Integrative omics of the genetic basis for wheat WUE and drought resilience reveal the function of TaMYB7-A1

Abstract

Improving wheat drought resilience and water use efficiency (WUE) is critical for sustaining productivity under increasing water scarcity. Here, we integrate genome-wide association study (GWAS), expression quantitative trait locus (eQTL) mapping, population-transcriptome analysis, and summary-data-based mendelian randomization (SMR), followed by functional validation using indexed EMS mutants and transgenic lines, to systematically identify key WUE regulators. GWAS across water conditions in 228 accessions identifies 73 quantitative trait loci (QTLs) for WUE-traits. Transcriptome profiling of 110 diverse accessions reveals 28 drought-responsive modules. eQTL mapping uncovers 146,966 regulatory variants, including condition-specific hotspots associated with key drought-related pathways. Integrative analysis underscores 85 high-confidence candidate genes, notably TaMYB7-A1. Overexpression of TaMYB7-A1 enhances photosynthesis, WUE, root development, and grain yield under drought condition by activating TaPIP2;2-B1 (water transport), TaRD20-D1 (stomatal regulation), and TaABCB4-B1 (root growth), reflecting reduced water loss and improved physiological resilience. Our study presents a comprehensive regulatory map and robust targets for wheat drought adaptation and resilient cultivar breeding.

Introduction

Common wheat (Triticum aestivum L.) stands as one of the world’s most vital staple crops, contributing approximately 20% of global dietary calories and protein intake¹. As the global population continues to rise and climate instability intensifies, ensuring sustainable wheat production becomes increasingly urgent. Water scarcity, one of the most formidable challenges to wheat productivity, already affects over 40% of global cropland and is projected to worsen due to changing precipitation patterns and rising evapotranspiration rates². Drought conditions can result in yield losses of up to 30% and significantly compromise grain quality, posing direct threats to food security and nutrition on a global scale³. As a result, enhancing drought resilience and water use efficiency (WUE) in wheat is a critical objective for crop improvement under both current and future climate scenarios⁴.

Wheat responds to water limitation through a range of morphological, physiological, and molecular adaptations aimed at improving water uptake or minimizing water loss⁵. A key morphological strategy is the modulation of root system architecture—specifically deeper rooting or increased root biomass—to access water from deeper soil layers⁶, involving genes like TaVSR1-B⁷, TaRNAC1⁸, and TaCLE24b⁹. On the physiological front, drought triggers rapid stomatal closure to reduce transpiration, which helps conserve water but also reduces CO₂ assimilation, thereby creating a fundamental trade-off between water conservation and photosynthetic efficiency¹⁰. This balance is tightly regulated by complex hormonal signaling networks, especially those involving abscisic acid (ABA). ABA serves as a central hub in mediating drought responses by activating signaling cascades that stomatal behavior and gene expression¹¹. In wheat, several core ABA signaling components—including TaPYL1-1B¹², TaPYL4¹³, TaPYL9¹⁴, DIW1/TaPP2C158¹⁵, TaSnRK2.10¹⁶, and TaABF2¹⁷—have been shown to contribute significantly to drought tolerance. Crosstalk between ABA and other hormones, such as jasmonic acid (JA) via TaOPR3 and TaMYC2, further enhances adaptive capacity¹⁸. Exogenous application of JA or methyl jasmonate (MeJA) has been reported to support wheat growth under water-limited conditions¹⁹.

In addition to hormonal pathways, drought stress induces oxidative stress by triggering the accumulation of reactive oxygen species (ROS), which can cause cellular damage by oxidizing membranes, DNA, and proteins³. To mitigate these effects, wheat activates robust antioxidant defenses—including enzymes such as superoxide dismutase, catalase, and peroxidase—along with osmoprotectants like proline, soluble sugars, and inorganic ions, to stabilize cellular functions and maintain osmotic balance²⁰. Growing evidence in plants indicates that aquaporins (AQPs) are responsible for cellular water homeostasis maintenance and consequently play a critical role in drought stress tolerance²¹. Intracellular calcium signaling also plays an essential role, as drought-induced changes in cytosolic calcium ion (Ca²⁺) concentrations activate downstream transcriptional regulators and stress-responsive pathways, contributing to both short-term protection and long-term acclimation²². Notably, while drought tolerance mechanisms ensure plant survival under water deficit, improving WUE aims to sustain high yield with minimal water input—an especially important distinction for sustainable agriculture²³.

Advancements in quantitative trait loci (QTL) mapping, genome-wide association studies (GWAS), and mutation screening have facilitated the identification of factors linked to drought tolerance and WUE in wheat^2,4. Key transcription factor (TF) families implicated in drought response include NAC (e.g., TaNAC071-A²⁴, TaSNAC8-6A²⁵), AP2/ERF (e.g., TaDTG6-B²⁶, TaERF87²⁷), MYB/MYC (e.g., TaMYB30-B²⁸), bZIP (e.g., TabZIP156²⁹), NF-Y (e.g., TaNF-YA7-5B³⁰, TaNF-YB11³¹), and WRKY (e.g., TaWRKY31³², TaWRKY1-2D³³). Additionally, other functional proteins target drought adaptation via various mechanisms: LATERAL ROOT DENSITY (LRD)³⁴ and SINA-type E3 ubiquitin ligase TaSINA2B³⁵ aid root growth, vacuolar cation/H⁺ antiporter TaNHX2 modulates stomatal aperture³⁶, alkane biosynthesis ECERIFERUM1-6A (TaCER1-6A) alters cuticle permeability³⁷, and WD40 protein TaWD40-4B.1³⁸ and Valine-glutamine (VQ) motif-containing protein TaVQ4-D³⁹ support ROS scavenging. Despite these advances, the functional characterization of gene candidates remains limited, hampering molecular breeding efforts to improve drought resilience and WUE in wheat^4,40.

Unlike crops such as maize and rice, where rapid linkage disequilibrium (LD) decay enables GWAS to resolve traits to single genes, the extended LD blocks⁴¹ and highly repetitive, polyploid nature of the wheat genome⁴² pose significant challenges for fine-mapping trait-associated loci. To tackle these limitations, integrative multi-omics strategies—combining genomic variation, transcriptomic profiling, and regulatory network inference—have emerged as powerful tools for unraveling the complexity of polygenic traits. Moreover, the use of large-scale mutant populations—particularly ethyl methanesulfonate (EMS)-induced mutant collections—provides critical resources for functional validation of candidate genes in wheat⁴³. Recent efforts in crops such as maize, soybean, cotton, and wheat have demonstrated the utility of such integrative approaches in elucidating key gene networks for diverse traits, including fiber development⁴⁴, seed biosynthesis⁴⁵, inflorescence architecture⁴⁶, and abiotic stress responses^47,48,49. However, comprehensive analyses focused on WUE and drought resilience in wheat—linking natural variation, gene expression, and trait performance—remain scarce.

Here, we present an integrative multi-omics approach to elucidate the genetic basis underlying WUE and drought resilience in wheat. Using a diversity panel of 228 wheat accessions, we generated a comprehensive dataset encompassing genotypic variation, transcriptome profiles, and population phenotypes—WUE, shoot biomass, root biomass, and root surface area (RSA)—under contrasting water conditions. Through combined GWAS, eQTL, and summary-data-based Mendelian randomization (SMR) analyses, we underscored candidate causal genes and identified core regulatory components. Functional validation further supported the role of TaMYB7-A1, an R2R3-MYB TF, in enhancing drought resistance by modulating water conservation, root architecture, and osmotic balance. Our study provides a comprehensive analysis and valuable genetic resources to accelerate the improvement of WUE and drought resilience in wheat.

Results

Genome-wide association analysis identifies key loci linked to WUE under varying water conditions

To define optimal conditions for evaluating drought response and WUE at the seedling stage, we first conducted a soil moisture gradient experiment using representative wheat cultivars (Supplementary Data 1). Biomass and plant-biomass based WUE (WUE_p, calculated as dry biomass per unit water consumed) were measured across a range of relative soil water content (RSWC) values in drought-sensitive, tolerant, and moderately-tolerant varieties⁵⁰. As expected, biomass increased with rising RSWC. Drought-sensitive genotypes exhibited a pronounced reduction in biomass under 6% RSWC, whereas drought-tolerant lines maintained relatively high biomass even at 15% RSWC (Fig. 1a). WUE_p showed an inverse relationship to soil moisture, decreasing with increasing RSWC and forming a characteristic S-shaped response curve. The inflection point occurred at approximately 9% RSWC, delineating a transition zone between moderate and severe water deficit. Based on this physiological response, we defined three soil moisture conditions: 15% RSWC as well-watered (WW), 9% RSWC as moderate drought stress (DS1), and 6% RSWC as severe drought stress (DS2).

Fig. 1: GWAS identifies genetic loci associated with water use efficiency in wheat.

a Representative images of wheat seedlings under a gradient of relative soil water content (RSWC) from 3% to 18%, alongside biomass–RSWC and WUE_p–RSWC curves for drought sensitive, tolerant, and moderately-tolerant groups (DS, DT, and DM, respectively). Uniform 1-week-old seedlings (11 seedlings per replicate) from five accessions per group (see Supplementary Data 1) were used. RSWC levels used in subsequent experiments were indicated by red arrows. Error bars represent standard deviation (SD) from five accessions. Scale bar, 10 cm. Significant differences were marked in red (DS vs. DT), blue (DS vs. DM), and black (DM vs. DT). **P < 0.01 (two-sided Student’s t-test). b Overview of the genotyping, phenotyping, and transcriptome profiling workflow. A total of 228 accessions were phenotyped and genotyped using the Wheat660K SNP array. A core set of 110 accessions was selected for whole-exome sequencing (WES) and RNA-seq. WUE_p (calculated as dry biomass per unit water consumed), shoot dry weight (DW.S), root dry weight (DW.R), and root surface area (RSA). c Radar chart of population phenotypes changing trend under water stress conditions, presented relative to the maximum observed value, and means were indicated. d Manhattan plots showing GWAS results for WUE_p across multiple environments. The horizontal dashed line indicates the genome-wide significance threshold at −Log₁₀(P value) = 4.0. Dots in different colors represent different environments. Wheat homologs of known drought-resistance genes within significant intervals were labeled. e Regional Manhattan plot and linkage disequilibrium (LD) heatmap for the qWUE_p_WW-2A locus. Red diamond marks SNP significantly associated with WUE_p_WW above the horizontal dashed line. The candidate gene TaHXK3 was highlighted in blue. Heatmap color intensity (white to red) corresponds to D′ values from 0 to 1. f Box plot comparing WUE_p_WW between two groups in 228 accessions based on the leading SNP in qWUE_p_WW-2A. Boxes indicate the interquartile range (IQR) with the median line, and whiskers extend to 1.5 × IQR. ***P = 1.30e-08 (Wilcoxon rank-sum test). g Summary of conditional and non-conditional QTLs identified for WUE_p, DW.S, DW.R, and RSA. h Scatterplot showing the relationship between WUE_p and the accumulation of superior alleles across accessions. The red line represents the linear regression fit with a 95% confidence interval in gray shading. Pearson correlation coefficient (r) and P value were shown.

Using this framework, we phenotyped a genetically diverse panel of 228 wheat accessions, broadly representing Chinese landraces and modern cultivars⁵¹, for four WUE-related traits–WUE_p, shoot dry weight (DW.S), root dry weight (DW.R), and RSA–were collected under WW, DS1, and DS2 conditions (Fig. 1b, Supplementary Fig. 1a, and Supplementary Data 2). Strong positive correlations were observed between WUE_p, DW.S, and DW.R, whereas correlations with RSA were weaker, accompanying significant natural variation across accessions (Supplementary Fig. 1b, c). On average, DW.S, DW.R, and RSA declined with increasing drought severity, reflecting growth suppression under water limitation. In contrast, WUE_p increased under drought conditions, with the greatest variation observed under DS1 and DS2, highlighting the potential for genetic dissection of this trait (Fig. 1c and Supplementary Fig. 1d).

To explore the genetic variant contributing to WUE, all 228 accessions were genotyped using the Wheat660K single-nucleotide polymorphism (SNP) array⁵², resulting in 323,741 high-quality SNPs after stringent filtering (see “Methods”). Population structure analysis grouped the panel into three major sub-populations (Supplementary Fig. 2a–c), and the average LD decay was estimated at ~3.0 Mb, consistent with prior reports in wheat⁵³ (Supplementary Fig. 2d). GWAS across all measured traits and water conditions revealed 2112 significant SNPs (P value < 1.00e-4) (Fig. 1d, Supplementary Fig. 3a and Supplementary Data 3). Several known drought-associated genes and their homologs were recovered, such as TaCLE24b⁹, TaGAD1³⁶, TaHXK3-2A⁵⁴, TaAKS1²⁷, TaKCS3⁴⁹, TaNF-YB11³¹, and TaMYBsm1⁵⁵, validating the analysis pipeline. For instance, TaHXK3-2A⁵⁴—a hexokinase gene previously implicated in stomatal index—was located within the LD block of a WUE_p-associated QTL under WW condition (qWUE_p_WW), with leading SNP AX-109946724 (Fig. 1e). Accessions carrying the A allele (n = 101) of this SNP had significantly higher WUE_p than those with the G allele (n = 116) (Fig. 1f), supporting its functional relevance.

We identified 49 non-conditional QTLs (associated with absolute trait values) and 24 conditional QTLs (reflecting trait ratios across water conditions), of which candidate QTLs were merged within 3 Mb of significant SNPs when detected in multiple environments (Fig. 1g, Supplementary Fig. 3b and Supplementary Data 4). The number of candidate genes within each QTL interval varied widely, ranging from a handful to over a hundred, which was partly related to local gene density. In total, 5169 genes were located within significant intervals. To evaluate the additive effect of favorable alleles, we compared the top and bottom 20 accessions ranked by WUE_p. Accessions with high WUE_p harbored significantly more superior alleles across the detected loci compared to low WUE_p accessions (Fig. 1h and Supplementary Fig. 3c), indicating an additive genetic contribution to improved WUE performance.

Together, these results identify significant SNPs and QTLs associated with WUE-related traits, including both known drought-responsive factors and novel candidates. However, the broad LD in wheat limits gene-level resolution, necessitating integrative approaches to refine and validate true candidate genes.

Population transcriptome reveals coordinated gene expression dynamics underlying WUE diversity

Gene expression variation is a major driver of phenotypic diversity in complex traits, making population-level transcriptome profiling an essential strategy for linking DNA variation, transcriptional regulation, and trait performance⁵⁶. To explore how transcriptional plasticity contributes to drought response, we performed RNA-seq on shoot and root tissues from 110 representative accessions, capturing the breadth of WUE_p variation in the broader 228-line panel (Fig. 1b and Supplementary Fig. 2b). Samples were collected under three soil moisture conditions—WW, DS1, and DS2, yielding 638 high-quality transcriptome datasets, with an average of 6.69 million reads per sample (Supplementary Data 5).

Principal component analysis (PCA) demonstrated strong tissue-specific expression patterns and clear separation of transcriptomes by water condition within each tissue type, indicating robust drought-induced transcriptional reprogramming across genotypes (Fig. 2a and Supplementary Fig. 4a). As drought severity increased, the coefficient of variation (CV) in gene expression decreased, suggesting a convergent response among accessions. Expression fold changes (FC) and their CVs were notably higher between drought-treated and well-watered conditions than between DS1 and DS2, confirming that transcriptomic variation reliably reflects the imposed water gradient (Fig. 2b and Supplementary Fig. 4b).

Fig. 2: Population-level transcriptome dynamics of wheat in response to water scarcity.

a Principal component analysis (PCA) of the population-level transcriptome of root tissue under WW, DS1, and DS2 conditions. Each dot represents a sample from one accession, with conditions distinguished by different colors and symbols. b Comparison of the expression coefficient of variation (CV) (left), fold-change (FC) (middle), and CV of expression FC (right) among expressed genes in root tissue of 110 representative accessions among different water conditions. Boxes indicate the interquartile range (IQR) with the median line, and whiskers extend to 1.5 × IQR. ns no significance; ***P ≤ 2.20e-14 (Wilcoxon rank-sum test). c Distribution of expressed genes in population accessions. Genes expression in > 95% of accessions was classified as core (orange), <5% as unique (blue), and 10-90% as other (gray). d Enriched Gene Ontology (GO) terms for core, unique, and other genes. Bar length represents fold enrichment, and color indicates −Log₁₀(P.adj) from Fisher’s exact test. e Heatmap of k-means clustered root DEGs (left), with known drought-responsive genes (center), and enriched GO terms for each cluster (right). Bar length represents fold enrichment, and color indicates −Log₁₀(P.adj) from Fisher’s exact test. f Hierarchical clustering dendrogram of population DEGs in root tissue using weighted correlation network analysis (WGCNA) (top) and heatmap of top module–trait correlations (bottom). Color represents Pearson correlation coefficients (values in squares) and asterisk indicates significance (P < 0.05, two-sided Student’s t-test). g Comparison of the gene expression CV of 110 representative accessions between two grouped modules, which were significantly associated with traits or not. Boxes indicate the interquartile range (IQR) with the median line, and whiskers extend to 1.5 × IQR. ns no significance; ***P ≤ 1.70e-15 (Wilcoxon rank-sum test). h GO enrichment analysis of the root top trait-associated modules in (f). Dot size represents fold enrichment, and color indicates −Log₁₀(P value) from Fisher’s exact test. i Comparison of the expression level of the Red module in root tissue among water conditions. Each gene in the module was plotted as a dot, and its expression levels under different water conditions were linked with gray lines. ns no significance; *P < 0.05; **P < 0.01 (Wilcoxon rank-sum test). j Co-expression network for the hub genes of the Red module in root tissue. Node (circle) size reflects the connectivity, and osmoregulatory-related genes were highlighted in blue color and red symbol.

Genes were classified by population-wide expression frequency into three categories: “core” genes expressed in >95% of accessions, “unique” genes expressed in <5%, and “other” genes expressed in between (Fig. 2c). Core genes exhibited higher average expression levels, lower expression CVs, and stronger but more discrete drought responses compared to unique genes (Supplementary Fig. 4c, d). Gene Ontology (GO) analysis revealed that core genes were mainly linked to enzymatic activity, supporting conserved drought responses, while unique genes enriched in ion channel activity and developmental regulation, likely underpin accession-specific adaptations (Fig. 2d).

Differential expression analysis (see “Methods”) identified 17,096 and 8645 differentially expressed genes (DEGs) in roots and shoots, respectively, in response to water deficits, with 3281 DEGs shared between the two tissues (Supplementary Fig. 5a and Supplementary Data 6). Clustering of DEG revealed four dominant expression patterns in both tissues: in roots, CR1 and CR2 were upregulated, CR3 was down-regulated, and CR4 showed inapparent change under DS1 or DS2 compared to WW; in shoots, expression increased in CS1, decreased in CS2, and remained unchanged in CS3 and CS4 (Supplementary Fig. 5b). These clusters included numerous known drought-responsive genes, such as TaERF73⁵⁷ and TaRNAC1⁸ in root, TaCER1³⁷ and TaPP2C6⁵⁸ in shoot and TabHLH27⁵¹ in both tissues (Fig. 2e and Supplementary Fig. 5c). GO enrichment of root DEGs revealed distinct biological roles: CR1 (translation), CR2 (morphogenesis), CR3 (hormone and stimulus response), and CR4 (ion transport) (Fig. 2e). Notably, expression levels of CR1 and CR3 clusters were negatively correlated across accessions, reflecting the classic trade-off between growth and stress response programs⁵⁹ (Supplementary Fig. 5d and Supplementary Data 7).

To further examine how coordinated gene expression and its dynamics contribute to trait variability, we conducted weighted correlation network analysis (WGCNA) across the transcriptome datasets (Fig. 2f and Supplementary Fig. 6a). Genes were clustered into modules based on similar expression trends with specific module eigengene, and modules were assessed for associations with WUE-related traits (Supplementary Fig. 6b–d). Genes in trait-associated modules consistently displayed higher expression CVs, except for DW.S trait, supporting the view that transcriptional dynamics contribute to phenotypic diversity (Fig. 2g). Focusing on the top three associated modules per trait, we identified five top trait-associated modules in roots and eight in shoots (Fig. 2f, Supplementary Fig. 6e and Supplementary Data 8). GO enrichment of these modules revealed functional categories central to drought response, including water deprivation, hormone signaling, osmotic stress response and photosynthesis (Fig. 2h and Supplementary Fig. 6f).

Among these, the Red module in root tissue stood out for its significant association with all four WUE-related traits, a particularly strong correlation with WUE_p, and marked downregulation under water deficit, especially in DS2 (Fig. 2f, i). This module included several known osmoregulatory genes, such as TaALDH7B4⁶⁰, TaPRX52⁶¹, TabZIP20⁶², and TaABA4⁶³, which formed a highly connected hub network (Fig. 2j). In shoots, the Darkred module was upregulated under drought conditions and exhibited enriched oxidative stress functions (Supplementary Fig. 6g, h), reinforcing the role of osmotic homeostasis in WUE improvement.

In summary, population-scale transcriptome profiling uncovered dynamic, tissue-specific gene expression programs that drive natural variation in WUE-related traits. It underscores the potential roles of coordinated osmotic regulation and oxidative stress responses, via key hub genes and co-expression modules, in contributing to WUE under drought stress in wheat.

eQTL mapping and SMR analysis identify expression-driven regulatory loci underlying WUE and drought-related traits

Understanding how genetic variation modulates gene expression and how this contributes to phenotypic diversity is essential for unraveling the regulatory architecture of complex traits⁶⁴. To pinpoint regulatory elements that shape WUE and drought responses, we performed eQTL mapping using both Wheat660K array SNPs and whole-exome sequencing data from 110 selected representative wheat accessions (see “Methods”).

This analysis identified 146,966 significant eQTLs associated with the expression of 8218 genes (hereafter “eGenes”) across all environments, with the highest eQTL region density at 1.91 Kb size (Supplementary Figs. 7 and 8a and Supplementary Data 9). These eQTLs were broadly distributed across the genome, predominantly located in distal intergenic regions, exons, and promoters within 1 Kb upstream of transcription start sites (Fig. 3a). Using a 2.4-Mb LD threshold (Supplementary Fig. 8b), 11.27% of eQTLs were classified as cis-eQTLs and 88.73% as trans-eQTLs, averaged across tissues. Trans-eQTLs, especially inter-chromosomal ones, showed a subgenome-biased regulatory pattern, predominantly involving B-to-D genome interactions and relatively few A-to-A interactions (Supplementary Fig. 8c). Notably, 27.54% of eGenes were regulated solely by cis-eQTLs, 24.05% solely by trans-eQTLs, and nearly 50% by a combination of both (Fig. 3b). Consistent with previous findings^45,47, cis-eQTLs existed higher determination coefficients and tended to explain a greater proportion of expression variance (Fig. 3c). Drought conditions led to highly dynamic eQTL behavior. Only 10.99% of all detected eQTLs were shared across WW, DS1, and DS2 conditions, while the vast majority exhibited condition-specific regulatory patterns (Supplementary Fig. 8d). These results highlight the context-dependent nature of regulatory variation under abiotic stress.

Fig. 3: eQTL and SMR analyses reveal genetic modulation of gene expression diversity and trait associations.

a Genome-wide distribution of eQTLs under different water conditions. b Proportions of eGenes classified by eQTL type across conditions. c Comparison of eQTL explanatory power (R² value) between cis- and trans-eQTLs across conditions. Boxes indicate the interquartile range (IQR) with the median line, and whiskers extend to 1.5 × IQR. ***P ≤ 3.54e-10 (Wilcoxon rank-sum test). d Circos diagram showing genome-wide eQTLs, hotspots, and GWAS signals across tissues and conditions. From outer to inner tracks: eGene numbers, chromosome names, eQTLs in root and shoot, eQTL hotspots in root and shoot, and GWAS signals in root and shoot. The innermost track depicts hotspot regulatory interactions, with drought-response related GO terms highlighted in blue and labeled around the circle. e, f Comparison of −Log₁₀(P value) and phenotypic variance explained (PVE) of SNPs in GWAS, grouped by SNP within non-eQTL, eQTL, and eQTL hotspot. Boxes indicate the interquartile range (IQR) with the median line, and whiskers extend to 1.5 × IQR. ns no significance; *P < 0.05; **P < 0.01 (Wilcoxon rank-sum test). g Manhattan plot summarizing SMR analysis for four WUE-related traits. SMR scores (y-axis) were plotted against genomic positions (x-axis). The horizontal dashed lines indicate the significance threshold (−Log₁₀(P value) = 2.0). Known drought- or WUE-related genes were labeled in distinct colors and shapes for corresponding traits. Above the x-axis, positive correlations; below the x-axis, negative correlations. h Comparison of GWAS and eQTL effect sizes between SMR significance and SMR non-significance groups. Boxes indicate the interquartile range (IQR) with the median line, and whiskers extend to 1.5 × IQR. ***P < 0.001 (Wilcoxon rank-sum test). i, j SMR association of candidate genes within qWUE_p.7A.1 GWAS intervals with P value in (i), and correlation between GWAS and eQTL effect size shown in (j).

Although most eGenes were regulated by a limited number of eQTLs (only 1–3 statistically) (Supplementary Fig. 8e), several prominent eQTL hotspots, genomic regions affecting the expression of numerous genes, were identified (Supplementary Fig. 9). These eQTL hotspots overlapped with identified GWAS signals, and their target eGenes were significantly enriched in pathways critical for drought adaptation, including stomatal movement, water deprivation response, response to redox state, ABA signaling, and photosynthesis (Fig. 3d). Importantly, SNPs located within eQTL hotspots had stronger detection power and phenotypic impact in GWAS (Fig. 3e, f), and regulated genes with higher expression level and greater expression variability (Supplementary Fig. 10a, b), reinforcing their functional significance in mediating phenotypic plasticity under water-limited conditions.

To directly link gene expression with phenotypic variation, we applied SMR to explore likely causal genes for WUE-related traits. This approach identified 830, 473, 464, and 324 genes significantly associated with WUE_p, DW.S, DW.R, and RSA, respectively (P value < 0.01), yielding 1709 unique genes with trait associations (Fig. 3g and Supplementary Data 10). Of these, 82.09% were associated with a single trait, while 17.91% showed pleiotropic associations with two or more traits (Supplementary Fig. 10c). Known drought-related regulators such as TaSAP5⁶⁵, TaTIP41⁶⁶, and TaCML20⁶⁷—involved in degradation of DREB2A interacting protein, ABA-induced stomatal closure, and calcium-dependent root growth—were recovered (Fig. 3g), validating the analytical framework.

Global statistical analysis revealed that genes significantly associated with SMR exhibited stronger GWAS and eQTL effects (Fig. 3h), aiding in the identification of key genes within significant GWAS intervals. For example, within the GWAS-identified interval qWUE_p.7A.1 (containing 61 genes), four genes exhibited significant SMR-associated signals. These included TaCML29⁶⁸ (a calcium-binding protein) and TaAKS1²⁷ (a regulator of osmotic adjustment), which also showed strong eQTL and GWAS effect sizes (Fig. 3i, j). In general, genes significantly associated with SMR displayed higher association metrics across other datasets, including eQTL and GWAS, supporting the robustness of the integrative approach.

Collectively, we defined the transcriptional regulatory landscape of drought response and WUE in wheat by combining eQTL and SMR analyses. The identification of eQTL hotspots and 1709 expression-associated candidate genes provides a valuable foundation for narrowing causal gene sets and advancing functional dissection of WUE and drought adaptation mechanisms in wheat.

Multi-omics integration and mutant validation identify high-confidence regulators of WUE and drought resilience

Building on GWAS, eQTL, and SMR analyses, we next aimed to underlie high-confidence causal genes underlying WUE-related traits by integrating genomic variation, gene expression profiles, and phenotypic diversity with functional validation. To systematically delineate regulatory relationships, a correlation network based on genomic co-localization across all conditions was constructed (see “Methods”), incorporating 65 GWAS bins, 188 SMR-associated genes, and 324 eQTL regions (Supplementary Fig. 11). This network captured 256 unique regulatory interactions (Supplementary Data 11), with observed connectivity for WUE_p, DW.S, and DW.R.

Among the SMR-associated genes within this integrated network, many corresponded to orthologs previously implicated in drought responses, including pathways related to water deprivation, osmotic stress, stomatal movement, calcium ion, and root development in rice and Arabidopsis thaliana (Fig. 4a, b). To further refine candidate regulators, we intersected network genes with tissue-specific DEGs identified from the population transcriptome, yielding 85 high-confidence genes with both genomic and transcriptional evidence of involvement in drought response (Fig. 4c, d and Supplementary Data 12). Notably, 28 of these genes (32.94%) were orthologous to previously characterized drought-responsive genes, including TabZIP72⁶⁹ (stress signaling), TaFSD2⁷⁰ (osmotic homeostasis), TaCML29⁶⁸ (Ca²⁺ signaling), TaBPS1⁷¹ (root development), and TaAKS2⁷² (stomatal regulation) (Fig. 4d). The remaining novel candidates spanned diverse functional categories, including TFs (AP2, ERF, B3, WRKY), enzymes (oxidases, kinases, hydrolases), transporters, molecular chaperones, and structural proteins such as chloroplast-associated and microtubule-related components. Several classes—e.g., GDSL-type esterase/lipase⁷³, U-box proteins⁷⁴, chloroplast proteins⁷⁵, dirigent proteins⁷⁶, and VQ motif proteins³⁹—have been implicated in stress adaptation, lending further support to their potential functional relevance in wheat (Fig. 4d).

Fig. 4: Integrative multi-omics underscores candidate causal genes for drought resistance and WUE in wheat.

a The co-localization network of WUE_p was constructed by integrating GWAS, eQTL, and SMR analyses under pooled water conditions. An SMR gene was considered co-localized with a GWAS signal if within the same LD region, or its regulating eQTL was located in that region. GWAS bins, eQTL regions, and SMR genes were indicated by different symbols. Edges indicate relationships: regulatory links between eQTLs and SMR genes, and LD-based co-localization between GWAS signals and SMR genes or eQTLs. b Overlap among GWAS signals, eQTLs, and SMR genes with orthologs in rice or Arabidopsis thaliana implicated in drought resistance or WUE. Genes were grouped into five functional categories, shaded in different colors. SMR genes co-localized with GWAS signals were filled according to −Log₁₀(P value); those within GWAS signals were outlined in red, others in yellow. c Bar graphs showing the number of genes identified in the co-localization network for each trait. Candidate genes, also identified as DEGs from population-level transcriptome data, were highlighted in purple. d The 85 high-confidence candidate genes identified through multi-omics analysis, including known genes and novel genes, along with respective proportion and functional category. e Representative images of wild-type KN9204 (WT) and gene-indexed homozygous mutant lines from quadruplicate-verified seedling drought assays. Images show seedlings after 7 days of water withholding (top) and 7 days of rewatering (bottom). The red arrows indicate tender and vibrant green leaves with 36 seedlings per plot. Scale bars, 5 cm. f Bar graphs of survival rate (left) and fresh weight (right) for WT (gray) and mutant lines (green and brown) after 7 days of recovery. Data represent mean ± S.D. of four independent replicates (36 seedlings per replicate). ns no significance; *P < 0.05; **P < 0.01 (two-sided Student’s t-test). g Scatterplot of survival rate and fresh weight for WT (gray triangle) and mutant lines (green and brown circles). Vertical and horizontal dashed lines indicate the thresholds for significant differences compared to WT. h Hierarchical regulatory network of 85 key candidate genes. Diamonds represent transcription factors (TFs) and ovals represent functional genes. Gene expression tissue specificity was color-coded: shoot-specific (green), root-specific (blue), and shared (brown). Arrows indicate regulatory relationships: red for positive and blue for negative, and which intensity reflects the strength of expression correlation. TaMYB7-A1 were highlighted in red.

To experimentally validate gene function, we leveraged a gene-indexed EMS-mutagenized population of the elite wheat cultivar KN9204⁴³. Homozygous mutant lines for five randomly selected candidate genes were subjected to seedling drought assays and compared with wild-type controls (Fig. 4e and Supplementary Data 13). Of these, mutants for two genes exhibited significantly higher survival rate (SR) (27.10% in WT, 96.00% in G1, 77.93% in G2, p = 1.90e-05 and 0.001), and mutants for two others showed reduced SR (7.11% in G4, 0.00% in G5, p = 0.008 and 0.007) after rewatering. Additionally, fresh weight (FW) increased in one mutant (23.30 mg in WT, 32.93 mg in G2, p = 0.045) and decreased in another (16.41 mg in G5, p = 0.022) (Fig. 4f). In total, four out of the five tested genes (80%) exhibited altered responses to drought, either positive or negative, which supports their functional roles in drought adaptation and validates the predictive power of the multi-omics approach (Fig. 4g).

Further transcriptional regulatory network analysis of the 85 high-confidence genes revealed a hierarchical regulatory structure involving TFs with tissue-specific and overlapping activity. Specifically, three TFs were active in shoot networks, three in root networks, and one TF TaMYB7-A1 differentially expressed in both tissues and existed regulatory roles in the shoot. This interdependent regulatory module appears to orchestrate WUE-related responses across tissues (Fig. 4h).

In conclusion, by integrating genetic association, transcriptomic regulation, and functional genomics, we identified a core set of 85 candidate causal genes for WUE and drought resistance in wheat. Functional validation using indexed mutants confirmed the predictive power of this pipeline, bridging multi-omics discoveries with mechanistic gene function.

Functional validation of TaMYB7-A1 reveals its role in drought resilience and yield stability

Among the high-confidence candidates identified, TaMYB7-A1 (TraesCS2A02G554200), an ortholog of OsMYB6 in rice, emerged as a key regulator due to its central position in the integrative regulatory network, with expression activity detected in both shoot and root (Fig. 4h). OsMYB6 is known to enhance salt and drought resistance in rice⁷⁷, further supporting the relevance of its wheat ortholog. TaMYB7-A1 is located within a 2.31-Mb LD block on chromosome 2A (Block2, Chr2A: 758.42–760.73 Mb), which was significantly associated with DW.S (Fig. 5a). Among the 45 genes in this region, only TaMYB7-A1 was an SMR-associated gene and population-level transcriptome DEG under drought conditions (Fig. 5b).

Fig. 5: TaMYB7-A1 enhances drought resistance and yield performance under water-limited conditions in wheat.

a Manhattan plot of GWAS result for shoot dry weight (DW.S), alongside a heatmap of linkage disequilibrium (LD) among SNPs within a 5.98-Mb interval surrounding the peak SNPs on chromosome 2A. The y-axis represents −Log₁₀(P value) values of SNP associations, and the x-axis shows chromosomal positions. The horizontal dashed line indicates the GWAS significance threshold of −Log₁₀(P value) = 4.0. LD intensity is color-coded from yellow to red (D’ = 0–1). b Regional plot of Block2 of qDW.S.2A.1 in (a), from bottom to top: annotated genes, GWAS signals (lead SNP in blue), SMR association (significantly associated genes in green), and population transcriptome DEGs (significant genes in GFOLD and edgeR colored by orange and yellow, respectively). Y-axis: −Log₁₀(P value) for GWAS, −Log₁₀(P value) for SMR, accession proportion of FC in population transcriptome DEG identification. Dashed lines indicate significance thresholds for GWAS (P value = 1.00e-4), SMR (P value = 0.01), and population transcriptome DEG (Log₂FC > 1 in at least 15% of accessions). TaMYB7-A1 was shaded by a pink vertical line, and the eQTL of TaMYB7-A1 colored by purple. c Expression pattern of TaMYB7-A1 in shoot and root tissue under well-watered (WW) and water-limited (WL) conditions. RT-qPCR was performed with TaTubulin as the internal control. Expression was normalized to shoot tissue under WW (set to 1.0). Data represent mean ± S.D. of three independent replicates. *P < 0.05; **P < 0.01 (two-sided Student’s t-test). d Comparison of expression level (top) and DW.S (bottom) between two groups of 110 representative accessions based on SNP Chr2A_759463340, an eQTL regulating TaMYB7-A1. Boxes indicate the interquartile range (IQR) with the median line, and whiskers extend to 1.5 × IQR. **P < 0.01 (Wilcoxon rank-sum test). e Correlation between TaMYB7-A1 expression and DW.S under drought (DS1) conditions. Each point represents one accession. Marginal distributions of gene expression and DW.S were shown along the top and right axes, respectively. The orange line represents the linear regression fit with a 95% confidence interval (light orange range). Pearson correlation coefficient (r) and P value were shown. f Drought resistance assay of TaMYB7-A1 overexpression (OE) lines and WT under WW (top) and WL (withholding then rewatering) (bottom) conditions. Photographs and measurements of survival rate and fresh weight per plant were taken after 7 days of recovery under water-limited conditions. Scale bars, 10 cm. Data represent mean ± S.D. of three independent replicates with 28 seedlings per replicate. Representative images from triplicate-verified seedling drought assays were used. *P < 0.05; **P < 0.01 (two-sided Student’s t-test). g Representative images of grain morphology in OE lines and WT under WW and WL in field conditions. Red lines indicate the length of ten WT grains and the width of twenty WT grains. Scale bars, 1 cm. h–j Comparison of 1000-grain weight (h), spike number per plant (i), and grain yield per plant (j) between OE lines and WT under WW and WL conditions in field conditions. Data represent mean ± S.D. (n = 6 for grain traits, each with above 200 seeds; n = 40 for spike number per plant). Boxes indicate the interquartile range (IQR) with the median line, and whiskers extend to 1.5 × IQR. ns no significance; *P < 0.05; **P < 0.01; ***P < 0.001 (two-sided Student’s t-test).

Expression analysis revealed that TaMYB7-A1 was significantly upregulated in both shoot and root tissues of the drought-sensitive wheat variety Fielder under water-limited (WL) conditions (Fig. 5c), consistent with a stress-responsive role. A downstream SNP (Chr2A_759463340), located near the gene and acting as a proximal regulatory eQTL, was identified as a key variant influencing TaMYB7-A1 expression. Based on this SNP, the population was classified into two groups: C-type (n = 94) and T-type (n = 14). Accessions carrying the T-type allele exhibited significantly higher TaMYB7-A1 expression (p = 4.97e-08) and greater shoot biomass (DW.S, p = 0.0023) compared to those with the C-type variant (Fig. 5d). Additionally, TaMYB7-A1 expression was positively correlated with DW.S across accessions (Fig. 5e), reinforcing its role in biomass accumulation and WUE-related traits.

To experimentally validate its function, we generated independent TaMYB7-A1 overexpression (OE) lines driven by the Ubiquitin promoter in the Fielder background⁷⁸. Three OE lines with elevated transcript levels were selected for further analysis (Supplementary Fig. 12a). Under well-watered (WW) conditions at the seedling stage, TaMYB7-A1 OE lines were phenotypically similar to wild-type (WT) Fielder controls. However, under water-limited (WL) conditions, TaMYB7-A1 OE plants showed significantly improved SR (44.05% in WT, 69.05%-79.76% in OE lines, p = 0.022, 0.007, and 0.010) and FW compared to WT (40.25 mg in WT, 59.44-80.57 mg in OE lines, p = 0.035, 0.004, and 0.001) (Fig. 5f).

Field trials were conducted to evaluate the agronomic performance of TaMYB7-A1 OE lines under both WW and WL conditions. Under WW conditions, TaMYB7-A1 OE lines did not differ significantly from WT in grain size or weight (Fig. 5g, h and Supplementary Fig. 12b, c). Minor variations were observed in spike number (OE1) (15.93 vs. 13.05, p = 0.0009) and spikelet number (OE3) (18.20 vs. 19.10, p = 0.012), but these did not negatively impact yield (except OE2; 12.55 vs. 12.75 g, p = 0.014) (Fig. 5i, j and Supplementary Fig. 12d). Under WL conditions, all OE lines produced significantly more spikes per plant (except OE1) (8.80 in WT, 10.93 in OE2, 9.93 in OE3, p = 4.80e-05 and 0.045), more spikelets per spike (17.35 in WT, 17.85–18.73 in OE lines, p = 7.80e-13, 0.013 and 3.80e-09), and larger grains (grain weight: 33.94 g in WT, 36.67–38.38 g in OE lines, p = 0.02, 0.0012 and 0.0042; grain length: 6.65 mm in WT, 6.83–6.84 mm in OE lines, p = 0.031, 0.047 and 0.04; grain width: 3.58 mm in WT, 3.66–3.77 mm in OE lines, p = 0.0066, 0.048 and 0.0063), resulting in significantly higher grain yields per plant (10.19 g in WT, 10.84–11.30 g in OE lines, p = 1.90e-07, 4.70e-11 and 5.70e-12) relative to WT controls (Fig. 5g–j and Supplementary Fig. 12b–d).

Comprehensive phenotypic analysis under WL conditions showed higher seedling SR and advantageous field traits (superior seeds, more spikes, and increased spikelets), demonstrating that TaMYB7-A1 enhances drought resistance and yield performance under water-limited conditions without compromising productivity under optimal water supply.

TaMYB7-A1 enhances drought resilience and WUE through coordination of transpiration, root development, and osmotic adjustment

To investigate the molecular mechanisms by which TaMYB7-A1 enhances drought resistance and WUE, we performed transcriptomic profiling of TaMYB7-A1 OE lines and WT under WW and WL conditions (Supplementary Fig. 13). PCA revealed clear transcriptomic separation between WW and WL samples, indicating robust drought induction and genotype-specific transcriptional reprogramming (Fig. 6a).

Fig. 6: RNA-seq reveals the molecular regulatory role of TaMYB7-A1 in wheat drought resistance and WUE.

a Principal component analysis (PCA) of transcriptomes from shoot (left) and root (right) samples of OE lines and WT under WW and WL conditions. Samples of different genotypes were indicated by different colors, and samples from each condition with different symbols. b Venn diagram displaying the overlap of drought response genes in WT (blue) and TaMYB7-A1 affected genes under WW (green) and WL (red) conditions. The TaMYB7-A1 affected drought response genes were marked in red. c Enriched Gene Ontology terms for DEGs in shoot and root tissue. Dot size represents fold enrichment and color indicates −Log₁₀(P value) from Fisher’s exact test. d Venn diagram displaying the overlap of DEGs with predicted TaMYB7-A1-binding motif, population-transcriptome DEGs, and response to PEG-simulated drought stress (publicly available datasets), resulting in 40 genes shared by all three groups and considering as potential downstream direct targets. Of them, eleven genes have known orthologs involved in drought-resistance or WUE were shown. e Relative water loss of detached leaves of OE lines and WT. Data represent mean ± S.D. of three independent replicates. Statistical significance was labeled with the color corresponding to the line. *P < 0.05; **P < 0.01 (two-sided Student’s t-test). f Leaf surface temperature of OE lines and WT under WW and WL conditions was investigated using infrared thermography. In the infrared imaging photographs (upper panel), blue to red colors indicate temperatures from low to high. Representative images from triplicate-verified seedling drought assays were used. Scale bars, 5 cm. Data represent mean ± S.D. (n = 20). Statistical significance was labeled using Tukey’s HSD multiple comparisons test. g Comparison of the Fv/Fm (maximum PSII efficiency) and WUE_i between OE lines and WT under WW and WL conditions. Data represent mean ± S.D. of eight independent replicates. Boxes indicate the interquartile range (IQR) with the median line, and whiskers extend to 1.5 × IQR. ns no significance; *P < 0.05; **P < 0.01 (two-sided Student’s t-test). h Phenotypic characteristics of the root systems in OE lines and WT under WW and WL conditions. Representative images from triplicate-verified seedling drought assays were used. Scale bars, 10 cm.

By overlapping genes drought-responsive (WL vs. WW) and those regulated by TaMYB7-A1 (OE vs. WT, differentially expressed in all three OE lines), we identified 548 and 688 DEGs in shoot and root tissues, respectively (Fig. 6b and Supplementary Data 14). GO enrichment analysis revealed involvement in stomatal regulation, photosynthesis, osmotic stress response and root development (Fig. 6c). Notably, several marker gene were identified, including TaASR5⁷⁹ and TaGS2⁸⁰ related to stomatal regulation and photosynthesis; TaZFP182⁸¹ and TaZFP252⁸² involved in osmoregulatory accumulation; TaPIP2;2⁸³ and TaPIP1;3⁸⁴ encoding aquaporin proteins, as well TaEXPB2⁸⁵ and TaFBX257⁸⁶ associated with root development (Supplementary Fig. 14a). These findings implied a regulatory function of TaMYB7-A1 in enhancing water-uptake, osmotic adjustment and root development.

To further identify potential direct targets, we searched for the TaMYB7-A1-binding motif, based on its closest Arabidopsis orthologs (Supplementary Fig. 14b, c), in the promoters of overlapping population-transcriptome DEGs and drought-responsive genes. This analysis revealed 40 putative direct targets (Fig. 6d and Supplementary Data 15), including genes involved in stomatal regulation (TaRD20-D1⁸⁷, TaKAT-B1⁸⁸), osmotic adjustment (TaPIP2;2-A1/B1⁸³, TaPIP1;3-D1⁸⁴, TaOXS3-A1⁸⁹, TaPER25-D1⁹⁰ and TaP5CS-A1⁹¹) and root development (TaAS1-A1⁹², TaABCB4-B1⁹³, TaRALF1-D1⁹⁴) (Fig. 6d).

Physiological assays further supported these findings. TaMYB7-A1 OE lines exhibited reduced relative water loss and increased leaf temperature under WL conditions, indicating tighter stomatal control and lower transpiration rates (Fig. 6e, f). Importantly, TaMYB7-A1 OE lines maintained significantly higher photosystem II efficiency (Fv/Fm) and instantaneous water use efficiency (WUE_i, the ratio of CO₂ assimilation to transpiration) compared to WT plants under WL, without differences under WW condition (Fig. 6g). Root system architecture was also markedly enhanced in TaMYB7-A1 OE lines, with increased FW, volume, average diameter, number of forks and crossings (Fig. 6h and Supplementary Fig. 15a, b), as well root vitality (Supplementary Fig. 15c) under WL condition, supporting improved water acquisition. TaMYB7-A1 OE plants also displayed reduced physiological damage under WL condition, evidenced by reduced ROS levels (DAB staining, H₂O₂, and O₂⁻ content) (Fig. 7a, b and Supplementary Fig. 16a), lower malondialdehyde (MDA) accumulation (Supplementary Fig. 16b), and enhanced osmoprotectants accumulation, including proline and soluble sugars (Fig. 7c, d).

Fig. 7: Osmotic regulation and regulatory mechanism of TaMYB7-A1 in wheat.

a, b Comparison of the ROS level staining by DAB (a) and H₂O₂ content (b) between OE lines and WT under WW and WL conditions. Grayscale values in images were extracted to quantify staining intensity; higher values indicate lighter coloration and lower ROS levels. Data represent mean ± S.D. (n = 12 for grayscale value of DAB staining; n = 6 for H₂O₂ content; with three seedlings per replicate). Boxes indicate the interquartile range (IQR) with the median line, and whiskers extend to 1.5 × IQR. ns no significance; **P < 0.01; ***P < 0.001 (two-sided Student’s t-test). c, d Comparison of the proline content (c) and soluble sugar content (d) involving osmotic adjustment between OE lines and WT under WW and WL conditions. Data represent mean ± S.D. (n = 4, with three seedlings per replicate). ns no significance; *P < 0.05; **P < 0.01; ***P < 0.001 (two-sided Student’s t-test). e, f Electrophoretic mobility shift assay (EMSA) showing direct binding of TaMYB7-A1 to the promoter regions of TaRD20-D1 (e) and TaABCB4-B1 (f). In the assay, “+” and “−” denote the presence and absence of the probe or protein, respectively. The term “Shift” denotes the protein-bound DNA probe bands, while “Free probe” indicates the unbound DNA probe bands. A representative photograph of three independent replicates was shown. g Quantify of Dual-luciferase (LUC) reporter assays in tobacco leaves demonstrating the transcriptional activation of TaMYB7-A1 on TaPIP2;2-B1 (left), TaRD20-D1 (middle), and TaABCB4-B1 (right). The relative value of LUC/REN (Renilla) was normalized with the value in green fluorescent protein (GFP) set as 1.0. Data represent mean ± S.D. of five independent replicates. *P < 0.05; **P < 0.01 (two-sided Student’s t-test). h Expression patterns of TaPIP2;2-B1 (left), TaRD20-D1 (middle), and TaABCB4-B1 (right) in OE lines and WT under PEG-simulated drought stress. RT-qPCR was used to determine their expression patterns, with TaTubulin as the internal control. The expression levels were normalized with WT-CK set as 1.0. Data represent mean ± S.D. of three independent replicates. ns no significance; *P < 0.05; **P < 0.01 (two-sided Student’s t-test). i A model of TaMYB7-A1 enhances drought resilience and WUE through coordination of transpiration, root development, and osmotic adjustment.

To validate the direct regulatory targets of TaMYB7-A1, electrophoretic mobility shift assays (EMSA) and luciferase (LUC) reporter assays in tobacco leaves confirmed that TaMYB7-A1 directly binds and activates TaPIP2;2-B1 (aquaporin), TaRD20-D1 (stomatal movement), and TaABCB4-B1 (root development) (Fig. 7e–g and Supplementary Fig. 16c, d). Additionally, RT-qPCR analyses corroborated the significantly differential expression of these genes to response drought in TaMYB7-A1 OE lines (Fig. 7h).

Collectively, TaMYB7-A1 orchestrates a multifaceted drought response by modulating water transport, root architecture, and osmotic adjustment (Fig. 7i). These mechanisms collectively contribute to improved WUE and drought resistance, demonstrating the potential utility in breeding wheat varieties better adapted to water-limited environments.

Discussion

Unraveling the genetic and molecular architecture of drought resilience and WUE is crucial for breeding crops adapted to water-limited environments⁴. However, the polygenic and environment-dependent nature of these traits poses considerable barriers to gene discovery and trait improvement. Here, we developed an integrative multi-omics framework—combining GWAS, transcriptome, eQTL mapping, SMR causal inference, and mutant validation—to dissect these complex traits in wheat. By anchoring trait variation to gene regulatory variation, we overcame the resolution barrier posed by wheat’s extended LD⁴¹ and polyploidy⁴².

The integration of eQTL and SMR analyses with GWAS greatly enhanced our ability to pinpoint likely causal genes from broad QTL intervals. This strategy identified 85 trait-associated genes, including regulators involved in Ca²⁺ signaling²², osmotic stress⁷⁰, root development⁷¹, and stomatal movement⁷² (Fig. 4). Notably, the identification of eQTL hotspots—regions where genetic variants modulate the expression of multiple genes—provides valuable entry points into gene networks controlling stress adaptation. These hotspots were enriched for drought-response pathways and aligned with GWAS peaks, supporting their functional relevance (Fig. 3).

Our approach builds on a growing precedent in crops such as cotton⁴⁴, soybean⁴⁵, and maize⁴⁷, where integrated omics has been proven effective in illuminating key regulators of agronomic traits. Combined transcriptome and eQTL-GWAS strategies in soybean have uncovered regulators of seed weight and oil content (e.g., GmRWOS1⁴⁵), while similar strategies identified regulators of secondary cell wall development⁴⁴ and stress balance⁴⁷. In wheat, such approaches remain rare due to genome complexity, and this study significantly expands functional gene resources for WUE and drought resistance.

Moreover, functional validation using gene-indexed EMS mutants⁴³ demonstrated that 80% of tested candidates significantly influenced drought resistance, thus bridging the gap between genomic signal and phenotypic effect (Fig. 4). This resource complements CRISPR-based validation systems⁹⁵ and could accelerate translational breeding in wheat and related cereals.

Drought stress responses are shaped not only by genetic architecture but also by transcriptional flexibility and coordination⁹⁶. Our population-level transcriptome analysis revealed extensive tissue-specific and condition-dependent gene expression variability, with distinct gene clusters associated with growth, stress responses, and hormonal regulation. Genes in WUE-associated expression modules, especially those related to osmotic adjustment⁹⁷, ROS detoxification⁹⁸, and hormone signaling⁹⁹, displayed strong correlations with trait variation under drought, reinforcing the role of gene regulatory networks in shaping phenotypes¹⁰⁰ (Fig. 2).

Expression modules showed functional coherence, echoing findings from Arabidopsis, where co-expression networks are linked to drought memory¹⁰¹ and ABA sensitivity¹⁰². Similarly, in rice, network-based studies have implicated TFs in orchestrating responses to water limitation^103,104. Our discovery of similar regulatory modules in wheat, including drought-enriched root (Red) and shoot (Darkred) modules, emphasizes the evolutionary conservation of stress-regulatory architecture across plant lineages (Fig. 2).

Our eQTL mapping further highlighted the plasticity of gene regulation under drought. Most eQTLs were condition-specific, suggesting that transcriptional rewiring plays a key role in adaptation⁴⁷. Subgenome interactions, especially B-to-D dominance in trans-eQTLs, reflect the regulatory complexity unique to hexaploid wheat (Supplementary Fig. 8). Such insights align with findings in polyploid cotton¹⁰⁵ and Brassica napus¹⁰⁶, where regulatory partitioning among subgenomes contributes to stress resilience.

Together, these results affirm that dynamic transcriptional regulation is a critical layer linking genotype to drought-tolerant phenotypes. Capturing this plasticity, through eQTL and co-expression networks, provides a system-level view essential for predictive breeding and functional validation.

The MYB TF family, particularly the R2R3-MYB subfamily, plays pivotal roles in mediating plant stress response and tolerance¹⁰⁷. In wheat, several MYB genes (e.g., TaMYB30-B²⁸, TaMYB33¹⁰⁸, TaMYBsm1⁵⁵, and TaMYB44-5A¹⁰⁹) have been shown to confer drought responsiveness when allogeneic expression in Arabidopsis thaliana. Additionally, a few wheat transgenic evidences confirmed functional roles for MYB, like TaMYBL1²⁴, TaMYB96-2D/5D³⁷, and TaMpc1-D4¹¹⁰. In our study, TaMYB7-A1 emerged as a key regulator of WUE and drought resistance in wheat. It was supported by multiple layers of evidence—GWAS, SMR, eQTL, and DEGs—and validated by both transgenic morphological and physiological analyses. Overexpression of TaMYB7-A1 conferred enhanced SR, biomass, and grain yield per plant under water-limited conditions without compromising performance under well-watered conditions, aligning with the goals of climate-resilient wheat breeding (Fig. 5).

Mechanistically, TaMYB7-A1 orchestrated improvements in transpiration, root development, and osmotic protection. It directly activated key targets including TaPIP2;2-B1 (aquaporin), TaRD20-D1 (stomatal movement), and TaABCB4-B1 (root auxin transport), linking transcriptional control to physiological outcomes (Figs. 6 and 7). This multifaceted role parallels that of other MYB genes in crops: GmMYB118 improves soybean drought resistance by modulating ROS homeostasis¹¹¹, AtMYB60 regulates stomatal aperture in Arabidopsis¹¹², and Robust Root System 1 (OsRRS1) represses root growth to enhance drought resistance¹¹³. Given its multifaceted role in different tissues, the tissue-specific expression pattern of TaMYB7-A1 and its transcriptional response to environmental water conditions would be an interesting research direction in the future.

TaMYB7-A1’s regulation of core physiological processes—stomatal conductance, photosynthesis, root system architecture, and osmoprotectants accumulation—positions it as a hub for balancing water conservation and carbon gain under drought. The phenotypic effects of TaMYB7-A1 are diverse and physiologically integrated. To further dissect its mechanisms for yield improvement, it is worthy to use chemical inhibitors that disrupt specific pathways—such as ABA antagonists, DPI (NADPH oxidase inhibitor), catalase inhibitors, and jasmonate treatments (JA or MeJA)—to differentiate the contributions of stomatal control, root development, and antioxidant regulation. Furthermore, exploring allelic variation of TaMYB7-A1 in natural populations may guide allele mining and marker-assisted selection strategies for drought-resilient breeding.

Beyond MYBs, several other TFs have been implicated in drought resistance in wheat, such as TaNAC071-A²⁴, TaDTG6-B²⁶, and TabZIP156²⁹ via ABA-dependent or ABA-independent pathways. A deeper understanding of the interactions among key factors will help elucidate the modular network architecture of WUE regulation in wheat, supporting the design of breeding that integrates multiple genes for maximal effect, and targeting such master regulators may offer more robust breeding strategies than focusing on single downstream effectors.

In conclusion, our study demonstrates the power of multi-omics integration in elucidating the regulatory basis of WUE and drought resilience in wheat. By linking DNA variants, gene expression, and phenotypic variation, we identified 85 high-confidence candidate genes—including TaMYB7-A1—that contribute to trait resilience under water deficit. Our findings highlight that transcriptional plasticity, regulatory hotspots, and master regulators collectively shape adaptive phenotypes. These insights pave the way for precision breeding and gene-editing approaches to develop wheat cultivars optimized for future climate challenges.

Methods

Plant materials and growth conditions

The nationwide collection of 228 wheat accessions in the GWAS panel was cultivated in sealed sand plots for 3 weeks at Shijiazhuang, Hebei Province (37.89°N, 114.69°E) during 2020 and 2021 (Supplementary Data 2). The T₃ generation of TaMYB7-A1 overexpression transgenic lines—generated by amplifying the coding sequence, integrating into pLGY-UBI vectors, and transforming into cv. Fielder⁷⁸—was confirmed through genotyping and expression analysis. These wheat transgenic lines and gene-indexed KN9204 TILLING mutants (Supplementary Data 13) were multiplied and tested in a greenhouse under conditions of 22 °C, 16 h/18 °C, 8 h (light/dark), and 40% relative humidity. The yield-related traits were evaluated during harvest season on the field of the Institute of Genetics and Developmental Biology at Changping (40.22°N, 116.23°E), Beijing. Tobacco (Nicotiana benthamiana) was grown in a mixed soil of vermiculite, nutrient soil, and floral soil in a ratio of 1:1:1 for luciferase reporter assay in a greenhouse.

Water stress treatment across different experimental systems

For the water stress experiment on wheat population accessions, three water gradient treatments were ultimately performed: 15% RSWC (well-watered, WW), 9% RSWC (moderate drought stress, DS1), and 6% RSWC (severe drought stress, DS2). Each experiment consisted of 11 plants per pot with three replicates. Shoot dry weight (DW.S), root dry weight (DW.R), RSA, and water use efficiency (WUE_p, calculated as dry biomass per unit water consumed) were measured with excluding outliers, and collectively referred to as WUE-related traits.

To evaluate drought resistance at the seedling stage, the method of water withholding and then rewatering was used³⁶. Germinated seeds of TaMYB7-A1 transgenic lines alongside its wild type (WT, cv. Fielder) were transplanted into cultivation boxes (32 × 16 × 12 cm, length × width × depth), and germinated seeds of KN9204 TILLING mutants were transplanted into square boxes (8 × 8 × 8 cm, length × width × depth), which all boxes were filled with a uniform soil mixture (Pindstrup substrate to vermiculite in a ratio of 1:1) in the greenhouse. Soil moisture content was monitored every 3 days using a Moisture Hygrometer (LY-201), and rewatering was carried out when the soil moisture dropped to ~4% with recording SR and FW 7 days later. 28 plants per line in cultivation boxes and 36 plants per mutant in square boxes were compared in each test.

To assess each seedling morphological and physiological phenotypes, germinated seeds were transplanted into chain paper tube nursery bags (11 × 8 cm, diameter × depth) filled with a uniform soil mixture (Pindstrup substrate to vermiculite in a ratio of 1:1). When soil moisture content decrease to ~20% (defined as water-limited condition, WL), materials were sampled for morphological phenotype, physiological measurements, expression analysis and RNA-seq, and were compared with those under well-watered (WW) condition.

To further examine the expression profiles of targeted genes, 1-week-old seedlings of TaMYB7-A1 transgenic lines and WT were transferred to 15% PEG-6000 (m/v) treatment. Samples were collected, immediately frozen in liquid nitrogen, and stored at −80 °C for RT-qPCR analysis.

The trait evaluation was also applied in the field: transgenic lines were planted in six 1.5 m-long rows under both WW and WL conditions. For the WL treatment, irrigation was withheld after the flowering stage, while management of the WW condition followed local cultivation practices (watering 3–4 times).

Phenotype evaluation under water stress conditions

Various morphological and physiological phenotypes were examined under WW and WL conditions. Leaves were detached and incubated at room temperature, and the relative water loss rate was calculated by weighing each sample at the specified time points. The leaf temperature was captured using an InfraRed Camera R500 and analyzed with InfReC Analyzer NS9500 Standard software. The Fv/Fm was measured using a MINI-PAM-II after dark adaptation and analyzed with WinControl-3 software. Photosynthetic parameters were assessed using the LI-6400 Portable Photosynthesis System, and the instantaneous water use efficiency (WUE_i) was calculated as the ratio of net CO₂ assimilation rate to transpiration rate. After photographing whole seedlings, the maximum length and FW of shoots and roots were measured separately. Roots were thoroughly washed and scanned using an Epson Expression 13000XL, and root system architecture traits were then extracted using WinRHIZO 2008 software.

Relevant physiological phenotypes were measured using spectrophotometry following the provided instructions with minor adjustments, including MDA content (BC0025, Solarbio), root vitality (BC5275, Solarbio), proline content (BC0295, Solarbio), soluble sugar content (BC0035, Solarbio), H₂O₂ content (BC3595, Solarbio), and O₂⁻ content (BC1295, Solarbio). To evaluate ROS accumulation, seedlings were stained with DAB (G4815, Solarbio), and grayscale values were extracted using ImageJ (v1.53k), where higher grayscale values indicate lower ROS levels.

Genotype obtaining and filtering

Genotyping of the 228 wheat accessions was performed using the Affymetrix Wheat660K SNP arrays by CapitalBio Corporation. High-quality SNP markers were filtered based on the following criteria: (1) Minor allele frequency ≥ 5%; (2) Missing genotype rate ≤ 10% across the population; (3) Heterozygosity rate ≤ 5%; (4) Uniquely mapped to the reference genome IWGSC RefSeq v1.0. A total of 323,741 high-quality SNPs were retained for further analysis⁵¹.

In addition to genotyping with the Affymetrix Wheat660K SNP arrays, Whole-Exome Sequencing (WES) was performed on 110 selected representative accessions by Tcuni (Chengdu, China). Quality control of raw reads was conducted using fastp (v0.22.0), followed by alignment to the IWGSC Refseq v1.1 using BWA (v0.7.17). Reads with “QUAL < 100.0, GQ < 20.0, FS > 60.0, ReadPosRankSum < –8.0, QD < 2.0, MQRankSum < –12.5, MQ < 40.0 and SOR > 3.0” were removed. BAM files were sorted, and PCR duplicates were removed using samtools (v1.6). SNP calling followed the GATK (v4.1.8.1) best practice workflow. The “HaplotypeCaller” and “GenotypeGVCFs” were used to generate and merge each sample to all samples with a raw VCF file, and the resulting VCF file was further filtered using bcftools (v1.9) with “DP < 5 or 3 (Homozygous depth < 5, Heterozygous depth < 3), and GQ < 3.”

The final combined genotype dataset for the 110 selected representative accessions integrated SNP data from the Affymetrix Wheat660K SNP arrays and WES. This combined dataset was further filtered using bcftools (v1.9) to exclude SNPs with a missing rate > 10%, minor allele frequency < 5%, and heterozygosity rate > 50%. After filtering, a total of 697,570 high-quality SNPs were retained for subsequent analysis.

Genome-wide association study (GWAS) analysis

Wheat population structure was evaluated using the Structure (v2.3.4) programs. The neighbor-joining (NJ) phylogenetic tree was constructed using MEGA7 software and visualized with the ggtreeExtra (v1.8.1) package in R. Principal components analysis (PCA) was conducted with GAPIT and visualized using the scatter3d function in R. LD decay was calculated and visualized using PopLDdecay (v3.42), and LD blocks were visualized with LDBlockShow (v1.40). GWAS analysis between high-quality SNPs and WUE-related traits was performed using a mixed linear model implemented in GEMMA (v0.98.3)¹¹⁴, with PCA calculated via the GAPIT package (v3.4.0) as a fixed effect and the Kinship matrix as a random effect. The P value threshold was calculated using a modified Bonferroni correction (Genetic Type I Error Calculator, v0.2) with a suggested threshold of P = 1/Ne (Ne = effective SNP number)¹¹⁵. Our results indicated that the suggested P value thresholds at the chromosome level ranged from 1.95e-04 to 8.87e-04 (Supplementary Data 3), and thus, we considered the value of 1.00e-04 as the criterion for genome-wide significance in this study^41,116. The candidate QTLs were merged within 3-Mb regions around significantly associated SNPs and required to be detected in more than one environment. Manhattan plots were generated using the CMplot package (v4.5.1) and the ggplot2 package (v3.5.2) in R.

RNA extraction, sequencing, and population-transcriptome analysis

Total RNA (110 selected representative wheat accessions, three water gradient stresses, root and shoot two tissues) was extracted using the Quick RNA Isolation Kit (2407#22D, HUAYUEYANG BIOTECHNOLOGY CO., LTD), following the manufacturer’s instructions. Next, the 3′ RNA-seq libraries were constructed⁴⁵. In order, fragmenting RNA (~2 μg), synthesizing the first-strand cDNA and the double-stranded DNA, performing end repair, dA-tailing, and adapter ligation, and finally PCR amplification. The libraries were sequenced on the NovaSeq platform (Illumina) for 150 bp paired-end reads. Raw data were processed using fastp (v0.22.0) to remove adapters and low-quality bases, and clean data were then mapped to the IWGSC Refseq v1.0 reference genome using hisat2 (v2.2.1), resulting in 638 high-quality RNA-seq samples for subsequent analysis (Supplementary Data 5). Gene count matrices were generated using featureCounts (v2.0.3), and the expression level [Counts Per Million (CPM)] was calculated for each gene by the edgeR package (v3.42.4) in R.

Based on the distribution characteristics of the data, genes with decile expression levels more than 1 (CPM > 1) across the population were considered expressed, and then were transformed to a normal distribution using the R function “qqnorm” (v3.6.1) for subsequent analysis. DEGs were identified using the edgeR package (v3.42.4) with the criteria (Log₂FC > 1, P value < 0.05) and GFOLD (v1.1.4) (Log₂FC > 1), which occurred in at least 15% of accessions under any water stress condition, respectively⁴⁸. K-means clustering and heatmap visualization were conducted using the ComplexHeatmap package (v2.14.0) in R, and GO enrichment analysis was performed with the clusterProfiler package (v4.6.2) in R.

Expression quantitative trait locus (eQTL) and hotspot analysis

The mixed linear model implemented in the MatrixEQTL package (v2.3)¹¹⁷ and GEMMA (v0.98.3), which, considering the population structure, genetic relatedness, and estimated confounding factors, was simultaneously used to perform association analysis between the combined genotype data and the normalized expression profiles of expressed genes. The significantly associated SNPs (P value < 4.11e-07, determined jointly by 0.05/effective SNP number) were identified using MatrixEQTL and GEMMA¹¹⁵. A revised two-step method was employed to address associations involving multiple SNPs for a single gene expression and to define eQTL region: (1) all of the associated SNPs were grouped into one cluster if the distance between two consecutive SNPs was less than 10 Kb, and the clusters with at least three significantly associated SNPs were considered as candidate eQTLs represented by their lead SNP which had the most significant P value. (2) Candidate eQTLs in LD (r² > 0.1) with other more significant candidate eQTLs for the same expression trait were identified as LD-driven associations and subsequently removed. The cutoff of identifying cis-eQTLs and trans-eQTLs was set at 2.4 Mb, which is the LD decay distance of the combined genotype dataset.

eQTL hotspot was considered as a genomic region that regulates the expression of a certain number of genes, and the gene number threshold was calculated from the distribution of the maximum number of associated genes in 1000 random permutations (P value < 0.01). The Circos plot was displayed using TBtools-II (v2.309).

Weighted correlation network analysis (WGCNA)

The weighted correlation network analysis (WGCNA), based on the concept of scale-free networks, was applied to identify gene modules with highly correlated expression patterns among population accessions¹¹⁸. The analysis was performed using the WGCNA package (v1.72.5) in R following these steps: removing outlier samples and genes, calculating similarity matrix, determining soft threshold (SFT, β power), converting similarity matrix to adjacency matrix, converting adjacency matrix to topological overlap matrix (TOM), constructing modules by one-step method (minModuleSize = 50, mergeCutHeight = 0.25), analyzing correlation between modules and phenotype traits, excavating interested modules and hub genes, and visualizing co-expression network by Cytoscape (v3.7.2).

Summary-data-based mendelian randomization (SMR) analysis

After data formatting and preparation, SMR analysis (v1.3.1)¹¹⁹ was applied to evaluate the association between gene expression and trait variation under different environments. The summary-level statistic was analyzed using 10,000 Kb and 5000 Kb windows for cis-eQTLs and trans-eQTLs, respectively. Heterogeneity in dependent instruments (HEIDI) outlier test was applied to separate pleiotropy, which association due to linkage. SMR associations were declared significant if the SMR P value < 0.01 and the HEIDI-test P value > 0.05^120,121.

Construction of the genomic co-localization network and transcriptional regulation network

The GWAS bin was defined as the candidate interval formed by extending 3 Mb around significant SNPs (based on the LD decay distance), around each SNP surpassed the significant threshold in the GWAS analysis. The co-localization network incorporated two types of relationships: (1) the significant SMR-associated genes were located directly within the GWAS bin; (2) the significant SNPs within the GWAS bin exhibited eQTL regulatory effects on the SMR-associated causal genes. Each trait was analyzed independently, and a physical location-based co-localization network was ultimately constructed.

For candidate genes identified through co-localization and DEGs of population-level transcriptome, a potential transcriptional regulatory network was constructed based on the significant expression correlation (P value < 0.05) and predicted TF motif binding (P value < 1.00e-04 in FIMO scanning) across different tissues. The matched motif information used for TF-target prediction was obtained from the PlantTFDB database (https://planttfdb.gao-lab.org/).

Expression analysis

Total RNA was extracted using the Quick RNA Isolation Kit (2407#22D, HUAYUEYANG BIOTECHNOLOGY CO., LTD). First-strand complementary DNA (cDNA) was synthesized from 2 μg of DNase I-treated total RNA using the FastKing RT kit (KR116, TIANGEN BIOTECH CO., LTD). RT-qPCR was performed using the ChamQ Universal SYBR qPCR Master Mix (Q711-03, Vazyme Biotech Co., Ltd., China) by QuantStudio^TM 5 (Applied Biosystems, Thermo Fisher Scientific). The relative expression levels of target genes were normalized to the TaTubulin (TraesCS1D02G353100) using the 2^−ΔΔCT method, and primer sequences were listed in Supplementary Data 16.

RNA-seq analysis of TaMYB7-A1 transgenic lines

The shoot and root samples under WW and WL conditioin were performed, and libraries were sequenced using a DNBSEQ-T7 platform (BGI-Shenzhen, China) for 150 bp paired-end reads. After adapters removing, low-quality reads filtering, reference genome mapping (IWGSC Refseq v1.1), and counts calculating, transcripts per million (TPM) were normalized. The PCA and cluster heatmaps were performed and visualized by FactoMineR (v2.11) and ComplexHeatmap package (v2.14.0) in R. DEGs were identified using DESeq2 (v1.34.0) in R with the thresholds of “Log₂ FC > 1 and P.adjust < 0.05.”

The evolutionary tree was constructed in MEGA7 on sequence similarity to identify matched motif for scanning potential downstream genes targeted by TaMYB7-A1. FIMO (v5.5.5) was used to scan the 2-Kb promoter sequences of TaMYB7-A1-regulated genes with default parameters (P value = 1.00e-04). Further refining the list of candidate downstream targets, genes were also evaluated for whether differential expression in population-transcriptome and response to drought based on publicly available datasets (SRP045409 and SRP068165)¹²².

Luciferase reporter assay and electrophoretic mobility shift assay (EMSA)

The 2.5 Kb promoters of TaPIP2;2-B1, TaRD20-D1, and TaABCB4-B1 were amplified from Chinese Spring (CS), and fused into the CP461-LUC vector to construct the ProTaPIP2;2-B1::LUC, ProRD20-D1::LUC, and ProTaABCB4-B1::LUC reporter vectors. The coding sequence of TaMYB7-A1 from CS was cloned into the pSUPER-GFP vector (PRI101) as an effector. These effector and reporter plasmids were transformed into Agrobacterium GV3101 and co-infiltrated into Nicotiana benthamiana leaves (at 6–8 leaf stage) in various combinations. After being cultivated in a greenhouse at 22 °C temperatures and a 12 h light/12 h dark cycle for 2–3 days, the Firefly luciferase (LUC) and Renilla luciferase (REN) activities were measured using the Dual-Glo^® Luciferase Assay System (E2940, Promega Biotech Co., Ltd). Fluorescence signals were quantified using the dual-LUC assay reagent on a SpectraMax iD3 (MOLECULAR DEVICES), and relative LUC activity was calculated as the ratio of LUC to REN. Primer sequences used for vector construction were listed in Supplementary Data 16.

For EMSA analyses, TaMYB7-A1 proteins were fused with MBP tags and transformed into E. coli BL21. After 24 h induction at 28 °C with 0.5 mM IPTG, proteins were purified using MBP beads, transferred to a nylon membrane, and detection of biotin-labeled probes (synthesized at BGI, sequences listed in Supplementary Data 16) was performed using LightShift Chemiluminescent EMSA Kit (20148, Thermo Scientific) according to the manufacturer’s instructions.

Statistics and data visualization

Unless specified, R (https://cran.r-project.org/; version 4.2.2) and GraphPad Prism 8 were used to compute statistics and generate plots. Pearson correlation coefficient analysis was used in Figs. 1h, 2f and 5e, Supplementary Figs. 1b, c, 5d, 6d, e and 13. The Fisher’s exact test was used in Figs. 2d, e, h, 3d and 6c, Supplementary Figs. 5c and 6f. The two-sided Student’s t-test was used in Figs. 1a, 4f, 5c, f, h–j, 6e, g, 7a–d, g, h, Supplementary Figs. 12a–d, 15b, c and 16a, b. The Wilcoxon rank-sum test was used in Figs. 1f, 2b, g, i, 3c, e, f, h and 5d, Supplementary Figs. 3c, 4b–d, 5b, 6g and 10a, b. Tukey’s HSD multiple comparisons test was used in Fig. 6f and Supplementary Fig. 1d.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.