Forest tree genomics: 10 achievements from the past 10 years and future prospects

Abstract

Key message

This review highlights some of the discoveries and applications made possible by “omics” technologies over the last 10 years and provides perspectives for pioneering research to increase our understanding of tree biology.

Context

A decade after the first forest tree genome sequence was released into the public domain, the rapidly evolving genomics and bioinformatics toolbox has advanced our understanding of the structure, functioning, and evolution of forest tree genomes.

Aims and methods

This review highlights some of the discoveries and applications that “omics” technologies have made possible for forest trees over the past 10 years.

Results

In this review, we start by our current understanding of genome evolution and intricacies of gene regulation for reproduction, development, and responses to biotic and abiotic stresses. We then skim over advances in interactome analysis and epigenomics, the knowledge of the extent of genetic variation within and between species, revealing micro- and macro-evolutionary processes and species history, together with the complex architecture of quantitative traits. We finally end with applications in genetic resource conservation and breeding.

Conclusion

The knowledge gained through the use of these technologies has a huge potential impact for adapting forests to the main challenges they will have to face: changing demand from ecosystem services with potentially conflicting strategies in terms of conservation and use, as well as climate changes and associated threats. Genomics will undoubtedly play a major role over the next decade and beyond, not only to further understand the mechanisms underlying adaptation and evolution but also to develop and implement innovative management and policy actions to preserve the adaptability of natural forests and intensively managed plantations.

1 Introduction

Forest system currently faces a number of major challenges, including increases in the demand for wood, pressure to conserve forest areas, and global climate change and associated threats. The adaptation of these systems in response to these challenges will require a multifaceted approach, in which genomic sciences have an important role to play. The aim is to apply a series of technologies from the fields of genetics, molecular and cell biology fostered by bioinformatics, and robotics to analyses of the structure, function, and evolution of sets of genes, right up to complete genomes, with the use of high-throughput methods. These methods include the structural characterization of entire genomes, genes, mRNA, and proteins, the genome mapping and sequencing (genomics), the study of mRNA, proteins, and metabolite abundances across different environmental conditions and/or developmental stages (functional genomics, i.e., transcriptomics, proteomics, metabolomics), the analysis of epigenetic modifications (epigenomics), the inference of evolutionary mechanisms (comparative and population genomics related to macro- and micro-evolutionary processes, respectively), and the study of complex biological systems (community genomics, metagenomics). Since the publication of the first tree genome sequence (black cottonwood, Tuskan et al. 2006), technologies and genetic resources facilitating research into forest tree genomics have advanced our understanding of the tree growth and development (resulting in large and long-lived organisms), the responses of trees to intrinsic (ontogenic phase change) and extrinsic (biotic and abiotic) factors, the remarkable buffering capacity (plasticity) of trees, enabling them to cope with chronic stresses and extreme events, the molecular basis of genetic variation within and between species, and the way in which this variation has been shaped by evolutionary forces and its relationship to phenotypic variation and adaptation. The objective of this review, which is not intended to be exhaustive, is to outline the tremendous progress achieved in forest tree genomics over the last 10 years. This progress is illustrated by considering a series of 10 achievements. We will begin by skimming over the knowledge acquired from the sequencing of forest tree genomes. We will then move on to the key developmental traits underlying the biology of woody perennials, the molecular mechanisms driving the responses of trees to biotic and abiotic stresses, and we will tackle two emerging fields (molecular interactions, epigenetics) that promise to improve our understanding of the functioning of tree genomes considerably. Finally, we consider the knowledge gained from the description and interpretation of naturally occurring genetic variation within and between species. This knowledge has provided us with an understanding of the contemporary and historical evolutionary processes that have contributed to the observed patterns of geographic and phenotypic variation and the molecular basis of quantitative trait variation. We will conclude by considering the potential impact of genomic studies on the conservation and improvement of forest tree genetic resources. This review is accessible to readers from diverse backgrounds keen to acquire a basic understanding of the opportunities for tacking the complex issues facing the world’s forests provided by this discipline. A glossary of terms for readers that are not necessarily familiar with genomics or other specialized jargon was added as electronic supplemental material (ESM_1.doc).

2 Accomplishments in forest tree genomics

2.1 What have we learnt from the sequencing of tree genomes?

Over the last 10 years, forest tree genomics has benefited considerably from advances in next-generation sequencing technologies, making it possible to investigate the role of hundreds of genes, to access sequence-based markers for breeding at the genome scale, and to study the evolutionary history of tree species (Neale and Kremer 2011). Even if the sequencing of forest tree genomes lags far behind that of fruit trees and annual crops (http://en.wikipedia.org/wiki/List_of_sequenced_plant_genomes), more than 20 tree genome sequencing projects are currently underway (Neale et al. 2013), and six completed forest tree genomes have yet been released and published. In 2002, the US Department of Energy (DOE, USA) set up an international initiative to sequence the Populus trichocarpa genome, the first tree genome (Tuskan et al. 2006) and only the third plant genome, after Arabidopsis thaliana, the model dicot for plant biology (Arabidopsis Genome Initiative 2000), and rice, the major economic crop that became a model monocot (Goff et al. 2002; Yu et al. 2002; IRGSP 2005). An additional poplar species has since been sequenced, Populus euphratica (Ma et al. 2013), together with a second angiosperm tree species, Eucalyptus (Myburg et al. 2014), and three conifers, Norway spruce (Nystedt et al. 2013), white spruce (Birol et al. 2013), and loblolly pine (Neale et al. 2014; Zimin et al. 2014). In the first section of this review, we will briefly summarize the new findings obtained by exploring these first forest tree genomes, in terms of specific aspects of tree biology, such as angiosperm/gymnosperm wood formation and life history traits.

The Populus genome sequence provided the first insight into the genome structure and functional biology of a specific tree. The poplar genome is about four times larger than that of Arabidopsis. This larger size results mostly from a complex history of whole-genome duplications, chromosomal rearrangements, and tandem duplications, as shown by Tuskan et al. (2006). The poplar genome contains 1.6 times as many genes as the Arabidopsis genome, and the expansion of several gene families associated with tree-specific traits may also have contributed to the evolution of tree biology. The annual formation of wood is just one of a number of processes exclusive to trees (Plomion et al. 2001). Several genes associated with cellulose or lignin biosynthesis were found to occur in duplicated pairs in Populus, whereas only single copies were present in Arabidopsis. For example, the oxidative polymerization of monolignols, the precursors of lignins, involves two different, non-redundant types of oxidative enzymes—laccases and peroxidases—as recently demonstrated for Arabidopsis vascular development (Zhao et al. 2013a). The laccase gene family of Populus is much larger than that of Arabidopsis (51 versus 17 genes, respectively; Berthet et al. 2012). Moreover, phylogenetic and expression studies of these two multigene families showed that the 16 orthologs of the three Arabidopsis lignin-related laccases (AtLAC4, AtLAC11, and AtLAC17) present in poplar were expressed mostly in the stem xylem, but with some exceptions (expression in the root or other organs), suggesting a conservation of function, but with subfunctionalization for some duplicated poplar genes.

If we take a closer look at the process of wood formation, we find that secondary xylem development evolved between gymnosperms and angiosperms. One of the two main differences between these two groups is the lack of S units in conifer lignins. All the genes involved in monolignol biosynthesis in angiosperms were identified in the pine genome, except for the F5H homolog, which is crucial for the biosynthesis of S units in plants (Bonawitz and Chapple 2010, for review; Neale et al. 2014). The second difference concerns water transport and mechanical support functions, both of which are mediated principally by tracheids in conifers, whereas angiosperms have fibers for mechanical support and vessels for water transport. The number of genes encoding vascular-related NAC domain transcription factors, key regulators of xylem vessel differentiation in angiosperms, is much smaller (only two genes identified) in Norway spruce (Nystedt et al. 2013) than in Arabidopsis and poplar, in which 7 and 16 such genes, respectively, have been identified (Ohtani et al. 2011) suggesting a possible expansion and subfunctionalization of the NAC domain transcription factor family in angiosperms, particularly in woody angiosperms.

Unlike annual plants, which have a short lifespan, trees have had to develop long-term defense strategies against insects and pathogens. Interestingly, although the total number of NBS-LRR pathogen resistance genes is similar in Arabidopsis and poplar, some subfamilies are considerably larger in Populus (Tuskan et al. 2006; Kohler et al. 2008; Bresson et al. 2011). Interestingly, the TIR class of NB-LRR proteins (TNLs) is also much larger in conifers (780 genes in loblolly pine and 180 genes in Norway spruce) than in Arabidopsis (3 genes).

In addition to revealing functional differences between genomes, these recent works have increased our understanding of the life histories of angiosperm/gymnosperm trees. Populus, like Eucalyptus and grapevine, has undergone at least one whole-genome duplication (WGD) event. Analysis of the Populus genome led to the identification of three separate WGD events. The most recent burst of gene creation happened 60 to 65 million years (Myr) ago, due to a single whole-genome event. A second duplication, in a common ancestor of Populus (Fabid, also known as eurosid I) and Arabidopsis (Malvid, also known as eurosid II), was found to have occurred about 100 to 120 Myr ago. A weaker signal was also found for a third, more ancient duplication event, but this event was not dated by the authors (it probably corresponds to the palaeohexaploidy event, see below). A comparison of the estimated molecular clocks of Populus and Arabidopsis revealed a markedly slower rate of sequence evolution in Populus, possibly due to its perennial status, leading to recurrent contributions of “ancient gametes” from old individuals (Tuskan et al. 2006). An analysis of the recently sequenced Eucalyptus genome and phylogenetic studies based on 17 species have suggested that Populus should be reclassified as a Malvid and that Eucalyptus should be placed in a sister taxon to the eurosids (Myburg et al. 2014). A study of the evolutionary history of the Eucalyptus genome also revealed the ancient palaeohexaploidy event (∼130 to 150 Myr), which has also been discovered in grapevine and is common to all eudicots (Jaillon et al. 2007). This event has also been shown to be superimposed over a more recent lineage-specific palaeotetraploidy event (∼110 Myr). WGD is an important mechanism of genome size expansion. As conifers have genomes about 20 to 30 times larger than those of angiosperms, it was intriguing that no evidence for such a mechanism was found in Norway spruce (Nystedt et al. 2013); only the trace of a very ancient WGD predating the divergence of angiosperms and gymnosperms was detected by the authors (∼350 Myr), consistent with previous phylogenetic analysis (Jiao et al. 2011). Instead, it appears that conifer genomes grew mostly through the insertion of repeated elements, principally long terminal repeat-retrotransposons (LTR-RT; essentially the Ty3/Gypsy and Ty1/Copia superfamilies) (Nystedt et al. 2013). The authors proposed a model for genome evolution in conifers (six conifer species were investigated), in which retrotransposon activity began early in evolution and was not countered as efficiently as in angiosperms, resulting in larger gene/pseudogene numbers, and numerous long introns, with genes separated by large regions of transposable element-rich, highly polymorphic DNA in conifers but with the maintenance of synteny over large phylogenetic distances.

2.2 Developmental genomics

As reported in Section 2.1, the development of genomics-enabled research has been instrumental in the identification of genes underlying the traits characteristic of the biology of woody perennials. In this section, we review the progress made by the forest tree genomics community towards elucidating wood formation, bud dormancy, root development in interaction with microbial symbionts, and embryo development in conifers.

2.2.1 Wood formation

Wood or secondary xylem formation results from cambium activity. The secondary xylem fulfills highly specialized functions critical for tree growth and development, such as water conduction from the roots to the crown and mechanical support. Xylem cells—vessels, fibers, and tracheids—specialize in one or both of these functions. These cells differentiate, developing thick and lignified secondary cell walls, before undergoing programmed cell death. In addition, ray cells generated by the cambium connect the phloem to the inner wood. They have storage functions and are also involved in heartwood formation. Trees can give rise to different types of wood—juvenile and mature wood, early and late wood, and reaction wood—with very different properties, depending on their stage of development or in response to environmental cues (Déjardin et al. 2010; Plomion et al. 2001). Wood formation is, therefore, a highly regulated process, and our understanding of which has greatly improved, thanks to advances in genomics.

Genomic studies of wood formation in angiosperms and gymnosperms began more than 15 years ago (Allona et al. 1998; Sterky et al. 1998), with comparisons of transcriptomes and proteomes between different types of wood, leading to the establishment of correlations between the expression of a number of specific genes and specific wood characteristics, such as cell wall composition and properties (Plomion et al. 2000; Déjardin et al. 2004; Gion et al. 2005; Paux et al. 2005; Andersson-Gunneras et al. 2006; Qiu et al. 2008; Paiva et al. 2008a, b; Villalobos et al. 2012).

Furthermore, causal links were established through studies of natural mutants or, more frequently, genetically modified (GM) plants with modified expression of genes involved in cell wall component metabolism: mostly for lignin (van Holme et al. 2013; van Acker et al. 2014) but also for cellulose (Coleman et al. 2007; Coleman et al. 2009), hemicelluloses (Park et al. 2004; Baba et al. 2009; Nishikubo et al. 2011; Zhao et al. 2013b), pectins (Biswal et al. 2014), and other cell wall proteins, such as expansin (Gray-Mitsumune et al. 2008). Nowadays, such modified plants would ideally be characterized by global omics analyses, leading to the establishment of relationships between the phenotype and networks of molecular components, and this approach will undoubtedly lead to the construction of increasingly realistic models explaining wood formation.

The regulation of wood formation has been investigated in a number of studies, leading to the identification of transcription factors involved in secondary cell wall differentiation: this is the case, for example, for ARBORKNOX2 and KNAT7, two Populus homeobox genes (Du et al. 2009; Li et al. 2012), and for PtrMYB152 (Wang et al. 2014a), PtrHB7, a class III HD-Zip gene (Zhu et al. 2013b), and FPF1, a gene also involved in the regulation of flowering (Hoenicka et al. 2012). Likewise, Zhong et al. (2011) demonstrated a key role of PtrWNDs, wood-associated NAC domain transcription factors, as master switches regulating a battery of downstream transcription factors forming a transcriptional network controlling secondary cell wall biosynthesis during wood formation. As described in Section 2.4, the conjunction of several complementary techniques, such as xylem protoplast transfection, RNA-Seq, and ChIP-Seq, has paved the way for the identification of hierarchical gene regulatory networks directed by master transcription factors in wood formation (Lin et al. 2013).

MicroRNAs (miRNAs) have also been implicated in the regulation of wood formation. These miRNAs are short non-coding RNAs with critical regulatory functions. In a pioneering study, Lu et al. (2005) described a number of miRNAs regulated in wood in response to tension and compression stresses. The use of next-generation sRNA sequencing has led to the identification of several miRNAs that are particularly abundant in the xylem: these miRNAs are predicted to target genes known to be important in secondary growth, including the critical reaction wood enzyme xyloglucan endotransglycosylase/hydrolase and vascular system-related transcription factors (Puzey et al. 2012). Furthermore, transgenic poplar trees expressing either a miRNA-resistant POPCORONA or a synthetic miRNA targeting POPCORONA have been used to infer the function of POPCORONA during secondary growth: the synthetic miRNA-mediated knockdown of POPCORONA expression resulted in abnormal lignification in pith cells, whereas the overexpression of a miRNA-resistant POPCORONA delayed the lignification of xylem and phloem fibers during secondary growth. The misexpression of POPCORONA was also shown to result in a coordinated change in gene expression within a previously described transcriptional network regulating cell differentiation and cell wall biosynthesis and in the expression of hormone-related genes associated with fiber differentiation (Du et al. 2011).

In parallel to these important advances in molecular biology, progress has also been made towards the high-throughput phenotyping of cell wall properties. Optimization of the MS-based sequencing of lignin oligomers (Morreel et al. 2010), together with the phenol profiling of plants with lignin modifications, has provided a clear overview of the lignin biosynthesis gene network (van Holme et al. 2012a, b). Chemometric analyses of 2D NMR spectra for cell walls have made it possible to identify changes in cell wall components (Hedenström et al. 2009). Likewise, in situ images of the different chemotypes present in poplar cell walls can be obtained by MicroFTIR analysis (Gorzsás et al. 2011). Finally, a wealth of information has been generated by the development of cell wall polymer mapping with microarrays, using specific monoclonal antibodies and cellulose-binding modules (Moller et al. 2007).

The last few years have seen major technical improvements, leading to the generation of huge amounts of data and opening up new possibilities for network analysis. For example, text mining, co-expression network analysis, and comparative genomics are providing an ever-increasing number of opportunities to identify candidate genes for cell wall biosynthesis (Yang et al. 2011). A genome-wide metabolic pathway database was recently created for P. trichocarpa with pathway analysis tools (Zhang et al. 2010). In the future, a combination of microdissection and high-throughput analysis should make it possible to acquire large amounts of data for increasingly precisely defined samples corresponding to single cell types or to very specific developmental steps. Finally, studies of interactions between molecular components should greatly advance our understanding of wood formation.

2.2.2 Vegetative bud phenology

Perennial plants are immobile organisms that cannot migrate to cope with unfavorable winter conditions. They have developed a strategy for synchronizing their growth and reproductive phases with the favorable environmental conditions. This strategy, dormancy, enables trees both to protect themselves from cold injuries and to ensure an appropriate architecture (reviewed by Preston and Sandve 2013). The term “dormancy” was defined in 1987 by Lang et al. (1987) as a phase during which growth is temporarily suspended, in any plant structure containing a meristem (i.e., meristematic cells). In trees, dormancy occurs mostly in the buds or vascular cambium and can be divided into three main phases according to the factors controlling growth cessation (Chao et al. 2007). The first kind of dormancy is called paradormancy. During paradormancy, growth is stopped by physiological factors external to the affected structure (i.e., apical dominance or correlative inhibition). Paradormancy is followed by “winter dormancy,” which is divided into two main stages: endodormancy, the deepest phase of dormancy (reviewed by Horvath et al. 2003) induced by a decrease in day length, and ecodormancy, which is imposed by unfavorable environmental conditions. Endodormancy is maintained by internal factors specific to the bud itself. Endodormancy generally begins at the end of the summer and steadily increases in intensity during the fall, peaking at some time in November, the precise timing of the peak depending on the species considered (Naor et al. 2003). Endodormancy is broken when chilling requirements are fulfilled. Ecodormancy is established at the release of endodormancy. Ecodormancy occurs during late winter and early spring and is imposed exclusively by environmental factors unfavorable for growth (essentially cold temperatures).

Dormancy (i.e., endo- and ecodormancy) is tightly controlled by both photoperiod and temperature. Photoperiod strongly influences the induction of dormancy, whereas temperature is involved in its release (for both endodormancy and ecodormancy). Temperature thus plays a major role in the phenological cycle of these species. Global warming may strongly affect phenology, because increases in temperatures may extend the growing seasons of trees (reviewed by Menzel and Fabian 1999) and prevent endodormancy release if chilling requirements are not fulfilled during early winter. They may also favor ecodormancy release by accelerating bud cell growth during late winter and/or early spring. There is therefore an urgent need to decipher the mechanisms underpinning dormancy and to identify the genes/polymorphisms that matter for adaptation, because these changes may have severe effects in forest ecosystems, increasing the risk of early frost damage or of exposure to new pathogens.

Molecular mechanisms involved in dormancy induction and release remain poorly characterized in forest trees. Gene expression profiling has been performed in poplar (Rhode et al. 2007), spruce (Yakovlev et al. 2006), oak (Ueno et al. 2013), peach (Bassett et al. 2006), and apricot (Yamane et al. 2008) and has led to the identification of a set of candidate genes for dormancy regulation. In peach, Bielenberg et al. (2008), using the ever-growing mutant, have shown that MADS box genes were relevant candidate genes for dormancy regulation. Indeed, the ever-growing mutant is characterized by continuous growth of its apical meristem and does not respond to short-day signaling or low temperature. Ruttinkt et al. (2007) carried out a combination of transcriptome and metabolome profiling in poplar, to obtain the first molecular time table of apical bud formation and dormancy induction in a forest tree species. More recently, Ueno et al. (2013) provided the first insight into the gene networks involved in endo- and ecodormancy in European white oaks. They reported that genes overexpressed during endodormancy were related to dehydration, high light intensity, and abscisic acid, whereas those most strongly overexpressed during ecodormancy were related to metal ion binding, cellular transition, and fatty acid binding.

2.2.3 Root development

In woody perennial species, the optimal adaptation of root architecture to the soil is crucial to ensure solid anchorage of the plant in the soil and the efficient acquisition of water and nutrients. The mature root system of trees has a typical root architecture, including primary and lateral roots, resulting from the integration of multiple environmental signals such as symbiosis associations and water availability. The need to integrate signals from multiple pathways, therefore, complicates the dissection of the transduction pathways involved in root development. Only a few transcript profiling analyses have been conducted in woody plants, to investigate the events regulated during root development, and there is still no transcriptional roadmap. The development of genomic and transcriptomic tools, such as EST sequencing, microarrays, and next-generation sequencing technologies, and their application to different plant–fungal ectomycorrhizal associations have resulted in highly valuable information being obtained, including the identification of key genes involved in root architecture. For example, the research performed on the poplar/Laccaria bicolor association (Felten et al. 2009) and, more recently, on the oak/Pisolithus tinctorius association (Sebastiana et al. 2014) has improved our understanding of auxin-dependent pathways in roots.

Transcriptome analysis requires the detailed and precise annotation of sequenced plant genomes and thus provides an opportunity to identify genes regulated differently in the plant studied. The genomic sequences of P. trichocarpa (Tuskan et al. 2006) and, more recently, of E. grandis (Myburg et al. 2014) have been used to identify the members of gene families and to compare entire gene families between species. For example, some studies have reported a lack of expansion in poplar or in eucalyptus of certain gene families involved in hormone homeostasis and signal transduction. This is the case for families encoding cytokinin-related enzymes (Ramírez-Carvajal et al. 2008) and auxin response factor (ARF) gene family (Yu et al. 2014). Conversely, genome-wide analysis of the Populus PIN (auxin efflux transporters) family highlighted a more diversified expansion of this family in Populus than in Arabidopsis, indicating a potential role of these transporters in tree growth and development and, more importantly, in the development of roots and leaves (Liu et al. 2014).

Transcriptomic analyses performed at different stages of adventitious root formation in woody plants have revealed significant transcriptome remodeling during the formation of adventitious roots in Pinus taeda (Brinker et al. 2004), Populus sp. (Ramírez-Carvajal et al. 2009; Rigal et al. 2012), and Eucalyptus grandis (Abu-Abied et al. 2012) and allowed the identification of key genes involved in these developmental events. For example, recent studies on several woody species have generated interesting data, strongly implicating transcription factors in the successive steps of adventitious root establishment (reviewed by Legué et al. 2014) including GRAS (Sanchez et al. 2007) and AP2/ERF families (Ramírez-Carvajal et al. 2009; Rigal et al. 2012).

We also need to determine the links between changes in gene expression and alterations of biochemical and physiological functions and, ultimately, root development and the way in which the expression patterns of different genes are interconnected. The determination of gene function will require the construction and production of transgenic tree lines. This will involve a considerable technical effort on the part of laboratories. Several ambitious projects managed by several teams have been successfully carried out, providing the scientific community with a number of lines in which gene expression has been enhanced or decreased by RNA interference and gene tagging (Busov et al. 2005, 2010). Studies aiming to identify genes on the basis of their expression patterns, using enhancer-trap and gene-trap insertion lines, are also very useful, but fewer studies of this type have been carried out.

The transcriptional findings and integrative databases already available constitute a fundamental resource for future studies of molecular events and for the identification of key proteins involved in developmental processes. In parallel, transgenic lines are crucial for functional genomics studies and should be made more readily available to the research community.

2.2.4 Conifer embryo development

Conifers are the primary source for wood production worldwide (Canales et al. 2014). By unraveling the complexity of the regulated gene network involved in conifer embryo development, it should be possible to develop genomic and epigenomic tools for the early selection of improved varieties. This is a critical issue to face rapid socioeconomic and environmental changes. Plant growth can be determined early during embryogenesis (Yakovlev et al. 2014). This knowledge is also required to develop efficient clonal propagation methods of selected trees, such as somatic embryogenesis (Klimaszewska et al. 2011). This process has great potential for the deployment of new varieties in plantation forestry (Lelu-Walter et al. 2013; Klimaszewska et al. 2015).

It is difficult to sample manageable quantities of embryogenic masses during early zygotic embryogenesis (de Vega-Bartol et al. 2013; Elhiti et al. 2013). As somatic embryos (SEs) closely mimic zygotic embryos (ZEs) during maturation from early to late embryogenesis, they are considered as a model in vitro system to study the molecular biology of embryo development in conifers (Vestman et al. 2011; Yakovlev et al. 2014). Classical genetic approaches for the identification of embryogenesis-related genes are impracticable in conifers, due to their long generation time and large genome size. Since the discovery of somatic embryogenesis 30 years ago (Hakman et al. 1985; Klimaszewska et al. 2015), our knowledge of regulated genes in this system has thus remained highly fragmented and mostly based on expression studies of a few candidate genes (ESM_2A.doc). Current somatic embryogenesis protocols also essentially resulted from tedious “trial and error” strategies with low inputs from molecular studies. Microarray and RNA sequencing methods have recently provided critical advances for the genome-wide profiling of gene expression. Transcriptomics is developing rapidly in conifers, with the recent advent of large genomic resources (Lorenz et al. 2012; Raherison et al. 2012; Canales et al. 2014), including draft genomes for Picea (Birol et al. 2013; Nystedt et al. 2013) and Pinus (Neale et al. 2014).

Transcriptomic profiling is generating a growing body of information about coordinated gene expression during conifer embryo development (ESM_2B.doc, reviewed in Trontin et al. 2015). The number of transcribed genes appears to be 30 to 40 % larger than in any other tissue (Cairney and Pullman 2007; Yakovlev et al. 2014) with high relevance to the gene network in A. thaliana (300–450 genes, Cairney and Pullman 2007; Zhang et al. 2012a). Transcript profiles during zygotic embryogenesis are highly correlated between Pinus pinaster and A. thaliana, with only 3 % of the transcripts estimated to be gymnosperm-specific (de Vega-Bartol et al. 2013). Differences between angiosperms and gymnosperms are thought to arise principally from spatiotemporal variations in gene expression resulting partly from epigenetic modifications, which may act as an adaptive mechanism in such long-lived species (Cairney and Pullman 2007; Vestman et al. 2011; de Vega-Bartol et al. 2013). Yakovlev et al. (2014) specifically reported temperature-dependent differential transcriptomes in Picea abies embryogenic masses potentially associated with the formation of an epigenetic memory, with a delayed, persistent impact on seedling growth. Zhang et al. (2012b) also demonstrated the widespread occurrence of microRNAs in Larix kaempferi embryogenic masses, with predicted target genes involved in SE development. Transcriptome profiling confirmed that important processes are conserved in higher plants, including the apical–basal embryo patterning driven by polar auxin transport and the activation of the auxin-mediated response machinery during radial embryo patterning (Vestman et al. 2011; de Vega-Bartol et al. 2013). Transcriptomics has also highlighted the complexity of the processes and genes involved in the spatiotemporal development of conifer embryos from embryogenic induction (Elhiti et al. 2013; Rutledge et al. 2013) to the switch from embryonic phase to vegetative growth (Stasolla et al. 2003, 2004; Vestman et al. 2011; Morel et al. 2014a). An impressive picture of coordinated functions and genes has been obtained for SE maturation in P. abies (Stasolla et al. 2003, 2004; Vestman et al. 2011) and ZE formation in P. pinaster (de Vega-Bartol et al. 2013). It should be possible to model embryo development, by interpreting and integrating the large transcript, protein, and metabolite datasets (Kell et al. 2005; Vanderschuren et al. 2013; Wolfender et al. 2013). However, proteomic and metabolomic studies are currently scarce for conifers (ESM_2B.doc, reviewed in Trontin et al. 2015). There are still many limitations, at both experimental (Abril et al. 2011) and interpretational levels (Lippert et al. 2005; Saghatelian and Cravatt 2005; Teyssier et al. 2011, 2014). A recent study of embryo development in P. pinaster demonstrated the applicability of integrated approaches for the production of robust data (Morel et al. 2014a).

“Omics” approaches could offer practical offshoots, such as diagnostic tools for checking embryogenic potential or embryo quality. Various miRNAs with stage-specific expression have been described in L. kaempferi, suggesting possible modulation of embryogenic potential (Zhang et al. 2012b; Li et al. 2014a). The metabolic signature has been shown to accurately predict embryogenic potential in P. taeda (Robinson et al. 2009). Transcriptomics in P. taeda (Pullman et al. 2003) and proteomics in P. pinaster (Morel et al. 2014b) have provided strong evidence of differences between SEs and fully mature ZEs. Transcriptomics has also proved to be of practical value in Picea glauca to check SE quality in different maturation conditions (Stasolla et al. 2003). The transcriptomic and proteomic profiling of early maturing SEs in P. pinaster has also yielded robust diagnostic tools for detecting disturbances in pathways critical for normal embryo development (ESM_2C.doc, Morel et al. 2014a).

2.3 Molecular mechanisms involved in biotic and abiotic stress responses

We illustrate in this section how genomic technologies have not only improved our understanding of the structure and evolution of forest tree genomes but have also provided a suitable platform for obtaining knowledge about the molecular mechanisms involved in responses to biotic and abiotic cues. Considering that trees are not standing by themselves, we also considered in this section emerging researches aiming at drawing a holistic picture of the interactions between forest trees and their microbiome.

2.3.1 Abiotic stresses

Abiotic stresses decrease the growth and productivity of crops and forests. The physiological mechanisms of acclimation to environmental stresses have been extensively studied, but the analysis of their molecular bases started more recently (a decade ago in forest trees). Understanding these molecular mechanisms is of particular relevance in the frame of climate change, to achieve more rapid genetic gains in abiotic stress resistance by molecular breeding. As pointed out by Dubos et al. (2003), the molecular mechanisms involved in stress responses in trees, which have a long life cycle and specific tissues, should be considered separately and in addition to those of crops and model plants. Over the last 10 years, the number of studies analyzing the molecular basis of the response/acclimation/adaptation of trees to abiotic stresses has increased steadily. The release of the P. trichocarpa genome (Tuskan et al. 2006) led to poplar becoming the model tree species for functional genomics. About three quarters of “omics” publications concerning the response of trees to abiotic stresses concern a wide range of species from the genus Populus (ESM_3.xls). The remaining studies concern eucalyptus, pine, and, to a lesser extent, oak, beech, and Douglas fir. The context of global change may account for water deficit being the stress most frequently studied in functional genomics. Salinity, high atmospheric CO₂ concentration, hypoxia, heat, cold, ozone, nitrogen deprivation, and metal toxicity have also been studied, to a lesser extent.

Comparative approaches have been widely used through “omics” or quantitative trait locus (QTL) approaches. The genotype specificity of transcriptomic, proteomic, or metabolomic responses has been highlighted by many studies. Comparing six genotypes of Populus balsamifera, Hamanishi et al. (2010) showed that the growth response to drought was correlated with genetic responsiveness. In P. deltoides × nigra, the phenotypic response to moderate drought was found to be very similar for a stress-tolerant and a stress-sensitive genotype, and the differences in transcriptional responses were therefore attributed to intrinsic divergences in genome functioning (Cohen et al. 2010). In a similar study on Eucalyptus, the drought-tolerant hybrid displayed changes in the expression of a broader set of genes in response to water deficit, and the stress signaling cascade differed between the two genotypes studied (Villar et al. 2011). Similar genotype × environment interactions have been reported for proteomic studies (Bonhomme et al. 2009; Xiao et al. 2009; Bedon et al. 2012). The results obtained in a field experiment suggested that the better maintenance of productivity during the dry season by a drought-tolerant Eucalyptus genotype involved cell wall modification, ROS detoxification, and osmoregulation (Bedon et al. 2012). A comparison of the metabolomes of two Eucalyptus species showed that many low-abundance compounds may help plants to cope with water stress through non-osmotic functions (Warren et al. 2012). Comparisons of the transcriptomes and metabolomes of mature unstressed leaves from P. euphratica (salt-tolerant) and Populus canescens (salt-sensitive) suggested that the evolutionary adaptation of P. euphratica to saline environments involved the permanent activation of control mechanisms for osmotic adjustment, ion compartmentalization, and the detoxification of reactive oxygen species (Janz et al. 2010). Other molecular players and networks involved in acclimation to abiotic stresses are detailed in a recent review by Harfouche et al. (2014).

Few studies have investigated the effects of releasing an environmental stress. In P. euphratica, full re-irrigation after water deficit was found to lead to the recovery of most phenotypic traits and the reversal of transcriptional changes (Bogeat-Triboulot et al. 2007). In a similar experiment on oak, recovery following re-irrigation occurred in two steps, in which the observed transcriptional remodeling was consistent with physiology and growth (Spiess et al. 2012).

Other studies have provided interesting information about less studied abiotic stresses. Large differences were found between poplar and Arabidopsis, which is flood-sensitive, in terms of metabolite and transcript patterns in response to hypoxia, accounting for the ability of poplar to maintain its carbon and energy metabolism and, thus, its flood tolerance (Kreuzwieser et al. 2009). A QTL analysis associated with a study of transcriptional responses to ozone showed the involvement of key genes relating to ethylene production and response (Street et al. 2011). As for model plants, most molecular studies of the response of trees to abiotic stresses have been conducted on leaf material (ESM_3.xls). Those comparing transcriptomic or proteomic changes in different organs have highlighted the tissue specificity of the response (Bogeat-Triboulot et al. 2007; Cohen et al. 2013; Bedon et al. 2012). The proteins and genes identified are potential markers and targets for molecular breeding, but the diverse requirements for protecting and maintaining the function of different plant organs may render the engineering of stress tolerance in plants more difficult (Polle et al. 2006).

More attention should be paid to the molecular aspects of wood response, including xylem hydraulic adaptation to salt stress (Janz et al. 2012), drought and embolism (Berta et al. 2010; Secchi et al. 2011), and cadmium accumulation in bark and phytoremediation (He et al. 2013). Greenhouse and laboratory experiments are useful as they allow the control of environmental conditions, but field studies remain exception, despite their high value, and this approach should be developed further (Villar et al. 2011; Pandey et al. 2013). In addition to transcriptomic, proteomic, and metabolomic changes, epigenetic responses and non-coding microRNAs contribute to acclimation or adaptation to abiotic stresses. There is currently a lack of studies in this area (see Section 2.5), and further research is required (Harfouche et al. 2014). More integrated research will be necessary, if we are to unravel the complex molecular mechanisms and pathways underlying responses to abiotic stresses (Castell and Ernst 2012; Harfouche et al. 2014). The release of additional completed tree genome sequences (Myburg et al. 2014; Zimin et al. 2014) and next-generation sequencing should expand the range of research and help us to decipher the molecular mechanisms underlying the response of forest trees to abiotic stresses.

2.3.2 Biotic stresses

Several disturbances to natural forest ecosystems and forest plantations due to outbreaks of insect or pathogens have had major impacts, at the regional scale, on timber production or tree population sustainability (Sturrock et al. 2011). With current predictions of climate change, tree populations may become even more susceptible to outbreaks of existing and new pests and diseases. Structural and functional genomic approaches in forest trees have revealed a large diversity in genetic control of induced responses in the host and the complexity of molecular communications between the two partners (Duplessis et al. 2009) which both contrast with the Flor’s gene-for-gene model.

Like many other plants, forest trees have evolved two strategies for recognizing microorganisms including pathogens, symbionts, and endophytes (Guttman et al. 2014). Hereafter, the focus will be laid on local and systemic responses induced by a pathogen attack. Conserved microbial elicitors (PAMPs) and more specific pathogen effectors are recognized by host receptor proteins (PRR) present in the plasma membrane of tree cells and by-products of disease resistance genes (R genes) present in the cytosol, respectively (Jones and Dangl 2006). Several published studies on forest trees have focused on the effector-triggered immunity (ETI) initiated by these R genes, which leads to the hypersensitive response and a disease-free phenotype at whole-plant level (Rinaldi et al. 2007; Liu et al. 2013). Transcriptome analyses for the poplar–Melampsora interaction have shown that the defense reaction is triggered by specific signaling systems and includes the accumulation of transcripts encoding pathogenesis-related proteins (PRs), gluthathione S-transferases (GSTs), and a rust-induced secreted protein (RISP) specific to Populus (Duplessis et al. 2009).

Several such qualitative resistances with oligogenic control and a major impact on damage levels were genetically mapped for different forest pathosystems soon after the development of the first genetic maps: Pinus sp.–Cronartium sp. (Wilcox et al. 1996), Eucalyptus–Puccinia (Junghans et al. 2004), and Poplar sp.–Melampsora sp. (Newcombe et al. 1996). R genes are frequently overcome by pathogen populations (Kinloch et al. 2004; Dowkiw et al. 2010). Forest tree pathologists and breeders have, therefore, focused on the genetic and molecular basis of quantitative resistance, which is thought to be under more complex genetic control and to be more durable. The results of many fine mapping, candidate region sequencing, genome annotation, and transcriptional studies have blurred the distinction between qualitative and quantitative resistances: (i) major QTLs for quantitative resistance can account for more than 40 % of the observed phenotypic variation in field or laboratory experiments (Jorge et al. 2005; Freeman et al. 2008); (ii) in artificial inoculation tests controlling for pathogen diversity, some major QTLs display significant strain specificity, resembling that of R genes (Dowkiw et al. 2010); (iii) qualitative resistance factors from Populus deltoides and quantitative resistance factors against leaf rust inherited from P. trichocarpa and Salix viminalis have been fine-mapped to genomic regions rich in R genes (Bresson et al. 2011; Samils et al. 2011); and (iv) the difference in the timing of the activation of similar defense responses between susceptibility and partial or full resistance in the poplar–Melampsora interaction is consistent with the signal conversion model described for Arabidopsis (Nimchuk et al. 2003; Duplessis et al. 2009).

Mapping studies have revealed a major contribution of both additive and non-additive (epistasis) genetic variation to disease resistance, supporting the hypothesis of complex interaction and possible successful clonal selection in which all genetic effects can be readily captured (Jorge et al. 2005; Alves et al. 2012). Thanks to association genetics, La Mantia et al. (2013) identified five variants in orthologs of Arabidopsis genes with known functions in plant defense each accounting for smaller proportions of phenotypic variation for leaf rust severity in P. trichocarpa.,

For tree–insect interactions, the availability of reference genomes for Populus and Eucalyptus led to attention being focused on comparative analyses of the PR protein-encoding genes induced by different pathogens and insects in forest trees (reviewed by Veluthakkal and Dasgupta 2010). Analyses of the diversity of the nucleotide sequences encoding protease inhibitors (PIs) in P. balsamifera and Populus tremula revealed few signs of selection but differences in adaptive histories both within a single species and between closely related species (Neiman et al. 2009; Bernhardsson and Ingvarsson 2012).

In association with other omics studies, the completion of several new tree genomes and of new pathogen and insect genomes should contribute in the near future to the discovery of processes underlying severe disturbances such as mountain pine beetle attack involving different pine host species, a bark beetle species, and a tree-killing fungus (http://www.thetriaproject.ca/).

2.3.3 Trees are holobionts

Almost all plant tissues harbor microorganisms (Turner et al. 2013). Trees do not escape the rule (Hacquard and Schadt 2015) and can be considered as “superorganisms” or holobionts (Margulis 1991). Over the past decade, high-throughput sequencing technologies have unraveled the huge diversity of microbial communities associated to trees and forest ecosystems. Buée et al. (2009), Öpik et al. (2009), and Jumpponen and Jones (2009) pioneered in this field by using 454 pyrosequencing of barcode regions to study fungal diversity associated to forest soils, roots, and leaves, respectively. They described these communities in unprecedented detail because high-throughput sequencing technologies enable the detection of non-cultivable microorganisms (Hibbett et al. 2009). They opened the way to numerous studies highlighting variations in tree-associated microbial communities at various spatial scales. For instance, a fine-scale study showed that bacterial communities of oak rhizosphere differ from those of the surrounding soil (Uroz et al. 2010). Less contrasted results were obtained for fungal communities of beech rhizosphere (Coince et al. 2013). Fine-scale studies were also conducted for above-ground communities. For instance, Cordier et al. (2012a) showed within-canopy variations in foliar fungal communities of beech. Leff et al. (2015) obtained similar results in the case of bacterial communities associated to leaf and bark of Gingko biloba trees. By opening the microbial world to ecologists, the next-gen revolution also gave rise to larger scale studies, aimed at assessing the influence of climate and other global change components on tree-associated microbial communities. The analysis of beech-associated fungal communities along elevation gradients suggested that the air temperature is a major structuring factor of foliar communities (Cordier et al. 2012b) but not of their below-ground counterparts (Coince et al. 2014). A comparison between urban and non-urban stands revealed significant effects of anthropogenic activities on foliar fungal communities of Quercus macrocarpa (Jumpponen and Jones 2010). High-throughput sequencing technologies have also fostered the study of the relationship between the genetic variability of trees and the variability of their microbial communities. There is now a large body of evidence showing that microbial communities are influenced by the genetic variability of host trees, both at intra-specific (Redford et al. 2010; Cordier et al. 2012a; Bálint et al. 2013) and inter-specific (Redford et al. 2010; Kembel et al. 2014; Kembel and Mueller 2014) levels. A current challenge is to assess relative effects of tree genotype and environment by using common-garden experiments (Bálint et al. 2015) and to decipher the genetic architecture of microbial communities. Significant advances have been made recently on this latter topic on the model plant A. thaliana. The use of mutant lines and genome-wide association mapping revealed that plant loci responsible for defense, cell wall integrity, and cuticular wax composition influence the foliar microbiota (Riesberg et al. 2013; Horton et al. 2014). Another challenge is to go beyond the identification of the genes and environmental factors structuring microbial communities, by elucidating the effects of these hyperdiverse communities on tree growth, health, and reproduction. We are indeed far from a proper understanding of the outcomes of the interactions between microorganisms and their host (Borer et al. 2013), even for model tree species such as poplars (Hacquard and Schadt 2015). It is thus time to move from meta-barcoding approaches to meta-“omics” approaches enabling the identification of gene transcripts, proteins, and metabolites expressed by tree-associated microbial communities. Such functional approaches are still rare for microbial communities associated to plants and soils (Knief et al. 2012; Damon et al. 2012), and deserve further investigation.

2.4 Molecular interactions

The completion and annotation of plant genome sequences have revealed that the functions of most genes are still unknown. As most proteins function in macromolecular complexes, the identification of protein partners or other molecular interactors (nucleic acids, carbohydrates, lipids, etc.) is particularly relevant when assigning a function to a given protein. Post-genomic high-throughput studies have been carried out, to accelerate functional studies through the identification of plant interactomes, i.e., the whole set of molecular interactions in a given cell or tissue, including both physical interactions between molecules and indirect interactions between genes (genetic interactions). For example, the first experimental Arabidopsis interactome (AI-1) led to the identification of 2,700 proteins and 6,200 interactions, generating hypotheses about the molecular functions of several thousand unknown proteins (Braun et al. 2011). The molecular interactions identified can then be displayed as networks. Depending on the underlying biological question and the data available, these networks may illustrate transcriptional and post-transcriptional regulations (transcription factor/DNA or miRNA/target gene interactions), signal transduction pathways, or fluxes in metabolic pathways. Most traits of interest are quantitative and, therefore, particularly suitable for modeling by network approaches.

Both high-throughput experimental techniques and computational predictions are required to decipher the interactome. Experimental data remain very scarce in the plant kingdom, with the exception of the model plant Arabidopsis. Indeed, several large-scale protein interaction studies have been carried out with different experimental approaches, mostly based on yeast two-hybrid (Y2H) screening with a cloned ORFeome and tandem affinity purification/mass spectrometry (for a review, see Braun et al. 2013). The Arabidopsis interactome was further expanded by prediction methods based on statistical learning methodology and/or the transfer of interaction annotation based on homology with other species (human, yeast, nematode, fruitfly) (De Bodt et al. 2009; Geisler-Lee et al. 2007; Lin et al. 2011). In tree species, no large-scale systematic study has yet been undertaken. However, for P. trichocarpa, a dedicated biomass ORFeome was cloned from 374 selected ORFs found to be more strongly expressed in xylem tissues than in phloem and was used in a binary Y2H or a Y2H cDNA library screening: interacting proteins were found for 74 baits (http://xylome.vbi.vt.edu/index.html), but further validation, with alternative methods like co-immunoprecipitation or bimolecular fluorescence complementation, is required as Y2H is prone to artifacts. For monolignol biosynthesis, a predictive kinetic metabolite-flux model has been recently established for the 21 enzymes and 24 metabolites of the pathway in P. trichocarpa differentiating secondary xylem, based on both in vivo mass spectrometry quantification of all the isoforms in the pathway and kinetic parameters measured in vitro for functional recombinant proteins (Wang et al. 2014b). Predictions derived from the model were validated in transgenic poplars with altered monolignol biosynthesis. For transcription factors, it is now possible to identify DNA targets at the genome level, by combining the immunoprecipitation of chromatin with NGS sequencing (ChIP-SEQ), thus facilitating the construction of regulation networks at the genome scale. The ChIP-SEQ technique has been shown to be applicable on poplar cambial or xylem tissues and will certainly provide informative data about gene regulation networks in years to come (Li et al. 2014b). ChIP-PCR was successfully used on poplar secondary xylem, to validate potential DNA targets of secondary wall-associated NAC domain 1 (SND1), a transcription factor controlling wood formation (Lin et al. 2013). Predictive approaches have proved highly successful for tree species. Rodgers-Melnick et al. (2013) presented the first predicted interactome for P. trichocarpa, generated by the computational prediction of protein–protein interactions from primary sequence data only (conserved protein domains and predicted subcellular localization as input features).

The deciphering of molecular interactions is still in its infancy for tree species but will certainly increase in importance in the near future. We must first establish the critical datasets required, together with experimental evidence, to accelerate the inference of gene function and the construction of gene networks. Interactome data are not particularly straightforward to obtain, particularly for non-model species, but they can also be coupled to transcriptomic data, for the generation of informative biological networks. Biological network approaches have highlighted the existence of highly connected proteins, called hubs: these proteins are of particular interest in functional studies as they must play a key role in plant biology. In addition, biological network modeling could be used to predict the phenotypic changes resulting from changes to gene expression, thereby accelerating hypothesis-driven research for the development of new breeding applications, as already described in the field of medical research (Hood et al. 2004).

2.5 Epigenomics in trees: a new dimension to phenotype prediction in a changing environment

In addition to the genetic component, epigenetic variation is now proposed to contribute to phenotypic plasticity, adaptive capacity, and evolutionary trajectories in both natural and cultivated plant populations (Bossdorf et al. 2008; Nicotra et al. 2010). Indeed, genome sequencing and genotyping were expected to identify the genetic components of common traits, but these approaches were not entirely successful, suggesting that there must be other sources of the missing heritability (Maher 2008). Epigenetics is the study of meiotically or mitotically heritable changes in gene function that do not result from changes in DNA sequence (Allis et al. 2007). At the molecular level, epigenetic phenomena are mediated by reversible marks, such as DNA methylation and histone modifications, including methylation, acetylation, phosphorylation, and ubiquitination, and by small RNAs that can alter regulatory states of genes or genomic regions. The study of epigenetic patterns at the genome-wide level is referred to as “epigenomics.” As for spontaneous mutations in DNA, errors in the maintenance of methylation state result in the accumulation of single methylation polymorphisms (SMPs) over an evolutionary timescale. If the rates of SMP formation are orders of magnitude greater than those of spontaneous mutations, changes in regional methylation levels occur at similar frequencies. The regions concerned, which are known as differentially methylated regions (DMRs), correspond to genomic regions with different methylation profiles in different samples (tissues, cells, individuals, or others). DMRs are regarded as possible functional regions involved in the regulation of gene transcription and could act as QTLepi in natural populations, thereby constituting a measureable component of the so-called missing heritability (Cortijo et al. 2014). Epigenomic data have recently been reported in forest trees (Bräutigam et al. 2013).

DNA methylation in trees was first evaluated by determining global DNA methylation percentages by HPLC or HPCE, after the hydrolysis of DNA to generate nucleosides or nucleotides (Gentil and Maury 2007). Variations of global DNA methylation have been reported in several tree species, in different populations with different origins, in various organs, at different developmental stages, in different culture conditions, and in response to several environmental constraints, such as water availability and temperature (Hasbún et al. 2008; Monteuuis et al. 2009; Gourcilleau et al. 2010; Mankessi et al. 2011; Teyssier et al. 2014). However, it was not possible to identify the genomic context of these variations in these analyses. In recent years, epigenomics has emerged in parallel to the development of next-generation sequencing (NGS) techniques, genomic resources, and associated bioinformatics and biostatistics packages, together with the use of specific methods to identify DNA methylation marks. Three main methods for detecting DNA methylation at the genome-wide level have been applied to trees (Mensaert et al. 2014; Rodriguez et al. 2012; ESM_4.doc, ESM_5.doc): (1) methylation-sensitive amplification polymorphism (MSAP), (2) methylated DNA immunoprecipitation (MeDIP), and (3) whole-genome bisulfite sequencing (WGBS).

MSAP revealed DNA methylation polymorphism in different families of trees, between individuals/populations, in response to environmental changes, and during in vitro culture (ESM_4.doc). MeDIP and WGBS have been used to identify thousands of DMRs at the genome-wide level but only in Populus, the first tree to be sequenced (Tuskan et al. 2006). These epigenomic studies have significantly advanced our knowledge of structural and functional aspects of genomics. Indeed, determination of the poplar methylome revealed features particular to this tree, such as a higher CHG methylation level than reported for other plants (Feng et al. 2010) and low levels of methylation at DNA recombination hotspots (Slavov et al. 2012). Furthermore, gene-body DNA methylation is extensive in the open chromatin state, linked to structural gene characteristics (gene size and copy number) and correlated with tissue-specific gene expression (Vining et al. 2012; Lafon-Placette et al. 2013; Vining et al. 2013). DNA methylation is also involved in regulating stress response genes (Liang et al. 2014; Maury and Lafon-Placette, unpublished data). In addition, the methylation patterns of the parents are partially and dynamically passed onto their hybrid offspring (Gao et al. 2014). Finally, the first hypomethylated poplar trees (ESM_4.doc; Zhu et al. 2013a) were obtained by using a RNAi strategy to silence the DDM1 gene, the product of which mediates the methylation of transposable elements and genes. These mutants represent an interesting model for investigating the role of DNA methylation under conditions of environmental variation. Insights into epigenomics will improve our understanding of adaptive tree responses to fluctuations in the environment, particularly in a context of global climate change.

2.6 Ecological genomics: genomic answers to ecological questions

Ever since Darwin (1859) and Ford (1964), the question of the heritable basis of adaptation to environmental conditions has been a fundamental issue in population genetics and ecology. Environmental parameters vary continually over space and time, so “adaptation” is intrinsically “local.” Forest trees have large, often continuous populations and long life cycles. The question of how they adapt to changing environmental conditions is thus of considerable fundamental importance in addition to having implications for forest management. Genomics constitutes a powerful approach for obtaining answers to long-standing ecological questions, by shedding light on the role of selection in shaping patterns of genetic diversity and identifying environmental drivers of selection, in particular.

The power of genomics in forestry has already been described by González-Martínez et al. (2006). The relative durability of tree stands makes it possible to associate ecological variables with genotype frequency patterns in a reliable manner (Sork et al. 2013). Genomic methods can then be used to screen large numbers of loci for association with environmental parameters. This approach has led to the identification of polymorphisms associated with environmental gradients at the regional scale in P. taeda (Eckert et al. 2010) and at the local scale (Jump et al. 2006; Csilléry et al. 2014) in Fagus sylvatica. Additional studies are underway in several species (Pinus halepensis, Eperua falcata, P. pinaster, Symphonia globulifera, etc.); preliminary results for candidate loci and traits have already demonstrated signatures of microgeographic adaptation in E. falcata (Audigeos et al. 2013; Brousseau et al. 2013) and P. halepensis (Hernandez-Serrano et al. 2013). Generally speaking, two kinds of approach have been applied to data analysis, depending on the nature of the information about the structure of environmental variation. In cases in which variation is considered to be continuous, the methods applied are mostly based on regression between environmental variables and allele frequencies (e.g., SAM; Joost et al. 2007). By contrast, in cases in which environmental variation is considered to be discrete (or is discretized during data collection, e.g., by sampling at different positions along a continuous gradient), the preferred methods are those based on the detection of divergence outliers (e.g., Beaumont and Balding 2004). However, it should be noted that the consideration of environmental variation as continuous or discrete often depends on the experimental design and constraints. With the exception of particular cases, in which one or more environmental variables change abruptly over space (e.g., over a cliff), or in which the patches available for tree growth are themselves discontinuous (e.g., on islands), environmental variation is generally continuous. This raises the question of a further merger, between ecological genetics (or genomics) and landscape genetics (or genomics). Depending on the extent to which ecological contrasts give rise to continuous patterns, there are good reasons for treating dispersal, migration, and adaptation as a unified process (Sork et al. 2013; Schoville et al. 2012).

Most early population genomic studies on forest trees focused on a limited number of polymorphisms of a few candidate genes. However, more realistic mechanisms of evolutionary change, such as polygenic and epistatic selection, must be considered in selection tests. To this end, population genomics will benefit from the ongoing development of pan-genomic approaches.

2.7 Integrating genomics into phylogeography and phylogeny

The phylogenetic relationships between species, the delimitation of closely related species, and the genetic structure of populations within species provide key information for decision-making in the conservation and sustainable use of forest tree germplasm. Early genetic studies on isozymes or plastid DNA revealed major footprints of past range dynamics in temperate trees (e.g., Petit et al. 2003), which found applications in genetic provenance discrimination of forest reproductive material. Plastid DNA markers were also applied for species delimitation in trees (Kress and Erickson 2008); however, their power remained limited because of weak reproductive barriers and frequent interspecific gene flow in trees (Petit and Hampe 2006). In the last decade, the increased use of multiple nuclear genetic markers and high-throughput genomics has made it possible to provide more precise information on past population history, more efficient delimitation of species, and inference of phylogenetic relationships between them and to characterize adaptive evolution at the molecular level in forest trees. This revolution in genetic markers has been accompanied by major transitions in data analysis, notably a shift from descriptive to hypothesis testing approaches, yielding valuable information for a more informed management of forest genetic resources.

In the last decade, phylogeographic studies have increasingly used multiple unlinked nuclear loci in combination with population genetic models using coalescent theory (Nielsen and Beaumont 2009). This has allowed discriminating between alternative scenarios of population genetic history within species and in closely related species (Heuertz et al. 2006; Gao et al. 2012; Cornille et al. 2013) and has permitted a reliable estimation of population genetic parameters, such as divergence times between lineages (Budde et al. 2013; Couvreur et al. 2008; Morris et al. 2008; Scotti-Saintagne et al. 2013a; Scotti-Saintagne et al. 2013b). The inferred demographic history has been used as a robust baseline information to detect gene loci under adaptive evolution (Grivet et al. 2011; Källman et al. 2014). Comparative phylogeographic approaches have given insights into the congruence of demographic history across species through time, first in temperate tree species (e.g., Jaramillo-Correa et al. 2010; Petit et al. 2003) and more recently in tropical species (Dauby et al. 2014; Heuertz et al. 2014; Jones et al. 2013; van der Merwe et al. 2014). Important predictive power has been gained in phylogeography by integrating species distribution modeling based on spatially interpolated climatic data for different time periods (Carstens and Richards 2007; Cornille et al. 2013). Significant advances in phylogeography could further come from new disciplines, such as “geogenomics,” which involves the use of large-scale genetic data to constrain geological hypotheses (Baker et al. 2014).

Applications of massively parallel high-throughput sequencing (HTS) have emerged in recent years, facilitating cost-effective marker development (McPherson et al. 2013; Micheneau et al. 2011; Slavov et al. 2012). HTS technologies hold great promise for disentangling evolutionary relationships in complex groups, especially in tropical tree taxa in which botanical knowledge remains limited and cryptic species are common (Heuertz et al. 2014; Turchetto-Zolet et al. 2013). The first HTS studies reconstructed the complete organellar genomes of tree taxa, which resulted in a higher resolution in phylogeography (McPherson et al. 2013; van der Merwe et al. 2014), taxonomy, and phylogenetics, particularly for tropical taxa (in Chrysobalanacae, Malé et al. 2014; in Malpighiales, Xi et al. 2012). HTS studies of the nuclear genome are still rare in phylogenetics and phylogeography of trees (but see Carstens et al. 2013; Stölting et al. 2013). They require the cost-effective parallel sequencing of hundreds or thousands of homologous DNA regions across hundreds of individuals, using appropriate technology for the targeted phylogenetic depth. Genome reduction methods involving enzymatic digestion are suitable at the within-species level (e.g., Stölting et al. 2013), whereas sequence capture for conserved nuclear regions is recommended for phylogenomics (reviewed in McCormack et al. 2013).

In phylogenomics, new Bayesian methods make it possible to infer phylogenetic relationships from multilocus nuclear data while accounting for intraspecies polymorphism and incomplete lineage sorting (Heled and Drummond 2010). In addition, Bayesian dating methods that incorporate uncertainty in evolutionary rate variation alongside time constraints based on fossils hold promise to improve dating (Dos Reis et al. 2012).

The characterization of genetic boundaries between closely related species remains challenging when reproductive barriers are incomplete. Interspecific gene flow can have important evolutionary consequences, such as the rapid introgression of beneficial variants (Morjan and Rieseberg 2004). This facilitates adaptation from standing genetic variation, which is particularly relevant during rapid range expansion (Keller et al. 2010; Lascoux and Petit 2010). Bayesian clustering algorithms (reviewed in François and Durand 2010) have proved useful for tree species delimitation (Duminil et al. 2012; Guichoux et al. 2013). However, population sampling must account for the presence of gene flow by including evolutionarily important areas, such as secondary contact zones or hybrid zones (Eckert et al. 2008; Scotti-Saintagne et al. 2013a).

Finally, phylogeographic and ecological genetics studies will undoubtedly become more integrated in the future. Historical inference is important not only for an understanding of the evolutionary past of a particular species or ecological community but also as a prerequisite to test for selection on sequence data (Carstens et al. 2013; Källman et al. 2014) and analyses of phenotypic evolution in different populations (Keir et al. 2011; Stone et al. 2011).

2.8 Genotype–phenotype association

Unraveling the genetic architecture of traits of economic or ecological importance, which are usually quantitative, is crucial for forest tree improvement and management. This can be achieved by identifying and localizing the genomic regions controlling the variation of quantitative traits (QTLs). Large-effect QTLs could theoretically be used in marker-assisted breeding schemes, but this approach has not yet been used in forest trees (Muranty et al. 2014), mostly because the most relevant traits for forest tree breeding are highly complex and probably controlled by many small-effect QTLs. Furthermore, most of the QTLs detected have large confidence intervals and have not been validated. Consequently, the identification of causal molecular polymorphisms controlling quantitative traits remains a great challenge that must be met before such information can be effectively transferred to breeding programs (ESM_6.pdf).

Since the first QTL mapping experiment in forest trees in the 1990s (Bradshaw and Stettler 1995), the techniques used to identify QTLs have changed radically. The advent of genomics and the democratization of sequencing due to NGS have generated thousands of single nucleotide polymorphism (SNPs) for genetic mapping, a prerequisite for QTL identification in controlled crosses. As a result, the number of markers on genetic maps has greatly increased, from a few hundred to several thousand, thanks to the extensive use of SNP arrays in many species (e.g., Chancerel et al. 2013) and, more recently, the direct use of sequencing for genotyping (Neves et al. 2014). Nevertheless, in order to improve the resolution of the QTLs mapped, it is necessary that the increase in the number of molecular markers is accompanied by an increase in the number of recombination events within the population studied, which can be achieved by increasing its size. This approach has been successfully used for the fine mapping of a major QTL for rust resistance in poplar (Bresson et al. 2011), but it remains time-consuming and expensive.

Association mapping (also known as linkage disequilibrium or LD mapping), which involves the detection of QTLs in more complex populations, offers an attractive alternative to the fine-mapping of QTLs in forest trees (Neale and Kremer 2011). Indeed, most forest trees are outcrossing species and this, together with their almost undomesticated status, implies a rapid decay of LD, making it possible to detect polymorphisms in the close physical vicinity of the causal variants or even the functional variants themselves. As a result, such approaches have become very popular in forest trees over the last decade, and many associations have been reported. However, despite the great promise of association mapping, most of the associations reported to date have accounted for only a very small proportion of the genetic variation, and this has greatly hindered their use in breeding programs. An obvious explanation for such disappointing results is that association studies in forest trees have not yet exhaustively screened the entire genome, as they have focused only on candidate genes and/or regions. Here again, the advent of NGS opens up new possibilities of screening for almost all SNPs within the gene space or even within the entire genome, resulting in an exhaustive genome-wide scan (Evans et al. 2014). Moreover, sequencing approaches will enable a genome-wide discovery and typing of structural variation (presence/absence and copy number variants), with effects on quantitative traits that remain largely unexplored (Muchero et al. 2014).

Another way of gaining insight into the genetic determinism of complex traits is the genetic analysis of intermediate non-organismal phenotypes, such as transcriptomic, proteomic, or metabolomic data. Such systems biology approaches are appealing for highly complex traits, as they provide an intermediate step between the genotype and the phenotype, thereby facilitating the deciphering of genetic architecture for such traits. Only a few such studies have been carried out for forest trees, and these studies mostly involved QTL mapping for mRNA data obtained with microarrays (Kirst et al. 2005; Drost et al. 2010). With the advent of “omics” approaches, systems biology studies should and will undoubtedly become more widespread for forest trees in the near future. They will allow the detection of expression, protein, and metabolite QTLs and the construction of gene networks, which, together with genome-wide variation data, may make it possible to move from associations to causal links, through dedicated statistical modeling (Marjoram et al. 2014).

2.9 Conservation genomics

The science of conservation genomics is directly derived from conservation genetics, i.e., the use of genetic methods for understanding the impact of habitat modification on genetic structures and fitness and designing conservation strategies in practice, particularly in rare and endangered populations and species. Because quantitative genetic methods were complex and costly to implement, conservation genetics mostly focused on the use of molecular markers to decipher the demographic history of taxa and their phylogeography, thereby identifying groups of populations to be given priority for conservation efforts (see the concepts of ESU, evolutionary significant unit, as defined by Moritz 1994, and MU, management unit, as defined by Palsbøll et al. 2007).

Conservation schemes based on genetic methods have always recognized that neutral markers told only part of the evolutionary story (albeit an important one) and that access to parts of the genome undergoing selection was crucial. Before genomic tools became available, this was best achieved by comparing phenotypes of at least partly known ancestry in controlled environments (common gardens and reciprocal transplants). However, more frequently, environmental surrogates were used, to indicate the potential existence of natural selection. If both phenotypic or environmental divergence and neutral genetic or phylogeographic structure were found, such distinct populations were considered to be of high conservation priority (Lesica and Allendorf 1995; Allendorf et al. 2013). The European networks for the conservation of forest genetic resources are based on these strategies (Koskela et al. 2013; Lefevre et al. 2013).

With advances in genomics, and increasing access to many genes and potentially complete genome sequences for trees, it is becoming increasingly possible to compare surrogate phenotypic and ecological information with genetic information. Genomics is making it possible to revise our approach to conservation science and conservation strategies. In some cases, the change is just a question of degree, such as the much larger number of markers available now than in the past, making effective size and demographic estimates more precise. In other cases, the change is revolutionary, as, for example, for the comparison of DNA sequences responding to demography with those responding to selection (potentially in different ways, depending on the location of the populations within the ecological niche of the species) and for predicting the ability of populations to adapt to environmental changes and new patterns and thresholds of ecological disturbances (Allendorf et al. 2010). Even over short spatial scales (such as a single mountain), genomic studies have demonstrated the existence of local adaptation and significant differentiation for genes involved in key adaptive traits (phenology and resistance to drought and cold), of clear utility for the design of conservation strategies (Lalagüe et al. 2014).

There have been many calls for the use of genomic data for conservation purposes in forest trees (e.g., González-Martínez et al. 2006). However, few examples of the practical use of genomic data for strengthening conservation networks are available for forest trees as of yet, although many studies have demonstrated or confirmed that some populations or particular regions deserve protection. In Pinus sylvestris for example, genomic data from a relatively small number of adaptive genes have confirmed the high level of differentiation and unusual evolutionary history of populations from Scotland (Wachowiak et al. 2011). In P. trichocarpa, an extensive genome scan revealed previously unnoticed small- and large-scale geographic differentiation patterns in western North America (Slavov et al. 2012). As genome scan techniques are becoming cheaper and are now technically affordable for many different laboratories, they will undoubtedly soon be widely used for characterizing the evolutionary ecology and history of forest tree species, finally making it possible to include both demographic and adaptive processes in the design of conservation networks and to prioritize conservation actions. However, this era has yet to arrive.

2.10 Genomics and breeding

Most forest tree breeding programs were launched in the 1950s and have focused on species with relatively short rotation periods, either conifers (pine, spruce, larch, Douglas fir, etc.) or broadleaf trees (eucalyptus, poplar, wild cherry, sycamore, etc.). Considerable genetic gains have been achieved for the economic and adaptive traits of most of these species (reviewed by Pâques 2013), despite the inherent difficulties involved in the breeding of these large, long-lived organisms. Given their key ecological and economic roles, forest trees have also been the object of rapid developments in genetic and association mapping for QTL discovery in the last 20 years (see Section 2.8). Despite these parallel developments in breeding and QTL mapping, few breakthrough applications for forest trees using a synergistic combination of these two approaches have been reported. This situation was quantitatively assessed and discussed in a review work by Muranty et al. (2014), which focused particularly on marker-assisted selection. Indeed, forest trees are among the genetically improved species most likely to benefit from the use of gene- or marker-based information in breeding. Indeed, reducing the long generation intervals of forest tree breeding programs by early marker-based evaluation is the most evident benefit, with other benefits including the limitation of phenotyping costs, an increase in the precision of evaluation of difficult traits (wood properties, phenology, biotic and abiotic responses to stresses, see previous sections), and the explicit management of genetic diversity. The large levels of polymorphism often harbored by forest tree species and their complex genomes have undoubtedly been limiting factors preventing the accumulation of genomic resources to the point required for the revolutionization of breeding programs (see Section 2.1). However, a number of initiatives worldwide are dealing with this issue, including dedicated European (reviewed in http://www.forestry.gov.uk/fr/euframeworkprojects) and North American (American Conifer Translational Genomics Network: https://dendrome.ucdavis.edu/ctgn/, SMarTForests: http://www.smartforests.ca/) projects.

A number of marker-based applications are paving the way for genomics-assisted breeding. Some of the simplest, in terms of marker requirements, do not require modifications to existing breeding programs and concern the management of breeding populations through fingerprinting (identity and pedigree checking, genetic diversity estimation) and the optimization of genetic gain deployment (seed quality control). Other applications have low requirements for genomic resources but potentially large impacts on existing breeding programs. These applications aim to minimize the uncertainties in the assessment of relatedness in pedigree-based genetic evaluations, through the use of marker-based estimates, and to improve precision. Markers can be used to recover full parentage in open or polymix mating regimes (El Kassaby and Lstiburek 2009), thereby increasing the precision of evaluations without the need for a costly control-cross regime. This strategy is currently being evaluated in several breeding programs, for Scots pine in Sweden (Rosvall 2011) and maritime pine in France (Bouffier et al. in preparation). In a more general approach known as G-BLUP, a marker-based relationship matrix is used alone or together with one from a pedigree, in the statistical mixed model for genetic evaluation. This approach can finely capture relatedness at within-family levels, thereby increasing the precision of evaluations.

Ultimately, the achievement of sufficiently high levels of genome coverage by dense genotyping or sequencing provides not only highly precise relatedness estimates but also information about any relevant genetic variation from underlying causal mutations. This is the principle behind genomic selection (GS, Meuwissen et al. 2001), which paves the way for the early evaluation of genotyped candidates without phenotype data. Simulation studies (Grattapaglia and Resende 2011, Iwata et al. 2011; Denis and Bouvet 2013) have already demonstrated the potential of GS for forest trees in diverse breeding and genetic scenarios. These theoretical studies have been complemented by the first empirical studies in forest trees (Zapata-Valenzuela et al. 2012; Zapata-Valenzuela et al. 2013; Resende et al. 2012a; Resende et al. 2012b), which have yielded medium- to high-level accuracies, even with limited numbers of markers (less than 5,000 SNPs). However, these preliminary results must be interpreted with caution, as they relate to populations with small effective sizes and thus favorably high linkage disequilibrium levels. In large populations, such as those in most forest tree breeding populations, GS may be less accurate, given the marker density currently attainable in these species (Beaulieu et al. 2014). Nevertheless, GS is clearly an alternative for the forest tree breeding programs of tomorrow.

3 Conclusion and perspectives

The sections above clearly illustrate how, in the 10 years since the DNA sequence of the poplar genome was made publicly available, the forest tree genomics community has fully embraced the rapidly evolving tools of genomics and bioinformatics (i) to study the genetic and molecular changes underlying the complex developmental traits characteristic of the biology of woody perennials, (ii) to determine how the information encoded in the genome of individual trees responds to external cues and to identify the evolutionary forces responsible for shaping the phenotypic variation we see at both the population and species levels in natural conditions, and (iii) to identify the causal genetic polymorphisms underlying phenotypic differences. By harnessing the power of genomics, this community has shown that it is committed to applying this knowledge in management practices, conservation, and breeding programs, to help natural and planted forests adapt to the rapid pace of current and projected climate changes. It is difficult to see where events will take us in the next 10 years, given the currently exponential increase in the amount of DNA sequence data, likely decreases in the cost of sequencing, and the emergence of new technologies. This research community has matured, but its sustainable development will require the mobilization of adequate funding and human resources, improvements to international collaboration to address global challenges, and integration with other disciplines.

3.1 What is to come in the next 10 years?

Over the next decade, we will undoubtedly continue to see an accumulation of genomic resources (including reference genome and epigenome sequences) and basic understanding about genome structure and evolution, the distribution of gene, non-protein-coding transcribed fragments, and transposable elements across the genome and their interactions. The sequencing of large heterozygous tree genomes remains challenging, and work is still in the exploratory phase for most such genomes (e.g., chestnut and oak) or requires refinement (e.g., conifers) with new sequencing technologies and bioinformatic approaches (Faino and Thomma 2014). However, the recent results obtained in this area are more than encouraging and open the way for the resequencing of thousands of genotypes, a prerequisite for the description of sequence variation within and between species, the identification of causal variants underlying phenotypes of interest, and the provision of knowledge about the evolutionary history of tree populations. Beyond the apparent completeness of genome sequence information, much remains to be done concerning our understanding of gene regulation. Multidisciplinary groups are required to make use of this wealth of resources to derive fundamental insights into the biology of woody perennials through the integration of “omics” technologies into research activities. MicroRNAs constitute a specific class of noncoding molecules worthy of attention, given their fundamental biological role. Clearly, the next 10 years will see improvements in descriptions of the number and biological role of these RNAs, particularly as concerns tree-specific features. Moreover, systems biology, by providing a holistic approach, may help to uncover the molecular players, their complex networks of interaction and key hubs underpinning developmental processes, and responses to external disturbances. Advances in genomic technologies, statistical, mathematical, and computational methods are very promising and should enable us to meet this challenge. RNA interference-based screening and gene tagging approaches are still underdeveloped for forest trees but will probably be instrumental in achieving these ends. However, the logistic complexities associated with high-throughput screening in such large organisms may hamper the development of this area of research. Another field that will undoubtedly change our view on how trees develop, grow, and adapt throughout their extended life span is that of epigenetics. Reversible epigenetic marks contribute to phenotypic plasticity and, therefore, constitute an essential factor in the adaptive capacity of these long-lived organisms. This is clearly a hot topic in the framework of rapid climate change. Moreover, as epigenetic marks may differ between genotypes and may be heritable, a fraction of phenotypic variation shaped by epigenetic mechanisms may potentially be targeted by natural selection and, therefore, contribute to the evolutionary trajectory of populations. Clearly, assessing the relative contributions of epigenetic modifications and changes in allele frequency (the “classical” mechanism of adaptation operating at the population level) will be a major challenge in the coming years. Finally, two other research areas should benefit from the discovery of gene regulatory networks underlying trait variation. The hunt for the so-called missing heritability (proportion of phenotypic variance unaccounted for by single nucleotide polymorphisms) is one of these areas (Maher 2008). Understanding the nature of genetic adaptation to environmental heterogeneity over space and time is the other.

3.2 Will genomics change the landscape of forest tree breeding?

Proofs-of-concept of genomic prediction in forest trees have been obtained and show that this technology should work (i.e., it should improve estimated breeding value accuracy and, therefore, result in a genetic gain per unit time), and the genomic revolution holds great promise for economic benefit in the forest industry. However, further research is required to confirm that this technology could be readily implemented. We can see at least three complementary actions that could potentially favor a rapid and efficient implementation of genomic selection in the next decade. One is the construction of consensus reference populations for each species across countries, with shared efforts for the required phenotypic evaluation. Another action is the rational implementation of genotyping, notably through the use of different marker coverages at different population levels and of imputation for equaling density coverages. Useful lessons can be learned from existing success stories (as in dairy cattle and pig breeding), but the economic viability of incorporating genome-enabled selection into forest tree breeding programs remains to be demonstrated. Thus, a third action will be the demonstration that the magnitude of the estimated breeding value accuracy improvement will be large enough to counterbalance the genotyping/sequencing costs. The next decade will certainly see an accumulation of proof-of-concept studies but, more importantly, researchers and breeders will have to work together to demonstrate the economic viability of this methodology before its implementation.