.) is an important cool season legume, grown in more than 70 countries around the world. Canada is the largest lentil producer, contributing nearly 41% of the global production (FAO, 2017). Lentil seeds are a rich source of proteins, carbohydrates, vitamins, minerals, fibers and antioxidants (Roy et al., 2010). They also contain non-nutritional compounds

#### Edited by:

Andreia Figueiredo, Universidade de Lisboa, Portugal

#### Reviewed by:

Alessandra Durazzo, Consiglio per la Ricerca in Agricoltura e l'Analisi dell'Economia Agraria (CREA), Italy Galina Suvorova, All-Russia Research Institute of Legumes and Groat Crops, Russia Marisa Porrini, Università degli Studi di Milano, Italy

> \*Correspondence: Navid Bazghaleh navid.bazghaleh@usask.ca

#### Specialty section:

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

> Received: 13 April 2018 Accepted: 13 July 2018 Published: 03 August 2018

#### Citation:

Bazghaleh N, Prashar P, Purves RW and Vandenberg A (2018) Polyphenolic Composition of Lentil Roots in Response to Infection by Aphanomyces euteiches. Front. Plant Sci. 9:1131. doi: 10.3389/fpls.2018.01131

**133**

such as tannins and phytic acid that serve as a defense mechanism against pathogens, insects and parasites (Urbano et al., 2007; Constabel et al., 2014; Sánchez-Chino et al., 2015). Lens species differ in the size of seeds (5–90 mg per seed) and the color and pattern of seed coats, which vary from yellow or gray to dark brown (Ladizinsky, 1979a,b; Ferguson et al., 2000). As a legume crop, lentil provides agro-ecological services by incorporating atmospheric nitrogen into the soil through biological fixation, and promoting the diversity of soil microbial communities (Borrell et al., 2016).

Aphanomyces euteiches is a soil-borne oomycete that causes Aphanomyces root rot (ARR) in many legume crops such as lentil, pea (Pisum sativum L.), alfalfa (Medicago sativa L.), common bean (Phaseolus vulgaris L.), and red clover (Trifolium pratense; Wicker et al., 2001; Chatterton et al., 2016). This pathogen has recently been reported in western Canada (Armstrong-Cho et al., 2014; Chatterton et al., 2016). ARR can cause major crop losses especially under environmental conditions that are conducive to pathogen growth and disease development (McPhee, 2003; Gossen et al., 2016). Susceptible crop plants generally have a limited range of genetic resistance. No effective seed treatment is available for protection against A. euteiches, and its spores are long-lived in soil (Gossen et al., 2016). These characteristics make ARR the most difficult and serious root pathogen of susceptible legumes. Crop rotation and the use of cultivars with partial resistance are the only efficient ways to control ARR (Wicker et al., 2001; Moussart et al., 2013).

Polyphenols are the largest group of secondary metabolites and have very diverse structures (Cheynier, 2012) that make them unique and multifunctional natural products in plants (Quideau et al., 2011). They often have important ecological roles and are involved in a range of functions in plant growth, development, and defense. Polyphenols can attract, repel or protect plants against insects, fungi, bacteria, and viruses (Bennett and Wallsgrove, 1994; Daayf et al., 2012; Olivoto et al., 2017). Synthesis of polyphenols is a complex process that is associated with the shikimate, malonate, and phenylpropanoid pathways (Lattanzio, 2013) and is encoded by multiple genes (Vermerris and Nicholson, 2008; Olivoto et al., 2017). A detailed understanding of how these polyphenols fit into the biochemical pathways for disease resistance is necessary to develop strategies to control plant diseases.

The roots of plants are able to produce a diversity of compounds that directly or indirectly influence microbial species (Lanoue et al., 2010; Baetz and Martinoia, 2014; Bazghaleh et al., 2016). The production of antimicrobial polyphenols have been reported in various plant species (Puupponen-Pimiä et al., 2001; Deng et al., 2015; Shalaby and Horwitz, 2015; Pagliarulo et al., 2016) which could be independent or part of a plant's response to infection. Some polyphenols accumulate in plant tissues even prior to an active defense response (Nicholson and Hammerschmidt, 1992; Balmer et al., 2015). Knowledge of the components of the range of polyphenols in the root tissues and root extracts of crop plants and the potential effect of those compounds on pathogenic microbial species could be used as a first step for identifying plant genotypes that are more resistant to plant diseases (Wink, 1988; Reuveni et al., 1991; Wille et al., 2018).

Polyphenols have been detected in plants roots and their effects on soil pathogens described (Evidente et al., 2010), however, little is known about the composition of these metabolites, particularly in legume plants. This study tested the hypothesis that lentil genotypes have different compositions of polyphenols in their root tissues and root extracts. We also hypothesized that infection by A. euteiches alters the accumulation of specific polyphenols in the root tissues of lentil, and that some of these might inhibit the hyphal growth of A. euteiches. We compared the polyphenol profiles found in root tissues and root extracts of selected genotypes of cultivated (L. culinaris) and wild (Lens ervoides) lentil, including a lowtannin cultivar, using liquid chromatography-mass spectrometry. The effect of the individual polyphenols on the growth of A. euteiches was investigated in vitro. The aim of this study was to identify potential polyphenols that could be involved in providing resistance to ARR in lentil.

## MATERIALS AND METHODS

# Composition of Polyphenols in Root Extracts of Lentil

#### Experimental Design and Plant Growth Conditions

Seeds of the L. culinaris cultivars ZT-4 (low-tannin), CDC Maxim (gray seed coat), Eston (green seed coat), and L. ervoides genotype L01-827A (wild type seed coat) were obtained from the Crop Development Centre at the University of Saskatchewan (Saskatoon, SK, Canada). The lentil genotypes were selected based on differences in the color and polyphenolic profile of their seed coats (Mirali et al., 2016a,b, 2017). The seeds were scarified by puncturing the seed coats using a small blade. The seeds were surface sterilized by successive immersion in 95% ethanol for 30 s, in sterile distilled water for 30 s, in 2.5% Javex <sup>R</sup> bleach solution (sodium hypochlorite) for 2 min, and then in distilled water for 5 min. The seeds were germinated on moist, sterile filter paper in Petri dishes in the dark at 25◦C for 72 h prior to use. Eight germinated seeds of each genotype were transplanted into a 10 cm plastic pot filled with porous ceramic media (Profile <sup>R</sup> Greens GradeTM, BrettYoung, Edmonton, AB). The planting holes in the media were treated with a peat-based Rhizobium leguminosarum inoculant (Nitragin Nitrastick GC <sup>R</sup> , Nitragin Inc., Brookfield, WI, USA). The pots were arranged as a randomized complete block design in four replicates. Plants were grown under 22/16◦C d/night temperatures and 16/8 h d/night length for 30 d in a growth room (GR48, Conviron, Winnipeg, MB) located in the controlled environment facilities of the College of Agriculture and Bioresources at the University of Saskatchewan. Distilled water was added to pots as needed, and plants fertilized weekly with 100 mL of the half strength modified Hoagland's nutrient solution.

**Abbreviations:** ARR, Aphanomyces Root Rot; CDC, Crop Development Centre; LC, Liquid Chromatography; MS, Mass Spectrometry; SRM, Single Reaction Monitoring; ZT, Zero Tannin.

#### Collection of Roots and Extraction of Polyphenols

All plants in each pot were gently removed 30 d after transplanting. The intact roots were collected, weighed, and transferred into 50 mL tubes containing 10 mL 80% MeOH. Polyphenols that had been released from root tissues were extracted by soaking roots in 80% MeOH for 72 h. The tubes were stored at −20◦C until further use. After 3 d, the tube was thawed at room temperature, vortexed, and 1.5 mL was transferred into a 2 mL centrifuge tube, which was centrifuged for 5 min at 12,000 rpm. Then 1 mL of the supernatant was removed and dried using a benchtop centrifugal vacuum concentrator (Labconco, Kansas City, Mo) for approximately 2 h and immediately reconstituted in 200 µL of the recon solution (90:10 H2O:MeOH v/v) to reconstitute the extract.

## Composition of Polyphenols in Root Tissues of Healthy and Diseased Lentil Plants

#### Experimental Design and Plant Growth Conditions

A factorial pot experiment with four lentil genotypes and two infection levels was conducted in a controlled temperature environment. Seeds of the lentil genotypes ZT-4, CDC Maxim, Eston, and L. ervoides L01-827A were scarified, surface sterilized, and germinated as described for experiment 1. Eight germinated seeds of each genotype were transplanted into 10 cm plastic pots filled with peat and vermiculite-based media, Sunshine Mix 3 R (Sun Gro Horticulture Canada Ltd.). The planting holes in the media were treated with a peat-based R. leguminosarum inoculant (Nitragin Nitrastick GC <sup>R</sup> , Nitragin Inc., Brookfield, WI, USA). The pots were arranged in four replicates. Plants were grown under 22/16◦C d/night temperatures and 16/8 h d/night length for 30 d in the growth room (GR48, Conviron, Winnipeg, MB). Distilled water was added to pots as needed, and plants were fertilized weekly with 100 mL of the half strength modified Hoagland's nutrient solution. Roots of the infected plants were inoculated with zoospores of A. euteiches. The experiment was repeated two times.

#### Preparation of Inoculum of Aphanomyces euteiches and Root Infection

Zoospores of A. euteiches were produced following a protocol developed in the Plant Pathology Laboratory, University of Saskatchewan, Canada (Banniza, pers. comm.). After counting the zoospores using a haemocytometer, the final concentration was adjusted to 5 × 10<sup>3</sup> zoospores mL−<sup>1</sup> with sterile deionized water. Ten days after planting, a 5-mL aliquot of the zoospore suspension was injected into the soil near the base of the lentil plants.

#### Collection of Roots and Extraction of Root Polyphenols

Fourteen days after inoculation, the plants were uprooted and the adhering media (Profile <sup>R</sup> Greens GradeTM, Winnipeg, MB) were removed from the roots by soaking the roots in the tap water. The roots were then visually evaluated for infection on the basis of discoloration and root decay. The roots were stored at −80◦C until use.

Polyphenols were extracted from roots of healthy and diseased plants using a procedure similar to that of Mirali et al. (2014) with a few modifications. One hundred fifty milligram of fresh root tissue was placed in 2 mL Sarstedt microtube and then freeze dried over 24 h. The freeze-dried samples were weighed, and then one ¼- inch ceramic bead (MP Bio, Cat. No. 6540-412) was placed in each tube along with 1 mL of an extraction solvent containing 70:30 v/v acetone:water plus internal standards. The root tissues were pulverized using a Mini-Beadbeater-16 (BioSpec Products, Inc. US) for 2 × 2.5 min followed by shaking using an Eppendorf mixer for 1 h at 1,400 rpm and 23◦C. After centrifuging the tubes at 12,000 rpm for 5 min, 500 µL of the supernatant was pipetted into 1.5 mL Eppendorf tubes and centrifuged again at 12,000 rpm for 5 min. From the Eppendorf tube, 200 µL of the supernatant was pipetted into a new 1.5 mL Eppendorf tube and dried down for approximately 2 h using a benchtop centrifugal vacuum concentrator (Labconco, Kansas City, MO). The supernatant was then reconstituted by adding 200 µL of 90:10 MiliQ-water:MeOH to each tube followed by vortexing for 20 sec and shaking on an Eppendorf mixer for 30 min at 1,400 rpm at 23◦C. The tubes were centrifuged at 12,000 rpm for 5 min and then 150 µL of the supernatant was transferred into LC vials and used for LC-MS analysis.

## LC-MS Analysis

Polyphenols were purchased from Sigma-Aldrich (Missouri, USA), and Extrasynthese (Genay, France) except for the deuterated compounds, which were purchased from Toronto Research Compounds (Toronto, Canada).

Root tissues and root extracts of the lentil plants were analyzed for polyphenols using a diode-array detector (DAD) for UV-vis detection and a targeted LC-MS method. Since all polyphenols absorb UV-vis, the DAD detector detects all polyphenols in the sample and the MS gives the m/z of the ion(s) within each peak. The targeted method (list of polyphenols analyzed is given in **Table 1**) uses LC-selective reaction monitoring (SRM) and is based on the method initially developed by Mirali et al. (2014), that has been more recently updated (Mirali et al., 2016a; Purves et al., 2016). Note that catechin-13C3, 4-hydroxybenzoic acid-<sup>13</sup>C7, ferulic acid-d3, resveratrol 4-hydroxyphenyl-13C6, vanillin ring-13C3, and quercetin-d<sup>3</sup> were used as internal standards. The m/z values used for the parent and fragment ions for each of the polyphenols are given in **Table 1**.

Previous chromatographic conditions optimized by Mirali et al. (2016a) were applied with some modifications. An Agilent 1290 UPLC (Agilent Technologies, Santa Clara, CA) equipped with an auto sampler (G4226A), a binary pump (G4220), a thermostatted column compartment (G1316), and a diode array detector (G4212) was used. Compounds were separated using a Phenomenex (Torrance, CA) core-shell Kinetex pentafluorophenyl (PFP) column (100 mm × 2.1 mm, 2.6µm particle size). The mobile phase employed a 30 min binary gradient shown in **Table 2**; solvent A consisted of H2O: CH2O<sup>2</sup> 99:1, and solvent B was ACN:H2O: CH2O<sup>2</sup> 90:9:1 (v/v). The injection volume was 5 µL and the flow rate was 0.35 mL/min. Polyphenols exiting the column were detected by the DAD and then subsequently with a Thermo Fisher TSQ Vantage TABLE 1 | Polyphenols analyzed in the root extracts and root tissues of cultivated lentil genotypes ZT-4, CDC Maxim, Eston, and Lens ervoides L01-827A using liquid chromatography-mass spectrometry.

#### TABLE 1 | Continued



\*Internal standard.

<sup>a</sup>Negative ionization mode.

<sup>b</sup>Positive ionization mode.

triple quadrupole mass spectrometer equipped with a heated electrospray ionization (HESI) interface.

Data were processed by comparing peak areas of polyphenols normalized to the peak area of an internal standard (IS) using Thermo Xcalibur 2.2 software. Each compound (**Table 1**) was quantified using a 4-point calibration curve that was obtained using serial dilutions. Kaempferol dirutinoside and

(Continued)

catechin-3-glucoside were not commercially available, however, their presence has been reported in lentil seeds, and they were examined here based on previous reports (Aguilera et al., 2010; Mirali et al., 2016a). The standard curves of kaempferol 3 rutinoside and catechin were used to estimate the quantities of these compounds, respectively. Concentration values were converted to ng/g based on dry sample weight.

#### Bioassay of the Effect of Polyphenols on Mycelial Growth of Aphanomyces euteiches

The polyphenols found in the root tissues and root extracts were evaluated for their effect on mycelial growth of the pathogen A. euteiches (AE1) isolated from Saskatchewan soils.


Solvent A consists of water: formic acid (99;1, v/v), and solvent B consists of water: acetonitrile: formic acid (9:90:1, v/v/v).

Commercially available polyphenols were used to prepare the stock solutions. A 1 mm well was created in the center of 5% corn meal agar (CMA) plates. An aliquot of 20 µL of 1 g L−<sup>1</sup> of each polyphenol was added to the well.

Control plates received 20 µL of sterilized deionized water and a 1 mm plug of A. euteiches was cut from the margins of an actively growing culture plate and placed in the well. Inoculated plates were incubated in the dark at 22◦C and radial growth was measured daily for up to 5 d. Percentage inhibition or stimulation of radial growth was calculated using the formula:

%age growth inhibition/stimulation = (C − T)/C×100

Where C = growth in the control plate and T = growth in the treated plate (Dickinson and Skidmore, 1976). The experiment was repeated three times with three replicated plates per treatment.

Polyphenols found to be active against the pathogen A. euteiches were selected for further examination of their minimum inhibitory concentration. Using the same protocol described above, three concentrations of the selected polyphenols were examined for their effect on radial growth of the pathogen. Each concentration included a 10 times dilution level. We used a wide range of concentrations for each polyphenol to detect the possible inhibitory effects. The experiment was repeated three times with three replicated plates per treatment.

# Statistical Analysis

Analysis of variance (ANOVA) was performed to test the significance of the effect of genotype, treatments and the interactions between genotype and treatment on the

concentration of polyphenols in roots tissues. The least significant difference (LSD) test was used to determine differences in the concentration of the compounds among genotypes and between the treatments. LSD was also used to test differences among genotypes in the concentration of the compounds in the root extract. A p-value of 0.05 was used as the threshold below which the null hypothesis was rejected. Comparisons of mean values and principal component analysis (PCA) were completed using R package (R, v.3.3.2) (www.r-project.org). A Dunnett's test was used to compare the A. euteiches mycelial inhibition of each polyphenol/concentration with control (P < 0.05).

#### RESULTS

Since all polyphenols absorb in the UV-vis region, the combination of LC-DAD with LC-MS is a powerful approach for investigating polyphenols. **Figure 1** shows a comparison of LC-DAD spectra (UV-vis 250–600 nm) obtained using extracts of (A) healthy roots and (B) diseased root tissues of lentil genotype ZT-4. The figure shows that the most significant changes in the spectra are observed at 9.75 and 12.41 min, with less significant changes being observed at other retention times. The compound at 9.75 min has an m/z of 755 (negative mode MS) and no standard has this m/z and retention time in our targeted LC-SRM method. This compound was observed to have a similar product ion spectrum compared with kaempferol 3-O-rutinoside-4′ -glucoside (m/z 755, tr = 9.6 min included in the targeted method) and therefore is consistent with being a structural isomer of kaempferol 3-O-rutinoside-4 ′ -glucoside. Similarly, the compound at 12.41 min has an m/z of 901 (negative mode) and no standard has this m/z and retention time in our targeted method. This unknown compound was observed to have many of the same product ions as kaempferol dirutinoside (m/z 901, tr = 8.69 min,

analyzed using LC-SRM.

included in the targeted method) and therefore is consistent with being a structural isomer of kaempferol dirutinoside. Thus, the UV-vis spectra not only detected large changes in the profile but also the presence of polyphenols not included in the targeted SRM method. For this study, we are focusing on quantifying polyphenols already present in our library.

#### Biochemical Analysis of Root Extracts

Fourteen polyphenols detected by LC-SRM were common to the root extracts of all 4 genotypes of lentil. Variation in the level of these polyphenols among the 4 genotypes of lentil was shown in **Figure 2**. Lens ervoides L01-827A had elevated amounts of most of the polyphenols in its root extract (**Table 3**).

#### Biochemical Analysis of Healthy and Diseased Root Tissues

The LC-SRM analysis detected 17 polyphenols in the root tissues of all 4 lentil genotypes. The effect of genotype, infection and the interactions between genotype and infection on the concentration of each of these polyphenols was described in **Table 4**. Concentrations of some of the polyphenols varied among the genotypes in healthy plants, whereas in diseased plants, the concentration of most of the polyphenols varied among the genotypes (**Table 5**). Concentration of several compounds varied between diseased and healthy root tissues of the same genotype. The variation observed in the concentration of polyphenols between healthy and diseased roots was larger in L. ervoides L01-827A whereas it was smaller in ZT-4 and CDC Maxim (**Figure 3**). The concentration of apigenin, kaempferol, and naringenin was higher in diseased root tissues for all four lentil genotypes (**Table S1**). Diosmetin, hesperetin 7-rutinoside, kaempferol 3-galactoside, kaempferol 3-rutinoside, naringenin 7-rutinoside, and vanillic acid were only significantly more abundant in diseased roots of L. ervoides L01-827A (**Table S1**). Coumaric acid was only detected in diseased roots of L. ervoides L01-827A and Eston. 3,4-dihydroxybenzoic acid was detected in

TABLE 4 | Statistical significance of the effect<sup>a</sup> of genotype and infection on the composition of some phenolic compounds in the root tissues of cultivated lentil genotypes ZT-4, CDC Maxim, Eston, and Lens ervoides L01-827A determined by liquid chromatography-mass spectrometry (n = 4).


<sup>a</sup>Significant effects were detected using analysis of variance (ANOVA) (n = 4).

TABLE 3 | Concentration of phenolic compounds (ng/g root) detected in the root extracts of cultivated lentil genotypes ZT-4, CDC Maxim and Eston and Lens ervoides L01-827A using liquid chromatography-mass spectrometry.


<sup>1</sup>Standard deviation.

<sup>2</sup>Significant differences were detected using a Least Significant Difference (LSD) test (n = 4) (p < 0.05). Different letters indicate significant differences between means.

TABLE 5 | Concentration of phenolic compounds (µg g−<sup>1</sup> root) detected in the root tissues of cultivated lentil genotypes ZT-4, CDC Maxim and Eston and Lens ervoides L01-827A using liquid chromatography-mass spectrometry.


<sup>1</sup>Significant differences were detected using a Least Significant Difference (LSD) test (n = 4) (p < 0.05). Different letters indicate significant differences between means. <sup>2</sup>Due to a co-eluting interference in the SRM channel, the value of vanillic acid was estimated by LC-UV.

diseased roots of all and in healthy roots of L. ervoides L01-827A (**Table S2**).

compounds at 0.1 g L−<sup>1</sup> (10x dilution), or at lower concentrations of any of these polyphenols (**Table 6**).

#### In vitro Assay Testing the Effect of Polyphenols on Mycelial Growth of Aphanomyces euteiches

Seven polyphenols including vanillic acid, 4-hydroxybenzoic acid, 3,4-dihydroxybenzoic acid, 4-aminosalicylic acid, coumaric acid, delphinidin 3-glucoside and phloretin inhibited the mycelial growth of A. euteiches at 1 mg mL−<sup>1</sup> in vitro (p < 0.0001). Coumaric acid and phloretin also inhibited A. euteiches at 0.1 g L−<sup>1</sup> (10x dilution). No inhibition was observed for other

# DISCUSSION

We used semi-quantitative mass spectrometry to investigate genotypic variations within the composition of polyphenols that naturally exist in the root tissues and root extracts of lentil. We analyzed a wide range of polyphenols and demonstrated that the qualitative and quantitative composition of polyphenols not only differs among the lentil genotypes, but also between the healthy and A. euteiches infected roots.

The higher concentration of polyphenols in the root extracts of the L. ervoides genotype (tertiary gene pool) compared to the cultivated L. culinaris genotypes (primary gene pool) could be an evolutionary divergence or a consequence of selection for larger plants and seeds during the domestication process (Alo et al., 2011; Wong et al., 2015). It was shown that the selection of new cultivars for specific traits or under optimal growth conditions may inadvertently lead to the loss of specific genes and phytochemicals (Hättenschwiler and Vitousek, 2000; Dixon, 2001) including those involved in the synthesis of polyphenols.

Lentil genotype ZT-4 is characterized by the expression of the single recessive gene, tan, that affects the color and thickness of the seed coat (Vaillancourt et al., 1986). The seed coats of lowtannin lentils lack many of the polyphenols found in normal seed coats, causing them to be thinner and more fragile compared to the seed coat of normal genotypes (Matus et al., 1993; Mirali et al., 2016a). This study shows that the tan gene is also expressed in the root tissues and results in a lower concentration of most of the polyphenols.

Variations observed in the levels of specific polyphenols between healthy and diseased root tissues may be due to defensive responses through the inactivation of microbial enzymes, reinforcement of plant structural components or creation of strategic barriers at points of entry (Sylvia and Sinclair, 1983; Beckman, 2000; Vermerris and Nicholson, 2008). In this study, the elevated concentration of apigenin, kaempferol, and naringenin in the diseased roots of all four lentil genotypes (**Table S1**) may suggest that these three polyphenols are associated with a general response against A. euteiches. TABLE 6 | Inhibitory effect of bioactive polyphenols<sup>a</sup> detected in lentil root tissues on the in vitro<sup>b</sup> radial growth of Aphanomyces euteiches<sup>c</sup> .


<sup>a</sup>20 µL of phenolic compound was applied.

<sup>b</sup>Corn meal agar (CMA) medium (5%).

<sup>c</sup>Mycelial growth was measured 5 d after inoculation.

<sup>d</sup>Values represent percent inhibition of radial growth of Aphanomyces euteiches compared to control.

<sup>e</sup>Significant differences between treatments and control were detected using Dunnett's test (n = 3) (\*\*\*p < 0.001, \*p < 0.05).

<sup>f</sup> Not significant.

Support for this hypothesis can be found in previous studies in which naringenin and apigenin were shown to have inhibitory effects on the bacterial pathogens and nematodes (Baidez et al., 2006; Treutter, 2006). Since such inhibitory effect was not detected against A. euteiches in our bioassay, these polyphenols might have other potential roles in a response against A. euteiches. Similar levels of several compounds in healthy and diseased roots of the low-tannin genotype ZT-4 may suggest that the expression of the gene tan is one of the factors influencing the composition of polyphenols in responses to phytopathogens. The differences observed in the concentration of polyphenols between healthy and diseased roots in different genotypes (**Tables 3**, **4**) may indicate that the synthesis of defensive metabolites is genotype dependent. The variation was larger in L. ervoides L01-827A (**Figure 3**), suggesting that this wild genotype might have a wider capacity for a defensive response to plant pathogens.

In this study, the lentil genotypes were treated by the endosymbiont R. leguminosarum. It is notable that some polyphenols act as chemoattractants for rhizobium, improving the root colonization (Fisher and Long, 1992; Begum et al., 2001; Cooper, 2004) and, rhizobium may aid the plant to resist against pathogen by mediating the polyphenolic content of root tissues (Mishra et al., 2006). However, rhizobium may interact differently with the lentil genotypes resulting in different degrees of root colonization, and different responses to plant pathogens.

Certain phytochemicals were shown to attract the zoospores or inhibit the hyphal growth of A. euteiches (Yokosawa et al., 1986; Smolinska et al., 1997; Cannesan et al., 2012). In our study, the first involving lentil, some of the root polyphenols that inhibited the hyphal growth of the A. euteiches were more concentrated in specific lentil genotypes. For instance, the concentration of vanillic acid and 3,4-dihydroxybenzoic acid were higher in L. ervoides L01-827A which has a marbled gray seed coat. Coumaric acid was detected only in diseased roots of L. ervoides L01-827A, and in the green seed coat of the cultivar Eston (Mirali et al., 2016b). Differences in the quantity and quality of the phenylpropanoid metabolites in different genotypes of lentil suggests there could be useful genotypic variation in the level of resistance to A. euteiches in lentil. Since the majority of these A. euteiches inhibitors were organic acids, the organic acid content of root tissues and extracts appears to be associated with some level of resistance to ARR. Organic acids comprise a major portion of root extracts in most plant species (Neumann and Römheld, 2001; Cawthray, 2003; Sandnes et al., 2005). Organic acids and their derivatives were detected in both root tissue and root extracts of lentil. The observation that some organic acids, including caffeic acid, gallic acid, and 3-hydroxy-4-methoxycinnamic acid, did not affect the hyphal growth of A. euteiches suggests that the inhibitory effect on A. euteiches is limited to specific organic acids. The concentrations of root polyphenols could be variable in real conditions depending on various factors. However, an accumulative concentration of the bioactive polyphenols could be achieved and be effective to inhibit the root rot pathogen. Polyphenols, particularly those released from roots, are involved in a wide range of agroecological processes, and could affect specific ecological niches in the rhizosphere (Aira et al., 2010; Doornbos et al., 2012). The selective release of organic acids and other polyphenols that inhibit A. euteiches may improve soil health by reducing the abundance of the pathogen in soil, and the severity of ARR in subsequent susceptible crops in rotation.

To date there is no report of detection of effective resistance to A. euteiches in lentil. Among the genotypes used in this study, ZT4 was more susceptible to A. euteiches (Knowpulse Report, 2015), while L01-827A and Eston have shown a moderate level of resistance to A. euteiches (Banniza, pers. comm.; Saskpulse, 2015). One reason for their better tolerance to A. euteiches could be the elevated levels of specific of polyphenols. Such variations could increase the understanding of pathways of biochemical resistance to plant diseases and could be used as a starting point for identification of lentil genotypes that are resistant to root rot diseases in breeding programs.

Our study revealed that genotypic variations for the composition of polyphenols exist in lentil, and that there is a relationship between the composition of polyphenols and tolerance to A. euteiches. The tan gene in lentil influences the composition of polyphenols in root extracts. A number of studies have described the potential role of polyphenols in providing resistance against root rot diseases in different plant species (Farkas and Kiraaly, 1962; Lagrimini et al., 1993; Daayf et al., 2012; Aksoya et al., 2017). Such studies usually focus on a small number of compounds and thus cannot provide a clear image of the function of polyphenols. Here, we analyzed a wide range

#### REFERENCES


of polyphenols in response to root rot disease. Understanding of the role of polyphenols in providing resistance against root rot diseases, and their potential application in breeding programs is still not well defined. In lentil breeding programs, the exploitation of polyphenol content has mainly targeted the seed coats in order to improve seed quality, storage, marketability and nutritional values (Vaillancourt et al., 1986; Mirali et al., 2016b, 2017; Ganesan and Xu, 2017). Further research is required to identify and quantify all the polyphenols that are differentially expressed in uninfected and infected root tissues of lentil species. It is necessary to detect correlations between the concentration of specific polyphenols and other metabolites in the root tissues in order to identify mechanisms of resistance to A. euteiches. The induction of resistance in host plants can be influenced by a number of factors, including plant genotype, nutrition and environmental conditions (Walters et al., 2013). Therefore, it is necessary to understand the influence of plant nutrition on the polyphenolic content of roots, since this may override the influence of plant genetics on the composition of polyphenols. We are currently investigating unknown peaks in the UV spectra to identify new compounds of importance for roots that will be added to the polyphenol library in future studies.

# AUTHOR CONTRIBUTIONS

NB grew plants and collected the root materials. NB and RP contributed to LCMS analyses. PP prepared the inoculum and conducted the in-vitro plate assay. All authors contributed to experimental design and to preparation of the manuscript.

# ACKNOWLEDGMENTS

The authors acknowledge financial assistance from the NSERC Industrial Research Chair program, and Saskatchewan Pulse Growers. We also acknowledge the assistance provided by Thermo Fisher Scientific (San Jose, CA) through instrumentation made available as part of a collaboration between Thermo Fisher and the University of Saskatchewan. Additional support was provided by the University of Saskatchewan Health Sciences core mass spectrometry facility, and by the Pulse Research Crew at the Crop Development Centre, University of Saskatchewan.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2018. 01131/full#supplementary-material

maize rhizosphere microbial communities. Soil Biol. Biochem. 42, 2276–2281. doi: 10.1016/j.soilbio.2010.08.029


and phylogeny in wild and domesticated lentil. J Hered. 102, 315–329. doi: 10.1093/jhered/esr015


in the rhizosphere. A review. Agron. Sustain. Dev. 32, 227–243. doi: 10.1007/s13593-011-0028-y


Substances at the Soil–Plant Interface, eds R. Pinton, Z. Varanini, and P. Nannipieri (New York, NY: Dekker), 23–72.


the ectomycorrhizal fungus Laccaria laccata. Phytopathology 73, 390–397. doi: 10.1094/Phyto-73-390


**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 Bazghaleh, Prashar, Purves and Vandenberg. 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) and the copyright owner(s) 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.

# Phenolic Profile and Susceptibility to Fusarium Infection of Pigmented Maize Cultivars

Jamila Bernardi<sup>1</sup> , Lorenzo Stagnati<sup>1</sup> , Luigi Lucini2,3 \*, Gabriele Rocchetti<sup>4</sup> , Alessandra Lanubile<sup>1</sup> , Carolina Cortellini<sup>5</sup> , Giovanni De Poli<sup>5</sup> , Matteo Busconi1,3 and Adriano Marocco1,3 \*

<sup>1</sup> Department of Sustainable Crop Production, Università Cattolica del Sacro Cuore, Piacenza, Italy, <sup>2</sup> Department for Sustainable Food Process, Università Cattolica del Sacro Cuore, Piacenza, Italy, <sup>3</sup> Research Centre for Biodiversity and Ancient DNA, Università Cattolica del Sacro Cuore, Piacenza, Italy, <sup>4</sup> Department of Animal Science, Food and Nutrition, Università Cattolica del Sacro Cuore, Piacenza, Italy, <sup>5</sup> Terra Srl, Cremona, Italy

#### Edited by:

Andreia Figueiredo, Universidade de Lisboa, Portugal

#### Reviewed by:

Paula Branquinho Andrade, Universidade do Porto, Portugal Wieslaw Wiczkowski, Institute of Animal Reproduction and Food Research (PAN), Poland Andrei Mocan, Iuliu Ha¸tieganu University of Medicine and Pharmacy, Romania

#### \*Correspondence:

Luigi Lucini luigi.lucini@unicatt.it Adriano Marocco adriano.marocco@unicatt.it

#### Specialty section:

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

Received: 19 February 2018 Accepted: 25 July 2018 Published: 14 August 2018

#### Citation:

Bernardi J, Stagnati L, Lucini L, Rocchetti G, Lanubile A, Cortellini C, De Poli G, Busconi M and Marocco A (2018) Phenolic Profile and Susceptibility to Fusarium Infection of Pigmented Maize Cultivars. Front. Plant Sci. 9:1189. doi: 10.3389/fpls.2018.01189 Maize is a staple food source in the world, whose ancient varieties or landraces are receiving a growing attention. In this work, two Italian maize cultivars with pigmented kernels and one inbred line were investigated for untargeted phenolic profile, in vitro antioxidant capacity and resistance to Fusarium verticillioides infection. "Rostrato Rosso" was the richest in anthocyanins whilst phenolic acids were the second class in abundance, with comparable values detected between cultivars. Tyrosol equivalents were also the highest in "Rostrato Rosso" (822.4 mg kg−<sup>1</sup> ). Coherently, "Rostrato Rosso" was highly resistant to fungal penetration and diffusion. These preliminary findings might help in breeding programs, aiming to develop maize lines more resistant to infections and with improved nutraceutical value.

Keywords: flavonoids, Fusarium, metabolomics, anthocyanins, FER, phytoalexins

# INTRODUCTION

Maize is the most cultivated cereal grain throughout the world, considering both yield and harvested area (Organization for Economic Cooperation and Development-OECD and FAO, 2015). Maize is a staple crop in the African region and South America, while in developed countries is used mainly to feed livestock as forage, silage and grain rather than as biofuel and for industrial uses. A new study by FAO and OECD estimates that global consumption of cereals will increase by 390 million tons between 2014 and 2024. The core driver of the increase will be the rising demand for animal feed, of which about 70% is maize, accounting for more than half of the total (FAO, 2014).

Endosperm is the main storage tissue in maize kernel, accumulating carbohydrates, such as starch (90–95%), and storage proteins, such as prolamins (10–12%) (Wu and Messing, 2014). Considering its nutritional value, other important components in maize kernels are: carotenoids, flavonoids and hydroxycinnamic acids. Carotenoids are the common pigments in maize, insoluble compounds that accumulate in the endosperm and confer the typical orange color. Most of the cultivated maize has yellow kernels, but some varieties possess the ability to pigment different tissues (i.e., anthers and roots), especially in response to stresses. Phenolic compounds such as flavonoids, are responsible for the red, purple, blue and black coloration of kernels and other parts of the plant. In maize seed, accumulation of pigments can occur in two tissues, namely the pericarp, the maternal-derived tissue, rather than the aleurone that is the peripheral part of

the endosperm. Anthocyanins are water-soluble flavonoids that accumulate in vacuoles of the aleurone. Other colored flavonoids in maize are the red pigments phlobaphenes, polymers of the flavan-4-ols apiforol and luteoforol that accumulate in the seed pericarp and cob glumes of maize (Sharma et al., 2012). Flavonoids, like other phenolics, have the power to protect the kernel from biotic and abiotic stresses, and have been associated with the beneficial effects of the Mediterranean diet, given their potent antiangiogenic, anti-inflammatory, and anticarcinogenic activities (Petroni and Tonelli, 2011). In rats fed with anthocyanin-rich maize, the amount of cardiac tissue that was damaged following ischemic conditions was reduced by approximately 30% compared to rats fed with anthocyanin-free maize (Toufektsian et al., 2011). Anthocyanins from purple corn also prevent weight gain and obesity in mice under high fat diet, and they can reduce severe diabetic complications (Tsuda, 2012). Phenolic compounds that accumulate in the maize endosperm and pericarp may contribute to resistance against insect damages, Fusarium ear rot (FER) and fumonisin contamination (Sharma et al., 2012; Atanasova-Penichon et al., 2016). Indeed, flavonoids could act as physical impediment against fungal attack (in particular when accumulated in the pericarp) by hardening maize kernel thus reducing the spread of mycelium in the inner parts of the seeds (Atanasova-Penichon et al., 2016). Other flavonoids are known to reduce insect attack like the flavone maysin (a C-glycosyl luteolin derivative) that can decrease the growth of earworm larvae in maize (Sharma et al., 2012).

In the last years, there was a growing demand of consumers in increasing the phenolic content of vegetables by rediscovering ancient cultivars that possess a natural pigmentation, because they are expected to have higher nutritional value (Falcone Ferreyra et al., 2012; Casas et al., 2014). With this regard, metabolomics has been proposed as a powerful tool to achieve a comprehensive picture of the phenolic signature in crop foods (Rocchetti et al., 2017, 2018a). Nonetheless, the actual phenolic profile is also supposed to play a range of physiological roles in plant, including protection toward both abiotic (e.g., UV radiation or oxidative stress via radical scavenging) and biotic stresses (Nakabayashi and Saito, 2015; Talhaoui et al., 2015). Extensive literature can be found on phenolics in nonpigmented maize, whereas most of the work on pigmented maize referred to blue genotypes from Mexico. However, redpurple genotypes received limited attention to date, although they have been recently reported to possess a favorable nutritional profile (Rocchetti et al., 2018b). Furthermore, previous works investigated phenolic profile of maize through targeted approaches, and therefore they might have not comprehensively screened the whole profile, including eventual conjugates and glycosylated compounds.

Therefore, the aim of this work was to study some maize genotypes featured by red-purple kernel pigmentation, according to their field performance, phenolic profile, in vitro antioxidant capacity and resistance to Fusarium infection. Assumed that phenolics alone are not the unique component in the resistance to Fusarium infection, it becomes relevant to test to which extent these compounds can interfere with fungal spread. On these bases, the phenolic profiles and resistance to Fusarium of three genotypes were compared to a non-pigmented commercial maize hybrid. In fact, these genotypes received less attention than other pigmented maize genotypes, even they might have a double attitude both in disease resistance and functional ingredients. A recent study on starch digestibility after cooking highlighted that the three genotypes "Nostrano della Val di Non" and "Rostrato Rosso," and a "Purple B73" possessed distinctive and diverse pigmentation patterns that could be linked to the modulation of starch digestibility (Rocchetti et al., 2018b). On these bases, the ultimate aim of this work was to investigate the potential exploitation of the above-mentioned cultivars in breeding programs rather than gaining information on their viability as functional foods.

# MATERIALS AND METHODS

#### Maize Germplasm

Two pigmented cultivars "Nostrano della Val di Non" and "Rostrato Rosso", and a "Purple B73" line, were provided by ISTA, Agroalimentare Sud S.p.A. (Lodi, Italy). The yellow maize hybrid Agrister (Limagrain, Saint-Beauzire- France) and the B73 inbred line (available as our stock) were used as non-pigmented control (**Figures 1Aa–Dd**). The former non-pigmented genotype was used as reference for field evaluation, phenolic profile and antioxidant capacity investigations. However, the latter was a fungicide-free inbred line to be used as control for both in vitro and field infection assays.

# Field Evaluation

The maize genotypes were sown on April 2015 in randomized blocks, with four replicates per genotype, including the commercial hybrid Agrister, in the experimental field located in San Damiano (PC), Italy (44◦ 540 10.700N, 9◦ 410 25.600E). Plots consisted of six rows 5 m long and spaced 0.75 m apart, planting 25 seeds per row. Each plot was spaced apart by four rows of the commercial hybrid Agrister and by a 5 m block of hybrid on the edge. Standard agricultural practices were followed. Phenotypic evaluation was performed according to the UPOV characters (ASFIS, 1992). Ears were harvested at maturity, kernels were sampled and kept at −20◦C for further analyses on phenolic profile and antioxidant capacity.

A further experiment was carried out on April 2017, using the same genotypes, to investigate resistance to Fusarium. The field trials were located at CERZOO "Centro Di Ricerche Per La Zootecnia E L0Ambiente S.C.R.L." facilities (San Bonico 45.003624N, 9.705179E, Piacenza, Italy). Plot scheme and agricultural practices were the same as for previous trials. The "Rostrato Rosso" and "Nostrano della Val di Non" plants derived from a second-cycle of selfing.

#### Resistance to Artificial Infection

In the field trial, plants were hand self-pollinated and at 15 days after pollination ears were artificially inoculated with a spore suspension of F. verticillioides ITEM10027 (MPVP 294, 1 × 10<sup>6</sup> spores per mL) according to the pin-bar method (Maschietto et al., 2017). Ears were harvested at maturity and phenotypically

evaluated for FER severity using a seven-point scale (Maschietto et al., 2017). Ears not inoculated were harvested at maturity and kernels used for the further investigation through in vitro infection, according to the rolled towel assay (RTA) (Lanubile et al., 2015). With this aim, seeds with similar size and shape, preferably flattened and without visible damage, were selected from each maize genotype. To minimize at most the presence of contaminating fungi, the seeds were surface-sterilized by shaking them in 50-mL tubes at room temperature with 70% ethanol for 5 min, washed by sterilized bi-distilled water for 1 min, then with commercial bleach solution for 10 min, and finally rinsed three times with sterile distilled water for 5 min each time (Lanubile et al., 2015). Two towels of germination paper (Anchor Paper, Saint Paul, MN, United States) were moistened with sterilized distilled water. For each genotype, 20 seeds were placed on the germination paper about 10 cm from the top of the towel with the embryo side facing out. Kernels were inoculated on the embryo side near the pedicel with 100 µL of a 1 × 10<sup>6</sup> spores per ml suspension of F. verticillioides ITEM10027 (MPVP 294) and covered with another moistened towel. Towels were rolled up and arranged vertically in a 25-L bucket covered with a black plastic bag and kept for 7 days at 25◦C in the dark. For each genotype, a control RTA was prepared as previously described, but avoiding the inoculation step.

Seedlings were rated using a five-point severity scale adapted from previous research on soybean seedlings (Lanubile et al., 2015). According to this scale, the scores were as follows: 1 = healthy, germinated seedlings with no visible signs of colonization; 2 = germination and colonization of the kernel near the pedicel; 3 = germination with widespread colonization of the kernel and browning of the coleoptile; 4 = germination with reduced seedling development, complete colonization of the kernel, and lesions and abundant mold on the shoot, 5 = no germination due to complete rotting of the kernel. Severity after inoculation (SEV\_I) and in the control assays (SEV\_C) were recorded for each genotype.

#### Extraction of Phenolic Compounds

The maize samples were randomly collected from four ears within each plot of the same genotype, obtaining six replicates. Kernels were grinded in a laboratory mill equipped with a 1-mm screen. Thereafter, 10 ml of a hydro-alcoholic solution (80% methanol, acidified with 0.1% formic acid, v/v) was used to extract phenolic compounds from 1 g of each replicate, by means of an Ultra-Turrax (25,000 rpm for 3 min). Suspensions were centrifuged 6,000 × g, and then 5% trichoroacetic acid (TCA) was added to the liquid phase to precipitate proteins. The solutions were stored overnight in freezer, at −18◦C, filtered by means of a 0.22 µm syringe cellulose filter, and stored in dark vials at −18◦C until further analyses. Six individual replicates were extracted from each genotype.

# UHPLC-ESI/QTOF Screening of Phenolic Compounds

The comprehensive phenolic profile of maize samples was investigated through an untargeted metabolomic approach, using ultra-high-pressure liquid chromatography (UHPLC) in combination to a quadrupole-time-of-flight (QTOF) mass spectrometer. In more detail, the instrumentation consisted of a

1290 UHPLC coupled to a G6550 mass spectrometer (all from Agilent Technologies, Santa Clara, CA, United States) via a JetStream dual electrospray ionization source. Chromatographic and mass spectrometric instrumental conditions to screen phenolic compounds in the samples were optimized in previous works (Borgognone et al., 2016; Rocchetti et al., 2017). Briefly, a Knauer BlueOrchid C18 column (100 mm × 2 mm, 1.8 µm) was used for chromatographic separation, using a binary mixture of methanol and water as mobile phase (LCMS grade, VWR, Milan, Italy). The gradient elution was designed to increase methanol from 5 to 95% within 34 min with a flow rate of 220 µL min−<sup>1</sup> and a 3.5 µL injection volume. The mass spectrometer was set up to positive SCAN mode, detecting masses in the range 100–1000 m/z. Each extract was injected once, as single instrumental replicate.

The software Profinder B.07 (from Agilent Technologies) was used to elaborate raw data, and polyphenols annotation was carried out using the database Phenol-Explorer 3.6 (Rothwell et al., 2013), considering the whole isotopic pattern. The "find-by-formula" algorithm, that includes monoisotopic mass, isotopes spacing and ratio, was used for annotation (mass accuracy tolerance < 5 ppm). Thereafter, mass and retention time alignment and filter-by-frequency (features not present in at least one treatment in 100% of replications were discarded) were applied.

In order to provide quantitative information, phenolic compounds were firstly ascribed into classes and sub-classes, and then cumulative intensity for each phenolic sub-class were converted in mg kg−<sup>1</sup> equivalents, by means of calibration curves from nine polyphenol standards (from Extrasynthese, Lyon, France). Furofuran lignans were quantified as sesamin, dibenzylbutyrolactone lignans as matairesinol, phenolic acids as ferulic acid, anthocyanins as cyanidin, tyrosols and low-molecular-weight phenolics as tyrosol, alkylresorcinols as 5-pentadecylresorcinol (also known as cardol), stilbenes as resveratrol, flavanols and flavonols as catechin, and flavones as luteolin equivalents. Calibration curves were built using a linear fitting (un-weighted and not forced to axis-origin) in the range 0.05–500 mg L−<sup>1</sup> ; a coefficient of determination R <sup>2</sup> > 0.97 was used as acceptability threshold for calibration purposes.

#### In vitro Antioxidant Capacity Assays

Antioxidant capacity assays were carried out on the same extracts used for phenolic profiling. The in vitro antioxidant capacity of each maize sample was evaluated as radical scavenging ability (DPPH assay) and ferric reducing power (FRAP assay), as previously described (Ghisoni et al., 2017). Briefly, 1.5 mL of maize extract was combined to the same volume of an ethanol solution of DPPH (1.0 mM). The absorbance was recorded at 5-min intervals (until the steady state) to 517 nm using a Perkin Elmer Lambda 12 spectrophotometer (Ontario, Canada). The results were finally expressed as gallic acid equivalents (GAE).

The FRAP antioxidant assay was carried out by means of a clinical analyzer ILAB 600 (Instrumentation Laboratory, Lexington, MA, United States). The FRAP working reagent consist of acetate buffer (300 mM, pH 3.6), TPTZ (10 mM) in 40 mM HCl and FeCl3 (20 mM), in the ratio 10:1:1 (v/v). Each extract (100 µL) was combined to 3 ml of FRAP working reagent, and the absorbance was recorded at λ = 600 nm, after 243 s of incubation at 37◦C. The FRAP results were expressed as GAE.

#### Statistical Analysis

The analysis of variance for the agronomic traits and the statistical analysis of artificially infected samples was performed using the R software. Phenotypic values collected in the RTA experiment were square-root transformed and mean values of severity were calculated. FER phenotypes, scored as percentages, were arccosine transformed before performing statistical analysis. The Kruskal–Wallis one-way analysis of variance and the Kruskal–Dunn post hoc test, available in the R package PMCMR (Pohlert, 2014) were applied to detect significant differences between maize accessions tested in the RTA experiment; the Welch one-way ANOVA and the Games–Howell post hoc test, available in the R packages one waytes (Dag et al., 2017) and user friendly science (Peters, 2017) were applied for FER field evaluation.

Analysis of variance for in vitro antioxidant capacity (one-way ANOVA, P < 0.05) and correlations between antioxidant capacity and concentration of different phenolic classes (Pearson, two-tails) were carried out in IBM SPSS statistics 24.

Metabolomics data were elaborated by means of Agilent Mass Profiler Professional B.12.06 software, as previously described (Rocchetti et al., 2017). Abundance of phenolic compounds was normalized at the 75th percentile and corrected for the corresponding median in all samples. Volcano plots, carried out combining ANOVA (p < 0.01, Bonferroni multiple testing correction) and fold-change (FC) analysis (cut-off = 5), and Venn diagrams were also generated. The not-supervised statistical approach Hierarchical Cluster Analysis was then carried out as previously described (Rocchetti et al., 2017). Finally, the raw data were elaborated into SIMCA 13 (Umetrics, Malmo, Sweden) for the supervised statistical approach orthogonal projection to latent structures discriminant analysis (OPLS-DA). A confidence limit of 95 and 99% was used to investigate the presence of outliers in the model (according to Hotelling's T2 approach), while cross validation (CV-ANOVA, p < 0.01) together with a permutation test (N = 100) to exclude overfitting, were then carried out. The goodness-of-fit and prediction ability of the model (i.e., R <sup>2</sup>Y and Q <sup>2</sup>Y, respectively) were also investigated, adopting cut-off values provided in literature (Rombouts et al., 2017). Finally, the VIP analysis was carried out to investigate the variable's importance in projection, i.e., considering those phenolic metabolites with the highest discrimination potential (VIP score > 1.2).

# RESULTS

#### Phenotyping of Pigmented Maize Genotypes

Three colored genotypes ("Nostrano della Val di Non," "Rostrato Rosso" and "Purple B73") and a yellow hybrid used as reference ("Agrister"), were characterized both for vegetative and reproductive traits. Agronomic measurements of the main traits

are summarized in **Table 1** whereas all the parameters are listed in **Table 2**.

#### "Purple B73"

"Purple B73" is a medium-short inbred line; given this, plants are phenotypically uniform, with a very strong purple pigmentation of the stalk, leaves, glumes, bracts and tassel while silks and anthers are white (**Table 2**). Ears are classified as short of 15–18 cm with violet – deep-gray kernels arranged in 16 rows (**Figure 1Aa**). This inbred line reached silking at 804 GDD and physiological maturity at 1327 GDD (**Table 2**). The purple corn is resistant to stalk lodging and the percentage of barren plants was around 30%.

#### "Rostrato Rosso"

The "Rostrato Rosso" plants are medium-high tall with a short cycle, silking was at 709 GDD and physiological maturity at 1291 GDD. The variety is variable for anthocyanin pigmentation of the leaf sheath that can be deep green or with purple strikes, also the tassel attitude may vary between erect or pendulous (**Table 2**). Ears are morphologically uniform with flint violet–black kernels with a pronounced rostrum (**Figure 1Bb**). The ear is slightly conical and longer than 20 cm, kernels are arranged in 12–14 rows. Forty-one percent of plants were earless and lodging was noted for this variety. Lodging can be a consequence of the very high ratio of ear insertion related to plant height (**Table 2**).

#### "Nostrano della Val di Non"

The "Nostrano della Val di Non," hereafter called "Val di Non" has a low-medium height and reached mid silk 1 week before the other genotypes getting the physiological maturity at 1234 growing-degree days (GDD; **Table 2**). Neither leaves nor the stem are pigmented while silks and kernels are pigmented (**Table 2**). The kernel pigmentation can vary from deep orange to dark red (**Figure 1Cc**). Variability has been observed in the anthocyanin pigmentation of tassel glumes that could be violet, red or intermediate colors. Nonetheless the number of barren plants was quite high (62%) the production was good since this plant did not present lodging symptoms (**Table 1**).

## Phenolic Profile of Pigmented and Yellow Genotypes

The phenolic profile in pigmented genotypes (i.e., "Purple B73," "Rostrato Rosso," and "Val di Non") and in non-pigmented references (i.e., "Agrister" and "B73 line") was investigated using an untargeted metabolomics approach based on UHPLC-ESI/QTOF mass spectrometry. Overall, phenolic profile was diverse, with flavonoids being the most frequently detected class detected (152 annotated compounds: 46 anthocyanins, 48 flavanols and 58 flavones), followed by phenolic acids (55 compounds), tyrosols (43 compounds), alkylphenols (14 compounds) and other phenolics (20 lignans and 6 stilbenes). All the comprehensive information regarding phenolic compounds identified across the different maize samples are provided as supporting information (**Supplementary Tables S2**, **S3**), including annotations (raw formula, identification scores) and composite spectra (masses and abundances).

The Agrister yellow maize was used as control for both phenolic profiling and related antioxidant capacity, since this former was grown together with pigmented genotypes, i.e., under the same pedoclimatic conditions. Unsupervised hierarchical cluster analysis was then produced considering the foldchange heat map, highlighting a substantial change of the phenolic profile moving from the Agrister yellow maize (control) toward the pigmented genotypes (**Figure 2**). The output of the heat map showed two main clusters; the first one was represented by the line "Purple B73," while the second cluster included "Rostrato Rosso" and "Val di Non," together with the control. However, this latter showed a distinct phenolic profile, when compared to pigmented varieties, being in a separate sub-cluster. Nevertheless, the heat map of the hierarchical cluster analysis clearly showed that the abundance of several phenolics tends to disappear moving from yellow to pigmented varieties.

Subsequently, the Venn diagrams were carried out in order to shed light on differentially and common phenolic compounds, considering pigmented varieties and Agrister yellow maize (**Figure 3**). Overall, the output of Venn diagrams showed that the three pigmented genotypes had 30 common phenolic compounds. Nevertheless, "Val di Non" possessed 8 exclusive compounds (**Supplementary Table S1**), being above all phenolic acids (coumaric acid, coumaric acid 4-O-glucoside, and hydroxycaffeic acid), while "Rostrato Rosso" and "Purple B73" reported 22 and 17 exclusive phenolics, respectively. Interestingly, among phenolic compounds characterizing the "Rostrato Rosso," several cyanidin-derivatives forms were detected, i.e., pelargonidin 3-O-glucosyl-rutinoside/sophoroside,


Within the same column, means followed by the same letter are not significantly different at P ≤ 0.05 according to Fisher's test.

TABLE 2 | Morphological and physiological traits of all the genotypes according to the UPOV characters.


<sup>a</sup>VdN, Nostrano della Val di Non; <sup>b</sup>RR, Rostrato Rosso.

cyanidin 3,5-O-diglucoside and cyanidin 3-O-sambubioside, delphinidin 3-O-xyloside, and malvidin 3-O-(600-acetylgalactoside). "Purple B73" showed an abundance of flavonoids as exclusive compounds, being characterized by flavones, flavonols and anthocyanins.

Considering specific compounds, apiforol that is a precursor of phlobaphenes was up accumulated in "Val di Non" and "Rostrato Rosso" respect to both "Purple B73" and yellow maize. This evidence suggests that these two cultivars could accumulate also phlobaphenes in kernels, additionally to the other phenolics detected. Furthermore, the compound maysin was up accumulated in "Rostrato Rosso" respect to other pigmented genotypes.

Starting from these differences, the relative abundance of each phenolic subclass was investigated according to the curves from the respective phenolic standards (**Figure 4A** and **Table 3**). Remarkably, the flavonoid profile was very explicative, allowing to clearly discriminate among the four maize genotypes; "Rostrato Rosso" and "Purple B73" were the richest in anthocyanins, being 4399.4 and 3167.9 mg kg−<sup>1</sup> , respectively, while "Val di Non" and the yellow maize (used as control) showed the lowest values in anthocyanin equivalents (752.5 and 205.4 mg kg−<sup>1</sup> , respectively). The second phenolic class recorded in abundance was that of phenolic acids, with comparable values detected for "Purple B73" (1305.9 mg kg−<sup>1</sup> ), "Rostrato Rosso" (1149.7 mg kg−<sup>1</sup> ), and the control sample (979.6 mg kg−<sup>1</sup> ), while

the lowest values were recorded in "Val di Non" (589.2 mg kg−<sup>1</sup> ). Furthermore, "Rostrato Rosso" showed the highest content of tyrosol equivalents (822.4 mg kg−<sup>1</sup> ) when compared to the other maize samples.

Subsequently, the OPLS-DA analysis was performed in order to better account for markers of the differences observed in phenolic profile. The predictive model clearly discriminated among maize cultivars (**Figure 4B**), showing that the pigmented genotypes, i.e., "Rostrato Rosso" and "Purple B73" samples possessed a completely differentiated phenolic profile when compared to the other samples, the latter clustering together onto the OPLS-DA hyperspace. The characteristics of the OPLS-DA model were excellent, with R <sup>2</sup>Y and Q <sup>2</sup>Y being 0.97

and 0.88, respectively. No outliers could be identified using Hotelling's T2, whereas OPLS-DA over fitting could be excluded by both CV-ANOVA (correlation p 6.2 10−17) and permutation testing (**Supplementary Figure S1**). Afterwards, the variable's importance in the OPLS-DA model was evaluated using the VIP analysis and exporting the VIP scores for each phenolic compound detected through the untargeted UHPLC-ESI/QTOF-MS approach. The VIP score summarizes the contribution that a variable makes to the OPLS-DA model. The phenolic compounds with the highest recorded VIP scores (>1.2) are reported in **Table 4**; the 42 phenolic compounds detected could be considered the most important and contributing variables in class discrimination. In line with the previously reported evaluations from unsupervised multivariate statistics and Volcano analysis, the most abundant phenolics identified by the VIP analysis could be ascribed to flavonoids (i.e., anthocyanins and flavones) and phenolic acids (above all, hydroxycinnamics), thus confirming that these two phenolic

non-pigmented kernel control.

TABLE 3 | Semi-quantitative data for the different phenolic classes, as phenolic equivalents (±standard deviation) following UHPLC/QTOF profiling in each of the four maize genotypes.


#### TABLE 4 | Compounds better discriminating between different maize genotypes, as selected by VIP analysis following OPLS-DA.


Compounds are provided together with VIP scores ± standard error (measure of variable's importance in the OPLS-DA model).

classes are the most explicative in determining the differences observed in phenolic profile.

#### In vitro Antioxidant Capacity of the Maize Kernels

In this work, the in vitro antioxidant capacity of different maize genotypes was evaluated by means of two different methods, being the DPPH radical scavenging and the FRAP reducing power, since the aforementioned methods are based on different reaction kinetics, i.e., a hydrogen atom transfer (DPPH method) and a single electron transfer (FRAP). The results for antioxidant capacity are reported in **Table 5**. Remarkably, both assays provided essentially the same information, when compared to the cumulative abundances by UHPLC-ESI/QTOF-MS for each phenolic class equivalent. DPPH values were in the range 18.8 – 150.1 mg kg−<sup>1</sup> GAE, whereas FRAP values were comprised between 4364.6 and 18616.2 mg kg−<sup>1</sup> GAE. The FRAP results were consistent with the semi-quantitative values, recording the highest values in "Rostrato Rosso" followed by "Purple B73," "Val di Non" and the Agrister yellow maize samples.



Within the same column, means followed by the same letter are not significantly different at P ≤ 0.05 according to Fisher's test.

# Response to Artificial Infection With F. verticillioides

The inbred line B73 was chosen as a yellow kernel reference because of its common use in association panels to screen for FER resistance (Zila et al., 2013). Previous literature on the concentration of beneficial phytochemicals in harvested grains of yellow maize highlighted a moderate variability of phenolic profile, mainly ascribable to a year × genotype interaction (Butts-Wilmsmeyer et al., 2017). Our UHPLC-ESI/QTOF-MS phenolic profiling of the B73 non-pigmented maize, followed by quantification as phenolic sub-classes, resulted in a substantially comparable profile, as compared to the Agrister genotype used as reference for antioxidant activity (**Supplementary Tables S2**, **S3**). The modest differences we observed between the two yellow maize samples could be likely ascribed to the above-mentioned year × genotype interaction. Nonetheless, the use of a commercial hybrid such as Agrister (i.e., the genotype used as reference in phenolic profiling) would not have been suitable for in vitro infection assays, because these seeds were commercialized as fungicide-coated kernels. All maize genotypes in the control RTA were free from F. verticillioides presence, with the exception of "Purple B73" and "Val di Non" (mean value of 1.25). After artificial inoculation different responses could be observed: "Rostrato Rosso" was the less infected with SEV\_I means of 2.1 while "Val di Non" and "Purple B73" were the most favorable for fungal development (3.3 and 4.2 of mean values, respectively) (**Figures 5A,B**). Analysis of variance of SEV\_I phenotypic values, performed according to the Kruskal–Wallis test, resulted in a p-value = 4.196 × 10−<sup>9</sup> . Therefore, samples were compared each other to shed light to significant variations between genotypes (Kruskal–Dunn post hoc test). "Purple B73" was significantly different with respect to the other pigmented and not pigmented varieties. Likewise, "Rostrato Rosso" was found to be significantly different from all other maize genotypes, while "Val di Non" response to infection was not different from that of the control. These results suggest that within colored maize cultivars different levels of resistance can be evidenced: "Rostrato Rosso" was the most resistant and "Purple B73" was the most susceptible (**Table 6** and **Figures 5A,B**).

#### Field Evaluation for FER Resistance

Field-grown maize genotypes were visually scored for FER severity following artificial inoculation, and variation between accessions is shown in **Figures 5C,D**. Because ears of the cultivar "Val di Non" were poorly pollinated, they were not used for Fusarium infection analysis in order to prevent an incorrect FER evaluation (Walker and White, 2001).

ANOVA after Welch test resulted in significant differences (p = 0.00102) and thus the Games–Howell post hoc test was applied to check for homogenous classes (**Table 6**). All the genotypes were significantly different each other, with the highest significance values considering comparisons to "Rostrato Rosso" (**Table 6**).

## DISCUSSION

Pigmented maize genotypes were characterized by different contents in anthocyanins, and in other not pigmented phenolic compounds. Even though anthocyanins were the highest in "Rostrato Rosso" and "Purple B73," phenolic acids were the second class in abundance, having the highest content in "Rostrato Rosso," "Purple B73" and control. Tyrosols and flavones were highest in "Rostrato Rosso," whereas "Purple B73" showed the highest flavonols content. Therefore, the difference in phenolic profile among the considered genotypes was far beyond what observed at phenotype level. With this regard, an untargeted metabolomic profiling approach appears to be more appropriate in describing the actual phenolic profile from a holistic perspective. Indeed, VIP analysis from OPLS-DA multivariate statistics highlighted that compounds belonging to flavonoids such as flavones and flavonols, rather than hydroxycinnamic acids, were also responsible of differences in phenolic profile between cultivars, together with anthocyanins. Previous literature on phenolics in pigmented maize mainly focused on anthocyanins profile, highlighting as the malonyl and succinyl derivatives are the predominant one (Carmelo-Méndez et al., 2016). The results obtained confirm the importance of acylated anthocyanins, whereas total anthocyanins in our samples were in the same range of other pigmented maize (Lopez-Martinez et al., 2009) and one order higher than values reported for Mexican blue maize (Mora-Rochín et al., 2016). However, to the best of our knowledge, no comparative information has been reported regarding the phenolic content of other classes of phenolic compounds.

A confirmation on the relevance of colorless phenolics also in pigmented maize can be gained looking at Pearson's correlation values (**Table 7**) between the in vitro antioxidant capacity and the phenolic classes content. Tyrosols and anthocyanins were highly correlated to FRAP reducing power (0.87 and 0.86, respectively, p < 0.01) whereas DPPH radical scavenging capacity correlated with flavones, tyrosols and anthocyanins (0.85, 0.80, and 0.70, respectively, p < 0.01). A significant but weak correlation was observed between DPPH and FRAP values (0.47, p < 0.01); this result is not surprising, considering that the different antioxidant tests rely on different mechanisms and should be better considered as complementary rather than alternative (Przygodzka et al., 2014). Indeed, the in vitro antioxidant capacity should not be determined based on a single antioxidant test

TABLE 6 | P-values after Kruskal–Dunn post hoc test on severity of infection phenotypic values following RTA inoculation with F. verticillioides (SEV\_I) and in field inoculation (FER).


Significant comparisons are marked by an asterisk.

TABLE 7 | Two-tails Pearson correlation between in vitro antioxidant capacity assays and content of the main phenolic classes.


Significant correlations are indicated by asterisks (∗p < 0.05, ∗∗p < 0.01).

model (Alam et al., 2013). Interestingly, the spread of DPPH values was much narrower than FRAP reducing power values, suggesting that DPPH assay might be less informative for the description of antioxidant capacity in the maize samples analyzed. Nonetheless, our results regarding antioxidant capacity values were consistent with previous studies. In particular, the antioxidant capacity of pigmented maize samples and by-products has been previously studied (Bello-Pérez et al., 2015), with anthocyanins giving the most important contribution to the FRAP, DPPH, and ABTS recorded values (Bello-Pérez et al., 2015; Carmelo-Méndez et al., 2017). However, most of the available studies in literature investigated the use pigmented maize varieties as potential ingredient for the development of functional foods, but focusing the attention above all on blue maize.

Moving toward the involvement of phenolic profiles in resistance toward Fusarium infection, interesting differences could be pointed out from both RTA and field inoculation experiments. The genotype "Rostrato Rosso" was the most resistant after F. verticillioides infection, both in vitro and in field conditions. Interestingly, this cultivar was featured by a significantly higher phenolic content. "Purple B73" presented a more severe infection following in vitro and field inoculation with Fusarium, as compared to pigmented genotypes.

With this regard, it is important to highlight that plants use a complex and interconnected defense system against pests and pathogens; the production of low molecular weight secondary metabolites having antimicrobial activity, collectively known as phytoalexins, is part of these defense mechanisms (Ahuja et al., 2012). Previous literature reported that the kernel color can be associated to Fusarium resistance and that the accumulation of flavonoids pigments in kernel is able to reduce the accumulation of fumonisin B1 (Pilu et al., 2011). Nonetheless, the localization of pigments in maize kernel might have a prominent role in the actual degree of resistance to Fusarium infection. Recently, it was reported that flavonoids can inhibit fumonisin accumulation in Fusarium-inoculated maize ears, even though pigmented pericarp alone was ineffective in preventing the accumulation of the mycotoxin (Venturini et al., 2016).

#### CONCLUSION

The different pigmented maize cultivars selected in this work appear to be interesting both in terms of phenolic profile and

#### REFERENCES


antioxidant capacity. Therefore, the findings support further studies in order to formulate food matrices enriched in phenolic compounds, and thus with beneficial health-promoting effects.

Given the fact that the highest resistance to Fusarium infection was observed in the cultivar having the highest phenolic content, further studies should be made to confirm the association between specific phenolic compounds or classes of phenolics, and fungal resistance. Undeniably, pigmented maize cultivars could be important both in finding breeding strategies in the framework of sustainable agriculture as well as in developing foods/food ingredients with higher nutraceutical value.

#### AUTHOR CONTRIBUTIONS

All authors contributed to revise work critically, gave the final approval of the version to be published and agreed on all aspects of the work. In particular, AM designed the study, in cooperation with JB and LL. JB, LS, and MB carried out the field experiments and the artificial infections, contributed to interpretation of data, drafted and revised the manuscript. LL and GR developed the mass spectrometric method, performed statistics, and drafted the manuscript. AL, MB, CC, and GDP drafted and critically revised the manuscript.

### FUNDING

This work is a part of the Italian project "Pasta e Nuovi Prodotti Alimentari ad Alta Qualità" (PAQ) funded by the Italian Ministry of Economic Development. GR was recipient of a Ph.D. fellowship from the AgriSystem School at Università Cattolica del Sacro Cuore. The authors acknowledge Fondazione Invernizzi for the contribution to the funding of the metabolomics platform, Paola Battilani and Paola Giorni for kindly providing F. verticillioides strain.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2018.01189/ full#supplementary-material

Fusarium and mycotoxin accumulation. Front. Microbiol. 7:566. doi: 10.3389/ fmicb.2016.00566



**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 Bernardi, Stagnati, Lucini, Rocchetti, Lanubile, Cortellini, De Poli, Busconi and Marocco. 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(s) 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.

# Identification of Biomarkers for Defense Response to Plasmopara viticola in a Resistant Grape Variety

Giulia Chitarrini1,2, Evelyn Soini<sup>1</sup> , Samantha Riccadonna<sup>3</sup> , Pietro Franceschi<sup>3</sup> , Luca Zulini<sup>4</sup> , Domenico Masuero<sup>1</sup> , Antonella Vecchione<sup>4</sup> , Marco Stefanini<sup>4</sup> , Gabriele Di Gaspero<sup>5</sup> , Fulvio Mattivi1,6 and Urska Vrhovsek<sup>1</sup> \*

<sup>1</sup> Food Quality and Nutrition Department, Fondazione Edmund Mach, San Michele all'Adige, Italy, <sup>2</sup> Department of Agricultural and Environmental Sciences, University of Udine, Udine, Italy, <sup>3</sup> Computational Biology Unit, Fondazione Edmund Mach, San Michele all'Adige, Italy, <sup>4</sup> Genomics and Biology of Fruit Crop Department, Fondazione Edmund Mach, San Michele all'Adige, Italy, <sup>5</sup> Istituto di Genomica Applicata, Udine, Italy, <sup>6</sup> Center Agriculture Food Environment, University of Trento, Trento, Italy

#### Edited by:

Marta Sousa Silva, Universidade de Lisboa, Portugal

#### Reviewed by:

Michaela Griesser, University of Natural Resources and Life Sciences, Vienna, Austria Luigi Bavaresco, Università Cattolica del Sacro Cuore, Italy

> \*Correspondence: Urska Vrhovsek urska.vrhovsek@fmach.it

#### Specialty section:

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

Received: 03 July 2017 Accepted: 21 August 2017 Published: 05 September 2017

#### Citation:

Chitarrini G, Soini E, Riccadonna S, Franceschi P, Zulini L, Masuero D, Vecchione A, Stefanini M, Di Gaspero G, Mattivi F and Vrhovsek U (2017) Identification of Biomarkers for Defense Response to Plasmopara viticola in a Resistant Grape Variety. Front. Plant Sci. 8:1524. doi: 10.3389/fpls.2017.01524 Downy mildew (Plasmopara viticola) is one of the most destructive diseases of the cultivated species Vitis vinifera. The use of resistant varieties, originally derived from backcrosses of North American Vitis spp., is a promising solution to reduce disease damage in the vineyards. To shed light on the type and the timing of pathogen-triggered resistance, this work aimed at discovering biomarkers for the defense response in the resistant variety Bianca, using leaf discs after inoculation with a suspension of P. viticola. We investigated primary and secondary metabolism at 12, 24, 48, and 96 h post-inoculation (hpi). We used methods of identification and quantification for lipids (LC-MS/MS), phenols (LC-MS/MS), primary compounds (GC-MS), and semi-quantification for volatile compounds (GC-MS). We were able to identify and quantify or semiquantify 176 metabolites, among which 53 were modulated in response to pathogen infection. The earliest changes occurred in primary metabolism at 24–48 hpi and involved lipid compounds, specifically unsaturated fatty acid and ceramide; amino acids, in particular proline; and some acids and sugars. At 48 hpi, we also found changes in volatile compounds and accumulation of benzaldehyde, a promoter of salicylic acid-mediated defense. Secondary metabolism was strongly induced only at later stages. The classes of compounds that increased at 96 hpi included phenylpropanoids, flavonols, stilbenes, and stilbenoids. Among stilbenoids we found an accumulation of ampelopsin H + vaticanol C, pallidol, ampelopsin D + quadrangularin A, Z-miyabenol C, and α-viniferin in inoculated samples. Some of these compounds are known as phytoalexins, while others are novel biomarkers for the defense response in Bianca. This work highlighted some important aspects of the host response to P. viticola in a commercial variety under controlled conditions, providing biomarkers for a better understanding of the mechanism of plant defense and a potential application in field studies of resistant varieties.

Keywords: biomarkers, Plasmopara viticola, hybrid, plant pathogen, Bianca, grapevine, resistance, metabolomics

# INTRODUCTION

fpls-08-01524 September 2, 2017 Time: 15:7 # 2

Downy mildew is one of the most destructive diseases of the grapevine caused by the biotrophic oomycete Plasmopara viticola (Berk. and Curt.) Berl. & de Toni. This pathogen was introduced from North America into Europe in the second half of the nineteenth century. The cultivated species Vitis vinifera is susceptible to P. viticola. Disease management strategies rely on the use of fungicides potentially harmful to humans and the environment (Negatu et al., 2016; Christen et al., 2017; Rortais et al., 2017). In some situations, chemical protection is also economically challenging, due to the costs of synthetic fungicides and the labor involved in spraying.

The pathogen is able to infect green tissues and establish biotrophism widely across the Vitis genus. Unlike the European V. vinifera, some accessions in North American wild species have evolved host resistance. Resistant accessions are able to activate defense responses upon pathogen infection, which culminate in localized necrosis, resulting into lower rates of sporangia release compared to susceptible accessions (Bellin et al., 2009; Polesani et al., 2010).

Resistant accessions of wild species have been crossed with cultivated varieties to introgress resistance. The use of resistant varieties is a promising strategy for viticulture to cope with downy mildew (Bisson et al., 2002). Among these, the variety Bianca is widely cultivated in Hungary, Moldova, and Russia and is one of the few resistant accessions in which the genetic basis of resistance has been elucidated (Bellin et al., 2009). Bianca is an hybrid between Bouvier and the resistant grapevine Villard Blanc. It was obtained in 1963 (Csizmazia and Bereznai, 1968), and officially registered for use in wine production in 1982 (Kozma and Dula, 2003). A large part of the resistance phenotype of Bianca is explained by the Rpv3 locus, located in chromosome 18. In Bianca and in all known resistant descendants of the Villard Blanc the Rpv3 locus controls the ability to trigger a localized hypersensitive response (HR) soon after the initiation of the infection (Bellin et al., 2009; Di Gaspero et al., 2012). HR in the proximity of infection sites confines biotrophic pathogens, restricting their endophytic growth (Jones and Dangl, 2006). Early inducible responses include cell wall deposition, release of reactive oxygen species and hypersensitive cell death (HR) at the infection site, controlled by interactions between avirulence gene products and plant receptors, and it can be the result of multiple signaling pathways (Heath, 2000).

Plant defense responses require energy and activation of signaling molecules, primarily supplied by primary metabolism of carbohydrates, organic acids, amines, amino acids, and lipids (Bolton, 2009; Rojas et al., 2014). HR also stimulates the expression of defense responses near the infected area and the onset of systemic acquired resistance (Greenberg and Yao, 2004). Several studies have shown the importance of secondary metabolites for expressing plant defense, often related to specific functions such as toxicity against pathogens, or acting as signal molecules after stress (Bennett and Wallsgrove, 1994; Gershenzon and Dudareva, 2007). The induction of stress-related metabolites known as phytoalexins contributed to the inhibition of biotrophic pathogens in resistant grapevines (Dercks and Creasy, 1989; Derckel et al., 1999; Slaughter et al., 2008; Godard et al., 2009; Ferri et al., 2011; Gessler et al., 2011). Stilbenes is a class of phytoalexins that provided active compounds with antifungal activity against various pathogens, including P. viticola (Dercks and Creasy, 1989). The pattern of stilbene accumulation upon P. viticola infection differs among Vitis species. Stilbene concentration showed earlier and higher increase in resistant varieties as compared to susceptible ones. In other cases, downy mildew resistance was observed in the absence of stilbene accumulation (Keller, 2015). This suggests the necessity to investigate which secondary metabolites play a key role in resistance and which of them are reliable biomarkers of the defense response in resistant varieties.

We expect that several classes of primary and secondary metabolites are modulated in Bianca during the defense response to P. viticola. In this scenario, metabolomics is the most suitable approach for monitoring a wide range of molecules. Indeed, several metabolomics studies have been already reported in grapevine. Some of them aimed at highlighting intervarietal variation in berry composition (Mulas et al., 2011; Gika et al., 2012; Degu et al., 2014; Teixeira et al., 2014; Bavaresco et al., 2016). Other studies aimed at identifying metabolite changes in infected leaves (Batovska et al., 2009; Ali et al., 2012; Becker et al., 2013; Algarra Alarcon et al., 2015). However, the metabolite changes that are brought about by the resistance mechanism have not yet been fully described.

In this work, we monitored metabolite changes in leaf discs of the resistant variety Bianca after infection with a suspension of P. viticola, with the aim of discovering biomarkers for specific stages of the host defense. In particular, we evaluated both primary and secondary metabolism at 12, 24, 48, and 96 h post-inoculation (hpi). We used existing protocols of LC-MS/MS for identification and quantification of lipids and phenols, and GC-MS for identification and semi-quantification of volatile organic compounds (VOCs). Moreover, we validated a new GC-MS protocol for the identification and quantification of primary compounds, including organic acids, amino acids, amines, sugars, and lipids, which yielded 48 metabolites in Bianca leaf discs.

#### MATERIALS AND METHODS

#### Plant Material

Metabolite analyses were performed using leaves from twonode cuttings of the cultivar Bianca. The mother plants were held at Fondazione Edmund Mach grape collection, San Michele all'Adige, Italy (46◦ 120 0 <sup>00</sup>N, 11◦ 8 0 0 <sup>00</sup>E). Own-rooted vines (n = 45) were grown in potted soil in controlled greenhouse conditions. Water was supplied by drip irrigation in order to avoid premature infections of downy mildew on leaves. At the stage of 12-leaf shoots, the plants were sorted into three homogenous groups; each group represented a biological replicate. At the time of the experiment plants were healthy, with no evidence of foliar diseases.

## Artificial Inoculation of Leaf Discs and Incubation under Controlled Conditions

The third, fourth, and fifth fully expanded leaves beneath the apex were detached from each plant, rinsed with ultrapure water. From each leaf, 1.1 cm diameter discs were excised with a cork borer and placed randomly onto wet paper in Petri dishes with the abaxial side up. Around 100 discs per condition per time point (i.e., 12 hpi, inoculated, biological replicate 1 = 100 discs) were used. Leaf discs were left to equilibrate at 21◦C for 12 h after punching and prior to inoculation. P. viticola spores were collected from natural infected leaves in an untreated vineyard in 2014 and immediately frozen at −20◦C. They were propagated by infecting a susceptible variety and collecting fresh sporulation. After sporulation, the fresh spores were immediately used to prepare the experiment suspension. Discs were sprayed with P. viticola inoculum suspension at 1 × 10<sup>6</sup> sporangia/mL. Sealed Petri dishes were incubated in a growth chamber at 21◦C until sampling. Mock inoculated control were prepared with ultrapure water. Leaf discs were sampled at 12, 24, 48, and 96 hpi/mock, then ground under liquid nitrogen to obtain a frozen powder. Three biological replicates were sampled at each time point.

# Targeted Primary Compound Analysis and Method Validation

#### Sample Preparation

The extraction of primary metabolites was carried out according to Fiehn et al. (2008) with some modifications. Briefly, 0.1 g of fresh leaf powder was subjected to extraction by adding 1 mL of cool (−20◦C) extraction solvent, composed of isopropanol/acetonitrile/water (3:3:2 v/v/v). A 20 µL aliquot of a solution containing palmitic-D3, nicotinic-D4, and glucose-D7 (1000 mg/L) was added as an internal standard. The extraction mixture was vortexed for 10 s, shaken at 4◦C for 5 min and centrifuged at 12,000 g for 2 min at 5◦C. A second round of extraction was carried out following the same procedure. The two supernatants were merged and re-suspended in a final volume of 5 mL using the extraction solvent. A total of 250 µL of supernatant was placed in a 1.5 mL Eppendorf tube and evaporated to dryness under N2. The residue was re-suspended in 500 µL of acetonitrile/water (50:50 v/v), vortexed for 10 s, sonicated and centrifuged at 12,000 g for 2 min. The supernatant was then transferred into a 1.5 mL Eppendorf tube and dried out under N2. The dried extract was subject to derivatization, first by adding 20 µL of methoxamine hydrochloride in pyridine (20 mg/mL) to inhibit cyclization of reducing sugars and shaken at 30◦C for 1 h; then by adding 80 µL of N-methyl-Ntrimethylsilyl-trifluoroacetamide with 1% trimethylchlorosilane for trimethylsilylation of acidic protons and shaken at 37◦C for 30 min. Finally, 5 µL of a solution containing decane and heptadecane (1000 mg/L) were added in order to monitor the chromatographic analysis and the instrumental conditions. The derivatized extract was then transferred into vials for analysis. One microliter of derivatized extract was injected for GC/MS analysis.

#### Instrumental Conditions

Analyses were performed using a Trace GC Ultra combined with a mass spectrometer TSQ Quantum GC and an autosampler Triplus (Thermo Electron Corporation, Waltham, MA, United States). A RXI-5-Sil MS w/Integra-Guard <sup>R</sup> (fused silica) (30 m × 0.25 mm × 0.25 µm) column was used for compound separation. Helium was used as the carrier gas at 1.2 mL/min and the injector split ratio was set to 1:10. The injector, transfer line and source temperature were set to 250◦C. The initial oven temperature was kept at 65◦C for 2 min, increased by 5.2◦C/min to 270◦C and held at 270◦C for 4 min. These conditions were shown to represent a good compromise in order to obtain a not excessively long chromatographic run, a high number of compounds and good peak separation. The mass spectrometer was operated in electron ionization mode. Data acquisition was performed in full scan mode from 50 to 700 m/z. Data processing was performed using XCALIBURTM 2.2 SOFTWARE.

#### Method Validation

The method for primary metabolites was validated according to the currently accepted US Food and Drug Administration (FDA) bio-analytical method validation guide (US Department of Health and Human Services, 2001). Validation assays were established on calibration standards and quality control (QC) samples prepared as a pool of grape samples, extracted and derivatized according to the procedure described above. QC samples were used to evaluate the recovery of each compound and the stability of sample, intra- and inter-day variability, and to evaluate the efficiency of the extraction procedure. The standard mix was used to determine the limit of detection (LOD), limit of quantification (LOQ), and linearity range for each compound. Matrix calibration curves built using QC samples were compared with solvent calibration curves. Matrix effect (ME) values were determined using the slope ratios: ME% = 100 × (1 − slope solvent calibration curve/slope matrix calibration curve) (Kwon et al., 2012). LOQ and LOD were evaluated at the concentration in which the quantifier transition presented a signal-to-noise (S/N) ratio of >10 and >3, respectively. Intra- and inter-day variability were evaluated using the coefficient of variation (CV%) of QC samples injected 10 times on 1 day and then for 5 consecutive days. The recovery test was estimated on 10 spiked grape samples and calculated as the average of the "measured value/expected value" ratio (%). Each compound was identified and quantified against the standard, using one, or in the case of a few compounds, two specific m/z characteristics for the individual metabolite (extracted ion monitoring) and excluding saturated fragments. The fragments used for quantitation and the linear retention index (RI) are reported in Supplementary Table S1. Compounds were expressed as mg/kg of fresh leaves.

#### Targeted Lipid Compound Analysis

Lipid analysis was carried out according to Della Corte et al. (2015), using Folch's extraction method (Folch et al., 1957; Della Corte et al., 2015) with some modifications. Briefly, 0.3 mL of methanol were added to 0.1 g of fresh leaf powder and vortexed for 30 s, then 0.6 mL of chloroform containing butylated hydroxyl

toluene (500 mg/L) were added, followed by the addition of 10 µL of internal standard (docosahexaenoic acid 100 µg/mL). Samples were placed in an orbital shaker for 60 min. After the addition of 0.25 mL of water, samples were centrifuged at 3600 rpm for 10 min. The total lower lipid-rich layer was collected and re-extracted by adding 0.4 mL of chloroform/methanol/water 86:14:1 v/v/v. The samples were centrifuged at 3600 rpm for 10 min, the total lower lipid-rich layer was collected. Both chloroform fractions were merged and evaporated to dryness under N2. Samples were re-suspended in 300 µL of acetonitrile/2 propanol/water (65:30:5 v/v/v) containing the internal standard cholesterol at a concentration of 1 µg/mL and transferred into a HPLC vial. Separation was performed using a UHPLC Dionex 3000 (Thermo Fisher Scientific, Germany), with a RP Ascentis Express column (15 cm × 2.1 mm; 2.7 µm C18) purchased from Sigma, following a 30 min multistep linear gradient following Della Corte et al. (2015). The UHPLC system was coupled with an API 5500 triple-quadrupole mass spectrometer (Applied Biosystems/MDS Sciex) equipped with an electrospray ionization (ESI) source. Compounds were identified using Analyst Software based on their true reference standard, retention time and qualifier and quantifier ion, and were quantified using their calibration curves and expressed as mg/kg of fresh leaves.

#### Targeted Phenolic Compound Analysis

Phenolic compounds were determined according to Vrhovsek et al. (2012), with some modifications. Briefly, 0.4 mL of chloroform and 0.6 mL of methanol:water (2:1) were added to 0.1 g of fresh leaf powder. A 20 µL aliquot of gentisic acid (50 mg/L) and rosmarinic acid (50 mg/L) were added as internal standards. The extraction mixture was shaken for 15 min in an orbital shaker, then centrifuged for 5 min at 15,000 g at 4 ◦C. The upper aqueous-methanolic phase was collected. The extraction was repeated by adding 0.6 mL of methanol and water (2:1 v/v) and 0.2 mL of chloroform; the samples were centrifuged for 5 min at 15,000 g at 4◦C. The aqueous-methanolic phase was collected and combined with the previous one. Both fractions were merged and evaporated to dryness under N2. Samples were re-suspended in 500 µL of methanol and water (1:1 v/v), centrifuged and transferred carefully into an HPLC vial. Chromatographic analysis was performed using a Waters Acquity UPLC system (Milford) with a Waters Acquity HSS T3 column (100 mm × 2.1 mm; 1.8 µm) following Vrhovsek et al. (2012). Mass spectrometry detection was performed on a Waters Xevo triple-quadrupole mass spectrometer detector (Milford) with an electrospray ionization (ESI) source (Vrhovsek et al., 2012). Compounds were identified based on their reference standard, retention time and qualifier and quantifier ion, and were quantified using their calibration curves and expressed as mg/kg of fresh leaves. Data processing was performed using Waters MassLynx V4.1 software.

#### Volatile Compound Analysis

Volatile compounds were extracted with solid phase microextraction, using a method adapted from Matarese et al. (2014) and Salvagnin et al. (2016). The extraction was carried out with some modifications; briefly, 0.1 g of fresh leaves were placed in 10 mL glass vials with 2 mL of buffer (0.1 m Na2HPO<sup>4</sup> and 50 mM citric acid; pH 5), 0.2 g of NaCl, and 5 µL of 1-heptanol (25 mg/L) as internal standard. Samples were kept at 60◦C for 20 min and compounds in the headspace were captured for 35 min at 60◦C. A Trace GC Ultra gas chromatograph coupled to a Quantum XLS mass spectrometer (Thermo Electron Corporation, Waltham, MA, United States) was used to separate the compounds with a fused silica Stabilwax <sup>R</sup> -DA column (30 m × 0.25 mm i.d. × 0.25 µm) (Restek Corporation, Bellefonte, United States). The headspace was sampled using 2-cm DVB/CAR/PDMS 50/30 µm fiber from Supelco (Bellefonte, PA, United States). The compounds were desorbed in the GC inlet at 250◦C for 4 min. The GC oven parameters were set following Salvagnin et al. (2016). The MS detector was operated in scan mode (mass range 40–450 m/z) with a 0.2 s scan time and the transfer line to the MS system was maintained at 250◦C. Data processing was performed using XCALIBURTM 2.2 SOFTWARE. For the identification of volatile compounds we used letter "A" for compounds with comparable mass spectra and retention time to those of the pure standard, "B" for those with a RI match on a similar phase column with the database NIST MS Search 2.0, and "C" for those identified in the mass spectral database NIST MS Search 2.0 (Sumner et al., 2007). The experimental linear temperature RI of each compound was calculated using a series of n-alkanes (C10-C30) in the same experimental conditions as the samples. The results were expressed in a semi-quantitative manner and expressed in µg/kg using 1-heptanol as the internal standard.

# Data Analysis

Statistical analysis and data visualization were performed with custom R scripts (R Core Team, 2017). Missing values were imputed with a random value between zero and LOQ. The concentrations were transformed using the base 10 logarithm, in order to make data distribution more normal-like (van den Berg et al., 2006). Principal component analysis (PCA) was performed on the obtained multidimensional dataset, after mean centering and unit scaling, using the FactoMineR and Factoextra R packages (Lê et al., 2008; Kassambara and Mundt, 2017). The t-statistic was computed using the Stats package (R Core Team, 2017), while network visualization exploited the ggraph package (Pedersen, 2017).

# RESULTS

In leaf discs inoculated with P. viticola and in mock-inoculated controls, we identified 176 compounds (Supplementary Table S2) belonging to acids (18), amino acids (13), amines and others (3), sugars (14), carnitines (1), sterols (3), fatty acids (14), glycerolipids (4), glycerophospholipids (4), sphingolipids (1) prenols (1), benzoic acid derivates (4), coumarins (2), phenylpropanoids (6), dihydrochalcones (1), flavones (1), flavan-3-ols (9), flavonols (11), stilbenes and stilbenoids (14), and other phenolics (2). All these metabolites were annotated with identification level 1 (with standards) and their concentration was expressed as mg/kg of fresh leaves. The volatile acids

(3), alcohols (7), aldehydes (9), benzenoids (4), ketones (4), terpenoids (14), other VOCs (5), and unknown VOCs (5) were semi-quantified as the equivalent of the internal standard (1-heptanol) and their concentration was expressed as µg/kg of fresh leaves (Supplementary Table S2). The concentration reported represents the average value of three biological replicates ± standard error. For the identification of VOCs, we reported the confidence levels for metabolite identification defined by the Metabolomics Standards Initiative (Sumner et al., 2007): level A is assigned to compound for which the mass spectrum and the retention time match with the one of the pure standard; level B indicates that the RI of the compound and of the reference standard matches on a similar phase column; level C is assigned when the compounds mass spectrum is available into mass spectra databases (Supplementary Table S2).

# Validation Results of the Primary Compound Method

Unlike for lipid, phenolic, and volatile compounds, a validated protocol for identification and quantification of primary metabolites in grapevine leaves was missing at the beginning of this study. We thus adopted a method established by Fiehn et al. (2008) on grape berries and performed a validation step to confirm the identity of each compound in a leaf matrix. All the standards were injected to obtain their fragmentation patterns and to calculate their retention indices. The calculated retention indices and mass spectra were compared with the NIST MS Search 2.0 database. The method was validated with the injection of relative standards for 96 compounds: 29 acids, 17 amino acids, 12 amines and others, 24 sugars and 14 fatty acids (Supplementary Figure S1). All the validation results are summarized in Supplementary Table S1. The ME values evaluated by comparing the calibration curves (matrix and solvent) were in the range between −20 and 20%, except for salicylic acid, citric acid, glycine, beta-alanine, tyrosine, fructose, and myoinositol, which slightly exceeded the limit of ±20% established by the validation method guide; this value can be considered as insignificant, because it is close to the relative standard deviation values of repeatability (European Commission, 2011). Intra- and inter-day repeatability were evaluated for each compound and expressed as CV%. The value should not exceed 15% for intra-day and 20% for inter-day; again in this case we had very good results, except for oxalic acid (intra-day 18.2%; inter-day 26.4) and malonic acid (intra-day 15.2%; inter-day 44.6%). The recovery ranges were over 90% for 74 compounds, between 80 and 90% for 13 compounds, between 70 and 80% for four compounds, and between 50 and 70% for five compounds. Using solvent calibration curves we evaluated the linearity ranges and the LOD and LOQ limits for each compound reported in Supplementary Table S1. In general, we obtained good validation results for the method, which make us confident about the possibility of applying the method for accurate quantification of primary compounds in different matrices.

The fatty acid derivatization step can modify the profile, with the formation of oxidation or isomerization products (Rigano et al., 2016) and as previously reported, the best option is to use trimethylsilyl diazomethane, with the production of methyl esters (FAMEs), avoiding the poor separation of fatty acid compounds and substantial interference (Topolewska et al., 2015). In our method, we validated all the compounds following the derivatization used by Fiehn et al. (2008), but we found a consistent residue in blank injections of some compounds, such as palmitic acid, stearic acid, and arachidic acid in particular during the sample runs. Due to this interference, we were not able to correctly quantify fatty acid compounds in our matrix using the GC-MS method therefore their quantification was performed using LC-MS/MS.

# Metabolite Changes during the Defense Response

Global metabolite changes in the resistant host upon pathogen inoculation were first visualized by using PCA (**Figure 1**). In the plot, the position of the three biological replicates and their average is reported for each time point (12, 24, 48, and 96 h) for inoculated and not inoculated leaves. PCA of all compounds revealed good separation between the factors of the study and the temporal evolution was clearly captured by the first PCA component (Dim 1), accounting for 35.8% of total variance. The second component (Dim 2), accounting for 13.4% of total variance, discriminated leaf discs undergoing a defense response to P. viticola from mock inoculated controls (**Figure 1**).

In order to identify which class of metabolites was responsible for this separation, we performed PCA (**Figure 2**) separately for primary metabolites (**Figure 2A**), lipids (**Figure 2B**), phenolic compounds (**Figure 2C**), and VOCs (**Figure 2D**). Again the time trend was clearly distinguishable (captured by the first component), and also a good separation between the two conditions can be noticed for specific time points for the different classes of compounds. Indeed, we observed for primary compounds a clear separation between the two conditions along the second dimension (which captured 24.1% of the total variance) at 48 hpi and, looking at the two components (explaining a total of 55.3% of the variance), possibly at 24 hpi (**Figure 2A**). Lipids showed the greatest differences at 24 hpi, where the inoculated and control samples are separated mainly along the second component, explaining 16.4% of the total variance (**Figure 2B**). Phenols were involved in the plant response only later, at 96 hpi, with the first component capturing 52.3% of the variance and possibly explaining both the time course and the differences between the two conditions (**Figure 2C**). Finally, PCA of VOCs separated inoculated and not inoculated samples at 48 hpi, and at 96 hpi, mainly along the first component, which explains 51.5% of the variance due to both the time trend and the differences between the two conditions in the last two time points (**Figure 2D**).

In order to select biomarkers for each specific stage of the defense response, we computed the t-statistic for all the metabolites for each time point, since it takes into account both the difference between the means and the estimate of the biological variability. To concentrate on the compounds most different between inoculated and control samples, we focused on a subset of metabolites (176) and time points (4) having an arbitrary absolute value for the t-statistic greater than 3 (|t| > 3): 64 values of the t-statistic satisfied our constraint.

In terms of compounds we identified 53 metabolites, which were different between inoculated and control samples in at least one time point (Supplementary Figure S2 and Table S3). The results of this analysis are represented in the network of **Figure 3**. The network contains 53 nodes, each one representing one metabolite. A link is drawn between two metabolites only if both metabolites have |t| > 3 at the same time point. In the same visualization the class of the compound is highlighted by the color of the node and the time course information by the color of the link (**Figure 3**). Time point specific cliques are characterizing the structure of the network: the metabolites shown in one of these cliques show differences between the two conditions at that specific time point. The smaller number of nodes in the network at 12 hpi indicates that metabolic changes were minimal at this time point, only involving a small group of volatile compounds, glycine, ampelopsin D + quadrangularin A, trans-resveratrol, kaempferol-3-O-rutinoside, and pallidol. Several lipids and primary metabolites were highly modulated at 24 hpi, as already shown in **Figure 2**. No lipids were highly modulated at 48 hpi, and many polyphenols were highly modulated at 96 hpi. Moreover, it is apparent that ceramide and trans-piceid show a central position in the network, meaning that these compounds were highly modulated at all the time points. Some interesting results are represented by metabolites connected with two different colored links; these were different at two time points, as exemplified by transresveratrol (**Figure 3**).

To further investigate this interesting subset of compounds and as further check for both the selection criterion and the visualization proposed, we explored the trends of log10 concentration over time for some key metabolites. These plots show results consistent with the network representation (Supplementary Figure S3). Ceramide concentration in inoculated leaves was already higher at 12 hpi compared with controls and reached the highest concentration at 96 hpi (0.32 mg/kg; Supplementary Table S2). Trans-piceid concentration was already high at 12 hpi and reached the highest concentration at 48 hpi (5.29 mg/kg) (Supplementary Table S2). Among the polyphenols, trans-ε-viniferin was the compound that was modulated earliest at 48 hpi, while other trimeric and tetrameric stilbenoids, such as ampelopsin H + vaticanol C, pallidol, ampelopsin D + quadrangularin A, Z-miyabenol C and α-viniferin for example, were modulated at 96 hpi (Supplementary Figure S3). Accumulation of trans-resveratrol occurred early after infection (12 hpi), and was followed by a decrease in concentration at 24 and 48 hpi, and a resumption of accumulation at 96 hpi (Supplementary Figure S3).

A considerable number of compounds belonging to each class increased in concentration over time in both inoculated and not inoculated samples (Supplementary Table S2); this would explain the high variance in the first dimension of PCA, which is associated with the time course (**Figure 1**). The progressive accumulation of stress-related compounds in leaf discs, regardless pathogen inoculation, can be explained by other stresses affecting the tissues as a consequence of leaf removal, punching of the leaf lamina, and artificial conditions of leaf disc incubation. We found accumulation of some lipid compounds, such as arachidic acid, oleanolic acid, and uvaol, in inoculated

and control samples (Supplementary Table S2). In polyphenols we also observed an accumulation of flavonols and some trimers and tetramers belonging to the stilbene and stilbenoid class during the first 48 h, irrespective of pathogen infection, and then differentiation at 96 hpi (Supplementary Table S2). These results are consistent with previous reports of metabolite changes caused by mechanical wounding (Chitarrini et al., 2017).

# DISCUSSION

Grapevine and P. viticola interaction is still poorly understood in terms of metabolites: there is the need to improve the knowledge about how the plant system is perturbed after stress. In this study, a metabolomic approach has revealed major changes in primary and secondary metabolism of a resistant grape variety during the defense response to P. viticola. The identification of biomarkers, specific of four stages of the defense response, from 12 to 96 hpi, reflected a progression of physiological events that bring about resistance.

In addition to the importance of secondary metabolites in the fight against pathogens, the role of primary metabolism needs to be taken into account, since it is not only an energy provider but also regulates defense responses in plants in the presence of potential pathogens or pathogen-derived elicitors (Rojas et al., 2014). We expected that a defense response against an endophytic biotroph could be triggered only after the establishment of intimate contact between pathogen haustoria and host plasma membranes. In fact, we observed minimal metabolite changes in the host within 12 hpi, compatibly with a scenario in which P. viticola oospores/zoospores take several hours to germinate on the leaf lamina, target the stomata, form appressoria, break through the cell wall of mesophyll cells and develop functional haustoria. We identified a few biomarkers for this very early stage of host–pathogen interaction. Most of them were volatile compounds, which may interfere with the pathogen endophytic invasion of mesophyll air spaces.

The classes of biomarkers specific to 24 and 48 hpi suggested that early host responses to P. viticola were being set in place during those stages. We detected a sharp shift in primary metabolism. Leaf discs undergoing the defense response showed a cumulative amount of sugars, organic acids, and amino acids 6.7% higher than controls at 12 hpi, 9.4% higher at 24 hpi, 14.1% higher at 48 dpi, and 11% lower at 96 dpi. These data suggest that most of the metabolic effort for containing pathogen infection was carried out by 48 hpi and the metabolic cost for this effort was paid at 96 hpi. In leaf discs undergoing the defense response, organic acids were 13.3% higher than controls at 12 hpi and 24.6% higher at 24 hpi. Vice versa, sugars and amino acids were consistently lower at the same time points in leaf discs

undergoing the defense response (−2.3 and −12.1% sugars at 12 and 24 hpi, respectively; −8.9 and −19.9% amino acids at 12 and 24 hpi, respectively). At 48 hpi sugars and amino acids were 33.9 and 42.3% higher in leaf discs undergoing the defense response compared to controls. The cost of expressing defense has been shown in barley as a peak of respiration rate during the expression of host resistance to Blumeria graminis (Brown and Rant, 2013).

Primary metabolism is important for energy supply but it also has a role providing precursors of secondary metabolites, building blocks of PR proteins, and components of the defense signaling cascade (Rojas et al., 2014). Less et al. (2011) found different regulation of specific genes in Arabidopsis related to primary metabolism, due to abiotic and biotic stress response; in particular, up-regulation of genes involved in energy production processes and down-regulation of genes associated with assimilatory processes was found (Less et al., 2011). We found changes in primary compounds at 24 and 48 hpi, in particular we observed an interesting modulation for proline. In Arabidopsis, both supply and catabolism of proline are components of salicylic acid-mediated resistance, contributing to cell death in response to Pseudomonas (Deuschle et al., 2004; Cecchini et al., 2011). The role of proline in the Bianca grapevine variety after P. viticola infection should be elucidated with further experiments, however, for the moment we can identify this molecule as a putative biomarker.

Lipids represent a class of compounds with structural diversity and complexity. They are critical components of plant cell membranes and provide energy for metabolic activities. We found changes at 24 hpi in particular with a faster decrease in some unsaturated fatty acids after P. viticola infection. Ceramide started accumulating very early in infected samples compared to the control, and continued to accumulate after biotic stress up to 96 hpi; it was previously reported that ceramides can be essential as signaling molecules in the activation of defense-related plant programmed cell death (Kachroo and Kachroo, 2009; Berkey et al., 2012).

Subsequently, secondary metabolism was affected more strongly by the pathogen, with changes in the volatile compounds at 48–96 hpi and at the latest at 96 hpi in phenolic compounds. Some phenolic compounds, such as phenylpropanoids and flavonoids, have previously been identified and considered responsible for distinguishing the resistant cultivar Regent from the susceptible Trincadeira (Ali et al., 2012). We found higher concentrations of these compounds in our infected samples compared with the control at 96 hpi (Supplementary Table S2); this result suggests their involvement as biomarkers of

resistance to the pathogen in the Bianca grapevine. Transresveratrol production in grapevine leaves after pathogen infection was identified by Langcake and Pryce (1977). It has been demonstrated that trans-resveratrol is a precursor of fungal toxicity compounds identified as phytoalexins; these compounds can be produced by grapevine leaves after abiotic and biotic stress and can be used in the grapevine as a marker of resistance against pathogens (Jeandet et al., 2002). The accumulation of trans-resveratrol at 12 hpi in our infected samples can reflect the role of this molecule as a precursor of other toxic molecules, and the very early trans-resveratrol accumulation in Bianca is probably due to a rapid response to the pathogen. In our study we found a major increase in some molecules deriving from resveratrol, such as trans-ε-viniferin at 48 hpi and subsequently trans- and cis-piceid, isorhapontin, ampelopsin H + vaticanol C-like isomer, α-viniferin and pallidol at 96 hpi. During the first hpi, a low accumulation of viniferins (grapevine specific stress related metabolites) was found after pathogen infection, probably due to their accumulation at a later stage (4–7 days after inoculation), as previously described (Pezet et al., 2004; Jean-Denis et al., 2006; Slaughter et al., 2008). The time course of accumulation of these viniferins is in full agreement with several experiments reviewed by Bavaresco et al. (2012), beginning with the synthesis of resveratrol and progressing with the formation of dimers and then the higher oligomers. Such a path requires growth through the subsequent addition of one resveratrol unit to an existing dimer, leaving one part of the initial structure unchanged. The biosynthesis of dimers and higher oligomers appears to be important for resistance, in agreement with the observations of Malacarne et al. (2011) in a segregating population of Merzling × Teroldego. This paper indeed highlighted a negative correlation between the content of different oligomers and the percentage of sporulation upon infection, while this was not the case for the monomers trans-resveratrol and transpiceid, which were also found in sensitive genotypes with high sporulation. Moreover, the importance of viniferin oligomers is further confirmed by the concentration values required to induce inhibition of mildew development recently reported by Gabaston et al. (2017).

In our study, the peak of accumulation of phenolic compounds at 96 hpi was anticipated at 48 h by the accumulation of phenylalanine in inoculated samples (**Figure 3** and Supplementary Table S2). Phenylalanine is the precursor of the phenylpropanoid pathway, leading to the synthesis of flavonoids and stilbenes by stilbenes, two classes of compounds that increased at 96 (Sparvoli et al., 1994; Flamini et al., 2013).

Among the volatile compounds, we found an increase in benzaldehyde production at 48 and 96 hpi in inoculated samples (Supplementary Figure S2) Benzaldehyde is considered as a growth suppressor and spore inhibitor, with activity against Botrytis cinerea, also at a low concentration (Martínez, 2012). Benzaldehyde also promotes salicylic acid accumulation, induces expression of PR proteins and increases TMV resistance in tobacco (Ribnicky et al., 1998). The higher concentration we found in infected Bianca samples at 48 and 96 hpi (around 1.5 times higher compared to the control) suggests its involvement as a putative biomarker against P. viticola growth or diffusion.

Based on our results, we can argue that all the compounds significantly differentiated in infected samples have a role in Bianca-P. viticola interaction. In particular, 53 metabolites have been identified as putative biomarkers in hybrid Bianca grapevine leaves after P. viticola infection. Some of them are known biomarkers of resistance (viniferins). Among the others, some are likely to be putative biomarkers of resistance in Bianca leaf discs after P. viticola infection, such as benzaldehyde and proline.

To the best of our knowledge, this is the first time that an extensive metabolomic study has been undertaken using a resistant grape variety to better understand metabolic perturbation after P. viticola infection, finding early stage biomarkers for different chemical classes of metabolites. These results can represent a starting point for better understanding grapevine resistance and can lead to discoveries regarding new mechanisms for plant–pathogen interaction between the grapevine and P. viticola.

We also obtained good results for method validation in relation to the identification and quantification of 97 primary compounds belonging to different chemical classes: acids, amino acids, amines, sugars, and fatty acids, using a GC-MS method for separation and identification. The method can easily be applied to further analysis for the identification and quantification of primary compounds in different matrices.

# AUTHOR CONTRIBUTIONS

GC, LZ, AV, MS, GD, FM, and UV designed the experiment. GC, GD, LZ, and AV performed the experiment. GC, ES, and DM did the extractions, analytical analysis, and data treatment. GC, ES, and DM developed and validated the method. SR and PF conducted all the statistical analyses. All authors discussed the results and implications and commented on the manuscript at all stages.

# FUNDING

This research was supported by the FIRST International Ph.D. program and ADP 2016 project founded by the Autonomous Province of Trento.

# ACKNOWLEDGMENT

Cesare Lotti is acknowledged for his assistance in GC-MS analysis.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fpls.2017.01524/ full#supplementary-material

#### REFERENCES

fpls-08-01524 September 2, 2017 Time: 15:7 # 10


Franco-American hybrid. Acta Hortic 603, 457–463. doi: 10.17660/ActaHortic. 2003.603.58


contribution to plant defense. Front. Plant Sci. 5:17. doi: 10.3389/fpls.2014. 00017


**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 Chitarrini, Soini, Riccadonna, Franceschi, Zulini, Masuero, Vecchione, Stefanini, Di Gaspero, Mattivi and Vrhovsek. 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.

# Identification of Lipid Markers of *Plasmopara viticola* Infection in Grapevine Using a Non-targeted Metabolomic Approach

Lise Negrel, David Halter, Sabine Wiedemann-Merdinoglu, Camille Rustenholz, Didier Merdinoglu, Philippe Hugueney and Raymonde Baltenweck\*

SVQV, Institut National de la Recherche Agronomique, Université de Strasbourg, Colmar, France

The Oomycete Plasmopara viticola is responsible for downy mildew, which is one of the most damaging grapevine diseases. Due to the strictly biotrophic way of life of P. viticola, its metabolome is relatively poorly characterized. In this work, we have used a mass spectrometry-based non-targeted metabolomic approach to identify potential Plasmopara-specific metabolites. This has led to the characterization and structural elucidation of compounds belonging to three families of atypical lipids, which are not detected in healthy grapevine tissues. These lipids include ceramides and derivatives of arachidonic and eicosapentaenoic acid, most of which had not been previously described in Oomycetes. Furthermore, we show that these lipids can be detected in Plasmopara-infected tissues at very early stages of the infection process, long before the appearance the first visible symptoms of the disease. Therefore, the potential use of these specific lipids as markers to monitor the development of P. viticola is discussed.

Keywords: grapevine, downy mildew, metabolomics, lipids, biomarkers

# INTRODUCTION

Plasmopara viticola is an obligate biotrophic Oomycete responsible for downy mildew, which is one of the most damaging grapevine diseases. As downy mildew outbreaks may result in significant economic losses, control of the disease is most often achieved by fungicide treatments, whose impacts on environment and health raise more and more concerns. The relationships between grapevine and P. viticola have therefore been subjected to thorough investigations in order to better understand the infection process and to identify potential targets for alternative control strategies (Gessler et al., 2011). Grapevine responses to downy mildew infection have been characterized in both compatible and incompatible situations, through the study of susceptible Vitis vinifera cultivars and resistant species such as Vitis riparia, V. rupestris, or Muscadinia rotundifolia. Transcriptomic analyses have shown that downy mildew infection induces a strong and rapid transcriptional reprogramming in host tissues, including the induction of pathogenesisrelated proteins and enzymes required for the synthesis of phenylpropanoid-derived compounds (Kortekamp, 2006; Polesani et al., 2008, 2010; Wu et al., 2010; Gessler et al., 2011; Legay et al., 2011; Lenzi et al., 2016). Similarly, metabolomic profiling experiments have revealed profound changes in grapevine tissues upon downy mildew infection, which affect both the primary and the secondary metabolism (Figueiredo et al., 2008; Ali et al., 2009; Buonassisi et al., 2017). One of the most prominent metabolic change is the biosynthesis of large amounts of stilbene phytoalexins

#### *Edited by:*

Andreia Figueiredo, Universidade de Lisboa, Portugal

# *Reviewed by:*

Luigi Bavaresco, Università Cattolica del Sacro Cuore, Italy Michele Perazzolli, Fondazione Edmund Mach, Italy

#### *\*Correspondence:*

Raymonde Baltenweck raymonde.baltenweck@inra.fr

#### *Specialty section:*

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

> *Received:* 12 November 2017 *Accepted:* 05 March 2018 *Published:* 21 March 2018

#### *Citation:*

Negrel L, Halter D, Wiedemann-Merdinoglu S, Rustenholz C, Merdinoglu D, Hugueney P and Baltenweck R (2018) Identification of Lipid Markers of Plasmopara viticola Infection in Grapevine Using a Non-targeted Metabolomic Approach. Front. Plant Sci. 9:360. doi: 10.3389/fpls.2018.00360 in both compatible and incompatible interactions (Pezet et al., 2003; Alonso-Villaverde et al., 2011), specific patterns of stilbene accumulation being associated with increased resistance to the pathogen (Malacarne et al., 2011; Duan et al., 2015; Chitarrini et al., 2017). However, metabolomic analyses of grapevine-downy mildew interactions have mostly resulted in the characterization of metabolites from grapevine, the metabolome of P. viticola being paradoxically relatively poorly characterized. This is probably partly due to the strictly biotrophic way of life of P. viticola, which thus cannot be obtained in pure culture, making the characterization of its metabolome difficult.

In this work, we have used a high-resolution mass spectrometry-based non-targeted metabolomic approach to characterize metabolites differentially accumulated in grapevine and in P. viticola. The rationale behind this strategy was to look for prominent ions present in P. viticola and absent in grapevine healthy tissues, which may be derived from Plasmopara-specific molecules. This has led to the characterization and structural elucidation of compounds belonging to three families of atypical lipids, which are not detected in healthy grapevine tissues. Furthermore, we show that these lipids can be detected in Plasmopara-infected tissues from very early stages of the infection process. Therefore, the potential use of these specific lipids as markers to monitor the development of P. viticola is discussed.

# MATERIALS AND METHODS

#### Chemicals and Standards

LC-MS grade methanol and chloroform were from Roth Sochiel (Lauterbourg, France), water was provided by a Millipore water purification system. Arachidonic acid, eicosapentaenoic acid and N-Arachidoyl-D-sphingosine Cer(d18:1/20:0) were purchased from Sigma–Aldrich (Saint-Quentin Fallavier, France). N-Hexadecanoyl-D-erythro-C16-sphingosine Cer(d16:1/16:0) was from Abcam (Paris, France). Triarachidonoyl-glycerol and trieicosapentaenoyl-glycerol were purchased from Larodan Fine Chemicals (Malmö, Sweden). AG, DAG, EPG, and DEPG were obtained from AA and EPA by ZnCl2-catalyzed esterification of glycerol (Spring and Haas, 2004; Mostafa et al., 2013).

#### Plant Material and Downy Mildew Inoculation

The isolate SC of P. viticola used in this work and the leaf disc bioassay were described previously (Peressotti et al., 2010). Inoculum (sporangia) was prepared from infected leaves of V. vinifera cv Muscat Ottonel treated with Topsin 70 WG fungicide (50 mg/L) (Nisso Chemical, Germany), in order to avoid fungal contamination (Peressotti et al., 2010). Before inoculation leaves were surface-sterilized with bleach (2% active chlorine) for 2 min, followed by three washes with sterile water. In this work, the leaf disc bioassay was conducted with the following grapevine varieties and species: V. vinifera cv Cabernet-Sauvignon, cv Syrah, cv Cinsault, V. rupestris, V. riparia, and V. rotundifolia, using 15 leaf discs for each variety. Kinetic analysis of Plasmopara-specific lipids accumulation was performed on green cuttings of V. vinifera cv Syrah, the hybrid variety Bianca and V. riparia, grown in a greenhouse. The sixth leaf counted from the apex of 3.5 months old plants, were harvested, surface-sterilized, and washed with sterile water. Leaves were infected by flotation of their abaxial surface on a suspension of sporangia (10<sup>5</sup> sporangia mL−<sup>1</sup> ) for 5 h. Control leaves were floated on sterile water. Three leaves were used for each treatment. The leaves were then transferred to wet paper with the abaxial surface up and closed in transparent plastic bags. Leaves were stored in a growth chamber at 21◦C, with 16 h of light per day. 0, 24, 48, 72 h, and 6 days following inoculation, discs (diameter 1.5 cm) were punched from infected and control leaves, weighed, placed in a 2 mL tube, frozen into liquid nitrogen and stored at −80◦C until analysis.

#### Preparation of Sporangia Extracts From *P. viticola*

The abaxial faces of the leaves of 2 month-old Muscat Ottonel plants were sprayed with a suspension of sporangia (10<sup>5</sup> sporangia mL−<sup>1</sup> in water). Mock-inoculated leaves were sprayed with sterile water. Three inoculated and three control plants were then placed separately in growth chambers at 21◦C, with 16 h of light per day. Seven days later, two leaves per plant were carefully cut and randomly distributed in three groups. Sporulating leaves were shaken in 35 mL ultrapure water to recover sporangia. The three suspensions of sporangia were adjusted to 10<sup>5</sup> sporangia mL−<sup>1</sup> and 30 mL of each suspension was centrifuged at 30,000 g for 20 min. Pellets of sporangia were extracted with 2 mL of MeOH/CHCl<sup>3</sup> (1/1, v/v), by sonicating for 15 min in an ultrasound bath. Two mL of ultrapure water was added to allow phase separation. After centrifugation at 15,000 g for 10 min, the chloroform phase was concentrated to 100 µL and diluted with 100 µL of MeOH. Final extracts corresponded to 1.5.10<sup>7</sup> sporangia mL−<sup>1</sup> . Mock inoculated leaves were extracted as described below.

#### Metabolite Extraction From Grapevine Leaves

Leaf discs were freeze-dried and weighed. After grinding the discs using a bead mill (TissueLyser II, Qiagen, Courtaboeuf, France), metabolites were extracted with MeOH/CHCl<sup>3</sup> (1/1, v/v) using 15 µL per mg dry weight. The suspension of leaf powder in MeOH/CHCl<sup>3</sup> was then sonicated for 15 min in an ultrasound bath. Two milliliters of ultrapure water was added to allow phase separation. After centrifugation at 15,000 g for 10 min, 100 µL of the chloroform phase was recovered, diluted with 100 µL of MeOH and analyzed by UHPLC-MS.

### UHPLC-HRMS

Analyses were performed using a Dionex Ultimate 3000 UHPLC system (Thermo Fisher Scientific, San Jose, USA). The chromatographic separation was performed on a Nucleodur C18 HTec column (50 × 2 mm, 1.8µm particle size; Macherey-Nagel, Düren, Germany) maintained at 20◦C. The mobile phase consisted of methanol with formic acid (0.1%, v/v) in isocratic elution at a flow rate of 0.30 mL/min. The sample volume injected was 1 µL. The UHPLC system was coupled to an Exactive Orbitrap mass spectrometer (Thermo Fischer Scientific) equipped with an atmospheric pressure chemical ionization (APCI) source operating in positive mode. Parameters were set at 300◦C for ion transfer capillary temperature and the corona discharge current was set at 5 µA. Nebulization with nitrogen sheath gas and auxiliary gas were maintained at 15 and 10 arbitrary units, respectively, and the nebulizer temperature was maintained at 400◦C. The spectra were acquired within the m/z mass range of 100–1,200 atomic mass units (amu), using a resolution of 50,000 at m/z 200 amu. The system was calibrated internally using dibutylphthalate as lock mass at m/z 279.1591, giving a mass accuracy lower than 1 ppm. The instruments were controlled using the Xcalibur software and data was processed using the XCMS software package (Smith et al., 2006). Raw data were converted to the mzXML format using MSconvert before analysis. Settings of the xcmsSet function of XCMS were as follows: method = "centWave", ppm = 2, noise = 50,000, mzdiff = 0.001, prefilter = c(5,15,000), snthresh = 6, peakwidth = c(6,35). Peaks were aligned using the obiwarp function using the followings settings of the function group.density: bw = 10, mzwid = 0.0025. This allowed the alignment of 2,156 peaks in the positive mode. Ion identifiers were generated automatically by XCMS as MxxxTyyy, where xxx is the m/z and yyy the retention time in seconds. Selection of major differential ions (peak area > 10<sup>6</sup> , fold > 15) between sporangia and leaf extracts led to a final set of 123 major ions. The integration of each peak was checked manually before validation. T-tests were performed in R with the t.test function. Differential ions with p ≤ 0.05 were considered for further characterization. The exact m/z and retention time of each metabolite were used for targeted metabolomic analyses using the Excalibur software. For kinetic analysis of P. viticola development, seven major Plasmopara-specific lipids [AA, DAG, EPA, DEPG, TEPG, Cer(d16:1/18:0), Cer(d16:1/20:0)] were selected.

#### RESULTS

#### Characterization of Candidate *Plasmopara*-Specific Metabolites Using Non-targeted Metabolomics

As an obligate biotrophic pathogen, P. viticola cannot be grown in pure culture. However, sporangia can easily be collected from infected grapevine leaves. In order to characterize potential Plasmopara-specific metabolites, we compared extracts of sporangia and uninfected grapevine leaves using ultra high-performance liquid chromatography coupled to highresolution mass spectrometry (UHPLC-HRMS). This differential metabolomic approach has led to the selection of a set of 123 major ions (peak area > 10<sup>6</sup> , p ≤ 0.05), which are at least 15 times more abundant in sporangia extracts than in leaf extracts (Supplementary Table 1). The level of contamination of sporangia extracts by leaf-derived material was estimated by quantifying chlorophyll a. Chlorophyll a content in sporangia extracts (1.7.10<sup>4</sup> ± 1.2.10<sup>4</sup> arbitrary units) was 4,000 times lower than in leaf extracts (6.8.10<sup>7</sup> ± 4.1.10<sup>7</sup> arbitrary units), indicating very low levels of contamination. Submission of these 123 ions to the Metlin database (Smith et al., 2005) led to the putative identification of two compounds of particular interest. Indeed, one of the proposed structures for compounds giving the ion n ◦ 10 (M303T44) with the mass-to-charge ratio (m/z) 303.2316 was eicosapentaenoic acid (EPA). Similarly, ion n◦ 12 (M305T48) with m/z 305.2473 was potentially associated to arachidonic acid (AA) (**Table 1**). Both EPA and AA had previously been identified in mycelial extracts of Phytophthora infestans, another phytopathogenic Oomycete causing the late blight of potato (Bostock et al., 1981). Commercial EPA and AA standards were therefore analyzed by UHPLC-HRMS using the same conditions as those used for P. viticola extracts. The retention times (RTs) and fragment patterns of the standards were identical to those of M303T44 (n◦ 10) and M305T48 (n◦ 12) (Supplementary Figure 1), confirming their identities as EPA and AA, respectively.

# Structural Elucidation of EPA-Containing Lipids

In both the P. viticola extracts and the commercial standard, in source fragmentation of EPA produced a characteristic ion with m/z 285.2212 (C20H29O) corresponding to a loss of water. Extracted ion chromatogram (EIC) for m/z 285.2212 showed that other metabolites gave rise to this particular ion in P. viticola sporangia extracts (**Figure 1**), indicating that these molecules may contain EPA. A more detailed analysis of the mass spectra of some of these metabolites showed that they were actually present in our list of major Plasmopara-specific ions (**Figure 1**, **Table 1**). This prompted us to characterize these putative EPAcontaining compounds further. The mass and the formula of some of these compounds indicated that they may correspond to glycerides, constituted by glycerol (C3H8O3) partially or totally esterified by EPA. For example, the ion M377T39 (n◦ 5) with m/z 377.2683, corresponding to the formula C23H37O<sup>4</sup> (average mass error, AME 0.8 ppm), was putatively identified as eicosapentaenoyl-glycerol (EPG). Indeed, it gave rise to a first dehydration fragment with m/z 359.2579 (C23H35O3), as well as to the characteristic dehydration fragment of EPA (m/z 285.2210). The first fragment corresponded to the ion M359T39 (n◦ 4). The same reasoning could be applied to the ions M661T85 (n◦ 16) and M946T474 (n◦ 117), which where therefore putatively identified as dieicosapentaenoyl-glycerol (DEPG) and trieicosapentaenoyl-glycerol (TEPG), respectively. The identification of EPA glycerides was confirmed by comparing their characteristics to those of the corresponding synthetic standards (see below). Altogether, EPA and EPA glycerides give rise to 15 of 123 the major Plasmopara-specific ions selected in Supplementary Table 2.

# Structural Elucidation of AA-Containing Lipids

In source fragmentation of the AA standard gave rise to a characteristic ion with m/z 287.2369 (C20H31O, AME 0.1 ppm) corresponding to a loss of water. Using the same strategy as with EPA, EIC for m/z 287.2369 revealed several potential AA-containing molecules in P. viticola sporangia extracts (Supplementary Figure 2), some of them being listed among the Plasmopara-specific compounds (**Table 1**). Similarly, the formula

TABLE 1 | Plasmopara-specific ions identified by non-targeted metabolomics, which are referred to in the text.


Major compounds are indicated in bold. Identifications confirmed using the corresponding standards are marked by a gray background. Ions are numbered according to their retention time and are organized as derivatives of EPA, AA, and ceramides, indicated in green, blue, and red, respectively. Putative identifications are indicated with italics. For all ions, the identifier, the m/z, and retention time (RT) are indicated. For fragment nomenclature, see Supplementary Figure 3. The original list of ions identified by non-targeted metabolomics on one experiment including three biological replicates of extracts of P. viticola sporangia and of mock-inoculated leaves is presented in Supplementary Table 1 and the full list of 51 ions identified in this work is presented in Supplementary Table 2.

and the fragmentation pattern of some of them indicate that they may be AA glycerides. For example, the ion M379T42 (n◦ 8, **Table 1**) with m/z 379.2839 (C23H39O4, AME 1 ppm) could be putatively identified as arachidonoyl-glycerol (AG), as it produced a dehydration product with m/z 361.2736 (C23H37O3, AME 0.1 ppm) together with the characteristic dehydration ion of AA (m/z 287.2369, C20H31O). Applying the same reasoning, the ion M666T124 (n◦ 30) was putatively identified as diarachidonoyl-glycerol (DAG). As TEPG was abundant in P. viticola sporangia, we logically looked for triarachidonoylglycerol (TAG), although no ion potentially derived from this compound was found in **Table 1**. TAG was indeed detected, albeit with a retention time of 21.81 min, exceeding the 10 min limit originally set for the XCMS analysis, and therefore explaining its absence in **Table 1**. In order to confirm the identity of the fatty acid glycerides detected in P. viticola extracts, the corresponding standards were synthesized by ZnCl2-catalyzed esterification of glycerol (Mostafa et al., 2013). Using EPA and AA, mixtures of glycerides of the corresponding fatty acids were obtained and analyzed by UHPLC-HRMS using the same conditions as those used for sporangia extracts. The synthetic AG, DAG, EPG, and DEPG exhibited the same RT and fragment patterns as the corresponding compounds detected P. viticola extracts, therefore confirming their identity. The identities of the triglycerides were confirmed using the corresponding commercial TAG and TEPG standards.

#### Characterization and Structural Elucidation of Further *Plasmopara*-Specific Lipids

Among putative Plasmopara-specific metabolites, the ion with m/z 236.2371 (putative formula C16H30N) appeared three times at different RTs (ions n◦ 33, 66, and 93, at RT 125, 165, and 223 s, respectively; **Table 1**). For all RTs, this ion seemed to belong to a group of ions corresponding to fragments of larger metabolites. Indeed, EIC for m/z 236.2371 revealed four different peaks at RTs 96, 125, 165, and 223 s (**Figure 2A**). Analysis of the major ions present at these RTs revealed four compounds, which are also present in **Table 1**, namely M492T96, M521T125, M549T166, M577T223 (n◦ 19, n◦ 37, n◦ 71, n◦ 95). Each of these ions exhibited a mass difference of 28.031 atomic mass units (amu) with the

following one, this mass increase being compatible with their belonging to a family of compounds that differentiate themselves by an increasing number of ethylene (C2H4) groups. Therefore, the putative formula of M492T96 (n◦ 19) was C32H62O2N, and the following metabolites corresponded to the putative formulas C34H66O2N, C36H70O2N, C38H74O2N, respectively. Further analysis of the mass spectra revealed that, alongside the major ions listed above, we could detect minor ions corresponding to the same formula with either a loss of water or one additional water molecule (H2O). The latter ions corresponded to m/z 510.4877, 538.5189, 566.5502, and 594.5818, respectively. Search of the Metlin database for compounds possibly corresponding to these ions (with 1 ppm AME) retrieved candidates belonging to the ceramide family, composed of a C16:1 or C18:1 sphingosine, linked to a C14, C16, C18, C20, or C22 saturated fatty acid (Supplementary Table 3). In order to validate these potential candidates, a commercial standard of the ceramide N-arachidoyl-D-sphingosine Cer(d18:1/20:0), which may correspond to the ion M577T223 (n◦ 95), was analyzed by UHPLC-HRMS using the same conditions as those used for P. viticola extracts. Its mass spectrum showed a [M+H]<sup>+</sup> pseudo molecular ion with m/z 594.5814 corresponding to C38H76O3N, together with several fragments with m/z 264.2683 (C18H36N), m/z 282.2791 (C18H38ON) and m/z 576.5713 (C38H74O2N), which was the major fragment. Ions coming from the fatty acid part were detected too (m/z 312.3258, C20H42ON and m/z 354.3363, C22H44O2N), in agreement with the fragmentation of ceramides described previously in several studies (Levery et al., 2000; Hsu and Turk, 2002; Murphy, 2015) and obtained with this standard (Supplementary Table 3). However, neither its RT nor

its fragmentation corresponded to any of the ions found in P. viticola, excluding the hypothesis that some of these ions may be derived from C18:1 sphingosines.

A second set of candidates retrieved from the Metlin database contained C16:1 sphingosine. Indeed, the four compounds present in P. viticola extracts shared the same specific fragments with m/z 236.2371 (C16H30N) and 254.2476 (C16H32ON) (**Figure 2B**). The formulas of these fragments could be compatible with their being derived from a C16:1 sphingosine (Lhomme et al., 1990; Pivot et al., 1994). Therefore, a commercial standard of the ceramide Cer(d16:1/16:0) was analyzed by UHPLC-HRMS. Both its RT and its fragmentation pattern corresponded to that of the ion M492RT96 (n◦ 19) found in P. viticola (Supplementary Table 3). Furthermore, M492RT96 and Cer(d16:1/16:0) exhibited the same RT in three different chromatographic conditions (data not shown), showing that the ion M492RT96 (n◦ 19) is indeed derived from Cer(d16:1/16:0) present in P. viticola extracts (**Figures 2A**, **3**). The ions M492T96, M521T125, M549T166, and M577T223 (n◦ 19, n◦ 37, n◦ 71, n ◦ 95) all share common fragments with m/z 236.2373 and 254.2476, corresponding to C16H30N et C16H32ON, respectively, which are derived from the C16:1 sphingosine. As for the commercial standards Cer(d16:1/16:0) and Cer(d18:1/20:0), three ions coming from the fatty acid part are detected for all four ceramides (Supplementary Table 3). For Cer(d16:1/18:0) (n◦ 37), these characteristic fragments termed A, B, and C are indicated in Supplementary Figure 3. Therefore, the ions M521T125, M549T166, and M577T223 (n◦ 37, n◦ 71, n◦ 95) are very likely to be derived from a family or ceramides with a C16:1 sphingosyl group bound to a saturated fatty acid with 18–22 carbon atoms, namely Cer(d16:1/18:0), Cer(d16:1/20:0), and Cer(d16:1/22:0), respectively (**Figures 2A**, **3**). Detailed analysis of the fragments derived from these compounds is presented in Supplementary Table 3.

Altogether, we could propose an identification for 51 (Supplementary Table 2) of the final set of 123 major ions, which are at least 15 times more abundant in sporangia extracts than in leaf extracts (with p ≤ 0.05, Supplementary Table 1).

### *Plasmopara*-Specific Lipids Can Be Used for Phenotyping Resistance to Downy Mildew

The accurate quantification of pathogen development is important for a better understanding of plant-pathogen

interactions. In the case of P. viticola, methods based on cell counting or image analysis are routinely used to quantify sporulation, especially as a phenotyping tool to characterize resistance sources in the Vitis genus (Peressotti et al., 2011; Divilov et al., 2017). Our differential metabolomic approach has led to the characterization of lipids, which are abundant in the sporangia of P. viticola and not detected in healthy grapevine leaf tissues. In order to evaluate the possibility to use these compounds for phenotyping resistance to P. viticola, analysis of lipid markers was compared to image analysis for quantifying sporulation in a leaf disc assay routinely used to assess resistance to P. viticola (**Figure 4A**; Peressotti et al., 2011). Correlations between sporulation and quantification of all individual lipid markers were very good (**Figure 4B**), all associated p-values being lower than 10−<sup>9</sup> . This demonstrates the specificity of the lipid markers and confirms that they are perfectly suited for phenotyping resistance to P. viticola. In particular, EPA and AA are giving the best correlation coefficients, meaning that they are very valuable markers for phenotyping resistance.

# Early Detection and Kinetic Analysis of *Plasmopara*-Specific Lipids Accumulation During Downy Mildew Infection

Beside end-point quantification of sporulation, an accurate monitoring of disease progression is important for precise description of disease susceptibility phenotypes. Therefore, we evaluated the possibility to use Plasmopara-specific lipids to quantify P. viticola in grapevine tissues before sporulation, i.e., as early and dynamic markers of downy mildew infection. Comparison of infected and mock-inoculated leaves showed that all ceramides and fatty acid derivatives described above were easily detected in the extract of infected leaves before sporulation, all these compounds being absent in control leaves (**Figure 5**). This prompted us to evaluate the potential of the Plasmoparaspecific lipids to monitor the kinetic of P. viticola infection in three grapevines genotypes, which differ in their susceptibility to downy mildew: Syrah, Bianca, and V. riparia. Lipid analysis at different times post-inoculation allowed effective monitoring of disease development throughout the infection process, all lipids being efficiently detected even at early infection stages (24 h postinoculation; **Figure 6**, Supplementary Figure 4, Supplementary Table 4). Different patterns and timings of Plasmopara-specific lipids accumulation were obtained in grapevine varieties with different susceptibility to downy mildew. However, patterns of Plasmopara-specific lipid accumulated in infected grapevine leaves differed significantly from that of isolated sporangia.

# DISCUSSION

### *P. viticola* Contains Unusual Lipids, Which Are Not Detected in Healthy Grapevine Leaf Tissues

Looking for potential markers of downy mildew infection using a non-targeted metabolomic approach, we could identify derivatives of Plasmopara-specific lipids, which include EPA and AA-containing lipids, in addition to ceramides. Both EPA and AA have previously been isolated from extracts of P. infestans, as elicitors of the accumulation of antimicrobial stress metabolites in potato tubers (Bostock et al., 1981). EPA and AA have been detected in a number of other Oomycetes (Gandhi and Weete, 1991; Spring and Haas, 2002; Pang et al., 2015), and more recently, the role of EPA and AA as microbial associated molecular patterns (MAMPs) has been extended to other plant-Oomycete interactions (Savchenko et al., 2010; Bostock et al., 2011). Due its remarkable abundance in Plasmopara halstedii, EPA has been proposed as a marker for downy mildew contamination in sunflower seeds (Spring and Haas, 2004). Similarly, fatty acids, including AA, have been used to detect Phytophthora sojae in soybean fields (Yousef et al., 2012). However, lipid analysis in Oomycetes has mostly targeted fatty acids released after total lipid hydrolysis and the information on complex lipids is very scarce. Here, we have used UHPLC-HRMS to directly characterize complex lipids without the modification usually needed for GC-MS analysis (hydrolysis, derivatization). In particular, optimized mild atmospheric pressure chemical ionization (APCI) has allowed the identification of ions corresponding to the whole molecules, as well as characteristic fragments. The use of APCI for the characterization of complex lipids has been reported very recently (Mutemberezi et al., 2016), and this approach is likely to gain in popularity in the current context of the development of lipidomics.

UHPLC-HRMS has led to the characterization of two families of glycerides derived, respectively, from EPA and AA, which had not been formally identified in Oomycetes to date. Furthermore, occurrence of TAG and TEPG is poorly documented and very few natural sources of these triglycerides have been described. TAG has been reported in the red algae Gracilaria verrucosa (Kinoshita et al., 1986). Several organisms such as Mortierella sp. (Shimizu et al., 1988) and the Oomycetes Pythium sp. (Gandhi and Weete, 1991; O'Brien et al., 1993) and Saprolegnia sp. (Shirasaka and Shimizu, 1995) have been reported to produce EPA-containing triglycerides. However, to our knowledge, TEPG has not been formally identified in these organisms.

In addition to EPA and AA derivatives, sporangia of P. viticola contain a family of C16:1 ceramides including Cer(d16:1/16:0), whose identification could be confirmed using a commercial standard. Although we could not identify them formally due to the lack of standards, other major C16:1 ceramides of P. viticola are very likely to include Cer(d16:1/18:0), Cer(d16:1/20:0), and Cer(d16:1/22:0) (**Figure 3**). Several shingophospholipids containing a C16:1 sphingosine have been detected in different plant Oomycete pathogens (Lhomme et al., 1990; Kim et al., 1998; Moreau et al., 1998). However, Cer(d16:1/16:0), Cer(d16:1/18:0), Cer(d16:1/20:0), and Cer(d16:1/22:0) had never been reported in Oomycetes before this work.

None of the lipids characterized above were detected in significant amounts in healthy grapevine leaf tissues. In addition, using an array of grapevine varieties, hybrids, and Vitis species with different susceptibility to downy mildew, correlation between sporulation and quantification of individual lipid markers was excellent (**Figure 4**), thus demonstrating that, in the context of grapevine-downy mildew interaction, these lipids


FIGURE 4 | Correlation between image analysis-based quantification of sporulation and quantification of lipid markers of P. viticola. Leaf disc bioassay was performed with Cabernet-Sauvignon, Syrah, Cinsault, V. rupestris, V. riparia, and V. rotundifolia, using 15 leaf discs for each variety. (A) Typical quantification of sporulating area 6 days post inoculation using image analysis (Peressotti et al., 2011). (B) Correlation and associated p-value of individual lipid marker quantification with sporulating area, using Pearson's correlation test.

can be considered as Plasmopara-specific. Their potential use as markers of downy mildew infection was therefore investigated.

## Lipids of *P. viticola* Can Be Used as Early Markers of Downy Mildew Infection in Grapevine Leaves

Most of the lipids characterized in sporangia extracts could be detected at early stages of downy mildew infection, indicating that these lipids are also present in the mycelium of P. viticola (**Figure 5**). The presence of seven major Plasmopara-specific lipids could easily be detected 24 h post-inoculation (**Figure 6**), long before the first external symptoms, which usually start to appear 4 days post-inoculation on susceptible grapevine varieties (Smith et al., 2006; Unger et al., 2007; Peressotti et al., 2011). The patterns of lipid accumulation were different in susceptible and resistant grapevine varieties. The accumulation of Plasmoparaspecific lipids was unexpectedly higher 24 h post-inoculation in V. riparia than in the more susceptible varieties, suggesting

P. viticola. For easier comparison, infected (A) and control leaves (B) are presented at the same intensity scale (3.10<sup>6</sup> arbitrary units). Peaks corresponding to lipids of P. viticola are numbered from 1 to 6 and the corresponding retention times are indicated. Peak 1: AA, EPA, AG, EPG; peak 2: DAG, DEPG, Cer(d16:1/16:0); peak 3: Cer(d16:1/18:0); peak 4: Cer(d16:1/20:0); peak 5: Cer(d16:1/22:0); peak 6: TEPG.

a rapid development of hyphae from zoospores, even on the resistant variety (**Figure 6**). This is consistent with previous studies showing that P. viticola can form primary hyphae in the context of compatible, incompatible, and even non-host interactions (Díez-Navajas et al., 2008). However, the infection process was quickly stopped presumably as the result of plant defense reactions, the amount of detectable lipids reaching a plateau 24 h post-inoculation. Lipid accumulation in the fully susceptible variety Syrah was significantly higher than in Bianca, consistent with the partial resistance of this variety (Bellin et al., 2009).

Interestingly, the pattern of lipid accumulation was modified along the infection process. At early stages, AA and EPA were rather accumulated as fatty acids, whereas at late stages, the corresponding triacylglycerols, and especially TEPG, were massively accumulated (**Figure 6**), suggesting that the lipid pattern may be used as an indicator of infection developmental step. Indeed, the pattern of Plasmopara-specific lipids observed on Syrah leaves at late infection stage (6 days post inoculation) differed substantially from that of sporangia (**Figure 6**), although, at this stage, leaves were covered with an intense downy sporulation. This suggests that, even at the sporulation stage, the major part of Plasmopara-specific lipids comes from mycelial material localized inside grapevine leaves, this material being characterized by its high TEPG content. Altogether, these results show that Plasmopara-specific lipids may be used to monitor the development of downy mildew throughout the infection process, in both susceptible and resistant grapevine varieties.

#### Future Development of Mass Spectrometry-Based Methods for Monitoring Early Steps of Grapevine-Downy Mildew Interaction

In this work, we have used a non-targeted metabolomic approach to characterize unusual P. viticola-specific lipids. Although AA, EPA, and shingophospholipids have been found in Oomycetes before (Bostock et al., 1981; Lhomme et al., 1990; Pivot et al., 1994; Judelson, 2017), the major triglycerides TAG and TEPG and the ceramide family characterized in this

Bianca (partially resistant) and V. riparia (resistant). (A) Quantification of the 7 major lipids 0, 24, 48, 72 h, and 6 days post inoculation (dpi) with P. viticola. As a comparison, a typical lipid profile of sporangia of P. viticola is indicated. (B) Enlargement for early infection times (0–48 h). Three biological replicates were analyzed for each condition and means are presented. A detailed analysis of Plasmopara-specific lipids in these samples is presented in Supplementary Figure 4 and the associated statistics are presented in Supplementary Table 4.

work had not been unambiguously identified in Oomycetes. Beyond the identification of new lipids from P. viticola, we show that some of these compounds can be used to efficiently monitor the infection process of downy mildew, long before the appearance of any external symptom of the disease. The availability of reliable methods for analysis of early steps of pathogen development contributes to a better understanding of plant–pathogen interactions. Genetic analysis of quantitative resistance to plant pathogens require methods allowing a precise quantification of pathogen development for a reliable detection of the genomic regions involved in resistance. Assessment of resistance to grapevine downy mildew has been traditionally performed by visually scoring disease symptoms or by measuring sporulation using a cell counter (Bellin et al., 2009). More recently, high-throughput imaging-based methods to quantify sporulation have been developed (Peressotti et al., 2011; Divilov et al., 2017). Although such high-throughput methods are very useful to screen breeding populations, they do not allow monitoring of disease progression. Pathogen growth can be monitored by quantitative PCR-based methods (Brouwer et al., 2003; Anderson and McDowell, 2015). These methods are sensitive, however, they require costly reagents and rather complicated sample preparation. Pre-symptomatic detection of downy mildew can be achieved using chlorophyll fluorescence imaging (Cséfalvay et al., 2009), but this approach is not quantitative. Through the characterization of lipid markers of P. viticola, we provide targets for the development of LC-MSbased methods allowing a sensitive, continuous and quantitative monitoring of downy mildew progression inside infected leaves. These methods will be likely to been successfully applied to a wide range of grapevine genotypes, including varieties and Vitis species. Optimization of chromatographic conditions will allow the development of fast (5 min or less) separation methods targeting the most relevant compounds, compatible with medium to high-throughput quantification. Finally, the same strategy could be applied to other important Oomycete or fungal plant pathogens, in order to develop high-sensitivity LC-MS-based methods for quantification of pathogen growth and disease progression. These methods could be used for accurate analysis of host-pathogen interaction and for the assessment of resistance to fungicides.

# AUTHOR CONTRIBUTIONS

DM and PH: conceived the original research plans and raised the funding; RB: supervised experiments and identifications of unknown compounds; LN: performed most of the experiments; DH and CR: participated in data and statistical analyses; SW-M: supervised the experiments with downy mildew; LN, DM, PH, and RB: wrote the article with contributions of all the authors.

#### REFERENCES


## FUNDING

LN was supported by a doctoral fellowship from the Conseil Interprofessionnel des Vins d'Alsace, the Conseil Interprofessionnel du Vin de Bordeaux, and the Comité Interprofessionnel du Vin de Champagne and from the Région Alsace. This work was co-funded by the European Regional Development Fund (ERDF) in the framework of the INTERREG IV and V Upper Rhine programs (BACCHUS and VITIFUTUR projects), and by the Fondation Jean Poupelain (Javrezac, France) through the HealthyGrape2 program.

# ACKNOWLEDGMENTS

We thank the staff of UE0871 Service d'Expérimentation Agronomique et Viticole (SEAV, INRA, Colmar) for maintenance of plant material. We are grateful to Marie-Annick Dorne and Marie-Céline Lacombe (INRA, Colmar) for help with downy mildew infections and to Andy Schneider (INRA, Colmar) for help with lipid extraction and analysis. We thank Patrick Schaub, Florian Wüst, and Peter Beyer (University Freiburg, Germany) for MS2 analyses.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2018. 00360/full#supplementary-material


mildew upon limited deployment of a resistant variety. BMC Plant Biol. 10:147. doi: 10.1186/1471-2229-10-147


**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 Negrel, Halter, Wiedemann-Merdinoglu, Rustenholz, Merdinoglu, Hugueney and Baltenweck. 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.