Evolution, 58(7), 2004, pp. 1536±1548

WHEN VICARS MEET: A NARROW CONTACT ZONE BETWEEN MORPHOLOGICALLY CRYPTIC PHYLOGEOGRAPHIC LINEAGES OF THE RAINFOREST SKINK, CARLIA RUBRIGULARIS

BEN L. PHILLIPS,1 STUART J. E. BAIRD,2 AND CRAIG MORITZ3 Department of Zoology and Entomology, The University of Queensland, St Lucia, Queensland, 4067 Australia

Abstract. Phylogeographic analyses of the fauna of the Australian wet tropics rainforest have provided strong evidence for long-term isolation of populations among allopatric refugia, yet typically there is no corresponding divergence in morphology. This system provides an opportunity to examine the consequences of geographic isolation, independent of morphological divergence, and thus to assess the broader signi®cance of historical subdivisions revealed through mitochondrial DNA phylogeography. We have located and characterized a zone of secondary contact between two long isolated (mtDNA divergence Ͼ 15%) lineages of the skink Carlia rubrigularis using one mitochondrial and eight nuclear (two intron, six microsatellite) markers. This revealed a remarkably narrow (width Ͻ 3 km) hybrid zone with substantial linkage disequilibrium and strong de®cits of heterozygotes at two of three nuclear loci with diagnostic alleles. Cline centers were coincident across loci. Using a novel form of likelihood analysis, we were unable to distinguish between sigmoidal and stepped cline shapes except at one nuclear locus for which the latter was inferred. Given estimated dispersal rates of 90±133 m ϫ genϪ1/2 and assuming equilibrium, the observed cline widths suggest effective selection against heterozygotes of at least 22±49% and possibly as high as 70%. These observations reveal substantial postmating isolation, although the absence of consistent deviations from Hardy-Weinberg equilibrium at diagnostic loci suggests that there is little accompanying premating isolation. The tight geographic correspondence between transitions in mtDNA and those for nuclear genes and corresponding evidence for selection against hybrids indicates that these morphologically cryptic phylogroups could be considered as incipient species. Nonetheless, we caution against the use of mtDNA phylogeography as a sole criterion for de®ning species boundaries.

Key words. Carlia rubrigularis, hybrid zone, likelihood pro®les, phylogeography, secondary contact, speciation.

Received August 23, 2002. Accepted March 29, 2004.

There is a general debate among evolutionary biologists lying patterns of genetic divergence (Ballard et al. 2002; about the nature of genetic differences that result in repro- Irwin 2002; Hudson and Turelli 2003). ductive isolation and the relative roles of isolation, genetic Analysis of zones of secondary contact between lineages drift, and divergent selection in speciation (e.g., Orr and can shed light on the extent and nature of reproductive iso- Smith 1998; Turelli et al. 2001; Gavrilets 2003). It is widely lation (Barton and Hewitt 1985, 1989). The majority of con- accepted that allopatric isolation combined with divergent tact zones concern taxa that are distinct phenotypically or selection on ecological or sexually dimorphic traits can result chromosomally and in many cases there is evidence for sub- in speciation, and it has also long been proposed that inci- stantial pre- and postzygotic isolation (Barton and Hewitt dental genetic divergence in allopatry can result in substantial 1985; Harrison 1993). However, despite the recent promi- postzygotic isolation (Dobzhansky 1937). Evidence for the nence given to phylogeographic structure as a criterion for latter comes in particular from experimental crosses of al- species recognition (e.g., Goldstein and DeSalle 2000; Tem- lopatric and morphologically cryptic sibling species (e.g., pleton 2001; Wiens and Penkrot 2002; Sites and Marshall Dobzhansky 1970). The recent emphasis on mtDNA phylo- 2003) and in conservation biology (Avise 2000; Moritz geography has revealed numerous cases of apparently deep 2002), there are remarkably few detailed studies of secondary historical subdivisions within species, many of which were contacts between lineages that are strongly differentiated in morphologically cryptic or even discordant with subspecies terms of mtDNA phylogeography, but which are morpho- boundaries (Avise 2000). However, it remains unclear wheth- logically cryptic (Hewitt 2001). er divergent mtDNA phylogroups represent organismal lin- The ``Wet Tropics'' rainforests of northeastern Australia eages that are independently evolving and perhaps intrinsi- offer an excellent natural laboratory for such studies. Com- cally isolated, or ones that will simply merge on secondary plementary biogeographic, palynological, and phylogeo- contact (Avise and Walker 1998; Wake and Schneider 1998). graphic data indicate that under cool-dry conditions that were Further, it is possible that, for one reason or another, mtDNA typical for much of the Quaternary, rainforest in this region phylogroups misrepresent organismal history and/or under- existed as two major isolates, each consisting of several smaller refugia (Webb and Tracey 1981; Nix 1991; Kershaw 1 Present address: Department of Biological Sciences, University 1994; Schneider et al. 1998; Hugall et al. 2002). The same of Sydney, New South Wales, 2006 Australia; E-mail: phillips@ bio.usyd.edu.au. evidence supports a rapid expansion of the rainforest during 2 Present address: Centre de Biologie et de Gestion des Popu- a cool-wet phase commencing approximately 8000 years ago lations, INRA-CBGP, Campus International de Baillarguet, CS 30 (Kershaw 1986; Walker 1990; Hopkins et al. 1993). Although 016 34988, Montferrier/Lez cedex, France; E-mail: stuart@ these data suggest long periods of isolation, it is also likely holyrood.ed.ac.uk. 3 Present address: Museum of Vertebrate Zoology, 3101 Life Sci- that populations restricted to these rainforest isolates have ences Building, University of California, Berkeley, California been subject to intermittent secondary contact during brief 94720-3160; E-mail: [email protected]. intervals when conditions were optimal for expansion of the 1536 ᭧ 2004 The Society for the Study of Evolution. All rights reserved. SECONDARY CONTACT IN CARLIA RUBRIGULARIS 1537 cool upland rainforests that harbor most of the endemic, rain- 2000). In the initial stages of any contact a cline in character forest-restricted species. This history of habitat contraction state frequencies must exist. The shape of the cline is deter- is re¯ected by a repeated pattern of strong mtDNA phylo- mined by the degree of selection against hybrids, the scale geographic structure between the two major isolates (Joseph of dispersal, and the time since contact (Endler 1977; Arnold et al. 1995; Schneider et al. 1998; Schneider and Moritz 1999; 1994). Neutral mixing and relatively large dispersal distances Hugall et al. 2002) with levels of intraspeci®c sequence di- will quickly result in a wide shallow cline whereas strong vergence sometimes approaching interspeci®c levels for the selection against hybrids, coupled with relatively small dis- taxa concerned (Moritz et al. 1997; Stuart-Fox et al. 2002). persal distances, will result in short, steep clines between Nevertheless, in all cases so far examined the highly diver- populations. A strong barrier to gene ¯ow between lineages gent phylogeographic lineages are morphologically cryptic would elicit the following characteristics: (1) substantial de- or separable only in multivariate space. (Schneider and Mo- viations from Hardy-Weinberg equilibrium at diagnostic loci ritz 1999; Schneider et al. 1999; M. Cunningham, unpubl. due to assortative mating and/or selection against hybrids; data). (2) a positive statistical association between loci for alleles These observations raise the question of whether the highly characteristic of each lineage (linkage disequilibrium) due to divergent phylogeographic lineages display any reproductive the dispersal of parental types into the zone and/or, the se- isolation, either pre- or postmating, or whether they simply lective removal of recombinant genotypes; and (3) clines that will merge when they come into contact. To address this remain narrow rather than widening under neutral diffusion. question, we analyzed a zone of secondary contact between Here we survey populations between the known locations divergent mtDNA phylogeogroups of the rainforest skink of the northern and southern lineages of C. rubrigularis to Carlia rubrigularis. These common, small (Ͻ52 mm snout/ locate a zone of secondary contact. We then characterize the vent), red-throated, and heliothermic skinks are endemic to zone using multiple molecular markers and analyze the data the Wet Tropics region (Nix and Switzer 1991; Cogger 2000). for patterns of genetic disequilibria and cline shape and The species is distributed across a wide variety of wet forest width. As cline width scales with dispersal, we also estimate habitats, typically at altitudes of less than 900 meters, and per generation dispersal distances from disequilibria within is most abundant in light gaps and along rainforest margins. the zone, and independently from isolation-by-distance Populations from the northern and southern wet tropics are among populations removed from the contact zone. reciprocally monophyletic for mtDNA with Ͼ15% net se- quence divergenceÐsimilar to that seen between recognized METHODS species of Carlia (Stuart-Fox et al. 2002). Multivariate anal- ysis of ®ve ecologically relevant morphological traits re- Sampling vealed that the northern and southern lineages of Carlia rub- The phylogeography presented in Schneider et al. (1999) rigularis are morphologically very similar, especially relative suggested that the northern and southern lineages of C. rub- to variation among habitats within each lineage (Schneider rigularis would be in contact somewhere in the southern et al. 1999). However, a mtDNA phylogeny for the genus Lamb Range area. We sampled populations along the length (Stuart-Fox et al. 2002) indicated that these lineages are not of the Lamb Range and found that C. rubrigularis was at all sister taxa; rather the southern lineage appears more closely sites other than those above 900 m altitude. Animals were related to Carlia rhomboidalis, a morphologically similar but captured by hand, measured, and a small piece of tail tip was blue-throated species found in wet forests from the southern removed from each individual and stored in 80% ethanol. extremity of the wet tropics to mideastern Queensland (In- Each locality was sampled up to a maximum of 20 individ- gram and Covacevich 1989; Cogger 2000). Thus, the two uals. These samples were initially screened at the mitochon- phylogroups of C. rubrigularis plus C. rhomboidalis consti- drial marker (see below). Upon locating populations with tute a complex of three morphometrically similar mtDNA mixed northern and southern mtDNA haplotypes (popula- lineages, replacing each other from north to south and with tions 2 and 6, see below) the surrounding area was sampled the two southern elements (southern C. rubrigularis and C. intensively. This process resulted in a 4-km transect through rhomboidalis) being most closely related. Some authors (e.g., the contact zone consisting of nine populations with an av- Wiens and Penkrot 2002) would recognize each of the three erage of 18 individuals per population (Fig. 1). lineages as independent, but related, species. However, given our emphasis on evolutionary processes (Harrison 1998), we Markers prefer to retain the current taxonomy until the nature and level of genetic isolation among these lineages is established. In the laboratory total cellular DNA was extracted follow- The nonsister status of northern and southern C. rubrigularis, ing standard Phenol-Chloroform extraction protocols. A di- together with extensive paleoecological evidence supporting agnostic restriction endonuclease (RE) assay for northern and contraction of their habitat to isolated refugia during the cold southern haplotypes was designed using the mtDNA se- dry periods of the Pleistocene, demonstrates that the clines quence data in Schneider et al. (1999). The RE used (HaeIII) examined here are due to secondary contact rather than par- cut both northern and southern haplotypes in a diagnostic apatric divergence (c.f. Endler 1982a). manner (see Appendix). Polymerase chain reaction (PCR) Our analyses of the contact zone between northern and ampli®cation of the relevant Cytochrome-B (CytB) fragment southern lineages of C. rubrigularis are based on theory pre- was undertaken as described in Schneider et al. (1999) and dicting cline structure and patterns of genetic disequilibria RE fragment patterns were scored following digestion and (Barton and Hewitt 1985; Barton and Gale 1993; Barton agarose gel electrophoresis. 1538 BEN L. PHILLIPS ET AL.

FIG. 1. Map of the study area. Sampling sites and allele/haplotype frequency data are represented by pie charts. (A) shows CytB haplotype frequencies throughout the Lamb Range. (B) shows 900 m contour and both Aldolase allele (left half of pie) and CytB haplotype (right half of pie) frequencies around the area of contact. Numbers 1±9 represent sites sampled within the contact (Pies shown only for sites 1 and 9). Coordinates for sites are as follows: Site 1: 145Њ33.44ЈE, 17Њ9.12ЈS; Site 2: 145Њ33.86ЈE, 17Њ9.38ЈS; Site 3: 145Њ34.13ЈE, 17Њ9.22ЈS; Site 4: 145Њ34.68ЈE, 17Њ9.16ЈS; Site 5: 145Њ34.86ЈE, 17Њ8.84ЈS; Site 6: 145Њ35.16ЈE, 17Њ8.89ЈS; Site 7: 145Њ35.15ЈE, 17Њ8.65ЈS, Site 8: 145Њ35.4ЈE, 17Њ8.59ЈS; Site 9: 145Њ35.71ЈE, 17Њ8.30ЈS. Sites contributing to isolation by distance analysis are as follows: Northern groupÐBG: 145Њ38.57ЈE, 16Њ50.78ЈS; CR: 145Њ36.74ЈE, 16Њ56.17ЈS; DC: 145Њ36.63ЈE, 17Њ2.0ЈS; WH: 145Њ36.00ЈE, 17Њ4.85ЈS; Site 1 (above). Southern groupÐB2: 145Њ38.36ЈE, 17Њ5.89ЈS; B1: 145Њ38.65ЈE, 17Њ5.49ЈS; T2: 145Њ37.69ЈE, 17Њ17.12ЈS; T1: 145Њ37.74ЈE, 17Њ7.64ЈS; NT: 145Њ37.27ЈS, 17Њ7.85ЈN; LE: 145Њ37.49ЈE, 17Њ9.62ЈS.

Additionally, two diagnostic nuclear intron markers were assay. The remaining southern allele could not be distin- developed (Aldolase intron 1 (Ald) and Rh2 Opsin intron 4 guished from the common allele using restriction enzymes (Rho)). For each of these loci double stranded sequences were (Appendix). Populations within and adjacent to the contact obtained from C. rubrigularis specimens outside the zone of were screened at both intron loci. Fragment patterns for in- contact spanning much of the Wet Tropics (data not shown). trons were scored following electrophoresis on a nondena- Sequences were obtained using standard sequencing proto- turing 8% polyacrylamide gel. Populations inside and outside cols and speci®cally designed primers (Appendix). Sequenc- the zone of contact were also screened at six microsatellite ing of intron markers revealed that Ald showed a ®xed dif- loci (A11, A16, D4, D13, D19, D20; Appendix). One of these ference in alleles between northern and southern populations. loci (A16) showed a semidiagnostic shift in allele frequency One northern allele and two southern alleles were identi®ed between lineages (Table 1). The remaining ®ve loci showed from sequences, all of which were diagnosable by RE assay a broad overlap in allele sizes between northern and southern (see Appendix). Sequencing at the Rho locus identi®ed two populations (data not shown). Although these loci are non- alleles private to each of the northern and southern popula- diagnostic there is no reason to discard the information they tions and one common to both. Both of the northern alleles do contain, thus they were combined to make multilocus and one of the southern alleles were distinguishable by RE hybrid index estimates (see below). Eleven previously sam- SECONDARY CONTACT IN CARLIA RUBRIGULARIS 1539

pled pure populations (mtDNA haplotypes either 100% north- ern or 100% southern), spanning approximately 40 kms out- 0 1 0.4 0.6 0.79 0.97 0.03 side the contact zone (Fig. 1), were screened at these micro- 0.00 (40) 0.21 (38) 0.00 (20) 0.00 (36) satellite loci allowing an isolation-by-distance estimate of dispersal. ed were BG, CR, 0 0.9 0.38 0.62 0.62 0.93 0.03 Analyses 0.00 (40) 0.38 (40) 0.10 (20) 0.05 (40) After calculation of FIS values all markers were collapsed to two allele sytems (i.e., ``northern'' alleles and ``southern'' alleles). Clines were ®tted to population allele frequency data 0.1 0.6 0.5 0.23 0.55 0.72 0.05 through the contact using the compound tanh and exponential 0.5 (20) 0.13 (40) 0.40 (40) 0.23 (40) model of Szymura and Barton (1986) implemented in the ``Analyse'' application (Barton and Baird 1995). Sigmoid clines are a very general product of gene ¯ow, with or without 0.7 0.1 0.07 0.25 0.54 0.77 0.67 various forms of selection (Barton and Gale 1993; Kruuk et 0.14 (28) 0.23 (30) 0.33 (15) 0.20 (30) al. 1999). A sigmoid cline in allele frequencies (p) is de- scribed as a function of its width and center such that, p ϭ (1 ϩ tanh[2(xϪc)/w])/2, (1) 0 0.06 0.11 0.53 0.24 0.32 0.29 S), NT, and B2. 0.31 (36) 0.76 (38) 0.68 (19) 0.71 (38) Ј where x is the distance from the center of the cline, c is the position of the center of the cline, and w is the width of the

14.76 cline (de®ned as the inverse of the maximum slope) (Szymura Њ and Barton 1986). We also wished to explore a more complex 0.21 0.06 0.29 0.12 0.18 0.38 0.06

E, 17 scenario of stepped clines. Theory predicts that when a num- 0.44 (34) 0.88 (34) 0.82 (17) 0.56 (34) Ј ber of selected characters change simultaneously across a zone of contact, association between them can cause a sharp 38.37 Њ step at the center of each cline. We follow the approach of 0 0.3 0.1 0.17 0.08 0.13 0.03 Szymura and Barton (1986) for stepped cline hypotheses. 0.9 (20) 0.83 (40) 0.63 (40) 0.85 (40) Outside the central region where these associations are lo- calized we assume that there is uniform impediment to in- S), LB (145

Ј trogression, such that clines in characters will decay expo- 0 0 0.1 0.04 0.36 0.07 0.04 nentially towards their tails as 5.97 Њ 0.50 (28) 1.00 (28) 0.93 (14) 0.96 (28) p ϰ exp(Ϫ4x␪1/2/w) (2) E, 17 Ј and the corollary for (1-p), where ␪ is the rate of exponential 0 0 0 0 0.11 0.15 0.08 decay. The central ``barrier'' region of a contact zone is 36.56 Њ 0.74 (38) 0.92 (38) 1.00 (19) 1.00 (38) speci®ed by equation (1) and encompasses a step in character state û dp/dx. The exponential decay curve equation (2) is used to describe the distal regions under reduced selection as tails with initial gradient dp/dx. The strength û of the 0 1 0.53 0.47 0.95 0.84 0.16 central selective barrier to hybridization is most conveniently 0.00 (118) 0.05 (216) 0.00 (109) 0.00 (174) expressed as the physical distance that would result in an equivalent change in character state in the absence of a bar- rier, and thus has units of distance. The decay rate of the tails is proportional to the level of selection acting on a char- 0 0 0 0 0.13 0.31 0.02 acter outside the central region. 0.56 (80) 0.98 (200) 1.00 (153) 1.00 (258) The present study system lends itself to a one dimensional cline analysis because of the narrowness of the strip of habitat between the altitudinal limit of Carlia distribution and the physical limit of the shore of Lake Tinaroo (Fig. 1). Cline ®tting was undertaken along a best ®t axis through sampled populations, equating to a compass bearing of 58.8Њ. Locality samples were weighted by effective sample size, a measure of the number of independent sampling events given observed genetic disequilibria. Deviation of the population from Har- Northern 1 Northern 2 Southern 1 Southern 2 &Northern Common Southern Northern Southern Northern Southern 1 Southern 2 dy-Weinberg equilibrium means the states of the two alleles 1. Frequency of diagnostic alleles used in this study at northern and southern sites and at sites within the contact zone. Northern populations screen sampled at a locus are not statistically independent. The same argument applies when there is linkage disequilibrium be- ABLE Locus Allele Northern pops Southern pops Site 1 Site 2 Site 3 Site 4 Site 5 Site 6 Site 7 Site 8 Site 9 Rho A16 DC, WH, and Site 1. Southern sites screened were LE, T6 (145 T CytB Ald tween loci, or reduced variance due to relatedness among 1540 BEN L. PHILLIPS ET AL. individuals in a sample (FST). Effective sample sizes (Ne), those solutions in which the estimated length of the stepped taking into account maximum-likelihood estimates (MLEs) portion of the cline was less than our lower bound for dis- of FIS and FST, were calculated for each sample at each locus persal. as (N. H. Barton, pers. comm.); Within the contact zone (populations 1±9) Hardy-Wein- 2N berg equilibrium (HWE) was assessed for each marker using Ne ϭ , (3) MLE FIS in Analyse. Hardy-Weinberg equilibrium was also 2NFSTϩ (1 ϩ F IS) assessed at each site for each marker within the contact zone. where N is the number of diploid individuals sampled. Ne is Average linkage disequilibrium (D) through the cline was scaled such that complete relatedness results in effective sam- assessed for diagnostic markers by partitioning the variance ple size ഠ1, whereas no relatedness but complete hetero- in ``hybrid index'' summed across diagnostic loci (Barton zygote de®cit results in effective allele sample size N. The and Gale 1993). average Ne/N ratio across sites ranged from ten to forty per- Two methods were used to estimate dispersal distance per cent across loci. Clines were ®tted to each locus indepen- generation. The ®rst used the relationship between pairwise dently using a maximum-likelihood (ML) approach that is FST (represented as FST/[1 Ϫ FST]) and the natural logarithm novel in some respects. We make comparisons within and of distance between sites at six microsatellite loci for 25 between two classes of hypotheses: simple contact, repre- population pairs in the two dimensional range of habitat out- sented by symmetrical sigmoid clines (``Sig,'' two param- side the contact. Rousset (1997) showed that the slope b of eters: w, c); and contact with a central barrier to gene ¯ow, a regression through such data under a two dimensional mod- represented by stepped clines. Barrier hypotheses are further el should yield an estimate of the product of density (d) and subdivided into simpler symmetric clines (``Sstep'' four pa- dispersal (␴2) such that 1/b ϭ 4d␴2. Pairwise estimates from rameters: w, c,û,␪) and more complex asymmetric clines ®ve populations ®xed for the northern mtDNA lineage (10 (``Astep'' six parameters: w, c,ûns, ␪ns,ûsn, ␪sn; subscripts pairs) spanning ca. 35 kms, and from six populations ®xed indicate south/north polarity). The models are nested; thus, for the southern mtDNA lineage (15 pairs) spanning about when two models are compared and one has more free pa- 7 kms, were pooled for this analysis. Least squares regression rameters (and therefore greater potential to ®t well), the par- was used to determine the slope and its standard error for simonious model is accepted over the more complex model these data thus allowing the approximate probability distri- if their difference in log likelihood (⌬LL) is not signi®cant bution of d␴2 to be calculated (using the standard errors and (chi-squared on two⌬LL with ⌬ parameters degrees of free- assuming a normal distribution of errors). Given a lack of dom) (Edwards 1972; see Barton 2000 for discussion). rigorous density estimates, ␴2 bounds were calculated for a Our approach to comparing clines extends existing max- density range of 20±450 individuals per hectare, which en- imum-likelihood cline ®tting procedures (Barton and Baird compasses published estimates for ecologically similar spe- 1995) by, for each model and each locus, exploring the like- cies (e.g., Patterson 1984, 416 haϪ1; Braithwaite 1987, 316 lihood surface stepwise along axes for both center position haϪ1; Sumner et al. 2001, 20 haϪ1). c and width w with the other parameters free to vary at each The second method for estimating dispersal uses the re- point. In this way the likelihood pro®les (Hilborn and Mangel lationship between maximum-linkage disequilibrium and the 1997) for both c and w can be constructed. Working with width of a cline (Barton and Gale 1993) and is appropriate these likelihood pro®les allows intuitively straightforward to the transect through the effectively one-dimensional (lin- inference about biologically relevant aspects of multilocus ear) habitat along the north shore of Lake Tinaroo. Under 2 hybrid zones. For example, coincident versus staggered cline this method, ␴ ϭ Drw1w2 where r is the rate of recombi- center hypotheses can be compared as follows: Summing log nation between two markers and w1 and w2 are the widths of likelihood c pro®les over a set of L loci results in the log- the clines for each marker. This method is independent of likelihood pro®le for the ML shared center of the L loci. The density and has been shown to be robust for weak to moderate center coincidence ML can be compared with the sum of levels of selection within a zone (Barton and Gale 1993). noncoincident c pro®le MLs using a likelihood ratio test (Hil- The ®ve microsatellite loci that did not show diagnostic dif- born and Mangel 1997). Summing log likelihoods over loci ferences between northern and southern populations did dif- implicitly assumes that the information at each locus is in- fer in allele frequency, enabling a multilocus assignment test dependent, so that the tightness of support bounds on the to correctly assign individuals from non hybrid populations location of a consensus center may be overestimated. How- to their respective lineage (Rieseberg et al. 1998). Using the ever if associations among loci are all of the same order of same procedure, individuals from hybrid populations were magnitude this will have little effect on the likelihood value analyzed using pooled northern and southern populations as at the consensus MLE, which is the basis for the comparison reference points. The most likely position on a linear vector of coincident center and staggered cline hypotheses. between these reference points was calculated for each in- When considering stepped clines, estimates and bounds for dividual's genotype with respect to the ®ve microsatellite ␤ and ␪ were calculated by an exhaustive search within the markers using the approach outlined in Rieseberg et al. two unit bound domain of our best cline ®t. Due to the low (1998). This position is analogous to a ``hybrid index'' cal- sampling density at the cline center (see Results) some so- culated by summation over diagnostic loci, and we refer to lutions for some loci were able to yield cline widths ap- it here as the nondiagnostic hybrid index (NDHI). Log like- proaching zero. However the area across which selection acts lihoods were summed over all individuals at each sampling in concert with linkage disequilibrium must be larger than locality, giving a ML estimate and support surface for the the dispersal distance of the organism. Therefore we omitted NDHI through the contact zone. The NDHI must be treated SECONDARY CONTACT IN CARLIA RUBRIGULARIS 1541 with the same caution as we would treat a hybrid index summed over ®ve diagnostic loci. Although it might be rea- sonable to assume that the ®ve loci have the same cline cen- ters, there is little reason to suppose they would all have the same width or cline shape (see Discussion). Taking a con- sensus over clines of different widths and symmetries tends to lead to an abrupt change in the central region where all clines concur. Therefore, the NDHI was deemed appropriate for determining the center position of the zone, but unsuitable for the estimation of other cline parameters.

RESULTS

Identi®cation of Diagnostic Markers and Location of the Contact Zone Screening of intron markers con®rmed that the Ald locus showed a ®xed difference between northern and southern populations. The Rho locus showed two alleles private to each population with the common (ϩ southern 2) allele class present at approximately 30% in northern populations (Table 1). At the A16 microsatellite marker, ``northern'' alleles ranged from approximately 98% frequency in northern pop- ulations to less than 5% in southern populations (Table 1). The other ®ve microsatellite loci showed extensive overlap in allele size between southern and northern populations, al- though there were signi®cant differences in allele frequency at all loci (data not shown). The fact that there were multiple northern and southern alleles at some of the diagnostic loci allowed us to test for Hardy-Weinberg equilibrium at these loci in populations out- side the contact zone. There was no signi®cant deviation in populations outside the contact zone for the Ald (MLE FIS ϭ 0.07, two unit bounds 0.0±0.37) or Rho (MLE FIS ϭ 0.00, two unit bounds 0.0±0.09) markers. However a signi®cant FIS (MLE FIS ϭ 0.18, two unit bounds 0.04±0.34) was de- tected at the A16 locus in pure southern populations sug- gesting that null alleles may complicate inference at this lo- cus. However, low FIS in the populations ¯anking the center of the hybrid zone makes it unlikely that null alleles are common in the localities central to our analysis. Initial haplotype screening revealed mixed populations within a discrete area of the Lamb Range (Fig. 1). Further sampling and screening of markers showed an essentially complete change from northern to southern types within four kilometers at all four diagnostic loci along a narrow strip of FIG. 2. (A) Best ®t clines in population allele frequency under a suitable habitat (Fig. 1) de®ned by the 900 m contour to the constrained center for the consistent barrier scenario. (B) Consensus north and the edge of a reservoir (established in 1958) to the FIS estimates for A16 and Rho through the cline. (C) Average link- south. Within this area we sampled nine sites with an average age disequilibrium estimates through the cline. Hybrid index fre- quency histograms are shown for sites 4, 5, and 6 (insets: x-axis of 18 individuals per site (Figs. 1, 2A). Two of these sites is the hybrid index, y-axis the frequency). In all cases, error bars (2 and 6), being the populations that initially revealed the represent two unit support bounds and the c-axis measures distance presence of the contact, were sampled two months earlier (m) through the cline (consensus center located at c ϭ 2840). than other populations.

Genotypic and Allelic Disequilibrium in the Contact Zone 2A). For the ®ve nondiagnostic microsatellite loci there was a rapid shift in NDHI within the zone (Fig. 2A). Nuclear loci with diagnostic differences between northern Estimates of FIS indicated de®ciencies of heterozygotes at and southern mtDNA lineages showed rapid transition in al- two loci, Rho (FIS ϭ 0.42; 2 unit bounds 0.23±0.59) and A16 lele frequencies across the 4 km contact zone in broad agree- (FIS ϭ 0.13; two unit bounds 0±0.28). By contrast, analysis ment with the location of the mtDNA cline (Table 1; Fig. of genotype frequencies at the third diagnostic nuclear locus, 1542 BEN L. PHILLIPS ET AL.

TABLE 2. Observed and expected number of heterozygotes totaled TABLE 3. Maximum likelihood of Sig and Astep models for each across all sites in the contact. Heterozygotes have been split into locus. G-statistics (twice the difference in log likelihood) are used across-lineage (i.e., N ϫ S) and within-lineage (N1 ϫ N2 and/or to assess the signi®cance of the improvement offered by the Astep S1 ϫ S2) classes. model. No improvement between the models was shown by the NDHI data. Within lineage Within lineage Across lineage Across lineage heterozygotes heterozygotes heterozygotes heterozygotes Locus Sig ML Astep ML G df P Locus (expected) (observed) (expected) (observed) A16 Ϫ6.637 Ϫ5.449 2.376 2 0.305 Ald 6.7 5 36.5 38 A1d Ϫ11.854 Ϫ4.126 15.456 4 0.003 Rho 40.2 39 45.2 26 Rho Ϫ4.113 Ϫ1.146 5.934 4 0.204 A16 24.3 20 51.5 44 CytB Ϫ4.168 Ϫ1.567 5.202 4 0.267

Ald, produced an FIS estimate of zero (two unit bounds 0± 0.16). Plotting consensus FIS estimates for A16 and Rho against distance along the transect suggested that maximum de®ciency of heterozygotes occurred at site 4, just to the left of cline centers (Fig. 2B). Within the zone, comparison of (total) expected versus observed values for individual ge- notypes indicated strong heterozygote de®cit for combina- tions of N and S alleles, but not N ϫ NorSϫ S alleles at Rho and an approximately equal de®ciency of heterozygotes in both classes at the A16 locus (Table 2). Across the three diagnostic nuclear loci, signi®cant average pairwise linkage disequilibrium (D) was detected with high- est values at sites 4 and 6 near the center of the zone (Fig. 2C). In this context, the lower estimate of D at site 5 appears anomalous. Post hoc inspection of hybrid index distributions for sites 4, 5, and 6 (Fig. 2C, insets) showed that high linkage disequilibrium was generated at sites 4 and 6, by the presence of individuals with extreme hybrid index values. Interest- ingly, these extreme values were representative of northern and southern types at sites 4 and 6 respectively, as would be expected if linkage disequilibrium was generated primarily by immigration of parental genotypes into the zone. Given the noise inherent in this parameter, we constructed a con- servative estimate for the maximum linkage disequilibrium in the zone by averaging the estimates of sites 4 and 6 and taking the lowest and highest support bounds over these sites. This yielded an estimate of maximum average pairwise link- age disequilibrum of 0.0885 (38% of its maximum possible value) with limits ranging from 0.0538 to 0.128.

Likelihood Inference about the Nature of the Contact Zone We plotted the log-likelihood pro®les of center position for each cline model and for each diagnostic marker locus (Fig. 3). This process revealed several observations of note: First, for all loci except A16, the Sstep model provided neg- ligible improvement over the Sig model and was therefore dropped from further analyses. Second, in all cases the Astep model provided an increase in likelihood over the Sig model (Fig. 3, Table 3). The magnitude of this improvement de- pended upon where the center position was located and in the individual locus tests the difference was signi®cant for the Ald locus (P ϭ 0.003, Table 3), but was not signi®cant for Rho or A16. The stepped nature of the ALD cline and the heterozygote de®cits at the other two nuclear loci, Rho and A16, are clear evidence of selection against hybrids, and FIG. 3. Marginal likelihood surfaces along the center position (c) axis. All three models (Sig, Sstep, and Astep) are compared for thus imply a barrier to gene ¯ow at the center of the zone. each locus. Likelihood values are for the best ®tting model at each The detection of signi®cant effects of a barrier at one locus, particular center position (all other parameters free to vary). but nonsigni®cant effects at the other loci suggests two sce- SECONDARY CONTACT IN CARLIA RUBRIGULARIS 1543

TABLE 4. Summary of best-®t cline parameters under the consistent barrier and mixed holistic scenarios with a consensus center. Asterisks denote situations where our data could not provide robust estimates/bounds.

w (m) ␤SN qSN ␤NS qNS ML Mixed (c ϭ 2600) Cyt 2u Min. 1650 Ð Ð Ð Ð MLE 2240 Ð Ð Ð Ð Ϫ4.087 2u Max. 3150 Ð Ð Ð Ð Ald 2u Min. 760 460.45 0.05 0* 0.00* MLE 1040 1150±1430 0.20±0.23 39.3-* 0.00±1* Ϫ8.861 2u Max. 1490 3421 0.65 * 1* A16 2u Min. 90 Ð Ð Ð Ð MLE 450 Ð Ð Ð Ð Ϫ6.815 2u Max. 1230 Ð Ð Ð Ð Rho 2u Min. 670 Ð Ð Ð Ð MLE 1070 Ð Ð Ð Ð Ϫ2.438 2u Max. 1790 Ð Ð Ð Ð Consistent Barrier (c ϭ 2840) Cyt 2u Min. 640 121 0.03 43* 0.00* MLE 1121±1135 809±937 0.33 4452-* 0.00±0.80* Ϫ1.75 2u Max. 1412 8225 0.66 * 1* Ald 2u Min. 568 104 0.08 157 0.01* MLE 1030±1050 306±344 0.31±0.32 4813- 0.01±0.99* Ϫ9.317 2u Max. 1980 683 1 * 1* A16 2u Min. 222 13 0.12* 14825 0.31 MLE 588±610 13±21 0.99±1* 14825- 0.33±0.80 Ϫ6.512 2u Max. 1550 234* 1* * 0.8 Rho 2u Min. 210 78 0 116 0.00* MLE 509±539 598 0.05 63872±69500 0.88±0.92* Ϫ2.646 2u Max. 1280 1830 0.99* * 1* narios in which to assess overall properties of the hybrid sistent barrier scenario, cline widths were more similar across zone (see discussion): (1) A mixed model (one locus ϭ Astep; loci (MLEs 509±1135 m) and narrowest at Rho and Ald, the the rest ϭ Sig), where a barrier to gene ¯ow affects only the two loci with heterozygote de®cits (Table 4). Overall, esti- chromosome segment marked by the Ald locus. All other loci mates of cline width (mixed w Ͻ2240 m, two unit bounds have sigmoid clines, and their higher likelihoods under a 90±3150 m; consistent barrier w Ͻ1135 m, two unit bounds barrier model are assumed to be due to the greater degrees 210±1980 m) were narrow compared to the geographic scale of freedom of that model. (2) A consistent barrier model (all of many hybrid zones (65% of zones reported in Barton and loci ϭ Astep), where the barrier affecting Ald is also affects Hewitt (1985) had widths greater than our maximum esti- other loci, but their increased likelihood under the barrier mate). model is not signi®cant due to insuf®cient power to detect weaker effects than that those operating on Ald. Estimates of Dispersal Distance and Selection Strength Our choice of descriptive model (mixed vs. consistent bar- rier) does not affect the qualitative conclusion that selection Comparison among pure northern and pure southern pop- is operating against hybrids, but does in¯uence estimates of ulations at the microsatellite loci revealed pairwise FST values cline width and consequently the strength of that selection. range from 0.012 to 0.068 (over 35 km) among northern Visual inspection of likelihood pro®les (Fig. 3) suggest ap- populations and 0.008 to 0.046 (over 7 km) among southern proximate coincidence of cline centers under both mixed and populations. Using the combined pairwise estimates, regres- consistent barrier scenarios. Constraining all loci to a com- sion of FST/(1 Ϫ FST) and log distance revealed an isolation mon center does not result in a signi®cant decrease in like- by distance effect with a slope of 0.00857 (R2 ϭ 0.26, SE lihood under either model (mixed, P ϭ 0.20; consistent bar- ϭ 0.0031, P ϭ 0.010; because FST is imperfectly measured, rier, P ϭ 0.77). Under the mixed scenario, cline widths were the signi®cance of the slope may be an overestimate and variable (MLEs 450±2240 m; Table 4), but under the con- con®dence limits underestimated). These values were trans- 1544 BEN L. PHILLIPS ET AL. formed to a probability density function of d␴2. This gave a rainforests (Nix 1991; Hugall et al. 2002) and preliminary median value for neighborhood size of 9.28 individuals and models speci®c for C. rubrigularis (B. Phillips, unpubl.) sug- a (minimum) 95% con®dence range of 5±35 individuals. As- gest that the major refugia for this species were in the Thorn- suming Carlia densities lie in the range d ϭ 20±450 indi- ton Uplands, 100 km north of the contact zone and the eastern viduals per hectare yields ␴ values ranging approximately Atherton uplands 25 km south of the zone, and coastal regions 11±133 m ϫ genϪ1/2. adjacent to each area. Thus, the contact zone is much closer Using the estimate of maximum D and its bounds from the to the presumed southern refugium than to the one in the contact zone (see above) and the estimates of cline width for north. This contrasts with Endler's (1982b) assumption that our narrowest clines under each scenario (because these are the zones of secondary contact should be equidistant from re- ones most likely to be contributing to disequilibrium, Table 4) fugia. Interestingly, zones of secondary contact between yields ␴ values ranging from 112 to 173 m ϫ genϪ1/2 under northern and southern phylogeographic lineages have been the mixed scenario and 90 to 140 m ϫ genϪ1/2 under the con- located for several other species in this region: the tropical sistent barrier scenario. This assumes that there is no physical bettong Bettongia tropica, 5 km to the north (Pope et al. linkage between the two markers and that linkage disequilibrium 2000); the microhylid frog Cophixalus ornatus,17kmtothe is primarily due to migration rather than selection. Each method south (C. Hoskin, unpubl. data); the snail Gnarosophia bel- for estimating dispersal is subject to considerable uncertainty lendenkerensis, 12.5 km to the south (Hugall et al. 2002); and makes several untested assumptions. Nonetheless, the two and the hylid frog Litoria genimaculata, 15±38 km north (B. independent approaches give reasonably consistent ranges of Phillips and C. Hoskin, unpubl. data). Thus, the Lamb Range values and in the following we use the overlap, 112±133 m ϫ represents a ``phylogeographic'' suture zone (Remington genϪ1/2 (mixed model) and 90±133 m ϫ genϪ1/2 (consistent bar- 1968), as described by (Hewitt 2000) for the European Alps, rier model), as our best estimate of dispersal rate (␴) in the although here with spatial congruence at a much ®ner spatial contact zone. scale. Pollen cores from two nearby lakes (Lake Euramoo Given these estimates of cline width and dispersal, and and Lake Barrine, Fig. 1) provide evidence for the displace- assuming clines have reached equilibrium between selection ment of dry open forests by rainforest across this area be- and migration, we can calculate an effective level of selection tween 6100 and 7500 years ago (Walker 1990). Although C. s*. This is the selection against heterozygotes which would rubrigularis could well have been at the forefront of rainforest be required to maintain a cline of the same width were the expansion because of its preference for edge habitats, it is hybrid zone a single locus system; s* ϭ 8(␴/w)2 (Barton and also possible that the contact examined here may not have Gale 1993). Using our estimate of dispersal rate and the above arisen until the warmer wetter period 3600±5000 years ago equation, we estimate s* as lying between 0.50±0.70 (mixed) (Nix 1991). Prior to this time, much of the Lamb Range may or 0.22±0.49 (consistent barrier). Given the uncertainties and have acted as a physical barrier to contact (much as the 900 assumptions involved, these should be treated only as indi- m contour does today) due to the prevailing cooler, wetter cations of the order of magnitude of selection in the zone. conditions (Nix 1991) that are avoided by C. rubrigularis. Nevertheless, the width of the clines relative to our estimate Considering these factors, the contact observed here could of dispersal indicates that selection against hybrids is strong be as old as 3600±7500 years. irrespective of which cline model is chosen. Under the con- Given the evidence from heterozygote de®cits of nonneu- sistent barrier model, further information can be obtained trality (i.e., selection acting against hybrids), we cannot sen- from the shape of the tails of the clines. The Astep model sibly estimate the age of the zone directly from its width. estimates barrier strength (␤) and the rate of exponential de- The presence of mixed populations to the south of Lake Tin- cay (␪) for northern to southern (geographically west to east, aroo (Fig. 1) implies that contact at least predates the creation Fig. 1) and southern to northern lineages simultaneously. For of the lake in 1958 and the paleoecological evidence suggests all diagnostic markers the MLE selective barrier north to that it could be thousands of years old. With moderate to strong selection against hybrids, as appears to be the case south (␤ns) is greater than the geographic extent of the hybrid zone, and greater than the corresponding barrier south to here, the equilibrium cline width should be determined pri- north (Table 4). Intense sampling in the tails of clines is marily by dispersal distance (Barton and Gale 1993) and necessary for precise estimates of the decay parameters (Bar- equilibrium should thus have been attained rapidly. We there- ton and Gale 1993), and as a result support limits here are fore assume that the hybrid zone is at equilibrium. wide. We restrict our inference to the following: Decay pa- Subsequent to locating the zone, it was characterized ge- rameter estimates (␪) for all markers except A16 are consis- netically for eight nuclear loci, three of which showed ®xed tent with relatively weak selection outside the barrier region (Ald) or nearly ®xed (Rho, A16) differences between north- (lower support limits Ͻ 10%). A16 not only has a strong ern and southern mtDNA lineages. Key observations from central barrier, its decay estimates indicate it may also be this analysis are: in¯uenced by selection outside the center of the zone. (1) The zone is narrow, with width estimates of 450 m to 2.2 km and cline centers that are coincident. (2) There is substantial linkage disequilibrium and some, DISCUSSION but not all, diagnostic loci show strong de®ciencies of het- We have located and characterized genetically a zone of erozygotes. Both heterozygote de®cits and linkage disequi- secondary contact between northern and southern mtDNA librium show peak values on the northern side of cline cen- lineages of C. rubrigularis in the southern reaches of the ters. Lamb Range. Paleoclimatological modeling of mesothermal (3) At least one locus shows a stepped and asymmetrical SECONDARY CONTACT IN CARLIA RUBRIGULARIS 1545 cline, but we are unable to distinguish rigorously between between northern and southern lineages of C. rubrigularis, the mixed model in which some loci have sigmoidal clines especially if selection against hybrids was weaker or if the and others are stepped, and the consistent barrier model in contact was maintained for longer. The contact zone could which all loci are stepped. be considered as an evolutionary ®lter, allowing exchange of (4) Under the consistent barrier scenario, all clines are alleles that are bene®cial, but rarely ones that are neutral or found to be asymmetric in the same direction and estimates subject to divergent selection (Martinsen et al. 2001). of cline width tend to decrease with an increase in observed The observation of strong selection against hybrids be- heterozygote de®cits. tween morphologically similar but phylogeographically dis- Taken together, these observations on the genetic structure tinct lineages of C. rubrigularis is consistent with the view of the contact zone suggest strong selection against hybrids that phylogeographic lineages can represent stages in the es- that, in the absence of heterozygote de®cit at one of the tablishment of reproductively isolated lineages via gradual, diagnostic loci, we assume to be primarily postzygotic rather incidental divergence in allopatry (Avise and Walker 1998). than prezygotic. Further, the consistent asymmetry of any The establishment of reproductive isolation by this means is barriers considered in the zone implies differences in ®tness expected to be slow (Gavrilets 2003). In contrast to the mor- or effective gene ¯ow rate with greater introgression from phologically conservative skinks studied here, it is notable the southern to northern lineage than vice versa. that some sister taxa of insects from these same rainforests, Whether there is a de®cit of hybrid genotypes depends on and which have experienced an analogous history of habitat which locus is evaluated (cf. Ald vs. Rho and A16), despite ¯uctuation, have attained complete reproductive isolation strong average pairwise linkage disequilibrium. This obser- within this suture zone (e.g., Bell et al. 2004). Additional vation suggests that strong postzygotic selection against hy- studies of contact zones involving morphologically cryptic brid genotypes is occurring in association with some loci and phylogroups are needed to explore the relationship between not others. The contact zone is positioned within an essen- interlineage divergence and levels of selection against hy- tially homogeneous habitat with no obvious dispersal bar- brids and how this might vary across taxonomic groups (Tur- riers. It follows that the shape and width of the clines are elli et al. 2001). As sequence divergence between mtDNA largely a function of dispersal and the degree of endogenous phylogroups can be a poor predictor of divergence time (Ed- selection against hybrids. wards and Beerli 2000; Hudson and Turelli 2003), such stud- There have been numerous studies of cline properties at ies would bene®t from a multilocus perspective. That there secondary contacts between phenotypically and/or chromo- is evidence for partial reproductive isolation between mtDNA somally distinct taxa (reviewed by Barton and Hewitt 1985; phylogroups of C. rubrigularis does not necessarily support Arnold 1994; Morgan-Richards and Wallis 2003). However the view that mtDNA differences alone can be used to di- few studies have combined quantitative estimates of dispersal agnose species (e.g., Wiens and Penkrot 2002). The diversity and cline width to estimate effective selection strength (s*) of views on concepts of species results in part from varying under the tension zone model. In comparison to such esti- emphases on diagnosis of taxa versus the processes by which mates involving distinct species (Bombina, s* ϭ 22%, Szy- independent evolutionary lineages arise (de Queiroz 1998; mura and Barton 1991; Heliconius, s* ϭ 20±30%, Mallett et Harrison 1998). Inevitably, this debate spills over into one al. 1990; Pontia, s* ϭ 47±64%; Porter et al. 1997) and chro- about explicit procedures and criteria for diagnosing species mosome races (e.g., Sceloporus, s* ϭ 30%, Marshall and (Sites and Marshall 2003). The inherent ambiguity is ex- Sites 2001), our analyses suggest relatively strong effective empli®ed by the complex studied hereÐthe two lineages of selection (s* ϭ 22±49% for consistent barrier model; s* ϭ C. rubrigularis plus C. rhomboidalis. Using phylogeographic 50±70% for mixed model) against hybrids between phylo- (Wiens and Penkrot 2002) or phylogenetic (Cracraft 1983; geographic lineages within C. rubrigularis. Our estimates rest Goldstein and DeSalle 2000) approaches, the two (nonsister) on the assumptions of the models (above) and the validity lineages of C. rubrigularis clearly are diagnosable and would of the dispersal estimate, the latter being a weak point in be regarded as separate species. By contrast, they could be many previous studies (Barton and Hewitt 1985). In the pre- considered conspeci®c (albeit paraphyletic) under the cohe- sent case, two independent genetic approaches gave broadly sion species concept (Templeton 2001) as the null hypothesis consistent estimates of dispersal rate and, notwithstanding of ecological exchangeability (using ecomorphology as a sur- uncertainty about population density, produced results that rogate; Schneider et al. 1999) is not rejected. Their status are plausible in relation to studies of other rainforest skinks under a genealogical species concept (Baum and Shaw 1995) (e.g., Sumner et al., 2001). is unclear as the attainment of genealogical concordance be- In the absence of assortative mating, these high values of tween even 50% of nuclear loci and mtDNA is a slow and effective selection against hybrids do not necessarily imply highly stochastic process (Hudson and Coyne 2002). Also, that the northern and southern lineages are evolving inde- there is potential for introgression of adaptive alleles (Barton pendently. The current selection barrier is suf®cient to im- and Hewitt 1985; Porter et al. 1997). More generally, the pede both negatively selected and neutral alleles, but alleles suggestion (Wiens and Penkrot 2002) that mtDNA represents that are selectively favored in both lineages should pass an ideal marker for diagnosis of species boundaries suffers through this barrier relatively quickly, in a time related to from three signi®cant limitations. First, as Wiens and Penkrot the degree of increased ®tness they confer, relative to the (2002) acknowledge, as a maternally inherited locus, mtDNA strength of selection against hybrids (Pialek and Barton can indicate diagnostic differences between populations de- 1997). Further, it is possible that there has been episodic spite substantial exchange of nuclear genes, especially if there introgression during earlier (interglacial) periods of contact is male-mediated gene ¯ow or polygyny (e.g., Birky et al. 1546 BEN L. PHILLIPS ET AL.

1989; Hoelzer 1997). Second, as a single gene, estimates of Braithwaite, R. W. 1987. Effects of ®re regimes on lizards in the population history derived from mtDNA alone are subject to Wet-Dry Tropics of Australia. J. Trop. Ecol. 3:265±275. Cogger, H. G. 2000. Reptiles and amphibians of Australia. Reed, considerable error arising from stochastic among-locus var- Melbourne. iance in coalescent processes (Edwards and Beerli 2000; Ir- Cracraft, J. 1983. Species concepts and speciation analysis. Curr. win 2002; Hudson and Turelli 2003). Third, selection acting Ornith. 1:159±187. on nuclear and/or mitochondrial sequences can cause sub- de Queiroz, K. 1998. The general lineage concept of species, species criteria, and the process of speciation: a conceptual uni®cation stantial differences in both the rate of divergence and spatial and terminological recommendations. Pp. 57±75 in D. J. Howard patterns of diversity, such that mtDNA can be no more ef- and S. H. Berlocher, eds. Endless forms. Species and speciation. ®cient, or perhaps even misleading as an indicator of lineage Oxford Univ. Press, Oxford, U.K. boundaries (e.g., Garcia-Paris et al. 2003). For these reasons Dobzhansky, T. 1937. Genetics and the origin of species. Columbia we, along with others (e.g., Jockusch and Wake 2002) urge Univ. Press, New York. ÐÐÐ 1970. Genetics and the evolutionary process. Columbia that diagnosis of species be based on multiple lines of evi- Univ. Press, New York. dence, that is, phenotypes or nuclear loci in addition to Edwards, A. W. F. 1972. Likelihood. Cambridge Univ. Press, Cam- mtDNA. bridge, U.K. Edwards, S. V., and P. Beerli. 2000. Perspective: gene divergence, population divergence and the variance in coalescence time in ACKNOWLEDGMENTS phylogeographic studies. Evolution 54:1839±1854. The authors wish to thank P. Fox, D. O'Connor, J. Dickson, Endler, J. A. 1977. Geographic variation, speciation and clines. Princeton Univ. Press, Princeton, NJ. C. Hoskin, C. Larroux, and S. Phillips for help in the ®eld. ÐÐÐ. 1982a. Problems in distinguishing historical from ecolog- Special thanks to A. Hugall for advice at all stages of this ical factors in biogeography. Am. Zool. 22:441±452. project. C. Schneider and M. Lara provided the microsatellite ÐÐÐ. 1982b. Pleistocene forest refuges: fact or fancy? Pp. 641± primers and associated technical assistance. D. Wake and J. 657 in G. Prance, ed. Biological diversi®cation in the tropics. Columbia Univ. Press, New York. Alexandrino provided valuable discussion. The manuscript Garcia-Paris, M., M. Alcobendas, D. Buckley, and D. B. Wake. was improved by comments from C. Schneider, C. Graham, 2003. Dispersal of viviparity across contact zones in Iberian A. Porter, and anonymous reviewers. The research was fund- populations of ®re salamanders (Salamandra) inferred from dis- ed by grants from the Cooperative Research Center for Trop- cordance of genetic and morphological traits. Evolution 57: ical Rainforest Ecology and Management (Australia), the Na- 129±143. Gavrilets, S. 2003. Models of speciation: what have we learned in tional Science Foundation, and the Australian Research 40 years. Evolution 57:2197±2215. Council. Goldstein, P. Z., and R. DeSalle. 2000. Phylogenetic species, nested hierarchies, and character ®xation. Cladistics 16:364±384. Harrison, R. G. 1993. Hybrid zones and the evolutionary process. LITERATURE CITED Oxford Univ. Press, Oxford, U.K. Arnold, M. L. 1994. Natural hybridisation and evolution. Oxford ÐÐÐ. 1998. Linking evolutionary pattern and process: The rel- Univ. Press, Oxford, U.K. evance of species concepts for the study of speciation. Pp. 19± Avise, J. C. 2000. Phylogeography: the history and formation of 31 in D. J. Howard and S. H. Berlocher, eds. Endless forms. species. Harvard Univ. Press, Cambridge, MA. Species and speciation. Oxford Univ. Press, Oxford, U.K. Avise, J. C., and D. Walker. 1998. Pleistocene phylogeographic Hewitt, G. M. 2000. The genetic legacy of Quaternary ice ages. effects on avian populations and the speciation process. Proc. Nature 405:907±913. R. Soc. Lond. B 265:457±463. ÐÐÐ. 2001. Speciation, hybrid zones and phylogeographyÐor Ballard, J. W. O., B. Chernoff, and A. C. James. 2002. Divergence seeing genes in space and time. Mol. Ecol. 10:537±550. of mitochondrial DNA is not corroborated by nuclear DNA, Hey, J. 2001. The mind of the species problem. Trends Ecol. Evol. morphology, or behavior in Drosophila simulans. Evolution 56: 16:326±329. 527±545. Hilborn, R., and M. Mangel. 1997. The ecological detective: con- Barton, N. H. 2000. Estimating multilocus linkage disequilibria. fronting models with data. Princeton Univ. Press, Princeton, NJ. Heredity 84:373±389. Hopkins, M. S., J. Ash, A. W. Graham, J. Head, and R. K. Hewett. Barton, N. H., and S. J. E. Baird. 1995. Analyse: an application for 1993. Charcoal evidence of the spatial extent of the Eucalyptus analysing hybrid zones. Freeware, Edinburgh. Available at woodland expansions and rainforest contractions in north http://helios.bto.ed.ac.uk/evolgen/mac/analyse. Queensland during the late Pleistocene. J. Biogeogr. 20:59±74. Barton, N. H., and K. S. Gale. 1993. Genetic analysis of hybrid Hoelzer, G. A. 1997. Inferring phylogenies from mtDNA variation: zones. Pp. 13±45 in R. G. Harrison, ed. Hybrid zones and the mitochondrial-gene trees versus nuclear-gene trees revisited. evolutionary process. Oxford Univ. Press, Oxford, U.K. Evolution 51:622±626. Barton, N. H., and G. M. Hewitt. 1985. Analysis of hybrid zones. Hudson, R. R., and J. A. Coyne. 2002. Mathematical consequences Annu. Rev. Ecol. Syst. 16:113±148. of the genealogical species concept. Evolution 56:1557±1565. ÐÐÐ. 1989. Adaptation, speciation, and hybrid zones. Nature Hudson, R. R., and M. Turelli. 2003. Stochasticity overrules the 341:497±503. ``three-times rule'': genetic drift, genetic draft and coalescence Baum, D. A., and K. L. Shaw. 1995. Genealogical perspectives on times for nuclear loci versus mitochondrial DNA. Evolution 57: the species problem. Pp. 289±303 in P. C. Hoch and A. G. 182±190. Stephens, eds. Experimental and molecular approaches to plant Hugall, A., C. Moritz, A. Moussalli, and J. Stanisic. 2002. Rec- biosystematics. Missouri Botanical Garden, St. Louis, Missouri. onciling paleodistribution models and comparative phylogeog- Bell, K. L., D. K. Yeates, C. Moritz, and G. B. Monteith. 2004. raphy of the Wet Tropics rainforest land snail Gnarasophia bel- Molecular phylogeny and biogeography of the dung beetle genus lendenkerensis (Brazier 1875). Proc. Natl. Acad. Sci. USA 99: Temnoplectron Westwood (Scarabaeidae: Scarabaeninae) from 6112±6117. Australia's wet tropics. Mol. Phylogenet. Evol. In press. Ingram, G., and J. Covacevich. 1989. Revision of the genus Carlia Birky, C. W., Jr., P. Fuerst, and T. Maruyama. 1989. Organelle (Reptilia: Scincidae) in Australia with comments on Carlia bi- gene diversity under migration, mutation and drift: equilibrium carinata of New Guinea. Mem. Queensl. Mus. 27:443±490. expectations, approach to equilibrium, effects of heteroplasmic Irwin, D. 2002. Phylogeographic breaks without geographic barriers cells and comparison to nuclear genes. Genetics 121:613±627. to gene ¯ow. Evolution 56:2383±2394. SECONDARY CONTACT IN CARLIA RUBRIGULARIS 1547

Jockusch, E. L., and D. B. Wake. 2002. Falling apart and merging: determined using mtDNA and microsatellites. Mol. Ecol. 9: diversi®cation of slender salamanders (Plethodontidae: Batra- 2041±2053. choseps) in the American West. Biol. J. Linn. Soc. 76:361±391. Porter, A. H., R. Wenger, H. Geiger, A. Scholl, and A. M. Shapiro. Joseph, L., C. Moritz, and A. Hugall. 1995. Molecular support for 1997. The Pontia daplidice-edusa hybrid zone in northwestern vicariance as a source of diversity in rainforest. Proc. R. Soc. Italy. Evolution 51:1561±1573. Lond. B 260:177±182. Remington, C. L. 1968. Suture-zone of hybrid interaction between Kershaw, A. P. 1986. Climatic change and Aboriginal burning in recently joined biotas. Evol. Biol. 2:321±428. north-east Australia during the last two glacial/interglacial cy- Rieseberg, L. H., S. J. E. Baird, and A. M. DesRocher. 1998. Pat- cles. Nature 322:47±49. terns of mating in wild sun¯ower hybrid zones. Evolution 52: ÐÐÐ. 1994. Pleistocene vegetation of the humid tropics of north- 713±726. eastern Queensland, Australia. Paleogeog. Paleoclim. Paleoecol. Rousset, F. 1997. Genetic differentiation and estimation of gene 109:399±412. ¯ow from F-statistics under isolation by distance. Genetics 145: Kruuk, L. E. B., S. J. E. Baird, N. H. Barton, and K. S. Gale. 1999. 1219±1228. A comparison of multilocus clines maintained by environmental Schneider, C. J., M. Cunningham, and C. Moritz. 1998. Comparative adaptation or by selection against hybrids. Genetics 153: phylogeography and the history of endemic vertebrates in the Wet Tropics rainforests of Australia. Mol. Ecol. 7:487±498. 1959±1971. Schneider, C. J., and C. Moritz. 1999. Rainforest refugia and evo- Mallett, J., N. Barton, G. M. Lamas, J. C. Santisteban, M. M. Mu- lution in Australia's Wet Tropics. Proc. R. Soc. Lond. B 266: sedas, and H. Eeley. 1990. Estimates of selection and gene ¯ow 191±196. from measures of cline width and linkage disequilibrium in Hel- Schneider, C. J., T. B. Smith, B. Larison, and C. Moritz. 1999. A iconius hybrid zones. Genetics 124:921±936. test of alternative models of diversi®cation in tropical rainfo- Martinsen, G. D., T. G. Whitham, R. J. Turek, and P. Keim. 2001. rests: ecological gradients vs. rainforest refugia. Proc. Nat. Acad. Hybrid populations selectively ®lter gene introgression between Sci. USA 96:13869±13873. species. Evolution 55:1325±1335. Sites, J. W., Jr.,and J. C. Marshall. 2003. Delimiting species: a Marshall, J. C., and J. W. Sites. 2001. A comparison of nuclear and renaissance issue in systematic biology. Trends Ecol. Evol. 18: mitochondrial cline shapes in a hybrid zone in the Sceloporus 462±470. grammicus complex (Squamata: Phrynosomatidae). Mol. Ecol. Stuart-Fox, D. M., A. F. Hugall, and C. Moritz. 2002. A molecular 10:435±449. phylogeny of rainbow skinks (Scincidae: Carlia): taxonomic and Morgan-Richards, M., and G. Wallis. 2003. A comparison of ®ve biogeographic implications. Aust. J. Zool. 50:39±51. hybrid zones of the weta Hemideina thoracica (Orthoptera: An- Sumner, J., F. Rousset, A. Estoup, and C. Moritz. 2001. ``Neigh- ostostomatidae): degree of cytogenetic differentiation fails to bourhood'' size, dispersal, and density estimates in the prickly predict zone width. Evolution 57:849±861. forest skink (Gnypetoscincus queenslandiae) using individual ge- Moritz, C. 2002. Strategies to protect biological diversity and the netic and demographic methods. Mol. Ecol. 10:1917±1927. evolutionary processes that sustain it. Syst. Biol. 51:238±254. Szymura, J. M., and N. H. Barton. 1986. Genetic analysis of a hybrid Moritz, C., L. Joseph, M. Cunningham, and C. J. Schneider. 1997. zone between the ®re-bellied toads, Bombina bombina and B. Molecular perspectives on historical fragmentation of Australian variegata near Cracow in southern Poland. Evolution 40: tropical and sub-tropical rainforests. Pp. 442±454 in W. Laur- 1141±1159. ÐÐÐ. 1991. The genetic structure of the hybrid zone between the ence and R. Bierragaard, eds. Tropical forest remnants: ecology, ®re-bellied toads Bombina bombina and B. variegata: Compar- management and conservation of fragmented communities. isons between transects and loci. Evolution 45:237±261. Univ. of Chicago Press, Chicago, IL. Templeton, A. R. 2001. Using phylogeographic analyses of gene Nix, H. 1991. Biogeography: pattern and process. Pp. 11±39 in H. trees to test species status and processes. Mol. Ecol. 10:779±791. Nix and M. Switzer, eds. Rainforest animals: atlas of vertebrates Turelli, M., N. H. Barton, and J. A. Coyne. 2001. Theory and spe- endemic to Australia's Wet Tropics. Australian Nature Conser- ciation. Trends Ecol. Evol. 16:330±343. vation Agency, Canberra. Wake, D. B., and C. J. Schneider. 1998. Taxonomy of the pleth- Nix, H., and M. Switzer. 1991. Rainforest animals: atlas of ver- odontid salamander genus Ensatina. Herpetologica 54:279±298. tebrates endemic to Australia's Wet Tropics. Australian Nature Walker, D. 1990. Directions and rates of tropical rainforest pro- Conservation Agency, Canberra. cesses. Pp. 22±32 in L. Webb and J. Kikkawa, eds. Australian Orr, M. R., and T. B. Smith. 1998. Ecology and speciation. Trends tropical rainforests: scienceÐvaluesÐmeaning. Commonwealth Ecol. Evol. 13:502±506. Scienti®c and Industrial Research Organization Australia, Can- Patterson, G. B. 1984. The effect of burning-off tussock grassland berra. on the population density of common skinks Leiolopisma-nigri- Webb, L., and J. G. Tracey. 1981. Australian rainforests: pattern plantare-maccanni. N.Z. J. Zool. 11:189±194. and change. Pp. 605±694 in A. Keast, ed. Ecological biogeog- Pialek, J., and N. H. Barton. 1997. The spread of an advantageous raphy of Australia. Junk, The Hague. allele across a barrier: the effects of random drift and selection Wiens, J. J., and T. A. Penkrot. 2002. Delimiting species using DNA and morphological variation and discordant species limits against heterozygotes. Genetics 145:493±504. in spiny lizards (Sceloporus). Syst. Biol. 51:69±91. Pope, L. C., A. Estoup, and C. Moritz. 2000. Phylogeography and population structure of an ecotonal marsupial, Bettongia tropica, Corresponding Editor: J. Wiens 1548 BEN L. PHILLIPS ET AL.

APPENDIX The diagnostic loci used in this study. The tables below indicate (A) the allele classes distinguishable and the speci®c fragment patterns generated by the REs used; and (B) the primers and their sequences used for each locus.

(A) Locus Enzyme Northern 1 Southern 1 Southern 2 Southern 2 Common Cytochrome B HaeIII 151 128 ÐÐÐ 271 143 151 Aldolase NlaIII 88 88 45 ÐÐ 92 92 66 111 111 88 92 NlaIV 17 17 17 ÐÐ 99 274 99 175 175 Rhodopsin HinfI 151 151 524 151 151 373 373 373 373 AluI 28 51 51 51 51 44 90 61 61 61 51 133 72 72 72 61 250 90 90 90 90 250 250 250 250 (B) Marker Primer 1 Primer 2 CytB (MVZ04, Ph-1) GCAGCCCCTCAGAATGATATTTGTCCT GACCCCAATACGAAAAACCACCC Aldolase (Ald1CR, Ald2CR) AAGAAGGATGGAGCTGACTTTGC GCCATTCTGTAACACAACAGCCAA Rhodopsin (Rho3CR, Rho4CR) CCTTGCCTGGACACCCTATGCTG CTCTGGAATAAAGGAGAGGGTCTCT A16 (rubA16a, rubA16b) GTTTACAATATAAGTTACCTTGAACAG CTTGACTGACTTGTCCCTTCG A11 (A11a, A11b) GTTTAGGACTGACAGAATTTGCCTT GACACCTATTTTGAGAACCACTGGA D4 (D4a, D4b) GTTTCTGCCATTTGTCTCATATGATTT CAGGATGGGGCATAAAGATTTTCC D13 (D13a, D13b) GTTTCCTCACGAGACACCGCATC CTGCTGGTTTTTCACTTTCCTTG D19 (D19a, D19b) GTTTCACAACAACCCTTTGAGATCAG AATGTCTGTTCTAGCGGCTTCTGCGG D20 (rubD20a, rubD20d) GGGTCTTTGGAAAGAACTGGAAG GTGGTAAGTAGTCCAGAGCC