4 DISCUSSION
The present study of shifts from heterostyly to homostyly provides novel insights into the molecular basis of intraspecific mating system transitions and their evolutionary consequences. Previous studies suggesting independent origins of homostyly used molecular markers not associated with the genetic control of heterostyly (Ness et al., 2010; Shao et al., 2019; Yuan et al., 2017; Zhou et al., 2017). By leveraging new knowledge on the molecular basis of heterostyly and extensive sampling of populations with different frequencies of homostyles, this is the first study to demonstrate that heterostyly has been repeatedly lost likely
via multiple, presumably loss-of-function mutations of the S-locus gene controlling stigma position (
CYPᵀ). Additionally, we show that the transition to homostyly promotes a change of mating system from outcrossing to selfing, with ensuing population genetic consequences. Furthermore, we discuss the potential roles of gene flow in the spreading of homostyly. Finally, we propose that inbreeding depression might explain why homostyly, despite its apparent advantages, has not become fixed in
P. vulgaris. Thus, the present contribution provides a benchmark for similar studies on the loss of heterostyly across angiosperms.
4.1 Homostyles replace short-styled individuals within populations
The variation in the frequency of floral morphs among populations provides an opportunity to elucidate how homostylous phenotypes originate and spread. According to previous evolutionary models, the origin of homostyles from S-individuals should cause an initial reduction of the latter, triggering an increase in the number of homostyles (Crosby, 1949; 1959). When homostyles first arise, their pollen can be used for both self- and cross-fertilization of L-individuals, but not for fertilization of S-individuals, due to lack of spatial reciprocity between high anthers of homostyles and low stigma of S-individuals and incompatibility between homostyle pollen and S-morph stigmas (Crosby, 1949; Figure 1). Furthermore, homostyles can be fertilized by compatible pollen of S-morphs, but this cross is thought to occur rarely in nature due to stigma clogging with self-pollen in homostyles (Crosby, 1959; but see Bodmer, 1959). Hence, homostyles have a reproductive advantage over L- and S-morphs stemming from their ability to fertilize both themselves and the L-morphs (Richards, 2003). Finally, self-fertilization of homostyles (Figures S3A and B) and occasional crosses from homostyles to L-individuals produce homostylous and L-progeny, but no S-progeny (Figures S3C and D).
Congruently with Crosby’s predictions (1949) and findings from previous studies (Curtis & Curtis, 1985), our results demonstrate that, as homostyles increase, more S- than L-individuals are lost within populations (Table 1 and Figures 2A and B). After 72 and 36 years since Crosby’s (1949) and Curtis and Curtis (1985) observations, respectively, we also found that homostyly has not become fixed in Somerset populations. Furthermore, the fact that L-individuals are maintained at low frequencies, even in the absence of S-individuals, suggests that they are occasionally fertilized by homostyles. Experiments aimed at quantifying pollen delivery and receipt among the three floral morphs would help to clarify the selective regimes favoring the increase of homostyles. Overall, our results are consistent with the hypothesis that long-homostyles originate from S-individuals and explain why the latter occur at lower frequencies than L-individuals across populations with all three morphs.
4.2 Multiple losses of heterostyly within species are associated with different molecular changes in the CYPᵀ gene
Recent advancements in genomics have enabled the identification of loci responsible for repeated losses of phenotypic traits (Albalat & Cañestro, 2016; Sharma et al., 2018). For instance, multiple loss-of-function mutations were detected in
SCR or
SCR-like genes corresponding to independent losses of self-incompatibility in
Arabidopsis thaliana and
Laevenworthia alabamica (Chantha et al., 2013; Shimizu et al., 2008). In the case of homostyly, recent experimental studies established that
CYPᵀ disruption causes a shift to homostyly (Huu et al., 2016; Kappel et al., 2017). Accordingly, all S-individuals of
P. vulgaris analyzed in the present study share the same functional
CYPᵀ allele (
CYPᵀ-1), whereas private alleles with nonsynonymous mutations or indels (
CYPᵀ-2 to -
9) are found exclusively in homostyles (Figures 3A, 3B and 4). These mutations cause either changes in the polarity of the amino acid side chain (
CYPᵀ-3 and
-4) or premature stop codons (
CYPᵀ-2,
-5 and
-6), both types of mutation presumably inducing loss or reduction of function (Zhang, 2000). In one case (
CYPᵀ-7), a nonsynonymous mutation did not change amino acid side-chain polarity; thus, this mutation may or may not affect protein function (Figure 3A). Notably, the alleles found in our study differ from the ones previously reported in two long-homostyles of
P. vulgaris (
CYPᵀ-8 and -
9, originally named
CYPᵀ SLH1 and
SLH2, respectively; Li et al., 2016). Thus, a total of eight different putative loss-of-function mutations in
CYPᵀ have been identified so far in homostylous individuals of
P. vulgaris. Future work should focus on elucidating how the found nonsynonymous mutations affect
CYPᵀ function
. Finally, the groupings of the haplotype network were congruent with those of the phylogeny (Figures 3B and 4), as expected given the lack of recombination in the S-locus, hence
CYPᵀ. Overall, our results show that a diversity of changes in the same gene is associated with multiple, independent origins of homostyly within species.
Loss of phenotypic traits can be caused by mutations not only in coding regions of specific genes but also in their promoter regions or by structural rearrangements. For instance, some individuals of self-compatible
A. thaliana were characterized by mutations in
cis-regulatory regions, inversions, or splicing variants, causing loss of function in
SCR or
SRK genes controlling self-incompatibility (Dwyer et al., 2013; Shimizu et al., 2008; Shimizu & Tsuchimatsu, 2015). In
P. vulgaris, a surprising result is that six out of 27 homostyles have the same
CYPᵀ allele as the S-individuals (
CYPᵀ-1; Figures 3A, 3B and 4). These six homostyles occur in two geographically close populations (M01 and T08; Figures 2A). The occurrence of a
CYPᵀ allele with no mutations in homostyles enables us to suggest for the first time that additional mechanisms might be responsible for the inactivation of
CYPᵀ, including mutations in its promoter region or structural rearrangements that cannot be detected
via exon sequencing.
In general, this result underscores that mutations outside genes controlling specific traits might represent essential sources for repeated trait losses.
The interpretation of CYPᵀ sequencing results presented above is compatible with multiple origins of homostyly in P. vulgaris, but alternative scenarios involving a single origin should be considered. For example, as discussed above, the initial homostylous phenotype could have arisen via mutations in the promoter region, structural rearrangements, or changes in the expression of CYPᵀ but not in the CYPᵀ coding region per se, as in the six homostyles carrying CYPᵀ-1, the haplotype of S-morphs. Such homostyle could have originated once in a large ancestral population that later became fragmented and/or migrated to neighboring populations, gaining nonsynonymous mutations and indels in CYPᵀ after the gene had already been rendered functionless. However, in the scenario of a single origin of homostyly triggered by a mutation outside CYPT, we would expect homostyles carrying CYPᵀ-1 to co-occur in the same population with other homostyles carrying mutated CYPᵀ alleles (i.e., CYPᵀ-2 to -9), but our findings do not match this prediction (Figure 4). Thus, the most parsimonious interpretation of the haplotype network and phylogeny inferred from CYPᵀ sequences (Figures 3 and 4) is congruent with multiple origins of homostyly via independent mutations at CYPᵀ. Furthermore, CYPᵀ-9 (SLH2 from Li et al., 2016) was found exclusively in a homostyle from Chiltern Hills located approximately 200 km from Somerset. Given the limited dispersal abilities reported for P. vulgaris (Cahalan & Gliddon, 1985), this finding further supports independent origins of homostyly. Future transcriptomic or RT-PCR analyses of homostyles with different CYPᵀ alleles aimed at testing the expression of both mutated and unmutated CYPᵀ should further elucidate the question of single vs. multiple origins of homostyly in P. vulgaris. Finally, our microsatellite data do not enable us to determine whether mutations in CYPᵀ accumulated before or after the fragmentation of a hypothetical ancestral population. Whole genome sequences from multiple heterostyles and homostyles aimed at improving phylogenetic resolution and demographic analyses are necessary to elucidate the potential role of fragmentation in the evolution of P. vulgaris homostyles.
The multiple CYPᵀ mutations found in the homostyles of a small geographic region in Somerset could be viewed as a puzzling result. However, loss of gene function is usually followed by degeneration through accumulation of mutations, or even gene loss (Sharma et al., 2018). Additionally, lack of recombination and hemizygosity of the S-locus should both decrease the efficacy of purifying selection on its genes, including CYPᵀ, possibly contributing to accumulation of mutations (Becher, Jackson, & Charlesworth, 2020; Gossmann, Woolfit, & Eyre-Walker, 2011). Furthermore, a higher mutation rate was reported for CYPᵀ than GLOᵀ (Huu et al., 2020). Thus, the multiple CYPᵀ mutations of homostyles detected here are compatible with both theoretical expectations and previous empirical evidence. Future sequences of other S-locus genes, their paralogs, and additional genes outside the S-locus from both P. vulgaris and other primroses would provide a broader context to compare selection on CYPᵀ vs. other genes in both homostyles and heterostyles.
4.3 Consequences of shifts to homostyly on mating system and population genetics
Changes in floral morphology can have profound effects on the mating system (Opedal, 2018). Specifically, the reduction of anther and stigma separation and loss of self-incompatibility of homostylous morphs should increase selfing compared to heterostyles. Evidence of this mating system transition in homostyles was reported for
Eichhornia (Husband & Barrett, 1993),
Turnera (Belaoussoff & Shore, 1995),
Amsinckia (Schoen et al., 1997), and some
Primula species (Yuan et al., 2017; Zhou et al., 2017; Zhong et al., 2019). In
P. vulgaris, previous estimates of outcrossing rates in homostyles
ranged broadly from 0.05 to 0.80
(Bodmer, 1958, 1984; Crosby, 1958, Piper et al., 1986).
Our results from progeny arrays indicate that outcrossing rates in homostyles are significantly lower than in heterostyles (tm = 0.14 vs. 1.0, respectively), as expected. Such outcrossing rates were estimated based exclusively on germinated seeds. However, ungerminated seeds were likely those with higher homozygosity, hence higher inbreeding depression, thus their inclusion would have further decreased our estimates of outcrossing rates. Also as expected, the frequency of homostyles was associated with an increase in population-level selfing rates (Figure 5A). Nevertheless, the fully homostylous population (M01) had lower selfing rates than expected, possibly caused by increased error rates due to its small sample size (Redelings et al., 2015). The estimated outcrossing rates of homostyles firmly place them in the category of selfers (i.e., tm ≤ 0.2; sensu Goodwillie et al., 2005; Schemske & Lande, 1985). Overall, our findings clarify previous conflicting results about the effects of homostyly on the mating system of P. vulgaris, confirming that transitions to homostyly increase selfing.
Mating systems shape how genetic diversity is partitioned within and among populations (Barrett, 2010; Wright et al., 2013). In concordance with theoretical expectations, we found that intrapopulation homostyle frequency significantly increased inbreeding coefficients (Figure 5B). These results conform with the reduction of genetic diversity associated with transitions from heterostyly to homostyly documented in other systems (Husband & Barrett, 1993; Ness et al., 2010; Yuan et al., 2017; Zhong et al., 2019; Zhou et al., 2017). Moreover, our population genetic results indicate that increased frequency of homostyles in populations increases homozygosity, thus lowering intrapopulation genetic diversity and elevating inter-population genetic differentiation. Altogether, our findings confirm the population genetic consequences predicted for the transition to selfing.
4.4 Does homostyly spread among populations and why has it not become fixed in Primula vulgaris?
Gene flow plays a central role in spreading advantageous mutations (Morjan & Rieseberg, 2004; Ralph & Coop, 2010) and could favor the migration of homostylous alleles among populations. In
P. vulgaris, it has been proposed that homostyles could have migrated from initial places of origin to neighboring populations through pollen or seed, the former being more likely (Crosby, 1949; 1960). Previous studies estimated that dispersal is restricted to a maximum of a few hundred meters from the parental plant in
P. vulgaris (Cahalan & Gliddon, 1985). However, occasional pollen flow over 1–3 kilometers has been reported (Van Geert, Van Rossum, & Triest, 2010). Indeed, our population genetic results support a model of moderate gene flow among populations (
Nm = 1.96; Figure S2) consistent with moderate genetic differentiation among populations (
Fst = 0.083; Figures 6A and B), the latter being lower than previous estimates for
P. vulgaris (
Fst = 0.086–0.508; van Geert et al., 2015). Furthermore, we did not detect isolation by distance (Mantel test,
P = 0.457), further favoring the hypothesis of gene flow. Moreover, the geographic distribution of
CYPᵀ alleles in our study indicates that 12 homostylous individuals from five populations (D*11, T03, T04, T07 and T10) separated by 3-18 Km shared the same
CYPᵀ allele (
CYPᵀ-2; Figure 4), suggesting that some homostyles could have migrated between neighboring populations. To summarize, Crosby (1949; 1959) proposed a single origin of homostyles followed by their spreading
via migration. While supporting multiple origins of homostyly, our results imply that it can also spread
via gene flow, providing partial corroboration of Crosby’s hypothesis.
A crucial question remains as to why homostyly has not become fixed in
P. vulgaris despite automatic selfing advantage (Fisher, 1941; 1949) and reproductive assurance (Piper et al., 1986). In fact, the opposite has been found, since previous studies recorded the loss or decrease of homostyles in revisited populations (Boyd et al., 1990; Curtis & Curtis, 1985). Crosby (1949) proposed that selfed progeny from homostyles homozygous for the S-locus should have reduced fitness, likely caused by potentially deleterious effects of dominant homozygosity in S-locus genes. Such negative effects were suggested by crossing experiments with S-morphs of
Primula sinensis (Mather & de Winton, 1941) and
P. oreodoxa (Yuan et al., 2018). However, whether homostyles with one or two copies of the S-locus have differences in fitness has not been experimentally tested.
An additional explanation for the failure of homostyly to completely replace heterostyly is that the negative effects of inbreeding depression (Goodwillie et al., 2005; Richards 1984) can outweigh the advantages of homostyly (Boyd et al., 1990; Piper et al., 1984). Theoretical models suggest that inbreeding depression values above 0.5 should prevent the invasion of selfing individuals in an outcrossing population (Lande & Schemske, 1985). Two recent studies investigated the relationship between inbreeding depression and the spread of self-compatibility, finding contrasting results. In experimental populations of
Linaria cavanillesi, self-compatible individuals with no inbreeding depression displaced self-incompatible individuals in just three generations (Voillemot & Pannell, 2017; Voillemot, Encinas-Viso, & Pannell, 2019). In contrast, inbreeding depression values of 0.54 prevented the spread of self-compatible plants in experimental patches of self-incompatible plants in
Laevenworthia alabamica (Layman, Fernando, Herlihy, & Busch,
2017). Our estimate of inbreeding depression at seed germination stage (0.58) in
P. vulgaris is closer to the latter study and above the threshold value of 0.5 predicted to prevent the invasion of selfing (Lande & Schemske, 1985). The effect of inbreeding depression in
P. vulgaris might be even higher than in our current estimates, for it could also act at later stages of the life cycle. Altogether, our results imply that inbreeding depression in homostyles is sufficiently high to prevent the fixation of homostyly within populations.
While theory proposes that increased selfing should purge inbreeding depression over time by eliminating recessive deleterious alleles from populations (Schemske & Lande, 1985), transitions to homostyly in
P. vulgaris may be too recent for the purging process to have occurred. Moreover, recessive deleterious alleles could have been re-introduced into populations
via gene flow, thus slowing the decrease of inbreeding depression. Our study estimated moderate levels of gene flow among populations of
P. vulgaris (
Nm = 1.96; Supplementary Figure S2), implying that re-introduction of genetic load can occur. In conclusion, our findings suggest that inbreeding depression contributes to maintaining heterostylous morphs within populations despite the reproductive advantages of homostyly.