HYBRID ZONE ORIGINS, SPECIES BOUNDARIES, AND THE EVOLUTION OF WING-PATTERN DIVERSITY IN A POLYTYPIC SPECIES COMPLEX OF NORTH AMERICAN ADMIRAL BUTTERFLIES (NYMPHALIDAE: LIMENITIS)

Authors

  • Sean P. Mullen,

    1. Department of Ecology and Evolutionary Biology, Corson Hall, Cornell University, Ithaca, New York 14853
    2. Department of Biological Sciences, Iacocca Hall, Lehigh University, Bethlehem, Pennsylvania 18015
    3. Department of Organismal and Evolutionary Biology, Harvard University, Cambridge, Massachusetts 02138
    Search for more papers by this author
  • Erik B. Dopman,

    1. Department of Ecology and Evolutionary Biology, Corson Hall, Cornell University, Ithaca, New York 14853
    2.  E-mail: sem307@lehigh.edu
    Search for more papers by this author
  • Richard G. Harrison

    1. Department of Ecology and Evolutionary Biology, Corson Hall, Cornell University, Ithaca, New York 14853
    Search for more papers by this author

Abstract

Hybrid zones present opportunities to study the effects of gene flow, selection, and recombination in natural populations and, thus, provide insights into the genetic and phenotypic changes that occur early in speciation. Here we investigate a hybrid zone between mimetic (Limenitis arthemis astyanax) and nonmimetic (Limenitis arthemis arthemis) populations of admiral butterflies using DNA sequence variation from mtDNA and seven nuclear gene loci. We find three distinct mitochondrial clades within this complex, and observe a strong overall concordance between wing-pattern phenotypes and mitochondrial variation. Nuclear gene genealogies, in contrast, revealed no evidence of exclusivity for either wing-pattern phenotype, suggesting incomplete barriers to gene exchange and/or insufficient time for lineage sorting. Coalescent simulations indicate that gene flow between these two subspecies is highly asymmetric, with the majority of migration occurring from mimetic into nonmimetic populations. Selective sweeps of alleles responsible for mimetic phenotypes may have occurred more than once when mimetic and nonmimetic Limenitis occurred together in the presence of the model (Battus philenor).

Studies of hybridization between closely related taxa provide important insights into the nature of species and the process of speciation (Barton and Hewitt 1985; Hewitt 1988; Barton and Hewitt 1989; Harrison 1990, 1993; Jiggins and Mallet 2000; Barton 2001; Howard et al. 2003). The causes and consequences of hybridization have been most widely studied in the context of hybrid zones, areas of interaction between genetically distinct groups of individuals that result in the production of at least some offspring of mixed ancestry (Harrison 1990). Hybrid zones are frequently narrow relative to the geographic distributions of the parental populations, and typically associated with steep clines in phenotypic and genetic variation (Barton and Hewitt 1985, 1989).

Zones of hybridization can arise via either primary intergradation or secondary contact (Mayr 1942). Primary hybrid zones result from differentiation among a series of connected populations that can lead to striking, and often concordant, clinal variation in multiple characters. Hybrid zones of secondary origin, in contrast, form when populations that have diverged in allopatry subsequently come back into contact. As Barton (2001) noted, hybridization can affect divergence in a number of ways (e.g., reinforcement, hybrid speciation, homogenization, or adaptive divergence), and the ultimate outcome of hybridization for any given species pair will largely depend on the balance between selection and dispersal. Thus, hybrid zones are “natural laboratories” (Hewitt 1988) uniquely suited to investigate the impact of gene flow on the formation and maintenance of species boundaries during divergence.

North American admiral butterflies (genus Limenitis) represent a small but phenotypically diverse radiation of nymphalid butterflies that consists of several geographically distinct species and wing-pattern forms (Platt 1983). Although best known from studies of the evolution of mimicry (Poulton 1909; Brower 1958a,b; Platt et al. 1971; Platt 1975; Ritland 1991; Ritland and Brower 1991; Ritland 1998; Prudic et al. 2002), these butterflies have also, historically, been of interest because hybridization between nominal species is widespread (see Platt 1983 for review). In addition, several stable hybrid zones between wing-pattern races have been previously described (Platt and Brower 1968; Remington 1968; Porter 1989, 1990; Boyd et al. 2000), the most dramatic example involving hybridization between mimetic and nonmimetic populations of the polytypic Limenitis arthemis-astyanax species complex (Fig. 1; Platt and Brower 1968; Remington 1968).

Figure 1.

Wing pattern diversity in the Limenitis arthemis species complex. Specimens in the left-most column are nonmimetic L. a. arthemis, specimens on the far right are mimetic L. a. astyanax, and intermediate individuals representing the range of variation observed in hybrid populations are shown in column 2.

WING-PATTERN VARIATION AND SPECIES BOUNDARIES IN THE L. ARTHEMISASTYANAX COMPLEX

White-banded admiral butterflies (L. arthemis arthemis) occur in the northeastern United States and throughout Canada as far west as Alaska (Scott 1986). These butterflies are characterized by broad, white bands that traverse both the dorsal and ventral surfaces of the wing and by a series of submarginal red spots. In contrast, Red-spotted Purples (L. arthemis astyanax), which are Batesian mimics of the Pipevine Swallowtail (Battus philenor; Brower and Brower 1962; Platt et al. 1971; Platt 1975), lack any white band and possess vibrant blue to blue-green iridescent scales along the outer portion of the hindwing. These butterflies are distributed throughout the southeastern United States. Hybridization between the two forms occurs in a narrow geographic band across New England, southern Ontario, and the Great Lakes, and individuals with intermediate phenotypes have been known for at least 140 years (Edwards 1865, 1877; Scudder 1889).

Significant geographic variation in wing pattern also exists within both white-banded and mimetic populations. For example, white admiral butterflies west of Lake Superior have a brick red band along the margins of the ventral hindwing, and these northwestern populations are sometimes considered a separate subspecies, L. a. rubrofasciata (Barnes and McDunnough 1916). In addition, isolated populations of a mimetic subspecies (L. a. arizonensis) occur in the southwestern United States, and differ from L. a. astyanax in the shape of their hindwings (more pointed) and color of their submarginal chevrons (white vs. iridescent). Given the extensive geographic variation in wing pattern in this group and the phenotypic evidence for hybridization between the two eastern wing-pattern forms, it is not surprising that species boundaries in this complex have long been debated (Grey 1879; Scudder 1889; Edwards 1891; Field 1904, 1914; Nakahara 1924; Gunder 1932; Hovanitz 1949; Ford 1953; Fisher 1958; Remington 1958; Platt and Brower 1968; Remington 1968; Collins 1991).

Remington (1968) noted that L. a. arthemis and L. a. astyanax hybridize in a “suture-zone” that corresponds to the geographic position of numerous other hybrid zones (see also Swenson and Howard 2004). The larval forms of these butterflies feed predominantly on tree species in the plant family Salicaceae (e.g., aspen, poplar, willow), many of which are restricted to either southern deciduous forests or northern boreal forests. The transition between these two forest biomes occurs across the geographic “suture-zone” identified by Remington (1968) and is entirely coincident with the phenotypic hybrid zone between L. a. arthemis and L. a. astyanax. Remington further observed that in addition to using hosts in the family Salicaceae, northern, nonmimetic populations of L. a. arthemis also feed extensively on yellow birch (Betula alleghaniensis) whereas southeastern populations of L. a. astyanax frequently use larval host plants in the family Rosaceae (e.g., black cherry, Prunus serotina). Based on these observations, Remington (1968) argued that hybridization between the two eastern wing-pattern forms of the Limenitis arthemis–astyanax complex was the result of secondary contact between two lineages that had diverged with respect to host plant use. Platt and Brower (1968), in their landmark study, however, found no evidence of assortative mating between L. a. arthemis and L. a. astyanax based on Hardy–Weinberg expectations for wing-pattern phenotypes/genotypes. They concluded that the hybrid zone was primary in origin, therefore, and due to an abrupt reversal of selective forces for mimicry and disruptive coloration (see Silberglied et al. 1980; Cuthill et al. 2005 for a discussion of disruptive coloration).

The debate over the origins of the hybrid zone and species boundaries in this complex stems largely from a lack of knowledge about the relationship between patterns of genetic and phenotypic variation. Specifically, it is unclear whether the extensive racial variation in wing pattern observed among the described subspecies reflects a history of population differentiation in allopatry or a response to current geographic variation in selection pressures related to mimicry. The model in this system, B. philenor, occurs throughout the southeastern and southwestern United States, and the ranges of the mimetic subspecies (L. a. astyanax and L. a. arizonenesis) closely overlap the range of Battus (Brower and Brower 1962). To differentiate alternative hypotheses (primary intergradation vs. secondary contact), we examined patterns of mitochondrial DNA variation from populations spanning the entire geographic range of this polytypic species, and used these data to evaluate genetic population structure and look for evidence of current introgression. We then reconstructed the evolutionary relationships among the described subspecies, using both mitochondrial and nuclear gene genealogies, to make inferences about the origins of the hybrid zone and the evolution of mimicry. Finally, we used a coalescent-based model to estimate time since divergence and historical rates of gene flow between the two eastern wing-pattern races (L. a. arthemis and L. a. astyanax). We discuss our results in the context of the speciation process, recent examples of “divergence with gene flow” (e.g., Emelianov et al. 2004; Turner et al. 2005), and claims about the “reality” of species (Mallet et al. 2007).

Methods

SAMPLES AND SEQUENCES OBTAINED

Specimens of Limenitis were sampled from 39 populations across the geographic range of this complex (Table 1), including multiple populations of each wing-pattern form (L. arthemis arthemis; L. a. astyanax, and L. a. arizonensis). Individuals were assigned to subspecies based on visual inspection of their wing patterns together with collection locality in the context of previously published range maps (Scott 1986). Butterflies that possessed intermediate wing patterns, which were collected from populations within the recognized hybrid zone, were considered intergrades. Although western populations of white-banded admirals have historically been considered as a distinct subspecies, L. a. rubrofasciata (Barnes and McDunnough 1916), current taxonomy treats all white-banded butterflies in this species complex as a single subspecies, L. a. arthemis.

Table 1.  Voucher specimen locality data, sample codes, and sample sizes for individuals sequenced (Specimen codes in bold were sequenced for nuclear loci).
SubspeciesLocationCodeSample SizeLatitude (N)
L. a. rubrofasciataNorth Star County, AlaskaAK 464° 45′
L. a. rubrofasciataEdmonton, Alberta – CanadaAB 453° 34′
L. a. rubrofasciataWeyburn, Sask. – CanadaSK 450° 26′
L. a. arthemisSW Manitoba – CanadaMB1 651° 60′
L. a. arthemisSE Manitoba – CanadaMB21349° 54′
L. a. arthemisSE Ontario – CanadaON1 648° 22′
L. a. arthemisSW Ontario – CanadaON2 448° 20′
L. a. arthemisPrince Edward Isle – CanadaPE 146° 26′
L. a. arthemisOneida County, WisconsinWI1 546° 36′
L. a. arthemisBayfield County, WisconsinWI2 846° 35′
L. a. arthemisMarquette County, MichiganMI1 546° 20′
L. a. arthemisMontreal, Quebec – CanadaQC 245° 23′
L. a. arthemisMackinac County, MichiganMI2 545° 55′
L. a. arthemisVilas County, WisconsinWI3 445° 55′
L. a. arthemisForest County, WisconsinWI4 545° 25′
L. a. arthemisHancock County, MaineME 244° 54′
L. a. arthemisTaylor County, WisconsinWI5 544° 20′
L. a. arthemisWinona County, MinnesotaMN 343° 20′
L. a. arthemisSauk County, WisconsinWI6v543° 30′
L. a. arthemisMerrimack Co., New HampshireNH 143°40′
IntergradesTompkins County, New YorkNY 542° 25′
IntergradesMontgomery Co, Penn.PA 640° 24′
L. a. astyanaxMason County, IllinoisIL1 540° 05′
L. a. astyanaxChampaign Co., IllinoisIL2 440° 10′
L. a. astyanaxMontgomery Co., MarylandMD 339° 42′
L. a. astyanaxColes County, IllinoisIL3 539° 20′
L. a. astyanaxFranklin County, OhioOH 139° 10′
L. a. astyanaxWarren County, VirginiaVA 538° 55′
L. a. astyanaxJefferson County, KentuckyKY 538° 13′
L. a. astyanaxMarion County, ArkansasAR1 536° 20′
L. a. astyanaxMadison County, ArkansasAR2 435° 48′
L. a. astyanaxCumberland Co., N. CarolinaNC 235° 50′
L. a. astyanaxCharleston Co., S. CarolinaSC 433° 10′
L. a. astyanaxCalhoun County, MississippiMS 533° 58′
L. a. astyanaxFort Bend County, TexasTX1 229° 23′
L. a. arizonensisBrewster County, TexasTX2 229° 20′
L. a. arizonensisCochise County, ArizonaAZ1 332° 01′
L. a. arizonensisSanta Cruz County, ArizonaAZ2 334° 18′
L. a. arizonensisNorthern Pima County, ArizonaAZ3 231° 55′
Outgroups
L. lorquiniBenton County, OregonN.A. 144°40′
L. weidemeyeriiBoulder County, ColoradoN.A. 140°15′
L. archippusTompkins County, New YorkN.A. 142°25′

We sequenced a 2.02-kb region of mtDNA spanning all of COI, tRNA-Leu, and a large portion of COII for 166 individuals, including three outgroup taxa (L. lorquini, L. weidemeyerii, and L. archippus). PCR was performed using universal insect mtDNA primers C1-J-1751 (5′-TCCAATGCACTAATCTGCCATATTA-3′) and C2-N-782 (5′-GGAGGATTTGGAAATTGATTAGTTCC-3′) (Simon et al. 1994; Willet et al. 1997), under the following cycle conditions: 35 cycles of 60 sec at 95°C, 60 sec at 50°C, and 120 sec at 72°C. Successfully amplified products were cleaned for sequencing using Exonuclease I and Shrimp Akalaine Phosphatase (Exo/Sap). Exo/Sapped PCR fragments were sequenced on an ABI PRISM 377 automated sequencer using BigDye terminator labeling (Applied Biosystems, Foster City, CA). Five additional internal primers were used to sequence the entire 2-kb fragment in both directions (see Mullen 2006).

In addition, we amplified and sequenced seven nuclear loci from individuals in 12 populations (Table 1) representing a subset of the individuals sampled for the broader phylogeographic analysis. The sampled loci included kettin and lactate dehydrogenase (Ldh), known to be sex-linked in other butterflies or moths (Koike et al. 2003; Dopman et al. 2004), as well as five anonymous nuclear loci described below. Sequences of kettin and Ldh were obtained using PCR primers designed from Manduca and Papilio sequences (P. Andolfatto, pers. comm.; see Table 2). The anonymous nuclear markers were developed by amplifying, cloning, and sequencing anonymous restriction enzyme fragments from a genomic DNA pool. DNA was isolated from individual butterflies using Qiagen (Valencia, CA) extraction kits and ∼25 μg of flight muscle. Ten microliters of genomic DNA was digested overnight at 37°C using 1 μl (5 units) of MboI (New England Biolabs, Beverly, MA) in a total reaction volume of 20 μl. SNX linkers (Hamilton et al. 1999) were ligated to the digested fragments and 1 μl of the digest was used as template for a PCR reaction with SNX specific primers (SNXF 5′-CTAAGGCCTTGCTAGCAGAAGC-3′; SNXR 5′-pGCTTCTGCTAGCAAGGCCTTAGAAAA-3′). PCR reactions were performed using the following amplification profile: 35 cycles of 60 sec at 95°C, 60 sec at 50°C, and 120 sec at 72°C. PCR fragments were visualized on a 1.0% agarose gel and size selected for ∼0.5–1 Kb via gel extraction (QIAquick Gel extraction column). Cleaned fragments were then blunt-end cloned into pUC19 with T4 ligase (Invitrogen, Carlsbad, CA). The ligation reaction was purified using a QIAquick PCR purification column and transformed into chemically competent One-Shot(tm) E. coli cells (Invitrogen Topo-TA® Cloning Kit). Clones were grown upon ampicillin plates, 48 positive clones were isolated and subjected to a boiling prep extraction (96°C for 10 min) and insert size was assessed by PCR using M13 forward and reverse primers. Amplified clones were sequenced as described above and NCBI's Blastn was then used to compare the sequences with GenBank. Clones that showed high sequence similarity to known housekeeping genes, such as Actin, were excluded because these genes are known to be under significant functional constraint and are not often useful for analyzing recent divergence events. We chose 12 clones that showed no significant similarity to sequences in GenBank and designed primers using PrimerSelect (DNAstar, Madision, WI). Ultimately, we chose to use five anonymous loci (Table 4) that amplified well from a test panel of six individuals. These amplifications followed the general cycling profile: 35 cycles of 60 sec at 95°C, 60 sec at 48–60°C, and 30–90 sec at 72°C (∼60 sec/kb of product length). Because direct sequencing of nuclear gene regions is often difficult due to heterozygosity and the presence of insertions and deletions, we cloned each PCR product (using Invitrogen Topo-TA Cloning Kit) to isolate homogeneous sequencing templates. Sequencing multiple clones for each locus from a single butterfly was done to assess intraindividual variation. In no case did we observe more than two alleles. Therefore, we sequenced a single insert for each nuclear gene per individual specimen with universal M13 forward and reverse primers.

Table 2.  List of PCR primers.
LocusPrimer Sequences
mtDNA
 C1-J-17515′- TCCAATGCACTAATCTGCCATATTA-3′
 C2-N-7825′- GGAGGATTTGGAAATTGATTAGTTCC-3′
Nuclear
 Kettin-F5′-TGAAATCCCGGCGAACCAGTAACA-3′
 Kettin-R5′-TGAAATCCCGGCGAACCAGTAACA-3′
 LDH-F5′-ATCGCCAGTAACCCCGTGG-3′
 LDH-R5′-CGATAGCCCAGGAAGTGTATCCCTTC-3′
 Anon6-F5′-CAGAAGCCCGACGTACAGGTG-3′
 Anon6-R5′-CTTCCAGCTCCATCATCACATTCT-3′
 Anon10-F5′- ATGCTGATAAACCTTCCTCTTGA-3′
 Anon10-R5′-AAGGCATACCGCAAAATAAGTC-3′
 Anon14-F5′-GGCAGGCTTCCCACACC-3′
 Anon14-R5′-TGCCGCCCACAGACATCA-3′
 Anon15-F5′-ACTTGGAGAGGTAACGGGAATA-3′
 Anon15-R5′-GTGCAAAGTCGGGGTAGGT-3′
 Anon17-F5′-TAAAGGATTAATGCAAGGTGCTATC
  TTATC-3′
 Anon17-R5′-TAACATCAAACGTCTTAATATTAGAGGA
  GGAA-3′
Table 4.  Nucleotide polymorphism within the L. a.arthemis species complex (S= No. of segregating sites; π= the average number of nucleotide differences among individuals; θ= nucleotide diversity; Rm= minimum number of recombination events; Fu and Li's D* based on the number of singleton mutations vs. S; Tajima's D compares θs and π). Bolded values represent significant departures from neutral expectations.
LocusNo. of sequencesTotal sitesSπθRmD*D
  1. Values of θ and π are given per site. Rm is the minimum number of recombination events per locus determined by Hudson and Kaplan's (1985) four-gamete test.

COI-II
 L. a. arthemis 0.005030.01139 3671.84
 L. a. astyanax16619081550.015150.013400−1.93 0.49
 L. a. arizonesis 0.001890.00231 −0.62−0.86
Kettin
 L. a. arthemis 12880 220.005470.008302−1.39−1.51
 L. a. astyanax 10  130.003950.005641−1.44−1.11
Ldh
 L. a. arthemis 12715 250.007450.0125602.271.81
 L. a. astyanax  9 1100.073310.066755−0.07 0.74
Anon10
 L. a. arthemis 12595 300.029630.035655−0.75−0.92
 L. a. astyanax  9  150.031190.035786−0.49−0.57
Anon14
 L. a. arthemis 12600 140.005630.007030−1.46−1.17
 L. a. astyanax  9  160.008610.009812−1.08−0.59
Anon6
 L. a. arthemis 11493  60.003710.004180−1.03−0.45
 L. a. astyanax  8  190.015270.015140−0.12 0.05
Anon15
 L. a. arthemis 11518 400.017600.027543−1.83−1.47
 L. a. astyanax 10  380.017050.005251−1.62−1.49
Anon17
 L. a. arthemis 14668 300.008610.014271−2.03−1.70
 L. a. astyanax  9  150.005460.008350−1.93−1.67

POLYMORPHISM AND DIVERGENCE

Population genetic parameters were determined using DnaSP (Rozas et al. 2003). For each of the three subspecies (arthemis, astyanax, and arizonensis), we calculated mitochondrial nucleotide diversity (θ) and the average number of nucleotide differences (π) among the sampled taxa (Nei and Li 1979). In addition, we estimated Tajima's D (Tajima 1989) and Fu and Li's D* (Fu and Li 1993) to test deviations from neutrality and equilibrium expectations (Kimura 1983). Tajima's D (1989) compares two estimates of 4Neμ, based on estimates of θs and π. The independent estimates should be equal for neutrally evolving loci in equilibrium populations. Significant departures may indicate demographic effects or the action of selection. Similarly, Fu and Li's D* (1993) compares estimates of 4Neμ based on the number of singleton mutations (hs) and the number of segregating sites (S), and can also be used to detect demographic or selective events. All of these tests, however, assume that no population structure occurs within subspecies, and the existence of unrecognized structure could potentially bias these statistics. Therefore, an analysis of molecular variance was performed on the mitochondrial data using Arlequin ver. 2 (Schneider et al. 2000) to test for population structure both among and within wing-pattern races. Finally, descriptive statistics were determined for each of the sampled nuclear loci, for populations of L. a. arthemis and L. a. astyanax that were used in the coalescent analyses of gene flow. These data were also used to test an isolation model of speciation (see below).

GENE GENEALOGIES

MtDNA sequences were aligned with CLUSTALW in MEGALIGN (DNASTAR, Madison, WI) and checked by eye to produce the final alignment (1911 bp). To explore the general topology and branch lengths of the mtDNA gene tree, a neighbor-joining tree was constructed in PAUP* 4.0b10 (Swofford 2003) using an HKY + I +Γ model of DNA substitution, which was identified as the most appropriate model via hierarchical likelihood-ratio tests implemented in MODELTEST (Posada and Crandall 1998). Subsequent tree searching was done via a Bayesian analysis performed using MrBayes 3.1.1 (Ronquist and Huelsenbeck 2003). We set the general lset parameters in MRBAYES to the equivalent of an HKY + I +Γ model (lset nst = 2 rates = invgamma) for our Bayesian analyses. Four chains, one cold and three heated incrementally (0.2 temp), were employed in duplicate for a series of MCMC searches of varying lengths sampled every 100 generations. We ran the final search for 5 million generations, which resulted in split frequencies of ∼0.0012; in general, simulations should be run until split frequencies, a measure of similarity of sampled trees, are less than 0.01. Trees were then imported into PAUP*4.0b (Swofford 2003) and the first 10,000 trees were discarded as burn-in. The posterior probabilities for the final mitochondrial gene genealogy were determined by building the consensus topology from the remaining 40,001 sampled trees. To compare these Bayesian posteriors with more familiar bootstrap values (Felsenstein 1985), we conducted 1000 maximum-likelihood heuristic bootstrap replicates using the same DNA substitution model in the genetic algorithm for rapid likelihood inference program (GARLI) developed by Zwickl (2006). Bootstrap values were determined by importing the resulting tree file into PAUP* and compiling a majority-rule consensus tree.

Neighbor-joining trees based on uncorrected P-distances were also constructed for each nuclear locus in PAUP* 4.0b10 (Swofford 2003). Unambiguous insertions and deletions were coded as a fifth base and regions that were difficult to align with confidence were trimmed. Application of Hudson and Kaplan's (1985) four-gamete test to our data indicates that all seven loci have been influenced by recombination. Therefore, each locus represents a mosaic of genealogical histories (Maddison 1997), and defining relationships among alleles based on overall similarity is a reasonable approach. To further evaluate the robustness of the neighbor-joining topologies, we explored both maximum-likelihood and Bayesian approaches for each locus using a GTR + I +Γ model of DNA substitution and the same bootstrap replication and tree visualization methods described above for the mitochondrial analyses.

TESTING AN ISOLATION MODEL

To explore whether patterns of shared polymorphism and polyphyly observed for L. a. arthemis and L. a. astyanax in our nuclear gene trees are due to gene flow or incomplete lineage sorting, we fit an isolation model of speciation to the multi-locus datasets using the program WH (Wakeley and Hey 1997; Wang et al. 1997). The model assumes that two species, or populations, arose from a single ancestral species t generations in the past and that both the ancestral and descendent populations have constant effective population sizes. Furthermore, the model assumes zero migration and no selection. The program provides estimates of basic population parameters including, the population mutation rates for the ancestral and descendant populations (θ1, θ2, θA: where θ= 4Nu) and an estimate of the time since splitting (τ= 2ut). To test the quality of the fit between our data and levels of polymorphism expected under an isolation model we conducted 10,000 coalescent simulations based on the estimated population parameters. (Wang et al. 1997; Kliman et al. 2000), and compared the distribution of shared and fixed differences to our observed values.

COALESCENT ESTIMATES OF TIME SINCE DIVERGENCE AND MIGRATION

To investigate patterns of gene flow among the nonmimetic (L. a. arthemis) and mimetic (L. a. astyanax) wing-pattern races, we estimated demographic parameters using Hey and Nielsen's (2004) coalescent simulation IM (Isolation Model version 07/01/06). IM was chosen because the program can implement sophisticated demographic models, including changes in population size and because it allows a large number of loci to be studied simultaneously. However, the program assumes no selection, no recombination, and no hidden population substructure, and interpretations may be sensitive to violations of these assumptions.

We used a total of eight loci for these analyses, collected from a subsample of the individuals used in the broader-scale mitochondrial survey (Table 1), including: mitochondrial COI and COII (1908 bp), kettin (880 bp), Ldh (715 bp), Anon6 (493 bp), Anon10 (595 bp), Anon14 (600 bp), Anon15 (518 bp), and Anon17 (668 bp). The presence of recombination between sites was checked using the Hudson and Kaplan algorithm (1985) included in DnaSP 4.01 (see Table 4). Because intragenic recombination was detected, we followed Hey and Nielsen's (2004) advice and used only the largest region of each locus for which the minimum number of recombination events (Rm) was equal to zero (i.e., the largest region with no evidence of recombination). The trimmed dataset that was ultimately used for analysis included: mtDNA (1908 bp), kettin (583 bp), Ldh (294 bp), Anon6 (406 bp), Anon10 (152 bp), Anon14 (315 bp), Anon15 (339 bp), and Anon17 (606 bp).

Three loci (mtDNA, Anon6, and Anon14) followed the HKY mutation model of evolution, whereas the other gene regions followed the Infinite Sites model. First runs used parameter values recommended by Hey and Nielsen (2004) for priors of upper bounds of parameters. Convergence and mixing of runs were evaluated by examining the level of autocorrelation between final and initial parameter values, the estimated value of the Effective Sample Size (ESS), and the consistency of results over different runs using the same model (three replicates were performed).

We focused on estimating the parameter values for a model of isolation such that at time t a single ancestral population with population mutation rate θa diverged into two daughter populations, such that θ1≠θ2≠θa. After time t, each deme exchanged migrants at rate m1 and m2 (m1m2), where m1 and m2 are migration rates into deme 2 and 1, respectively. All loci shared the same two migration rate parameters, and the inheritance scalars (effective copy number) were held constant. In final runs, priors were set to 30 for θ, 5 for t, and m1= 10 and m2= 5. Optimal chain swapping rates and ESS numbers were obtained with 10 chains and a geometric increment model with terms g1= 0.7, g2= 0.8, and the window size for t updating was reduced to v = 5. The burn-in period was set to 100,000 iterations and convergence of runs was reached after 6,000,000 chain steps.

Results

Bayesian analysis identified three distinct and strongly supported mitochondrial DNA clades, representing 147 unique haplotypes, within the polytypic Limenitis arthemis species complex (Fig. 2). Clade I includes all of the white-banded butterflies sampled, with the exception of a small number of individuals that fall basal in the genealogy and share a mitochondrial haplotype with L. lorquini (one of the outgroups). In addition, Clade I contains some of the mimetic individuals sampled from L. a. astyanax populations throughout the southeast. Clades II and III represent phenotypically uniform groups of mimetic individuals sampled from the southeastern (II) and southwestern (III) United States. The relationship among these three major mitochondrial clades is unresolved.

Figure 2.

Bayesian genealogy showing relationships among mtDNA haplotypes sampled from the Limenitis arthemis complex.. Bayesian posterior probabilities are given above branches, majority-rule ML bootstrap values are in bold below branches (average pairwise distances among clades is shown in Table 3). Also note that codes for OTUs indicate postal code of the region or state in which a specimen was collected, the population number from that postal code, and the haplotype for each individual at that locality (i.e., ON1.1 indicates the first haplotype from Ontario population 1).

Within Clade I are several strongly supported subclades, all but one of which consist of exclusively mimetic or nonmimetic individuals (Fig. 2). Subclade IA includes only white-banded L. a. arthemis, which are primarily but not exclusively rubrofasicata individuals sampled from regions north of the Great Lakes and throughout western Canada and Alaska (Figs. 2, 3). In contrast, subclade IB is comprised only of mimetic L. a. astyanax from the southeastern United States. These individuals are broadly sympatric with con-subspecifics falling in clade II (Figs. 2, 3). Of 16 populations represented in clades IB and II, 10 contain representatives in both clades, in spite of the small sample sizes from each locality. These two major mitochondrial lineages within L. a. astyanax differ on average by more than 2.4% (Table 3). Like subclade IA, subclade IC consists only of nonmimetic (white-banded) individuals, and includes butterflies possessing the eastern L. a. arthemis phenotype as well as individuals with the western rubrofasciata phenotype. Therefore subclade IB, comprised only of mimetic butterflies, is nested within a larger clade (IA + IB + IC) that otherwise consists entirely of nonmimetic individuals. Taken together, this larger clade is sister to subclade ID, the only clade that includes both mimetic and nonmimetic individuals. The basal members of subclade ID are nonmimetic, but the majority of individuals within the subclade are butterflies of both phenotypes from populations in the eastern United States near or adjacent to the known hybrid zone between L. a. arthemis and L. a. astyanax (Figs. 2 and 3)

Figure 3.

Map of geographic ranges of the four subspecies of the L. arthemis complex showing the relative frequency of mtDNA haplotype groups sampled at each locality.

Table 3.  Average pairwise genetic distances for mitochondrial clades within the L. arthemis complex and three outgroup taxa.
 IAIBICIDIIIIILorquinWeidemeyerViceroy
IA 
IB0.0060  
IC0.00470.0044 
ID0.00900.00990.0084  
II0.02520.02700.02520.0220 
III0.02570.02480.02460.02360.0257  
Lorquin0.03140.03300.03190.02970.02410.0251 
Weidemeyer0.03250.03370.03350.03150.02520.02620.0194  
Viceroy0.03980.04130.04190.03770.03300.03720.03300.0309

EVOLUTIONARY RELATIONSHIPS AMONG WING-PATTERN FORMS BASED ON NUCLEAR GENE GENEALOGIES

A subset of butterflies sequenced for mtDNA (Table 1) was sampled for each nuclear gene, but we failed to obtain clean sequence data for all individuals for all loci (sample sizes ranged from 19 to 23, Table 4). Three phylogenetic approaches (distance, maximum-likelihood, and Bayesian) recovered highly concordant topologies for each locus in all but one case (see below); we present only the neighbor-joining trees (Fig. 4; tree files from GARLI and MrBayes are available from the authors). The seven nuclear gene loci reveal different genealogical relationships among the subspecies in this complex, and none of the nuclear gene trees reveal concordance between wing-pattern phenotypes and sampled alleles (Fig. 4). When trees are rooted with L. archippus, even presumed outgroups, L. lorquini and L. weidemeyeri, fall within the ingroup for all of the nuclear loci except Anon14 (and Anon17 for which L. archippus outgroup sequence was unavailable). NJ trees for four of the sampled loci (Anon15, Anon6, Kettin, and Ldh) place L. a. arizonensis sequences in the basal clade; however, the Bayesian and maximum-likelihood consensus trees for Kettin place L. a. arizonensis in an unresolved basal polytomy.

Figure 4.

Figure 4.

Nuclear gene genealogies showing patterns of species-level polyphyly and absence of exclusivity among nonmimetic and mimetic subspecies.

Figure 4.

Figure 4.

Nuclear gene genealogies showing patterns of species-level polyphyly and absence of exclusivity among nonmimetic and mimetic subspecies.

PATTERNS OF MOLECULAR VARIATION

Patterns of polymorphism and divergence within the three subspecies of the L. a. arthemis complex are summarized in Table 4. Tajima's D and Fu and Li's D* test statistics are significantly negative for the mitochondrial sequences sampled from the white-banded, nonmimetic, subspecies (L. a. arthemis-–eastern and western populations). These statistics are negative but not significant for the allopatric, mimetic, subspecies in the southwestern United States (L. a. arizonensis). Mimetic populations of L. a. astyanax, sampled throughout the southeast, in contrast, had a weakly positive Tajima's D value but also showed a nonsignificantly negative Fu and Li's D* statistic. Nucleotide diversity (θ) varied among the nuclear loci by a factor of 10 (0.007–0.067), with substantial variation in both levels of nucleotide polymorphism (Table 4) and the prevalence of insertions and deletions. However, consistent with the mitochondrial data, Tajima's D and Fu and Li's D* statistics were negative for each locus in nearly all comparisons (Table 4), but only for Ldh were values significantly different from neutral/equilibrium expectations. Finally, an analysis of molecular variance (AMOVA) found significant differences among subspecies (L. a. arthemis, including what some call rubrofasicata, and L. a. astyanax; P < 0.001;% of variation = 37.92) but no significant differences among populations sampled within subspecies (P > 0.05). Our population sample sizes, however, are small and the possibility of substructure within subspecies remains a possibility, particularly among white-banded populations (see Fig. 3). Limenitis arthemis arizonensis was not included in the AMOVA because our sample sizes are small, and previous phylogenetic analyses suggested that this subspecies is genetically distinct from the two eastern forms (Mullen 2006).

TESTING AN ISOLATION MODEL OF SPECIATION

The observed pattern of shared polymorphism between populations of L. a. arthemis and L. a. astyanax sampled for the multi-locus nuclear gene genealogies suggests either that these two wing-pattern races have experienced significant levels of recent gene exchange or there has been insufficient time since divergence to permit lineage sorting. As pointed out by Wang et al. (1997), for loci that are not experiencing gene flow, the number of fixed differences and the number of shared polymorphisms are expected to negatively covary with increasing divergence time. Under a scenario of migration, however, this relationship is expected to break down because gene flow reintroduces polymorphisms. The results of our model fitting simulations (parameter estimates not shown) indicate that the observed levels of shared polymorphism are significantly higher than expected under a simple model of speciation by isolation (test statistic = 63.0, P= 0.0335). Note that the test statistic calculated by WH is one-tailed because it focuses on detecting a departure from the model in the direction expected under a scenario of historical gene flow (Kliman et al. 2000).

ESTIMATES OF MIGRATION RATES AND DIVERGENCE TIMES

Estimates of the mean mutation rate per gene per generation and the number of generations per year are required to convert results of the IM analysis into demographic units. Mutation rates per site in nuclear genes are generally expected to be around ∼10−9 per nucleotide site per generation, or 10−6 per (entire) gene per year (Ridley 1996; Futuyma 1998). We used a mutation rate of 10−8 per site per generation and a generation time (g) of 0.5 year (i.e., 2 generations/year) to be conservative with respect to our estimated time of divergence (i.e., a faster mutation rate will produce shallower divergence estimates). The mutation rate per gene per year (μ) for each gene was calculated and the geometric mean of these rates was used to generate a global per year mutation rate of 4.436−6 (μ= 2.218−6/generation) for the multilocus analysis, following Hey and Nielsen (2004).

Results were highly consistent over the three replicates (Fig. 5). The maximum-likelihood estimate (MLE) for the effective ancestral population size (Na) was approximately ∼350 k individuals. The MLEs for L. a. arthemis (nonmimetic, N1) and L. a. astyanax (mimetic, N2) were much higher, corresponding to ∼1.97 million and ∼2.5 million individuals, respectively (see Table 5 for estimates of the 90% highest posterior density interval—HPD90% interval—resulting from the different runs). The estimated divergence time, t, varied between 132,000 and 397,000 ybp, with an MLE of 235,000 ybp. The MLEs for migration rates m1 and m2 (migration rates per year of genes moving from L. a. arthemis into L. a. astyanax or vice versa, as we move backward through time) were 1.61945−07 and 5.54607−08. ML estimates of the population migration rate (2nm) (moving forward in time) between the two taxa were highly asymmetric. The migration rate per generation from L. a. astyanax into L. a. arthemis was 3.2, whereas the MLE for the rate of migration from L. a. arthemis into L. a. astyanax was 0.14 (the HPD90% interval for m2 encompasses zero migration).

Figure 5.

Marginal posterior probability distributions for demographic parameters obtained by IM. The three replicates are indicated by red, green, and blue. The estimate for the effective population size of L. a. arthemis1) is in column 1 top left, L. a. astyanax2) is in column 2, top right, and ancestral population size (θA) is in column 1, middle left. Time since divergence for L. a. arthemis and L. a. astyanax is shown in column 2, middle right. Finally, migration rate from L. a. arthemis into L. a. astyanax is shown in column 1, bottom left and from L. a. astyanax to L. a. arthemis is at the bottom right, column 2.

Table 5.  Estimates of the 90% highest posterior density interval for each demographic parameter generated for each of three independent IM runs. (N= estimate of effective pop. size; t= time since population divergence; 2nm= population migration rate/year). The bolded values represent the mean value of the best maximum-likelihood estimate of each parameter from the three replicates.
ParameterRun0.050.25MLE0.750.95
N11744,8631,455,3431,913,7192,692,9526,818,326
2813,6161,409,4991,982,4722,555,4455,557,794
3767,7731,432,4212,028,3042,601,2776,291,196
Mean775,4131,432,4211,974,8322,616,5586,222,443
N211,059,1731,772,2792,443,4293,240,4239,029,158
21,059,1731,898,1112,485,3733,198,4796,868,868
31,017,2171,898,1112,506,3563,240,4237,540,030
Mean1,045,1881,856,1672,478,3863,226,4387,812,689
NA1148,968286,486355,240446,915676,098
2126,058240,643355,240401,083630,266
3148,968240,643355,240401,083653,187
Mean141,327255,924355,240416,364653,187
t (years)1126,779193,267231,583283,421390,479
2139,175200,029234,964283,421407,383
3131,287202,283238,344285,675391,606
Mean132,413198,519234,964284,165396,497
2n1m112.670.392.675.0415.90
20.041.103.555.8318.09
30.040.663.375.3019.15
Mean0.920.723.205.3917.71
2n2m210.030.030.033.1616.52
20.030.030.193.1114.98
30.030.030.193.3315.09
Mean0.030.030.143.2015.53

Discussion

Hybridization between L. a. arthemis and L. a. astyanax has led to a long-standing debate over the nature of species boundaries and the origins of mimicry within this polytypic species (Remington 1958; Platt and Brower 1968; Remington 1968; Platt 1987). Our mitochondrial DNA genealogy reveals that the history of wing-pattern diversification in this system has been complex (Fig. 2). Each of the wing-pattern forms consists of several distinct mtDNA clades, and neither white-banded nor mimetic phenotypes are monophyletic. However, haplotype sharing between subspecies is only evident in populations from eastern North America that are within or adjacent to the hybrid zone. Mimetic populations of Red-spotted Purples found in the southwestern United States and Mexico (Clade III; L. a. arizonensis) and the corresponding mimetic form from the southeast (Clades I and II; L. a. astyanax) do not share mitochondrial haplotypes.

The mitochondrial data also reveal that white-banded butterflies (L. a. arthemis) sampled across the northern United States and Canada predominantly fall into a single major clade (I), that includes distinct subclades representing (1) northwestern populations of white-banded butterflies (L. a. arthemis/rubrofasciata; subclades IA + IC, Fig. 2), (2) northeastern samples of white-banded butterflies (L. a. arthemis; subclade ID), and (3) mimetic individuals sampled from southeastern populations of red-spotted purples (L. a. astyanax; subclade IB) that are entirely sympatric with other mimetic butterflies that have mitochondrial haplotypes in clade II (Figs. 2, 3). A small number of white-banded individuals also group with one of the outgroup taxa, L. lorquini, with which it is known to occasionally hybridize. Thus, substantial population structure, at least with respect to mitochondrial haplotypes, occurs among wing-pattern forms. This finding is consistent with a secondary origin of the current contact zone between white banded and mimetic populations.

Discriminating between histories of primary intergradation versus secondary contact is notoriously difficult (Barton and Hewitt 1985). However, if the contact zone arose via primary differentiation, there is no a priori reason to expect such striking concordance between the geographic boundaries of mitochondrial haploptyes and the geographic boundaries of wing-pattern phenotypes. Although it is true that unrecognized effects of endosymbionts (Hurst and Jiggins 2005) and/or other forms of selection acting on mtDNA (e.g., epistasic interactions with nuclear loci; see Dasmahapatra et al. 2002) could potentially produce strong associations between mitochondrial haplotypes and wing phenotypes, we currently have no evidence to suggest that the geographic patterns of mitochondrial variation in this complex have arisen through nonneutral processes. In addition, several other lines of evidence support a secondary rather than primary origin for this contact zone. First, there is abundant paleoclimatic data (pollen records, isotope data, radiolarian fossils, etc) indicating that dramatic climatic oscillations have occurred at regular intervals over the last 2.4 million years (e.g., Williams et al. 2004). As a result, many of the species currently occupying boreal and temperate habitats in Europe and North America have experienced historical population fragmentation and have, subsequently, come back into secondary contact following postglacial population expansion (Hewitt 2000, 2001). Second, Tajima's D values for each of the loci sampled (Table 4) are uniformly negative, and together with our multilocus estimates of effective population sizes for nonmimetic (L. a. arthemis) and mimetic (L. a. astyanax) populations support a history of population expansion. In fact, our IM results indicate that both subspecies have experienced more than a fivefold increase in effective population size (compare Na to N1 and N2, Table 5). Therefore, if mimetic and nonmimetic populations of these butterflies persisted in geographically disjunct glacial refugia, it is likely that postglacial population expansion gave rise to the current hybrid zone between wing-pattern phenotypes. Interestingly, pollen records for the larval host plants of these butterflies (Populus, Salix, Prunus, etc.) support this hypothesis and suggest that separate refugia occurred in Beringia, along the northeastern Atlantic Coast, and in regions of southeastern United States (Williams et al. 2004; http://www.ncdc.noaa.gov/paleo/pollen/viewer/webviewer.html). Finally, the zone of phenotypic intermediacy between mimetic and nonmimetic populations of Limenitis arthemis coincides closely with the ecological transition from temperate broadleaf forests to boreal forest, where numerous other secondary hybrid zones, including two Lepidopteran examples (Papilio glaucus×P. canadensis,Scriber et al. 2003; Hylaphora columbia×H. cercropi,Remington 1968), occur (Remington 1968; Swenson and Howard 2004). Taken together, the phenotypic, genetic, biogeographic, and paleoclimatic evidence all suggest that current hybridization between L. a. arthemis and L. a. astyanax represents relatively recent secondary contact between already divergent lineages.

WING-PATTERN EVOLUTION AND THE ORIGIN OF MIMICRY

A second major goal of this study was to gain a better understanding of the historical context in which wing-pattern variation evolved. If mimicry increases fitness, then natural selection is expected to favor the evolution and/or maintenance of the mimetic phenotype whenever Limenitis is sympatric with the distasteful model B. philenor. The genetic variation underlying mimetic phenotypes in L. a. astyanax and L. a. arizonensis, might be explained as a consequence of common ancestry, independent mutational events, or gene flow. To understand the origins of mimicry in this system, therefore, we need to explain both the origin of the two mimetic subspecies (L. a. astyanax and L. a. arizonensis) and the presence of two evolutionarily divergent mtDNA clades within populations of L. a. astyanax.

Phylogenetic relationships among Neartic species of Limenitis suggest that the ancestral phenotype of the North American forms was a white-banded butterfly with a wing pattern similar to the nonmimetic subspecies L. a. arthemis (Mullen 2006). Under a single origin of mimicry scenario, southwestern and southeastern populations of mimetic admirals are expected to be more closely related to each other than to the northern white-banded subspecies, L. a. arthemis. Lack of resolution at the base of our mtDNA gene tree precludes a confident assessment of the relationships between these two mimetic subspecies, L. a. astyanax and L. a. arizonensis (Fig. 2), but the nuclear gene genealogies (Fig. 4) and previous phylogenetic work (Mullen 2006) suggest that the initial divergence events were between populations in the southwest (L. a. arizonensis) and the remaining subspecies. Unfortunately, support for this topology is weak (Fig. 4) and polyphyletic relationships are observed for many of the nuclear gene genealogies. Species-level polyphyly is a common observation among closely related taxa (Funk and Omland 2003) and is often attributed to incomplete lineage sorting (ILS). However, the results of our model fitting simulations reject a simple isolation model of speciation. Therefore, although ILS may have contributed to the observed genealogical patterns, the nuclear gene genealogies also likely reflect the influence of historical and contemporary hybridization between the two wing-pattern races. Thus, it is difficult to resolve whether the shared mimetic phenotypes of L. a. astyanax and L. a. arizonensis are the result of a single de novo origin of mimicry in the common ancestor of these populations or due to convergence on a similar mimetic pattern via independent mutation events. Adaptive phenotypic convergence is widespread among mimetic insects (Müller 1879; Berger and Kaster 1979; Brower 1996; Wittkopp et al. 2003; Joron et al. 2006), but resolution of this ambiguity in Limenitis will likely require additional phylogenetic data from more rapidly evolving markers (e.g., AFLPs) or the identification of the gene or gene regions underlying the wing-pattern polymorphism for use in phylogeny reconstruction.

The phenotypically uniform mitochondrial lineage of mimetic butterflies (subclade IB) that occurs within a broader mitochondrial clade of primarily nonmimetic, white-banded individuals (Fig. 2) must also be a consequence of either independent origins or gene flow. The mimetic individuals in subclade IB are completely sympatric with other Red-spotted Purples (L. a. astyanax) collected in the southeastern United States (Clade II), but differ by more than 2% mtDNA sequence divergence (Table 3). This genealogical pattern could reflect independent mutational origins of the mimetic phenotype within two allopatric mtDNA clades, followed by secondary sympatry. However, historical hybridization between mimetic and nonmimetic populations (in the presence of the model) provides a more likely explanation.

As with other examples of mimicry (e.g., Heliconius; Papilio-Scriber et al. 1996) the genetic architecture of the mimetic phenotype in the L. arthemis complex is relatively simple. Medial white banding in admiral butterflies is controlled by a single autosomal gene with two major alleles and several additional loci that modify the penetrance of alleles at the major locus (Remington 1958; Platt and Brower 1968; Robinson 1971; Platt 1983). Previous work has demonstrated that the Batesian mimcry relationship between L. a. astyanax and B. philenor (Pipevine Swallowtail) confers an advantage to unbanded individuals in areas in which the model (Battus) is common (Brower and Brower 1962; Platt et al. 1971), and theory (Bates 1862; Fisher 1958) predicts that there will be strong selection for the unbanded (mimetic) alleles in the presence of the model. Therefore, hybridization between nonmimetic and mimetic Limenitis is expected to lead to a “sweep” of the alleles underlying mimicry when both phenotypes co-occur in areas where the model is abundant.

Imagine the following scenario. Two populations come into contact. One is mimetic and carries mtDNA haplotype A. The other is nonmimetic and carries mtDNA haplotype B. The model is present and the mimetic/nonmimetic difference is due to two alleles at a single locus. Assuming neutrality of mtDNA (no selection on A and B haplotype difference) and selection for the mimetic allele in the presence of the model, we expect to see a single population that still harbors both mtDNA haplotypes (A and B) but is now represented only by the mimetic type (which has swept to fixation). If we then sampled this new population for mitochondrial variation we would find that haplotype B (originally associated only with nonmimetic types and still in that association outside the area of hybridization and where the model is not present) is associated with the mimetic phenotype in the hybridizing population. Furthermore, if we build a mitochondrial gene tree, we would find that the mtDNA B clade is primarily nonmimetic (the “ancestral condition”), but has a clade (or clades) within it that are mimetic (those representatives of the B haplotype that were converted to the mimetic phenotype in the hybridizing population). In fact, this is the pattern we observe (Fig. 2). This suggests that historical hybridization between mimetic individuals in mitochondrial clade II with nonmimetic individuals in subclade IB may have resulted in a “sweep” of the mimetic phenotype via introgression of mimicry alleles into the subclade IB population. Indeed, it is possible that nonmimetic lineages of these admiral butterflies have repeatedly been converted to the mimetic phenotype due to introgression and strong natural selection for mimicry whenever the two wing-pattern forms historically came into contact in the presence of the model (i.e., sharing the alleles underlying adaptation via hybridization).

HYBRIDIZATION AND SPECIES BOUNDARIES

The debate over species boundaries in the L. arthemis-astyanax complex is due, largely, to a lack of detailed knowledge about historical and contemporary levels of gene flow between mimetic and nonmimetic populations. Although many researchers have viewed these two wing-pattern races as incipient species on the verge of complete reproductive isolation (Hovanitz 1949; Fisher 1958; Remington 1958, 1968), others have argued just as strongly that this system simply represents clinal phenotypic variation in a single, panmictic species (Nakahara 1924; Platt and Brower 1968; Platt 1975, 1983). Our data suggest that species boundaries in this system actually lie somewhere between these two extremes.

For example, although mtDNA haplotype sharing between L. a. arthemis and L. a. astyanax suggests that the hybrid zone is a conduit for ongoing gene flow between these two subspecies (Fig. 3), haplotype sharing between mimetic and nonmimetic forms is restricted to a single subclade in the eastern portion of the hybrid zone (ID; see Figs. 2 and 3). The lack of concordance between mitochondrial DNA and wing-pattern phenotypes in this part of the range of these butterflies could be explained as mitochondrial introgression from northeastern white-banded populations into mimetic L. a. astyanax adjacent to the hybrid zone (Fig. 3). However, our multilocus coalescent estimates of migration between L. a. arthemis and L. a. astyanax indicate that, historically, gene flow has been strongly asymmetric (Table 5) and predominately from the mimetic subspecies into white-banded populations of L. a. arthemis. Given this pattern, we hypothesize that the phenotypic hybrid zone between the two butterfly wing-pattern races has shifted northward following secondary contact. The shift might be a consequence of expansion of the model's host plant (Aristolochia) in response to warming temperatures, a pattern that is also seen in Papilio (Scriber and Ording 2005). If this scenario is correct, then the haplotype sharing observed in the mtDNA genealogy for populations sampled in the eastern portion of the hybrid zone may reflect historical introgression of the alleles underlying the mimetic phenotype into L. a. arthemis rather than recent mitochondrial gene flow from L. a. arthemis into mimetic populations. A similar situation, in which a phenotypic hybrid zone has shifted northward leading to apparent discordance between mitochondrial haplotypes and phenotype, has been reported for Black-capped (Poecile atricapillus) and Carolina (P. carolinensis) chickadees (Gill 1997; Reudink et al. 2007).

The fact that mtDNA haplotype sharing occurs only in a geographically restricted group of mimetic and nonmimetic individuals suggests that, although gene flow is apparently ongoing between these populations in the eastern United States, some barrier to gene exchange exists between the two wing-pattern forms in the western portion of the hybrid zone. Of course, this conclusion is sensitive to the scope of our sampling. Waldbauer et al. (1988) also found that even across a relatively small geographic distance (8 km), at the Straits of Mackinac in Michigan, phenotypic evidence of gene flow between mimetic and nonmimetic populations was rare. Together with our finding of very limited mitochondrial introgression in the western part of the hybrid zone (Fig. 3), these results imply that intrinsic and/or extrinsic barriers to gene exchange are more pronounced in this region. Interestingly, phenotypic evidence for differentiation between eastern and western L. a. arthemis has previously been reported, and was evident in our sampling. Northwestern individuals of white-banded admirals are sometimes treated as a separate subspecies (L. a. rubrofasicata) on the basis of subtle wing-pattern variation (Barnes and McDunnough 1916), and an examination of the geographic distribution of mitochondrial haplotypes makes clear the subdivision among nonmimetic populations (Fig. 3). Therefore, differences in the extent of introgression between nonmimetic and mimetic populations may be a consequence of intrinsic differences between eastern and western L. a. arthemis populations, perhaps tracing back to differences arising during periods of allopatry in North Atlantic and Beringial refugia.

Barriers to gene exchange in this system are clearly incomplete. However, the dramatically different wing-pattern phenotypes corresponding to L. a. arthemis and L. a. astyanax are not simply the result of primary differentiation, as suggested by Platt and Brower (1968), but reflect a complex history of divergence and gene flow between two partially isolated evolutionary lineages. Geographic, demographic, and selective factors have likely all played important roles in the origins and maintenance of wing-pattern diversity in this group, and it is possible that selection related to mimicry both maintains the current distribution of wing-pattern forms and limits gene flow between these two wing-pattern races. If so, then this system may represent another of the growing number of examples of natural selection acting on adaptive phenotypic traits and incidentally contributing to reproductive isolation between ecologically divergent populations (Noor 1995; Funk 1998; Jiggins et al. 2001; Podos 2001; Schluter 2001; Via 2001; Allender et al. 2003; McKinnon et al. 2004; Rundle and Nosil 2005; Doebeli et al. 2005; see Funk et al. 2006 for a cross-taxon comparative study of ecological divergence and reproductive isolation).

Conclusions

Our results indicate that hybridization between mimetic and nonmimetic wing-pattern races of the polytypic L. arthemis-astyanax species complex is the result of secondary contact between evolutionarily distinct lineages that have experienced substantial divergence in the face of intermittent gene exchange. They also suggest that the alleles underlying the mimetic phenotype have, historically, spread through previous nonmimetic populations during periods of sympatry with both the mimic and the model. Although it remains to be demonstrated that natural selection acting on mimicry is driving and/or maintaining divergence between these wing-pattern races, examples of divergence with gene flow in insects are becoming more common (Feder et al. 2003; Emelianov et al. 2004; Turner et al. 2005; Bull et al. 2006; Kronforst et al. 2006; Michel et al. 2006; Machado et al. 2007).

Indeed, although the recognition that gene flow may continue throughout the speciation process has recently increased, examples of “divergence with gene flow” have been known from the hybrid zone literature for more than 70 years. Few systems, however, provide as many examples of this phenomenon as heliconiine butterflies. In fact, Mallet et al. (2007) recently reviewed rates of natural hybridization in Heliconius and used these and other data to argue that the discreteness and “reality” of species have been eroded both below and above the level of species. The reality of species, in this case, apparently refers to complete reproductive isolation between species, or a strict interpretation of Mayr's (1942) Biological Species Concept. At the root of this argument is the belief that there is a conflict between these two equally valuable views of species, as either reproductively isolated groups or as stable clusters of phenotypes that maintain their distinctiveness in the face of gene flow and hybridization. In fact, we see no conflict; as Harrison noted (1998), disagreements over species boundaries are often related to where in the “life history” of a species one chooses to focus attention. With respect to Limenitis, the issue of species boundaries resolves to whether one focuses on patterns of gene flow and the lack of complete barriers to gene exchange or on the evolution and maintenance of the unique phenotypic wing-pattern variation. In this case, the two wing-pattern races clearly represent an intermediate stage in the evolutionary transition from a single, panmictic population to fully isolated biological species and, therefore, are probably best viewed as geographic races or subspecies. Ultimately, however, we are less interested in whether L. a. arthemis and L. a. astyanax are different species than in the evolutionary processes that produced and maintain their dramatically different wing-pattern phenotypes.

Associate Editor: D. Funk

ACKNOWLEDGMENTS

This work was supported by an National Science Foundation DDIG Grant (DEB-0407499) to RGH and SPM, as well as by funding to SPM from the Kieckhefer Adirondack Fellowship, the Theodore Roosevelt Memorial Fund (AMNH), the Edna Bailey Sussman Fund for environmental field research, Sigma-Xi, and a Joan Mosenthal DeWind Award (Xerces Society). Specimens were provided by a large number of individuals but J. Gilardi deserves special recognition. Finally, we thank Dr. D. Funk and two anonymous reviewers for comments on an earlier draft of this manuscript.

Ancillary