An F₈+ recombinant inbred line (RIL) population (n=137) was derived by single seed descent from the F₂ of a cross between the wild P. sativum ssp. elatius line JI1794 (a representative of the northern “humile” subgroup; Hellwig et al., 2022) and the cultivar NGB5839. This population was described previously described by Weller et al. (2012) and Williams et al. (2022). This population was grown under long-day (LD) conditions (16 hours light – 8 hours dark), with 4 replicate plants per genotype. Plants were grown in a 1:1 gravel:vermiculite mixture, topped with sterilized potting mix which included controlled release fertilizer. Seeds were harvested after plants had completely senesced and dried, and were stored for at least a month to ensure they had fully matured prior to their use in permeability analyses.

The harvested seeds of the RIL population and parental lines were assessed for several traits potentially related to physical dormancy ( Table 1 ). To estimate seed coat permeability, we measured time to fully imbibe with water by submerging seed in water and retrieving, drying and weighing at regular intervals until there was no further increase in mass due to water uptake. This was measured in both young (1 month old mature dried) and old (2 year old mature dried) seed. The relative water uptake capacity of the dry seed was characterized by the relative increase in weight of the seeds at full imbibition. The thickness of the testa was initially measured on detached fragments using a micrometer, but to verify the precision of these measures, further seed coat dimensions were determined from transverse sections mounted on slides, from three representative seeds for each genotype. Images were captured at 10X and 40X magnification to allow for accurate delineation of cell layers and anatomical zones. The total thickness of the testa and the width of distinct component layers ( Supplementary Figure S1 ) were measured using the line tool in ImageJ (Schneider et al., 2012), while roughness was measured as the ratio of the outer cuticle length over the light line length ( Figure 1A , Supplementary Figure S1 ). Mean seed weight was recorded from ten representative seeds per genotype. Variation in seed pigmentation traits was also recorded, including for Mendel’s A locus governing flower and seed anthocyanin content and several other classical seed pigmentation loci.

Genomic DNA was extracted from young leaflets of each genotype using the CTAB extraction protocol (Doyle and Doyle, 1987), and DarTseq markers (Sansaloni et al., 2011) generated using Diversity Array Technology Pty. Ltd. (Canberra, Australia). A total of 4,599 markers (4,575 DArT and 24 gene-based markers) were used to construct a linkage map (previously described by Williams et al., 2022). The linkage map was subsequently reduced to 3073 markers for quantitative trait loci (QTL) analysis by removing every third marker, which improved computational efficiency while having minimal effect on mapping resolution, and some linkage groups inverted to maximize synteny with the Pisum genome assembly (Kreplak et al., 2019).

QTL analysis was performed using MapQTL v6 (Van Ooijen, 2009), following Williams et al. (2022). In brief, QTLs were defined by a > 3 LOD score and identified using the interval mapping (IM) function. Iterative searches for additional QTLs were performed using the restricted Multiple QTL Model (rMQM) function, which increases the power of QTL analysis by reducing residual variances attributed to previously identified QTLs (cofactors). The amount of variation explained by each QTL was estimated using the coefficient of determination (R²) which is represented as the Phenotypic Variance Explained (PVE).

QTL regions of interest were refined in segregating F₃, F₄ and F₅ progeny derived from specific individuals in the original F₂ population. Where possible, these progenies were selected to be homozygous (fixed) for other relevant loci influencing the focus trait. Mapping resolution within these selected QTL regions was increased through use of additional high-resolution melt (HRM) markers, either already available or newly developed from pea transcript sequences with selection guided by DArT marker positions in the RIL linkage map or the high-density consensus map of Tayeh et al. (2015). Progenies of these advanced generations were grown under the conditions described above, and phenotyped for permeability, roughness (as a binary presence/absence trait) and testa thickness (via micrometer).

A comparison between the wild P. s. humile line JI1794 and the domesticated P. s. sativum cultivar NGB5839 illustrates the significantly thicker testa in the wild line, and the undulating outer surface characteristic of the ‘gritty’ phenotype ( Figure 1A ). Recently-matured (one-month-old; “young”) dry seed of JI1794 did not imbibe or germinate even after more than 100 days of immersion in water, whereas similar seed of NGB5839 started to take up water within one hour of immersion, reached 50% imbibition within 4h ( Supplementary Figure S2 ) and reached 100% imbibition after 24h ( Figure 1B ). To examine the potential influence of seed age on imbibition, we also compared JI1794 seed of various ages post-harvest. Unlike young seed, 80% of older seed that had been stored for two years in a relatively stable, cool and low-humidity indoor environment commenced imbibition within 6h of immersion and reached the fully-imbibed state by 24h ( Figure 1B , failure to imbibe not shown). In contrast, this same period of storage had no effect on imbibition of NGB5839. A comparison of storage duration on permeability in JI794 (in this case detecting partial imbibition after 24h of immersion) revealed that the initial complete impermeability persisted for between 6 and 9 months, after which imbibition increased to reach 100% after 4 years of storage at relatively constant temperature and humidity ( Figure 1C ). This suggests that the impermeability of wild seed coats is lost as the seed ages.

We next examined variation for seed coat traits potentially related to dormancy and domestication in the JI1794 x NGB5839 RIL population. This included the thickness of the seed coat and its component layers, permeability (time to fully imbibe) and the roughness or ‘grittiness’ of the seed coat, as the main traits potentially related to dormancy. We also examined seed weight, water uptake capacity and various pigmentation features. Mean data for all traits and a correlation matrix are provided in Supplementary Table S1 and Supplementary Figure S3 , with notable correlations discussed below.

Total testa thickness varied from 76 – 160 µm and exhibited a near-normal distribution in the population ( Figure 2A ). To examine whether specific anatomically distinct layers within the testa might vary in thickness and how this might contribute to the total thickness, we also defined specific zones within the transverse section (testa sections TT-A through TT-F; Supplementary Figure S1 ) and measured these individually. The largest component was section TT-D of the macrosclereid cell layer, which is the main component of the testa. This section contributed 38 – 64% of the total variation for testa thickness across the population. Testa thickness showed a positive correlation with water uptake capacity (r = 0.20, p < 0.05), along with a positive correlation with seed coat roughness (r = 0.28, p < 0.01). Testa thickness also showed a moderate negative correlation with seed weight (r = -0.26, p < 0.01), but interestingly a much stronger negative correlation with seed weight was found for the testa sections TT-A, TT-B and TT-C (r = -0.54 - -0.77, p < 0.001). Likewise, the presence of flower pigmentation had a moderately positive correlation with testa thickness (r = 0.30, p < 0.001), but this correlation was much higher when considering the thickness of testa section D (r = 0.49, p < 0.001).

Permeability varied widely, with imbibition time ranging from 2 hours to more than 70 days, heavily skewed towards a shorter time to imbibe. To adjust this skew prior to analysis, permeability scores were log transformed, creating a bimodal distribution ( Figure 2B ). This measure of permeability was moderately correlated to total testa thickness, with thicker testa leading to a longer time to imbibe (r = 0.34, p < 0.001, Figure 2C ). Again, this effect was more pronounced when considering the thickness of testa section D (r = 0.41, p < 0.001, Figure 2D ). Permeability was not affected by seed coat roughness (r = 0.01, p > 0.05). Likewise, no correlation was detected between permeability and water uptake capacity (r = -0.04, p > 0.05), suggesting that seed water capacity was unrelated to imbibition rate. Time to fully imbibe was also positively correlated with presence of flower pigmentation (r = 0.55, p < 0.001).

Variability in seed permeability to water was found to be under genetic control in this population, with two QTLs discovered on chromosomes 6 (qPERM6) and 7 (qPERM7), respectively explaining 33.0% and 9.1% of the observed variation ( Table 2 ). The qPERM6 locus was located over the region containing Mendel’s A ( Figure 3 ), raising the possibility that A itself might influence permeability via its effects on seed-coat composition. In contrast qPERM7 was in a region distinct from any known loci likely to have an influence on seed coat properties.

Three QTL were detected for total testa thickness (TT) as measured by sectioning, on chromosomes 2, 6 and 7. Of these, the qTT2 locus explained 25.5% of the observed variation and was located near the top of chromosome 2. qTT7 was located in a region distinct from qPERM7, but qTT6 was closely collocated with qPERM6. Collectively, these comparisons indicate that the genetic control of physical dormancy is at least in part independent of testa thickness, but also imply the possibility of a relationship between these traits. Additional QTL were detected when testa thickness was measured by micrometer, but the presence of collocated QTL for other traits at these additional loci suggests a measurement precision issue (see below).

In order to examine whether variation for testa thickness might reflect a specific contribution from certain cell layers or structural features we also looked at the genetic control of the thickness of the six distinct testa layers described above ( Supplementary Figure S1 ). Significant loci were found only for testa layers D, E and F, with five QTL for qTT-D2 explaining 78.1% of the variation. The three strongest qTT-D QTL collocated with the qTT2, qTT6 and qTT7 loci, and explained 53% of the variation. Two weaker TT-D loci were detected on chromosomes 5 and 3, together explaining another 15% of the variation. The only locus detected for layer E was in a position on chromosome 6 distinct from that of qTT6 and qTT-D6, while the two loci detected for layer F co-located with the qTT2/qTT-D2 and qTT7/qTT-D7 loci. Overall, these results indicated that the observed variation in total testa thickness was primarily due to variation in layer D, and revealed the contribution of three minor loci on chromosomes 3, 5 and 6.

In addition to Mendel’s A, the four seed pigmentation traits measured all related to well-known classical loci, and major QTL were detected in the expected genomic locations when analyzed as qualitative traits ( Figure 3 , shown as peak position). These included the hilum pigmentation locus Pl on chromosome 1 (Lamprecht, 1948; Balarynová et al., 2022), the seed-coat “marbling” locus M on chromosome 5 (Lamprecht, 1948; Ellis et al., 2023), the seed coat speckling locus Fs on chromosome 3 (Lamprecht, 1961; Ellis et al., 2023) and the seed coat ground color locus GLA on chromosome 4 (Lamprecht, 1959; Blixt, 1962). Each of these also had a secondary QTL which collocated with Mendel’s A, reflecting their likely dependence on the action of A. However, with the exception of the A locus itself, none of these were collocated with effects on testa thickness or permeability.

Seed coat roughness has long been noted as a key distinguishing feature of wild and domesticated pea seeds; a difference attributed to a major Mendelian locus GRITTY located on chromosome 1. Assessment of this trait has typically been made somewhat subjectively based on the degree of friction encountered when rubbing seeds together (Ellis et al., 2023). We quantified it more objectively, examining seed coat sections under the light microscope and taking quantitative measurements of the surface undulations. A major QTL explaining 45% of the variation (qGRIT1) was detected on chromosome 1, in the expected position of the GTY locus. A second, minor locus (qGRIT7) explaining a further 8.8%, was located near the cluster of TT loci on chromosome 7. These results suggest that the GRITTY trait has no major genetic association with testa permeability, and only a minor association with testa thickness. Interestingly, an additional QTL for testa thickness was discovered tightly linked to qGRIT1 when measuring testa thickness using a micrometer. Given this QTL was not present when using the more accurate sectioning measurement, this likely reflects the additional apparent thickness conferred by the undulations of the seed coat when roughness is present, highlighting the need for precision in measurements when working at this scale.

Genetic control of seed water uptake and seed weight was also detected in this cross, with QTL for these traits often collocated with testa thickness QTL ( Figure 3 ). A common locus for both was found on chromosome 5, with an additional seed weight QTL present in the QTL cluster on chromosome 6, and an additional water uptake QTL collocating with the testa thickness section E QTL on chromosome 6 (in a distinct location). There was also an independent QTL for water uptake on chromosome 3, suggesting additional physiological phenomena beyond testa thickness affecting this trait.

As the largest permeability QTL was located on chromosome 6 in a region also featuring Mendel’s A locus and QTL for physical seed characteristics, we attempted to refine the position and determine whether it might be distinct from the A gene. We also further refined the position of the main seed roughness QTL (GRITTY) as a next step in status as a key domestication trait.

The presence of the wild allele at QTL6 conferred significantly increased testa thickness (t₁₂₉ = -3.49, p < 0.001) and lower permeability (t₁₁₂ = -6.75, p < 0.001) compared to the domesticated, conforming to findings in the RIL population for this QTL ( Figure 4A ). To validate this QTL, further advanced segregation populations were developed using single seed descent from an F₂ line which was segregating for the qPERM6 locus (markers LF and A) and fixed at all other loci of interest (specifically fixed for the domesticated allele at qPERM7), thereby allowing the analysis of this QTL as a near Mendelian trait in the F₄ and F₅ populations developed. For greater resolution around qPERM6 region (spanning between 38.4cM and the A_1 marker) eight additional markers ( Supplementary Table S2 ) were genotyped across these populations ( Figure 4B ). Markers were ordered based on recombinant frequency or (if there was no recombination) by order in the Medicago (v4.0, Tang et al., 2014) or P. sativum (v1a, Kreplak et al., 2019) reference genomes.

Results showed all individuals in these populations were fixed for wild alleles between RPS27 and CWF and segregating at all other loci. Phenotyping these populations (separately, due to the destructive nature of the phenotyping) confirmed the segregation of testa thickness in the F₄ and permeability in the F₅ ( Figure 4C ). Homozygosity for the domesticated allele at the Mendel’s A gene resulted in significantly thinner testa (t₄₇ = 13.1, p < 0.001) and higher permeability (t₄₇ = 3.4, p = 0.001) compared to the heterozygous and homozygous wild individuals, indicating RPS27 as the upper boundary for this QTL effect. Further, when only individuals fixed for wild Mendel’s A were analyzed, genotype at NAD had no significant effect on permeability in the F₅ ( Figure 4D ), indicating NAD as the lower boundary for this QTL effect.

Of the additional markers previously genotyped in the RIL population (Williams et al., 2022), the markers FULa, AGO1 and CABB were found to span the peak of the seed coat roughness QTL ( Figure 5A ). To confirm the validity of this QTL, advanced F₄ populations (and a subsequent F₅ population) were generated by single seed descent from lines in the F₂ cross which were segregating for the region between CABB and FULa on chromosome 1 but fixed for the wild allele at qPERM6 and the domesticated allele at qPERM7. Phenotyping these populations showed segregation of seed coat roughness, with the QTL effect able to be localized between FULa and CABB, a region that spans 37 MB and contains 331 genes ( Figure 5B , Supplementary Table S3 ). Notably, genotype at AGO1 had no effect on seed permeability ( Figure 5C ), confirming that seed coat roughness had no detectable impact on the permeability of the seed coat to water.