Comparison of multi-stress resilience in wild and domesticated Cowpea

Abstract

A common hypothesis in plant science is that wild relatives of crops possess greater resilience to multiple stresses because they retain adaptive traits that have been diminished through domestication. Such reductions in resilience in cultivated crops are often attributed to trade-offs, whereby selection for agronomic performance decreases investment in stress tolerance. Alternatively, weakened multi-stress resilience may arise from relaxed natural selection in managed environments, where crops experience fewer ecological pressures. Vigna unguiculata, cowpea, is crucial for food security in arid and semi-arid regions due to its inherent stress tolerance, low input requirements, and nitrogen-fixing capacity. To test whether wild cowpea genotypes exhibit stronger multi-stress resilience than cultivated forms, we conducted a controlled full-factorial experiment using geographically and genetically diverse germplasm. Fourteen wild and cultivated genotypes spanning gradients in temperature, precipitation, and dry-season length were exposed to simulated drought and herbivory. We quantified growth responses and assessed trait variability using the coefficient of variation. Drought, herbivory, and their interaction significantly reduced productivity across all Vigna genotypes. However, wild types exhibited lower phenotypic variability and more consistent performance across stress treatments, suggesting that domestication may have reduced the stability of multi-stress responses. These findings improve our understanding of stress adaptation in cowpea and highlight the value of wild germplasm for breeding climate-resilient legumes.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-35860-4.

Introduction

Sustaining and developing resilient food systems that provide sustainable food security will be one of the greatest challenges of the 21st century, as climate change already poses and will further pose a serious threat to farming systems globally^1,2. Extreme weather events such as droughts and floods, rising temperatures, increasing soil salinity, and emerging patterns of pests and diseases place significant stress on crops and cropping systems, leading to reduced productivity and yield losses^3,4.

In this context, increasing attention is being paid to the use and integration of underutilised and neglected crops to enhance climate resilience and thus food security in local food systems^5,6. As these crops and varieties are often well adapted to specific environmental conditions and stresses, identifying local varieties with high plasticity can enhance both crop resilience and agro-biodiversity^6–8.

The traditional view of domestication often emphasizes trade-offs: as plants are selectively bred for desirable agronomic traits such as yield, uniformity, or palatability they may lose ancestral abilities to withstand environmental and biological stresses^9–11. This trade-off model suggests that improved performance under managed conditions comes at the expense of resilience to specific stressors, both biotic (e.g., pests, pathogens) and abiotic (e.g., drought, heat)^12,13. However, this model may oversimplify the complexities of stress adaptation⁹. The reduction in multi-stress resilience during domestication may not result from direct trade-offs, but rather from the passive loss of traits due to the relaxation of natural selection in controlled agricultural environments. In these settings where irrigation, pest control, and soil management buffer environmental extremes there is less evolutionary pressure to retain diverse stress-resistance mechanisms¹⁴. Thus, the decline in resilience may be better understood as a consequence of shifting selective environments rather than inevitable trade-offs¹⁵.

Crop responses to multiple stresses differ from their responses to single stresses¹⁶. Multi-stress resilience in crops refers to a plant’s ability to maintain stable growth and reproduction when simultaneously or sequentially exposed to multiple environmental stressors, both abiotic (e.g., drought, heat, salinity) and biotic (e.g., pests, pathogens)¹⁷. This resilience involves complex physiological, morphological, and molecular responses that allow the crop to buffer or recover from stress impacts without significant performance loss. Unlike tolerance to a single stressor, multi-stress resilience reflects a broader, more integrated capacity to function under variable and often unpredictable conditions reflecting real-life agricultural conditions^18,19.

Experimental studies involving multiple and combined stress scenarios can help uncover the true extent of resilience in wild germplasm and clarify the domestication process. Under this hypothesis, wild relatives are expected to exhibit broad resilience to both biotic and abiotic stressors, shaped by long-term exposure to fluctuating and challenging environments. These genotypes may not be specialized for any single stress but rather possess generalized robustness against diverse challenges such as drought, pests, and nutrient-poor soils. In contrast, domesticated varieties may show narrower stress responses, often non-linear and highly context-dependent, further challenging the simplicity of trade-off models. By testing both wild and cultivated genotypes under simultaneous stressors (e.g., drought + pathogen, heat + herbivory), we can assess whether wild lines are more resilient to all stress types. This approach also opens opportunities to identify stress-resilient genotypes for reintegration into breeding programs.

Due to their adaptability to diverse climatic and agronomic conditions, short life cycles, as well as their multiple uses, Vigna species are increasingly recognized in agricultural and nutrition research as climate-resilient crops that are essential to local food and production systems, particularly in rural (sub-) tropical regions^20–22. Their adaptation to harsh environmental conditions, low input requirements, and nitrogen-fixing capacity, combined with their nutritive content, makes them especially valuable in subsistence farming systems^23,24. The genus Vigna, a member of the Fabaceae family, encompasses between 100 and 150 species indigenous to the (sub-) tropical regions of both the Old and New World. Ten species have been domesticated primarily in Africa and Asia, serving as crucial food crops, forage, and cover crops, holding considerable economic importance in developing countries^25,26. In addition to their climate resilience, their use as natural fertilizers and catch crops makes them especially suitable for low-input systems, including subsistence farming systems, making them an essential component of many local agricultural and food systems across sub-Saharan Africa and Asia^23,25. Although species within the genus Vigna are primarily used for food, feed, they also serve many specific purposes in diverse geographic and cultural contexts. Depending on the setting, different edible parts (pods, seeds, leaves, and even roots) are consumed based on local culinary preferences, resulting in different cultivation lines and breeding efforts. While most Vigna consumption stems from their production in predominantly subsistence farming systems, Vigna species are also harvested from the wild in many areas, classifying them both as underutilized crops and wild edible plants^2,27.

Most cowpea production in sub-Saharan Africa and Asia is carried out by smallholder farmers under marginal conditions²⁸. To address the challenges posed by climate change for these systems, it is essential to adapt agricultural practices and incorporate crop varieties with resistance traits into breeding programmes²⁹. Wild relatives of Vigna are known to possess such traits, having evolved under a wide range of environmental conditions and biotic pressures. Through continuous adaptation to local climates, these species have developed mechanisms to tolerate drought, such as altered flowering times and the ability to maintain leaf water content under stress^30–32. In addition to drought, insect herbivory is a major constraint on Vigna production. Key pests include leaf-feeding beetles, several Spodoptera caterpillar species, and other flower and foliar pests³³. In response to these pressures, wild Vigna species have evolved a wide range of defensive strategies against insect herbivores^34,35. This natural diversity in defence traits makes wild Vigna particularly valuable for improving food security in semi-arid regions of sub-Saharan Africa²⁴. Incorporating the genetic diversity of wild Vigna into cultivated lines therefore offers substantial potential to enhance the adaptive capacity of existing breeding materials to both abiotic and biotic stressors^36,37.

Vigna unguiculata (L.) Walp., commonly known as cowpea, is a domesticated species that thrives in temperatures ranging from 25 °C to 35 °C and annual rainfall between 750 and 1100 mm. It can grow in various soil types and is predominantly cultivated in warm climates across the tropics of Africa, South Asia, and Latin America. The high number of wild subspecies of V. unguiculata found in Africa supports the hypothesis that the species originated on the continent³⁸. This is especially evident in the southernmost regions of Africa, where the greatest concentration of subspecies and genetic diversity is found, and where more diversity may yet remain undiscovered^27,39,40. Taxonomic revisions based on morphological and molecular traits have identified 10 perennial and 1 annual subspecies. The annual subspecies is further divided into two varieties: ssp. unguiculata var. unguiculata (the domesticated cowpea) and ssp. unguiculata var. spontanea (Schweinf.). The latter, var. spontanea, typically grows near cultivated fields and roadsides and is recognized as the wild progenitor of domesticated cowpea⁴⁰.

This study investigates the drought and herbivory tolerance of both wild and cultivated Vigna unguiculata genotypes. By evaluating plant performance under drought stress, herbivory, and their combined effects. Specifically, we assess how these stressors influence key growth traits across diverse Vigna genotypes, compare responses between wild and cultivated forms, and explore the extent to which environmental origin shapes adaptive capacity. These insights will contribute to a deeper understanding of plant defence strategies and inform breeding programs targeting improved resilience in legume crops.

Results

Climatic variables and trait coefficients of variation (CVs)

We tested whether environmental origin predicted genotype-level variability in proportional stress responses, quantified as CV(log-ratio). Cultivated accessions showed no significant correlations between dry-season length and CV(log-ratio) for any trait (biomass: R = − 0.6; length: R = 0; leaves: R = − 0.6; shoots: R = − 0.6; Fig. 1A,C,E,G). Wild accessions exhibited consistently positive but mostly non-significant associations (biomass: R = 0.52, p = 0.12; length: R = 0.52, p = 0.12; leaves: R = 0.41, p = 0.24; Fig. 1B,D,F). A significant correlation was detected for shoot number (R = 0.71, p = 0.021; Fig. 1H), indicating higher variability in shoot responses in genotypes from more seasonal environments. No significant relationships were found between maximum rainfall and CV(log-ratio) for cultivated (Fig. 2A,C,E,G) or wild accessions (Fig. 2B,D,F,H). A correlation heatmap confirmed the overall weakness of climate effects (Supplementary Fig. S1).

Fig. 1.

Relationship between dry-season length (months) and the coefficient of variation of log-ratios (CV of log-ratio) for four vegetative traits in Vigna unguiculata (Biomass, Length, Number of leaves, Number of shoots). Panels show cultivated (left column) and wild accessions (right column) for each trait. CV(log-ratio) was calculated as the variability of log-responses of each genotype to the three stress treatments (Herbivory, Drought, Combined) relative to its own control. Points represent genotypic means; lines show linear regressions with 95% confidence ribbons. Spearman correlation coefficients (R) and associated P-values are shown in each panel. Colours indicate domestication type.

Fig. 2.

Relationship between maximum monthly rainfall (mm) at the collection site and the coefficient of variation of log-ratios (CV(log-ratio), for four vegetative traits in Vigna unguiculata (Biomass, Length, Number of leaves, Number of shoots). Panels show cultivated (left column) and wild accessions (right column). CV(log-ratio) quantifies genotypic variability in log-responses to the three stress treatments relative to control conditions. Points represent genotypic means; lines show linear regressions with 95% confidence ribbons. Spearman correlation coefficients (R) and P-values are displayed in each panel.

Effects of drought and herbivory

Across all vegetative traits, i.e., biomass, shoot length, number of leaves, and number of shoots, stress treatments produced large decreases in performance (Figs. 3, 4, 5 and 6a). Bayesian hierarchical models detected strong and credible treatment effects (i.e., a 95% credible interval that excludes zero, so the effect is statistically meaningful) for raw values, log-ratios, and genotype-level coefficients of variation (CV), with all 95% credible intervals excluding zero (Supplementary Table S2). Biomass declined markedly under herbivory, drought, and combined stress, with the Bayesian model estimating a median treatment effect of − 4.41 (95% Credible Interval CrI: −7.93 to − 1.32; Fig. 3A). Shoot length showed the largest absolute reductions (median effect = − 11.14, 95% CrI: −12.73 to − 9.60; Fig. 4A), while leaf number (− 2.28, 95% CrI: −2.87 to − 1.70; Fig. 5A) and shoot number (median effect = − 0.97, 95% CrI: −1.11 to − 0.83; Fig. 6A) also decreased consistently across all stress treatments. Log-ratio distributions (Figs. 3B, 4, 5 and 6B) confirmed substantial proportional performance losses relative to control values (e.g., biomass log-ratio effect = 1.19, 95% CrI: 0.41–4.06; shoot length: 0.25, 95% CrI: 0.20–0.30; leaves: 0.10, 95% CrI: 0.08–0.13; shoots: 0.20, 95% CrI: 0.17–0.23).

Fig. 3.

(A) Violin–boxplots show the distribution of Biomass for Cultivated (red) and Wild (blue) accessions across the four treatments. Violin shapes represent the full data distribution, while boxplots summarize medians and interquartile ranges. (B) Density plots show the distribution of Biomass log-ratios, which express each plant’s proportional change in biomass relative to the mean control value within its domestication type. This standardizes responses and allows comparison of treatment effects across types. (C) Boxplots display the Coefficient of variation (CV) of biomass computed within each Genotype × Type × Treatment group, summarizing standardized variability in biomass responses across treatments.

Fig. 4.

(A) Violin–boxplots show the distribution of Length for Cultivated (red) and Wild (blue) accessions across the four treatments. Violin shapes represent the full data distribution, while boxplots summarize medians and interquartile ranges. (B) Density plots show the distribution of Length log-ratios, which express each plant’s proportional change in biomass relative to the mean control value within its domestication type. This standardizes responses and allows comparison of treatment effects across types. (C) Boxplots display the Coefficient of variation (CV) of Length computed within each Genotype × Type × Treatment group, summarizing standardized variability in biomass responses across treatments.

Fig. 5.

(A) Violin–boxplots show the distribution of No. of leaves for Cultivated (red) and Wild (blue) accessions across the four treatments. Violin shapes represent the full data distribution, while boxplots summarize medians and interquartile ranges. (B) Density plots show the distribution of No. of leaves log-ratios, which express each plant’s proportional change in biomass relative to the mean control value within its domestication type. This standardizes responses and allows comparison of treatment effects across types. (C) Boxplots display the Coefficient of variation (CV) of No. of leaves computed within each Genotype × Type × Treatment group, summarizing standardized variability in biomass responses across treatments.

Fig. 6.

(A) Violin–boxplots show the distribution of No. of shoots for Cultivated (red) and Wild (blue) accessions across the four treatments. Violin shapes represent the full data distribution, while boxplots summarize medians and interquartile ranges. (B) Density plots show the distribution of No. of shoots log-ratios, which express each plant’s proportional change in biomass relative to the mean control value within its domestication type. This standardizes responses and allows comparison of treatment effects across types. (C) Boxplots display the Coefficient of variation (CV) of No. of shoots computed within each Genotype × Type × Treatment group, summarizing standardized variability in biomass responses across treatments.

Trait stability, as indicated by the CVs, also decreased under stress. Genotype-level CVs increased across treatments for biomass (median effect = 5.06, 95% CrI: 1.70–8.86; Fig. 3C), length (median effect = 9.60, 95% CrI: 5.62–13.55; Fig. 4C), leaves (median effect = 3.40, 95% CrI: 1.37–6.18; Fig. 5C), and shoots (median effect = 2.93, 95% CrI: 1.00–5.80; Fig. 6C), indicating lower performance consistency across stress combinations.

Effect of type of genotype

Domestication type had a consistent effect across traits (Supplementary Table S2. Supplementary Figure S2). Wild accessions exhibited higher performance than cultivated ones across all treatments for biomass (median effect = 0.50, 95% CrI: 0.03–1.80), shoot length (median effect = 2.19, 95% CrI: 0.12–6.07), leaf number (+ 1.72, 95% CrI: 0.16–3.64), and shoot number (median effect = 0.43, 95% CrI: 0.05–0.93), as shown in Figs. 3, 4 and 5 A.

Wild genotypes demonstrated smaller proportional reductions in performance under stress. Log-ratio type effects were positive and statistically credible for all traits (biomass: median effect = 0.07, 95% CrI: 0.00–0.24; length: median effect = 0.08, 95% CrI: 0.00–0.29; leaves: median effect = 0.09, 95% CrI: 0.00–0.27; shoots: median effect = 0.05, 95% CrI: 0.00–0.17), indicating that wild plants experienced less severe declines than cultivated plants (Figs. 3B, 4, 5 and 6B).

Domestication type altered the magnitude of stress responses

Type × Treatment interaction effects were consistently non-zero across all traits and response types (Supplementary Table S2, Supplementary Figure S2), indicating that cultivated and wild accessions did not respond to the stress treatments in the same way. In other words, domestication status altered both the strength and the pattern of stress responses. For raw traits, interaction effects were moderate but credibly positive for all four variables, i.e., biomass (median effect = 0.43, 95% CrI: 0.13–0.86), shoot length (median effect = 2.04, 95% CrI: 0.75–3.56), leaf number (median effect = 1.76, 95% CrI: 1.25–2.29), and shoot number (median effect = 0.32, 95% CrI: 0.19–0.45). Similar patterns occurred for log-ratios (median |β| = 0.05–0.11 across traits) (Supplementary Figure S2). These interaction effects show that cultivated genotypes experienced steeper declines under stress and showed greater sensitivity to the combination of stressors than wild genotypes, which responded more uniformly across treatments. Because cultivated accessions consistently had lower trait values, the positive interaction terms indicate that their performance declined more sharply under stress, particularly under combined drought and herbivory. In contrast, wild accessions showed smaller proportional reductions and lower CVs, reflecting more uniform and stable responses across treatments.

Multivariate analysis

Multivariate PERMANOVA across the four compared traits revealed a dominant effect of stress treatment (R² = 0.682, P = 0.001), with smaller but significant effects of domestication type (R² = 0.020, P = 0.001) and the Type × Treatment interaction (R² = 0.021, P = 0.001) (Supplementary Table S3). A canonical analysis of principal coordinates (CAP) further illustrated these patterns (Supplementary Figure S3). Individuals clustered strongly by treatment along the primary constrained axis, while cultivated and wild accessions separated along a secondary axis. Biomass and shoot number contributed most strongly to the constrained ordination space, followed by shoot length and leaf number.

Discussion

Understanding how domestication has shaped crop responses to environmental challenges is critical in an era of rapidly changing climate conditions⁴¹. This study examined the long-standing hypothesis that domesticated crops, through selective breeding for agronomic traits, may have sacrificed resilience to biotic and abiotic stressors, a phenomenon often attributed to selection trade-offs. However, a complementary perspective to this is that such losses in stress tolerance may not be the result of direct trade-offs but rather the passive relaxation of natural selection under managed agricultural systems. In this context, multi-stress resilience, defined as the capacity of a genotype to maintain stable performance under simultaneous or sequential environmental stressors, offers a more realistic and integrated measure of plant robustness in fluctuating field conditions. In our study, by comparing a selection of wild and cultivated accessions across multiple traits, we aimed to test whether domesticated varieties indeed show reduced and less stable performance under stress, as predicted by the trade-off model, or whether wild genotypes demonstrate broader and more consistent stress resilience.

Compared to the control groups, drought, herbivory and their combination resulted in significant reductions in performance of all Vigna genotypes. All tested accessions showed significantly reduced biomass, shoot length, number of leaves, and number of shoots when subjected to drought and herbivory individually. When both stressors were applied simultaneously, these reductions were even more pronounced. The response to these stressors varied significantly depending on the genotype and domestication status (cultivated vs. wild). This may be related to the fact that during domestication, selection is predominantly focused on the architecture of plant species, for example fruit size or number of leaves, and phytochemical content [e.g.,^42,43].

These overall reductions in trait performance highlight the sensitivity of Vigna genotypes to both individual and combined stressors. However, beyond average performance losses, an important distinction emerged in how consistently genotypes responded to these stressors. When we examined the variability of trait responses, clear differences between wild and cultivated types became apparent. Notably, wild genotypes showed a smaller variability in response to stressors compared to cultivated genotypes. This means that, when exposed to stress, the response in traits of wild varieties was more predictable and consistent than that of cultivated varieties. Furthermore, wild varieties showed less variability in response and thus a more predictable response with an increasing length of the dry season, while this was not the case for cultivated varieties. One of the most significant indicators for climate-resilient cropping systems is growth and yield stability in variable and extreme climatic conditions^44,45. Plant performance stability can be conceptually divided into static and dynamic components. Static stability refers to a genotype’s ability to maintain consistent trait expression (e.g., growth, yield) across environments, while dynamic stability reflects the capacity to adjust trait performance proportionally and predictably in response to environmental variability⁴⁶. In the context of climate variability, such predictability is crucial as it enables farmers to anticipate plant behavior under stress and make informed management decisions. Thus, the stability of trait responses under multiple stressors can serve as a valuable proxy for climate resilience. In our study, wild Vigna genotypes demonstrated more consistent and less variable responses under stress conditions compared to cultivated ones, indicating a higher level of resilience and robustness to environmental fluctuations.

Plant performance under climate variability is increasingly assessed in major staple crops like maize, rice, and wheat using advanced prediction tools such as crop models, machine learning, and artificial intelligence [e.g., ^47,48]. These approaches enable precise forecasting and informed decision-making for large-scale, homogeneous cropping systems. However, such predictive frameworks are far less applicable to subsistence farming systems, which often rely on a wide array of locally adapted, underutilized, and neglected crops, such as Vigna, grown in highly heterogeneous environments. In these diverse and data-scarce settings, predictive modeling is limited or unfeasible. As a result, the identification and use of inherently resilient crop varieties become critical strategies for enhancing food system stability in the face of both biotic and abiotic stressors.

Resilience to climate change requires adaptation to multiple stressors and their interactions^18,19. Besides the rising global temperature, the intensification of the water cycle, leading to increasingly erratic rainfall patterns worldwide, is one major result of climate change, which increases the likelihood of both floods and droughts occurring^49,50. While heat and drought stress pose enormous threats to global agricultural systems⁵¹, a secondary consequence of climate-induced stress is that plants, including crops, become more susceptible to additional stressors, such as pests and diseases^52,53. Our results show that the most severe productivity reductions for all traits occurred in response to the interaction between drought and herbivory, compared to individual stressors. Wild genotypes showed less variability in their responses than cultivated genotypes. This suggests that multi-stressor resilience plays an important role in plants’ resilience to climate change effects. Additionally, experimental evidence proves that plant responses to combinations of stressors significantly diverge from responses to individual stresses¹⁶. In addition to focusing on crops’ and wild crops relatives’ responses to individual stresses, more research should focus on plant phenotypical responses, their variability, and the mechanisms of and trade-offs between resilience to interactions of stressors⁷. Given the lower variability in the response of the tested wild Vigna genotypes to the interaction of stressors, their multi-stress resilience and thus general robustness can be considered higher than compared to cultivated varieties.

Our results thus suggest trade-offs for cultivated genotypes between desirable traits and multi-stress resilience. Historically, breeding efforts have focused on optimizing yield in predictable conditions. However, due to the complexity, unpredictability, and location specificity of climate change effects, its unfolding challenges are expected to place greater emphasis on climate resilience in breeding programmes, in which crop wild relatives can play a crucial role^54,55. Wild Vigna genotypes should thus be considered important sources of genetic material for the development of (climate) resilient crops and cropping systems⁵⁶. They should be considered valuable crop wild relatives with great potential to increase climate change resilience of farming systems when integrated into cropping systems as such or used as sources of wild genetic material for introgression into cultivated breeding lines. As the response variability of wild genotypes decreased with increasing dry season length in our study, our findings suggest that environmental predictability, particularly water availability, plays a role in shaping trait variability, with potential implications for the evolution of drought adaptation strategies. Therefore, further research should explore this relationship, and the potential of prioritising wild genotypes already adapted to extreme precipitation patterns in breeding efforts.

Vigna germplasm collections of specific species, or of a broad diversity of species such as the collection held by the Meise Botanic Garden, including wild genotypes, hold great potential for the development of climate resilient Vigna breeding lines⁵⁷. Legume crops, including Vigna, have experienced less genetic improvement partly due to the underutilization of wild relatives by breeders⁵⁸. Identifying and characterizing new genotypes of both wild and cultivated species is essential to determine their utility as genetic resources⁵⁹. Studies like those by Takahashi et al. (2016)⁶⁰ and van Zonneveld et al. (2020)²¹ have contributed to understanding the phylogenetics and cross-compatibility of Vigna species, facilitating their categorization within gene pools. However, these collections are often not phenotyped at all or only to a limited extent. The results of this research highlight the importance of phenotyping these collections to assess and compare the responses of different genotypes under varying conditions, including exposure to multi-stressors that are relevant in current agricultural and climatic contexts. Apart from the phenotyping of the major staple crops and some globally economically important vegetables, such as tomato^61,62, large scale phenotyping of germplasm collections of underutilized and neglected crops, could validate existing collections and unlock their potential contribution to the development of climate-resilient crops, and food security in a changing climate.

Conclusion

This study provides insight into the responses of Vigna genotypes to drought, herbivory, and their interaction, with the selected wild genotypes showing more predictable responses to stressors than cultivated ones. These findings suggest tradeoffs between domestication and stress resilience, suggesting that selection for productivity may reduce adaptability under variable conditions. Predictability in trait responses, particularly under multiple stressors, is crucial for ensuring yield stability, especially in subsistence and low-input farming systems affected by climate change. Systematic phenotyping and the inclusion of wild genotypes in breeding programs and farming systems should be considered for building sustainable and resilient food systems in the face of climate change.

Methods

Selection of vigna accessions

Fourteen Vigna accessions (= populations) were selected for this study i.e., four cultivated and ten wild relatives. These accessions were selected from the germplasm bank of Meise Botanic Garden, with a focus on the rainfall conditions and length of the dry season in their collection locations, as provided by a climatic database⁶³. African and Asian genotypes were selected from regions that experienced scanty rainfall and longer drought periods, suggesting potential for better growth under dry conditions (Supplementary Table S1). Seeds of the different accessions were stored for up to 10 years at 15% RH and − 20 °C, before being used. Previous experiences have shown that effects of storage under optimal conditions on seed germinability and plant vigour are negligible³².

Seed germination

To speed up and synchronise seed germination, seeds were first scarified to break dormancy using a scalpel. They were then placed between two filter papers in a petri dish, moistened with tap water and placed in an incubator. The incubation conditions were set at 30 °C during the day and 20 °C at night for a 16-hour cycle, continuing until sufficient germination was achieved. Per accession, four Petri dishes with 20 seeds each were germinated under controlled conditions. The germinated seeds selected for the experiment had all germinated within two weeks after imbibition.

Experimental design and conditions

The experiment was conducted in the controlled greenhouse facilities of ILVO and Ghent University (Flanders Research Institute for Agriculture, Fisheries and Food, Melle, Belgium). Supplemental lighting (Philips Hortigrow 600 W) was automatically activated when natural light fell below 100 W/m², for up to 10 h per day. Night-time temperatures were kept at 17 °C, while daytime temperatures generally exceeded 21 °C due to solar radiation and the supplemental lighting.

On 17 January 2023 (day 0), more than 60 seeds per accession were sown, from which 60 uniform and healthy seedlings were selected and transplanted into 400-ml pots filled with Peltracom “Universale potgrond.” The plants were randomly assigned to four treatments i.e., control (C), drought (D), herbivory (H), and combined drought–herbivory (DH), with 10 plants per accession per treatment. To minimise positional effects in the greenhouse, pots were arranged on four flooding tables following a 4 × 4 Latin square design, in which each treatment appeared exactly once on each table and once in each row position. This layout ensured that any small spatial variation in light or microclimate was evenly distributed across treatments. During the first 36 days (17 January–22 February), all plants were maintained under identical conditions and watered 2–3 times per week to keep soil moisture close to field capacity. As plants grew, bamboo stakes were added to support climbing.

Drought stress was initiated on 22 February (day 36) by stopping flooding for all D and DH pots. Soil moisture was monitored using a pot-weight approach following Verheyen et al. (2024)³². Each pot was weighed at saturation to determine its fully watered weight, and the wilting threshold was defined as the pot weight at which reversible leaf wilting first appeared. When drought-treated pots approached this threshold, a small volume of water (20–60 ml, depending on plant size) was supplied to bring them just above the threshold and prevent irreversible damage. On 3 April (day 76), D and DH plants received a single final watering of approximately 40 ml. C and H plants continued receiving water ad libitum via table flooding every 2–3 days. After 3 April, drought plants received no further water until harvest (days 90–94).

Simulated herbivory was applied twice to mimic chewing damage^65–67. The first clipping occurred on 22 February (day 36), concurrent with drought onset. Plants in the H and DH treatments were clipped 1 cm above the first node, or 1 cm above the side branch when lateral growth was already present. A second identical clipping was applied on 3 April (day 76).

At harvest, all plants were measured for response traits. Plant length (Length) was recorded as the length of the longest shoot. To capture drought-induced morphological plasticity known in Vigna, we also counted the number of leaves and lateral shoots (NO_shoots). Lateral shoots were defined strictly as shoots originating from the main (longest) shoot. After measurements, all aboveground material was harvested, oven-dried at 70 °C for four days in a forced-air oven, and weighed to obtain aboveground biomass (Biomass). From the initial 560 plants, we harvested 528 plants corresponding to 94.3% of the initially planned plants. Exclusions were due to early mortality during the course of the experiment, caused by either wilting or technical issues.

Statistical analysis

All statistical analyses were performed in R (RStudio v.2023.06.1)⁶⁴. Data handling and visualization used tidyverse, and vegan and brms packages were used for Bayesian modelling and multivariate analyses. Raw trait values (biomass, shoot length, number of leaves, number of shoots) were analysed alongside two derived response types. First, log-ratios relative to the control treatment were calculated to quantify the direction and proportional magnitude of each plant’s response to stress. Biologically, log-ratios express how strongly a genotype’s performance deviates from its unstressed baseline, providing a scale-free measure of stress sensitivity. Second, genotype-level coefficients of variation (CV = SD/mean × 100) were computed for each trait and for the log-ratios across the three stress treatments (drought, herbivory, and drought and herbivory). In biological terms, CV captures the response stability: lower CV indicates more consistent, predictable performance under stress, whereas higher CV reflects greater instability or context-dependent responses. In the context of resilience, genotypes with small changes in log-ratio (i.e., small performance loss) and low CV (i.e., stable responses across stressors) are interpreted as exhibiting greater multi-stress robustness.

Climate variables (dry-season length and maximum monthly rainfall) were extracted based on the collection location for each accession from WorldClim v2⁶³, representing key aspects of the environments in which the wild and cultivated genotypes originate. These variables are ecologically relevant indicators of water availability and seasonality, both of which are expected to influence tolerance strategies and phenotypic variability in legumes. To test whether genotypes from harsher or more seasonal environments exhibit greater or lower variability in their stress responses, we evaluated associations between these climate variables and CV(log-ratio). This metric represents the coefficient of variation of log-transformed stress responses for each genotype and quantifies the extent to which a genotype’s proportional response to the three stressors (herbivory, drought, combined) varies relative to its own control baseline. Higher CV(log-ratio) values indicate more variable, and thus less consistent, proportional stress responses, whereas lower values reflect more stable and resilient responses across stress treatments.

To test the effects of domestication type (Cultivated vs. Wild), stress treatment (Herbivory, Drought, and Drought and Herbivory), and their interaction, we fitted a suite of ANOVA-style Bayesian hierarchical models using brms. Separate models were fitted for each trait (i.e., biomass, shoot length, number of leaves, number of shoots) and for each of the three response types (raw values, log-ratios, and genotype-level CVs). All models followed the structure Response variable ~ Type × Treatment + (1|Genotype) + (1|Table). Genotype and experimental table were included as independent random intercepts to account for repeated measurements within accessions and for block effects arising from the experimental layout. Categorical predictors (domestication type and treatment) were coded using sum-to-zero contrasts, which centre factor levels around the overall mean. With this coding, the estimated main effects represent the average difference between levels of a factor after accounting for all treatments, and interaction terms quantify how treatment effects differ between domestication types. Fixed-effect coefficients were assigned weakly informative Normal (0, 5) priors, and random-effect standard deviations were assigned weakly informative Student-t (3, 0, 5) priors to provide mild regularisation without imposing strong assumptions. Models were fitted using four Markov chain Monte Carlo (MCMC) chains with 3,000 iterations per chain (1,000 warm-up) via the rstan backend, with sampling controls (adapt_delta = 0.95, max_treedepth = 12) to avoid divergent transitions. Convergence was assessed through R̂ values (< 1.01), effective sample sizes, and visual inspection of trace plots. Posterior samples for fixed effects were summarised using medians and 95% credible intervals to describe the magnitude and uncertainty of main and interaction effects. To obtain population-level estimates (i.e., excluding random effects), we generated marginal posterior predictions for each Type × Treatment combination. Pairwise contrasts between cultivated and wild genotypes within each treatment were computed directly from the posterior samples so that uncertainty was fully propagated; contrasts whose 95% credible intervals excluded zero were interpreted as credibly different. Finally, to assess whether domestication type, treatment, and their interaction produced coordinated changes across multiple traits, we analysed multivariate responses using PERMANOVA (adonis2) based on Euclidean distances of standardized (z-scored) trait values. This approach tests for differences in overall multivariate trait composition rather than examining each trait separately, allowing us to detect joint trait responses that may not be apparent from univariate models. The model included Type, Treatment, and their interaction, with permutations (n = 999) stratified by experimental table to respect the blocking structure of the experiment. To visualise these multivariate patterns, we used distance-based redundancy analysis. This ordination provides an interpretable representation of how domestication type and treatments structure variation in the multivariate trait space and highlights which traits contribute most strongly to the observed separation.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

EDLP, FV, FVC designed the experiment, FV provided the /Vigna/ genotypes, EDLP, FVC and EDM established and conducted the experiment, FVC collected experimental data, EDLP did the statistical analysis. EDM wrote the discussion in consultation with EDLP and FV. All authors contributed to the final manuscript. This research was supported by the Ghent University BOF Starting Grant to EDLP “Integrating multitrophic interactions into sustainable crop production and diversified cropping systems” (project code BOF/STA/202502/002).

Funding

This research was supported by the Ghent University BOF Starting Grant to EDLP “Integrating multitrophic interactions into sustainable crop production and diversified cropping systems” (project code BOF/STA/202502/002).

Data availability

All data generated or analysed during this study are included in this published article (and its Supplementary Information files).

Declarations

Competing interests

The authors declare no competing interests.