Species population genetics could be an important factor explaining variation in clade species richness. Here, we use newly generated amplified fragment length polymorphism (AFLP) data to test whether five pairs of sister clades of Costa Rican orchids that differ greatly in species richness also differ in average neutral genetic differentiation within species, expecting that if the strength of processes promoting differentiation within species is phylogenetically heritable, then clades with greater genetic differentiation should diversify more. Contrary to expectation, neutral genetic differentiation does not correlate directly with total diversification in the clades studied. Neutral genetic differentiation varies greatly among species and shows no heritability within clades. Half of the variation in neutral genetic differentiation among populations can be explained by ecological variables, and species-level traits explain the most variation. Unexpectedly, we find no isolation by distance in any species, but genetic differentiation is greater between populations occupying different niches. This pattern corresponds with those observed for microscopic eukaryotes and could reflect effective widespread dispersal of tiny and numerous orchid seeds. Although not providing a definitive answer to whether population genetics processes affect clade diversification, this work highlights the potential for addressing new macroevolutionary questions using a comparative population genetic approach.

One of the biggest mysteries in biology is why groups vary so much in diversity. Species are unevenly distributed between sister clades at all hierarchical levels (Dial and Marzluff 1989; Marzluff and Dial 1991). This must be the result of variation between groups in speciation or extinction rates or in limits to group diversity (Rabosky 2009; Kisel et al. 2011), but beyond this it remains uncertain what factors control diversity within and between groups. Previous studies have focused pragmatically either on easily obtainable traits such as body size (e.g., Owens et al. 1999; Isaac et al. 2005) or on statistical inference from the shapes of phylogenetic trees (e.g., Phillimore and Price 2008). Key population genetic parameters that underpin theories of speciation, and are therefore likely to have a large impact on rates and patterns of speciation, have been neglected largely because of the difficulty of measuring them across a broad enough sample of species to detect an impact on diversity (Barraclough and Nee 2001).

Theories of speciation focus on the processes causing populations within a species to differentiate sufficiently to be considered new species. In particular, divergent selection is thought to drive population differentiation, whereas gene flow counteracts differentiation (Felsenstein 1981; Gavrilets 2003). Genetic drift and mutation also contribute, but genetic drift is generally considered to be of minimal importance in driving speciation, and mutation-order speciation (driven by the fixation of different, incompatible mutations in different populations) has not been empirically demonstrated (Sobel et al. 2010). If the strengths or balance of these processes depended on characteristics that are heritable within clades, such that species within clades tended to share similar levels of population differentiation, then population genetic differentiation could be a species-level trait influencing rates of diversification (Jablonski 2008). For example, more species-rich clades might be those with, on average, lower rates of gene flow between populations and correspondingly higher levels of differentiation between populations. Alternatively, population genetic differentiation might not influence clade diversification if it is not heritable within clades, or if other processes are more important in limiting diversification (such as extinction or ecological limits on diversity).

Few studies have tested the link between species population genetics and clade diversification. A handful of studies have tested the link between dispersal ability and diversification rates, and dispersal ability is generally a good proxy for rates of gene flow (Zera 1981; Govindaraju 1988; Bohonak 1999). Most of these studies have found greater diversification associated with poorer dispersal as expected (e.g., Jablonski 1986; Belliure et al. 2000). In some cases, greater diversification has been found to be associated with greater dispersability, although this link is likely driven by the ability of species to colonize new regions, rather than by the level of gene flow between established populations (e.g., Owens et al. 1999; Phillimore et al. 2006). In a few cases, greater diversification has been found to be associated with intermediate dispersal (e.g., Price and Wagner 2004; Paulay and Meyer 2006) or not associated with dispersal at all (e.g., Vrba 1984; Herrera 1989). Fewer studies have tested whether variation in the strength of diversifying selection might explain patterns of species richness; the exception being those showing a relationship between diversification and the strength of sexual selection, as measured by the divergence of mating characteristics between males and females (Barraclough et al. 1995; Stuart-Fox and Owens 2003; Seddon et al. 2008).

A better approach than using surrogate measures is to use genetic data to test directly for the effects of population genetic processes on diversification. To our knowledge, only one previous study has taken this approach. Kisel and Barraclough (2010) compiled estimates of population differentiation from literature studies, as a surrogate measure for gene flow, and showed that major taxa with less gene flow within species were able to speciate within smaller oceanic islands than taxa with greater gene flow. Measures of differentiation from neutral markers are influenced by factors other than gene flow, such as population age and demography (Marko and Hart 2011), but gene flow seemed the most likely explanation for the scaling with island area found. However, Kisel and Barraclough (2010) did not address whether variation in levels of gene flow or genetic differentiation explained diversity differences between related clades.

Here, we test whether the level of neutral population genetic differentiation within species is phylogenetically heritable and whether it affects clade diversification, using new AFLP data from 647 individuals from 17 exemplar species of tropical orchids belonging to five pairs of sister clades that differ greatly in species richness. We test whether neutral genetic differentiation is associated with diversification by comparing average within-species genetic differentiation between species-rich and species-poor sister clades, taking into account possible confounding influences of species’ ecology and range size. We use sister group comparisons to control for clade age in comparing diversification and to provide replicated, phylogenetically independent tests of the hypothesized relationship. In addition, sister groups often share many traits, reducing the number of variables that can confound conclusions about traits of interest (Barraclough et al. 1998).

We expect neutral genetic differentiation within species to be partly phylogenetically heritable. This is because neutral differentiation is affected by many species characteristics that are phylogenetically heritable (such as flower, pollen, and seed morphology; preferred habitat; and plant size), but is also affected by external factors (such as geography, pollinator availability, weather, and chance events) and variable characteristics of populations (such as population size, which is likely to affect pollinator behavior). At larger taxonomic scales, when comparing very different taxonomic groups (e.g., birds and snails; Kisel and Barraclough 2010, or distantly related plants; Hamrick 1983; Duminil et al. 2007), it is clear that heritable differences in neutral differentiation exist, but this has not been examined at finer taxonomic scales. For this reason, we aim to test whether neutral genetic differentiation is heritable at a finer taxonomic scale—in this case, among clades of orchids.

We also predict that increased diversification within clades should be associated with stronger neutral genetic differentiation within species. This could be the result of greater diversification being driven by reduced gene flow, increased mutation, and/or increased genetic drift. We do not expect such an association to result if diversification is driven mainly by local adaptation, as in this case only small parts of the genome should be affected, and not overall neutral differentiation. In any case, we expect an association between diversification and genetic differentiation to be possible regardless of the usual mode of speciation, whether allopatric, parapatric, or sympatric, and whether “ecological” or “non-ecological” (Sobel et al. 2010), as long as levels of gene flow, mutation, and/or drift are heritable to some extent and affect the probability of speciation.

Orchids are a good study group because their many species and great variability allow many independent tests of factors hypothesized to affect diversity patterns. They abound in pairs of sister clades that differ greatly in diversity—for example, the most dramatic pairs in this dataset compare clades with 42 and 140 species to sister clades with 1066 and 1387 species, respectively. They also present an unsolved biodiversity mystery: with 26,000+ described species (Dressler 2005; Govaerts et al. 2010), orchids are probably the largest family of angiosperms. Furthermore, although they have attracted a devoted research community stretching back to Darwin (1862) and earlier, the reasons for their diversity are still highly debated (Gravendeel et al. 2004; Cozzolino and Widmer 2005; Tremblay et al. 2005).

We use AFLPs to generate the genetic data analyzed here because they are the most accurate fingerprinting method available that does not require developing individual markers de novo for each species. Microsatellites and other codominant markers can be more reliable than AFLPs and give more information, as they produce a complete genotype for each individual, allowing for more precise estimates of population genetic measures (Mueller and Wolfenbarger 1999). However, they take months to develop for each new species and the costs of developing them for multiple species often outweigh the rewards. As they are typically used—as purely dominant markers—AFLPs give less precise estimates of population measures such as FIS and FST, although more powerful analysis methods have been developed that also take AFLP band intensities into account (e.g., Foll et al. 2010). However, they are more reliable than restriction fragment length polymorphisms (RFLPs), randomly amplified polymorphic DNA, and other types of fragment-based markers and are quick to develop for new species, making them the marker of choice for sets of species that have not previously been studied genetically, such as the 17 species in this study (Mueller and Wolfenbarger 1999; Bensch and Åkesson 2005). Furthermore, AFLPs have proven to provide useful and robust insights for population genetic studies (e.g., Grahame et al. 2006; Jorgensen et al. 2006; Hensen et al. 2011).

We measure the level of genetic differentiation within each species with ΦST, an analog of FST and thus a measure of the proportion of total neutral genetic variation between (rather than within) populations. Because the calculation of ΦST is based on distances between haplotypes (numbers of differences between individuals, over all loci) rather than allele frequencies or heterozygosity at each locus (Excoffier et al. 1992), it is particularly suited to the analysis of datasets with many loci, such as AFLP datasets. As with FST, ΦST ranges from 0 to 1, with high values indicating clear population differentiation and low values indicating little differentiation (Wright 1931; Slatkin 1985). When gene flow is at equilibrium with genetic drift, ΦST like FST should increase with geographic distance between populations, as migration is more frequent between nearby populations (Hutchison and Templeton 1999). This makes the scale of sampling an important factor in population genetic studies. For this reason, we evaluate patterns of genetic isolation by distance (IBD) in addition to analysing ΦST alone.

In addition, we calculate three additional measures of neutral differentiation—FST, GST (Hedrick 2005), and D (Jost 2008)—and use them to repeat key analyses, to confirm that our conclusions are not biased by choice of differentiation metric. FST, the classic measure first defined by Wright (1931, 1951) and in some cases equivalent to Nei's GST (1973), is a simple ratio of expected heterozygosities/gene diversities. It is based on the idea that completely differentiated populations are fixed for different alleles, so that when differentiation is high, heterozygosity is low within populations but high overall. However, because of how FST is calculated, it is limited by within-population homozygosity: if heterozygosity within populations is high, FST must be low, even if populations have different sets of alleles. GST is a corrected version of FST that was developed to address this problem. It is calculated by dividing FST or GST by the maximum FST or GST value possible considering the level of within-population homozygosity (Hedrick 2005). D is another newly developed measure of differentiation, still based on heterozygosities, that uses a different mathematical definition of genetic diversity (Jost 2008).

We also account for two possible covariates affecting our comparisons: species range size and ecology. The relationship between range size and diversification is complex, as the former may directly influence speciation and extinction rates, yet is also a product of lineages’ evolutionary histories (Rosenzweig 1995; Webb and Gaston 2003; Pigot et al. 2010; Birand et al. 2012). Similarly, it is unclear to what extent ecological traits of species generally affect diversification, but some association is expected (e.g., Phillimore et al. 2006); for example, niche breadth should affect the probability that a species expands its range, speciates, and goes extinct (Funk et al. 2002; McPeek 2008; Birand et al. 2012). Furthermore, there is evidence that both range size and ecology are associated with variation in population differentiation (Hamrick and Godt 1996; Morjan and Rieseberg 2004). For this reason, we include both a restricted and a widespread species for as many study clades as possible and explicitly test the strength of associations between species ecology and range size with genetic differentiation and diversification.



Five pairs of sister study clades were chosen from two subtribes of the tribe Epidendreae (subfamily Epidendroideae) with recently published phylogenetic analyses and complete sampling at the genus level: Pleurothallidinae (Pridgeon et al. 2001) and Laeliinae (van den Berg et al. 2009). If possible, we would have included more sister clade pairs, allowing for more comparisons of the relationship between species richness and within-species population genetics. However, the extensive sampling needed for each study clade (multiple individuals from multiple populations from multiple species) limited the number of sister clade pairs we could consider. Nevertheless, we believe five comparisons are sufficient for a first test of our hypothesis.

Sister clades were chosen from well-resolved portions of the published phylogenetic trees. The only uncertainty in the composition of chosen clades was whether the genus Meiracyllium belongs in the Brassavola clade or was placed there spuriously (van den Berg et al. 2009). We assumed that it belongs in the Brassavola clade; Meiracyllium contains only two species and should not affect results greatly either way. Sister clades were chosen that have as large differences as possible in species richness, using species numbers for genera from the World Checklist of Orchidaceae (Govaerts et al. 2008). All pairs chosen differ in species richness by at least fivefold. Clade species richness and genera included in each clade are listed in Table 1.

Table 1.  Study clade pairs, with genera they include and currently accepted species richness according to the World Checklist of Orchidaceae (Govaerts et al. 2010). [Correction added to Table 1 after online publication May 22, 2012].
Species-rich cladeNo. of spp.Species-poor sister cladeNo. of spp.
Masdevallia clade  751 Trisetella clade  23
 Masdevallia Ruiz & Pav.  582  Trisetella Luer  23
 Diodonopsis Pridgeon & M. W. Chase   5  
 Dracula Luer  126   
 Porroglossum Schltr.  38  
Lepanthes clade 1066 Lepanthopsis clade  42
 Lepanthes Sw.1066  Lepanthopsis (Cogn.) Ames 42
Platystele clade  276  Dryadella clade  53
 Platystele Schltr.  99 Dryadella Luer 53
 Scaphosepalum Pfitzer in H. G. A. Engler & K. A. E. Prantl (eds.)   46   
 Specklinia Lindl. 131  
Scaphyglottis clade   77 Jacquiniella clade  13
 Scaphyglottis Poepp. & Endl.  68 Jacquiniella Schltr. 12
 Dimerandra Schltr.    9  Acrorchis Dressler   1
Epidendrum clade 1387 Brassavola clade 140
 Epidendrum L. 1325  Brassavola R. Br. in W. T. Aiton  21
 Barkeria Knowles & Westc.  15 Cattleya Lindl.111
 Caularthron Raf.    4  Guarianthe Dressler & W. E. Higgins   4
 Laelia Lindl.  24 Meiracyllium Rchb. f.  2
 Myrmecophila Rolfe   10  Rhyncholaelia Schltr.   2
 Orleanesia Barb. Rodr.   9  

Study species were chosen from one genus from each study clade using four criteria. Species had to be epiphytic and easy to find and identify. In addition, species within clade pairs were chosen to have similar habitats, and when possible at least one restricted and one widespread species were chosen for each clade. Further details are given in the Supporting Information. Study species with descriptions of their ranges and habitats are listed in Table 2.

Table 2.  Study species, with distributions and habitats. Species are arranged by genus, with genera from sister clades together. No. of L= number of locations sampled. No. of S= number of samples genotyped.
GenusSpeciesRestricted or widespreadGeographic rangeTypical habitatElevation range (m)No. of LNo. of S
Masdevallia nidifica Rchb. f.WidespreadNicaragua to N. PeruVery humid, rain or cloud forest700–2000534
  rafaeliana Luer Restricted Costa Rica and Panama Cloud or oak forest 2600–3000 2 30
Trisetella triglochin (Rchb. f.) LuerWidespreadCosta Rica to S. tropical AmericaVery humid, rain or cloud forest200–1900270
Lepanthes ciliisepala Schltr. Restricted Costa Rica and possibly Venezuela Cloud or oak forest 1400–2050 2 36
  elata Rchb. f.WidespreadCosta Rica to W. ColombiaCloud or oak forest1500–2600352
Lepanthopsis floripecten (Rchb. f.) Ames Widespread S.E. Mexico to S. tropical America Rain forest 1900–2000 5 20
Platystele propinqua (Ames) GarayRestrictedCosta RicaCloud or oak forest1400–1900324
  stenostachya (Rchb. f.) Garay Widespread Mexico to S. tropical America Very humid or rain forest 0–1900 3 50
Dryadella odontostele LuerWidespreadCosta Rica, Panama, ColumbiaVery humid forest50–150312
Scaphyglottis jimenezii Schltr. Restricted Costa Rica and W. Panama Very humid, rain or cloud forest 700–2400 3 47
  fusiformis (Griseb.) R.E. Schult.WidespreadCosta Rica to S. tropical AmericaVery humid or rain forest50–1400442
Jacquiniella aporophylla (L.O. Williams) Dressler Restricted Costa Rica and Panama Rain forest 800–1500 4 37
  teretifolia (Sw.) Britton & P. WilsonWidespreadMexico to northern S. AmericaHumid, very humid, rain or cloud forest1100–1850546
Epidendrum exasperatum Rchb.f. Restricted Costa Rica and Panama Very humid, rain, cloud or oak forest; pastures and slopes 900–2500 3 27
  laucheanum Bonhof ex RolfeWidespreadMexico to ColombiaVery humid or cloud forest1300–2100531
  vulgoamparoanum Hágsater & L.Sánchez Restricted Costa Rica and Panama Very humid or rain forest 0–350 4 38
Brassavola nodosa (L.) Lindl.WidespreadMexico to S. tropical AmericaDry, humid, very humid or scrub forest; rocks or mangroves0–100850


All samples were collected in Costa Rica with kind help from the Lankester Botanical Garden (University of Costa Rica). Sample collection occurred in two field seasons, April to May 2008 and March to May 2009. Four samples were from plants previously collected by D. Bogarin and maintained in the Lankester living collection (see Table S2, available in Dryad: doi:10.5061/dryad.55md6hj5). Voucher specimens were deposited at the Lankester Botanical Garden (Table S1, available in Dryad).

We sampled the largest number of individuals, populations, and species possible while ensuring that each clade was represented by multiple populations of at least one species. Because some species were difficult to locate in the field, and because of the effort required to sample 17 species at a population level, sampling for each species was not as comprehensive as for typical single-species studies. However, as our aim was to investigate general patterns of neutral genetic differentiation and not to describe the population genetics of each study species in detail, we believe that this level of sampling was appropriate for this study.

Field expeditions were oriented toward finding populations for each species representing as wide a range of distances between populations and as large a portion of the Costa Rican distribution of each species as possible. Further details regarding sampling locations are given in the Supporting Information. Maps of sampling locations for each species are given in Figures S1–S3 and details of sampling locations for each species are given in Table S2 (available in Dryad).

To our best knowledge, each sample was taken from a separate individual. Where it was unclear if adjacent shoots were part of the same plant, only one was sampled. In general, samples were taken from widely spaced plants to reduce the chance of sampling one clone multiple times. Leaf or floral tissue samples were put into labeled plastic bags in the field and kept in a refrigerator or ice chest until they could be cut and dried in silica gel for preservation (Chase and Hills 1991).


DNA was extracted using the Qiagen DNeasy Plant Kit (Qiagen, Crawley, West Sussex, UK) following the manufacturer's protocol. Except when too little sample material was available (for small species), approximately 20 mg silica-dried material was used for each extraction. Flowers were used in preference to leaves when available; flowers dry more quickly in silica gel and are more easily homogenized.

Trials of up to 12 selective primer combinations were carried out separately for each species using the AFLP Regular Genome Plant Mapping Kit from Applied Biosystems (Carlsbad, CA). Primers were then chosen to maximize number of peaks per sample, number of polymorphic peaks per species, evenness of spread of peak sizes (even if this required choosing a primer that produced fewer peaks), and profile repeatability.

AFLP reactions were carried out following Vos et al. (1995) with minor modifications using EcoRI, T4 Ligase, and ligase buffer from Promega (Madison, WI), MseI from New England BioLabs (Ipswich, MA), MseI and EcoRI adaptor pairs and primers from Applied Biosystems and PCR mastermix from Fermentas (St. Leon-Rot, Germany). A different set of primers (and thus, AFLP loci) was used for each species. The AFLP genotypes were sequenced using a GeneScan 500 ROX size standard and capillary sequencer from Applied Biosystems. Details are given in the Supporting Information. Details of the protocol variations used for each species, including the selective primers used, are listed in Table S3 (available in Dryad).


AFLP scoring was carried out separately for each species using GeneMapper version 4.0 (Applied Biosystems) to manually identify bins and AFLPScore version 1.4a (available at http://www.sheffield.ac.uk/molecol/software~/aflpscore.html; Whitlock et al. 2008) to optimize scoring parameters and objectively create a binary genotype table for each species. Details are given in the Supporting Information. Details of scoring parameters used and error rates for each species and primer combination are given in Table S4 (available in Dryad).


For all species, populations used as units for analysis were defined by distance: samples collected within a diameter of 1.5 km were treated as a single population. This distance was chosen because it was the maximum length of contiguous stands of the same species that were sampled, and it corresponds to the scale at which previous studies of tropical orchids have found low genetic differentiation (Trapnell and Hamrick 2005).

Before carrying out analyses, the AFLP dataset for each species was checked for outlier loci potentially under the influence of selection using the software DfDist (available from http://www.rubic.rdg.ac.uk/~mab/stuff/; modified to allow dominant data from Beaumont and Balding 2004). Further details are given in the Supporting Information. Over all species, only 26 loci were excluded.

Some final AFLP datasets included populations with only one or a few individuals due to difficulties with collecting in the field or AFLP genotyping. To deal with this problem, all analyses were carried out both on the full dataset including all populations and on a reduced dataset that excluded populations with less than three individuals. In the case of Lepanthopsis floripecten, which had only one population with three or more samples, the reduced dataset only excluded populations with one individual. Both full and reduced final AFLP datasets are available in Dryad.


The level of neutral genetic differentiation within each species was estimated in four different ways. In all cases, negative values were replaced with 0 for further analyses (Long 1986). Except where stated otherwise, all analyses were carried out using R version 2.8.1 (R development core team 2008) or higher.

First, ΦST values were estimated separately for each species using Arlequin version 3.5 (Excoffier et al. 1992; Excoffier and Lischer 2010), with between- and within-group variances calculated using analysis of molecular variance of genetic distances between sample haplotypes. The significance of each overall, species-level ΦST value was tested through permutation of the original haplotype table.

Second, FST values were estimated for each species using AFLP-SURV version 1.0 (Vekemans 2002), based on allele frequencies estimated using a Bayesian method and the approach of Lynch and Milligan (1994). Hardy–Weinberg equilibrium was assumed for all species because no data on inbreeding were available. Allele frequencies were calculated using a nonuniform prior distribution when sampling was good enough to permit; otherwise a uniform prior distribution was used (for both datasets of Trisetella triglochin, and full datasets of Jacquiniella aporophylla, Lepanthopsis floripecten, Masdevallia nidifica, and Platystele propinqua).

Third, GST was calculated using heterozygosities estimated with AFLP-SURV (using the same settings as described above) and equation 4b from Hedrick (2005). The FST value given by AFLP-SURV was used for GST in the equation; Hw from AFLP-SURV was used for Hs; and number of populations sampled was used for k.

Finally, D was calculated using heterozygosities estimated with AFLP-SURV (using the same settings as described above) and equation (11) from Jost (2008). The Ht value given by AFLP-SURV was used for Ht in the equation; Hw from AFLP-SURV was used for Hs; and number of populations sampled was used for n.

Because the geographic arrangement of sampled populations varied between species, the relationship between pairwise ΦST values and distance was investigated for each species with more than two sampled populations using Mantel tests. Details are given in the Supporting Information. A table of pairwise ΦST values and distances between populations is available in Dryad.


The phylogenetic tree of study species was reconstructed using newly generated matK sequences (sequences deposited in GenBank—accession numbers JQ771559-JQ771575 and detailed in Table S5, also available in Dryad; tree deposited in TreeBase—http://purl.org/phylo/treebase/phylows/study/TB2:S12523). Details of sequencing and tree reconstruction methods are given in the Supporting Information.

Phylogenetic heritability was estimated for overall ΦST, FST, GST, and D by calculating the phylogenetic signal using λ (Pagel 1999). λ varies from 0 to 1, where 0 means a trait evolves independently of the phylogenetic tree (is not heritable along lineages), and 1 means trait values are entirely determined by the tree (are completely heritable). The maximum likelihood value of λ for each differentiation measure was calculated using the CAIC package in R (version 1.0.4–94; Orme et al. 2008) and the matK species tree. Likelihood profiles for λ over the interval 0 to 1 were examined graphically using the caper package in R (version 0.5; Orme 2012). Likelihood ratio tests were used to test whether each λ value was significantly different from 0 or 1, by computing the likelihood ratio between a model optimizing λ and a model fixing λ to 0 or 1, and computing the P-value using a chi-squared distribution (Freckleton et al. 2002).

We also conducted a randomization test to evaluate whether differentiation values within study clades are more similar to one another than differentiation values between clades. This test evaluated the phylogenetic heritability of genetic differentiation within our terminal clades, as our hypothesis requires differentiation to be heritable only within clades, not necessarily between sister clades. This test was conducted by shuffling pairwise ΦST values randomly among all population comparisons, then calculating F from an analysis of variance (ANOVA) with “clade” as the explanatory factor for ΦST. The P-value for rejecting the null hypothesis of no significant differences among clades was the proportion of F values from the randomizations that equaled or exceeded the observed value. To prevent the analysis being biased by the number of sampled populations within each species (which varied widely), we weighted each population comparison by the reciprocal of the number of population comparisons for that species, so that each species contributed a weight of 1 to the analysis. The pairwise ΦST values are available in Dryad.


The effect of neutral genetic differentiation on diversification was tested using two-tailed Wilcoxon signed rank tests comparing mean overall ΦST, FST, GST, and D values between sister clades. Mean values of each differentiation measure were calculated for each clade over all sampled species.

The effect of the interaction between range size and differentiation on diversification was tested using two-tailed Wilcoxon signed rank tests comparing mean overall ΦST, FST, GST, and D values of widespread species between sister clades. Values from restricted species were not compared because only one restricted species from a species-poor clade was collected.

Mean species range size, elevation range, and number of habitats were compared between sister clades using two-tailed Wilcoxon rank sum tests to investigate whether differences in range size or ecology could explain differences between clades in diversification. For these tests, range sizes for all species in all genera in each clade were taken from the World Checklist of Orchidaceae (Govaerts et al. 2007) as number of TDWG Level 2 regions (http://www.tdwg.org/standards/109/) for which species occurrence was noted. Species elevation and habitat ranges were compiled from the Manual de Plantas de Costa Rica, Vol. III (MPCR; Dressler 2003) for all species native to Costa Rica from each clade. Elevation and habitat data were compiled only for species native to Costa Rica as this way all data came from a single source and were comparable. Mean range sizes, elevation ranges, and habitat ranges of study clades are given in Table S6 (available in Dryad). Tables of all range sizes, elevation ranges, and habitat ranges used for these analyses are also available in Dryad.

Finally, pairwise ΦST values were examined using a linear model including ecological variables and clade species richness in order to identify which ecological factors are most strongly associated with genetic differentiation and whether any affect the relationship between differentiation and diversification. We used all ecological variables that were available to us from the MPCR and fieldwork carried out for this study. The full model included at the population level, geographic distance, elevation difference, and difference in average branch circumference between populations; at the species level, number of TDWG Level 2 regions occupied, elevation range, number of habitats, minimum elevation, mean branch circumference, variance in branch circumference, number of months with flowering recorded (from the MPCR), occurrence in cloud forest (yes or no), and occurrence in disturbed areas (yes or no); and at the clade level, clade size (big or small), and number of species in the clade. Variance in branch circumference, elevation range and number of habitats were included as proxies of ecological specificity/niche breadth. The full model was used to examine the variance in ΦST explained by each variable, and multimodel comparison over the full set of possible variable combinations was used to quantify the significance of each variable (Burnham and Anderson 2002). In addition, an ANOVA of ΦST versus species was used to measure the amount of variance explained by differences between species. Note that because we used pairwise ΦST values, the linear model might overestimate the degrees of freedom for comparisons. To account for this, any variables found to be important in the linear models were verified using Mantel tests, using the ade4 package in R (Dray et al. 2007). Elevation and habitat ranges of study species are given in Table 2. Branch circumference mean and variance for each species are given in Table S7 (available in Dryad). Pairwise ΦST values and distances, elevation differences, and branch circumference differences between populations are available on Dryad, as are branch circumference data for all samples and additional ecological data for all study species (number of months with flowering, occurrence in cloud forest, occurrence in disturbed areas).


A total of 647 plants, from 32 locations, from 17 species in five sister clade pairs were sampled and successfully genotyped (details of sampling locations given in Figures S1–S3 and Table S2, which is available in Dryad).

All differentiation measures vary widely between species, ranging from zero, indicating no differentiation, to maximums of 0.358 (ΦST), 0.321 (FST), 0.492 (GST), and 0.296 (D), indicating clear differentiation between populations (Table 3). Values of FST, GST, and D are strongly correlated with each other (for both datasets, adjusted r2 values ranging from 0.79 to 0.97), but not with values of ΦST (adjusted r2 values ranging from −0.001 to 0.025 for full dataset, and from 0.12 to 0.16 for reduced dataset). The relationship between ΦST and geographic distance is not significant for any species, and only near significance for one (Epidendrum laucheanum, P= 0.051; other P-values ranging from 0.11 to 1). Plots of ΦST against geographic distance for all species are given in Figure 1 for the full dataset and Figure S4 for the reduced dataset.

Table 3.  Measures of neutral genetic differentiation for all study species. N is the number of populations in each dataset. Hw is the mean within-population expected heterozygosity/gene diversity. Ht is the total gene diversity or expected heterozygosity in the species as a whole. Significance of overall ΦST is indicated by asterisks: *P < 0.05; **P < 0.005; ***P < 0.0005.
ComparisonClade diversitySpecies N Hw Ht Overall ΦSTAny pairwise ΦST > 0.2Overall FSTOverall GSTOverall D
1)High Masdevallia nidifica—full dataset50.410.410.36***Yes0.010.010.01
   Masdevallia nidifica—reduced dataset 2 0.26 0.32 0.35*** Yes 0.21 0.35 0.18
   Masdevallia rafaeliana 20.270.270No000
  Low Trisetella triglochin 2 0.15 0.15 0.06** No 0.02 0.02 0.01
2)High Lepanthes ciliisepala
   Lepanthes elata 3 0.09 0.09 0.07*** No 0.04 0.05 0.01
 Low Lepanthopsis floripecten—full dataset60.410.420.14Yes0.010.020.01
   Lepanthopsis floripecten—reduced dataset 4 0.20 0.29 0 No 0.32 0.43 0.16
3)High Platystele propinqua—full dataset30.300.310.17***Yes0.020.030.01
   Platystele propinqua—reduced dataset 2 0.16 0.17 0.12*** No 0.06 0.08 0.02
   Platystele stenostachya***Yes0.10.120.02
  Low Dryadella odontostele 3 0.33 0.46 0.18* Yes 0.28 0.49 0.30
4)High Scaphyglottis jimenezii 30.140.140No0.010.010.003
   Scaphyglottis fusiformis 4 0.15 0.17 0.24*** Yes 0.10 0.13 0.03
 Low Jacquiniella aporophylla—full dataset40.350.350No000
   Jacquiniella aporophylla—reduced dataset 2 0.19 0.19 0 No 0 0 0
   Jacquiniella teretifolia—full dataset50.100.100No0.010.010.002
   Jacquiniella teretifolia—reduced dataset 3 0.10 0.10 0.01 No 0 0 0
5)High Epidendrum exasperatum**No0.030.040.01
   Epidendrum laucheanum 5 0.16 0.18 0.045 No 0.11 0.13 0.03
   Epidendrum vulgoamparoanum
  Low Brassavola nodosa—full dataset 8 0.13 0.14 0.06** No 0.07 0.08 0.01
   Brassavola nodosa—reduced dataset70.140.140.05**No0.030.040.10
Figure 1.

Relationship between Φst and distance for all study species for the full dataset. Data are presented by sister-clade pair. Species from the species-rich clade in each case are shown in black; species from the species-poor clade are shown in red (grey). Restricted species are shown with open symbols and widespread species with filled symbols. The Epidendrum clade includes two restricted species: E. exasperatum is represented by open circles, whereas E. vulgoamparoanum is represented by open triangles. Each point represents Φst calculated for one pair of populations.

Over the clades examined here, neutral genetic differentiation is not strongly phylogenetically heritable (Figures 2 and S5). The maximum likelihood value of λ for overall ΦST for both datasets, and for FST, GST, and D for the reduced dataset, is below 0.1 and not significantly different from 0 (P-value for difference > 0.05). For FST, GST, and D for the full dataset the likelihood of lambda taking any value between 0 and 1 is almost constant (flat likelihood profiles), precluding accurate estimation of λ and suggesting a bias caused by poorly sampled populations (Table 4). The randomization test does not support ΦST values being more similar within clades than between (full dataset, P= 0.32; reduced dataset, P= 0.11).

Figure 2.

Associations of different measures of overall neutral genetic differentiation and species range size with species phylogenetic relationships (data from the full dataset). Boxes are shaded according to species’ overall values of Φst, Fst, Gst, or D and circles are sized proportionally to the log of species range size measured as number of TDWG Level 2 regions. Sister clades are connected by black branches, and in each sister clade pair, the larger clade is indicated by species names in bold. Species names are abbreviated by the first letter of genus and species. Numbers at nodes are bootstrap support values.

Table 4.  Results of main statistical analyses, of phylogenetic heritability, association with diversification, and association with diversification and range size, of all studied measures of neutral genetic differentiation.
MeasureMaximum likelihood value of λ P-value, λ≠ 0 P-value, λ≠ 1 P-value, association with diversification P-value, association with diversification, widespread species only
ΦST, full dataset6.6×10−511.9×10−80.310.44
ΦST, reduced dataset 6.6×10−5 1 1.8×10−8 0.12 0.13
F ST, full dataset6.6×10−51110.81
F ST, reduced dataset 6.6×10−5 1 4.3×10−6 0.81 1
GST, full dataset6.6×10−51110.81
GST, reduced dataset 0.027 0.91 1.8×10−6 0.81 0.81
D, full dataset10.12111
D, reduced dataset 0.061 0.8 3.8×10−6 0.81 1

There is no support for the hypothesis that genetic differentiation is associated with diversification in these orchids. There is no significant difference between species-rich and species-poor sister clades in mean overall ΦST (full dataset, P= 0.31, reduced dataset, P= 0.12), FST, GST, or D (same results for all three measures: full dataset, P= 1, reduced dataset, P= 0.81) (Table 4). In addition, there is no consistent distribution among clades of high values of any differentiation measure (Table 3 and Fig. 2). For example, for overall ΦST, which of all the differentiation measures shows the most variation among species, three of the five clade comparisons show the same pattern, in which all species have low ΦST values, except one species from the large clade in each case (Masdevallia nidifica, Scaphyglottis fusiformis, and Epidendrum exasperatum). The opposite pattern is seen in the LepanthopsisLepanthes comparison: the only species with a high ΦST value is Lepanthopsis floripecten, from the small clade. Finally, in the PlatysteleDryadella comparison, all species have moderately high ΦST values.

There is some evidence of an interaction between species range size and neutral differentiation in determining diversification. In both datasets, the only species with any differentiation measure over 0.2 (under ideal equilibrium conditions indicating gene flow reduced enough to allow independent evolution of populations; Wright 1931; Slatkin 1985) are widespread species. Furthermore, in both datasets, the only species with overall ΦST over 0.2 are widespread species from large clades (Masdevallia nidifica, Platystele stenostachya, and S. fusiformis). These associations can be seen in Figure 2. However, there are an equal number of widespread species from large clades with low ΦST, and there is no significant difference between the genetic differentiation of widespread species from small and large sister clades (Table 4). Furthermore, the proportion of species with overall ΦST over 0.2 is not significantly higher in large clades than in small clades (Fisher's exact test, P= 0.24).

Neither species range size nor ecology shows any association with clade diversification. Mean range size does not differ between large and small sister clades (P= 0.10); neither do mean elevation range (P= 0.63) nor mean number of habitats occupied (P= 0.13) differ.

The analysis of pairwise ΦST values illustrates the importance of species-level traits, rather than population or clade-level factors, in determining levels of genetic differentiation in these study species. In a nested ANOVA including species as an explanatory factor, differences between species account for 34% of variation in pairwise ΦST values. In a model including all potential explanatory variables, of the variation explained by the model (44% of the total variation in ΦST), 82.5% is explained by species-level traits and only 11% by population-level traits and 6.5% by clade-level traits. Model averaging shows that the only explanatory variables with no consistent effect (slope confidence intervals including zero) are geographic distance, elevation difference, and number of species in a clade. However, of the variables with consistent effects, only difference in branch circumference has a relative importance above 0.95, indicating that the best model is highly likely to include this trait: pairs of populations with a greater difference in branch circumference have significantly higher ΦST. Detailed results from the model averaging analysis are given in Table 5. The correlation between ΦST and difference in branch circumference did not remain significant using univariate Mantel tests (within-species correlations between −1 and 1, P-values between 0.15 and 1; Fisher's combined P= 0.35), but this may be a function of the small numbers of populations for each species.

Table 5.  Model averaging-based relative importance and slopes of explanatory variables for ΦST. Variables with a ‘*’ in the “Consistent” column have a 95% confidence interval for their slope that does not include 0, suggesting that they have a consistent effect on ΦST.
ParameterRelative importanceModel averaged slope95% CI lower bound for slope95% CI upper bound for slopeConsistent
  1. circ.diff = difference in average branch circumference between compared populations; circ.var = variance within a species in branch circumference; elev.range = elevation range of a species; geo.dist = geographic distance between compared populations; disturbed.areas = species occurrence in disturbed areas (yes or no); min.elev = minimum elevation at which a species occurs; circ.mean = mean branch circumference for a species; cloud.forest = species occurrence in cloud forest (yes or no); num.habitats = number of habitats occupied by a species; clade.size = clade size (big or small); num.fl.months = number of months in which flowering has been recorded for a species; clade.numspp = number of species in a clade; sp.numregions = number of TDWG Level 2 regions occupied by a species; elev.diff = elevation difference between compared populations.

circ.var 1.00 0.0024 0.0024 0.0025  
geo.dist 0.54 0.0100 −0.0655 0.0856  
min.elev 0.49 −0.0062 −0.0071 −0.0053  
cloud.forest 0.46 0.0024 0.0022 0.0025  
clade.size 0.35 0.0262 0.0054 0.0470  
clade.numspp 0.31 0.0345 −0.0052 0.0742  
elev.diff 0.14 0.0101 −0.0230 0.0431  


There is no support for the hypothesis that levels of neutral genetic differentiation within species control clade diversification. Across the species studied here, neutral genetic differentiation varies widely between species, is not phylogenetically heritable and does not differ significantly between species-poor and species-rich sister clades. However, there is weak evidence for a relationship between neutral genetic differentiation, species range size and diversification: high ΦST values indicating independent evolution of populations were observed only in widespread species from large clades. Some variation in neutral genetic differentiation can be explained by ecological traits, but even when including ecological covariates in a global model with population differentiation and clade size, there is no relationship between genetic differentiation and clade diversification.

The most conservative interpretation of these results is that levels of neutral genetic differentiation within species do not affect higher clade diversification in any way. Even though genetic differentiation is required for speciation, there are multiple situations in which it would not be expected to affect diversification rates. First, speciation rates may be limited more by other steps in the speciation process than by the development of genetic differentiation. For these orchids, for instance, speciation may be most limited by the rate at which species colonize new regions or by the diversity of potential pollinators available. Second, variation in diversification rates may be driven mainly by extinction rather than speciation. This is especially likely if new species tend to have small ranges or population sizes (e.g., with models of speciation as in Mayr 1982; Rosindell et al. 2010). Third, current patterns of diversity may be driven by variation in diversity limits (limits to the number of related species that can coexist in a region), rather than diversification rates (Rabosky 2009; Vamosi and Vamosi 2010; Kisel et al. 2011). Fourth, the level of genetic differentiation within a species may be too labile over evolutionary timescales to consistently affect diversification rates (perhaps like many population-level phenomena; Losos 2011). Finally, speciation in these orchids may be driven primarily by selection on a limited number of traits (whether through local adaptation, adaptation to new pollinators, etc.) in the face of gene flow, and thus may neither affect nor be affected by neutral population differentiation. One way to test this would be quantifying within- and between-population variation in morphological traits; in the case of the species studied here, a preliminary assessment of population differentiation in leaf shape (Y. Kisel, unpubl. data) showed no association between diversification and local adaptation.

There are also some scenarios in which a link between genetic differentiation and diversification might exist but not be measurable in a study such as this. First, average levels of genetic differentiation within species may be meaningful predictors of diversification only at higher taxonomic levels, where differences between clades outweigh the variability within them or where bigger differences exist between clades. This possibility could be tested by repeating this study with higher level clades and data for more species.

Second, it is possible that within the clades compared here all lineages do not contribute equally to clade diversification. In this case, even a strong link between genetic differentiation and speciation rates could be hidden by the lumping together of lineages with different characteristics. This possibility could be addressed by measuring the association between genetic differentiation and diversification rates using data for all species within a single clade and whole-tree analyses of diversification.

This highlights the main limitation affecting our study, namely the number of samples we could feasibly collect and analyze. Optimally, more species and sister-clade pairs would have been included. We attempted to select study species that were representative of their clades in ecology and morphology. However, it is possible that some study species do not represent the rest of their clade well in some traits that affect genetic differentiation, such as rates of selfing versus outcrossing (mating system). Mating system has been identified as the primary life-history trait affecting population genetic structure in plants, through its effects on both pollen-mediated gene flow and genetic drift through inbreeding (Duminil et al. 2007, 2009). In general, neutral genetic differentiation is higher in plants with higher rates of selfing (and thus less gene flow and more inbreeding and drift). Unfortunately, we lack the information to test whether variation in genetic differentiation among our species is associated with variation in mating system: most of the numerous species of epiphytic orchids are virtually unstudied beyond their taxonomy. Mating system has only been studied for one of our study species, Jacquiniella teretifolia, which is known to self-pollinate often (Hietz et al. 2006). Another possible factor affecting inferred genetic structure is if any study species are recent polyploids, as polyploidy complicates the estimation of population genetics parameters from genetic data (Obbard et al. 2006; Clark and Jasieniuk 2011). We used estimators of genetic differentiation designed to be calculated from diploid data; if any of our study species were polyploid, this would have added a large error component to our results. At this time, chromosome numbers are not known for any study species or their near relatives.

In addition, ideally at least five populations would have been sampled for each species, extending across the species range, and genetic data would have been available for at least 10 individuals for each population. Particularly for the hard-to-find species Lepanthopsis floripecten and Dryadella odontostele, limited sampling could mean that our data do not reflect the actual species population genetics well. These constraints could be addressed with collaborative efforts such as the IntraBioDiv consortium (Gugerli et al. 2008) or with meta-analyses of published population genetic data. However, even considering these limitations, we believe that the data we present here are sufficient to draw general conclusions about the study species we included. Population genetics studies, especially multispecies studies (e.g., Alsos et al. 2007; Alvarez et al. 2009; Skrede et al. 2009) commonly sample fewer than 10 individuals per population. Furthermore, we sampled populations representative of our study species’ usual ecology and habitat, and so the differentiation measures we estimated should be broadly representative of the general patterns for each species.

Another factor that could confound the signal for a correlation between genetic differentiation within species and diversification is the tendency for the speciation process itself to change levels of genetic differentiation. As genetic differentiation is an emergent trait at the species level (Jablonski 2008) and quantified by comparing populations, its level within species should generally be decreased by speciation in a manner analogous to species range size. This is because speciation is likely to divide the parent species range where population differentiation is highest, so that most speciation events result in daughter species with decreased average differentiation and decreased range size compared to the parent. As a result, younger species should tend to have both smaller ranges and less differentiation between populations, although the relationships between species age, range size, and differentiation will depend on how quickly ranges expand and differentiation develops after speciation (Pigot et al. 2010). In this case, although clades with a tendency for greater genetic differentiation might speciate more rapidly, new species would always show reduced population differentiation, weakening any observed relationship between genetic differentiation and clade diversity. Our results offer some support for this scenario: well-differentiated populations only occur in widespread species, which would make sense if widespread species are usually older than restricted species. Although species ages were not available to test this idea, such a complex scenario cannot be ruled out and would be worth exploring by studying further the relationships between population differentiation, species range size, and species age within more completely sampled clades. Diversification models such as QuaSSE (FitzJohn et al. 2009) and GeoSSE (Goldberg et al. 2011), which simultaneously estimate rates of trait evolution and diversification, as well as allowing traits to evolve differently at speciation events, might also provide a way forward.

It may also be the case that diversification is greater in clades with greater variation among species in levels of differentiation, rather than higher average within-species differentiation. In this scenario, clades with higher rates of evolution for traits affecting within-species genetic differentiation are expected to include more species with characteristics favoring differentiation and speciation, driving higher diversification rates for the clade as a whole. We did not have data for enough species to test this possibility effectively, but it would be an interesting analysis to try if population genetic data were available for most species from a set of closely related clades.

Finally, our results could have been biased against finding a link between genetic differentiation and diversification by the existence of cryptic species or incorrect species delimitations, which would cause us to record high levels of within-species differentiation erroneously. However, we believe this is highly unlikely. Only one of our study species, Brassavola nodosa, has ever had an infraspecific variant recognized. Caribbean populations in Costa Rica have also been known as B. nodosa var. grandiflora or B. grandiflora based on fragrance differences (Williams 1981) and slight morphological differences, but current taxonomic understanding is that all these populations represent B. nodosa, the same species as on the Pacific coast of Costa Rica (Dressler 2003). Furthermore, we found relatively low levels of genetic differentiation for this species (Table 3) rather than the high values expected with cryptic species or incorrect species delimitation.

One surprising finding is the lack of phylogenetic heritability for neutral genetic differentiation across these study species, given that previous studies have found a phylogenetic or taxonomic component for variation in population genetic differentiation. Early studies, for example, a review by Hamrick (1983), showed that plant species in the same broad taxonomic category (e.g., gymnosperm, monocot, or dicot) tend to have similar levels of population genetic structure. Duminil et al. (2007) confirmed these results with a similar taxonomic approach, using nested ANOVAs to show that almost 80% of the variation in population genetic differentiation among a range of plant species can be explained by phylogenetic effects at the genus, family, or order levels. Furthermore, they found a significant phylogenetic signal in the same data using Abouheif's (1999) test for serial independence. The main difference between these previous studies and ours is taxonomic scale: these studies compared species across a wide range of families and higher taxonomic ranks, whereas we compared relatively closely related species, all from the same tribe and subfamily. This may mean that neutral genetic differentiation is phylogenetically conserved only at a broad scale and not at the fine scale we investigated here. However, too few studies have investigated the phylogenetic signal of genetic differentiation so far to make any conclusions; further studies of this topic would be valuable.

Another striking finding is the lack of a significant relationship between genetic differentiation and geographic distance for any species, even those with high genetic differentiation. Orchids are capable of long-distance dispersal as a result of their abundant, tiny seeds (Arditti and Ghani 2000), and this could explain the lack of IBD. However, frequent long-distance dispersal should also lead to low population differentiation (Peterson and Denno 1998). One potential explanation is that orchids display a pattern similar to the “everything is everywhere but the environment selects” hypothesis for microorganisms (Baas-Becking 1934; Finlay 2002). This proposes that microorganisms are able to disperse anywhere because of their great abundance and long-distance dispersal ability, but that they only found new populations in suitable habitat patches and a single genotype tends to monopolize each patch. Under these circumstances, neutral genetic differentiation can be high but without any correlation with geographic distance (Fontaneto et al. 2008). Orchids might follow this model, as their seeds are on the same scale as microorganisms covered by the “everything is everywhere” hypothesis (< 2 mm), and single seed pods can contain millions of seeds (Arditti and Ghani 2000).

The finding that genetic differentiation tended to be greater between populations growing on different sized tree branches adds further support to the “everything is everywhere” hypothesis. Branch circumference has been identified as an important niche trait for epiphytic orchids because it is correlated with branch age and light and water availability (Chase 1988; Gravendeel et al. 2004). The relationship between branch circumference and differentiation could indicate that populations of these orchids differentiate ecologically by becoming specialized on particular branch types, rather than geographically as expected if distributions are limited by dispersal.

However, additional ecological data are needed to fully understand the links between the ecology and population genetics of these orchids. Data on pollinators and mycorrhizal fungi, neither of which are well known, would be particularly useful as both could be major drivers of orchid population structure (Waterman et al. 2011). The pollen dispersal potential of different orchid pollinators, for example, gnats versus hawkmoths, varies greatly. Additionally, all orchid seeds require a mycorrhizal fungal partner to germinate (Benzing 1987; Rasmussen 1995), and so orchid distributions could be limited by orchid species’ mycorrhizal specificity and the distributions of mycorrhizal partners (Swarts et al. 2010). Among the species studied here, only broad generalizations are possible: Brassavola species are pollinated by large sphingid moths (perhaps explaining low differentiation in B. nodosa); Lepanthes probably by fungus gnats, although pleurothallid orchids tend to be fly-pollinated; and Epidendrum species tend to be pollinated by Lepidoptera, although there is much variability in the group (Pijl and Dodson 1966; Cingel 2001). Mycorrhizae of most tropical orchid species have not been studied, but those that have been examined have not been shown to associate with specific fungal partners (Hadley 1970; Tupac Otero et al. 2002). For the study genera, multiple Epidendrum species tend to associate with one fungal genus, Epulorhiza (Zettler et al. 1998, 2007; Nogueira et al. 2005; Pereira et al. 2009); and in Masdevallia, for which mycorrhizal associations were investigated using roots sampled from the same plants used in this study, the widespread species Masdevallia nidifica (with strong genetic differentiation) appears to be more specific in mycorrhizal associations than the narrow endemic M. rafaeliana (with weak genetic differentiation; Renshaw 2010). More data on both interactions are needed to determine their role in explaining our findings. In addition, it would be useful to have more information about other factors known to limit epiphytic orchid distribution, such as light level, substrate type (e.g., bark or humus), host tree identity, position on host tree, local humidity, etc. (see Johansson 1975; Benzing 1987; Zotz and Hietz 2001; Burns and Zotz 2010). The variables we included in our analysis of variation in Φst, such as branch circumference and population elevation, correlate with some of these factors, but not all. It is likely that some of these additional ecological factors could prove to be strong predictors of population genetic differentiation if included in similar studies in the future.

To conclude, our study shows that comparative population genetics can be used to address macroevolutionary questions. Comparative population genetics studies so far have focused mainly on exploring the range of natural variation in population genetics and testing how much variation can be explained by species traits (Loveless and Hamrick 1984; Hamrick and Godt 1996; Morjan and Rieseberg 2004). As population genetic data accumulate and become easier to generate, potential applications for such data increase. Rather than exploring only factors that control population genetics, it is now possible to study how much population genetics control other processes. In this study, no association was found between population genetics and diversification, but unraveling complexities of the relationship between the two and in the role of population genetics in speciation theory is a research area worth exploring further.

Associate Editor: J. Vamosi


For invaluable help with fieldwork in Costa Rica, we thank J. Warner, F. Pupulin, M. Muñoz, Rafa, and Rei and the rest of the staff at the Jardin Botanico Lankester; SINAC staff including J. Guevara, R. Blanco, and O. Masis; UCR reserve director R. Sanchez; and the owners of Bosque de Paz Reserve and Rara Avis Reserve. We also thank F. Bonilla and C. Piedra for hosting YK and field assistants in Costa Rica, and we thank J. Hu, R. Phillips, M. Turjak, P. Renshaw, and K. Castillo for volunteering in the field. Thanks to V. Savolainen, R. Cowan, and O. Paun for help with the molecular laboratory work and to R. Butlin and A. Purvis for insightful and encouraging comments on an early version of the manuscript. Finally, we thank the Associate Editor and three anonymous reviewers for their help in improving the manuscript. YK was supported by a U.S. National Science Foundation Graduate Research Fellowship, a Deputy Rector's Award from Imperial College London and fieldwork grants from Sigma Xi, the Bentham-Moxon Trust at RBG Kew, National Science Foundation GRFP, and the Central Research Fund of the University of London.