Neutral transcriptome rewiring promotes quantitative disease resistance evolvability at the species level

Abstract

Quantitative disease resistance (QDR) is an immune response limiting pathogen damage in plants. It involves transcriptomic reprogramming of numerous genes, each having a small contribution to plant immunity. Despite the broad-spectrum nature of QDR, the evolution of its underlying transcriptome reprogramming remains largely uncharacterized. Here, we analyzed global gene expression in response to the necrotrophic fungus Sclerotinia sclerotiorum in 23 Arabidopsis (Arabidopsis thaliana) accessions of diverse origin and contrasting QDR phenotypes. Over half of the species pan-transcriptome displayed local responses to S. sclerotiorum, with global reprogramming patterns incongruent with accession phylogeny. Due to frequent small-amplitude variations, only ∼11% of responsive genes were common across all accessions, defining a core transcriptome enriched in highly responsive genes. Coexpression and correlation analyses showed that QDR phenotypes result from the integration of the expression of numerous genes. Promoter sequence comparisons revealed that variation in DNA-binding sites within cis-regulatory regions contributes to gene expression rewiring. Finally, transcriptome–phenotype maps revealed abundant neutral networks connecting diverse QDR transcriptomes with no loss of resistance, hallmarks of robust and evolvable traits. This navigability associated with regulatory variation in core genes highlights their role in QDR evolvability. This work provides insights into the evolution of complex immune responses, informing models for plant disease dynamics.

The evolution of quantitative disease resistance is supported by transcriptional responses conserved at the species level and variation in DNA-binding sites within cis-regulatory regions.

Introduction

Plants actively respond to pathogen attacks by modulating their physiology, using a range of molecular mechanisms to detect pathogens and defend against them. Gene-for-gene resistance is an immunity mechanism that involves single dominant resistance genes belonging to the nucleotide-binding domain and leucine-rich repeat or receptor-like protein families (Dodds and Rathjen 2010). Quantitative disease resistance (QDR) is another form of plant immunity that limits the damage caused by pathogen infection, which is nearly universal, often durable and broad spectrum (Roux et al. 2014; Corwin and Kliebenstein 2017). QDR involves genes from very diverse families, each having a small effect on resistance (Gou et al. 2023), which makes the identification and characterization of QDR genes difficult. Interestingly, QDR is the only form of plant immunity that reduces disease symptoms caused by many organisms that actively kill host cells for infection, such as the necrotrophic pathogen Sclerotinia sclerotiorum (Niks et al. 2015; Wang et al. 2019). This devastating pathogen is responsible for white and stem mold diseases on a broad range of dicot plants (Bolton et al. 2006) and notably causes severe damage on Brassicaceae, including Arabidopsis thaliana, when conditions are favorable (Derbyshire and Denton-Giles 2016).

QDR involves reprogramming of a large number of genes (Wu et al. 2016; Zhang et al. 2019; Sucher et al. 2020; Pink et al. 2023). Their individual contribution to the resistance phenotype is largely unknown but is likely low, in agreement with an infinitesimal model of complex polygenic traits (Nelson et al. 2013; Boyle et al. 2017). Yet, stress-responsive genes and the transcription factors (TFs) regulating them may play crucial roles in plants adaptation to pathogen attacks (Coolen et al. 2016; Shankar et al. 2016; Groen et al. 2020). Understanding the causes and limits of population divergence in phenotypic traits, particularly how they contribute to adaptive responses to environmental changes, is crucial for understanding the evolutionary dynamics of complex traits like QDR (Opedal et al. 2023). This understanding can help document the diversity and functioning of QDR genetic determinants and shed light on the transcriptome-level processes involved in plant adaptation to pathogens.

Upon pathogen attack, plant lineages display diversified metabolic and transcriptomic responses that converge on essential hubs (Han and Tsuda 2022). Convergent pathways, exemplified by the intricate hormone signaling network, drive the expression of core genes across long evolutionary timescales (Aerts et al. 2021) and buffer metabolic and transcriptomic plasticity of plant defense (Zhang et al. 2017). The evolution of the plant immune system is determined by variation both in coding sequences (CDSs) and in gene expression (Tsuda and Somssich 2015; Monteiro and Nishimura 2018). Local diversity in gene expression pattern across species and populations can be responsible for adaptive evolution (Harrison et al. 2012; Lasky et al. 2014). Several studies compared resistant with susceptible plant genotypes (Westrick et al. 2019; Mwape et al. 2021) or used expression quantitative trait loci (Moscou et al. 2011) to point at genomic loci that control gene expression differences and identify determinants and regulators of S. sclerotiorum resistance. However, studies on global expression diversity in response to pathogens at the species level remain scarce, and the extent to which diverse patterns of transcriptome reprogramming can lead to similar QDR phenotype is unclear. Transcriptional differences within species primarily rely on DNA sequence variations, including single nucleotide polymorphisms, insertions or deletions (indels), and transposable elements (TEs) (Studer et al. 2011; Ando et al. 2018; Soltis et al. 2020). Variation in TF-binding motifs in 5′-regulatory regions modulate transcriptomic defense responses, such as in Brassicaceae where the emergence of WRKY TF-binding motifs associates with specific response to the bacterial molecular pattern flg22 (Winkelmüller et al. 2021). Cis-regulatory variants could be a major source of transcriptome adaptation during plant evolution by acting on TF-binding sites (Filichkin et al. 2011; Yu et al. 2015; Wang et al. 2016), but their contribution to QDR variation at the species level is currently elusive.

Quantitative resistance is often linked to long-term durability, and partial resistance tends to be prevalent in plant populations (Corwin and Kliebenstein 2017). However, the mechanisms driving the emergence, maintenance, and evolutionary dynamics of QDR transcriptome reprogramming remain poorly understood. According to neutral evolution theories, stochastic processes may play a substantial role in shaping transcriptome evolution, particularly in plants (Broadley et al. 2008). One of the key challenges is predicting evolutionary outcomes within this neutral framework, which posits that diverse genotypes can converge to produce similar phenotypes (van Nimwegen et al. 1999). Conserved transcriptome profiles suggest that selective pressures, such as purifying or positive selection, help to maintain beneficial transcriptomic patterns while avoiding deleterious phenotypes (Rifkin et al. 2005; Melo and Marroig 2015; Melo et al. 2016). The plasticity of transcriptome reprogramming, facilitated by mutations in both cis- and trans-regulatory elements, may be a key driver of expression variation within and between species (Zheng et al. 2011; Hill et al. 2021). Understanding the accessible evolutionary pathways and how they are navigated in the context of QDR could provide valuable insights on both natural evolutionary dynamics and potential avenues for directed evolution.

Despite the widespread nature of QDR and its importance for the durable management of plant diseases, our understanding of QDR evolution is limited to knowledge on specific genes or gene families that were associated to this form of immunity (Badet et al. 2017; Barco et al. 2019). The conservation of global QDR responses at the species level offers a unique opportunity to study the recent evolution of transcriptomic patterns underlying complex traits such as QDR. In this paper, we explored the intraspecific variation in A. thaliana transcriptomic responses to S. sclerotiorum across 23 accessions with diverse resistance phenotypes and geographical origins. We identified a set of ∼2,000 genes strongly responsive to S. sclerotiorum inoculation and under purifying selection, thus defining the A. thaliana core transcriptome. Cross-accession comparison of promoter sequences highlighted recent presence/absence polymorphisms in cis-regulatory elements associated with variations in modules of coexpressed genes and genes whose expression correlated with QDR. Finally, we constructed transcriptome–phenotype maps relating global transcriptome reprogramming and QDR phenotype at the species level to explore the evolution of transcriptomic response to a pathogen. We reveal abundant neutral networks connecting diverse transcriptomes with no loss of disease resistance. These landscapes support a role for the core transcriptome in the robustness and evolvability of QDR. These findings advance our understanding of the evolutionary dynamics of complex immune responses at the species level.

Results

QDR phenotype and expression similarity are decorrelated in A. thaliana admixture groups

To explore the evolution of S. sclerotiorum resistance at the species level, we collected 23 natural A. thaliana accessions covering a broad range of genetic diversity, geographic distribution, and climatic origins (Fig. 1A; Supplementary Fig. S1). A phylogenetic tree based on CDS derived from pseudo-genomes identified 8 of the 9 major ADMIXTURE genetic ancestry groups or genetic clusters described in Alonso-Blanco et al. (2016). To relate genetic origin with disease resistance variation at the species level, we evaluated the susceptibility to S. sclerotiorum. Susceptibility ranged from a mean lesion doubling time of 307.1 min (Se-0) to 104.1 (Co-1), representing approximately a 3-fold increase in susceptibility (Fig. 1B). The clusters included accessions with contrasted susceptibility levels, indicating that QDR varied largely independently within the groups. To document A. thaliana species pan-transcriptome diversity, we performed RNA-sequencing in 23 accessions. To assess the global gene expression profile during infection, total mRNA of healthy leaves and leaves colonized by S. sclerotiorum were collected 24 h postinoculation (hpi) and sequenced. Each sample was collected in independent triplicates to allow the detection of accession-specific differentially expressed genes (DEGs). We obtained an average of 27,467,257 mapped reads per sample, with 99% mapping to Araport11 reference transcripts in healthy leaves, and an average of 43% mapping to Araport11 and 57% mapping to the S. sclerotiorum genome in infected leaves (Supplementary Table S3 and Fig. S2). A consistent 60.92% ± 1.72% of complete transcriptomes was expressed across the 23 accessions in healthy plants and 54.71% ± 3.51% in infected plants. To identify S. sclerotiorum–responsive transcripts, we determined DEGs in infected versus healthy leaves for each accession (Supplementary Fig. S3). The number of DEGs ranged from 6,712 to 9,269, with a majority of DEGs downregulated in every accession (upregulated/downregulated ratio of 0.72 ± 0.11, Supplementary Tables S4 and S5). DEGs represented an average 39.56% ± 2.80% of expressed genes in each accession. Overall, 17,340 genes corresponding to 52.36% of A. thaliana transcriptome were differentially expressed in at least in one accession. Comparison between phylogenetic tree of accessions and transcriptome similarity dendrogram did not show clear congruence (Supplementary Fig. S3). To estimate the size of S. sclerotiorum–responsive transcriptome at the species level, we performed random accession sampling and logarithmic regression fit. This predicted a maximum of 22,643 DEGs (68.4% of the predicted transcriptome) across the whole A. thaliana diversity with R²=0.947 (Supplementary Fig. S4). Across accessions, 51.5% DEGs showed a maximum log fold change (LFC) variation <2, and 2.6% had a maximum LFC variation >10 (Supplementary Fig. S5), indicating that drastic changes of expression profiles across accessions are limited to a few genes (Fig. 1C). To assess the diversity of transcriptional responses to S. sclerotiorum at the species level, we compared the expression profile of the 17,340 DEGs across 23 accessions (Fig. 2). DEGs with a differential expression in one accession only (accessory) ranged from 36 to 292 per accession, making a total 2,202 accessory DEGs (12.7%) at the species level. There were 13,181 DEGs (76%) differential in 2 to 22 accessions (shell). We identified 1,957 DEGs (11.3%) differential in all accessions (core) including 1,049 upregulated core DEGs and 908 downregulated core DEGs. Among the 17,340 DEGs, 1221 (7.0%) showed opposite regulation patterns (upregulated vs. downregulated) in at least 2 accessions. The number of DEGs with opposite regulation decreased from 13% for DEGs differential in 2 accessions to 0% in core DEGs (Fig. 2).

Figure 1.

Global differential gene expression patterns and phylogeny do not correlate with S. sclerotiorum resistance across A. thaliana accessions. A) Phylogenetic tree of the 23 accessions used in this work based on CDSs. Rectangles indicate the 8 main ADMIXTURE groups, and the ADMIXTURE group for Mt-0 was not determined. Numbers along branches represent confidence values based on the UFBoot2 bootstrap method with 1,000 bootstrap iterations. The scale bar represents genetic distances, with 0.001 substitutions per site. B) Lesion doubling time (LDT) in minutes representing susceptibility of accessions at 24 h after inoculation with S. sclerotiorum, plot in log scale. Measurement were obtained from n = 28 to 112 leaves in 3 experiments, with the black horizontal bar representing the mean speed. Kernels are colored according to mean susceptibility, from the most resistant (blue) accessions to the most susceptible (red). C) Heatmap displaying the distribution of the 1,000 DEGs upon S. sclerotiorum inoculation with the highest coefficient of variation across accessions. Accessions are ordered based on genetic proximity in all panels.

Figure 2.

A total of 17,340 genes are differentially expressed in at least one accession. A) Number of DEGs according to the number of accessions in which they are differential. The graph shows DEGs upregulated in all accessions where they are differential (red), DEGs downregulated in all accessions where they are differential (blue), DEGs upregulated in at least one accession and downregulated in at least one accession (purple), and the total number of DEGs (gray). B) Synthetic representation of DEG classes, with the number of DEGs per class between brackets.

These results show that S. sclerotiorum infection triggers reprogramming of a majority of A. thaliana species pan-transcriptome, with moderate expression diversity across accessions and no clear link between the phylogenetic proximity of accessions and their transcriptomic response to S. sclerotiorum.

A small set of strongly responsive genes defines A. thaliana core response to S. sclerotiorum

To characterize QDR responses conserved at the species level, we examined the expression, functional annotation, and conservation of core DEGs. Core DEGs were consistently upregulated or downregulated in all 23 accessions and showed a high expression stability, with 79% of core DEGs having an LFC standard deviation below 2 (Fig. 3A). Upregulated/downregulated DEGs ratio was 1.16 in core DEGs, 0.76 in accessory DEGs, and 0.60 in shell DEGs, suggesting the accumulation of genes upregulated upon S. sclerotiorum inoculation during A. thaliana evolution. Using regression analysis, we estimated a minimum core DEGs set of 719 upregulated and 352 downregulated genes across the whole A. thaliana diversity (Supplementary Fig. S6). The mean LFC values per gene across accessions show that core upregulated and downregulated DEGs are among the most highly regulated genes in response to S. sclerotiorum. Of the top 500 DEGs with the highest mean LFC values, 72% are core upregulated DEGs (6.54-fold enrichment compared with all upregulated genes), while 45.5% of the top 500 downregulated DEGs are core downregulated (6.78-fold enrichment compared with all downregulated genes) (Fig. 3B). To determine the functional landscape of core DEGs, we performed a Gene Ontology (GO) enrichment analysis. GOs corresponding to 254 biological processes were significantly enriched among the upregulated core DEGs (Fig. 3C). The majority of these GOs corresponded to ontologies related to primary metabolism, secondary metabolism, molecules transport, and defense responses. Precisely, GOs related to secondary metabolism included camalexin biosynthetic process (GO:0010120), chorismate metabolic process (GO:0046417), and lignin biosynthetic process (GO:0009809) and GOs related to defense responses included pattern recognition receptor signaling pathway (GO:0002221), response to ethylene (GO:0009723), and jasmonic acid–mediated signaling pathway (GO:0009867). Core upregulated DEGs included known defense related genes such as ABCG40, CERK1, PAD3, RBOHD, and RLP30 (Stotz et al. 2011; Zhang et al. 2013; Zhou et al. 2013; Sucher et al. 2020; Pi et al. 2023). GOs corresponding to 154 biological processes were significantly enriched among the downregulated core DEGs (Fig. 3D), including responses to light, developmental responses, or photosynthesis. To study the evolutionary dynamics of core DEGs, we performed a phylostratigraphic analysis using Phytozome orthogroups. A total of 1,039 upregulated core DEGs and 899 downregulated core DEGs were included in Phytozome orthogroups. Core S. sclerotiorum–responsive genes showed a clear gene age pattern with dominant old genes (Viridiplantae and Embryophyta) and young genes (Brassicaceae and A. thaliana) to a lesser extent, creating skewed U-shaped distributions (Fig. 3E). These distributions are similar to those of all A. thaliana genes, indicating that A. thaliana transcriptome is predominantly composed of old and young genes. Compared with all A. thaliana genes, core DEGs included significantly fewer young genes and more old genes (+17.6% for upregulated core DEGs and +21.6% for downregulated core DEGs). Relative to A. thaliana whole genome, accessory DEGs included fewer old genes (−1.7%), while shell DEGs were enriched in old genes (+4.2%). These patterns suggest that intraspecific expression variation is better tolerated in young genes. To explore the relationship between transcriptome conservation and genetic diversity, we calculated nucleotide diversity (π and θ) in the 23 A. thaliana accessions. Core DEGs exhibited significantly lower θ nucleotide diversity compared with shell DEGs and accessory DEGs (median: −14.1% and −13.9%, respectively) (Fig. 3F). The median nonsynonymous to synonymous nucleotide diversity ratio (π_N/π_S ratio) was 0.74 in core DEGs, 0.93 in shell DEGs, 1.03 in accessory DEGs, and 0.82 in genes expressed during infection not differential, indicating that most core DEGs are under purifying selection compared to shell DEGs and accessory DEGs (Fig. 3G). No bias due to genomic position was identified for these metrics (Supplementary Figs. S7 and S8). Therefore, conservation of gene expression patterns is linked to selective forces reducing nucleotide diversity and nonsynonymous mutations. We conclude that purifying selection at the species level drove the long-term sequence and function conservation, high and robust expression of A. thaliana core response genes to S. sclerotiorum.

Figure 3.

A total of 1,957 core DEGs are differentially expressed in all 23 A. thaliana accessions upon S. sclerotiorum inoculation. A) Heatmap displaying the distribution of 1,957 core DEGs across the 23 accessions after Sclerotinia inoculation, with an absolute |LFC| ≥ 2. Red panels correspond to 1.049 genes significantly upregulated, and blue panels correspond to 908 downregulated genes. B) Mean LFC by gene across the 23 accessions. Red bars correspond to 1.049 core upregulated DEGs, and blue panels correspond to 908 core downregulated DEGs. GO enrichment analysis of the 1,049 core upregulated DEGs C) and the 908 core downregulated DEGs D), plotted using the BINGO module from the Cytoscape software. Enrichment was calculated using BINGO and visualized on the GO hierarchy using Cytoscape. Circle size represents the number of genes in one GO category. E) Proportion of DEGs according to gene age. Gene ancestry information was retrieved from the Phytozome database. Light purple corresponds to core DEGs (upregulated and downregulated), dark gray to accessory DEGs, gray to shell DEGs, and light gray to the total genes in the Arabidopsis Phytozome database. Statistical analyses were conducted using the chi-squared test. Asterisks indicate the following P-values: *P < 0.05, **P < 0.01, ***P < 0.001, and ns = not significant. F and G) Genetic population metrics in core DEGs (purple), shell DEGs (gray), accessory DEGs (dark gray), and genes expressed not differential (light gray). F) Nucleotide diversity (θ) and G) the nonsynonymous to synonymous nucleotide diversity ratio (π_N/π_S ratio), with the y axis scale plotted logarithmically. Letters identify significantly different gene groups. The mean difference between groups was estimated using 10,000 bootstrap replicates of randomly sampled subsets of the same size, and a 95% confidence interval was calculated. One representative bootstrap replicate from the 10,000 replicates is shown in F and G). The center line of the boxplots represents the median, the box limits correspond to the upper and lower quartiles, and the whiskers extend to 1.5 times the interquartile range. Outliers are shown as points outside this range.

Transcriptome-wide coexpression clustering reveals the rewiring of gene modules across accessions

To compare the diversity of transcriptomic responses to S. sclerotiorum between accessions, we used LFC values for all expressed genes to correlate transcriptomes across accessions. Pearson correlation was calculated between each transcriptome to assess the overall proximity between accessions. Based on the similarity of global gene expression patterns, we identified subsets of accessions that were largely consistent with the DEG-based dendrogram (Fig. 4A; Supplementary Fig. S2). We found no evidence for a link between correlation matrix clustering and susceptibility to S. sclerotiorum, or with environmental parameters at the geographic origin of the accessions (Supplementary Table S6). Additionally, we observed no significant correlations between transcriptome subsets and phylogenetic clusters, indicating the absence of a phylogenetic signal in defining transcriptome proximity among accessions.

Figure 4.

Identification of 48 gene coexpression modules in the response of A. thaliana accessions to inoculation by S. sclerotiorum.  A) Pearson correlation matrix of the 23 accessions based on global transcriptome profiles upon S. sclerotiorum inoculation. Color circles indicate susceptibility to S. sclerotiorum for the 23 accessions according to Fig. 1B. B) UMAP of TOM matrix obtained with the WGCNA coexpression analysis. Each point represents a single expressed gene, and each color represents a distinct coexpression module. The dotted lines illustrate the accession-specific module diverging from other gene coexpression modules. Genes belonging to accession-specific modules are enclosed by black circles. C) Heatmap showing the expression profile of genes associated with the 4 largest coexpression modules across 23 accessions. Green indicates LFC > 0, and purple indicates LFC < 0. D) DAP cis-motif enrichment comparing 500 bp promoter regions of genes from coexpression Modules 1–4 in the 2 major accession groups defined by the first split of the accession dendrogram. P-value corresponds to enrichment tests obtained with SEA from the MEME Suite, comparing the promoter genes from the smallest accession groups against the larger accession group described by the heatmap dendrogram in C), with TF corresponding to transcriptome binding of the DAP-seq motif.

To further investigate transcriptome similarity in response to S. sclerotiorum across accessions, we performed transcriptome-wide coexpression module detection. Weighted correlation network analysis (WGCNA) identified 48 coexpression modules, comprising between 3 and 2,826 genes each, with a median of 314.15 genes per module (Fig. 4B). Based on accession clustering, 25 coexpression modules were predominantly influenced by gene expression changes across multiple accessions, whereas 23 modules were driven by major gene expression changes specific to a single accession (Supplementary Fig. S9). Transcriptome similarity between accessions was not consistent across gene modules as illustrated by gene expression patterns from the 4 major modules (Fig. 4C). For instance, the Spearman correlation between the accession dendrogram for Module 1 and Modules 2, 3, and 4 was weak, with values of 0.22, 0.04, and 0.33, respectively (Supplementary Fig. S10). GO enrichment analysis in the 4 major modules identified GOs related to response to stress in Module 1, hormone signaling and primary metabolism in Module 2, development and gene expression regulation in Module 3, and immune responses in Module 4 (Supplementary Fig. S11). This highlights the combination of diverse processes with contrasted quantitative contribution to the QDR response of each accession.

Enrichment in gene expression regulation ontologies in Module 3 suggests that variation in cis may contribute to distinct QDR responses across A. thaliana accessions. To get support for cis-regulation variation into the diversity of A. thaliana QDR responses, we analyzed the distribution of DNA affinity purification (DAP) cis-motifs in the promoter of genes from each module. We compared motifs in the promoter regions of all genes within each coexpression module, using accessions grouped into 2 major similarity clusters as defined by the first split of the accession dendrogram for each module (Supplementary Table S7). DNA-binding sites for the TFs AT5G66940, MYB81, RAV1, and MYB65 were differentially enriched in Module 1 genes (Fig. 4D). A few motifs were differentially enriched in genes from several modules, indicated no strong signal of common DNA-binding site variation across coexpression modules. But some motifs like MYB81, MYB96, and ERF3 motifs were uniquely enriched in genes from Modules 1, 3 and 4, respectively.

We conclude that regulatory variation targeting modules of shell DEGs, partly driven by cis-motif divergence between accessions, underlie diversity of the QDR transcriptome at the species level.

Correlation between gene expression and disease susceptibility is marginal at the species level

To relate gene expression with disease susceptibility at the species level, we tested for correlation between gene expression and accessions susceptibility. First, we tested the correspondence between distance matrices for accession phenotypes and accession transcriptome similarity for each of the coexpression modules identified previously. Seven coexpression modules significantly explained the variation in resistance at the species level: 5 exhibited a positive correlation (Modules 36, 25, 26, 4, and 8) and 3 showed a negative correlation (Modules 20, 31, and 37) (Fig. 5A). However, significant correlation values only ranged from 0.15 to 0.21 for positively correlated modules and −0.12 to −0.15 for negatively correlated modules. Except for coexpression Modules 4 and 25, these correlations were driven by gene expression changes in a single accession (Rub, Co-1, Se-0, Wa-1, Ws-0, or Oy-0) (Supplementary Fig. S9). Therefore, variations in disease susceptibility at the species relate to the regulation of coexpression gene modules only in a limited number of accessions.

Figure 5.

Analysis of species-wide correlation between gene expression and disease resistance phenotypes. A) Spearman correlation between dendrogram-derived distance matrices from gene expression clustering and phenotypic distance matrices for coexpression modules identified in Fig. 3. P-values were calculated using a Spearman permutation test (n = 9,999). Coexpression modules with P-values <0.05 are highlighted in red. B) Correlation between expression profiles obtained by RNA-seq and disease resistance in 23 accessions for all expressed genes. Blue dots represent genes with |R²| > 0.4 and RMSE < 0.05. The red dotted line indicates RMSE = 0.05. C) Correlation between LFC (x axis) and disease resistance (y axis, plotted on a logarithmic scale) in 23 accessions for the 32 best correlated genes. For each gene, the 21 of 23 LFC values with the lowest FDR were retained. Genes with |R²| > 0.4 and RMSE < 0.05 are presented in the figure. The blue lines show linear regression of the data with R² and P-value labeled, and the 95% confidence interval shown as a gray area. D) DAP cis-regulatory motifs with importance scores identified by the Boruta package as significant contributors to LFC variation across 23 accessions in the 32 best correlated genes.

Next, we tested if the expression of individual genes associated with QDR across accessions (Fig. 5; Supplementary Fig. S12). For this, we used expression LFC for all expressed genes and disease resistance phenotype for correlation and regression analysis. With a |R²| > 0.4 and rmse < 0.05, the expression of 32 genes was significantly correlated with resistance variation, none of which had been previously associated with disease resistance to S. sclerotiorum (Fig. 5, B and C). To identify genes associated with disease resistance or susceptibility, we described the overall expression pattern of the genes by calculating the mean LFC per gene across accessions and analyzing it in relation to the corresponding R² value. Among those, 22 genes had an expression correlated with disease resistance (i.e. R² < 0 and mean LFC > 0 or R² > 0 and mean LFC < 0), such as genes encoding the TPR9 tetratricopeptide repeat protein and the AT1G56660 MAEBL protein (Fig. 5C). Ten genes had an expression correlated with disease susceptibility (i.e. R² > 0 and mean LFC > 0 or R² < 0 and mean LFC < 0), including gene encoding the AT2G40820 actin binding-like protein and the AT5G01175 microtubule controlling protein (Fig. 5C). Among the 32 QDR-associated genes, 22 were shell DEGs differentially expressed in 2 to 22 accessions with an average |LFC| from 0.51 to 5.31. Ten genes had |LFC| ≤ 2 in all 23 accessions. Two mutants for 2 of these genes showed a significant alteration in their response to S. sclerotiorum (Supplementary Fig. S13). This approach suggests that only a few shell DEGs were associated with QDR variation at the species level, potentially due to the limited contribution of individual genes or complex genetic interactions.

To investigate the regulatory mechanisms driving transcriptome variation at the species level and the acquisition of expression profiles of the 32 QDR-associated genes, we conducted a comparative analysis of the cis-motif content within the 500-bp promoter regions of these genes across 23 accessions. We identified 497 cis-motifs that showed at least one presence/absence polymorphism in the 500-bp promoter region the 32 QDR-associated genes. Next, we used a random forest approach to identify a set of 46 cis-motifs whose presence/absence polymorphism in promoter regions robustly associated with LFC variation in at least one of the 32 genes (Fig. 5D; Supplementary Table S8). Specifically, the ATHB21 and SGR5 cis-motifs were linked to LFC variation in 3 genes (AT4G09490, AT4G24930, and AT4G24930 or AT2G02510, AT2G39280, and AT4G09600, respectively), the 4 cis-motifs AGL25, AT1G20910, AT5G60130, and REM19 were linked to LFC variation in 2 genes (Supplementary Table S8), while 40 cis-motifs found in relation to LFC variation in a single gene. Notably, the impact of these cis-motifs varied between genes, indicating a lack of consensus in cis-motif acquisition at the species level.

Core genes make the transcriptome–phenotype map of A. thaliana QDR navigable

To get insights into how global gene expression constrains the evolution of QDR at the species level, we constructed a transcriptome–phenotype map of A. thaliana QDR against S. sclerotiorum. For this, we clustered the 23 accessions based on the LFC of all expressed genes using principal component analysis (PCA) and used Gaussian interpolation to connect experimentally determined disease resistance phenotypes by a smooth surface (Fig. 6, A and B). This map describes the expected QDR phenotype for every point of the transcriptome space covered by our sampling of accessions. It is composed of 2 major sectors: a susceptibility basin connecting accessions Co-1, Ang-0, Mt-0, Kyoto, Eds-1, Gy-0, Bla-1, Bor-1, Nok-3, and Hs-0 and a resistance plateau connecting accessions Es-0, Ct-1, Van-0, Col-0, Ws-0, Wa-1, Ra-1, Ei-2, Rub, Oy-0, and Ll-0. Accessions Se-0 and Rub form 2 resistance peaks at 2 sides of the map, indicating that divergent transcriptomes can produce similar levels of disease resistance. Within this map, accessions distant in transcriptome space can be connected through similar phenotypes. This defines the equivalent of neutral networks of accessions connecting diverse transcriptomes with a similar fitness, approximated here by resistance to S. sclerotiorum. To estimate the differential regulation required for movement in transcriptome space, we assessed transcriptome reprogramming changes between accessions by evaluating differences in DEGs through pairwise comparisons (Supplementary Fig. S14). An increase in accession distance on the transcriptome–phenotype map (in the x–y plane) correlates with an increase in the number of unique DEGs in pairwise comparisons between accessions.

Figure 6.

Navigability within the transcriptome-resistance map based only on core DEGs is correlated with susceptibility. A) Transcriptome-resistance map showing accession proximity in the transcriptome space in response to S. sclerotiorum inoculation (x–y plane) and resistance to S. sclerotiorum (z axis, color scale). The x and y axes represent Components 1 and 2 of a PCA using LFC data for all expressed genes. The color gradient represents disease resistance as indicated by the lesion doubling time (LDT) in minutes, from more susceptible (blue) to more resistant (red), with experimentally studied accessions shown as small labeled spheres. Lines of equal resistance (white) are provided for legibility. B) Top view of the transcriptome-resistance map shown in A). C) Top projections of the transcriptome–phenotype map for all expressed genes (left) and core DEGs only (right) showing sectors of the map accessible to Bor-1 accession without loss of resistance (navigability) delineated by green borders. D) Relationship between navigability of the transcriptome-resistance map obtained for all expressed genes (y axis) and navigability of the map obtained for shell DEGs (x axis, left) and core DEGs (x axis, right) in 15 A. thaliana accessions. E) Relationship between resistance against S. sclerotiorum (y axis), represented by lesion doubling time (LDT) in minutes, and navigability of the transcriptome-resistance map obtained with shell DEGs (x axis, left) and core DEGs (x axis, right) in 15 A. thaliana accessions. The black lines show linear regression of the data with R² and P-value labeled, and the 95% confidence interval shown as a gray area.

Genotype variants connected only through fitness increase define the navigability of fitness landscapes. We calculated the QDR landscape navigability for each accession, defined as the percentage of the transcriptome map accessible with no loss of resistance. Navigability values ranged from 69.14% (Eds-1) to 0.02% (Se-0) (Supplementary Table S9). The resistance maxima observed in our study for the Se-0 and Rub genotypes can be reached by 14 genotypes without passing through susceptible valleys, due to the large intermediate resistance plateau (Fig. 6B). This suggests that transcriptomic responses to S. sclerotiorum are highly evolvable in A. thaliana, tolerating significant changes with no resistance loss.

To evaluate the contribution of core and shell DEGs to A. thaliana QDR evolvability through potential regulatory adjustments, we generated transcriptome–phenotype maps based solely on the corresponding DEG sets. We calculated the QDR landscape navigability for each accession on the core DEGs map and all expressed genes map (Supplementary Table S9). QDR landscape navigability increased in 14 accessions when using transcriptome–phenotype maps based solely on the expression of core DEGs, compared with maps using all expressed genes. For example, the Bor-1 accession shows a navigability of 30.4% with the core DEGs-only map, compared with 15.6% with the all-expressed-genes map (Fig. 6C; Supplementary Fig. S15). Navigability on the whole transcriptome map correlated with navigability on the shell DEG map (R² = 0.63, P-value = 0.00038) but not with the navigability on the core DEG map (R² = 0.02, P-value = 0.58), consistent with the prevalent contribution of shell DEGs to the overall response to S. sclerotiorum (Fig. 6D). Resistance to S. sclerotiorum correlated with navigability on the core DEG map (R² = 0.83, P-value = 2.6 e−6) but not with the navigability on the whole transcriptome map (R² = 0.01, P-value = 0.77) and the shell DEG map (R² = 0.13, P-value = 0.19) (Fig. 6E) indicating that QDR evolvability is related to regulatory variation in core DEGs in this fitness landscape.

Discussion

QDR provides partial and durable protection against various pathogenic microbes across the green lineage. Genetic mechanisms associated with the short- and long-term evolution of QDR remain largely obscure (Kou and Wang 2010). In a previous study, comparative transcriptome analysis across plant species highlighted exaptation through regulatory divergence as a path toward the evolution of QDR against S. sclerotiorum (Sucher et al. 2020). Here, we analyzed patterns of QDR transcriptome evolution at the species level through a comprehensive RNA-seq analysis of 23 A. thaliana accessions displaying various degrees of resistance to S. sclerotiorum.

How complex is plant QDR?

The omnigenic model of complex traits proposes that most genes expressed in an infection context should contribute to the heritability of disease resistance, either through a role in immunity or in its regulation (Boyle et al. 2017). Interspecific comparison found an average 32.63% genes responsive to S. sclerotiorum across Pentapetalae, an estimate resulting from the analysis of a single genotype of each species (Sucher et al. 2020). Here, we assessed infra-specific variation in A. thaliana gene responsiveness to S. sclerotiorum. In each accession, DEGs represented an average 39.6% of expressed genes, together accounting for 52.4% of A. thaliana pan-transcriptome responsive to S. sclerotiorum infection. This number is consistent for instance with 50.4% of A. thaliana genes differentially regulated upon Fusarium oxysporum inoculation (Guo et al. 2021). We have chosen to map RNA-seq reads to the Araport11 reference genome for a more accurate quantification of gene expression. This approach does not allow to identify accession-specific transcripts that are absent from the reference genome and probably slightly underestimates the overall diversity of S. sclerotiorum responsive transcripts in A. thaliana.

How predictable is QDR?

There was no detectable phylogenetic signal associated with global transcriptome patterns and resistance phenotype in our 23 A. thaliana accession panel, indicating that the heritability of these traits at the species level is complex. Previous analyses reported that transcriptome variation tends to recapitulate evolutionary relationships accurately at the interspecies level but loses precision for closely related species (Winkelmüller et al. 2021; Wu et al. 2022). This may be due to transcriptional noise being prevalent in a subset of QDR genes or to the independent evolution of sequence and regulatory polymorphisms at QDR loci (Tirosh and Barkai 2008; Harrison et al. 2012; Uebbing et al. 2016). Contrasted contributions of regulatory divergence and sequence variation to the evolution of quantitative plant immunity may indeed result in independence between these polymorphisms at the species level. QDR phenotypes may therefore be difficult to predict from sequence information alone (Gentzbittel et al. 2019). By training machine learning models on plant transcriptomes, Sia et al. (2023) found that information on expression of only 0.5% of genes was sufficient to predict classes of disease phenotypes across multiple pathosystems. Here, we identified 32 genes whose expression correlated with QDR phenotypes at the species level. They have not been associated with disease resistance to date but may encode putative signaling proteins controlling known defense pathways and could potentially be used as predictors of QDR phenotypes at the species level. In addition to transcriptome regulation, constitutive mechanisms may also play a role in QDR, such as cell wall composition (Molina et al. 2021). To further investigate the role of constitutive gene expression in predicting QDR, we also examined the correlation between gene expression during S. sclerotiorum infection and QDR phenotypes. This analysis revealed 7 genes, not previously linked to plant immunity, whose expression patterns showed significant correlation with QDR phenotypes (Supplementary Fig. S12). Interestingly, while gene expression in infected conditions alone does not provide a reliable means of predicting QDR phenotypes, the coordinated evolution of gene expression at the subspecies level could contribute to phenotypic adaptation (Mezey et al. 2008) and impede phenotypic predictions unless the genetic diversity of the species has been investigated with sufficient depth (Crossa et al. 2010).

What is the contribution of the variable transcriptome to A. thaliana QDR?

Regulatory divergence across A. thaliana accessions was generally of low magnitude, questioning the adaptive value of this diversity. By contrast to tightly regulated core genes, shell and accessory genes identified in our work may exhibit a high level of transcriptional noise, possibly underlying a bet-hedging strategy to improve immunity (Viney and Reece 2013; Urban and Johnston 2018). In such a strategy, noise may lead to stochastic activation of diversified immune responses in cell subsets. This local immune response may be sufficient to limit pathogen colonization or may propagate through cell to cell communication. Over long evolutionary time, the plasticity of gene expression may also facilitate the acquisition of precise temporal and spatial expression patterns underlying QDR (Groen et al. 2020; Jones and Vandepoele 2020). The recent advent of single-cell RNA-sequencing (Tang et al. 2023) will allow testing cell population immunity mechanisms that were not accessible to bulk RNA analyses. The existence of 32 genes whose expression correlated with QDR at the species level suggests that minor transcriptional variations in key regulators may play a significant role in phenotypic evolution due to their interactions within the broader gene networks (Liu et al. 2019; Jenull et al. 2021). To initiate the characterization of these genes, we shown that 2 mutants corresponding to 2 of the 32 QDR-associated genes showed alterations to their response to S. sclerotiorum (Supplementary Fig. S13). Further investigation of these genes, alongside exploring the regulatory mechanisms and functions of genes whose expression varies across species, presents an opportunity to identify additional contributors to QDR.

What is the contribution of core DEGs to A. thaliana QDR phenotype?

We identified 1,049 core upregulated Among S. sclerotiorum–responsive genes in A. thaliana in all accessions, suggesting either a role in mediating resistance across various genetic backgrounds (Sucher et al. 2020; Yang et al. 2021) or broad-spectrum manipulation by the fungus. GO analysis associated core upregulated DEGs with general responses to fungal pathogens, such as the biosynthesis of camalexin, chorismate metabolism, and lignin biosynthesis (Zhu et al. 2013; Zhang et al. 2017). These processes, conserved at the species level, typically contribute to a broad-spectrum defense against fungal pathogens, including Botrytis cinerea, Colletotrichum higginsianum, and Phytophthora parasitica (La Berre et al. 2017; Soltis et al. 2020; Zhu et al. 2023). Core DEGs form a backbone network where fine-tunable transcriptome modules play a pivotal role in responding to pathogen genetic diversity (Kim et al. 2014; Hillmer et al. 2017). Mutants pad3-1, abcg40-2, and rlp30-1 altered in core DEGs show increased susceptibility to S. sclerotiorum, indicating clear contribution of some core DEGs to QDR (Zhang et al. 2013; Sucher et al. 2020; Kusch et al. 2022). However, the relative contribution of many core DEGs to disease resistance remains to be tested, and why the global pool of core DEGs contributes to different levels of resistance across accessions is unclear. The contribution of core genes to QDR may be regulated at the posttranscriptional level in a contrasted manner across accessions. Alternatively, specific core transcript ratio may be required for efficient QDR, or the precise timing of core transcripts activation may be crucial for QDR, emphasizing the dynamic role of transcriptional reprogramming as a major contributor to fungal resistance (Niks et al. 2015; French et al. 2016; Yang et al. 2021).

How robust and evolvable is QDR in A. thaliana?

A fundamental question in biology is how genotypes influence the phenotypes and functions of organisms, which in turn impact their evolutionary success. The shape of genotype fitness maps strongly influences the trajectory of evolution and our understanding of genetic robustness and evolvability (Visser and Krug 2014). Because of missing data and lack of information on the direction of evolution, previous attempts at estimating a fitness landscape map for complex traits extrapolated experimental data using averaging, discretization, and Gaussian kernel in the principal component space (Lorenzo-Redondo et al. 2014; Iwasawa et al. 2022). With a similar approach, our analysis emphasizes redundancy in the transcriptome–phenotype map, where multiple global transcriptomes map onto the same level of disease resistance to S. sclerotiorum, defining neutral networks (Greenbury et al. 2022). Neutral networks intrinsically support phenotypic robustness to (epi)genetic interference. Our fitness landscape model predicts that 15 accessions can navigate >1% of the transcriptome space with no loss of resistance. Most of the accessions tested had intermediate resistance allowing strong navigability, in agreement with QDR being predominant in natural populations (Corwin and Kliebenstein 2017). High transcriptome navigability could be a crucial component for the durability of QDR, by preventing extreme fitness valleys or peaks that limit transcriptome flexibility. A corollary of this finding is that globally similar transcriptomes can produce contrasted phenotypes (Bonnet et al. 2017). This may be explained by a prevalent role of regulation at the posttranscriptional level in QDR, although the massive transcriptional reprogramming induced by S. sclerotiorum inoculation argues against this hypothesis. Alternatively, the QDR response network may be conserved and redundant, or structured to canalize similar downstream responses in spite of variable inputs (different pathogen genotypes, environmental conditions, or plant genetic background) (Zhang et al. 2017). Neutral networks also allow populations to access a wide variety of transcriptomes and novel phenotypes without intermediate fitness penalty, promoting evolvability (Greenbury et al. 2016). Broad neutral networks of QDR transcriptome support high evolvability of this trait at the species level. It should be noted that our focus on a limited sampling of 23 accessions may underestimate the full range of the navigable transcriptome space or conversely have missed fitness valleys that limit navigability. Besides, our transcriptome–phenotype maps only represent the 2 principal components of the transcriptome space, and navigability may vary when considering additional dimensions of this space (Zagorski et al. 2016).

How did the regulation of fungal-responsive genes evolve?

Variation in transcriptome may be facilitated by regulatory elements or epigenetic mechanisms that alter gene expression even before genetic variants arise in the population (Alvarez et al. 2015). The acquisition of a new expression pattern in a limited number of genes may then significantly alter QDR phenotypes (Chen et al. 2013). The cross-accession comparison of DAP motifs in the promoter of genes forming coexpression modules and genes whose expression correlated with QDR identified presence/absence polymorphisms associated with regulatory variation. This variation often targeted a few accessions only. Our analyses point toward the acquisition of WRKY binding motifs, known for their involvement in transcriptional regulation (Wani et al. 2021) as a potential mechanism for the stabilization of core gene expression at the species level. When comparing the promoter sequences of 32 QDR-associated gene between the 23 accessions, we identified DAP motifs whose presence or absence is significantly associated with 21 gene expression patterns, such as the cis DNA-binding motif recognized by the TF WRKY7 for the QDR-associated gene AT4G09490. Most of the variation in cis-motifs is unique to each QDR-associated gene, suggesting a possible recent reconfiguration of neutral transcriptome by accessions through mutations in the promoter regions (van Kooten et al. 2021). Given the unique DAP motifs enriched specifically in QDR-responsive genes of each accession, we propose that the gain or loss of DNA TF-binding motifs contributes significantly to transcriptional variation at the species level. Contrary to our expectations, we did not find TE families significantly associated with QDR transcriptome modulation at the species level. Cis-regulatory variations arise from the emergence of novel enhancer sequences (Long et al. 2016) or from genome rearrangements that recruit additional genes to transcriptionally active regions. Cis-regulatory variations may also serve as a source for generating allele-specific transcripts involved in QDR (Fukuoka et al. 2014; Yang et al. 2021). Novel enhancer activities often result from the cooption of TF-binding sites already present in ancestral enhancers (Wittkopp and Kalay 2012; Macquet et al. 2022). The acquisition of cis-regulatory enhancers in promoter regions is known to drive the evolution of gene expression, shaping transcriptome patterns related to plant resistance (Ogasahara et al. 2022). The divergence of cis-regulatory motifs is suggested to be a key contributor to the rapid transcriptional reprogramming essential for the broad-spectrum efficiency of plant immune responses (Mine et al. 2018). Consequently, variation in cis-regulatory elements across genotypes depends on various selective forces, including abiotic and biotic environmental factors (Ricci et al. 2019; Marand et al. 2023).

Which properties of the QDR transcriptome space promote robustness and evolvability?

The accumulation of multiple and independent mutations in the promoter region may lead to the rewiring of a neutral transcriptome (Greenbury et al. 2022), supporting high evolvability dynamics. Subtle differences in evolvability could play an important role in shaping the long-term success of transcriptomic variants (Ferrare and Good 2024). The navigability of our core DEG transcriptome–phenotype map correlated with accession susceptibility, suggesting that purifying selection stabilized core DEGs expression to serve as a network backbone for QDR phenotype and genetic architecture to evolve. On the other hand, lineage-specific genes often participate in adaptation to biotic and abiotic stress (Tian et al. 2012). Thus, genes from the most recent age classes may contribute relatively directly to QDR, including a few with expression directly correlated to the QDR phenotype. The evolution of robust QDR phenotypes at the species level may involve concerted regulatory variation between the stable and reduced set of core genes and the extensive and plastic pathogen-responsive transcriptome, thus indicating the need to study evolution of the defense network as a whole (Kahlon and Stam 2021). Future research should evaluate the impacts of environmental constraints on these evolutionary dynamics to explore both past and future trajectories of QDR evolution in changing environments.

Materials and methods

Plant material and RNA sampling

A. thaliana accessions listed in Supplementary Table S1 were obtained from The Versailles Arabidopsis Stock Center. General information regarding collection location and climate data of these accessions was retrieved from https://1001genomes.org and https://www.worldclim.org. ADMIXTURE genetic groups corresponding to the 23 accessions were recovered from Alonso-Blanco et al. (2016). Arabidopsis T-DNA mutant lines listed in Supplementary Fig. S13 were obtained from the Nottingham Arabidopsis Stock Centre; Homozygous T-DNA insertions were verified by PCR performed with primers given in Supplementary Fig. S13C. Plants were grown in Jiffy pots under controlled conditions at 22 °C with a 9-h light period at an intensity of 120 μmol/m²/s. Five-week-old plants were inoculated with S. sclerotiorum strain 1980 (ATCC18683) on 3 leaves using 0.5-cm-wide plugs of potato dextrose agar (PDA) medium colonized by the fungus. The agar plugs containing the fungal pathogen were placed on the adaxial surface of leaves, and plants were incubated in trays sealed with plastic wrap to maintain 80% relative humidity. Twenty-four hours after inoculation, leaf rings including the periphery of disease lesions were collected as in Peyraud et al. (2019). Material obtained from 3 leaves per plant were pooled as one sample, and samples were collected in 3 to 5 replicates. Leaves from noninoculated plants under the same conditions were collected as control at the same time.

RNA extraction, sequencing, and mapping

Samples were ground using glass beads (2.5 mm), and RNA was extracted using NucleoSpin RNA PLUS extraction kits (Macherey-Nagel) following the manufacturer's protocols. The quality of the extracted RNA was assessed using Agilent bioanalyzer nanochips. Libraries preparation and mRNA sequencing were conducted at the GeT-Plage facility (INRAE Castanet Tolosan, France) using the Illumina TrueSeq Stranded mRNA kit, 2 × 150 bp reads sequencing with S4 chemistry on a NovaSeq instrument. For each experimental condition (inoculated or noninoculated), 3 to 5 independent biological samples were processed, and their respective Illumina paired-end reads were obtained (Supplementary Table S2). These reads were simultaneously aligned to the Arabidopsis Columbia-0 reference genome Araport11 and the S. sclerotiorum strain 1980 version 2 genome (Derbyshire et al. 2017), using the pipeline nf-core/rnaseq version 3.12.0 described in https://doi.org/10.5281/zenodo.1400710. Consistency between RNA-seq samples was evaluated using Pearson correlation between samples, calculated with the cor() function from the stats package (version 4.0.4) (Supplementary Fig. S2).

Identification and analysis of DEGs

Normalization of read counts per gene was performed using the edgeR package considering total read count by library (Robinson et al. 2010). Differential expression analysis was conducted using edgeR package v 3.32.1 in R 4.0.4. using mock treated plants as a reference for each accession with ∼replicates + treatment as the design formula. Genes with |LFC| ≥ 2, a Benjamini–Hochberg-adjusted P-value of <0.05, and a false discovery rate (FDR) <0.05 were considered significant for differential expression. Raw and processed gene expression data were deposited in NCBI's Gene Expression Omnibus under accession number GSE248079. Heatmaps were generated using the heatmap.2 package in R 4.2.1. Hierarchical clustering was performed using the default parameters of hclust, and the robustness of the branches was evaluated with pvclust v2.2-0. GO enrichment analysis was performed using BINGO module 3.0.5 from Cytoscape 3.9.1, assessing overrepresentation of GO terms in our gene lists compared all GO terms from A. thaliana.

Log regression analyses to extrapolate the number of responsive and core genes were done using “drc” package v3.0-1 in R, and the log-logistic (LL.4) model was employed to characterize the relationship between the number of DEGs (n_genes) and categorical grouping factors (n_accessions). The maximal gene expression level was estimated at 10¹² for “n_accessions.”. Visualizations were created using the “ggplot2” package, showcasing boxplots and overlaid red curves representing the fitted exponential trends.

Phenotypic characterization of Arabidopsis accessions in response to S. sclerotiorum

The 0.5-cm-wide plugs of PDA agar medium containing S. sclerotiorum strain 1980 grown for 72 h at 20 °C on 14 cm Petri dishes were placed on the adaxial surface of detached leaves. These leaves were positioned in Navautron system (Barbacci et al. 2020); records were done using high-definition (HD) cameras “3MP M12 HD 2.8–12 mm 1/2.5 IR 1:1.4 CCTV Lens.” In total, 33 to 135 leaves from 3 experiments were inoculated per accession. Kinetics of S. sclerotiorum disease lesions were analyzed using INFEST script v1.0 (https://github.com/A02l01/INFEST). Statistical analyses of disease phenotypes were conducted using Tukey test in R 4.2.1 as described in Barbacci et al. (2020). Lesion doubling time in minutes was calculated using the formula 10 × ln(2)/β, where β is the speed obtained with the INFEST script v1.0.

Phylogenetics tree construction

Phylogenetic tree of the accessions were constructed from all CDSs recovered from the 23 corresponding pseudo-genomes available at the 1001 Genome Project Data Center. For this, pseudo-genomes were annotated using A. thaliana functional annotations (TAIR version 10) with liftoff (v1.6.1) (Shumate and Salzberg 2021). Annotations were converted to CDSs and protein sequences with gffread (Pertea and Pertea 2020). A total of 21,787 orthology groups were inferred with OrthoFinder version 2.5.4 (Emms and Kelly 2019). Orthogroups with a unique copy in each accession were aligned with Mafft (v7.313) (Nakamura et al. 2018). A super alignment was built from the orthogroup alignment by OrthoFinder (Supplementary Data Set 1). The accession tree was inferred with IQ-TREE (v2.5.4) (Nguyen et al. 2015). The best model according to the Bayesian Information Criterion computed by ModelFinder (Kalyaanamoorthy et al. 2017) was GTR+F+I+G4. The Newick file utilized for plotting the phylogenetic tree is provided in Supplementary Data Set 2. Branch support values were computed by 2 different methods: SH-aLRT (Guindon et al. 2010) and UFBoot (Hoang et al. 2018). Accession clustering was done using the cophenetic function from the R package ape v5.7.1. The tanglegram in Supplementary Fig. S3 was generated using phylogenetic tree data and the DEGs heatmap accession dendrogram using cophylo function from phytools v2.1-1.

Ortholog identification and population genetics metrics in Arabidopsis

Arabidopsis orthologous gene groups were retrieved using the Phytozome v13 database (Goodstein et al. 2012). These orthogroups were used to assign relative gene age (from the A. thaliana node to the Viridiplantae node), based on the species tree node to which the most recent common ancestor of the species represented in the orthogroup maps. Enrichment of gene sets in gene age categories were analyzed with chi-square tests at a significance threshold P-value of <0.05.

To investigate genetic diversity at the species level across 23 accessions, population genetics analyses were conducted using Variant Call Format (VCF) files obtained from https://1001genomes.org. These files were processed using Python version 3.8.19, with data handling and result exporting performed using Pandas (version 2.0.3). Scikit-allel (version 1.3.7) was employed to calculate population genetics metrics, including nucleotide diversity (π), Watterson's Theta (θW), and Tajima's D. Genetic variants among the VCF files were annotated using SnpEff v5.2c to classify variants as synonymous (no amino acid change) or nonsynonymous (amino acid change) based on the Araport11 Arabidopsis annotation. To extract mutation types, the annotated VCF file was parsed with SnpSift, which facilitated the identification of synonymous and nonsynonymous mutations based on variant effect annotations. BCFtools (version 1.2) was utilized to manage the VCF files and analyze trait variants, Vcftools (version 0.1.16) was employed to calculate nucleotide diversity for nonsynonymous mutations (πN) and synonymous mutations (πS). Group comparisons were performed using 10,000 bootstrap replicates with the boot package in R. The mean difference between groups was estimated, and a 95% confidence interval was calculated. A difference was considered significant if the interval did not contain zero (Johnston and Faulkner 2021).

Regression analysis, disease score correlation with expression for each gene

To identify genes whose expression correlates with disease phenotype within accession subsets, the mean disease index by accession (the growth speed of the fungus referred as speed hereafter) was fitted with LFC of expression for each gene (hereafter “LFC”) in corresponding accessions by linear regression. We utilized all LFC data, considering the 21/23 LFC values with the lowest FDR values. This approach preserves the 21 most relevant LFC values, to avoid replacing weak LFC values with NA or zero, which could introduce bias in correlation analyses. We employed R² and root mean square error (RMSE) as evaluation metrics. R² quantifies the proportion of variation in dependent variables that can be accounted for by the predictors, while RMSE represents the model's average error. R² and RMSE were calculated in R using ggplot2 v3.4.2 with the following formulas: R² = [cor(speed, LFC)]^2, and rmse = sum{[speed−cov(speed, LFC)/var(LFC) × LFC−mean(speed)−cov(speed, LFC)/var(LFC) × mean(LFC)]^2}, a rmse threshold of 0.05 was used. Based on R² > 0.4 and RMSE < 0.05, 33 genes fitting these characteristics were identified. Due to unclear annotation of the promoter region of AT4G31405 in all accessions except Col-0, this gene was removed from our analysis.

Promoter sequence analyses

A total of 500-pb promoter sequences were retrieved from for the 23 accessions using gft files generated by liftoff (v1.6.1) (Shumate and Salzberg 2021) for phylogenetic tree construction. To conduct DAP cis-motif enrichment analysis, we employed MEME Suite 5.5.7 (Bailey et al. 2015) with the Simple Enrichment Analysis (SEA) tool, using the Arabidopsis DAP motifs database from (O’Malley et al. 2016). For the identification of DAP cis-motifs predicting gene expression variation, we used the Boruta feature selection algorithm (Boruta v.8.0.0) implemented in R (v.4.0.5) with significance threshold P-value of <0.05. Tentative features were resolved using the TentativeRoughFix method. Gene-specific datasets were prepared by extracting LFC values and DAP cis-regulatory motif data from 500 bp promoter regions, identified using the Find Individual Motif Occurrences tool, based on the complete DAP motifs database from O’Malley et al. (2016). The Boruta algorithm was applied to assess the importance of motifs in explaining LFC variation across accessions. Final feature importance was determined by the mean importance score and the algorithm's final decision for each motif.

Comparative analysis of transcriptome similarity among accessions

Pearson correlation matrix to correlate transcriptomes across accessions, was performed using the cor function excluding auto-comparisons from stats package v4.3.1, considering the LFC for all expressed genes. The resulting matrix was subjected to hierarchical clustering using the pheatmap library v1.0.12 in R, with a specified number of clusters (k = 4) and the “ward.D” method. The same clustering pattern was obtained using a distance matrix with dist function from stats package v4.3.1.

Coexpression gene modules identification and analysis

We employed Weighted Gene Coexpression Network Analysis (WGCNA) (R package WGCNA, version 1.75-5) to identify gene coexpression modules across 23 accessions. To cluster genes with similar patterns of differential expression changes and to directly compare expression changes between accessions, we used the LFC data of all expressed genes. Hierarchical clustering was performed on the Topological Overlap Matrix (TOM), and dynamic tree cutting (R package dynamicTreeCut, version 1.63-1) was used to identify coexpression modules. Finally, the Uniform Manifold Approximation and Projection (UMAP) dimensionality reduction technique (R package umap, version 0.2.10.0) was applied to visualize the module structure. Data points with UMAP1 and UMAP2 coordinates exceeding absolute thresholds of 12 were excluded from the UMAP plot to improve readability. We calculated Spearman correlations between dendrogram-derived distance matrices (from gene expression clustering) and phenotypic distance matrices. For each module, gene expression data were clustered using hierarchical clustering (Euclidean distance, complete linkage), and the resulting cophenetic distance matrix was compared to phenotypic distance using a Spearman permutation test (n = 9,999).

Transcriptome–phenotype map construction and navigability

Transcriptome-resistance maps were generated using the 2 major principal components of PCA analysis (prcomp function, stats package v4.3.1), based on all LFC values calculated from all expressed genes from the 23 accessions. The x axis corresponds to the first principal component (PC1), and the y axis corresponds to the second principal component (PC2), explaining 16.7% and 10.0% of the total variance, respectively. Resistance values (lesion doubling time in minutes) for each accession were added as the 3rd dimension on the z axis. For representational purposes, resistance was calculated based on the fungus's propagation rate. To maintain consistency in units, resistance was computed for each accession as the absolute difference between the progression rate and the maximum value of this rate measured for the given accession. These scattered data were then adjusted using multilevel B-Splines to reconstruct the transcriptome-resistance maps (R package MBA: Multilevel B-Spline Approximation version 0.1-0). To assess the navigability associated with an accession, we estimated the ratio of the transcriptome-resistance map area covered by all trajectories starting from the given accession. The navigable area was obtained using a constrained random walker, with the starting point corresponding to the accession's position on the map. At each iteration, the constrained random walker could move randomly to any of the 8 adjacent positions, provided the trajectory leads to a position with at least equal resistance. When the constrained random walker could no longer move, the length of its trajectory was recorded. A new trajectory was then estimated by resetting the walker to its initial position. This process was repeated 5,000 times for each accession. The navigability associated with the accession was then calculated as the union of all trajectories, normalized by the total surface area of the map.

Accession numbers

Sequence data from this article can be found in NCBI Gene Expression Omnibus (GEO) under accession number GSE248079.

Author contributions

S.R. designed the research. F.D. performed the research. F.D., M.K., R.S., A.B., and D.R. analyzed the data. F.D. and S.R. wrote the paper.

Supplementary data

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

Supplementary Figure S1. Twenty-three natural accessions covering a broad range of the species geographic diversity.

Supplementary Figure S2. Pearson correlation matrix between RNA-seq samples.

Supplementary Figure S3. Phylogenetic tree of accessions and gene expression dendrogram did not show clear congruence.

Supplementary Figure S4. A maximum of 22,643 DEGs (68.4% of the predicted transcriptome) is estimated across the whole A. thaliana diversity.

Supplementary Figure S5. Maximum distance of LFC by gene.

Supplementary Figure S6. A maximum of 719 and 352 DEGs is estimated for, respectively, core upregulated DEGs and core downregulated DEGs considering the whole A. thaliana diversity.

Supplementary Figure S7. Categorization of DEGs did not reveal differences in Watterson's θ based on gene position on chromosomes.

Supplementary Figure S8. Categorization of DEGs did not reveal differences in π_N/π_S ratio based on gene position on chromosomes.

Supplementary Figure S9. Genes responding to S. sclerotiorum can be classified into at least 48 coexpression modules.

Supplementary Figure S10. Transcriptome similarity among accessions varies across gene coexpression modules.

Supplementary Figure S11. GO processes related to stress response, developmental processes, primary metabolism, and immune responses are enriched in coexpression modules 1 to 4.

Supplementary Figure S12. Analysis of species-wide correlation between gene expression during infection and disease resistance phenotypes.

Supplementary Figure S13. Two mutants for 2 genes exhibited a significant alteration in their response to S. sclerotiorum.

Supplementary Figure S14. Accession distance in the transcriptome landscape depends on the number of unique DEGs in pairwise comparisons between accessions.

Supplementary Figure S15. Transcriptome-resistance map showing accession proximity in the transcriptome space using only core DEGs.

Supplementary Table S1. List of A. thaliana accessions used in this study with corresponding names, initial collection locations, and lesion doubling times during S. sclerotiorum infection.

Supplementary Table S2. Number of biological samples per condition and per Arabidopsis thaliana accession.

Supplementary Table S3. Total mapped reads per sample.

Supplementary Table S4. General transcriptome statistics.

Supplementary Table S5. LFC of A. thaliana DEGs (|LFC| > 2) 24 hours after S. sclerotiorum infection across the 23 accessions.

Supplementary Table S6. Köppen–Geiger climate classification for the 23 A. thaliana accessions.

Supplementary Table S7. Accession groups used for promoter sequence comparison across modules.

Supplementary Table S8. Boruta analysis results on presence/absence polymorphisms in promoter regions strongly associated with LFC variation in at least one of 32 genes.

Supplementary Table S9. Navigability by accessions across the transcriptome-resistance map, considering only accessory DEGs, core DEGs, shell DEGs, or all genes for transcriptome proximity among accessions.

Supplementary Data Set 1. Super alignment obtained with OrthoFinder.

Supplementary Data Set 2. Newick file used to plot the phylogenetic tree.

Funding

This work was supported by the French Laboratory of Excellence project TULIP (ANR-10-LABX-41 and ANR-11-IDEX-0002-02), the French Agence Nationale de la Recherche (ANR) (ANR-19-CE20-15 and ANR-21-CE20-30), the European Research Council (ERC-StG-336808), and the Plant Health and Environment Division of INRAE.

Data availability

The data underlying this article are available in the article and in its online Supplementary material.

Dive Curated Terms

The following phenotypic, genotypic, and functional terms are of significance to the work described in this paper:

• ATHB21 Gramene: AT2G02540

• ATHB21 Araport: AT2G02540

• MYB65 Gramene: AT3G11440

• MYB65 Araport: AT3G11440

• MYB96 Gramene: AT5G62470

• MYB96 Araport: AT5G62470

• RAV1 Gramene: AT1G13260

• RAV1 Araport: AT1G13260

• SGR5 Gramene: AT2G01940

• SGR5 Araport: AT2G01940

• AT4G09490 Gramene: AT4G09490

• AT4G09490 Araport: AT4G09490

• AT4G24930 Gramene: AT4G24930

• AT4G24930 Araport: AT4G24930

• AT2G02510 Gramene: AT2G02510

• AT2G02510 Araport: AT2G02510

• AT2G39280 Gramene: AT2G39280

• AT2G39280 Araport: AT2G39280

• AT4G09600 Gramene: AT4G09600

• AT4G09600 Araport: AT4G09600

• ERF3 Gramene: AT1G50640

• ERF3 Araport: AT1G50640

• WRKY7 Gramene: AT4G24240

• WRKY7 Araport: AT4G24240

• CERK1 Gramene: AT3G21630

• CERK1 Araport: AT3G21630

• PAD3 Gramene: AT3G26830

• PAD3 Araport: AT3G26830

• RbohD Gramene: AT5G47910

• RbohD Araport: AT5G47910