Selection of parents for generating RILs

The selection of parental lines from the panel of 48 cultivars was based on identifying maximally contrasting lines for both breeders’ priority traits and genetic distance, using prior phenotypic data gathered from field trials of the panel of cultivars and genetic marker diversity data, respectively. A GGE biplot analysis (Yan et al., Reference Yan, Hunt, Sheng and Szlavnics2000) of the panel, based on a PCA of data collected as part of breeders’ trials and relating to phenotype scores for four traits: overall yield, standing ability, downy mildew resistance and seed protein concentration, is shown alongside supporting data in online Supplementary Table S2. Fig. 1 shows an analysis of genotype data for the cultivar set, based on scores for 152 genetic (PDR1 SSAP) markers. Genotype data were collected (as SSAP marker band presence/absence scores) for the cultivar set plus the JI germplasm reference accessions (designated JI lines), the latter of which included the parents of diverse mapping populations, described previously (Ellis and Poyser, Reference Ellis and Poyser2002; Vigeolas et al., Reference Vigeolas, Chinoy, Zuther, Blessington, Geigenberger and Domoney2008; Ellis et al., Reference Ellis, Hattori, Cheema, Donarski, Charlton, Dickinson, Venditti, Kaló, Szabó, Kiss and Domoney2018), and included biological replicates for several lines. An example gel used for genotyping is shown in online Supplementary Fig. S1. Fig. 1(a) shows the phylogenetic relationship of the cultivars within the panel in relation to 10 JI reference lines. Most but not all of the JI reference lines are separated and shown at the upper edge of the tree (Fig. 1(a)). The data indicated that the cultivars could be distinguished genetically from each other and clearly from JI 15 and JI 281 (Fig. 1(a)), which represent the very diverse parents of sets of RILs involving JI 15, JI 281, JI 399 and JI 1194 (Hall et al., Reference Hall, Parker and Ellis1997a, Reference Hall, Parker, Ellis, Turner, Knox, Hofer, Lu, Ferrandiz, Hunter, Taylor and Baird1997b; Ellis et al., Reference Ellis, Hattori, Cheema, Donarski, Charlton, Dickinson, Venditti, Kaló, Szabó, Kiss and Domoney2018). The relationship between two JI germplasm genotypes, JI 1194 and JI 1201, should be noted as being closely adjacent. These are two near-isogenic lines (developed by G.A. Marx), with contrasting alleles for three loci that regulate leaf development (afila, af; stipules-reduced, st, tendril-less, tl). It is noteworthy that JI 813 lies close to cultivars of the marrowfat class (Fig. 1(a)); JI 813 is derived from the marrowfat cv. Vinco.

Fig. 1. (a) Phylogenetic relationships among the panel of 48 cultivars in comparison with 10 JI reference lines (JI numbered lines and cv. Birte), based on the genetic analysis of 152 PDR1 SSAP markers. Where there were marker disagreements between duplicate samples of the same accession, these are numbered separately (JI 15a, b; JI 399-1, -2). The positions adopted by the three cultivars chosen as parents are highlighted by the green arrows. The bar indicates the distance matrix scale, as determined from neighbour-joining phylogenetics. (b). Relationships among the panel of 48 pea cultivars, based on PCO analysis of the genetic marker data, with a projection of the genetic variance data onto planes of the two leading dimensions. The positions adopted by the lines chosen as parents are indicated (black circles). The % variation explained by the two dimensions is indicated (32% for PC1, 29% for PC2).

The dataset comprising 152 polymorphic markers was used to calculate a distance matrix of (dis)similarity. Compression by principal coordinate analysis (PCO, Fig. 1(b)) showed that at least two major groups of accessions could be distinguished, one of which included marrowfat types (e.g. the cultivars Maro, Princess, Kahuna and Samson, clustered in the right-hand side of the plot). In this plot, the proportion of variance in the first two dimensions is similar and accounts for about 60% of the genetic variation among the cultivars.

Analysis of the marker data obtained for the cultivars, using the population genetics programme ‘Structure’ (Pritchard et al., Reference Pritchard, Stephens and Donnelly2000; Jing et al., Reference Jing, Vershinin, Grzebyta, Shaw, Smýkal, Marshall, Ambrose, Ellis and Flavell2010), facilitated a comparison of the chosen parents with the cultivars as a whole (Fig. 2). The ‘Structure’ programme takes an objective approach to propose common progenitor populations for a given set of genotypes, based on estimations of the number of progenitor populations (K) and their relative contribution to each individual genotype. The value of K is estimated by multiple runs of the programme for different values of K and by investigating how the statistic ln(K|D) varies with K, where ln(K|D) provides an estimate of the likelihood of the data given the modelled K. From the analysis shown in Fig. 2(a), K values of 2, 3 and 4 were investigated further and the correlations of their Q groups are shown (Fig. 2(b)). Fig. 2(c) shows the Q plots, illustrating the contribution of each presumed progenitor to an individual genotype, for K = 2, K = 3 and K = 4. Although K = 2 was best supported according to the method of Evanno et al. (Reference Evanno, Regnaut and Goudet2005), the K = 3 plot shows best how the three selected cultivars represent distinct subgroups within the panel. The cultivars with a substantial contribution from sub-population K3,3 (shown in green) are predominantly marrowfat types with some large blues, and all marrowfat lines showed this contribution (Fig. 2(c)).

Fig. 2. Structure v 2.1 analysis, based on the genetic marker data obtained for a set of 48 pea cultivars using the default parameter set with the admixture model and comprising 10,000 Markov Chain Monte Carlo (MCMC) runs after a ‘burn-in’ of 10,000. K values in the range 1–10 were examined with five runs for each K value. (a) On the basis of the Evanno et al. (Reference Evanno, Regnaut and Goudet2005) analysis, the values of K = 2, 3 and 4 were investigated. (b) The correlations between Q groups within each K were calculated in Excel to establish correspondences; non-self correlations between corresponding Q groups ranged from r = 0.993 to 0.998 for K = 3 and r = 0.987 to 0.996 for K = 4. Within each K value, Q values for the three most correlated runs were averaged and the correlations between these are presented. From these correlations, the way the groups split as K increases has been deduced. (c) The averaged Q values for K = 2, 3 and 4 are plotted (top to bottom) and each of the varieties is identified. The three parents of the RIL populations are marked with arrows and additionally shown to the left of the Q plots.

This ‘Structure’ analysis (distribution on the Q plots, Fig. 2(c)), together with the marker PCO plot (Fig. 1(b)), was used to select three cultivars that were as distinct as possible on the basis of the genetic marker analysis, while being constrained by also showing contrasting phenotypes (online Supplementary Table S2). In this way, the derived RILs were expected to segregate for traits of interest and to be amenable to genetic analysis. One additional constraint was placed on the final selection of lines: that they should not differ phenotypically because of the allele at the af locus since this trait is likely to have major pleiotropic effects that would dominate the characterization of any resulting RIL population. The afila (af) gene affects leaf morphology (wild-type leafed versus so-called semi-leafless phenotypes) and is likely to be relevant to many agronomic traits, including overall field performance (Burstin et al., Reference Burstin, Marget, Huart, Moessner, Mangin, Duchene, Desprez, Munier-Jolain and Duc2007); the specific effects of this gene are best investigated in near-isogenic lines. On this basis, the cv. Minerva was ruled out as a parent, even though it is very distinct from most of the recommended list varieties which were analysed (Figs. 1, 2). The lines finally selected as parents (the cultivars Brutus, Enigma and Kahuna) corresponded to different market classes (large blue, white and marrowfat types, respectively) and all were af lines. The parental lines are shown to the left of the Q plots in Fig. 2(c) to highlight the relative contribution of their (conjectured) progenitors. The three parents capture 63% of the alleles identified in the cultivars. The frequency of the dominant alleles identified in the three selected lines is strongly correlated with their frequency in cultivars as a whole (r ² ~0.8).

In summary, the cvs. Brutus, Enigma and Kahuna were selected as semi-leafless varieties of contrasting market classes for the generation of mapping populations. The parents represented the phenotypic (online Supplementary Table S2) and genotypic (Figs. 1, 2) variation available within the elite pea gene pool. The constraints placed on their selection meant that the parental lines were asymmetrically placed on the phenotypic analysis (online Supplementary Table S2). The consistency of genotype data among the seed lots available for the chosen parental lines was checked, using one SSAP primer combination (online Supplementary Fig. S2).

Establishment of crosses and development of genetic maps

Reciprocal crosses were carried out between pairs of the chosen parents, yielding three populations of RILs, and 220 F₂ seeds (110 for each reciprocal cross) were sown for every cross (Brutus × Enigma, Brutus × Kahuna, Enigma × Kahuna as BE/EB, BK/KB and EK/KE RILs, respectively). The three parental lines had contrasting seed traits (yellow or green cotyledon colour; large- or medium-sized seeds). The F₁ seeds and/or plants were checked to prove that they were true hybrids. Hybrid status was confirmed by phenotype (cotyledon colour when Kahuna or Brutus had been the maternal parent, where green cotyledon colour (i) is recessive to yellow (I) in Enigma, and by genotype, using SSAP marker analysis where markers from both parents were apparent in heterozygous plants; see online Supplementary Figure S2 for parental polymorphisms scored). Online Supplementary Figure S3 (A, B) shows examples of the phenotypes scored for parental and F₁ hybrid seeds, where the cv. Kahuna (a marrowfat) was a parent. The phenotypes of the F₁ seeds obtained for these two crosses indicated that the marrowfat trait may be maternally determined (online Supplementary Fig. S3). The combined results confirmed that the crosses had been successful and allowed the efficient generation of the F₂ populations.

The genetic map data obtained for the three sets of RILs at F₆ are shown in online Supplementary Figs. S4–S6. Alignment of SSAP marker data across diverse populations, including wide crosses, facilitated the map development. Due to the much greater genetic similarity between the cultivar parents, there were as expected far fewer markers available for most linkage groups (LG) in the cultivar-derived RILs than in those derived from wide crosses. It was notable that, in some cases, there was a severe paucity of genetic marker data, potentially indicative of a common origin of chromosomal segments within the relevant parents. This is particularly true for the BE/EB RILs, where LG IV and VII have two markers each (online Supplementary Fig. S4). In such cases, these common LG regions could be largely discounted as having an association with the control of quantitative traits evident in the derived RILs. In contrast, where the cv. Kahuna is a parent, a much greater number of polymorphic markers was evident for LG VII, in particular (online Supplementary Figs. S5, S6). This possibly indicates much greater distinctness of this LG in the marrowfat class of pea, compared with the other combining varieties.

Trait and quantitative trait locus (QTL) analysis in RILs

At F₈ (F₆ bulks), single plot data were generated for the RILs (year 1, Y1) but, thereafter, triplicate plots were sown for every RIL (Y2-4). Throughout the trials conducted on the three sets of RILs, the principal traits scored were overall yield, standing ability and thousand seed weight. Although susceptibility to downy mildew was additionally considered as a relevant trait to score, this disease was only in evidence to any great extent in year 3 at the PGRO site, where it was associated with generally poor performance due to waterlogging in very wet weather. Equally, standing ability or lodging, a trait that is often scored by its components (creep, followed by erect growth, as opposed to canopy collapse), was not always in evidence. The datasets collected were analysed genetically in two ways: as means of the raw data values and as adjusted data, according to accepted practices for national and recommended list trials, when part plots had been damaged, lost or otherwise affected by non-standard problems, such as invasive weeds.

Fig. 3(a) shows an example of the range of variation for thousand seed weight, as measured in one season (Y4) for EK/KE RILs and parental lines. The low standard error of the mean (SEM, Fig. 3(a)) was typical of measurements for this trait across all populations. Although the range of trait values varied according to the season for all populations (not shown), the parental values fell consistently at either end of the seed weight spectrum, indicating a multi-gene control and little transgressive segregation (Fig. 3(a)). QTL (quantitative trait locus/loci) analysis of thousand seed weight data revealed a consistent pattern of genetic marker association across years (Table 1, Fig. 3(b)). Two genetic loci were associated with thousand seed weight on LG I: one of these (top of LG I) was apparent when the cv. Kahuna was involved in the cross (BK/KB and EK/KE RILs) and the second (bottom of LG I) was consistent among years for the BE/EB RILs (Fig. 3(b)). The cvs. Kahuna and Enigma contribute positively to the trait at the QTL on the top and bottom of LG I, respectively. Two additional genetic regions were associated with variation in thousand seed weight when Kahuna was a parent, with one of these also apparent in the BE/EB population (Table 1). The QTL on LG IV fell just over the LOD threshold in the BK/KB population (not shown), whereas it was very significant for the EK/KE population (Table 1). A QTL on LG V was evident for two populations, EK/KE and BE/EB (Table 1). Overall, the cv. Kahuna contributed positively to the seed weight trait over three distinct genetic loci, explaining up to 96% of the variation in seed weight (Table 1).

Fig. 3. (a) An example of the range of variation for thousand seed weight (TSW), as measured for the EK/KE population (year 4, triplicate plots of 166 lines, PGRO site). The positions of the parental lines are indicated by arrows (blue, cv. Enigma; purple, cv. Kahuna) with values of 215.0 ± 5.8 and 323.3 ± 5.1 (mean ± SE), respectively. (b) Major QTL on linkage group (LG) I for TSW in three RIL populations across 3 years (red, Y1; green, Y2; brown, Y4). Peaks are shown above the LOD threshold (dotted vertical lines, top scale; LOD 2.6-2.7 for BK/KB, 2.4 for BE/EB, 2.5–2.8 for EK/KE) at the bottom of LG I for BE/EB and at the top of LG I for EK/KE and BK/KB populations. The additive genetic effect is shown (blue line, bottom scale), with the parent contributing positively to the trait indicated in every case. The linkage groups are aligned, using genetic markers in common within the populations analysed (blue highlight) and with additional wide crosses.

Table 1. Summary of QTL identified in three populations (BE/EB, EK/KE, BK/KB) for thousand seed weight (TSW) and yield traits over 4 years

The maximum peak LOD scores, LOD threshold, % variation explained by the locus, the parental line contributing positively to the trait, linkage group (LG) and close genetic markers are listed for the trait QTL.

Fig. 4(a) shows an example of the range of variation observed for overall yield in one population. Both parents (cvs. Enigma and Kahuna) showed values at the upper end of the yield spectrum, as would be expected for two commercially cultivated lines, but with an appreciable number of lines showing higher or particularly lower yields than either parent, indicative of transgressive segregation. Although the yield data typically showed much higher SEM values than for other traits (see Fig. 3(a) for example), the maximum yield potential for the RILs was shown to be in excess of 5 t/ha, dependent on the RIL and the season. For overall yield, association with genetic marker data showed more variability, as expected for a complex trait. Nonetheless, some consistency of QTL associations was observed (Table 1, Fig. 4(b)). Two QTL were evident on LG I, with the parent cv. Enigma contributing positively to yield at each locus. One of the loci was on the upper end of LG I (Fig. 4(b)) in a region also associated with thousand seed weight in populations involving cv. Kahuna (Fig. 3(b)). The significance of this yield QTL was enhanced by analysis of data adjusted for non-standard plot effects (Fig. 4(b)). The QTL for yield that was detected towards the lower end of LG I for the BE/EB RILs (Fig. 4(b)) was not coincident with that influencing thousand seed weight in the same cross (Fig. 3(b)). The QTL detected for overall yield on LG III identified a similar region of the LG in all three populations (Fig. 4(b), Table 1). A QTL for yield on LG V was evident in the EK/KE population in 1 year only (Table 1).

Fig. 4. (a) An example of the range of variation in yield, as determined for the EK/KE population (year 2; triplicate plots of 166 lines, for which triplicate plot data were returned for 159; PGRO site). The positions of the parental lines are indicated by arrows (blue, cv. Enigma; purple, cv. Kahuna) with values of 3.923 ± 0.52 and 3.627 ± 0.37 t/ha (mean ± SE), respectively. (b) QTL on linkage groups I and III (LG I, LG III) for yield in two RIL populations across one (EK/KE; red, Y4 raw; black dashes, Y4 adjusted data) or two (BE/EB LG I; red, Y2; green, Y3; BE/EB LG III; green, Y3; brown, Y4 raw; purple, Y4 adjusted) years. Adjusted data are corrected for areas of plots affected by non-standard influences. Peaks are shown above the LOD threshold (dotted vertical lines, top scale; LOD 2.4 for BE/EB (LG I), 2.5 for EK/KE (LG I), 2.4–2.5 for BE/EB LG III). The parent cv. Enigma contributed positively to the trait for each QTL. The linkage groups can be aligned, using genetic markers in common within the populations analysed (blue highlight) and with additional wide crosses.

A further experiment aimed to establish the components of yield that contributed to the major QTL identified in the three populations. A subset of lines, selected on the basis of relative consistency of yield, was subjected to trials alongside the high-yielding cultivar, Prophet, using commercially-relevant plot size and sowing density. A very strong correlation (R ² = 0.92) between overall yield and standing ability was apparent in one such trial, where some lines (including cv. Prophet) yielded in excess of 5 t/ha (online Supplementary Fig. S7).

Candidate genes for traits

The genetic location of some of the QTL for thousand seed weight data in this work prompted an investigation into candidate genes within the genetic regions identified. This included two candidates, AgpS2 (Fig. 3b) and subtilisin, the latter of which mapped close to af (leaf phenotype) on the lower end of LG I in additional crosses (cv. Princess × JI 185, not shown) and to the syntenic region of chromosome 5 in Medicago truncatula (D'Erfurth et al., Reference D'Erfurth, Le Signor, Aubert, Sanchez, Vernoud, Darchy, Lherminier, Bourion, Bouteiller, Bendahmane, Buitink, Prosperi, Thompson, Burstin and Gallardo2012).

The predicted amino acid sequences for the small subunit 2 of ADP-glucose pyrophosphorylase gene (AgpS2) in 10 pea lines, including the three parental lines from this study, revealed one amino acid difference in cv. Kahuna, compared with other lines (online Supplementary Fig. S8). Although this substitution might be significant (K454I), it is not present in a second marrowfat line, cv. Princess, included in the analysis.

Although genetic variation in subtilisin has been associated with significant differences in mean seed weight in two legume species (D'Erfurth et al., Reference D'Erfurth, Le Signor, Aubert, Sanchez, Vernoud, Darchy, Lherminier, Bourion, Bouteiller, Bendahmane, Buitink, Prosperi, Thompson, Burstin and Gallardo2012), the gene sequences determined for the entire coding region of subtilisin in cvs. Brutus, Enigma and Kahuna (2290 bp) showed no nucleotide polymorphisms. Some few polymorphisms were apparent in comparisons with two additional lines (JI 281, JI 185; not shown).

The association between yield and genetic markers on LG III identified the region containing the ‘La Della’ gene marker as being of interest. The marker is based on the gene encoding the pea putative gibberellin (GA) signalling DELLA protein LA (GenBank: DQ848351.1; Weston et al., Reference Weston, Elliott, Lester, Rameau, Reid, Murfet and Ross2008). This genetic region in the middle of LG III also contains the marker A001 associated with lodging resistance (a component of standing ability) (Tar'an et al., Reference Tar'an, Warkentin, Somers, Miranda, Vandenberg, Blade, Woods, Bing, Xue, DeKoeyer and Penner2003; Reference Tar'an, Warkentin, Somers, Miranda, Vandenberg, Blade and Bing2004). The linkage between the A001 and La Della markers was checked in a wide mapping population (JI 15 × JI 399) and three recombinants were identified out of 85 lines scored.