# ORCHID GENOMICS AND DEVELOPMENTAL BIOLOGY

EDITED BY : Jen-Tsung Chen and Katharina Nargar 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 978-2-88963-975-5 DOI 10.3389/978-2-88963-975-5

#### 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

# ORCHID GENOMICS AND DEVELOPMENTAL BIOLOGY

Topic Editors:

Jen-Tsung Chen, National University of Kaohsiung, Taiwan Katharina Nargar, Commonwealth Scientific and Industrial Research Organisation (CSIRO), Australia

Citation: Chen, J.-T., Nargar, K., eds. (2020). Orchid Genomics and Developmental Biology. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-975-5

# Table of Contents


Maria Carmen Valoroso, Rómulo Sobral, Giuseppe Saccone, Marco Salvemini, Maria Manuela Ribeiro Costa and Serena Aceto

*34 Maltose Processing and Not* b*-Amylase Activity Curtails Hydrolytic Starch Degradation in the CAM Orchid* Phalaenopsis

Nathalie Ceusters, Mario Frans, Wim Van den Ende and Johan Ceusters

*46 Dissecting the Function of MADS-Box Transcription Factors in Orchid Reproductive Development*

Zhi Wei Norman Teo, Wei Zhou and Lisha Shen


Mariana Ferreira Alves, Fabio Pinheiro, Marta Pinheiro Niedzwiedzki and Juliana Lischka Sampaio Mayer


Young-Kee Kim, Sangjin Jo, Se-Hwan Cheon, Min-Jung Joo, Ja-Ram Hong, Myounghai Kwak and Ki-Joong Kim

#### *163 Corrigendum: Plastome Evolution and Phylogeny of Orchidaceae, With 24 New Sequences*

Young-Kee Kim, Sangjin Jo, Se-Hwan Cheon, Min-Jung Joo, Ja-Ram Hong, Myounghai Kwak and Ki-Joong Kim

#### *164 Phylogeny and Historical Biogeography of* Paphiopedilum *Pfitzer (Orchidaceae) Based on Nuclear and Plastid DNA*

Chi-Chu Tsai, Pei-Chun Liao, Ya-Zhu Ko, Chih-Hsiung Chen and Yu-Chung Chiang

# Editorial: Orchid Genomics and Developmental Biology

Jen-Tsung Chen1\* and Katharina Nargar 2,3

<sup>1</sup> Department of Life Sciences, National University of Kaohsiung, Kaohsiung, Taiwan, <sup>2</sup> Australian Tropical Herbarium, James Cook University, Cairns, QLD, Australia, <sup>3</sup> National Research Collections Australia, Commonwealth Industrial and Scientific Research Organisation (CSIRO), Canberra, ACT, Australia

Keywords: orchid, biotechnology, functional genomics, developmental biology, evolution

#### Editorial on the Research Topic

#### Orchid Genomics and Developmental Biology

Orchidaceae is the second largest family of flowering plants with more than 27,000 species inhabiting nearly every habitat worldwide. Orchids exhibit an exceptional morphological and ecological diversity, and are highly valued on the global horticultural market. Orchids possess unique morphological and physiological characteristics, such as highly reduced seeds with an immature embryo, complex flower structures such as the gynandrium and labellum, and have evolved crassulacean acid metabolism and mycoheterotrophy multiple times independently. This range of traits renders orchids prime non-model plants for studying different aspects of evolution through mechanistic studies considering gene function, physiology, and phylogenetic relationships. As summarized below, this Research Topic presents recent advances in orchid biology and consists of 12 publications in the fields of reproductive development, evolution, biotechnology, and photosynthesis.

#### Edited by:

Natalia Pabo´ n-Mora, University of Antioquia, Colombia

#### Reviewed by:

Serena Aceto, University of Naples Federico II, Italy Mariana Mondrago´ n-Palomino, University of Regensburg, Germany

#### \*Correspondence:

Jen-Tsung Chen jentsung@nuk.edu.tw

#### Specialty section:

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

> Received: 13 May 2020 Accepted: 22 June 2020 Published: 10 July 2020

#### Citation:

Chen J-T and Nargar K (2020) Editorial: Orchid Genomics and Developmental Biology. Front. Plant Sci. 11:1013. doi: 10.3389/fpls.2020.01013 FLOWER ORGAN DEVELOPMENT

The study of transcription factors (TFs), proteins that bind to the regulatory DNA sites of specific genes and are involved in the process of transcription, is an important tool for understanding flower development in orchids. Valoroso et al. studied the expression and interaction of three myeloblastosis (MYB) TFs that are part of a regulatory module, including DIVARICATA (DIV), RADIALIS (RAD), and DIV-and-RAD-Interacting-Factor (DRIF), on flower development of the Italian orchid, Orchis italica, native to the Mediterranean region. The results showed higher levels of gene expression in the labellum than in lateral inner tepals of the O. italica flower. In addition, yeast two-hybrid analysis revealed that the Orchis-MYB TF OitDRIF1 interacts with OitDIV and OitRAD. Hence, the results support the hypothesis that these three MYB TFs have a crucial role in controlling zygomorphy in orchid flowers.

Lai et al. investigated the function of a SHINE-Like TF, namely PeERF1, in several Phalaenopsis orchids. The authors conclude that PeERF1 is crucial for the development of labellum epidermis, particularly in the formation of nanoridges.

**5**

In model plants, MADS-box TFs have been shown to play crucial roles in diverse developmental processes. Based on a comprehensive review of the literature, Teo et al. summarize reproductive developmental processes in model plants including floral transition and patterning, and discuss unique aspects of orchid floral development such as the modified ABC model and the perianth code.

An overview on orchid flower development is provided by Wang et al. summarizing recent advances in our understanding of floral induction, transition, and flowering in popular orchid genera such as Dendrobium, Oncidium, and Phalaenopsis. The authors propose a hypothetical pathway for flowering regulation in P. aphrodite under low ambient temperature, and discuss models of the orchid code and perianth code in P. equestris and O. "Gower Ramsey".

The anatomical features of reproductive organs at different developmental stages of the mycoheterotrophic orchid Pogoniopsis schenckii was investigated by Alves et al. The authors compared characteristics of the seed coat in P. schenkii to photosynthetic species of the family. P. schenckii was found to have a seed coat that originates from the nucellar epidermis and not from the outer integument as in the other species.

#### EVOLUTION

Family-wide comparative studies on plastid structure and gene loss facilitate important insights into plastid evolution in Orchidaceae. Kim et al. compared structural variation and gene content of 124 complete orchid plastomes and investigated gene loss in orchids within a phylogenomic framework. The authors provide recommendations for additional molecular studies to increase our understanding of plastome evolution, in particular in mycoheterotropic orchids.

The deceptive flowers of Lady slipper orchids (subfamily Cypripedioideae) possess a slipper shaped labellum that traps insects to facilitate pollination. Tsai et al. investigated evolutionary relationships, intrageneric classification, and historical biogeography of the Venus slipper orchids, genus Paphiopedilum, based on nuclear (ITS) and three plastid markers (trnL intron, trnL-F, atpB-rbcL) for 78 taxa. The results supported the current intrageneric classification of the genus and provide insights into biogeographic origin and the spatiotemporal evolution of the genus.

Being one of the largest flowering plant families, orchids have evolved a great variety of pollination strategies. Fernández et al. profiled and compared floral transcriptomes of two European orchid genera with different pollination strategies, the sexually deceptive Ophrys and Gymnadenia with a food-rewarding pollination strategy. Further, phylogenomic analysis was carried out based on transcriptome data to resolve evolutionary relationships within the two genera.

#### BIOTECHNOLOGY

In orchids, protocorm-like bodies (PLBs) induced from explants or callus in vitro have long been recognized to resemble somatic embryos (SEs) in some species. Based on a molecular study, Chen et al.reported that PLBs do not derivefrom SEs at least in P. equestris. Further, the authors induced SEs by overexpressing P. aphrodite LEAFY COTYLEDON1 (PaLEC1) and suggested that induction of SEs may be a useful tool for clonal propagation of orchids.

Orchids rely on symbiotic relationships with fungi for seed germination and subsequent seedling growth. Sisti et al. characterized fungal communities in roots and fruits of a mycoheterotrophic orchid, P. schenckii, and carried out germination trials with non-mycorrhizal endophytic fungi isolated from P. schenckii under in vitro conditions. The authors observed that germination was stimulated by inoculation of P. schenckii seeds with the isolated non-mycorrhizal fungi.

#### PHOTOSYNTHESIS

The photosynthetic pathway known as crassulacean acid metabolism (CAM), found in different plant lineages including orchids, is adaptive in arid conditions due to the high water-use efficiency it confers on the plant. Ceusters et al. studied the metabolite and enzyme patterns of starch degradation in the CAM orchid Phalaenopsis "Edessa". Two routes of degradation were assessed: the hydrolytic route (involving cytosolic D-enzyme, plastid D-enzyme, maltase, and b-amylase), and the phosphorolytic route (involving starch phosphorylase) to compare enzyme activities as well as starch degradation rates. Ceusters et al. concluded that the phosphorolytic pathway might be the main route of starch degradation in CAM orchid Phalaenopsis.

#### CONCLUSION AND PERSPECTIVES

This Research Topic on Orchid Genomics and Developmental Biology provides only a snapshot of this exciting and growing field of research. We hope that with the rapid development of molecular tools and approaches, particularly several orchid genomes and hundreds of transcriptomes that are now available together and with high-throughput technologies and integrative multi-omics, emerging biotechnology such as CRISPR gene editing, researchers will be able to further unravel the secrets of orchid biology with emphasis on mechanistic insights into flowering regulation, flower organ development, embryo/protocorm formation and development, phylogenomics, symbiosis, and photosynthetic pathways, in the future.

We greatly appreciate the invaluable contribution of all authors and reviewers as well as the efforts of Chief Editor Neelima Roy Sinha of Frontiers in Plant Science: Plant Development and EvoDevo.

#### AUTHOR CONTRIBUTIONS

J-TC and KN drafted the manuscript. All authors contributed to the article and approved the submitted version.

Conflict of Interest: 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 © 2020 Chen and Nargar. 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.

# Floral Induction and Flower Development of Orchids

*Shan-Li Wang1‡, Kotapati Kasi Viswanath2‡, Chii-Gong Tong1, Hye Ryun An3, Seonghoe Jang4\*†, Fure-Chyi Chen2\*†*

*1 Biotechnology Center in Southern Taiwan (BCST) of the Agricultural Biotechnology Research Center (ABRC), Academia Sinica, Tainan, Taiwan, 2 Department of Plant Industry, National Pingtung University of Science and Technology, Pingtung, Taiwan, 3 National Institute of Horticultural and Herbal Science (NIHHS), Rural Development Administration (RDA), Wanju-gun, South Korea, 4 World Vegetable Center Korea Office (WKO), Wanju-gun, South Korea*

#### *Edited by:*

*Jen-Tsung Chen, National University of Kaohsiung, Taiwan*

#### *Reviewed by:*

*Lee Jeong Hwan, Chonbuk National University, South Korea Federico Valverde, Institute of Plant Biochemistry and Photosynthesis (IBVF), Spain*

#### *\*Correspondence:*

*Seonghoe Jang seonghoe.jang@worldveg.org Fure-Chyi Chen fchen@mail.npust.edu.tw*

#### *†ORCID:*

*Seonghoe Jang orcid.org/0000-0001-5018-3480 Fure-Chyi Chen orcid.org/0000-0001-5475-9248*

*‡These authors have contributed equally to this work*

#### *Specialty section:*

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

*Received: 09 May 2019 Accepted: 10 September 2019 Published: 10 October 2019*

#### *Citation:*

*Wang S-L, Viswanath KK, Tong C-G, An HR, Jang S and Chen F-C (2019) Floral Induction and Flower Development of Orchids. Front. Plant Sci. 10:1258. doi: 10.3389/fpls.2019.01258*

Orchids comprise one of the largest, most highly evolved angiosperm families, and form an extremely peculiar group of plants. Various orchids are available through traditional breeding and micro-propagation since they are valuable as potted plants and/or cut flowers in horticultural markets. The flowering of orchids is generally influenced by environmental signals such as temperature and endogenous developmental programs controlled by genetic factors as is usual in many flowering plant species. The process of floral transition is connected to the flower developmental programs that include floral meristem maintenance and floral organ specification. Thanks to advances in molecular and genetic technologies, the understanding of the molecular mechanisms underlying orchid floral transition and flower developmental processes have been widened, especially in several commercially important orchids such as *Phalaenopsis*, *Dendrobium* and *Oncidium*. In this review, we consolidate recent progress in research on the floral transition and flower development of orchids emphasizing representative genes and genetic networks, and also introduce a few successful cases of manipulation of orchid flowering/flower development through the application of molecular breeding or biotechnology tools.

Keywords: floral transition, flower development, molecular genetics, orchids, orchid biotechnology, transgenic orchids

## INTRODUCTION

The orchid family is one of the largest families of angiosperms. The flowers of Orchidaceae exhibit a high degree of speciation, with wide variations in floral features such as morphology, color, size and fragrance for attraction of pollinators (Peakall, 2007). Orchids are traded worldwide as cut flowers and potted flowering plants. The flowering of orchids can generally be divided into two steps: floral transition and flower development. In floral transition, juvenility, ambient to cool temperature and/ or photoperiod are crucial in determining the time that orchids initiate flowering with respect to ontogeny and season. Although molecular and genetic approaches have made it possible to shed some light on the mechanisms underlying floral transition in model plants such as *Arabidopsis*  (*Arabidopsis thaliana*) and rice (*Oryza sativa*) (He, 2009; Sun et al., 2014), orchid flowering is still not well-understood. Flower formation is initiated by activities of genes known as flowering time genes which regulate the conversion of the vegetative meristem to the floral meristem. Then, floral meristem identity genes regulate the formation of the flower. Once flowering initiates, cadastral genes which govern the whorl formation in flowers are expressed, and the structures of the whorls and their occurrence at the right position are regulated by homeotic genes. Although extreme variation in flower morphology is found across the angiosperms, four types of relatively simple floral organs can be distinguished, namely sepals, petals, stamens and carpels (Theissen and Melzer, 2007; Heijmans et al., 2012). Based mainly on studies with *Arabidopsis*, the ABCDE model explains the establishment and maintenance of floral organ specification through an interactive network of MADS-box transcription factors. Herein, we review recent research on floral transition and flower development of a few representative orchid species from the perspective of molecular genetics. Some successful cases of the use of biotechnological tools to modulate orchid floral traits are also highlighted.

### FLORAL TRANSITION OF ORCHIDS

#### Effect of Ambient Temperature

The Orchidaceae plants are widely distributed across the globe, thus the mechanisms of flowering control among different genera may have developed according to their natural habitats. Apostasioideae, Cypripedioideae, Vanilloideae, Epidendroideae and Orchidoideae are the subfamilies of Orchidaceae. Among these five subfamilies, the Epidendroideae is the largest containing more than 500 genera with around 20,000 species (Freudenstein and Chase, 2015). Different orchids of the Epidendroideae require different conditions of ambient temperature and photoperiod to induce flowering (Hsiao et al., 2011).

Most orchids take several years to finish the juvenile stage. *Phalaenopsis* usually reach maturity after three to five leaves. Usually, the floral spike (inflorescence) emerges from the axillary buds of the fourth node below the apical leaf whereas other axillary buds are maintained in a dormant state during the flowering season (Sakanishi et al., 1980). The mechanism of differential commitment of distinct axillary buds for flowering is unknown. At the beginning of flowering, the axillary buds are enlarged and then protrude from the base of a leaf, which takes about 3–4 weeks. After that, the bud is elongated to become a floral spike.

The flowering of several orchids is influenced by change in ambient temperature. The flowering of *Phalaenopsis* is promoted by low ambient temperature, usually less than 26 °C (Blanchard and Runkle, 2006); and can be reversed if the ambient temperature is elevated. The flowering activity of *Dendrobium* is also promoted by low ambient temperature (Campos and Kerbauy, 2004); however, flowering of some hybrid cultivars of *Dendrobium–* such as *Dendrobium* Chao Praya Smile and *Dendrobium* Madame Thong-In, are promoted by high ambient temperature. The flowering of *Miltoniopsis* and *Zygopetalum* are promoted by cool temperature 11–14 °C (Lopez et al., 2003; Lopez and Runkle, 2006). For the *Oncidium*, another popular orchid, change in ambient temperature enhances or is neutral to their flowering activity. For *Cypripedium* species, vernalization below 5 °C or subzero temperature is required for flowering (Zhang et al., 2014).

In the model plant *Arabidopsis*, flowering activity is delayed by low ambient temperature (16 °C) but promoted by vernalization. In low ambient temperature, an alternatively spliced form of FLOWERING LOCUS M (FLM), FLM-β associates with SUPPRESSOR OF VEGETATIVE PHASE (SVP) to suppress the floral integrator gene *FLOWERING LOCUS T* (*FT*) (Lee et al., 2013). The expression of *FLM* is down-regulated *via* alternative splicing coupled with nonsense-mediated decay (AS-NMD) in response to elevated temperatures (Sureshkumar et al., 2016), indicating that *FLM* plays an important role in flowering initiation by modulating thermo-sensitivity. The *FT* expression is also suppressed by a MADS-box gene *FLOWERING LOCUS C* (*FLC*). Constant expression of *FLC* prevents the winterannual *Arabidopsis* from floral transition before winter and this repression is released by the vernalization pathway signaling (He, 2009). The vernalization pathway suppresses the *FLC* activity *via* the trimethylation of histone 3 lysine 27 (H3K9me3) of *FLC* chromatin. Vernalization is an obligatory requirement for the flowering of lily (*Lilium longiflorum*). In high ambient temperature, a fully developed floral bud is present inside the bulb. A period of prolonged cold temperature in winter is required for the switch of shoot apical meristem (SAM) to inflorescence meristem (IM). The candidate genes involved in vernalization response in lily have been elucidated using transcriptome analysis (Villacorta-Martin et al., 2015; Li et al., 2016) and several orthologs of *Arabidopsis* flowering genes were identified. Among them, *SVP* and *VERNALIZATION 1* (*VRN1*) were down- and up-regulated respectively in floral buds when the bulbs were treated with low ambient temperature. Furthermore, ectopic expression of lily *SVP* or *VRN1* in *Arabidopsis* resulted in delayed or earlier flowering, respectively.

Understanding how ambient temperature regulates the floral transition of orchids is valuable not only for the horticultural market but also for the comprehension of flowering regulation in different plant species. Using molecular genomics tools to investigate flowering in popular ornamental orchids has made it possible to identify many flowering-related genes; however, how such genes regulate flowering in an ambient temperaturedependent manner is still not clear.

### Effect of Photoperiod

In *Arabidopsis* and rice, photoperiod is crucial for flowering control (He, 2009; Sun et al., 2014; Song, 2016). Under longday conditions, *Arabidopsis FT* expression is activated in vascular tissues of leaves in a circadian rhythmic manner by the transcriptional regulator CONSTANS (CO) whose expression and activity are controlled by light signaling pathways and the circadian clock (Song, 2016). In rice, two orthologs of *FT*, *Heading date 3a* (*Hd3a*) and *RICE FLOWERING LOCUS T* (*RFT*) are mainly responsible for floral induction under short-days (SD) and LD conditions, respectively (Sun et al., 2014).

Floral initiation has been shown to be regulated by photoperiod in a few orchids. In *Doritis pulcherrima* (now *Phalaenopsis pulcherrima*), under the 30 °C/20 °C (day/night) condition, floral spikes are initiated more efficiently by 9-h light than 12-h light treatment (Wang et al., 2003). In *Miltoniopsis* orchids, SD incubation at 23 °C before shifting to cool temperature (11–14 °C) facilitates flowering (Lopez and Runkle, 2006). However, during cool temperature treatment, different photoperiods have no significant effect on flowering suggesting that ambient temperature may play a major role in flowering of *Miltoniopsis*. On the other hand, the formation of floral spikes in *Psygmorchis pusilla* is positively correlated with the increase in day length implying that *P*. *pusilla* is a quantitative LD plant (Vaz et al., 2004). The effect of photoperiod on flowering seems to be various among different orchid species which have great diversity in adaptation. Furthermore, most orchids are native to tropical areas where day length does not change dramatically throughout the year. Thus, it is reasonable to anticipate that photoperiod has limited effects on flowering of orchids.

### Effect of Phytohormones

The effects of phytohormones on orchid flowering have been studied (Goh and Yang, 1978). A synthetic cytokinin, 6-benzylaminopurine (BA), stimulated the flowering of monopodial (e.g., *Phalaenopsis*) and sympodial (e.g., *Dendrobium*) orchids whereas auxin suppressed the BA effect. The positive effect of BA on flowering is likely to be enhanced when combined with gibberellic acid (GA3) although GA3 applied alone does not have an influence on floral induction (Hew and Clifford, 1993). In *Phalaenopsis* and *Doritaenopsis*, plants sprayed with BA produced visible inflorescences 3 to 9 days earlier than those without BA treatment although the application of BA could not replace the inductive low temperature in *Phalaenopsis* (Blanchard and Runkle, 2008). Interestingly, even though GAs do not induce flowering, optimum levels of endogenous GAs in the flowering shoot tips are required for flower development in *Phalaenopsis*. Indeed, injection of GAs can rescue the blockage of flower development under high temperature (Su et al., 2001). The existence of abscisic acid (ABA) in different tissues of *Phalaenopsis* has been examined. Relatively higher level of free ABA has been detected in dormant axillary buds whereas free or bound forms of ABA were not found in floral shoots (Wang et al., 2002). In addition, exogenous application of ABA to the stem of *Phalaenopsis* inhibited the formation of floral spikes even under inductive low ambient temperature conditions implying that ABA plays an inhibitory role in orchid floral transition.

Taken together, developmental maturity and ambient temperature are likely the main endogenous and environmental factors, respectively, controlling orchid flowering together with the combined action of phytohormones. Many orchids are epiphytic plants, meaning they are adapted to face frequent periods of nutrient and water scarcity. Although this kind of orchid uses the crassulacean acid metabolism (CAM) as the strategy for carbon fixation like other CAM plants such as succulents and pineapples to adapt themselves to arid conditions (Silvera et al., 2009), flowering consumes a lot of energy. Therefore, the right timing of bud commitment for floral induction with the best physiological situation is crucial for the maintenance of the species through successful sexual propagation. In addition, most orchids have particular pollinators in the wild environment; the timing of floral transition should be consistent with the appearance of their pollinators. Thus, sensing ambient temperature seems to be a good strategy to achieve this aim if their pollinators only appear in a particular season.

### Genes Involved in Flowering in *Phalaenopsis*

Based on expression analyses and functional studies with heterologous expression systems, candidate genes for flowering regulatory networks in orchids have been reported. Recently, genome-wide analyses have also been taken into consideration for the identification of Orchidaceae-specific and/or speciesspecific key genes controlling flowering/flower development of orchids (Huang et al., 2016; Lin et al., 2016; Wen et al., 2017).

*P*. *aphrodite FT1* (*PaFT1*) has been isolated and functionally characterized (Jang et al., 2015). Flowering of *P*. *aphrodite* is induced by low ambient temperature (≤25 °C) but prohibited by high ambient temperature (≥28 °C). Moreover, photoperiod has no significant influence on flowering of *P*. *aphrodite*. Expression patterns of *PaFT1* reflect the flowering behavior of the orchid; its flowering is induced by low ambient temperature but not by different light regimes. Reduced expression level of *PaFT1* by virus-induced gene silencing (VIGS) methods resulted in delayed flowering under inductive low ambient temperature, while ectopic expression of *PaFT1* suppressed the late flowering phenotype caused by the induced expression of *SVP* and active *FRIGIDA* (*FRI*), a *FLC* activator, in *Arabidopsis.* Moreover, PaFT1 is able to physically interact with PaFD reminiscent of the *Arabidopsis* FT-FD module (He, 2009). *LEAFY* (*LFY*) is another floral integrator in *Arabidopsis* (He, 2009). The *LFY* gene of *P*. *aphrodite* (*PhapLFY*) has also been characterized (Jang, 2015). Expression of *PhapLFY* under the control of *Arabidopsis LFY* promoter rescued the abnormal floral structure of *Arabidopsis lfy* mutant and caused early heading by overexpression in rice. Recently, a *CO*-like gene, *PhalCOL* has been identified in *P*. *hybrida* (Wedding Promenade) (Zhang et al., 2011). *PhalCOL* has significant sequence similarity to *CO* of *Arabidopsis*. Overexpression of *PhalCOL* in tobacco caused an early-flowering phenotype suggesting the functional convergence of *CO* genes in flowering among different plant species. To discover candidate genes responsible for the flowering control of *P*. *aphrodite*, transcription profiles in axillary buds of plants treated with low and high ambient temperature were analyzed and compared to each other (Huang et al., 2016). The result showed that, in addition to *FT*, *LFY*, *APETALA1* (*AP1*) and *OVEREXPRESSION OF CONSTANS 1* (*SOC1*) homologs, genes involved in GA biosynthesis were also up-regulated by low ambient temperature. Another expression analysis using *Phalaenopsis* Fortune Saltzman found that transcripts of *KNOX1*, R2R3-like *MYB*, adenosine kinase 2, S-adenosylmethionine synthetase, dihydroflavonol 4-reductase, and naringenin 3-dioxygenase were accumulated at a higher level in spikes grown under warm day/cool night (28 °C/21 °C) compared with those grown at daily warm temperature (28 °C/26 °C) under natural light conditions (Li et al., 2014).

### Genes Involved in Flowering in *Dendrobium*

Flowering of *D. nobile* is promoted by low ambient temperature (Campos and Kerbauy, 2004). Orthologs of *FT* and *MOTHER OF FT AND TFL1* (*MFT*), a homolog of *FT*, have been identified in *D. nobile* Lindl. (Li et al., 2012). Expression of *DnFT* is up-regulated in leaves but down-regulated in axillary buds under low temperature treatment (12 °C/9 °C, day/night) whereas *DnMFT* expression is not affected by the low temperature treatment. Transgenic *Arabidopsis* ectopically expressing *DnFT* exhibited early flowering. In addition, an ortholog of *SOC1*, *DOSOC1* has been identified from *Dendrobium* Chao Praya Smile (Ding et al., 2013). Increased expression of *DOSOC1* was detected during floral transition. Moreover, overexpression of *DOSOC1* in both *Arabidopsis* and *Dendrobium* resulted in earlyflowering phenotypes. *DOAP1*, an ortholog of *AP1* has also been identified and characterized from the orchid (Sawettalake et al., 2017). The role of *DOAP1* is similar to that of *AP1* in *Arabidopsis*. Overexpression of *DOAP1* rescued the floral defect of *ap1* mutant and resulted in early flowering in wild-type *Arabidopsis*. Differential gene expression in the SAM during *in vitro* transition from vegetative to reproductive growth has been investigated in *Dendrobium* Madame Thong-In (Yu and Goh, 2000a). Several transcription factors, including a MADS-box gene of the *AP1*/ *AGL2* family, a class I *KNOX* gene and a homolog of the *Drosophila SEVEN-UP* gene were differentially expressed in vegetative and transitional SAM. The *KNOX* gene plays an important role in the function of SAM, and encodes a KNOTTED1-like homeobox (Knox) protein later designated as *DOH1* (*Dendrobium Orchid Homeobox 1*) (Yu et al., 2000). In tissue culture conditions, the expression of *DOH1* was gradually increased in the apical meristem during vegetative growth whereas it was decreased during the progress of reproductive growth. Overexpression of antisense *DOH1* resulted in early flowering in *Dendrobium*. On the other hand, *DOMADS1*, a MADS-box gene of the *AP1*/*AGL9* superfamily, was identified from the cDNA library of transitional SAM (Yu and Goh, 2000b). Expression of *DOMADS1* was up-regulated in transgenic plants harboring p*35S*: antisense *DOH1* suggesting that *DOH1* is a possible upstream repressor of *DOMADS1* in the flowering control of *Dendrobium* orchids. Recently, transcriptomes of *D. nobile* were analyzed under cold (4 °C) or exogenous cytokinin (thidiazuron) treatment (Wen et al., 2017). The results showed that *SOC1*, *LFY* and *AP1* genes were induced by both low temperature and thidiazuron treatment whereas *DnVRN1* and *FT* were induced only by cold treatment. Also, some marker genes for the GA signaling pathway were up-regulated under both low temperature and thidiazuron treatments. Further investigation is needed to uncover the cytokinin-GA signaling network in the inductive temperature condition underlying the floral transition in *Dendrobium*.

#### Genes Involved in Flowering in *Oncidium*

High ambient temperature (30 °C) accelerates the flowering of *Oncidium*. The flowering response to changes in ambient temperature in *Oncidium* is opposite to that of *Phalaenopsis*. The flowering promoter *FT* and repressor *TERMINAL FLOWER 1* (*TFL1*) both encode proteins belonging to phosphatidylethanolamine-binding protein (PEBP) family and have also been identified in *Oncidium* Gower Ramsey (Hou and Yang, 2009). *OnFT* was expressed in axillary buds, leaves, pseudo-bulb and flowers while *OnTFL1* was only expressed in axillary buds and pseudo-bulb. The expression of *OnFT* was regulated by photoperiod, with highest expression from the 8th to 12th hour of the light period and lowest expression at dawn. However, expression of *OnTFL1* is not influenced by photoperiod. Ectopic expression of *OnFT* resulted in an earlyflowering phenotype in *Arabidopsis*, and late-flowering of *Arabidopsis ft-1* mutant was rescued by *OnFT* overexpression (Hou and Yang, 2009). On the contrary, ectopic expression of *OnTFL1* in *Arabidopsis* displayed late flowering. *OMADS1*, a homolog of *Arabidopsis AGL6* has also been identified in *Oncidium* Gower Ramsey (Hsu et al., 2003). *OMADS1* transcripts were detectable in the apical meristem and floral organs, and ectopic expression of *OMADS1* in *Arabidopsis* caused early flowering with up-regulation of *FT*, *SOC1*, *LFY* and *AP1*. Overexpression of *OMADS1* in *Oncidium* also resulted in precocious flowering (Hsu et al., 2003). Recently, ascorbate (AsA) content has been shown to play an important role in floral transition in response to thermal stress (30 °C over a 14-day period) in *Oncidium* Gower Ramsey (Chin et al., 2014). Under thermal stress conditions, the level of reactive oxygen species (ROS; e.g., H2O2) was significantly increased and the AsA redox ratio [AsA to dehydroascorbate (DHA, the oxidized form of AsA)] was decreased with prominent up-regulation of cytosolic ascorbate peroxidase (*cytAPX1*) expression. The oxidation of AsA to DHA by ascorbate peroxidase is the key reaction to remove H2O2. The report by Chin and colleagues suggests that the AsA/DHA redox ratio may act as one of the endogenous signals that induce the flowering of *Oncidium* in response to high ambient temperature. Furthermore, it has been shown that reduced GSH redox ratio caused by down-regulation of GSH metabolism-related genes such as glutathione reductase (*GR1*), γ-glutamylcysteine synthase (*GSH1*) and glutathione synthase (*GSH2*) was linked to the decrease in the AsA redox ratio for flowering of *Oncidium* orchid (Chin et al., 2016).

#### Genes Involved in Flowering in *Dortiaenopsis* and Other Orchids

*DhFVE*, a *Dortiaenopsis* ortholog of *Arabidopsis FVE*, which is a component of the autonomous flowering pathway, has been identified (Sun et al., 2012). Flowering of *Dortiaenopsis* (now *Phalaenopsis*) is promoted by low ambient temperature (22 °C/18 °C, day/night) (Luo et al., 2011) and the expression of *DhFVE* is increased in the stem during floral transition. In addition, *EARLY FLOWERING 4* (*EFL4*) family genes, *DhEFL2*, *DhEFL3* and *DhEFL4* have been identified in *Dortiaenopsis* (Chen et al., 2015). *Arabidopsis EFL4* is known to affect flowering through photoperiod perception and circadian regulation (Doyle et al., 2002). Ectopic expression of *DhEFL2*, *DhEFL3* or *DhEFL4* delayed flowering of *Arabidopsis*. Moreover, *GIGANTEA* (*GI*), an upstream activator of *CO* (Sawa et al., 2007), has been identified as *DhGI1* in *Dortiaenopsis* hybrid (Luo et al., 2011). Expression of *DhGI1* is up-regulated by low temperature and may play a role in flowering initiation in *Dortiaenopsis* hybrid. Recently, many putative flowering genes have been identified in *Cymbidium* and *Erycina* through transcriptome analyses. However, their functions still remain to be examined (Li et al., 2013; Lin et al., 2016).

### A Hypothetical Model of Flowering Regulation in *P. aphrodite*

The regulatory networks of floral transition in different orchids may be divergent since the required inductive conditions are not all the same. In addition, the genetic tools for orchid research are still limited; therefore, it is time-consuming and also difficult to reveal the molecular mechanisms of flowering control in orchids. The popular ornamental orchid *P. aphrodite* has a particular requirement for floral transition, i.e., low ambient temperature, which makes this orchid an interesting target for a case-study of flowering regulation in orchids. Furthermore, a transformation system has been established in orchids including *P. aphrodite* (Hsing et al., 2016). Thus, the molecular mechanism of flowering regulation could be investigated more thoroughly.

The hypothetical gene regulatory network in flowering of *P. aphrodite* is illustrated in **Figure 1** based on the published results (data adapted from research on orchids). The expression of *PaFT1* is induced by low ambient temperature, and PaFT1 interacts with PaFD to possibly activate the downstream genes required for floral induction. Reflecting the *Arabidopsis* model, PaFT1 protein is likely transported from leaves to axillary buds to induce spiking in *Phalaenopsis* orchid. If the PaFT1 moves to dormant/juvenile buds, there must be a repressive mechanism against PaFT1 action or other unknown floral co-activators are still absent in dormant/juvenile buds. However, we cannot exclude the possibility that the pavement for the PaFT1 movement to the dormant/juvenile buds is not available. Based on the transcriptomic analyses of *Phalaenopsis* orchids, *KNOX1*, *SOC1* and *FVE* genes are induced by low ambient temperature (Sun et al., 2012; Li et al., 2014; Huang et al., 2016) although *Dendrobium KNOX1* gene is down-regulated by low ambient temperature (Yu and Goh, 2000a; Yu et al., 2000). Thus, the role of *KNOX1* in floral transition of *Phalaenopsis* needs to be further examined. In *Phalaenopsis*, *SOC1* may activate *LFY* and *AP1* during floral transition as is the case in *Arabidopsis*. Recently, *FVE* has been suggested to be an upstream activator of *SOC1* in *P. aphrodite* (Koh et al., 2018). The *CO* in *Phalaenopsis* may also activate *FT* irrespective of photoperiod (Zhang et al., 2011; Kaewphalug et al., 2017). In addition, *ELF* genes in *Phalaenopsis* may repress the *CO* activity (Doyle et al., 2002; Chen et al., 2015). Moreover, the GI-FLAVIN-BINDING, KELCH REPEAT, F BOX 1 (FKF1) and CYCLING DOF FACTOR (CDF) may also regulate the *CO* activity in *Phalaenopsis* as is also the case in *Arabidopsis* (Song, 2016). Since *SVP* and *FLM* are critical flowering regulators responding to changes in ambient temperature in *Arabidopsis*, it is also possible to anticipate that the two orthologs of *SVP* and *FLM* also act in floral induction of *Phalaenopsis* in various ways. Actually, *SVP* and *FLM* repress the expression of *FT* at low ambient temperature in *Arabidopsis* (Lee et al., 2013). Thus, the working mechanism of those orthologs in *Phalaenopsis* might be distinct from that of *Arabidopsis*. In addition, further studies are required to reveal the mechanisms underlying phytohormone-dependent flowering pathways linked to changes in ambient temperature in orchids. Even taking these findings together, it can be seen that there is a long way to go to achieve a better understanding of the flowering regulatory network in *P*. *aphrodite*. As one of the shortcuts to reach this goal, high throughput analyses can be applied to identify candidates involved in the floral transition of *Phalaenopsis*. Also, collection of information on flowering networks in various plant species would be very useful in the interpretation of large-scale experimental results such as omics data. Most importantly, forward or reverse genetic studies are

FIGURE 1 | The hypothetical flowering regulation in *P*. *aphrodite* under low ambient temperature. Low ambient temperature drives *PaFT1* activity in leaves and then PaFT1 protein is transported selectively to mature axillary buds or both mature and juvenile axillary buds. A complex formed by the interaction between PaFT1 and PaFD may activate downstream genes such as *AP1* required for floral meristem identity. An example of unknown factors for flowering in the meristem is presented by a green circle with a question mark inside. In juvenile axillary buds, some flowering activators required for floral induction may be absent. *PaFT1* can also be activated by *CO* pathway and the *CO* activity—*PhalCOL* (*Phalaenopsis CO Like*), here—which is likely to be regulated by *EFL* and *GI* genes. This regulation should not be dependent on photoperiod. Low ambient temperature also activates *PaFVE*. Consequently, *PaSOC1* is activated by the *PaFVE*. Moreover, the reduced expression of a KNOTTED1-like homeobox (Knox) gene may lead to the activation of *SOC1* through other MADS box genes such as *DOMADS1*. The orchid *LFY*  genes such as *PhapLFY* can also be induced by *PaSOC1*. The solid lines represent pathways that have been reported in orchids and the dotted lines indicate hypothetical regulations based on studies using other plant species. References for each regulation are presented in the figure.

required for the functional confirmation of candidate genes in orchid flowering. The homologs of flowering genes in a few orchid species are listed in **Table S1**.

### FLOWER DEVELOPMENT OF ORCHIDS

#### Floral Development of *Arabidopsis* and Monocot Plants

Floral pattern formation is one of the significant features of angiosperms, the mechanisms of which can be explained using the ABC or ABCDE model (**Figure 2A**). The identity of various floral organs is determined by MADS-box transcription factors or their complexes during flower development. According to the ABC model, in *Arabidopsis*, sepals in the first whorl are characterized by the expression of A-class genes; petals in whorl 2 are characterized by co-expression of A- and B-class genes, stamens in whorl 3 are characterized by co-expression of Band C-class genes, and carpels in whorl 4 are determined when C-class genes are solely expressed (Jack, 2001; Theissen, 2001). The ABC model has been extended to the ABCDE model with additional functions of D- and E-class genes in floral organ identity determination (Theissen and Melzer, 2007; Heijmans et al., 2012). D-class genes are involved in ovule development while E-class genes are involved in the identity of all floral verticils. Most genes in the ABCDE model are MADS-box genes. *AP1* and *AP2* (the only non-MADS-box genes) are A-class genes; *AP3* and *PISTILLATA* (*PI*) are B-class genes. *AGAMOUS* (*AG*) is a C-class gene; *SEEDSTICK* (*STK*) is a D-class gene. The D-class genes were first identified in *Petunia*, namely *FLORAL BINDING PROTEIN 7* (*FBP7*) and *FBP11*. Finally, *SEPALLATA 1* (*SEP1*)*, SEP2, SEP3* and *SEP4* formally known as *AGAMOUS-LIKE 2* (*AGL2*), *AGL4*, *AGL9* and *AGL3*, respectively, are characterized as E-class genes. According to the floral quartet model, MADS-box proteins form whorl-specific tetrameric complexes during floral organ determination. The tetrameric transcriptional factor complexes recognize specific *cis*-regulatory elements termed CArG-boxes (CC-AT rich-GG), and different complexes are postulated to function in controlling expression of different target genes (Jetha et al., 2014). Recently, a reliable and efficient two-stage approach has been developed for angiosperms MADS-Box genes classification using machine learning methods to overcome errors in phylogenetic tree construction (Chen et al., 2019).

The MADS-box genes involved in the floral development in monocot plants such as rice, maize (*Zea mays*), wheat (*Triticum aestivum*), barley (*Hordeum vulgare*) and lily have also been elucidated (Callens et al., 2018). MADS-box protein structure is conserved between diverse plants. Comparing the expressions and functions of the ABCDE MADS-box genes among different monocot plants provides the opportunity to elucidate their roles in determining the floral development during evolution.

In rice, the A-class gene, the *FUL*-like (*FRUITFULL* locus of *Arabidopsis*) *OsMADS14*, *OsMADS15*, *OsMADS18* and *OsMADS20*, are expressed to determine the whorl 1 structures of lemma and palea; co-expression of A and B-class genes, *DEF*-like (*DEFICIENS*: an *Antirrhinum* B-class gene) *SUPERWOMEN1* (*SPW1 or OsMADS16*) and *GLO*-like (*GLOBOSA*: another *Antirrhinum* B-class gene) *OsMADS2* and *OsMADS4*, determines the whorl 2 structure of lodicules (Yoshida and Nagato, 2011); co-expression of B and C-class genes, *OsMADS3* and *OsMADS58*, determines the whorl 3 structure of stamens. Another C function gene *DROOPING LEAF* (*DL*) have been identified in carpel specification and floral meristem determinacy (Shige-Hiro et al., 2019). C and D-class genes, *OsMADS13* and *OsMADS21*, are expressed to determine the whorl 4 structure of the carpel and ovule, respectively. In addition, the E-class genes, *OsMADS1*, *OsMADS5* and *OsMADS34*, are involved in the development of all four whorl structures (**Figure 2A**).

The lily has a floral character most similar to orchids, with almost identical sepals and petals. Flowers from the lily family *Liliaceae* have three sepals in whorl 1, three petals in whorl 2, six stamens in whorl 3 and three fused carpels in whorl 4. Three A-class genes from lily, *LMADS5*, *LMADS6* and *LMADS7* have been analyzed (**Figure 2A**) (Chen et al., 2008). They were all expressed in the vegetative stem and inflorescence meristem. With regard to the floral organs, *LMADS5* and *LMADS6* are mainly expressed in the carpel, whereas *LMADS7* is only expressed in the vegetative stem and inflorescence meristem. Early flowering phenotype and homeotic conversions of sepals to carpel-like and of petals to stamen-like structure were observed when they were ectopically expressed in *Arabidopsis*. For the B-class gene, mRNA of *AP3*-like *LMADS1* could be detected in all four whorls but the protein was detected only in the petal and stamen (Tzeng and Yang, 2001). The *PI*-like *LMADS8* and *LMADS9* are expressed in whorl 1 and 2 during all stages of development and are expressed in the stamen only in young flower buds (Chen et al., 2012a). The ectopic expression of *LMADS8* and *LMADS9* partially converted the sepals into petaloid structures in *Arabidopsis*. The *PI*-like gene *LFGLOA* and *LFGLOB* identified from *L.* x *formolongi* are also expressed in the first, second and third whorl of lily (Akita et al., 2008). Several C-class genes have been identified from different lilies including *LMADS10*, *LFAG1* and *LLAG1* (Benedito et al., 2004; Akita et al., 2008; Hsu et al., 2010). They are expressed in whorl 3 and 4, and ectopic expression of C-class genes in *Arabidopsis* resulted in carpel-like septal and stamen-like petal phenotypes. In addition, a double-flowered cultivar "Aphrodite" of *L.* x *formolongi* with petal-like stamen may be caused by reduction in the expression of *LFAG1* in whorl 3 (Akita et al., 2008). The lily D-class gene, *LMADS2* completely accumulated in the carpel and more specifically in the ovules (Tzeng et al., 2002). Ectopic expression of *LMADS2* in *Arabidopsis* converted the sepals into a stamen-like structure in whorl 2. *LMADS3* and *LMADS4* have been identified as E-class genes in lily (Tzeng et al., 2003). Both *LMADS3* and *LMADS4* expression are expressed in the inflorescence meristem, floral buds and all four whorls of the flower organ. Ectopic expression of *LMADS3* in *Arabidopsis* showed extremely early flowering, which was associated with up-regulation of *FT*, *SOC1*, *LUMINIDEPENDENS* and flower meristem identity gene *LFY* and *AP1*. Taken together, the different expression patterns of A- and B-class genes from that of the dicot plants such as *Arabidopsis* suggest a modified ABCDE model for flower development in lily.

### Floral Organ Identity Genes in Orchids

The structure of the orchid flower has a zygomorphic nature in contrast to most plant groups leading to precise interaction with the pollinator (Cubas, 2004). Orchid flowers generally contain an outer whorl with three sepals, an inner whorl with three petals, and a single column in the center. The sepals in most orchids are enlarged and look like petals. Two of the petals are displayed in a normal shape and one becomes a highly specialized structure called the lip or labellum. The sepals and petals are usually called tepals since they are very similar in appearance. The column, also called the gynostemium, is the reproductive organ of orchids, which combines the gynoecium (female organ) and androecium (male organ). On the top of the column is the male anther which contains packets of pollen called pollinia. Many floral MADSbox genes have been identified from different orchids. According to their expression patterns and putative roles in determining the different floral organs, advanced models of flower development for orchids have been proposed. Also, these studies offer an opportunity to understand the relationships between the diverse array of floral MADS-box genes and flower development.

### A-Class Genes

A- and E-class genes belong to the same *AP1*/*AGL9* superfamily group. A-function genes come from the *SQUA*-like (*SQUAMOSA*: an *Antirrhinum* A-class gene) subgroup, which can be divided into *FUL*/*AGL8*-like and *euAP1*-like clades. E-function genes come from the *SEP*-like subgroup containing *AGL9*-like and *AGL2*/*3*/*4*-like clades. *AGL6* is a member of the *AP1*/*AGL9* group between the *SQUA*-like and *SEP*-like subgroups (Purugganan et al., 1995; Chang et al., 2009). Two *AP1*-like genes, *ORAP11* and *ORAP13*, were isolated and characterized in *Phalaenopsis*  Formosa Rose. Both genes are highly expressed during the early stages of floral buds and vegetative organs (Chen et al., 2007). At later stages of flower development, *ORAP11* is expressed only in the column and *ORAP13* expression is absent from all of the floral organs. *PaAP1-1* identified from *P. aphrodite* is expressed in the inner whorls of the pollinia and pedicel whereas *PaAP1-2* is expressed in the pedicel only (Su et al., 2013). This expression pattern suggests that *PaAP1-1* and *PaAP1-2* play a role in the development of the pollinia and gynoecium rather than being involved in the perianth formation like that in *Arabidopsis*. Furthermore, *PaAP2-5* is mainly expressed in whorl 1 and 2*, PaAP2-7* is specifically expressed in pollinia, and *PaAP2-11* is expressed at a low level among all floral organs. Novel functions of homeotic genes may be acquired based on the diversified expression in *Phalaenopsis* through evolutionary processes. Two *AP1/FUL*-like genes in *Phalaenopsis* hybrid "Athens," *PhaMADS1* and *PhaMADS2* are highly expressed in the ovary before pollination, but a low level of expression is detected in the perianth and gynostemium (Acri-Nunes-Miranda and Mondragon-Palomino, 2014). *DOMADS2* identified from *Dendrobium* Madam Thong-in, and *DthyrFL1*, *DthyrFL2* and *DthyrFL3* from *D*. *thyrsiflorum* have been characterized as *SQUA*-like and *FUL*-like genes, respectively (Yu and Goh, 2000b; Skipper et al., 2005). The expression of *DOMADS2* could be detected in the SAM during floral transition but it was limited to the column at the later stages of flower development. Low level expression of *DthyrFL1*, *DthyrFL2* and *DthyrFL3* was observed in the vegetative tissues, but higher levels were detected in the inflorescences and ovules. The *AP1* homolog of *Oncidium* Gower Ramsey, *OMADS10* is expressed in vegetative leaves, lip and carpel of mature flowers (Chang et al., 2009). An *AP2*-like gene of *D*. *crumenatum*, *DcOAP2* is expressed in all floral organs (Xu et al., 2010). Weak expression of three *AP1*-like genes, *EpMADS10*, *EpMADS11* and *EpMADS12* from *Erycina pusilla* has also been observed in all floral organs (Dirks-Mulder et al., 2017). An *AP2 like* gene of *C. ensifolium*, *CeAP2*, is expressed mainly in sepals and petals with negative regulation by miR172 (Yang et al., 2015).

### B-Class Genes

B-class genes in Orchidaceae have been surveyed and have 11 species characterized into four *AP3*- and two *PI*-duplicated homologs (Pan et al., 2011). *PI* homologs are uniformly expressed in all floral whorls whereas different clades of *AP3* homologs may have different expression patterns within floral organs. In *P*. *equestris*, four *DEF*-like genes, *PeMADS2*, *PeMADS3*, *PeMADS4* and *PeMADS5* have been identified. (Tsai et al., 2004). They are expressed in floral organs with distinct patterns. *PeMADS2* and *PeMADS5* are expressed in sepals, petals, lip and column; *PeMADS3* is expressed in petals, lip and column, whereas *PeMADS4* is expressed in lip and column only. In peloric (liplike petal) mutant, expression of *PeMADS5* is absent from all floral organs, and expression of *PeMADS4* is extended to the liplike petal. This indicates that, among the four *DEF*-like genes, *PeMADS4* may determine the lip formation in *Phalaenopsis*. In addition, these four *DEF*-like MADS genes are able to interact with a *GLO*-like gene *PeMADS6* (Tsai et al., 2005; Tsai et al., 2008). Ectopic expression of *PeMADS6* caused petaloid sepals in *Arabidopsis* flowers. Furthermore, these heterodimers could bind the CArG *cis*-element. A single copy of *PI*-like gene of *Phalaenopsis* hybrid, *PhPI10* has been also characterized (Guo et al., 2008). Expression of *PhPI10* is restricted to the lip of flowers. Several *AP3*-like genes *PaAP3-1*, *PaAP3-2*, *PaAP3-3* and *PaAP3-4* were identified in *P*. *aphrodite* and their expression was preferentially detected in sepals, petals and the lip/column (Su et al., 2013). Another *PI-like* gene, *PaPI*, has also been identified but its transcripts are detectable in all floral organs. In *Dendrobium crumenatum*, *DcOAP3A*/*B* and *DcOPI* were identified as *AP3* like and *PI*-like genes, respectively (Xu et al., 2006). Transcripts of *DcOAP3A* and *DcOPI* are accumulated in all floral organs and their proteins form heterodimers. Abnormal flowers containing petaloid sepals were produced when *DcOPI* was overexpressed in *Arabidopsis*. In addition, the phenotypic alteration of B-function mutants has also been observed by overexpressing *DcOAP3- SRDX* repressor fusion construct implying a dominant negative effect on the B-function *via* the hetero-dimerization with its interacting partners. The *Oncidium* Gower Ramsey *AP3*-like gene, *OMADS3* is also expressed in all floral organs (Hsu and Yang, 2002). Another study characterized the B-class MADS-box genes of *Oncidium* including the *AP3*-like *OMADS3*, *OMADS5* and *OMADS9*, and *PI*-like *OMADS8* (Chang et al., 2010). *OMADS5* is expressed in sepals and petals; *OMADS9* is expressed in petals and lip; *OMADS8* is expressed in all four floral organs like *OMADS3*. In lip-like petals and lip-like sepals of peloric mutant flowers, expression of *OMADS5* is down-regulated suggesting that *OMADS5* negatively regulates the lip formation. Recently, the perianth formation in *Cymbidium goeringii* was determined through the complete analysis of expression levels of B-class genes along with co-expression of A-class and E-class genes (Xiang et al., 2018). *CgDEF1* is expressed in sepals and petals but not lip; *CgDEF3* and *CgDEF4* are highly expressed in lip and lip-like petals whereas *CgDEF2* is expressed in all floral organs. Several B-class genes have been isolated from *Habenaria* (Kim et al., 2007), *Orchis* (Salemme et al., 2011) and *Erycina* (Dirks-Mulder et al., 2017). *HrDEF* expression is exhibited in the petals and column whereas that of *HrGLO1* and *HrGLO2* is detected in all floral organs. In the petaloid-sepal mutant, expression of *HrDEF* was extended to the petaloid sepals suggesting that distinctive expression of *HrDEF* determines the differentiation of sepals and petals of *H. radiata* flowers. In *Orchis italica*, the *PI*/*GLO*like genes *OrcPI* and *OrcPI2* are expressed in all floral organs of immature floral buds but their expression is restricted to the lip in the mature flower. In *E. pusilla*, *EpMADS13*, *EpMADS14* and *EpMADS15* are *AP3*-like genes while *EpMADS16* is a *PI*-like gene. Their expression is detectable in almost all floral organs but has diverse patterns. In *Rhynchostylis gigantea*, *RgAP3* an *AP3* like and *RgPI* a *PI*-like genes were cloned and expression levels of *RgAP3* were noticed only in the petal and sepal, and *RgPI* was expressed in every part of the floral organs (Zhang et al., 2013).

## C- and D-Class Genes

The reproductive organ, the gynostemium, is another unique floral structure in orchids in addition to the labellum. In the ABCDE model, C-class genes are important for the development of stamens and carpels, and D-class genes are required for ovule development. Both C- and D-class genes belong to the *AG* subfamily of MADS-box genes (Theissen et al., 2000; Kramer et al., 2004). In orchids, C- and D-class genes involved in the development of the gynostemium and ovule have been also identified. In *P*. *equestris*, *PeMADS1* and *PeMADS7* were characterized as C- and D-class genes, respectively (Chen et al., 2012b). Spatial expression analyses of *PeMADS1* and *PeMADS7* demonstrated that they are specifically expressed in the gynostemium of the flower. Development of ovules is initiated after pollination in orchids and *PeMADS1* and *PeMADS7* transcripts are accumulated in the ovules after pollination. Moreover, expression of *PeMADS1* could be detected in petals of *gynostemium-like petal* (*gylp*) mutant suggesting that *PeMADS1* functions in gynostemium development. PeMADS1 and PeMADS7 could form a homodimer or heterodimer *via*  the PeMADS8, an E-class protein. Ectopic expression analyses showed that *PeMADS1* could rescue the phenotype of *AG* mutant, and *PeMADS7* in *Arabidopsis* produced characteristic phenotypes of the D-class gene family without homeotic conversions. Another two studies also characterized C- and D-class genes in *Phalaenopsis* Hatsuyuki and *Phalaenopsis* Athens including *AG*-like *PhlAG1*, *PhaMADS8*, *PhaMADS10* and *STK*-like *PhlAG2*, *PhaMADS9* (Song et al., 2006; Acri-Nunes-Miranda and Mondragon-Palomino, 2014). Expressions of *PhlAG1* and *PhlAG2* could be detected in the lip, column and ovule, and *PhaMADS8*, *PhaMADS9* and *PhaMADS10* were specifically expressed in the gynostemium and ovary. Expression of *PhaMADS9* but not *PhaMADS8* and *PhaMADS10* is increased in the gynostemium of peloric mutants suggesting that the expression level of *STK*-like gene is crucial for gynostemium development. The C- and D-class genes in *D. crumenatum* and *D. thyrsiflorum* were also characterized; *DcOAG1* and *DthyrAG1* are C-class genes; *DcOAG2* and *DthyrAG2* are D-class genes (Skipper et al., 2006; Xu et al., 2006). *DcOAG1* is expressed in all floral organs while *DcOAG2* is mainly expressed in the ovary and in the envelope cells of pollinia. Overexpression of *DcOAG1* in *Arabidopsis* transformed the sepals and petals into carpel-like and stamen-like structures, respectively. *DthyrAG1* and *DthyrAG2* are expressed in the inflorescences and ovules after pollination, and *DthyrAG2* is believed to play more important roles in later stages of ovule development in *D. thyrsiflorum* since its expression was higher than that of *DthyrAG1* in ovules. The C- and D-class genes in *Oncidium* Gower Ramsey, namely *OMADS4* and *OMADS2* respectively, were also characterized (Hsu et al., 2010). *OMADS4* is expressed in the stamens and carpels whereas expression of *OMADS2* is restricted to the stigmatic cavity and ovary of carpels. In addition, yeast two-hybrid analyses showed that *OMADS4* and *OMADS2* are able to form homodimers by themselves or form heterodimers with each other. Ectopic expression of *OMADS4* and *OMADS2* caused only early or moderately early flowering in *Arabidopsis* without homeotic conversion of floral organs. C- and D-class genes characterized in *Cymbidium* (Wang et al., 2011b), *Orchis* (Salemme et al., 2013) and *Erycina* (Dirks-Mulder et al., 2017), include C-class genes *CeMADS1*, *CeMADS2*, *OitaAG*, *EpMADS20*, *EpMADS21* and *EpMADS22*, and D-class genes *OitaSTK* and *EpMADS23*. *CeMADS1* is only expressed in the column but *CeMADS2* is expressed in all floral organs. In the multitepal mutant whose male and female reproductive organs are replaced by a newly emerged flower, expression of *CeMADS1* is lost, suggesting that *CeMADS1* is associated with the development of the gynostemium. *O. italica OitaAG* and *OitaSTK* are also expressed in the reproductive organs in both the early and late stages. Among the *E. pusilla* MADS-box genes, *EpMADS20* is expressed in all floral organs while others, such as *EpMADS21*, *EpMADS22* and *EpMADS23* are mainly expressed in the gynostemium.

### E-Class Genes

E-class genes show unique interaction with other floral organ identity genes to determine all floral organs. *Arabidopsis* has four *SEP* genes: *SEP1*, *SEP2*, *SEP3*, and *SEP4*. The flowers of a triple mutant of *sep1 sep2 sep3* consist entirely of sepal-like organs (Pelaz et al., 2000) while the quadruple mutant of *sep1 sep2 sep3 sep4* produces leaf-like organs instead of floral organs (Ditta et al., 2004). Moreover, simultaneously reduced expression of four rice *SEP*-like genes (*OsMADS1*, *OsMADS5*, *OsMADS7* and *OsMADS8*) caused the conversion of all floral organs except the lemma into leaf-like structures (Cui et al., 2010). In orchids, a number of genes belonging to the E-class have been characterized at the molecular level. A *SEP3*-like gene, *OM1* was the first E-class gene identified, from the mature flower of *Aranda* Deborah, and its expression was identified in sepals and petals (Lu et al., 1993). In *Phalaenopsis* hybrid Athens, three *SEP*-like genes *PhaMADS4*, *PhaMADS5* and *PhaMADS7* have been identified and these are expressed in the sepal, petal and labellum (Acri-Nunes-Miranda and Mondragon-Palomino, 2014). In *P. aphrodite*, *PaAGL6-1* is expressed in the lip, whereas the *PaAGL6-2* is expressed throughout all floral organs (Su et al., 2013). Four *SEP*-like genes *PeSEP1*, *PeSEP2*, *PeSEP3* and *PeSEP4* were characterized from *P. equestris* (Pan et al., 2014). These *PeSEP* genes are expressed in all floral organs and VIGS of *PeSEP3* resulted in production of leaf-like tapels. Down-regulation of *PeSEP2* expression alone by VIGS has minor effects, but silencing of both *PeSEP2* and *PeSEP3*  expression caused reduced expression of B-class genes such as *PeMADS2, PeMADS3, PeMADS4, PeMADS5* and *PeMADS6* suggesting an association between *PeSEP* functions and B-class gene expression. *DOMADS1* and *DOMADS3* from *Dendrobium* Madame Thong-In are homologous genes of *SEP3* and *SEP4*, respectively, and *DcOSEP1* from *D. crumenatum* is a homolog of *SEP3* (Yu and Goh, 2000b; Xu et al., 2006). Expressions of these genes are constantly activated during floral transition and continue into the mature floral stage. Four E-class genes belonging to the *AP1*/*AGL9* superfamily, *OMADS6*, *OMADS7*, *OMADS10* and *OMADS11* have been identified from *Oncidium*  Gower Ramsey (Chang et al., 2009). *OMADS6* is expressed in all floral organs except the stamen, and expression patterns of *OMADS7* and *OMADS11* are similar to that of *OMADS6*. Unlike *OMADS6*, *OMADS10* is expressed in vegetative leaves although its transcripts are specifically accumulated in the lip and carpels of mature flowers. Moreover, overexpression of *OMADS6*, *OMADS7* and *OMADS11* resulted in extremely early flowering but *OMADS10* overexpression caused moderately early flowering in *Arabidopsis*. Several *AGL6*-like genes including *EpMADS3*, *EpMADS4* and *EpMADS5* have been identified in *E. pusilla*. They are expressed in floral organs with various expression levels in distinct floral organs suggesting that multiple *AGL6*-like genes may also contribute to the development of floral organs (Dirks-Mulder et al., 2017). Seven E-class genes *CgSEP1*, *2*, *3*, *4* and *CgAGL6-1, -2, -3* were characterized in *C. goeringii* (Xiang et al., 2018). The expression level of *CgSEP1* wasincreased in the peloric mutant lips and decreased in the peloric mutant sepals. High expression levels of *CgSEP2* and *CgAGL6-1* could be detected in the sepals, but rarely in the lips and columns of the wild-type and the peloric mutant. An increased expression level of *CgSEP3* was noticed in peloric mutant lip-like petals and lips, but expression level of *CgSEP4* was decreased in the peloric mutant lip-like petals and lips. The expression level of *CgAGL6-2* is increased in all floral organs of the wild-type and peloric mutant, but increased level of *CgAGL6-3* could only be detected the in lip and lip-like petals of wild-type and peloric mutant. Based on expression patterns and phenotypic alterations caused by the E-class genes, it is likely that E-class genes are required for petal, stamen, and carpel formation in both dicot and monocot plants. Of note, a recent report has shown that the greenish flower phenotype of a mutant orchid cultivar in *Habenaria radiata* is due to the absence of SEP (*HrSEP-1*) function (Mitoma and Kanno, 2018). The characterized genes of flower development in orchids are listed in **Table S2** (Aceto and Gaudio, 2011; Mondragon-Palomino, 2013; Teixeira da Silva et al., 2014).

### Connective Codes for Perianth Formation

Genetic models related to the regulatory patterning formation of actinomorphic flowers in *Arabidopsis* and *Antirrhinum* are well characterized. However, orchid flowers usually have zygomorphic symmetry with a prominent, well-differentiated labellum in the inner tepals, and it is morphologically distinct from tepals of other angiosperms (Salemme et al., 2011). The perianth formation in orchids cannot be solely explained by the ABCDE model. Many divergent genes with novel functions are responsible for the regulation of perianth formation in orchids. The model "Orchid Code" has been proposed to explain how the various expression levels of duplicated *AP3*/*DEF*-like genes regulate the perianth morphogenesis in orchids (**Figure 2B**) (Mondragon-Palomino and Theissen, 2008; Mondragon-Palomino and Theissen, 2009). In this model, duplicated *DEF*like paralogous are grouped into four clades phylogenetically and their expression patterns determine the formation of different flower organs. The development of outer tepals (sepal) is regulated by the co-expression of clade 1 (*PeMADS2*-like) and clade 2 (*PeMADS5-like* or *OMADS3*-like) genes whereas the two lateral petals are regulated by the co-expression of clade 1, clade 2 and clade 3 (*PeMADS3*-like) genes. The lip structure is regulated by the co-expression of clade 1, clade 2, clade 3 and clade 4 (*PeMADS4*-like) genes. In contrast, *GLO*-like genes form a single clade and express in all four floral whorls. In other words, the divergent *DEF*-like gene is the key that leads to the morphological diversity of the flowers in orchids. Later, the orchid code model was refined by extensive analysis of *DEF*-like genes in different subfamilies of orchids (Mondragon-Palomino and Theissen, 2011). According to this refined orchid code, four clades of *DEF*-like genes are all expressed in lateral petal and lip. However, lower expression level of clade 3 and clade 4 is displayed in lateral petal whereas lower expression level of clade 1 and clade 2 genes is displayed in the lip. Moreover, a combination of four clade genes with differential expression level may also contribute to the developments of gynostemium and ovary.

Development of different floral organs of orchids has also been proposed by a homeotic orchid tepal (HOT) model (Pan et al., 2011). Twenty-four *AP3*-like genes were identified from 11 species of orchids and grouped into four clades (*PeMADS3* like, *PeMADS4*-like, *PeMADS2*-like and *PeMADS5*-like). Similar to the orchid code model, the sepal organ expresses *PeMADS2* like and *PeMADS5*-like genes whereas the lateral petal expresses *PeMADS2*-like, *PeMADS5*-like and *PeMADS3*-like genes. In the lip organ, all four clade genes are expressed. In addition, the HOT model also addresses the temporal change of *PeMADS3*-like and *PeMADS4*-like expressions in different stages of inflorescence development. In the early floral organ primordial stage, they are expressed in all whorl organs. In the late floral organ primordial and floral bud stages, *PeMADS3*-like genes are restricted in petal, lip and column whereas the *PeMADS4*-like genes are restricted in the lip and column. Combinations of *PI*-like genes and other MADS-box genes also determine the identity of lip and column.

Another study also based on the expression analysis proposed a "Perianth code" (P code) hypothesis to explain the floral identity of orchids (**Figure 2B**) (Hsu et al., 2015). Two clades of *AP3*-like genes (*OAP3-1* and *AP3-2*) in *Oncidium* associate with *OPI* and different clades of O*AGL6*-like genes (E-class). Different heterotetradmers determine different organs. The SP (sepal/petal) complex (*OAP3-1*/*OAGL6-1*/*OAGL6-1*/*OPI*) specifies the sepal/ petal formation whereas the L (lip) complex (*OAP3-2*/*OAGL6-2*/ *OAGL6-2*/*OPI*) is exclusively required for lip formation. The P code model suggests that the diverse E-class genes are also important for development of lip structure of orchids. Diverse roles of E-class genes in floral identity have also been suggested in lily (Tzeng et al., 2003). Two *AGL2*-like genes, *LMADS3* and *LMADS4*, were identified in lily, and they were expressed in all four whorls of floral organs. Ectopic expression of *LMADS3* but bot *LMADS4* in *Arabidopsis* resulted in reduced plant size, early flowering and loss of floral determinacy. In *H. radiata*, two E-class genes (*HrSEP-1* and *HeSEP-2*) were characterized recently (Mitoma and Kanno, 2018). The expression of *HrSEP-1* is lost in a mutant cultivar "Ryokusei" that has greenish petals and a smaller lip in whorl 2, and several septaloid organs and a ventral column in whorls 3 and 4. Moreover, the expression of *HrSEP-2* is up-regulated in Ryokusei. Alternative expressions of B-class genes by silencing of E-class genes have also been revealed in *Phalaenopsis* (Pan et al., 2014). This indicates a feedback regulation between the floral identity genes may also play an important role in the development of floral organs.

All these models specify that the diverse roles of duplicated *AP3*/*DEF*-like and other floral identify genes are important for specialized flower morphogenesis, especially for the lip development of orchid flowers. This idea is supported by the recent study of whole-genome sequencing of *Apostasia shenzhenica*, a genus of Apostasioideae, which is regarded as being phylogenetically plesiomorphic in the orchid family (Zhang et al., 2017). The *Apostasia* has a radially symmetrical (actinomorphic) flower without lip and complex column structures. Comparing *A. shenzhenica* to other subfamilies of Orchidaceae, *A. shenzhenica* seems to have fewer B-class *AP3* genes.

## Orchid Biotechnology

Conventional breeding efforts were initiated more than one century ago and helped propel the growth of the ornamental orchid market. In order to obtain profitable horticultural traits, successful breeding usually proceeds by preserving or selecting advantageous parents or progenies. Application of a wellcontrolled cultivation management system is another important factor for successful commercial orchid farming. In the worldwide market, hundreds and thousands of successful cultivars have been generated using the conventional methods through enthusiastic professional and amateur breeders. The discovery of native tetraploid *Phalaenopsis* orchids that have better flower shapes and color intensities initiated the creation of numerous significant commercial hybrids through polyploidy breeding. Over the last two decades, a successful program to convert 20 diploid *Phalaenopsis* species to tetraploid has been developed by Chen et al. (2011). This program facilitated a breeding process that feeds the needs of the nursery business. Polyploidization can typically be achieved by introduction of colchicine in many agricultural or horticultural crops including orchids (Griesbach, 1981; Caperta et al., 2006; Azmi et al., 2016; Tuwo and Indrianto, 2016). One of the most successful product series, "Big White Flower," is comprised of tetraploid *Phalaenopsis* hybrids developed through polyploidization with the parents *Phalaenopsis* Doris. Among them, *P*. Sogo Yukidian 'V3' has been very popular in the market. At present, it is still one of the most profitable hybrids in this group. Chen et al. concluded that the importance of polyploidy to the improvement of orchid cultivars may rely on the increase in the number of sets of genes (Chen et al., 2011). The accumulation of additive genetic and heterotic effects of these genes could potentially improve the varieties possible. Currently, polyploidy is still a major technique that drives orchid breeding programs to commercial success. An alternative strategy for polyploid breeding is the phenomenon of unreduced gametes. By analyzing sporad types, Bolaños-Villegas et al. observed a certain percentage of dyads (2n gametes) and suggested the pollination from individual of these dyad may produce polyploid progeny as revealed in other plants (Bolaños-Villegas et al., 2008).

In the 1970s, the creation of recombination DNA molecules brought about revolutionary genetic engineering, also called genetic transformation, which constitutes the direct manipulation of an organism's genome using biotechnology. It is a set of technologies that is used to change genetic makeup, including the transfer of genes within and across species boundaries to produce improved or novel traits. In 2013, the blue *Phalaenopsis*  orchids created by the research team of Professor Mii at Chiba University, Japan were exhibited at the 11th Asian Pacific Orchid Conference. These true-blue orchids were genetically transformed using a flavonoid 3',5'-hydroxylase gene derived from *Commelina communis* which was incorporated into the *Phalaenopsis* genome and expressed in the flowers to produce delphinidin, a blue anthocyanin pigment that is also in the flowers of delphinium (larkspur). In 2016, the white *Oncidium* orchids created by a research group from the National Taiwan University, Taiwan were exhibited at the Taiwan International Orchid Show. The genetically engineered white orchids were transformed by a flower specific promotor driving the carotenoid cleavage dioxygenase gene that degrades carotenoids, which led to white petals.

Similar to many other crops, two major transformation systems were successfully used to transport foreign genes into orchid genomes: particle bombardment and *Agrobacterium*-mediated transformation systems. The early genetic transformation studies in orchids were confined to the biolistic-mediated transformation which needed a so-called gene gun to deliver the target genes (Kuehnle and Sugii, 1992; Chia et al., 1994). The first successful genetic transformation reported using the *Agrobacterium*mediated method was the transfer of *gus* gene into *Phalaenopsis*  orchid (Belarmino and Mii, 2000). Many years later, this system also successfully transformed other genera of orchids such as *Dendrobium*, *Cymbidium* and *Oncidium* (Yu et al., 2001; Liau et al., 2003; Chin et al., 2007). Target explants for transformation using protocorms, protocorm-like bodies (PLB) and calluses have been reported so far (Belarmino and Mii, 2000; Mishiba et al., 2005; Chin et al., 2007). Usually, the genetic transformation efficiency has been low for orchid plants and the transformation process is also time-consuming. The general slow growth rate could mean the entire transformation process could take up to 2 or 3 years to reach the stage of two-leaf young plantlets. Recently, a protocol using protocorms that could shorten the transformation process to 8 months was reported for *Phalaenopsis* orchids (Hsing et al., 2016). It will be worth investigating whether a similar time frame can be achieved in other orchid genera. Genome editing, a new and revolutionary genetic engineering technology, is an approach in which a specific target DNA sequence of the genome is altered by adding, removing, or replacing DNA bases in a highly precise manner (Gaj et al., 2013). One particular tool, the CRISPR/Cas system has been developed at an accelerated pace. The application of the CRISPR/Cas system to create engineered crops or plants has yielded fruitful results for further investigation. The technology development and application have been widely reviewed (Bortesi and Fischer, 2015; Yin et al., 2017). One crucial criterion to precisely edit orchid genomes is to rely heavily on the availability of the whole genome sequence information. We predict that the combination of the genome editing technology with bioinformatics analyses will create revolutionary breeding programs for orchid research and development in the future.

## PERSPECTIVES

For annual plants such as *Arabidopsis*, rice and wheat, precise control of flowering time in response to environmental cues is necessary for successful reproduction. For perennial plants like orchids, the flowering network may not be so complicated since vernalization is not required and the effect of photoperiod is limited. The changes in ambient temperature may play a more important role in the regulation of floral transition in orchids. Recently, regulations of microRNAs and anti-florigens such as *TFL1* have been shown to be involved in the flowering control of *A. alpina*, a perennial relative of *A. thaliana* (Wang et al., 2011a; Bergonzi et al., 2013). Although further investigation is required to discover whether orchid *TFL1* also plays a negative role in flowering in orchids, it is no exaggeration to say that *TFL1* may regulate the dormancy of juvenile axillary buds in the flowering season of orchids. Furthermore, regulation of *PHYTOCHROME INTERACTING FACTOR 4* (*PIF4*) *via* phytohormone pathways and sugar metabolism may also be involved (Kumar et al., 2012; Bolouri Moghaddam and Van den Ende, 2013). Flower development studies revealed that multiplicated floral homeotic genes (**Table S2**) are the basis of the specific morphology of orchids (**Figure 2**). However, how the high order complex of these proteins regulates floral organ identity needs to be further investigated. In addition, the feedback regulation between different floral homeotic genes needs further investigation to explain the floral development of orchids in detail. The A and E-class genes are involved in both floral initiation and flower development. It will be interesting to investigate whether A and E-class genes regulate the expressions of other MADS-box genes. Finally, further advances in functional studies on key genes for flowering/flower development may rely on a breakthrough in orchid transformation technology which leads to more efficient results. Recently, the genome sequences of several orchids have been determined. Understanding of the flowering control and flower development of the CAM-using plants may provide an insightful view into plant evolution.

### AUTHOR CONTRIBUTIONS

All the authors wrote and reviewed the manuscript.

#### FUNDING

This work was supported in part by grants from World Vegetable Center Korea Office (WKO #10000379) and core donors to the

#### REFERENCES


World Vegetable Center: Republic of China (ROC), UK aid from the UK government, United States Agency for International Development (USAID), Australian Center for International Agricultural Research (ACIAR), Germany, Thailand, Philippines, Korea and Japan and, a grant from the Ministry of Science and Technology, Taiwan (MOST 106-2321-B-020-002-) to F-C. C. and a grant from Council of Agriculture, Taiwan (COA 106AS-8.6.3-FD-Z1(1)) to F-C. C.

#### ACKNOWLEDGMENTS

The authors are grateful to Miranda Loney for English editing.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2019.01258/ full#supplementary-material


**Conflict of Interest:** 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 Wang, Viswanath, Tong, An, Jang and Chen. 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.*

# Evolutionary Conservation of the Orchid MYB Transcription Factors DIV, RAD, and DRIF

*Maria Carmen Valoroso1, Rómulo Sobral2, Giuseppe Saccone1, Marco Salvemini1, Maria Manuela Ribeiro Costa2 and Serena Aceto1\**

1 Department of Biology, University of Naples Federico II, Naples, Italy, 2 BioSystems & Integrative Sciences Institute (BioISI), Plant Functional Biology Centre, University of Minho, Campus de Gualtar, Braga, Portugal

The MYB transcription factors DIVARICATA (DIV), DIV-and-RAD-Interacting-Factor (DRIF), and the small interfering peptide RADIALIS (RAD) can interact, forming a regulatory module that controls different plant developmental processes. In the snapdragon Antirrhinum majus, this module, together with the TCP transcription factor CYCLOIDEA (CYC), is responsible for the establishment of floral dorsoventral asymmetry. The spatial gene expression pattern of the OitDIV, OitDRIF, and OitRAD homologs of Orchis italica, an orchid with zygomorphic flowers, has suggested a possible conserved role of these genes in bilateral symmetry of the orchid flower. Here, we have identified four DRIF genes of orchids and have reconstructed their genomic organization and evolution. In addition, we found snapdragon transcriptional cis-regulatory elements of DIV and RAD loci generally conserved within the corresponding orchid orthologues. We have tested the biochemical interactions among OitDIV, OitDRIF1, and OitRAD of O. italica, showing that OitDRIF1 can interact both with OitDIV and OitRAD, whereas OitDIV and OitRAD do not directly interact, as in A. majus. The analysis of the quantitative expression profile of these MYB genes revealed that in zygomorphic orchid flowers, the DIV, DRIF1, and RAD transcripts are present at higher levels in the lip than in lateral inner tepals, whereas in peloric orchid flowers they show similar expression levels. These results indicate that MYB transcription factors could have a role in shaping zygomorphy of the orchid flower, potentially enriching the underlying orchid developmental code.

#### Keywords: DIVARICATA, RADIALIS, DRIF, MYB, Orchidaceae

### INTRODUCTION

The MYB proteins DIVARICATA (DIV), RADIALIS (RAD), and DIV-and-RAD-Interacting-Factor (DRIF) are part of a regulatory module involved in distinct developmental processes of plants (Machemer et al., 2011; Raimundo et al., 2013). *DIV* and *DRIF* belong to ancient gene families that emerged in the green algae lineage, whereas the *RAD* genes are more recent as their origin can be dated back to gymnosperms (Raimundo et al., 2018). Canonical DIV transcription factors have two MYB domains (MYBI and II) (Galego and Almeida, 2002), in contrast to RAD and DRIF, both containing a single MYB domain (Corley et al., 2005; Raimundo et al., 2013). In addition to the N-terminal MYB domain, DRIF proteins share the conserved DUF3755 domain at the C-terminus, found only in this protein family and whose ability to bind WUSCHEL-RELATED HOMEOBOX

#### Edited by:

Jen-Tsung Chen, National University of Kaohsiung, Taiwan

#### Reviewed by:

Minsung Kim, University of Manchester, United Kingdom Chao Bian, Beijing Genomics Institute (BGI), China

> \*Correspondence: Serena Aceto serena.aceto@unina.it

#### Specialty section:

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

Received: 04 June 2019 Accepted: 02 October 2019 Published: 01 November 2019

#### Citation:

Valoroso MC, Sobral R, Saccone G, Salvemini M, Costa MMR and Aceto S (2019) Evolutionary Conservation of the Orchid MYB Transcription Factors DIV, RAD, and DRIF. Front. Plant Sci. 10:1359. doi: 10.3389/fpls.2019.01359

(WOX) and KNOTTED1-LIKE HOMEOBOX (KNOX) proteins has been recently described in *Populus trichocarpa* (Petzold et al., 2018). During evolution, the MYB domain has undergone successive rearrangements resulting in the acquisition of specific interaction abilities: the MYB domain of DRIF can interact with the MYBI domain of DIV or with the MYB domain of RAD (Machemer et al., 2011; Raimundo et al., 2013; Raimundo et al., 2018). In such interaction module, the small RAD proteins (less than 100 amino acids in size) have an antagonistic effect on the formation of the DIV/DRIF complex and thus have been classified as small-interfering peptides (siPEP) or microproteins (Seo et al., 2011; Staudt and Wenkel, 2011; Eguen et al., 2015).

The function of the DIV, DRIF, and RAD proteins has been described in different plant species, where they control distinct developmental processes. For example, RAD-like proteins regulate photomorphogenesis and floral transition of *Arabidopsis thaliana* (Hamaguchi et al., 2008; Li et al., 2015), DIV-like proteins are involved in sugar and hormone regulation of *Oryza sativa* (Lu et al., 2002), and the protein complexes DIV/ DRIF and RAD/DRIF control cell expansion of the fruit pericarp of *Solanum lycopersicum* (Machemer et al., 2011). However, the majority of studies regarding the *DIV*, *DRIF*, and *RAD* genes focused on their role in the establishment of flower zygomorphy, an evolutionary novelty that emerged several times in flowering plants from the ancestral condition of radial symmetry (Citerne et al., 2010; Endress, 2012). The first comprehensive analysis of the molecular pathway underlying floral symmetry was conducted in the snapdragon *Antirrhinum majus* (**Figure 1**), showing that mutations of the genes *CYCLOIDEA* (*CYC*), *DIV*, and *RAD* have an effect on symmetry of the flower. The TCP transcription factor *CYC* is expressed in the dorsal part of the flower and activates the expression of *RAD* (Luo et al., 1996; Cubas et al., 1999; Luo et al., 1999; Corley et al., 2005; Costa et al., 2005) through the interaction with 5′-GGNCCC-3′ binding sites in the *RAD* promoter and intron (Costa et al., 2005). The *DIV* and *DRIF* genes are expressed both in the dorsal and ventral domains of the flower of *A. majus* (Almeida et al., 1997; Galego and Almeida, 2002; Raimundo et al., 2013). In the ventral domain, the protein complex DIV/DRIF controls downstream genes involved in the ventralization of the flower. In the dorsal domain, the siPEP RAD binds to DRIF and prevents its interaction with DIV, thus inhibiting ventralization (Raimundo et al., 2013). In addition to its ability to activate ventralization genes, the DIV/DRIF protein dimers can bind the sequence 5′-GATAA-3′ (Raimundo et al., 2013) within the *DIV* promoter, possibly autoregulating its transcriptional activity (Sengupta and Hileman, 2018).

The role of the *CYC*, *DIV*, and *RAD* genes in controlling flower bilateral symmetry outside *A. majus* has been reported in other Lamiales (Zhou et al., 2008; Preston et al., 2009; Reardon et al., 2009; Preston et al., 2011; Reardon et al., 2014; Su et al., 2017) and in Dipsacales (Howarth and Donoghue, 2009; Boyden et al., 2012), whereas only limited knowledge about these genes is available in monocots and basal angiosperms. However, a recent study suggested the recruitment of this molecular network for the establishment of floral zygomorphy before the diversification between monocots and dicots (Madrigal et al., 2019).

proteins are expressed both in the dorsal and ventral parts of the flower. In the dorsal domain, the CYC protein interacts with the promoter and intron of the RAD gene, activating its expression. The RAD protein binds to DRIF and prevents the formation of the DIV/DRIF complex, thus inhibiting ventralization. In the ventral domain, the absence of RAD permits the formation of the DIV/ DRIF complex that controls downstream genes involved in the ventralization.

Among monocots, Orchidaceae is one of the most speciesrich families, adapted to many different habitats (Cozzolino and Widmer, 2005; Aceto and Gaudio, 2011). Most of the orchid species have zygomorphic flowers sharing a common organization of the perianth into three outer and three inner tepals. Zygomorphy of the orchid flower is evident in the diversified and complex morphology of the median inner tepal (labellum or lip; **Figure 2**) (Rudall and Bateman, 2002). The vast majority of studies concerning orchid flower development focused on MADS-box transcription factors (Salemme et al., 2011; Aceto et al., 2014; De Paolo et al., 2014; Cai et al., 2015; Lin et al., 2016; Zhang et al., 2016; Zhang et al., 2017; Valoroso et al., 2019), with particular attention to the *DEFICIENS* (*DEF*) and *AGAMOUS-LIKE 6* (*AGL6*) genes. The pivotal role of these genes in the evolution and formation of the orchid perianth is well explained by the "orchid code" theory and its successive modifications (Mondragon-Palomino and Theissen, 2009; Mondragon-Palomino and Theissen, 2011; Pan et al., 2011; Hsu et al., 2015; Dirks-Mulder et al., 2017). On the contrary, the MYB transcription factors are largely understudied in orchids and their potential involvement in the establishment of orchid flower symmetry has been only recently proposed in *Orchis italica* and *Cattleya trianae* (Valoroso et al., 2017; Madrigal et al., 2019).

To date, eight *DIV*, four *RAD*, and two *DRIF* genes have been reported in *O. italica* and, among them, the corresponding homologs responsible for floral symmetry of *A. majus* have been identified (Valoroso et al., 2017). Phylogeny and genomic organization of the orchid *DIV* and *RAD* genes has also been studied (Valoroso et al., 2017; Madrigal et al., 2019), whereas a description of the *DRIF* gene family is still missing.

The aim of the present study was to expand knowledge on *DIV*, *RAD*, and *DRIF* genes of orchids and to obtain more evidence supporting their involvement in the establishment of flower zygomorphy. We firstly focused on the orchid *DRIF* genes, searching for homologs within the orchid genomes and reconstructing their phylogeny. Then, we scanned the putative promoter and intron of the *DIV* and *RAD* genes to identify known *cis*-regulatory elements conserved between orchids and snapdragon. Finally, we analyzed the interaction ability of the OitDIV, OitRAD, and OitDRIF1 proteins of *O. italica* and examined their transcript abundance in the perianth tissues of zygomorphic and peloric orchid flowers.

### MATERIALS AND METHODS

### Plant Material

The orchids used in this study were grown under natural light and temperature in the greenhouse of the Department of Biology of the University of Naples Federico II (Napoli, Italy). *O. italica* Poir. plants are part of the Orchidaceae collection of the Department of Biology. *Phalaenopsis equestris* (Schauer) Rchb.f. and *Phalaenopsis* Joy Fairy Tale (*Phal*. Ho's Princess Arai × *Phal.* Coral Isles) are commercially available orchids (Giulio Celandroni Orchidee, San Giuliano Terme, Pisa, Italy). *O. italica* and *P. equestris* display flower zygomorphy as the second floral whorl is clearly distinguished into two lateral inner tepals and one median inner tepal (lip) (**Figures 2A**–**E**). The peloric perianth of *Phalaenopsis* Joy Fairy Tale shows two lip-like structures in substitution of the lateral inner tepals, conferring radial symmetry to the flower (**Figures 2F**, **G**).

Single flowers from three different plants of each orchid were collected before (single floret length, ∼1 cm) and soon after anthesis (**Figure 2**). The perianth tissues (outer tepals, inner lateral tepals, and lip) were dissected and stored in RNA-later (Ambion) until RNA extraction.

#### Sequence Retrieval and Phylogenetic Analysis

In order to identify *DRIF* transcripts expressed in flower tissues of *O. italica*, the amino acid sequences corresponding to the DUF3755 domain of the known *O. italica* OitDRIF1 and two proteins (GenBank accession numbers MK834277 and MK834278, respectively) (Valoroso et al., 2017) were used as queries to scan the inflorescence transcriptome of *O. italica* (De Paolo et al., 2014) through tBLASTn searches. Using the same approach, *DRIF* homologs were searched within the transcriptomes of other orchids present in the *Orchidstra* database (Chao et al., 2017) and within the genome of *Apostasia shenzenica*, *P. equestris*, and *Dendrobium catenatum* (Cai et al., 2015; Zhang et al., 2016; Zhang et al., 2017). Recently identified *DRIF* nucleotide and amino acid sequences (Raimundo et al., 2018) were downloaded from SustainPine (http://www. scbi.uma.es/sustainpinedb/home\_page), Monocots PLAZA (https://bioinformatics.psb.ugent.be/plaza/versions/plaza\_v4\_ monocots/), NCBI (https://www.ncbi.nlm.nih.gov/), and Araport (https://www.araport.org/). The species and the corresponding accession numbers of the DRIF sequences used in the present work are listed in **Table S1**.

The amino acid sequences of the DRIF homologs identified were aligned with MAFFT (Katoh and Standley, 2013) and the resulting alignment was manually adjusted. Poorly aligned positions were removed using GBLOCKS (Talavera and Castresana, 2007) and maximum likelihood (ML) phylogenetic tree was constructed with RAxML v 8.2.10 (Stamatakis, 2014) using the default settings, with 1,000 bootstrap replicates.

#### Analysis of Conserved Transcription Factor Binding Sites

Approximately 3 kb noncoding sequences upstream of the translation start site of the *DIV* and *RAD* genes of the orchids *A. shenzenica*, *D. catenatum*, and *P. equestris* were downloaded from the corresponding genomes deposited at NCBI (**Table S1**). These putative promoter sequences were scanned for the presence of conserved transcription factor binding sites (TFBSs) using the *PlantPAN 3.0* database (Chow et al., 2019). The Multiple Promoter Analysis search mode was applied to identify known conserved plant TFBSs shared within each gene group of putative promoters. Specific nucleotide motifs known as TFBSs of CYC (5′-GGNCCC-3′) (Kosugi and Ohashi, 2002; Costa et al., 2005; Yang et al., 2012; Gao et al., 2015; Sengupta and Hileman, 2018) and DIV (5′-VGATAMSV-3′) (Raimundo et al., 2013; Sengupta and Hileman, 2018) of *A. majus* were searched within the orchid *RAD* and *DIV* putative promoters, respectively. In addition, the intron sequences of the *RAD* genes of *O. italica*, *P. equestris*, *D. catenatum* (Valoroso et al., 2017), and *A. shenzenica* were scanned for the presence of the CYC TFBS, as described above.

### Expression Analysis

Total RNA was extracted from the perianth tissues (outer tepals, inner tepals, and lip, before and after anthesis) of *O. italica*, *P. equestris*, and *Phalaenopsis* Joy Fairy Tale using Trizol (Ambion) followed by DNase treatment. After RNA extraction and quantification, 500 ng of total RNA from each tissue were reverse-transcribed using the Advantage RT-PCR kit (Clontech) and a mix of oligo dT and random hexamer primers.

In order to validate the nucleotide sequence of the four *OitDRIF* transcripts identified in the inflorescence transcriptome of *O. italica*, specific primer pairs were designed (**Table S2**) and used to amplify the cDNA of *O. italica* inflorescence. The amplification products obtained were cloned into pSC-A-amp/kan vector (Agilent), sequenced

using the T3 and T7 primers, and analyzed using an ABI 310 Automated Sequencer (Applied Biosystems). Their sequence was compared with that of the transcripts identified in the transcriptome of *O. italica*.

Relative expression of the orchid *DIV*, *RAD*, and *DRIF1* genes was evaluated by real-time PCR experiments, using 18S as reference gene, as previously described (De Paolo et al., 2015; Valoroso et al., 2017). Primer pairs are listed in the **Table S2**. Reactions were conducted in biological triplicates and technical duplicates. Mean ± SEM was calculated for each duplicate and biological triplicate. Gene relative expression level in inner lateral tepals and lip was normalized relative to outer tepals. Two-tailed *t* test was conducted to assess the statistical significance of the relative expression differences between lateral inner tepals and lip of each species, before and after anthesis.

## Yeast Two-Hybrid Analysis

The coding sequences (CDSs) of the *OitDIV* (KY089088), *OitRAD* (KY089097), and *OitDRIF1* (MK834277) homologs of *O. italica* were PCR amplified using the primer pairs listed in **Table S2** and 500 ng of cDNA of *O. italica* inflorescence. To analyse protein– protein interactions between OitDIV, OitRAD, and OitDRIF1, the GAL4-based yeast two-hybrid (Y2H) system (Matchmaker two-hybrid system; Clontech) was used. The amplified CDSs of *OitDIV*, *OitRAD*, and *OitDRIF1* were cloned into bait (pGBT9) and prey (pGAD424) vectors (Clontech). All the prey and bait recombinant vector combinations were used to transform *Saccharomyces cerevisiae* strain AH109 (Gietz et al., 1995), conducting each experiment in triplicate. Plasmid presence after double yeast transformations was checked by growing cells in Synthetic Defined (SD) medium lacking tryptophan and leucine. Protein ability to interact with each other was evaluated in SD medium lacking tryptophan, leucine, and histidine. Possible transcriptional activation activity of OitDIV, OitRAD, and OitDRIF1 proteins fused to the binding domain of GAL4 (pGBT9 vector) was verified by monitoring growth of yeast transformed cells in SD medium without histidine, in the presence of 10 mM 3-aminotriazole. Empty vectors pGBT9 or pGAD424 were transformed in combination with the recombinant vectors as negative controls.

## RESULTS AND DISCUSSION

#### Identification and Phylogenetic Analysis of the Orchid DRIF Genes

To date, the *DRIF* genes of orchids have been identified only in *O. italica*, where the expression pattern of *OitDRIF1* and *OitDRIF2* was analysed in floral tissues (Valoroso et al., 2017). Evolutionary analysis has demonstrated the ancient origin of the *DRIF* genes: they have been found (together with the *DIV* genes) from green algae to angiosperms. In angiosperms, the *DRIF* homolog number in the examined species is generally five (Raimundo et al., 2018). These findings led us to search for other *DRIF* genes expressed in *O. italica*, to identify their homologs in other orchids and to verify the number of *DRIF* genes within the available genome of orchid species, currently restricted to the only subfamilies Epidendroideae and Apostasioideae.

Among plants, the DUF3755 domain is unique to DRIF proteins (Raimundo et al., 2018), and when we used it as query to identify other *DRIFs* expressed in the inflorescence transcriptome of *O. italica* we found two different transcripts, in addition to those previously identified, named *OitDRIF3* and *OitDRIF4* (accession numbers MK834279 and MK834280, respectively). All the four *OitDRIF* transcripts of *O. italica* encode for proteins containing the N-terminal MYB domain and the C-terminal DUF3755 domain. Within the genome of *P. equestris* and *D. catenatum* (both belonging to the subfamily Epidendroideae), we found four and three *DRIF* genes, respectively. In the genome of *A. shenzenica*, belonging to the basal subfamily Apostasioideae, we found three *DRIF* genes. All these orchid *DRIF* genes encode for proteins containing the MYB and the DUF3755 domain.

To cover all the five subfamilies of Orchidaceae, we scanned the transcriptomes of *Ophrys sphegodes*, belonging to the same subfamily of *O. italica* (Orchidoideae), *Cypripedium formosanum* (Cypripedioideae), and *Vanilla planifolia* (Vanilloideae) present in the database *Orchidstra* (Chao et al., 2017), a transcriptomics collection for Orchidaceae. In *C. formosanum* we found four *DRIF* transcripts, three in *O. sphegodes* and two in *V. planifolia*. Although some of them are not full-length transcripts, missing part of the N-terminus, all contain both the MYB and DUF3755 domains. The graphical view of the amino acid alignment of the orchid DRIF proteins is reported in **Figure S1**, where the consensus sequences of the MYB and DUF3755 domains are shown.

We reconstructed the genomic organization of the orchid *DRIF* genes from the assembled genomes of *P. equestris*, *D. catenatum*, and *A. shenzenica* and compared it to that of the *DRIF* genes of *A. majus*, whose genome assembly has been recently released (Li et al., 2019). Based on the exon/ intron number, it is possible to divide the *DRIF* genes into two structural types: seven exons–six introns, shared by the *DRIF1-4* genes of *A. majus* and one *DRIF* gene of *A. shenzenica*, *D. catenatum*, and *P. equestris*; eight exons–seven introns, displayed by the *DRIF5-6* genes of *A. majus*, two *DRIF* genes of *A. shenzenica* and *D. catenatum*, and three of *P. equestris* (**Figure 3** and **Table S3**). Exon size is quite well conserved both among orchids and between orchids and snapdragon, whereas intron size is variable, with very large introns in orchids, reflecting a common feature of the orchid genomes due to the high number of transposable elements (Salemme et al., 2013a; Salemme et al., 2013b; Cai et al., 2015; Zhang et al., 2016; Zhang et al., 2017).

To understand the evolutionary relationships among the orchid DRIF proteins and the DRIFs of other plant species, we constructed the phylogenetic tree shown in **Figure 4**. The DRIF proteins of Tracheophyta are included in the clade I, whereas the three DRIF sequences of the moss *Physcomitrella patens* (Bryophyta) belong to the ancestral clade II, in agreement with the DRIF phylogeny recently described (Raimundo et al., 2018). Within the clade I, the orchid DRIFs (in the phylogenetic tree highlighted in different shades of pink) form three distinct groups, two belonging to the subclade Ia and one to the subclade Ib (**Figure 4**). Based on the tree topology, the orchid DRIFs seem to have originated by duplication events predating the diversification between monocots and dicots. In fact, within both subclades Ia and Ib, the orchid branches are grouped with other monocots (*O. sativa*, highlighted in green) and dicots (e.g., *A. majus* and *A. thaliana*, highlighted in different shades of blue). However, lineage-specific duplications have occurred during the evolution of DRIFs in orchids, leading to the formation of two paralog groups within the subclade Ia. In particular, an orchidspecific duplication generated the paralog groups that include, among others, OitDRIF2 and 3. The orchid homologs of DRIF1 of *A. majus* belong to the subclade Ib, where orchid-specific duplication events seem to be absent. In fact, within the subclade Ib, a single DRIF is present for each orchid species, including

FIGURE 3 | Genomic organization of the DRIF genes of Phalaenopsis equestris, Dendrobium catenatum, Apostasia shenzenica (Orchidaceae), Antirrhinum majus (Lamiales), and Physcomitrella patens (Bryophyta). Light blue boxes and black lines represent exons and introns, respectively. Ama, A. majus; Ash, A. shenzenica; Dcat, D. catenatum; Peq, P. equestris; Ppa, P. patens. Clades Ia, Ib, and II are referred to the main groups detected in the DRIF phylogeny (see figure). The code number following the abbreviation of the species name is the accession number of the DRIF sequences deposited in public databases (Table S1).

majus (light blue); Ata, Arabidopsis thaliana (blue); Atr, Amborella tricopoda (red); Ash, Apostasia shenzenica; Cfo, Cypripedium formosanum; Dca, Dendrobium catenatum; Nnu, Nelumbo nucifera (dark green); Oit, Orchis italica; Osa, Oryza sativa (green); Osp, Ophrys sphegodes; Peq, Phalaenopsis equestris; Ppa, Physcomitrella patens (yellow); Ppi, Pinus pinaster (gray); Vpl, Vanilla planifolia. The code number following the abbreviation of the species name is the accession number of the DRIF sequences deposited in public databases (Table S1).

OitDRIF1 of *O. italica*, and the orchid DRIF phylogeny reflects the evolutionary relationships existing within Orchidaceae: Apostasioideae, Cypripedioideae, and Vanilloideae are the most basal subfamilies, Epidendroideae and Orchidoideae the most derivates (Givnish et al., 2015). The phylogenetic relationships among DRIF proteins coincide with the division proposed based on the genomic organization of the *DRIF* loci examined, showing that the genes belonging to the subclade Ia have eight exons, whereas those belonging to the subclade Ib have seven exons. The genomic organization of the *DRIF* genes of the moss *P. patens*, belonging to the basal clade II, shows the presence of seven exons and six introns (**Figure 3** and **Table S3**). The exon/intron structure shared by genes belonging to clade II and subclade Ib indicates that this genomic organization might be the ancestral condition and that the subclade Ia might have originated through a split of exon 7, followed by lineage-specific evolution of exon size.

Compared to the *DRIF* genomic organization, *DIV* and *RAD* genes have a significantly different structure with two exons and one intron (Valoroso et al., 2017). This difference supports the hypothesis previously proposed on the evolutionary origin of the *DIV*, *RAD*, and *DRIF* genes based on the comparison of their MYB domain (Raimundo et al., 2018). The *RAD* genes might have originated through the loss of the region of the *DIV* genes encoding for the MYBII domain. The *DIV* and *DRIF* genes might have evolved from a common ancestral gene through lineage-specific rearrangements (e.g., duplications or gene fusion events), resulting in different exon/intron organizations and acquisition of the region encoding for the MYBII (*DIV* lineage) or DUF3755 domain (*DRIF* lineage). Phylogenetic trees of the DIV and RAD proteins of orchids and other plant species are reported in **Figures S2** and **S3**, respectively. Both the DIV and RAD trees have been produced by the same approach used for the construction of the DRIF tree (see *Materials and Methods*).

#### Analysis of the Conserved TFBS

Some aspects of transcriptional regulation of the genes involved in floral bilateral symmetry are known for *DIV* and *RAD* of *A. majus*. In snapdragon, the expression of the *RAD* gene is activated by direct interaction of the TCP transcription factor CYC possibly through the binding to three conserved TFBSs, two located within the promoter and one within the intron of *RAD* (Costa et al., 2005). The presence of conserved CYC TFBSs within the promoter of *RAD* has been recently reported also in other Lamiales species (Sengupta and Hileman, 2018).

Within the genomic sequence upstream of the translation start site of the *RAD* gene of *P. equestris*, *D. catenatum*, and *A. shenzenica*, we found distinct conserved TFBSs (**Table S4**), among which TCP binding sites. The sequence 5′-GGNCCN-3′, very similar to the *A. majus* CYC consensus binding site 5′-GGNCCC-3′, is present in the putative *RAD* promoter of *A. shenzenica* (three sites) and *D. catenatum* (two sites) (**Table 1**). Its absence in *P. equestris* is possibly due to the lack of a complete sequence information of the upstream region (only 958 bp are currently available) of the *RAD* gene in the corresponding genomic scaffold. Within the *RAD* intron, the sequence 5′-GGNCCN-3′ is present in *D. catenatum* (four sites), *P. equestris* (one site), and *O. italica* (two sites), whereas in *A. shenzenica* it is not present. In *D. catenatum* and *P. equestris* one of the sites exactly matches the canonical CYC TFBS of *A. majus*. The presence of putative CYC target sequences within the promoter and intron of the *RAD* gene (**Table 1**) may suggest a conserved direct transcriptional regulation of *RAD* by CYC in zygomprphic orchid flowers. In *P. equestris*, *D. catenatum*, and possibly *O. italica* both promoter and intron *cis*-regulatory motifs might be necessary to activate the transcription of *RAD* in the specific spatial domain linked to zygomorphy, as in *A. majus*. The absence of these regulatory sequences within the *RAD* intron of *A. shenzenica* is in agreement with this hypothesis, *A. shenzenica* being a basal orchid species with radially symmetric perianth (Zhang et al., 2017). The putative CYC binding sequence of orchids diverged to some extent from that of *A. majus* and other Lamiales and possibly the orchid CYC protein co-evolved to recognize slightly different sequences.

In *A. majus*, two putative DIV binding sites have been identified *in silico* within the *DIV* promoter, suggesting the existence of an autoregulatory loop that maintains the transcription of *DIV*

TABLE 1 | Predicted binding sites of CYC (5′-GGNCCC-3′) and DIV (5′-VGATAMSV-3′) of Antirrhinum majus in the putative promoter and intron of RAD and in the putative promoter of DIV, respectively, in the orchids Phalaenopsis equestris, Dendrobium catenatum, and Apostasia shenzenica.


The analysis of the RAD intron was conducted also in Orchis italica. The nucleotide positions of the putative promoters are indicated with negative numbers, considering as +1 the first nucleotide of the translation start site codon ATG. The nucleotide positions of the orchid RAD introns are numbered considering as +1 the first nucleotide of the intron sequence. The binding sites exactly conserved among orchids and snapdragon are in bold.

(Sengupta and Hileman, 2018). Within the genomic sequence upstream the translation start site of the *DIV* gene of *P. equestris*, *D. catenatum*, and *A. shenzenica* many conserved TFBSs are present, among which MYB binding sites (**Table S4**). The target binding sequence of DIV of *A. majus* 5′-VGATAMSV-3′ is present in *A. shenzenica* (three sites) and *D. catenatum* (one site), whereas the three sequences found in *P. equestris* have A or T instead of C or G in the seventh position (**Table 1**). Although the DIV binding site of orchids is only partially conserved, these results suggest that also in orchids the transcriptional activity of *DIV* might be regulated by a positive feedback.

#### Protein Interactions and Expression Pattern

In flowering plants, the involvement of the DRIF/DIV and DRIF/ RAD complexes in floral zygomorphy has been demonstrated in *A. majus* (Raimundo et al., 2013) and inferred in a few other species (Garces et al., 2016; Madrigal et al., 2019).

To date, the interaction among the DRIF, DIV, and RAD proteins in orchids has never been tested. We used the Y2H assay and found that in yeast the OitDRIF1 protein of *O. italica* can interact both with OitDIV and OitRAD, whereas OitDIV and OitRAD do not directly interact (**Figure 5**). This result is in agreement with the ancient evolutionary origin of this interaction module. In fact, the ability of the DIV and DRIF proteins to interact has evolved early, coincident with their origin in the green algae lineage. Later, with the emergence of the *RAD* genes in gymnosperms, the DRIF–RAD interaction has evolved (Raimundo et al., 2018).

To obtain evidences about the possible involvement of the orchid *DIV*, *RAD*, and *DRIF* genes in zygomorphy of orchid flowers, we examined their expression profile in the perianth tissues of *O. italica* and *P. equestris*, both with zygomorphic flower, and in a peloric *Phalaenopsis* missing the bilateral symmetry due to the presence of three lips in the second floral whorl (**Figure 2G**). **Figure 6** shows the expression levels, before and after anthesis, of the orchid *DIV*, *RAD*, and *DRIF1* genes in inner tepals and lips normalized with respect to outer tepals. In the lip of *O. italica* and *P. equestris*, the expression level of the orchid *DIV*, *RAD*, and *DRIF1* is significantly higher than in inner tepals after anthesis. This same pattern is observed before anthesis for the three genes in *P. equestris* and only for *RAD* in *O. italica*, where *OitDIV* and *OitDRIF1* are expressed at similar levels in lip and inner tepals. This expression profile is in agreement with the previous reports about the expression of these genes in *O. italica* and *C. trianae* that suggested a possible involvement of the orchid *DIV*, *RAD*, and *DRIF1* in zygomorphy of the orchid perianth (Valoroso et al., 2017; Madrigal et al., 2019). Interestingly, the expression levels of *DIV*, *RAD*, and *DRIF1* in the peloric *Phalaenopsis* are similar in the lip and in the lip-like structures present in substitution of inner tepals, both before and after anthesis. The absence of a clear morphological differentiation among the structures of the second whorl in this peloric *Phalaenopsis* is associated with similar expression levels of the three genes, in particular of *RAD*. These results support the model in which in the zygomorphic flower of *O. italica* and *P. equestris*, DIV and DRIF, expressed in all the perianth organs as in *A. majus*, can interact in inner tepals, in which *RAD* expression is lower than

in lip, and control ventralization. Regardless of DIV and DRIF1 levels, in the lip, RAD competes with DIV for the binding to DRIF1 and prevents the formation of the DIV–DRIF complex, thus inhibiting ventralization (Valoroso et al., 2017). Both in *O. italica* and *P. equestris*, the levels of *RAD* are lower in inner tepals than in lip. Consequently, in inner tepals, DIV could interact with DRIF1 and activate ventralization. On the contrary, the presence of higher levels of *RAD* in lip could allow the formation of the RAD/DRIF1 complex and inhibit the interaction between DIV and DRIF1, thus preventing ventralization. In the peloric *Phalaenopsis*, there is competition between DIV and RAD also in the lip-like structures of the second whorl and consequently ventralization is suppressed. The apparent rotation of the orchid

Phalaenopsis Joy Fairy Tale. The relative expression Rn (expressed as log10 of the mean between technical duplicates of three biological replicates) is normalized with respect to 18S endogenous gene and outer tepal tissue. The bars represent SEM and the asterisks indicate significant difference in relative expression between lateral inner tepals and lip assessed by t test. Early and late indicate before and after anthesis stage, respectively.

model with respect to that of *A. majus* is due to resupination: the 180° rotation of the pedicel shifts the lip (a dorsal structure) to a ventral position during orchid flower development.

The results here obtained suggest the involvement of the orchid *DIV*, *RAD*, and *DRIF* genes in the zygomorphy of the orchid perianth and their conserved function, in addition to their conserved interaction ability, in species displaying flower zygomorphy.

Previous studies have demonstrated the role of the MADS-box genes in orchid flower development and various models have been proposed to explain the evolution and formation of the orchid perianth (Aceto and Gaudio, 2011; Salemme et al., 2013a; Acri-Nunes-Miranda and Mondragon-Palomino, 2014; Hsu et al., 2015; Dirks-Mulder et al., 2017; Valoroso et al., 2019). All these models propose interaction among different MADS-box transcription factors and attribute a crucial role to the expression levels of four *DEF* and three *AGL6* genes that permit or prevent the formation of specific protein quartets that drive the formation of specific parts of the perianth (outer tepals, inner tepals, lip) (Mondragon-Palomino and Theissen, 2009; Mondragon-Palomino and Theissen, 2011; Pan et al., 2011; Hsu et al., 2015; Dirks-Mulder et al., 2017). Our findings suggest the existence of a second, MYB-based pathway underlying flower organ arrangement in orchids.

The very recent advances in functional genetic studies of orchids (Kui et al., 2017) open the road to investigate the potential link between these two different molecular pathways (MADS- and MYBbased) in the formation of the orchid flower and its zygomorphy and to clarify if and what kind of interaction exists, to obtain an integrated view of this complex developmental process.

#### DATA AVAILABILITY STATEMENT

The datasets generated for this study can be found in the NCBI nucleotide, DRIF1: MK834277, DRIF2: MK834278, DRIF3: MK834279, DRIF4: MK834280.

### AUTHOR CONTRIBUTIONS

MV performed the research, analyzed the data, and participated to write the paper. RS performed the research and analyzed the data. GS, MS, and MMRC analyzed the data and participated in the paper writing process. SA designed the research, analyzed the data, and wrote the paper.

#### REFERENCES


#### FUNDING

This study was financially supported by grant Ricerca dipartimentale 2018 from the University of Naples Federico II and by Fundação para a Ciência e Tecnologia/Ministério da Ciência, Tecnologia e Ensino Superior through national funds (Programa de Investimento e Despesas de Desenvolvimento da Administração Central) with a project grant PTDC/BIA-PLA/1402/2014 and by FCT/MCTES/PIDDAC (Portugal) under the project PEst-OE/BIA/UI4046/2014; UID/ MULTI/04046/2013.

### ACKNOWLEDGMENTS

We thank Prof. Giovanni Scopece for plant material and Prof. Luciano Gaudio for critical reading of the manuscript.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2019.01359/ full#supplementary-material


**Conflict of Interest:** 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 Valoroso, Sobral, Saccone, Salvemini, Costa and Aceto. 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.*

# Maltose Processing and Not β-Amylase Activity Curtails Hydrolytic Starch Degradation in the CAM Orchid Phalaenopsis

#### *Nathalie Ceusters1, Mario Frans1, Wim Van den Ende2 and Johan Ceusters1,3\**

1 KU Leuven, Department of Biosystems, Division of Crop Biotechnics, Research Group for Sustainable Crop Production & Protection, Campus Geel, Geel, Belgium, 2 KU Leuven, Department of Biology, Laboratory of Molecular Plant Biology, Leuven, Belgium, 3 UHasselt, Centre for Environmental Sciences, Environmental Biology, Diepenbeek, Belgium

#### Edited by:

Katharina Nargar, Commonwealth Scientific and Industrial Research Organisation (CSIRO), Australia

#### Reviewed by:

Ian Joseph Tetlow, University of Guelph, Canada Alberto A. Iglesias, National University of the Littoral, Argentina

> \*Correspondence: Johan Ceusters johan.ceusters@kuleuven.be

#### Specialty section:

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

Received: 09 August 2019 Accepted: 08 October 2019 Published: 14 November 2019

#### Citation:

Ceusters N, Frans M, Van den Ende W and Ceusters J (2019) Maltose Processing and Not β-Amylase Activity Curtails Hydrolytic Starch Degradation in the CAM Orchid Phalaenopsis. Front. Plant Sci. 10:1386. doi: 10.3389/fpls.2019.01386

Crassulacean acid metabolism (CAM) is one of the three photosynthetic pathways in higher plants and is characterized by high water use efficiency. This mainly relies on major nocturnal CO2 fixation sustained by degradation of storage carbohydrate such as starch to provide phosphoenolpyruvate (PEP) and energy. In contrast to C3 plants where starch is mainly degraded by the hydrolytic route, different observations suggested the phosphorolytic route to be a major pathway for starch degradation in CAM plants. To elucidate the interplay and relevant contributions of the phosphorolytic and hydrolytic pathways for starch degradation in CAM, we assessed diel patterns for metabolites and enzymes implicated in both the hydrolytic route (β-amylase, DPE1, DPE2, maltase) and the phosphorolytic route (starch phosphorylase) of starch degradation in the CAM orchid Phalaenopsis "Edessa." By comparing the catalytic enzyme activities and starch degradation rates, we showed that the phosphorolytic pathway is the major route to accommodate nocturnal starch degradation and that measured activities of starch phosphorylase perfectly matched calculated starch degradation rates in order to avoid premature exhaustion of starch reserves before dawn. The hydrolytic pathway seemed hampered in starch processing not by β-amylase but through insufficient catalytic capacity of both DPE2 and maltase. These considerations were further corroborated by measurements of enzyme activities in the CAM model plant Kalanchoë fedtschenkoi and strongly contradict with the situation in the C3 plant Arabidopsis. The data support the view that the phosphorolytic pathway might be the main route of starch degradation in CAM to provide substrate for PEP with additional hydrolytic starch breakdown to accommodate mainly sucrose synthesis.

Keywords: hydrolytic starch degradation, phosphorolytic starch degradation, DPE2, maltase, crassulacean acid metabolism, Phalaenopsis

## INTRODUCTION

Crassulacean acid metabolism (CAM) is one of the three photosynthetic pathways present in higher plants and is characterized by an optimized water use efficiency (WUE) by taking up CO2 predominantly at night when evapotranspiration rates are low. It is convenient to recognize four distinct phases of gas exchange in CAM plants, which are also used to describe the photosynthetic performance. Sequestration of CO2 at night (Phase I) occurs *via* the enzyme phosphoenolpyruvate carboxylase (PEPC) with the 3-C substrate phosphoenolpyruvate (PEP) provided by the glycolytic breakdown of carbohydrate. The final 4-C product, malic acid, is stored in a large central vacuole. At the start of the next day, in Phase II, stomata will gradually close and external CO2 is mainly fixed by ribulose-1,5 bisphosphate carboxylase-oxygenase (rubisco). Gas exchange is curtailed by stomatal closure during the middle of the day (Phase III), thereby reducing transpirational water losses and improving WUE. During this phase, malic acid exits the vacuole and decarboxylation releases CO2 which is re-fixed by rubisco. In Phase IV stomata open again towards the end of the day and external CO2 is mainly sequestered *via* rubisco (Osmond, 1978).

Because of the tight relationship between PEP availability and nocturnal CO2 fixation (Dodd et al., 2003; Ceusters et al., 2019), the provision of carbon skeletons for the synthesis of PEP represents a significant sink for carbohydrate. Besides starch, also glycolysis driven by sucrose hydrolysis can attribute to PEP formation. Various studies with CAM bromeliads showed diurnal accumulation of both starch and soluble sugars followed by nocturnal decrease to fuel the dark reactions of CAM (Christopher and Holtum, 1998; Popp et al., 2003; Ceusters et al., 2008; Ceusters et al., 2009a; Ceusters et al., 2009b; Ceusters et al., 2011; Ceusters et al., 2014). Therefore, transitory starch plays a crucial role in many CAM plants. In contradiction to the starch biosynthesis pathway for which the major steps have been well characterized, the process of transitory starch breakdown in CAM has not yet been elucidated (Lloyd and Kossmann, 2015). In starch storing CAM species different modes of starch degradation might be deployed for nocturnal starch breakdown, i.e. the phosphorolytic and/or the hydrolytic route (Borland et al., 2016; Brilhaus et al., 2016). Both routes are assumed to share the same enzymes to induce the initial attack to starch granules. Glucan water dikinase (GWD) initially phosphorylates the starch granule which consists of branched amylopectin and linear amylose. Debranching enzymes limit dextrinase (LDA) and isoamylase (ISA) bring about cleavage of the branch points in amylopectin. Different enzymes are subsequently used for further processing of linear amylose into soluble sugars (Weise et al., 2011). Whilst phosphorolytic starch degradation in the chloroplast mainly delivers glucose-6-phosphate (Glc6P) for export, the export product resulting from the hydrolytic starch breakdown is maltose (Kore-eda and Kanai, 1997; Weise et al., 2004).

Several observations indicate the phosphorolytic pathway to be the main route in CAM. The facultative CAM plant *Mesembryanthemum crystallinum* (common ice plant) is able to switch from C3 to CAM under drought or salt stress and has emerged as a useful model for exploring CAM. By shifting to the CAM mode, a considerable export of Glc6P out of the chloroplast was observed whilst maltose export dramatically diminished (Neuhaus and Schulte, 1996). In addition, consistent increases in transcript abundances and in the activity of enzymes involved in the phosphorolytic route were observed when switching to CAM mode (Paul et al., 1993; Dodd et al., 2003; Cushman et al., 2008; Taybi et al., 2017). Phosphorolytic degradation of starch is mediated *via* starch phosphorylase, which releases glucose 1-phosphate (Glc1P) from the non-reducing ends of glucan chains. Subsequently, phosphoglucomutase (PGM) catalyses the conversion of Glc1P to Glc6P which is exported to the cytosol *via* the plastidic Glucose 6-Phosphate: Phosphate Translocator (GPT). GPT transcript abundance has been found to show a > 70-fold upregulation in *M. crystallinum* upon CAM induction (Haüsler et al., 2000; Cushman et al., 2008; Kore-eda et al., 2013). In C3 plants the phosphorolytic mode of starch degradation has been proposed to provide carbon solely for metabolism inside the chloroplast *via* the pentose phosphate pathway, particularly under stress conditions when photorespiration is elevated (Weise et al., 2006; Zeeman et al., 2007).

The hydrolytic route of starch degradation is mainly mediated *via* β-amylase, leading to the production of maltose. Besides maltose, a range of short malto-oligosaccharides will emerge as by-products. These require further processing by chloroplastic glucanotransferase disproportionating enzyme (DPE1) to yield glucose and a spectrum of larger, linear oligosaccharides which can further be metabolized again by β-amylase. Notwithstanding α-amylase also possesses the capacity to degrade transitory starch, no phenotype has been detected in knockout mutants of Arabidopsis (Yu et al., 2005). To reach the cytosol glucose and maltose require the dedicated transporters Maltose Excess Protein1 (MEX1) and Plastidic Glucose Transporter (pGlcT) (Niittylä et al., 2004; Smith et al., 2005; Cho et al., 2011). In Arabidopsis, the major export product out of the chloroplast has been found to be maltose and further processing in the cytosol requires a cytosolic disproportionating enzyme (DPE2), which transfers one glucose unit from maltose to an acceptor (a form of cytosolic heteroglycan) and releases the other glucose molecule (Lu and Sharkey, 2004). Based on calculations of starch breakdown in Arabidopsis wild types and DPE2 mutants, direct hydrolysis by maltase (α-glucosidase) has been considered of minor importance (Chia et al., 2004). The potential contribution of maltase in CAM plants is yet unknown. In the cytosol, cytosolic hexokinase catalyses the conversion from glucose to Glc6P, which together with the Glc6P derived from the phosphorolytic degradation, can be used for further sucrose synthesis (Borland et al., 2016).

Several studies on the common ice plant (*M. crystallinum*) demonstrated the upregulation of transcript abundances and activities of a range of starch-degrading enzymes implicated in both the hydrolytic and phosphorolytic routes of starch degradation (α-amylase, β-amylase, starch phosphorylase, and disproportionating enzyme (DPE1 and DPE2)) following CAM induction (Paul et al., 1993; Neuhaus and Schulte, 1996; Haüsler et al., 2000; Dodd et al., 2003; Cushman et al., 2008). These observations pose important questions about 1) the relative contribution of both routes of starch degradation in CAM and 2) whether specific enzymes might be rate limiting in either phosphorolytic and hydrolytic pathways. Exploiting the phosphorolytic pathway might confer specific advantages for CAM plants as Glc6P not only functions as an allosteric activator for PEPC but its glycolytic conversion in the cytosol also provides ATP, which adds in energizing nocturnal accumulation of malate in the vacuole (Holtum et al., 2005; Weise et al., 2011; Cheung et al., 2014). However, it has recently been postulated that the use of both the hydrolytic and phosphorolytic starch degradation pathway in CAM plants might provide a means to accommodate carbohydrate partitioning between competing forces of growth and nocturnal carboxylation during the diel cycle (Borland et al., 2016). To elucidate the interplay and relevant contributions of the phosphorolytic and hydrolytic pathways for starch degradation we assessed diel *in vitro* activity patterns for enzymes implicated in either the hydrolytic route (β-amylase, DPE1, DPE2, maltase) or the phosphorolytic route (starch-phosphorylase) of starch degradation in the CAM orchid *Phalaenopsis* "Edessa." These analyses were further complemented with measurements of diel leaf gas exchange and important metabolite dynamics (starch, malic acid, sucrose, glucose, fructose, maltose, Glc6P, Glc1P). By comparing *in vitro* catalytic activities and starch degradation rates we were able to gain more insight about the relative contributions of both pathways and postulated a bottleneck limiting hydrolytic starch conversion.

### MATERIALS AND METHODS

#### Plant Material and Sampling

*Phalaenopsis* "Edessa" is an obligate starch-storing CAM plant and belongs to the family of the Orchidaceae. In starch-storing CAM plants starch is the main carbohydrate to accommodate nocturnal carboxylation whilst soluble sugar storing species employ sucrose, glucose, and/or fructose. Vegetative plants of 16 weeks old were cultivated in a growth room with a constant temperature of 28°C, a relative humidity of 75% and a 12h photoperiod (zeitgeber time ZT0-ZT12) with photosynthetic photon flux density (PPFD) of 100 µmol m-2 s-1. Watering was performed twice a week; once with a nutrient solution Peters 20N-8.7P-16.6K of 1 mS cm-1 and once with water. After six weeks, leaf samples (n = 5) were taken from the upper onethird of young fully expanded source leaves during a cycle of 24 h starting from 08.00 h (ZT0) every 2 h until 08.00 h the next morning (ZT24). The samples from 08.00 h (ZT0 and ZT24) were taken when the lights were turned on whilst the samples taken at 20.00 h (ZT12) were taken in the dark under green safety light.

At ZT12 samples were also taken from leaf pair six of the obligate model CAM plant *Kalanchoë fedtschenkoi* in order to provide comparison with the *Phalaenopsis* enzyme activity determinations. These plants were also grown in a growth room with a temperature of 25 °C and 19 °C during the light and dark periods respectively, with a relative humidity of 75% and a 12-h photoperiod (zeitgeber time ZT0-ZT12) with photosynthetic photon flux density (PPFD) of ~250 µmol m-2 s-1. Plants were grown under these conditions for at least 8-10 weeks before sampling at which time all plants had around 10 to 12 leaf pairs.

All samples were immediately frozen in liquid nitrogen, powdered and stored at -80°C.

### Gas Exchange Measurements

Net CO2 exchange was measured on the youngest fully expanded leaves, using a LCi Portable Photosynthesis System (ADC BioScientific Ltd., UK; https://www.adc.co.uk/). The top part of the leaf was enclosed in a broad leaf chamber (6.25 cm2) and the incoming air was passed through a 20-l bottle to buffer short-term fluctuations in the CO2 concentration. After six weeks, gas exchange data were collected over a 24-h period with measurements obtained at 15-min intervals (n = 3).

## Chemical Analyses of Metabolites

Soluble sugars (glucose, fructose, sucrose, and maltose) were extracted using hot water (80°C), as described by Tarkowski et al. (2019), and quantified by high performance anion exchange chromatography with pulsed amperometric detection, as described by Verspreet et al. (2013).

Extraction for measurements of starch, malic acid, and phosphorylated sugars (Glc6P and Glc1P) was performed as described by Chen et al. (2002), but with modifications. Approximately 180 mg of powdered tissue was mixed with 450 µl of ice-cold 4% (v/v) HClO4. The mixture was allowed to thaw slowly on ice for 30 min. The resulting suspension was then centrifuged at 4°C for 10 min at 16,200 g. The insoluble residue from the perchloric acid extraction was used to determine starch content spectrophotometrically at 340 nm as glucose equivalents (Genesys 10S UV-VIS, Thermo Scientific, USA; https://www.thermofisher. com), following digestion with a mix of amyloglucosidase (EC 3.2.1.3) and α-amylase (EC 3.2.1.1). The analyses were conducted as earlier described by Ceusters et al. (2008). The supernatant from the HClO4 extraction was neutralized at 4°C with 5 M K2CO3, and the resulting potassium perchlorate precipitate was removed by 10 min centrifugation at 16,200 g and 4°C. Five mg activated charcoal was added to the supernatant, and after 15 min at 4°C, removed by new centrifugation. The supernatant was used for measurement of malic acid, Glc6P, and Glc1P. Malic acid was measured in a 500-µl reaction mixture (EnzytecTM code n° E1215) containing: glycylglycinebuffer, NAD, glutamate oxaloacetate transaminase (GOT, EC 2.6.1.1). Analysis was performed spetrophotometrically by determining the change in absorbance at 340 nm after adding L-malate dehydrogenase (L-MDH, EC 1.1.1.37). The phosphorylated sugars Glc6P and Glc1P were measured in a 500-µl reaction mixture containing: 100 mM HEPES-KOH, pH 7.6, 4 mM MgCl2, 0.2 mM NADP and 1 unit glucose-6-phosphate dehydrogenase (G6PDH, EC 1.1.1.49) for Glc6P, then 1 unit glucose phosphate isomerase (GPI, EC 5.3.1.9) and 1 unit PGM (EC 5.4.2.2) for Glc1P (Mohanty et al., 1993). Analysis was performed spectrophotometrically by determining the change in absorbance at 340 nm.

#### Enzyme Activities

All extraction steps were performed by homogenizing leaf material at 4°C in extraction buffer containing: 0.3 M N-(2-hydroxyethyl)

piperazine-N′-(2-ethane sulphonic acid) (HEPES, pH 7.4), 20 mM MgCl2, 1 mM EDTA, 1 mM EGTA, 4.6 mM DTT, 80 mM benzamidine, 1 mM phenylmethylsulfonyl fluoride (PMSF), and 0.1 mM Triton X-100.

The extraction of β-amylase was based on the method described by Zeeman et al. (1998), but with modifications. After extraction, the homogenate was centrifuged at 4°C for 10 min at 16,200 g. The supernatant was incubated at 30˚C for 30 min in a 250 μl reaction mixture containing 50 mM 3-(N-Morpholino) propanesulfonic acid (MOPS, pH 6.4), 5 mM EDTA, 5 mM DTT, and 15 mg ml–1 soluble potato starch. After incubation, the reaction was stopped by boiling for 5 min and maltose was hydrolyzed by incubation in a 135 µl reaction mixture, containing 74 mM 2-(N-morpholino) ethanesulfonic acid (MES, pH 6.8), and 2 units maltase (EC 3.2.1.20), for 2 h at 37˚C. The reaction was stopped by boiling for 5 min and the resulting glucose was determined in a 500 µl reaction mixture containing: 25 mM HEPES (pH 7.9), 1 mM MgCl2, 1.2 mg ml-1 ATP, 1.0 mg ml-1 NAD, and 2 units hexokinase (EC 2.7.1.1). Analysis was performed spectrophotometrically (Genesys 10S UV-VIS, Thermo Scientific, USA) by determining the change in absorbance at 340 nm after adding 4 units G6PDH.

The extraction of maltase (α-glucosidase) and plastidic D-enzyme (4-α-glucanotransferase: DPE1) was based on the methods described by Kruger and ap Rees (1983) and by Takaha et al. (1993) respectively, but with modifications. After extraction, the homogenate was centrifuged at 4°C for 10 min at 16,200 g. The supernatant was incubated at 30˚C for 30 min in a 250 μl reaction mixture containing 50 mM sodium acetate (pH 5.2), 90 mM maltose for maltase and 50 mM MOPS-NaOH (pH 6.8), and 60 mM maltotriose for plastidic D-enzyme. The reaction was stopped by boiling for 1 min and 100 µl of pure extract was also boiled for 1 min as control. Subsequently glucose was determined in a 500 µl reaction mixture containing: 25 mM HEPES (pH 7.9), 1 mM MgCl2, 1.2 mg ml-1 ATP, 1.0 mg ml-1 NAD and 2 units hexokinase (EC 2.7.1.1). Analysis was performed spectrophotometrically by determining the change in absorbance at 340 nm after adding 4 units G6PDH.

The extraction and assay of cytosolic D-enzyme (DPE2) was based on the method described by Chia et al. (2004), but with modifications. After extraction, the homogenate was centrifuged at 4°C for 10 min at 16,200 g. Subsequently the supernatant was desalted by passing twice through a 0.5-ml column of Sephadex G-25, equilibrated with 100 mM Tris-HCl (pH 7.5 at 4°C), 1 mM DTT, 1 mM benzamidine, and 5% (w/v) glycerol. Three series of the same samples were incubated at 30°C for 2 h. The first series for assay of glucose production contained: 31 mM MOPS (pH 7), 10% (v/v) glycerol, 2.5% (w/v) oyster glycogen, and 30 mM maltose in a 200 µl reaction mixture. The two remaining series are incubated lacking either maltose or glycogen to exclude glucose not from the reaction catalzsed by DPE2. The reaction was stopped by boiling for 2 min and analysis was performed spectrophotometrically by determining the change in absorbance at 340 nm after adding 4 units G6PDH. The activity was calculated as the difference between the amount of glucose produced in the first incubation series, and the sum of the amounts of glucose produced in the two-control series.

The extraction of starch phosphorylase (α-glucan phosphorylase) was based on the method described by Kumar and Sanwal (1982), but with modifications. After extraction, the homogenate was centrifuged at 4°C for 10 min at 16,200 g. The supernatant was incubated at 30°C for 30 min in a 200 µl reaction mixture containing 0.2 M MOPS (pH 7.1), 15 mg ml-1 soluble potato starch, and 16 mM phosphate mix Na2HPO4/KH2PO4. The reaction was stopped by boiling for 1 min and the supernatant was assayed in a 500 µl reaction mixture: 0.1 M HEPES (pH 7.6), 4 mM MgCl2, and 1.5 mg ml-1 NADP. The resulting Glc1P was determined spectrophotometrically by calculating the change in absorbance at 340 nm after addition of 3 units PGM (EC 5.4.2.2) and 3 units G6PDH.

The extraction and assay of PEPC was based on the method described by Borland and Griffiths (1997). About 200 mg leaf material was homogenized in 1 ml extraction buffer at 4°C containing: 200 mM Tris-HCl (pH 8.0), 2 mM EDTA, 1 mM dithiothreitol (DTT), 2% (w/v) polyethylene glycol (PEG) 20,000, 1 mM benzamidine, and 10 mM malic acid with 240 mM NaHCO3. The homogenate was centrifuged for 2 min at 16,200 g. The extract was then desalted by passing twice through a 0.5 ml column of Sephadex G-25, equilibrated with 100 mM Tris-HCl (pH 7.5 at 4°C), 1 mM DTT, 1 mM benzamidine, and 5% (w/v) glycerol. The Ki of PEPC for malic acid was estimated using different malic acid concentrations (0.25, 0.5, 2, 8, 16 mM) in a reaction mix (500 µl) containing: 65 mM Tris-HCl (pH 7.5), 5 mM MgCl2, 0.2 mM NADH, 10 mM NaHCO3, and 2.5 mM PEP. Production of oxaloacetate by PEPC was coupled to oxidation of NADH by the high endogenous NAD-dependent MDH activity in the extracts. The reaction was initiated by the addition of 50 µl of extract and change in absorbance at 340 nm was measured for 4 min at 25°C. Preliminary experiments confirmed a linear decrease of NADH for during at least 6 min.

#### Data Analyses

Where appropriate, data were analyzed using the statistical software package IBM SPSS Statistics V23. Before carrying out statistical tests, normality of the data was checked by means of the Kolmogorov-Smirnoff statistic (p > 0.05). Means are compared by independent sample t-test (α = 0.05). All replicates considered in our study were independent biological replicates originating from different plants.

### RESULTS

#### Nocturnal CO2 Uptake and Starch Degradation

To investigate nocturnal starch degradation and the associated enzyme activities in more detail, nocturnal starch breakdown was divided into two distinct periods based on the calculated rates of breakdown (**Figure 1A**). During the first period of the night (from ZT12 to ZT20), nocturnal degradation of starch proceeded nearly linearly (y = -20.21 x + 131.16; R2 = 0.998) with a mean rate of 67 ± 13 nmol g-1FW min-1, providing 32 ± 7 µmol Glc eq. g-1FW which is equivalent to the provision of 64

FIGURE 1 | Diel patterns of starch (µmol g-1FW, A), malic acid (µmol g-1FW, B), leaf gas exchange (µmol m-2 s-1, C), and Ki of PEPC for malic acid (mM, D) for young fully developed leaves of Phalaenopsis 'Edessa'. The black arrow in panel (A) indicates the division of the nocturnal period based on the calculated starch degradation rates. The dark period is indicated in grey. Data are means ± SD (n = 5 for A, B and D; n = 3 for C).

µmol g-1FW of PEP. Moving closer to dawn, from ZT20 to ZT24, nocturnal starch degradation was also linear but (y = -11.48 x + 40.10; R2 = 0.974) proceeded with a significantly lower rate of 38 ± 6 nmol g-1FW min-1 and provided 9 ± 2 µmol Glc eq. g-1FW which is equivalent to 18 µmol g-1FW of PEP. In each of the considered periods, starch degradation provided enough substrate to sustain nocturnal malic acid synthesis (**Figure 1B**) with an initial accumulation of 56 ± 10 µmol g-1FW followed by another 20 ± 7 µmol g-1FW at the end of the night. In accordance, nocturnal CO2 uptake (**Figure 1C**) was about 96 ± 8 mmol m-2 for the first nocturnal period and 35 ± 9 mmol m-2 for the second period. The pattern of leaf gas exchange showed increasing CO2 uptake during the first three hours (ZT12 – ZT15) of the dark period whilst a decreasing CO2 assimilation rate was noticed during the final three hours (ZT21 – ZT24) of the dark period (**Figure 1C**). Measurements of Ki PEPC for malic acid fluctuated during the day around a low mean value of 6 ± 3 mM. During the first period of the night (from ZT12 to ZT20) Ki values strongly increased to a maximum value of 16 ± 1 mM (ZT20), and slightly decreased during the second period of the night to 11 ± 1 mM (ZT24) (**Figure 1D**).

#### Diel Patterns of Sugars and Sugar Intermediates

The diel pattern of sucrose showed a rather stable diurnal phase followed by a significant (p < 0.05) nocturnal decrease to a minimum value of 3 ± 1 µmol g-1FW, before rising again in the later part of the night to predusk levels (**Figure 2A**). Glucose and fructose concentrations were relatively stable during the diel cycle and fluctuated around 2 ± 1 µmol g-1FW (**Figures 2B**, **C**). Leaf maltose contents remained rather stable during the photoperiod, varying around 22 ± 10 nmol g-1FW, and doubled during the dark period, reaching a peak value of 47 ± 13 nmol g-1FW (ZT22) (**Figure 2D**). Glc6P (**Figure 2E**) concentrations generally increased during daytime until a dramatic decrease at the day-night transition (ZT12) from 86 ± 9 nmol g-1FW to 44 ± 6 nmol g-1FW. After two hours in the dark, values restored again to 82 ± 6 nmol g-1FW (ZT14) and remained relatively stable during the remainder of the night, until a strong decrease during the last two hours of the dark period (ZT22 – ZT24). A similar diel pattern was observed for Glc1P (**Figure 2F**). During daytime concentrations generally increased, until a dramatic decrease at the day-night transition (ZT12) from 7 ± 1 nmol g-1FW to 4 ± 1 nmol g-1FW. After two hours in the dark,

FIGURE 2 | Diel patterns of sucrose (µmol g-1FW, A), glucose (µmol g-1FW, B), fructose (µmol g-1FW, C), maltose (nmol g-1FW, D), glucose 6-phosphate (nmol g-1FW, E), and glucose 1-phosphate (nmol g-1FW, F) for young fully developed leaves of Phalaenopsis 'Edessa'. The dark period is indicated in grey. Data are means ± SD (n = 5).

values restored to 5 ± 1 nmol g-1FW, and remained relatively stable during the remainder of the night, followed by a gradual decrease during the last four hours of the dark period (ZT20 – ZT24).

#### Diel Activity Patterns of Enzymes Implicated in Starch Degradation

In order to gain more insight into the nocturnal starch degradation process, total extractable activities of a range of

enzymes implicated in starch degradation were measured over the complete diel cycle (**Figure 3**). Starch phosphorylase activity significantly (p < 0.05) decreased during daytime from 95 ± 15 nmol g-1FW min-1 to 48 ± 6 nmol g-1FW min-1 (**Figure 3A**). Compared to these daytime values, nocturnal starch phosphorylase activity was higher, fluctuating around a mean of 68 ± 8 nmol g-1FW min-1 (**Figure 3A**). β-amylase showed the highest activity of all enzymes but without any clear pattern over the 24-h cycle (**Figure 3B**). Heavily fluctuating activities between

0.21 ± 0.01 µmol g-1FW min-1 and 19 ± 6 µmol g-1FW min-1 were observed during both day and night. The chloroplastic disproportionating enzyme (DPE1) showed an increasing activity during daytime from 44 ± 16 nmol g-1FW min-1 (ZT0) to 200 ± 28 nmol g-1FW min-1 (ZT10) (**Figure 3C**). Except for the observed high activity at ZT20 (317 ± 60 nmol g-1FW min-1), nocturnal DPE1 activity was quite stable around a mean value of 156 ± 12 nmol g-1FW min-1. For cytosolic disproportionating enzyme (DPE2) high activities were observed during the photoperiod at ZT2 and ZT4, i.e. 124 ± 16 nmol g-1FW min-1 and 112 ± 4 nmol g-1FW min-1 respectively. However, during the remainder of the day and night, the enzyme showed an overall low activity of only 8 ± 4 nmol g-1FW min-1 (**Figure 3D**). Maltase activity remained relatively stable during the photoperiod with a mean activity of 28 ± 8 nmol g-1FW min-1 (**Figure 3E**). During the dark period, maltase activity significantly (p < 0.05) increased from a minimum activity of 30 ± 6 nmol g-1FW min-1 (ZT12) to a maximum activity of 44 ± 5 nmol g-1FW min-1 (ZT24).

#### Nocturnal Starch Degradation and Calculated Enzymatic Capacities

The nocturnal *in vitro* activities of the enzymes involved in starch degradation (starch phosphorylase, β-amylase, DPE1, DPE2, and maltase) were considered relative to the upper and lower limits of the observed starch degradation rate i.e. 67 ± 13 nmol g-1FW min-1 from ZT12-ZT20 and 38 ± 6 nmol g-1FW min-1 from ZT20-ZT24. Registered minimum, maximum, and mean activities were all considered for each enzyme in comparison with the calculated nocturnal starch degradation rate and symbols were used to indicate whether enzymatic capacity was theoretically sufficient (√) or insufficient (X) to meet the observed starch degradation rates (**Table 1**). These considerations indicated that even the lowest recorded values of either β-amylase or DPE1 activity were significantly higher (p < 0.05) than the calculated nocturnal starch degradation rates for both distinct periods. Exceeding starch degradation rates by multiples of two for DPE1 to even 6–80 for β-amylase, these particular enzymes were not ratelimiting at all in the starch degradation process. In contrast even the maximum values of DPE2 activity were far below (p < 0.05) the calculated nocturnal starch degradation rates for both periods and could only account for about 15% of the calculated starch degradation. Whilst maltase activity was not limiting starch degradation during the second period (p > 0.05) it could only account for about 50% of the calculated starch degradation during the first period of the night (p < 0.05). Considering the summed activity of DPE2 and maltase, it could still only account for about 65% of the calculated starch degradation during the first period of the night (p < 0.05). For both considered periods the calculated starch degradation rates perfectly matched the observed min-max range for starch phosphorylase activities. Only during the first nocturnal period minimum observed activities of starch phosphorylase might potentially limit starch degradation accounting for about 75% of starch breakdown (p < 0.05).

#### Starch Degradative Enzyme Activities in Phalaenopsis Match With Those in Kalanchoë But Are Clearly Different From Published Arabidopsis Values

To strengthen our study, we further compared our mean values of enzyme activities from *Phalaenopsis* with either a negative control C3 photosynthesis species such as Arabidopsis (based on published data) and a positive control CAM species such as *K. fedtschenkoi* (**Table 2**; to allow direct comparison the *Phalaenopsis* values were recalculated to mass basis). The data clearly indicated that starch phosphorylase activity in the model CAM plant *K. fedtschenkoi* could also easily accommodate the observed starch degradation whilst DPE2 + maltase fell short. These observations

TABLE 1 | Comparison between nocturnal starch degradation rate (nmol g-1FW min-1) and calculated nocturnal activity (minimum, maximum and mean) of starch degradative enzymes (nmol g-1FW min-1) in young fully developed leaves of Phalaenopsis 'Edessa', for both nocturnal periods.


To investigate nocturnal starch degradation and the associated enzyme activities in more detail, the nocturnal part was divided into two distinct periods i.e. Period 1 from ZT12 to ZT20 and Period 2 from ZT20 to ZT24. Symbols are used to indicate if the enzymatic capacity is sufficient (√) of insufficient (X) to accommodate the observed starch degradation.

Data are means ± SD (n = 5) and asteriks indicate significant differences between the calculated enzymatic activity and the nocturnal starch degradation rate of the same period at \*P < 0.05 according to the independent sample t-test.


TABLE 2 | Comparison of nocturnal starch degradation rate (nmol g-1FW min-1) and mean activities of starch degradative enzymes (nmol g-1FW min-1) in leaves of Phalaenopsis 'Edessa', Kalanchoë fedtschenkoi and Arabidopsis thaliana.

Data are means ± SD (n = 5). Values for Arabidopsis originate from Chia et al. (2004). Different colors have been used to indicate whether the relevant enzymatic capacities in either starch phosphorolysis (starch phosphorylase) or hydrolysis (DPE2 + maltase) are adequate (green) or inadequate (red) to accommodate the observed starch degradation.

strongly contradicted with Arabidopsis showing a completely different situation with high DPE2 activity, supporting starch hydrolysis, and insufficient starch phosphorylase activity. Moreover, high DPE2 and low starch phosphorylase activities have also been reported by Borland et al. (2016) for Arabidopsis whilst the reverse was true for *M. crystallinum* (in the CAM mode) under identical growth conditions.

#### DISCUSSION

#### Regulation of Nocturnal Starch Degradation to Avoid Premature Exhaustion of Carbohydrate

Carbohydrate availability represents a key limiting factor for CAM productivity, which makes the diel breakdown and resynthesis of a transient pool of carbohydrate a central requirement for CAM homeostasis (Borland and Dodd, 2002; Cushman et al., 2008). To account for the observed gradual decrease in starch degradation rate towards the end of the night in the leaves of *Phalaenopsis* "Edessa" we considered two starch degradation rates in our study i.e. (1) 67 ± 13 nmol g-1FW min-1 from ZT12 to ZT20 and (2) 38 ± 6 nmol g-1FW min-1 from ZT20 to ZT24 (**Figure 1**). Avoiding premature exhaustion of reserve carbohydrate availability before the onset of dawn is consistent with observations in the C3 plant Arabidopsis where the rate of degradation of transitory starch during the night was enhanced or declined following either shortened or lengthened photoperiods (Lu et al., 2005; Graf et al., 2010). The exact physiology behind the mechanisms that allow the plant to match starch degradation to its needs and the length of night are still exciting topics of current investigation (Fernandez et al., 2017; Flis et al., 2019). In *Phalaenopsis* "Edessa" nocturnal starch breakdown during the 4 final hours was reduced with ca. 33% and accompanied by a threefold reduction in nocturnal CO2 fixation and malic acid accumulation (**Figure 1**). It is very reasonable that a reduced availability of carbon skeletons for PEP synthesis at the end of the night will lead to both the reduced rate and total amount of CO2 fixed by PEPC in CAM plants. Carbohydrate limited gas exchange towards the end of the dark period is also consistent with the fact that the Ki of PEPC for malic acid showed an upward trend during the major part of the night but started to decrease during the 4 final hours (**Figure 1**). High values of Ki are indicative of phosphorylated PEPC through PEPC-kinase action, rendering PEPC less sensitive for malic acid inhibition (Nimmo et al., 1986; Borland and Griffiths, 1997; Borland et al., 1999). Expression of the kinase itself is not likely to bring about less phosphorylation towards dawn as similar declines in Phase-I CO2 assimilation well before the onset of dawn have been noticed in other CAM plants such as *M. crystallinum*, *K. daigremontiana* and *Clusia rosea*, whilst transcript abundance of PEPC kinase remained unaffected during the second nocturnal period (Borland et al., 1998; Borland et al., 1999; Dodd et al., 2003).

#### The Phosphorolytic Pathway Can Accommodate the Observed Nocturnal Starch Degradation

Both the phosphorolytic and hydrolytic pathways of starch degradation are assumed to be deployed by CAM species to balance the sink demands of CAM with growth and maintenance (Borland et al., 2016). This strongly contrasts with C3 plants where starch is degraded at night *via* the hydrolytic route (Niittylä et al., 2004; Weise et al., 2004; Stitt and Zeeman, 2012), whilst phosphorolytic starch degradation mainly provides substrate for internal chloroplast metabolism (Zeeman et al., 2004; Weise et al., 2006). To increase our insights in the possible relative contributions of either pathway in the process of CAM starch degradation we compared minimum and maximum enzyme activities of a range of enzymes during the night with different rates of observed nocturnal starch degradation in leaves of the CAM orchid *Phalaenopsis* "Edessa" (**Table 1**). Starch degradation is a complex process that requires different enzymes working in sequence and/or concert and attacking different substrates. Chloroplastic starch phosphorylase could potentially account for nearly all starch degradation during the whole night with mean activities matching exactly the calculated starch breakdown and as such avoiding premature exhaustion of starch. In accordance with a decreased starch degradation during the latter part of the nocturnal period, starch phosphorylase activity also slowed down towards dawn. High levels of Glc6P were present during the major part of the night, almost three times higher compared to the observed levels of maltose (i.e. 89 ± 7 nmol g-1FW and 31 ± 4 nmol g-1FW respectively). This is consistent with an important role for starch phosphorylase in starch degradation, since Glc6P is the major export product from phosphorolytic starch degradation. Glc6P is not only considered an allosteric activator of PEPC but can also easily be converted to PEP in the cytosol providing ATP by substrate-level phosphorylation and offering energetic advantages (Holtum et al., 2005; Cheung et al., 2014). According to recently developed diel flux balance models for CAM this would make a marked difference to help offset the energetic costs of running the CAM cycle (Shameer et al., 2018).

#### Hydrolytic Starch Degradation Is Limited by DPE2 and Maltase Activity

The other starch degrading enzymes residing in the chloroplast and belonging to the hydrolytic starch conversion pathway, i.e. β-amylase and DPE1, showed *in vitro* activities far exceeding those required for *in vivo* starch degradation by multiples of 2 to 80 (**Table 1**). Those high activities could potentially lead to complete exhaustion well before the end of the dark period without any further regulation. However, further processing of maltose, generated by degradation of starch *via* β-amylase, in the cytosol is dependent on either cytosolic disproportionating enzyme (DPE2) or maltase. DPE2 showed only very low nocturnal activities in *Phalaenopsis* and could theoretically meet only about 15% of the observed starch breakdown. Low DPE2 activities have also been reported in *M. crystallinum* and have previously been hypothesized to potentially represent a possible bottleneck in the hydrolytic starch degradation process (Borland et al., 2016). This is in marked contrast with the critical importance of DPE2 in C3 plants such as Arabidopsis where DPE2 activity, which is about 20 times higher than in *Phalaenopsis*, can easily account for the measured starch degradation (**Table 2**). Whilst maltase activities in Arabidopsis were considered of minimal importance and about seven-fold lower compared to DPE2 activity (**Table 2**), our measurements showed a four-fold higher mean maltase activity than DPE2 during the night in *Phalaenopsis* "Edessa," which was still insufficient to accommodate nocturnal starch degradation. However, it can be assumed that both enzymes, DPE2 and maltase, can act in concert to accommodate maltose processing in the cytosol. Considering the sum of both enzymatic activities, the calculated minimum, maximum, and mean activity were still significantly lower (p < 0.05) than required to account for the observed nocturnal starch degradation (**Table 1**). In the model CAM plant *K. fedtschenkoi* DPE2 activity was higher but still insufficient to account for the measured starch breakdown in concert with maltase, whilst β-amylase was not rate limiting (**Table 2**). The bottleneck effect of the maltose processing enzymes in regulating β-amylase mediated starch breakdown was also

#### REFERENCES


depicted by a significant nocturnal accumulation (p < 0.05) of its substrate, maltose, from 27 ± 5 nmol g-1FW (ZT12) to 47 ± 13 nmol g-1FW (ZT22) in *Phalaenopsis* leaves.

In conclusion we compared *in vivo* nocturnal starch degradation and *in vitro* enzyme activity measurements during the whole night in leaves of the CAM orchid *Phalaenopsis* "Edessa" to shed more light into the interplay of the phosphorolytic and hydrolytic processes of starch degradation in CAM. We showed that the phosphorolytic pathway is the major route to accommodate nocturnal starch degradation and that measured activities of starch phosphorylase perfectly matched calculated starch degradation rates in order to avoid premature exhaustion of starch reserves before dawn. The hydrolytic pathway seemed hampered in starch processing not by β-amylase but through insufficient catalytic capacity of both DPE2 and maltase. These considerations were further corroborated by measurements of enzyme activities in the CAM model plant *Kalanchoë fedtschenkoi* and strongly contradict with the situation in the C3 plant Arabidopsis. The data support the view that the phosphorolytic pathway might be the main route of starch degradation in CAM to provide substrate for PEP with additional hydrolytic starch breakdown to accommodate mainly sucrose synthesis.

### DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/ supplementary material.

### AUTHOR CONTRIBUTIONS

NC, WV and JC proposed the conceptual framework for the study and performed the data collection and analysis. NC and MF performed the experimental analyses. NC and JC interpreted the data and wrote the manuscript.

### FUNDING

This research was supported by the Research Fund KU Leuven.

## ACKNOWLEDGMENTS

Microflor NV is acknowledged for supplying plant material. The authors also acknowledge Kim Vekemans (Geel) and Timmy Reijnders (Leuven) for assistance in the lab.


biochemical characterization and intercellular localization of carboxylation processes in three species which show different degrees of CAM. *Planta* 205, 342–351. doi: 10.1007/s004250050329


and starch degradation in the facultative CAM plant, *Mesembryanthemum crystallinum*. *J. Plant Physiol.* 219, 144–154. doi: 10.1016/j.jplph.2017.07.021


**Conflict of Interest:** 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 Ceusters, Frans, Van den Ende and Ceusters. 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.*

# Dissecting the Function of MADS-Box Transcription Factors in Orchid Reproductive Development

*Zhi Wei Norman Teo1,2, Wei Zhou1 and Lisha Shen1\**

1 Temasek Life Sciences Laboratory, National University of Singapore, Singapore, Singapore, 2 Department of Biological Sciences, Faculty of Science, National University of Singapore, Singapore, Singapore

The orchid family (Orchidaceae) represents the second largest angiosperm family, having over 900 genera and 27,000 species in almost all over the world. Orchids have evolved a myriad of intriguing ways in order to survive extreme weather conditions, acquire nutrients, and attract pollinators for reproduction. The family of MADS-box transcriptional factors have been shown to be involved in the control of many developmental processes and responses to environmental stresses in eukaryotes. Several findings in different orchid species have elucidated that MADS-box genes play critical roles in the orchid growth and development. An in-depth understanding of their ecological adaptation will help to generate more interest among breeders and produce novel varieties for the floriculture industry. In this review, we summarize recent findings of MADS-box transcription factors in regulating various growth and developmental processes in orchids, in particular, the floral transition and floral patterning. We further discuss the prospects for the future directions in light of new genome resources and gene editing technologies that could be applied in orchid research and breeding.

#### Edited by:

Jen-Tsung Chen, National University of Kaohsiung, Taiwan

#### Reviewed by:

Adriana Garay, National Autonomous University of Mexico, Mexico Philip Ruelens, Wageningen University & Research, Netherlands

> \*Correspondence: Lisha Shen lisha@tll.org.sg

#### Specialty section:

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

Received: 02 August 2019 Accepted: 23 October 2019 Published: 15 November 2019

#### Citation:

Teo ZWN, Zhou W and Shen L (2019) Dissecting the Function of MADS-Box Transcription Factors in Orchid Reproductive Development. Front. Plant Sci. 10:1474. doi: 10.3389/fpls.2019.01474

Keywords: orchid, MADS-box transcription factors, floral transition, floral patterning, development

## INTRODUCTION

The orchid family (Orchidaceae) is currently the second largest angiosperm family, having over 900 genera and 27,000 species in almost all parts of the world except Antarctica. New genera in the orchid family are being discovered at a rate of around 13 per year for over the past decade (Schuiteman, 2004; Chase et al., 2015). Nevertheless, many wild species are at the brink of extinction because of illegal trading activities (Wijnstekers, 2001; Hossain et al., 2013). Orchids have their own ecological niches through their relationships with mycorrhizal fungi, specialized pollinators and host trees (Fay and Chase, 2009). Of all the orchid species, 70% are epiphytic (growing on trees), 25% are terrestrial (growing on ground), and the remaining 5% are found on various supports such as rocks (Atwood, 1986). Thus, it is important to conserve orchid species through generating awareness and increasing our understanding on the species physiology and diversity (Cribb et al., 2003).

Being a class of valuable ornamental plants with distinct and attractive flowers, orchid is viewed as a high value commodity in the global flower cultivation and landscaping industries. They are also highly sought for as food and traditional medicine (Arditti, 1992; Bulpitt et al., 2007). Besides their great economic values, orchids are also exclusive genetic resources for studying plant developmental processes, including floral transition, floral development, flower pigmentation, and senescence, because of the specialized reproductive structures and the unique strategies for

1 **46** reproduction (Yu and Goh, 2001; Gutiérrez, 2010; Da Silva et al., 2014). However, the orchid research as well as orchid breeding have been challenging due to the long vegetative developmental period before switching to flower development, and technical limitations in transformation and obtaining transgenic lines in various orchid species. Currently, to investigate the function of orchid genes, several methods are being used including the heterologous expression of gene of interest under the strong constitutive CaMV *35S* promoter in *Arabidopsis* (*Arabidopsis thaliana*) or tobacco and virus-induced gene silencing (VIGS). Sometimes, transgenic orchids are also generated to study gene function. So far, the genetic transformation of orchids using the *Agrobacterium*-mediated approach on protocorm-like bodies or rhizomes has been reported in *Cymbidium*, *Oncidium*, *Dendrobium*, *Phalaenopsis*, and other orchids (Belarmino and Mii, 2000; Yu and Goh, 2000; Chai et al., 2002; Chen, 2002; Sjahril and Mii, 2006; Shrestha et al., 2007; Zhang et al., 2010; Ding et al., 2013).

The orchids including *Phalaenopsis*, *Dendrobium*, *Cymbidium*, and *Oncidium* from the Epidendroideae subfamily are used as orchid plant models for research and biotechnology. Most of these orchids are predominantly found to be growing in tropical Asia to Australia with the exception of *Oncidium* in the West (**Table 1**). Orchids are either monopodial or sympodial in their growing habits. Monopodial orchids, such as *Phalaenopsis*, grow as a single erect "stem" with alternating leaves on opposing parts of the center. They store water in their thick leaves and roots but

#### TABLE 1 | Orchid model plants and their growing characteristics.


have no pseudobulbs. Sympodial orchids, such as *Dendrobium*, *Cymbidium*, and *Oncidium*, grow from a horizontal stem called rhizome and have pseudobulbs to store water and grow new leaves. After blooming, the plant will resume growth at axillary buds at the base of the previous pseudobulbs.

Recent findings from different orchid species have elucidated that a group of MADS-box transcription factors sharing a greatly conserved N-terminal DNA binding domain (MADSbox) exert important functions in controlling orchid growth and development, in particular, the floral transition and floral patterning. In this review, we discuss the biological roles of these MADS-box proteins and the mechanisms how they contribute to flowering and floral organ formation in orchids. We further elaborate about the prospects for research and development in light of new genome resources and gene editing technologies that could be applied in orchid research and breeding.

#### THE MADS-BOX PROTEIN FAMILY

In orchids and other angiosperms, there is a family of MADSbox transcription factors that have been identified to control many plant developmental processes, including floral transition, floral patterning, as well as male and female gametophyte development (Coen and Meyerowitz, 1991; Weigel and Meyerowitz, 1994; Yu and Goh, 2000; Acri-Nunes-Miranda and Mondragón-Palomino, 2014; Valoroso et al., 2019). The MADS-box family proteins are conserved in nearly all eukaryotes. The MADS-box acronym is derived from the yeast *MINICHROMOSOME MAINTENANCE 1* (*MCM1*) (Passmore et al., 1988), the *Arabidopsis AGAMOUS* (*AG*) (Yanofsky et al., 1990), the *Antirrhinum DEFICIENS* (*DEFA*) (Schwarz-Sommer et al., 1990), and the mammalian *SERUM RESPONSE FACTOR* (*SRF*) (Norman et al., 1988; Gramzow et al., 2010). All identified MADS-box proteins each contain a MADS-box domain of ~58 amino acid at the N-terminus that binds to a consensus CC[A/T]6GG sequence, termed as the "CarG-box" motif (Hayes et al., 1988; Riechmann et al., 1996). Interestingly, flowering plants (angiosperms) have more of these genes (e.g. 107 in *Arabidopsis*; 51 in *Phalaenopsis equestris*) compared to yeast (e.g. 4 in *Saccharomyces cerevisiae*) and mammals (e.g. 5 in *Homo sapiens*) (Becker and Theißen, 2003; Messenguy and Dubois, 2003; Par˘enicová et al., 2003; Cai et al., 2015). The MADS-box family of genes form two major lineages, namely type I of *SRF*like genes and type II of *MEF2-*like genes, which is resulted from an ancient event of gene duplication prior to the divergence of the kingdoms of plants and animals (Alvarez-Buylla et al., 2000). In plants, type II genes of MADS-box, also called as MIKC-type genes, feature four distinct protein domains arranged from the N-terminal end to C-terminal end. They are the highly conserved DNA binding MADS-box (M) domain, the less-conserved intervening domain (I) for conferring interaction specificity between different MADS-box transcription factors and/or other proteins, the keratin-like coiled-coil (K) domain for conferring protein–protein interactions, and a highly variable C-terminal (C) domain for regulating gene transcription or multimeric protein complexes formations (Shore and Sharrocks, 1995; Theißen et al., 1996; Riechmann and Meyerowitz, 1997; Honma and Goto, 2001; Becker and Theißen, 2003; Hill et al., 2008).

The MIKC-type genes are specifically present in plants and many of these genes have been shown to control key processes of plant development including vegetative growth and reproductive organ development with complex cascades of events and networks (Theißen, 2001; Kaufmann et al., 2005; Adamczyk and Fernandez, 2009). Particularly, their functions in determining plant reproductive development are more remarkable as they regulate the development of consecutive reproductive processes, namely the floral transition, floral meristem specification, floral patterning, pollen growth, and development of ovules and seeds (**Figure 1**). The MADS-box genes that act in the regulation of flowering time and floral patterning in plants will be elaborated more in the latter sections (**Table 2**). The floral homeotic genes that play crucial roles in specifying reproductive floral organ identities are among the best characterized MADS-box genes. The extensive study of mutants with floral homeotic defects has resulted in the birth of the "ABCE model," which explains how the genes of A, B, C, D and E classes act jointly to determine floral organs identities (Coen and Meyerowitz, 1991; Weigel and Meyerowitz, 1994; Theissen and Saedler, 2001). All floral homeotic genes in *Arabidopsis* belong to the MADS-box family except for the A-class gene, *APETALA2* (*AP2*). In addition to determining the identities of floral organs, MADS-box proteins also regulate floral meristem specification, the process of which involves four meristem identity genes, namely *LEAFY* (*LFY*) and three MADS-box genes closely related to each other, *APETALA1* (*AP1*), *CAULIFLOWER* (*CAL*), and *FRUITFULL* (*FUL*) (Ferrándiz et al., 2000; Ng and Yanofsky, 2001). Furthermore, studies of MADS-box proteins have also demonstrated their function in seed and silique growth. For example, three MADSbox genes, *SHATTERPROOF 1* (*SHP1*), and its close homologs *SHP2* and *SEEDSTICK* (*STK*) contribute to normal growth and development of carpels and fruits (Liljegren et al., 2000; Pinyopich et al., 2003). In addition, several MADS-box genes belonging to type I, such as *PHERES 1* (*PHE1*), *AGAMOUS-LIKE 80* (*AGL80*), *DIANA* (*AGL61*), etc., are involved in embryo and seed growth (Kohler et al., 2003; Portereiko et al., 2006; Bemer et al., 2008; Colombo et al., 2008; Kang et al., 2008).

### MADS-BOX PROTEINS IN ORCHID FLOWERING

### The Floral Transition of Orchid

The floral transition, a developmental transition from vegetative to reproductive phase, is one of the key developmental transitions in the plant life cycle. The timing of floral transition greatly affects the success of plant reproduction. In the model plant *Arabidopsis*, the vegetative shoot apex, from which leaves are generated, is converted into the inflorescence meristem, from which flowers are generated, during the floral transition. In orchids, the length of the vegetative phase can vary from one to thirteen years between different species, but the average time for most species is between two to three years (Hew and Yong, 2004). The process of flowering is induced in the meristem of dormant axillary buds. In sympodial orchids such as *Dendrobium* and *Oncidium*, the formation of bud primordia occurs at the axils of the leaves. In the orchid *Cymbidium*, the inflorescence will be developed from the dormant axillary buds at the base of the pseudobulb. For monopodial orchids such as *Phalaenopsis* and *Vandas*, they typically have at least two dormant bud primordia at each leaf axil which can grow into inflorescences or keikis (new orchid plantlets).

With the increasing demand for whole orchids and cut flowers, modern horticulturists and breeders are learning about the flowering behavior of different species to control the time of blooming so as to maximize their economic value. Each orchid species has a time of the year when it will bloom naturally. Most orchids will grow in the abundance of sunlight and moisture during summer and bloom in the fall, winter, or spring. Significant progress has been made to determine the effects of environmental conditions, such as temperature and day length, in inducing flowering in different species (**Table 3**). Studies on *Cymbidium*, *Oncidium*, *Dendrobium*, and *Phalaenopsis* have shown that a low night temperature of 13°C and a large fluctuation of 10°C to 14°C in daily diurnal temperature are sufficient to induce flowering (Rotor, 1952; Rotor and Withner, 1959). A high day temperature of more than 28°C for 8 h or longer promotes vegetative growth and inhibit the process of flowering in *Phalaenopsis* (Blanchard and Runkle, 2006; Newton and Runkle, 2009). However, a prolonged exposure to a constant elevated temperature of 30°C induces flowering by activating the thermal stress response (Chin et al., 2014). Since orchids are likely to be shaded by leaves on trees, the length of daylight is not known to influence flowering with the exception of *Dendrobium phalaenopsis* flowering under short days (Rotor, 1952; Rotor and Withner, 1959; Lopez and Runkle, 2004). It is noteworthy to mention that flowering is an intricate process and many environmental conditions including light intensity and humidity can affect the initiation of floral spikes.

### Floral Pathway Integrators

For the model plant *Arabidopsis*, the timing of floral transition is regulated by a complex system consisting of several flowering pathways—photoperiod, vernalization, thermosensory, gibberellins (GA), autonomous, and age—that perceive both environmental and endogenous flowering signals (Mouradov et al., 2002; Simpson and Dean, 2002; Blázquez et al., 2003; Boss et al., 2004; Wang et al., 2009). For environmental flowering signals, the photoperiod pathway perceives the daylength in seasonal changes; the vernalization pathway measures the period of the plant's exposure to cold; the thermosensory pathway mediates the ambient temperature effect. For endogenous flowering signals, the GA pathway promotes flowering under non-inductive photoperiod, while the autonomous pathway is a photoperiod-independent pathway that induces flowering through perceiving the internal signals at various stages of development. These genetic pathways regulate the transcription of two main integrators of floral pathways, *FLOWERING* 

FIGURE 1 | Function of MADS-box proteins in the whole plant life cycle. (A) MADS-box genes regulate Arabidopsis development throughout its life cycle. Many MADS-box genes mediate the transition to flowering. The flowering time repressor genes, including FLC (Michaels and Amasino, 1999), FLM (Ratcliffe et al., 2001), MAF2-5 (Ratcliffe et al., 2003; Gu et al., 2013), SVP (Hartmann et al., 2000; Li et al., 2008), and AGL15/18 (Adamczyk et al., 2007), are shown in red color, whereas the flowering time promoter genes, including SOC1 (Lee et al., 2000), AGL24 (Yu et al., 2002), AGL6 (Yoo et al., 2011), XAL1/AGL12 (Tapia-Lopez et al., 2008), AGL17 (Han et al., 2008), AGL19 (Schonrock et al., 2006), AGL28 (Yoo et al., 2006), and AGL42/71/72 (Dorca-Fornell et al., 2011), are shown in green color. All the identified floral organ identity genes except AP2 encode MADS-box transcription factors. MADS-box genes are also involved in root growth (e.g. XAL1, XAL2, AGL21, ANR1, SHP1,2, and STK) (Zhang and Forde, 1998; Tapia-Lopez et al., 2008; Moreno-Risueno et al., 2010; Garay-Arroyo et al., 2013; Yu et al., 2014), vegetative growth (e.g. AGL16's function in stomata development) (Kutter et al., 2007), pollen maturation and tube growth (AGL65/66/104) (Adamczyk and Fernandez, 2009), ovule development (e.g. AGL13, AGL23, AGL62, SHP1,2 and STK) (Liljegren et al., 2000; Pinyopich et al., 2003; Colombo et al., 2008; Kang et al., 2008; Hsu et al., 2014), and embryo and seed development (e.g. DIA, AGL80, AGL23, and PHE1) (Kohler et al., 2003; Portereiko et al., 2006; Bemer et al., 2008; Colombo et al., 2008). (B) Functions of MADS-box genes in orchid development. Orchid MADS-box proteins have been shown to regulate flowering and floral organ formation. AG, AGAMOUS; AGL6, AGAMOUS-LIKE 6; AGL15, AGAMOUS-LIKE 15; AGL16, AGAMOUS-LIKE 16; AGL17, AGAMOUS-LIKE 17; AGL18, AGAMOUS-LIKE 18; AGL19, AGAMOUS-LIKE 19; AGL21, AGAMOUS-LIKE 21; AGL23, AGAMOUS-LIKE 23; AGL24, AGAMOUS-LIKE 24; AGL28, AGAMOUS-LIKE 28; AGL42, AGAMOUS-LIKE 42; AGL65, AGAMOUS-LIKE 65; AGL66, AGAMOUS-LIKE 66; AGL71, AGAMOUS-LIKE 71; AGL72, AGAMOUS-LIKE 72; AGL80, AGAMOUS-LIKE 80; AGL104, AGAMOUS-LIKE 104; ANR1, ARABIDOPSIS NITRATE REGULATED 1; AP1, APETALA1; AP3, APETALA3; CAL, CAULIFLOWER; CO, CONSTANS; DIA, DIANA; FLC, FLOWERING LOCUS C; FLM, FLOWERING LOCUS M; FT, FLOWERING LOCUS T; FUL, FRUITFULL; MAF2- 5, MADS AFFECTING FLOWERING 2-5; PHE1, PHERES1; PI, PISTILLATA; SEP1-4, SEPALLATA1-4; SHP1,2, SHATTERPROOF1,2; SOC1, SUPPRESSOR OF OVEREXPRESSION OF CONSTANS 1; STK, SEEDSTICK; SVP, SHORT VEGETATIVE PHASE; XAL1, XAANTAL 1; XAL2, XAANTAL 2.

TABLE 2 | A summary of MADS-box regulators involved in reproductive development in the model plant Arabidopsis and orchids.


#### TABLE 2 | Continued


aFM, floral meristem.

bMADS-box genes whose function has been examined by stable or transient overexpression or silencing in orchids.

*LOCUS T* (*FT*) and the MADS-box gene *SUPPRESSOR OF OVEREXPRESSION OF CONSTANS 1* (*SOC1*), which then activate the expression of *AP1* and *LFY*, two floral meristem identity genes, to start the process of floral meristem formation (**Figure 2**) (Kardailsky et al., 1999; Kobayashi et al., 1999; Blázquez and Weigel, 2000; Lee et al., 2000; Samach et al., 2000; Liu et al., 2009). Several MADS-box proteins including SOC1, AGAMOUS-LIKE 24 (AGL24), AGL6, and AGL17, promote flowering (Lee et al., 2000; Yu et al., 2002; Han et al., 2008; Yoo et al., 2011), whereas MADS-box regulators including FLOWERING LOCUS C (FLC), SHORT VEGETATIVE PHASE (SVP), MADS AFFECTING FLOWERING 1/FLOWERING



LOCUS M (MAF1/FLM), and MAF2/3/4/5 form various complexes to repress flowering (**Figure 2**) (Helliwell et al., 2006; Searle et al., 2006; Li et al., 2008; Gu et al., 2013; Lee et al., 2013; Pose et al., 2013; Mateos et al., 2015). Another three MADSbox transcription factors that are closely related to each other, AP1, CAL, and FUL are also involved in the activation of *LFY* in promoting flowering and floral meristem specification as the triple mutant of these genes generates leafy shoots in place of floral organs (Ferrándiz et al., 2000).

In orchids, the process of the floral transition occurs in the axillary buds where the bud primordia will develop into a more convex shape upon entering reproductive phase. Recent works have identified and examined many orthologs of flowering integrators and other MADS-box genes in orchids (**Figure 2**). As

one of the major floral pathway integrator genes, *FT*, encoding a small globular protein, is transcriptionally activated by CONSTANS (CO) in companion cells in the leaf veins and the FT protein moves to the shoot apical meristem (Kardailsky et al., 1999; Kobayashi et al., 1999; An et al., 2004; Corbesier et al., 2007; Liu et al., 2012; Nakamura et al., 2014; Zhu et al., 2016). In orchids for instance, *Oncidium*, *Dendrobium*, and *Cymbidium*, the expression of *FT* orthologs was predominantly expressed in the leaves and axillary buds. In addition, the expression of *FT* orthologs has been found to be influenced by daylength in *Oncidium* and *Cymbidium*, showing a similar photoperiodic pattern like *FT* in *Arabidopsis* (Hou and Yang 2009; Huang et al., 2012). Ectopic expression of *FT* orthologs, *OnFT*, *DnFT*, *DOFT*, *CeFT*, *CgFT*, *CsFT*, and *PaFT1* from the orchids *Oncidium*,

FIGURE 2 | Biological roles of MADS-box genes in controlling flowering in the model plant Arabidopsis and orchid. In Arabidopsis, the MADS-box genes including SOC1, FLC, SVP and AGL24 integrates signals for flowering from environmental and endogenous cues. In orchid, orthologous genes of SOC1, AGL6, SVP, and AP1 have been isolated and functionally characterized either in heterologous system (e.g. Arabidopsis) or orchid and shown to be involved in promoting flowering. MADS-box transcription factors that function as flowering activators and suppressors are shown in green and red, respectively, whereas other flowering regulators are shown in black. Promoting and repressive effects are indicated by black arrows and orange T bars, respectively. The dashed lines with arrows indicate possible positive regulation based on the studies using heterologous systems. Double-ended diamond arrows indicate protein–protein interactions. AGL6, AGAMOUS-LIKE 6; AGL17, AGAMOUS-LIKE 17; AGL19, AGAMOUS-LIKE 19; AGL24, AGAMOUS-LIKE 24; AP1, APETALA1; CO, CONSTANS; FLC, FLOWERING LOCUS C; FLM, FLOWERING LOCUS M; FT, FLOWERING LOCUS T; FTIP1, FT-INTERACTING PROTEIN 1; FUL, FRUITFULL; LFY, LEAFY; MAF2, MADS AFFECTING FLOWERING 2; SOC1, SUPPRESSOR OF OVEREXPRESSION OF CONSTANS 1; SVP, SHORT VEGETATIVE PHASE.

*Dendrobium*, *Cymbidium*, and *Phalaenopsis*, respectively, results in a precocious flowering phenotype in transgenic plants of *Arabidopsis* or tobacco (Hou and Yang, 2009; Huang et al., 2012; Li et al., 2012; Xiang et al., 2012; Jang, 2015; Wang et al., 2017). More importantly, downregulation of *DOFT* delays flowering in *Dendrobium* orchids, whereas overexpression of *DOFT* accelerates flowering in orchids (Wang et al., 2017). Interestingly, low temperature treatment specifically induces the expression of *FT* in leaves in both *Dendrobium* and *Phalaenopsis*, suggesting *FT* is the main floral inducer under floral inductive low temperature regime (Li et al., 2012; Jang et al., 2015).

### MADS-Box Genes and Orchid Flowering

*SOC1* encodes a MADS-box transcription factor that is a member of the *Tomato MADS-box gene 3* (*TM3*)-like genes subfamily from angiosperms and gymnosperms (Lee et al., 2000; Becker and Theißen, 2003; Cseke et al., 2003; Nakamura et al., 2005). *SOC1* expression is detected in both leaves and shoot apices and is regulated by several floral pathways (Borner et al., 2000; Lee et al., 2000; Samach et al., 2000; Moon et al., 2003). In *Dendrobium nobile*, the expression of a close *SOC1* ortholog, *DnAGL19*, has been found to be increased after vernalization (Liang et al., 2012). In the orchid *Dendrobium* Chao Praya Smile, the expression of the *SOC1* ortholog *DOSOC1* is highly detected in reproductive organs, such as inflorescence apex, pedicel, floral buds and open flowers (Ding et al., 2013). *DOSOC1* expression is upregulated in the whole seedlings upon the floral transition (Ding et al., 2013). Overexpression of *DOSOC1* shows early flowering in both *Arabidopsis* and *Dendrobium* orchids, implying the evolutionary conserved functions of *SOC1*-like genes as activators of flowering (**Figure 2**) (Ding et al., 2013). Moreover, *DOSOC1* expression is downregulated in *DOFT* knockdown *Dendrobium* orchid, whereas its expression is upregulated in *DOFT* overexpression orchid (Wang et al., 2017), indicating a conserved regulatory mechanism of *SOC1*-like genes expression. Intriguingly, *DOSOC1* overexpression in *Dendrobium* results in abnormal floral organ development with formation of immature perianth organs only, indicating the role of *DOSOC1* in maintaining the identity of floral meristem and formation floral organs (Ding et al., 2013). This is in line with the function of *SOC1*-like genes in flower development in some plant species (Liu et al., 2013; Teo et al., 2014). Additionally, another MADS-box gene *FUL* has been shown to act redundantly with *SOC1* in regulation of flowering time in *Arabidopsis* (Melzer et al., 2008). Mutations in *ful* only slightly delay flowering, while in combination with *soc1* mutants, the flowering is further delayed as compared with both single mutants. Three *FUL*-like genes have been isolated in the orchid *Dendrobium thyrsiflorum*, namely *DthyrFL1/2/3* (Skipper et al., 2005). These three genes are upregulated during orchid reproductive development, yet their involvement in orchid flowering remains unknown.

*SOC1* expression is repressed by a floral repressor protein complex formed by FLC and SVP, which are also MADS-box proteins (Li et al., 2008). The orchid *SVP* orthologs have been reported in *Cymbidium* orchids, whereas no *FLC* orthologs have been isolated so far in monocots. The expression levels of *CgSVP1/2/3*, *SVP* orthologs, are greatly reduced upon cold treatment in *Cymbidium goeringii* (Yang et al., 2019). Moreover, transient overexpression of *CgSVP2* results in retarded flower bud growth, indicating its role as a repressor of flower bud formation (Yang et al., 2019). However, the involvement of *SVP* orthologs in orchid floral transition and its potential regulation of *SOC1* need further investigation.

*AGL6* encodes another MADS-box transcription factor which regulates the transition to flowering in *Arabidopsis* (Yoo et al., 2011). Knockdown of *AGL6* by artificial microRNA leads to late flowering, in contrast, *agl6-1D* wherein *AGL6* is activated by the *35S* enhancer shows early flowering. Two *AGL6*-like genes, *OMADS1* (*OAGL6-2*) and *OMADS7* (*OAGL6-1*), have been found in the *Oncidium* Gower Ramsey orchid, and overexpression of either gene leads to early flowering in *Arabidopsis* (Hsu et al., 2003; Chang et al., 2009), implying a conserved function of *AGL6*-like genes in mediating flowering.

As mentioned above, AP1, a MADS-box protein, specifies the identity of floral meristem as well as sepal and petal identity (Irish and Sussex, 1990; Mandel et al., 1992; Bowman et al., 1993). In *ap1* mutants, flowers exhibit a homeotic transformation that sepals develop into bracts and petals fail to develop. By comparison, *AP1* overexpression causes early flowering and conversion of the inflorescence meristem into a determinate floral meristem (Mandel and Yanofsky, 1995). Orthologs of *AP1* have been identified and characterized in *Oncidium*, *Dendrobium*, and *Cymbidium*. The expression of *AP1* orthologs can be detected in both vegetative tissues and reproductive structures such as floral buds and pedicel (Yu and Goh, 2000; Chen et al., 2007; Chang et al., 2009; Tian et al., 2013; Sawettalake et al., 2017). Overexpression of *AP1* orthologs, such as *OMADS10* from *Oncidium* Gower Ramsey and *DOAP1* from *Dendrobium* Chao Praya Smile, causes early flowering as well as conversion of inflorescence meristems to determinate floral meristems in *Arabidopsis* (Chang et al., 2009; Sawettalake et al., 2017). Transgenic *Dendrobium* orchids overexpressing *DOAP1* also show accelerated flowering as compared to wild-type orchid and conversion of inflorescence meristems to determinate floral meristems (Sawettalake et al., 2017). In addition, *DOAP1* expression is promoted by *DOFT* in *Dendrobium* orchids, which is a conserved regulation as that in *Arabidopsis* (Wang et al., 2017). These studies suggest that orchid *AP1* orthologs have conserved functions in promoting the floral transition and determination of floral meristems.

### MADS-BOX PROTEINS IN ORCHID FLORAL PATTERNING

### The Orchid Flower

In angiosperms, the flowers are usually composed of four types of structures, which form two parts, namely the vegetative part and the reproductive part. While the morphology and elaboration can differ greatly among different species, the diversification of floral patterning has taken place in a relatively conserved manner in having similar general organization of four types of structures arranged in four concentric whorls. In *Arabidopsis*, the flower consists of four concentric whorls of floral organs from the outer to inner whorls: sepals (four), petals (four), stamens (six), and two fused carpels. In orchids, the flowers are usually bilaterally symmetrical (zygomorphic) with three outer sepals, two inner petals, and a highly specialized inner median petal named lip or labellum which acts as the main pollinator attractant (**Figure 3A**). In a number of orchid species, the outer sepals and inner petals are named as tepals as they cannot be distinguished from each other morphologically. The reproductive structure gynostemium or column is composed of fused male (stamen/ anther) and female (carpel/pistil) organs.

## ABCE and Floral Quartet Model

Understanding the specification of the distinct floral organs through genetic study in *Arabidopsis* and *Antirrhinum majus* (snapdragon) has resulted in the birth of "ABCE model" and "floral quartet model" (**Figure 4A**) (Meyerowitz et al., 1989; Schwarz-Sommer et al., 1990; Coen and Meyerowitz, 1991; Weigel and Meyerowitz, 1994; Irish, 2010; Theißen et al., 2016). In the classical "ABC model," a combination of three gene classes specifies the four types of floral organs: sepal, petal, stamen, and carpel (Coen and Meyerowitz, 1991; Weigel and Meyerowitz, 1994). In *Arabidopsis*, the A-class genes (*AP1* and *AP2*) determine the sepal identify in the outermost whorl, A-class and B-class (*AP3* and *PI*) genes together specify petals in the second

whorl, B-class and C-class (*AG*) genes determine the identity of the male reproductive organ stamen in the third whorl, and the C-class gene specifies the female reproductive organ carpel in the innermost whorl. The expression domains of A-class and C-class genes are mutually exclusive. After discovering that the E-class genes (*SEP1–4*) are essential for the determination of all of the four whorls of floral organs, the classical "ABC model" was then extended to the "ABCE model." Additionally, D-class genes (*STK* and *SHP1/2*) are needed for determining the identity of ovule (Pinyopich et al., 2003). Interestingly, except for *AP2,* all these genes belong to MADS-box gene family, and their proteins form tetrameric complexes. This was coined as the "floral quartet model" as the tetrameric complexes functions as a whole to direct the development of specific floral organs (**Figure 4A**) (Theissen and Saedler, 2001). In orchids, the study of MADSbox transcription factors related to floral patterning has been challenging as recent functional analyses have shown that these genes have functionally diversified in their own lineages, making the prediction of function on the basis of orthology difficult (Irish and Litt, 2005).

#### MADS-Box Genes and Orchid Floral Patterning A-Class Genes

In orchids, there are several *AP1* orthologs isolated in *Cymbidium*, *Oncidium*, *Dendrobium*, and *Phalaenopsis* (**Table 2**). In *Dendrobium Madame Thong-In*, the *AP1* ortholog *DOMADS2* is expressed during the transitional phase and floral development (Yu and Goh, 2000). In mature flowers, *DOMADS2* is detected in the column and ovary but not in the pedicel, sepal, or petal. The differential expression pattern compared with *AP1* expression in the sepal and petal in *Arabidopsis* suggests the functional divergence of *AP1*-like genes during floral patterning. In the *Dendrobium* Chao Praya Smile orchid, the *AP1* ortholog *DOAP1* is detected at high levels in the inflorescence meristem as well as flowers. In addition, the overexpression of *DOAP1* can partially complement the *Arabidopsis ap1* mutant in restoring petal formation, suggesting *DOAP1* functions as a homeotic gene (Sawettalake et al., 2017). In *Oncidium* Gower Ramsey, the *AP1* orthologs *OMADS10* is expressed in the leaves, lip and carpel (**Figure 3B**). Overexpression of *OMADS10* induces early flowering without any floral organs defects in *Arabidopsis* (Chang et al., 2009). Two *AP1* orthologs, *PaAP1-1* and *PaAP1-2* has been isolated in the moth orchid *Phalaenopsis aphrodite* (Su et al., 2013b). *PaAP1-1* is mainly expressed in the pollinia and pedicel, whereas *PaAP1-2* is specifically expressed in the pedicel. The expression patterns of orchid *AP1* orthologs are unlike the A-class genes in *Arabidopsis* which are only present in the sepals and petals, but is somehow similar to the monocot lily *AP1* orthologs, *LMADS5/6*, which are expressed in the vegetative leaves and the innermost whorl carpel (Chen et al., 2008). This indicates the divergent function of orchid *AP1*-like genes in floral organ development.

#### B-Class Genes

The B-class genes are necessary for determining the identity of petals and stamens. There are two B-class genes in *Arabidopsis*, *AP3* and *PI,* analogous to the *A. majus DEFICIENS* and *GLOBOSA*, respectively. Mutations in either *AP3* or *PI* lead to similar phenotypes wherein petals and stamens are transformed into sepals and carpels, respectively (Bowman et al., 1989; Hill and Lord, 1989). Many studies have identified various numbers of B-class genes and studied their expression patterns in several orchid species (**Table 2**) (Hsu and Yang, 2002; Tsai et al., 2004; Tsai et al., 2005; Xu et al., 2006; Mondragón-Palomino and Theißen, 2008; Chang et al., 2009; Chang et al., 2010; Su et al., 2013b; Tsai et al., 2014; Hsu et al., 2015; Mao et al., 2015; Yang and Zhu, 2015). In *Dendrobium crumenatum*, the expression of *DcOAP3A* and *DcOPI* is detected in all parts of the mature flowers, but the expression of *DcOAP3B* is present in petals, lips, Teo et al. MADS-Box Proteins in Orchids

anthers, and column only (Xu et al., 2006). DcOAP3A/B can form heterodimers with DcOPI. In *Oncidium* Gower Ramsey, *OMADS3* (*OAP3-3*), *OMADS5* (*OAP3-1*), *OMADS9* (*OAP3- 2*), and *OMADS12* (*OAP3-4*) belong to the *AP3* lineage, while *OMADS8* (*OPI*) belongs to the *PI* lineage. *OMADS3* and *OMADS8* are expressed in both vegetative tissues and all floral organs of mature flowers (**Figure 3B**) (Chang et al., 2010). *OMADS5* is detected in both sepals and petals but not in the lip, whereas *OMADS9* is detected in the petals and lips (Chang et al., 2010). *OMADS12* is detected in the orchid reproductive floral organs including stamens and carpels, but not in the sepals, petals, and lips (Hsu et al., 2015). Overexpression of truncated *OMADS3* in *Arabidopsis* results in *ap2*-like flowers with homeotic conversion from sepals and petals to carpel-like and stamen-like organs (Hsu and Yang, 2002), while overexpression of *OMADS8,* but not *OMADS5/9,* causes the transformation of sepals to expanded petal-like structures (Chang et al., 2010). While OMADS3, OMADS5, and OMADS9 can assemble into both homodimers and heterodimers within the same group, OMADS8 can only form heterodimer with OMADS3 (Chang et al., 2010). It has been proposed that *OMADS3/5/8/9* is probably needed for the specification of sepals and petals and *OMADS3/8/9* but the absence of *OMADS5* leads to the formation of lips (Chang et al., 2010). In *P. equestris*, the MADS-box genes *PeMADS2*, *PeMADS3*, *PeMADS4,* and *PeMADS5* belong to the *AP3* lineage and *PeMADS6* belongs to the *PI* lineage. They are all expressed in lips and columns with *PeMADS2* also found in sepals and petals and *PeMADS3* in petals (Tsai et al., 2004; Tsai et al., 2005). Similarly to those B-class genes in *Dendrobium* and *Oncidium*, PeMADS2-5 interacts with PeMADS6 to form heterodimers and binds to CarG boxes on DNA (Tsai et al., 2005).

#### C- and D-Class Genes

In *Arabidopsis*, the C-class gene *AG* is required for the normal development of the stamens and carpels found in the third and fourth whorls, respectively. Mutations in *AG* cause the homeotic transformation of stamens and carpels into petals and sepals (Yanofsky et al., 1990). In addition, since *AG* is also necessary for floral meristem determinacy, the flowers of *ag* mutants are indeterminate and show the "flower within a flower" phenotype of sepal-petal-petal reiteration. D-class genes are required for regulating ovule identify. C and D-class genes are members of the *AG*-like family and are resulted from an ancient gene duplication event (Becker and Theißen, 2003). Both C and D-class genes have been identfied from several orchid speices (**Table 2**).

In *P. equestris*, *Phalaenopsis* sp. "Hatsuyuki," *Cymbidium ensifolium*, and *Oncidium* Gower Ramsey, the C-class genes *PeMADS1*, *PhalAG1*, *CeMADS1*, and *OMADS4,* respectively, are highly expressed in the floral buds and column in mature flowers (**Figure 3B**) (Song et al., 2006; Hsu et al., 2010; Wang et al., 2011; Chen et al., 2012). *OMADS4* in the *Oncidium* Gower Ramsey orchid is specifically detected in the stamens and carpels, similar to the expression pattern of LMADS10 from Lilium longiflorum (Hsu et al., 2010). Both *OMADS4* and *LMADS10*, when overexpressed in *Arabidopsis*, result in early flowering, whereas *LMADS10* overexpression also leads to curly leaves and floral organ conversions, indicating the probable functional diversification of the monocot C-class genes. Several C-class genes from other orchid species have broader expression pattern. For examples, the *D. crumenatum* and *D. thyrsiflorum AG* orthologs, *DcOAG1* and *DthyrAG1*, respectively, are expressed in all kinds of floral organs and are not confined to the reproductive organs (Skipper et al., 2006; Xu et al., 2006). This expression pattern is similar to the *AG* homolog from *Illicium floridanum* that is also expressed in the tepals and reproductive organs (Kim et al., 2005), suggesting the regulatory mechanisms involved in the regulation of the expression of these C-class genes have evolved independently. The ectopic expression of *DcOAG1* in *Arabidopsis* accelerates flowering with abnormal floral organs in the first and second whorls (Xu et al., 2006).

*OMADS2* in the orchid *Oncidium* Gower Ramsey, a D-class gene, is specifically detected in stigmatic cavity and ovary (Hsu et al., 2010). This expression pattern is close to that of LMADS2 from *L. longiflorum*, which is exclusively present in the carpel (Tzeng et al., 2002). OMADS2 forms homodimers and heterodimers with OMADS4. Overexpression of *OMADS2*  in *Arabidopsis* leads to early flowering without any floral organ converstion (Hsu et al., 2010).

#### E-Class Genes

The members of the *SEP* MADS-box subfamily belong to the E-class genes that are necessary for the formation of all floral organs and floral meristem determinacy in *Arabidopsis*. The triple mutant in *SEP1/2/3/4* genes produce flowers with all floral organs converted to leaf-like organs (Pelaz et al., 2000; Ditta et al., 2004). *SEP* genes are present in angiosperms, but not gymnosperms, indicating that *SEP* genes may have been important for the existence of flowers (Nam et al., 2003). In *D. crumenatum*, the *SEP* ortholog *DcOSEP1* is detected in all floral organs, similarly to *Arabidopsis SEP*s. DcOSEP1 is able to interact with the DcOAP3A-DcOPI and DcOAP3B-DcOPI heterodimers, but not with DcOAP3A and DcOPI individually, indicating that DcOSEP1 is able to form a higher order protein complex with DcOAP3A-DcOPI or DcOAP3B-DcOPI, similar to their counterparts in *Arabidopsis* (Honma and Goto, 2001; Theissen and Saedler, 2001; Xu et al., 2006). In the *Dendrobium* Madame Thong-In orchid, *DOMADS1* and *DOMADS3* encode MADS-box proteins closely related with SEP1 and SEP3, respectively. *DOMADS1* is present in all the floral organs similarly to *DcOSEP1*, while *DOMADS3* is only present in the pedicel (Yu and Goh, 2000). In the *Oncidium* Gower Ramsey orchid, *OMADS6* (*OSEP3*) and *OMADS11* (*OSEP1*) encode MADS-box proteins homologous to *SEP3* and *SEP1/2*, respectively. Both genes are highly present in the sepal, petal, lip, and carpel, with weaker and undetectable expression in stamens for *OMADS6* and *OMADS11*, respectively (**Figure 3B**) (Chang et al., 2009). Overexpression of *OMADS6* in *Arabidopsis* leads to homeotic transformation of sepals into carpelliod structures and petals into stamen-like organs (Chang et al., 2009). In *P. equestris*, the four *SEP*-like *PeSEP* genes are expressed in flower buds with the expression of *PeSEP2* higher in floral stalk and column and *PeSEP3* in petals (Pan et al., 2014). Like the SEP proteins in *D. crumenatum*, PeSEP2, PeSEP3, and PeSEP4 proteins cannot interact with the B-class proteins PeMADS2, PeMADS4, PeMADS6, or the D-class protein PeMADS7 individually, but can form multimeric complexes with PeMADS2/6, PeMADS4/6 and PeMADS6/7. Only PeSEP1 is able to interact with PeMADS2, PeMADS4, PeMADS6, and PeMADS7 individually, and with PeMADS2/6, PeMADS4/6, and PeMADS6/7 (Pan et al., 2014). Silencing of *PeSEP3* by VIGS results in the conversion of tepal to leaf-like organ in *Phalaenopsis*, whereas silencing of *PeSEP2* does not greatly affect flower development (Pan et al., 2014), suggesting that these *PeSEPs* have divergent functions in orchid flower development.

The *AGL6*-like genes are similar to *SEP*-like genes and the *AGL6*-like gene in petunia functions like *SEP* genes in floral patterning (Rijpkema et al., 2009). It has been proposed to add *AGL6*-like genes to class-E genes. As described above, the *Arabidopsis AGL6* functions as a flowering promoter (Yoo et al., 2011). *Arabidopsis* has another *AGL6*-like gene called *AGL13*, which acts similarly to E-class *SEP* genes in specifying male and female gametophytes (Hsu et al., 2014). The orchid *Oncidium* Gower Ramsey also has two *AGL6*-like genes, *OMADS1* and *OMADS7*. *OMADS7* is expressed in all the floral organs, similar to that of E-class gene including *OMADS6* (**Figure 3B**) (Chang et al., 2009). *OMADS1* shows a different expression pattern which is in the lip and carpel, but not in other floral organs (**Figure 3B**) (Chang et al., 2009). Besides being early flowering, the flowers of *OMADS1* or *OMADS7* overexpression show homeotic transformation of sepals into carpel-like structures (Chang et al., 2009), indicating their dual roles in promoting floral transition and regulating floral organ formation. In *P. aphrodite*, the *AGL6* like gene *PaAGL6-1* is expressed specifically in the lip, suggesting that *PaAGL6* may play an important role in lip formation (Su et al., 2013b). Subsequent studies have further revealed that the orchid *AGL6*-like genes play important roles in determining sepal/petal/lip formation (discussed in the following section). Together, *AGL6*-like genes may have diverse function in all four whorls of floral organs.

## The Orchid Perianth Code

Flowers in orchids and several other monocots such as lily, the sepals and petals are morphologically similar and are also collectively called tepals. This is different from *Arabidopsis* and other dicots flowers, in which sepals and petals have distinguished morphologies. To explain this difference in perianth organs specification, the modified ABC model has been proposed, in which the expression domain of B-class genes are extended to the outermost whorl of floral organs in many orchid species (**Figure 4B**) (Van Tunen et al., 1993; Bowman, 1997; Kramer et al., 2003; Mondragón-Palomino and Theißen, 2008). In the *Oncidium* Gower Ramsey orchid, the *AP3*-like gene *OMADS3* and the *PI*-like gene *OMADS8* are detected in all perianth organs (**Figure 3B**) (Chang et al., 2010).

However, the orchid flower has a median petal called lip, which has a highly diversified morphology and acts as the main attractor of pollinators. The specification of the lip cannot be simply explained by the modified "ABC model". Several years of molecular studies of MADS-box proteins and orchid floral patterning have led to the discovery of the model of formation of perianth organs: the Perianth (P) code (**Figure 4C**) (Hsu et al., 2015). Based on this model, the two tetrameric MADSbox protein complexes, SP (sepal/petal) complex (OAP3-1/ OAGL6-1/OAGL6-1/OPI) and L (lip) complex (OAP3-2/ OAGL6-2/OAGL6-2/OPI), compete to promote the development of sepal/petal and lip, respectively. Different copies of B-class *AP3*-like genes (*OAP3-1* and *OAP3-2*) and *AGL6*-like genes (*OAGL6-1* and *OAGL6-2*) have different whorl-specific or whorlbiased expression patterns, providing the basis for the formation of SP and L complexes. Moreover, the relative levels of the two complexes may also determine the formation of various forms of intermediate lips or distinct lips in orchid.

### CONCLUDING REMARKS AND FUTURE PERSPECTIVES

The orchid family is the second largest family of angiosperms and has delighted cultivators for their unsurpassed beauty and complexity. The study on orchids has come a long way since people began gathering and propagating them under controlled environment. Recent findings in orchids, mainly *Cymbidium*, *Oncidium*, *Dendrobium*, and *Phalaenopsis*, have revealed that MADS-box proteins play critical roles in orchid flowering and floral patterning. The unraveled molecular mechanisms underlying orchid flowering and floral development can be applied to both classical orchid breeding and targeted manipulation of orchids for desired flowering traits and floral patterns. Orchids have many MADS-box genes, for example, 51 in *P. equestris* and 63 in *Dendrobium catenatum* (Cai et al., 2015; Zhang et al., 2016), however, only several of the MADSbox genes have been characterized, and most of them are shown be involved in orchid flowering or floral development. A recent study shows that MADS-box regulators might be relevant with the development of seeds without endosperm and epiphytism in orchids (Zhang et al., 2017), but revealing the functions of MADS-box genes in other develomental processes needs more future endeavor. In addition, in this review, the comparisons of orchid MADS-box genes to *Arabidopsis* have been included as many orchid genes are named and share similar functions to their closest orthologs in *Arabidopsis*. With the increasing number of MADS-box genes being studied in model monocots such as rice, wheat, barley, maize, and lily (Callens et al., 2018), more detailed comparisons of MADS-box gene functions between orchid and other monocots can be performed and discussed in the near future.

Nowadays, the function of orchid MADS-box genes, in many cases, is studied in heterologous plant systems (e.g. *Arabidopsis* and tobacco). In only several studies, the function of MADS-box genes has been examined by using transient overexpression/knockdown in orchids by VIGS or generating transgenic orchid overexpressing or knocking down of target genes (**Table 2**). For examples, silencing of *OAGL6-2* by VIGS has been done in *Oncidium* Gower Ramsey and *Phalaenopsis amabilis* hybrid to test the Perianth code, and *DOAP1* when overexpressed in *Dendrobium* Chao Praya Smile, leads to early flowering. To better understand the function of orchid genes, there is a need for more reliable and faster genetic transformation systems of different orchid species in orchid study and targeted orchid breeding with desired traits. Moreover, with modern genomic editing tools such as CRISPR-Cas9 (Clustered Regularly Interspaced Short Palindromic Repeats–Caspase 9), it is now possible to generate orchid mutants for *in vivo* functional characterizations (Kui et al., 2017). Indeed, CRISPR-Cas9 has been successfully used to create multiple mutants of MADS genes in the orchid *P. equestris* very recently (Tong et al., 2019).

With the advent of sequencing technologies, five orchid genomes have been released, including *Apostasia shenzhenica*, *D. catenatum*, *Dendrobium officinale*, *P. equestris*, and *Vanilla planifolia* (Cai et al., 2015; Yan et al., 2015; Zhang et al., 2016; Zhang et al., 2017; Hu et al., 2019). Moreover, the transcriptomes of several orchid species from different subfamilies are freely available in online databases, such as Orchidstra 2.0 (http://orchidstra2. abrc.sinica.edu.tw), OrchidBase 3.0 (http://orchidbase.itps.ncku. edu.tw) and OOGB (http://predictor.nchu.edu.tw/oogb) (Chang et al., 2011; Fu et al., 2011; Su et al., 2013a; Tsai et al., 2013; Chao et al., 2017; Tsai et al., 2017). Recently, the transcriptome of a Mediterranean orchid *Orchis italica* inflorescence have also been analyzed (Valoroso et al., 2019). Genome sequences and transcriptomic data have provided valuable information

### REFERENCES


in aiding basic research and genomics-assisted horticultural breeding. The advent of molecular tools has allowed genomic analysis to determine the underlying mechanisms behind many morphological characteristics and developmental processes. With modern genomic editing tools available, it is now feasible to generate mutants or novel varieties in orchids. This would greatly help not only in the molecular genetic research of orchid biology, but also in generating novel orchid varieties with various desirable traits through targeted gene editing.

#### AUTHOR CONTRIBUTIONS

ZT and LS wrote the manuscript. WZ made the drawings of *Arabidopsis* and orchid plants. All authors read and approved of the manuscript.

### FUNDING

Preparation of the review has been supported by the intramural funding from Temasek Life Sciences Laboratory and National University of Singapore.

specify the central cell in *Arabidopsis* ovules. *Plant Cell* 20, 2088–2101. doi: 10.1105/tpc.108.058958


subfamilies in the morphological evolution of eukaryotes. *J. Mol. Evol.* 43, 484– 516. doi: 10.1007/BF02337521


**Conflict of Interest:** 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 Teo, Zhou and Shen. 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.*

# Phalaenopsis LEAFY COTYLEDON1- Induced Somatic Embryonic Structures Are Morphologically Distinct From Protocorm-Like Bodies

Jhun-Chen Chen1,2, Chii-Gong Tong1,2, Hsiang-Yin Lin1,2 and Su-Chiung Fang1,2\*

<sup>1</sup> Biotechnology Center in Southern Taiwan, Academia Sinica, Tainan, Taiwan, <sup>2</sup> Agricultural Biotechnology Research Center, Academia Sinica, Taipei, Taiwan

#### Edited by:

Jen-Tsung Chen, National University of Kaohsiung, Taiwan

#### Reviewed by:

Sonia Gazzarrini, University of Toronto, Canada Guangdong Wang, Nanjing Agricultural University, China

> \*Correspondence: Su-Chiung Fang scfang@gate.sinica.edu.tw

#### Specialty section:

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

Received: 01 July 2019 Accepted: 13 November 2019 Published: 29 November 2019

#### Citation:

Chen J-C, Tong C-G, Lin H-Y and Fang S-C (2019) Phalaenopsis LEAFY COTYLEDON1-Induced Somatic Embryonic Structures Are Morphologically Distinct From Protocorm-Like Bodies. Front. Plant Sci. 10:1594. doi: 10.3389/fpls.2019.01594 Somatic embryogenesis is commonly used for clonal propagation of a wide variety of plant species. Induction of protocorm-like-bodies (PLBs), which are capable of developing into individual plants, is a routine tissue culture-based practice for micropropagation of orchid plants. Even though PLBs are often regarded as somatic embryos, our recent study provides molecular evidence to argue that PLBs are not derived from somatic embryogenesis. Here, we report and characterize the somatic embryonic tissues induced by Phalaenopsis aphrodite LEAFY COTYLEDON1 (PaLEC1) in Phalaenopsis equestris. We found that PaLEC1-induced somatic tissues are morphologically different from PLBs, supporting our molecular study that PLBs are not of somatic embryonic origin. The embryonic identity of PaLEC1-induced embryonic tissues was confirmed by expression of the embryonic-specific transcription factors FUSCA3 (FUS3) and ABSCISIC ACID INSENSITIVE3 (ABI3), and seed storage proteins 7S GLOBULIN and OLEOSIN. Moreover, PaLEC1-GFP protein was found to be associated with the Pa7S-1 and PaFUS3 promoters containing the CCAAT element, supporting that PaLEC1 directly regulates embryo-specific processes to activate the somatic embryonic program in P. equestris. Despite diverse embryonic structures, PaLEC1-GFP-induced embryonic structures are pluripotent and capable of generating new shoots. Our study resolves the long-term debate on the developmental identity of PLB and suggests that somatic embryogenesis may be a useful approach to clonally propagate orchid seedlings.

Keywords: somatic embryogenesis, Phalaenopsis orchids, LEAFY-COTYLEDON1, protocorm-like-body, Phalaenopsis equestris

## INTRODUCTION

Plant totipotency provides a means of clonally propagating a wide variety of plant species (Ikeuchi et al., 2013; Kareem et al., 2016; Radhakrishnan et al., 2017). In vitro regeneration bypasses the embryogenic program and directly generates seedlings with identical genetic makeup. One of the most common in vitro regeneration methods is somatic embryogenesis (Zimmerman, 1993; Pulianmackal et al., 2014). Somatic embryogenesis is crucial for establishing genetic transformation platforms for many non-model plant species and for clonal propagation of numerous high-value plants. For example, somatic embryos are used as transformation materials for alfalfa, American chestnut, cassava, cotton, grapevine, maize, mango, melon, Norway spruce, papaya, rose, tea tree, and walnut (Umbeck et al., 1987; Mcgranahan et al., 1988; Robertson et al., 1992; Fitch et al., 1993; Li et al., 1996; Brettschneider et al., 1997; Trinh et al., 1998; Mondal et al., 2001; Akasaka-Kennedy et al., 2004; Chavarri et al., 2004; Li et al., 2006; Polin et al., 2006; Vergne et al., 2010). In addition, the regeneration capacity of somatic embryos has made somatic embryogenesis a common method through which to clonally propagate economically important trees or herbal plants (Joshee et al., 2007; Nordine et al., 2014; Guan et al., 2016; Kim et al., 2019).

Embryogenesis is a defined developmental program during which the zygote grows and develops into a mature embryo. Somatic embryogenesis, on the other hand, activates the embryogenesis program in the absence of gamete fusion (von Arnold et al., 2002; Braybrook and Harada, 2008; Yang and Zhang, 2010; Feher, 2015). Zygotic embryogenesis and somatic embryogenesis programs not only share similar morphogenesis and maturation phases, they also share similar if not completely identical genetic and molecular networks (Zimmerman, 1993; Mordhorst et al., 2002; Gaj et al., 2005). Moreover, ectopic expression of several key embryo-associated transcription factors (TFs) is capable of inducing the embryogenesis program in somatic tissues (Lotan et al., 1998; Hecht et al., 2001; Stone et al., 2001; Boutilier et al., 2002; Zuo et al., 2002; Harding et al., 2003; Kwong et al., 2003; Gaj et al., 2005; Wang et al., 2009), demonstrating the developmental plasticity of plant tissues.

Orchids evolve specialized developmental programs including the co-evolution of diverse floral structures and pollinators (Waterman and Bidartondo, 2008), formation of pollen dispersal units (pollinia) (Pacini and Hesse, 2002), lack of cotyledon organogenesis during embryogenesis (Kull and Arditti, 2002; Yeung, 2017), and mycorrhizal fungi-assisted seed germination (Rasmussen, 2002), and all of these developmental processes contribute to their distinct morphology and physiological characteristics. These unique developmental strategies have not only fascinated many evolutionary and plant biologists; the beauty of the resulting floral structures is also enthusiastically admired by the general public. Much effort has been put into tissue culturebased clonal propagation of elite orchids over the past decades and this technology has transformed the orchid business into a multimillion-dollar orchid biotechnology industry (Winkelmann et al., 2006; Liao et al., 2011; Hossain et al., 2013).

Generally, embryogenesis of angiosperm plants starts from morphogenesis with continuous changes in embryo morphology and establishment of shoot-root polarity followed by maturation and desiccation processes (Bentsink and Koornneef, 2008; Braybrook and Harada, 2008). One of the characteristic features that defines the somatic embryo is the formation of the embryonic cotyledons. Even though orchid embryos go through a maturation and desiccation process, they lack characteristic cotyledons (organogenesis) and fail to establish a shoot-root axis during embryogenesis (Arditti, 1992; Dressler, 1993; Burger, 1998). Instead, a tubular embryo structure with an anterior meristem is formed. Upon germination, a tubular embryo emerges as a protocorm and new leaves and roots are generated from the anterior meristem of the protocorm (Nishimura, 1981).

Protocorm-like body (PLB)-based regeneration is commonly used to produce huge amounts of orchid seedlings of elite cultivars (Arditti and Krikorian, 1996;Chen et al., 2002;Arditti, 2009;Chugh et al., 2009; Yam and Arditti, 2009; Paek et al., 2011; Yam and Arditti, 2017). For years, much effort has been devoted to develop in vitro protocols to induce PLB and somatic embryo development either directly or indirectly (via the callus tissue) from explants to improve micropropagation in Phalaenopsis orchids (Tokuhara and Mii, 2001; Tokuhara and Mii, 2003; Kuo et al., 2005; Chen and Chang, 2006; Gow et al., 2009; Gow et al., 2010; Pramanik et al., 2016). PLBs are often induced from somatic tissues such as protocorms, floral stalk internodes, leaves, and root tips (Chen et al., 2002; Park et al., 2002; Chen and Chang, 2004; Chen and Chang, 2006; Teixeira da Silva et al., 2006; Zhao et al., 2008; Guo et al., 2010; Paek et al., 2011). Because embryogenesis produces tubular embryos, which, upon germination, develop into protocorms, and PLBs resemble protocorms morphologically, initiation, and development of PLBs is often regarded as somatic embryogenesis (Jones and Tisserat, 1990; Begum et al., 1994; Ishii et al., 1998).However, our recent study providesmolecular evidence to argue that PLBs are not somatic embryos. Instead, PLB development seems to adopt a shoot pole-related organogenesis program (Fang et al., 2016). So if PLBs are not somatic embryos, what does the somatic embryo of orchids look like? Are somatic embryos of orchid species also prolific in propagating clonal plants? To address these questions and to explore the potential of using orchid somatic embryos for clonal propagation, we generated transgenic plants overexpressing Phalaenopsis aphrodite LEAFY COTYLEDON1 (PaLEC1) gene in Phalaenopsis equestris. LEC1 encodes a HAP3 subunit of CCAAT-binding TF required for embryo development (Lotan et al., 1998). Moreover, ectopic expression of LEC1 isolated from Arabidopsis and other plant species has been demonstrated to be a potent inducer of the somatic embryogenesis program (Lotan et al., 1998; Yazawa et al., 2004; Garcês et al., 2014; Zhu et al., 2014;Min et al., 2015; Fang et al., 2016). Consistently, we have shown that PaLEC1 is capable of inducing somatic embryogenesis in Arabidopsis (Fang et al., 2016). Here, we report PaLEC1-induced embryonic tissues in P. equestris and demonstrate their embryonic identity by validation of expression of the embryonic-specific genes. To the best of our knowledge, this is the first time that somatic embryonic tissues was shown to be induced by the embryonic factor, PaLEC1, in the transgenic orchid plants. Our studies reveal that PaLEC1 may be a useful tool for large scale propagation of clonal plants in orchid biotechnology.

### MATERIALS AND METHODS

### Plant Materials and Growth Conditions

P. equestris plants were grown and maintained in a growth chamber with alternating 12 h light (23°C)/12 h dark (18°C) cycles. Hand pollination was conducted for each flower, as described previously (Chen and Fang, 2016).

Seeds were allowed to germinate in germination medium (0.1% tryptone, 1.8% sucrose, 2% potato, 0.75% agar, 1.08 g/l MS salt, 1X MS vitamin pH 5.7). For regeneration assay, the protocorm segments (explants) were transferred to T2 medium containing 0.1% (w/v) tryptone, 2% (w/v) sucrose, 2% (w/v) potato homogenate, 2.5% (w/v) banana homogenate, 0.01% (v/v) citric acid, 0.1% (w/v) charcoal, and 1% (w/v) agar adjusted to pH 5.5 (Chen et al., 2009) to induce PLBs.

#### Generation of Transgenic Phalaenopsis equestris Plants

Protocorms or PLBs of P. equestris were used for transformation. Agrobacterium-based transformation was conducted as described previously with slight modification (Liau et al., 2003). Briefly, protocorms or PLBs were co-cultured with Agrobacterium tumefaciens strain EHA105 culture (OD600 = ~ 0.8) containing the 35S:PaLEC1-GFP or pH7FWG2 (vector only control) construct in the presence of 100 mM acetosyringone overnight (for protocorms) or 3 days (for PLBs) in the dark. After removing Agrobacterium inoculum, the infected tissues were washed four to six times with distilled water containing 40 mg/L meropenem (China Chemical & Pharmaceutical). The infected tissues were allowed to recover in New Dogashima medium (NDM) (Belarmino and Mii, 2000) containing 10 g/L maltose, 0.1 mg/L naphthaleneacetic acid (NAA), and 0.4 mg/L benzyladenine (BA) and supplementedwith 40mg/L meropenemfor 1 week in the dark. All the NDM media used in this study contained 10 g/L maltose, 0.1 mg/L NAA, and 0.4 mg/L BA. The recovered tissues were then selected on NDM agar plates supplemented with 20 mg/L hygromycin and 40 mg/L meropenem at 23°C in the dark. The infected tissues were selected continuously by sub-culturing tofresh NDM plates supplemented with 20 mg/L hygromycin and 40 mg/L meropenem at 2- to 4-week intervals until new leaves emerged. The hygromycin-resistant transgenic plants were then transferred to T2 (Chen et al., 2009) medium supplemented with 20 mg/L hygromycin and 40 mg/L meropenem to encourage growth.

#### Deoxyribonucleic Acid Isolation and Southern Blotting

Genomic DNA from young seedlings (1 to 1.5 cm in size) of T1 transgenic plants and wild-type plants was isolated by cetyltrimethylammonium bromide (CTAB)-based method (Murray and Thompson, 1980; Saghai-Maroof et al., 1984) with some modifications. Briefly, freeze-dried tissue was ground in the presence of liquid nitrogen, resuspended in 1 g/10 ml of the CTAB extraction buffer [100 mM Tris, pH 8.0, 1.4 M NaCl, 20 mM EDTA, 2% CTAB, 0.2% (w/v) polyvinylpyrrolidone 40, and 0.2% b-mercaptoethanol], and incubated at 65o C for 60 min with occasional gentle vortexing. Equal volume of chloroform: isoamylalcohol (24:1) was then added and mixed by inversion for 15 min. Plant extract was then centrifuged at 3220 × g for 15 min at 4o C. The aqueous phase was extracted with equal volume of chloroform:isoamylalcohol (24:1) one more time. The aqueous phase was transferred to a new tube and equal volume of isopropanol was added, mixed, and incubated at ‑20<sup>o</sup> C overnight. DNA was precipitated by centrifugation at 3,220 × g at 4o C for 10 min. DNA was washed once with 75% ethanol followed by 99% ethanol. DNA was then resuspended in H2O supplemented with 10 µg/ml RNase A and incubated at 37o C overnight to remove RNA. This method yielded approximately 150‑350 µg per g of orchid tissue. Approximately 15 µg DNA was digested and Southern blotting and hybridization were carried out as described previously (Fang et al., 2006).

#### Antibody Generation and Western Blotting

To generate polyclonal antibodies specific for PaLEC1 protein, PaLEC1 peptide YLHRYRELEGDHRGSIRG (LTK BioLaboratories, Taiwan) was synthesized to generate rabbit polyclonal antisera. Polyclonal antibodies raised against PaLEC1 were affinity purified for western blotting analysis.

To prepare total protein extract, tissues were ground to a fine powder in the presence of liquid nitrogen using a pestle and mortar. One gram of ground tissues was homogenized in 1 ml 1X Urea-based protein extraction buffer (250 mM Tris-HCl, pH 6.8, 3.5% sodium dodecyl sulfate (SDS), 10% glycerol, 1 M urea) supplemented with 1X protease inhibitor cocktail (Sigma-Aldrich), 1 mM phenylmethylsulfonyl fluoride by vortexing. Cell debris was removed by centrifugation at 16,100 × g for 10 min at 4°C. Concentration of protein lysates was determined by Bio-Rad DC Protein Assay Kit (Bio-Rad).

Protein lysates were resolved by a standard 10% SDSpolyacrylamide gel electrophoresis (PAGE) and transferred to Immobilon-P PVDF membrane (Merck Millipore). Blots were blocked in 1 × TBST (20 mM Tris base, 150 mM NaCl, 0.1% Tween 20) supplemented with 5% non-fat milk at room temperature (RT) for 1 h and incubated with anti-GFP antibodies (Roche) diluted 1:2,000 in 1 × TBST supplemented with 5% non-fat milk at 4°C overnight. For detection of PaLEC1 and PaRbCL proteins, blots were incubated with PaLEC1 (1:2,000) and RbCL (1:5,000, Agrisera) antibodies diluted in 1 × TBST with 5% non-fat milk for 1 h at RT. Blots were washed three times for 15 min each time and then incubated with horseradish peroxidase (HRP) conjugated goat-anti-mouse-IgG (1:20,000, Jackson ImmunoResearch Laboratories), or goat-antirabbit-IgG (1:20,000, PerkinElmer) for 1 h at RT. After washing in 1 × TBST for 15 min three times, the blots were processed for chemiluminescence detection using Amersham ECL select Western Blotting detection reagent. The images were acquired by a ChemiDoc XRS+ imager (Bio-Rad).

#### Ribonucleic Acid Isolation and Quantitative Real-Time Polymerase Chain Reaction

Plant tissues (young seedlings or callus tissues) were homogenized by pestle and mortar in the presence of liquid nitrogen. RNA was isolated using RNA extraction reagent (3-Zol, MDBio) according to the manufacturer's instructions. Total RNA was treated with RNase-free DNase (Qiagen) to remove DNA followed by RNeasy column purification (Qiagen) according to the manufacturer's instructions.

Approximately 5 mg DNA-free RNA was reverse transcribed in the presence of a mixture of oligo dT and random primers (9:1 ratio) using the GoScript Reverse Transcription System (Promega) according to the manufacturer's instructions. Ten microliters of quantitative RT-PCR reaction was set up as follows: 2.5 ml of 1/20 diluted complementary DNA, 0.2 mM of primers, and 5 ml of 2X KAPA SYBR FAST master mix (KAPA Biosystems). The PCR program was used for DNA amplification: 95°C for 1 min, 40 cycles of 95°C for 5 s, and 58°C, 60°C, or 62°C for 20 s. PCR was performed in triplicate. Standard error was calculated from three technical replicates. Fold change in expression was calculated as 2 ‑△△CT. A melting curve of each PCR was examined to ensure no spurious products were present. Primer pairs used for quantitative PCR are listed in Supplementary Table S1. A ubiquitin gene, PaUBI (Lin et al., 2014), was used as an internal control.

#### Light Microscopy

Tissues were fixed in paraformaldehyde fixative and sectioned to 10 mm in thickness using the hybrid-cut method as previously described (Chen et al., 2016). Tissues sections were then stained with hematoxylin solution and photographed on a Zeiss Axio Scope A1 microscope equipped with an Axio-Cam HRc camera (Zeiss, Germany).

Fresh tissues of transgenic plants were hand sectioned and fluorescence images were photographed on a LSM 710 Confocal Microscope (Zeiss).

#### Chromatin Immunoprecipitation-Polymerase Chain Reaction

Chromatin immunoprecipitation was conducted as described previously (Villar and Köhler, 2010) with slight modification. One gram of young seedling (1 to 1.5 cm in size, pooled from approximately 10 T1 seedlings) tissue was fixed in 37 ml of 1% formaldehyde and vacuum-infiltrated for 20 min. Two and half milliliters of 2 M glycine was added to stop crosslinking. Fixed tissues were then homogenized by pestle and mortar and chromatin was extracted and filtered by a 70 mM cell strainer (Biologix Group Limited) as described (Villar and Köhler, 2010). Chromatin was resuspended in 200 ml of nuclei lysis buffer (50 mM Tris-HCl, pH 8.0; 10 mM EDTA, 1% SDS, and 1X protease inhibitor cocktail) before sonication. Chromatin was then transferred to a micro TUBE-500 AFA Fiber Screw-Cap tube and sonicated for 12 min or 20 min under the condition: peak incident power 150 S; duty factor 2%, cycles per burst 200, in a Covaris S220 system (Covaris). The sheered chromatin was precleared by incubating with 40 ml protein A Magnetic Sepharose beads (Merck) at 4°C for 1.5 h. In the meantime, 5 mg PaLEC1 polyclonal antibody or histone H3 antibody (Agrisera) was incubated with protein A beads at 4°C for 1.5 h. After preclearing, protein A/antibody mix was added to each sheered chromatin sample and incubated at 4°C overnight. Magnetic beads were then washed as described (Villar and Köhler, 2010) and eluted by adding 150 ml of elution buffer (1% SDS and 0.1% NaHCO3) and incubating at 65°C for 15 min. Elution was repeated one more time and elutes were pooled into a single tube. Twelve microliters of 5 M NaCl was added into 200 ml elutes and incubated at 65°C overnight. Proteins were removed by incubating with 6 ml of 2 mg/ml proteinase K, 6 ml of 0.5 M EDTA, and 12 ml of 1 M This-HCl (pH 6.5) at 45°C for 2 h. Immunoprecipated DNA was then purified by QIAquick PCR Purification Kit (Qiagen) following the manufacturer's instructions.

Twenty microliters of PCR reaction was set up as follows: 2 or 3 ml of chromatin immunoprecipitation (ChIP) DNA, 0.5 mM of primers, 2 ml of 10X PCR buffer, 3% dimethyl sulfoxide, and 1.25 unit Power Taq polymerase (Genomics, Taiwan). The PCR program was used for amplification of immunoprecipitated DNA: 95°C for 2 min, 42 cycles of 94°C for 5 s, 58–60°C for 20 s, and 72°C for 20 s. Primers used for ChIP-PCR are listed in Supplementary Table S1. ChIP-PCR was repeated three times using three independent biological samples.

## RESULTS

#### Generation of Transgenic Phalaenopsis equestris Plants Overexpressing PaLEC1- GFP

To explore the possibility ofinducing a somatic embryonic program in Phalaenopsis orchids and to investigate whether somatic embryonic potential can be utilized for micropropagation, transgenic P. equestris plants overexpressing the embryonic maker PaLEC1 that has been shown to induce somatic embryogenesis in Arabidopsis (Fang et al., 2016) were generated. After transformation, we successfully obtained multiple independent transgenic lines carrying the 35S:PaLEC1-GFP transgene with the hygromycin phosphotransferase (hpt) marker (Supplementary Figure S1A). Expression of PaLEC1-GFP was also confirmed in four out of seven independent transgenic lines by western blotting using the GFP antibody (Supplementary Figure S1B). However, multiple transformation attempts failed to generate transgenic plants carrying the empty pH7FWG2 vector. It is not clear why only 35S:PaLEC1-GFP transgenic plants were obtained. Overexpression of the embryonic genes BABYBOOM has been reported to enhance embryogenic callus proliferation, overall transformation efficiency, and regeneration capacity in several monocot plants recalcitrant for transformation (Lowe et al., 2016). Hence, it is possible that ectopic PaLEC1 enhances transformation efficiency by increasing embryogenic callus formation and regeneration ability.

#### Altered Protocorm Development in Transgenic Plants Overexpressing PaLEC1-GFP

The 35S:PaLEC1-GFP transgenic plants were allowed to grow in growth chamber for approximately 5 years before reaching maturity for flowering. 35S:PaLEC1-GFP transgenic lines #1, #2, and #3 started to flower in March of 2018 and were therefore used in this study. The flowers were hand-pollinated and T1 seeds were allowed to produce in developing capsules. Seeds were sown and allowed to germinate on medium containing hygromycin.

As expected, seeds harvested from wild-type control plants failed to grow and were arrested on hygromycin-containing plates 5 months after sowing (MAS, Figure 1A). T1 seeds from 35S: PaLEC1-GFP transgenic lines #1 and #2, on the other hand, started to expand and turned green at 2 MAS (Figure 1B). The germinated protocorms continued to grow and produce new shoots on hygromycin-containing plates at 5 MAS (Figure 1B). Transgenic 35S:PaLEC1-GFP line #3 failed to produce viable seeds and was therefore omitted from further study. Southern blotting was used to confirm independent integration of the transgene in 35S:PaLEC1- GFP#1 and 35S:PaLEC1-GFP #2 lines (Supplementary Figure S1C).

Protocorms from wild-type plants continued to develop and grow into young seedlings at approximately 80–90 days after sowing (DAS, Figures 2A, B). Instead of growing into individual seedlings, protocorms of the transgenic 35S:PaLEC1-GFP #1 plants developed into various structures (Figure 2A) that did not completely resemble seedlings of wild-type seedlings at 80– 90 DAS (Table 1). For example, approximately 25% germinated protocorms underwent uncontrolled cell division and formed unorganized callus-like structures (callus type, Figure 2C), which is similar to somatic embryonic calli commonly observed in other plant species (Zimmerman, 1993;Yang and Zhang, 2010;Ikeuchi et al., 2013).Around 48% of protocorms, on the other hand, developed into a fan-like structure with multiple meristems forming on the surface (multiple-meristem type, Figure 2C). Around 27% protocorms formed a dorsal-fin-like structure with newly emerging leaves growing from it (dorsal-fin type, Figure 2C). The callus-like structure was capable of acquiring meristematic fate and developed into shoots (Supplementary Figure S2). Even though only one shoot was developed from the dorsal-fin type structure, new shoots could occasionally emerge from the base of the dorsal-fin type structure.

Even though PaLEC1-GFP protein could not be detected in T0 35S:PaLEC1-GFP #2 transgenic plants, some of the germinated T1 seeds also developed into structures that were similar to those observed in 35S:PaLEC1-GFP #1 line 80–90 DAS (Supplementary Figure S3). Approximately 35.2% of protocorms grew and became normal seedlings as seen for wild-type

FIGURE 1 | Selection of T1 35:PaLEC1-GFP #1 transgenic plants on hygromycin plates. (A) Wild-type protocorms failed to develop on hygromycin-containing medium. (B) 35:PaLEC1-GFP #1 protocorms continued to grow and developed into seedlings 5 months after sowing. MAS, month after sowing. Black scale bar = 100 mm. Red scale bar = 0.5 cm.

fin-like structures. MAS, month after sowing; DAS, day after sowing; Red scale bar = 500 mm. Blue scale bar = 1 mm.

protocorms (Table 1). Around 56.3% protocorms developed into multiple-meristem type structures and 8.5% protocorms developed into callus-like structures. No dorsal-fin type structure was observed in the T1 35S:PaLEC1-GFP #1 line. Quantitative RT-PCR and western blotting were conducted to validate and re-evaluate the expression of PaLEC1-GFP mRNA and its encoded protein in 35S:PaLEC1-GFP #1 and 35S:PaLEC1- GFP #2. As compared to 35S:PaLEC1-GFP #1 seedlings, 35S: PaLEC1-GFP #2 had much less PaLEC1-GFP mRNA than 35S: PaLEC1-GFP #1 line (Figure 3). Because GFP monoclonal antibody may be less sensitive than polyclonal antibodies in detecting PaLEC1-GFP protein, PaLEC1 polyclonal antibodies


TABLE 1 | Different types of structures were formed from germinating protocorms of 35S:PaLEC1-GFP transgenic lines. Number of the represented type (number) and percentage of the represented type (%) of structures are shown.

were generated. Similar to hygromycin-resistant T0 35S:PaLEC1- GFP #1 parent, PaLEC1-GFP protein was expressed in high abundance in callus-like, multiple-meristem, and dorsal-fin types of structures germinated from T1 seeds (Figure 3). Using the PaLEC1 polyclonal antibodies, PaLEC1-GFP protein was also detected in callus-like tissues but was barely detectable in multiple-meristem-like structures and wild-type-like seedlings of 35S:PaLEC1-GFP #2 transgenic line (Figure 3), indicating PaLEC1-GFP protein was expressed but accumulated in very low abundance. We noticed that PaLEC1-induced callus tissues had a reduced amount of ribulose bisphosphate carboxylase large subunit (RbcL) protein (Figure 3), that is consistent with the pale-yellow appearance of PaLEC1-induced callus tissues.

To gain further structural details of the PaLEC1-induced structures, histological section and histochemical staining were conducted. As described previously (Yeung, 2017), shoot apical meristem was initiated from the apical end of a protocorm in the wild-type plants (Figure 4). In the 35S:PaLEC1-GFP transgenic plants, active cell division appeared to take place on the outer cell layer (epidermal cells) of developing protocorms (early stage, Figure 4). Protocorms then grew into friable and unstructured cellular masses with small cells showing dense cytoplasm on the surface (late stage, Figure 4). As for multiple meristem-type structures, the dividing cells on the surface of developing protocorms gradually organized into the shoot apical meristems at the early stage (Figure 4). As development proceeded, leaf primordia started to emerge from the organized shoot apical meristems (late stage, Figure 4) and multiple shoots were formed.

#### Explants of Transgenic Plants Overexpressing PaLEC1-GPF Induce Tissues That Are Dissimilar to Protocorm-Like Bodies

Overexpression of LEC1 gene is known to be a potent inducer of somatic embryos (Lotan et al., 1998; Mu et al., 2008; Horstman et al., 2017a). We were therefore interested in asking what types of somatic embryonic structures could be induced by PaLEC1 in P. equestris. To do so, protocorms were used as the explants for induction of somatic embryonic tissues. As described previously (Zhao et al., 2008; Mahendran and Bai, 2012; Fang et al., 2016), cutting protocorms of wild-type plants induced green PLBs (Figure 5). Cutting protocorms of T1 35S:PaLEC1-GFP #1 transgenic line, on the other hand, induced callus-like tissues and pale yellow-green protrusions (Figure 5). The appearance of pale yellow-green color is consistent with a previous report stating that overexpressing LEC1 downregulates light responsive genes in Arabidopsis (Junker et al., 2012). Protrusions and callus-like tissues continued to grow and develop into unorganized thick and flat tissues, which then eventually grew into multiple-meristem type structures that were capable of producing multiple shoots (Figure 5). Using protocorms of T1 35S:PaLEC1-GFP #2 line as explants also induced callus-like and multiple-meristem type structures (Figure 5). Taken together, we conclude that PaLEC1-induced tissues have the ability to establish new shoot apical meristems for shoot regeneration.

To evaluate whether PaLEC1-GFP protein was regulated post-transcriptionally as somatic embryonic calli programmed to initiate shoot formation, PaLEC1-GFP protein was monitored by western blotting. As shown in Supplementary Figure S4, PaLEC1-GFP protein remained stable throughout the process.

#### PaLEC1 Is Sufficient to Activate Embryonic Marker Genes in Seedlings of PaLEC1-GPF Transgenic Plants

To evaluate the embryonic identity of PaLEC1-GFP overexpressing somatic tissues, expression of previously identified embryonic-specific genes (Fang et al., 2016) were examined in T1 35S:PaLEC1-GFP #1 protocorm-derived young seedling tissues. Expression of the embryonic TFs ABSCISIC ACID INSENSITITVE3-like1 (PeABI3L1), PeABI3L2, BABY BOOM (PeBBM), FUSCA3 (PeFUS3), and WRINKLED1 (PeWRI1) and seed storage proteins, 7S GLOBULIN1 (Pe7S-1), Pe7S-2, OLEOCIN1 (PeOLE1), and PeOLE2 genes were confirmed to be preferentially expressed in ovary tissues collected at 130 days after pollination (DAP) of P. equestris (Figure 6). As expected, overexpression of PaLEC1-GFP caused significant induction of embryonic TFs PeFUS3 and PeABI3L1, and seed storage proteins, Pe7S-1, Pe7S-2, and PeOLE2 genes (Figure 6). Overexpression of PaLEC1-GFP protein only led to mild elevation of PeABI3L2 mRNA and did not affect expression of PeBBM, PeWRI1, and PeOLE1 genes (Figure 6), suggesting that PaLEC1 activates only part of the embryonic program. Similar expression patterns were also confirmed in independently collected young seedlings and callus tissues except PeABI3L2 whose expression was not induced in

FIGURE 3 | Expression of PaLEC1-GFP protein in 35:PaLEC1-GFP transgenic lines. (A) Quantitative RT-PCR analysis of expression of PaLEC1-GFP in wild-type (wt) and two 35:PaLEC1-GFP transgenic lines. (B) Western blotting showing expression of PaLEC1-GFP and PaRbcL proteins in callus-type, multiple-meristem type, and dorsal fin-like embryonic tissues of 35:PaLEC1-GFP#1 plants. (C) Western blotting showing expression of PaLEC1-GFP and PaRbcL proteins in callus-type, embryonic tissues of 35:PaLEC1-GFP#2 plants. Blots stained with Ponceau S staining were used to visualize total loaded protein. wt, wild-type plants; C, callus-like tissues; MM, multiple meristem-like structures; DF, dorsal fin-like structures; N, normal wt-like tissues.

PaLEC1-induced callus tissues (Supplementary Figures S5A, B). It is likely that the moderate level of PaLEC1-GFP (Supplementary Figure S5B) in callus was not enough to induce PeABI3L2. Activation of PeFUS3 was also confirmed in leaf tissues of five independent T0 35S:PaLEC1-GFP transgenic plants (Supplementary Figure S6).

FIGURE 4 | Histological examination of embryonic tissues of 35:PaLEC1-GFP#1 transgenic plants. (A) Light micrograph of 10 mm longitudinal section of developing wild-type protocorm stained with hematoxylin. (B) Light micrograph of a 10 mm section of the early and late stages of callus-like tissues derived from protocorms of the T1 35:PaLEC1-GFP#1 seeds. (C) Light micrograph of a 10 mm section of the early and late stages of multiple meristem-like tissues derived from protocorms of the T1 35:PaLEC1-GFP#1 seeds. Shoot apical meristem (SAM) with the emerging leaf primordia are marked by a white asterisk. For panel (B and C), low and high magnification of structures marked with red and black rectangles are shown. Red scale bar = 1 mm. Black scale bar = 100 mm.

#### PaLEC1 Is Predominantly Localized in the Nucleus and Is Associated With Target Promoters Containing the CCAAT Element

Nuclear Transcription Factor-Y (NF-Y) is an evolutionarily conserved, DNA-binding, trimeric protein complex composed of NF-YA, NF-YB, and NF-YC subunits (Dolfini et al., 2012). LEC1 encodes an atypical subunit of the (NF-YB) CCAATbinding TF. To validate subcellular localization of PaLEC1, confocal microscopy was used to examine the localization of PaLEC1-GFP protein in transgenic plants. As predicted, PaLEC1-GFP was predominantly localized in the nuclei in the cells of the 35S:PaLEC1-GFP#1 line (Figure 7). The nuclear localized GFP signal was detected in all of the cells investigated in the 35S:PaLEC1-GFP#1 line. PaLEC1-GFP protein, however, was only detected in a few cells in callus tissues and was therefore much less abundant in 35S:PaLEC1-GFP#2 transgenic line (Figure 7), supporting the low expression level shown by western blot. Hence, PaLEC1 protein is a nuclear protein.

It has been reported that LEC1 regulates many embryonicspecific TFs and storage proteins by directly binding to their promoters through the CCAAT DNA element (Lee et al., 2003; Gnesutta et al., 2017; Pelletier et al., 2017). To investigate whether PaLEC1 activates its targets in the same manner, we retrieved the 1 kb DNA sequences upstream of the start codon of PeFUS3, PeABI3L1, PeABI3L2, PeBBM, PeWRI1, Pe7S-1, Pe7S-2, PeOLE1, and PeOEL2 (Cai et al., 2015) and surveyed the presence of the CCAAT element. Because of a large gap within the promoter regions of PeABI3L2, it was omitted from further study. Intriguingly, two to five CCAAT DNA elements were found within the 1 kb promoter sequences of PeFUS3, PeABI3L1, Pe7S-1, Pe7S-2, and PeOLE2 (Supplementary Figure S7) whose expression was induced by overexpression of PaLEC1 (Figure 6). None or only one CCAAT element was present in promoters of PeBBM, PeWRI1, and PeOLE1 (Supplementary Figure S7) whose expression was not induced in the PaLEC1 overexpressor (Figure 6). The correlation between the number of CCAAT elements and activation capability of PaLEC1-GFP suggests that PaLEC1-GFP protein may activate its targets in a CCAAT DNA element dosage-dependent manner.

To test whether PaLEC1-GFP binds the CCAAT element directly, ChIP using PaLEC1 antibody followed by PCR analysis was conducted. Because PeFUS3 and Pe7S-2 mRNAs were significantly upregulated by PaLEC1 and tandem CCAAT elements are present in their promoters (Figure 8 and Supplementary Figure S8), PeFUS3 and Pe7S-2 were chosen for further ChIP analysis. Histone H3 antibody was used as a positive control and no antibody was included as the negative control. We found that PaLEC1-GFP protein interacted with the PeFUS3 promoter region containing the CCAAT element but did not interact with PeBBM promoter that lacks the CCAAT element (Figure 8). Similarly, PaLEC1-GFP was found to interact with the Pe7S-2 promoter carrying the CCAAT element but not to the PeBBM promoter lacking the element (Figure 8). These results were confirmed in three independent biological samples. Taken together, it is very likely that PaLEC1 activates PeFUS3 and Pe7S-2 through binding to the CCAAT element-containing promoters.

#### DISCUSSION

#### PaLEC1 Induces Diverse Somatic Embryonic Structures

Our recent comparative transcriptome study provided evidence that PLB regeneration does not follow the somatic embryogenesis program (Fang et al., 2016) and raises a

question about the morphology and anatomy of orchid somatic embryos. In this study, overexpression of PaLEC1 was utilized to induce the somatic embryonic program in P. equestris. Similar to PLB induction, active cell division is a prerequisite for PaLEC1 induced somatic embryonic structures (Figures 4B, C). For PLB development, the polarized growth of proliferating cells gives rise to small protuberances with a distinct gradient of cell size, with the smaller cells occupying the future shoot pole and larger and vacuolated cells forming the base of the structure (Lee et al., 2013). For PaLEC1-induced somatic embryonic tissues, on the other hand, diverse types of somatic embryonic structures were formed from protocorms or explants of the two 35S:PaLEC1- GFP transgenic lines regardless of the large difference in the amounts of PaLEC1-GFP protein accumulated. The most common type (~50% in both transgenic lines) was tissue clusters containing multiple shoot apical meristems (multiplemeristem type). For this type of regeneration, multiple meristems were initiated directly from developing protocorms and establishment of unorganized callus tissue did not seem to be a prerequisite. It is important to note that this type of regeneration is very different from direct shoot regeneration during which leaf primordia are generated directly (Košir et al., 2004; Wu and Chen, 2008; Yam and Arditti, 2017). In addition to producing multiple meristems for shoot regeneration, new shoots could also be produced indirectly from callus-type tissue (Supplementary Figure S2). The frequency of obtaining the callus-type tissue was higher in the strong PaLEC1-GFP overexpressors than in the weak ones (Table 1), indicating formation of the embryonic callus may be PaLEC1 dosage-dependent. Dosage-dependent induction of somatic embryogenesis has also been reported for Arabidopsis BBM and PLT2 (Horstman et al., 2017b). Intriguingly, the fin-like structure was only observed in the

FIGURE 7 | Nuclear localization of PaLEC1-GFP. (A) No GFP signal was detected in wild-type plants. (B) PaLEC1-GFP was detected in nuclei of cells in roots of 35:PaLEC1-GFP#1 transgenic line. Nuclear localization of PaLEC1-GFP was confirmed in multiple (more than three) independent seedlings. (C) PaLEC1-GFP was detected in very few nuclei within callus cells of 35:PaLEC1-GFP#2 transgenic line. This sparsely distribution of nuclear PaLEC1-GFP signal was confirmed in multiple (more than three) protocorm-derived callus tissues. Photos with lower magnification are shown in the upper panel in panels (B and C). Photos with higher magnification are shown in the bottom panel in panels (B and C). GFP, photographs taken in the GFP channel. RFP, photographs taken in the RFP channel. DIC, differential interference contrast images. Merge, differential interference contrast images of cells superimposed with PaLEC1 (GFP) and chloroplast (RFP) channels. Scale bar = 20 mm.

strong overexpressor (PaLEC1-GFP #1) and normal protocormderived seedlings were only observed in the weak overexpressor (PaLEC1-GFP #2). Taken together these results indicate that PaLEC1-induced embryonic tissues were morphologically distinct from PLBs. Regardless of the type of structures induced by PaLEC1, all embryonic structures documented here are totipotent and capable of generating clonal orchid seedlings.

elements (red rectangles) in 1 kb upstream of the start codon of PeFUS3 and Pe7S-2 genomic DNAs. No CCAAT element was found in 1 Kb promoter region of PeBBM. Primers used for PCR reactions were shown as black arrows. (B) PCR of chromatin immunoprecipitation isolated DNA fragments to confirm interaction between PaLEC1-GFP protein and the CCAAT-containing promoters of the indicated genes. C, total cell lysate; "-", no antibody control; PaLEC1, PaLEC1 antibody precipitated chromatin; H3, histone H3 antibody precipitated chromatin. The amplified PCR fragments were verified by DNA sequencing. The same result was validated in two other independent biological samples (Supplementary Figure S8).

### PaLEC1 Activates Part of the Embryonic Program

The embryonic identity of PaLEC1-induced embryonic tissues was verified by expression of embryonic TFs, PeFUS3, PeABI3L1, PeABI3L2, as well as seed storage proteins Pe7S-1, and Pe7S-2. In Arabidopsis, FUS3, ABI3, and WRI1 are reported to be the direct targets of LEC1 (Pelletier et al., 2017). However, overexpression of PaLEC1 failed to induce PeWRI1. LEC1 is reported to directly activate WRI1 or indirectly acts through upregulation of FUS3 and LEC2 (Baud et al., 2007; Mu et al., 2008; Shen et al., 2010; Wang and Perry, 2013; Pelletier et al., 2017). Despite a high level of PaLEC1 protein and elevation of PeFUS3 mRNA in 35S: PaLEC1-GFP#1 transgenic line, we did not observe induction of PeWRI1, suggesting a PaLEC1-independent pathway is required to induce PeWRI1 in Phalaenopsis orchids. Similar to Arabidopsis on the other hand, PaLEC1 cannot activate PeBBM and only induces a subset of storage proteins (West et al., 1994; Pelletier et al., 2017). In Arabidopsis, BBM acts upstream of the LEC1-FUS3-ABI3-LEC2 network to induce the somatic embryogenesis program (Horstman et al., 2017b). Whether a similar regulatory module exists in Phalaenopsis orchids remains to be tested. Taken together, despite being a potent inducer of somatic embryogenesis in Phalaenopsis orchids, the PaLEC1-dependent regulatory network is similar but not identical to that reported in Arabidopsis.

LEC1 belongs to a distinct subgroup of the NF-YB family that is diverged from non-LEC1-type NF-YB genes in non-seed vascular plants (Xie et al., 2008). It has been proposed that LEC1-type NF-YB is recruited into a seed-specific regulatory network during evolution. Similar to Arabidopsis LEC1-type NF-YB TFs (Calvenzani et al., 2012; Laloum et al., 2013), PaLEC1 also binds to the CCAAT-containing promoters of its downstream genes, PeFUS3 and Pe7S-2. Therefore, PaLEC1 is very likely to regulate its targets by directly acting through binding of the CCAAT DNA element. Even though one CCAAT element was present in the 1 kb promoters of PeWRI1 and PeOLE1 genes (Supplementary Figure S4), PeWRI1 and PeOLE1 mRNAs were not activated by overexpression of PaLEC1-GFP protein, indicating the CCAAT DNA element may not be sufficient for PaLEC1 binding. This notion is supported by studies of Arabidopsis and soybean that the G-box (CACGTG), abscisic acid response element (ABRE)-like (C/ G/T)ACGTG(G/T)(A/C), and RY (CATGCA) motifs are significantly overrepresented in LEC1 target genes (Pelletier et al., 2017). The involvement of G-box, ABRE-like, and RY motifs in PaLEC1 functions remain to be determined.

In conclusion, our study of PaLEC1-induced somatic embryogenesis has demonstrated that somatic tissues of Phalaenopsis orchids do not resemble PLB, a result that is consistent with our previous comparative transcriptome studies. In addition to confirming the role of PaLEC1 in activating the somatic embryonic program, we obtained evidence supporting a direct role for PaLEC1 in controlling its targets and revealed the potential of PaLEC1-induce somatic embryonic tissues for orchid propagation. Conservation of the developmental processes and gene regulatory networks controlled by PaLEC1 is consistent with the notion that LEC1 is a major regulator of embryogenesis.

#### REFERENCES


Arditti, J. (1992). Fundamentals of orchid biology (New York: John Wiley & Sons). Arditti, J. (2009). Micropropagation of Orchid (Oxford, UK: Wiley-Blackwell).


#### DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/ Supplementary Material.

### AUTHOR CONTRIBUTIONS

S-CF designed the experiments, coordinated the studies, and wrote the manuscript. J-CC maintained and characterized transgenic orchid plants, and developed the ChIP protocol. J-CC and H-YL carried out the molecular biology experiments. C-GT developed and coordinated the orchid transformation process. All the authors have read and approved the final manuscript.

#### FUNDING

This work was supported by the Innovative Translational Agricultural Research Grants (to S-CF); and in part by a grant (to S-CF) from the Biotechnology Center in Southern Taiwan, Academia Sinica.

#### ACKNOWLEDGMENTS

We thank Dr. Yao-Cheng Linfor his assistancewith bioinformatics, Ms. Miao-Ju Wei for her technical support during the initial work on transgenic orchid plants, the Confocal Microscopy Core, and Greenhouse Core Facilities for their supporting services, and Ms. Miranda Loney for English editing.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2019. 01594/full#supplementary-material


Conflict of Interest: 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 Chen, Tong, Lin and Fang. 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.

# The Role of Non-Mycorrhizal Fungi in Germination of the Mycoheterotrophic Orchid Pogoniopsis schenckii Cogn.

*Laís Soêmis Sisti1\*, Denisele Neuza Aline Flores-Borges1, Sara Adrián López de Andrade2, Samantha Koehler3, Maria Letícia Bonatelli4 and Juliana Lischka Sampaio Mayer1\**

1 Laboratory of Plant Anatomy, Department of Plant Biology, Institute of Biology, State University of Campinas, Campinas, Brazil, 2 Laboratory of Plant Molecular Physiology, Department of Plant Biology, Institute of Biology, State University of Campinas, Campinas, Brazil, 3 Laboratory of Plant Taxonomy, Department of Plant Biology, Institute of Biology, State University of Campinas, Campinas, Brazil, 4 Laboratory of Genetics of Microorganisms, Department of Genetics, College of Agriculture "Luiz de Queiroz," University of São Paulo, Piracicaba, Brazil

#### Edited by:

Jen-Tsung Chen, National University of Kaohsiung, Taiwan

#### Reviewed by:

Kei Hiruma, Nara Institute of Science and Technology (NAIST), Japan Pablo Delgado-Sánchez, Universidad Autónoma de San Luis Potosí, Mexico

#### \*Correspondence:

Laís Soêmis Sisti laisoemis@hotmail.com Juliana Lischka Sampaio Mayer mjimayer@yahoo.com.br

#### Specialty section:

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

Received: 25 July 2019 Accepted: 12 November 2019 Published: 29 November 2019

#### Citation:

Sisti LS, Flores-Borges DNA, Andrade SALd, Koehler S, Bonatelli ML and Mayer JLS (2019) The Role of Non-Mycorrhizal Fungi in Germination of the Mycoheterotrophic Orchid Pogoniopsis schenckii Cogn. Front. Plant Sci. 10:1589. doi: 10.3389/fpls.2019.01589

Endophytic fungi are those that inhabit within organs and tissues without causing damage, while mycorrhizal fungi develop hyphal complexes called pelotons within cortical cells of orchid roots. Although abundant and frequent in all plant organs, the role of endophytic fungi has been neglected in relation to orchid's early development. Pogoniopsis schenckii Cogn. is an aclorophyllated and mycoheterotrophic (MH) orchid. This study aimed at i) investigating the endophytic fungal community in organs of P. schenckii and its mycorrhizal fungi associated; ii) evaluating the ability of isolated fungus in the in vitro germination of the seeds of the species, and iii) describing the development of P. schenckii protocorm, analyzing the ultrastructure of the infected cells. Six genera of fungi were isolated and identified through the partial sequencing of the internal transcribed spacer region, all belonging to the phylum Ascomycota. Also, Tulasnellaceae was identified through uncultured technique as potentially mycorrhizal in this MH orchid. Some isolates of the genera Trichoderma, Fusarium, and especially Clonostachys presented germinative potential on P. schenckii seeds, causing rupture of the external tegument. The protocorms showed complete absence of peloton formation, but fungal hyphae were clearly observed within living cells. This is the first report of germination of a MH and aclorophyllated orchid species stimulated by the presence of non-mycorrhizal endophytic fungi isolated from fruits and roots of the same species.

Keywords: endophytic fungi, symbiotic germination, plant anatomy, aclorophyllated plant, ultrastructure, protocorm, Orchidaceae, Tulasnellaceae

### INTRODUCTION

Mycoheterotrophic (MH) plants evolved independently in several locations and represent one of the most extreme forms of mycorrhizal dependence (Leake, 1994). These plants remain aclorophyllated throughout their life cycle and are totally dependent on their mycorrhizal partners for their survival (Peterson et al., 2004). There are approximately 235 MH species within the Orchidaceae family (Merckx, 2013). One of the main characteristics of mycorrhizal orchids is the formation of a hyphal complex, also called peloton, which develops within the parenchyma cells of the roots and rhizomes

1 **80** of such plants (Peterson et al., 2004; Rasmussen and Rasmussen 2009). During the germination in *Orchidaceae*, the embryo swells and promotes the seed coat rupture, thus forming a cone-shaped structure (Arditti, 1967), also known as protocorm (Bernard, 1909, qt. in Leroux et al., 1997). This structure is considered an intermediate phase between the embryo and the seedling (Leroux et al., 1997). After seed coating rupture, absorbent trichomes appear, the protocorm increases in size, the apical meristem is installed, and the first leaves are formed, all that followed by the development of adventitious root (Arditti, 1967; Harrison, 1977).

Under natural conditions, even photosynthetic orchids undergo a phase of mycorrhizal fungi dependence at their early stages of growth and development, not being able to germinate in their absence (McKendrick et al., 2000; Leake et al., 2004). Some species of green orchids can be germinated *in vitro,* in an asymbiotic way, if soluble carbohydrates and other organic compounds are supplied (Rasmussen and Rasmussen, 2009). Otherwise, it is necessary to establish a symbiotic association with an appropriate fungal partner and a complex carbohydrate source (Zettler, 1997; Peterson et al., 2004). During the symbiotic germination, anatomical analyzes allows the observation of pelotons in protocorm cells of MH species, and transmission electron microscopy is important to investigate and confirm that plant cells containing hyphae remain alive and with intact organelles, ensuring that the association with the fungus is not harmful to the plant. In addition, the latter allows verifying if the hyphae have clamp connections, dolipores, and parenthesomes, typical characteristics of Basidiomycota. As described by Roy et al. (2009), in MH orchid *Epipogium aphyllum*, the authors observed, in electron micrographs, the presence of clamp connections, dolipores, and parenthesomes, and confirmed the identification of *Inocybe*, a basidiomycetous symbiont, by means of molecular techniques.

Orchids can relate to a great diversity of fungal taxa and combine it with different nutritional strategies (Rasmussen, 2002; Peterson et al., 2004). Photosynthetic orchid species are generally associated with rhizoctonia-like fungi group belonging to the families Ceratobasidiaceae, Tulasnellaceae, and Sebacinaceae (Otero et al., 2002; Rasmussen, 2002). However, the heterotrophic species of temperate regions often associate with ectomycorrhizal basidiomycetes, as those of the families Thelephoraceae and Russulaceae (Julou et al., 2005; Roy et al., 2009). In addition, MH species from tropical regions appear to exhibit greater diversity in their symbiotic associations and less degree of specificity when compared with species from temperate regions, associating with predominantly saprophytic fungi (Martos et al., 2009; Selosse et al., 2010). It has been shown in a study by Yagame et al. (2007), which revealed the association of the Asiatic orchid *Epipogium roseum* with saprophytic fungi belonging to *Psathyrella* or *Coprinus* in Coprinaceae, isolated from root and rhizome. Even though most MH species belong to tropical regions, most of the studies include species belonging to temperate regions as models (Selosse et al., 2010).

Although mycorrhizal associations with orchids are predominantly related to the fungi of the phylum Basidiomycota, groups of Ascomycota have already been described forming such association. In a previous work by Selosse et al. (2004), it was possible to verify the formation of pelotons by fungi belonging to the order Pezizales, previously identified as ascomycetes, present in root cells of the green *Epipactis microphylla*, analyzed by light and transmission electron microscopy. According to the authors, an endophyte was found forming this association with the orchid, and it belongs to a fungal group known as ectomycorrhizal in tree roots. Endophytic microorganisms, unlike pathogens, do not cause damage to the host and are known for establishing beneficial or neutral associations with plants (Petrini, 1991). They inhabit the internal part of organs and tissues of plants, and can confer protection against pathogens, or cause the production of plant growth factors (Azevedo, 1998; Ma et al., 2015). Although their role in orchids is rarely addressed, nonmycorrhizal endophytic fungi can be found in all organs of the plant, encompassing more than 110 genera, predominantly belonging to the phylum *Ascomycota* (Ma et al., 2015).

The genus *Pogoniopsis* belongs to the subfamily Vanilloideae, Pogoniae tribe (Cameron, 2009). It is composed only of two species, *Pogoniopsis schenckii* Cogn. and *Pogoniopsis nidusavis* Rchb.f. & Warm., and both are MH. Both develop under organic matter in dense tropical and subtropical forests, are aclorophyllated, and have a pale-yellow coloration. The plants of the genus have short and fasciculate roots, with bracts that cover the floral stem. *P. schenckii* is a poorly-known species, with collections described in several states of Brazil (Bittencourt and de Gasper, 2016). Despite being widely distributed, it is considered a rare species, and it is included in the Red List of the Threatened Flora of Paraná State (1995), and classified as a vulnerable species by the Red List of Threatened Flora of São Paulo State, both in Brazil (2004) (CNCFlora, 2012). In addition, in a recent study, projections on future climate changes indicated that with global warming, the species *P. schenckii* could have its ideal niche reduced by up to 30% of its current extent (Kolanowska et al., 2017), severely compromising its survival. Therefore, studies to understand its germination and biology, as well as strategies to enable its establishment *in vitro,* can be of great value for the conservation of populations of this species.

In a previous analysis, it was possible to observe the presence of fungal hyphae inside the mature fruits of *P. schenckii* (data not shown), and this finding promoted the hypothesis that these fungi could be related to the germination of the species, in addition to those present in the roots. Moreover, there is lack of studies on the endophytic fungi present in other organs of MH orchids and not only in the roots, especially when dealing with their function for germination of these plants (Ma et al., 2015). Therefore, other organs of the species (floral stem and fruits) were also included in the studies of isolation and identification of fungi in order to test such hypothesis. This work was generally aimed at i) investigating the endophytic fungal community of *P. schenckii* and the noncultivable mycorrhizal fungi present in their roots; ii) evaluating the role of these fungi in germination of the seeds of *P. schenckii*, and iii) monitoring the development of the protocorm after germination.

#### MATERIAL AND METHODS

#### Vegetal Sampling Collection

Roots, fruits, and floral stems samples of *P. schenckii* were collected from three populations located at three trails: *Trilha do Poço do*  *Pito* (TPP), *Trilha do Pirapitinga* (TP), and *Trilha do Garcez* (TG) (**Figure 1**). All trails belong to the Santa Virgínia Nucleus (SVN), located in the Serra do Mar State Park, in the outskirts of the municipalities of São Luiz do Paraitinga and Natividade da Serra, São Paulo State, Brazil. Access to the study area and collection permission was authorized by COTEC through permission (291/2018 D83/2018 PM) and São Paulo State Environment Secretariat (SMA) permission 260108 – 005.510/2014.

#### Isolation of Fungi and Analysis of Frequency of Isolation

In order to investigate the community of endophytic and mycorrhizal fungi of *P. schenckii*, roots, floral stems, and fruit portions of the three populations were isolated according to the methodology described by Araújo et al. (2001). Healthy plant samples were washed in running water, and then washed for 1 min with 70% ethanol solution, followed by a 3 or 5 min commercial sodium hypochlorite (2% active chlorine) washing, and a 30 s 70% alcohol washing. Finally, two washes were performed in autoclaved distilled water, and an aliquot of the water from the last wash was plated in commercial PDA (potato dextrose agar) medium to verify the efficacy of surface disinfection. After, fragments of about 5 mm were sectioned and inoculated in PDA with addition of ampicillin (75 μg/ml) and left to grow at room temperature in the dark for 5 days. After the incubation period, the grown hyphae were striated in WA (water agar: 7 g/L agar) and kept at room temperature and in the dark. After 3 days, fragments of the WA containing the grown fungal mycelium were inoculated at the center of PDA plates and incubated for 7 days under the same conditions previously described. The isolated fungi were placed in mineral oil and stored.

In order to calculate the frequency of isolation (FI) of fungi, the calculation described in Araújo et al. (2014) was performed, which consists in evaluating the number of plant fragments that presented fungal growth in relation to the total number of fragments sampled.

### Total Deoxyribonucleic Acid Extraction and Identification of Isolated Fungi

For identification of the isolated fungi, the methodology of total fungal DNA extraction described in Raeder and Broda (1985), with modifications by Araújo et al. (2014), was used. The fungi were inoculated in PDA and grown for approximately 10 days at room temperature. The grown fungal mycelium was scraped from the surface of the culture medium, crushed in liquid nitrogen, and approximately 200 mg of the crushed mycelium was transferred to a 2 ml tube. Then, 1 ml of extraction buffer (1% SDS, 25 mM EDTA, 250 mM NaCl, and 200 mM Tris-HCl [pH 8.0]) was added to the tube, vortexed, and incubated at 65°C for 20 min. The solution was centrifuged at 10,000 xg for 10 min at 4°C and the supernatant was transferred to a new tube containing 800 μl of phenol. The suspension was homogenized by inversion and centrifuged at 10,000 x*g* for 10 min at 4°C. The supernatant was transferred to a new tube and 400 μl phenol and 400 μl chloroform (1:1) were added. The solution was homogenized by inversion and centrifuged at 10,000 x*g* for 10 min at 4°C. The upper phase was transferred to a new tube containing 800 μl of chloroform, homogenized by inversion, and centrifuged at 10,000 xg for 10 min at 4°C. The supernatant was transferred to a new tube and 450 μl of isopropanol was added. The solution was homogenized by inversion, incubated for 5 min at room temperature and centrifuged at 10,000 x*g* for 5 min at 4°C. The supernatant was discarded and the precipitate (DNA) was washed with 500 μl of 80% ethanol and centrifuged at 10,000 xg for 5 min at 4° C. Then, the ethanol was discarded, the DNA was dried at 37°C for 30 min and finally resuspended in 40 μl of deionized distilled autoclaved water and kept in a refrigerator overnight. The DNA suspension was quantified in a spectrophotometer (NanoDrop 2000c, Thermo Scientific) and stored in a freezer at −20°C.

For amplification of the fragment of interest, amplification of the ITS1-5,8S-ITS2 region was performed using the primers ITS1 (5'-TCCGTACCTCAACCTGCGG-3') and ITS4

FIGURE 1 | Representation of the Serra do Mar State Park (A) (in green) and location of the trials (triangles) where Pogoniopsis schenckii (B) individuals were collected.

(5'-TCCTCCGCTTATTGATATGC-3') (White et al., 1990). Polymerase chain reaction (PCR) was performed with 3.7 mM MgCl2, 1 mM of each deoxynucleoside triphosphate, 0.4 μM of each primer, 2.5 U of Taq DNA polymerase, 1 X buffer, 5 ng DNA in a final volume of 50 μl. The ITS1-5.8S-ITS2 region of recombinant DNA (rDNA) was amplified by 24 cycles of PCR reaction with initial denaturation of 4 min at 94°C, followed by 24 cycles of 30 s at 94°C, 1 min at 55°C; 30 s at 72°C, and a final extension of 7 min at 72°C in thermal cycler (Peltier Thermal Cycler 200, MJ Research). The reaction product was analyzed by using agarose gel (1% w/v) along with the 1 Kb/100 pb DNA molecular weight marker (Invitrogen). Amplified (400 pb) rDNA ITS1-5, 8S-ITS2 region fragments were sequenced at the company Macrogen (Seoul, South Korea) and the sequences obtained were used for phylogenetic identification on the NCBI (National Center for Biotechnology Information—www.ncbi. nlm.nih.gov) database through Blastn.

A phylogenetic tree was built using the MEGA 10.0 program (Kumar et al., 2018). The alignment was made with MUSCLE default settings. The evolutionary history was inferred according to the neighbor-joining method (Saitou and Nei, 1987), considering bootstrap test (1,000 replicates) (Felsenstein, 1985), and the evolutionary distances were computed using the Jukes-Cantor method (Jukes and Cantor, 1969) (**Figure 2**). Sequences were deposited in GenBank under the accession numbers MN611256 to MN611298.

#### Total Root Deoxyribonucleic Acid Extraction and Identification of Basidiomycetes

Roots of *P. schenckii.* were used order to investigate the presence of Basidiomycetes that could not be cultivated. For this purpose, DNA extractions were made from root samples of TPP and TP populations, which were collected and promptly stored at −80°C. The methodology used was described in (Tel-Zur et al., 1999), with modifications. In this procedure, the samples were macerated and placed in a microtube, and then 1.5 ml of sorbitol (0,35 M sorbitol, 100 mM Tris-HCl, 5 mM EDTA) was added. The samples were vortexed, incubated in the refrigerator for 20 min, and centrifuged for 10 min at 10,000 x*g* at 4°C. The supernatant is discarded and 800 μl of cetyl trimethylammonium bromide and 30 μl of sakosyl 30% were added. The samples were then vortexed and incubated at 65°C for 1 h. Then 600 μl of chloroform-isoamyl alcohol was added and the samples were vortexed again and centrifuged for 10 min at 10,000 x*g*. The supernatant was placed in a new microtube with 30 μl of ammonia acetate and 500 μl of isopropanol, and the samples were incubated at −20°C overnight. In the next day, samples were

Saccharomyces cerevisiae (MN545451.5) was used as the out-group. Bootstrap values (n = 1,000 replicates) are shown at the intersections.

centrifuged at 15,000 x*g* for 30 min at 4°C. The supernatant was discarded, 500 μl of 70% ethanol were added and centrifuged for 10 min at 15,000 x*g* at room temperature. The supernatant was discarded again, the samples were incubated at 65°C for 30 min, 50 μl of deionized distilled water is added, and the samples were incubated at 37°C for 2 h. The extraction products were analyzed with 1% agarose gel with the 1 kb molecular weight marker.

The amplification of the region of interest was performed through the primers ITS1-OF and ITS4-OF (Taylor and McCormick, 2008) and following the methodology indicated by the authors. An aliquot of the product of this first reaction was used in a second reaction for the primers ITS4-tul and ITS1, according to their recommendation. The amplified samples were sequenced at the company Macrogen (South Korea) and the sequences obtained were used for phylogenetic identification on the NCBI (National Center for Biotechnology Information www.ncbi.nlm.nih.gov) database through Blastn.

#### Symbiotic, Asymbiotic, and In Situ Germination

For symbiotic germination trials, fruits from two isolated populations of *P. schenckii* were superficially disinfected according to the methodology described in Pereira et al. (1993) and Araújo et al. (2001). After disinfection, the fruits were sectioned and the seeds removed. The seeds were disinfested by immersion in sodium hypochlorite (1% active chlorine) for 7 min and washed three times with autoclaved distilled water. The seeds were inoculated on filter paper fragments arranged on plates containing OMA-oatmeal agar (2 or 4 g/L oat flour and 7 g/L agar [pH 6.0]), as suggested by Zettler (1997). Each plate was inoculated with the mycelium of fungal representatives isolated from root or fruit of *P. schenckii*; the plates were sealed and incubated in the dark at room temperature. For control purposes, plaques with seeds were maintained in the same culture medium without inoculation of fungi. The seeds were transferred monthly to new plates with oat meal agar and analyzed periodically, and the development of the protocorm was classified in phases, according to their size.

For the asymbiotic germination, the same methodology of disinfestation of the seeds described for the symbiotic germination trial was used. The culture medium used was the MS, proposed by Murashige and Skoog (1962), with modifications. The concentration of micro and macro-nutrients was reduced by half and the medium supplemented with 30 g/L sucrose and 0.5 mg/L nicotinic acid, 0.5 mg/L pyridoxine HCl, 0.1 mg/L thiamine HCl, 2.0 mg/L glycine, and 100 mg/L myo-inositol (pH 6.0). The plates were sealed, incubated in the dark in a growth chamber at 27°C, and analyzed periodically.

For the *in situ* germination trial, a methodology proposed by Rasmussen and Whigham (1993) was used. Several mature fruits of a population of *P. schenckii* were placed in small packages made of nylon, whose pores allow the passage of water and nutrients and soil bacteria and hyphae of filamentous fungi. After being properly sealed, the packages were tied to a stake fixed to the ground and buried superficially under litter in the TPP site, near the same population of plants of their origin. The fruits buried were monitored and recovered periodically for seed germination analysis.

### Anatomical, Ultrastructure, and Surface Analysis

Samples of protocorms at different stages, as well as samples of buried field structures, were fixed with 2.5% glutaraldehyde in 3% sodium cacodylate buffer (pH 7.25) for 24 h at 4°C. Postfixation was performed with 1% aqueous osmium tetroxide (OsO4) overnight. The samples were then washed three times in distilled water and dehydrated in a rising ethyl series. After dehydration, the samples were soaked in hydrophilic acrylic resin LR White® Hard Grade (EMS) and polymerized in gelatin capsules in an oven at 60°C for 12 h. The ultrafine sections were contrasted with uranyl acetate (Watson, 1958) and lead citrate (Reynolds, 1963) and examined under a Philips EM 100 transmission electron microscope operated at 60 Kv in the Laboratory of Electron Microscopy of the Institute of Biology of UNICAMP.

For analyzes of the anatomical structures of the protocorms, as well as in the transmission electron microscopy analysis, samples of different sizes were processed, ultrafine sections were stained with 0.05% toluidine blue (Sakai, 1973) in phosphate and citrate buffer pH 4.5, and analyzed under a light microscope. To record the results, the images were captured using an Olympus DP71 video camera coupled to an Olympus BX 51 microscope.

To investigate the surface of the samples, protocorms in different phases and fixed in solution of Karnovsky (Karnovsky, 1965) were used. After fixation, the samples were dehydrated in ethyl series and dried by the CO2 critical point method in a Balzers CPD 030 Critical Point Dryer. The material was then mounted on metal supports and covered with colloidal gold for 220 s with a Bal-Tec SCD 050 Sample Sputter Coater. The analysis and electromicrographic recording were performed using a LEO VP 435 scanning electron microscope at 20 kV, at the Institute of Biology of UNICAMP.

#### RESULTS

#### Isolation and Identification of Fungal Community From Pogoniopsis schenckii Organs

Thirty-three fungi were isolated from the roots, 19 from the fruits, and 16 from the floral stems of *P. schenckii.* The fungal isolates were purified and stored, totalizing 68 fungi from all populations and different plant organs (**Figure 3**). None of the surface disinfestation methodologies generated fungal contamination in the water of the last wash. Forty-two isolates were identified, all belonging to the phylum Ascomycota. Among the fungi isolated from the roots, the genera *Xylaria* and *Coniochaeta* were found as endophytic in *P. schenckii* individuals from TG population, and only the genus *Trichoderma* were isolated from individuals of TP and TPP populations. As for floral stem and fruit portions of TP and TPP individuals, the genera *Fusarium, Clonostachys*, and *Colletotrichum* were reported in both organs. Further details on isolation and identification of the fungal community can be found in **Supplementary Table 1**.

FIGURE 3 | Isolated fungal representatives of different organs of Pogoniopsis schenckii. Fusarium sp. and Colletotrichum sp., respectively, isolated from fruit (A, B). Clonostachys sp., both isolated from floral stem (C, D).

A total of three isolation procedures were carried out at different stages of plant development. Initially, the plant material used for the isolation were root portions of two populations of *P. schenckii*, TG, and TP, through the modified surface disinfestation methodology of Pereira et al. (1993) and Araújo et al. (2001), aiming at increasing the efficiency of disinfestation (5 min in hypochlorite). However, this first trial resulted in the isolation of 15 fungi only from TG population, with isolation frequency (IF) of 12.5%. Therefore, in a second isolation, the authors' methodology was used without modification (3 min in hypochlorite) for isolating fungi from the roots of the plant. In this experiment, FI of 38.8 and 41.6% were obtained for TPP population and TP population, respectively, both located in the SVN, and a total of 18 fungi were isolated from the *P. schenckii* populations. Due to the higher IF obtained in the second experiment, the unmodified methodology was adopted for a third isolation round. In this isolation, fruit and floral stem portions of NSV individuals were used without discriminating the populations, obtaining 100% IF for both organs.

### Identification of Non-Cultured Root Basidiomycetes

Identification of the sequences obtained from the roots of *P. schenckii* with the specific primers ITS4-tul and ITS1 revealed the presence of the fungus from genus *Tulasnella*, belonging to family Tulasnellaceae and Basidiomycota phylum.

### Germination Trials

In a first trial, beginning in April 2016, a total of 18 fungal representatives isolated from the roots and 13 fungi from the fruits of *P. schenckii* were selected according to their morphology to be inoculated in seeds of TP and TPP populations. After about 20 days of plate incubation, it was possible to observe the external seed coat rupture of both populations in the presence of two fungal isolates from the roots of five from the fruits. At 40 days of experiment, two other fungi, one of root and another of fruit, were able to cause the external seed coat rupture of both populations. However, among the total of nine fungi, three isolates of root, and six of fruit, only the isolates F38 and F39 caused visible changes in the development of the protocorms after the external tegument rupture, both fungi of fruits of *P. schenckii*. The experiment exceeded 15 months, presenting minimal changes only in the size of the protocorms, under magnifying glass. Among the fungi that showed to be effective in germination, they comprise the genera *Trichoderma* sp. (F16, F25, and F32), *Fusarium* sp. (F34, F37, and F40) and *Clonostachys* sp. (F38, F39, and F43).

The mature seeds of *P. schenckii* had a rounded shape and they were slightly tapered at one end. Its external tegument is extremely rigid and dark brown in color (**Figure 4A**). At the initial stage of germination, the seed underwent a slight swelling accompanied by the whitening of the external tegument, almost imperceptible changes. After this stage, the testa rupture occurred (**Figures 4A**, **B**), and the protocorm began to increase in size (**Figures 4C–E**), at a very slow growth rate, assuming a slightly elongated shape (**Figure 4F**).

No changes were observed in the seeds during the whole period of incubation of the asymbiotic germination plates in MS medium, which took about 15 months, suggesting the inefficiency of germination in the absence of fungi and under such cultivation conditions. The same occurred with the sowing and seed recovery trials *in situ*, which did not demonstrate efficacy in the germination of seeds of *P. schenckii* either.

#### Anatomical, Ultrastructure, and Surface Analysis

Due to its ability to cause changes in protocorm development after external tegument rupture, different stages of seed development incubated with the isolate F38 (*Clonostachys* sp.) were analyzed by scanning electron microscopy. The seeds have an oval shape and a wavy external tegument. At their initial germination stage, it is possible to observe the proliferation of fungal hyphae exclusively in the funicular region of the seeds (**Figure 5A**). Then, the seeds increase in size and the external tegument ruptures (**Figures 5B–D**). The protocorm is exposed, fully covered by hyphae and the development of tector trichomes begins (**Figures 5E, F**). In this phase, the protocorm continues to increase in size, taking an elongated form, and scarce tector trichomes can be observed

(**Figures 5G**, **H**). The hyphae growth allowed only partial visualization of the vast majority of protocorms, making it difficult to analyze the samples and collect data.

The light microscopic analysis of the seeds with the onset of external tegument rupture allowed observing that the plant cells were alive and intact. However, no evidence was found of the formation of pelotons or even the growth of hyphae within the cells at this early stage of germination. The living cells have numerous vacuoles widely distributed in their interior and a large mass of fungal hyphae was also observed on the surface of the tegument (**Figure 6A**).

It was possible to verify the presence of intact plant cells through the analysis of transmission electron microscopy of protocorms resulting from the trials of symbiotic germination at different stages of development. As observed in light microscopy, the surface of the protocorm is covered by a mass of fungal hyphae, and fragments of the ruptured external tegument were found (**Figures 6B**, **C**). The cells of the observed protocorm

tegument scarification, and massive growth of hyphae on its surface after 20 days in co-culture with the isolate F38 (A). Overview of the periphery of the protocorm covered by hyphae (B) and detail of fragment of the external integument cracked (C). Integral cell of the protocorm (D) and, in detail, cell with evidenced nucleus and hyphae in transverse section (E). Hyphae in longitudinal and transverse section (F) and fungal cell septum detail (G). Mass entanglement of hyphae growing from inside the protocorms after phase 5 (H). hy, hyphae; et, external tegument; nu, nucleus; cw, cell wall; se, septum.

had the nucleus and cell wall intact and fungal structures were found occupying much of the interior of the cells, being several times smaller than the first ones (**Figures 6D**, **E**). The hyphae, surrounded by the membrane of the plant cell (**Figure 6E**), presented several structures with distinct patterns, and some hyphae were filled with dense cytoplasmic content and others with little dense content (**Figure 6F**). It was possible to note the absence of clamp connections, dolipores and parenthesomes in the hyphae found (**Figure 6G**). No formation of pelotonswas observed by the fungal isolate of the genus *Clonostachys*.

After phase 5, the growth of the protocorm was no longer observed, which started being degraded by the fungi. Instead, a cluster of fungal hyphae was found occupying the entire interior of the protocorm and expanding beyond the external tegument of the seeds, forming massive structures (**Figure 6H**). It is believed that at some point in development, protocorm structures were degraded and absorbed by the inoculated fungus, even on plates where protocorms were transferred to new culture media on a monthly basis.

### DISCUSSION

#### Endophytic and Mycorrhizal Community

The community of endophytic cultivable fungi of the MH species *P. schenckii* was characterized for the first time. All the fungi isolated from root, floral stem, and fruit belong to the phylum *Ascomycota* and were grouped in the taxon Sordariomycete*s* (Ma et al., 2015). The genera *Xylaria* and *Coniochaeta*, reported for TG population roots, were not found in the plants of TPP and TP populations, which presented only the presence of the genus *Trichoderma*. It is interesting to note that TG population is miles away from the other two, and there is a road separating them geographically. As for floral stem and fruit, the genera found were the same for the two organs, and the genera *Fusarium*, *Clonostachys*, and *Colletotrichum* were isolated. According to Ma et al. (2015), the diversity of non-mycorrhizal endophytic fungi of orchids would be more related to the locality of the orchid collection than to other factors, and it may vary from population to population to the detriment of the geographical differences between them.

The genera *Xylaria*, *Trichoderma*, and *Fusarium* are nonmycorrhizal endophytes often found in association with orchids (Ma et al., 2015). The taxon *Xylariaceae* has been repeatedly found in *Dendrobium* sp. (Yuan et al., 2009; Chen et al., 2011; Chen et al., 2013), and *Xylaria* sp. has also been reported in the green orchid *Anoectochilus setáceus,* from which it was possible to extract an important antibacterial compound (Ratnaweera et al., 2014). Antimicrobial properties have also been observed in *Trichoderma* sp., associated with Cupressaceae (Mahdieh and Soltani, 2014). The abundance of compounds with pharmacological properties found in orchids may have the production, to some extent, related to the great diversity of fungal metabolites present in their tissues, which would represent a benefit to the plant in this association (Ma et al., 2015). The production of metabolites by the endophytic fungi isolated from *P. schenckii* was not investigated in this work, and it is a point to be explored. To our best knowledge, *Coniochaeta* was reported only once in recent work by Lee and Eom (2017), associated with a native orchids from Korea, *Coniochaeta mutabilis*.

Some plant pathogens may appear as asymptomatic endophytes in orchids, such as *Trichoderma*, related to diseases in cotton (Lutfunnessa and Shamsi, 2011), and *Fusarium* (Pecoraro et al, 2013). Although some species of *Colletotrichum* have already been listed as pathogens in orchids such as *Oncidium flexuosum*, *Bulbophyllum cylindraceum* and *Coelogyne cristata* (Tao et al., 2013), they have also been found in healthy tissues of *Lepanthes* and *Dendrobium* (Bayman et al., 1997; Chen et al., 2011) as in *P. schenckii.* In this study, the isolations were only performed in tissues and organs of healthy plants, without any apparent disease.

Differently from what is expected from tropical MH orchids, *P. schenckii* is associated with *Tulasnella* sp., belonging to the groups frequently found for green orchids from temperate regions (Rasmussen, 2002; Rasmussen et al., 2015). Fungi belonging to *Tulasnellaceae* have also been reported previously associated with two orchids of the genus *Neottia* (Těšitelová et al., 2015). For green orchids from the Atlantic Forest, there were identification of fungi belonging to the orders *Sebacinales* and *Cantharellales,* described in Oliveira et al. (2014), in addition to the genera *Ceratorhiza* and *Ephulorhiza* reported in a conservation study conducted by Pereira et al. (2005). The methodology used for the molecular identification of the DNA extracted from the roots of *P. schenckii* is quite specific for *Tulasnellaceae*, bringing into question the fact that other species of fungi may be related to this orchid besides the one reported in the present study. This analysis represented a brief investigation to verify the presence of this taxon, known as being a rhizoctonia associated with the roots of *P. schenckii,* considering the absence of mycorrhizal fungi in the isolation trials by means of culture.

Some genera of mycorrhizal fungi may not be able to grow in culture medium or be isolated through conventional techniques, which may explain their absence among the isolated fungi of *P. schenckii*. In their work, Zelmer et al. (1996) had reported the difficulty of isolating mycorrhizal fungi in *Cypripedium acaule* and *Malaxis monophyllos* and, despite the presence of pelotons in root tissues, no mycorrhizal fungi were isolated in their work. For *P. schenckii*, the absence of typical pelotons was also confirmed for the root cells and subterranean system of the species (unpublished data—Flores-Borges, D.N.A). Although the presence of degraded pelotons cannot be discarded. Moreover, no hyphae were observed with clamp connections, dolipores or parenthesomes, typical characteristics of Basidiomycota (Roy et al., 2009), in fungi of roots (unpublished data—Flores-Borges, DNA), as well as of fruits of *P. schenckii* (unpublished data—Alves, MF). Among the difficult cultivable fungi, there are those not related to rhizoctonia (Zelmer et al., 1996). It was not true for the fungus found in this work, *Tulasnella* sp., which, despite being closely related as the group, as well as Ceratobasidiaceae and Sebacinales, was not possible to be verified by means of the culture dependent identification methodology employed in this study. The PDA medium is very nutrient rich and greatly favors the development of fast-growing fungi, which can lead to the non-development of slower-growing fungi such as Tulasnellaceae (Zettler and Corey, 2018).

#### Germination Trials

Regarding the fungi tested, three genera were able to cause seed germination of *P. schenckii*, namely: *Trichoderma*, *Fusarium*, and *Clonostachys*. Among these, only the genus *Clonostachys* stimulated the development of the protocorm beyond the seed testa rupture. *Fusarium* has previously been reported in the literature as being able to cause seed germination of three green orchids, *Cypripedium reginae*, *Cypripedium parviflorum*, and *Platanthera grandiflora* (Vujanovic et al., 2000), and a century earlier, by Bernard (1900, 1903, 1904), qt. in Vujanovic et al. (2000). Other taxa of Ascomycota have also been related to the germination of terrestrial orchid seeds belonging to Neottieae and Orchideae tribes (Rasmussen et al., 2015). According to Vujanovic et al. (2000), *Fusarium* would be able to produce structures very similar to the shape and distribution of monilioid cells, typical of rhizoctonia, and easily confused with them. As far as it is known, there are no reports in the literature about the germination of orchids by fungi of the genus *Trichoderma* or *Clonostachys*.

In preliminary studies by Yagame et al. (2007), it was not possible to induce the germination of the aclorophyllated orchid *E. roseum* through symbiotic germination in oat meal agar medium despite the use of a mycorrhizal fungus, due to the excessive growth of the fungi in the culture medium, as also reported in the first trials with *P. schenckii*. In subsequent trials, altering oat concentration of the culture medium as well as its renewal by means of periodic transfers of the seeds and protocorms to new plates, it was enough to solve this obstacle. The germination methodology initially cited in this paragraph has been successfully used for several species of green orchids (Stewart and Kane, 2007; Steinfort et al., 2010; Pereira et al., 2015) and, in some cases, for MH ones (Bruns and Read, 2000). The methodology used may not have been adequate to meet the requirements necessary for the complete development of protocorm and establishment of the seedling of this species, or the association with other mycorrhizal fungal taxa may be necessary for this to occur. It is not clear what prevented the later development of the protocorm of *P. schenckii* beyond that reached in this work, and further studies are necessary.

In *in situ* germination experiments by Rasmussen and Whighan (1993), none of the five terrestrial orchid species germinated before 6 months of experiment; two of them, *Liparis liliifolia* and *Tipularia discolor*, were not able to germinate in the 12 months from the beginning of their experiment, and the seeds of *P. schenckii* did not germinate in a period of 18 months. During this period, the seeds remained intermixed by hyphae within the fruits, but without presenting increase in volume or testa rupture (data not shown). As expected, the asymbiotic germination of *P. schenckii* also did not produce results and, to date, no MH orchid species have been able to germinate in the absence of mycorrhizal fungi (Bruns and Read, 2000; McKendrick et al. 2002; Leake et al., 2004; Yagame et al., 2007). Although the *in vitro* symbiotic germination assays show only the initial development of protocorms of *P. schencki*i, and their development is not possible until the establishment of the seedling, this is probably the first report of the germination of a species of heterotrophic and aclorophyllated orchid that is stimulated by the presence of non-mycorrhizal endophytic fungi isolated from roots and, specifically, from fruits.

### Protocorm Development

The mycorrhizal interactions within the Orchidaceae family occur from the association between specific fungal taxa of the phylum Basidiomycota and the orchids (Peterson et al., 2004). This interaction can be verified by the ability of the fungus to cause germination of the plant seeds, besides of its capacity to form pelotons inside the living cells of the plant (Peterson et al., 2004; Rasmussen and Rasmussen, 2009). In this study, despite of causing the germination of the seeds of *P*. *schenckii,* the formation of pelotons in the cells of the protocorm was not observed for the isolate F38, of the genus *Clonostachys*, being that this fungus does not belong to a taxa known as mycorrhizal. Therefore, the interaction between both cannot be classified as mycorrhizal. However, MH orchids are not known to be able to germinate or even develop independently from a mycorrhizal fungus (Rasmussen and Rasmussen, 2009), and even less in the form of culture to which the seeds of *P. schenckii* were submitted, in culture medium with complex carbon sources (Zettler, 1997). Therefore, it is assumed that some degree of interaction between some of the tested fungal isolates and the orchid protocorms has occurred that seed germination and initial protocorm development was stimulated, although it is not of mycorrhizal nature.

## CONCLUSION

Identifying fungal endophytes from MH orchids is a key step in understanding species interactions in tropical conditions. All the endophytic fungi present in the roots of *P. schenckii* belonged to the phylum Ascomycota and varied in the detriment of the geographic distance of the populations, as commonly observed for the endophytes of other orchids. The methodology employed for fungal isolation was not able to isolate potentially mycorrhizal fungi from the Tulasnellaceae, found by other means, as well as no other taxon of Basidiomycota.

Differently from what is observed for most of the MH species of tropical regions, *P. schenckii* is associated with fungi commonly found in green orchids. Three different genera of isolated nonmycorrhizal fungi were able to induce the germination of the seeds of the species under the asymbiotic condition and one of them, *Clonostachys* sp., in addition to the external tegument rupture, promoted the initial development of the protocorm. *P. schenkii* protocorms in co-culture with this isolate showed fungal hyphae within living plant cells, but without the formation of typical mycorrhizal pelotons.

*P. schenckii* was not able to germinate in the absence of fungi or in seeding and seed recovery trials *in situ*. This is the first report of stimulation of the germination of the MH orchid species in the presence of non-mycorrhizal endophytic fungi isolated from above and belowground organs of this orchid species.

### DATA AVAILABILITY STATEMENT

All datasets for this study are included in the article/ **Supplementary Material**.

## AUTHOR CONTRIBUTIONS

JM and LS initiated and designed the study. LS carried the anatomical analysis, analysis of scanning microscopy, isolation and analysis of fungi, germination trials and writing the manuscript. MB and LS developed the phylogenetic tree. DF-B carried transmission microscopy and SA and SK assisted molecular analysis of fungi and contributed to the drafts. MB and JM supervised the work and assisted and interpreted of the results.

### FUNDING

This work was financed by the São Paulo Research Foundation (FAPESP—2015/26479-6), and the National Council for Scientific and Technological Development (CNPq - 447453/2014-9; 310184/2016-9). We thank CNPq and the Coordination for the Improvement of Higher Education Personnel (CAPES) for the master's degree scholarships to the first author and doctoral scholarship to the second author, respectively. This study was funded in part by CAPES— Finance Code 001.

#### ACKNOWLEDGMENTS

We thank the Instituto Florestal (Parque Estadual da Serra do Mar, Núcleo Santa Virginia and Núcleo Picinguaba) for

#### REFERENCES


the development of the study on a protected public land. We also thank the Laboratory of Electron Microscopy (LME) of Unicamp for assisting in the preparation of samples and microscope analysis. The authors also thank Espaço da Escrita—Pró-Reitoria de Pesquisa—UNICAMP—for the language services provided.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2019.01589/ full#supplementary-material


roots of endangered native orchids from the Atlantic Forest, Brazil. *Mycorrhiza* 24 (1), 55–64. doi: 10.1007/s00572-013-0512-0


**Conflict of Interest:** 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 Sisti, Flores-Borges, Andrade, Koehler, Bonatelli and Mayer. 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 Phylogenomic Analysis of the Floral Transcriptomes of Sexually Deceptive and Rewarding European Orchids, Ophrys and Gymnadenia

Laura Piñeiro Fernández 1,2\*, Kelsey J. R .P. Byers 2,3, Jing Cai 2,4, Khalid E. M. Sedeek 2,5,6, Roman T. Kellenberger 2,7, Alessia Russo 1,2,8, Weihong Qi <sup>9</sup> , Catharine Aquino Fournier <sup>9</sup> and Philipp M. Schlüter 1\*

<sup>1</sup> Institute of Botany, University of Hohenheim, Stuttgart, Germany, <sup>2</sup> Department of Systematic and Evolutionary Botany, University of Zurich, Zurich, Switzerland, <sup>3</sup> Department of Zoology, University of Cambridge, Cambridge, United Kingdom, <sup>4</sup> Center for Ecological and Environmental Sciences, Northwestern Polytechnical University, Xi'an, China, <sup>5</sup> Laboratory for Genome Engineering and Synthetic Biology, Division of Biological Sciences, King Abdullah University of Science and Technology, Thuwal, Saudi Arabia, <sup>6</sup> Agricultural Genetic Engineering Research Institute (AGERI), Agriculture Research Centre, Giza, Egypt, <sup>7</sup> Department of Plant Sciences, University of Cambridge, Cambridge, United Kingdom, <sup>8</sup> Department of Plant and Microbial Biology, University of Zurich, Zurich, Switzerland, <sup>9</sup> Functional Genomics Centre Zurich, Zurich, Switzerland

#### Edited by:

Jen-Tsung Chen, National University of Kaohsiung, Taiwan

#### Reviewed by:

Ashley N. Egan, Aarhus University, Denmark Alejandra Vázquez-Lobo, Universidad Autónoma del Estado de Morelos, Mexico

#### \*Correspondence:

Laura Piñeiro Fernández laura.pineiro@systbot.uzh.ch Philipp M. Schlüter philipp.schlueter@uni-hohenheim.de

#### Specialty section:

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

Received: 20 July 2019 Accepted: 07 November 2019 Published: 29 November 2019

#### Citation:

Piñeiro Fernández L, Byers KJRP, Cai J, Sedeek KEM, Kellenberger RT, Russo A, Qi W, Aquino Fournier C and Schlüter PM (2019) A Phylogenomic Analysis of the Floral Transcriptomes of Sexually Deceptive and Rewarding European Orchids, Ophrys and Gymnadenia. Front. Plant Sci. 10:1553. doi: 10.3389/fpls.2019.01553 The orchids (Orchidaceae) constitute one of the largest and most diverse families of flowering plants. They have evolved a great variety of adaptations to achieve pollination by a diverse group of pollinators. Many orchids reward their pollinators, typically with nectar, but the family is also well-known for employing deceptive pollination strategies in which there is no reward for the pollinator, in the most extreme case by mimicking sexual signals of pollinators. In the European flora, two examples of these different pollination strategies are the sexually deceptive genus Ophrys and the rewarding genus Gymnadenia, which differ in their level of pollinator specialization; Ophrys is typically pollinated by pseudocopulation of males of a single insect species, whilst Gymnadenia attracts a broad range of floral visitors. Here, we present and describe the annotated floral transcriptome of Ophrys iricolor, an Andrena-pollinated representative of the genus Ophrys that is widespread throughout the Aegean. Furthermore, we present additional floral transcriptomes of both sexually deceptive and rewarding orchids, specifically the deceptive Ophrys insectifera, Ophrys aymoninii, and an updated floral transcriptome of Ophrys sphegodes, as well as the floral transcriptomes of the rewarding orchids Gymnadenia conopsea, Gymnadenia densiflora, Gymnadenia odoratissima, and Gymnadenia rhellicani (syn. Nigritella rhellicani). Comparisons of these novel floral transcriptomes reveal few annotation differences between deceptive and rewarding orchids. Since together, these transcriptomes provide a representative sample of the genus-wide taxonomic diversity within Ophrys and Gymnadenia (Orchidoideae: Orchidinae), we employ a phylogenomic approach to address open questions of phylogenetic relationships within the genera. Specifically, this includes the controversial placement of O. insectifera within the Ophrys phylogeny and the placement of "Nigritella" type morphologies within the phylogeny of Gymnadenia. Whereas in Gymnadenia, several

**93**

conflicting topologies are supported by a similar number of gene trees, a majority of Ophrys gene topologies clearly supports a placement of O. insectifera as sister to a clade containing O. sphegodes.

Keywords: phylogenomics, orchids, Ophrys, Gymnadenia, transcriptome, pollination strategy

#### INTRODUCTION

Orchidaceae and Asteraceae constitute the largest families of flowering plants. Over 800 orchid genera and 25,000 species have been described, with an average rate of 500 species and 13 genera described per year (Cribb et al., 2003; Chase et al., 2015). Orchids have colonized a great variety of geographical ranges, from Scandinavia to Tierra del Fuego (Antonelli et al., 2009; Domínguez and Bahamonde, 2013), although the vast majority of species occur in tropical and neotropical areas (Dressler, 1993). The key to their success has variously been hypothesized to reside in their epiphytic habitat (for tropical orchids) or in their high level of pollinator specialization (Gravendeel et al., 2004; Cozzolino and Widmer, 2005). About two thirds of orchid species present rewards to their visitors, in most cases, nectar (Dafni and Ivri, 1979; Bell et al., 2009; Johnson et al., 2013). These rewarding species are commonly generalized in their pollination, attracting a wide range of pollinators (Brantjes, 1981; Claessens and Kleynen, 2017). However, the ability to produce nectar is missing in one third of species across the family. Instead, they have developed alternative mechanisms based on deception (Ackerman, 1986; Jersáková et al., 2006; Schiestl and Schlüter, 2009; Johnson and Schiestl, 2016). Some of these mechanisms target generalist pollinators, e.g., food deception, where orchids attract pollinators by advertising floral cues that resemble those from rewarding plants (Salzmann et al., 2007; Braunschmid et al., 2017). On the other hand, orchids have also developed mechanisms such as sexual deception to attract highly specialized pollinators. Sexually deceptive flowers produce chemical signals that mimic the sexual pheromones of pollinators, and thus, lead the pollinators to "pseudo-copulate" with the flowers (Kullenberg and Bergström, 1976; Paulus and Gack, 1990; Schiestl et al., 1999). Examples of such behaviour occur in the Australian Chiloglottis spp. (Mant et al., 2002; Schiestl et al., 2003), or the recently discovered sexually deceptive Caladenia abbreviata (Phillips and Peakall, 2018).

In the European flora, one can find representatives of the aforementioned pollination strategies in the sexually deceptive genus Ophrys and the rewarding genus Gymnadenia, both within the subtribe Orchidinae (subfamily Orchidoideae) (Inda et al., 2012). Orchids from the Mediterranean genusOphrys attract male pollinators by means of sexual deception (Paulus and Gack, 1990; Ayasse et al., 2000; Schiestl et al., 2000). Attractiveness to pollinators in the genus is highly species-specific, that is, each Ophrys species normally attracts a single pollinator species (Paulus and Gack, 1990; Paulus, 2018) by releasing chemicals (for solitary bees, mostly alkenes) mimicking the female sex pheromones (Schiestl et al., 2000; Schlüter and Schiestl, 2008; Xu et al., 2012). This high specificity acts as a pre-zygotic barrier and facilitates reproductive isolation between orchid species (Xu et al., 2011; Xu et al., 2012; Paulus, 2018). Ophrys is a recently diverged genus (crown age estimated ca. 5Ma) with ancestral wasp pollination (Breitkopf et al., 2015), but extant species are commonly pollinated by solitary bees, e.g. Eucera or Andrena (Paulus and Gack, 1990; Gaskett, 2011). Successful floral isolation and species divergence in the genus may easily be achieved by shifts between similar pollinators, where small changes in genes involved in the pheromone profiles can lead to attraction of new, related pollinators (Schlüter et al., 2011; Sedeek et al., 2014; Schlüter, 2018). For instance, after two independent shifts to (mostly) Andrena solitary bee pollination (Breitkopf et al., 2015), two parallel adaptive radiations have taken place simultaneously within the last ca. 1 Ma, yielding two major clades, the Ophrys sphegodes and the Ophrys fusca species complexes. In line with its recent radiation, a large amount of genetic polymorphism is shared across closely related species within the O. sphegodes complex, which has been attributed to common ancestry rather than independent mutations or recent hybridization, although a hybridization event prior to radiation seems distinctly possible (Sedeek et al., 2014; Roma et al., 2018; Cozzolino et al., 2019). Coalescence theory predicts that in the case of a radiation, the time of coalescence of these polymorphic alleles will often predate the split of species (Takahata, 1989). Yet, or maybe because of this, phylogenetic relationships within Ophrys remain controversial, with different markers in the genome potentially painting different pictures of relationships (Cozzolino et al., 2019). Phylogenetically, the ca. 10 main Ophrys lineages are split into three major clades (where clade a includesOphrysinsectifera, b includes theO. fusca s.l. lineage and g includes the O. sphegodes s.l. lineage) and the relationships among major lineages within these clades are relatively clear, although one major question remains unclear. In particular, the placement of the wasp-pollinatedO. insectifera L. (clade a) within theOphrys phylogeny has been suggested to be either the earliest-branching lineage [topology: (a,(b,g))] or more closely related to the O. sphegodes lineage [topology: (b,(a,g))] (cf. e.g. Breitkopf et al., 2015; Bateman et al., 2018b, and references therein).

The Eurasian genusGymnadenia is characterized by fragrant, purple towhite, resupinateflowers thatmainly attract diurnal and nocturnal Lepidoptera species offering nectar as a reward. Although they attract a wide range of Lepidoptera, and some species are found in sympatry, pollinator overlap is minimal between most species (Vöth, 2000; Huber et al., 2005; Claessens andKleynen, 2011) and strong pollinator-mediated reproductive isolation has been reported between the putative sister species G. odoratissima (L.) Richard and Gymnadenia conopsea (L.) Brown (Sun et al., 2015). The latter species is strongly genetically differentiated from the morphologically similar taxon G. densiflora (Wahlenberg) Dietrich (Stark et al., 2011). Finally, the Alpine G. rhellicani (Teppner & E. Klein) Teppner & E. Klein (syn. Nigritella rhelliani) represents a morphologically distinct lineage within the genus, characterized by extremely dense inflorescences, generally dark red and without resupination, i.e. the labellum remains pointing upwards as opposed to rotated downwards as in otherGymnadenia species. The former genus Nigritella was merged into Gymnadenia only following molecular phylogenies (Hedrén et al., 2000). Previous phylogenetic analysis have shown that Gymnadenia odoratissima is sister to Gymnadenia conopsea, and Gymnadenia densiflora forms a clade with Gymnadenia rhellicani (Bateman et al., 2003; Sun et al., 2015). However, these relationships remain contentious, since other studies support a sister-group relationship among Nigritella and the "classical" genus Gymnadenia (Hedrén et al., 2000; Brandrud et al., 2019). Hence, further attention is warranted, especially to clarify the position of Nigritella. The age of the most recent common ancestor shared among all Gymnadenia/Nigritella species is estimated to be around 2.5–3 Ma (Inda et al., 2012).

Due to the high taxonomic complexity of Orchidaceae, reconstructing phylogenetic patterns to understand relationships in the family remains challenging. In the last decades, phylogenetic studies in orchids moved from a morphological (Chittka and Menzel, 1992; Gravendeel et al., 2004) to a molecular approach aiming to provide a better insight into orchid relationships (Cameron et al., 1999; Stark et al., 2011; Inda et al., 2012; Breitkopf et al., 2015; Givnish et al., 2015; Bateman et al., 2018a). Previously, the focus of these analyses was at the level of using few genetic markers, e.g. ITS, to reconstruct phylogenies. However, this approach can be problematic as some markers are chosen by their relevance or suitability in a certain taxonomic group, even though they could present low resolution for certain taxonomic groups (Capella-Gutiérrez et al., 2014). Moreover, this approach generally focuses on estimating one coherent tree (e.g. by concatenating sequences), which ignores the fact that different loci can have different phylogenetic histories. Especially when dealing with recently diverged groups with incomplete lineage sorting (Pamilo and Nei, 1988), a genomic approach focusing on understanding patterns on different gene genealogies, may allow the quantification of the different phylogenetic scenarios and thus, be more informative on the evolutionary history of a group (Pease and Hahn, 2015; Pease et al., 2016). Orthologous genes, described as homologous genes that originated from a common ancestral gene as a result of the speciation process (Fitch, 1970), tend to retain the original function from the common ancestor over evolutionary time (Jensen, 2001). Thus, groups of orthologous genes within gene families, together with a genome-wide approach, are perfect candidates to resolve orchid phylogeny and effectively clarify their relationships in an evolutionary framework (Li et al., 2003; Deng et al., 2015).

Here, we present the novel floral transcriptome of the Mediterranean sexually deceptive orchid Ophrys iricolor Desf., a representative of the genus Ophrys in the Aegean area, which is considered to be a member of the O. fusca group (clade b) and represents the evolutionarily distinct abdomen-pollinated members of the genus (previous section Pseudophrys) (Schlüter et al., 2009). In addition, we present several floral transcriptomes of both rewarding and deceptive orchids of the subtribe Orchidinae, particularly the rewarding orchids G. conopsea, G. densiflora, G. odoratissima, and G. rhellicani, together with the sexually deceptive O. insectifera, Ophrys aymoninii (Breistroffer) Buttler, and finally, an updated transcriptome of O. sphegodes s.l. (Sedeek et al., 2013). Using a set of orthologous genes, we employ a genome-wide approach to phylogenetic analysis of these novel floral transcriptomes together with published orchid transcriptomic/genomic data, to compare the transcriptomes of deceptive and rewarding orchids. Furthermore, as these transcriptomes cover the genus-wide taxonomic diversity within Ophrys and Gymnadenia, our objectives are to elucidate (1) the placement of the O. insectifera complex within the three major clades in the Ophrys phylogeny, (2) the placement of the morphologically distinct G. rhellicani (and presumably other members of subgenus Nigritella) within the phylogeny of Gymnadenia and (3) whether there is evidence of introgression due to shared pollinators in distinct Ophrys lineages.

#### MATERIAL AND METHODS

#### Plant Material

The novel Ophrys iricolor s.l. (O. iricolor s.s. and Ophrys mesaritica H.F. Paulus, C. Alibertis & A. Alibertis) crossspecies transcriptome is presented here. Data from the putative sister species O. iricolor s.s. and O. mesaritica (Schlüter et al., 2009) were assembled into a single transcriptome due to expected high levels of allele sharing among the group, as seen in the O. sphegodes complex (Sedeek et al., 2013; Sedeek et al., 2014). Sample size (Ophrys iricolor s.l., N = 16 biological replicates; O. sphegodes s.l., N = 37) and provenance are listed in Table 1. The previously published cross-species O. sphegodes s.l. (O. exaltata subsp. archipelagi (Gölz & H.R. Reinhard) Del Prete, O. garganica Nelson ex O. & E. Danesch, and O. sphegodes Miller) transcriptome (Sedeek et al., 2013) is here updated with data from additional samples, including from O. incubacea Bianca (samples from Sedeek et al., 2014) within the same species complex that is characterized by the aforementioned high levels of allele and transcript sharing among species (Sedeek et al., 2013; Sedeek et al., 2014) and is hence covered in a single cross-species transcriptome assembly. Additionally, O. insectifera and O. aymoninii transcriptomes are also presented here. Data from O. insectifera and O. aymoninii (collected in Gervasi et al., 2017), were assembled into separate transcriptomes because these species are pollinated by different types of pollinators (O. insectifera is wasp-pollinated, while O. aymoninii is Andrena-pollinated) and the assumption of high levels of within-group allele sharing cannot be made. Finally, sampled flowers from the clearly distinct species G. conopsea, G. densiflora, G. odoratissima and G. rhellicani (from Kellenberger et al., 2019) were used to create individual transcriptome assemblies for these species to complement the published cross-species Gymnadenia transcriptome assembly (N = 10,


2References are R1: Schlüter et al. (2011), R2: Sedeek et al. (2014), R3: Gervasi et al. (2017), R4: Kellenberger et al. (2019), R5: Sedeek et al. (2013);

3Raw sequencing data for this column represents the sum of data from all Gymnadenia samples;

aSpecies are I: O. iricolor, M: O. mesaritica; bone O. iricolor sample failed to produce results; cspecies are E: O. exaltata subsp. archipelagi, G: O. garganica, I: O. incubacea, S: O. sphegodes; dfor O. exaltata, O. garganica and O. sphegodes, one sample each was derived from labella at bud stage (same biological individuals as used for open flowers), and one sample each of these species was field-collected whereas the remaining samples were grown under greenhouse conditions (Schlüter et al., 2011); efor O. incubacea, 3 were sampled under greenhouse conditions and 5 were collected in the field and were added in a second sequencing batch; fspecies are C: G. conopsea, D: G. densiflora, O: G. odoratissima, R: G. rhellicani; gO. mesaritica was sampled at Pirgos, Crete (28 February 2013; accessions PMS540 A,D,K,N,O,Q,R,T) and O. iricolor at Kato Chorio, Crete (8 April 2013; accessions PMS558 E, I), Vasiliki, Crete (9 April 2013; PMS560 A, C) and at Jouchtas, Crete (10 April 2013; PMS561 A, I, H, O), all under permit number 125001/95 issued on 28 January 2013 by the Hellenic Republic Ministry Of The Environment, Energy & Climate Change, Athens, Greece. hRawsequencingdatapublishedbyKellenbergeretal.(2019).iFigshareratherthanTSAidentifier.

TABLE 1 | Statistics of

transcriptomic

 data for each

species/assembly.

Table 1) (Kellenberger et al., 2019). As far as it was possible to ascertain pollination status (not always possible for Gymnadenia flowers), all samples used in this study were from unpollinated flowers of diploid individuals. Flowers were flash-frozen and stored at ‑80°C until RNA extraction was conducted as detailed by Kellenberger et al. (2019). Since polyploids are known from Gymnadenia and (occasionally) Ophrys and to ensure that all samples sequenced were diploid, ploidy levels of O. iricolor and O. mesaritica were checked via flow cytometry of pollinia as described by Xu et al. (2011) using a Cell Lab Quanta™ SC-MPL flow cytometer (Beckman Coulter, Fullerton, Canada). Phaseolus coccineus "Scarlett Emperor" (sativa Rheinau SG, Switzerland) leaf material was used as internal standard. Ploidy levels were previously described for O. sphegodes s.l. (Sedeek et al., 2014), O. insectifera and O. aymoninii (Gervasi et al., 2017) and the four Gymnadenia species (Kellenberger et al., 2019) used in this study, including all sequenced individuals.

#### RNA Extraction, Library Preparation and Sequencing

Total RNA was extracted separately for each biological individual and tissue with TRIzol reagent (Thermo Fisher Scientific, Massachusetts) according to the manufacturer's protocol followed by a purification step using Qiagen RNeasy MinElute Cleanup Kit (Qiagen, Netherlands). Quality of the isolated RNA was determined with a Qubit® (1.0) Fluorometer (Life Technologies, California, USA) and a Bioanalyzer 2100 (Agilent, Waldbronn, Germany). Paired-end sequencing was performed on the Illumina HiSeq 2000 or 2500 platforms (Illumina, Inc, California, USA) for Ophrys and Gymnadenia samples (Table 1), generating separate files for each biological sample.

### Transcriptome Assemblies and Functional Annotation

Individual reads were first aligned to PhiX Control library (Illumina) sequences using bowtie2 v2.2.4 (Langmead and Salzberg, 2012) to remove sequencing control reads. Filtered reads were trimmed using Trimmomatic v. 0.36 (Bolger et al., 2014) to remove any Illumina adapters. Surviving reads were then de-novo assembled to transcripts using Trinity r20140717/ v. 2.0.618 (Grabherr et al., 2011). In the case of O. sphegodes, where a previous assembly based on 454, Solexa and Sanger data was available (Sedeek et al., 2013), additional Illumina HiSeq reads were assembled with Trinity as described above and then merged with the published assembly using cd-hit-est (Li and Godzik, 2006; Fu et al., 2012) (95% sequence identity threshold with full length alignment coverage for the shorter sequence). Protein coding regions were analysed using TransDecoder r20140704 (http://transdecoder.github.io) (Haas et al., 2013). The assembled contigs were annotated with the standard Trinotate annotation pipeline (https://trinotate.github.io/) (Grabherr et al., 2011) against Swissprot (Boeckmann et al., 2003), Pfam (Finn et al., 2014), TmHMM (Krogh et al., 2001), Gene Ontology (Ashburner et al., 2000) and SignalP (Petersen et al., 2011). Due to high levels of overlap among the four singlespecies Gymnadenia transcriptomes (Figure S1B), we annotated only the cross-species Gymnadenia transcriptome from all four species. For purposes of comparison, we also updated the annotation of the previously published, updated (v.2) transcriptome of O. sphegodes (Sedeek et al., 2013) with Trinotate. Finally, to estimate the completeness of the transcriptomes, we performed a BUSCO v3.1.0 assessment ( Simão et al., 2015 ) with the lineage databases embryophyta\_odb10 and liliopsida\_odb10 (Figures 1A, B).

transcriptomes (from inside to outside), where the first three (blue) categories together are taken as an estimation of transcriptome "completeness". (A) BUSCO results with embryophyta\_odb10 and (B) the larger liliopsida\_odb10 databases.

#### Phylogenomic Analysis

OrthoMCL v2.0.9 (Li et al., 2003) was used under the MySQL v14.14 server to identify orthologous groups based on annotated coding sequences (CDS) (where no annotated CDS were available, they were derived by TransDecoder as above) of 15 members of the Orchidaceae family including the above described Ophrys and the four Gymnadenia single-species transcriptome assemblies together with the transcriptomes/ genomes of Apostasia shenzhenica and Phalaenopsis equesteris (Zhang et al., 2017), Dactylorhiza fuchsii (Balao et al., 2017), Chiloglottis trapeziformis (Wong et al., 2017), Dendrobium catenatum (Zhang et al., 2016), and Platanthera clavellata and Goodyera pubescens (retrieved from the 1KP project; http://www. onekp.com/). Following the TranslatorX pipeline (Abascal et al., 2010), sequences were aligned using Mafft v7.407 (Katoh and Standley, 2013). To construct phylogenetic trees, a pipeline as described in Xu et al. (2017) was followed. In brief, poorly aligned sequences were removed using trimal v1.2 (Capella-Gutiérrez et al., 2009). Selection of the best-fit models of nucleotide substitution was performed with jModelTest 2.1.10 (Santorum et al., 2014), with parameters: -f -i -g 4 -a -AIC -s 3. This allowed the inclusion of models with unequal base frequencies, a proportion invariable sites, rate variation among sites and set 4 categories, model-averaged phylogeny for each active criterion. Moreover, it used AIC (Akaike Information Criterion) for model selection and accounted for 3 substitution schemes. Maximum likelihood trees of the best-fit models were calculated with phyML 3.3 (Guindon and Gascuel, 2003). For each taxonomically fully sampled orthologous group, tree topologies from Ophrys and Gymnadenia single-copy gene branches were extracted. In addition, we also extracted topologies where one Ophrys species was missing. The extraction of tree topologies was automated with an in-house R script. Moreover, for both Ophrys and Gymnadenia, we extracted topologies where gene duplications happened only within a monophyletic group of a given species. In the latter case, all but one of the duplicate tips was dropped from the phylogeny (keep.tip function from the package ape for R v3.5.0) (R Core Team, 2001). After retrieving (rooted) topologies of target groups, we compared these topologies with Robinson-Foulds distances, where a distance of 0 indicates that topologies are in full agreement with each other (Robinson and Foulds, 1981), using the package phytools (Revell, 2012) for R. Tree visualization was performed using the Bioconductor package Ggtree (Yu et al., 2017) for R. Finally, we compared the annotation, particularly the GO Plant Slim terms, of the different topologies observed for Ophrys and Gymnadenia.

#### RESULTS

#### Transcriptome Assemblies and Functional Annotation

All Ophrys individuals were diploid (Figure S2 for O. iricolor s.l.), consistent with previous studies (Xu et al., 2011; Sedeek et al., 2014). After sequencing, a total of 493.5 million paired-end (PE) reads from O. iricolor and 191.9 million from Gymnadenia were produced (Table 1). All the raw sequencing data (totalling 431.8 Gbp from 2111 million PE reads) are available in the Sequence Read Archive (SRA) of the National Center for Biotechnology Information (NCBI) under the accession numbers in Table 1. We successfully produced 131,528 and 589,218 contigs (Table 1) for O. iricolor and for the Gymnadenia cross-species assembly, respectively, corresponding to 88,664 and 174,633 Coding Sequences (CDSes) (Table 2). The remaining sequences did not match any known gene from the databases queried. The annotation tables can be downloaded from figshare (links in Table 2). Based on the three main Gene Ontology categories (biological process, cellular component, and molecular function), we compared the 14 most common GO Plant Slim terms (Clark et al., 2005) in the Gymnadenia spp. cross-species, O. iricolor and the updated O. sphegodes transcriptomes (Table 1, Figure 2). To avoid overrepresentation of general terms such as "metabolic" or "cellular" processes, we omitted the first 7, 3, and 3 terms for Biological Process, Cellular Component and Molecular Function, respectively. Overall, the three transcriptomes are very similar in GO terms. The main differences between O. sphegodes and O. iricolor are the lack of terms related to "response to endogenous stimulus" in O. iricolor, and the presence of terms related to "vacuole" in O. sphegodes (Figures 2A, B). On the other hand, the Gymnadenia transcriptome differs from the Ophrys transcriptome by showing a high number of genes related to "nucleus" processes and an absence of those related to "membrane" processes (Figure 2B). Finally, BUSCO assessments with the embryophyta lineage database indicated that the completeness of the transcriptomes was 93.4, 91.9, and 94.1% for O. iricolor, O. sphegodes v.2, and cross-species Gymnadenia transcriptomes, respectively (Figure 1A). These results therefore suggest a reasonably high assembly quality of our floral transcriptomes, especially when compared with fully sequenced orchid genomes (encoding the transcripts of all tissues), i.e. the Apostasia genome with a 93.62% completeness, 94.45% in Phalaenopsis equestris and 95.49% in Dendrobium catenatum (all using the embryophyta database) (Zhang et al., 2017). Also, with 87.6, 85.5, and 87% for the larger BUSCO liliopsida lineage database (Figure 1B), for O. iricolor, O.


FIGURE 2 | GO Plant Slim functional annotation. GO Plant Slim annotation of the three transcriptomes for the most common (A) Biological Process, (B) Cellular Component and (C) Molecular Function terms.

sphegodes v.2 and cross-species Gymnadenia transcriptomes, respectively, our transcriptomes appear relatively complete with respect to monocot-specific genes.

### Phylogenomic Analysis

Overall, we found the 15 orchid species included in this study to share a total of 1,749 gene families. From these gene family phylogenies, 226 contain Ophrys monophyletic groups with no gene duplications sampled from all Ophrys species and 160 contain Gymnadenia monophyletic groups with no gene duplications sampled from all Gymnadenia species separately. In addition, 116 contain informative topologies with one Ophrys species missing; and 153 and 318 topologies contain gene duplications (or alleles) within single-species monophyletic groups for Ophrys and Gymnadenia, respectively. For Ophrys and Gymnadenia, we found 5 and 6 of the 15 possible rooted topologies for four taxa, respectively. In Ophrys, the most common topology (77% of the trees) suggests that O. insectifera s.l. (with O. aymoninii) is not the basalmost clade, but instead places it in a clade with O. sphegodes, whereas O. iricolor takes the basal position (Figures 3A, B). This is also evident from the consensus tree over all orthologous gene groups (Figure 3C) and from all (100%) of the trees missing one Ophrys species (Figure 3B). In the case of Gymnadenia, the distribution of topologies is more even. Yet, the most common topology, supported by 33% of the trees, places G. rhellicani at the basal position in the Gymnadenia tree (Figure 4). Overall, a total of 48% of evaluated Gymnadenia genes show a topology that places G. rhellicani as a sister to all other species. Also, strikingly, only 33% of gene topologies support a sister-species relationship among G. conopsea and G. odoratissima. We compared the GO annotations of each topology in Ophrys and Gymnadenia, but despite some annotation differences between topologies, there is no clear pattern with respect to putatively pollinatorrelevant features (Figures S3 and S4). Although not significant, the two most common Ophrys topologies also show the highest average branch length (Figure S5A), whereas two less common topologies have the longest branch lengths in Gymnadenia (Figure S5B, non-significant); these are not united by a common phylogenetic theme (e.g. with respect to G. rhellicani).

### DISCUSSION

This study provides significant new transcriptome sequence resources aimed to improve our knowledge about the highly complex Orchidaceae family. Specifically, we present novel floral transcriptomes of several members of the subtribe Orchidinae of the Orchidoideae subfamily, covering both sexually deceptive and rewarding orchids. Overall, there were no striking differences between sexually deceptive and rewarding orchids when comparing the most common annotation terms based on Gene Ontology categories. This is not a surprise, because the GO Slim categories approach, although providing a large vocabulary to describe the functional categories, also suffers from a lack of clarity and too broad definitions, resulting in only a vague overview of molecular biology (Smith et al., 2003). At the same time, the phylogenetic proximity of Ophrys and Gymnadenia provides a plausible explanation for the lack of strong differentiation in terms of GO categories and suggests that differences in pollination strategy do not require fundamental changes in the genome-wide repertoire of florally expressed genes. This is in line with the phylogenetic lability of pollination strategies reported within the Orchidinae (Inda et al., 2012).

However, clear differences between Ophrys and Gymnadenia are apparent in terms of the transcriptome-wide distribution of gene tree topologies. For phylogeny reconstructions, rather than concatenating sequences, we evaluated multiple gene family trees separately. Trees derived from concatenated sequences do not reveal discrepancies between individual genes that are expected under a standard coalescent process, i.e., the more recent a species split is, the more tree topologies are expected due to incomplete lineage sorting (Takahata, 1989). Disagreement between gene trees and species trees has been observed in an increasing number of studies suggesting that the combination of a large amount of ancestral polymorphism and post-speciation gene flow between taxa can lead to large systematic differences between gene and species trees (Green et al., 2010; Novikova et al., 2016; Filiault et al., 2018; Malinsky et al., 2018).

Interestingly, the two most common gene tree topologies recovered for Ophrys reflect previous published phylogenetic reconstructions, our topologies 1 and 2 (Figure 3) corresponding to the phylogenies published most recently by Bateman et al. (2018b) and Breitkopf et al. (2015), respectively. Breitkopf et al.'s reconstruction suggested the O. insectifera group (clade a, including O. aymoninii) as the basal clade on the tree. By contrast, the phylogenetic reconstruction by Bateman et al. places O. insectifera closer to the O. sphegodes group, whereas a lineage containing the O. fusca complex (clade b, here represented by O. iricolor) is the earliest diverged. Our results, with a consensus of 85% of gene topologies, overwhelmingly support the inner placement of O. insectifera, rather than a basal position (Figure 3C). However, with the wasp-pollinated O. insectifera sister to the clade containing O. sphegodes and the wasp-pollinated O. speculum sister to the clade containing O. iricolor/O. fusca, the phylogeny's implication for the ancestral mode of pollination remains unchanged; the inference of ancestral wasp pollination in the genus Ophrys (Breitkopf et al., 2015) therefore seems unaffected by our findings. Nevertheless, it is striking that we found no strong evidence for discordant phylogenies throughout the genome.

Since the O. insectifera-group member O. aymoninii, a narrow endemic in southern France, is Andrena-pollinated (Paulus and Gack, 1990; Gervasi et al., 2017), phylogenies placing O. aymoninii together with the other Andrena-

Gymnadenia.

pollinated linages, O. sphegodes and/or O. iricolor could be (but need not be) an indication of hybridization and introgression via Andrena pollinators. Although our analysis recovers phylogenies (Figure 3A, topologies 3 and 5) consistent with this hypothesis, with only 4% of the gene trees overall, support for pollinatormediated introgression is weak at best.

Unlike Ophrys with a clearly predominant phylogeny across the transcriptome, Gymnadenia presents a much less clear picture of species relationships. The sister relationship between G. conopsea and G. odoratissima has been supported in several previous studies (e.g. Bateman et al., 2003; Sun et al., 2015) including by a recent genome-wide RAD-Seq (concatenated) SNP data set (Brandrud et al., 2019). This relationship is here supported by the most common topology in the transcriptome (Figure 4, topology 3). Yet this is also the only topology that supports this relationship, accounting for only 33% of orthologous gene groups evaluated. We must therefore conclude that, from a genomic perspective, the sister relationship of G. conopsea and G. odoratissima is not beyond doubt.

The genus Gymnadenia now typically includes its former sister genus Nigritella as a subgenus. Initial hypotheses built on morphological data (Wucherpfennig, 2002), anthocyanin pigments (Strack et al., 1989), or AFLP markers (Ståhlberg, 1999) suggested the separation of the two genera. Early molecular phylogenies (usually based solely on ITS) typically sampled only G. conopsea, G. odoratissima, and a single member of Nigritella, which was generally the outgroup to the sister Gymnadenia species (Hedrén et al., 2000). When additional species were sampled and added to this basic phylogeny, G. densiflora (or, depending on the sampling, G. borealis) was shown to be the sister taxon to members of Nigritella, arguing for combining the genera (Pridgeon et al., 1997; Bateman et al., 1997; Bateman et al., 2003; Stark et al., 2011; Efimov, 2013). Addition of three nuclear genes did not change this topology (Rey, 2011; Sun et al., 2015). Interestingly, where authors considered multiple phylogenetic methods, conflict seems to arise in tree construction, with parsimony showing Nigritella as the outgroup to G. conopsea/G. densiflora/G. odoratissima/G. borealis, while Bayesian and maximum likelihood analyses demonstrate a sister relationship between Nigritella and either G. borealis or G. densiflora (Rey, 2011; Inda et al., 2012). In a major upgrade to the generic phylogeny, Brandrud et al. (2019) performed RAD-Seq, with contrasting results to the ITS-based phylogenies. Their phylogeny shows four Nigritella species as the outgroup to five Gymnadenia species, with no sister relation between G. densiflora and Nigritella, and the relevant nodes show high support.

Given the often contradictory results of earlier circumscription attempts, it is perhaps not too surprising that the different Gymnadenia gene topologies are relatively evenly distributed and that we see no single Gymnadenia phylogeny standing out as the best supported tree. However, the most common gene tree topology shows G. rhellicani as the outgroup to the other three sampled species (Figure 4, topology 3), in agreement with the recent RAD-Seq-based concatenated SNP analysis by Brandrud et al. (2019). Nonetheless, overall support for versus against a basal position of G. rhellicani is equivocal, at 48% of gene trees for (topologies 3 and 4) versus 52% against a basal position. The prevalence of other supported topologies (generally with G. odoratissima rather than G. rhellicani as the outgroup) suggests a complex population genetic history within the genus, perhaps partially due to gene exchange and incomplete lineage sorting. Neither gene annotation (Figure S4) nor average gene tree branch lengths (Figure S5) for topologies with basal G. rhellicani placement stand out as an indication of adaptive processes. Although Gymnadenia and Nigritella have produced one stable hybrid offspring, the apomict G. runei (Teppner and Klein, 1989) and other hybrids may be found, some dispute about their frequency exists (Claessens and Kleynen, 2011; Brandrud et al., 2019). Taken together, our analysis of Gymnadenia hints at a complex relationship among species that we are only beginning to understand.Whether this apparently more complicated pattern of genome-wide relationships in Gymnadenia as compared to Ophrys is due to the difference in pollination systems is currently unclear, although Gymnadenia's less specialized pollination strategy would certainly present more opportunities for hybridization.

Using multiple gene family trees instead of one concatenated tree has proven to be a useful approach (Boussau et al., 2013; One Thousand Plant Transcriptomes Initiative, 2019). Concatenation of sequences implies that loci with a larger number of phylogenetically informative sites can bias the inference such that it may not be representative of patterns of unlinked genes throughout the genome. Also, such an approach holds no explicit information about the specific other topologies that may be useful for disentangling more complex evolutionary patterns of relationships throughout the genome, as would clearly be of interest in cases such as Gymnadenia. This problem is likely to be more severe in phylogenies of closely related species where excessive incomplete linage sorting may be expected and where a more sophisticated coalescent-based analysis may be valuable. Additionally, a consensus tree of individual gene trees (e.g. Figure 3C) is informative of the proportions of those genes in the genome that support a certain species relationship. Moreover, it is important to note that unlike a bootstrap pseudoreplicate approach, this allows for real quantification of proportions of independently segregating loci and/or functional genes and is thus more biologically meaningful. So far, our analysis only covers a small part of the genome. However, given a high-quality genome reference, future integration of this approach along chromosomes may be able to reconstruct the ancestry of individual chromosomal fragments and thereby shed light on the detailed evolutionary patterns and the role of selection (see Filiault et al., 2018) in shaping lineage divergence. The significant new sequence resources provided in this study may be a first step towards realizing this goal for European orchids in the future.

### DATA AVAILABILITY STATEMENT

The datasets analysed for this study can be found in the NCBI accessions PRJNA574279 and PRJNA504609, and as indicated in Table 1 and 2.

### AUTHOR CONTRIBUTIONS

Designed the project: PS. Drafted the manuscript: LP, with assistance from PS, KB, and RK. Revised the manuscript: all authors. Extracted material and prepared the libraries: JC, KS, and RK. Sequenced and processed the raw data: WQ, CA. Assembled and annotated the transcriptomes: LP, KB, JC, AR, and WQ. Conducted phylogenomic analysis: LP. Interpreted the results: LP, KB, RK, and PS. Acquired funding: KB, RK, and PS.

### FUNDING

This work was financially supported by the Swiss National Science Foundation (SNF) (31003A\_155943 to PS). Additional support came from the University of Zurich Research Priority Programme "Evolution in Action" (PS/AR and F. Schiestl/RK), a University of Zurich Forschungskredit grant to RK, a PLANT FELLOWS Postdoctoral Fellowship grant to KB (European Union: FP7-PEOPLE-2010-COFUND Proposal 267243), and several travel grants from the Georges & Antoine Claraz Foundation Zurich (to LP, RK, KB, and PS).

### ACKNOWLEDGMENTS

We wish to thank Kriton Kalantidis, University of Crete, for access to his lab and support during sampling. We thank Karin Gross, Danae Laina and Daniel Gervasi for help with field work and sample collection, Shuqing Xu for sharing his scripts, and Florian Schiestl for support and sharing data. We also thank our funding sources as well as members of the Schlüter and Schiestl labs for assistance and helpful discussions.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2019.01553/ full#supplementary-material

### REFERENCES


Conflict of Interest: 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 Piñeiro Fernández, Byers, Cai, Sedeek, Kellenberger, Russo, Qi, Aquino Fournier and Schlüter. 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.

# First Record of Ategmic Ovules in Orchidaceae Offers New Insights Into Mycoheterotrophic Plants

*Mariana Ferreira Alves\*, Fabio Pinheiro, Marta Pinheiro Niedzwiedzki and Juliana Lischka Sampaio Mayer\**

Departamento de Biologia Vegetal, Instituto de Biologia, Universidade Estadual de Campinas, São Paulo, Brazil

The number of integuments found in angiosperm ovules is variable. In orchids, most species show bitegmic ovules, except for some mycoheterotrophic species that show ovules with only one integument. Analysis of ovules and the development of the seed coat provide important information regarding functional aspects such as dispersal and seed germination. This study aimed to analyze the origin and development of the seed coat of the mycoheterotrophic orchid Pogoniopsis schenckii and to compare this development with that of other photosynthetic species of the family. Flowers and fruits at different stages of development were collected, and the usual methodology for performing anatomical studies, scanning microscopy, and transmission microscopy following established protocols. P. schenckii have ategmic ovules, while the other species are bitegmic. No evidence of integument formation at any stage of development was found through anatomical studies. The reduction of integuments found in the ovules could facilitate fertilization in this species. The seeds of P. schenckii, Vanilla planifolia, and V. palmarum have hard seed coats, while the other species have seed coats formed by the testa alone, making them thin and transparent. P. schenckii, in contrast to the other species analyzed, has a seed coat that originates from the nucellar epidermis, while in other species, the seed coat originates from the outer integument.

Keywords: anatomy, integument, Epidendroideae, saprophytic, Vanilloideae

### INTRODUCTION

Flowers are highly variable structures, resulting in a great morphological diversity and a variety of adaptive processes in angiosperms (Endress, 1994; Friis et al., 2011). Variations in flower size and number of whorls, besides the presence or absence of fused floral parts, are caused by differences that occur during the development of floral organs. Plants exhibit open organization, which means that their organs are generally exposed, and that they do not have any organ or parts of organs internalized, with the exception of carpels (Endress, 2015). In turn, carpels can be free or united, becoming curved during their initial development, with edges getting closed or sealed when they are fully developed (Endress, 2006). While most floral organs are exposed, mainly due to the action of pollinating agents (e.g., animals, wind, and water), ovules are completely enclosed in the carpel a condition known as angiospermy (Endress, 2006).

Ovules are female reproductive structures that develop in the seeds (Bouman, 1984; Endress, 2011). Despite their relatively stable basic structure, ovules have a wide diversity of form, varying in terms of

#### Edited by:

Jen-Tsung Chen, National University of Kaohsiung, Taiwan

#### Reviewed by:

David Smyth, Monash University, Australia Dennis William Stevenson, New York Botanical Garden, United States

#### \*Correspondence:

Mariana Ferreira Alves marianafealves@gmail.com Juliana Lischka Sampaio Mayer jimayer@yahoo.com.br

#### Specialty section:

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

Received: 05 July 2019 Accepted: 17 October 2019 Published: 29 November 2019

#### Citation:

Alves MF, Pinheiro F, Niedzwiedzki MP and Mayer JLS (2019) First Record of Ategmic Ovules in Orchidaceae Offers New Insights Into Mycoheterotrophic Plants. Front. Plant Sci. 10:1447. doi: 10.3389/fpls.2019.01447

1 **106** their position in the ovary, size, curvature, number and thickness of integument, funiculus length, and degree of vascularization (Endress, 2011). For angiosperms, there are records of bitegmic, unitegmic, and ategmic species (Bouman, 1984; Endress, 2011). Although most angiosperms are bitegmic, variation in the number and thickness of integuments can be observed at different taxonomic levels, such as in families and genera. For example, in Olacaceae, there are described bitegmic, unitegmic, and ategmic species (Brown et al., 2010). In Melastomataceae, ovules are bitegmic; however, in species of the same genus, the number of the outer integument layers can vary from two to many (Caetano et al., 2018).

Previous studies have described the main function of integuments as the delimitation of the micropyle, and protection to the embryo sac and embryo (Herrero, 2001); however, they may also have other functions in species of different families. For example, the inner epidermis of the inner integument can function as a secretory tissue, playing a role in the nutrition of the embryonic sac. This layer of cells is known as integumentary tapetum (Kapil and Tiwari, 1978). Another hypothesis is that the number of integument layers could be related to the fruit type and seed dispersal mode. A study performed with several species of Melastomataceae tried to confirm if there was a relationship between ovules with multiseriate outer integument and fleshy fruits (Caetano et al., 2018). The data obtained did not confirm this relation; however, ancestral state reconstruction shows a tendency for ovules with multiseriate outer integument to occur in fleshy fruit clades. Recent studies conducted with *Arabidopsis* show that the number of ovule integument layers is related to gene and hormone expression (Bencivenga et al., 2012; Gomez et al., 2016; Coen and Magnani, 2018) and may be responsible for the seed coat diversity observed in angiosperms. After fertilization, the integument layers go through different pathways to establish a protective barrier for the embryo (Windsor et al., 2000). There is an immense diversity in seed structure, such as size, color, texture, and shape; this diversity is related to dispersal and germination strategies (Boesewinkel and Bouman, 1984), and may have been initially determined by the arrangement and number of ovule integuments.

In most angiosperms, the formation of ovules is complete when anthesis starts. In Orchidaceae, however, a different pattern is observed, where in the development of ovules and their respective placental proliferation are conditioned to the pollination event (Swamy, 1949a). In general, orchids have low reproductive success because of low pollination rates (Cozzolino and Widmer, 2005). Thus, ovules will be produced only if there is guaranteed seed formation, in order to prevent unnecessary energy expenditure (Arditti, 1992). The formation of the integuments in orchids occurs simultaneously with the formation of the embryonic sac. To date, in most species studied, the embryonic sac is bitegmic (Swamy, 1949a; Yeung and Law, 1997; Mayer et al., 2011). However, four unitegmic species, which are all mycoheterotrophic, have been described in previous studies (Abe, 1976; Arekal and Karanth, 1981; Krawczyk et al., 2016; Li et al., 2016).

Mycoheterotrophic plants are aclorophyllated and are completely dependent on carbon available through their association with fungi throughout their life cycle (Leake, 1994). Recent phylogeny using plastid and mitochondrial genomes in Orchidaceae show that mycoheterotrophic species evolved several times independently (Li et al., 2019). In the family, 235 species with this condition are described (Merckx et al., 2013), and little is known about the reproductive process of these species. Owing to the importance of the seed coat in the life cycle of plants, and because it is considered a stable characteristic, understanding its structure and development can reveal information relevant to its functional aspects, such as dispersal and seed germination (Bouman, 1984; Windsor et al., 2000; Endress, 2011). Thus, the objective of this work was to analyze the origin and development of the seed coat of the mycoheterotrophic orchid *Pogoniopsis schenckii* Cogn, and to compare this development with that of other species in the family that have chlorophyll and present different mechanisms of seed dispersal, *Polystachya estrellensis* Rchb.f., *Elleanthus brasiliensis* Rchb.f., *Isochilus linearis* (Jacq) Barb.Rodr., and *Cleistes libonii* (Rchb. f.) Schltr.—species that exhibit anemochory, and *Vanilla planifolia* Jacks. ex Andrews and *Vanilla palmarum* (Salzm ex. Lindl.) Lindl.—species showing evidence of zoochory (Cribb, 1999). *Pogoniopsis schenckii* is an endemic mycoheterotrophic species found in the Brazilian Atlantic Forest. Prior studies indicate a tendency of reduction in the number of integuments in species of mycoheterotrophic plants, including orchids (Abe, 1976; Arekal and Karanth, 1981; Maas and Ruyters, 1986; Bouman et al., 2002; Endress, 2011; Krawczyk et al., 2016; Li et al., 2016). Thus, our hypothesis is that *P. schenckii* also exhibits reduction in the number of integuments, leading to a greater exposure of the ovule and simplification of the seed coat involving the embryo, which may facilitate the penetration of fungal hyphae. In this context, structural information on the reproductive organs of mycoheterotrophic species, especially *P. schenckii*, can contribute to the elucidation of processes related to the symbiosis between fungi and mycoheterotrophic species. In addition, since the mode of seed dispersal of *P. schenckii* is not known, characterization of the stages of development of its seeds can contribute to the understanding of the ecological interactions involved in the dispersal and colonization of new habitats.

### MATERIAL AND METHODS

### Species Studied and Literature Review

*Pogoniopsis schenckii* Cogn. -Epidendroideae- is aclorophyllated and remains underground for almost its entire life cycle. During its reproductive phase, a floral stem appears above ground level; afterwards, flowers and fruits develop. *Polystachya estrellensis* Rchb. f., *Elleanthus brasiliensis* Rchb. f., *Isochilus linearis* (Jacq) Barb. Rodr., belonging to the subfamily Epidendroideae, and *Cleistes libonii* (Rchb. f.) Schltr., *Vanilla planifolia* Jacks. ex Andrews, and *Vanilla palmarum* (Salzm ex. Lindl.) Lindl., belonging to the subfamily Vanilloideae, are photosynthetic species that are found in all Brazilian regions, and in different phytogeographical domains, such as the Cerrado, Atlantic Forest, and Amazon. Voucher specimens were deposited at the Herbarium of the University of Campinas (HUEC), Campinas, São Paulo, Brazil, and the registration numbers are: 196921, 205027, 161354, 197343, 205047, 205028, and 20502.

A literature review was carried out to verify the number of species that developed bitegmic ovules, and the number of species that developed unitegmic ovules. The following keywords were used to search publication databases: embryo development in Orchidaceae, embryology in Orchidaceae, and integuments in Orchidaceae.

#### Anatomic Analyses

To analyze the integument development freshly opened flowers off all species were collected. For *E. brasilliensis* and *V. palmarum* fruits from natural pollinations at different developmental stages were collected. For *P. schenckii* freshly opened flowers were marked and monitored. For the other species we carried out experimental self-pollination in flowers during the first day of anthesis. The flowers were then monitored and fruits at different stages of development were collected with 15, 20, 25, 30, 60, and 90 opening flower/days after pollination. All material was fixed in Karnovsky (Karnovsky, 1965), dehydrated in serial dilutions of ethanol, and were infiltrated with hydroxyethylmethacrylate (Gerrits and Smid, 1983). The samples were sectioned at 4 μm thickness using a Leica RM2245 rotary microtome, stained with Toluidine Blue 0.05% in phosphate buffer, pH 4.5 (Sakai, 1973), and mounted using Entellan® synthetic resin (Merck®). The slides were analyzed under an Olympus BX51 optical microscope and photographed with an Olympus DP71 digital camera.

#### Scanning Electron Microscopy

Botanical material was fixed in Karnovsky's solution (Karnovsky, 1965), dehydrated using a serial dilution of ethanol and critical point dried under carbon dioxide (CO2) in a Balzers model CPD 030 Critical Point Dryer. The material was then mounted on metal supports and coated with colloidal gold for 220 s on the Bal-Tec model SCD 050 equipment. Analysis and electron micrograph recordings were performed using a LEO VP 435 scanning electron microscope at 20 kV, at the Institute of Biology/UNICAMP.

#### Transmission Electron Microscopy

To analyze the changes occurring during the development of *P. schenckii* seed coat, ovules, and seeds at different stages of development were fixed using 2.5% glutaraldehyde in 0.2 M sodium cacodylate buffer, pH 7.25, for 24 h (Mc Dowel and Trump, 1976). Post-fixation was performed with 1.0% osmium tetroxide (OsO4) in sodium cacodylate buffer for 12 h in the dark (Gabriel, 1982). The material was dehydrated in a series of increasing concentrations of acetone solution and soaked in LR White® resin according to the manufacturer's instructions. The ultrafine sections were prepared with Leica ultramicrotome using diamond knife. The ultrafine sections were contrasted with uranyl acetate (Bozzola and Russel, 1998) and lead citrate (Hanaich et al., 1986), examined under a Philips EM 100 transmission electron microscope at 80 kV, and documented with Eastman Kodak 5302 (35 mm) film.

### RESULTS

Literature review showed that 97 species of orchids of different subfamilies, 2 species of Vanilloideae, 8 species of Cypripedioideae, 31 species of Orchidoideae, and 56 species of Epidendroideae, had been evaluated till date. Regarding these species, 93 presented bitegmic ovules, and 4 presented unitegmic ovules (**Table 1** and

**Supplementary Table 1**). With the 6 new species included in this study, a total of 103 species have been evaluated in terms of the type of ovule integuments.

All species analyzed in this study are teninucellate and bitegmic, except for *P. schenckii* (**Figures 1A**–**J**), which has an ategmic ovule. *P. schenckii* is the first species within Orchidaceae possessing this characteristic. There is no growth of integuments at any stage of development (**Figures 1A**, **B**, **I**, **J**). The embryonic sac is only covered by the nucellar epidermis, which exhibit cells with high metabolism, thin walls, and evident nucleus (**Figures 1A**, **B**, **E**, **F**). Owing to the absence of integuments, there is no micropyle formation in the ovule of the species (**Figures 1B**, **I**, **J**, **K**).

In other species, the development of integuments occurs from periclinal divisions in the epidermal cells located at the base of the ovules, which is seen in both the subfamily Epidendroideae (**Figures 2A**–**P**) and in the subfamily Vanilloideae (**Figures 3A**–**Q**). During ovule development, the inner integument grows and recovers the nucellar epidermis. In *P. estrellensis*, *E. brasiliensis*, *I. linearis,* and *C. libonii*, the outer integument grows and covers the inner integument (**Figures 2A**–**D, G**–**L, M**–**O** and **3A**–**E**). In *P. estrellensis*, the outer integument has three layers (**Figure 2C**). The initial development of the integuments in *V. planifolia* and *V. palmarum* show that the outer integument also has three layers (**Figure 3P**). In other species, the outer integument has two layers, and in all the species analyzed, the inner integument has two layers (**Figures 2M** and **3B**).

After fertilization, changes are observed in the integument of all species. In *P. schenckii* (i.e., about 25 days after the floral opening), the seed coat becomes hard and forms from the nucellar epidermis itself (**Figures 1C**, **D**, **G**, **H**, **L**). It is possible to observe that during development of the seed coat there is an accumulation of substances that confers the cytoplasm a dense aspect (**Figures 1G, H**). When mature, the seed presents a brown-colored integument (**Figure 4A**). In *P. estrellensis*, *E. brasiliensis,* and *I. linearis*, the inner integument is fully absorbed, and the outer integument undergoes elongation. In the outer integument, the inner layer is absorbed, and the outer layer gives rise to the seed testa, which, when mature, is impregnated with lignin and surrounds the embryo (**Figures 2D–F, P** and **4B**–**D**). In *C. libonii*, at the beginning of seed development (i.e., about 40 days after fertilization), the inner layer of the inner integument begins to possess a dense cytoplasm and an evident nucleus (**Figures 3E**, **F**). Sixty days after fertilization, it is possible to observe the presence of a natural yellow-colored secretion surrounding the embryo (**Figures 3G**, **H** and **4E**). In the mature seed, only the outer tegument and this secreted layer that covers the embryo remain as coat, whereas the inner layer of the internal integument is reabsorbed (**Figures 2E**, **P** and **4E**). This substance is probably secreted by the cells of the inner layer of the inner integument. In *V. planifoli*a and *V. palmarum*, the mature seed has a hard dark-colored coat (**Figures 3M**, **Q** and **4F**, **G**).

### DISCUSSION

For the first time the presence of ategmic ovules, as observed in *P. schenckii*, are described in Orchidaceae. Bitegmic ovules are commonly in orchids (Swamy, 1949a), but a reduction in ovule integuments are commonly observed in mycoheterotrophic


#### TABLE 1 | Continued


Ateg, ategmic ovules; biteg, bitegmic ovules; uniteg, unitegmic ovules.

species, which are unitegmic (Tohda, 1967; Abe, 1976; Arekal and Karanth, 1981; Krawczyk et al., 2016; Li et al., 2016). Anatomical analyses show that the ovules of *P. schenckii* develop normally, and that there is no evidence of development of integument in the ovules of the species at all time points. In this way, the nucellar epidermis is responsible for surrounding the embryo sac, and in the mature seed, for surrounding the embryo. This result differs from the pattern found in other species analyzed in this study, which had seed coats originating from the outer integument.

Reduction of integuments occurs independently in different groups. They have been described in mycoheterotrophic species of Gentianales (Gentianaceae), parasite species of Santalales (Balanophoraceae, Loranthaceae, Olacaeae, and Santalaceae), and in a photosynthetic species of Aquifoliales (Cardiopteridaceae) (Maas and Ruyters, 1986; Bouman et al., 2002; Brown et al., 2010; Polli et al., 2016; Sato and Maria Gonzalez, 2016; Suaza-Gaviria et al., 2016; Tobe, 2016; Gonzalez et al., 2019). Molecular studies show that in ategmic ovules of Santalales, the genes associated with the expression of the integument are expressed in the periphery of the ovary, and that the reduction found in these species is the result of the fusion between the integument and the nucellus (Brown et al., 2010). In *P. schenckii*, the reduction of integuments leads to a total loss of the micropyle. However, this structural reduction does not seem to compromise reproduction, since the synergids continue to secrete substances for pollen tube attraction. The absence of integuments could facilitate the penetration of the synergids (**Figure 1B**) and subsequent fertilization. Mycoheterotrophic orchids have ovules with simpler structures, and the absence of a distinct micropyle is common in unitegmic species (Tohda, 1967; Abe, 1976; Arekal and Karanth, 1981; Li et al., 2016). The micropyle is responsible for directing the pollen tube; moreover, both the micropyle and secretions released by the synergid that promotes pollen tube attraction facilitate fertilization (Cheung and Wu, 2001; Okuda et al., 2009; Chen and Fang, 2016).

FIGURE 1 | Longitudinal sections of ovules and seeds of Pogoniopsis schenckii. (A) Megaspore mother cell. Note that there is no formation of integuments. (B) Penetrated synergid, egg cell, and polar nuclei. (C) Young seed. (D) Embryo with five cells. (E–H) Transmission microscopy electromyography. (E) Embryo sac and nucellar epidermis. (F) Details of the nucellar epidermis which presents cells with thin walls and evident nucleus. (G) Seed coat that originates from the nucellar epidermis. (H) Details of the hard seed coat. Note their accumulation of substances that confers the cytoplasm a dense aspect. (I–L) Scanning microscopy electromyography. (I–K) Ovule in development. Note that there is no formation of integuments. (L) Seed. cw, cell wall; ec, egg cell; em, embryo; f, funiculus; mc, megaspore mother cell; ov, ovule; ne, nucellar epidermis; nu, nuclei; pn, polar nuclei; ps, penetrated synergid; sc, seed coat; se, seed; sy, synergid; va, vacuolo. Scale bars A, C, I = 20 µm; B, D, K, L = 50 µm; E, J = 10 µm; F–H = 2 µm.

The differentiation of ovules in the species studied occurred after the stimulation of pollination, and the development of the integuments in *P. estrellensis*, *I. linearis*, *E. brasiliensis*, *C. libonii, V. planifolia,* and *V. palmarum* occurs simultaneously with the events of the megasporogenesis, as observed in other species of the family (Swamy, 1949a; Sood, 1985; Sood, 1986; Sood and Rao, 1986; Mayer et al., 2011; Li et al., 2016; Duarte et al., 2019). In most orchids, the outer integument has two layers of cells (Swamy 1949a; Wirth and Withner, 1959). However, in *P. estrellensis,* the outer integument was observed to have three layers, and the *Vanilla* species presented ovules with outer integuments that had three to four layers of cells in *V. imperialli*s, and four to six layers of cells in *V. planifolia* (Swamy, 1947; Nishimura and Yukawa, 2010; Kodahl et al., 2015). It is believed that the outer multiseriate integument in *Vanilla* would be related to the larger size of the seed found in the species of the genus (Kodahl et al., 2015).

Of all species analyzed, *P. schenckii*, *V. planifolia,* and *V. palmarum* have seeds with hard coat. Preliminary results show that in *P. schenckii*, dispersal in the species is very restricted (personal data). It was found to not be related to anemochory; moreover, dispersal by animals was not observed. Based on this, it seems that the hard coat has other unknown functions. Population genetics studies have been conducted, seeking to understand how this restricted dispersal can affect the dynamics of the populations of the species (Alves, unpublished data). Besides the hard coat *P. schenckii* presents seed with a large funiculus, differing from the other analyzed species. Preliminary analyzes show that the funiculus assists in the penetration of fungal hyphae after dispersion (personal

FIGURE 2 | Longitudinal sections of ovules and seeds of Epidendroideae species. (A–F) Ovules and seeds of Polystachya estrellensis. (D, F) Scanning microscopy electromyography. (A) Differentiation of the initial archesporial cell. (B) Megaspore mother cell and formation of integuments. (C) Embryo sac. \*indicates the outer integument with three layers. (D) Embryo sac. (E, F) Seed. (G–L) Ovules of Elleanthus brasiliensis. (J, L) Scanning microscopy electromyography. (G) Differentiation of the initial archesporial cell. (H) Formation of integuments indicated by arrows. (I) Megaspore mother cell. (J–K) Formation of integuments. (L) Embryo sac with the outer integument developed. (M–P) Ovules and seeds of Isochilus linearis. (M) Megaspore mother cell and formation of integuments. (N, O) Embryo sac with the integuments developed. P. Seed. ac, initial archesporial cell; em, embryo; f, funiculus; ii, inner integument; mc, megaspore mother cell; mi, micropyle; ne, nucellar epidermis; oi, outer integument; wt, wall thickening. Scale bars A–C; E; G-K; M-P = 20 µm; D, F = 50 µm; L = 10 µm.

data). Seeds with hard coat have also been described for other mycoheterotrophic orchids. For *Cyrtosia japonica*, seeds with coats originating from the outer integument and inner integument (Yang and Lee, 2014) are registered. In *C. japonica*, the seed presents an outer integument with four layers, and the outermost layer later becomes sclerified (Yang and Lee, 2014). It is suggested that the observed lignification protects the embryo when it passes through the alimentary tract of its dispersers (Rodolphe et al., 2011; Yang and Lee, 2014). In *Yonia japonica,* the seed also presents a lignified coat; however, the fruits and seeds of the species are dispersed by insects (Suetsugu, 2018). Similar to *Cyrtosia,* it is believed that the lignified seed coat in *Y. japonica* is an adaptation that protects the seed during digestion (Suetsugu, 2018).

It is assumed that *V. planifolia* and *V. palmarum* undergo endozoochory dispersal (Cribb, 1999; Kodahl et al., 2015). In these species, as in others belonging to Vanilloideae, seeds with hard coats exist as a strategy for the dispersal of the genus (Kodahl et al., 2015); in addition, hard coats would protect seeds that can be dispersed over long distances. In other species, seeds were observed to have a thin and transparent coat, which is seemingly a common feature in Orchidaceae. Seeds from Orchidaceae have small sizes, and are called "dust seeds" (Swamy, 1949b; Arditti and Ghani, 2000). The rather small size observed in orchid seeds was traditionally thought to be an adaptation to long-distance wind dispersal events (Arditti and Ghani, 2000). However, recent molecular studies have shown discordant patterns that show the orchid seeds ability to reach long distances (Cozzolino et al., 2003; Trapnell and Hamrick, 2004). Many other species present dispersal patterns limited to a few meters (Chung et al., 2004; Chung et al., 2005; Ren et al., 2017).

FIGURE 3 | Longitudinal sections of ovules and seeds of Vanilloideae species. (A–I) Ovules and seeds of Cleistes libonii. (C, D, I) Scanning microscopy electromyography. (A) Differentiation of the initial archesporial cell. Arrows indicate the initial formation of integuments. (B) Megaspore mother cell and formation of integuments. (C, D) Development of integuments. (E) Initial development of embryo. (F–G) Embryo. \*indicate the extracellular exudate. (H–I) Seeds. (J–N) Ovules and seeds of Vanilla planifolia. (L, N) Scanning microscopy electromyography. (J–K) Megaspore mother cell and formation of integuments. (L) Development of integuments. (M, N) Seeds with hard dark-colored coat. (O–Q) Ovules and seed of Vanilla palmarum. (O) Differentiation of the initial archesporial cell. Arrows indicate the initial formation of integuments.(P) Megaspore mother cell and formation of integuments. Note the outer integument with three layers. (Q) Seed with hard dark-colored coat. ac, initial archesporial cell; em, embryo; f, funiculus; ii, inner integument; il, inner layer of the inner integument; mc, megaspore mother cell; mi, micropyle; ne, nucellar epidermis; oi, outer integuments; ol, out layer of the inner integument; wt, wall thickening. Scale bars A, B, E–G, J–M, O–Q = 20 µm; C = 50 µm; D, H, N = 100 µm; I = 200 µm.

Dark-colored hard seed coats, as observed in *P. schenckii*, *V. planifolia,* and *V. palmarum,* have already been described for *Apostasia* (Swamy, 1947). Occurrence of phytomelanin deposition in Asparagales seeds is described in the literature (Dahlgren et al., 1985). Phytomelanin is a dark and insoluble pigment that is found in different parts of plants and exhibit distinct transport load and structural stability (Nicolaus et al., 1964; Cordero and Casadevall, 2017). Phytomelanin's main function is to confer protection to different conditions, such as environmental variations, harmful radiation, extreme temperatures, and chemical and mechanical stress (Roulin, 2014; Cordero and Casadevall, 2017). The darkcolored integument in the species studied may result from phytomelanin deposition. However, studies are still needed to clarify this issue.

The results obtained show novelties in the development of the seed coat in Orchidaceae. *P. schenckii* has an ategmic ovule and has a hard seed coat that originates from the nucellar epidermis. Mycoheterotrophic plants have numerous modifications in their morphology, reproductive biology, and physiology (Leake, 1994; Bidartondo, 2005), the most prominent among loss of photosynthetic function and severe ruptures in the plastid genome (Graham et al., 2016). The genomic losses observed may be related not only to photosynthetic processes, but also to the absence of genes that present other functions, such as genes related to reproductive functions. Anatomical analyses show that there is no evidence of integument development in the ovules of *P. schenckii.* Thus, the reduction of the integuments found in the species may be due to the absence of gene expression, or even the absence of genes linked to the development of the integument; however, molecular studies are necessary to elucidate this issue.

#### DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

### REFERENCES


## AUTHOR CONTRIBUTIONS

MA carried the anatomical analysis, analysis of scanning and transmission microscopy, and writing the manuscript. FP supervised the work and writing the manuscript. MN carried anatomical analyses in *Cleistes libonii* and JM was responsible for collecting the material, carried anatomical analyses, writing the manuscript, and supervised the work.

### FUNDING

This study was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil (CAPES) - Finance Code 001. Juliana Lischka Sampaio Mayer thank FAPESP (2015/26479-6), FAEPEX 0944/14, CNPq (447453/2014-9), and CNPQ (310184/2016-9) for funding support.

### ACKNOWLEDGMENTS

We thank *The Espaço da Escrita - Pró-Reitoria de Pesquisa - UNICAMP - for the language services provided* and the access to equipment and assistance provided by the Electron Microscope Laboratory (LME/UNICAMP). We thank the Instituto Florestal (Parque Estadual da Serra do Mar, Núcleo Santa Virginia and Núcleo Picinguaba) for the development of the study on protected public land and Carlos Eduardo Pereira Nunes for his help in field activities.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2019.01447/ full#supplementary-material

SUPPLEMENTARY TABLE 1 | List of the type of integuments in species of Orchidaceae. Ateg = ategmic ovules; biteg = bitegmic ovules; uniteg = unitegmic ovules.


the transcription factor ATS. *Plant Physiol.* 172, 2403–2415. doi: 10.1104/ pp.16.01231


**Conflict of Interest:** 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 Alves, Pinheiro, Niedzwiedzki and Mayer. 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.*

# PeERF1, a SHINE-Like Transcription Factor, Is Involved in Nanoridge Development on Lip Epidermis of Phalaenopsis Flowers

Pei-Han Lai <sup>1</sup>† , Li-Min Huang1† , Zhao-Jun Pan<sup>2</sup> , Wann-Neng Jane<sup>3</sup> , Mei-Chu Chung<sup>3</sup> , Wen-Huei Chen<sup>4</sup> and Hong-Hwa Chen1,4,5\*

#### Edited by:

Jen-Tsung Chen, National University of Kaohsiung, Taiwan

#### Reviewed by:

Hirokazu Tanaka, Meiji University, Japan Chao Bian, Beijing Genomics Institute (BGI), China

#### \*Correspondence:

Hong-Hwa Chen hhchen@mail.ncku.edu.tw † These authors have contributed equally to this work

#### Specialty section:

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

Received: 23 July 2019 Accepted: 04 December 2019 Published: 30 January 2020

#### Citation:

Lai P-H, Huang L-M, Pan Z-J, Jane W-N, Chung M-C, Chen W-H and Chen H-H (2020) PeERF1, a SHINE-Like Transcription Factor, Is Involved in Nanoridge Development on Lip Epidermis of Phalaenopsis Flowers. Front. Plant Sci. 10:1709. doi: 10.3389/fpls.2019.01709 <sup>1</sup> Department of Life Sciences, National Cheng Kung University, Tainan, Taiwan, <sup>2</sup> Institute of Ecology and Evolutionary Biology, National Taiwan University, Taipei, Taiwan, <sup>3</sup> Institute of Plant and Microbial Biology, Academia Sinica, Taipei, Taiwan, <sup>4</sup> Orchid Research and Development Center, National Cheng Kung University, Tainan, Taiwan, <sup>5</sup> Institute of Tropical Plant Sciences, National Cheng Kung University, Tainan, Taiwan

Phalaenopsis orchids have a spectacular floral morphology with a highly evolved lip that offers a landing platform for pollinators. The typical morphological orchid lip features are essential for the special pollination mechanism of Phalaenopsis flowers. Previously, we found that in the lip, a member of the AP2/EREBP protein family was highly expressed. Here, we further confirmed its high expression and characterized its function during lip development. Phylogenetic analysis showed that AP2/EREBP belongs to the Va2 subgroup of ERF transcription factors. We named it PeERF1. We found that PeERF1 was only expressed at stage 5, as flowers opened. This coincided with both thickening of the cuticle and development of nanoridges. We performed knockdown expression of PeERF1 using CymMV-based virus-induced gene silencing in either the AP2 conserved domain, producing PeERF1\_AP2-silenced plants, or the SHN specific domain, producing PeERF1\_SHN-silenced plants. Using cryo-SEM, we found that the number of nanoridges was reduced only in the PeERF1\_AP2-silenced group. This change was found on both the abaxial and adaxial surfaces of the central lip lobe. Expression of PeERF1 was reduced significantly in PeERF1\_AP2-silenced plants. In cutin biosynthesis genes, expression of both PeCYP86A2 and PeDCR was significantly decreased in both groups. The expression of PeCYP77A4 was reduced significantly only in the PeERF1\_AP2-silenced plants. Although PeGPAT expression was reduced in both silenced plants, but to a lesser degree. The expression of PeERF1 was significantly reduced in the petal-like lip of a big-lip variant. PeCYP77A4 and PeGPAT in the lip were also reduced, but PeDCR was not. Furthermore, heterologous overexpression of PeERF1 in the genus Arabidopsis produced leaves that were shiny on the adaxial surface. Taken together, our results show that in Phalaenopsis orchids PeERF1 plays an important role in formation of nanoridges during lip epidermis development.

Keywords: AP2/EREBP, cutin biosynthesis genes, lip, nanoridge, orchid, Phalaenopsis, transcription factor

#### INTRODUCTION

Phalaenopsis orchids are renowned for their unique and elegant floral morphology and long florescence duration. Recently, they have become the model Orchidaceae research plants. Two databases for genetic information have been established, OrchidBase 3.0 and Orchidstra 2.0 (Fu et al., 2011; Su et al., 2013a; Tsai et al., 2013; Chao et al., 2017). The floral morphology of Phalaenopsis orchids includes three sepals, three petals, and one column. The column is formed by fusion of the style and a part of the androecium. The outer two perianth whorls are typically petaloid and are referred to as tepals. Instead of producing three uniform petals, Phalaenopsis flowers have a highly evolved, modified, and resupinated inner medium petal, the lip (Rudall and Bateman, 2002; Tsai et al., 2004; Tsai and Chen, 2006; Tsai et al., 2008; Pan et al., 2011; Pan et al., 2014). This lip is understood to play an important role both in pollination and evolution (Robinson and Burns-Balogh, 1982; Cozzolino and Widmer, 2005; Mondragón-Palomino and Theißen, 2008), as it provides a platform for pollinators.

Lip morphogenesis consists of five stages, from the embedded stage 1 to the open flower of stage 5. There is no division of the lip at stage 1. However, in stage 2, the lip divides quickly into three distinct parts. There are two lateral lobes, one central lobe, and one callus. The split lip forms a tunnel-like structure in the mature flower (Figure 1A, Li), a feature related to evolved pollination strategies (Cozzolino and Widmer, 2005). Although Phalaenopsis orchids exhibit unique floral morphological features, Cryo-scanning electron microscopy (Cryo-SEM) has revealed that the perianth lip epidermis has a unique functional morphology not found in sepals or petals (Pan et al., 2011; Hsieh et al., 2013a; Hsieh et al., 2013b; Pan et al., 2014; Hsu et al., 2015b). Heavy and dense nanoridges cover the lip epidermis. These undulated nanostructures, also called cuticular folds, are assumed to contain cuticular lipids (Koch et al., 2008; Koch et al., 2009a; Koch et al., 2009b).

Diverse perianth and floral epidermis adaptations have evolved in both eudicots and monocots. In many flowering plants, the epidermal surfaces of sepals and petals display a range of patterns in combination with diverse micro- and nanostructures (Kay et al., 1981; Whitney et al., 2009; Whitney et al., 2011b; Kourounioti et al., 2013). More than 75% of petal epidermal cells of angiosperms are conical or papillate, usually on the adaxial side where potential pollinators would be found (Kay et al., 1981; Whitney et al., 2009; Whitney et al., 2011a; Whitney et al., 2011b). Moreover, in many plants, sepal and petal epidermal cells are covered with various density and orientation of nanoridges (Jeffree, 2006). These structures on the surface of sepal and petal epidermal cells are believed to attract pollinators and enhance pollination success through visual signals (Whitney et al., 2009; Whitney et al., 2011b; Kourounioti et al., 2013; Moyroud et al., 2017) and act as tactile signals affecting pollinator movement (Prüm et al., 2011; Rands et al., 2011; Prüm et al., 2012; Prüm et al., 2013; Adachi et al., 2015). Moreover, cell surface cuticle structures can strengthen cells and thereby function many ways in plant development as well as survival and defense in unfavorable environments, such as under biotic or abiotic stress (e.g., dehydration, pathogens, UV light, frost, and insect attacks) (Koch et al., 2008; Koch et al., 2009a; Koch et al., 2009b).

The first identified transcription factors (TFs) that regulate cuticle biosynthesis are SHINEs/WAX INDUCERs (SHNs/ WINs), members of the V group of the ethylene responsive factor (ERF) subfamily of the apetala2/ethylene response element binding protein (AP2/EREBP) TF family (Aharoni et al., 2004; Broun et al., 2004; Nakano et al., 2006). In Arabidopsis SHINE gain-of-function mutant (shn) and plants overexpressing AtWIN1/SHN1, AtSHN2, or AtSHN3 have shiny leaves and increased accumulation of epidermal wax on the top of leaves as compared with wild type (Aharoni et al., 2004; Broun et al., 2004). By co-silencing all three AtSHN clade members, SHNs redundantly regulate the formation of petal surface nanoridges and also cell elongation, adhesion, and separation (Shi et al., 2011). Recently, increased number of SHN-like TFs that belong to the ERF-V group have been identified and they exhibit various functions during plant physiological processes.

The ERF-V group includes two subgroups: Va and Vb. The Va subgroup contains two conserved motifs of the conserved middle motif (CMV-1) and C-terminal motif (CMV-2), which the Vb subgroup does not contain (Nakano et al., 2006). The ERF-Va subgroup is further divided into two subgroups, Va1 and Va2, containing a complete or incomplete CMV-1 motif, respectively (Nakano et al., 2014). Functional characterization of genes in the Va1 subgroup from several plants indicates that they are involved in cuticle development. Arabidopsis AtSHNs regulate cuticle formation (Aharoni et al., 2004; Broun et al., 2004; Shi et al., 2011).

Similar research has been done with several plant species. In barley it was found that HvNud is involved in the lipid biosynthesis of the grain surface, which produces hulled caryopses (Taketa et al., 2008). Tomato SlSHINE3 is involved in cutin metabolism of fleshy fruit epidermal cells for patterning the epidermal surface (Shi et al., 2013) and SlSHN1 is involved in wax accumulation of leaf epidermal cells, which enhances drought tolerance (Al-Abdallat et al., 2014). Rice OsAP2/ERF- "N-22" is involved in wax biosynthesis and also enhances drought resistance (Mawlong et al., 2014), while wheat TdSHN1 is involved in the cuticle formation of leaf surfaces (Jäger et al., 2015). Eucalyptus EgrSHN1 and EgrSHN2 are involved in cell wall biosynthesis of flowers (Marques et al., 2013). Hence, the complete CMV-1 and CMV-2 motifs of ERF-Va1 genes are deemed the SHINE domains. SHINE domains are considered to be important in cuticle development. The subgroup Va1 is identified as the SHINE clade (Aharoni et al., 2004).

In contrast, the Va2 subgroup, with an incomplete CMV-1 motif, is involved in various physiological processes: Arabidopsis At5g25190 is induced by 1-aminocyclopropane-1-carboxylicacid (ACC) and salt (NaCl) and was named the ethylene- and saltinducible ERF gene (AtESE3) (Zhang et al., 2011), but its

Abbreviations: AF, anticlinal field; CF, central field; Cryo-SEM, Cryo-scanning electron microscopy; ERF, ethylene responsive factor; RT-PCR, reverse transcription–polymerase chain reaction; SHN/WIN, SHINE/WAX INDUCER; TFs, transcription factors; VIGS, virus-induced gene silencing.

Co, column. (A) Schematic diagrams of the cell structural division of a single cell surface. The central field (CF) is in the center middle of the cell surface and the anticline field (AF) is between the middle area and the boundary of the cell. Anticlines directly connected to the cell area (modified after Koch et al., 2009a). (A) Schematic diagrams of epidermis that built up the cell hierarchical structures with 2-D smooth wax films or decorated with protuberance, nanoridges, and nodelike nanoridges. (D–O) Cyro-SEM top view and side view of epidermal cells of perianth organs. White arrows indicate a protuberance on the top of the cell. Black arrows indicate node-like nanoridges in the CF of the cell. White arrowheads indicate parallel nanoridges in the AF of the cell. Scale bars, 30 mm. (P–Q) Cross section of adaxial epidermal cells of petal and lip. Scale bars, 10 mm. (R–S) Lipophilic dye "Sudan Black B" staining on the adaxial epidermal cells of petal and lip. Scale bars, 100 mm.

overexpression confers no typical SHINE phenotype (Aharoni et al., 2004). Tomato LeERF1 regulates fruit ripening and softening (Li et al., 2007). Populus PtaERF003 is involved in lateral root formation (Trupiano et al., 2013). Eucalyptus Egr33m and Egr40m are involved in wood cell wall biosynthesis (Marques et al., 2013). Tomato SlERF52 regulates flower pedicel abscission (Nakano et al., 2014). Whereas, NvERF045 in berries regulates berry ripening and is also involved in cuticle development (Leida et al., 2016). Moreover, the Vb subgroup At5g19790 does not contain CMV-1 and CMV-2 motifs and is important in low potassium signaling (Kim et al., 2012).

Gene associated with cutin biosynthesis for epidermal nanoridge formation include CYP86A and CYP77A, members of a cytochrome P450 family, glycerol-3-phosphate acyltransferase 6 (GPAT), and defective in cuticular ridges (DCR) (Kannangara et al., 2007; Li-Beisson et al., 2009; Panikashvili et al., 2009; Shi et al., 2011; Shi et al., 2013; Petit et al., 2016; Mazurek et al., 2017). GPAT6 and CYP77A6 are for the formation of floral cutin in Arabidopsis thaliana (Li-Beisson et al., 2009). CYP86A4 has been reported as one of the downstream target genes SHN (Shi et al., 2011; Shi et al., 2013). It has been shown that DCR-deficient plants have defective cuticle formation with altered epidermal cell differentiation (Panikashvili et al., 2009). This defective formation in reproductive and vegetative tissues was correlated with low abundance of 9(10),16-dihydroxyhexadecanoic acid in the cutin polymer of DCR (At5g23940)-deficient plants (Panikashvili et al., 2009).

We previously identified several unigenes dominantly expressed in the Phalaenopsis lip (Hsiao et al., 2013). Among them, one member of the AP2 family, P. equestris ethylene responsive factor 1 (PeERF1), was found to be most similar to the SHINE clade homolog At5g25190.

Here, to extend our understanding of the function of PeERF1 in Phalaenopsis orchids, we analyzed its spatial and temporal gene expression; downregulated PeERF1 expression by using CymMV-based virus-induced gene silencing (VIGS) in orchids; and examined heterologous overexpression of PeERF1 in Arabidopsis for comparison. The abnormal phenotypes of nanoridge sculpture patterns of lip epidermal cells were observed in the somaclonal variant, P. "Join Big foot TH365" containing enlarged petal-like lip mutants. We further investigated the relationship of putative Phalaenopsis orthologs of known cutin biosynthetic genes with the cuticle formation in Phalaenopsis lips. These genes include two cytochrome P450s, PeCYP86A2 and PeCYP77A4, and two putative acyltransferases, PeGPAT and PeDCR. We hope that these results will contribute to the understanding of transcriptional regulation of late-stage orchid lip formation during floral morphogenesis.

#### MATERIALS AND METHODS

#### Plant Materials

We conducted gene spatial expression analysis on specimens of Phalaenopsis equestris obtained from the Taiwan Sugar Corp. (Tainan, Taiwan), and both gene temporal expression and VIGS experiments on the commercial cultivar P. OX Red Shoes "OX1408" obtained from Oxen Biotechnology Corp. (Tainan, Taiwan).

The flower buds of P. OX Red Shoes "OX1408" were divided into five stages by size (Figure 3C). A single raceme spike inflorescence embraces 8–10 flowers. The smallest flower bud is embedded in the tip of an inflorescence and is stage 1 (< 0.5 cm), followed by successive stage 2 (0.5–1 cm), stage 3 (1–2 cm), stage 4 (2–3 cm), and blooming flowers are stage 5 (floral diameter of 12 cm) (Figure 3C). For comparison, we used wild-type flower and big-lip variant flower (petal-like lip) of P. hybrid "Join Big foot" (P. Yu Pin Easter Island x P. I- Hsin Diamond "Join White of Love"). These were provided by Join Orchids Incorporation (Tainan, Taiwan). For VIGS experiments, plants were kept in the greenhouse at the Tainan District Agricultural Research and Extension Station, Council of Agriculture, under a controlled temperature of 27°C/22°C (day/night). Other mature orchid plants were maintained in the greenhouse at National Cheng Kung University under natural light and controlled temperature from 23°C to 27°C.

#### Cryo-SEM

We examined changes in cellular morphology from the 1st to the 8th blooming flowers of silenced plants (stage 5, floral diameter of 12 cm) using Cryo-SEM. Sample preparation and Cryo-SEM examination follows previous research (Hsieh et al., 2013a; Pan et al., 2014). Fresh samples were dissected and loaded on the stub, which was subsequently frozen with liquid nitrogen slush, and then quickly transferred to a sample preparation chamber at -160°C for 5 min. After that time the temperature was raised to -85°C and sublimed for 15 min. Samples were then coated with platinum (Pt) at -130°C and transferred to the cryo-stage in an SEM chamber and observed at -160°C using Cryo scanning electron microscope (FEI Quanta 200 SEM/Quorum Cryo System PP2000TR FEI) with 20 kV.

Images were taken under a Cryo stage at < -160°C. We measured the nanoridge area on 40 flowers, 5 flowers from each of 8 plants using ImageJ (http://rsb.info.nih.gov/ij/). Mean data were compared by Duncan's multiple-range test, using SPSS v17.

### Cloning and Characterization of PeERF1

Using TRIsure reagent (Bioline, UK) total RNA was extracted and then treated with RNase-free DNaseI (Invitrogen, USA) to remove residual DNA. We cloned the full-length cDNA of PeERF1 (accession no. MG948436) using a SMART rapid amplification of cDNA ends (RACE) kit (Clontech, USA). We randomly selected 6 to 8 positive clones for sequencing. For gene expression analysis, quantitative real-time RT-PCR (qRT-PCR) was performed in triplicate and repeated independently three times as previously described (Hsu et al., 2015a). Primers for all the PCR and qRT-PCR experiments are in Supplementary Table S1.

For qRT-PCR, the cDNA template was mixed with 2X SYBR Green PCR master mix (Applied Biosystems, Norwalk, CT, USA) in an ABI 7300 instrument (Applied Biosystems) with three biological replicates. For gene quantification, qRT-PCR was performed at stage 5 flowers of PeERF1-silenced plants in triplicate, and repeated in three silenced plants independently. For PCR reaction, each sample was analyzed in triplicate. Reactions involved incubation at 50°C for 2 min, then 95°C for 10 min, and thermal cycling for 40 cycles (95°C for 15 s and 60°C for 1 min). The relative quantification was calculated according to the manufacturer's instructions (Applied Biosystems). To control the integrity of RNA and normalize target RNA copy numbers in gene-silenced and mock-treated flowers, the housekeeping gene PeActin4 (AY134752) was recruited as an internal control for normalization (Chen et al., 2005).

#### Sequence Alignment and Phylogenetic Analysis

Multiple sequence alignment was generated by using AlignX (Vector NTI advance 11, Invitrogen). The protein sequences of SHN homologous TFs were obtained from the National Center for Biotechnology Information (NCBI) databases (http://www. ncbi.nlm.nih.gov/), and accession numbers are as follows: A. thaliana AtSHN1 (NP\_172988), AtSHN2 (NP\_196680), AtSHN3 (NP\_851073), At5g25190 (NP\_197901.1), and At5g19790 (NP\_197480.1); tomato (Solanum lycopersicum) SlSHN1 (XP\_004235965), SlSHN3 (XP\_004240977), and SlERF52 (BAO18577); tomato (Lycopersicon esculentum) LeERF1 (AAL75809); berry (Vitis vinifera) VvERF045 (ANT73695); barley (Hordeum vulgare L.) HvNud (BAG12386); wheat (Triticum aestivum L.) TdSHN1 (ANY98960); rice (Oryza sativa) OsAP2/ERF-"N-22" (ACU44657), and Populus (P. tremulax P. alba) PtaERF003 (Potri.018G021900). Eucalyptus grandis SHN homologous TFs can be accessed in the Phytozome database (http://www. phytozome.net/cgi-bin/gbrowse/eucalyptus/): EgrSHN1 (Eucgr.C04221.1), EgrSHN2 (Eucgr.C01178.1), Egr33m (Eucgr.C02719.1), and Egr40m (Eucgr.C03947.1) (Marques et al., 2013). These sequences were used to construct phylogenetic trees by using MEGA5.0 (Tamura et al., 2011). Phylogenetic relationships were inferred by the neighbor-joining method and evolutionary distances were computed by the Poisson correction method. Bootstrap values were calculated with 1,000 replicates.

#### Virus-Induced Gene Silencing (VIGS)

VIGS experiments with PeERF1 were performed as previously described (Hsieh et al., 2013a; Pan et al., 2014); 142-nt and 175 nt fragments for the conserved AP2 domain and the specific incomplete "SHINE domains," i.e., CMV-1 and CMV-2 motifs of PeERF1, respectively, were constructed into the pCymMV-Gateway plasmid (Lu et al., 2007). The constructed VIGSsilencing plasmids for producing the PeERF1\_AP2-silenced and PeERF1\_SHN-silenced plants were named pCymMVGateway-PeERF1\_AP2 and pCymMV-Gateway-PeERF1\_SHN, respectively. These plasmids were transformed into Agrobacterium (strain EHA105). For infiltration, Agrobacterium tumefaciens strain EHA105 containing pCymMVGateway-PeERF1\_AP2 or mMV-Gateway-PeERF1\_SHN were grown overnight at 28°C to OD600 = 1. After centrifugation, bacterial cell pellets were resuspended by adding 300 ml MS medium containing 100 mM acetosyringone and allowed to stand at room temperature for 0.5 h. Two methods of Agro-infiltration were used: inflorescence injection and leaf injection. For inflorescence injection, suspensions were injected into the stalk of the raceme with eight internodes and one visible floral bud (extruding out of its bract) by use of a 1-ml syringe with a needle in all silencing treatments. The raceme stalk usually emerges from the stem between the third and fourth leaves. For leaf injection, suspensions were injected into the leaf directly above the emerging inflorescence. Mock-treated plants were recruited as the negative control. They were handled the same and contained an empty vector of a Cymbidium mosaic virus infectious clone with a Gateway system vector. Transformed EHA105 was injected into both inflorescence spikes and the leaf directly above the emerging inflorescence. For VIGS, eight independent PeERF1\_AP2-silenced and eight PeERF1\_ SHN-silenced plants as well as eight mock control plants were generated, and repeated twice independently. qRT-PCR was used to examine the knockdown expression of cuticle biosynthesis-related genes in the 5th floral buds (stage 3, length of 1–2 cm) after agro-infiltration in triplicate and repeated three times independently; Cryo-SEM was used to examine the changes in cellular morphology from the 5th to the 8th blooming flowers of silenced plants (stage 5, floral diameter of 12 cm).

#### Ectopic Expression of PeERF1 in Arabidopsis

A. thaliana ecotype Columbia was used for transformation experiments as previously described (Chen et al., 2012). Fulllength cDNA of PeERF1 was cloned into the pBI121 vector under the control of the constitutive Cauliflower mosaic virus (CaMV) 35S promoter, and the resulted plasmid was named pBI121- PeERF1. We then introduced pBI121-PeERF1 plasmid into Agrobacterium tumefaciens (strain GV3101) and transformed into wild-type Arabidopsis by the floral dip method (Clough and Bent, 1998). In total, 60 kanamycin-resistant T1 seedlings were obtained and grown at 23°C in a growth chamber under long-day conditions (16-h light/8-h dark). A total of 457 T2 heterozygous seedlings were obtained with a segregation ratio of 3:1 as analyzed by chi-square test. For gene expression analysis, RNA samples of T1 heterozygous plants were extracted to confirm the successful expression by qRT-PCR in triplicate and repeated three times independently. The primers were listed in Supplementary Table S1.

### RESULTS

#### Phalaenopsis Flowers Showed Unique Lip Cuticle Features Relative to Other Perianth Epidermal Morphology

To investigate the detailed ultrastructure of orchid floral epidermal cells, we first examined the floral epidermal morphology of a native species, P. aphrodite subsp. formosana (Supplementary Figure S1) and the commercial cultivar, P. OX Red Shoe "OX1408" (Figure 1A) by using Cryo-SEM. The schematic diagrams of different regions and hierarchical structures on the cell surface of Phalaenopsis flowers are shown in Figures 1B, C. The "central field" (CF) and "anticlinal field" (AF) represent the inner and outer parts of cells, respectively, and the boundaries of two perpendicular cell walls are "anticlines" (Figure 1B). The epidermal cells of the Phalaenopsis orchid perianth have a variety of cell morphology, including convex or papilla cell shapes. The cell surface of the perianth epidermal cells can be smooth, covered with 2-D wax films or decorated with protuberances, nanoridges, or other node-like nanoridges (Figure 1C).

The adaxial epidermal cells of sepals and petals in P. OX Red Shoe "OX1408" flowers featured a papilla cell shape with a protuberance on the top (Figures 1D, E, J, K, white arrow), and the abaxial epidermal cells of sepals and petals had a convex cell shape with smooth 2-D wax films (Figures 1G, H, M, N). In contrast, heavy and dense nanoridges were found on the lip epidermis. The lip adaxial epidermal cells had convex cell shape with the appearance of node-like nanoridges in the CF (Figures 1F, L, black arrow) and parallel and radial nanoridges in the AF (Figures 1F, L, white arrowhead) of cell surfaces. The lip abaxial epidermal cells had a convex cell shape with parallel nanoridges in the CF (Figures 1I, O, white dashed arrow) and AF of cell surfaces (Figures 1I, O, white arrowhead). The same heavy nanoridges were also observed on the lip epidermis of P. aphrodite subsp. formosana (Supplementary Figure S1D).

Cross sections of petal and lip adaxial epidermal cells showed markedly different thickness of cuticles (Figures 1P, Q). The petal adaxial epidermal cells were covered with smooth 2-D wax films (Figure 1P) as compared with the complex and heavy nanoridges on the lip adaxial epidermal cells (Figure 1Q). Histochemistry staining with the lipophilic dye "Sudan Black B" revealed a lipid layer on the petal cuticle and lip adaxial epidermal cells (Figures 1R, S).

#### PeERF1 Displayed Lip Development-Associated Gene Expression Patterns

We analyzed and confirmed the spatial expression patterns of PeERF1 in both vegetative (root, leaf, and stalk) and reproductive organs (pedicle, bud, sepal, petal, lip, and column) of P. equestris (Figures 2A, B) by qRT-PCR. PeERF1 was highly expressed in reproductive organs (pedicle and flower bud) with lower expression in roots, and very low expression in leaf and stalk tissue. As expected, PeERF1 was highly expressed in the lip and column, with threefold expression in lip as compared with sepals and petals (Figure 2C).

#### Development of Cuticles on Lip Epidermal Cells Concomitant With the PeERF1 Gene Expression at Late Stage of Lip Morphogenesis

Temporal expression of PeERF1 during lip morphogenesis was examined in five stages of lip development (Figures 3A–C). PeERF1 showed low and increasing expression from stage 1 to stage 4, with a sharp increase at stage 5 (Figure 3D).

FIGURE 3 | Temporal expression patterns of PeERF1 and the ultrastructure of lip epidermal cells during various developmental stages of P. OX Red Shoes "OX1408" flowers. (A–C) Various stages of lip development during floral morphogenesis. S1, stage 1, flower bud (0–0.5 cm); S2, stage 2, flower bud (0.5–1 cm); S3, stage 3, flower bud (1–2 cm); S4, stage 4, flower bud (2–3 cm); S5, stage 5, flowering. Scale bars, 2 cm (A–B) and 1 cm (C). (D) Temporal expression patterns of PeERF1 at various lip developmental stages. Total RNA were extracted from the various lip developmental stages in three independent plants. Three technical repeats were performed for each sample. Data are mean ± SD. Numbers above the bars are expression levels after normalization with the internal control (PeActin4). (E–V) Cyro-SEM of adaxial and abaxial sites of epidermal cells of different parts of lip organs during developmental stages. Black rhombus-head lines indicate the secreted bubble-like preliminary cuticle components. White rhombus-head lines indicate the traces of preliminary nanoridges. Black arrows indicate node-like nanoridges in the CF of the cell. Black dashed arrows indicate irregular nanoridges in the CF of the cell. White arrowheads indicate parallel nanoridges in the AF of the cell. Scale bars, 10 mm.

The ultrastructure of the lip epidermis during various developmental stages was examined under Cryo-SEM. Both adaxial and abaxial surfaces of epidermal cells in the lip-central lobe showed secreted bubble-like preliminary cuticle components and heavy cuticle layers from stages 1 to 3 (Figures 3E, F, G, H, K, L, black rhombus-head line). Traces of preliminary nanoridges started to form on the lip epidermis from stages 2 to 3 (Figures 3G, H, L, O, P, S, T, white rhombushead line). These then thickened and the number of nanoridges increased from stages 4 to 5 (Figures 3I, J, M, N, Q, R, U, V).

The final mature forms of nanoridges varied in different parts of the lip. The adaxial epidermal cells of the central lobe showed anode-like nanoridges in the CF (Figure 3J, black arrow), and parallel and radial nanoridges in the AF (Figure 3J, white arrowhead). Similar nanoridges on adaxial epidermal cells of the central lobe were also found on the adaxial and abaxial epidermal cells of lateral lobes (Figure 3R, V). In contrast, the abaxial epidermal cells of the central lobe showed looser parallel nanoridges in the CF (Figure 3N, white dashed arrow) and AF (Figure 3N, white arrowhead). Thus, the typical lip epidermal features containing nanoridges are important markers of the lip morphological identity. These results suggest that PeERF1 may have a correlation with nanoridge formation during lip morphogenesis.

#### PeERF1 is Phylogenetically Assigned to the Va2 Subgroup of the ERF Subfamily

Phylogenetic analysis showed that PeERF1 belongs to the Va2 subgroup of the ERF subfamily of the AP2/EREBP family (Figure 4A), and the sequence alignment indicated that PeERF1 contains incomplete "SHINE domains" (CMV-1 and CMV-2 motifs) (Figure 4B). Of note, the protein sequences of incomplete "SHINE domains" of PeERF1 are distinguished from other members of the ERF-Va2 subgroup, which results in a separate branch from the other ERF-Va2 members and closer to the ERF-Va1 subgroup (Figure 4).

#### Silencing of PeERF1 Resulted in Abnormal Sculpture of Nanoridges on the Lip Epidermis

To assess the role of PeERF1 in lip cuticle development, CymMVbased VIGS was used to silence its expression with a 142-nt conserved AP2 domain (139–281 nt) and the 175-nt incomplete "SHINE domains" (346–521 nt) in P. OX Red Shoes "OX1408." This resulted in plants that were PeERF1\_AP2-silenced and PeERF1\_SHN-silenced (Supplementary Figure S2A). Using Cryo-SEM we examined the perianth epidermis micromorphology from PeERF1\_AP2-silenced plants. Lip epidermis nanoridges showed continuously altered distribution of structures within one plant from the 1st to the 7th flowers. We also found more severely altered phenotypes, with looser, thinner, uneven, and hollowed distribution of nanoridge structures especially on the lip epidermis of the 6th and the 7th flowers (Supplementary Figure S3). Flowers on these silenced plants showed no obvious difference in floral morphology as compared with mock-treated plants (Supplementary Figures S2B–D).

The most severe phenotype was observed on the 7th blooming flowers in PeERF1\_AP2-silenced plants (Figure S3D). This was concomitant with the expression of PeERF1 at late stage of lip morphogenesis. We then further examined the cell surface on the lip epidermis of the 7th blooming flowers of both types of silenced plants (Figures 5A–L, top view; Supplementary Figure S4, side view). Looser and fewer nanoridges in the AF of adaxial and abaxial surfaces of central lobe epidermal cells were observed as compared with mock-treated plants (Figures 5A–F, white arrowhead). In addition, the AF showed uneven distribution of parallel and irregular nanoridges, which was hollow by reducing the coverage of nanoridges on adaxial and abaxial surfaces of lateral lobe epidermal cells (Figures 5H, I, K, L, white arrowhead; Figure 5M). Moreover, PeERF1\_AP2-silenced plants showed unique, thicker, and extended nanoridges across anticlines of both the adaxial and abaxial surface of lip-lateral lobe epidermal cells (Figures 5H, K, black spherical-head line).

In contrast, the sepals and petals of all blooming flowers of PeERF1-silenced plants otherwise showed no obvious phenotypic changes in the adaxial and abaxial surface of epidermal cells (Supplementary Figure S5). PeERF1 and cutin biosynthesis genes were downregulated in both silenced plants; however, these cells do not have cuticle on their cell surface (Supplementary Figure S6).

#### Off-Target Silencing Effects in VIGS Phenotype

Both PeERF1-silenced plants showed downregulation of PeERF1 (Figures 6A and S6A). However, the transcript level of PeSHN1 was much lower than that of PeERF1 (Figures 6A, B). We speculate that the downregulation of PeSHN1 was due to offtarget effect since there are 71.8% and 47.1% identity between the AP2 and SHN domains of PeERF1 and PeSHN1 genes, respectively (Supplementary Figure S7). However, it is also possible that transcription of PeSHN1 is regulated by PeERF1 to some extent (Figures 6A, B and Supplementary Figure S6B).

#### Reduced Expression of Cutin Biosynthesis Genes in the PeERF1-Silenced Plants

Next, we examined whether the expression of cuticle biosynthesis genes was altered upon the reduction of PeERF1 expression in the VIGS plants. Expression of PeERF1 in the lip was reduced significantly in PeERF1\_AP2-silenced plants, but less so in PeERF1\_SHN-silenced plants. PeCYP86A2 and PeCYP77A4, PeGPAT, PeGPAT, and PeDCR are putative Phalaenopsis orthologs of known cutin biosynthetic genes, which contribute to the cuticle formation in Phalaenopsis lips. Similar to the reduced expression of PeERF1 in the PeERF1\_SHN-silenced plants, the expressions of PeCYP86A2, PeDCR, and PeCYP77A4 were reduced significantly in PeERF1\_AP2-silenced plants (Figures 6C, D). The expressions of PeCYP86A2 and PeDCR were reduced, while the expression of PeCYP77A4 was not affected in the PeERF1\_SHN-silenced plants (Figures 6C, D). The expression of PeGPAT was reduced in both PeERF1\_AP2-silenced and PeERF1\_SHN-silenced plants, but to a less extent (Figure 6F). Therefore, both PeCYP77A4 and

were calculated with 1,000 replicates. PeERF1 is highlighted with a star (\*). (B) Multiple alignment of amino acid sequences of PeERF1 and ERF-V group proteins. According to the existence of "SHINE domains" (CMV-1 and CMV-2 motifs), the ERF-V group was classified into two groups: Va group, with SHINE domains, and Vb group, without SHINE domains. In addition, the Va group was further divided into two subgroups: Va1 subgroup, with complete SHINE domains (deemed as "SHINE clade"), and Va2 subgroup, with incomplete SHINE domains.

nanoridges in the AF of the cell. Black spherical-head line indicates thicker and extended nanoridges across anticlines of cells. Scale bars, 30 mm. (M) The coverage of nanoridges on the top of adaxial and abaxial lip epidermal cells of mock-treated and PeERF1-silenced flowers. For Cryo-SEM examination, eight plants were injected with control, specific domain and conserved domain separately, and repeated twice. Data are mean ± SD (n = 15); the same letters above the bars indicate no statistical difference by Duncan's multiple range test (P < 0.05).

PeGPAT were found to be crucial for cuticle formation, and their expressions were regulated by other TFs in addition to PeERF1 or PeSHN1.

#### Nanoridges Disappeared in the Petal-Like Lip of the "Join Big Foot" Variant

From population of the P. hybrid "Join Big foot" grown from seedlings we selected two plants each with wild-type flower (normal split-lip organ) and with the big-lip enlarged petal-like lip flower. We observed these in order to confirm the association of PeERF1 and cuticle formation during lip morphogenesis (Figure 7A). Cryo-SEM revealed that cell morphology and ultrastructure were severely changed in the petal-like lip epidermal cells of the big-lip variant as compared with wild type (Figures 7B–Q). In wild type, the epidermal cells of split-lip organs showed an abnormal, uneven, and hollowed distribution of parallel nanoridges in the AF of cells (Figures 7B, D, F, H). In the big-lip variant, all adaxial and abaxial surfaces of the lip epidermal cells displayed similar characteristics of petal adaxial epidermal cells with a papilla cell shape with smooth 2-D wax

FIGURE 7 | Floral morphology and ultrastructure of lip epidermal cells of somaclonal variants with normal lip or enlarged petal-like lip mutants. (A) The flowers of somaclonal variants of P. hybrid "Join Big foot TH365" (P. I-Hsin Diamond "Join White of Love" x P. Yu Pin Easter Island). The lip morphology of "big foot mutant" flower showed morphological conversions to petal-like structure. Scale bars, 2 cm. (B–Q) Top view and side view of adaxial and abaxial sites of lip central and lateral lobe epidermis in "wild-type" and "big foot mutant" of somaclonal variants of P. "Join Big foot TH365" flowers. White arrows indicate a protuberance on the top of the cell. Black arrows indicate node-like nanoridges in the CF of the cell. White dashed arrows indicate parallel nanoridges in the CF of the cell. Black dashed arrows indicate irregular nanoridges in the CF of the cell. White arrowheads indicate parallel nanoridges in the AF of the cell. Black arrowheads indicate irregular nanoridges in the AF of the cell. Black spherical-head line indicates thicker and extended nanoridges across anticlines of the cells. Different cell shape types of floral epidermal cells are in the lower left corner of each panel. Scale bars, 30 mm.

films and a protuberance on the top of the cell (Figures 7C, E, G, I, K, M, O, Q; Supplementary Figures S8 and S9). The phenotypic observations were done in three big-foot variant plants. Lip morphology was altered to various levels, from severe to mild, yet the phenotypic change for nanoridge formation was the same (data not shown).

#### Reduced Expression of PeERF1 and Cutin Biosynthesis Genes Associated With Big-Foot Variant

To further understand the molecular mechanisms of petal-like lip formation, we investigated gene expression of cutin biosynthesis genes PeERF1, PeSHN1 (PeCYP86A2, PeCYP77A4, PeDCR, and PeGPAT) (Figure 8), as well as expression of floral morphogenesis genes (B-class [DEFICIENS (DEF)/APETALA3 (AP3)-like (PeMADS2-5) and GLOBOSA (GLO)/PISTILLATA (PI)-like (PeMADS6)], AGAMOUS-LIKE6a (PeAGL6a), and Eclass [SEPALLATA1-4 (PeSEP1-4)] MADS box genes)

(Supplementary Figure S10). Expression of PeERF1 was significantly reduced in the lip of the big-foot variant as compared to the wild type. In addition, the expressions of PeCYP77A4 and PeGPAT were reduced in the bigfoot variant lip compared to that of the wild-type plant, while expression levels of PeCYP86A2 and PeDCR were not significantly different between the big-foot variant and the wild type. Interestingly, expression of PeSHN1 was nearly unaffected for petal-like lip in big-foot variant as compared to that of lip in wild-type plant. These results suggest that PeERF1, PeCYP77A4, and PeGPAT were involved in the nanoridge formation in the orchid lip.

For MADS box genes, highest expressions of PeMADS3, 4 and 6, and PeAGL6a were detected in the lip of "wild-type" flowers, whereas PeMADS2 expressed higher in the petals (Figure 8A). In contrast, only PeMADS2 and PeMADS3 expressed higher in petal-like lip compared to the petal in the big-foot variant (Supplementary Figure S10). Intriguingly, we found little or no difference in transcriptional levels of PeMADS4~6 between the petal-like lip and the petals of the big-foot variant (Supplementary Figure S10).

#### Ectopic Expression of PeERF1 in Arabidopsis Resulted in a Typical SHINE Phenotype

To further characterize the biological function of PeERF1, we performed ectopic overexpression of PeERF1 under the control of CaMV 35S promoter by Agrobacterium-mediated transformation in Arabidopsis. Two batches of overexpression were performed. In the first batch, a total of 60 T1 transgenic lines were generated, and in the second batch of overexpression, a total of 48 transgenic lines were generated. Most of overexpressing lines showed SHINE phenotypes. Three overexpressing lines (L24, L37, and L52) from the first batch and three overexpression lines (P12, P13, and P15) from the second batch were used for further analysis (Figure 9). These plants have enhanced brilliant, shiny green color and curveddown edges rosette leaves as compared to wild-type plants (Figures 9A–C). For seedlings of 35S:PeERF1 transgenic lines, the adaxial surface of the second pair of true leaves show a shiny surface with few trichomes as compared with the wild type (Figures 9D, E). At the cellular level, the adaxial surfaces of rosette leaf epidermal cells of 35S:PeERF1 transgenic plants showed ectopic nanoridge formation and wax deposition on the surface, in contrast to the wild-type leaf with a smooth surface (Figures 9F–H, black arrow).

In addition, qRT-PCR was performed to examine the gene expression in the two transgenic lines with obvious phenotype, P12 and P15 from the second batch of overexpression. Ectopic expression of PeERF1 was accompanied by upregulation of cuticle biosynthesis genes, including AtCYP86A4, AtDCR, and AtGPAT6 in two independent lines (Figure 9I). Furthermore, the ectopic overexpression of PeERF1 in Arabidopsis did not disturb the expression of the three endogenous AtSHNs (AtSHN1~AtSHN3) as well as the PeERF1 orthologous gene At5g25190 in all ectopic overexpression lines except the expression of At5g25190 in the line P12 (Figure 9I). These results suggest that PeERF1, with similar SHINE phenotypes, might be assigned to one of the SHINE-like TFs and can be ectopically overexpressed and functionally involved in leaf cuticle development in Arabidopsis.

### DISCUSSION

#### PeERF1 Exhibits Partial SHINE Functions but Contains Incomplete SHINE Domains

ERF-Va2 subgroup genes are not known to have a typical SHINE phenotype; rather, they are involved in various developmental and physiological processes (Aharoni et al., 2004; Li et al., 2007; Marques et al., 2013; Trupiano et al., 2013; Nakano et al., 2014; Leida et al., 2016). Recently, berry VvERF045 SHINE domains were found to be involved in berry ripening and epidermal cuticle development. Transgenic grapevine lines overexpressing VviERF045 show stunted growth, discolored, and smaller leaves, with reduced gene expression for epidermal wax decoration and wax biosynthesis. This indicates that VviERF045 is a potential repressor in epidermis patterning and cuticle development (Leida et al., 2016).

In this study, PeERF1\_AP2-silenced plants showed significantly reduced expressions of PeERF1 and PeSHN1 with loose and uneven nanoridges on the lip epidermal surface, accompanied by drastically reduced expression of cutin biosynthesis genes including PeCYP86A2, PeCYP77A4, and PeDCR. Expression of PeGPAT, on the other hand, was affected only to a less extent (Figure 6). These results suggest that both PeERF1 and PeSHN1 are important for cuticle formation. Yet the expression level of PeSHN1 was much lower than that of PeERF1. Similar to the reduced expression of PeERF1 in the PeERF1\_SHN-silenced plants, the expressions of PeCYP86A2, PeDCR, and PeCYP77A4 were reduced significantly in PeERF1\_AP2-silenced plants. The expression of PeCYP77A4 was not affected in the PeERF1\_SHN-silenced plants (Figure 6D). The expression of PeGPAT was reduced in both

FIGURE 8 | Expression patterns of PeERF1, PeSHN1, cutin-biosynthetic genes in petal and lip of "wild-type" and "big foot mutant" flowers of P. hybrid "Join Big foot TH365." Transcript level of genes related to lip epidermis development PeERF1 (A), PeSHN1 (B), and cutin metabolism-related genes (PeCYP86A2, PeCYP77A4, PeGPAT, and PeDCR) (C–F) in flowers of "wild-type" and "big foot mutant" of P. "Join Big foot TH365" were examined. P, L, and PPL were represented as petal, lip, and petal-like lip, respectively. mRNA of petal and lip were extracted from the 2nd floral bud in three independent plants of big foot mutant and wild type. Three technical repeats were performed for each sample. Data are mean ± SD. Numbers above the bars are expression levels after normalization with internal control (PeActin4). \*P < 0.05 by one-tailed t-test.

PeERF1\_AP2-silenced and PeERF1-SHN plants, but to a lesser extent (Figure 6F). Therefore, it appears that PeCYP77A4 and PeGPAT are regulated by PeERF1 and/or PeSHN1 and downregulation of these genes might explain the cuticle formation defects in the PeERF1-silenced plants. Based on our observations, it is also suggested that proper expression of PeCYP77A4 and PeGPAT might involve additional TF(s). Thus, our results suggest that PeERF1 as a SHN-like homolog is a potential activator in the lip epidermal cell patterning of Phalaenopsis flowers. Furthermore, transgenic Arabidopsis lines overexpressing PeERF1 showed typical SHINE phenotypes similar to AtSHN-overexpressing Arabidopsis (Figures 9A–C).

FIGURE 9 | Phenotype analysis and expression patterns of transgenic Arabidopsis plants ectopically expressing PeERF1. (A) Wild-type and 35S:PeERF1 transgenic lines (the first batch, L24, L37, and L52) at 45 days old, and (B) wild-type and 35S:PeERF1 transgenic lines (the second batch, P12, P13, and P15) at 60 days old. Scale bars, 5 cm. (C) Shiny and curved-down edges of the rosette leaves of 45-day-old wild-type and 35S:PeERF1 transgenic plants of L24, L37 and L52, and 60 day-old wild-type and 35S:PeERF1 transgenic plants of P12, P13, and P15. Scale bars, 2 cm. (D–E) Seedlings of wild-type and 35S:PeERF1 transgenic plants. Black arrow indicates the top view of the second pair of true leaves with a shiny surface. (F–H) Micrograph images of adaxial surface of rosette leaf epidermal cells of wild-type and 35S:PeERF1 transgenic line 37. Black arrow indicates ectopic wax deposition in the AF and anticlines of the cell. Scale bars, 10 mm. (I) Expression patterns of PeERF1 and Arabidopsis cuticle-associated genes (AtSHN1-3, At5g25190, AtCYP86A4, AtCYP77A6, AtGPAT6, and AtDCR6) in wild type and 35S: PeERF1 overexpressing lines P12 and P15 were determined by qRT-PCR and normalized to the expression level of the Actin gene as an internal control. Fold induction of PeERF1 and 8 genes associated with cuticle biosynthesis in 35S:PeERF1 transgenic plants and wild-type plants, as compared to control. Asterisks were used to indicate statistically significant difference compared with wild-type plants. Three technical replicates were performed for each overexpression line and repeated in two different overexpression lines independently. Data are mean ± SD. \*P < 0.05 by one-tailed t-test.

The PeERF1 overexpressing plants enhanced the expression of AtCYP86A4, AtDCR, and AtGPAT6 without disturbance of the expression of endogenous AtSHNs and the homologous gene At5g25190 except the expression of At5g25190 in the line P12 was increased (Figure 9I). With the fact that At5g25190 does not have the function for cuticle formation (Aharoni et al., 2004), the phenotype observed in the overexpressor P12 was due to the PeERF1 per se. Therefore, our results suggest that PeERF1, as a SHINE-like TF, increase the Arabidopsis cuticle-associated genes and result the shiny surface of rosette leaves.

Recent studies have indicated the importance of CMV-1 and CMV-2 motifs of SHINE domains. A highly conserved valine (V) residue in the complete CMV-1 motif in barley has been found to be associated with lipid biosynthesis of grain (Taketa et al., 2008). When V is changed to aspartic acid (D) in the complete CMV-1 motif, instead of the typical hulled caryopsis in barley, a naked one results. This is associated with lipid formation of caryopsis and hull adhesion (Taketa et al., 2008). In addition, the existence of a C-terminal 30-amino-acid region of the CMV-2 motif is also important for transcriptional activation of SlERF52 (Nakano et al., 2014). The protein sequences of PeERF1 were more distinguished from other members of the ERF-Va2 subgroup (Figure 4). We found that PeERF1 has a serine (S) residue at position 5 of the incomplete CMV-2 motif, whereas several members of the Va2 subgroup have a histidine (H) residue at the same position (Figure 4B).

#### PeERF1, PeSHN1, and Other TFs Together Regulate Cutin Biosynthesis Genes for Nanoridge Formation

Studies of SHINE genes modulating cuticle permeability and epidermal cell patterning have revealed the requirement of the downstream synthesis of cutin polyesters (Kannangara et al., 2007; Li-Beisson et al., 2009; Panikashvili et al., 2009; Shi et al., 2011; Shi et al., 2013; Petit et al., 2016; Mazurek et al., 2017). A recent model was formulated of the association between cutin biosynthesis and nanoridge formation of the petal cuticle (Mazurek et al., 2017). However, although several SHN putative downstream target genes related to cuticle formation have been reported, SHN TFs do not bind directly to most of their presumed targets. In fact, they require an interacting partner for SHN-mediated target regulation (Kannangara et al., 2007). So far, only the promoter regions of CYP86A cytochrome P450s (AtCYP86A4, AtCYP86A7, and SlCYP86A69) and GSDLmotif lipases (AtRXF26) have been found to be activated by SHNs in Arabidopsis and tomato (Shi et al., 2011; Shi et al., 2013).

When PeERF1 was silenced, we found associated downregulation of cutin biosynthesis gene expressions and reduced numbers of nanoridges on lip abaxial and adaxial surfaces. However, even though the expressions of PeCYP86A2 and PeDCR were significantly reduced, that of PeGPAT was only mildly reduced in lip epidermis in both PeERF1-silenced plants (Figure 6). Thus, PeCYP86A2 and PeDCR may be downstream genes of PeSHN1, and PeCYP77A4 a downstream gene of PeERF1. PeGPAT was regulated by PeERF1, PeSHN1, and non-SHN like TFs. Hence, PeERF1 as a SHN-like TF is involved in decorating the floral organ epidermal surface by regulating downstream cutin biosynthesis genes.

#### PeERF1, as a Downstream Target of Floral Morphogenesis Genes, is Involved in the Late Stage of Lip Morphogenesis

Recently, the roles of B- and E-class MADS-box genes were revealed for orchid tepal development (Tsai et al., 2004; Tsai et al., 2005; Lu et al., 2007; Hsieh et al., 2013a; Su et al., 2013b; Pan et al., 2014; Hsu et al., 2015b; Huang et al., 2015; Huang et al., 2016). We have shown that knock-down expression of Eclass PeSEP1-4 genes reduced the expression of PeERF1. PeSEP1 silenced plants had a changed ultrastructure in terms of nanoridge formation, of the floral epidermal cells (Pan et al., 2014). This result suggests that PeERF1 could be a downstream gene of E-class MADS box genes.

In this study, we found that the expression of PeERF1 was reduced to almost zero in the big-foot somaclonal variants with a petal-like lip, along with the loss of gene expression of B-class (PeMADS4, PeMADS5, and PeMADS6) and E-class (PeAGL6a) MADS box genes (Supplementary Figure S10). These results suggest that the developmental program of lip morphogenesis is very complex, and PeERF1 is a downstream target gene of B- and E-class MADS box genes and is involved in lip morphogenesis.

#### The Timing for VIGS Phenotype and Off Target VIGS

Our previous VIGS results of MADS box silencing data showed the highest silencing efficiency at 4–7 weeks post-silencing with the first four flowers blooming after viral inoculation (Hsieh et al., 2013a; Hsieh et al., 2013b; Pan et al., 2014; Hsu et al., 2015a). This indicates that MADS box genes are induced in early floral morphogenesis.

PeSHN1 as an SHN ortholog contained the complete SHINE domains and was assigned to the ERF-Va1 subgroup (SHINE clade) (Figure 4). However, PeERF1 was expressed at late stage lip morphogenesis so that VIGS phenotype was more distinct on the late emerged floral buds. The identity of PeSHN1 with PeERF1\_AP2-silenced and PeERF1\_SHN-silenced regions were 71.8% and 47.1%, respectively. Yet, no contiguous matches were longer than 11 nt for the PeSHN1 coding region and the VIGS fragment (Supplementary Figure S7). A nucleotide identity of less than 11 nt on target mRNA has previously been shown to reduce the chances of silencing induction (Senthil-Kumar and Mysore, 2011; Hsieh et al., 2013a). The off-target effects of PeSHN1 silencing that occurred in all PeERF1-silenced plants may be due to sequence similarity. The combined effects of specific silencing of PeERF1 and off-target silencing of PeSHN1 together resulted in a change of nanoridge formation in lip epidermal cell patterning in the PeERF1-silenced plants, yet PeSHN1 expressed much lower than PeERF1 in the Phalaenopsis orchids.

#### Distinct Sculpture on the Lip Epidermal Surface Reveals the Special Deceit Pollination Strategy in Phalaenopsis

To contribute to successful sexual reproduction in higher plants, the perianth of flower creates various cues, such as tactile, visual, and olfactory signals, to reward or not reward (deceive) pollinators. In Orchidaceae, approximately one-third of orchid species have a deceit pollination strategy by using general floral signals without rewarding pollinators with nectar or pollen (Ackerman, 1986; Nilsson, 1992; Jersáková et al., 2006). Although numerous Phalaenopsis species are scentless, with a diversity of colorful perianths, floral morphology, and floral scents, bees are the major pollinators of Phalaenopsis flowers via food-deceptive pollination (Roman Kaiser, 1993; Hsiao et al., 2006; Tsai et al., 2008).

In this study, we examined the floral epidermal cell surfaces of Phalaenopsis species (Figure 1; Supplementary Figure S1); the tissues of sepals and petals had a papillae shape with a protuberance on the smooth epidermal surface, whereas the lip harbored numerous nanoridges on the epidermal surface. The biological function of nanoridges on the floral organs for pollinator attraction has been linked with the unique visual and tactile signals they produce (Whitney et al., 2009; Prüm et al., 2011; Rands et al., 2011; Whitney et al., 2011b; Prüm et al., 2012; Kourounioti et al., 2013; Prüm et al., 2013; Adachi et al., 2015; Moyroud et al., 2017).

We propose three steps of a special deceit pollination strategy in the scentless Phalaenopsis species. First, the papillae cell shape on the sepal and petal epidermis creates big brilliant visual cues to attract pollinators from a distance. Then, the convex cell shape with heavy nanoridges on the top of the lip epidermal cells generates more "flashy" visual cues (similar to guide lights on an aircraft runway) and direct pollinators to land on the lip instead of sepals and petals. Finally, the distinctive split lip with central lobes and lateral lobes creates a tunnel-like structure, and the nanoridges on the lip epidermis generate tactile cues that help pollinators walk and explore the area. During this process, because of the heavy and dense nanoridges on the top of the lip epidermis, the pollinators may slip toward the column and attach to the pollinia. While visiting subsequent flowers, successful pollination may occur when attached pollinia are placed into the stigmatic cavity underneath the column. The slippery quality of nanoridges for beetles has been reported (Prüm et al., 2011), and the slip-and-fall pollination mechanism related to the ultrastructural characterization of the floral lip was also shown for Gongora bufonia (Orchidaceae) (Adachi et al., 2015).

### CONCLUSION

In conclusion, our results demonstrate that PeERF1, as an SHNlike TF, was involved in lip epidermal cell morphological formation at the last flowering stage by regulating lip nanoridge development in Phalaenopsis flowers. In addition, the heavy nanoridges on the lip epidermis, as typical lip features, may be essential for the pollination mechanism of Phalaenopsis. This study gives a better understanding of the transcriptional regulation of the late stage development of lip morphogenesis and the special pollination mechanism of Phalaenopsis.

### DATA AVAILABILITY STATEMENT

The datasets generated for this study can be found in the PeERF1, MG948436; PeSHN1, XM\_020736987.1; PeCYP86A2, XM\_020732683.1; PeCYP77A4, XM\_020725159.1; PeGPAT, XM\_020727266.1; PeDCR, XM\_020725429.1.

### AUTHOR CONTRIBUTIONS

P-HL, W-HC, and H-HC conceived the research plans. P-HL performed most of the experiments, analyzed the data, and wrote the article with contributions from all the authors. L-MH assisted in the identification and expression analysis of cutin biosynthesis genes. Z-JP assisted with the performance of VIGS experiments. W-NJ and M-CC performed Cryo-SEM analysis and provided service for Cryo-scanning electron microscope.

### FUNDING

This work was supported by grant no. MOST-107-2313-B-006- 003-MY3 from the Ministry of Science and Technology, Taiwan.

### ACKNOWLEDGMENTS

We thank Dr. Ming-Hsien Hsieh (Tainan District Agricultural Research and Extension Station, Council of Agriculture, Taiwan) for providing the greenhouse and for assistance with operating the VIGS experiment, and Dr. Shau-Ting Chiu (Biology Department, National Museum of Natural Science, Taichung, Taiwan) for helpful discussion of floral epidermal morphology. We also thank Oxen Biotechnology Corp. (Tainan, Taiwan) and Join Orchids Corp. (Tainan, Taiwan) for providing the plant materials for of P. OX Red Shoes "OX1408" and P. "Join Big foot TH365," respectively.

### SUPPLEMENTARY MATERIALS

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls. 2019.01709/full#supplementary-material

## REFERENCES


Conflict of Interest: 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 © 2020 Lai, Huang, Pan, Jane, Chung, Chen and Chen. 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.

# Plastome Evolution and Phylogeny of Orchidaceae, With 24 New Sequences

Young-Kee Kim<sup>1</sup> , Sangjin Jo<sup>1</sup> , Se-Hwan Cheon<sup>1</sup> , Min-Jung Joo<sup>1</sup> , Ja-Ram Hong<sup>1</sup> , Myounghai Kwak <sup>2</sup> and Ki-Joong Kim1\*

<sup>1</sup> Division of Life Sciences, Korea University, Seoul, South Korea, <sup>2</sup> Department of Plant Resources, National Institute of Biological Resources, Incheon, South Korea

In order to understand the evolution of the orchid plastome, we annotated and compared 124 complete plastomes of Orchidaceae representing all the major lineages in their structures, gene contents, gene rearrangements, and IR contractions/expansions. Fortytwo of these plastomes were generated from the corresponding author's laboratory, and 24 plastomes—including nine genera (Amitostigma, Bulbophyllum, Dactylorhiza, Dipodium, Galearis, Gymnadenia, Hetaeria, Oreorchis, and Sedirea)—are new in this study. All orchid plastomes, except Aphyllorchis montana, Epipogium aphyllum, and Gastrodia elata, have a quadripartite structure consisting of a large single copy (LSC), two inverted repeats (IRs), and a small single copy (SSC) region. The IR region was completely lost in the A. montana and G. elata plastomes. The SSC is lost in the E. aphyllum plastome. The smallest plastome size was 19,047 bp, in E. roseum, and the largest plastome size was 178,131 bp, in Cypripedium formosanum. The small plastome sizes are primarily the result of gene losses associated with mycoheterotrophic habitats, while the large plastome sizes are due to the expansion of noncoding regions. The minimal number of common genes among orchid plastomes to maintain minimal plastome activity was 15, including the three subunits of rpl (14, 16, and 36), seven subunits of rps (2, 3, 4, 7, 8, 11, and 14), three subunits of rrn (5, 16, and 23), trnC-GCA, and clpP genes. Three stages of gene loss were observed among the orchid plastomes. The first was ndh gene loss, which is widespread in Apostasioideae, Vanilloideae, Cypripedioideae, and Epidendroideae, but rare in the Orchidoideae. The second stage was the loss of photosynthetic genes (atp, pet, psa, and psb) and rpo gene subunits, which are restricted to Aphyllorchis, Hetaeria, Hexalectris, and some species of Corallorhiza and Neottia. The third stage was gene loss related to prokaryotic gene expression (rpl, rps, trn, and others), which was observed in Epipogium, Gastrodia, Lecanorchis, and Rhizanthella. In addition, an intermediate stage between the second and third stage was observed in Cyrtosia (Vanilloideae). The majority of intron losses are associated with the loss of their corresponding genes. In some orchid taxa, however, introns have been lost in rpl16, rps16, and clpP(2) without their corresponding gene being lost. A total of 104 gene rearrangements were counted when comparing 116 orchid plastomes. Among them, many were concentrated near the IRa/b-SSC junction area. The plastome phylogeny of

#### Edited by:

Jen-Tsung Chen, National University of Kaohsiung, Taiwan

#### Reviewed by:

Aleksey Penin, Lomonosov Moscow State University, Russia Kenji Suetsugu, Kobe University, Japan

> \*Correspondence: Ki-Joong Kim kimkj@korea.ac.kr

#### Specialty section:

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

Received: 24 September 2019 Accepted: 10 January 2020 Published: 21 February 2020

#### Citation:

Kim Y-K, Jo S, Cheon S-H, Joo M-J, Hong J-R, Kwak M and Kim K-J (2020) Plastome Evolution and Phylogeny of Orchidaceae, With 24 New Sequences. Front. Plant Sci. 11:22. doi: 10.3389/fpls.2020.00022 124 orchid species confirmed the relationship of {Apostasioideae [Vanilloideae (Cypripedioideae (Orchidoideae, Epidendroideae))]} at the subfamily level and the phylogenetic relationships of 17 tribes were also established. Molecular clock analysis based on the whole plastome sequences suggested that Orchidaceae diverged from its sister family 99.2 mya, and the estimated divergence times of five subfamilies are as follows: Apostasioideae (79.91 mya), Vanilloideae (69.84 mya), Cypripedioideae (64.97 mya), Orchidoideae (59.16 mya), and Epidendroideae (59.16 mya). We also released the first nuclear ribosomal (nr) DNA unit (18S-ITS1-5.8S-ITS2-28S-NTS-ETS) sequences for the 42 species of Orchidaceae. Finally, the phylogenetic tree based on the nrDNA unit sequences is compared to the tree based on the 42 identical plastome sequences, and the differences between the two datasets are discussed in this paper.

Keywords: Orchidaceae, plastome evolution, gene loss, IR contraction/expansion, genome rearrangement

### INTRODUCTION

Orchidaceae is one of the most flourishing flowering plants and contains about 736 known genera and 28,000 species worldwide (Christenhusz and Byng, 2016). Recent studies recognize five subfamilies within Orchidaceae (Apostasioideae, Vanilloideae, Cypripedioideae, Orchidoideae, and Epidendroideae) as a monophyletic group (Chase et al., 2015). The most recently differentiated subfamily, Epidendroideae, includes about 505 genera and 20,600 species, and accounts for most of Orchidaceae (Chase et al., 2015). Orchidaceae is widely distributed throughout the world, and most members in temperate regions have terrestrial life forms, but orchids in tropical rainforests are known to have mainly epiphyte life forms (Givnish et al., 2015). Non-photosynthetic mycoheterotrophic orchids are found in a total of 43 genera and belong to three subfamilies: Vanilloideae, Orchidoideae, and Epidendroideae (Merckx et al., 2013).

Complete Orchidaceae plastomes have been reported in 38 genera and 118 species (NCBI GenBank, June 30, 2019). There are only five genera (Corallorhiza, 10 spp.; Cymbidium, 9 spp.; Dendrobium, 40 spp.; Holcoglossum, 11 spp.; and Neottia, 7 spp.) in which at least five plastomes per genus have been decoded (Logacheva et al., 2011; Barrett and Davis, 2012; Yang et al., 2013; Barrett et al., 2014; Feng et al., 2016; Niu et al., 2017b; Barrett et al., 2018; Kim et al., 2018). Among them, Corallorhiza and Neottia have been subjected to extensive evolutionary studies of their plastomes because both photosynthetic and nonphotosynthetic species occur in a congeneric group (Barrett and Davis, 2012; Barrett et al., 2014; Feng et al., 2016; Barrett et al., 2018). In addition to the two genera, evolutionary studies have been carried out on the plastomes of some species of other orchid genera. For examples, extensive gene losses have been reported in several independent mycoheterotrophic orchid lineages—i.e., Aphyllorchis (Feng et al., 2016), Cyrtosia (Kim et al., 2019), Epipogium (Schelkunov et al., 2015), Gastrodia (Yuan et al., 2018), Hexalectris (Barrett and Kennedy, 2018) and Rhizanthella (Delannoy et al., 2011). In addition, ndh deletion and pseudogenization are assumed to be phenomena that occur independently in many orchid lineages such as Apostasia (Lin et al., 2017; Niu et al., 2017a), Calypso (Barrett et al., 2018), Cattleya (da Rocha Perini et al., 2016), Cephalanthera (Feng et al., 2016), Cremastra (Dong et al., 2018), Cymbidium (Yang et al., 2013; Kim et al., 2018; Wang et al., 2018), Dendrobium (Niu et al., 2017b), Epipactis (Dong et al., 2018), Eulophia (Huo et al., 2017), Holcoglossum (Li et al., 2019), Limodorum (Lallemand et al., 2019), Liparis (Krawczyk et al., 2018), Neuwiedia (Niu et al., 2017a), Oncidium (Wu et al., 2010; Kim et al., 2015a), Paphiopedilum (Niu et al., 2017b; Hou et al., 2018), Phalaenopsis (Chang et al., 2006), Phragmipedium (Kim et al., 2015a), Platanthera (Dong et al., 2018), Vanilla (Lin et al., 2015), and Vanda (Li et al., 2019). On the other hand, full ndh genes have been reported in members of Anoectochilus (Yu et al., 2016), Calanthe (Dong et al., 2018), Cypripedium (Kim et al., 2015b; Lin et al., 2015), Habenaria (Lin et al., 2015; Kim et al., 2017b), Masdevallia (Kim et al., 2015a), Ophrys (Roma et al., 2018), Pleione (Shi et al., 2018), and Sobralia (Kim et al., 2015a). Gene relocations within a plastome often occur in the reduced plastome of orchids such as Cyrtosia (Kim et al., 2019), Hexalectris (Barrett and Kennedy, 2018), and Rhizanthella (Delannoy et al., 2011). Similar gene rearrangement events within a plastome have been reported in Campanulaceae (Cosner et al., 1997), conifers (Hirao et al., 2008), Fabaceae (Cai et al., 2008), Geraniaceae (Chumley et al., 2006), and Oleaceae (Lee et al., 2007). Contractions of the small single copy (SSC) region similar to that of Geraniaceae were also reported in Paphiopedilum (Kim et al., 2015a; Niu et al., 2017b; Hou et al., 2018) and Vanilla (Lin et al., 2015; Amiryousefi et al., 2017).

Phylogenetic studies of Orchidaceae using entire plastomes are in a relatively early stage because of limited available plastome sequences (Givnish et al., 2015; Kim et al., 2015a; Lin et al., 2015; Niu et al., 2017b; Dong et al., 2018). But the relationships among major orchid lineages determined using whole plastomes agree well to the large-scale phylogenetic studies of Orchidaceae using two or three genes (Górniak et al., 2010; Freudenstein and Chase, 2015). Therefore, several outstanding phylogenetic problems in Orchidaceae will be resolved if more plastome sequences are accumulated.

In this study we first completely decoded the plastomes of 24 taxonomic groups of Orchidaceae, including nine genera (Amitostigma, Bulbophyllum, Dactylorhiza, Dipodium, Galearis, Gymnadenia, Hetaeria, Oreorchis, and Sedirea) for which the plastomes had not yet been decoded previously. Second, this study re-annotated and compared the entire plastome sequences of 129 taxa comprising 124 Orchidaceae and five outgroups to investigate evolutionary directions in orchid plastomes, such as sizes, gene contents, gene losses, gene rearrangements, and inverted repeat (IR) expansions/contractions. Third, three stages of gene loss patterns were inferred from mycoheterotrophic orchids and the minimum genes required for plastid maintenance were inferred. Fourth, the phylogenetic trees of Orchidaceae were constructed using whole plastome sequences, and the times at which each taxonomic group differentiated were inferred. Among the 124 orchid plastomes completely decoded, 42 were produced by the NGS method in the laboratory of the corresponding author, and the nuclear ribosomal RNA gene unit (18S-ITS1-5.8S-ITS2-28S-NTS-ETS) was also annotated at the same time. Finally, the common features and differences among the 42 species were compared and explained by comparing their plastome trees and nrDNA gene unit trees.

### MATERIALS AND METHODS

#### Plant Materials and DNA Extraction

Plant leaf materials used in this study and their voucher information are given in Table 1. Fresh leaf samples were ground into fine powder with liquid nitrogen in a mortar. Ground samples were used to extract genomic DNA using a Gspin™ II Genomic DNA Extraction Kit (Intron, Seoul, Korea). The quality of DNA was checked by a UVVIS spectrophotometer. Extracted DNAs were deposited into the Plant DNA Bank in Korea (PDBK) and voucher specimens were deposited into the Korea University (KUS) herbarium and National Institute of Biological Resources (NIBR) herbarium.

#### NGS Sequencing, Plastome Assembly, and Annotation

Four samples—Calanthe bicolor, Dendrobium moniliforme, D. moniliforme "Royal Dream," and D. moniliforme "Sangeum" were sequenced by Illumina HiSeq 2000. Twenty other samples were sequenced by Illumina MiSeq. The resulting raw reads from HiSeq were trimmed by Geneious 6.1.8 (Kearse et al., 2012) with a 0.05 error probability limitation. Dendrobium officinale (NC024019) was used as a reference sequence to perform reference-guided assembly using a Geneious assembler. Contigs from the reference-guided assembly were used as reference

TABLE 1 | NGS data status and general features of the 24 newly sequenced Orchidaceae plastomes.


sequences to perform reference-guided assembly repeatedly until complete plastome sequences were obtained.

The resulting raw reads from MiSeq were trimmed by BBDuk 37.64, implemented in Geneious 11.1.5 (length: 27 kmer). BBNorm 37.64 was used to normalize trimmed reads (target coverage level: 30; minimum depth: 12). Normalized reads were used to perform de-novo assembly by Geneious assembler. Resulting de-novo assembly contigs were used as reference sequences to perform reference-guided assembly to obtain complete plastome sequences when the de-novo assembly did not produce complete plastome contigs. All complete plastome sequences were used as reference sequences to gather chloroplast reads from trimmed read sets. Gathered reads were reused to perform de-novo assembly and validate the level of completeness. Due to its unusual plastome structure, Gastrodia elata was assembled again by SPAdes 3.10.0 (Bankevich et al., 2012) using an error correction tool and assembly module. NGS results are given in Table 1.

Complete plastome sequences were annotated with BLASTn, tRNAscan-SE 2.0 (Lowe and Chan, 2016), ORF finder, and find annotation function in Geneious 11.1.5 (Habenaria radiata plastome sequences, NC035834, used as a reference). Alternative start codons—such as ACG and TTG—were also included in the ORF finder. Genes with many stop codons in the middle of sequences, uncorrectable frame shift mutations, and several large abnormal indels, were judged to have a pseudogene. Gene sequences with complete CDS or few (five or fewer) internal stop codons and for which RNA editing was possible were analyzed to have a gene. These criteria were based on a character reconstructions study of the gene status of ndh in Orchidaceae (Kim et al., 2015a). ORFs from the plastome sequence of four mixotrophic orchids (Corallorhiza maculata var. maculata, Dipodium roseum, Gastrodia elata, and Hetaeria shikokiana) were translated into protein sequences. The translated sequences were exported in the fasta format files to perform psi-blast (Altschul et al., 1997) based on several Orchidaceae plastome databases. Circular plastome maps were constructed using the OGdraw web server (Lohse et al., 2007). All downloaded NCBI plastome data used in our study were reannotated because the published data contained many annotation errors. Finally, we manually edited some ambiguous annotation area using our comparative alignments of orchid sequences.

#### Phylogenetic Analysis

Ninety-nine plastome sequences were downloaded from the NCBI to perform phylogenetic analysis (Table 2; 94 Orchidaceae sequences, four Asparagales sequences, and one Liliales sequences). Eight to 79 CDS and two to four rRNA genes were extracted from plastome sequences. Each extracted region was aligned by MAFFT (Katoh et al., 2002) and all alignments were checked manually. Each aligned gene sequences were subjected to jModeltest (Darriba et al., 2012) in CIPRES Science Gateway (Miller et al., 2010) to obtain the best model. GTR-I-G-X or GTR-G-X were the best-fit models for all genes except rrn5 (short sequence), for which SYM-G was the best-fit model. Eighty-three alignments were concatenated to a length of 87,399 bp. Concatenated alignments were used to perform jModeltest (Darriba et al., 2012) in CIPRES Science Gateway (Miller et al., 2010) to obtain the best model. GTR+G+I was the best-fit model for concatenated data. Lost genes were treated as missing data because they do not affect the phylogenetic signals in the remaining genes (Lam et al., 2016). A maximum likelihood (ML) tree was constructed using RaxML-HPC2 on XSEDE in CIPRES Science Gateway with a GTR+G+I model and 100 bootstrap replicates (−672774.637133) of ML optimization likelihood. Concatenated alignment was also used to construct a Bayesian inference (BI) tree. MrBayes\_CIPRES api (Miller et al., 2015) was used to construct a BI tree with a Markov chain Monte Carlo (MCMC) chain length of 1,000,000 and GTR +G+I model. The trees obtained were treated graphically using Treegraph2 (Stöver and Müller, 2010).

IRScope (Amiryousefi et al., 2018) was used to describe SC-IR junction regions among 41 plastome sequences generated in the laboratory (Gastrodia elata was excluded due to its lack of IR). The length information of 24 new plastome sequences was visualized by ggplot2 package in R (Wickham et al., 2016). Gene contents of 129 plastome sequences were displayed as a heatmap. The ProgressiveMauve algorithm (Darling et al., 2010) was used to check plastome rearrangements among 116 Orchidaceae plastome sequences, excluding nonphotosynthetic orchids with an extremely short plastome (Cyrtosia, Epipogium, Gastrodia, Lecanorchis, and Rhizanthella). The resulting LCBs (locally collinear blocks) from ProgressiveMauve were extracted and numbered to visualize their features (Tables S1 and S2).

The nuclear rDNA region—which contains 18S rRNA, internal transcribed spacers, 5.8S rRNA, 28S rRNA, NTS, and ETS—was generated in 42 Orchidaceae species (Table 2). Fortytwo rDNA sequences were aligned by MAFFT (6,043 bp long). The alignment was tested with jModeltest in CIPRES Science Gateway. RaxML-HPC2 on XSEDE in CIPRES Science Gateway was used to construct an ML tree with a GTR+G+I model and 100 bootstrap replicates. The plastome sequence gene data of 42 species were also used to construct an ML tree. Eighty-three CDS and rRNA genes were extracted and aligned by MAFFT. Eightythree alignments were concatenated into one (80,798 bp long). Concatenated alignments were used to perform jModeltest. An ML tree was constructed by RaxML-HPC2 on XSEDE with a GTR+G+I model with 100 bootstrap replicates. The trees obtained were treated by Treegraph2.

#### Divergence Time Estimation

The alignments used in the ML tree construction were also used to estimate divergence time. The GTR estimated model was selected to build a time divergence tree, following the results of PartifionFinder v2.1.1 (Lanfear et al., 2012). An XML file was prepared by BEAUti 2.5.2 (Bouckaert et al., 2019). The XML file was submitted to the CIPRES Science Gateway to perform BEAST2-XSEDE (Bouckaert et al., 2019). A relaxed clock log normal model (Drummond et al., 2006) and Yule model were chosen to perform MCMC with a chain length of 300,000,000. Logs and trees were collected every 5,000 generations, and three independent runs were performed. Three fossil data

#### TABLE 2 | Information on the sequences used in this study.


#### Kim et al. Evolution and Phylogeny of Orchidaceae

#### TABLE 2 | Continued


#### TABLE 2 | Continued


#### TABLE 2 | Continued


\*Red scientific name indicates that the species is mixotroph or obligate (full) mycoheterotroph throughout its life cycle. The bold NCBI accession numbers for nrDNAs and plastomes indicate that the species' NGS data were generated in this study.

(Asparagales, normal distribution, sigma 8.0, mean 105.3; Dendrobium, log-normal distribution, sigma 2.0, offset 23.2; and Goodyera, log-normal distribution, sigma 2.0, offset 15.0) were used to calibrate nodes (Ramírez et al., 2007; Conran et al., 2009; Gustafsson et al., 2010; Iles et al., 2015).

Three log and tree files were concatenated by Logcombiner v2.5.2 (Rambaut and Drummond, 2014) by discarding 20% of files. The concatenated log files were checked by Tracer v1.6 (Rambaut et al., 2014) to validate the effective sample size (ESS). Major parameters—including posterior, likelihood, and the prior—exceeded an ESS of 100, and all other parameters exceeded an ESS of 50. The concatenated tree files were treated by Treeannotator (Rambaut and Drummond, 2007) in CIPRES Science Gateway with an option of 0.95 posterior probability. The concatenated maximum clade credibility tree generated by Treeannotator was treated by FigTree v1.4 (Rambaut, 2012) and "phytools" and "ape" packages in R.

#### RESULTS

#### General Features of 24 New Orchidaceae Plastomes

The taxonomic positions, NGS methods, raw read numbers, trimmed read numbers, plastome lengths, coverage depth, voucher information, etc. of the 24 new plastomes are summarized in Table 1. The coverage depths ranged from 220x (Amitostigma gracile) to 1,734x (Dactylorhiza viridis var. coreana), so that each plastome sequence was sequenced at least several hundred times.

Among the plastomes of the 24 newly decoded species, the length of the plastome of Gastrodia elata, a non-photosynthetic species, was the shortest at 35,056 bp and that of Cremastra unguiculata, a photosynthetic species, was the longest at 159,341 bp. Gastrodia elata is unique in that it only has a single copy of the plastome because the IR was lost. The plastomes of the remaining 23 species have quadripartite structures consisting of an large single copy (LSC), an SSC, and two IR regions. The total lengths of the plastomes and the relative lengths of the IR, SSC, and LSC of the 24 newly decoded species were compared, as shown in Figure 1.

Among the 24 species, the plastomes of 13 (Amitostigma gracile, Calanthe aristulifera, C. bicolor, Dactylorhiza viridis var. coreana, Epipactis thunbergii, Galearis cyclochila, Goodyera rosulacea, Habenaria chejuensis, H. flagellifera, Liparis auriculata, L. makinoana, Oreorchis patens, and Platanthera mandarinorum) had all the same genes as typical plant plastids. Various subunits of the ndh gene class were found to be pseudogenized or lost in eight species (Bulbophyllum inconspicuum, Cremastra unguiculata, Dendrobium moniliforme, D. moniliforme "Sangeum", D. moniliforme "Royal Dream", Dipodium roseum, Gymnadenia conopsea, and Sedirea japonica). In addition, petL was lost in Cremastra unguiculata, psbD was pseudogenized in Dipodium roseum, and trnG-UCC was lost in Gymnadenia conopsea. In the case of the non-photosynthetic species Corallorhiza maculata var. maculata, ccsA, cemA, ndhB, ndhC, ndhD, ndhG, ndhH, ndhI, ndhJ, ndhK, petA, petD, petG, psaA, psaB, psbA, psbC, rbcL, rpoA, rpoB, rpoC1, and ycf1 existed as pseudogenes and ndhA, ndhE, ndhF, psbB, psbJ, psbL, psbM, and rpoC2 were lost. In the case of Hetaeria shikokiana, another non-photosynthetic species, many genes were lost or pseudogenized. Among the genes, cemA,

FIGURE 1 | Plastome length variation in 24 newly sequenced orchid species. The plastome of Gastrodia elata is 35,056 bp long and consists of only one single copy region. The species names in red indicate mycoheterotrophic species.

ndhA, ndhB, ndhC, ndhH, petA, petD, petG, petN, psaB, psaI, psbC, psbF, psbN, rpoA, rpoC1, rpoC2, ycf3, and ycf4 were pseudogenized, and atpA, atpB, ccsA, ndhD, ndhE, ndhF, ndhG, ndhI, ndhJ, petB, psaA, psaC, psbA, psbB, psbJ, psbL, psbM, rbcL, rpl23, and rpoB were lost. Gastrodia elata, which has the shortest plastome of the 24 species examined, only had the following genes: accD, clpP, matK, rpl2, rpl14, rpl16, rpl20, rpl36, rps2, rps3, rps4, rps7, rps8, rps11, rps12, rps14, rps18, rps19, rrn5, rrn16, rrn23, ycf1, and ycf2. Two pseudogenes (psaI and psbK) remained, and all the other genes were lost.

#### Comparative Analyses of Orchidaceae Plastomes

The Orchidaceae plastomes that have been completely decoded and can be used in comparative studies comprise 60 genera, 142 species, and 146 accessions, including the 24 new plastomes (NCBI database, June 30, 2019). Since the main purpose of this study was to identify the evolutionary trends in plastomes of the entire Orchidaceae, there were three genera—Cymbidium (9 spp.), Dendrobium (40 spp.), and Holcoglossum (11 spp.)—for which the plastomes of many species were decoded, but only three species each were included in the comparative study. In the case of Corallorhiza (10 spp.) and Neottia (7 spp.), the plastomes of all species were included in the comparative study, even though the plastomes of many species were decoded, because these genera include both photosynthetic and non-photosynthetic species. In the case of the remaining genera, all available plastomes were included in the comparative study except for one species of Cypripedium, in which many problems were found in the reannotation process. Therefore, five subfamilies, 17 tribes, 60 genera, 118 species, and 124 accessions were used in the comparative study in this paper. In addition, five outgroup plastomes were also used for comparison. Therefore, the GenBank accession numbers, taxonomic classification, habitats, plastome sizes, LSC lengths, IR lengths, SSC lengths, GC contents, and numbers of genes and pseudogenes of the 129 plastomes used in this study, along with whether or not their corresponding species is photosynthetic, are listed in Table 2. The plastomes of land plants usually have an AT-biased base composition. Furthermore, highly reduced plastomes have more AT-biased substitutions than typical plastomes because they experience relaxed selection. The average GC content of orchid plastomes in this study was 36.40 ± 1.71%. The three highly reduced orchid plastomes—Epipogium aphyllum, E. roseum, and Rhizanthella had 32.8%, 30.6%, and 34.2% GC contents, respectively. The CDS of E. aphyllum., E. roseum, and Rhizanthella had 38.4%, 34.3%, and 38.1% GC contents, respectively. They showed 2–6% more AT-biased substitutions than general orchid plastomes. Extreme AT bias was reported in the plastome of Rhopalocnemis phalloides, which had a GC content of only 13.2% (Schelkunov et al., 2019). Among the 124 Orchidaceae plastomes used in the comparative study, 42 were decoded using the NGS method in the laboratory of the corresponding author. In addition to the plastomes, the base sequences of nuclear ribosomal RNA repeating unit DNA (nrDNA, 18S-ITS1-5.8S-ITS2-28S-NTS-ETS), which are present as tandem repeats, could also be assembled in these 42 species (Figure 2). Therefore, the GenBank accession numbers of these species are also published in Table 2 for the first time. Consequently, these 42 species were used to compare nrDNA repeating unit-based trees and plastome-based trees.

Among the 124 Orchidaceae plastomes, 121 had an IR region. In the case of Aphyllorchis montana, the IR region was lost and the plastome existed only as a single copy (Table 2), similarly to Gastrodia elata 1 and 2. The plastome of Aphyllorchis montana is small at 94,559 bp. Another example of the SSC region being lost and the plastome consisting of the LSC and two IRs was reported in Epipogium aphyllum (Figure 3). These three species are examples of the quadripartite structure of a plastome not being maintained. The quadripartite structure was maintained in all the remaining species. Epipogium aphyllum is the only example in which the IRs appear consecutively without any SSC. However, in the case of Epipogium roseum, which is a related species, an SSC region, although short, existed at 890 bp, and the IR region was shortened to 261 bp (Table 2, Figure 3).

The largest plastome in Orchidaceae was 178,131 bp, in Cypripedium formosanum, and the smallest was 19,047 bp, in Epipogium roseum (Figure 3). The plastome sizes of all photosynthetic orchids were at least 140 kb. On the other hand, the plastome sizes o f non-photosynthetic mycoheterotrophic orchids were mostly smaller than 150 kb and showed high positive correlations with gene numbers (Figure 4A). In addition, unlike the plastome sizes of epiphytic orchids, which were at least 140 kb, those of terrestrial orchids varied greatly because they include mycoheterotrophic species (Figure 4B). Although the plastome sizes of photosynthetic orchids mostly ranged from 145–160 kb, those of Cypripedium had a much larger derange; this was highly correlated with the expansion of the LSC region, and not correlated much with the expansion of IRs (Figures 4C, D). On the other hand, in the case of mycoheterotrophic species, plastome sizes showed high correlations with the size of both the LSC and IR.

FIGURE 3 | Six representative plastomes of Orchidaceae. Cypripedium formosanum has the longest plastome (178.1 kb), while Epipogium roseum has the shortest. Gastrodia elata and Epipogium roseum hold 27 genes in their plastome, even though they have substantially different plastome sizes. The heavy vertical bars indicate inverted repeat (IR) regions and the broken lines among plastomes indicate the boundaries of IR and single copy regions.

#### Gene Content Evolution of Orchidaceae

The gene contents of the 124 orchid species showed high variation (Figure 5). A plastome usually contains a total of 113 genes, comprising 6 atp, 11 ndh, 6 pet, 9 rpl, 4 rpo, 12 rps, 4 rrn, 5 psa, 15 psb, 30 trn, and 11 ungrouped genes. Among the 124 species of Orchidaceae, 27 (22%) have all 113 genes existing in an active status. In eight species, one to three genes were pseudogenized or lost. These 35 species fall into the category of having plastomes with almost all their plastid genes, and they make up 28% of all 124 species in this study. There are 69 species where four to 11 of the 11 genes in the ndh gene class or all 11 ndh genes plus one or two other gene (s) do not function, indicating that the plastid ndh gene class does not function in 56% of all 124 species. Furthermore, ndh genes, photosynthesis light reaction genes (pet, psa, psb), and rpo gene were shown to be lost in 13 species. Such cases are also accompanied by the loss of rpo genes. Finally, in the case of Lecanorchis (two species), Epipogium (two species), Gastrodia (one species, two accs.), and Rhizanthella (one species), many housekeeping genes such as rpl, rps, and trn were also lost.

The presence of genes and pseudogenization in the 124 Orchidaceae and five outgroup species are set forth in Figure 5 by subfamily, tribe, species, and gene class. In the case of photosynthetic orchids, up to 13 genes were lost. In most of these cases, 11 ndh genes and one or two other genes were lost. In an exceptional case with photosynthetic orchids, 19 genes were lost in Vanilla aphylla with epiphytic habitats. The lost genes seem to consist of 11 ndh genes and eight other pseudogenized genes. However, given that only 11 to 12 genes were lost from the plastids in other Vanilla species, it is inferred that the plastid gene annotation of this species is problematic. On the other hand, among the species known to be mycoheterotrophic species, the fewest genes were lost in Limodorum abortivum, where the loss of seven genes comprising cemA, five ndh genes, and rpl22 was observed. Limodorum abortivum was followed by Cymbidium macrorhizon, in which 10 ndh genes were lost, Eulophia zollingeri, where 11 ndh genes and trnG-UCC were lost, and Dipodium roseum, where 11 ndh genes and a psbD gene were lost. However, although these four species have the mycoheterotrophic nutritional mode, they also have chlorophyll, and are reported or observed to be orchids that carry out low levels of photosynthesis. In the case of Corallorhiza, both mycoheterotrophic and photosynthetic species exist, and the levels of photosynthesis vary according to variety or habitat, even in the same species. When seen based on the degree of gene loss, species in which 11 ndh genes and one to three other genes were lost are very likely to preserve photosynthetic activity. When the gene contents of the 124 orchid plastomes were compared, a large gap was found in the area where 14–25 genes were lost. Therefore, whether photosynthetic ability was lost or not can be divided around this range (Figures 4 and 5).

Among the plastomes that are classified into completely nonphotosynthetic orchids because at least 25 genes were lost, two groups are recognized based on the numbers of lost genes and preserved genes. That is, they are a group of plastomes in which the number of preserved genes is similar to or larger than the number of lost genes (13 plastomes), and a group of plastomes in which the number of lost genes is overwhelmingly larger (seven plastomes). Among the 124 plastomes, the Orchidaceae species

that belong to the latter are Epipogium aphyllum (86 lost, 27 preserved), Epipogium roseum (84 lost, 29 preserved), Gastrodia elata 1 and 2 (85 lost, 28 preserved), Lecanorchis kiusiana (80 lost, 33 preserved), Lecanorchis japonica (81 lost, 32 preserved), and Rhizanthella gardneri (80 lost, 33 preserved). These seven plastomes have 27–33 genes preserved, and among them, 20 genes are common (Table 3). These 20 genes are inferred to be the minimum common plastid genes necessary to maintain nonphotosynthetic plastids. However, among these 20 plastid genes, five are missing in any of the remaining 117 Orchidaceae plastomes. Therefore, 15 genes are common to all orchid plastids. The 15 genes comprise 14 housekeeping genes (rpl14, rpl16, rpl26, rps2, rps3, rps4, rps7, rps8, rps11, rps14, rrn5, rrn16, rrn23, and trnC-GGA) and clpP.

To determine the patterns of plastome gene loss over Orchidaceae evolution, we plotted the gene loss patterns on the ML phylogenetic trees using Bayesian estimation approaches (Figure 6). Apostasioideae, Vanilloideae, Cypripedioideae, and Orchidoideae are shown in Figure 6A, and Epidendroideae is shown in Figure 6B. In the case of the ndh gene class, gene loss or pseudogenization occurred independently in almost all lineages of Epidendroideae. On the other hand, in the case of the four remaining subfamilies, the loss of ndh genes, although relatively rare, was observed independently in at least five lineages. The loss of genes directly involved in the photosynthetic light reaction—such as psa, psb, and pet—was observed in six lineages of Epidendroideae, two lineages of Orchidoideae, and one lineage of Vanilloideae. The loss of housekeeping genes such as rpl, rps, and trn occurred independently in three lineages: Gastrodia-Epipogium of Epidendroideae, Lecanorchis of Vanilloideae, and Rhizanthella of Orchidoideae.

Sixteen genes in the typical land plant plastomes (atpF, ndhA, ndhB, petB, petD, rpl2, rpl16, rpoC1, rps12, rps16, trnA-UGC, trnG-UCC, trnI-GAU, trnK-UUU, trnL-UAA, and trnV-UAC) generally have one intron, and two genes (clpP and ycf3) have

of housekeeping genes such as the rps, rpl, or trn gene classes occur independently in four genera, Epipogium, Gastrodia, Lecanorchis, and Rhizanthella. The species names in red indicate mycoheterotrophic orchids.

two introns. The absence or presence of these 20 introns are summarized (Figure 7). The majority of intron losses are associated with the loss of their corresponding gene; the exceptions are rpl16, rps16, and clpP(2), in which the introns were lost but the corresponding genes were not in some terminal clade of the tree.

To identify gene rearrangements among Orchidaceae plastomes, gene block analysis was carried out using MAUVE. Among the 124 Orchidaceae plastomes, only 116 taxonomic groups were analyzed; eight were excluded—Cyrtosia septentrionalis, Epipogium roseum, Epipogium aphyllum, Gastrodia elata 1, G. elata 2, Lecanorchis kiusiana, L. japonica, and Rhizanthella gardneri—because they had severe gene losses and were therefore meaningless in gene block analysis. A total of 134 gene blocks were identified. The aligned lengths (kb), gene regions, forward/reverse orientations, and names of taxa distributed in individual gene blocks are listed in Tables S1 and S2. In addition, the orientations of the 134 gene block and distribution patterns by taxonomic group (Figure 8A), and the lengths of individual blocks and the number of shared taxonomic groups (Figure 8B), were set forth. Among the 134 blocks, 104 were identified as genome rearrangement blocks because they showed variations, and 30 were identified as constant blocks. In terms of plastome regions, 48 blocks were found in the LSC region and among them, 34 were identified as rearrangement blocks and 14 were identified as constant blocks. Only three blocks were found in the IR region, and they were rearranged blocks. Eighty-three blocks (~60%) were found in the SSC region and 24 small blocks were identified in the ycf1 gene region existing at the IR/SSC boundary.

Based on block length, 81 blocks shorter than 1 kb, 35 blocks 1 to 5 kb long, six blocks 5 to 10 kb long, and 12 blocks longer


TABLE 3 | Minimum genes required for seven orchid species with extremely degraded plastomes (Epipogium aphyllum, E. roseum, two Gastrodia elata, Lecanorchis japonica, L. kiusiana, and Rhizanthella gardneri).

Bold gene name with an asterisk indicates that all 124 orchid species share the gene.

than 10 kb were found. The longest gene block is 52.63 kb in length, located between petD-rps12 in the LSC region that has rearranged. This block was followed by trnE-UUC-ycf3 (40.6 kb, rearrangement), rpoC2-rpoB (29.49 kb, rearrangement), trnG-UUC-rpoC2 (22.42 kb, rearrangement), psbA-trnK-UUU (21.00 kb, rearrangement), cemA-trnP-UGG (18.15 kb, rearrangement), trnV-UAC-atpB (16.92 kb, rearrangement), psbB-petD (16.64 kb, rearrangement), trnL-UAA-ndhJ (14.42 kb rearrangement), psaJ-clpP (14.42 kb, rearrangement), rrn16-rrn5 (13.19 kb, rearrangement), and rps16 (12.32 kb, rearrangement). Among the 12 gene blocks longer than 10 kb long, 10 are distributed in the LSC region and two in the IR region (Figure 8B).

In the gene block analysis, to compare structures in the vicinity of the IR/SSC junctions where many small rearrangements are concentrated, the SC-IR junctions of 42 species of Orchidaceae (including 24 new plastome sequences) decoded in this laboratory were schematized by taxonomic group and expressed as shown in Figure S2. Here, the IR of Gastrodia elata was deleted. All the LSC-IRb junctions of 19 taxonomic groups, not including non-photosynthetic taxonomic groups (Cyrtosia, Hetaeria, Lecanorchis), were formed in the vicinity of rpl22. The LSC-IRb junction of Cyrtosia septentrionalis was found in the vicinity of rps19, that of Lecanorchis was found in the vicinity of rpl2, and that of Hetaeria shikokiana was found in the vicinity of rpl2. As with LSC-IRb junctions, LSC-IRa junctions were identified to have been formed in the IGS before psbA in all taxonomic groups except for three nonphotosynthetic taxonomic groups (Hetaeria shikokiana, Lecanorchis kiusiana, and L. japonica). The IRa-SSC and IRb-SSC junctions of the remaining taxonomic groups—except nonphotosynthetic taxonomic groups, seven species of Vanilloideae (in Cyrtosia, Lecanorchis, Pogonia, and Vanilla) and one species of Orchidoideae (Hetaeria)—were located on ndhF and ycf1, respectively. Seven taxonomic groups belonging to Vanilloideae commonly showed a phenomenon of a shorter SSC and among them, in the case of non-photosynthetic taxonomic groups

FIGURE 6 | Evolution of gene losses in the phylogenetic tree of Orchidaceae. A total of 129 taxa—124 Orchidaceae and five outgroup taxa—were the subject of tree reconstruction. The sequences of 83 protein coding genes were concatenated to a length of 87,399 bp. A maximum likelihood (ML) tree was constructed using RaxML-HPC2 with a GTR+G+I model (ML = −672774.637133 of ML optimization likelihood). All the genes were then plotted on the tree node using parsimony criteria under the condition of no parallel gains of the same gene. The species names in red indicate mycoheterotrophic orchids. (A): The basal portion of the tree showing the subfamilies Apostasioideae, Vanilloideae, Cypripedioideae, and Orchidoideae. (B): The upper portions of the tree showing the subfamily Epidendroideae.

Cyrtosia and Lecanorchis, the rRNA genes generally present in the IR were relocated to the SSC. Although Hetaeria shikokiana belongs to Orchidoideae, it has a shortened SSC (2,320 bp), similar to the taxonomic groups of Vanilloideae (Figure 1).

### Phylogenetic Relationships Among the Major Lineages of Orchidaceae

The phylogenetic relationships connected to Apostasioideae {Vanilloideae [Cypripedioideae (Orchidoideae, Epidendroideae)]}

indicated in gray. All large rearrangements longer than 15 kb are located on the LSC. The species names in red indicate mycoheterotrophic orchids. (A): Forward and reverse orientations of each gene block in the orchid species. (B): Locations, lengths, and frequencies of 134 gene blocks.

were identified based on a ML tree made using 83 genes (Figure S3). Although bootstrap values were high at most nodes, those of the nodes of Anoectochilus emeiensis and Goodyera velutina, and G. procera and G. fumata of Cranichideae, were shown to be relatively low at 69.4% and 77.6%, respectively, as was the bootstrap value of the node dividing Neottia camtschatea, N. listeroides, and N. nidus-avis at 63.3%. The results of a BI tree made using the same data matrix are shown in Figure S4. The BI value of Cranichideae, which had low bootstrap values in the ML tree, showed a high degree of support at 1.0, but the BI value of Corallorhiza's branches showed a degree of support of 0.5. Despite the bootstrap and BI support problems in several sub-taxonomic groups, the relationships among the tribes (Pogonieae, Vanilleae, Cranichideae, Diurideae, Orchideae, Neottieae, Sobralieae, Nervilieae, Gastrodieae, Arethuseae, Malaxideae, Collabieae, Epidendreae, Cymbidieae, and Vandeae) represented by the data matrix used were identified to be strongly supported in both the ML and BI trees. In addition, both ML and BI trees had very long branch lengths leading to the taxa with highly reduced plastomes, such as Epipogium, Gastrodia, and Rhizanthella (Figures S3 and S4).

#### Comparing Chloroplast and Nuclear Datasets

Among the 42 Orchidaceae species decoded in this laboratory, both plastome sequences and nrDNA units (18S-ITS1-5.8S-ITS2-28S) about 10 kb long exist (Table 2). Therefore, phylogenetic trees using the nrDNA regions and phylogenetic trees using the base sequences of the 83 plastid genes were made using the same method and compared (Figure 9). The best nrDNA tree was obtained at ML = −35673.363572 and the best plastid tree was obtained at ML = −332414.394814. These phylogenetic trees included only three subfamilies— Vanillioideae, Orchidoideae, and Epidendroideae—and the mutual relationships among the three subfamilies and the tree topologies in Vanilloideae and Orchidoideae were identical. However, differences between the two trees were found in the tree topologies in Epidendroideae. In particular, the phylogenetic positions of Gastrodia elata, Bulbophyllum inconspicuum, Calanthe bicolor, C. aristulifera, Cremastra unguiculata, Corallorhiza maculata var. maculata, Oreorchis patens, Sedirea japonica, and Thrixspermum japonicum differed between the two trees (Figure 9).

#### Time Estimation

A phylogenetic tree was constructed using 83 plastid genes in 124 Orchidaceae species and five outgroup species, and the divergence times of individual tree nodes were estimated using BEAST2. The results are shown in Figure 10. The inferred branching times of 21 main clades are set forth in Table 4. For easy comparison with existing studies, literature materials are also presented in Table 4. In this study, Orchidaceae was inferred to have diverged from other Asparagales at 99.20 (83.78–114.92) mya; Apostasioideae, which is a basal subfamily, was inferred to have diverged from other subfamilies at 79.91 (59.65–99.00) mya; and Vanilloideae was inferred to have diverged at 69.84 (52.59–89.63) mya. It was inferred that Pogonieae and Vanilleae of Vanilloideae diverged at 57.52 (37.88–77.68) mya, Cypripedioideae diverged at 64.97 (48.54–84.93) mya, and Orchidoideae diverged from

FIGURE 9 | Comparison of a plastid tree and nrDNA tree for the same 42 orchid species. The 83 aligned protein coding genes were 80,798 bp long. A maximum likelihood (ML) tree was constructed using RaxML-HPC2 with a GTR+G+I model with 100 bootstrap replicates. The best plastid tree was obtained with ML = −332414.394814. The nrDNA unit (18S-ITS1-5.8S-ITS2-28S) was approximately 10 kb long. The tree reconstruction methods for nrDNA were identical to those of the plastid tree. The species names in red indicate mycoheterotrophic orchids. The lines between the two trees indicate the topological differences between them.

FIGURE 10 | Fossil data showing the estimated divergence time of each node. Three fossil data were used to calibrate nodes (Asparagales—mean 105.3 mya, Dendrobium—23.2 mya, and Goodyera—15.0 mya). Orchidaceae diverged from its sister family at 99.20 mya, and then five subfamilies subsequently diverged in the order of Apostasioideae (79.91 mya), Vanilloideae (69.84 mya), Cypripedioideae (64.97 mya), Orchidoideae (59.16 mya), and Epidendroideae (59.16 mya). However, several specious subtribes within the Epidendroideae diverged in relatively short time periods (39.57–28.13 mya). The species names in red indicate mycoheterotrophic orchids.

Epidendroideae at 59.16 (43.99–78.66) mya. Cranichideae of Orchidoideae diverged from Orchideae at 46.64 (29.49–66.26) mya and Diurideae diverged from Cranichideae at 37.59 (23.13– 57.30) mya. Neottieae diverged at 55.06 (40.55–73.82) mya and other taxonomic groups of Epidendroideae and Sobralieae diverged at 51.46 (37.65–69.44) mya. In addition, Gastrodieae and Nervilleae diverged from other Epidendroideae at 49.21 (35.89–66.60) mya, and the two tribes diverged from each other at 38.27 (27.59–51.84) mya. Arethuseae diverged at 39.57 (27.20– 54.68) mya, Malaxideae at 34.95 (24.93–47.40) mya, Collabieae at



32.10 (23.20–44.64) mya, Epidendreae at 29.96 (21.36–41.98) mya, and Cymbideae and Vandeae at 28.13 (18.92–39.72) mya (Figure 10, Table 4).

### DISCUSSION

#### Evolution of the Plastome Structure

Most plastomes of flowering plants have quadripartite structures consisting of LSC and SSC regions with two IR regions between the two SCs (Bock, 2007). The IR regions are generally known to play a role in the structural stability of plastomes (Palmer and Thompson, 1982), and Orchidaceae plastomes are no exception. When the 124 Orchidaceae plastomes were compared, all but Aphyllorchis montana, Epipogium aphyllum, and Gastrodia elata (2 accessions) had quadripartite structures (Table 2). Gastrodia elata (Epidendroideae-Gastrodieae) is a special case in which the plastome exists only as one single copy because the IR region was lost when the plastome contracted to about 35.1 kb (Figure 1). In the case of Aphyllorchis montana (Epidendroideae-Neottieae), the plastome also exists as only one single copy region because the IR region was lost when the plastome size decreased to 94.6 kb (Feng et al., 2016). In the case of Epipogium aphyllum (Epidendroideae-Nervilieae), the plastome has a tripartite structure consists of the SC and IR because the two IR regions were combined to head-to-head formation when the SSC region disappeared and the plastome contracted to 30.6 kb (Schelkunov et al., 2015). However, in the case of Epipogium roseum in the same genus, the SSC region exists at 890 bp (shorter than average) and the IR decreased to 261 bp (Figure 3). Although these structural variations are related to gene losses associated with mycoheterotrophic habitats, the phenomenon of broken quadripartite structures appears only in the three species mentioned above of the 28 mycoheterotrophic species compared. Plastomes exist as only SC regions due to the loss of IR in several gymnosperms (Wu et al., 2011), Trifolium (Fabaceae), Medicago (Fabaceae), and Cicer (Fabaceae) (Cai et al., 2008). However, all these are photosynthetic species, which are different from the Orchidaceae mentioned above.

Another feature that can be found frequently in Orchidaceae plastomes is IR expansion/shift toward the SSC region. Therefore, the SSC size has been greatly reduced. Due to this feature, which can be found in Pogonia and Vanilla of Vanilloideae, the lengths of IR became 29,808–32,683 bp because both IRa and IRb expanded toward the SSC (Lin et al., 2015; Amiryousefi et al., 2017). In contrast, the lengths of the SSC were shortened by as much as 1,254–5,387 bp. The similar SSC contractions were also found in three species of Cypripedioideae: Paphiopedilum armeniacum, P. dianthum, and P. niveum (Table 2). The same phenomenon was observed in the newly decoded plastome of Hetaeria shikokiana, and the length of the IR and the SSC were found to be 31,960 bp and 2,320 bp, respectively (Figure 1). Among the four genera in which SSC contraction occurred, Paphiopedilum, Pogonia, and Vanilla carry out photosynthesis, while Hetaeria does not. As for life forms, three genera (Hetaeria, Paphiopedilum, and Pogonia) have terrestrial life forms, while one genus (Vanilla) has an epiphytic life form. In all four genera, ndh genes were deleted in most cases. This is thought to be a widely occurring phenomenon in orchids due to parallel evolution and is not considered attributable to the direct effect of SSC contraction. Since ndh genes exist in the vicinity of IR-SSC—such as ndhA, ndhB, and ndhH in Pogonia and Hetaeria, which are pseudogenes—ndh genes are not a direct cause of SSC contraction. However, many small sequence block rearrangements appear in many Orchidaceae lineages near the IR-SSC boundary, indicating that the ndhF and ycf1 genes at the IR-SSC boundary may affected the stability of Orchidaceae plastomes. This is a common phenomenon, especially in species that have lost their ndh gene class or the capacity for photosynthesis.

On the other hand, no IR shift toward the IR-LSC is observed in the Orchidaceae plastomes (Figure S2). The largest plastome size in Orchidaceae was 178,131 bp in Cypripedium formosanum, but there was no IR expansion toward the LSC in this case either. The increase in plastome sizes in photosynthetic plant species is mainly related to IR expansion (Chumley et al., 2006; Weng et al., 2014; Blazier et al., 2016). However, in the case of Cypripedium, the genome size increased due to scattered ATrich repeats among IGS in the LSC region without IR expansion (Figure S2) (Kim et al., 2015b). Mycoheterotrophic orchids have plastome sizes below 150 kb and show high correlations between length of the LSC region and the IR length (R2 = 0.9571 and R2 = 0.8213, respectively). On the other hand, the plastome sizes of all photosynthetic orchids were at least 140 kb and showed a weak correlation with the length of the LSC region (R<sup>2</sup> = 0.5606), but little correlation with the IR length (R<sup>2</sup> = 0.1251) (Figures 4C, D). This means that, in mycoheterotrophic orchids, genome contraction occurs regardless of region, but genome size expansion is affected more by the expansion of the LSC region than the expansion of the IR region.

Gene block analysis was conducted using MAUVE to identify gene rearrangements among Orchidaceae plastomes and, according to the results, 104 out of 134 blocks had rearrangements in at least one taxonomic group (Figure 8, Tables S1 and S2). Of the 104 rearrangements, 34 were distributed in the LSC region, three were distributed in the IR region, and the remaining 67 were concentrated on the SSC region. Among the LSC rearrangements, 11 were at least 10 kb long. On the other hand, the SSC rearrangements were mostly concentrated on ndhF and ycf1 in the IR/SSC boundary region, and most were short rearrangements that did not exceed 1 kb (Figure 8, Table S2). Many of the IRa-SSC junction rearrangements are assumed to be attributable to the unequal crossing over among repeats during the processes of pseudogenization and gene loss of ndhF and ycf1. In the IR region, IR shift occurred as the genes in the SSC region were lost and became smaller, and the region containing the rrn4.5-rrn23 gene moved to the SSC region. Their position changes have been reported in mycoheterotrophs with large gene losses, such as Aphyllorchis montana, Cyrtosia septentrionalis, Hexalectris warnockii, Lecanorchis japonica, L. kiusiana, and Rhizanthella gardneri (Delannoy et al., 2011; Feng et al., 2016; Barrett and Kennedy, 2018; Kim et al., 2019). Given that the relevant shift did not occur in other mycoheterotrophs (Corallorhiza, Dipodium, Hetaeria, or Neottia), this is considered to be an independent evolutionary phenomenon associated with gene loss.

Based on taxonomic groupings, although some species in the genera that include both photosynthetic and non-photosynthetic nutritional modes such as Cephalanthera, Cymbidium, Epipactis, and Platanthera (Merckx et al., 2013) shared rearrangements, species of other genera in the same category (e.g., Cremastra) did not share any rearrangements, indicating that the gene relocation in the plastome occurred after development of non-photosynthetic mycoheterotrophy. Therefore, the plastid gene rearrangements are judged to have developed independently in many orchid lineages, and this is supported by the fact that each of photosynthetic groups (Apostasia, Bulbophyllum, Dendrobium, Elleanthus, Oberonia, Ophrys, Palmorchis, Pleione, Pogonia, Sobralia, and Vanilla) has multiple rearrangements. Furthermore, since a number of rearrangements have been reported in Aphyllorchis montana, Cyrtosia septentrionalis, Hexalectris warnockii, Lecanorchis japonica, L. kiusiana, and Rhizanthella gardneri, which are mycoheterotrophs that reached stage 2 of gene loss, it is assumed that rearrangements played some role in gene loss. However, given that there is no gene rearrangement in mycoheterotrophs (Corallorhiza, Hetaeria, and Neottia) (Delannoy et al., 2011; Feng et al., 2016; Barrett and Kennedy, 2018; Kim et al., 2019), whose gene losses reached stage 1 or 2, further comparative studies are necessary to determine the relationships between gene relocation and gene loss.

#### Evolution of Mycoheterotrophy and Gene Loss in Orchidaceae

Almost all orchid species are initially mycoheterotrophs at an early stage of development, then subsequently develop into full autotrophs. Some orchids exhibit both autotrophic and mycoheterotrophic (mixotrophic) nutritional modes, even at adult stages. Finally, several orchid species maintain an obligate (full) mycoheterotrophic (non-photosynthetic) nutritional mode throughout their life cycle. Forty-three genera in Orchidaceae are known to include non-photosynthetic mycoheterotrophs (Merckx et al., 2013). Among them, 17 were used in this study (Aphyllorchis, Cephalanthera, Corallorhiza, Cremastra, Cymbidium, Cyrtosia, Dipodium, Epipogium, Eulophia, Gastrodia, Hetaeria, Hexalectris, Lecanorchis, Limodorum, Neottia, Platanthera, and Rhizanthella), and only photosynthetic species were analyzed in the case of three of these genera (Cephalanthera, Cremastra, and Platanthera). However, when the plastomes were checked, it was found that Cymbidium macrorhizon, Eulophia zollingeri, Dipodium roseum, and Limodorum abortivum only lost the ndh gene (Figure 5, Figure S1). In addition, after many literature reviews and firsthand observation by these researchers, it was found that, regarding these four species, chlorophylls exist in stems and flowers and low-level photosynthesis is carried out during a certain period in the life cycle (Blumenfeld, 1935; Girlanda et al., 2006; Kim et al., 2017a; Suetsugu et al., 2018). Therefore, it is reasonable to treat these four species as mixotrophs rather than obligate mycoheterotrophs. Among the species thought to be non-photosynthetic, Corallorhiza, which was studied the most extensively, is very likely to have photosynthetic capability if only ndh genes had been lost. In addition, cases have been reported in which members of the same species of Corallorhiza had different, mixotrophic nutritional modes depending on living environments (Barrett and Davis, 2012; Barrett et al., 2014; Barrett et al., 2018). The estimated divergence times of these mixotrophs are 7.44 (2.41–15.37) mya (Cymbidium), 17.34 (8.87–26.76) mya (Dipodium and Eulophia), and 17.77 (3.68– 31.79) mya (Limodorum), and the evolution of these genera are thought to be relatively recent events. Furthermore, the divergence times of the obligate mycoheterotrophic groups were also estimated as follows: 23.61 (16.05–33.12) mya (Corallorhiza and Cremastra), 21.56 (11.77–33.40) mya (Hexalectris), 38.27 (27.60–51.84) mya (Gastrodia and Epipogium), 28.11 (15.75–40.61) mya (Aphyllorchis, Cephalanthera, Epipactis, and Neottia), 5.32 (1.71–9.80) mya (Platanthera), 37.59 (23.13–57.30) mya (Rhizanthella), 7.48 (2.26–13.66) mya (Hetaeria), and 27.82 (13.31–45.95) mya

(Cyrtosia and Lecanorchis) (Figure 10). It can be seen that, except for Gastrodieae (Gastrodia), Nervilleae (Epipogium), and Diurideae (Rhizanthella), all these species diverged more recently than 30 mya (Figure 10). When estimated based on the foregoing, there are two possibilities: the obligate mycoheterotrophic taxa are relatively recently evolved taxonomic groups within Orchidaceae, and information about the older obligate mycoheterotrophic taxonomic groups is not yet available. To test these, further studies should be conducted on other species in Gastrodieae and Nervilleae, whom are estimated to have diverged for long times ago.

While epiphytic orchids consisted only of photosynthetic species and all had genomes at least 140 kb long, terrestrial orchids showed a large variation in plastome sizes and had much different numbers of functional genes because they included nonphotosynthetic species (Figure 4A). In addition, the correlation between genome size and the number of functional genes was high (R2 = 0.9069) in mycoheterotrophs, and relatively low (R<sup>2</sup> = 0.3320) in photosynthetic species (Figure 4B). Also, mycoheterotrophs are distributed into three clusters: one with about 100 functional genes, one with 55–90 functional genes, and one with 25–30 functional genes (Figures 4B and 5). This means that in the evolutionary process of mycoheterotrophic species, gene loss occurred step by step in three stages rather than on a continuous spectrum.

The first stage of gene loss was the pseudogenization or loss of the ndh gene class. There are 11 ndh gene subunits in each plastome. ndh gene losses are observed in all five subfamilies of Orchidaceae, and appear at especially high frequencies in Cypripedioideae, Epidendroideae, and Vanilloideae. On the other hand, the frequencies of non-functionalization are low in Apostasioideae and Orchidoideae. The ndh genes of the epiphytic orchids (Bulbophyllum, Cattleya, Dendrobium, Erycina, Gastrochilus, Holcoglossum, Neofinetia, Oberonia, Phalaenopsis, Pelatantheria, Pendulorchis, Sedirea, Thrixspermum, Vanda, and Vanilla) used in the analysis were lost or pseudogenized (Figures 5 and 6). Although the loss or pseudogenization of ndh genes was also observed in terrestrial taxonomic groups (Apostasia, Calypso, Cephalanthera, Cremastra, Cymbidium, Epipactis, Goodyera, Kuhlhasseltia, Limodorum, Liparis, Neuwiedia, Oncidium, Paphiopedilum, Phragmipedium, Platanthera, and Pogonia), the frequencies are low compared to epiphytes. Although the loss or pseudogenization of ccsA, cemA, and infA is often observed in addition to ndh gene non-functionalization, but no clear trend has been found (Barrett et al., 2014; Kim et al., 2015a; Feng et al., 2016; Niu et al., 2017b; Hou et al., 2018). In the phylogenetic tree, the non-functionalization of ndh appears to occur gradually and independently in the process of pseudogenization in many independent lineages (Figure 6). However, given that most orchid species in which ndh gene non-functionalization occurred retain photosynthetic capacity, it is inferred that the function of this gene class may be affected by the nuclear or mitochondrial genomes. Of course, the gene function may be maintained by RNA editing after pseudogenization, but the possibility is low because most pseudogenization entail not only base changes but also indels. However, given that ndh genes were lost in all non-photosynthetic species, the loss of the ndh gene class is considered to be a precondition for the development of mycoheterotrophs.

The second stage of gene loss is the loss of functions of genes involved in the photosynthetic light reaction, such as pet (six genes), psa (five genes), and psb (15 genes). This includes the rpo (four genes) gene class, which includes housekeeping genes. Gene losses at this stage are shown to have progressed independently in at least 10 clades including six independent lineages of Epidendroideae (Aphyllorchis, Corallorhiza, Epipogium, Gastrodia, Hexalectris, and Neottia), two Orchidoideae clades (Hetaeria and Rhizanthella), and one Vanilloideae clade (Cyrtosia and Lecanorchis) (Figure 6). In particular, in the case of Corallorhiza and Neottia, gene losses progressed independently depending on species, even within the same genus, indicating that these are important taxonomic groups for understanding second stage gene losses. In addition, the plastomes of Cyrtosia and Lecanorchis of Vanilloideae degraded significantly so that ccsA, cemA, rbcL, ycf3, and ycf4 were commonly deleted in addition to ndh, pet, psa, psb, and rpo, and it was identified that some subunits of the housekeeping genes, such as rpl and rps, were also deleted from Lecanorchis (Figure 6A). The degradation of Cyrtosia and Lecanorchis of Vanilloideae is thought to be part of the transition process from plastome degradation stage 2–4 (Wicke et al., 2016). On the other hand, Hetaeria shikokiana of Orchidoideae and Corallorhiza maculata var. maculata of Epidendroideae correspond to plastome degradation stage 2 because some of their ndh, psa, psb, pet, and rpo genes were deleted or pseudogenized. However, since plastome degradation stages 2 and 3 (Wicke et al., 2013; Wicke et al., 2016) appeared to occur simultaneously in Orchidaceae, this stage was defined as the second stage of gene loss.

The third stage of gene loss in Orchidaceae includes cases where only 27 to 33 out of 113 unique plastome genes remain, meaning that further gene loss has occurred than in stage 2 cases. Since at least 55 genes are preserved in stage 2 and fewer than 33 genes are preserved in stage 3, a large gap exists between the two stages (Figure 4B). In stage 3, all the photosynthesisrelated gene functions were lost and most housekeeping genes such as trn, rpl, and rps—were also lost. In addition, the fact that pseudogenes do not exist or are limited to three or fewer is also a characteristic of stage 3. This stage is observed in a total of four lineages: two lineages of Epidendroideae (Epipogium, 27 to 29 conserved genes; Gastrodia, 28 conserved genes), one of Orchidoideae (Rhizanthella, 33 conserved genes), and one of Vanilloideae (Lecanorchis, 32–33 conserved genes) (Figures 5 and 6, Table 2). Regarding the number of genes, the 27 in Epipogium aphyllum are the minimum number of genes found in Orchidaceae plastomes. Furthermore, 20 genes that commonly exist in the four lineages of seven plastomes, which are in gene loss stage 3, were identified, comprising rpl (12, 14, 16, 36), rps (2, 3, 4, 7, 8, 11, 14, 18, 19), rrn (5, 16, 23), trnC-GCA, trnfM-CAU, accD, and clpP (Table 3). The fact that the same 20 genes have been preserved even though the extremely contracted orchid plastomes evolved from four independent mycoheterotrophic lineages means that those genes selectively remained because they perform the minimum functions necessary to maintain the plastomes. Since five (rpl12, rps18, rps19, trnfM-CAU, and accD) of these 20 genes were lost in one to four other orchid species (Table 3), it was concluded that there are 15 common orchid plastome genes.

### Comparisons of Plastome and nrDNA Trees

Thus far, phylogenetic studies using a portion of the plastome genes have been common (Cameron et al., 1999; Givnish et al., 2015), and phylogenetic studies using the entire plastome genes have also progressed thanks to the development of NGS technology (Yang et al., 2013; Kim et al., 2015a; Feng et al., 2016). The use of chloroplast genes in phylogenetic studies has many advantages such as ease of use, but this has been pointed out to be vulnerable to lineage sorting and problems such as plastome capture due to hybridization because it only tracks maternal lineages. Although NGS technology and transcriptome analysis are developing, there are still cost limitations to using the entire nuclear genome for phylogenetic studies. Therefore, nuclear ribosomal ITS regions have been extensively used in studies of the relationships between species or allied genera. An advantage of the genome skimming NGS is that, for plastome sequences with high depths of coverage (see Table 1), it can decode about 10 kb of regions that exist as tandem repeats (18S-ITS1-5.8S-ITS2-28S-NTS-ETS) (Figure 2). Although it has difficulties in recovering entire NTS-ETS when the plastome coverage depth is below then 500x, genome skimming for NGS can easily recover the 6 kb region of the 18S-ITS1-5.8S-ITS2-28S region. Therefore, in this study, an attempt was made to compare phylogenetic trees constructed using all the same plastome genes and phylogenetic trees constructed using the nrDNA unit sequences. Orchid plastomes from 42 species, representing three subfamilies of Orchidaceae, were produced in the laboratory of the corresponding author and directly compared (Figure 9).

According to the results, the relationships among the three subfamilies were the same as (Vanilloideae (Orchidoideae, Epidendroideae)). Furthermore, the associations among the tribes and species in Vanilloideae and Orchidoideae were completely identical. However, the topologies of the two trees were different for the relationships among Epidendroideae taxa. In the Epidendroideae, there was no difference in generic or tribal grouping, but the positions of Gastrodia, in which the plastome reduced greatly, and Bulbophyllum, which is a photosynthetic species, were very different. In addition, clades with bootstrap support values of less than 50% had different topologies. However, it is not clear from this comparative analysis alone whether these differences are due to lineage sorting, ancient plastome capture, other evolutionary histories, or sampling issues. Although plastome data for diverse orchid species are currently known, the sequence data for nrDNA repeats for the 42 species in this study are presented for the first time. Although some nodes in Epidendroideae show discrepancies, major parts of the trees are generally identical to each other. In addition, these data are thought to be easily obtainable because nrDNA repeats can be recovered by simply carrying out additional data mining while conducting plastome sequencing using the genome skimming NGS method. If more data are accumulated for the same sample, then the evolution of orchids can be more easily understood.

#### Diversification of Major Lineages of Orchidaceae

The results of estimating the divergence times of major clades of Orchidaceae using all 83 genes in the plastome were similar to the findings of previous studies conducted using two or three genes at some points, and different at others (Figure 10, Table 4). First, in this study, the time of origin of orchids was estimated to be 99.20 (83.78–114.92) mya, which is similar to the previous estimate of 104 mya using two genes, and not significantly different from the 111.38 mya estimated using three genes (Gustafsson et al., 2010; Givnish et al., 2015). However, the finding in this study shows the divergence time of the basal taxonomic groups ranging from Orchidoideae and Cypripedioideae to Epidendroideae to be a little earlier than the previous finding estimated using two genes (1.84–6.03 mya differences), and a little more recent than the result using three genes. That is, the times estimated in the two previous studies are very different and the one in this study fell between the other two. However, the origin time of terminal tribes in Epidendroideae was inferred to be more recent by this study than the previous studies, which only used two or three genes.

Orchidoideae is estimated to consist of 198 genera and 4,931 species, and the Epidendroideae is estimated to consist of 505 genera and 20,606 species (Chase et al., 2015; Christenhusz and Byng, 2016). These two clades account for most of Orchidaceae and diverged 71.85 mya (Figure 10, Table 4). The divergence times of Arethuseae (25 genera, 723 species), Collabieae (20 genera, 453 species), Cymbideae (165 genera, 3,997 species), Epidendreae (99 genera, 6,935 species), Malaxideae (16 genera, 4,631 species), Podochileae (27 genera, 1,292 species), and Vandeae (136 genera, 2,340 species)—tribes of Epidendroideae consisting of many species—are later than 39.57 (27.20–54.68) mya, which is relatively recent. These data mean that many species differentiated in a short time. In studies that analyzed diversification rates, the relevant tribes had higher rates compared to the basal lineage in most cases, thereby supporting the hypothesis that many species underwent evolutionary radiation in a short time (Givnish et al., 2015).

#### Perspective

Sixty genera, 140 species, 146 accessions of Orchidaceae plastomes have been completed decoded and can be used in comparative studies of plastomes, including the 24 plastomes newly reported in this study (NCBI database, June 30, 2019). In the present study, the evolutionary trends of plastomes were compared and analyzed to understand the evolutionary processes of orchid plastomes and mycoheterotrophic orchids. Only three different types of species were selected in the case of Cymbidium (9 spp.), Dendrobium (40 spp.), and Holcoglossum (11 spp.) because many species in these genera have been studied. Therefore, the plastomes of 60 genera, 118 species, and 124 accessions were analyzed. In addition, for the 42 species decoded in the corresponding author's laboratory, a plastome tree and nrDNA tandem repeat tree were compared to discuss evolution. Orchidaceae is known to include about 736 genera and 28,000 species. Therefore, those genera for which their plastomes have been decoded thus far make up approximately 8% of all the known genera of the Orchidaceae, and those species whose plastomes have been decoded thus far comprise less than 1% of all the known species of the Orchidaceae. There are 43 genera known to include nonphotosynthetic mycoheterotrophs in Orchidaceae. In this study, plastome gene loss patterns of 17 genera that correspond to 40% of the foregoing genera were compared and analyzed to derive third-stage gene losses and 15 common genes. However, if further studies are conducted with more genera and species, Orchidaceae plastome gene losses can be understood better. In particular, genera such as Corallorhiza and Neottia, in which photosynthetic species coexist with non-photosynthetic mycoheterotrophs, should be intensively studied with all species, including several populations per species. Studies of the taxonomic groups in which many species were recently differentiated should be continuously conducted to elucidate the mechanism of evolutionary radiation. In this regard, the use of both plastid and nuclear genomes using NGS technology will increase for orchid evolution studies. If high-quality NGS data are accumulated by 50% at the genus level and 10% at the species level, then the evolution of orchids can be better understood.

### REFERENCES


### DATA AVAILABILITY STATEMENT

The datasets generated for this study can be found in the National Center for Biotechnology Information (NCBI), please find accession numbers in Table 2.

### AUTHOR CONTRIBUTIONS

K-JK and MK designed research. Y-KK, SJ, S-HC, J-RH, and MK collected the research materials. Y-KK and SJ performed research. Y-KK, SJ S-HC, J-RH, and M-JJ analyzed data and deposited the data to data libraries. Y-KK and K-JK wrote the manuscript. MK and K-JK secured the research funds.

### FUNDING

This work was supported by the National Research Foundation of Korea (NRF) under grant no. NRF-2015M3A9B8030588 to K-JK and by the National Institute of Biological Resources (NIBR) under the genetic evaluation of vascular plants IV-1 (2018, grant no. NIBR201803102) to MK and K-JK.

#### ACKNOWLEDGMENTS

We thank to two anonymous reviewers for their helpful comments to improving the manuscript. We would like to thank Noah Last of Third Draft Editing for his English language editing. We also thank to Dr. Sanghun Oh of Daejeon University for the leaf materials of Calanthe aristulifera and Dr. Kyeongwon Kang of Babo Orchid Farm for the materials of Habenaria chejuensis and H. flagellifera. We thank the curator of the Korea University Herbarium (KUS) for voucher specimen preparation. The genomic and chloroplast DNAs are deposited in the Plant DNA Bank in Korea (PDBK).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2020.00022/ full#supplementary-material


transitional clade of heterotrophic orchids, and implications for heterotrophic angiosperms. Mol. Biol. Evol. 31, 3095–3112. doi: 10.1093/molbev/msu252


Yuan, Y., Jin, X., Liu, J., Zhao, X., Zhou, J., Wang, X., et al. (2018). The Gastrodia elata genome provides insights into plant adaptation to heterotrophy. Nat. Commun. 9, 1–11. doi: 10.1038/s41467-018-03423-5

Conflict of Interest: 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 © 2020 Kim, Jo, Cheon, Joo, Hong, Kwak and Kim. 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.

# Corrigendum: Plastome Evolution and Phylogeny of Orchidaceae, With 24 New Sequences

Young-Kee Kim<sup>1</sup> , Sangjin Jo<sup>1</sup> , Se-Hwan Cheon<sup>1</sup> , Min-Jung Joo<sup>1</sup> , Ja-Ram Hong<sup>1</sup> , Myounghai Kwak <sup>2</sup> and Ki-Joong Kim<sup>1</sup> \*

*<sup>1</sup> Division of Life Sciences, Korea University, Seoul, South Korea, <sup>2</sup> Department of Plant Resources, National Institute of Biological Resources, Incheon, South Korea*

Keywords: Orchidaceae, plastome evolution, gene loss, IR contraction/expansion, genome rearrangement

#### Approved by:

**A Corrigendum on**

*Frontiers Editorial Office, Frontiers Media SA, Switzerland*

\*Correspondence:

*Ki-Joong Kim kimkj@korea.ac.kr*

#### Specialty section:

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

> Received: *03 March 2020* Accepted: *04 March 2020* Published: *20 March 2020*

#### Citation:

*Kim Y-K, Jo S, Cheon S-H, Joo M-J, Hong J-R, Kwak M and Kim K-J (2020) Corrigendum: Plastome Evolution and Phylogeny of Orchidaceae, With 24 New Sequences. Front. Plant Sci. 11:322. doi: 10.3389/fpls.2020.00322*

#### **Plastome Evolution and Phylogeny of Orchidaceae, With 24 New Sequences**

by Kim, Y.-K., Jo, S., Cheon, S.-H., Joo, M.-J., Hong, J.-R., Kwak, M., et al. (2020). Front. Plant Sci. 11:22. doi: 10.3389/fpls.2020.00022

In the original article, there was a mistake in the legend for Figure 4 as published. The word "**terrestrial**" at the end of the legend for Figure 4A should be "**epiphytic.**" The correct legend appears below.

Figure 4. Relationships between plastome lengths and gene numbers. **(A)**: Terrestrial orchids show a wider range of variation than epiphytic orchids. **(B)**: Mycoheterotrophic orchids show a wider range of variation than photosynthetic orchids. **(C,D)**: Plastome lengths are more strongly correlated with LSC lengths than IR lengths in both mycoheterotrophic and photosynthetic orchids.

The authors apologize for this error and state that this does not change the scientific conclusions of the article in any way. The original article has been updated.

Copyright © 2020 Kim, Jo, Cheon, Joo, Hong, Kwak and Kim. 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.

# Phylogeny and Historical Biogeography of Paphiopedilum Pfitzer (Orchidaceae) Based on Nuclear and Plastid DNA

Chi-Chu Tsai 1,2† , Pei-Chun Liao3† , Ya-Zhu Ko4† , Chih-Hsiung Chen<sup>5</sup> and Yu-Chung Chiang4,6\*

#### Edited by:

Jen-Tsung Chen, National University of Kaohsiung, Taiwan

#### Reviewed by:

Allison Miller, Saint Louis University, United States Xun Gong, Chinese Academy of Sciences, China

\*Correspondence:

Yu-Chung Chiang yuchung@mail.nsysu.edu.tw † These authors have contributed equally to this work

#### Specialty section:

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

> Received: 23 July 2019 Accepted: 28 January 2020 Published: 27 February 2020

#### Citation:

Tsai C-C, Liao P-C, Ko Y-Z, Chen C-H and Chiang Y-C (2020) Phylogeny and Historical Biogeography of Paphiopedilum Pfitzer (Orchidaceae) Based on Nuclear and Plastid DNA. Front. Plant Sci. 11:126. doi: 10.3389/fpls.2020.00126 <sup>1</sup> Kaohsiung District Agricultural Research and Extension Station, Pingtung, Taiwan, <sup>2</sup> Department of Biological Science and Technology, National Pingtung University of Science and Technology, Pingtung, Taiwan, <sup>3</sup> School of Life Science, National Taiwan Normal University, Taipei, Taiwan, <sup>4</sup> Department of Biological Sciences, National Sun Yat-sen University, Kaohsiung, Taiwan, <sup>5</sup> Department of Botany, National Museum of Natural Science, Taichung, Taiwan, <sup>6</sup> Department of Biomedical Science and Environment Biology, Kaohsiung Medical University, Kaohsiung, Taiwan

The phylogeny and biogeography of the genus Paphiopedilum were evaluated by using phylogenetic trees derived from analysis of nuclear ribosomal internal transcribed spacer (ITS) sequences, the plastid trnL intron, the trnL-F spacer, and the atpB-rbcL spacer. This genus was divided into three subgenera: Parvisepalum, Brachypetalum, and Paphiopedilum. Each of them is monophyletic with high bootstrap supports according to the highly resolved phylogenetic tree reconstructed by combined sequences. There are five sections within the subgenus Paphiopedilum, including Coryopedilum, Pardalopetalum, Cochlopetalum, Paphiopedilum, and Barbata. The subgenus Parvisepalum is phylogenetic basal, which suggesting that Parvisepalum is comprising more ancestral characters than other subgenera. The evolutionary trend of genus Paphiopedilum was deduced based on the maximum likelihood (ML) tree and Bayesian Evolutionary Analysis Sampling Trees (BEAST). Reconstruct Ancestral State in Phylogenies (RASP) analyses based on the combined sequence data. The biogeographic analysis indicates that Paphiopedilum species were firstly derived in Southern China and Southeast Asia, subsequently dispersed into the Southeast Asian archipelagoes. The subgenera Paphiopedilum was likely derived after these historical dispersals and vicariance events. Our research reveals the relevance of the differentiation of Paphiopedilum in Southeast Asia and geological history. Moreover, the biogeographic analysis explains that the significant evolutionary hotspots of these orchids in the Sundaland and Wallacea might be attributed to repeated migration and isolation events between the south-eastern Asia mainland and the Sunda Super Islands.

Keywords: Paphiopedilum, molecular phylogeny, biogeography, evolutionary trend, dispersal events

### INTRODUCTION

The orchid genus Paphiopedilum Pfitzer belongs to the subfamily Cypripedioideae Lindley. This subfamily has been considered a distinct lineage since Lindley (1840) separated them from other orchids based on the characteristic of having two separated fertile anthers [see (Cribb, 1998)]. This subfamily includes only five genera: Cypripedium, Mexipedium, Paphiopedilum, Phragmipedium, and Selenipedium. Mexipedium and Selenipedium are monotypic genera (Albert and Chase, 1992), which was a finding supported by ITS sequence analysis (Cox et al., 1997). These five genera are distributed in separate and restricted geographical ranges (Cribb, 1998). Paphiopedilum is distinguished from genera Cypripedium and Selenipedium by its conduplicate coriaceous leaves, as opposed to the plicate persistent leaves of the latter two genera. Furthermore, Paphiopedilum differs from Phragmipedium and Mexipedium, as they display imbricate sepal vernation, different chromosome base numbers and a unilocular ovary (Albert and Chase, 1992; Albert, 1994).

The systematics of the genus Paphiopedilum proposed by Cribb (1997b) are largely consistent with Atwood (1984), except that Cribb placed the Parvisepalum group within subgenus Brachypetalum. Cribb (1997b) accepted the suggestion of Karasawa (1982) and Karasawa and Saito (1982) to promote the Parvisepalum group (e.g., Parvisepalum delenatii, Parvisepalum armeniacum, Parvisepalum micranthum, Parvisepalum malipoense, and Parvisepalum emersonii) to the subgeneric rank, since the two relatively new species (i.e., P. malipoense and P. emersonii) found in this group have been described. According to the classification of Cribb (1997b), the genus Paphiopedilum comprised of approximately 69 species worldwide. Cribb divided this genus into three subgenera, Parvisepalum, Brachypetalum, and Paphiopedilum, which are mainly based on the morphological characteristics of flower inflorescence, leaf type, floral morphology, and molecular data on ITS sequences (Cox et al., 1997). Recently, several new species and treatment have been described for this genus. The genus Paphiopedilum was described as containing approximately 98 species worldwide by the year 2000 (Koopowitz, 2000). In this genus, Paphiopedilum was divided into five sections: Coryopedilum, Pardalopetalum, Cochlopetalum, Paphiopedilum, and Barbata. Subgenera of the genus Paphiopedilum distribute in distinct geographic regions (Cribb, 1997b). The subgenera Parvisepalum and Brachypetalum, as well as the section Paphiopedilum of the subgenus Paphiopedilum, are found only in mainland Asia. The Parvisepalum subgenus is concentrated in southern China and Vietnam, the subgenus Brachypetalum is mostly found in Thailand (Figure 1). Among the subgenus Paphiopedilum, Paphiopedilum ranges from India to southern China, Thailand and Indo-China, and the species diversity found in southern China was the most concentrated. The section Cochlopetalum is restricted to the islands of Sumatra and Java. The section Pardalopetalum is widespread in Southeast Asia, the Malay Archipelago, as far east as Sulawesi, and Luzon in the Philippines. The enormous species diversity of the section Coryopedilum locates in Borneo, and this section range from the Philippines to Sulawesi in New Guinea. The section Barbata is widespread from eastern Nepal, across to Hong Kong and the Philippines, south to the Malay Archipelago, New Guinea, and the Solomon Islands (Cribb, 1998).

Paphiopedilum is a genus of tropical Asiatic origin, and its range extends eastward, reaching the Philippines, Southeast Asia, Borneo, and the Malay Archipelago, crossing Wallace's Line into Sulawesi, the Moluccas, New Guinea, and the Solomon Islands (Cribb, 1998). Tracking back to the geological history of Southeast Asian, the Palawan, Mindoro, Zamboanga, and the adjacent small islands are the older islands of the Southern Philippines. These regions are located on the border of the Eurasian Plate and have been shifting away from the mainland mass by tectonic collision since the early Miocene (~30 Mya) and the shell of the older plate was merged to Borneo until 5~10 Mya (Karig et al., 1986; Stephan et al., 1986; Hall, 1996). In contrast, most of the Philippine islands formed less than 5 Mya (Aurelio et al., 1991; Quebral et al., 1994). In addition, the Sundaland was comprised of the Malay Peninsula, Sumatra, Java, Sulawesi, and Borneo and merged with Bali, the Philippines, and even New Guinea/Australia into Sunda Superland interconnecting by land bridge during the last glacial period (0.01~1.8 Mya) (van Oosterzee, 1997). Since the last glacial period, species migrated forward and backwards between these regions and isolated after the last glacial maximum (LGM), causing the broken of Sunda Superland (Tsai et al., 2015).

The chloroplast primers for the atpB-rbcL, trnL-trnF spacer, and trnL intron are useful for phylogenetic studies at the intrageneric level. The primers for the trnL-trnF spacer and trnL intron developed by Taberlet et al. (1991) have been applied for inferring phylogenies at the intrageneric level (Goldblatt et al., 2002; Hodkinson et al., 2002; Mogensen, 1996; Van Raamsdonk et al., 2003), and have also been used successfully on Orchidaceae (Tsai et al., 2012). The atpB-rbcL regions are high length differences due to frequent occurrence of indels and are often used in combination with other primers to provide more information (Yoshinaga et al., 1992; Chiang et al., 1998; Von Konrat et al., 2010). Therefore, this study aims to further elucidate the phylogeny of Paphiopedilum through analysis ITS (internal transcribed spacer) sequences and three non-coding plastid DNA sequences (trnL intron, trnL-F, and atpB-rbcL spacers). In addition, the biogeography of this genus is clarified based on the phylogenetic tree derived from the molecular evidence.

#### MATERIALS AND METHODS

#### Plant Materials

Seventy-eight taxa of Paphiopedilum and two outgroups from genus Phragmipedium were used in this study (Table 1). All leaf materials were taken from living plants in the greenhouse of the Kaohsiung District Agricultural Improvement Station (KDAIS) in Taiwan.

#### DNA Extraction, PCR Amplification, and Sequencing

Total DNA was extracted from fresh etiolated leaves by using the cetyltrimethylammonium bromide (CTAB) method (Doyle and Doyle, 1987). Approximate DNA yields were determined by using the spectrophotometer (model U-2001, Hitachi).

The PCR reaction was used to amplify nuclear ribosomal ITS sequence and chloroplast (cp) DNA fragments trnL intron and the trnL-trnF spacer, atpB-rbcL spacer. ITS primers were designed from conserved regions of the 3' end of the 18S rRNA gene and the 5'end of the 26S rRNA gene using sequences from different species present in GenBank. Universal primers for trnL intron and the trnL-trnF spacer were referenced from Taberlet et al. (1991). Primer sequences for amplifying of the atpB-rbcL spacer were designed from the conserved regions of the 3' end of the atpB gene and the 5'end of the rbcL gene of chloroplast DNA using sequences of different species obtained from GenBank. Detailed amplification conditions and primer sequences are given in Supplementary Table S1. All PCR products were separated by agarose gel electrophoresis (1.0%, w/v in TBE) and were recovered using glassmilk (BIO 101, California).

PCR products were directly sequenced using the dideoxy chain-termination method on an ABI377 automated sequencer with the Ready Reaction Kit (PE Biosystems, California) of the BigDye™ Terminator Cycle Sequencing. The PCR reaction primer sequences were used as sequencing primers. Each sample was sequenced two or three times to confirm the sequences. Reactions were performed as recommended by the product manufacturers.

#### Sequence Alignment and Phylogenetic Reconstruction

The sequence alignment was determined using the ClustalW multiple alignment program in BioEdit (Hall, 1999), and four regions were combined for the following analysis. The alignment was checked, and apparent alignment errors were corrected by hand. Indels (insertion/deletions) were treated as missing data. For phylogenetic reconstruction, two Phragmipedium taxa treating as outgroups were sequenced to resolve whether all ingroup taxa formed a monophyletic lineage. The best-fitting substitution model was selected (Supplementary Table S2) by a model test using MEGA 6.0 (Tamura et al., 2013). Tamura 3 parameter model (T92) using a discrete Gamma distribution (+G) was selected for following neighbor-joining (NJ) phylogenetic reconstruction. The general time reversible (GTR) using a discrete gamma distribution (+G) and considering the proportion of invariable sites (+I) were chosen for following divergence time estimation using the Yule model methods in BEAST 1.8.0 (Drummond and Rambaut, 2007; Drummond et al., 2012). The phylogenetic tree for the combined multiple sequence datasets used equally weighted characters. Moreover, because the sequence data of the four genera (Mexipedium, Selenipedium, Cypripedium, and Goodyera) in NCBI is limited, only two sets of fragment data (ITS: Mexipedium xerophyticum-MK161260.1; Selenipedium aequinoctiale-JF825977.1; Cypripedium\_macranthos-KT338684.1; Goodyera\_procera-MK451741.1and trnL-trnF spacer: Mexipedium xerophyticum-FR851215.1; Selenipedium aequinoctiale-JF825975.1; Cypripedium\_macranthos-JF797026.1; Goodyera\_procera-MK451782.1) are used as an additional analysis and compared with the data using only genus Phragmipedium as outgroup. The results of six outgroups are showed in Supplementary Data. TABLE 1 | Names of specimens, geographical distribution, source, and GenBank accession numbers for sequences of the internal transcribed spacer (ITS) of ribosomal DNA (rDNA), the plastid trnL intron, the trnL-F spacer, and the atpB-rbcL spacer.


#### TABLE 1 | Continued


#### TABLE 1 | Continued


a The systematics of Phalaenopsis are based on Christenson (2001).

b Voucher specimens were deposited at the herbarium of the National Museum of Natural Science, Taiwan (TNM).

Genetic relationships were determined using NJ in the MEGA 6.0 (Tamura et al., 2013), maximum parsimony (MP) in PHYLIP 3.68 (Felsenstein, 2004), and maximum likelihood (ML) in MEGA 6.0 (Tamura et al., 2013). Bootstrapping (1,000 replicates) was carried out to estimate the support for NJ, MP, and ML topologies (Felsenstein, 1985; Hillis and Bull, 1993). The strict consensus parsimonious tree was then constructed by using the MEGA 6.0 (Tamura et al., 2013).

#### Divergence Time Estimation

The combined chloroplast DNA (cpDNA) dataset was used to estimate the divergence times using the Bayesian Yule model methods (BEAST version 1.7.5). The characteristic of uniparental inheritance in cpDNA prevents the interference of recombination introgression on phylogenetic reconstruction (Drummond et al., 2012). The general-time reversible (GTR) model with estimates of invariant sites (+I) and gamma-distributed among site rate variation (+G) in all matrices without partitions model was determined by the nucleotide substitution model test, conducted in MEGA 6.0 (Tamura et al., 2013).

To estimate the divergence time, two strategies, the strict and relaxed clock models, were adopted. For the relaxed clock, the calibration point at the most recent common ancestor (MRCA) of Paphiopedilum and Phragmipedium for 22 Ma (Gustafsson et al., 2010) were used to calculate the divergence times of each node. However, since there is no suitable fossil record to correct the calibration of divergence times for the ingroup, we recalculate the divergence time by strict clock model for consistency. For the strict clock, the mean substitution rate of 1.82 × 10−<sup>9</sup> subs/ site/year with the lower and upper limits 1.11 × 10−<sup>9</sup> subs/site/year and 2.53 × 10−<sup>9</sup> subs/site/year, respectively, were used for the cpDNA spacers in Phalaenopsis amabilis complex (Tsai et al., 2015).

We conducted four independent runs of a Yule prior and four Markov Chain Monte Carlo (MCMC) chains with a different starting seed. The first 10% simulations were discarded (burn-in) in a total of 10<sup>8</sup> generations. For thinning, one tree was reserved every 10,000 trees, and finally, 10,000 trees were left to calculate the posterior probability of each node. The effective sample size (ESS) > 200 was used as a criterion to check whether the sampling (simulations) is proper and is reaching a stationary distribution by Tracer v1.6 (Rambaut et al., 2018). Four independent-runs results, including the log file and tree file, were combined with the assistance of LogCombiner 1.6.1 (Drummond and Rambaut, 2007). TreeAnnotator 1.6.1 (Drummond and Rambaut, 2007) was used to summarize a consensus tree with a criterion of the maximum clade reliability using the mean heights option. The final consensus tree was drawn by FigTree 1.3.1 (Rambaut, 2009).

#### Biogeographic Inference Using Reconstruct Ancestral State in Phylogenies

The Statistical dispersal–vicariance analysis was used to assess the biogeographic patterns of Paphiopedilum species [Statistical Dispersal-Vicariance Analysis (S-DIVA), (Yu et al., 2010)]. Bayes–Lagrange Statistical dispersal–extinction–cladogenesis (S-DEC) model (Ree and Smith, 2008) was performed in Reconstruct Ancestral State in Phylogenies (RASP) 3.2 (Yu et al., 2015) to distinguish the events of vicariance, dispersal, and extinction. Five geographic areas were determined mainly according to Myers et al. (2000) with a little modification to illuminate the vicariance, dispersal and extinction events of Paphiopedilum species. The hotspot areas in South-Central China and Indo-Burma with the Malay Peninsula were combined as area A, consisting of China, Nepal, India, Bhutan, Burma, Thailand, Malaysia, Cambodia, Vietnam, and Laos. The hotspot "Sundaland" including Borneo, Java, and Sumatra were set as the area B. We move the Malay Peninsula from the area "Sundaland" to area A due to the integrity of the current landmass. The hotspots "Wallacea" (include Sulawesi and Moluccas) and "Philippines" were set as area C and area E. Respectively, islands of New Guinea eastern from the Wallacea are defined as area D. Species of outgroup were all defined as the area I. These two outgroup species are distributed in Ecuador, Peru, Costa Rica, Panama, and Colombia. The real distribution of outgroups is too far from the areas of the species in this study. Therefore, the ranges of outgroups are assigned to a new area in which none of the ingroup species occurs (Yu et al., 2014). The ML tree topologies were used in S-DIVA analysis.

### RESULTS

#### Sequence Alignment and Characteristics

The lengths of the ITS sequences obtained from the Paphiopedilum and outgroup samples were similar to those reported for a broad example of angiosperms (Baldwin, 1992; Baldwin et al., 1995). The alignment length of the ITS sequence is 735 nucleotides, of which 343 were identified as variable sites with 235 potentially parsimony informative sites. The average genetic distance between the 78 Paphiopedilum samples was 0.039 in ITS, and the average genetic distance between the 78 Paphiopedilum species was 0.01 in cpDNA. The alignment of combined plastid DNA fragments contained a total of 2,409 characters, of which 872 were identified as variable sites with 588 potentially parsimony informative sites. Since the sequences of three samples of every species are the same within species, only one sequence per species was used for the analyses and deposited in NCBI GenBank. The accession numbers of the nuclear ribosomal ITS sequences and the three fragments of plastid DNA from the 78 Paphiopedilum taxa and the two outgroup samples (from the genus Phragmipedium) are shown in Table 1.

#### Phylogeny Reconstruction

Both NJ and MP trees revealed a monophyletic relationship of 78 Paphiopedilum taxa with high bootstrap supports (Figure 2). Moreover, the use of six outgroups for phylogeny reconstruction showed similar bootstrap supports (Supplementary Figure S1). In the Paphiopedilum monophyletic clade, three subgenera Parvisepalum, Brachypetalum, and Paphiopedilum formed independent monophyletic clades with 100/100/100, 98/91/84, and 61/59/86% bootstrap supporting values in NJ/MP/ML trees, in which subgenus Parvisepalum was diverged firstly from other lineages (Figure 2, Supplementary Figure S1). In subgenus Paphiopedilum, sections Barbata, Cochlopetalum, Pardalopetalum, and Coryopedilum are monophyletic with 94/ 96/100, 79/81/94, 100/100/100, and 100/100/100% bootstrap supporting values in NJ, MP, and ML trees. Additionally, the use of six outgroups for phylogeny reconstruction showed similar patterns (Supplementary Figure S1). However, section Paphiopedilum showed a low support [56/51/54% bootstrap values (Supplementary Figure S2); 59% bootstrap values (Supplementary Figure S1)] for the monophyly.

In addition, the molecular data demonstrates that a newly described variety, P. micranthum var. eburneum, is closely related to P. malipoense based on the plastid DNA within subgenus Parvisepalum, which is inconsistent with the inference by nuclear ITS and combined data. In ITS tree,P.micranthumvar.eburneumis sister with P. micranthum var. micranthum (Figures 2 and 3). Therefore, we infer a hybridization event between the ancestor of P. micranthum and P. malipoense that lead to a plastid capture in P. micranthum var. eburneum.

#### Divergence Time Estimates

The coalescence time of the genus Paphiopedilum was estimated to be 7.09 Mya with 95% confidence intervals (95% CI) of 5.88– 8.41 Mya (Figure 4) according to the substitution rate referenced from Tsai et al. (2015). If calibrating by relaxed clock referring to Gustafsson et al. (2010), the estimated coalescence time of the genus Paphiopedilum was 5.72 Mya (Figure 4). In the genus Paphiopedilum, the coalescence times estimated by strict clock were 4.30 Mya (95% CI: 3.50–5.18 Mya), 2.47 Mya (95% CI: 1.73– 3.33 Mya), and 4.08 Mya (95% CI: 3.39–4.86 Mya) for subgenera Parvisepalum, Brachypetalum, and Paphiopedilum, respectively (Figure 4). After re-calibrating by a relaxed clock, the coalescence time was estimated to be 3.3 Mya, 2.24, and 3.38 Mya Mya for subgenera Parvisepalum, Brachypetalum, and Paphiopedilum, respectively (Figure 4). In addition, the strict clock suggested that the coalescence times were tracked back to 2.19 Mya (95% CI: 0.17–2.77 Mya), 1.54 Mya (95% CI: 0.93–2.27 Mya), 3.12 Mya (95% CI: 2.49–3.81 Mya), 2.48 (95% CI: 1.86–3.17 Mya), and 1.60 Mya (95% CI: 1.10–2.18 Mya) for clades of subgenera Barbata, Cochlopetalum, Paphiopedilum, Coryopedilum, and Pardalopetalum of subgenus Paphiopedilum, respectively. By relaxed clock, the coalescence time was estimated to be 1.94, 1.3, 2.59, 1.81, and 1.08 Mya for clades of subgenera Barbata, Cochlopetalum, Paphiopedilum, Coryopedilum, and Pardalopetalum of subgenus Paphiopedilum, respectively. Additionally, the use of six outgroups for divergence time estimation also showed similar supports (Supplementary Figure S2). In short, the estimates of the coalescence times by strict and relaxed clocks are similar, but the time calculated is slightly shorter by relaxed clocks. Regardless, the coalescence time of the genus Paphiopedilum will not be earlier than Upper Miocene.

#### Demographic History and Historical Biogeography Inference

Complicated evolutionary processes of continuous and episodic dispersal, vicariance, and extinctions determined the current geographic distribution of genus Paphiopedilum. Since the most probable ancestral areas located on continental Asia (area A in Figure 5), dispersal events seem to determine the extant distributions of subgenera largely. The results supported vicariance events on nodes 159, 146, and 109 shown in Table 2 and Figure 5, and on nodes 163, and 132 (Supplementary Table S3 and Figure S3). The node 147 revealed dispersal events among sectionCoryopedilum/Pardalopetalum of Paphiopedilum and other sections of Paphiopedilum causing by migration route from area A (China, Nepal, India, Bhutan, Burma, Thailand, Malaysia, Cambodia, Vietnam, and Laos) to B area (Sumatra, Borneo, and Java). Meanwhile, the nodes 145, 129, 114, and 102 also revealed dispersal events from north to south, according to Sundaland and Sunda Super Islands.

Furthermore, the use of six outgroups for dynamic historical inference showed similar patterns (Supplementary Table S3 and

FIGURE 2 | Phylogenetic relationships using neighbor-joining (NJ), maximum parsimony (MP), and maximum likelihood (ML) resulting from analysis of the combined data matrix (nuclear ribosomal ITS, plastid trnL intron, trnL-F spacer, and atpB-rbcL spacer) from 78 Paphiopedilum and 2 outgroup species. Only the strict consensus of all most parsimonious trees (MP trees) are showed in this figure, and the bootstrap values > 50% are shown on each branch for NJ/MP/ML between major lineage.

Figure S3). Only the nodes 146 and 109 were detected vicariance event causing by the geological separation between Indochina and Sumatra/Borneo/Java. In addition, in subgenus Paphiopedilum, 2 vicariance and 10 dispersal events were detected, which suggesting a significant dispersal process affected biogeographical patterns in shaping the current distribution in the subgenus Paphiopedilum. Areas A and B might be the two possible ancestral areas and likely shaped by several complicated dispersal events in the subgenus Paphiopedilum.

#### DISCUSSION

#### Systematics Revision of Genus Paphiopedilum

In general, our phylogenetic inference is mostly congruent with that of Cox et al. (1997), Cribb (1997b), and Guo et al. (2015). In the genus Paphiopedilum, tessellated leaves, single flowers with broad elliptic to subcircular petals, and a sizeable thin-textured lip characterize subgenus Parvisepalum in southwest China and Vietnam (Cribb, 1998). Within this subgenus, the phylogenetic topography and the divergence time of at least 4.30 Mya rejected the previous hypothesis of the sister-species relationship between P. armeniacum and P. delenatii (Cribb, 1983) (Figure 4). The geographical distribution of these two species is also separated (Yunnan, China for P. armeniacum, and Vietnam for P. delenatii) (Cribb, 1998).

Furthermore, a newly described variety, P. micranthum var. eburneum, is phylogenetically close to P. malipoense in maternalinherited plastid DNA but close to P. micranthum in biparentalinherited nuclear ITS sequences (Figure 5), suggesting that P. micranthum var. eburneum is a natural hybrid between the maternal parent P. malipoense and the paternal parent P. micranthum and experienced the event of chloroplast

FIGURE 3 | Parsimonious phylogenetic tree of the Paphiopedilum subgenus Parvisepalum derived from ITS sequences of nuclear ribosomal DNA (nrDNA) (A) and plastid DNA (B). The solid circle (․) represents a putative natural hybrid (based on molecular evidence).

with the lower and upper limits 1.11 × 10−9 subs/site/year and 2.53 × 10−9 subs/site/year (Tsai et al., 2015). The red numbers in each node are using the Gustafsson et al. (2010) fossil calibration time data to calibrate the divergence time.

capture. The overlap of the geographical distribution of these three taxa also supports this hypothesis (Cribb, 1998). In addition, ITS sequences are usually concertedly evolved via unequal crossing-over (Schlotterer and Tautz, 1994) and biased gene conversion (Hillis et al., 1991), which results in sequence homogeneity between paralogs (Maynard, 1989).

The monophyly of subgenus Brachypetalum inferred in this study is congruent with the inference by Cox et al. (1997). The subgenus Brachypetalum is geographically confined to Southeast Asia (Cribb, 1998). Albeit overlapping distribution with subgenus Parvisepalum (Cribb, 1998), subgenus Brachypetalum is phylogenetically separated, consistent with the distinguishable leaf anatomy between these two subgenera (Cribb, 1998). Both molecular and morphological evidences support the independent taxonomic treatment between subgenera Brachypetalum and Parvisepalum (Karasawa, 1982; Karasawa and Saito, 1982; Cribb, 1997b), but object with Atwood's (1984) opinion of taking the subgenus Parvisepalum as a synonym of Brachypetalum.

The monophyletic subgenus Paphiopedilum can be morphologically and phylogenetically subdivided into five sections: Coryopedilum, Pardalopetalum, Cochlopetalum, Paphiopedilum, and Barbata (Cox et al., 1997; Cribb, 1997b). Section Coryopedilum is characterized by its plain green, straplike leaves, and multi-flowered inflorescences, which flowers are blooming simultaneously (Cribb, 1998). This section distributes throughout Borneo, Sulawesi, New Guinea, and the Philippines (Cribb, 1998). Except placing Paphiopedilum parishii and Paphiopedilum dianthum into section Pardalopetalum from section Coryopedilum, Cribb (1997b) agreed with Atwood (1984) that section Pardalopetalum is independent from

section Coryopedilum taxonomically, according to the ITS analysis (Cox et al., 1997) and similar green strap-like leaves and staminodes (Cribb, 1997b), which is consistent with our phylogenetic inference. However, the only character that separates sections of Pardalopetalum and Coryopedilum is the morphology of staminode. Whether this single character is sufficient to characterize them as separating sections should be re-evaluated with more evidence.

Unlike the simultaneous bloom of section Coryopedilum, section Cochlopetalum flower in succession, and their flowers bear elliptic bracts, linear, spirally twisted, spreading, ciliate petals, and a pot-shaped spotted lip (Cribb, 1998). Section

and (I) outgroup], respectively. The green and blue circles indicate the vicariance and dispersal events obtained from the RASP analysis, respectively.

Cochlopetalum distributes in Sumatra and Java only (Cribb, 1998). The extensive section Barbata is the sister of section Cochlopetalum, also characterized by a solitary flower with a lip and prominent incurved side-lobes, but the leaf tessellated (Cribb, 1998). The morphological dissimilarity and reciprocally monophyletic relationship indicate that, despite recently diverged, these two sections should be independent taxonomically.

### Biogeography and Evolutionary Trends

The clade of genus Paphiopedilum is coalesced to 7.09 or 5.72 Mya, similar to the estimate of 7.62 Mya by Guo et al. (2015). The flower morphology of subgenus Parvisepalum is intermediate between other subgenera of Paphiopedilum and Cypripedium (Chen and Tsi, 1984), which could be explained by the earlier divergence of subgenus Parvisepalum in genus Paphiopedilum (Guo et al., 2012). Presently, the genus Cypripedium is distributed throughout worldwide temperate zones (Cox et al., 1997), with China as a center for species diversity (Cribb, 1997a). Therefore, genera Paphiopedilum and Cypripedium have most likely diverged in mainland Asia (Chen and Tsi, 1984).

However, genus Paphiopedilum was suggested as the sister with two American genera Phragmipedium and Mexipedium according to morphology, plastid rbcL (Albert, 1994), ITS (Cox et al., 1997), and both nuclear and plastid genes (Guo et al., 2012). These inferences are conflict to the hypothesis of the divergence between Paphiopedilum and Cypripedium in China, but implied the divergence of Paphiopedilum from the group of Phragmipedium + Mexipedium, by which the slipper orchids (Cypripedioideae) were hypothesized widespread throughout North America and Asia in the past (Atwood, 1984; Albert, 1994; Cox et al., 1997).

Subgenus Parvisepalum in southwest China and Vietnam diverged earlier from the other subgenera of genus Paphiopedilum. The coalescence time of subgenera Parvisepalum, Brachypetalum, and Paphiopedilum were tracking back to the Upper Miocene (Guo et al., 2015). Subgenus Brachypetalum in mainland Southeast Asia was descended from the subgenus Parvisepalum inferred by S-DIVA (Figure 5), which agrees with other disjunctions at the Southern China and Indochina (Guo et al., 2015) or Sunda Shelf and New Guinea/Australia (Tougard, 2001; Lohman et al., 2011; Tsai et al., 2015).

The subgenus Paphiopedilum is further descended and evolved quickly in the Sunda Shelf. A land bridge might connect Mindoro, Palawan, Borneo, the Malay Peninsula, Borneo, Sumatra, Java, Bali, and various parts of the Philippines when sea levels falling during Pleistocene (about 0.01~1.8 Mya) (van Oosterzee, 1997) (Figure 1). Surfaced land bridge connected these regions and was beneficial to the interisland and continent-island dispersal in Southeast and South Asia (van Oosterzee, 1997). The clade of Coryopedilum + Pardalopetalum was the first derived in subgenus Paphiopedilum based on the phylogenetic tree, which reflects in the sympatric distribution of subgenus Brachypetalum and section Pardalopetalum (Figure 6). Following this clade formation, sections Paphiopedilum and Barbata were subdivided and dispersed throughout Southeast Asian archipelagoes across the land bridge during the glacials. The southward expansion from continental Asia into the Greater Sunda Islands through the Indochina and Malay Peninsulas were also reported in other taxa, e.g., Lithocarpus (Fagaceae) (Yang et al., 2018). Such colonization events between continental Asia and the Greater Sunda Islands mostly occurred during Miocene and Plio-Pleistocene (de Bruyn

TABLE 2 | The ancestral areas and dispersal–vicariance analysis inferred through Reconstruct Ancestral State in Phylogenies (RASP). Ancestral areas for the node and the number of dispersal (Dis), vicariance (Vic), and extinction (Ext) events are shown.


et al., 2014). As a "corridor," Indochina reveals high flora diversity and the high species richness, which facilitates the in situ speciation (de Bruyn et al., 2014).

Another flora diversity hotspot is Borneo (de Bruyn et al., 2014), which is also important for the genus Paphiopedilum. The section Cochlopetalum, which is found only in Sumatra and Java, might represent a group derived from Borneo. The S-DIVA inferred multiple times dispersal events sourced from Borneo with two vicariances to illustrate the current distribution of the five sections within subgenus Paphiopedilum (Figure 5 and Table 2). The Borneo is the second original center of Paphiopedilum. The tropical forests and rugged topography harbor diversified niches opened to the speciation of organisms. The repeated submergence and emergence of land bridges could promote repeated genetic isolation and gene flow between closed related taxa. During the Plio-Pleistocene glacial oscillations. This process accelerates the diversification rates in the Sunda Super Islands.

Dispersal and vicariance events that exposed geographic isolation among taxa might be due to the land bridge submergence (Chiang et al., 2009; Chiang et al., 2013; Ge et al., 2012; Ge et al., 2015; Hsu et al., 2013) in Sunda Shelf and Sunda Super Island during the Pleistocene (Figure 5 and Table 2) (Guo et al., 2015; Tsai et al., 2015). Cenozoic collision accompanied by a cyclical climate (glacial oscillations) caused by the fragmentation of the Sunda Super Islands (de Bruyn et al., 2014). The Borneo, Java, Sumatra, and the southern Philippines belong to the Sunda plate, Sulawesi is composed of broken plates, and the Moluccas and New Guinea belong to the Australian plate (Hall, 1998). The deep trenches between these plates cause segregation of species between Sundaland and the islands in the east. Because of the seed germination relies on symbiotic fungi, the geographical isolation maybe not only influences orchid itself but also in symbiotic fungi. However, these inferences still need further verification.

### CONCLUSIONS

In summary, the origin and coalescence time of genus Paphiopedilum tracked back to Southern China/Eastern Indochina since late Miocene and early Pliocene, while the range expansion and species divergence were related to sea-level fluctuations during the Plio-Pleistocene glacial cycles. Historical geological barriers shaped a pattern of vicariance among disjunct distributed subgenera after isolated ancestral populations. The ancestral taxa of subgenus Paphiopedilum migrated from Southern China/Eastern Indochina to south which developed quickly in the Sunda Shelf. Due to the submergence of the Sunda Shelf and Sunda Super Island, species of subgenus Paphiopedilum dispersedwith isolation between islands as well as subsequentin situ speciation within islands from other taxa within section or subgenus, which accelerated species divergence in subgenus Paphiopedilum. Paphiopedilum distributes in four of 25 biodiversity hotspots (Myers et al., 2000), the Indo-Burma, Sundaland, Wallacea, and Philippines, where are also the "major evolutionary hotspots"(de Bruyn et al., 2014). It is suggests that rich and fascinating historical biogeographic events have created rich species diversity there, such as the case of Paphiopedilum. However, deforestation has caused the so-called "empty forest syndrome"(de Bruyn et al., 2014). We hope that these areas will not become extinction hotspots, even though they are almost now.

### DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

### AUTHOR CONTRIBUTIONS

Conceived and designed the experiments: C-CT and Y-CC. Performed the experiments: C-CT, P-CL, Y-ZK, C-HC, and Y-CC. Analyzed the data: C-CT, P-CL, Y-ZK, C-HC, and Y-CC. Contributed reagents/materials/analysis tools: C-CT, P-CL, Y-ZK, C-HC, and Y-CC. Wrote the paper: C-CT, P-CL, and Y-CC. Conceived of the study, edited the manuscript, and approved the final manuscript: C-CT, P-CL, Y-ZK, C-HC, and Y-CC.

### FUNDING

This research was supported by funding from the Ministry of Science and Technology, Taiwan (MOST 105-2621-B-110-003- MY3 and MOST 105-2621-B-110-001) to Y-CC and by partial financing (the Higher Education Sprout Project) of NSYSU.

#### ACKNOWLEDGMENTS

The achievement of this study is dedicated to the memory of the scientific contributions of the first author C-CT, who died of a stroke on November 05, 2015.We are grateful to Dr. F. H. Linfor his valuable comments and helpful discussion in the course of the study.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpls.2020.00126/ full#supplementary-material

SUPPLEMENTARY FIGURE S1 | Phylogenetic relationships using Maximum Likelihood resulting from analysis of the combined data matrix (nuclear ribosomal ITS, and trnL-F spacer) from 78 Paphiopedilum and 6 outgroup species.

#### REFERENCES


SUPPLEMENTARY FIGURE S2 | Results of calescence time estimations performed with BEAST 1.8.0 for the from 78 Paphiopedilum taxa and 6 outgroup species based on the combined data matrix (nuclear ribosomal ITS, and trnL-F spacer).

SUPPLEMENTARY FIGURE S3 | Ancestral distributions reconstructed by the Statistical dispersal–vicariance analysis (S-DIVA, Yu et al., 2010) and Bayes– Lagrange Statistical dispersal–extinction–cladogenesis (S-DEC) model (Ree and Smith, 2008) performed in RASP 3.2 (Yu et al., 2015). Phylogenetic relationships of the 78 Paphiopedilum species, plus the six outgroups, obtained from sequence judgments of the combined sequence and generated by BEAST. The distribution areas of extant the 78 Paphiopedilum species species are marked in capitals A–E and I ((A) China, Nepal, India, Bhutan, Burma, Thailand, Malaysia, Cambodia, Vietnam, and Laos; (B) Sumatra, Borneo and Java; (C) Sulawesi and Moluccas; (D) New Guinea; (E) Philippines; and (I) outgroup), respectively. The green and blue circles indicate the vicariance and dispersal events obtained from the RASP analysis, respectively.


Conflict of Interest: 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 © 2020 Tsai, Liao, Ko, Chen and Chiang. 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.