Variation in wild pea (Pisum sativum subsp. elatius) seed dormancy and its relationship to the environment and seed coat traits

Plant material

The investigated samples originated from various genebanks and contained 97 wild pea (Pisum sativum subsp. elatius (M. Bieb.) Asch. & Graebn) accessions. These span from the western Mediterranean, through the southeastern Europe to the Middle East (Table S1, Fig. 1). This material was selected based on passport data including GPS information and assessed for genetic diversity relationship (Smýkal et al., 2017; Smýkal et al., 2018a; Trněný et al., 2018). Plants were grown in 5 litre pots with peat-sand (90:10) substrate mix (Florcom Profi, BB Com Ltd. CZ), in glasshouse conditions (January–May 2016 and 2017) day/night temperature within the ranges 35–20/15–12 °C, and a natural photoperiod increasing from 6 to 14 h without supplementary light (UP campus, Olomouc, CZ). Cultivated and wild peas are largely self-pollinating (Smýkal et al., 2018a) and as samples were of single seed descent they are expected to be genetically uniform. After harvest, mature seeds were cleaned from pods, dried at room temperature and packed in paper bags. The germination testing was performed within the month from harvest.

Germination testing

To assess the dormancy status, seed imbibition and germination was assessed (Baskin & Baskin, 2014). Based on temperature data from soil sensors monitored at several sites with wild pea populations (Smýkal et al., 2017; Von Wettberg et al., 2018) we tested the effect of two oscillating temperature regimes on dormancy level: (i) day/night temperatures of 35/15°C and (ii) 25/15 °C, both at 14h/10 h periodicity and in the dark. Intact seeds were placed on water saturated filter papers (Whatman Grade 1, Sigma CZ) in 90 mm Petri dishes (P-Lab, CZ) in temperature controlled chambers (Laboratory Incubator ST4, BioTech CZ). For each treatment, 25 seeds per accession with two replicas were incubated. To prevent fungal growth the fungicide (Maxim XL 035 FS; containing metalaxyl 10 g and fludioxonil 25 g) was applied in amount of 1 ml l⁻¹ of tap water. Initial comparison of control and fungicide containing samples did not show any significant difference. Seeds were monitored at 24 h intervals for a total of 28 days (the longest time the seeds remained fungus-free in the 35/15 °C treatment) and water was added as needed usually every 3rd day. The plates were randomly rearranged during the scoring.

Germination data analysis

The number of imbibing seeds (i.e., swollen enlarged seeds upon the uptake of water) and also germinating seeds (i.e., radicle protrusion was the criterion for germination) was scored daily for the period of 28 days for each accession. Since we are primarily interested in study of physical dormancy (PY) executed by seed testa barrier of water entry (Smýkal et al., 2014) we have used the number of imbibed seeds as a measure of germination in all further calculations (thus our data contain both imbibed and fully germinated seeds, which can be separated by several days from each other, notably all imbibed seeds have germinated in course of experiment).

In order to capture the complexity of germination we have used a spline function (function defined piecewise by polynomials) fitted to the data (Ramsay & Silverman, 2005; Van den Boogaart, Egozcue & Pawlowsky-Glahn, 2014). This treats the original germination data (daily counts of germinated seeds) as discrete realizations of an asymptotically continuous process, i.e., for each sample the germination can be represented by a non-decreasing function called the absolute germination distribution function (AGDF) in the following text. The term absolute is used to distinguish AGDF from standard cumulative distribution function; unlike the latter one, AGDF deals with absolute values of cumulative germination. This function reflects well specific patterns related to germination of a given accession including final germination proportion and germination speed. In addition, we considered the area under AGDF, computed using the Riemann integral. From its construction, increasing value of this coefficient indicates both higher germination proportion and/or germination speed. For its numerical computation we used numerical integration, composed trapezoidal rule (Burden & Faires, 2011). AGDFs were obtained by approximation of the original germination data by smoothing with spline functions (De Boor, 1978). We used a generalized functional (Machalová, 2002) with the smoothing spline of degree k = 4 (degree of the respective basis polynomials used for construction of the splines) and the first derivative in L ₂ norm in Matlab (https://www.mathworks.com/) and further processed using R (R Development Core Team, 2018). Each spline is characterized by a set of real numbers, called B-spline coefficients. They represent coefficients of a linear combination of basis functions (so called B-splines) which are the same for all samples and each of them covers part of the domain of the germination function. Accordingly, a value of each specific coefficient indicates local behavior of the spline (and, consequently, of the original germination data). Moreover, B-spline coefficients can be also further tuned, like here to achieve better smoothing properties. B-spline coefficients for each accession (Table S2) were subjected to cluster analysis using UPGMA and Euclidean distance metric (Legendre & Legendre, 2012) separately for each temperature treatment, while areas under the curve were used in the case of combined temperature treatments dataset. Therefore, concrete values of B-spline coefficients were not important for interpretation purposes here, however, they were simply utilized for the above multivariate analyses. Cluster analyses were performed in the software PAST 3.21 (Hammer, Harper & Ryan, 2001).

Environmental variables

Environmental variables (climate: http://worldclim.org/version2; soil: https://soilgrids.org) were obtained using the ArcGIS Pro software (https://pro.arcgis.com). Species occurrence databases have often certain degree of uncertainty due to different spatial accuracy (Stine & Hunsaker, 2001; Murphey et al., 2004; Meyer, Weigelt & Kreft, 2016). In order to minimise the error caused by coordinates we developed a geoprocessing model to automate the calculation of mean values from 5 km buffer around the collection sites. Using this approach, we smoothed the uncertainty caused by imprecise localisation. Data were extracted using geoprocessing model from GeoTIFF rasters in the WGS-84 coordinate system (EPSG: 4326) with a spatial resolution of 30 arc-seconds (∼1 km). The bioclimatic variables (BIO1–BIO19), are derived from the monthly temperature and rainfall values (Fick & Hijmans, 2017). These represent annual trends (e.g., mean annual temperature BIO1, annual precipitation BIO12), seasonality (e.g., annual range in temperature and precipitation BIO4 and BIO15) and extreme or limiting environmental factors (e.g., the temperature of the coldest and warmest month BIO5 and BIO6, and amount of precipitation in the wet and dry quarters BIO16 and BIO17). Soil data were extracted from SoilGrids collection of updatable soil property and class maps of the world (Hengl et al., 2017). We used values described in Table S1 from 5 cm depth and resolution 1 km (same as WorldClim). For details, including basic descriptive statistics of mean tendency and variation of each environmental variable and legend see Table S1.

Testing of relationships among germination pattern, geography and environment

Matrix of environmental variables was firstly checked for the presence of the multicollinearity. Because of strong covariation between several environmental variables (pairwise Pearson r ≥ |0.90|), eight soil and nine climatic variables were removed from the dataset and all remaining analyses (except for niche analysis) were done with reduced matrix of environmental variables containing 10 climatic and nine soil variables (Table S1). To assess whether there is spatial correlation present in the data, Moran’s I spatial correlation statistics (Legendre & Legendre, 2012) was calculated for each environmental variable in the reduced matrix using the software PASSaGE v. 2.0 (Rosenberg & Anderson, 2011). Ten distance classes with unequal widths were created and Moran’s I was calculated for each distance class with largest class excluded. To test the global significance of each spatial correlogram, we check whether the correlogram contains at least one correlation statistics that is significant at the Bonferroni-corrected significance level (Legendre & Legendre, 2012).

The matrix of B-spline coefficients describing pattern of germination of the studied accessions in two temperature treatments (35/15 °C and 25/15 °C), considered as response variables, was analysed by Principal Component Analysis (PCA; Legendre & Legendre, 2012) to find the main gradients within the germination data-set. No standardization was applied on the matrix of B-spline coefficients prior to PCA. Moran’s I spatial correlation statistics was used with the same settings as for environmental variables to test the presence of spatial structuring of accessions scores on the first two resulting ordination axes (PC1, PC2). Set of climatic and soil variables and geographic coordinates (latitude, longitude) were used as supplementary variables and correlated with the first two principal components. To control of possible spatial autocorrelation between each environmental variable and principal components representing germination pattern, we used both standard and modified version of the t-test to assess the significance of correlation between two spatial processes. Modified version of the t-test (Dutilleul et al., 1993) was performed in PASSaGE v. 2.0 and PCA was performed using Canoco 5.10 (Ter Braak & Šmilauer, 2012).

Constrained ordination (redundancy analysis, RDA; Šmilauer & Lepš, 2014) was used to partition the variation in the matrix of B-spline coefficients explained by space (spatial position of sampling sites) and primary predictors (environmental variables), using PCNM method (Borcard & Legendre, 2002) without and with forward selection also for primary predictors. To identify spatial explanatory variables, we used principal coordinates of neighbouring matrices (PCNM; Borcard & Legendre, 2002), resulting in the PCNM variables. Only those PCNM variables with an adjusted P lower than 0.05 during forward selection procedure were selected to enter RDA. Subsequently, pure and shared effects of environment (groups of soil and climatic variables were tested separately) and spatial variables (PCNM) on the matrix of B-spline coefficients were analysed using RDA. We used a specific partitioning algorithm that provides unbiased estimates of explained variation in RDA (Peres-Neto et al., 2006). Here, in the first step, all environmental variables (either soil or climatic) from reduced data-set entered analysis as primary predictors. In the second step, a stepwise procedure using forward selection was done resulting in the subset of environmental predictors, best explaining the matrix of B-spline coefficients. P values were corrected using the False Discovery Rate adjustment (Benjamini & Hochberg, 1995). The significance of pure and shared (marginal) effects of the explanatory variables was tested by a Monte Carlo permutation test with 999 permutations. Analyses were performed in Canoco 5.10 (Ter Braak & Šmilauer, 2012; Šmilauer & Lepš, 2014) and the library vegan (Oksanen et al., 2018) in R.

Niche analysis

Using the geographic locations of the studied accessions and their categorized germination response based on 35/15 °C treatment, we constructed ecological niche models for each of the germination response categories. As the modelling approach, Maxent (version 3.4.1, Phillips, Dudík & Schapire, 2017), a maximum-entropy based machine learning method was used. Maxent was shown to perform better than other methods when samples sizes are small (Elith et al., 2006; Hernandez et al., 2006) and it estimates the potential niche instead of the realized distribution of the modelled entity (Phillips, Anderson & Schapire, 2006). A buffer zone of 300 km was defined as the background area for each group. As environmental predictors, 19 bioclimatic variables, extracted from WorldClim (Fick & Hijmans, 2017) at a resolution of 2.5 arc minutes, were used. The equivalency tests (Warren, Glor & Turelli, 2008) were performed with software ENMTools (version 1.4.4 Warren, Glor & Turelli, 2010). Two overlap indices were employed in these tests, Schoener’s statistic, D (Schoener, 1968) and modified Hellinger distance, I (Warren, Glor & Turelli, 2008). Data manipulation and map creation were performed in the R with the packages sp (Pebesma & Bivand, 2005), raster (Hijmans, 2017) and dismo (Hijmans et al., 2017).

Quantification of soluble and insoluble proanthocyanidins in the seed coat

Seeds of 86 accessions (Table S3) were analyzed for soluble proanthocyanidins (PAs, in acetone and 4-dimethylaminocinnamaldehyde [DMACA] reagent) and insoluble PAs (in butanol-HCl) using the method described by Pang et al. (2008). Briefly, separated seed coats were freeze dried and 20 mg was homogenized and extracted with 10 ml of acetone-acetic acid-water and 1 h sonication, followed by 5 min centrifugation at 10,000 rpm. To the 0.5 ml of extract, the 1.5 ml of 0.1% DMACA (Sigma-Aldrich, Karlin, Czech Republic) dissolved in HCl-ethanol-water solution, was added. The reaction was measured on spectrophotometer after 10 and 25 min at 640 nm using proanthocyanidin A (Sigma-Aldrich, Karlin, Czech Republic) as standard to calculate the amount of soluble PAs. The pellet after extraction of soluble PA was freeze dried and extracted with 5 ml of butanol-HCl (x − yv∕v), sonicated for 1 h, centrifuged (5 min/10,000 rpm) and measured at 550 nm. Thereafter the extract was boiled in a water bath for another 1 h, cooled down and the absorbance measured again at 550 nm using proanthocyanidin A as standard to calculate the amount of insoluble PAs. The total content of PAs among dormancy categories was compared by ANOVA. Ratio soluble/insoluble PA among dormancy categories was compared by two AN(C)OVA models, one with and the second without using total PA as a covariate. Analyses were performed in NCSS 9 (NCSS, Kaysville, UT, USA).

Seed size and seed coat thickness measurements

Ten individual seeds per accession of all together 86 accessions (Table S3) were photographed using an Olympus SZ61 stereo microscope (Olympus Corp., Tokyo, Japan) equipped with an Olympus E-410 digital camera (Olympus Imaging Corp., Tokyo, Japan). The photographs were processed by QuickPHOTO MICRO 3.0, supplemented by the Deep Focus 3.3 module (PROMICRA s.r.o, Prague, Czech Republic). Image analysis of seeds was done by SmartGrain (Tanabata et al., 2012) recording seed area (AS), perimeter length (PL), width (W), length (L), length-to-width ratio (LWR), circularity (CS) and distance between intersection of length and width and centre of gravity (DS). The weight of twenty five seeds per accession with three replicas was recorded and converted to hundred seeds weight (HSW). Seed coat thickness of the stripped seed coat from five seeds per accession was measured in triplicate using a precision micrometer (0–25 mm, precision 0.01 mm, Hommel Herculer, Germany) and sectioned on cryo-microtome according to Hradilová et al. (2017). For the analysis of dormancy with respect to morphological parameters of seeds we used ordinal logistic regression (Kleinbaum & Klein, 2010) implemented in R package MASS (Venables & Ripley, 2002). Regression parameters were estimated using the maximum likelihood principle. The relative quality of specific models was assessed by Akaike information criterion (Akaike, 1974).

Effect of temperature on pea seed germination

Ninety seven accessions were tested over a period of 28 days and the final percentage of germinated seeds ranged from 0% to 100% (Table S2). Cluster analysis of B-spline coefficients representing germination responses divided the accessions into two groups in the 25/15 °C temperature treatment (Fig. S1A). One cluster held 29 non-dormant (N) accessions with a mean germination of 92% and the second cluster had 68 dormant (D) accessions with a mean germination of 15% (Table S2). The non-dormant group was further divided into those that germinated quickly (N1, 13 accessions, with over 80% germinated at 10 days and 100% at 28 days) and more slowly (N2, 16 accessions germinating between 50 to 80% at 10 days and up to 80% at 28 days) (Table S2). The group of dormant accessions was further divided into D1 sub-group of 14 accessions with a mean germination of 31%, D2 subgroup of 47 with a mean germination of 6% and D3 subgroup of 7 accessions with a mean germination of 54%. In D2 subgroup there were 16 accessions which did not imbibed and germinated at all during 28 days of testing.

In the 35/15 °C temperature treatment, cluster analysis of B-spline coefficients resulted in two main clusters: one with 36 non-dormant (N) accessions with a mean germination of 95% and second with 61 dormant (D) accessions with a mean germination of 28%. The group of non-dormant accessions contained 29 accessions classified as non-dormant already at 25/15 °C, plus additional seven accessions (Fig. S1B).

When germination responses at 25/15 °C and 35/15 °C were analyzed together, UPGMA clustering identified three groups (Fig. 2). One, with 28 non-dormant (N) accessions largely corresponding to N group at 25/15 °C treatment. The R1 group with eight accessions having large positive response at 35/15 °C (with change from 15% at 25/15 °C to up to 80% germination at 35/15 °C). Second responsive R2 group had 33 accessions with moderate response from 16% at 25/15° C to up to 60% germination at 35/15 °C). Finally, there were 29 accessions which remained dormant (e.g., with final germination bellows 20% at 35/15 °C, with 19 accession having less than 10%).

Relationships between germination pattern, geography and environment

Each environmental variable was significantly spatially structured. Significant positive and negative spatial correlation coefficients were observed at small and large distance classes, respectively (Table S4).

The first principal component of the PCA captured 87.3% of the variation of the B-spline coefficients matrix used to describe the germination pattern under both experimental temperature treatments and showed a clear gradient with dormant accessions on the left through responsive ones in the middle to the non-dormant accessions on the right of the ordination diagram (Fig. 3). Scores of accessions along the first ordination axis were therefore considered to be an indicator of the germination response of these accessions to experimental germination treatments (‘germination responsivity’). Germination responsivity was found to be significantly spatially structured. Significant positive spatial correlation coefficients were observed at two smallest distance classes (up to 100 and 200 km, respectively), negative spatial correlation was observed at the distance class 300–600 km, while nonsignificant spatial correlations were found for larger distance classes (Fig. S2).

Only subset of the environmental variables containing four bioclimatic variables (BIO1, BIO6, BIO7, BIO8) and one soil variable (CRFVOL) projected onto the two dimensional PCA diagram (PC1 and PC2 axes) were significantly correlated with at least one ordination axis after correction on spatial autocorrelation (Fig. 3, Table S5). However, only BIO7 was significantly correlated with PC1 axis representing germination responsivity (Fig. 3, Table S5).

The analysis of pure and shared effects of space (PCNM variables) and environmental variables (soil and climatic variables tested separately) on B-spline coefficient matrix representing germination pattern showed that all variables had an overall low explanatory power (space + soil variables: 13.6%; space + climatic variables: 17.4% of total variation in B-spline coefficient matrix; Table S6). PCNM variables accounted for more than a half of explained variation and always had significant effect on B-spline coefficients matrix. Shared effects between space and environmental variables (either soil or climatic) were non-significant on B-spline coefficients matrix. Overall pure effects of soil variables were non-significant on B-spline coefficients matrix, while climatic variables had significant overall pure effects on B-spline coefficients matrix. Using a forward selection procedure resulted in the stop of the selection process after the selection of BIO7 as the first and significant candidate term. Final RDA model thus included PCNM variables and BIO7 and explained 11.1% of total variation in B-spline coefficients matrix (Table S6). However, pure effect of BIO7 on B-spline coefficients matrix was weak (4.7% of total variation) and when three dormancy categories classified on the basis of cluster analysis were projected into resulting diagram, only poor separation among categories were visible (Fig. 4). Specifically, non-dormant and responsive accessions covered almost whole range of BIO7 while dormant accessions avoided occurring at sites with high values of BIO7, i.e., with larger temperature annual range.

The ecological niche modelling of potential distributions of the dormant (D), temperature responsive (R) and the non-dormant (N) groups obtained at 35/15 °C testing showed the differences in spatial patterns (Fig. 5, Fig. S3). The dormant group (D) shows a more localized potential distribution. Results of the niche equivalence tests (Fig. S3) further suggest that any observed geographical differences among dormancy groups are not produced by an underlying divergence in the niche.

Seed size and seed coat thickness relationship to dormancy level

Except for LWR having weak positive relationship towards non-dormancy, no clear relationships between categorized germination responses in the 25/15 °C treatment and seed morphological characters were observed (Fig. S4A). Similar results were obtained even when ordinal logistic regression with each morphological trait as covariate was employed (according to AIC simple logistic regression models were preferred over one multiple model); again the only significant covariate was LWR with P = 0.008. The ordinary logistic regression revealed a slight (positive) impact of the LWR covariate (P = 0.064) on dormancy release at 35/15 ° C treatment (Fig. S4B). Seed coat thickness differed between seed dormancy categories in the 25/15  °C treatment (Kruskal–Wallis test, χ ² = 12.4, P < 0.001) with dormant accessions having thicker seed coat (median 139 µm, 95% CI [128–146] µm) than non-dormant ones (median 97 µm, 95% CI [65–120] µm) (Fig. 6A, Fig. 6B, Fig. 7, Figs. S5A, Figs. S5B). Similar results were obtained in the 35/15 °C treatment (Kruskal–Wallis test, χ² = 11.7, P = 0.003), with both dormant (median 138 µm, 95% CI [123–150] µm) and responsive accessions (median 140 µm, 95% CI [110–152] µm) having similar and thicker seed coat than non-dormant ones (median 84 µm, 95% CI [58–117] µm). However, non-dormant accessions showed more variable seed coat thickness, reaching also the high values typical of dormant accessions (>80 µm), while seed coat thickness of dormant and responsive accessions did not fall below ca 80 µm (Figs. 6A, 6B, 7, Figs. S5A, S5B).

Content of proanthocyanidins in the seed coat

The amount of soluble and insoluble PAs was measured in the isolated seed coat of mature dry seeds (Table S3). The total content varied from 1.21 to 4.70 mg g⁻¹ of dry seed coat and showed weak tendency to be higher in the seed coat of dormant (mean ± SE; 2.18 ± 0.13  mg g⁻¹) that those of non-dormant (mean ± SE; 1.77  ± 0.14 mg g⁻¹) and responsive accessions (mean  ± SE; 1.87  ± 0.12 mg g⁻¹; ANOVA, F = 2.64, P = 0.079). Soluble PA represented dominant component of total PA (94–99%) while insoluble PA were in minority (1–6%). However, ratio soluble/insoluble PA was not constant and increased with increasing total PA in all dormancy categories (Fig. S6). Correcting for total PA (covariate, F = 26.26, P < 0.001), dormancy categories recognized at 25/15 °C treatment significantly differed in sPA/inPA ratio (F = 4.6, P = 0.036) with dormant accessions having higher ratio (LS adjusted mean at mean total PA, 95% CI; 23.5, 20.8–26.4) than non-dormant ones (19.1, 16.3–22.2). Using the same model but with three dormancy categories (including responsive one; 35/15 °C treatment) showed similar trend to that of the previous model but the effect of dormancy category on sPA/inPA ratio was non-significant (total PA, F = 15.52, P < 0.001; dormancy categories, F = 2.59, P = 0.086) (Fig. S6). Custom comparison (D+R vs. N) showed that both dormant and responsive accessions have higher sPA/inPA ratio that non-dormant accessions (F = 4.46, P = 0.039).

The relationship between environmental conditions and seed dormancy: patterns, potential limitations and alternative explanations

Despite expected relationships between pattern of seed dormancy release observed in our experiments and in situ environmental conditions of tested pea accessions, our results showed that germination responsivity (i.e., dormancy release) was weakly related to one climatic variable only: temperature annual range (BIO7). Specifically, dormant accessions does not occur at sites with large temperature annual range (Figs. 1, 3A, 3B, 4, Table S5). Using niche analysis we also showed that the dormant group (D) shows a localized potential distribution compared to the responsive (R) and non-dormant (N) groups (Fig. 5). This is a rather unexpected result, especially in the Mediterranean seasonal environment. This is further reinforced by the fact that the high probability areas of predicted occurrence constitute only a very small fraction of the high climate seasonality areas. Consequently, results of PCA, RDA and niche modeling in pea are not in agreement with metadata collected on wide range of legume species (Rubio de Casas et al., 2017) concerning the relationship between dormancy and seasonality. That study found that dormancy should increase in more seasonal environments. Of course, the niche patterns of both studies (Rubio de Casas et al., 2017; our study) could in part be attributed to scale differences (both spatial and taxonomic), but the need for an explanatory framework of wild pea dormancy in the Mediterranean region remains. Similar relationship between dormancy and environment as in pea was found by Wagmann et al. (2012) studying wild beet seed dormancy release. In contrary, two recent studies of Mediterranean wild lupines (Berger, Shrestha & Ludwig, 2017) and perennial woody legume (Vachellia aroma) along a precipitation gradient (Ferreras, Zeballos & Funes, 2017) showed no significant relationship between dormancy release and rainfall or temperature gradients. How could the dormancy-environment pattern we observed in pea be explained?

One set of explanations is related to (i) methodological constraints related to inaccurate estimations of in situ environmental conditions and (ii) specific conditions used in experimental germination treatments that might leave aside other important factors affecting in situ dormancy release in pea. Firstly, it can be speculated that the character of WorldClim data (average in term of time and also space) masks the micro-ecological pattern (Smýkal et al., 2018a). In geomorphologically complex regions, environmental conditions change considerably over short spatial scales, such that neighbouring populations can be subject to different selective pressures. This was found in the study of seed dormancy of Swedish A. thaliana accessions, with general relationship to the latitude and climate variables, with northern lines generally being less dormant than southern ones (Kerdaffrec & Nordborg, 2017) but with detected exceptions related possibly to local scale. Secondly, in their original habitat the plants grow in very distinct environment (temperature and rainfall patterns) and consequently the dormancy status of produced seeds might differ from seeds we have been testing in our common garden experiments, as shown in Arabidopsis (Edwards et al., 2017). This would require on site seed collection and/or reciprocal common garden experiment (Wagmann et al., 2012; Exposito-Alonso et al., 2018), both being essentially prevented by current plant genetic resources handling methodology (Nagoya protocol, Secretariat of the Convention on Biological Diversity United Nations Environmental Programme, 2011). Since we have not tested seeds in natural environment, we do not know if there is a cyclic pattern of germination as reported for several legumes with physical dormancy (Taylor, 1981; Taylor, 1996; Van Assche, Debucquoy & Rommens, 2003; Taylor, 2005). Seasonal testing of germination capacity of 14 annual legumes (of Trifolium, Medicago, Melilotus, Vicia) suggested the importance of chilling (low winter temperatures), prior to the submission to oscillating temperatures (Van Assche, Debucquoy & Rommens, 2003). Most of these taxa are temperate species, while in the case of Mediterranean and middle East origin wild pea Pisum sp. (Smýkal et al., 2017) chilling is expected to play only minor role as seedling establishment in natural conditions occurs from early to late autumn (recorded observations).

It also remains to be elucidated how a combination of temperature and moisture treatments influence the water entry, as shown for Ipomoea lacunosa (Jayasuriya et al., 2009) and comparison of wild and cultivated pea (Janská et al., 2018). The effect of substrate moisture was far less studied, partly because of problems with its control and manipulation. In natural conditions, the oscillations of temperature are accompanied by soil moisture changes. These are not fully recapitulated in laboratory experiments. Commonly used PEG solution (Michel & Kaufmann, 1973) provides only static conditions. In our experimental system we did not manipulate the water potential, and tested only water saturated conditions. Moreover, we used fully matured dry seeds from glasshouse grown plants to minimalize the seed source variability and to allow the experimental testing of seed responses and analysis of seed coat properties. In our study we used fully mature dry seeds obtained from glasshouse grown plants to minimalize the influence of the environment (Chen et al., 2014; Springthorpe & Penfield, 2015; Penfield & MacGregor, 2017; Renzi, Chantre & Cantamutto, 2018). Finally, it was found that both seed-maturation temperature and the timing of dispersal, strongly influenced germination behaviour (Edwards et al., 2017). If such situation is present in dehisced pea seeds under natural conditions remains to be tested. In addition to abiotic environmental (such as tested temperature or moisture) also biotic factors can have significant role. Thus we cannot exclude the action of microbes and fungus as recently demonstrated for Lepidium didymium (Brassicaceae) seeds released from dormancy by fungal degradation of the pericarp (Sperber et al., 2017). However, pea seed coats contain numerous antimicrobial substances including phenolic compounds (Pourcel et al., 2007; Hradilová et al., 2017; Raviv et al., 2017) which might prevent this.

Possible alternative explanation can be that pea seed dormancy is a bet-hedging mechanism in fluctuating (unpredictable) environments, in agreement with Volis & Bohrer (2013). Most of the high probability areas of the dormant (D) group are restricted to regions such as Israel, which has high summer temperatures and low temperature seasonality, a pattern also supported by the PCA and RDA results. These areas are very close to semi-desert environments and are characterized by higher unpredictability compared to the wet edge seasonal Mediterranean environment, which is more predictable. This is in agreement with models of Holst, Rasmussen & Bastiaans (2007) where seasonal fluctuations in temperature are considered the main regulatory factor of annual dormancy cycling (Bewley et al., 2013), while in semiarid and arid regions with high inter-annual rainfall variability, soil moisture is a major determinant for seedling emergence. As such, it could be concluded that for wild pea, dormancy is a bet-hedging mechanism developed to overcome the most unpredictable seasonal environments of the Mediterranean region. This explanation corresponds with the view that if conditions at germination are predictive of future survival and reproduction, then the value of bet-hedging decreases (Clauss & Venable, 2000). However, there is an area of overlap between the predicted distributions of the dormant (D), responsive (R) and the non-dormant (N) groups. It seems that this overlap is a mechanism of risk spreading, obtaining a long term selective advantage, which is a very successful strategy to adapt rapidly to changing environments. According to Philippi & Seger (1989), bet-hedging strategies are of two main forms: conservative and diversified. Conservative bet-hedging strategies tend to produce ever safer phenotypes, while diversifying bet-hedging strategies distribute the risk among two or more phenotypes. If seed dormancy is considered as a bet-hedging mechanism, then it can be viewed as the conservative form, producing the safest phenotypes; if we consider the niche overlap of the three groups, then the diversified, risk-spreading form is also apparent. Furthermore, the absence of strong environmental determination of environment-dormancy relationship, which is in consistency with the results of the niche equivalency tests, suggests that this bet-hedging mechanisms is functioning to cope with fine unpredictable fluctuations in similar environmental conditions.

The high probability areas of the non-dormant (N) group coincide with the Middle East and the Fertile Crescent, where the agricultural revolution and pea domestication took place (Trněný et al., 2018). Two hypotheses can be formulated with regard to this pattern. According to the first, mentioned above, the non-dormant group could be potentially domesticated pea escaped from cultivation, and/or an introgression group between wild and domesticated pea (Trněný et al., 2018). However, genomic analysis has revealed that only nine non-dormant accessions might be indeed genetically related to cultivated pea (IG52518, 52520, IG52442, IG52596, JI1089, JI2105, JI2201, JI2546 and P014) while the remaining 20 non-dormant accessions have clear genetic relationship to various clusters of wild P. elatius and are distant to cultivated pea (Trněný et al., 2018). On the other hand, if we assume that non-dormancy was a pre-domestication trait of wild pea even in very low frequencies within a bet-hedging diversified mechanism, our data support the view that during pea domestication such genotypes were selected and fixed in the pea crop.

Seed size, seed coat thickness and proanthocyanidins content relationships to the dormancy

In our study we have found differences in total PA content and ratio of soluble to insoluble PAs in relationship to dormancy levels (Table S3, Fig. S6). Inhibitory effect of PAs in the seed coat on germination was found in Sapium sebiferum (Euphorbiaceae) where the seed coat extract or PAs significantly inhibited germination via alteration of gene expression (Afzal et al., 2017). This would be certainly the conclusion when non-pigmented cultivated pea varieties were compared to wild accessions (Hradilová et al., 2017). Moreover also in legumes, pigmented seeds imbibe slower than non-pigmented ones and also germinate later (Wyatt, 1977; Werker, Marbach & Mayer, 1979; Kantar, Pilbeam & Hebblethwaite, 1996). Notably, insoluble PAs were shown to contribute to Rubus seed dormancy (Wada, Kennedy & Reed, 2011). Variation in PA content in the cultivated pea seeds has been reported (Troszynska & Ciska, 2002) but not in comparison to wild peas. Our data of PA content (Fig. S5) are in agreement with early studies (Marbach & Mayer, 1974; Werker, Marbach & Mayer, 1979) reporting positive correlation in content of phenolics, activity of catechol oxidase and pea seed dormancy, although they compared only wild versus domesticated accessions. We can speculate about the alteration of seed coat composition, including the oxidation of fatty acids (Gamma-Arachchige et al., 2013; Cechová et al., 2017) or polyphenolic compounds (Hradilová et al., 2017) during the temperature treatment (Janská et al., 2018). Indeed in Medicago truncatula cells of the outer integument showed abundant accumulation of polyphenolic compounds which may impact seed permeability (Moïse et al., 2005). Our analysis was based on spectrophotometric measurements of total content and we cannot exclude that specific PAs can be more specifically associated with dormant seed coat, as revealed by detailed analysis of pea seed coat (Hradilová et al., 2017) as well as soybean (Zhou et al., 2010). It remains to be tested if there are any specific PA compounds related to dormancy levels in wild pea.

Intraspecific variation of seed size in relation to dormancy was so far only rarely tested. Comprehensive study of legume species (Rubio de Casas et al., 2017) showed that smaller, dormant seeds are favoured in the environments where the growing season is short, whereas in the tropics, large non-dormant seeds dominate. In our study, we have found no relationship between seed weight or seed size parameters and dormancy, except for length-to-width ratio (LWR) being slightly positively correlated with non-dormant seed types (Figs. S4A, S4B). There might be some influence of the shape of seeds (Fig. 7) but obviously not the size of seeds themselves. During crop domestication (Smýkal et al., 2018b) including the legumes (Berger, Shrestha & Ludwig, 2017), selection acted toward seed size increase. At intraspecific level, in the case of bird vetch (Vicia cracca) the positive relationship between seed size and germination (Eliášová & Münzbergová, 2014) was observed but in relation to the ploidy level. This is not a case of diploid Pisum. From a reproductive point of view, seed size trades off with seed number, small-seeded species clearly have the advantage in fecundity. The advantage of large seeds appears to be their tolerance of stresses such as shade or drought (Rees, 1996). If this operates in wild pea, it remains to be tested. Positive correlation between testa thickness and seed dormancy is in agreement with several reports (reviewed in Baskin & Baskin, 2014) and confirms our previous findings in pea (Hradilová et al., 2017; Janská et al., 2018).