The genetic architecture and evolution of life-history divergence among perennials in the Mimulus guttatus species complex

Abstract

Ecological divergence is a fundamental source of phenotypic diversity between closely related species, yet the genetic architecture of most ecologically relevant traits is poorly understood. Differences in elevation can impose substantial divergent selection on both complex, correlated suites of traits (such as life-history), as well as novel adaptations. We use the Mimulus guttatus species complex to assess if the divergence in elevation is accompanied by trait divergence in a group of closely related perennials and determine the genetic architecture of this divergence. We find that divergence in elevation is associated with differences in life-history, as well as a unique trait, the production of rhizomes. The divergence between two perennials is largely explained by few mid-to-large effect quantitative trait loci (QTLs). However, the presence of QTLs with correlated, but opposing effects on multiple traits leads to some hybrids with transgressive trait combinations. Lastly, we find that the genetic architecture of the ability to produce rhizomes changes through development, wherein most hybrids produce rhizomes, but only later in development. Our results suggest that elevational differences may shape life-history divergence between perennials, but aspects of the genetic architecture of divergence may have implications for hybrid fitness in nature.

1. Introduction

Ecological divergence is a major source of phenotypic differentiation among closely related species [1]. Phenotypic differentiation can promote species coexistence [2] and contribute to reproductive isolation [3]. Yet, the ease at which ecological divergence evolves depends, in part, on the genetic architecture of adaptive traits. The number of alleles controlling these traits, the distribution of effect sizes and gene action (e.g. dominance or epistasis) can impact how easily evolution can proceed [4]. If species diverge in suites of correlated traits, the extent of pleiotropy or close linkage between loci controlling multiple traits may restrict multivariate trait evolution to certain phenotypic axes [5]. Alternatively, if traits are controlled by many loci, only some of which have correlated effects, then evolution may be more flexible [6,7].

Differences between closely related species in ecologically important traits may also change with ontogeny. This can occur simply because as development proceeds, individuals have increased capacity to acquire and devote resources to particular traits [8], and/or because the fitness benefit of said traits may change across ontogeny [9]. For example, investment in anti-herbivore defensive traits frequently changes across development in plants [9,10], which may reflect the differential fitness effects of herbivore damage across development [10]. However, few studies have quantified the extent of trait divergence between closely related species throughout ontogeny or tracked the genetic architecture of ecologically relevant traits across development (but see [11]).

Differences in elevation between closely related plant species are often associated with phenotypic divergence [12,13]. Plant populations differ predictably in phenotypes along elevational gradients; both within species [14–19] and between close relatives [20–22]. Differences in water availability, herbivory, climate, growing season, soil substrate or a combination of factors may shape phenotypic evolution along elevational gradients [15,16,18,23,24]. Adaptations to high elevation environments often involve both changes in complex suites of traits, such as life-history, including correlated changes in the timing of reproduction, investment in vegetative biomass and the timing of senescence [14–16,21,25], as well as unique phenotypes, like shorter stature [19,26], changes in leaf physiology [27] or production of below-ground morphological structures [28,29]. In particular, investment in below-ground biomass in high elevation perennials may be important for the survival of extreme weather events [22] and/or may increase fecundity in subsequent years [30].

The Mimulus guttatus species complex (section Simiolus, DC, Phrymaceae) comprises about a dozen closely related species that inhabit a broad altitudinal range across the Pacific Northwest and exhibit substantial variation in life-history [31]. Considerable effort has been made to characterize life-history in this group, particularly within or between annual species [15,23,32–36], and between annuals and perennials [5,24,37–46]. Little is known about life-history divergence among perennials in this group, although they account for a sizable fraction of species and inhabit ecologically unique habitat relative to annuals.

Here, we characterize the geographic and ecological distributions of five perennial taxa within the M. guttatus species complex, assess the extent of variation in life-history across development and determine if life-history variation within and among perennials is correlated with differences in elevation. We then map the genetic architecture of life-history divergence between two ecologically and phenotypically divergent perennial species. We assess the extent to which loci affect multiple traits and characterize the genetic architecture of a novel life-history trait for this complex throughout development. Our findings contribute to our understanding of ecological and trait divergence between closely related species and shed light on the genetic architecture of divergence.

2. Material and methods

(a). Study species

Although the taxonomy of the M. guttatus species complex is debated [47,48], at least five perennial taxa are recognized: M. tilingii (sensu lato), M. decorus, M. corallinus and two ecotypes of perennial M. guttatus; coastal perennials (e.g. M. grandis; [47]) and inland perennials (e.g. M. guttatus sensu stricto; [47]). Mimulus decorus and M. tilingii both comprise multiple genetic lineages [47,49,50], but as these lineages form monophyletic groups, we include each as a single taxon.

Perenniality in the M. guttatus species complex is defined by the production of stolons, horizontally growing, vegetative stems that develop from axillary meristems and root at each node. Although vegetative at first, stolons have the ability to become reproductive. In addition to stolons, some perennials produce rhizomes [51,52]. Rhizomes, like stolons, are horizontally growing, vegetative stems that originate above ground, but dive underground early in development. Once underground, they become highly branched, and white, and repress leaf growth (electronic supplementary material, figure S1 and figure S7). Rhizomes can re-emerge above ground, wherein, their morphology reverts to that of a typical stolon, including the ability to become reproductive. Rhizomes thus represent a unique axis of investment in vegetative biomass and may play an important role in adaptation to harsh environments, but little is known about their ecological or phylogenetic distribution or their genetic basis.

(b). Estimating elevational distributions

To categorize elevational niches, we curated occurrence records for each species (see electronic supplementary material, methods for details). We compiled 1,675 M. guttatus, 209 M. tilingii, 101 M. grandis, 57 M. decorus and 12 M. corallinus geographically unique sites that were collected between 1960 and 2019. These ranges largely agree with previously published ranges [51], although we include more northerly occurrences for M. decorus compared to [51]. We then used the R Dismo package to construct random cross-validated maximum entropy species distribution models for each taxa ([53]; electronic supplementary material, figures S2, and S3; see electronic supplementary material, methods for details). We randomly sampled 300–500 locales from the predicted suitable habitat for each species and extracted elevation data for each locale. We merged these predicted occurrences with actual occurrence records and performed an ANOVA with elevation as the response variable, and species, the data source (e.g. predicted or herbarium occurrence), and their interaction as fixed effects (electronic supplementary material, table S5 and figure S4). We then used a pairwise T-test with Holm correction for multiple testing to determine what species differed in the elevational niche. All statistical analyses were performed in R [54].

(c). Quantifying life-history variation

In order to assess trait divergence among perennials of the M. guttatus species complex, we performed two common garden experiments: a population survey and a QTL mapping experiment. For both, seeds were cold stratified on moist Fafard 4P potting soil for one week at 4°C to break dormancy, then transferred to long-day conditions in the Duke University greenhouses (18 h days, 21°C days/18°C nights). On the day of germination, seedlings were transplanted to individual 4 inch pots and monitored daily.

(i). Population survey

To quantify phenotypic differences among species, we performed a common garden survey of 199 plants from 40 populations of five perennial taxa (average of eight populations per species and two maternal families per population; electronic supplementary material, table S1). For each maternal family, we grew three replicate individuals (one for each time point surveyed).

We surveyed plants at three developmental time points; three-week post-stratification (‘vegetative’ period), the day of the first flower (‘early reproductive' period) and six-week post-stratification (‘late reproductive' period). At each time point, we unpotted one replicate per maternal family, cleaned belowground biomass and measured the number of stolons and rhizomes produced. We measured both stolons and rhizomes (if present) for several stem traits (i.e. length, width, leaf length, branches per node and highest node of emergence). On the day of first flower, we scored the date and node of first flower, corolla length and width, and the length and width of the first true leaf. As phenotyping was destructive, we discarded individuals after. In the few cases in which not all maternal families produced three replicates, we preferentially measured replicates at later developmental times.

(ii). Analysis of population survey

To assess differences between species in life-history, we used PCA to summarize life-history variation across multi-trait space, restricting the PCA to include the total number of stolons, the highest node of stolon emergence, the number and proportion of stems that were rhizomes, and the length, width and leaf length of above-ground stolons. PC1 explained 49.82% of the variation and was the only significant PC according to a broken stick model. We performed linear mixed models using the lme4 and car packages in R [50,51] with PC1 as the dependent variable, species, time point and their interaction as fixed effects and population as a random effect. We found a significant species by time point interaction (electronic supplementary material, table S6) and therefore performed linear mixed models for each time point separately with species as a fixed effect and population as a random effect (see electronic supplementary material, table S6 for full model details). The significance of fixed effects was assessed using Type III Wald's χ²-tests using the Anova function in the car package in R [55,56].

To determine whether variation in life-history among populations was related to a population's source environment, we performed additional linear mixed models with the morphological PC1 as the dependent variable, elevation, species, time point and all possible interactions as fixed effects and population as a random effect. Again, we find a significant species by time point interaction (electronic supplementary material, table S7). We thus perform submodels for each time point, with PC1 as the dependent variable, species and elevation as fixed effects, and population as a random effect.

Lastly, we evaluated the extent to which life-history traits covary within perennial taxa. Strong associations between the timing of reproduction and investment in vegetative biomass have been previously described in Mimulus [5,39,40]. To determine if perennials exhibited an association between flowering time and stolon number, we performed a linear mixed model with the node of first flower as the dependent variable, and the number of stolons and species as fixed effects, and population as a random effect (electronic supplementary material, table S8).

(d). Determining the genetic architecture of life-history divergence

(i). F2 grow out

We grew 429 F2s derived from a single self-fertilized F1 between M. decorus (IMP) and coastal perennial M. guttatus (OPB). These species are two of the most phenotypically divergent perennials in the complex (figures 1 and 2; electronic supplementary material, figure S4), varying in several traits, including the total number of stolons, the presence of rhizomes and flowering time (electronic supplementary material, table S3 and figure S8). We additionally grew 24 replicates of each inbred parental line and F1s. Due to an asymmetric seed barrier between M. decorus and M. guttatus [45], M. guttatus was the maternal donor for all F1s.

Figure 1.

(a) Geographic ranges of each perennial. (b) Perennials vary in life-history, and these differences increase across development (panels). Greater values of PC1 indicate more investment in vegetative biomass. (c) An ML phylogeny of five perennial taxa (plus annual M. guttatus and annual M. nasutus) using whole genome sequences. Note deep population structure between southern and northern M. guttatus, discussed in [37]. Bolded red branches represent taxa that always produce rhizomes. Numbers within triangles reflect the number of genomes used. (d) Differences in elevation niche between perennials. Violin plots represent predicted suitable habitat from maxent models, while interior boxplots are the elevations of herbarium samples. (e) Life-history expression within perennial species is correlated with elevation of the source population across development (panels). (Online version in colour.)

Figure 2.

(a,b) Representative M. decorus (DEC) and coastal perennial M. guttatus (CP). Both plants make stolons, but only M. decorus produces rhizomes. (c) Proportion of rhizomes by species. COR = M. corallinus, CP = coastal-perennial M. guttatus, DEC = M. decorus, IP = inland perennial M. guttatus, TIL = M. tilingii. (Online version in colour.)

Individuals were surveyed at the same developmental time points as above, but not destructively. Three- and six-week post-stratification individuals were assessed only for the number of stolons and rhizomes. On the day of the first flower, we scored flowering time (date and node of the first flower), the length and width of the corolla and first true leaf, the highest node of stolon emergence, and stolons were measured for several traits (length, width, leaf length, number of branches per node, highest node of emergence). A total of 410 F2 individuals were phenotyped, as well as 24 of each parental line and F1. As replicate F1s are largely genetically identical, phenotypic variation among them is primarily due to environmental effects. We thus used the phenotypic variances (σ²) of F1s and F2s to estimate broad-sense heritability (h²; the proportion of phenotypic variance attributed to genetic variance) for each trait as

$$h^2=\frac{{\sigma }_{[F2s]}^2-{\sigma }_{[F1s]}^2}{{\sigma }_{[F2s]}^2}.$$

We used the phenotypic means (Z) of F1s and parental lines to estimate dominance coefficients for each trait:

$$d=\frac{Z_{\mathrm{PARENT}1}-Z_{F1}}{Z_{\mathrm{PARENT}1}-Z_{\mathrm{PARENT}2}}.$$

(ii). Quantitative trait loci mapping

We created reduced representation libraries for 384 F2s using a modified multiplexed shotgun genotyping approach with the enzyme Csp6I ([57]; see extended methods for details). We aligned and processed reads, then implemented GOOGA to estimate individual genotyping error rates, construct a linkage map and estimate genotype posterior probabilities for each individual at each marker ([58] see extended methods for details). The final dataset comprised 213 individuals with 1311 markers. Each individual was genotyped for an average of 974 markers, and 83% of individuals were genotyped at 50% or more markers. Each marker had, on average, genotype information for 74% of individuals.

We used Haley–Knott regressions in R/qtl [59,60] to identify LOD scores and set significance thresholds using 1000 permutations. For one QTL on LG4, we find a modest effect on the number of stolons, although this did not reach genome-wide significance. We include this QTL in our final map, as single marker analysis using the fitqtl function determined, there was a significant association (F = 2.33, p = 0.033). For traits with more than one significant QTL, we assessed non-additive interactions among QTLs using the addint function in R/qtl [59,60]. Except in one case, we do not recover any significant interactions among QTLs. Lastly, we used these models to calculate the effect size (e.g. a, the average difference between homozygotes and heterozygotes), dominance (d, the difference between a and the mean trait value for heterozygotes), the percentage of phenotypic variance among F2s explained (e.g. %PVE) and the percentage of the parental difference explained by each QTL (e.g. RHE; the relative homozygous effect; the difference between alternative homozygotes divided by the parental difference).

3. Results

(a). Perennial taxa vary in both elevational niche and life-history

Perennial taxa of the M. guttatus species complex differ in geographic and environmental distributions (figure 1; electronic supplementary material, figures S2 and S3; and table S5). Four of five taxa are relatively restricted geographically and/or ecologically, while inland perennial M. guttatus is widespread. Of the more restricted taxa, three inhabit almost exclusively mid to high elevation habitats, with M. tilingii exhibiting a more broad distribution across the Sierras, Cascades, Rockies and Pacific coastal ranges (electronic supplementary material, figure S2 and figure S3) By contrast, M. corallinus exclusively inhabits high elevation sites in the Sierras, and coastal perennial M. guttatus inhabits low elevation sites along the Pacific coast (figure 1; electronic supplementary material, table S5). Differences in elevation niche were accompanied by differences in climate, with M. tilingii, M. corallinus and M. decorus inhabiting generally wetter habitats with higher temperature seasonality (electronic supplementary material, figure S3 and table S5).

Perennials exhibited substantial variation in life-history (as measured by PC1 which explained 49.82% of the variance), and differences between species increased with development (figure 1b; electronic supplementary material, table S6). Greater values of PC1 were associated with greater investment in vegetative biomass via the production of more stolons, stolons that emerge at higher nodes and more branched stolons (electronic supplementary material, figure S5). Thus, M. corallinus and M. decorus make the greatest number of stolons, which emerge at higher nodes and are more branched, while inland perennial M. guttatus, coastal perennial M. guttatus and M. tilingii make fewer stolons. A PCA that included stolon and reproductive traits (e.g. flowering time, floral size) agreed with these results and indicated that perennial taxa that invested more in stolons also flowered later (electronic supplementary material, figure S6).

We next tested if variation in life-history in a common garden was related to the elevation of the source population. We find that life-history covaried with elevation, but only at later developmental time points (figure 1; electronic supplementary material, table S7). Populations originating from higher elevations produced more stolons, stolons that emerged at higher nodes and more branched stolons in a common garden (figure 1; electronic supplementary material, figure S5). For species that produced rhizomes, higher elevation populations also tended to produce more rhizomes.

We scored the presence of a unique life-history trait, the production of rhizomes. Rhizomes are thinner, shorter and more branched than stolons (electronic supplementary material, figures S1 and S9). They lack chlorophyll, are prone to breakage and repress leaf growth (electronic supplementary material, figures S1 and S9). Three of five perennial taxa produced rhizomes at appreciable rates (figure 2). Although we lack the phylogenetic power to formally test for an association between the presence of rhizomes and habitat, we note that all of the taxa that produce rhizomes are also those that live largely in mid to high elevation habitats (figure 1; electronic supplementary material, figure S7).

(b). Few mid-to-large effect quantitative trait loci can explain life-history divergence

We constructed an F2 population between M. decorus and coastal perennial M. guttatus in order to dissect the genetic architecture of life-history differences. These two parents vary in several life-history traits (electronic supplementary material, table S3 and figure S8), including flowering time and size, leaf size, investment in stolons and the presence of rhizomes, and this is representative of differences between species (figure 1 and figure 2; electronic supplementary material, figure S6). The final map consisted of all 14 chromosomes and was 1232.3 cM in length (range: 52.1–125.8 cM per linkage group), with an average marker spacing of 1 cM.

For each of the seven traits that differed significantly between parents, we identified one to three significant QTLs with mid-to-large effects (figure 3; electronic supplementary material, figure S10). Individual QTLs explained 9.3% of the trait variance within F2s (range: 4.3–17.2%) and 83.6% of trait difference among parents (range: 7.5– > 100%; electronic supplementary material, table S4) on average. The average cumulative effect of all QTLs across traits was 15.9% of the F2 variance, and greater than 100% the trait difference among parents (electronic supplementary material, table S4). The observation that the cumulative effects of all QTLs for each trait often surpassed the difference between parents is largely due to the fact that some large effect QTLs influenced traits in the opposite direction than predicted from the parental difference (figure 3; electronic supplementary material, table S4). In line with this observation, some F2s exhibited substantial transgressive phenotypes. Of the 12 QTLs identified, four showed effects opposite to the parental difference, three of which co-localize to LG4.

Figure 3.

QTL map of life-history traits between coastal perennial M. guttatus and M. decorus. Each coloured box represents the 1.5 LOD interval surrounding significant associations (thick horizontal line). Asterisks indicate that the effect of the QTL is opposite of expected. QTL highlighted in grey did not reach genome-wide significance, but was significantly associated based on individual marker analysis. (Online version in colour.)

Two life-history traits—flowering time (e.g. the node of the first flower) and number of stolons—strongly covary within perennials of the M. guttatus species complex (figure 4; electronic supplementary material, table S8). Correlations between these traits persisted among F2s, suggesting that they are genetic (figure 4; electronic supplementary material, table S2). In line with this, QTLs on LG4 and LG13 affected flowering time, the highest node of emergence and stolon number (figure 3; electronic supplementary material, table S4). However, the allelic effects of these two QTL are opposite; for LG4, the M. guttatus allele is associated with later flowering and more stolons (typically M. decorus traits), while the LG13 QTL shows effects in the predicted direction (figure 4; electronic supplementary material, table S4).

Figure 4.

(a) Phenotypic correlation between flowering time and stolon number in the population survey. COR = M. corallinus, DEC = M. decorus, TIL = M. tilingii, IP = inland perennial M. guttatus, CP = coastal perennial M. guttatus. Lines represent linear regressions for each species. (b) Genetic correlations between flowering time and number of stolons in F2s between M. decorus and coastal perennial M. guttatus. Grey circles = individual F2s, blue circle = the coastal perennial M. guttatus parent (OPB), turquoise diamond = F1s and green square = the M. decorus parent (IMP). (c,d) Phenotypic effects of two QTL on flowering time and stolon number. G = the coastal perennial M. guttatus allele, D = the M. decorus allele. Error bars are standard errors. (Online version in colour.)

Lastly, we find that the production of rhizomes changes through development in hybrids and consequently, so does the genetic architecture. Early on, the ability to produce rhizomes is largely recessive, with few hybrids making rhizomes, and we find no significant QTLs for the ability to produce rhizomes while plants are vegetative (figure 5). At reproduction, more hybrids produce rhizomes, but the trait remains mildly recessive. At this point in development, we identify two significant QTLs which exhibit significant non-additive effects (F = 4.24, p = 0.0026), such that trait expression in F2s largely requires the M. decorus allele at both loci. However, later in reproduction, most hybrids produce rhizomes. At this point, no interactive effects are detected (p = 0.89), and a single locus is sufficient to explain approximately 31% of the parental difference (10.5% of the F2 variation).

Figure 5.

(a–c) Trait distributions for each parent (OPB = coastal perennial M. guttatus and IMP = M. decorus), F1s, and F2s. Values in (a) indicate the dominance coefficients for the proportion of stems that are rhizomes (top, grey) and the ability to produce rhizomes (bottom, black). (d–f) Effect of two markers on the proportion of rhizomes. The significance of each marker and their interaction is listed in the top left corner. G = the coastal perennial M. guttatus (OPB) allele, D = the M. decorus (IMP) allele. (Online version in colour.)

4. Discussion

We assessed the extent of divergence in the elevational niche and life-history among perennials of the M. guttatus species complex and determined the genetic architecture of life-history differences between two perennial species. We find that perennials exhibited substantial variation in the elevational niche and life-history, and the extent of phenotypic differences between species increased with ontogeny. Additionally, the presence of a unique life-history trait—rhizomes—is largely restricted to species that occur at mid to high elevations, and within species, investment in vegetative biomass is associated with elevation. Together, these suggest that elevation, and covarying factors, may partially shape life-history evolution within and among perennials of this group. Lastly, we find that the difference between parents for most traits is largely explained by few mid-to-large effect QTLs, including the ability to produce rhizomes. However, aspects of the genetic architecture may have consequences for hybrid fitness in nature.

(a). Elevation may shape life-history variation within and among perennials

In a common garden, perennials of the M. guttatus species complex exhibited clines in life-history along elevational gradients, which is consistent with the hypothesis that elevation, and its associated environmental factors, may partially shape life-history. Individuals from higher elevations produced more stolons, stolons at higher nodes and more branched stolons. Investment in vegetative biomass increased with development, and thus differences between perennials became more apparent through ontogeny. These ontogenetic differences between species likely reflect the fact that as development proceeds, individuals are able to devote more resources to specific traits (for example, higher elevation species/populations can commit more meristems to stolons as more nodes becomes available), rather than differences in selective regimes between species/populations across development (as in [10]). Increased investment in vegetative biomass at higher elevations may increase overwinter survival and promote greater fecundity in later years, as each stolon has the potential to become reproductive [30]. In line with this, perennials have higher fitness than annuals in a high elevation common garden [22]. Similarly, adaptive clines in investment in growth and delayed reproduction exist along elevational gradients in many taxa [14,16,17,20,25]. Further manipulative common garden experiments are needed to assess the agents, strength and targets of natural selection in this system.

Curiously, M. tilingii invested the least in vegetative biomass in our common garden experiment (consistent with [49,61]), despite inhabiting the highest elevations. Mimulus tilingii invests substantially in vegetative biomass in nature [52]. Differences between wild and greenhouse plants likely stem from differences in environment and/or in plant age, wherein high elevation environments may promote investment in more vegetative biomass [19,30], and/or naturally collected plants may be older than the plants phenotyped herein and therefore have had more time to invest in vegetative biomass.

We find the taxa that are restricted to mid to high elevation habitats exhibit a novel trait relative to most M. guttatus populations, the presence of rhizomes. Rhizomes may increase the survival of perennials in environments with extreme cold temperatures and sudden freezing, as subsurface soil temperatures tend to be more stable and warmer than ambient air temperatures during freezing events. Rhizomes have previously been shown to increase survival under other extreme weather conditions [22]. While uncommon in our experiment, 4/48 inland perennial M. guttatus individuals produced rhizomes. These individuals hailed from two mid-elevation sites (1065 m and 866 m, respectively). Mimulus guttatus can co-occur with M. decorus and M. tilingii at high elevations, and this trait may have introgressed between species. Rhizomes may also have evolved independently in each species or existed as a shared polymorphism in their common ancestor (as discussed in [62,63]). Alternatively, rhizomes may represent an ancestral state, wherein their loss accompanied colonization of low elevation habitats in M. guttatus. Given the phylogenetic distribution of this trait, each of these scenarios is plausible, and further population genomic dissections are required to assess these scenarios.

(b). Few mid-to-large effect quantitative trait loci explain life-history divergence

We find that life-history divergence between two perennial species can largely be explained by a few largely additive, mid-to-large effect QTL (although, while we identify QTLs that explain as little as 4% of the variation among F2s (electronic supplementary material, table S4), given that our genetics maps are based on a modest number of F2s, it is likely that we failed to detect some smaller effect loci or more subtle interactions among loci). This is simpler than the genetic architecture of life-history divergence in other systems [5,33,35,38,64–66]. One explanation for why some genetic architectures are complex versus simple is how natural selection acts on the trait in question [67,68]. Strong, divergent natural selection between populations is expected to produce simpler genetic architectures than stabilizing or balancing selection within populations [67–69]. In the case of life-history divergence between coastal perennial M. guttatus and M. decorus, trait differences may have arisen due to divergent natural selection via elevation (and covarying factors).

We uncover some curious features of the genetic architecture of life-history divergence between coastal perennial M. guttatus and M. decorus. First, we find two QTLs that are associated with three highly correlated traits: stolon number, flowering time and the highest node of stolon emergence. QTLs that affect both flowering time and stolon number have been described between annual and perennial M. guttatus [5,38,40] and may represent a genetic control of early meristem commitment (e.g. production of floral buds versus stolons; [5,44,70]). The LG4 QTL exhibits opposite allelic effects than predicted by the parental difference. Consequently, some F2s exhibit transgressive phenotypes—delayed flowering and substantially more stolons than either parent. Such extreme phenotypes may have negative fitness consequences in nature, highlighting a potential for extrinsic postzygotic reproductive isolation among perennials.

Second, we find that the genetic architecture of the proportion of rhizomes changes with development, and relatedly, so does the dominance. Early in reproduction, the ability to produce rhizomes is recessive and is controlled by two interacting loci. However, later in development, the ability to produce rhizomes becomes dominant (with the proportion of rhizomes showing minimal dominance), and only one of the previously determined QTLs continues to influence this trait. One potential genetic model for the change in genetic architecture through development is if the LG11 QTL largely controls the ability to produce rhizomes, but the timing of its expression is modulated by the LG9 QTL. This has implications not only for understanding the genetic basis of species divergence, but also, potentially for hybrid fitness in nature. While most hybrids produce rhizomes, the fact that rhizomes develop later in hybrids than in M. decorus may influence their competitive establishment between hybrids and pure species in high elevation habitats. Further reciprocal transplants will be necessary to assess hybrid versus parental fitness in each species' habitat.

Data accessibility

Data for the common garden experiments, genotype and phenotype data from the mapping experiment, and occurrence records used for the MaxEnt modelling are available from the Dryad Digital Repository: https://doi.org/10.5061/dryad.qnk98sffj [71]. Sequences used to build the maximum-likelihood phylogeny are available via the NCBI SRA (BioProject: PRJNA692721; see electronic supplementary material, table S9 for details).

Authors' contributions

J.M.C. conceived of the idea for this project, performed all experiments and data analyses, and wrote the paper. M.W.B. helped with phenotyping. J.W. contributed to the intellectual development of this project.

Competing interests

We declare we have no competing interests.

Funding

This project was funded by an NSF DDIG (DEB-1501758), ASN Student research award and a SSE Rosemary Grant Award to J.M.C., and an NSF Rules of Life award (DEB-1856157) to J.W. J.M.C. was funded by an NSF Dimensions of Biodiversity award (DEB-1737752) to Daniel Matute.