Summary
Genetic correlations among different components of phenotypes, especially resulting from pleiotropy, can constrain (antagonistic) or facilitate (adaptive) trait evolution. These factors could especially influence the evolution of traits that are functionally integrated, such as those comprising the flower. Indeed, pleiotropy is proposed as a main driver of repeated convergent trait transitions, including the evolution of phenotypically-similar pollinator syndromes.
We assessed the role of pleiotropy in the differentiation of floral and other reproductive traits between two species—Jaltomata sinuosa and J. umbellata (Solanaceae)—that have divergent suites of floral traits consistent with bee- and hummingbird-pollination, respectively. To do so, we generated a hybrid population and examined the genetic architecture (trait segregation and QTL distribution) underlying 25 floral and fertility traits.
We found that most traits had a relatively simple genetic basis (few, predominantly additive, QTL of moderate to large effect), as well as little evidence of antagonistic pleiotropy (few trait correlations and QTL co-localization, particularly between traits of different classes). However, we did detect a potential case of adaptive pleiotropy among floral size and nectar traits.
These mechanisms may have facilitated the rapid floral trait evolution observed within Jaltomata, and may be a common component of rapid phenotypic change more broadly.
Introduction
One feature of phenotypic evolution is the broad variation in observed rates of trait change, both among traits within a single lineage and among lineages. How quickly phenotypes evolve and in what direction, depends not only on their genetic basis--both the number and effect size of causal loci--and the intensity and nature of selection acting upon them, but also on their associations with other traits. Because of genetic covariance, developmental and phylogenetic constraints, and/or correlated selection pressures (Agrawal and Stinchombe 2009), phenotypic traits often do not evolve independently from one another. Among these causal mechanisms, strong genetic covariance arises either because different traits have a shared genetic basis (pleiotropy) or because they are based on genes that are physically adjacent and therefore often co-inherited (via linkage). Strong pleiotropy is often proposed to constrain phenotypic evolution by preventing correlated traits from moving efficiently towards their own (different) fitness optima (antagonistic pleiotropy). However, such shared genetic control may also promote phenotypic change if covariation is aligned in the direction of selection (adaptive pleiotropy) (Agrawal and Stinchombe 2009; Wagner and Zhang 2011; Smith 2016). Nonetheless, despite some detailed studies (e.g. Ettensohn 2013; Shrestha et al. 2014; Manceau et al. 2011; Xu and Schluter 2015), the genetic architecture of many ecologically important traits remains unclear, including the prevalence of strong genetic associations that could shape the course of phenotypic evolution.
How shared architecture influences phenotypic change is especially relevant for suites of traits that are functionally integrated (Armbruster et al. 2014) -- such as the angiosperm flower (Armbruster et al. 2009) -- because the magnitude and direction of pleiotropy will directly shape if and how these co-varying traits respond to selection. Further, because pleiotropy can constrain or favor particular developmental trajectories, it is often proposed to be a main driver of convergent transitions of integrated traits in different lineages (Preston et al. 2011; Smith 2016). The flower is an especially promising model for assessing the role of pleiotropy in shaping phenotypic evolution. Because flowers mediate fitness through their critical reproductive role, their constituent traits (i.e. reproductive structures, perianth, and other attraction/reward features) are often highly functionally integrated (Conner 2002; Armbruster et al. 2009). Moreover, repeated transitions of phenotypically similar or convergent suites of floral traits have been identified both within and across groups (Fenster et al. 2004; Goodwillie et al. 2010; Wessinger et al. 2016). For example, multiple parallel shifts from bee-pollination to hummingbird-pollination are associated with parallel transitions to flowers with red petals, large amounts of dilute nectar, and narrow corolla tubes within Penstemon (Wessinger et al. 2016); similarly, the evolution of the ‘selfing syndrome’ (i.e. reduced overall size, herkogamy, and floral rewards) often accompanies transitions from outcrossing to predominantly selfing mating systems, and has been documented in multiple lineages (Stebbins 1974; Goodwillie et al. 2010). Such patterns provide an opportunity to assess the relative frequency of adaptive vs. antagonistic pleiotropy in shaping these repeated trait combinations, within a comparative phylogenetic context.
In addition to these ecological and evolutionary features, the known genetic and molecular bases of floral development (Rijpkema et al. 2006; Smaczniak et al. 2012) themselves suggest that pleiotropy might be a key component shaping floral phenotypic change, as well as provide a functionally-informed framework for identifying how changes in these mechanisms can contribute to the diversification of floral traits. Under the ABC(DE) model of floral development, the combinatorial action of different gene products -- primarily different MADS-box transcription factors -- control transitions to flowering and the specification of floral organ identities and organ maturation, by promoting or repressing different downstream targets (reviewed in O’Maoileidigh et al. 2014; Bartlett 2017). This combinatorial function, and the ability to regulate shared or partially shared downstream targets, provides a potential mechanistic explanation for strong pleiotropy among floral traits. Moreover, several key regulators of floral development also function during fruit and seed production (Smaczniak et al. 2012; O’Maoileidigh et al. 2014); such correlated effects on fertility traits are another potential way that pleiotropy could shape floral phenotypic evolution.
Several empirical approaches have been taken to assess the genetic architecture underlying floral trait specification and evolution, and to evaluate the strength and direction of pleiotropy. Classical quantitative genetic analyses have revealed varying degrees of genetic covariance among floral traits (Gottlieb 1984; Conner et al. 2014), while numerous QTL (quantitative trait locus) mapping studies have found that loci for at least some different floral traits appear to co-localize to the same genomic region(s) (reviewed in Smith 2016). Interestingly, such studies have identified more putative cases of adaptive pleiotropy (e.g. QTL affect more than one trait in the direction of parental trait values) than antagonistic, suggesting that adaptive pleiotropy may be a common mechanism contributing to rapid floral evolution. Because QTL generally span a genomic region that contains more than one gene, such QTL co-localization is consistent with, but not definitive evidence of, a role of pleiotropy in shaping floral trait co-variation (e.g. see Hermann et al. 2013). Nonetheless, identifying strong trait correlations and/or QTL co-localization represents a critical step in assessing how pervasive pleiotropy could be in shaping floral trait evolution.
In this study, we examined phenotypic (co)variation among, and the genetic architecture underlying, floral and other reproductive traits within a segregating hybrid population derived from two Jaltomata (Solanaceae) species with divergent floral traits (Figure 1). Despite only having diversified with the last 5 million years (Sarkinen et al. 2013; Wu et al. 2018), species of Jaltomata are highly phenotypically diverse, including extensive variation in the size, shape, and color of floral traits that is absent among their close relatives within Solanum and Capsicum. Indeed, phylogenetic analyses suggest numerous transitions in floral traits within the genus, including several instances of convergent evolution (Miller et al. 2011; Wu et al. 2018). Importantly, many of these transitions appear to involve parallel changes in several different traits within a lineage (e.g. from flat corollas with small amounts of lightly colored nectar to highly fused corollas with large amounts of darkly colored nectar), suggesting either a shared genetic basis and/or correlated selection (perhaps pollinator-mediated selection, Fenster et al. 2004) influences these trait associations.
Here, we identified QTL contributing to floral and other reproductive trait variation within a recombinant population to: 1) examine the genetic architecture underlying reproductive trait divergence; 2) assess the role of genetic linkage and/or pleiotropy (via strong trait correlations and overlapping QTL) in floral trait (co)variation; and 3) assess evidence for a shared genetic basis between different classes of floral trait and other reproductive (specifically fertility) traits. We found evidence for a relatively simple genetic basis underlying most of the examined traits, as well as positive correlations and significant QTL co-localization among traits within each of three trait classes (floral morphology, floral color, and fertility). Together, these features might facilitate rapid changes in these traits. In comparison, we found few associations between traits from different classes, and therefore little evidence for antagonistic pleiotropy among these classes. One striking exception was an association between flower size and nectar traits that acts in the direction exhibited by multiple species in the genus, suggesting that this could instead be an instance of adaptive pleiotropy.
MATERIALS AND METHODS
Study system
Jaltomata (Schlechtendal; Solanaceae) is the sister genus to the large and economically important Solanum (Olmstead et al. 2008; Sarkinen et al. 2013; Wu et al. 2019), and includes approximately 60-80 species distributed from the Southwestern United States to the Andean region of South America, in addition to several species endemic to the Greater Antilles and the Galapagos Islands (Miller et al. 2011; Mione et al. 2015). Species are highly phenotypically diverse and live in a variety of habitats (e.g. tropical rainforests, rocky foothills, and lomas formations), despite their recent divergence (<5MYA, Sarkinen et al. 2013).
Here, we focused on a closely related species pair, J. sinuosa and J. umbellata, that differ in a suite of floral traits that is representative of major floral suites found in other species throughout the genus (Figure 1; Table S1). Jaltomata sinuosa has large rotate flowers with purple petals and a small amount of concentrated nectar (consistent with bee pollination, Fenster et al. 2004), while J. umbellata has small short-tubular flowers with white petals and a large amount of dilute, but dark red, nectar that is visible through the corolla tube (consistent with hummingbird pollination, Fenster et al. 2004; Kostyun and Moyle 2017; Mione et al. 2017). Jaltomata sinuosa is distributed along the Andes in South America from Venezuela to Bolivia, while J. umbellata is restricted to lomas formations along the Peruvian coast. Both species are self-compatible (Kostyun and Moyle 2017; J.L. Kostyun and T. Mione, unpub.) and shrubby, but differ in leaf traits such as overall size and shape, and type of trichomes. This species pair also has several incomplete, intrinsic postzygotic reproductive barriers, including quantitatively reduced fruit set and hybrid seed viability (Kostyun and Moyle 2017).
Generation of BC1 population and plant cultivation
We developed a segregating hybrid population by crossing J. sinuosa and J. umbellata. Viable F1 individuals in both directions of the cross produce flowers that are phenotypically intermediate between the parental genotypes (Figure 1), and retain reduced but sufficient levels of fertility when back-crossed to parents (Kostyun and Moyle 2017). Because we were especially interested in the genetic basis of red nectar and a fused corolla tube (exhibited by J. umbellata and F1s but not J. sinuosa), we generated the mapping population by backcrossing a single J. sinuosa x J. umbellata F1 (as the ovule parent) to the original parental J. sinuosa individual (as the pollen parent). BC1 individuals were germinated in a growth chamber, and then moved to the Indiana University greenhouse and grown under the same conditions as the parental and F1 individuals (16 hour light cycle, watered twice daily, and fertilized weekly).
Trait measurements
We measured 25 floral and other reproductive traits within our mapping population, F1, and parental genotypes (Table 1; Table S2; Figure S1). Floral morphological traits were measured with hand-held calipers, and nectar volume per flower was measured to the nearest 1 µL using a pipette. To reduce the potential effect of daily environmental variation, nectar volume was always measured in the early afternoon following watering.
Petal and nectar color were quantified using digital photography (Kendal et al. 2013; Garcia et al. 2014): dissected petals and nectar drops were photographed on a standard background along with white and black color standards. Light conditions were standardized for all images using RAW Therapee (RAW Therapee Development Team 2012), and color space attributes were measured in ImageJ (Schneider et al. 2012). Because RGB color attributes are device-dependent (i.e. they can vary depending upon the specific camera used), color values were also converted into device-independent L*a*b color attributes, using the ImageJ Color Space Converter plugin (Schwartzwald 2012). For both petal and nectar color, this approach produced eight interrelated color attributes: Intensity, Red, Green, Blue, Composite RGB, Lightness (L), ‘a’ color (ranges from green to magenta), and ‘b’ color (ranges from cyan to yellow). Broadly, Intensity, Red, Green, Blue, and Lightness convey information about color brightness, while Composite RGB, ‘a’ color, and ‘b’ color convey information about hue.
We also measured nine fertility traits (Table 1) to assess the potential genetic overlap between floral and other reproductive traits, as well as to examine the genetic architecture underlying intrinsic postzygotic barriers between this species pair. Fruit and seed related traits were measured on 2-6 crossed fruit per individual (depending on fruit set). For F1 and BC1 individuals, crossed fruit were produced using pollen from the J. sinuosa parental individual. To determine seed germination rates (following Farooq et al. 2005), we soaked 10 seeds per individual in 50% bleach for 30 minutes (to soften the seed coat), rinsed thoroughly, and placed on moist paper within plastic germination boxes. A week after sowing, seed coats were nicked slightly and seeds were given a drop of 10 mM giberellic acid (Sigma) to break dormancy. We then scored germination every 2 weeks for 4 months. Pollen viability was estimated from three different flowers per individual, using established methods (Jewell et al. 2012): Briefly, for each sample all undehisced anthers from a flower were collected into an eppendorf tube containing aniline blue histochemical stain, gently ground with a pestle to release pollen, and viable pollen grains were counted under an EVOS FL Digital Inverted Fluorescence Microscope (Fisher Scientific). From these data, we also calculated the proportion of viable pollen as number of viable pollen grains/total pollen grains in the sample.
Statistical analyses on phenotypic data
Following Shapiro-Wilk tests to assess normality assumptions, we transformed traits that showed a skewed distribution and/or significantly non-normal residuals. In particular, we arcsine transformed proportional traits (proportion of corolla fusion, fruit set, and proportion of viable pollen), and log-transformed inflorescence size, corolla fusion, petal length, style length, herkogamy, nectar volume, all color attributes, and remaining fertility traits. Significant differences between parental species for all traits were assessed by t-tests (Table 1). Distribution plots for all traits are provided in Supplementary Materials (Figures S2-S6), while illustrative plots for 15 focal traits are provided in the main text (Figure 2). Similarly, phenotypic correlations within the BC1 population were examined among all traits (Table S3), while relationships among a subset of focal traits are presented in the main text (Figure 3). Given significant correlations among many of the floral morphological traits and among color attributes (Table S3), we also used principle component analyses (PCAs) to create separate composite metrics (principle components (PCs)) for three groups of traits (i.e. Morph PC1-PC3; Nectar Color PC1-3; Petal Color PC1-PC3; Table S4) as additional measures of floral variation. Trait correlations, including PCs, are provided in Table S5, and distribution plots are provided in Figures S3-S5.
Genotyping and linkage map construction
Genomic DNA was extracted from young leaf tissue from the 2 parental individuals, 13 F1s (including the F1 parent used to generate the population), and 269 BC1s, using the Qiagen DNeasy Plant Mini Kit. DNA quantity and quality were confirmed via Nanodrop (Fisher Scientific) and gel electrophoresis with λ DNA-HindIII Digest marker (New England BioLabs). Samples were then sent to Novogene Corporation (Beijing) for genotyping-by-sequencing (GBS). GBS libraries were prepared using optimized restriction enzymes (MseI and HaeIII), and following insert size selection, sequenced on an Illumina Hi-Seq to generate 150bp paired end reads. Raw reads were trimmed and filtered using Trimmomatic (Bolger et al. 2014), and read quality was checked pre- and post-trimming using fastqc (Andrews 2010). To identify SNPs, cleaned reads were mapped to the domesticated tomato genome (Tomato Genome Consortium 2012) using the mem function in BWA (Li 2013). Alignment files were then input into the STACKS refmap pipeline (Catchen et al. 2013) to determine genotypes. Reads and genotype data are available in NCBI SRA XXXXXX.
To construct the linkage map, we first removed markers that were genotyped in less than 35% of individuals, or showed significant segregation distortion (i.e. alleles at >80% or <20% frequency). The linkage map was constructed using the MST and Kosambi algorithms, implemented in the R package ASMap (Taylor and Butler 2017). To alleviate map expansion issues, we removed markers which consistently differed from neighboring markers in terms of genotype assignment, indicating a high likelihood of genotyping error. The linkage map was then finalized using the ripple function in R package R/qtl (Broman et al. 2003).
Identifying QTL
We implemented Haley-Knott regression in R/qtl to identify QTL contributing to each trait. To account for potential environmental contributions to trait variation, we included date of measurement (Month) and location within the greenhouse (Bench) as covariates in our QTL scans. Putative QTL were first identified using the scanone function, followed by permutations for genome-wide LOD significance thresholds. Two dimensional scans (scantwo function) were used in the stepwise qtl function to fit multiple QTL models. These models were used to identify significant QTL, their 1.5 LOD confidence intervals, their effect sizes (i.e. difference in phenotype mean between homozygotes and heterozygotes), and the total amount of phenotypic variance explained by each QTL, as well as interactions among QTL, and potential contributions of covariates. QTL were considered to be co-localized if their 1.5 LOD intervals overlapped. Significant co-localization was assessed by comparing overlap among identified QTL to overlap from 10000 randomly generated distributions, for traits within each category (morphological, color/physiological, or fertility), and between each trait category. Briefly, a custom Python script was used to generate random distributions of QTL (by randomly re-distributing the identified QTL among the 12 linkage groups), and the observed frequency of co-localization in each was recorded for each randomization to generated count distributions, in R. All code used to generate the linkage map, identify QTL, and assess QTL co-location, is available on GitHub (https://github.com/gibsonMatt/jaltomataQTL).
RESULTS
Segregation patterns suggest additive alleles underlie most traits
Most traits were significantly different between the two parental species (Table 1). F1 means were intermediate for most traits as well, except that petals were generally brighter (more white) than either parent (Figures S5+S7). Other than fruit set and seed germination rates, all traits were unimodally distributed within the BC1s; phenotypic values were intermediate between F1s and the recurrent parent (J. sinuosa) for many of these traits, consistent with additive effects (Figure 2; Figures S2-S6). Several traits (7 of 25) showed transgressive segregation within the BC1s, including some floral morphological traits, nectar volume and color, and seed viability and germination rates (Table S2; Figures S2-S6).
Significant correlations observed within – but generally not between – floral morphology, floral color, and fertility trait categories
Within the BC1s, most trait combinations were not strongly associated. Nonetheless, several correlations remained significant following multiple testing (Bonferroni) correction (Table S3), primarily associations that are expected biologically, including allometric relationships among floral organs and positive associations among related fertility traits. For instance, corolla diameter was significantly positively associated with most other morphological traits, suggesting shared genetic control of overall floral size (Figure 3), while corolla diameter was also significantly negatively correlated with proportion of corolla fusion (i.e. shorter corolla tubes had wider limbs and longer tubes had narrower limbs, r = −0.348, p = 8.68e−8). These relationships were recovered with PCA on morphology traits, in which PC1-PC3 explained 76% of the variance among BC1s (Table S4). Based on trait loadings, PC1 corresponds to floral width vs. depth, PC2 to overall floral size, and PC3 to relative reproductive organ dimensions. Similarly, related fertility traits also remained strongly correlated, such as fruit mass with seed set, and number of viable pollen grains with proportion of viable pollen (Table S3). Finally, color attributes within each of nectar color and petal color were strongly correlated with one another, but these attributes were not associated between nectar and petals (Table S3). From PCAs on nectar color and petal color attributes (separately), PC1-PC3 for each explained 97% and 94% of the variance among BC1s, respectively (Table S4).
In contrast, there were relatively few significant correlations among different trait categories. Notable exceptions, however, included a positive relationship between floral size and nectar volume as well as floral size and certain aspects of nectar color (Figure 3; Table S3). There were also significant positive correlations between pollen viability and each of several components of flower size (as well as Morph PC2 or “size”) (Tables S3+S5). This latter relationship seems to be explained by anther size: across 15 Jaltomata species, mean viable pollen count is significantly associated with anther size prior to dehiscence (F = 15.56, p = 0.0017) (J.L. Kostyun, unpub.).
Linkage map construction recovered 12 linkage groups
Mapping high quality reads to the tomato genome identified 25,136 SNPs that differentiated the two parental species. Following all subsequent filtering (removing markers genotyped in less than 35% of individuals, with high segregation distortion or non-Mendelian inheritance, or with high likelihood of genotyping errors), we retained 520 high quality markers. Linkage map construction recovered 12 linkage groups (LGs), which correspond to the number of chromosomes in the parental species (Mione et al. 1993; Chiarini et al. 2017). Based on orthology with tomato, we were able to confidently assign 5 of these LGs to chromosomes (Figure 4). Total map length was 1593.71 cM (65.17 cM – 324.48 cM per chromosome/LG), with an average of 2.92 cM between markers (Figure 4).
Few moderate-effect QTL underlie most traits, with little QTL co-localization between trait categories
We identified a total of 63 QTL for our 25 traits (with 4 additional loci for PC traits). Most traits had 2-4 QTL, with a range of 0-6 QTL (Table 2, Table S6). Alleles at 55 of 67 QTL (82%) acted in the direction consistent with parental values (i.e. the allele from paternal donor J. umbellata moved the phenotype of BC1s closer to its species mean) (Table 2), and the amount of phenotypic variation explained per QTL ranged from 2-28%. The latter range suggests that we had reasonable power to identify QTL with even relatively small effects--explaining as little as 2% of the variance. Consistent with observed trait segregation patterns, significant interactions among QTL within individual traits were identified in few cases: for ovary diameter and certain nectar color attributes (Table 2, Table S6).
Although every linkage group had at least one QTL, QTL were not distributed uniformly across the genome, with notable clusters on LG1 and LG9 (Figure 4, Table 2). We also identified several instances of QTL co-localization within trait categories, especially for morphology and fertility traits, which each had significantly more cases of co-localization (1.5 LOD overlap) than expected by chance (133 observed vs. upper bound of 115 expected overlaps, p = 7.5e−4; 8 observed vs. upper bound of 4 expected overlaps, p = 7.0e−6, respectively) (Table S7; Figure S8). These co-localization instances included QTL for biologically-related traits (e.g. petal length and corolla diameter, or fruit mass and seed set; Table S6), for which we already observed strong correlations (Table S3). In some cases, co-localized QTL share the same or a very close peak marker (e.g. petal length, corolla fusion, and ovary diameter on LG12; Table S6) which is suggestive of potential pleiotropy, however we note that the large 1.5 LOD intervals of some QTL will increase instances of incidental co-localization events.
In contrast, co-localization between different trait categories was never greater than expected by chance, with co-localization between morphology and color traits actually significantly less than expected (p = 0.016) (Table S7; Figure S8). This is consistent with mostly incidental occurrences of overlap between QTL for traits in different categories. Nonetheless, we did detect co-localized QTL at the same or very close peak markers for, for example, nectar color (RGB) and volume on LG3 (Figure 4; Table S6) and for nectar color (a), corolla fusion, and corolla depth on LG7, (Table S6), which provide intriguing cases of potential adaptive pleiotropy (i.e. alleles at QTL that simultaneously act to increase floral size, nectar darkness, and/or nectar volume).
DISCUSSION
Genetic correlations among different components of the phenotype, especially resulting from pleiotropy, can constrain or facilitate trait evolution (Agrawal and Stinchcombe 2009). Pleiotropy could have particularly strong effects on the evolution of traits that are functionally integrated, such as those comprising the flower (Armbruster et al. 2009; Smith 2016). To better understand the genetic architecture underlying floral trait evolution within florally diverse Jaltomata, including whether pleiotropy might have shaped observed variation, we examined patterns of genetic segregation and genetic architecture for 25 floral and fertility traits in a hybrid (BC1) population between species with highly divergent floral traits. We found that most of our examined traits have a relatively simple genetic basis, with few to moderate QTL with largely additive effects. We also identified strong correlations and significant QTL overlap within trait categories, but few associations across different types of traits. The exceptions however, between certain aspects of floral morphology and nectar traits, are consistent with existing trait associations that are observed across the genus, suggesting that these could be examples of adaptive pleiotropy. Overall, our data suggest that the rapid floral trait evolution observed in this group could have been facilitated by a relatively simple genetic basis for individual floral traits, and a general absence of antagonistic pleiotropy among different types of reproductive traits, especially morphology and color.
Few genetic changes could underlie floral trait shifts
The relatively simple genetic architecture that we detect for most of our floral traits might be one mechanism that has permitted rapid floral evolution within the genus. Indeed, our inference that few QTL controlling corolla traits agrees with comparative development data from these species (Kostyun et al. 2017) in which we observe that relatively simple heterochronic changes in corolla trait growth rates distinguish these rotate vs. tubular corolla forms. Interestingly, our findings are also consistent with previous studies of floral trait genetics between closely related species (Smith 2016). For instance, one or few QTL have been found for species differences in nectar volume in several other systems (e.g. Bradshaw et al. 1998; Stuurman et al. 2004; Wessinger et al. 2014; but see Nakazato et al. 2013), similar to our inference of a single QTL for this trait. For petal and nectar color, we identified 2 and 5 QTL, respectively, each with moderate to major effects (Table 2, Table S6), similar to other systems that generally identify few loci of large effect for petal color (e.g. Bradshaw et al. 1998; Wessinger et al. 2014). Perhaps unlike these cases, however, it is likely that loci controlling color differences in Jaltomata are regulators of pigment quantity rather than presence/absence biosynthesis, because both nectar and petal color show gradation in the BC1s rather than discrete color bins. Moreover, preliminary data from a VIGS (virus-induced gene silencing) pilot study in J. sinuosa indicate that the purple petal pigment is an anthocyanin (J.L. Kostyun and J.C. Preston, unpub.), whereas for nectar color, preliminary data suggest that an indole-flavin contributes to red pigment in J. umbellata (J.L. Kostyun and D. Haak, unpub.), consistent with our inference that color variation is unassociated between these different floral components.
In addition to relatively few contributing loci, many of the examined traits also appear to be underpinned by additive effects (Table 1; Figure 3; Figures S2-6), while epistatic effects were comparatively rare. Both are factors that might also facilitate more rapid responses to selection. Other studies have similarly found that floral size traits are often additive (Gottlieb 1984). Although several floral traits showed transgressive segregation within our BC1s, which could indicate epistatic interactions, similar patterns can result from unique combinations of additive alleles that have opposite effects in the parental species (e.g. deVicente and Tanksley 1993) and we identified individual QTL with these opposing effects for many of these traits. In comparison, significant interaction effects among morphological QTL were detected for ovary diameter only (Table 2), consistent with a general lack of epistatic interactions for this class of traits.
The notable exceptions to additivity involved many of the fertility traits (as well as some components of color, see below). BC1 individuals tended to have lower seed set and poorer quality seeds (decreased viability and response to germination-inducing stimuli), and the recombinant BC population contained a subset of highly sterile individuals. The segregation of recombinant individuals with reduced viability and fertility often occurs in hybrids (Baack et al. 2015), including in hybrids from additional Jaltomata species pairs (Kostyun and Moyle 2017). Such patterns are typically due to deleterious epistatic interactions between loci that have diverged between the two parental lineages, as has been shown in close relatives including tomatoes (Moyle and Nakazato 2008; Sherman et al. 2014). These observations in Jaltomata are similarly consistent with a specific role for epistasis among incompatible alleles in the expression of postzygotic reproductive isolation.
Reduced constraints may also have facilitated rapid floral trait evolution
Because rapid floral evolution may occur either through a lack of antagonistic pleiotropy or through adaptive pleiotropy, we assessed evidence for these potential mechanisms within Jaltomata. Within trait categories, we detected positive but modest associations between several floral size traits, and among biologically related fertility traits (e.g. fruit size and seed set) (Figure 3; Table S3), as well as significant co-localization of QTL for these groups of traits (Table S7). Morphological associations in particular suggest that shared growth regulators (e.g. Sicard and Lenhard 2011; Brock et al. 2012) contribute to - but do not completely determine - observed variation in floral organ sizes. In contrast, we detected fewer instances of strong trait correlations and QTL co-localization between different trait categories (Table S3, Table S7). This general lack of antagonistic pleiotropy among different classes of floral and fertility traits may have facilitated rapid floral evolution in this system by minimizing constraints on the available combinations of floral traits. Despite this general pattern, we did identify several instances of QTL co-localization that might represent adaptive pleiotropy. Most notably, larger flowers generally produced darker (more red) nectar as well as a greater volume of nectar, as reflected in co-localization of QTL underlying aspects of floral morphology and both nectar volume and color (Table 2; Figure 4; Table S6). Interestingly, this trait combination (large flowers with copious dark nectar) is actually not exhibited by either parental species used in this experiment (Figure 1; Table 1); however, it is found in numerous other Jaltomata species (see below; Miller et al. 2011; Kostyun and Moyle 2017) and is consistent with recognized pollination syndromes (e.g. Fenster et al. 2004).
Ecological context for rapid floral change in Jaltomata
Overall, our findings suggest potential mechanistic explanations for the evolution of remarkable floral trait diversity among Jaltomata species within the last 5 million years (Sarkinen et al. 2013). Traits with a relatively simple genetic basis that are uncoupled from other floral and fertility traits have fewer mechanistic constraints, and therefore can more rapidly respond to selective opportunities as they arise. Although we have not yet directly assessed the role of selection in shaping floral differences among species, several features of Jaltomata floral biology are consistent with pollinator-mediated selection on floral traits (van der Niet and Johnson 2012), likely in conjunction with mating-system related changes (Goodwillie et al. 2010). First, floral trait variation within Jaltomata shows clear hallmarks of selection imposed by pollinator differentiation. Nearly all species in the earliest diverging Jaltomata lineage have relatively small, ancestrally rotate flowers with small amounts of lightly colored nectar, and hymenopterans have been observed visiting several of these species (T. Mione, pers. comm.). In contrast, many species in the South American derived clade—including J. umbellata examined here--have attractive features associated with vertebrate pollination (Fenster et al. 2004). Several of these are visited by hummingbirds (T. Mione, per. comm.), notably those with larger flowers with a highly fused corolla (either campanulate or tubular) and copious amounts of darkly colored nectar; intriguingly, this repeated natural trait covariation is consistent with the genetic association between floral size and nectar traits we identified here.
These features indicate that pollinators are a likely source of selection for floral differentiation among species within Jaltomata, but they do not necessarily explain why Jaltomata as a genus has been uniquely responsive to this pollinator variation, especially in comparison to its most close relatives. Species from both Solanum and Capsicum are found within the same geographical regions as Jaltomata, and are therefore exposed to the same pollinator variation, but are nonetheless almost uniformly rotate and bee-pollinated (Knapp 2010). Interestingly, one important difference between Jaltomata and these two genera is in their predominant mating system. Self-incompatibility is the ancestral state in the Solanaceae (Steinbachs and Holsinger 2002) and is broadly persistent in both Solanum and Capsicum (Goldberg et al. 2010). In contrast, all examined Jaltomata species are self-compatible (Mione 1992; J.L. Kostyun and T. Mione, unpub.), indicating that gametophytic self-incompatibility was lost early in the evolution of this clade. Moreover, the presence of delayed selfing and strong herkogamy in many species (e.g. Mione et al. 2015; Mione et al. in review), in addition to field observations of pollinators (above; T. Mione, pers. comm.), indicate that species most likely employ a mixed mating strategy in their native ranges. The absence of genetically-determined self-incompatibility and the predominance of mixed mating strategies might have uniquely facilitated the evolution of new floral trait variation in Jaltomata, compared to either Capsicum or Solanum. Mixed mating strategies are generally observed to maintain the largest amount of floral trait variation, compared to predominant selfing or enforced outcrossing (Goodwillie et al. 2005; Rosas-Guerrero et al. 2014). In addition, they have been predicted to facilitate pollinator shifts--especially to pollinators that might be more efficient but potentially unreliable (such as hummingbirds)—because they allow reproductive assurance (via selfing, when pollinators are limited) and increase the expression of new floral trait variation controlled by recessive alleles (Goodwillie et al. 2005; Brys et al. 2013; Wessinger and Kelly 2018). Notably, our data indicate that dark/red colored nectar is at least partially recessive (Figure 2), and that red petal pigmentation is completely recessive (Figure S7), consistent with this novel variation being based on new recessive alleles.
CONCLUSIONS
Genetic correlations among floral traits, especially those due to pleiotropic effects, likely shape permitted trajectories of floral evolution. To assess how such genetic associations might have contributed to observed patterns of floral diversity in Jaltomata, we examined segregation patterns and genetic architecture of 25 floral and fertility traits in a hybrid (BC1) population generated from parents with divergent floral traits. Our data are consistent with several mechanisms that could have allowed rapid floral trait evolution in this system: a largely simple genetic basis underlying variation in most of our floral traits, a general absence of antagonistic pleiotropy constraining floral evolution, and a potential instance of adaptive pleiotropy governing floral size and nectar traits. This genetic architecture, in combination with pollinator-mediated selection on a background of self-compatible mixed mating, might have uniquely positioned this genus for the rapid floral diversification now evident within Jaltomata.
AUTHOR CONTRIBUTIONS
JLK and LCM designed the experiment, JLK generated experimental materials, JLK and CMK collected phenotypic data, JLK and MJSG analyzed the data, and JLK and LCM wrote the paper with input from CMK and MJSG.
ACKNOWLEDGEMENTS
We thank the IU greenhouse staff for plant care, CJ Jewell and David Haak for logistical support, undergraduate research assistants (especially Meret Thomas-Huebner, Devki Shukla, and Shachia Jackson) for assistance with data collection, and Tim Leslie for assistance with simulations. This work was supported by the IU Biology Department, National Science Foundation Award (NSF DEB 1136707) to LCM, and National Science Foundation Graduate Research Fellowship Program (NSF DEB 1342962) and Doctoral Dissertation Improvement Grant (NSF DEB 1601078) to JLK.