In several cases, estimates of gene flow between species appear to be higher than we might predict given the strength of interspecific barriers separating these species pairs. However, as far as we are aware, detailed measurements of reproductive isolation have not previously been compared with a coalescent-based assessment of gene flow. Here, we contrast these two measures in two species of sunflower, Helianthus annuus and H. petiolaris. We quantified the total reproductive barrier strength between these species by compounding the contributions of the following prezygotic and postzygotic barriers: ecogeographic isolation, reproductive asynchrony, niche differentiation, pollen competition, hybrid seed formation, hybrid seed germination, hybrid fertility, and extrinsic postzygotic isolation. From this estimate, we calculated the probability that a reproductively successful hybrid is produced: estimates of Phyb range from 10−4 to 10−6 depending on the direction of the cross and the degree of independence among reproductive barriers. We then compared this probability with population genetic estimates of the per generation migration rate (m). We showed that the relatively high levels of gene flow estimated between these sunflower species (Nem= 0.34–0.76) are mainly due to their large effective population sizes (Ne > 106). The interspecific migration rate (m) is very small (<10−7) and an order of magnitude lower than that expected based on our reproductive barrier strength estimates. Thus, even high levels of reproductive isolation (>0.999) may produce genomic mosaics.
Species evolve independently of their close relatives as a consequence of multiple reproductive barriers between populations. These barriers, acting before or after fertilization, determine the likelihood of gene exchange between diverging taxa. As a result, the combined action of these multiple barriers is expected to reduce gene flow between species (reviewed in Coyne and Orr 2004). Yet, in several cases, estimates of gene flow (Nm) are higher than we might expect from the interspecific barriers separating these species pairs (Scotti-Santaigne et al 2004; Machado et al. 2007; Yatabe et al. 2007; Cooper et al. 2010).
Genomic scans present a different picture of species evolution. In many cases, levels of genetic differentiation between two species vary across the genome, with some areas showing high levels of differentiation and others very little. This seems to be the case in sympatric oak species (Scotti-Saintagne et al. 2004; Gömöry and Schmidtová 2007), Populus species (Populus fremontii and P.angustifolia; Martinsen et al. 2001), the American irises Iris fulva and I. brevicaulis (Martin et al. 2005, 2006), Drosophila pseudoobscura and D. persimilis (Machado et al. 2007), Heliconius spp. (Bull et al. 2006), and the Hawaiian silverswords Dubautia arborea and D. cilioata (Lawton-Rauh et al. 2007). Areas of high genetic differentiation have been associated with areas of low recombination such as chromosomal breakpoints (e.g., Yatabe et al. 2007) and hypothesized regions under selection (Scascitelli et al. 2010). In the case of Drosophila, genome porosity seems to exist despite evidence for strong reproductive isolation (Dobzhansky 1973; Machado et al. 2007). Reconciling the evidence for reproductive isolation with observed levels of introgression requires precise estimates of both factors and as far as we are aware, these full data have not been reported for any species pair. Strong reproductive isolation should result in very low per generation migration rates (m) between two species. However, differentiation between two species depends on the levels of ancestral polymorphism, the time since divergence, and the overall level of gene flow, Nem. Given sufficient time for drift to eliminate ancestral polymorphism, two species that are completely isolated will become differentiated. However, if the effective population sizes are sufficiently large, even small rates of migration can be sufficient to eliminate differentiation at neutral loci between two species.
Helianthus petiolaris and H. annuus are related North American sunflower species that diverged ∼1.8 million years ago (reestimation of Strasburg and Rieseberg 2008, see results section). Both occur from the southern border of Canada to the north of Mexico (see Fig. 1). Both species have probably experienced recent range expansions due to human disturbance, particularly on the eastern and western coasts. However, both were likely widely distributed across the plains and Southwest prior to human settlement of North America, and these areas remain the sites of highest density, including in areas not subject to human disturbance: hybrid species derived from these two parentals and endemic to the Southwest are estimated to have diverged at least 100,000 generations ago (Schwarzbach and Rieseberg 2002). They occupy different habitats: Helianthus petiolaris is found on sandy xeric soils, whereas H. annuus grows in mesic clay-based soils (Heiser 1947). Despite these ecological differences, H. annuus and H. petiolaris co-occur across much of central North America, and multiple hybrid zones occur (Rieseberg et al. 1998). Data suggest that introgression between H. petiolaris and H. annuus has been common since the split of these species, with long-term estimates of Nem of approximately 0.5 (Strasburg and Rieseberg 2008). Genomic differentiation is limited to loci under divergent selection, as well as regions near the breakpoints of chromosomal rearrangements that separate the two species, where most sterility factors are found (Lai et al. 2005; Yatabe et al. 2007; Strasburg et al. 2009).
Despite the evidence of substantial introgression and overall low genomic differentiation, numerous strong reproductive isolating barriers exist between H. petiolaris and H. annuus, including pollen competition, hybrid pollen inviability, and reduced hybrid seed set (Rieseberg et al. 1995a; Rieseberg 2000). Lowry et al. (2008a) estimated that the cumulative effect of these barriers would result in complete reproductive isolation, in sharp contrast to the genomic data.
Here, we quantify the cumulative effects of reproductive barriers between H. annuus and H. petiolaris drawing upon previously published data by our group, the results of a field reciprocal transplant experiment and a seed germination experiment, and the estimation of ecogeographic and reproductive asynchrony barriers using public datasets. We calculate the strength of the total interspecific reproductive barrier and the probability of hybridization, which we then compare to revised coalescent-based population genetic estimates of interspecific gene flow.
Several reproductive barriers separate H. annuus and H. petiolaris. To quantify their compound effects on total reproductive isolation, we performed a detailed analysis of each barrier individually and in combination. We classified these barriers into pre- or postzygotic.
BARRIER STRENGTH BETWEEN H. PETIOLARIS AND H. ANNUUS
Helianthus annuus and H. petiolaris are known to use different ecological niches. However, their typical habitats often occur in the same geographical region. As a result, both species are in parapatry across much of North America (Fig. 1). To estimate ecogeographic isolation, we used locality information from four state databases that span the zone of highest population density: the Bessey Herbarium at University of Nebraska (NEB), the New Mexico Biodiversity Collections Consortium (NMBCC 2009), the University of Arizona herbarium (ARIZ), and the Wisconsin state herbarium (WBIS). Latitude and longitude were estimated when necessary using other locality information and Google Earth (earth.google.com). Public land survey coordinates were converted to latitude and longitude using a web utility (http://www.earthpoint.us).
Following the methods of Kay (2006), we randomly placed 10 km by 10 km squares (sites) 1000 times for each state dataset. This spatial scale was chosen based upon estimates of pollinator travel distance and seed dispersal (Arias and Rieseberg 1994), and is likely conservative: smaller squares would result in higher estimates of reproductive isolation. For those squares that included at least one herbarium record, we recorded whether one species or both were included. We calculated the ecogeographic isolation for each species as follows:
State datasets were treated as independent: even specimens from Arizona and New Mexico typically had different collectors. States were used as replicates to estimate the mean and standard error of ecogeographic isolation.
Although H. petiolaris flowers earlier than H. annuus, the flowering distribution of both species shows a significant overlap (Heiser 1947). To estimate the contribution of reproductive asynchrony to prezygotic reproductive isolation, we gathered herbarium data from several U.S. herbaria: the four mentioned above, plus the Colorado State University Herbarium (CMML), the Illinois State Museum Herbarium (ISM), and the University of Kansas Herbarium (KANU). The phenology of H. annuus varies with geography due to differences in day-length sensitivity (Blackman et al. 2011), so we expected phenological isolation to differ across the range. From each herbarium, we obtained collection dates for all flowering specimens of H. annuus and H. petiolaris occurring within that state and then eliminated records collected by the same collector on the same day. We took each herbarium record to represent the midpoint of 21 days of flowering: a single head is receptive for five to seven days, but both species produce multiple heads over the course of the season. Shorter estimates of flowering period result in larger measures of reproductive asynchrony. We followed Martin and Willis (2007) in estimating the proportion of flowering for each species on any given day, and the proportion of plants in flower that were heterospecific. This necessarily but unrealistically assumes that all plants produce equal number of flower heads on each day, that all flower heads produce equal amounts of pollen, and that each flower head is equally likely to be visited by pollinators. We calculated the frequency of heterospecific pollinations by summing the proportion of sampled plants flowering on any given day multiplied by the proportion of heterospecific pollinations for the day. The expected proportion of heterospecific pollination was calculated as the proportion of herbarium records for the other species out of the total number of herbarium records. We then calculated reproductive isolation for each species as follows:
States were used as replicates for the calculation of mean reproductive isolation and standard error.
Selection against immigrants.
If the two species are adapted to different habitats, this should help to preserve distinct ranges and reduce gene flow. Local adaptation will lead seeds that disperse across species boundaries to have a lower fitness than the local individuals. A reciprocal transplant in mesic (the native habitat for H. annuus) and xeric habitats (native habitat of H. petiolaris) using field-collected seeds from Utah populations was performed in 2003 at two sites in Utah as part of a larger experiment (Donovan et al. 2010). Sixty achenes (seeds hereafter) of each of the two species were sown into each of 10 plots at each site. Seeds and seedlings were monitored weekly for germination and survival. Both H. annuus and H. petiolaris have dormant seeds. To avoid mistakenly scoring differences in dormancy as differences in fitness, germination rates for each habitat were calculated by dividing by the maximum germination observed for each species in any habitat in this experiment. A parallel experiment was conducted simultaneously with seedlings at the same two sites to examine survival and reproduction. Ten seedlings of each species were planted into six blocks, for a total of 60 seedlings per site. Seeds were nicked to break dormancy, allowed to germinate on moist filter paper in petri plates, and then grown in a greenhouse at the University of Georgia for five weeks before being transplanted in the field in Utah. Seed production was estimated from each surviving plant based on the number of flowering heads and the number of seeds produced per head. Because different plants were used to estimate differences in germination and reproduction, overall fitness (w) for each species in each habitat was estimated as the product of seed germination, seed survival to flowering, and seed output per flowering plant. We estimated reproductive isolation as RIimm= 1 − (wimmigrant/wnative) for each species. Due to fitness being the product of several factors estimated with different sample sizes, standard errors for this estimate were not calculated.
Heterospecific pollen will compete with conspecific pollen for ovule fertilization. If conspecific pollen shows an advantage in fertilizing ovules, then pollen competition contributes to interspecific reproductive isolation. Pollen competition between these species was estimated in a previous study (Rieseberg et al. 1995a). Briefly, pollen mixtures of H. annuus and H. petiolaris with the proportions of 1:9, 1:1, and 9:1 were used to pollinate flower heads of either species. These different pollen mixtures were tested in two different genotypes of each species. The paternity of the progeny was determined with isozyme markers. The pollen mixture proportions provide the expected proportion of hybrids (as a result of heterospecific matings) and parental genotypes (as a result of conspecific matings) produced in the F1 offspring in the absence of any reproductive barrier. With the expected heterospecific and conspecific matings, the reproductive isolation due to pollen competition was estimated with the formula proposed by Martin and Willis (2007):
In this case, reproductive isolation due to pollen competition is independent of the pollen ratio (P= 0.2; Rieseberg et al. 1995a). The mean reproductive isolation was calculated for each genotype, and the two genotypes were used to estimate the mean and standard errors of this partial barrier strength for each species.
Intrinsic postzygotic isolation
Hybrid seed formation.
Hybrid embryos may fail to develop into mature seeds. To measure hybrid seed formation as a reproductive barrier, we considered the percentage of filled achenes of two genotypes of each parental species pollinated with conspecific compared to heterospecific pollen from the data available in Rieseberg et al. (1995a). The contribution of hybrid seed formation to the interspecific reproductive barrier strength was calculated with the formula: 1 −(Achetero1/Acsp), where Ac is the percentage of filled achenes. Here too we estimated the mean and standard errors of hybrid seed formation as a reproductive barrier by fitting a linear model to the data with JMP (SAS 2000).
Hybrid seed germination.
Hybrid seeds may differ in germination rates from parental seeds due to intrinsic genetic mechanisms. To quantify how much the germination rate differs between parents (H. annuus and H. petiolaris) and F1s, we scarified 12 three-year-old nonvernalized seeds each from 10 different families of H. annuus, H. petiolaris, and the reciprocal F1s; F1a (with H. annuus cytoplasm), F1p (with H. petiolaris cytoplasm). Three days later, the seeds were transplanted to four trays filled with nursery pot mix. Each tray was considered a randomized block and divided into four plots randomly assigned to each cross. Four seeds per family were planted in each sub-block. Trays were transferred into a growth chamber (16 h of light per day at 22°C). The number of germinated seeds was counted after emergence on the eighth day following scarification. The hypothesis that germination rates differ between genotypes with block as a random effect was tested with a likelihood ratio test in JMP (SAS 2000). The mean and standard errors of germination rate were obtained by applying a generalized model with the R's glm function (without blocks) and subsequently using the function predict.glm (R Development Core Team 2008). Hybrid germination rate as a reproductive barrier with respect to each species (sp) was calculated with the formula: 1 − (gmeanF1/gsp), where g is the germination rate. The barrier strength mean and standard error was calculated and reported as for immigrant inviability.
Even after a hybrid plant matures, gene flow can be impeded if the hybrid has lower fertility than either parent. We can partition these barriers due to hybrid fertility into (1) F1 pollen viability measured as the percentage of stainable pollen of hybrids compared to parentals, and (2) F1 seed set measured as the percentage of flowers pollinated by a parental plant that produced seeds in hybrids compared to parentals pollinated by a conspecific. The means and standard errors of the pollen fertility were estimated from a dataset of 16 interspecific crosses, four crosses within H. annuus, and four crosses within H. petiolaris (Rieseberg 2000). All crosses were reciprocal and were carried out in greenhouses at Indiana University, Bloomington, Indiana. Relative F1 seed set was estimated by comparing the seed set for 12 intraspecific crosses to 30 crosses between F1 plants, and dividing by pollen viability for the F1 plants (Independent trials pollinating F1s with H. annuus pollen produced similar results). The contribution of these interspecific reproductive barriers to total reproductive isolation was calculated with the formula: 1 − (vmean F1/vmean parental), where v is the frequency of viable pollen or viable seeds. Four crosses were used to estimate mean and standard errors of reproductive isolation.
TOTAL INTERSPECIFIC BARRIER STRENGTH
The total barrier strength was estimated for each species by discounting, at each barrier, the likelihood that hybrids will not be produced, will be inviable or maladapted to local environmental conditions, or will have lower fertility than parental species, as appropriate: as in Coyne and Orr (2004). RIi can be the probability of co-occuring, the probability of pollination by a conspecific relative to the heterospecific, the fitness of the local species relative to immigrant individuals, or the fitness of the local species with relative to F1s. For example, the probability that H. annuus and H. petiolaris find themselves in the same site is obtained with the ecogeographic isolation estimates. Only a fraction of these individuals will have the opportunity to mate exclusively with conspecifics due to reproductive asynchrony and this must be also discounted. If the H. annuus fitness in a xeric habitat is lower than that of H. petiolaris, the barrier strength will be proportional to this fitness difference. Martin and Willis (2007) criticized this multiplicative approach to calculate total reproductive isolation between species because it implicitly assumes independence between barriers, which is not necessarily true. Gavrilets and Cruzan (1998) examined the relationship between two barriers in a two-locus model and derived a formula for the barrier strength that was more complex. Although the true relationship between the different reproductive barriers is undoubtedly more complex, we currently lack information to decide if and how much dependence there is between barriers. We follow previous publications (e.g., Lowry et al. 2008a) in calculating the total barrier strength. In particular, we acknowledge that niche differentiation leading to immigrant inviability is probably not fully independent of ecogeographic isolation (Martin and Willis 2007). We include this as a distinct barrier, but also calculate reproductive isolation without including immigrant inviability as a distinct contributor.
The probability of the successful production of a fertile F1 hybrid can be estimated by Phyb= 1 −RITotal. We estimate Phyb with and without immigrant inviability as distinct from ecogeographic isolation.
POPULATION GENETIC ANALYSES
To reconcile estimates of intra- and interspecific gene flow from published data with calculations of barrier strength (herein), we compared the probability of the successful production of a fertile F1 hybrid, Phyb, with the estimates of effective population size (Ne), the proportion of the gene pools that are shared each generation (m), and their product, the population migration rate (Strasburg and Rieseberg 2008). An accurate estimate of reproductive barriers should result in an estimate of Phyb, that is, approximately equal to m. Gene flow estimates based on genetic dissimilarities do not take into account ancestral lineage sorting as a source of genetic differentiation and can thus overestimate gene flow. However, Strasburg and Rieseberg (2008) used a coalescent approach applied to sequence data from 18 anonymous nuclear loci to obtain these estimates. These data were analyzed within an “isolation with migration” framework that makes it possible to distinguish ongoing gene flow that followed the initial divergence from incomplete lineage sorting (Hey and Nielsen 2004, 2007). Analyses were implemented in the computer program IM (Hey and Nielsen 2004), which allows estimation of divergence time, current and ancestral effective population sizes, and reciprocal rates of gene flow based on Markov chain Monte Carlo simulations of gene genealogies at each locus. The simulations model demographic expansions or contractions, but assume a constant long-term migration rate (i.e., the method does not distinguish between allopatry with secondary contact and a low rate of gene flow during the entire divergence). Recombination is assumed to be absent within loci, but loci are assumed to be unlinked.
We estimated these demographic parameters for our two species of interest, H. annuus and H. petiolaris, and for a third species, H. argophyllus. The latter is the sister species to H. annuus and so might be expected to have weaker barriers to hybridization and more shared ancestral polymorphism. However, H. argophyllus was likely allopatric with both H. petiolaris and H. annuus until recently and so experienced fewer opportunities for gene flow. All sampled populations are locally allopatric from the other two species.
The demographic data for H. annuus and H. petiolaris we present here differ from the data reported in Strasburg and Rieseberg (2008) in two ways. First, data from two additional H. annuus populations from southern Texas have been added to the previous populations from the Southwest (total of 12 H. annuus populations), resulting in a ∼30% increase in the estimate of its effective population sizes as well as marginal increases in the divergence time and gene flow estimates; full details of these new analyses will be presented elsewhere (J. L. Strasburg and L. H. Rieseberg, unpubl. data). Second, the substitution rate used in Strasburg and Rieseberg (2008), 10−8 substitutions/site/year, was incorrectly used as an overall substitution rate rather than a synonymous substitution rate. Based on the average ratio of overall to synonymous nucleotide diversity across six Helianthus species, this synonymous substitution rate corresponds to an overall rate of 6.1 × 10−9 substitutions/site/year.
Because these sunflowers are annuals, a one-year generation time has been used for demographic estimates. However, given the evidence of seed dormancy (Alexander and Schragg 2003), the average generation time may be somewhat longer. Thus, we have also performed our calculations using a two-year generation time, which results in a 50% reduction in current and ancestral effective population size estimates. Estimates of divergence time and gene flow rates are not affected. Because the mutation rate is scaled per year rather than per generation, and genetic diversity is a function of effective population size and the mutation rate, longer generation times imply a higher mutation rate per generation, and so a lower effective population size to obtain the same genetic diversity.
BARRIER STRENGTH BETWEEN H. ANNUUS AND H. PETIOLARIS
The two species were locally allopatric the majority of the time in all four states: H. annuus was found only with H. annuus within 10 km random squares from 76% to 86% of the time, whereas H. petiolaris was locally allopatric from 77% to 78% of the time. The mean reproductive isolation due to geographic isolation for H. annuus and H. petiolaris individuals occurring without the other species relative to the total geographic distribution of each species are 0.797 (±0.024 SE) and 0.777 (+0.004 SE), respectively (Table 1 and Fig. 2).
Table 1. Means and standard errors of each reproductive barrier measured in this study estimated for both Helianthus annuus and H. petiolaris. The cumulative reproductive barrier strength based on the individual reproductive barrier strength estimates (RI) is reported at each step. This table also reports the total barrier strengths and the probability that a viable and reproductively successful hybrid is formed (Phyb) for both H. annuus and H. petiolaris considering all barriers as independent or omitting immigrant inviability from the calculation.
RI Ann cumulative
RI Pet cumulative
* SE not calculated.
Total prezygotic barriers without immigrant inviability
Hybrid seed formation
F1 pollen viability
F1 seed set
Total barrier strength with immigrant inviability
Total barrier strength without immigrant inviability
Phyb with immigrant inviability
Phyb without immigrant inviability
Overall, the reproductive isolation due to flowering asynchrony was small and similar for the two species. Flowering time differences were estimated to increase the proportion of conspecific pollen received by H. annuus from the expected 0.51 (if there were no flowering time differences) to 0.57; for H. petiolaris the shift was from 0.49 to 0.55. For H. annuus, phenological isolation results in reproductive isolation of 0.216 (±0.027 SE; range 0.13–0.36 across the seven states) whereas for H. petiolaris the reproductive isolation is 0.229 (±0.033 SE; range 0.13–0.33 across the seven states).
The two species showed evidence of local adaptation. Germination for H. annuus was 40% in its native habitat compared to 30% in the H. petiolaris habitat. Helianthus petiolaris had similar, low germination rates in both habitats: 13% in H. annuus habitat, 14% in its own habitat. Survival was similar for both species (∼40% in H. annuus habitat, ∼90% in H. petiolaris), but strong habitat-dependent differences were observed in seed production. Helianthus annuus produced 2000 seeds per plant in its habitat, whereas H. petiolaris produced 1600 seeds. In the H. petiolaris habitat, H. annuus produced 2300 seeds per plant whereas H. petiolaris produced 2400. The reproductive isolation due to selection against immigrants was 0.244 for H. annuus and 0.318 for H. petiolaris. Uncertainties were not estimated for this parameter, but are likely quite large: the estimates may not be significantly different from zero.
A low percentage of hybrid production (<10%) was observed even when pollen mixtures contained 90% heterospecific pollen (Rieseberg et al. 1995a). This indicates that conspecific pollen is much more likely to fertilize ovules than heterospecific pollen in both species when both pollen types are deposited in the same stigma. The mean and standard error of the reproductive barrier strength due to pollen competition are 0.900 ± 0.038 for H. annuus and 0.998 ± 0.038 for H. petiolaris (Table 1 and Fig. 2). Pollen competition as a reproductive barrier is asymmetric and stronger in H. petiolaris than H. annuus (Rieseberg et al. 1995a).
Intrinsic postzygotic isolation
Hybrid seed formation.
The mean and standard errors of the reproductive barrier due to hybrid seed formation— measured as the percentage of filled achenes of interspecific crosses compared to intraspecific crosses—were: –0.022 ± 0.031 and –0.024 ± 0.031 for H. annuus and H. petiolaris, respectively (Table 1 and Fig. 2). Although the negative values can be interpreted as a slight advantage of heterospecific pollen in hybrid seed formation, the fact that standard errors encompass zero indicates that this barrier is unimportant to total reproductive isolation between the species.
Hybrid seed germination.
The probability that seeds will fail to germinate (here denoted as q), significantly differed among genotypic classes (P= 0.006). Helianthus annuus seeds had the lowest probability of failing (q = 0.05 ± 0.02) followed by the hybrid with H. annuus mother, F1a (q = 0.16 ± 0.03) and H. petiolaris (q = 0.17 ± 0.03). Seeds from the hybrid with H. petiolaris mother F1p had the highest probability of failing (q = 0.22 ± 0.4). The reproductive barrier strength due to germination rate of a species compared to the hybrid means measured by germination after transplanting in a growth chamber was 0.147 ± 0.002 and 0.028 ± 0.002 for H. annuus and H. petiolaris, respectively (Table 1 and Fig. 2).
One possible caveat with the seed germination rate data is that the differences observed may include effects of both seed viability and embryo dormancy. We attempted to minimize the effects of embryo dormancy by employing three-year-old seeds (dormancy is considerably reduced within 12–16 weeks of maturation) and through scarification. However, we cannot completely rule out an effect of embryo dormancy on our results. Because the effect of seed germination on the total reproductive barrier is negligible (see below), we have not attempted to further disentangle these effects.
F1 pollen viability and seed set.
The proportion of pollinated flowers that produced viable seeds within a parental species is high and very similar between species; mean and standard errors of 0.860 ± 0.02 and 0.868 ± 0.02 for H. annuus and H. petiolaris, respectively (Rieseberg 2000). Likewise, the proportion of viable pollen produced by plants derived from intraspecific crosses is also very high and similar for both species; mean and standard errors of 0.943 ± 0.01 and 0.949 ± 0.01 for H. annuus and H. petiolaris, respectively (Rieseberg 2000). A severe reduction in seed set and pollen viability was observed in both reciprocal interspecific crosses. The mean and standard error of pollen viability was 0.047 ± 0.003 and 0.048 ± 0.003 for F1a and F1p, respectively (Rieseberg 2000). The reproductive barrier strength due to pollen viability was 0.949 ± 0.004 and 0.949 ± 0.004 for H. annuus and H. petiolaris, respectively (Table 1 and Fig. 2). The mean and standard error of seed set for crosses between F1 plants was 0.008 ± 0.002 and 0.007 ± 0.002 for F1a and F1p, respectively (Rieseberg 2000). When corrected for pollen inviability, the reproductive barrier strength due to seed set in hybrid plants was 0.824 ± 0.02 and 0.824 ± 0.02 for H. annuus and H. petiolaris, respectively (Table 1 and Fig. 2).
The compounded effect of these reproductive barriers between H. annuus and H. petiolaris is very strong. The barrier strength estimates are 0.999906 for H. annuus and 0.999998 for H. petiolaris assuming independence between barriers, or 0.999875 for H. annuus and 0.999997 for H. petiolaris assuming immigrant inviability is a portion of ecogeographic isolation. Thus, the probability of successful production of a viable, fertile F1 individual is very small for both reciprocal crosses. If we assume that immigrant inviability and ecogeographic isolation are independent from each other, we estimate that the probability of introgression is 9.41 × 10−5 from H. annuus into H. petiolaris and 2.10 × 10−6 from H. petiolaris into H. annuus. If we omit immigrant inviability from our estimate, then we estimate that introgression success probabilities are 1.25 × 10−4 from H. annuus into H. petiolaris and 3.07 × 10−6 from H. petiolaris into H. annuus.
Compared to the estimates in Strasburg and Rieseberg (2008), the lower mutation rate used in this study results in ∼65% increases in effective population sizes and divergence time. Effective population sizes of H. annuus and H. petiolaris are 3.9 × 106 and 3.8 × 106, respectively, based on a generation time of one year. These values are halved when a generation time of two years is employed (Table 2 and Fig. 3). Using the estimates of effective population size and total migrants per generation—Nm= 0.34 (H. petiolaris into H. annuus) and Nm= 0.76 (H. annuus into H. petiolaris)—we can calculate a per-generation migration rate, m (or the proportion of the gene pools that are shared each generation). For H. annuus, we calculated m= 8.7 × 10−8 and m= 1.7 × 10−7 for generation times of one and two years, respectively. For H. petiolaris, m= 2.0 × 10−7 and m= 4.0 × 10−7 for generation times of one and two years, respectively (Table 2 and Fig. 3). Although a longer generation time significantly reduces estimates of the effective population size, it does not affect estimates of divergence time or the number of migrants per generation, Nm (Table 2 and Fig. 3). Therefore, to reduce confusion and redundancy, we will restrict further discussion to estimates obtained with a generation time of one year, with the caveat that these may underestimate m.
Table 2. Population genetic parameters, effective population size, bidirectional gene flow, and time since divergence between Helianthus annuus and H. petiolaris, recalculated from Strasburg and Rieseberg (2008).
N: H. annuus
N: H. petiolaris
Nm: H. petiolaris→H. annuus
Nm: H. annuus→H. petiolaris
m: H. petiolaris→H. annuus
m: H. annuus→H. petiolaris
Divergence time (years)
Although conventional wisdom suggests that species separated by strong reproductive barriers should show high levels of genetic differentiation, population genetic data indicate that apparently strong reproductive barriers may not be effective at preventing gene flow between species (Dobzhansky 1973; Machado et al. 2007). This is the pattern we observe between H. petiolaris and H. annuus, which as far as we are aware represents the first case in plants where we have quantitative estimates of both total reproductive isolation and gene flow.
Overall, the strength of reproductive barriers between H. annuus and H. petiolaris appears to be higher than that between most other plant species whose barriers have been quantified (Lowry et al. 2008a). Five of the 19 species pairs analyzed by Lowry et al. were considered completely isolated (including H. annuus and H. petiolaris); that is with RITotal= 1. Because we showed here that the reproductive barrier between H. petiolaris and H. annuus is very strong but not complete, we suspect that other species considered completely isolated by Lowry et al. (2008a) may be found to be incompletely isolated if barrier strength were to be estimated to a precision beyond 10−3, as we have done in this study.
The contribution of each individual reproductive barrier to total interspecific reproductive isolation in Helianthus is highly variable. Our estimate of ecogeographic isolation is highly dependent on the method used. Nonetheless, H. annuus and H. petiolaris would have a large opportunity to hybridize across North America if spatial distribution was the sole source of isolation, as demonstrated by the existence of numerous hybrid zones across their range (Rieseberg et al. 1998). Although human disturbance has likely increased the number of hybrid zones, many areas exist where hybrid zones likely occurred without human disturbance, such as where rivers bisected sandy soils. An intraspecific population migration rate of Nem∼0.9 has been estimated for both H. annuus and H. petiolaris (Schwarzbach and Rieseberg 2002). Thus, alleles that introgress in regions where both species occur will move into regions where the species are not in contact. We did not include the eastern portion of the United States in our analysis. Although there are many geographically isolated H. annuus populations in this area, the populations are often much smaller than those observed in the Midwest and Southwest.
Reproductive asynchrony also seems to be a weak reproductive barrier between these species. To estimate reproductive asynchrony, we assumed that both species have the same census population sizes and pollen production. Based on what can be observed in nature, where populations of H. annuus tend to produce more flowering heads than H. petiolaris, these assumptions are probably violated. If H. annuus does produce more flowers and pollen than H. petiolaris, then the probability of H. annuus ovules being fertilized by conspecific pollen and H. petiolaris ovules by heterospecific pollen will increase relative to the assumptions made here. Thus, the degree of reproductive isolation due to reproductive asynchrony probably slightly overestimates the degree of isolation in H. petiolaris and slightly underestimates the degree of isolation in H. annuus. Unfortunately, there are no demographic data available on both species across the United States to obtain a more precise estimate of reproductive asynchrony.
Reproductive isolation due to selection against immigrants is not very strong between H. annuus and H. petiolaris when compared to Mimulus cardinalis and M. lewisii (Angert and Schemske 2005), or to other closely related sunflower species such as H. exilis, which possesses locally adapted ecotypes to contrasting habitats on- and off-serpentine sites in California (Sambatti and Rice 2006). In H. exilis, the estimated strength of the reproductive barrier due to immigrant inviability alone is 0.968 (Lowry et al. 2008a).
A recent review on reproductive barriers among closely related plant taxa found that prezygotic isolation overall is about twice as strong as postzygotic isolation (Lowry et al. 2008a). Helianthus is somewhat of an exception to this rule in that total postzygotic isolation is strong and quite similar to prezygotic isolation in both species studied (Table 1). Moreover, our analysis shows that of the three strongest barriers, pollen competition, F1 pollen viability, and F1 seed set (RI ≥ 0.8 in each direction), two are postzygotic (Table 1 and Fig. 2), and our data for pollen competition could be due to a combination of pre- and postzygotic factors. On the other hand, several authors have argued that because premating barriers act earlier than postmating barriers, they have disproportionate effects on isolation (Schemske 2000; Coyne and Orr 2004; Rieseberg and Willis 2007). Although this is correct, unless reproductive isolation is absolute, even very late acting barriers are important: even reductions in gene flow from 10−4 to 10−5 can have long-term population genetic consequences.
Finally, two barriers were not considered in this study. Pollinator behavior was not considered as a source of prezygotic reproductive barrier in our study. Pollinators can contribute to prezygotic barriers in two ways. First, the geographic distribution of pollinator species can correlate with the distribution of sunflower species such that pollination occurs preferentially between conspecifics. Second, some pollinators may show preference for visiting one of the sunflower species. Although these factors have been shown to contribute to reproductive isolation in Mimulus (Bradshaw and Schemske 2003), we view this as unlikely in Helianthus as both sunflower species are mostly pollinated by bumblebees, which show a broad geographic distribution and appear to be generalists with respect to visitation preferences (Hurd et al. 1980).
Extrinsic postzygotic isolation due to hybrids having lower fitness in parental habitat was also not considered. Our experiments examining this factor have thus far not included any years where reproduction occurred in H. petiolaris habitat (Sambatti et al. 2008), and so miss a portion of the life cycle. However, our results suggest that extrinsic postzygotic barriers are relatively weak, as the viability of early generation hybrids in parental habitats is comparable to or greater than that of parental populations, apparently due to heterosis.
REPRODUCTIVE BARRIER STRENGTH CALCULATION IN PLANTS
To calculate total reproductive barrier strength between two taxa, it is assumed that individual barriers are sequential (Coyne and Orr 2004). Plants violate this assumption because gene flow can take two parallel routes: pollen and seed flow. While heterospecific seeds have to establish, grow, and reproduce, gene flow through pollen bypasses seed immigrant inviability. Because pollen flow seems to be one order of magnitude greater than seed flow in many different plant species (Petit et al. 2005), we suspect that current estimates of total reproductive barrier strength in plants may be overestimated. This is of particular concern because extrinsic postzygotic barriers measured by F1 performance with respect to the parental species are usually weak or even negative due to heterosis (Burke et al. 1998; Lowry et al. 2008a). If local adaptation of the cytoplasm occurs (Kimball et al. 2008; Sambatti et al. 2008), this will further reduce the importance of gene flow through seeds. The estimation of the barriers to pollen flow must exclude immigrant inviability. However, the change in reproductive isolation due to excluding this factor is minimal in sunflowers (Table 1).
Population genetic data seem at first glance to be inconsistent with a strong interspecific reproductive barrier, with several different population genetic approaches showing that gene flow is high between these species. For example, Yatabe et al. (2007) compared genetic distances at 55 microsatellite loci between the distantly related sympatric species pair (H. annuus and H. petiolaris) and the much more closely related species pair (H. annuus and H. argophyllus), which were likely allopatric for much of their evolutionary history. Helianthus annuus and H. petiolaris were genetically more similar than either of the allopatric species pairs. This pattern has been confirmed in other studies (Kane et al. 2009; Strasburg et al. 2009).
We have reestimated the effective population sizes, divergence time, and long-term levels of interspecific gene flow between H. annuus and H. petiolaris, and for H. annuus and H. argophyllus (Table 2, and J. L. Strasburg and L. H. Rieseberg, unpubl. data). The estimated number of effective individuals migrating between species per generation (Nm) shows a noticeable but nonsignificant asymmetry: Nm (Pet into Ann) = 0.34 and Nm (Ann into Pet) = 0.76 (Table 2 and Fig. 3). As estimates of long-term interspecific gene flow accumulate, it is becoming clear that low, often asymmetric, rates of gene flow between species are not uncommon (Llopart et al. 2005; Bull et al. 2006; Lawton-Rauh et al. 2007). However, the multilocus estimates of 0.34–0.76 for Nefm for these two sunflower species are substantially higher than most other published estimates, especially for species that are close to two million years old (Strasburg and Rieseberg 2008). Although human disturbance may have led to increased opportunity for hybridization, the large effective population sizes make it unlikely that the long-term estimates of migration have been biased significantly by recent increases in gene flow.
According to our barrier strength estimates, the production of a reproductively successful F1 hybrid between H. petiolaris and H. annuus should be an extremely rare event, somewhere between 104 and 106 times less likely than successful intraspecific matings, depending on the assumptions we make about the independence between the intrinsic postzygotic barrier components. These low values are shared with other species pairs that diverged around 0.5–1 million years ago. For example, there is evidence that successful hybrid females between D. pseudoobscura and D. persimilis, two sibling species that diverged 0.55 million years ago (Wang et al. 1997), can be found in nature with a probability of 10−4 (Powell 1983, but see Powell 1991). With a divergence time of 0.5 million years, we can estimate from Crochet et al. (2003) that the gull species Larus argentatus and L. fuscus produce hybrids with a probability of 1.0 × 10−5– 1.2 × 10−6 per year. Although consistent with what we can find in the literature, the low probability of successful hybrid production between these sunflower species seems to contradict the population genetic analyses, which suggest that gene flow between these species occurs at rates as high as half of the gene flow observed between populations of the parental species (Morjan and Rieseberg 2004).
This apparent discrepancy appears to result from the combined effects of population size, Ne, and the proportion of the gene pools that are shared each generation, m, on gene flow and its ability to homogenize populations or species. Very low rates of gene flow are required to prevent populations with large effective sizes from differentiating via genetic drift, whereas substantially more gene flow is required to prevent differentiation between smaller populations (Templeton 2006). The estimates of population sizes for these two species are very large, so relatively little gene flow is needed to prevent their differentiation at neutral loci. Of course, this also means that little of the differentiation between the two species is likely to have resulted from genetic drift in the first place since their effective population size is greater than their divergence time (approximately 1.8 million years). Note that while the effective size estimates here are very large for plants, they are consistent with the high levels of genetic diversity in these two species; based on a recent survey (Lynch 2006), H. annuus and H. petiolaris are two of the most genetically diverse flowering plants for which data are available.
The magnitudes of the per generation migration rates—m= 8.7 × 10−8 from H. petiolaris into H. annuus, and m= 2.0 × 10−7 from H. annuus into H. petiolaris—are substantially below our estimates for fertile F1 production (10−4 and 10−6, respectively) based on the analysis of reproductive barriers. This discrepancy may stem from violations in the assumptions of the IM model that can bias parameter estimates, such as population structure in the current or ancestral species, recombination, and variation in migration rate with time (Becquet and Przeworski 2009; Strasburg and Rieseberg 2010). Ancestral population structure in particular can lead to downward biases in estimates of m (Becquet and Przeworski 2009). Similarly, temporal variation in gene flow levels could lead to even lower estimates of migration (Becquet and Przeworski 2009; Strasburg and Rieseberg 2011). In contrast, estimates of contemporary Ne seem to be robust to most biologically realistic violations of the assumptions of the IM model, and are in keeping with observations of populations in the field, particularly in southern New Mexico. Therefore, relatively high levels of per generation gene flow estimated between species can be due mainly to their large effective population sizes.
A second explanation for the discrepancy between estimates from reproductive barriers and coalescent models may be differences in time scale. Isolation with migration models estimate long-term average migration rates, which need not be strictly identical to contemporary hybridization probabilities. Although recent human disturbance has likely altered the opportunities for hybridization, given the population sizes involved this should not have had a significant effect on our gene flow estimates derived from allopatric populations. We have no reason to think that there have been substantial changes in reproductive isolation recently, but cannot rule this out.
A third possible explanation for this discrepancy is that our approach of compounding different isolating barriers may underestimate reproductive barrier strength because selection against introgression over multiple generations can substantially reduce gene flow due to genome-wide linkage disequilibrium (Barton and Bengtsson 1986).
Lastly, the discrepancy may arise from the fact that the IM model assumes two panmictic populations, whereas these sunflower species have large and fragmented geographic distributions. Intraspecific gene flow rates are high enough that neutral population structure is weak in both H. annuus and H. petiolaris (Schwarzbach and Rieseberg 2002; Kane et al. 2009); and simulations indicate that comparable or even higher levels of population structure have little effect on parameter estimation, including gene flow estimation, in IM (Strasburg and Rieseberg 2010). In the presence of population structure, IM estimates migration rates with contributions from both isolation by distance (both within and between species) and conventional reproductive barriers, which can make IM estimates of gene flow further depart from those calculated empirically from reproductive barriers. Note that our estimates of barrier strength only consider the between-species component of isolation by distance, and this is done fairly arbitrarily based on overlap within 10 km squares. Although we think that this is a conservative estimate based on our knowledge of the species’ distributions and dispersal rates, we cannot rule out the possibility that we have overestimated the importance of ecogeographic isolation.
REPRODUCTIVE ISOLATION AND THE PERSISTENCE OF HYBRIDISING SPECIES
The large effective population size of the two hybridizing sunflower species has two main consequences. First, it increases the efficacy of selection relative to genetic drift in generating divergence; and second, it reduces the efficacy of reproductive barriers in preventing homogenization at neutral loci. Given the relatively short divergence time between these two species, it seems likely that a mosaic genome would have been observed between these species even if there had been no gene flow. Nonetheless, comparisons with younger allopatric taxa in the same genus (e.g., Yatabe et al. 2007; Strasburg et al. 2009) imply that gene flow has accentuated the genomic mosaic between H. annuus and H. petiolaris. However, this could be due to the spread of advantageous alleles between species (which can occur just as readily in species with low Ne), as well as neutral introgression. Studies of patterns of genomic divergence in sympatric and allopatric species pairs of similar age and population size will be required to tease apart the relative contribution of these different process to patterns of genomic heterogeneity. Because of the extent of contact between H. annuus and H. petiolaris, and the large effective population size of the two species, recombination is likely to have freed neutral or advantageous alleles from sites under divergent natural selection, limiting the lengths of differentiated regions between the two genomes. Nonetheless, under some conditions, such as small effective population sizes, neutral sites may diverge as well (i.e., divergence hitchhiking) (Feder and Nosil 2010). This may contribute to variation seen in the mosaic structure of some hybridizing plant and animal genomes (Nosil et al. 2008; Via and West 2008), and might account for the apparent maintenance of neutral differences in some hybrid zones with high Nm (Barton and Hewitt 1985). Wu (2001) proposed that speciation may occur on the scale of loci rather than genomes: these data from Helianthus suggest how this might happen.
Associate Editor: J. Pannell
The authors would like to thank N. C. Kane, M. S. Barker, B. Gross, and M. Grote for helpful advice and comments. This research was funded by an NSERC grant (327475) to LHR, an Australia Research Council (DP0986175) grant to EJB. and DOB, a National Institutes of Health Ruth L. Kirschstein postdoctoral Fellowship (5F32GM072409–02) to JLS, and a Killam postdoctoral Fellowship to DOB.