Genetic and chemical divergence among host races of a socially parasitic ant

Abstract Host–parasite associations facilitate the action of reciprocal selection and can drive rapid evolutionary change. When multiple host species are available to a single parasite, parallel specialization on different hosts may promote the action of diversifying natural selection and divergence via host race formation. Here, we examine a population of the kidnapper ant (Polyergus mexicanus) that is an obligate social parasite of three sympatric ant species: Formica accreta, F. argentea, and F. subaenescens (formerly F. fusca). Behavioral and ecological observations of P. mexicanus have shown that individual colonies parasitize only one species of host and that new Polyergus queens maintain host fidelity when establishing new colonies. To successfully adapt to a particular host, Polyergus ants may mimic or camouflage themselves with the species‐specific chemical cues (cuticular hydrocarbons) that their hosts use to ascertain colony membership. To investigate the extent of host specialization, we collected both genetic and chemical data from P. mexicanus that parasitize each of the three different Formica species in sympatry. We show that host‐associated genetic structure exists for both maternally inherited mitochondrial DNA data and biparentally inherited microsatellite markers. We also show that P. mexicanus can be distinguished by chemical profile according to host due to partial matching with their host. Our results support the hypothesis that host race formation is occurring among lineages of P. mexicanus that use different Formica hosts. Thus, this system may represent a promising model for illuminating the early steps of divergence, accumulation of reproductive isolation, and speciation.


Authors
Torres, Candice W Tonione, Maria A Ramírez, Santiago R et al.
Unlike conventional ecto-parasites and endo-parasites, social parasites take advantage of the brood care provided by the host species. Avian brood parasites, for example, exploit the rearing behavior of other birds by laying eggs in their hosts' nests and provide convincing examples of host race formation (i.e., gentes, Gibbs et al., 2000;Marchetti et al., 1998) and even speciation in sympatry (Sorenson, Sefc, & Payne, 2003). Less well studied are the social parasites that exploit and manipulate entire colonies of ants, bees, wasps, or termites.
Social insect colonies provide ample opportunities for the evolution of cheating and parasitism. Thus, it is not surprising that at least 230 species of ants have evolved to be social parasites (Buschinger, 2009). One challenge of the parasitic lifestyle is the necessity for sufficiently abundant hosts (Foitzik & Herbers, 2001). The use of multiple different host species is one way to overcome this difficulty. However, host generalist species may experience lower efficacy when exploiting multiple hosts because traits that enable the parasite to use one host effectively may decrease its ability to use other hosts (Bauer, Bohm, Witte, & Foitzik, 2010;Guillem, Drijfhout, & Martin, 2014). For example, the precise colony-mate recognition systems that have evolved in social insects and have been refined by long-term coevolution between host and parasite lineages are a formidable barrier for some social parasites (Bonavita-Cougourdan, Provost, Riviere, Bagneres, & Dusticier, 2004; D'Ettorre, Mondy, Lenoir, & Errard, 2002;Lenoir, D'Ettorre, Errard, & Hefetz, 2001).
Thus, social parasites are likely to experience an evolutionary tradeoff between selection mediated by host scarcity (which should drive the evolution of generalist parasites) and selection mediated by host defenses (such as colony-mate recognition), which should favor specialization by the parasite on a narrow range of hosts.
Kidnapper ants in the genus Polyergus (also known as Amazon ants, slave-making ants, slave-raiding ants, or pirate ants) are obligate social parasites that rely on their closely related Formica hosts to perform all colony tasks including brood care, nest maintenance, defense, and foraging (Topoff, Cover, & Jacobs, 1989;Trager, 2013). There appears to be high host fidelity from generation to generation because new Polyergus colonies are initiated when newly inseminated Polyergus queens accompany Polyergus workers on a raid of a nearby Formica colony (Topoff et al., 1989;Trager, 2013). Although this reproductive life history could promote maternal differentiation (e.g., in mtDNA), the extent of assortative mating, which could promote genomewide differentiation, remains unknown. The young Polyergus queen takes over the Formica colony by killing and replacing the resident Formica queen (Topoff, Cover, Greenberg, Goodloe, & Sherman, 1988).
The Formica workers in this usurped colony then assist the new Polyergus queen in raising her first cohort of offspring. Eventually, these Polyergus workers begin to conduct raids on neighboring Formica colonies of the same species as the originally usurped colony, kidnapping Formica pupae, which then eclose in the Polyergus colony. This process replenishes the population of host workers in the Polyergus colony (Bono, Blatrix, Antolin, & Herbers, 2007;Topoff, LaMon, Goodloe, & Goldstein, 1985). Any single Polyergus colony only raids colonies of a single host species, even when other suitable host species are available (Bono et al., 2007;Goodloe, Sanwald, & Topoff, 1987;King & Trager, 2007), with the exception of the European P. rufescens (Trager, 2013). Therefore, there is likely to be a high degree of host fidelity from generation to generation within these lineages of Polyergus.
Here, we examine the molecular and chemical ecology of the socially parasitic kidnapper ant, P. mexicanus (formerly P. breviceps), which parasitizes three sympatric species of Formica at our study site: F. accreta, F. argentea, and F. subaenescens. Based on the apparent vertical transmission of host/parasite fidelity, we predict that the three lineages of P. mexicanus using these different hosts will exhibit patterns associated with host specialization, host race formation and, possibly, reproductive isolation. Specifically, we first test the hypothesis that lineages of P. mexicanus that use different host Formica will display significant genetic differentiation from each other. We test this hypothesis using both maternally inherited (mtDNA) and biparentally inherited (microsatellite) genetic markers.
Next, we test the hypothesis that these Polyergus lineages also display significant phenotypic differentiation from each other in the pheromones used for colony recognition (cuticular hydrocarbons).
We address this hypothesis by extracting cuticular hydrocarbons from field-collected ants and analyzing them using gas chromatography-mass spectrometry (GC-MS).

| Field site and collection information
We conducted this study at Sagehen Creek Field Station, a University of California Natural Reserve located 13.5 km north of Truckee, CA ( Figure 1). At this site, P. mexicanus parasitizes at least four different species of Formica in sympatry, including F. subaenescens, F. argentea, F. accreta, and, rarely, F. neorufibarbis (P.S. Ward, personal communication). During the summers of 2008, 2009, and 2010, we collected both Polyergus mexicanus workers and Formica host samples from 18 colonies (Supporting Information Appendix S1). For five of these colonies, the host colonies were identified as F. argentea, five were identified as F. subaenescens (formerly F. fusca), and eight were identified as F. accreta (but see Section ). In addition, we included ants collected from a P. mexicanus colony collected at Blue Canyon Lake (BCL), Tuolumne Co., CA, located 130 km south of Sagehen Creek Field Station (Figure 1; Supporting Information Appendix S1). Host species were identified using the morphological key for the Formica fusca group developed by Francoeur (1973).

| Genetic analysis
We extracted whole genomic DNA from the head or a single leg of P. mexicanus and Formica workers using the DNeasy Blood and Tissue Kit (Qiagen, Valencia, CA), following the recommended protocol and eluting the DNA in 200 μl of buffer AE. To determine whether P. mexicanus shows population structuring according to maternally inherited DNA, we amplified and sequenced two fragments of the mitochondrial gene, cytochrome oxidase I (COI) and NADH dehydrogenase subunit 1 (NADH1), from one or two individuals of each genus (Polyergus and Formica) from each colony.
To amplify NADH1, we used the primers from Liautard and Keller The thermocycler conditions were as follows: a 3-min initial denaturation of 94°C followed by 38 cycles of 45 s at 94°C, 45 s at 49°C and 1 min at 72°C, and ending with a 10 min extension step at 72°C.
We verified the amplification of PCR products on 1% agarose gels and performed PCR cleanup using ExoSAP-IT (USB). To sequence, we added approximately 33 ng of DNA from the purified PCR product, 0.46 pmol of primer, and 4 μl of BigDye Terminator v3.1 (ABI), and the following temperature program was used: 96°C for 1 min then 25 cycles of 96°C for 10 s, 50°C for 5 s, and 60°C for 4 min. PCR products were sequenced using a 96 capillary 3730xl DNA analyzer. We edited and aligned sequences using the program Geneious version 9.1.5 (Kearse et al., 2012). The final alignment including all P. mexicanus and Formica individuals and outgroup (specified below) was 648 bp of COI and 173 bp of NADH1.
We used the CIPRES Science Gateway (Miller et al., 2010) and MrBayes v 3.2.2 (Ronquist & Huelsenbeck, 2003) to build a single Bayesian majority rule consensus tree inferred from a combined total of 841 bp of COI and NADH1 from both P. mexicanus and Formica spp. hosts. We obtained Camponotus chromaiodes sequence from GenBank (COI: AY334392.1, NADH1: JX966368.1) as an outgroup to root the phylogeny. We determined the best fit nucleotide substitution model by Akaike information criterion (AIC) in PartitionFinder 1.1.1 (Lanfear, Calcott, Ho, & Guindon, 2012). The data were ana-   When testing for HWE at the levels of colony and host, we found no loci consistently in equilibrium, likely due to population fragmentation across colonies and the local scale of population sampling in this study.
For the entire population of P. mexicanus, the average number of alleles per locus per colony was 2.56 ± 0.48 (SD), and the average expected heterozygosity was 0.39 ± 0.08. The average number of private alleles per colony was 0.20 ± 0.21. The average fixation index (F) across all loci was 0.181 ± 0.16. When analyzing P. mexicanus using the species of host as a population unit, the average fixation indices per locus were as follows: 0.023 ± 0.34 (F. accreta), 0.027 ± 0.21 (F. argentea), and 0.060 ± 0.34 (F. subaenescens).
We found all loci were linked across all possible locus pairings when considering all sampled P. mexicanus as one population. Therefore, we designated populations according to host and then viewed occurrence of LD across all possible locus pairs to see whether LD in the whole population may be a result of population structuring according to host. P. mexicanus parasitizing F. accreta showed 95%, P. mexicanus parasitizing F. argentea showed 100%, and P. mexicanus parasitizing F. subaenescens showed 86% LD. Like tests for HWE, LD is also sensitive to demographic events, particularly coancestry (Kaeuffer, Reale, Coltman, & Pontier, 2007), which is likely within our population.
To determine whether P. mexicanus colonies are distinguishable genetically according to host, we performed two types of analysis on microsatellite data that examine population clustering: one in the program STRUCTURE v2.3.3 (Falush, Stephens, & Pritchard, 2003) and the other a discriminate analysis of principal components (DAPC) in the program R. In STRUCTURE, we explored a range of possible numbers of population clusters (K) from 2 to 10 (the total number of colonies sampled) using a burn-in length of 50,000 followed by 100,000 Markov chain Monte Carlo (MCMC) repetitions. We used both the admixture and the correlated allele frequency models under default settings. STRUCTURE has two options that allow the user to set up prior populations of origin using "popflag" and "popinfo." During initial runs with and without these options, we determined there were no detectable differences between them. Here, we present results from runs with both population options turned on and the species of host parasitized by the genotyped P. mexicanus individual set as prior "populations." Since we were particularly interested in whether P. mexicanus at our site were genetically grouped according to host, we performed an additional 10 runs each of K = 3 (for the three species of hosts identified morphologically) and K = 4 (to account for a possible additional cryptic host indicated in the mtDNA results, see below) at 100,000 burn-in length and 1,000,000 MCMC repetitions. We summarized the clustering patterns found in our runs of K = 3 and K = 4 using CLUMPP v1.1.2 (Jakobsson & Rosenberg, 2007) and visualized them using DISTRUCT v1.1 (Rosenberg, 2004).
If the assumptions of STRUCTURE are not met (i.e., departures from HWE and LD are not associated with population structure but instead with inbreeding or scoring errors), STRUCTURE may oversplit a population (Pritchard et al., 2010). Therefore, we also performed a DAPC analysis in the program R using the function adegenet 1.3-4 (Jombart, 2012). DAPC is a multivariate analysis method that combines the advantages of principal components analysis (PCA) and discriminant analysis (DA) to determine assignment of individuals to genetic clusters (Jombart, Devillard, & Balloux, 2010). Because this method transforms genetic data using PCA, the assumptions of HWE and no LD do not need to be met to explore genetic clustering.
DAPC requires groups be defined prior to conducting the analysis and biologically defined groups are recommended as the most useful way of examining group membership (Jombart, 2012). As such, clusters were defined according to species of host P. mexicanus workers were found with before implementing the function dapc in the adegenet package.

| Chemical data collection and analysis
We analyzed cuticular hydrocarbon profiles from 18 to 20 individual Polyergus and Formica hosts from each of 19 colonies. After collecting P. mexicanus and Formica workers from the field, we freeze-killed the specimens and soaked each individual for 10 min in 200 μl of chromatography grade hexanes in 9 mm glass GC vials (Agilent Technologies, Santa Clara, CA). Ants were then removed from the solvent, allowed to air-dry, and were transferred into 95% ethanol for genetic analysis (see previous section). The hexane extracts were stored at −20°C and kept on dry ice for transport back to the labora- We performed a nonmetric multidimensional scaling (NMDS) analysis on the relative peak areas of P. mexicanus profiles using the package vegan and the function metaMDS in the program R (https:// cran.r-project.org, v2.14.0). Chemical peaks that were not present or that could not be detected by the ChemStation integrator were counted as having an area of zero. We used the percent peak area to calculate a Bray-Curtis pairwise distance matrix and performed an analysis of similarity (ANOSIM) on this matrix to test for possible differences in the chemical profiles according to host using the function anosim in R. Next, we combined the chemical data from P. mexicanus with the data of their Formica hosts into a single NMDS analysis to see whether P. mexicanus workers clustered more closely with their resident host than with other available host species.
In total, we analyzed the relative abundance of 105 chemical peaks across all P. mexicanus and Formica individuals. To determine which of these 105 peaks most likely contributed to any separation of P. mexicanus by hosts, we performed a SIMPER (similarity percentage) analysis using PRIMER 6 (Clarke, 1993), which is based on decomposition of the Bray-Curtis dissimilarity index.

| Mitochondrial sequence data
The host (Formica) mitochondrial tree resolved four major monophyletic groups rather than three that matched the nominal species ( Figure 2, right). Two of the host species, F. argentea and F. subaenescens, were recovered as monophyletic groups. However, our samples of the third host species (F. accreta) were divided between two separate, well-supported clades. In the codon-partitioned tree, one of these (F. accreta "B") was placed as sister to F. subaenescens, whereas the other (F. accreta "A") was recovered as a clade with an unresolved relationship relative to F. argentea and F. accreta "B" + F. subaenescens (Figure 2, right). These same four groups were recovered in the gene-partitioned tree, but with F. accreta "A" placed as sister to the other three clades. Notably, two of the Formica specimens that fell within the F. accreta "A" (499_F and 537_F) were initially identified as F. subaenescens based on morphology.
The Formica sample from Blue Canyon Lake ("BCL_F" in Figure 2), which was collected ~130 km south of Sagehen Creek Field Station, was placed within F. accreta "A." Based on morphology, this specimen was initially identified as F. cf. argentea (P.S. Ward, pers. comm.).
Similar to the pattern observed for the Formica hosts, our reconstruction of the intraspecific relationships among Polyergus parasites yielded four well-supported mitochondrial lineages (Figure 2, left).
Remarkably, these Polyergus clades perfectly matched the four ob- Formica also fell with the corresponding clade of F. accreta "A"-enslaving Polyergus (Figure 2, left).

| Microsatellite data
The STRUCTURE analysis of P. mexicanus revealed an overall pattern of structuring that matched the morphological host species identification (at K = 3, Figure 3a). The division of Polyergus using F. accreta "A" and "B" as host that was seen in the Polyergus mitochondrial tree (Figure 2, left) was not evident at microsatellite loci ( Figure 3a).
Likewise, the DAPC plot showed a pattern of clustering of P. mexicanus according to host species (Figure 3b)

| Chemical analysis
Qualitatively, the chemical profiles of the three species of Formica parasitized by P. mexicanus showed species-specific differences  100 exception was F. accreta "B" colony P9 that appeared more similar to F. subaenescens than to F. accreta "A" (Figure 5a, red x symbols), a relationship reminiscent of the mtDNA relationships (Figure 2; right).
There was no significant chemical differentiation when the analysis was performed at the colony level ( Figure 6). This may be due, in part, to the fact that Formica from each Polyergus colony actually originate from a variety of different natal colonies, thus leading to high diversity among individuals from a single colony's enslaved host population.
There also appeared to be qualitative differences among the chemical profiles of P. mexicanus parasitizing the three different species of Formica (Figure 4,   the host species (Figure 5b), and the results from the ANOSIM analysis of individual chemical profiles confirmed these host-specific differences to be significant (R = 0.678, p = 0.001). Colony-level ANOSIM revealed significantly lower CHC diversity within versus among colonies (R = 0.876; p = 0.001), but there was no clear association among colonies that used the same host species (Figure 6).
Interestingly, and unlike the pattern seen in the Formica host, the P. mexicanus workers from colony P9 (host: F. accreta "B") possessed chemical profiles that were most similar to the rest of the Polyergus enslaving F. accreta "A" (red x symbols in Figure 5b). When all parasitized Formica were included with their parasites in a single NMDS analysis, P. mexicanus individuals tended to group with their host species, with the exception of P. mexicanus parasitizing F. argentea

| Overview
In contrast to null expectation of panmixia, we found extensive genetic differentiation among sympatric colonies of P. mexicanus social parasites, at both mitochondrial loci and nuclear microsatellites.

Remarkably, this differentiation matched host use, suggesting that
Polyergus host races have evolved to specialize on different host Formica. Consistent with this, samples from a geographically distant site grouped with the appropriate host race, rather than sister to all Sagehen Creek samples, as would be expected under a typical pattern of genetic isolation by distance. However, different patterns were evident between the mtDNA and nuclear microsatellites, suggesting different patterns of maternal versus paternal gene flow or different rates of molecular evolution. Analysis of chemical phenotypes that are crucial for ant social behaviors (cuticular hydrocarbon pheromones; CHCs) produced mixed results. When all individuals were included separately, Polyergus colonies that used different Formica hosts were clearly differentiated. However, these differences were not observed when the analysis was performed with individuals grouped into colonies.
Overall, these data, combined with the known maternal vertical transmission of host species identity during colony founding, suggest that differentiation among lineages that use different hosts has led to the evolution of sympatric Polyergus host races. In addition, differentiation of P. mexicanus with respect to host species was observed at the nuclear microsatellites, suggesting that assortative mating may also be occurring. Interestingly, these host races of P. mexicanus are cryptic lineages, showing no obvious F I G U R E 4 Representative chemical profiles by GC-MS from one Polyergus mexicanus (bottom of each chromatogram) and its respective host species from the same colony (top of each chromatogram)  Trager, personal communication). In future studies, it would be useful to assess the extent of reproductive isolation among these lineages and determine whether they represent the formation of incipient species in sympatry.

| Genetic differentiation among host races
The genetic differentiation that we found among Polyergus lineages is consistent with the evolution of host specificity and high host fidelity, leading to reduced gene flow among parasites that use different hosts (Archie & Ezenwa, 2011;Criscione et al., 2005).
F I G U R E 5 NMDS of chemical profiles of (a) enslaved Formica, (b) P. mexicanus social parasites, and (c) both host and parasite together. Each symbol represents an individual ant; Formica host species and Polyergus host races coded by color. Hosts and parasites from colony "P9" indicated by red "x" symbols The difference that we observed between mtDNA and microsatellites with respect to the F. accreta host race may indicate different patterns of male versus female gene flow. The natural history of Polyergus colony founding suggests a mechanism for this difference: New queens disperse locally, on foot, into colonies of the same host species as their natal colony, whereas males fly in from more distant surrounding colonies that potentially use a different host species. Although the pattern of differentiation was not identical between mtDNA and microsatellites, the clear nuclear differentiation among Polyergus using different Formica hosts indicates that male-mediated gene flow is not occurring randomly. Instead, it appears that some additional form of reproductive isolation, such as assortative mating, is also occurring. The exact dynamic of male versus female gene flow across host species remains unknown, but is likely to be an important force in determining the extent of host specialization.
An additional mechanism that might further limit gene flow is assortative mating between kidnapper ant sexuals, as P. mexicanus queens and/or males might preferentially select mates from colonies that parasitize the same Formica species as their natal nest. In P. mexicanus, queens disperse on foot and release pheromones from their mandibular glands to attract males for mating (Greenberg et al., 2007;Greenberg, Aliabadi, McElfresh, Topoff, & Millar, 2004).
The specificity of this queen sex pheromone is unknown, and little is known about the details of mating behavior in P. mexicanus.
Cuticular hydrocarbons may also play an important role in mate choice (Howard, Jackson, Banse, & Blows, 2003). For example, a study by Beibl, Buschinger, Foitzik, and Heinze (2007) found that sexuals of the socially parasitic ant, Chalepoxenus mullerianus, obtain some of their CHCs from their host workers. This type of "gestalt" model for CHC-sharing among nestmates is well known in other ant species (Crozier & Dix, 1979;Soroker, Vienne, & Hefetz, 1995). In this way, in host races could be a powerful approach for testing the potential roles of these pheromones in reproductive isolation.
Another potential (but not mutually exclusive) driver of differentiation could be postzygotic selection against "hybrid" kidnapper ant colonies produced from the matings of sexuals originating from nests that parasitized two different hosts. For example, worker offspring of such crosses might be less effective at raiding the maternal host species due to reduced CHC matching or the expression of aberrant behaviors during raids. Although we did not discover any apparent hybrids in this study, more extensive sampling might reveal such colonies.

| Chemical adaptation
The total dependence of P. mexicanus on Formica likely imposes selection on the parasite to elude the recognition system of its hosts. Mimicry of host recognition cues, such as CHCs, is a common adaptive strategy used by such social parasites (Bagnéres and Lorenzi, 2010). Because CHC cues are often species-specific (Emery & Tsutsui, 2016;Howard, 1993;Martin, Helantera, Kiss, Lee, & Drijfhout, 2009), we predicted that P. mexicanus specializing on a particular host species might have chemical profiles matching their host species and that this would result in P. mexicanus colonies becoming distinct from one another according to host species.
We observed patterns consistent with this when all individuals were analyzed together without taking into account their colony of origin.
Chemical profiles of P. mexicanus workers generally clustered together to form three groups corresponding to one of the three species of host they parasitize. However, this pattern was not evident in our analysis at the colony level. Although odor sharing between host and parasite may account for some of this grouping, the distinctly different patterns displayed by the F. accreta "B" host versus the  to two of the studies mentioned above (Brandt et al., 2005;Nash et al., 2008), where chemical specialization and separation by host species occurring in sympatry was not apparent. Further studies that examine the cuticular chemical profiles of free-living Formica from our field site and nearby populations of P. mexicanus that parasitize only one host would help clarify the extent of this chemical adaptation.

| Formica accreta "B"
Interestingly, both hosts and parasites from the F. accreta "B" clade displayed conflicting patterns. The hosts in this colony, morphologically identified as F. accreta, were more similar to F. subaenescens in mtDNA sequence and CHC phenotype. Further study may reveal that this lineage is a cryptic, undescribed species of Formica.
Similarly, the P. mexicanus that parasitize F. accreta "B" formed a distinct and well-supported monophyletic group based on mtDNA but, at nuclear loci, grouped with the other Polyergus that used F. accreta as host. It is difficult to determine exactly how this lineage originated. One possibility is that different patterns of maternal versus paternal gene flow have produced different patterns at nuclear and mitochondrial loci. Future studies may resolve this mystery by sampling from a broader geographic area and applying more extensive genetic and genomic analyses.

| Host race formation and incipient speciation
At present, we cannot determine whether the patterns observed here represent the occurrence of cryptic species that occur in sympatry (secondary contact) or sympatric speciation. Providing clear geographic and population genetic evidence for speciation and host race formation in sympatry is challenging (Fitzpatrick, Fordyce, & Gavrilets, 2008;Via, 2001). Support for host race formation in ant social parasites has not been previously reported and has been rarely examined (but see Fanelli, Henshaw, & Cervo, 2005). Because our genetic data indicate restricted biparental gene flow among the three P. mexicanus host races within a small geographic area, a tantalizing conclusion is that P. mexicanus at our study site is forming host races and undergoing the first steps of speciation in sympatry. However, it is also possible that the overlap of these three host races at our field site represents secondary contact among lineages that diverged elsewhere. Examination of populations across a larger geographic area and other possible host associations will clarify the evolutionary pathways that led to this distribution of sympatric host races.
Strong divergent selection on a particular trait may jumpstart the speciation process and result in reduced gene flow between subpopulations (Nosil et al., 2009). Although our results clearly reveal patterns of genetic divergence consistent with this differentiation, the phenotypic traits driving this process are not yet clear. Host race formation may be viewed as an early stage in the process of speciation, with incomplete genetic isolation and development of complete reproductive barriers not yet fully established (Drès & Mallet, 2002). Studies that focus on these early stages of differentiation are crucially important for understanding the speciation process, in part because differences that arise later become confounded with the actual early drivers of differentiation (Via, 2009). Thus, studies such as ours serve as important starting points for increasing our understanding of the conditions by which new species form.

ACK N OWLED G M ENTS
We thank Jan Buellesbach, Jon Tanaka Jeff Brown for hosting us there. We also thank three anonymous reviewers whose comments greatly improved this manuscript.

AUTH O R CO NTR I B UTI O N S
CWT designed the experiment, collected the data in the field and the laboratory, and performed data analysis. MAT collected and analyzed mtDNA sequence data. SRR assisted with data analysis.
JRS performed fieldwork and contributed samples. NDT contributed to experimental design and assisted with data analysis. CWT and NDT designed figures and wrote the manuscript.

DATA ACCE SS I B I LIT Y
Voucher specimens (thorax and abdomen) deposited in the Bohart Museum of Entomology, UC Davis. DNA sequences: GenBank Accession numbers MH633506-MH633555, MH645523-MH645572.