RESEARCH ARTICLE
Colonization, Baker’s law, and the evolution of gynodioecy in Hawaii: implications from a study of Lycium carolinianum Jill S. Miller1,3
, Caitlin M. Blank1,2, and Rachel A. Levin1
Manuscript received 8 December 2018; revision accepted 6 March 2019. 1 Department of Biology, Amherst College, Amherst, Massachusetts 01002, USA 2 Department of Biology, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA 3
Author for correspondence (e-mail:
[email protected])
Citation: Miller, J. S., C. M. Blank, and R. A. Levin. 2019. Colonization, Baker’s law, and the evolution of gynodioecy in Hawaii: implications from a study of Lycium carolinianum. American Journal of Botany 106(5): 733–743. doi:10.1002/ajb2.1279
PREMISE: As Baker’s law suggests, the successful colonization of oceanic islands is often associated with uniparental reproduction (self-fertility), but the high incidence of dimorphism (dioecy, gynodioecy) on islands complicates this idea. Lycium carolinianum is widespread, occurring on the North American mainland and the Hawaiian Islands. We examined Baker’s ideas for mainland and island populations of L. carolinianum and examined inbreeding depression as a possible contributor to the evolution of gynodioecy on Maui. METHODS: Controlled crosses were conducted in two mainland populations and two populations in Hawaii. Treatments included self and cross pollination, unmanipulated controls, and autogamy/agamospermy. Alleles from the self-incompatibility S-RNase gene were isolated and compared between mainland and island populations. Given self-compatibility in Hawaii, we germinated seeds from self-and cross-treatments and estimated inbreeding depression using seven traits and a measure of cumulative fitness. RESULTS: Mainland populations of Lycium carolinianum are predominately self- incompatible with some polymorphism for self-fertility, whereas Hawaiian populations are self-compatible. Concordantly, S-RNase allelic diversity is reduced in Hawaii compared to the mainland. Hawaiian populations also exhibit significant inbreeding depression. CONCLUSIONS: Self-compatibility in Hawaii and individual variation in self-fertility in mainland populations suggests that a colonization filter promoting uniparental reproduction may be acting in this system. Comparison of S-RNase variation suggests a collapse of allelic diversity and heterozygosity at the S-RNase locus in Hawaii, which likely contributed to mate limitation upon arrival to the Pacific. Inbreeding depression coupled with autonomous self-fertilization may have led to the evolution of gynodioecy on Maui. KEY WORDS Baker’s law, gametophytic self-incompatibility; gynodioecy; Hawaii, inbreeding depression, Lycium; Lycium carolinianum; male-sterility; self-compatibility; Solanaceae.
Although present in only 5-6% of angiosperm species (Renner and Ricklefs, 1995; Charlesworth, 2002), the evolution of separate sexes is a recurring transition in angiosperms (Geber et al., 1999; Charlesworth, 2002; Renner, 2014). The occurrence of two mating types in populations, for example, females and males in dioecious populations or females and hermaphrodites in gynodioecious populations, has direct effects on patterns of mating as well as important emergent consequences such as the evolution of sexual dimorphism (Barrett and Hough, 2013; Kamath et al., 2017) and sex ratio evolution (McCauley and Taylor, 1997; Van Etten and Chang, 2014; Rivkin et al., 2015). A common pathway to separate sexes is the origin and spread of male-sterility in otherwise hermaphroditic
populations (i.e., gynodioecy; Webb, 1999; Spigler and Ashman, 2011). Understanding the pathways to dimorphic sexual strategies and the selective forces responsible for the spread of unisexual plants in cosexual populations has been and remains an important focus in ecology and evolutionary biology (Lewis, 1942; Lloyd, 1975, 1982; Charlesworth and Charlesworth, 1978; Thomson and Barrett, 1981; Geber et al., 1999; Barrett, 2002). Mating systems (selfing, outcrossing, and mixed-mating) are intimately tied to the evolution of sexual strategies. For example, selection for outcrossing to avoid inbreeding depression has been documented as a driver of the evolution of gender dimorphism (Thomson and Barrett, 1981; Sakai et al., 1997, 2006; Vaughton and
American Journal of Botany 106(5): 733–743, 2019; http://www.wileyonlinelibrary.com/journal/AJB © 2019 Botanical Society of America • 733
734 • American Journal of Botany
Ramsey, 2003; Ramsey et al., 2006). Under this scenario, the expectation is that hermaphrodites possess the ability to self-fertilize and do so with some frequency. Likewise, inbreeding depression is expected to be substantial and, coupled with self-fertilization, facilitates the retention and spread of male-sterile plants (hereafter, females) in populations (Charlesworth and Charlesworth, 1978; Lloyd, 1982; Charlesworth, 1999). Theoretical work indicates that under nuclear control of male sterility, females persist and can spread in populations if they have a two-fold or greater fitness advantage as compared to hermaphrodites (Lloyd, 1975; Charlesworth and Charlesworth, 1978). The necessary fitness advantage for females can also include enhanced seed fertility relative to hermaphrodites, and a combination of greater seed production in females can work in conjunction with selfing and inbreeding depression in hermaphrodites to stabilize gynodioecy (Sakai et al., 1997). Finally, the underlying genetic control of sexual expression also affects the conditions under which females can be maintained in populations (Charlesworth, 1981). The evolution of sexual strategies and mating systems intersects with an idea put forth by Baker (1955) over 60 years ago. Namely, that successful establishment following long- distance dispersal is likely associated with uniparental reproduction. Baker proposed that species that are self-compatible (or capable of asexual reproduction) are more likely to be successful colonizers than those that are obligate outcrossers and rely on pollen transfer between plants (i.e., self-incompatible species). In support of Baker’s contention, a higher frequency of self-compatibility, as opposed to self-incompatibility, has been documented on island floras (McMullen, 1987, 1990; Webb and Kelly, 1993; Anderson et al., 2001; Chamorro et al., 2012; reviewed recently by Grossenbacher et al., 2017), and empirical work has continued to emphasize the importance of Baker’s rule in both plants and animals (Schueller, 2004; Busch, 2005; Trouve et al., 2005; Bramow et al., 2013). However, the generality of Baker’s ideas has also been questioned, especially given observations that complicate his assertion. For example, Carlquist (1966) initially took issue with Baker (1955) based primarily on the high incidence of dioecy on islands, especially Hawaii. More recently, Sakai et al. (1995a) documented the importance of both colonization by dimorphic ancestors and the evolution of dimorphism in situ for the Hawaiian flora, calling for more detailed study of individual lineages. Models of dispersal and mating system likewise have contradicted Baker’s ideas by predicting associations between dispersal and outcrossing, as opposed to selfing (Cheptou and Massol, 2009; Massol and Cheptou, 2011; reviewed in Busch, 2011 and Pannell, 2015). In their recent review of Baker’s law, Pannell et al. (2015) identify several contexts under which Baker’s ideas could be investigated, including a comparison of the capacity for self-fertilization on oceanic islands vs. mainland populations. Likewise, the authors suggest that robust tests of Baker’s law require determination of the ancestral sexual strategies of colonists; that is, does dimorphism evolve in situ or were the initial colonists dimorphic? Species and genera with variation in sexual strategies provide useful natural systems to investigate both the pattern and the underlying selective forces relevant in mating and sexual system evolution. The genus Lycium is one of the largest in the tomato family (Solanaceae), with a cosmopolitan distribution and approximately 90 species (Levin et al., 2011). Although the majority of Lycium species are hermaphroditic, gender dimorphism has been documented in a dozen species to date (Minne et al., 1994; Miller and Venable,
2000; Venter and Venter, 2003a, b; Venter, 2007; Blank et al., 2014; Levin et al., 2015) despite the presence of ancestral gametophytic self-incompatibility (Richman, 2000; Miller et al., 2008). One species, L. carolinianum, has an especially wide-ranging distribution and is found along the Gulf of Mexico in North America and on several islands in the Pacific, including the Hawaiian Islands. Extending Baker’s logic to the distribution of L. carolinianum, it is possible that its establishment in Hawaii was facilitated by the capacity for uniparental reproduction in the original colonists (e.g., the loss of ancestral self-incompatibility). Demonstration of self- incompatibility in mainland populations and self- compatibility in island populations would be consistent with Baker’s assertion. However, exceptions to Baker’s rule have been documented (Carr et al., 1986; Sun and Ritland, 1998). In particular, Miller et al. (2008) reported the retention of self-incompatibility, despite a severe bottleneck in the number of incompatibility alleles, following a long- distance, oceanic dispersal of Lycium from the Americas to Africa. By contrast, the geographic distribution of L. carolinianum provides the opportunity for a comparative study of mating systems in mainland and island populations of the same species. In addition, L. carolinianum is one of two species in the genus with documented among-population variation in sexual strategy (Blank et al., 2014; see also Miller et al., 2016; Kamath et al., 2017). Specifically, although Blank et al. (2014) reported that most populations consisted of hermaphroditic individuals in monomorphic populations, they identified male-sterility and gynodioecy in populations from the island of Maui, Hawaii and in the Yucatán peninsula, Mexico. The presence of both cosexual (hermaphroditic) and dimorphic (gynodioecious) populations of L. carolinianum thus offers a further opportunity to study the potential importance of self-fertility and inbreeding depression as selective forces relevant to the evolution of separate sexes. Here, we use controlled pollinations to investigate the capacity for self-fertilization in four populations of Lycium carolinianum, including two mainland North American populations and two populations (on two islands) in Hawaii. In addition, we complement our field experiments with genotyping at the S-RNase mating system locus in these same populations. The S-RNase locus is involved in gametophytically controlled self-incompatibility in Lycium (Savage and Miller, 2006; Miller et al., 2011) and other Solanaceae (Richman et al., 1996; Stone and Pierce, 2005; Igic et al., 2007), and these data can be used in conjunction with crossing studies to infer functional self-incompatibility using analyses of allelic variation and molecular evolution. Given that populations in Hawaii are self-compatible (this study) and that some populations in Hawaii have also evolved gynodioecy (Blank et al., 2014), we evaluate the hypothesis that inbreeding depression contributes to the maintenance of females in gynodioecious populations in Hawaii. METHODS Study species
Lycium carolinianum Walter occurs along coastal areas in southeastern and southern North America, ranging from coastal South Carolina to Florida and along the Gulf Coast region from Florida to Texas and south to the Yucatán peninsula in Mexico. Remarkably, L. carolinianum also occurs over 6000 kilometers away on several Pacific islands including Easter Island, the Hawaiian Islands, and
the Ogasawara Islands. Phylogenetic studies to date indicate the monophyly of L. carolinianum and suggest long-distance dispersal from North America (Levin and Miller, 2005; Levin et al., 2007). Plants are salt-tolerant woody shrubs ranging in height from 0.5 to 2 m and occur in sub-saline or rocky coastal habitats. Flowers range in color from white to purple and are insect-pollinated. Birds are known to consume and disperse the fleshy fruits of Lycium (Poulin et al., 1994), and the fleshy, red berries of L. carolinianum are no exception (Butzler and Davis, 2006). Species descriptions of Lycium carolinianum (including L. sandwicense A. Gray; Levin and Miller, 2005) indicate that populations are hermaphroditic (Hitchcock, 1932; Chiang-Cabrera, 1981; Wagner et al., 1999), as does a survey of this species in the Ogasawara islands (Abe, 2006). More recently, Blank et al. (2014) characterized sexual expression in 17 populations across the species range. Although the majority of populations were hermaphroditic (five North American populations in Texas and Florida and eight populations from four islands in Hawaii), gender dimorphism was documented in several populations on Maui, Hawaii, and in the Yucatán peninsula, Mexico. Specifically, in dimorphic populations, all plants had flowers with stigmas, styles, and ovaries, but some plants had abortive anthers lacking pollen (females), whereas other plants had fully functional anthers (hermaphrodites; Blank et al., 2014). Controlled pollinations and the capacity for self-fertilization
We conducted controlled pollinations to determine the compatibility status of populations of Lycium carolinianum from both the North American mainland and Hawaiian Islands. Pollinations took place in three consecutive years in four populations: Hanapepe, on the island of Kauai, Hawaii (21.895, −159.599) in December 2010, St. Marks National Wildlife Refuge, Florida (30.074, −84.179) in October 2011, Hawea and Nakalele Points, Maui (21.004, −156.665 and 21.030, −156.596, respectively) in November 2011, and Brazoria National Wildlife Refuge, Texas (29.061, −95.242) in October 2012. We placed fine mesh bags over floral buds to prevent insect visitation and potential contamination of pollination treatments. Buds were monitored daily, emasculated prior to anther dehiscence to prevent self-pollination, and re-covered until floral anthesis. During the next several days, open flowers in the bags received one of two pollination treatments: cross-pollination or self-pollination. The cross-pollination treatment included pollen collected from multiple (typically greater than five) individuals, whereas the self-pollination treatment included pollen from the maternal parent. Both treatments included sufficient quantities of pollen to permit maximal seed set, and flowers remained covered until fruit collection. On each plant, we also marked newly opened, unmanipulated, and uncovered flowers as controls to assess natural levels of fruit and seed production. In the Hawaiian populations, we included an autogamy/agamospermy treatment in which flower buds were marked and covered with fine mesh bags to preclude external pollination for the duration of their development. As the population on Maui was gynodioecious, the self-pollination treatment was not possible on female plants. Fruit set was monitored for all experimental flowers, and, 4 to 6 weeks following pollination, mature fruits were collected and seed number determined. Sixty-two experimental plants were used including 17 and 19 in mainland populations (Texas and Florida, respectively), and five and 21 (seven females and 14 hermaphrodites) in the Hawaiian populations (Kauai and
May 2019, Volume 106 • Miller et al.—Evolution of gynodioecy • 735
Maui, respectively). Pollinations included 570 and 546 flowers for the cross-and self-pollination treatments, respectively, and 831 flowers were marked as controls. In the Hawaiian populations, 203 buds on 26 plants were used to assess autogamy (in hermaphrodites) or agamospermy (in females). Seed number was counted in a total of 1136 fruits, including 451 and 178 for the cross and self treatments, respectively, and 436 control fruits. Seeds from 71 fruits were counted to quantify seed production in the autogamy/agamospermy treatment. We analyzed fruit production for each population using generalized linear models assuming a binomial distribution and a logit link function; models included an overdispersion parameter and Firth bias-adjusted estimates as implemented in JMP, version 10.0.2 (JMP, 1989-2019). The dependent variable was the number of fruits produced following a particular pollination treatment divided by the total number of flowers in that treatment on a given plant. Effects included in the model included plant and pollination treatment. In the dimorphic Maui population, hermaphrodites and females were analyzed separately. Planned contrasts between treatments assessed the compatibility of plants within populations (cross-vs. self-pollination), pollen limitation (hand-pollination vs. control), and the ability to set fruit autonomously (control vs. autogamy [hermaphrodites] or agamospermy [females]). Seed number per fruit was analyzed using the REML method in separate general linear mixed-models for each population. Models included the fixed effect of pollination treatment, the random effect of plant and their two-way interaction. Tukey’s honestly significant difference (HSD) tests (as implemented in JMP) were used to assess differences among treatments. Genotyping and molecular evolution of the S-RNase mating system gene
We genotyped 58 individuals at the S-RNase mating system locus in the same populations as those included in the controlled crossing experiment (30 individuals from Texas and Florida; 28 individuals from Kauai and Maui, Hawaii). Genotypes were determined following Miller et al. (2011) using either RNA extractions of stylar material (RNeasy Plant Mini kit, Qiagen, Inc., Hilden, Germany) followed by cDNA synthesis (First Strand cDNA Synthesis kit, EMD Millipore, Inc., Burlington, Massachusetts, USA), PCR using degenerate primers (designed from conserved regions in Solanum, Nicotiana, and Petunia; Richman et al., 1995) to target S-RNases, and cloning (AccepTor Vector Giga kit, EMD Millipore, Inc.), or PCR screening of genomic DNA extracted from leaf tissue (DNeasy Plant Mini kit, Qiagen, Inc.) using primers designed to amplify S-RNase alleles in different lineages. Allele identity was confirmed for all individuals by sequencing (Retrogen, Inc., San Diego, California, USA). All S-RNase sequences were processed and aligned in Mesquite version 3.2 (Maddison and Maddison, 2014) and Sequencher version 4.8 (Gene Codes Corp., Ann Arbor, Michigan, USA, 1991-2007) to determine allele identities. Once alleles were designated, individual plants were assigned genotypes. Previous studies in self-incompatible populations have demonstrated that models incorporating positive selection fit S-RNase allelic data significantly better than corresponding models that do not incorporate positive selection (Savage and Miller, 2006; Igic et al., 2007; Miller et al., 2008; Miller and Kostyun, 2011). Following these studies, we used the codeml package in PAML (Yang, 2007) as implemented in PAMLX (Xu and Yang, 2013) to
736 • American Journal of Botany
TABLE 1. Analysis of inbreeding depression for seven traits and cumulative fitness for self-compatible populations of Lycium carolinianum in Hawaii. General linear mixed-models included the fixed effects of cross type (self-or cross-pollination), population (Kauai or Maui), and their interaction. The percent of total variation explained by random effects is also given. Traits with two or more asterisks remained significant following correction for multiple tests. Source of variation
Percent of total variation
Dependent
Cross type
Population
Cross type × Population
Seed number
F1,28.7 = 0.47 P = 0.5001 F1,17 = 7.8 P = 0.0125* F1,20.6 = 19.37 P = 0.0003*** F1,18.1 = 10.5 P = 0.0045** F1,17.7 = 9.13 P = 0.0075* F1,15 = 0.84 P = 0.3743 F1,16.4 = 18.56 P = 0.0005*** F1,17 = 13.03 P = 0.0022**
F1,17.4 = 1.24 P = 0.2806 F1,17 = 2.45 P = 0.136 F1,23 = 2.74 P = 0.1114 F1,21.1 = 0.004 P = 0.9528 F1,18.4 = 0.29 P = 0.598 F1,15 = 11.8 P = 0.0037** F1,17 = 1.43 P = 0.2474 F1,17 = 4.32 P = 0.053
F1,28.7 = 0.07 P = 0.7882 F1,17 = 0.06 P = 0.811 F1,20.6 = 4.59 P = 0.0442* F1,18.1 = 1.29 P = 0.2711 F1,17.7 = 0.5 P = 0.4871 F1,15 = 0.68 P = 0.4234 F1,16.4 = 2.26 P = 0.1519 F1,17 = 0.3 P = 0.5902
Percent germination Days to leaf emergence Plant height, 30 days Change height, 30-60 days Percent survival Biomass Cumulative fitness
assess whether the pattern of selection on S-RNase was consistent with results from our controlled crosses and previous studies documenting self-incompatibility. Specifically, PAML uses likelihood ratio tests to evaluate models that incorporate positive selection (M2 and M8) versus models without this parameter (M1 and M7). Given that S-RNase alleles in self-incompatible populations are subject to negative frequency-dependent selection, the expectation is that models including positive selection will fit S-RNase allelic data from self-incompatible populations significantly better than models without positive selection. Where models incorporating positive selection fit the data significantly better, we estimated the non-synonymous to synonymous rate ratio (ω = dN/dS), and empirical Bayesian probabilities were used to determine the number and identity of sites under positive selection. By contrast, following the loss of self-incompatibility, S-RNase allelic diversity is expected to degrade (Igic et al., 2004, 2008). Thus, we compared mainland (self-incompatible, see Results) and island (self-compatible, see Results) populations for both the total number of alleles recovered and the presence of unique S-RNase genotypes. Inbreeding depression
In the self-compatible Hawaiian populations (Kauai and Maui; see Results), we compared the fitness of offspring (produced by hermaphrodites) following either cross or self-pollination and assessed inbreeding depression for seven traits (Table 1) and a measure of cumulative fitness. Seed number was counted from 283 fruits (130 and 153 in the cross-and self-treatments, respectively) from 19 individuals on Kauai and Maui (x̅ = 7.5 fruits per plant, range = 3 to 14). On 17 July 2013, 646 seeds from these 19 individuals (19 individuals × 2 treatments × 17 seeds/treatment) were planted in 107 mL Ray Leach Cone-tainers™ (Stuewe & Sons, Tangent, Oregon, USA) and arranged in 98-position racks in the greenhouse. Cones were positioned randomly within racks and racks were rotated weekly to randomize rack position under the misting system. Two weeks after the first seed germinated, racks were removed from the misting system and watered once
Family
Family × Cross type
Residual
55.1
1.2
43.7
79.6
–
20.4
0.0
5.6
94.4
0.0
14.2
85.8
0.0
9.0
91.0
25.1
–
74.9
2.5
15.2
82.3
19.5
–
80.5
daily. We monitored seeds and seedlings daily over four months and recorded the dates of seed germination, emergence of the first true leaf, and seedling death. Plant height (cm) was measured at 30 and 60 days post-germination. For a random subset of individuals (n = 191 plants, x̅ = 4.8 plants per parent per treatment), above ground biomass was determined by destructive sampling 120 days post germination; samples were dried at 60°C for 7 days and weighed to the nearest milligram (APX-200, Denver Instruments, Bohemia, New York, USA). Cumulative fitness was calculated using average values for each maternal family and cross type (cross-or self-pollination) as seed number × percent germination × inverse of days to leaf emergence × growth rate (estimated as the change in height between 30 and 60 days) × percent survival × plant biomass. We used general linear mixed-models in JMP to analyze the effects of cross type (cross-or self-pollination), population (Kauai or Maui), and maternal family on individual fitness measures (seven traits) and cumulative fitness. Fixed effects included cross type (cross vs. self), population, and the population by cross-type interaction. Random effects included maternal family and the family by cross-type interaction. However, for percent germination, percent survival, and cumulative fitness, there was no replication (i.e., offspring) within parental family, as these traits were determined as the fraction of seeds germinating (or seedlings surviving) or using means (cumulative fitness, see above). Thus, the family by cross-type interaction term was not included. Natural log (plant height, change in plant height, and biomass), square root (seed number, days to leaf emergence), and arcsine-square root (percent germination, percent survival) transformations were implemented to meet assumptions of models. Bonferroni corrections were used to control for multiple analyses. For individual traits, a significant main effect of cross type, coupled with crossed offspring outperforming selfed offspring, was taken as evidence of inbreeding depression. We expressed inbreeding depression (δ) for each trait and cumulative fitness as 1 − (WSELF ∕WCROSS ) using back-transformed means for the cross-and self-treatments. However, for the number of days to leaf emergence, lower values
May 2019, Volume 106 • Miller et al.—Evolution of gynodioecy • 737
TABLE 2. Likelihood Ratio Chi-Square values, degrees of freedom (in parentheses), and significance levels for main effects and contrasts in generalized linear models for fruit production. Effect Population Texas Florida Kauai Maui
Contrast
Mating type
Plant
Treatment
Cross vs. Self
Control vs. Cross
Autogamy vs. Control
H H H H F
24.3 (16), ns 45.5 (18)** 2.9 (4), ns 44.0 (13)*** 0.02 (6), ns
156.4 (2)*** 97.2 (2)*** 50.7 (3)*** 24.2 (3)*** 63.4 (2)***
156.4 (1)*** 91.7 (1)*** −0.10 (1), ns 0.01 (1), ns –
41.8 (1)*** 10.3 (1)** 5.7 (1)* 0.13 (1), ns 41.2 (1)***
– – 21.2 (1)*** 19.3 (1)*** 2.9 (1), ns
Notes: ***P < 0.0001, **P < 0.005, *P < 0.05, ns = not significant.
represented faster development; thus, inbreeding depression was calculated as 1 − (WCROSS ∕WSELF ). RESULTS Controlled pollinations
The main effect of pollination treatment significantly affected fruit production in all populations (Table 2), although populations differed in their capacities for self-fertilization and, by extension, their mating systems. Specifically, mainland populations from North America (Texas and Florida) were predominately self- incompatible. Fruit production was significantly higher in the cross-as compared to the self-pollination treatment in both mainland sites (cross vs. self contrast in Texas and Florida: Table 2, Fig. 1A). Cross-pollination resulted in higher fruit production than did self-pollination for all plants in Texas and all but a single individual from Florida. However, fruit set following self-pollination was non-zero in 28% of plants overall and ranged from 8 to 27% in Texas and 6 to 91% in Florida (Fig. 1C). In contrast to the mainland, populations in Hawaii (Kauai and Maui) were self-compatible, with no difference between the cross-and self-pollination treatments (cross-vs. self-contrast: Table 2, Fig. 1A). The cross-pollination treatment significantly increased fruit production, as compared to controls, in all cosexual populations (Texas, Florida, and Kauai). In the gynodioecious population on Maui, the pattern differed for females and hermaphrodites. Specifically, cross-pollination increased fruit production in females, but had no effect on fruit production in hermaphrodites (Table 2, Fig. 1A). Autogamous fruit production by hermaphrodites in the Hawaiian populations was significantly lower than in the controls, but the magnitude of autogamy differed between the populations (0.03 on Kauai and 0.63 on Maui; Fig. 1A). Females did not produce fruit in the absence of pollination. Results for seed number per fruit (Fig. 1B) largely mirror those for fruit production. Post hoc Tukey’s tests indicated significantly higher seed production in the cross-compared to self-pollination treatment (91.4 vs. 21.8 seeds) in the Texas population and marginally so in the Florida population (21.3 vs. 15.7 seeds; Fig. 1B). Although nearly a third of individuals were capable of producing some seeds following selfing, all plants in Texas and all but two plants from Florida produced more seeds following the cross-pollination treatment (Fig. 1C). One individual in the Florida population, in particular, was capable of high levels of self-fertilization and had high fruit production with many seeds following self-pollination (Fig. 1C). In contrast to the mainland populations, there were no differences in seed set between the cross-and self-treatments or the cross and control treatments in either island population (Fig. 1B).
Genotyping and molecular evolution of the S-RNase mating system gene
We isolated a total of 23 unique S-RNase sequences from 58 individuals in four populations (Genbank accession numbers: MK573176 to MK573198). The partial coding sequences ranged in length from 99 to 170 amino acids and were on average 51% divergent at the amino acid level. The vast majority of allelic diversity was recovered from populations on the mainland (Texas and Florida contained 20 alleles isolated from 30 plants), whereas only three sequences were isolated from 28 individuals in Hawaii (Table 3). Further, two of the three sequences recovered in Hawaii (CARO16, MK573197 and CARO25, MK573198) were remarkably similar, differing at only a single synonymous nucleotide position, which we confirmed across many individuals (Appendix S1). We recovered two unique heterozygous genotypes in Hawaii (Table 3; Appendix S1) and were able to isolate only a single allele in four plants. In contrast, eighteen unique, heterozygous genotypes were recovered on the mainland (Table 3); in eleven plants we isolated only one allele, which we interpret as a failure of degenerate primers to amplify highly divergent alleles (Appendix S1). In mainland populations, site-specific models incorporating positive selection fit the S-RNase sequence data significantly better than corresponding models lacking positive selection (M2 vs. M1: LRT = 18.45, P < 0.0001 and M8 vs. M7: LRT = 21.69, P < 0.0001; Appendix S2), and in the M8 model the majority of positively selected sites (seven of 12 positions with Bayes empirical Bayes probabilities ≥ 0.90) were in hypervariable regions associated with allelic specificity (Ioerger et al., 1991; Ida et al., 2001). Inbreeding depression
In the self-compatible island populations, cross type (cross-or self-pollination) did not influence either seed number or seedling survival (Table 1). By contrast, we detected a main effect of cross type (i.e., inbreeding depression) for five traits. However, following correction for multiple tests only three traits remained statistically significant, including the number of days to leaf emergence, plant height at 30 days, and plant biomass (Table 1). On average, seedlings following from the cross-treatment produced leaves more quickly (13.1 vs. 16.3 days for cross-and self-treatments, respectively) and were significantly taller at 30 days (1.9 vs. 1.3 cm) and heavier at the end of the experiment (0.45 vs. 0.25 g) than seedlings from the self-treatment. Estimates of inbreeding depression for these three traits were 0.194, 0.302, and 0.438, respectively (Fig. 2). Inbreeding depression was also detected for a measure of cumulative fitness (δ = 0.672; Table 1, Fig. 2). There were few differences between populations except for seedling survival, which was lower on Kauai as compared to Maui (F1,15 = 11.8, P = 0.0037; Table 1). Populations
738 • American Journal of Botany
Self
***
Fruit production per flower
A 1.0
Control
Cross
***
***
Autogamy (H), Agamospermy (F)
ns
***
ns
***
***
ns
***
***
0.8 0.6 0.4 0.2 0.0
Texas
B
Florida
***
Seed number per fruit
ns ns
