Gene expression and genetic divergence in oak species highlight adaptive genes to soil water constraints

Abstract

Drought and waterlogging impede tree growth and may even lead to tree death. Oaks, an emblematic group of tree species, have evolved a range of adaptations to cope with these constraints. The two most widely distributed European species, pedunculate (PO; Quercus robur L.) and sessile oak (SO; Quercus petraea Matt. Lieb), have overlapping ranges, but their respective distribution are highly constrained by local soil conditions. These contrasting ecological preferences between two closely related and frequently hybridizing species constitute a powerful model to explore the functional bases of the adaptive responses in oak. We exposed oak seedlings to waterlogging and drought, conditions typically encountered by the two species in their respective habitats, and studied changes in gene expression in roots using RNA-seq. We identified genes that change in expression between treatments differentially depending on species. These “species × environment”-responsive genes revealed adaptive molecular strategies involving adventitious and lateral root formation, aerenchyma formation in PO, and osmoregulation and ABA regulation in SO. With this experimental design, we also identified genes with different expression between species independently of water conditions imposed. Surprisingly, this category included genes with functions consistent with a role in intrinsic reproductive barriers. Finally, we compared our findings with those for a genome scan of species divergence and found that the expressional candidate genes included numerous highly differentiated genetic markers between the two species. By combining transcriptomic analysis, gene annotation, pathway analyses, as well as genome scan for genetic differentiation among species, we were able to highlight loci likely involved in adaptation of the two species to their respective ecological niches.

Combining RNA-seq and a genome scan of genetic divergence between species highlights potentially adaptive genes in European white oaks.

Introduction

Major goals in evolutionary biology are to understand the genetic architectures of adaptation and speciation. This means identifying the genes involved in adaptive traits and reproductive barriers, their effect sizes, and functions (Campbell et al., 2018). There is now a clear consensus among evolutionary biologists that speciation is a gradual process involving the accumulation of barriers to gene flow driven by genetic drift (Coyne and Orr, 2004), and by the adaptation of populations to local environmental conditions (Lenormand, 2012).

Identifying adaptive loci in closely related species that occur in different habitats while exchanging genes are interesting cases to study with real-world applications. Indeed multiple studies have documented extensive gene flow among closely related species (Harrison and Larson, 2014), identified islands of introgression in species genomes (see e.g. in humans; Vernot and Akey, 2014) and highlighted introgression as a source of adaptive genetic variation (Suarez-Gonzalez et al., 2018). In the context of global change, there is great interest in using gene flow among closely related species adapted to different environments to help create new adaptive genetic combinations and accelerate the adaptive process (Burgarella et al., 2019). For instance, in oaks (i.e. Quercus), a botanical group in which numerous species still hybridize, Cannon and Petit (2020) proposed that forest management should favor interspecific gene flow to increase genetic diversity and produce new genetic combinations to face environmental changes and restore declining populations.

A common strategy to study genetic differentiation among species is to use genome-wide scans contrasting allele frequencies between species. The results, however, can be difficult to interpret, especially because most species have experienced successive periods of isolation and contact in the past (Bierne et al., 2011). In these cases, genetic differentiation of closely related species that are still exchanging genes, is expected in (1) loci involved in intrinsic reproductive barriers (i.e. preventing reproduction or survival of hybrid offspring independently of the environment), (2) loci directly involved in the adaptation of the species to their respective environments and where allelic combinations may reduce fitness in hybrids or even generate intrinsic barriers to reproduction, hence limiting gene flow, and (3) loci where differentiation between species is due to drift or background selection in regions of low recombination (reviewed in Wolf et al., 2010; Ravinet et al., 2017) (Figure 1, points 1–5). While genome scans of genetic differentiation can to some extent evidence the effect of drift, the results are generally too confounded to differentiate between the other possible sources of genetic differentiation. As a complementary approach to scans of genomic variation aiming to quantify the level of sequence divergence between sister species (reviewed by Storz, 2005), investigating contrasting patterns of gene expression among sister species adapted to contrasting environment may help identify loci involved in adaptive differentiation between species in regions of high genetic differentiation (Figure 1).

Figure 1.

Conceptual framework. In this diagram, we consider two closely related species exchanging gene flow (either continuously or during secondary contact, which is the case for sessile and PO as shown in Leroy et al. (2020)). In such situation, genetic differentiation (1) is expected to arise at loci along the genome because of different processes including drift, demographic history, or background selection in regions of low recombination (2). In addition, loci directly involved in intrinsic barriers to reproduction (i.e. those preventing reproduction or rendering offspring nonviable) are also expected to show differentiation between the two species (3). As those species adapt to different environmental conditions (evolve toward different evolutionary optima), loci related to the fitness of each species are also expected to accumulate genetic differentiation as a result of divergent natural selection (4). Some of these adaptive loci can lead to ecological barriers to gene flow when mal-adaptive alleles at these loci decrease the fitness of hybrids. The circles corresponding to each category of differentiated loci overlap, because loci can belong to two or more of these categories. In an experiment where individuals from the two species are confronted to different environmental constraints, characteristic of their respective ecological niches, genes will display differential expression following interpretable patterns (5). Two patterns will likely be informative to identify adaptive genes. The expression of genes following the first pattern changes in responses to environmental constraints, but differently for each species (6). We hypothesized that these genes are likely related to the adaptation of the two species to their respective ecological niche in response to the selective pressures (or environmental constraints) tested in the experiments. Genes following the second pattern (7) are more difficult to interpret, but would likely include genes for which constitutive expression at different levels are adaptive to the species respective niches, maybe to untested environmental constraints. Depending on the tissue studied, one could also expect some genes in this category to contribute intrinsic barriers to reproduction between the species. Among the genes following either of the two expression patterns, overlap is expected with loci displaying genetic differentiation among species if variation at those loci influence expression in cis. The other loci with differential expression that do not overlap with highly differentiated loci among species could be trans regulated. Abbreviations correspond to: Env., Environment, F_ST, Fixation index, a measure of population differentiation due to genetic structure.

To date, most of the transcriptomic studies contrasting gene expression among species were performed in animal models (Abzhanov et al., 2006; Jeukens et al., 2010; Martínez-Fernández et al., 2010; Renaut et al., 2010; Rajkov et al., 2021). In plants, contrasting patterns of gene expression among species have mostly been reported in hybrids. For example, in Senecio species, some studies have reported profound changes in gene expression in hybrids, favoring the survival of these plants in new environments not accessible to their parents (Hegarty et al., 2008; Chapman et al., 2013). In sunflower (Helianthus annuus, petiolaris, and deserticola), Lai et al. (2006) found that growth and reproductive success in a new environment was driven mostly by expression divergence (reviewed by Mack and Nachman, 2017). However, in forest trees, very few such studies have been performed (reviewed by Lexer et al., 2004; Neale and Kremer, 2011; Ingvarsson et al., 2016).

European white oaks represent a particularly interesting species complex in which both genetic divergence and differences in gene expression were reported. These species diverged recently (<10 million years ago; Hubert et al., 2014; Manos and Hipp, 2021) and have probably experienced multiple periods of allopatric isolation and contact associated with glacial retreats and postglacial recolonizations (Leroy et al., 2017, 2020). Among the species from this complex pedunculate oak (PO; Quercus robur L.) and sessile oak (SO; Quercus petraea Matt. Liebl.) are known to hybridize (Lepais and Gerber, 2011) but have different ecological preferences (Epron and Dreyer, 1990). In brief, SO is more tolerant of water shortage (Bréda et al., 1993), has higher water-use efficiency (Ponton et al., 2002), and tolerates soil water deficits better than PO (Bréda and Badeau, 2008). In contrast, PO can tolerate longer periods of waterlogging and is generally found in moist habitats.

Identifying the genetics underlying those ecological preferences is of great interest to forest managers. The prolonged rainfall deficit observed in mainland France in 1976 and 2003 triggered oak tree decline in Europe (Bréda et al., 2006; van der Werf et al., 2007). Remarkably, these events had contrasting effects on white oak species (Thomas and Hartmann, 1996), with SO displaying higher growth rates during these dry years (Becker et al., 1994; Lebourgeois, 2006; Bréda and Badeau, 2008; Friedrichs et al., 2009) and PO populations having a higher mortality rate (Durand et al., 1983; Lebourgeois et al., 2015). A potential strategy to maintain PO forests could be to use provenances with introgressions from SO for variants adaptive to drier conditions. At the genomic level, Leroy et al. (2020) recently made use of the oak genome sequence (Plomion et al., 2018) to perform scans for divergence between the four main European white oak species (Q. robur, Q. petraea, Quercus pyrenaica (Pyrenean oak), and Quercus pubescens (pubescent oak)). The regions identified included genes that, according to correlation of allele frequencies with climatic variation and functional annotations, were likely involved either in intrinsic reproductive barriers between species or adaptation to the species respective ecological niches.

Only a few studies have investigated the differences in gene expression patterns between sessile and PO. Porth et al. (2005) provided insight into species differences in mRNA levels between these two species in conditions of water deficit. They detected an upregulation of osmotic adjustment-related genes in SO. In a study based on suppressive subtractive hybridization, Le Provost et al. (2012) provided further insight into the genes potentially involved in the adaptation of PO to waterlogging in white roots. In particular, they identified genes with expression levels displaying strong species-by-environment interactions. These observations suggested that PO had evolved specific strategies to cope with waterlogging. They also showed that tolerance to waterlogging was driven by a rapid switch of PO metabolism to the fermentative pathway in hypoxic conditions, a strategy that could alleviate energy loss caused by oxygen deprivation.

In this context, we propose to (1) investigate in more detail the differences in gene expression between these two species in response to environmental stresses likely to occur in their respective ecological niches, (2) test the overlap between differentially expressed genes (DEGs) and regions of the genome differentiated between the two species, and (3) use the results to differentiate loci involved in adaptation from those where genetic differentiation is driven by other processes (Figure 1).

We subjected pedunculate and SOs seedlings to environmental conditions mimicking the ecological constraints (i.e. an excess or deficit of water) to which they are exposed in natural conditions. We then harvested root tips, quantified expression using RNA-seq, and compared gene expression between control and stressful conditions. We identified (1) genes with differential expression between species independently of the water constraint applied, (2) genes exhibiting similar expression response to water constraints in both species, and (3) genes displaying species-specific patterns of changes of expression when confronted to a water constraint. We hypothesized that genes differential expression between species and in particular those displaying species-specific expression changes when confronted to waterlogging or drought would shed light on the different molecular strategies underlying the ecological preferences of each species (Figure 1, points 4–6).

We found that the genes that displayed different patterns of expression between species included many markers with high genetic differentiation among species (F_ST, see Figure 1). Close inspection of gene annotations and functional networks analysis of genes with species-specific patterns of differential expression between treatments revealed functions likely related to the species adaptation to their respective ecological niches, as expected per our hypothesis. Surprisingly, and despite our study focusing on root tissue, genes that had different expressions between the two species independently of the treatment mostly included functions and pathways consistent with a role in intrinsic barriers to reproduction.

Hybridization between the two species was previously shown to be an important source of adaptive genetic variation. Understanding the mechanisms that limit gene flow between those two sympatric sister species and contribute to adaptation to their respective ecological niches will help design sustainable oak forest management that harness hybridization to foster the adaptive potential of oak populations in the context of rapid climate change.

Results

Characterization of the treatments applied

In this study, we submitted seedlings from two oak species, PO and SO, to waterlogging (waterlogging treatment) followed by a moderate water deficit (drought treatment). Waterlogging consisted in immersing all the seedlings up to 1-cm above the collar in deoxygenated water. We lowered O₂ levels in the water to ∼1 mg.L⁻¹ by bubbling N₂. After about 5 min of bubbling, O₂ concentrations became stable and were monitored throughout the experiment (Supplemental Text section 1). We sampled white roots on seedlings from the two species and three families per species (see “Materials and methods”) just before immersion (control), and after 6 h, 24 h (short-term response). and 9 days (long-term response) of immersion. On the ninth day, the pots of the remaining seedlings were drained and maintained at field capacity for 3 weeks. After this recovery period, we applied a water deficit to half of the remaining seedling by maintaining soil water content at 15% of field capacity during 9 days (the other seedlings served as controls). Predawn water potential fell to −0.9 MPa in the stressed samples compared to the −0.4 MPa in the controls and all seedlings reached the desiccation threshold at about the same time. We monitored the evolution of substrate water content daily as well as predawn and midday leaf water potential before each root sampling (Supplemental Text section 1). On the ninth day of this drought treatment, we sampled white roots on the seedlings from the two species and the three families per species, both in controls and drought conditions.

It should be noted that the waterlogging treatment was likely much more stressful for seedlings than the drought treatment. After 9 days of immersion, we observed the first hypertrophied lenticels, a clear sign of acclimation to soil hypoxia, in the tolerant species (PO) (Le Provost et al., 2016).

Global characterization of transcriptomic data

For further information, see Supplemental Text section 2. Within this file, we provide both a PCA (Principal component analysis) and a hierarchical clustering analysis to evaluate the RNA-seq data quality. Briefly, this descriptive analysis showed close clustering within biological replicates, attesting to the quality of the RNA-seq data. It also showed that the samples (species × treatment) clustered into different groups corresponding to the different combinations analyzed in this study.

Number and effect size of DEGs

We used an experimental design including the two species, two treatments (i.e. water constraints: waterlogging and drought) (see “Materials and methods”), with controls for each treatment. For each treatment, this experimental design allowed us to identify (1) genes with differential expression between species independently of the level of water constraint applied (hereafter “species effect”), (2) genes exhibiting plasticity in both species (hereafter “treatment effect”), and (3) genes displaying species-specific patterns of differential expression between seedlings under water constraint and the appropriate controls (hereafter “interaction effect”). In addition, we also considered genes with similar change of expression in both species in response to both treatments (waterlogging and drought) and all stress levels (short term and long term in the waterlogging treatment) (gene set #7 in “Materials and methods”). Our analysis focused on gene set #7 as well as on the gene sets displaying an “interaction” effect within each treatment (gene set #5 and #6), which together should capture genes involved in the adaptive molecular response of the two species to their respective ecological niche (Figure 1).

For the waterlogging experiment, 18,155 genes covered by more than 90 reads remained for abundance analysis. Most of these genes (18,146 genes, 99.9%) were covered by sequences from both species. We identified only five genes specific to PO and four genes specific to SO (Table 1). We also identified 820 genes with a significant treatment effect (P < 0.01), 1,167 genes with a significant species effect, and 319 genes with a significant interaction effect (Table 2;Supplemental Data Sets 1–3; Figure 2, A and B). Only 25 genes had all three types of effect (Figure 2B), but 183 genes had a significant treatment effect only, 522 had a significant species effect only and 278 genes had an interaction effect.

Table 1.

Overview of species-specific expressed genes for waterlogging or drought treatment

SPECIES	Number	Qrob_ID	Function
	Waterlogging
PO-RESPONSIVE GENES	5	Qrob_P0335670.2	Germacrene-D-Synthase
		Qrob_P0465260.2	Glucose-Induced Degradation Protein
		Qrob_P02281130.2	Kinase Alpha Hydrolase Domain
		Qrob_P0336470.2	PPPDE Thiol Peptidase Family Protein
		Qrob_P0125380.2	SABATH Methyltransferase
SO-RESPONSIVE GENES	4	Qrob_0362510.2	MYB Like HTH Transcriptional Regulator
		Qrob_P0119020.2 Qrob_P0598470.2 Qrob_P0605080.2	Cysteine Rich Repeat Secretory Protein 38
			Subtilase Family Protein T6P5
		14-3-3 Like Protein GF14 Omega
TOTAL	9
	Drought
PO-RESPONSIVE GENES	2	Qrob_P0374520.2 Qrob_P0084850.2	Unknown Protein
			Cyclic Nucleotide Gated Chanel 1
SO-RESPONSIVE GENES	9	Qrob_P0711010.2 Qrob_P0675500.2	Unknown Protein
		Qrob_P0092630.2	Aspartyl Protease Family Protein
		Qrob_P0627090.2	TIR NBS LRR Class Disease Resistance Protein
		Qrob_P0446600.2	HXXXD-Type Acyl-Transferase Family Protein
		Qrob_P0169890.2	NDR/HIN1 Like Protein
		Qrob_P0684400.2	Unknown Protein
		Qrob_P0362510.2	Cycloartenol Synthase
		Qrob_P0593920.2	MYB Like HTH Transcriptional Family Protein
			DUF946 Family Protein
TOTAL	11

Table 2.

Summary of the DESeq2 analysis for the two experiments (waterlogging and drought)

SPECIES	Treatment (M1–M2)	Species (M1–M3)	Interaction (M4)
	Waterlogging
PO	466	682	NA
SO	354	485	NA
TOTAL	820	1,167	319
	Drought
PO	62	179	NA
SO	60	216	NA
TOTAL	122	395	17

The number of genes in each cell was determined with the following parameters: adjusted P < 0.01 and fold-change ≥ 2 and indicate also in which species the genes is upregulated. For each effect, the compared models Mi (see “Materials and methods”) are indicated in parentheses.

Figure 2.

Illustration of the main results obtained from differential gene expression analysis. Left section: waterlogging experiment, Right section: drought stress experiment. A, Bar charts illustrating the number of DEGS obtained for each effect (The lower blue bar was used for PO, The upper red bar for SO). B, Venn diagram of the genes showing a significant treatment, species or interaction effect for each experiment.

After the drought treatment, 17,688 genes were considered according to the criteria defined above. We identified 11 genes specific to either PO or SO (Table 1). Application of the same statistical criteria as for the waterlogging experiment led to the identification of 122 genes with a significant treatment effect, 395 genes with a significant species effect, and 17 genes with a significant interaction (Table 2;Supplemental Data Sets 4–6; Figure 2, A and B). The overlap between these three types of effects is shown in the Venn diagram of Figure 2B. No gene displayed all three types of effect simultaneously, and we found 68 genes with a significant treatment effect only, 339 genes with a significant species effect only and 15 genes with a significant interaction. The lower number of DEGs identified after the drought treatment is probably due to the moderate water deficit applied.

We found that 184 of the genes displaying a “species” effect (i.e. 13.4% of the “species” effect genes) were differentially expressed between species in both treatments (waterlogging and drought; Supplemental Data Set 7). This overlap was highly significant (${\chi }^2$ on a 2 × 2 contingency table = 1154, P < 2.2e⁻¹⁶). These genes are listed in Supplemental Data Set 7.

Validation of RNA-seq results by RT–qPCR

For further information, see Supplemental Text section 3. In this file, we describe the reverse transcription–quantitative PCR (RT–qPCR) approach performed to validate our bioinformatic pipeline. For each experiment, we selected genes displaying a significant treatment, species, or interaction effect for RT–qPCR validation.

Gene set and subnetwork enrichment analyses

The main goal of this study was to highlight the molecular mechanisms involved in adaptation to waterlogging or drought, in PO and SO, respectively, rather than to identify genes which respond the same way in both species to these constraints. We, therefore, chose to focus the gene set and subnetwork enrichment analyses on the genes displaying a significant interaction (in either treatment, i.e. gene sets #5 and #6). Indeed, we hypothesized that these two gene sets should include genes involved in species-specific adaptive molecular strategies. We also scrutinized genes displaying a “species” effect in specific environmental conditions (gene sets #3 and #4), as well as in both conditions (intersect of gene sets #7), as higher basal levels of expression in the tolerant species may explain in part its better adaptation to the environmental constraints with which it is usually confronted in natural conditions. This hypothesis is supported by studies performed both in wheat (Triticum aestivum) (Zhang et al., 2018) and Drosophila (Horváth et al., 2022). For instance, Zang et al. (2018) reported in several genotypes, an association between the basal expression of PR protein and the tolerance level to Puccinia rust (Puccinia tritici). Similarly, In Drosophila, a higher basal expression of genes potentially involved in desiccation was identified by Horváth et al. (2022) using tolerant and sensitive strains. While this section focuses exclusively on the sets of genes displaying an “interaction” or a “species” effect in both treatments, the genes displaying a “species” effect in either waterlogging or drought treatment are presented in Supplemental Text section 4.

Gene set enrichment analysis was performed to identify classes of genes (based on gene ontologies, GOs) over expressed in our experimental settings. The results are shown only for the first 100 GO (ranked according to their P-value) in Supplemental Data Sets 2, 3, 5, 6, and 7. A graphical representation (using bubble plots) of the first 20 ontologies is also available in the same supplementary files for a clearer view of the ontologies regulated.

For genes with an “interaction” effect, the highest levels of enrichment during waterlogging were observed for the biological processes (BPs) “positive regulation of flavonoid biosynthetic process,” and “regulation of meristem structural organization,” and for the molecular functions (MFs) “phosphorelay response regulator activity” and “ubiquitin-protein transferase activity.” Additional information on the most enriched terms in the three main ontologies is also available in the supplemental Data Set 3 in both tabular and graphical forms.

For genes displaying a “species” effect whatever the treatment considered (Supplemental Data Set 7), BPs “oxidation-reduction process” and “response to stimulus” were found to be enriched, and enrichment was observed for the MFs “oxygen binding” and “ADP binding.” Additional information on the most enriched ontologies is available in Supplemental Data Set 7.

Subnetwork enrichment analysis was performed to highlight key molecular players involved in adaptation to waterlogging in PO or drought stress in SO. A network based on interaction-responsive genes was generated for each treatment. The functional networks are shown in Figure 3 for waterlogging and Figure 4 for drought stress. The genes with an interaction effect in waterlogged conditions revealed important hubs related to root growth (“root growth”), hormone signaling (“response to auxin stimulus” and “jasmonate response”), cell differentiation and expansion (“meristem function” and “cell expansion”), plant defense and cell death (Figure 3). These functions encompassed 34 out of 319 genes (i.e. 14%) identified in our DEG analysis. Different hubs were identified for drought stress (Figure 4), with MFs related to “defense response,” “response to auxin stimulus,” “growth yield,” and “seed yield.” A greater number of DEG (8/17, 47%) were detected in the network generated for the drought stress interaction responsive genes.

Figure 3.

Functional network for genes displaying a significant treatment-by-interaction effect in the waterlogging experiment. We used the k_means algorithm available in the Expender software to group the genes identified according to their expression profile. Abbreviations correspond to: Cont, control; Stress, long term response; CL, Cluster from the k_means analysis; +, positive action of the gene. The blue line was used for PO while the orange line for SO. Genes highlighted in light blue and dark blue belong to cluster #1 and cluster #2, respectively, while those in light red and dark red belong to cluster #3 and cluster #4, respectively. Ovals were used for genes encoding proteins while ovals with bars were used for genes encoding transcription factors. The gray color was used to highlight the molecular mechanisms that interact with the corresponding DEG.

Figure 4.

Functional network for genes displaying a significant treatment-by-interaction effect in the drought stress experiment. We used the k_means algorithm available in the Expender software to group the genes identified according to their expression profile. The blue line was used for PO while the orange line for SO. CL1: Cluster1, CL2: Cluster2. The cluster to which the gene belongs is also indicated (CL1: Cluster #1, CL2: Cluster #2). The gray color was used to highlight the molecular mechanisms that interact with the corresponding DEG.

The subnetwork generated from the genes displaying a “species” effect whatever the treatment analyzed is shown in Figure 5. We identified hubs related to: “flowering time,” “cell differentiation,” “cell proliferation,” “developmental process,” and “seed germination.” Meristematic cells generally display an enrichment in these MFs. It should be noted that most of the genes displaying a species effect were not identified in our subnetwork enrichment analysis (166/184, 90%).

Figure 5.

Functional network for genes displaying a significant species effect whatever the stress applied. Genes highlighted in blue and red have a higher expression in PO and SO, respectively. The intensity of the color is proportional to the expression level of the gene. The gray color was used to highlight the molecular mechanisms that interact with the corresponding DEG.

Overlap between loci showing a high degree of genetic differentiation between species and DEGs

For this comparison, we used a published dataset from a study investigating adaptive differentiation between PO and SO based on pool sequencing (Leroy et al., 2020). The mean F_ST value across the 39,644,639 single-nucleotide polymorphisms (SNPs) with F_ST > 0 was 0.0545, suggesting weak overall genetic differentiation between species. However, F_ST estimates ranged from 1.17e⁻⁶ to 1. We assessed enrichment in gene sets identified by differential expression analysis for SNPs in the 1%, 0.5%, 0.1%, 0.01%, and 0.001% right-hand tails of the genome-wide F_ST distribution, corresponding to thresholds of 0.14, 0.217, 0.454, 0.803, and 1, respectively (Figure 6). In each of the seven gene sets, we found significant enrichment in SNPs displaying a high degree of differentiation between the two oak species (Figure 6). Enrichment ratios exceeded six-fold for the species and treatment effects in the drought experiment. The lists of genes including highly differentiated markers for the various thresholds are provided in Supplemental Data Set 8.

Figure 6.

Genes differentially expressed between treatments and species in the drought and waterlogging experiments are enriched in highly differentiated SNPs between Q. robur (PO) and Q. petraea (SO). SNPs differentiation between the two species Q. robur and Q. petraea were obtained from Leroy et al. (2020). The y-axis represents the enrichment of the seven DEGs sets investigated in this study (x-axis). For example, an enrichment of above six for genes differentially expressed between treatments in the drought experiments indicates that this gene set included over 6 times more genes with highly differentiated SNPs than expected by chance. For each gene set, the five bars correspond to enrichment for genes including SNPs with F_ST values above the 0.9, 0.95, 0.99, 0.999, and 0.9999 quantile of the genome-wide FST distribution, respectively. Significance of enrichment estimates were assessed using 1,000 permutations and are indicated above each bars: ns, nonsignificant, “.”: P < 0.1, *P < 0.05, **P < 0.01, and ***P < 0.001.

We also found a significant enrichment in high-F_ST SNPs among genes displaying differential expression between species and across treatments (Figure 6, rightmost barplot).

Discussion

Here, we investigated the contribution of gene expression to the divergence of PO and SO. In our design, each species was subjected to two successive episodes of water stress (excess or deficit) corresponding to the environmental constraints that each species naturally encounter in its ecological niche. This made it possible to focus on genes displaying species-specific changes of expressions upon encountering ecological constraints (“species × treatment,” 319 DEG for waterlogging and 17 for drought stress) and on genes expressed differentially between species regardless of the treatment imposed (184 genes). We hypothesized that these genes likely included some involved in species-specific adaptive molecular features presumably involved in the ecological preferences of these species. In addition to these two sets of DEGs (the focus of the discussion), genes differentially expressed between the species in each treatment may also provide an important source of information, because genes with higher basal levels of expression in tolerant than in intolerant species may contribute to adaptation to a specific environmental condition. A dedicated discussion on these genes is provided in Supplemental Text section 4. It should be borne in mind that this study focused on root tips and that other tissues like leaves would also be of interest to understand the molecular mechanisms involved in the ecological preference of pedunculate and SOs. Studying changes in expression in leaves would likely lead to the discovery of different molecular mechanisms like the regulation of transpiration for example. We made the choice to study gene expression in the root tips because it is the first organ that senses the stresses applied in this study.

Molecular mechanisms underlying the ecological preferences of PO and SO

We focus here on the genes displaying only a significant treatment-by-species effect in the waterlogging and drought stress experiments. As stated in the “Introduction,” we hypothesized that genes with expression levels significantly affected by the species × treatment interaction would include genes involved in the respective ecological preferences of the two species.

What did we learn from the waterlogging experiment?

Waterlogging had a stronger effect than drought in terms of the number of DEGs (319) displaying “species × treatment” interactions. Our subnetwork enrichment analysis identified BPs (i.e. root growth, cell death, etc.) known to be regulated during aerenchyma and adventitious root formation in tolerant plant (Evans, 2003). In PO, Parelle et al. (2006) reported that morphological and anatomical changes (such as the formation of adventitious roots and aerenchyma) were essential to cope with hypoxia. They also reported the formation of larger amounts of these adaptive structures in PO than in SO. This formation of aerenchyma or adventitious roots in waterlogging-tolerant plants may be triggered by genes belonging to the main hubs identified here, suggesting that the molecular mechanisms identified here are particularly important for the adaptation of PO to waterlogging. Below we review some of the genes identified, reporting also their F_ST value.

(1) Programmed cell death leading to aerenchyma formation

Programmed cell death (PCD) is a major mechanism underlying plant development and adaptation to abiotic or biotic stress (Smith et al., 2015). In particular, it is involved in the modification of root architecture under drought stress and the formation of aerenchyma in hypoxic conditions. Tolerant plants typically produce aerenchyma in response to hypoxic conditions through the lysis and death of cells in the root cortex to produce additional gas space (Yamauchi et al., 2013). The tolerance to waterlogging observed in PO may be partially explained by its ability to produce more aerenchyma than SO (Parelle et al., 2006), a finding confirmed by our network analysis, which identified several genes related to cell death, a process essential to aerenchyma formation (Figure 3). The genes upregulated in PO included (1) Long-Chain Base 1 (LCB1; Qrob_P0060010.2, F_ST:0.43) encoding a LCB1 subunit of serine palmitoyltransferase. In Arabidopsis (Arabidopsis thaliana), this gene has been shown to regulate PCD by inducing reactive oxygen intermediates (Shi et al., 2007), (2) Beclin 1 protein (ATG6, Qrob_P0479390.2, F_ST:0.19) involved in the regulation of autophagy (Xu et al., 2017), (3) DNA-binding protein 9 (WRKY 9, Qrob_P0436950.2, F_ST: 0.24), which is known to be involved in cell death regulation in Nicotiana benthamiana (Liu and He, 2017), and (4) SKP1 (Qrob_P0747290.2, F_ST:0.2) encoding S-phase kinase-associated protein 1 and including multiple differentiated SNPs with a maximum F_ST value of 0.73. Thus, this report potentially links SKP1 protein to aerenchyma formation.

(2) Root growth and phytohormone biosynthesis

The formation of adventitious roots is another major morphological response to waterlogging in tolerant species. These roots contain aerenchyma (i.e. air channels), which favors the diffusion of gases in conditions of submersion (Evans, 2003). We identified important hubs potentially involved in the establishment of adventitious roots (root growth, cell expansion, and meristem function) (Figure 3). We also detected hubs related to phytohormones (response to auxin stimulus and jasmonate response) known to play a key role in the formation of adventitious roots.

Our analysis highlighted genes (upregulated in PO) involved in either primary root growth inhibition or lateral root development. A first subset of these genes is involved in auxin signaling (HDA6, WRKY23, Auxin Receptor F-BOX 3 [AFB3]) or jasmonate responses (Phytochrome B [PHYB], SKP1, 4-Coumarate:COA Ligase 2, MYC2 [basic helix–loop–helix protein 6]) consistent with the modification of the root system during hypoxic responses in PO being driven by phytohormones. For instance, MYC2 (Qrob_P0056680.2, F_ST: 0.24) encodes a basic helix–loop–helix DNA-binding protein. Kazan and Manners (2013) identified this gene as a master regulator of many aspects of jasmonate signaling playing a strong role in the formation of lateral and adventitious roots. EXO (Phosphate-responsive 1 family protein, Qrob_P0764920.2, F_ST: 0.14) is under the control of brassinosteroids. Its overexpression in Arabidopsis has been reported to promote shoot and root growth (Schröder et al., 2009).

Respiratory burst oxidase homolog protein B (RBOHB, Qrob_P0493470.2, F_ST: 0.28), which encodes a respiratory burst oxidase protein involved in the production of reactive oxygen species (ROS), is also upregulated in PO. The ROS produced by the product of RBOHB genes are involved in several developmental processes. Montiel et al. (2013) reported that the overexpression of this gene promotes lateral root growth, suggesting a potential role in adventitious root formation in PO. We also identified Novel Interactor of Jasmonic Acid domain 10 (NINJA, Qrob_P0009970.2, F_ST: 0.72) as being more strongly expressed in PO. This gene encodes an interactor of the jasmonate Zim domain 10. In a study of Arabidopsis mutants, Acosta et al. (2013) reported a downregulation of jasmonate signaling by NINJA that was associated with a large decrease in the size of the root system. We hypothesize that the overexpression of NINJA in PO is associated with its ability to maintain adventitious root formation.

Two other genes were found to be regulated in PO: HDA6 (Qrob_P0451700.2, F_ST: 0.82), which encodes an RPD3-like histone deacetylase, a key enzyme involved in the jasmonate response (Wu et al. (2008), and Cythochrome P450 protein (CYP94B3; Qrob_P0369790.2, F_ST: 0.49), which encodes a cytochrome P450 protein involved in jasmonate catabolism. Heitz et al. (2012) reported an association between CYP94B3 overexpression and insensitivity to jasmonate and a smaller lateral root system. These results highlight a stronger inhibition of the jasmonate response in PO, which is essential for lateral root formation, potentially accounting for the increase in lateral root system size in PO during waterlogging.

(3) Meristem function and response to auxin stimulation

The WRKY23 gene (Qrob_P0697030.2, F_ST: 0.59), which is involved in the auxin response, was more strongly expressed in PO. Grunewald et al. (2012) reported an important role for this gene in the establishment of an auxin gradient in the root system, allowing the maintenance of root meristematic activity and lateral root formation.

The AFB3 gene (Qrob_P0036220.2, F_ST: 0.75) was also upregulated in PO. AFB genes belong to a multigene family regulated by auxin and involved in several plant developmental processes. In A. thaliana, Vidal et al. (2010) showed that the AFB3 mutant was characterized by altered lateral root growth, whereas the primary root system was unaffected, suggesting a key role for AFB3 in modulating the architecture of the root system.

Overall, our network analysis highlighted several molecular mechanisms potentially involved in the formation of adaptive structures known to favor oxygen diffusion in the root system (aerenchyma and adventitious roots). This suggests that molecular mechanisms analogous to those observed in other tolerant plant species are more upregulated in PO relative to SO in hypoxic conditions.

What did we learn from the drought stress experiment?

The functional network obtained from the 17 genes displaying a significant drought-by-species interaction effect is presented in Figure 4. It should be noticed that the few genes identified in the drought stress experiment are involved in several BPs (Figure 4) suggesting that they are potentially key regulators of the drought stress response of oak. Three genes upregulated in SO were considered to be of particular importance (Phloem Protein 2-A1 [PP2-A1], CT-BMY (Beta Amylase), and Vacuolar Auxin Transport Protein (WAT1). The WAT1 gene (Qrob_P0331600.2, F_ST: 0.23) encodes a vacuolar auxin transporter. Auxin plays a crucial role in abiotic stress responses controlling vascular development (Ranocha et al., 2013). Zhang et al. (2017) reported that this gene was upregulated in drought-tolerant maize (Zea mays) lines during drought stress. We also identified a CT-BMY gene (Qrob_P0567500.2, F_ST: 0.25) encoding a beta-amylase known to be upregulated during abiotic stress. The upregulation of this gene is generally correlated with maltose accumulation, which protects proteins, membranes, and the photosynthetic electron transport chain during water stress (Kaplan and Guy, 2004). Finally, we identified PP2-A1 (Qrob_P0654080.2, F_ST: 0.36), a gene involved in the plant defense response. Zhang et al. (2011) reported a role for PP2-A1 in the activation of plant defense responses during biotic stress. Overall, our functional network analysis identified drought-responsive genes in SO related to three major pathways: ABA response, osmoregulation, and plant defense, all of which are known to be upregulated in drought-tolerant genotypes. These results suggest that SO has a better capacity to maintain both its primary root growth and osmoregulation, facilitating the maintenance of water metabolism during water shortage and greater tolerance to drought stress. In addition, most of the genes highlighted by our functional network presented at least one SNP above the 90% quantile of the genome-wide F_ST distribution, suggesting that our functional analysis yielded genes not only important for the response to stress in SO, but also probably subject to strong positive selection.

Genes with constitutive differential expression between species have functions consistent with intrinsic reproductive barriers

We focus here on the genes displaying a “species” effect regardless of the treatment imposed. We hypothesize that this gene set should also include key molecular players potentially involved in the adaptation of the species to their respective ecological niches. However, and despite our analysis being based on root tips, our functional analysis (Figure 5) highlighted several processes (i.e. “flowering time,” “cell proliferation,” and “cell differentiation”) which could be involved in intrinsic reproductive barrier between species rather than adaptation to their respective ecological conditions. Our transcriptomic data were obtained from roots rather than floral meristems, but the molecular mechanisms identified here probably reflect important functions of plant meristems generally. The interpretation that genes displaying stable differences in expression independently of the environmental conditions could be directly involved in limiting reproduction between species is also supported by the high level of differentiation of the genes identified in our biological network, ranging from 0.15 (F9L11.8) to 1 (BAS1). Moreover, 164 of the 184 genes identified in the whole dataset (89%) were also found in the 0.1% tail of the genome-wide F_ST distribution, with maximum F_ST values per genes ranging from 0.14 to 1 (Supplemental Data Set 8).

We considered eight genes in the biological network shown in Figure 5 to be of particular interest. Four of these genes relate to the “flowering time” process: BAS1 (Phyb Activation-Tagged Suppresor 1, Qrob_P0551020.2, F_ST: 1), AP3 (Floral Homeotic Protein APETALA 3, Qrob_P0454040.2, F_ST: 0.64), CYP94B3 (Cythochrome P450 protein, Qrob_P0369770.2, F_ST: 0.47) and F12M16.14 (Na⁺/Ca²⁺ exchanger-like protein, Qrob_P0657620.2, F_ST: 0.31). Sandhu et al. (2012) showed that the BAS1 gene encodes a protein involved in brassinosteroid catabolism and reported a key role for brassinosteroid inactivation during the floral induction of AP3, a key component of meristem identity in plants. AP3 interacts with AP1 to specify floral meristem identity during floral transition in A. thaliana (Liu et al., 2007). CYP94B3 encodes a cytochrome P450 protein. Bruckhoff et al. (2016) suggested that this gene might play an important role in controlling flowering time by inactivating the phytohormone jasmonyl isoleucine. We also identified two other CYP proteins with high F_ST values in our gene list (but not in our functional network): CYP72A9 (Qrob_P0084340.2, F_ST: 1) and CYP76G1 (Qrob_P0088450.2, F_ST: 0.75). Finally, we identified a F12M16.14 gene encoding a Na⁺/Ca²⁺ exchanger-like protein involved in maintaining Ca²⁺ homeostasis. Li et al. (2016) reported that transgenic Arabidopsis lines displayed alterations to flowering time due to changes in the expression of two major flowering genes (the Constans and Flowering loci).

Interestingly, three of the genes identified in our network analysis (ABA-responsive element-binding protein 3 [AREB3], Qrob_P0187720.2, F_ST: 0.75), TT2 (R2R3 MYB Transcription factor, Qrob_P0304680.2, F_ST: 0.66) and F4P9.36 (Dihydroflavonol-4-reductase protein, Qrob_P0554070.2, F_ST: 0.57)) belong to the “seed germination process” (Figure 5): (1) AREB3 encodes an ABA-responsive element-binding protein with a bZIP domain. Wang et al. (2015) reported a key role for this gene in germination, seed development, and embryo maturation. Moreover, Hoth et al. (2010) showed that AREB3 is co-expressed with a sugar transporter gene (AtSUC1) in pollen, (2) TT2, encoding an R2R3 MYB transcription factor involved in anthocyanin/proanthocyanin accumulation was also identified. Zhao et al. (2019) reported that Arabidopsis mutants with a downregulated TT2 gene had a shorter dormancy period, suggesting an important role for this gene in meristem functioning, and (3) F4P9.36, encoding a protein similar to dihydroflavonol-4-reductase. Østergaard et al. (2001) showed that this protein was involved in embryogenesis and seed maturation in A. thaliana.

Finally, we identified a SOT17 gene (Desulfoglucosinolate sulfotransferase protein, Qrob_P0168490.2, F_ST: 0.72) associated with cell proliferation. The SOT17 gene has been reported to be expressed in the shoot apical meristem and is potentially involved in cell proliferation in the root meristem via the glucose TOR signaling pathway (Kim et al., 2017; Xiong and Sheen, 2013). It is possible that this gene is implicated in several molecular processes like cell division, pollen germination, and tube growth, which are known to contribute to the establishment of intrinsic barriers to reproduction (Tian et al., 2021).

However, it should be pointed out that most of the genes (166 out of 184, 90%) were not found in the subnetwork enrichment analysis. Among them, genes with high F_ST values included:

• two CAS1 genes (Cycloartenol synthase 1Qrob_P0684390.2, F_ST:1, Qrob_P0095480.2, F_ST:0.63) encoding a cycloartenol synthase 1 involved in sterol biosynthesis. Babiychuk et al. (2008) showed that the CAS1 genes have a key role in the male gametophyte function and in the meristematic activity.

• a NOT2 gene (Polymerase II-dependent transcription protein, Qrob_P0689750.2, F_ST: 0.77) involved in male gametophyte initiation. Wang et al. (2013) reported that the downregulation of NOT2 in Arabidopsis caused severe defects in the male gametophyte.

• a F14M4.3 gene (F_ST: 0.92) similar to a short-chain dehydrogenase reductase (SDR) 5. Cheng et al. (2002) reported that the SDR super family is involved in both hormone biosynthesis in mammals and sex determination in maize.

• three beta glucosidase 17 (BGLU17) genes with F_ST values ranging from 0.97 to 1 (Qrob_P0491340.2, Qrob_P0491360.2, and Qrob_P0597130.2). BGLU17 encodes beta glucosidase proteins potentially involved in meristem functioning through the initiation of cell division. Brzobohaty et al. (1993) reported that beta glucosidase cleaved the cytokinin-O-glucoside conjugates, which are biologically inactive, to release active cytokinin, which is essential to initiate cell division in the meristem.

• a PAP2 gene (Qrob_P0010280.2, F_ST: 0.79) encoding a protein highly similar to phytochrome-associated protein 2. Kobayashi et al. (2010) showed that the pattern of meristem initiation was disorganized in rice (Oryza sativa) mutants for this gene. In the mutant transgenic lines, new meristems were unable to develop into spikelet meristems, highlighting a key role for PAP2 in the initiation of floral meristems.

• finally, a F13K3.9 (Qrob_P0012240.2, F_ST: 0.89) gene encoding a 2-oxoglutarate and Fe(II)-dependent oxygenase protein was identified. Ciannamea et al. (2006) showed that this gene was downregulated during vernalization, suggesting a possible role in floral meristem initiation.

Thus, our study highlighted several genes potentially involved in flowering time regulation, meristem functioning, embryo maturation, cell proliferation, male gametophyte initiation, and floral meristem identity. All the genes discussed in this section had very high F_ST values, suggesting that they are potential targets of natural selection. Overall, these results provide support for a potential role of these genes in maintaining intrinsic barriers to reproduction between the studied oak species.

Intrinsic barriers between species or adaptation to contrasting ecological niches

A surprising result of this study may be the nearly systematic enrichment of gene sets identify by the differential expression analysis in genes including high F_ST genetic polymorphisms. This supports our hypothesis that differential expression analysis in conditions mimicking major constraints contributing to the definition of sister species niches, may be a valid way to highlight adaptive loci. It also indicates that many genes detected in our experiments could be regulated in CIS by loci presenting strong allele frequency differentiation among SO and PO. Loci implicated in adaptation but regulated in TRANS, would only be detectable in a eQTL (Expression quantitative trait loci) study where numerous individuals from a mapping population are exposed to the different stresses and the variation in expression mapped. It is likely that the loci detected in such a study would also be enriched in high F_ST markers.

Excluding genes displaying differential expression among treatments, the highest enrichment values were observed for genes differentially expressed among species irrespective of the treatment (Figure 6). We predicted that some of these genes would also be involved in adaptation of the species to their ecological niches, but our functional analysis and the MFs described in the previous paragraph tend to support the idea that these genes could be involved in intrinsic barriers to reproduction between the two species (flowering time regulation, meristem functioning, embryo maturation, cell proliferation, male gametophyte initiation, and floral meristem identity). Obviously, our differential expression analyses were carried out on roots and many genes that may contribute to species delimitation are likely not expressed in this tissue. We predict that a differential expression analysis on flowers could yield an even stronger enrichment as additional genes involved in these functions are likely to be expressed in reproductive organs.

Genes for which expression changes between treatments were different between PO and SO (“Interaction” in Figure 6) also displayed enrichment in high F_ST polymorphisms, especially in the hypoxia treatment. Exposure to waterlogging is a key difference in the ecological niche of the two species (Truffaut et al., 2017), with PO capable of surviving much higher levels of hypoxia than SO. According to our conceptual framework (Figure 1), this gene set should include genes involved in the genetic adaptation of the two oak species to their respective ecological niches and the functional analysis supports this hypothesis. In particular, genes implicated in the jasmonate and auxin responses in PO have particularly high F_ST values and could be key adaptive regulatory loci.

Overall our results support the idea that differential expression analysis between sister species in treatments mimicking their respective ecological niches can help untangled the forces driving genetic differentiation among species. In this species complex where extensive gene flow is maintained in nature, many individuals are admixed, meaning that their genomes are a mosaic of the genomes from the PO and SO at different proportions. While mostly impossible to assess based on morphological criteria, this variation is potentially important to generate adaptive solutions in the context of rapid environmental changes. Identifying loci that are both differentiated between the parental species and likely adaptive to specific environmental conditions may become key to guide the introduction of relevant genetic variation in populations to accelerate adaptation in these long-lived species.

Materials and methods

Plant material

The plant material used here was described in a previous study by Le Provost et al. (2016). Briefly, we sampled half-sib progenies from three unrelated mother trees for both SO (Q. Petraea, located in “Forêt Domaniale” of Laveyron, latitude 43°45′49″N, longitude 0°13′11″W) and PO (Q. robur, from “Ychoux forest,” latitude 44°33′33″N, longitude −0°96′66″W), two sites representative of the differences in ecological niche between the two species. Before germination, we confirmed the species status of the mother trees with the diagnostic SNP markers described by Guichoux et al. (2011). We harvested the acorns in the fall of 2013 and sowed them in a 1:1 mixture of peat and sand in 0.2-L pots. After germination, we transferred the seedlings to a greenhouse (16-h photoperiod, 25°C during the day and 20°C at night) for 5 weeks, until they had three fully developed leaves.

Experimental design and measurement of physiological traits

Sixty homogeneous seedlings (i.e. with three fully developed leaves) were selected for each mother tree and species, corresponding to a total of 360 seedlings (60 half-sibs * 3 mother trees used as biological replicates * 2 species) and exposed to a waterlogging/drought stress cycle. An overview of the experimental design is presented in Supplemental Figure S1.

(1) Waterlogging treatment: The 360 seedlings were placed in five 100-L plastic containers corresponding to five blocks. The plastic containers were filled with water that was deoxygenated directly by bubbling with N₂. The seedlings were immersed in the water such that the water level reached 2-cm above the collar of the seedlings. The O₂ concentration in each container was monitored daily with a portable O₂ electrode (Cellox 325, WRW, Weilheim, Germany). Each block comprised 12 seedlings from the same species and mother tree. We sampled white roots from 10 seedlings (i.e. two in each of the five blocks) for each species and each mother tree after 0 (control), 6 and 24 h (short-term stress) and 9 days (long-term stress) of hypoxia. The sampling was destructive so 10 seedlings per genotype and species were removed from the experimental design for each sampling point as described in Supplemental Figure S1. The roots were immediately frozen in liquid nitrogen and stored at −80°C until RNA extraction. The 10 seedlings from a same mother tree were pooled for further RNA-seq analysis.

(2) Drought stress treatment: We removed the remaining 120 seedlings from the deoxygenated solution on Day 10. Once the excess water had percolated through the substrate by gravity, the seedlings were maintained at field capacity for three weeks and were then subjected to drought stress (Supplemental Figure S1). Drought stress was achieved by keeping water level in the substrate at 15% of field capacity for 9 days. Before the start of the experiment, we determined the water retention capacity of the substrate by weighing the pots at field capacity and then again after drying the substrate (at 65°C for 24 h). The amount of substrate per pot was also determined before sowing the acorns, to make it possible to measure the maximum water content per pot (∼170 g) independently of the biomass produced. Drought stress was then applied in the greenhouse, as previously described (Marguerit et al., 2012). We monitored the amount of water in the substrate daily, by weighing the pots with a precision of ±1 g (Sartorius, Aubagne, France), until the water content in the pots reached 15% field capacity (corresponding to a weight loss of 144 g). Pots that dried faster were maintained at 15% water content for 9 days before the harvesting of white roots. In parallel, we maintained a subset of seedlings at field capacity to serve as a control. We harvested white roots from these seedlings at the same time as from the stressed sample to prevent confounding ontogenetic effects (Supplemental Figure S1). As for the waterlogging treatment, we also generated three biological replicates for gene expression analysis by pooling the seedlings from the same mother tree (i.e. 10 seedlings per mother tree; Supplemental Figure S1). We also measured leaf predawn and midday water potential both for the controls and treated seedlings before each sampling, using a Scholander-type pressure chamber as described by Hsiao (1990). Measurements were made on five seedlings each for the control and the drought-stressed treatment.

We note that in this study seedlings were first exposed to waterlogging before applying the drought stress. This strategy may have consequences on both the number and the expression level of genes identified in drought stress experiment. We made this choice to mimic as closely as possible the environmental condition encountered by the seedlings in the wild. Indeed, they are exposed to waterlogging events during early spring followed by drought stress events in early summer influencing probably the repartition of the two species in the wild. We used seedlings <2 months old. This stage is relevant as the environmental constraints we studied mostly impact forest stand-specific composition via seedling mortality.

RNA extraction and sequencing

RNA was extracted and purified as described by Le Provost et al. (2007). Residual genomic DNA was removed with an RNase-free DNase, RQ1 (Promega, Madison, WI, USA), according to the manufacturer’s instructions, before purification. We assessed the quantity and quality of the extracted RNA on an Agilent 2100 Bioanalyzer (Agilent Technologies, Inc., Santa Clara, CA, USA).

For the waterlogging treatment, we constructed nine libraries for each species, corresponding to three biological replicates × three treatments (control, short-term response assessed by pooling equimolar amounts of total RNA extracted after six and 24 h (with the main goal to capture a maximum of genes involved both in the perception and transduction of the signal), and long-term response assessed on RNA extracted after 9 days of waterlogging).

For the drought stress treatment, we generated six libraries for each species, corresponding to three biological replicates × two treatments (control and long-term response, i.e. after 9 days of water levels at 15% field capacity in the substrate).

We generated these 30 libraries as described by Le Provost et al. (2016) and according to the Illumina protocol (TruSeq Stranded mRNA Sample Prep Kit). Briefly, we selected mRNA from 2 µg of total RNA. The mRNA was then chemically fragmented and reverse transcribed with a random hexamer primer. The second strand of the cDNA was generated, 3′-adenylated, and Illumina adapters were added. We amplified DNA fragments (with adapters) by PCR (Polymerase Chain Reaction) with Illumina adapter-specific primers. We quantified the libraries with a Qubit Fluorometer (Life Technologies, NY, USA). We estimated their size with Agilent 2100 Bioanalyzer technology (Agilent). We then sequenced each library by 101 base-read chemistry, in a paired-end flow cell, on an Illumina HiSeq2000 (Illumina, San Diego, CA, USA). More than 24 million usable reads were generated for each library (Supplemental Table S1).

Cleaning, mapping, and identification of DEGs

We applied the following two-step procedure to identify DEGs. We first removed low-quality reads (Phredscore < 20). The high-quality reads were then aligned with the 25,808 oak gene models previously published with the reference oak genome (Plomion et al., 2018), with the BWA -MEM mapper version 0.6.1 aligner (Li and Durbin, 2009) and a maximum insert size of 600 bp, with four mismatches allowed. A gene model is defined as a full gene from the start to the stop codon. We then selected gene models with at least 90 reads (i.e. over all biological replicates) for the differential analysis. In the second step, we used the DESeq2 package (Love et al., 2014) to identify DEGs (i.e. significant genes) with a P < 0.01 after adjustment for multiple testing with a false discovery rate (FDR) of 5%. The expression level of each gene was quantified by the TMM normalization method (Trimmed mea of means). We considered only DEGs for which at least a two-fold change in expression was observed. For each experiment, we assessed the effects of treatment, species, and their interaction in likelihood ratio tests implemented in the DESeq2 package.

The treatment and species effects were assessed by comparing a model without interaction (M1) with two simplified models for the treatment (M2), and species (M3) effects.

M1: Y_ijk = μ + T_i + S_j +ε_ijk, where T_i is the treatment effect (i = “control,” “short-term” “long-term” for waterlogging and i = “control” or “long-term” for drought stress), and S_j is the species effect (j = “PO” or “SO”).

$$\mathrm{M}2:\mathrm{ }{Y}_{\mathrm{jk}}\mathrm{ }=\mathrm{ }\mathrm{\mu }\mathrm{ }+S_j\mathrm{ }+{\mathrm{\epsilon }}_{\mathrm{jk}}$$

$$\mathrm{M}3: Y_{\mathrm{ik}}\mathrm{ }=\mathrm{ }\mathrm{\mu }\mathrm{ }+T_i\mathrm{ }+{\mathrm{\epsilon }}_{\mathrm{ik}}$$

For interaction effects, we compared a complete model: Y_ijk = μ + T_i + S_j + (S*T)_ij + ε_ijk (M4) to M1.

Differential gene expression analyses, therefore, yielded six gene sets: (#1) genes differentially expressed between waterlogged and control conditions (across species); (#2) genes differentially expressed between drought and control conditions (across species); (#3) genes differentially expressed between species throughout the waterlogging experiment; (#4) genes differentially expressed between species in the drought stress experiment; (#5) genes displaying significant treatment-by-species interaction in the waterlogging experiment; and (#6) gene displaying significant treatment-by-species interaction in the drought stress experiment. An additional gene set was also generated from the intersect between genes displaying a “species” effect regardless of the treatment considered (gene set #7).

Annotations for the DEGs were recovered from the PO reference genome (Plomion et al., 2018).

Gene set and subnetwork enrichment analysis

We performed GO-term enrichment analysis for the seven gene sets described above with the topGO R package (Alexa and Rahnenfuhrer, 2022). We corrected P-value for FDR and considered ontology terms with a corrected P < 0.05 to be significantly enriched.

We then performed enrichment analysis with Pathway Studio software (Pathway Studio, Elsevier 2017), as described by Le Provost et al. (2016). Briefly, the analysis is a comparison of the input entity list with the entities within the subnetwork using Fisher’s exact test.

Since our main goal was to identify genes potentially involved in ecological preferences, we focused on the sets of genes presenting significant treatment-by-species interactions (gene sets #5 and #6). We also analyzed genes differentially expressed between species either across treatments (gene sets #7, likely revealing genes involved in reproductive isolation between PO and SO) or in specific environmental conditions (gene sets #3 and #4, i.e. presenting higher basal levels of expression in the tolerant species, namely PO for excess water and SO for water deficit).

RT–qPCR

For further information, see Supplemental Text section 3. A brief description of the methods used to perform RT–qPCR analysis is provided.

Relative genetic divergence between DEGs

We investigated whether DEGs presented evidence of potentially adaptive genetic divergence, by assessing the overlap between the seven gene sets and loci with high F_ST fixation indices between PO and SO. Divergence estimates were obtained with the pool-seq data recently described by Leroy et al. (2020). The SO population was a pool of 13 individuals located in the Laveyron Forest while the PO population was a pool of 20 individuals sampled in the Ychoux Forest located in “Landes of Gascognes” (South West of France). F_ST was calculated with the popoolation2 software suite (Kofler et al., 2011) for each SNP. We then assessed enrichment in high F_ST SNPs among the genes of each of the seven gene sets. We defined high-F_ST SNPs empirically by considering the 1%, 0.5%, 0.1%, 0.01%, and 0.001% right tails of the genome-wide F_ST distribution. We first calculated the proportion of genes overlapping with high-F_ST SNPs for each gene set and each threshold. For each gene set of size N, we then built a null distribution by randomly sampling N genes 1,000 times, each time calculating the proportion of randomly drawn genes overlapping with high-F_ST SNPs. Enrichment was calculated by dividing the proportion of genes from each gene set overlapping with high-F_ST SNPs by the mean proportion of random genes overlapping with high-F_ST SNPs in the corresponding null distribution. We determined the significance of enrichment by comparing the proportion of genes from each gene set overlapping with high-F_ST SNPs from the corresponding null distribution.

Data accessibility statement

All sequences generated in this study were deposited in the Short Read archive of NCBI under PRJEB17875 (Waterlogging experiment) and PRJED19536 (Drought stress experiment) accession numbers.

Accession numbers

Accession numbers of the major genes identified in this study are available in Supplementary Text section 5 for the TAIR database.

Supplemental data

The following materials are available in the online version of this article.

Supplemental Text.

Supplemental Data Set 1. DEGs displaying a significant treatment effect in the waterlogging experiment.

Supplemental Data Set 2. DEGs displaying a significant species effect in the waterlogging experiment.

Supplemental Data Set 3. DEGs displaying a significant treatment-by-species interaction effect in the waterlogging experiment.

Supplemental Data Set 4. DEGs displaying a significant treatment effect in the drought stress experiment.

Supplemental Data Set 5. DEGs displaying a significant species effect in the drought stress experiment.

Supplemental Data Set 6. DEGs displaying a significant treatment-by-species interaction effect in the drought stress experiment.

Supplemental Data Set 7. DEGs displaying a significant species effect whatever the treatment applied.

Supplemental Data Set 8. F_ST value for the DEGs and their associated effect.

Supplemental Figure S1. Schematic representation of the experimental design used in this study.

Supplemental Table S1. Overview of the cDNA libraries constructed in this study.

 

These authors contributed equally (G.L.P. and B.B.)

G.L.P. and C.P. designed the study. G.L.P., I.L., B.B., and C.P. wrote the manuscript. B.B. and T.L. were involved in FST enrichment analysis. I.L. performed bioinformatics analysis with G.L.P. G.L.P. and D.P. were involved with the RT–qPCR experiments. cDNA library construction and sequencing were performed by K.L. and J.M.A. All the authors read and approved the manuscript.

The author responsible for distribution of materials integral to the findings presented in this article in accordance with the policy described in the Instructions for Authors (https://academic.oup.com/plphys/pages/general-instructions) is Gregoire Le Provost (Gregoire.le-provost@inrae.fr).