Ovary and ovule morphometrics
Regressions between ovary and ovule lengths at meiosis (dyad to early tetrad) and at the 1-nucleate ES (ES1) and early 8-nucleate ES (ES8) stages across 25 accessions were highly significant. However, the regression equations explained <50% of the variability (r2) at each stage (Additional file 1). Hence, large and small ovaries contained either large or small ovules, depending on accession, and ovary length only poorly predicted germline stage across accessions. For example, ovaries 0.3 cm long contained ovules in the meiocyte stage to the maturing ES stage depending on accession (Additional file 2).
Mean (±SE) ovule curvatures and areas (Figure 1A) were determined at two developmental stages, meiocyte and ES1, for 115 diploid genotypes and one naturally occurring tetraploid (Additional file 3). ANOVA was used to determine which of these two ovule development variables (curvature or area) would most closely correlate with germline stage (meiocyte or ES1). The dependent variable, coefficient of variation (CV), was represented by the CV values of 460 means, 115 for each of the four (2 × 2) method-by-stage combinations (diploid genotypes only). At the meiocyte and ES1 stages, mean CV values (±SE) based on ovule curvature were 0.151 (±0.004) and 0.134 (±0.004), respectively. The corresponding CV values based on ovule area were significantly larger, 0.210 (±0.006) and 0.185 (±0.005), respectively. The main effects (method and stage) were significant (P < 0.001), but the interaction effect was not significant. This analysis indicated that ovule curvature was less variable than ovule area at each germline stage.
Two sets of ANOVA were conducted to determine if variation in mean ovule curvature, ovule area, and three ovule area components (per genotype) varied according to taxonomic group. In the first set, all 116 genotypes from 57 accessions (Additional file 3) were partitioned into seven taxonomic groups, which consisted of the five subsp. bicolor races, accessions of subsp. verticilliflorum, and a group (other) that contained breeding lines and hybrids (Figure 2). Again, ovule curvature was more effective than ovule area in differentiating taxonomic groups, especially at the meiocyte stage. However, distinct partitioning also occurred among taxonomic groups based on the percentage of ovule area represented by the nucellus and integuments (Figure 2). These data further indicate that ovule shape (ovule curvature and relative growth dynamics of the nucellus and integuments) is more tightly correlated with germline development than is ovule area.
At the meiocyte stage, ovule curvature was most advanced for genotypes of the verticilliflorum group (Figure 2). In addition to strong curvature, the verticilliflorum group also had the largest and smallest percentages of ovule area represented by integuments and nucellus, respectively. As ovules mature, the integuments grow rapidly around the ovule, and consequently a larger proportion of the ovule is composed of integument. These data indicate that onset of meiosis was delayed in the verticilliflorum group compared to other groups (Figure 2). The opposite was observed for the kafirs. Here, ovules were only slightly curved at the onset of meiosis, and the integuments and nucellus represented the smallest and largest percentages of ovule area, respectively (Figure 2). Hence, in the kafirs, germline development is accelerated compared to other taxonomic groups. Variation within taxonomic group was also observed as indicated by highly significant (P < 0.001) effects for genotypes nested within taxonomic group and for genotypes nested within accessions (Additional file 4). The only insignificant effect was the taxonomic group by germline stage interaction for the percentage of ovule area represented by the germ (Figure 2, Additional file 4).
Apospory in accessions and mapping populations
Nucellar cells normally die adjacent to the expanding embryo sac. In the present study, this progressive process of programmed nucellar cell death began shortly after megasporogenesis and continued until after fertilization when the nucellus was essentially consumed. In ovules of highly aposporous angiosperms, one or more nucellar cell(s) is re-programmed to undergo embryo sac formation. Early indications of this reprogramming include an abnormal doubling in size of the nucellar cell and nuclear enlargement [1, 21]. In the present study, cells assuming these traits were counted as i) aposporous initials (AI) when they occurred in the micropylar region of the nucellus (usually adjacent to the MMC, meiocyte, or degenerating megaspores (DM)), or ii) large stack cells (LSC) when they occurred in the chalaza proximal to the MMC, meiocyte, or functional megaspore (FM) (Figure 1B, D). LSC developed from cells at the nucellus chalaza interface and belonged to or were closely associated with the cell file (stack) from which the MMC formed. Generally, LSC were much more prevalent than AI (Additional file 3).
We defined the FM stage as onset of FM enlargement, which coincided with DM degeneration (Figure 1B). We defined the 1-nucleate ES stage as acquisition by the FM of a vacuole similar in size to the nucleus. Likewise an AI was referred to as an AES once it had produced a similarly large vacuole. AES only rarely formed from LSC (based on observed locations of AES). Most were derived from AI and formed in the micropylar region. Sexual ES and AES were further characterized by number of nuclei present (Figure 1C, E).
Some AI, LSC and AES did not form until the FM stage. Hence, to minimize underestimating apospory, only ovules ranging in development from the FM stage through the ES2 stage were used in determining AI, AES and LSC frequencies. The ES2 stage criterion was used because determining the origin of the ES (sexual or aposporous) in ovules beyond the ES2 stage was problematic. In these ovules, megaspores and nucellar cells adjacent to the enlarging ES had degenerated.
Frequencies of AI, LSC and AES were determined for 150 S. bicolor genotypes from 65 accessions (Additional file 3, 116 genotypes; Additional file 5, 34 genotypes), a mapping population consisting of 300 F2, and a mapping population consisting of 119 recombinant inbred lines (RIL [30]). Correlations between AES and AI and between AES and LSC were higher among genotypes of the accessions than among genotypes of the mapping populations (Figure 3). In all three populations, the frequency of AES formation was more highly correlated with the frequency of AI formation than with the frequency of LSC formation. Compared to the genetically diverse accessions, regression r2 values between LSC and AI were twice as high in the segregated F2 and RIL mapping populations (Figure 3). None of the regressions between percentage germline degeneration (measured for accessions only) and percentages of AI, AES or LSC (or combinations of these) was significant.
Eleven of the 150 diploid genotypes from 65 accessions exceeded 3% AES formation (Additional file 3). Five of these were from breeding lines of subsp. bicolor (5 of 30 lines) and five were from accessions of subsp. verticilliflorum (5 of 35 accessions). One, a caudatum, represented all other taxonomic groups (1 of 85 accessions). Two tests of equality of proportions were conducted. These matched the "other" group (1 of 85) against the breeding lines (5 of 30) and the "other" group against the verticilliflorum (5 of 35). Both tests were rejected (P < 0.001 and P < 0.01, respectively). Hence, apospory was most prevalent in wild land races of subsp. verticilliflorum and in breeding lines of subsp. bicolor.
Flow cytometry of leaf tissue was used to determine the ploidy of the 11 genotypes that exhibited ≥3% AES formation. Ten were diploid, but one, which exhibited the highest AES percentage (14% with 45% AI formation), was tetraploid (Figure 4). Three other genotypes of this accession (IS 12702, subsp. verticilliflorum) were diploid. These diploids had high AI levels relative to other accessions (Additional files 3, 5), but only one exhibited an AES frequency >3% (4.9%). Several other genotypes with >3% AES formation were from accessions in which multiple genotypes were analyzed but only one genotype exhibited the high AES level (Additional files 3, 5). Only two genotypes (from two different subsp. verticilliflorum accessions) exhibited >6% AES formation. Eight genotypes exhibited >6% AI formation, one caudatum, three from the breeding lines, and four from subsp. verticilliflorum.
Apospory and ovule morphometrics
An objective of the current study was to determine if tendencies for apospory in S. bicolor are associated with other morphometric ovule development variables. To accomplish this, k-means multivariate clustering was used to partition genotypes of accessions, F2, and RIL into 3-4 groups (per population) with similar frequencies of AI or AES. In all three populations, meiosis and sexual ES formation occurred precociously in the groups with the highest AES formation frequencies (Figure 5).
As noted above, 11 of 150 genotypes from 65 accessions exhibited an AES frequency >3%. Three of these grouped together to form the highest AES k-means cluster, and the remaining eight clustered together to form the second highest k-means group. Both groups underwent meiosis and sexual ES formation early (low ovule curvature values) compared to the other k-means groups (Figure 5A, see Additional file 6 for ANOVA results). Two of the three genotypes in the highest AES group were from a single breeding line and the third was a subsp. verticilliflorum genotype. In the second highest group (eight genotypes), three were from breeding lines, four were from subsp. verticilliflorum and one was a caudatum (subsp. bicolor). If earliness of meiosis and sexual ES formation promoted apospory, a higher frequency apospory should have been observed among the kafirs (Figure 2). However, the kafirs exhibited low AI and AES frequencies. In contrast, five of the 11 highest AES-forming genotypes belonged to subsp. verticilliflorum, which on average underwent meiosis later than most of the other taxonomic groups (Figure 2).
Ovule area values during meiosis were also significantly lower for the 11 highest frequency AES-forming genotypes (Figure 5A, more and most groups; Additional file 6). This was accompanied by significantly larger percentages of total ovule area represented by the meiocyte (Figure 5A, Germline). This indicates that in these relatively small non-curved ovules (of aposporously active genotypes), the sexual meiocyte was actively growing and dividing; and this occurred whether AES were present or not. In contrast, percentage values for ovule area represented by the nucellus and integuments for the two highest AES-forming groups were variable (Figure 5A). Note from Additional file 6 that variability among genotypes in clusters was significant. ANOVA were also performed for groups of genotypes defined by k-means clustering using AI frequencies, but significant differences in ovule curvature or area were not detected among these clusters.
Ovule curvature data for the meiocyte, ES1 and ES8 stages were collected for the 300 genotypes of the F2 mapping population (Figure 5B). As with the accessions, groups of F2 with the highest and the next to highest AES formation frequencies (nine and 25 genotypes, respectively) underwent meiosis earlier than the other groups. This precociousness persisted into the ES1 and ES8 stages only for genotypes from the highest AES formation group (Figure 5B, see Additional file 7 for ANOVA results). Mean ovule curvatures for k-means clusters based on AI frequencies did not differ significantly at any stage. Tests were conducted to determine if F2 plants with a low mean ovule curvature exhibited higher AES formation frequencies. For these tests, genotypes of the F2 population were clustered (k-means) by mean ovule curvature at the meiocyte, ES1 and ES8 stages, and ANOVA were performed to determine if differences existed among clusters in frequency of AES formation. The F-values for these analyses were not significant (Additional file 7).
Precociousness of meiosis and sexual ES formation in the highest AES and AI frequency clusters was more distinct among the well segregated F8 RIL (Figure 5C) than among the F2 (Figure 5B), and the degree of earliness in the two highest AES groups was similar to that observed among the genetically diverse accessions (Figure 5A). Genotypes with high AI frequencies generally had high AES frequencies (Figure 3). However, several exceptions were observed. Two of the eight RIL in the highest AI formation group were in the lowest AES formation group. Likewise one of six RIL in the high AES formation group was in the low AI formation group. Genotypes with several AI often did not exhibit AES formation, and some genotypes with relatively high AES formation apparently passed through the AI phase quickly as few AI were observed.
About 30% of the RIL clustered into the more and most AI and AES formation groups. In contrast, only about 10% of accessions and F2 clustered into the more and most groups. The high percentage of RIL in the high AES and AI formation groups affected the ovule curvature dynamics of the entire RIL population. This was detected by clustering RIL according to mean ovule curvature at the meiocyte and ES1 stages. Clusters of genotypes exhibiting the lowest ovule curvature values (developmentally precocious) exhibited significantly higher AI and AES frequencies (Figure 5C, see Additional file 8 for ANOVA results). As noted above, such analyses were not significant for the accessions or for the F2 population.