Genomic signatures of adaptation in seed traits in the wild plant Brassica incana

Abstract

Seed traits underlying germination, by determining the environment experienced by plants throughout their lifetime, can play a key role in shaping plants’ adaptive strategies. However, the genomic bases of adaptation in seed traits and its link with local biotic and abiotic environments remain largely unexplored. Here, we used a pool-sequencing approach combining genome-wide association (GWA), genome-environment association (GEA) and a genome wide scan of a spatial genomic differentiation index (X^TX) to identify putative signatures of adaptation in seed traits and to climate and pollinator community across 14 populations of the wild plant species Brassica incana. We observed a complex genetic architecture potentially involved in seed trait adaptation, which varied depending on the seed trait function. Also, we identified several candidate genes simultaneously linked to signals of adaptation in seed traits and local abiotic or biotic conditions. These results expand our understanding on the adaptive value of seed traits, on its interaction with environmental conditions and on its potential for shaping the evolutionary trajectory of wild plant populations.

A study conducted on the wild plant Brassica incana identifies a complex genetic architecture underlying seed trait adaptation and shows how several candidate genes overlap with those involved in adaptation to biotic and abiotic conditions.

Introduction

Germination is the first and most radical transition occurring in the lifetime of seed plants. Indeed, the transition from seed to seedling implies the shift from heterotrophy to autotrophy and the first exposure of plants to biotic and abiotic environmental conditions, without the protective covering of the seed coat. By setting the conditions that plants will experience throughout their life cycle, germination shapes seedling survival and post-germination traits (i.e., phenotypic traits expressed by plants at later stages of their life cycle) as well as environment-dependent selective pressures acting on those traits^1–3. Therefore, seed traits underlying germination are usually subjected to a strong natural selection². This strong selection has been documented in several studies focusing on seed traits either related to seed germination response under different environmental conditions^4–8 or to seed morphological traits involved in germination, such as seed mass and relative embryo size^9–11.

Seed mass is a multifunctional trait that reflects many aspects of species’ reproductive strategy, such as mother plant investment, and thus influences dispersal syndrome, seed persistence and early seedling survival^9,11, while relative embryo size is a key trait associated with offspring size and germination timing (e.g.¹⁰). Among seed traits, germination timing is known to be subjected to a particularly strong action of selection in wild plant populations since it can prevent germination under unfavorable environmental conditions for seedling establishment and survival^3,12–14. The range of conditions required for a plant to complete a successful transition from seeds to seedlings, including dormancy-breaking and germination requirements (i.e., the germination niche³), ensures that germination occurs at the appropriate time and place for successful seedling establishment^3,12,15–17. This is highly relevant for plants from seasonal climates^3,16 as those in the Mediterranean¹⁸, where there is a marked annual variation in environmental conditions such as temperature and rainfall (i.e., dry summer periods with high temperatures followed by wet, mild autumns¹⁹). Germination requirements will thus determine the biotic and abiotic environment experienced throughout their life and, consequently, the strength and direction of natural selection acting on post-germination traits⁴.

Despite these lines of evidence on the crucial importance of adaptation in seed traits underlying germination, little is known about its genetic architecture (but see ref. ²⁰) and even less is known on the interacting genomic bases of adaptation between seed traits and the abiotic and biotic selective pressures acting on post-germination traits. The few studies investigating the genomic bases of seed traits have been conducted on model species or on plants of agronomic interest with the main aim of identifying specific genes crucial for crop improvements, and thus without focusing on adaptive and evolutionary processes (e.g.^21–25). This leads to limited insights into the genomic mechanisms underlying seed trait adaptation in non-model wild species.

Quantitative phenotypes displaying natural variation such as seed traits might have different genetic architectures underlying adaptation ranging from many loci of small-effect (polygenic model), single loci (monogenic model), or few loci of large-effect (oligogenic model), or a combination of these three different models^26–29). Each model is characterized by theoretical expectations and competing ideas on which genetic architecture is more likely to underlie rapidly evolving phenotypes and thus local adaptation^30–32. However, as reported for many life-history traits, several seed traits can be expected to show a highly polygenic adaptation (e.g.^33,34). Although this cannot be extended to all the seed traits, some of the traits involved in seed establishment and survival can be considered to be composite characters correlated to multiple fitness-related traits and their evolution is often assumed to be highly polygenic, with many loci of small effect rather than a few loci of large effect shaping quantitative changes in traits^34–36. A polygenic architecture of adaptation, by involving complex interactions among genes, could open the room for pleiotropic genes with functions going far beyond germination and that can thereby affect other traits of the plant life cycle. Moreover, the combined effect, for instance, of epistasis³⁷ and gene-environment interactions³⁸ further complicates the genetic landscape associated with adaptation in seed traits, making this a key and yet surprisingly underinvestigated area of study for understanding plant adaptation. Thus, elucidating the genetic architecture of adaptation to heterogeneous environments in plant seed traits can push forward our comprehension of their adaptability and ecological resilience, two topics of primary importance in a context of climate change.

Here, to fill this knowledge gap, we investigated both the genetic architecture underlying local adaptation in seed traits (seed germination responses and seed morphological traits), and the genetic interaction with putative adaptive response to climate and pollinator community descriptors in the Mediterranean perennial cliff plant Brassica incana Ten. (Brassicaceae). B. incana is an entomophilous, self-incompatible species which represents a particularly useful study system for investigating signatures of local adaptation, as it is characterized by a highly fragmented distribution³⁹ and its populations are known to experience different and challenging local environmental conditions due to the seasonal climatic patterns of the Mediterranean^40–42. Recent studies conducted on multiple B. incana populations located in Southern Italy identified putative signals of local adaptation both to the biotic and abiotic environment^40–42. For instance, Frachon et al.⁴⁰ revealed signals of genomic adaptation to climate, soil and pollinator community in 21 wild B. incana populations, while in Laccetti et al.⁴², by analyzing 14 different populations, it has been observed a strong association between local climate and seed germination response. Here, by focusing on 14 wild B. incana populations, we combined environmental (i.e., climate and pollinator community descriptors) and genomic data, obtained through a pool-sequencing approach, from Frachon et al.⁴⁰ with germination data from Laccetti et al.⁴² and original data on seed morphological traits. Specifically, using germination data from Laccetti et al.⁴², we estimated new traits related to seed dormancy and germination response under a wide range of temperature conditions. Genomic, environmental and seed trait data were thus used in our study to perform genome association analyses. Particularly, we first performed a genome-wide association (GWA) analysis combined with a genome scan for signatures of spatial genomic differentiation (X^TX) to identify the putative genomic bases of local adaptation in seed traits across the 14 investigated natural populations. Then, we performed a genome-environment association (GEA) analysis to determine whether candidate genes associated with seed trait variation across the populations were also involved in the adaptive response to their local abiotic (climate) and biotic (pollinator community) environment.

Results and discussion

Through an ecological genomics approach, we described the genomic regions putatively underlying local adaptation in seed traits and their overlap with those involved in adaptation to climate and local pollinator community descriptors in a wild, non-model plant species. We employed this approach in 14 wild populations of B. incana (Fig. 1) by measuring seed morphological traits and using experimental germination data obtained from Laccetti et al.⁴² where seeds were exposed to different temperature- and light-controlled conditions. Using these data, we thus estimated 17 new seed traits including (1) three traits related to germination response of fresh seeds associated with seed dormancy, (2) 11 traits related to germination response of after-ripened, non-dormant seeds to temperatures within and outside temperature conditions experienced by seeds in their local environment, and (3) three traits related to seed morphological traits known to be involved in key seed functions such as dispersal, establishment in the environment and germination (seed mass, relative embryo size, seed coat thickness; see Supplementary Tables 1 and Supplementary Data 1). Also, we selected 13 environmental variables related to climate and pollinator community descriptors (see Supplementary Data 1 and Supplementary Table 2). Most of the investigated seed traits showed extensive natural variation (Fig. 2 and Supplementary Fig. 2), providing the opportunity to explore the genomic regions putatively underlying local adaptation in the 14 B. incana populations.

Fig. 1. Geographic distribution of Brassica incana populations and description of the ecological genomics approach.

Sampling locations of the 14 wild populations of Brassica incana and description of the ecological genomics approach combining genome-wide association (GWA) based on seed traits (seed morphological traits and seed germination responses) and genome-environment association (GEA) based on abiotic and biotic environmental variables (climate and pollinator community composition).

Fig. 2. Variation of key seed morphological traits across Brassica incana populations.

a Images obtained through scanning electron microscopy (SEM), scale bars = 200 µm, used to calculate relative embryo size (i.e., embryo area to seed area ratio) and seed coat thickness in 14 wild populations of Brassica incana. b Natural variation observed for each of the three seed morphological traits subsequently used in the genome-wide association (GWA) analysis. Error bars represent 95% confidence intervals.

Adaptation in seed traits is shaped by a complex genetic architecture

A genome-wide association (GWA) analysis was conducted to determine the genetic architecture and identify genetic loci potentially associated with adaptation in seed traits. Specifically, we performed a genome-wide scan to test for the association between standardized allele frequency variation along the genome of B. incana and the among-population variation of 17 seed traits, using a Bayesian hierarchical model⁴³ and corrected by the local score method⁴⁴. We observed neat and narrow association peaks for all the investigated seed traits (Fig. 3; Supplementary Fig. 1). However, the genetic architecture of adaptation highly differs depending on the trait category and thus on its function (Fig. 4, Supplementary Table 1). Indeed, seed mass, a morphological trait reflecting many aspects of species’ reproductive strategy including mother plant investment, displayed an oligogenic architecture underlying local adaptation (i.e., number of candidate genes per trait < 10, Fig. 4) and showed unique association peaks, i.e., peaks that are not shared with genomic regions associated with any other seed trait (Fig. 4). Consistently with our findings, an oligogenic architecture of adaptation in seed mass was found in the legume crop soybean⁴⁵. Differently, instead, a previous study conducted in the monocot crop Sorghum bicolor, identified a more polygenic architecture³³. This different genetic architecture involved in seed mass adaptation could be due to a different geographic scale of the study (local vs. worldwide) or to a phylogenetic signal. Furthermore, relative embryo size, a morphological trait related to germination timing, showed a highly polygenic architecture of adaptation (i.e., number of candidate genes per trait > 10) and is associated with unique genomic regions (Fig. 4). The same result was obtained for germination proportion at alternating temperatures in fresh seeds (Fig. 4). Moreover, the genetic architecture of adaptation in two traits reflecting the extent of seed dormancy, i.e., (1) germination proportion at constant temperatures in fresh seeds and (2) the ratio between germination proportion at constant and alternating temperatures in fresh seeds, was highly polygenic and was associated with unique genomic regions compared to those associated with other traits (Fig. 4). Lastly, traits quantifying the ability of after-ripened seeds to germinate under temperatures within and outside local temperature conditions mainly showed a polygenic architecture underlying local adaptation and shared genomic regions between each other (Fig. 4). Our findings suggest that the genetic architecture of adaptation in seed traits in wild species might be particularly complex and is likely shaped by the combination of oligogenic and, more frequently, polygenic models thus confirming the hypothesis that seed trait adaptation is mainly associated with a polygenic architecture. In our study, we also highlight how in wild plants this genetic architecture might be highly variable depending on which trait and function related to seed germination is investigated. Moreover, the presence of shared genomic regions among seed traits suggests the presence of pleiotropy which is known to play a key role in plant adaptation (e.g.^46–48). For instance, pleiotropy may allow the synchronous response of different seed traits to varying selective pressures thus promoting rapid adaptation to the environment^30,49. However, although sampling in natural populations has proven to be useful to characterize genomic regions involved in adaptation to environmental conditions in wild populations (i.e., GEA in refs. ^40,50,51), it is important to note that seed traits analyzed in this study were estimated from seeds directly collected in natural wild populations of B. incana. As such, while the genomic regions identified in our study represent plausible candidates for local adaptation, we cannot exclude that seed traits might have been also partly shaped by environmental effects (see^52,53), which could have influenced the inferred genetic architecture underlying adaptation (e.g.⁵⁴). In this context, for instance, the year of seed collection might have at least partly influenced our results, since the genomic regions found associated with among-population variation in seed traits under our specific environmental conditions (i.e., late spring 2023), could partly vary in a different sampling year. Therefore, to be cautious, our results should be interpreted as signals of putative adaptive association and complementary studies, preferably using a common garden setup, might be useful to strengthen our findings.

Fig. 3. Manhattan plots from genome-wide association analysis (GWA) conducted on four seed traits of Brassica incana populations.

a Relative embryo size, (b) ratio between germination proportion at constant and alternating temperatures in fresh seeds (Regime_ratio_fresh_seeds), (c) germination proportion at 35/30 °C in after-ripened seeds (Germ_ar_seeds_35/30), and (d) germination proportion at 10/5 °C in after-ripened seeds (Germ_ar_seeds_10/5). The x-axis indicates the position of the 3,952,000 SNPs along the 139 super-scaffolds, ordered according to their size, illustrated by alternating black and gray dots. The y-axis indicates the Bayes factor corrected by the local score method (Lindley score). The 0.05% of SNPs with the highest association score according to the local score method is highlighted in green. For all the Manhattan plots see Supplementary Fig. 1.

Fig. 4. Relationship among candidate genes associated with genomic local adaptation in seed traits and to local environmental conditions in Brassica incana populations.

The left panel shows the number of candidate genes (set size) identified in genomic local adaptation to the specific variable in genome-wide association analysis (GWA) on the upper part and genome-environment association analysis (GEA) on the lower part. On the top panel, the number of candidate genes associated with a specific variable (single black dot) or shared among variables (multiple black dots linked) are indicated. Seed morphological traits are denoted in green, germination responses related to seed dormancy in light blue, germination responses quantifying the ability of after-ripened seeds to germinate under multiple temperature conditions in orange, and environmental variables (climate and pollinator community descriptors) in pink.

Genomic regions underlying among-population variation in seed germination are subjected to strong natural selection

To test for the presence of natural selection acting on seed traits, we performed a genome-wide scan based on a spatial genomic differentiation index (X^TX) associated with an enrichment analysis which revealed signatures of selection across the B. incana genome for 16 out of 17 seed traits (Table 1). Among these traits, we found a particularly strong enrichment value (fold enrichment higher than 100) for two seed traits related to seed dormancy, i.e., germination proportion at constant temperatures in fresh seeds and the ratio between germination proportion at constant and alternating temperatures in fresh seeds. This result is in line with previous studies showing a strong natural selection acting on seed traits shaping germination timing and the degree of seed dormancy^1,2,5,14,55. Also, consistently with our findings, the degree of dormancy (i.e., seed germination response to alternating temperatures in our study system) has been proposed to represent an adaptation to specific environments such as the Mediterranean, where dormancy release during summer could avoid germination and seedling establishment during hot and dry periods (e.g.^3,16,18,42). Interestingly, we found signatures of selection for top SNPs associated with germination response to cold and hot temperatures outside the range of conditions experienced by seeds in their local environment, i.e., germination proportion of after-ripened seeds at 10/5 °C and at 35/30 °C. Accordingly, in a previous study conducted within the same study system, a high germination response to hot and cold extreme temperatures was explained by warmer local temperatures⁴². These results together with the evidence of selection acting on germination response to hot and cold extreme temperatures suggest that local environments could make populations pre-adapted to successfully face extreme temperatures predicted under a climate change scenario thus enhancing species’ ecological resilience.

Table 1.

Enrichment in signatures of selection for the 17 seed traits testing the over-representation of the 0.05% upper tail of the Lindley score distribution in the 0.05% upper tail of the genome-wide spatial differentiation (X^TX) distribution

Trait	Description	ntops	Enrichment	P-value
Seed_mass	Seed mass (g)	3	3.03	*
Relative_embryo_size	Embryo area to seed area ratio	6	6.07	*
Seed_coat_thickness	Seed coat thickness (μm)	6	6.07	**
Germ_fresh_seeds_Co	Germination proportion of fresh seeds at constant regimes	128	129.56	***
Germ_fresh_seeds_Al	Germination proportion of fresh seeds at alternating regimes	0	0	ns
Regime_ratio_fresh_seeds	Germ_fresh_seeds_Co/Germ_fresh_seeds_Al	100	101.22	***
Germ_ar_seeds	Germination proportion of after-ripened seeds	40	40.49	***
Germ_ar_seeds_Co	Germination proportion of after-ripened seeds at constant regimes	15	15.18	**
Germ_ar_seeds_Al	Germination proportion of after-ripened seeds at alternating regimes	56	56.68	***
Germ_ar_seeds_35/30	Germination proportion of after-ripened seeds at 35/30 °C	28	28.34	***
Germ_ar_seeds_10/5	Germination proportion of after-ripened seeds at 10/5 °C	8	8.1	*
Germ_ar_seeds_30/20	Germination proportion of after-ripened seeds at 30/20 °C	32	32.39	***
Germ_ar_seeds_25/15	Germination proportion of after-ripened seeds at 25/15 °C	39	39.47	***
Germ_ar_seeds_20/10	Germination proportion of after-ripened seeds at 25/10 °C	42	42.51	***
Germ_ar_seeds_25	Germination proportion of after-ripened seeds at 25 °C	43	43.52	***
Germ_ar_seeds_20	Germination proportion of after-ripened seeds at 20 °C	49	49.6	***
Germ_ar_seeds_15	Germination proportion of after-ripened seeds at 15 °C	42	42.51	***

***P-value ≤ 0.001, **0.001 < P-value ≤ 0.01, *0.01 < P-value ≤ 0.05, ns P-value > 0.05.

Candidate genes putatively involved in seed trait adaptation are linked to seed response to local environmental conditions

To identify candidate genes putatively involved in genomic adaptation, we retrieved genes located within genomic regions identified as significantly associated with seed traits in the GWA analyses through the local score approach, as well as the first gene upstream and the first gene downstream, following the approach of Libourel et al.⁵⁶. We were able to identify 124 candidate genes and their associated functions (Supplementary Data 2). Interestingly, we found that genes known for their involvement in seed development, dormancy-related processes and temperature acclimation colocalized with genetic loci detected in our study, thus validating our approach. For instance, the protein coding genes GCP1 and WDR5A, involved in embryo development in A. thaliana^57,58 were also associated with among-population variation in relative embryo size in B. incana. Also, this seed trait was associated with the transcription factor MYB73. MYB family transcription factors play important regulatory roles in diverse biological processes⁵⁹. In particular, OsMYB73 which is mainly expressed in the early developing pericarp and endosperm, has been found to play a key role in shaping grain size and endosperm development in rice⁵⁹. Moreover, several genes known for their role in breaking seed dormancy were associated with the adaptation in the ratio between germination proportion at constant and alternating temperatures in fresh seeds, i.e., a trait underlying the degree of dormancy in our study system (Supplementary Data 2). As an example, SUVH5, belonging to the SUV(R) group of SET domain proteins, is a positive regulator of light-mediated seed germination and may represses the expression of DOG genes, master regulators in control of physiological seed dormancy⁶⁰. Notably, SUVH5 is a histone methyltransferase, an enzyme involved in epigenetic regulation, which thus suggests a potential role for epigenetic mechanisms in modulating seed dormancy responses to environmental cues. This raises the possibility that epigenetic mechanisms, such as histone modification, may contribute to adaptive strategies in wild populations by enabling flexible and potentially reversible germination responses under varying temperature regimes, such as those predicted under a climate change scenario⁶¹. Furthermore, HOS15 and PUB50, responsible for cold tolerance in A. thaliana^62,63, were associated with adaptation in germination response to low temperatures (germination proportion at 10/5 °C). Interestingly, we also identified several genes associated with seed germination response of after-ripened seeds and known to be involved in salt tolerance (e.g., NEK2, AOX3^64,65) in line with a previous study on B. incana reporting its high tolerance to seawater stress⁶⁶. For seed coat thickness, although we found a significant enrichment in X^TX index, the number of SNPs associated with these regions was lower than our set threshold and has been discarded from the analyses, thus impeding the identification of candidate genes. These findings emphasize that candidate genes putatively associated with adaptation in seed traits play a key role in shaping seed response to environmental conditions encountered by wild populations in their local environment.

Multiple candidate genes are putatively involved both in the adaptation in seed traits and to local climatic conditions and pollinator community composition

Through the ecological genomics approach combining GWA and GEA, we identified five candidate genes putatively underlying adaptation in seed traits that were also associated with signals of adaptation to climate and pollinator community (Fig. 4). Specifically, we found two candidate genes (RCF3 and a protein with unknown function) putatively involved in the adaptation in germination proportion at 20 °C in after-ripened seeds and total germination proportion in after-ripened seeds under an alternating regime that were also involved in the adaptation to mean annual precipitation (MAP). This finding is consistent with the well-known association between local precipitation regime and seed sensitivity to alternating temperatures and suggests a major role for these genes in the adaptation of B. incana to local annual precipitation⁶⁷. Surprisingly, although germination is highly affected by local temperature, we were not able to identify shared candidate genes between seed traits and mean annual temperature (MAT). This can be related to the use of large-scale climate data in our analyses, which reflects average temperatures over the past decades and may not capture fine-grain temperature variations that might be more relevant in germination processes. Additionally, we identified three candidate genes associated with the among-population variation of two traits involved in seed dormancy (i.e., ratio between germination proportion at constant and alternating temperatures in fresh seeds, and germination proportion at constant temperatures in fresh seeds), which are also involved in the adaptive response of B. incana to the pollinator functional category long-tongued bees. Interestingly, one of these genes, PAP3, belongs to the fibrillin family, which is related to the expression of phenotypic traits known to be involved in pollinator attraction (i.e., floral organ development and chromoplast pigment accumulation), and is also involved in seed dormancy breaking mechanisms (i.e., its expression, mediated by light and circadian regulation, promote the transition of seeds from dormancy to active growth) in A. thaliana^68–70. Also, four candidate genes were found to be potentially involved both in relative embryo size adaptation and in B. incana adaptive response to the pollinator functional category small flies. Among these genes, the transcription factor MYB73 is involved in anthocyanin biosynthetic pathway, flowering time and petal size development, but also in the regulation of grain size and endosperm development in rice^59,71,72. Anthocyanin production, petal size and specific flowering period play a key role in shaping plant-insect interaction, by enhancing plant visibility and thus pollinator attractiveness or promoting phenological synchrony between flowering plants and their pollinating insects. Of particular interest is that, here and in a previous study conducted within the same study system⁴⁰, we found signals of adaptation to small flies which are generally reported as inefficient pollinators⁷³. This apparently surprising result might be explained as a by-product of adaptation in a key seed morphological trait on which strong selective pressures usually act and suggests a profound effect of adaptation in seed traits on traits expressed by plants at the end of their life cycle.

The overlap between candidate genes associated with adaptation in seed traits and to local biotic environmental conditions in B. incana might be explained by an indirect effect of environmental cues shaping adaptation in seed traits on the expression of plant phenotypic traits linked to pollinator attractiveness. Alternatively, selection mediated by pollinators might affect traits involved in seed establishment and germination via transgenerational effects. Such effects have been shown to play an important role in modulating offspring responses to environmental cues, potentially contributing to adaptive transgenerational strategies, particularly under gradually changing climates (e.g.^74,75). This overlap suggests that traits usually considered to be shaped by different ecological processes might be evolutionary coupled with important implications for the mechanisms underlying plant adaptation under a climate change scenario.

Our study thus points out the importance of considering spatial heterogeneity in ecological conditions to determine the genetic architecture involved in plant response to different environments and the associated quantitative trait variation. Furthermore, acknowledging the constraints of the experimental design, our results highlight how adaptation in seed traits has the potential to influence the evolution of traits expressed at different stages of plant life cycle.

Conclusions

In this study, we documented putative signals of adaptation in seed traits in a wild non-model plant species. Our findings reveal a complex genetic architecture underlying genomic adaptation in seed traits in wild plant populations. Indeed, we observed the association with both unique and shared genomic regions among the investigated traits, and a polygenic or oligogenic architecture of adaptation depending on the seed trait function. Specifically, the adaptation in a key seed morphological trait reflecting mother plant investment (i.e., seed mass) was shaped by an oligogenic and specific genetic architecture. Differently, the genetic architecture of adaptation for all the other seed traits was highly polygenic. However, while seed traits related to germination timing and seed dormancy mainly showed the association with unique genomic regions, seed traits describing germination response to temperatures within and outside local temperature conditions were associated with several shared genomic regions. Notably, for almost all the seed traits we found signatures of selection along the genome with those related to seed dormancy showing particularly strong signals of natural selection. Additionally, several candidate genes putatively underlying adaptation in seed traits overlapped with the ones involved in B. incana adaptive response to climate (precipitation) and pollinator community composition. Our results reinforce the importance of considering the adaptation in early-life plant stages to their local environment (i.e., considering the different ecological conditions interacting with the plant) to better understand the genomic adaptive trajectory of wild populations. Future studies involving experimental validation of the genes identified in this study could provide deeper insights into the underlying regulatory mechanisms. Such analyses would not only clarify the roles of key genes within the pathway but also contribute to the development of a comprehensive regulatory model, potentially enabling comparisons with well-established models in other plant species. This study also paves the way for testing how the identified candidate genes, filtered through seed germination, influence phenotypic traits expressed by plants at different stages of their life cycle.

Materials and methods

Study species and study system

Brassica incana Ten. is a wild plant species, native to Eastern and Southeastern Europe and mainly distributed along coastal cliffs of Central and Southern Italy, and Sicily³⁹. It is a perennial species with a generalist pollination system⁴⁰ which produces cylindrical, bent siliques with small, rounded seeds inside³⁹. B. incana has a highly fragmented distribution and is characterized by populations experiencing a wide range of different biotic and abiotic environmental conditions⁴⁰. Here, we focused on 14 wild populations distributed in Southern Italy to identify signatures of local adaptation in seed traits and the potential interacting effects with abiotic and biotic environmental conditions (climate and pollinator community composition; Fig. 1).

Characterization of seed traits

In each population, we characterized seed traits related to seed morphology and seed germination responses of fresh and after-ripened seeds under a wide range of temperature conditions. In order to characterize seed morphological traits known to be involved in key seed functions such as dispersal, establishment in the environment and germination^76,77, we collected seeds from each of the 14 wild populations at the end of the flowering season 2023 (Fig. 2). Specifically, we collected seeds from fully ripened fruits from ten maternal families per population (inter-family distance ≥ 2 m) within an area of approximately 250 m², to minimize the risk of collecting related plants. This experimental design allows minimizing the amount of genetic variance compared to a design where seeds with unknown parental origin are tested and is widely used to characterize the adaptive potential of natural populations for which controlled crosses are difficult to manage in situ^42,78,79.

Seeds of each maternal family were thus characterized, exclusively for the purpose of our study, by measuring three key seed morphological traits: (1) seed mass, (2) relative embryo size (i.e., embryo area to seed area ratio), and (3) seed coat thickness (see Table 1 and Supplementary Table 1 for trait description and functional significance). In our study, we estimated seed mass by weighting an average of 240 seeds per population, collected from ten different maternal families per population, using a precision balance (accuracy ± 0.01 mg). Relative embryo size and seed coat thickness were obtained, from the same seed batch, through scanning electron microscopy (SEM) at the CeSMA (Centro Servizi Metrologici Avanzati) of the University of Naples Federico II. For each population, twenty seeds, obtained from ten different maternal families, were cut to make the morphology of the seed coat and the embryo visible. Subsequently, samples were thus coated with gold or gold/palladium to ca. 30 nm for observation under a Zeiss Merlin VP Compact SEM or a Nova NanoSEM 450. Finally, seed traits were extracted from calibrated images using ImageJ 1.33 (https://imagej.nih.gov/ij/).

Seeds collected at the end of the flowering season 2023 as described above were also used in a previous germination experiment investigating seed thermal responses within and outside temperature conditions experienced by seeds in their local environment (ref. ⁴², Supplementary Fig. 2). Briefly, in Laccetti et al.⁴², germination experiments were set up using both freshly collected seeds and dry after-ripened seeds (i.e., seeds dry stored for 6 months after collection at ~20 °C 40% RH as a procedure to break dormancy) from the 14 investigated populations. Seeds of each maternal family of the 14 populations were exposed to different temperature regimes (i.e., alternating and constant), and multiple temperature conditions within and outside the range of conditions experienced by seeds in their local environment, for a total of six different treatments in the experiment conducted on fresh seeds and eight treatments in the experiment conducted on after-ripened seeds (Supplementary Fig. 2). An average of 171 and 256 seeds per population was used in the experiment on fresh seeds and after-ripened seeds, respectively. To perform the experiments, seeds, placed in 9 cm diameter plastic 1% agar Petri dishes, were allocated into different temperature- and light-controlled incubators. In this study, by using data collected in these two germination experiments, we obtained 14 new seed traits related to (1) germination response of fresh seeds associated with seed dormancy, (2) germination response of after-ripened seeds to temperatures within and outside temperature conditions experienced by seeds in their local environment (Supplementary Tables 1 and Supplementary Data 1). The germination response to each treatment was quantified as the ratio between germinated seeds and the total number of seeds in each population. For after-ripened seeds, the germination response was standardized by dividing the germination response to each temperature by the mean germination response of the population. We used this approach since our aim was to quantify germination response to specific temperature conditions and an overall decrease in germination due to a different seed longevity across populations could have biased our interpretation of the results.

Genomic characterization

Genomic characterization of the 14 B. incana populations was performed by using pooled sequencing data available in Frachon et al.⁴⁰. In brief, leaf tissue was collected from an average of 28 individuals per population and stored in a −80 °C freezer until DNA extraction (sampling and DNA extraction details are available in methods in Frachon et al. 2023). After DNA concentration measurement using ddDNA Qubit assay, we pooled equimolarity with an average number of 27.7 individuals and a median of 29 individuals per population (i.e., per pool). We sequenced each pooled population using Illumina sequencing TruSeq and reads were mapped against the B. incana reference genome (obtained in Frachon et al. (2023) and available in the NCBI database; project number PRJNA859008) using the mapping QC app in ezRun and Bowtie2 v2.4.1. The detailed description of sequencing and de novo genome assembly is provided in Frachon et al.⁴⁰. Since Frachon et al.⁴⁰ performed SNP calling on 21 populations, we re-performed the SNP calling on the 14 investigated populations using the freebayes-parallel script in Freebayes v1.2.0-4-gd15209e⁸⁰. To do this, as described in Frachon et al.⁴⁰, SNPs with variant quality above Q20 were retained for downstream analysis using bcftool v1.9 and annotated with de novo predicted gene models using SnpEff v4.2. Then, we trimmed the genomic data using VCFtools v0.1.15⁸¹, by keeping (1) biallelic loci (--min-alleles 2, --max-alleles 2) with a mean read depth ranging from 6 to 100 (--min-meanDP 6, --max-meanDP 100), (2) loci with missing values (--max-missing-count 2) in less than two populations and a minor allele frequency higher than 0.07 (--maf 0.07), and (3) 139 super-scaffolds. After trimming, we obtained an allele read count matrix of 3,952,000 SNPs.

Genome association studies

In order to identify the genetic variants associated with seed traits variation and unravel their putative involvement in the adaptive response of B. incana to its local environment, we performed both genome-wide association (GWA) and genome-environment association analysis (GEA) using previously pooled sequencing data (3,952,000 SNPs) from Frachon et al.⁴⁰. The two genome association analyses were based on a Bayesian hierarchical model implemented in the BayPass software v2.1⁴³. For the GWA analyses, we used a set of 17 seed traits described in the previous section (full list in Supplementary Tables 1 and Supplementary Data 1). For the GEA analyses, we used 13 ecological variables selected from Frachon et al. (ref. ⁴⁰; Supplementary Data 1 and Supplementary Table 2). In our study, we re-performed the GEA analyses since the number of populations had been reduced from 21 to 14 compared to Frachon et al.⁴⁰. We selected four climate variables related to temperature and precipitation, five pollinator functional categories (i.e., long-tongued bees, bumblebees, large bees, hoverflies, small flies) found to be under selection in Frachon et al.⁴⁰, and four plant-pollinator network indices. Climate data, i.e., mean annual temperature (MAT), mean annual precipitation (MAP), average autumn temperature (Tave_at), and average autumn precipitation (PPT_at) were extracted for each B. incana population from the ClimateEU database v4.63, at 1.25 arc-min resolution, following the methodology reported in Hamann et al.⁸². Pollinator community data were, instead, obtained from field observations conducted in two different flowering seasons (i.e., spring 2018 and 2019). In detail, the observed pollinators were assigned to 12 functional categories: bumblebees (genus Bombus), long-tongued bees (genus Anthophora), other large bees (mostly genus Andrena), small bees, honeybees, large wasps, small flies, large flies, hoverflies, small beetles, large beetles, butterflies (mostly genus Pieris). In each population, the total number of visited flowers by each functional category was recorded. These data were also used to estimate plant-pollinator network indices for each B. incana population using the bipartite R package⁸³. Specifically, we estimated the following indices: (1) species strength (i.e., sum of dependencies of each population aiming at quantifying a population relevance across all its partners); (2) d (i.e., specialization of each population based on its discrimination from random selection of partners); (3) normalized degree (i.e., number of partner species in relation to the potential number of partner species); (4) partner diversity (i.e., Shannon diversity index). To reduce the effect of population genetic structure, we conducted our study at a regional scale following the approach of Frachon et al.^40,50,51. Additionally, we performed a singular value decomposition (SVD) of the covariance matrix of allele frequencies (Ω), without correction for population structure in the BayPass software. We did not observe a strong population structure along the first axis explaining 87.8% of the genomic variance (compared to 94.3% in Frachon et al.⁴⁰; Supplementary Fig. 3). Population structure was slightly more pronounced along the second axis, which explained 7.5% of the genomic variance (Supplementary Fig. 3). However, we used a Bayesian hierarchical model based on scaled covariance matrix to mitigate this effect and correct for potential complex effects of demographic histories⁴³. As described in Frachon et al.⁵¹, the core model was used to estimate the Bayesian factor (BF_is in dB hereafter called BFdB) between the allelic frequencies along the genome, and both seed traits (GWA analyses) and environmental variables (GEA analyses; climate and pollinator community composition). The core model was run three times, and the final Bayesian Factor was calculated by averaging the three runs. Given the large number of SNPs involved, we subsampled the procedure to estimate the covariance matrix of population allele frequencies (Ω) as in Frachon et al⁵⁰. and divided the full dataset into 17 sub-datasets of ∼20,8000 SNPs each. The GWA and GEA for each seed trait and each environmental variable, respectively, and each genomic subset were performed in parallel and subsequently merged. Then, we corrected each BFdB by using a local score approach that detects the accumulation of similar P-value in a given region increasing the power of genomic analyses and decreasing false positives⁴⁴. As suggested in Bonhomme et al.⁴⁴ and Libourel et al.⁵⁶, for the local score method we fixed the parameter ξ at three. Since the Bayesian model does not generate P-value, we ranked the BFdB values in the R environment (rank function) from the highest to the lowest and normalized the rank by the total number of SNPs to artificially create P-values as described in Frachon et al.⁴⁰. To identify candidate genes, we retrieved genes located within genomic regions identified as significantly associated with seed traits or environmental variables in the GWA and GEA analyses through the local score approach, as well as the first gene upstream and the first gene downstream, as described in Libourel et al.⁵⁶. To do this, we considered only the 0.05% of SNPs with the highest association score after applying the local score method and kept only genomic regions containing more than three SNPs. Then, we checked whether candidate genes associated with seed traits overlapped with candidate genes associated with environmental variables. The UpsetR R package⁸⁴ was used to visualize the shared candidate genes. Genes functions were predicted using the SwissProt database (release 2019_03).

Detection of signals of selection for seed traits

To identify signatures of selection for seed traits, we performed a genome-wide scan of the spatial genomic differentiation index (X^TX) associated with an enrichment analysis, across the 14 populations of B. incana, in the BayPass software. This index is analogous to the traditional F_ST-based scan⁸⁵ and is based on allele frequencies corrected for population structure. Specifically, for a given SNP, the X^TX measures the variance of the standardized population allele frequencies, which results from a rescaling based on the covariance matrix of population allele frequencies (Ω). This allows correcting for the genome-wide effects of confounding demographic evolutionary forces⁴³ and has been proven to allow a robust identification of SNPs that are highly differentiated among natural populations^40,50,51. As described above, we used the local score approach to correct the X^TX fixing the parameter ξ at three. Then, we tested for a significant enrichment in the 0.05% extreme upper tail of the X^TX distribution for the SNPs exhibiting the highest association scores with the 17 seed traits. Specifically, this significance was tested by running 10,000 null circular permutations of the 0.05% SNPs showing the highest association score with seed traits.

Statistics and reproducibility

To ensure the reproducibility of our findings, we provide comprehensive descriptions of all procedures, from data collection to analysis in the Material and Methods section. All the data were collected for 14 wild populations of Brassica incana. Sample size varied depending on the investigated trait: (1) seed morphological measurements were performed on 20 seeds per population, (2) seed mass was estimated on 240 seeds per population, (3) the first germination experiment performed on fresh seeds was conducted on an average of 171 seeds per population and (4) the second germination experiment was conducted on an average of 256 seeds per population. Pool-sequencing data were obtained from a median of 29 individuals per population. SNP calling was performed using the freebayes-parallel script in Freebayes v1.2.0-4-gd15209e. SNPs were annotated with de novo predicted gene models using SnpEff v4.2. Genomic data were trimmed using VCFtools v0.1.15. Genome association analyses were performed using a Bayesian hierarchical model implemented in the BayPass software v2.1. The full details, including parameters used, are mentioned earlier in the Material and Methods section. Supplementary data 1 provides the information for plotting Fig. 2 and Supplementary Fig. 2. These data were then used to perform genome association analyses. Sequencing data and the scripts used to conduct the analyses are publicly available as described in the Data availability and Code availability statements.

Reporting summary

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

Author contributions

Conceptualization: L.L., L.F., G.S. Data curation: L.F., L.L. Investigation: L.L., L.F., G.S., A.C., M.R.B.L. Methodology: L.L., L.F., G.S., A.C., M.R.B.L. Formal analysis: L.L., L.F. Funding acquisition: G.S. Project administration: G.S., L.F. Software: L.L., L.F. Resources: L.F., G.S., A.C. Visualization: L.L., M.R.B.L. Validation: L.L., L.F., G.S., A.C., M.R.B.L. Supervision: G.S., L.F. Writing—original draft: L.L., G.S. Writing—review & editing: L.L, G.S., L.F., A.C., M.R.B.L.

Peer review

Peer review information

Communications Biology thanks the anonymous reviewers for their contribution to the peer review of this work. Primary Handling Editors: Showkat Ganie and David Favero.

Data availability

Sequencing data from Pacbio and Illumina used for this study are available at the ENA database (project number PRJEB54646). The bionano raw data and assembled optical maps are available at NCBI database (sample name PRJNA859008). Data on seed trait and ecological variables are provided in Supplementary material (Supplementary Data 1).

Code availability

All the scripts used to perform the genome association analyses are available in GitHub (https://github.com/lucrezialaccetti8/Genome-association-analyses) and through the Zenodo Digital Repository (10.5281/zenodo.16631395).

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s42003-025-08673-w.