Genetic variation and structure in native populationsof the fire ant Solenopsis invicta: evolutionary anddemographic implications
KENNETH G. ROSS1*, MICHAEL J. B. KRIEGER2, LAURENT KELLER3 andD. DEWAYNE SHOEMAKER4
1Department of Entomology, University of Georgia, Athens, GA 30602, USA2Swiss Federal Institute of Technology Zurich, 8092 Zurich, Switzerland3Department of Ecology and Evolution, University of Lausanne, 1015 Lausanne, Switzerland4CMAVE, USDA-ARS, Gainesville, FL 32608, USA
Received 12 May 2006; accepted for publication 10 January 2007
Crucial elements in unraveling a species’ evolutionaryhistory are the description of the distribution of popu-lation genetic variation and the inference of the his-torical demography and gene flow regimes from thisdescription (Avise, 2000, 2004). The task of convertingthe patterns etched in population genetic variation
into evolutionary and natural history narratives hasbecome increasingly relevant for organisms that arepests, vectors of human disease, endangered species,model organisms for research, or otherwise of specialconcern. This is because knowledge gleaned frompopulation genetic studies can aid in deciphering thepaths of adaptation and diversification of such organ-isms, thus providing a necessary context for under-standing the characteristics that make them special(Bohonak et al., 2001; Frankham, Briscoe & Ballou,*Corresponding author. E-mail: [email protected]
Biological Journal of the Linnean Society, 2007, 92, 541–560. With 8 figures
2002; Lee, 2002; Shimizu, 2002; Sivasundara & Hey,2003; Veuille et al., 2004). With the advent of multipletypes of molecular markers of nuclear and organellargenomes (Zhang & Hewitt, 2003; Avise, 2004), as wellas the appearance of advanced analytical techniquesto evaluate data generated from them (Pritchard,Stephens & Donnelly, 2000; Beerli & Felsenstein,2001; Hey & Machado, 2003; Beaumont, 2004), thereis now unprecedented power to extract useful infor-mation from population genetic data for synthesiswith other relevant results to enrich our evolution-arily knowledge of such organisms.
The fire ant Solenopsis invicta Buren is a highlysocial insect of considerable concern to biologists.The species occupies a vast native range in SouthAmerica, where it is an ecologically significant com-ponent of the ant fauna, and it recently has become aserious invasive pest in the USA and elsewhere(Lofgren, 1986; Callcott & Collins, 1996; Henshawet al., 2005). Largely because of the enormous amountof research devoted to it in newly-colonized areas,S. invicta has emerged as a useful model for an arrayof biological problems (Tschinkel, 2006). Nonetheless,the population genetics of native S. invicta and itsnearest relatives has received relatively little atten-tion. Despite this gap, it is clear that these antscomprise a group of closely-related species thatdisplay modest morphological and genetic differencesbetween them and exhibit, in some cases, substantialintraspecific population differentiation (Ross et al.,1997; Ahrens, Ross & Shoemaker, 2005; Ross & Shoe-maker, 2005; Shoemaker, Ahrens & Ross, 2006a;Pitts, McHugh & Ross, 2007). Compounding the com-plexity of relationships within the group is the appar-ent occurrence of interspecific hybridization in somecircumstances (Ross & Shoemaker, 2005). This com-plexity poses several important challenges that mustbe addressed to enhance the value of S. invicta as amodel species and promote research intended tounderpin management strategies for invasive popula-tions (Goodisman & Hahn, 2005; Ross & Shoemaker,2005). Foremost among these is the need to generatea detailed picture of the distribution of genetic varia-tion in nominal S. invicta based on appropriatemarkers and samples. With such data, the origin anddiversification of this species, the emergence of geneflow patterns within it, and the phylogenetic andbreeding relationships to its closest relatives canbegin to be resolved.
Earlier genetic studies of native S. invicta usednuclear and/or mitochondrial DNA (mtDNA) data,often derived from limited sampling, to generate pre-liminary views of the population genetics and phylo-geography of this ant (Ross & Trager, 1990; Rosset al., 1997; Ahrens et al., 2005; Ross & Shoemaker,2005), but no comprehensive effort to integrate data
for both genomes from samples collected widely overthe native range has yet been attempted. In thepresent study, we remedy this shortcoming by anal-ysing data for diverse classes of markers from appro-priate exemplar samples using a suite of traditionaland newer methods.
MATERIAL AND METHODSSAMPLES
A total of 568 nests were sampled from 13 populationsof nominal S. invicta located at ten localities in Braziland Argentina (Table 1, Fig. 1, inset). At two of theArgentine localities, Corrientes and Formosa, nests ofboth social forms of this species were abundant, so thesympatric forms are here considered as separatepopulations (Ross et al., 1997). The monogyne (M)social form is characterized by nests with a singleegg-laying queen, whereas the polygyne (P) form ischaracterized by nests with multiple such queens; thetwo forms differ also in a number of other importantreproductive and life-history traits (Ross & Keller,1995). Polygyne nests were rare or (usually) absentamong nests collected at the remaining sites (Mescheret al., 2003). At the Arroio dos Ratos site in Brazil,two apparently fully reproductively isolated entitiesreferred to as S. invicta coexist (Ross & Shoemaker,2005); these are also treated as separate populationsin this study (designated hereafter as the ‘Arroio X’and ‘Arroio Y’ populations). All specimens were iden-tified as nominal S. invicta by Dr James C. Trager orDr James P. Pitts (Trager, 1991; Pitts et al., 2007).
GENETIC MARKERS
A single specimen per nest was selected for geneticanalysis. Genotypes were scored at seven allozymeloci (Shoemaker, Costa & Ross, 1992; Ross et al.,1997) and seven microsatellite loci (Krieger & Keller,1997; Shoemaker et al., 2006b), yielding a total of 14polymorphic nuclear loci surveyed (Table 1). Thenumber of individuals per population for whichnuclear genetic data were obtained were in the range9–83 (mean = 43.7). Sequence data for a 920-bp frag-ment of the mtDNA that includes portions of thecytochrome oxidase subunit I and II genes (Ahrenset al., 2005) were obtained from a subset of the sameindividuals used to generate the nuclear data, withthe following exceptions: identical sets of individualsfrom the Arroio Y and Rosario populations were usedto generate both the nuclear and mtDNA data, andadditional individuals not available for nuclear analy-ses were sequenced for the mtDNA in the São Gabrieldo Oeste population. GenBank accession numbers formost of the mtDNA sequences are provided in Appen-dix II of Ahrens et al. (2005); accession numbers for
sequences of the Y haplotype clade are AY499580,AY499581, AY499589, and AY499590. All mtDNAsequences were aligned by eye using the alignmentof Ahrens et al. (2005) as a template. The numberof individuals per population for which mtDNAsequence data were obtained were in the range 9–66(mean = 35.1) (Table 1). A compilation of the geneticdata used in the present study is available uponrequest (from K.G.R.).
GENETIC DIVERSITY AND DISEQUILIBRIUM ANALYSES
Estimates of the extent of nuclear genetic diversity[allelic richness and expected heterozygosity (Hexp)]were obtained for each population using the programGENEPOP (Raymond & Rousset, 1995a); estimates ofallelic richness for the two populations with thelargest sample sizes (Ceu Azul, Rinco dos Cabrais)were corrected by taking the mean values from 100random resamplings of 35 individuals (Leberg, 2002).Estimates of mtDNA diversity [number of haplotypes,haplotype diversity (H), and nucleotide diversity (p)]were obtained using the program ARLEQUIN(Schneider, Roessli & Excoffier, 2000). The Arroio Ypopulation was excluded from comparisons of diver-sity among populations because of the small samplesize.
Departures from single-locus (Hardy–Weinberg)equilibrium at the population level were examined byestimating values of FIS for the combined nuclearmarkers in the hierarchical FST analyses describedbelow (conducted using the program GDA; Lewis &Zaykin, 2002). We also calculated values of FIS sepa-rately for each locus and population. Significant dis-equilibrium was inferred in all cases where the 95%confidence limits obtained by bootstrapping over loci(10 000 replicates) did not overlap zero.
Departures from linkage equilibrium between allpairs of nuclear loci in each population were examinedby conducting randomization tests to approximate theFisher’s exact test (Zaykin, Zhivotovsky & Weir, 1995)using GDA. Any departures from Hardy–Weinbergequilibrium are accounted for in this procedure.
GENETIC DIFFERENTIATION ANALYSES
To recognize genetically distinct clusters of individu-als and infer levels of population admixture, we
employed the Bayesian method of Pritchard et al.(2000) as implemented in the program STRUCTURE.The method makes use of individual multilocus geno-typic data to evaluate models assuming differentnumbers of clusters based on the posterior probabili-ties given the data, model, and prior information.Each sampled individual is probabilistically assignedto a reconstructed genetic cluster based on its multi-locus genotype and the allele frequencies estimatedfor each cluster. No prior information (e.g. samplelocation) was used in our initial runs of STRUC-TURE. The models employed assume some level ofpopulation admixture but allow allele frequencies tovary independently across populations. All othermodel parameter values were the defaults for theprogram. All simulations used 100 000 Markov chainMonte Carlo iterations in the burnin phase and300 000 iterations in the data collection phase, withfour independent runs conducted on each set of dataand parameter values to ensure equilibration by theend of burnin and consistency in estimation of theposterior probabilities. Selection of the number ofdistinct clusters was based on evaluation of the DKstatistic of Evanno, Regnaut & Goudet (2005). Theanalyses were conducted on all nuclear genes as wellas on the combined nuclear and mtDNA data. For thelatter analyses, mtDNA haplotypes were binned intoclasses corresponding to seven well-supported haplo-type lineages detected in a previous study of nativeS. invicta (Shoemaker et al., 2006a). Separate analy-ses of each geographical population also were con-ducted using the nuclear data to detect any lower-level structure (Evanno et al., 2005).
In a final set of STRUCTURE simulations, weincorporated the geographical localities and socialform of samples as priors to calculate values of FK foreach study population. This statistic can be inter-preted as an analogue of FST that measures the diver-gence of each extant population from a singlehypothetical ancestral population (Pritchard et al.,2000; Falush, Stephens & Pritchard, 2003).
The magnitude of genetic differentiation betweengeographical populations was evaluated by estimat-ing values of FST (Weir & Cockerham, 1984) for thetwo classes of nuclear markers separately and com-bined, rST (Rousset, 1996) for the microsatellites, andFST (Excoffier, Smouse & Quattro, 1992) for themtDNA using the programs GENEPOP and ARLE-
Figure 1. Assignment of Solenopsis invicta from each study population to genetic clusters inferred from STRUCTUREsimulations (based on average membership coefficient, Q). Results for nuclear markers only are shown in the left half ofthe figure (Brazilian and Argentine populations are in separate columns) and results for all markers combined are shownon the right. Bars representing clusters for which Q < 0.05 are coloured black. Inset: Locations of study populations. Thenative range of S. invicta is shown with grey shading. A major landform of presumed biogeographical importance, theMesopotamia floodplain, is indicated by stippling.�
QUIN. The latter two statistics take into accountpresumed mutational relationships of the variants,whereas the first does not. The statistical significanceof interpopulation differentiation was determined bymeans of exact tests (Raymond & Rousset, 1995b)using GENEPOP, with probabilities combined acrossnuclear loci using the Z-transform test (Whitlock,2005). The Arroio Y population was excluded fromthese comparisons because of the small sample size.We also estimated FST and FST simultaneously at twolevels (region, geographical population) using GDAand ARLEQUIN, respectively, to hierarchically parti-tion total genetic variation at the nuclear and mtDNAgenomes (Weir & Cockerham, 1984; Excoffier et al.,1992). Statistical significance of the differentiation ateach level was determined by bootstrapping over loci(10 000 replicates) for FST or permuting haplotypesacross individuals (20 000 replicates) for FST. The tworegions designated in these hierarchical analyses,Argentina and Brazil, were so conceived because aprevious mtDNA study reported a pronounced phylo-geographical break between ants from the two coun-tries (Ahrens et al., 2005).
Isolation-by-distance (IBD) analyses were con-ducted to learn whether genetic differentiation be-tween populations increases with their geographi-cal separation. Analyses of the relationships ofFST/(1 - FST) or FST/(1 - FST) with the natural loga-rithm of geographical distance were conducted for thenuclear markers and mtDNA, respectively, usingGENEPOP (Slatkin, 1993; Rousset, 1997). The twosocial forms at Corrientes and Formosa were pooledfor the nuclear but not the mtDNA analyses becauseof the similar nuclear compositions of sympatricforms (see below). Significance of IBD relationshipswas determined by means of Mantel tests based on10 000 data permutations coupled with estimation ofSpearman rank correlation coefficients. Residuals ofthe linear regression of FST on geographical distancewere plotted against geographical distance, and sig-nificance of this relationship was tested with aMantel test (Hutchison & Templeton, 1999).
We employed the ordination technique known asnonmetric multidimensional scaling (NMDS) to helpreveal higher-level patterns of genetic relationshipsamong populations. This technique reduces multidi-mensional allele or haplotype frequency relationshipsrepresented in a matrix of pairwise distances betweenpopulations to a few dimensions that explain mostof the original distance data (Lessa, 1990; Guiller,Bellido & Madec, 1998). We used the program VISTA(Young, 1996) to conduct NMDS analyses based onpairwise values of Nei’s genetic distance (D) for thenuclear loci and the net number of nucleotide differ-ences (DA) for the mtDNA obtained from the programsPHYLIP (Felsenstein, 2004) and ARLEQUIN, respec-
tively. We determined the best dimensionality foreach model by generating scree plots (Kruskal &Wish, 1978) to locate an elbow in the curve depictingthe total variance in the data explained with eachadded dimension; any dimensions beyond the elbowexplain relatively little additional variance and thuswere not retained. A stress statistic measuring thediscrepancy between the matrix of model distances inN-dimensional space and the original distance matrixwas calculated (Kruskal, 1964), and a method of itera-tive approximations was applied until values of thisstatistic declined to an asymptote, at which point themodel was accepted. We graphed projections of modeloutput in the first three dimensions, which in allcases jointly accounted for > 75% of the total variancein the original distance data, to distinguish clusters ofgenetically similar populations. Individual geographi-cal populations were subdivided for the nuclearanalyses if multiple clusters were inferred from theSTRUCTURE analyses.
GENE FLOW ANALYSES
Patterns of gene flow between sampled populationswere explicitly evaluated using the programsMIGRATE (Beerli & Felsenstein, 1999, 2001; Beerli,2004) and BAYESASS (Wilson & Rannala, 2003). TheBayesian coalescent approach in MIGRATE wasemployed to infer historical rates of gene flow at bothgenomes by analysing the allozyme and microsatellitedata, both separately and together, as well as byanalysing the mtDNA data separately. Model settingsfor the nuclear data were an infinite alleles model ofmutation, a uniform prior, random starting genealo-gies for each chain, and variable mutation ratesamong loci (the rates for the microsatellites were setat four times those of the allozymes). Results werecombined across three replicate chains, each of whichrecorded 5000 of 110 000 sampled genealogies forparameter estimation (the first 10 000 genealogies,representing the burnin, were discarded). Other set-tings were the defaults for the program. Model set-tings for the mtDNA sequences were an F84 mutationmodel, a transition/tranversion ratio of 6.7 (deter-mined empirically from previous data; Shoemakeret al., 2006a), a uniform prior, and random startinggenealogies for each chain. Results were combinedfrom ten replicate chains, each of which recorded10 000 of 210 000 sampled genealogies (again, thefirst 10 000 represented the burnin).
The Bayesian non-equilibrium approach in BAYE-SASS was used to infer recent nuclear gene flow ratesby analysing the allozyme and microsatellite dataseparately and together (the program requires diploiddata). The model was run for 12 000 000 iterations(the first 2 000 000 were discarded as burnin), with
5000 of these iterations sampled for parameter esti-mation. Delta values were set at 0.15 to achieveaccepted numbers of proposed changes to the Markovchain between 40% and 60% of the total number ofiterations.
RESULTSPOPULATION GENETIC DIVERSITY
Most measures of the extent of genetic diversitywithin the sample populations are significantly cor-related, both between the different nuclear markerclasses and between the nuclear markers and themtDNA (diversity values are provided in Table 1). Forexample, significant correlations were found betweenthe allozymes and microsatellites for each nucleardiversity measure (Spearman rank correlations;P = 0.041 for heterozygosity and P < 0.001 for allelicrichness). Also, mtDNA nucleotide diversity is sig-nificantly correlated with both overall nuclearheterozygosity and allelic richness (Spearman rankcorrelations; both P < 0.01; Fig. 2). Importantly,estimates of the latter three statistics tend to besignificantly greater for Argentine than Brazilianpopulations (Mann–Whitney tests; all P < 0.02;Fig. 2). More specifically, genetic diversity assessedfor both genomes by all measures tends to decreasewith geographical distance from Corrientes, Argen-tina (Fig. 3), the location with ants most closelyresembling a hypothetical ancestral S. invicta popu-lation (see below). Among the Brazilian populations,Rinco dos Cabrais invariably has the highest diversityacross all measures.
SINGLE-LOCUS AND LINKAGE DISEQUILIBRIUM
The lower bootstrap 95% confidence limit for theoverall FIS value based on all nuclear markers isgreater than zero, indicating a significant generaldeficiency of heterozygotes relative to Hardy–Weinberg expectations within the study populations.Examination of the individual-locus values revealedtwo microsatellite loci with atypically high FIS valuesin most populations (Sol-18 and Sol-20), suggestive ofscoring problems, null alleles, or effects of selection atthese loci. Their removal results in 95% confidencelimits for FIS that bracket zero. Exclusion of these locifrom the main analyses of the present study hadminimal effects on the reported results.
Bootstrap 95% confidence limits for FIS calculatedseparately for each population include zero for allpopulations except Rosario and Rinco dos Cabrais.When Sol-18 and Sol-20 were removed, the newlimits bracket zero for the former but not the latterpopulation. Thus, there appears to be a genuine defi-ciency of heterozygotes in Rinco.
Significant linkage disequilibria between pairs ofnuclear genes generally occurred at frequencies < 5%in each study population. Exceptions are the CampoGrande (16.4%) and Rinco (27.3%) populations. Fol-lowing removal of Sol-18 and Sol-20, the frequency ofsignificant linkage disequilibrium in the former popu-lation falls below 5%, whereas that in Rinco remainsrelatively high (21.8%). Thus, Rinco dos Cabraisstands out from the other populations in terms of themagnitude of its single-locus and linkage disequilib-rium, as well as its diversity at both genomes.
IDENTIFICATION OF GENETICALLY DISTINCTIVE
CLUSTERS
Application of the method of Evanno et al. (2005) tothe posterior probabilities obtained from the STRUC-TURE simulations showed a clear peak in DK valuesat 11 genetic clusters when just the nuclear markerswere considered, and at six clusters when the mtDNAdata were included as well. The average assignmentof individuals from each geographical population toeach of these clusters (membership coefficient, Q) isdepicted in Figure 1. Considering first the resultsfrom the nuclear data, most of the Brazilian popula-tions as well as the Rosario (Argentina) populationhave very high membership (> 80%) in a singleunique genetic cluster. That is, these geographicalpopulations tend to be genetically distinct and havelow levels of admixture with other populations.Exceptions are: (1) the neighbouring populations SãoGabriel do Oeste and Campo Grande, which sharedominant membership in the same cluster (with SãoGabriel also having an approximately 6% member-ship in the dominant cluster at Pedra Preta, the
Figure 2. Association of genetic diversity at the nuclearand mitochondrial DNA genomes in the study populations.Black symbols represent nuclear heterozygosity and whitesymbols represent nuclear allelic richness. Brazilian popu-lations are denoted with circles and Argentine populationswith squares.
population immediately north); (2) the Arroio Y popu-lation, which in addition to majority membership in aunique cluster displays substantial minority member-ship in several clusters dominant in neighbouringBrazilian and Argentine populations; and (3) theRinco population, which besides majority membershipin a unique cluster displays substantial admixture(approximately 13%) with the adjacent Arroio Y popu-lation. Also exceptional are the Formosa and Corri-entes populations from northern Argentina, whichappear to be heavily admixed. In Formosa, this is due
largely to the joint presence of two dominant uniqueclusters whereas, in Corrientes, there is one domi-nant unique cluster as well as substantial minorityrepresentation of clusters dominant elsewhere inArgentina and in the nearest Brazilian populations(Campo Grande, Ceu Azul, Arroio Y). STRUCTUREanalyses conducted separately on each geographicalpopulation revealed that only in Rinco and theFormosa P form do the posterior probabilities impli-cate the presence of multiple genetic clusters (two ineach case). Notably, the two clusters discerned within
Figure 3. Association of nuclear and mitochondrial DNA diversity with geographical distances of study populations fromCorrientes, Argentina. The Rinco dos Cabrais, Brazil population is indicated by the white square. Spearman rankcorrelation probabilities of no association between diversity and distance are shown (excluding the Rinco population inparentheses).
each of these populations are relatively similar intheir genetic composition (see NMDS analyses below).
Addition of the mtDNA data to the nuclear dataresults in consolidation of several of the clusters rec-ognized by STRUCTURE (Fig. 1). Most notably, all ofthe Argentine populations now have overwhelminglydominant membership in a single shared clusteralthough, in Corrientes, there is also substantialminority representation (approximately 10%) of acluster characteristic of the nearest Brazilian popu-lation, Ceu Azul. Also, the two northernmost Brazil-ian populations have dominant membership in aunique shared cluster. Finally, with the addition ofthe mtDNA data, the dominant cluster in the ArroioY population is now the same cluster that dominatesin Argentina, indicating an affinity of this Brazilianpopulation with the Argentine samples based on thetotal genetic evidence.
MAGNITUDE OF POPULATION GENETIC
DIFFERENTIATION
The magnitude of differentiation between geographi-cal populations is significantly correlated between theallozymes and microsatellites, with the highest cor-relation obtained when FST is used for both sets ofnuclear markers (Mantel test on Spearman rank cor-relation coefficient; P = 0.001; Fig. 4A). Thus, thesevery different sets of nuclear markers registercongruent patterns of population differentiation,although allozyme differentiation consistentlyexceeds microsatellite differentiation (compare axisscales in Fig. 4A). For the microsatellites, FST and rST
values also are highly correlated (P < 0.001). Impor-tantly, FST for all nuclear markers is significantlycorrelated with FST for the mtDNA (P = 0.006;Fig. 4B), indicating a general congruence in diver-gence between populations at the two genomes.Apparent outliers with respect to this intergenomiccongruence involve several populations that are rela-tively similar at the nuclear markers but highlydivergent at the mtDNA (Fig. 4B; see also below).
Results of the hierarchical analyses of FST and FST
are shown in Table 2. The results are congruentbetween the allozymes and microsatellites in suggest-ing no meaningful differentiation between regions(Argentina and Brazil) but substantial differentiationamong geographical populations within the regions.Again, the magnitude of among-population structureregistered by the allozymes exceeds that for the mic-rosatellites. A different overall pattern is observed forthe distribution of mtDNA variation, with very largeproportions occurring between regions as well asamong populations. This pattern is expected based onthe results of Ahrens et al. (2005), who describeda major mtDNA phylogeographical discontinuity
between the two regions that is superimposed onstrong differentiation among the component popula-tions. Even after accounting for the higher-levelstructure, the proportion of mtDNA variation residingamong populations far exceeds that for the nuclearmarkers, a pattern evident also in the single-levelanalysis (compare axis scales in Fig. 4B). Thus, geo-graphical populations of native S. invicta appear sub-stantially more differentiated at their mtDNA thantheir nuclear genomes.
Exact tests were conducted to pinpoint instancesof significant pairwise population differentiation.Considering all the nuclear data, all populations arehighly significantly differentiated, with only the twosocial forms in Formosa approaching nonsignificantdifferentiation (P = 0.012). Considering the mtDNAdata, with the exception of the two forms in Formosa(P = 0.287) all population pairs are highly signifi-cantly differentiated.
A
B
Figure 4. Association of pairwise population differentia-tion measures obtained for the allozymes and microsatel-lites (A) and for all nuclear markers and the mitochondrialDNA (B). Three apparent outliers in the bottom graph aredepicted with white symbols.
IBD analyses revealed that nuclear genetic differ-entiation increases significantly with geographicalseparation (Fig. 5A), whether all the markers areconsidered (Mantel test; P = 0.002), just the allozymesare considered (P = 0.01), or just the microsatellitesare considered (P = 0.004). No significant associationwas found between geographical distances and theresiduals from the linear regression of FST on geo-graphical distance (Mantel test; P = 0.464; Fig. 5B),suggesting that the observed IBD did not arise froma long-term balance between gene flow and geneticdrift (migration-drift equilibrium). Similar results forthe nuclear markers were obtained when only theeight Brazilian populations were analysed. On theother hand, only a marginally significant pattern ofIBD was found for the mtDNA for all populations(P = 0.052), and no such pattern was found for thisgenome for just the Brazilian populations (P = 0.166).
Judging from the pairwise FST values, levels ofnuclear differentiation among Brazilian popula-tions generally exceed those in Argentina (two-tailindependent-sample permutation test with 1000 per-mutations; P < 0.001), a finding upheld when theallozymes and microsatellites are considered sepa-rately (both P < 0.004). This pattern persists evenwhen the two northernmost or three southernmostBrazilian populations are excluded (both P < 0.001with all loci). A similar pattern of greater mtDNAdifferentiation in Brazil than Argentina based on FST
values also was found (P < 0.005), as reported previ-ously for a subset of our samples (Ahrens et al., 2005).Again, this result is robust to exclusion of the geo-graphically peripheral Brazilian populations.
HIGHER-LEVEL GENETIC RELATIONSHIPS
AMONG POPULATIONS
Relationships among populations were visualizedusing NMDS. The genetic similarity of populations at
the nuclear genome has a clear geographical compo-nent (Fig. 6). The northern populations of Pontes ELacerda and Pedra Preta form a distinct group, as dothe closely situated central-range populations of São
Table 2. Proportion of total genetic variance distributed at two levels of structure [regions (Brazil, Argentina) andpopulations] as assessed by hierarchical analyses of FST (nuclear markers) and FST (mitochondrial DNA)
Source of variance Proportion of variance
Nuclear markers Between regions 0.000 (0.000, 0.002)Among populations within regions 0.217* (0.373*, 0.172*)Within populations 0.783 (0.627, 0.826)
mtDNA Between regions 0.239*Among populations within regions 0.540*Within populations 0.221
Results for the nuclear data are shown for all markers combined, as well as for the allozymes and microsatellitesseparately (respectively, in parentheses).*Differentiation significant at P < 0.05 (nuclear markers) or P < 0.001 (mtDNA) at this level of structure.mtDNA, mitochondrial DNA.
A
B
Figure 5. Results of isolation-by-distance (IBD) analysesfor all nuclear markers. The relationship of pairwise popu-lation differentiation (FST) with geographical distance isdepicted in A, whereas the relationship of the residuals ofthe linear regression of FST on geographical distance withgeographical distance is depicted in B.
Gabriel and Campo Grande. The two genetic clustersin the Formosa P form inferred from the STRUC-TURE analyses group closely with one another andwith the Formosa M form, collectively constituting arelatively distinct assemblage. A large group compris-ing ants from the south-central part of the range isapparent (Ceu Azul, Rinco, Corrientes, Rosario), withthe Corrientes M and Rosario populations somewhatdistinct within it. Again, the two genetic clusters inRinco inferred by STRUCTURE are relatively similar.The two forms from Arroio dos Ratos on the south-eastern margin of the range are relatively distinctfrom each other and from the remaining samples,although the Y form appears to have closer affinitieswith ants from the neighbouring populations in CeuAzul, Rinco, and Corrientes than with sympatric con-specifics of the X form (see also Fig. 1). The latterform is quite divergent from all the other antsstudied.
A somewhat different picture of relationshipsemerges from the NMDS analyses of the mtDNA(Fig. 7). Four distinct groups of sequence variants areevident, and these correspond largely to major hap-lotype lineages identified by Shoemaker et al. (2006a).In parallel with results from the nuclear markers, ageographical component to the distribution of mtDNAvariation clearly exists, as shown by the restriction ofclade 2 to a group comprising the central Argentinapopulations, of clade 3 to all three Argentina popula-tions, of the Y clade/clade 4 lineage to the group ofcentral range populations, of clade 5 to the northernpopulations, and of clades 6 and 7 predominantly tosouthern Brazil samples. On the the other hand,these geographical patterns of mtDNA affinity do not
Figure 6. Projections of nonmetric multidimensionalscaling (NMDS) model output in the first three dimensionsfor the nuclear markers. The percentage of the totalvariance in the original distance data explained by eachdimension is shown in parentheses on the appropriateaxis. Coloring of drop lines and symbols depicts groups ofgenetically similar populations. Argentine samples aredistinguished by orange halos.
Figure 7. Projections of nonmetric multidimensionalscaling (NMDS) model output in the first three dimensionsfor the mitochondrial DNA. The percentage of the totalvariance in the original distance data explained by eachdimension is shown in parentheses. Coloring of drop linesdepicts groups of genetically similar populations, andcolouring of symbols indicates population haplotype com-positions (haplotypes present at frequencies < 5% notshown). Argentine samples are distinguished by orangehalos. Inset: Tree depicting relationships of major mtDNAhaplotype clades in native Solenopsis invicta (Shoemakeret al., 2006a). Clade colours correspond to colours ofsymbols used in NMDS model output.
always mirror closely the patterns of nuclear affinityrevealed by NMDS. For example, the Ceu Azul andArroio X populations are closely allied at the mtDNAbut very divergent at the nuclear genes, whereas SãoGabriel clusters with the more northerly populationsaccording to the mtDNA but with its southern neigh-bour Campo Grande based on the nuclear genes. TheSão Gabriel/Campo Grande comparison representsone of the three examples of neighbouring populationswith similar nuclear gene pools but dissimilarmtDNA gene pools shown as outliers in Figure 4B.The remaining two such outliers are the CorrientesM/Rinco and Corrientes P/Ceu Azul comparisons,which are of special interest because they suggestthat the Mesopotamia floodplain, a landform associ-ated with a phylogeographical discontinuity in themtDNA sequences (Ahrens et al., 2005), does not actas a significant barrier to contemporary nuclear geneflow.
HISTORICAL AND RECENT GENE FLOW
BETWEEN POPULATIONS
Bayesian analyses with the program MIGRATE wereused to infer historical levels of gene flow amongsampled populations. Rates based on the allozymesand microsatellites were found to be highly correlated(Mantel test on Spearman rank correlation coefficient;P < 0.0001). Overall nuclear gene flow rates are sig-nificantly lower among Brazilian than amongArgentine populations, even when several of thesouthernmost or northernmost Brazilian populationswere excluded from consideration (one-tail permuta-tion tests with 1000 permutations; all P < 0.0001).Trans-Mesopotamian nuclear gene flow rates betweenthe two countries are not significantly lower thanoverall rates within either country (both P > 0.775).This historical nuclear gene flow between the tworegions seems to have occurred predominantly in thedirection from Brazil to Argentina (two-tail permuta-tion test; P < 0.0001). Surprisingly, rates of mtDNAgene flow inferred by MIGRATE are no lower amongBrazilian than among Argentine populations (one-tailpermutation test; P = 0.471), but trans-MesopotamianmtDNA gene flow was found to be significantly lowerthan overall gene flow within either country (bothP < 0.01). Inspection of mtDNA rates for specific popu-lation pairs revealed that many of the highest trans-Mesopotamian rates involve the Formosa population,consistent with the finding of Ahrens et al. (2005) thatthis population at the northern edge of the land-form often was involved in long-distance dispersalacross it. Consideration of the direction of trans-Mesopotamian mtDNA gene flow revealed a patternopposite to that observed for nuclear gene flow, withrates from Argentina to Brazil significantly exceeding
those in the other direction (two-tail permutationtest; P < 0.0001).
Bayesian analyses with the program BAYESASSwere conducted to assess recent nuclear gene flowlevels. Again, rates based on the two classes ofnuclear markers are highly correlated (Mantel test onSpearman rank correlation coefficient; P < 0.0001)and, again, overall rates are lower among Brazilianthan among Argentine populations (one-tail permuta-tion test with 1000 permutations; P < 0.0001). As withthe estimates of historical nuclear gene flow, currenttrans-Mesopotamian gene flow tends not to be lowerin magnitude than gene flow within either Brazilor Argentina (one-tail permutation tests; bothP > 0.999). This ongoing interregional migrationappears asymmetrical, with Argentine populations(especially the northernmost Corrientes and Formosapopulations) more likely than southern Brazilianpopulations to receive high levels of trans-Mesopotamian immigration (two-tail permutationtest; P < 0.0001).
GENE FLOW INTO THE RINCO DOS CABRAIS
POPULATION
The Rinco population is unique in several respects. Ithas the greatest genetic diversity of any Brazilianpopulation, it is the only population that exhibitssubstantial single-locus and linkage disequilibrium,and it is one of only two populations for whichSTRUCTURE simulations indicate the presence oftwo distinct nuclear genetic clusters (see above). Thecomprehensive STRUCTURE analyses suggest thatthe minority cluster in Rinco represents individualswith nuclear ancestry in the neighbouring Arroio Ypopulation (Fig. 1), a conclusion supported by theBAYESASS analyses (data not shown). Inspection ofthe haplotype compositions reveals that the mtDNAlineage fixed in the Arroio Y form (Y clade) also occurscommonly in Rinco, and results of the MIGRATEanalyses implicate high mitochondrial gene flow fromArroio Y to Rinco as the cause of this pattern (datanot shown). These results taken together indicatethat the Rinco population is atypical because it expe-riences substantial immigration from the stronglydifferentiated, neighbouring Arroio Y populationwithout evident amalgamation of the immigrant andresident gene pools having occurred.
RELATIONSHIPS OF SAMPLED POPULATIONS TO
ANCESTRAL POPULATION
Values of FK were calculated from the STRUCTUREsimulations to infer the resemblance of each geo-graphical population to a hypothetical ancestral popu-lation of South American S. invicta. Results based on
just the nuclear markers suggest that the Corrientespopulations most closely resemble the ancestralpopulation, followed by the Formosa and Rosariopopulations (Fig. 8). Values of FK generally scalewith distance from Corrientes (Spearman correlation;P = 0.007 when sympatric social forms are pooled).Essentially identical results were obtained when themtDNA was included with the nuclear markers in theestimates of FK. These analyses thus suggest that apopulation resembling extant populations in north-central Argentina is ancestral to all the S. invictapopulations we sampled, and that the Brazilianportion of the range is likely to have been colonizedlater than the Argentine portion. This conclusion isconsistent with the mtDNA haplotype phylogeny pre-sented in Figure 7, in that a lineage that is sister toall remaining haplotype lineages (clade 2) is foundonly in Corrientes and Rosario, Argentina.
DISCUSSION
The objective of the present study was to use infor-mation from different classes of nuclear markers andthe mtDNA to generate a detailed picture of thenature and distribution of genetic variation over thenative range of the fire ant S. invicta. To that end,genotypic data from 14 nuclear loci and mtDNAsequence data were obtained from 568 nests sampledwidely over the South American range of this species,then analysed with a combination of traditional andnewer statistical methods. These analyses, in combi-nation with the results of more limited earlier studies(Ross & Trager, 1990; Ross et al., 1997; Ahrens et al.,2005; Ross & Shoemaker, 2005), allow us to infer
some key features of the historical and recent demog-raphy and dispersal biology of this important insectspecies, an essential task for reconstructing its evo-lution. Furthermore, the genetic information formsthe basis of such important applied goals as identify-ing the source populations from which invasive popu-lations around the world are derived (Van Driesche &Bellows, 1996; Tsutsui et al., 2001), a crucial task forfocusing control-orientated research in the nativerange. Our major findings are that: (1) the differentclasses of markers yield mutually informative andcomplementary information; (2) geographical popula-tions of native S. invicta are strongly genetically dif-ferentiated, especially in Brazil; (3) regional geneticaffinities among geographical populations exist; and(4) the native S. invicta we studied most likely arederived from an ancestral population resemblingextant ants from northern Argentina. These findingsare discussed in turn below.
COMPLEMENTARINESS OF MARKERS
The two classes of nuclear markers we employed,allozymes and microsatellites, yielded remarkablycongruent results with respect to estimates of theextent of diversity within and magnitude of differen-tiation between populations. Such congruence is notnecessarily expected in view of the very differentmutational mechanisms giving rise to detectablevariation, the substantially different levels of poly-morphism, and the different coding/noncoding statusof loci in each class (Ross et al., 1999; Buonaccorsi,McDowell & Graves, 2001; Mcelroy et al., 2003;Gaudeul et al., 2004). This congruence shows thatdemographic and dispersal events have left similarimprints on these distinctive components of thenuclear genome, so that these genome-wide forces,rather than gene-specific forces such as selection andmutation pressure, can be presumed to dominate theobserved structuring of nuclear variation in nativeS. invicta.
Although population genetic differentiation scalesin parallel at the two classes of nuclear markers,allozyme differentiation was found to consistentlyexceed that registered by the microsatellites whenassessed with FST estimates (Fig. 4A, Table 2; Rosset al., 1997). Geographically localized selection actingon the allozyme loci could play some role in thisdifference (Mitton, 1997; Neigel, 1997); however, amore likely explanation invokes a combination of twoother factors, the statistical property that FST esti-mates cannot exceed homozygosity levels (Hedrick,1999; Buonaccorsi et al., 2001; O’Reilly et al., 2004)and constraints on detectable microsatellite diver-gence with low migration due to the comparativelyhigh levels of electromorph homoplasy for these
Figure 8. Association of FK values estimated fromSTRUCTURE simulations conducted on the nuclear datawith geographical distances from Corrientes, Argentina.All Argentine populations are labelled.
markers (Balloux et al., 2000; Estoup, Jarne &Cornuet, 2002). To account for the differences in levelsof variation between the two marker types, we esti-mated values of a differentiation measure that isstandardized by the observed level of variation, G′ST
(Hedrick, 2005), for the entire set of populations.Similar estimates for the allozymes (0.492) and mic-rosatellites (0.546) suggest that differences in varia-tion explain at least some portion of the disparities inFST values between these two classes of markers.
Comparison of the results from the nuclearmarkers and mtDNA similarly shows some commonfeatures in the patterns observed for the twogenomes, while also revealing some substantive dif-ferences. Common features are the parallel variationin within-population diversity (with diversity decreas-ing with distance from Corrientes), the parallelvariation in population differentiation (with greaterdifferentiation between Brazilian than Argentinepopulations), and the co-occurring patterns of IBDregistered at the two genomes. Joint detection ofthese patterns at both genomes adds weight to theevolutionary and natural history inferences derivedfrom them.
Differences between the nuclear and mtDNA dataalso are evident and presumably can be useful forinferring demographic and gene flow patterns innative fire ant populations when interpreted in lightof the different properties of these genomes (Ballard& Whitlock, 2004). Distinctive distributions of geneticvariance were observed at two spatial scales, withanalyses of molecular variance (AMOVA) revealingconsiderably greater differentiation for the mtDNAthan the nuclear markers both among populationsand between regions (indeed, no evidence was foundfor nuclear differentiation between the Argentine andBrazilian regions). Also, distinctive patterns of popu-lation affinities were recovered for the two genomesusing NMDS analyses; whereas populations tend togroup with their geographical neighbours using eitherdata set, the makeup of such groups differs dependingon the genome considered. This discrepancy inhigher-level groupings is reflected in the strongnuclear similarity, but equally pronounced mitochon-drial dissimilarity, characterizing some neighbouringpopulations. A good example involves São Gabriel doOeste and Campo Grande, populations located lessthan 200 km apart in the east-central part of therange. Whereas their overall nuclear similaritycauses them to cluster together in the NMDS analy-sis, their strong mtDNA differentiation leads each togroup with a different set of populations when suchanalysis is applied to the mtDNA sequences.
These disparities in structure registered by the twogenomes are likely to result predominantly from theirdifferent effective population sizes and transmission
dynamics. Stronger structure may be expected for themtDNA at all scales of analysis given the lower effec-tive population sizes and correspondingly greatereffect of drift for this genome when populations arerelatively isolated (Avise, 2004). Moreover, greatervagility of males than queens can contribute to thisdisparity. Such sex-biased dispersal has been proposedpreviously for invasive S. invicta in the USA (Shoe-maker et al., 2006b) and has been inferred from popu-lation genetic data in other ants as well (Gyllenstrand& Seppa, 2003; Sanetra & Crozier, 2003; Sundström,Keller & Chapuisat, 2003; Clemencet, Viginier &Doums, 2005). Finally, the occurrence of the endosym-biotic bacterium Wolbachia in native S. invicta mayhave altered the distribution of mtDNA variationwithout affecting the nuclear DNA, because cytoplas-mic genomes are predicted to be influenced by indirectselection associated with this maternally transmittedmicrobe (Hurst & Jiggins, 2005). Although previousstudies found little evidence for such an effect onnative fire ant mtDNA (Ahrens et al., 2005), we cannotrule out the possibility that Wolbachia infection hasplayed some role in driving population mtDNA diver-gence in S. invicta (Ahrens & Shoemaker, 2005).
EXTENT OF POPULATION GENETIC DIFFERENTIATION
A major finding of this study is that geographicalpopulations of native S. invicta are highly geneticallydistinct. The evidence comes in several forms. First,STRUCTURE analyses of the nuclear genes showthat most populations represent genetically uniqueclusters with very low levels of admixture from otherclusters. Additionally, AMOVA analyses reveal that alarge proportion of the total genetic variance at boththe nuclear and mtDNA genomes resides amongpopulations (22% and 54%, respectively). Finally,BAYESASS analyses suggest that populations typi-cally are composed of at least 95% non-immigrantnuclear genes (data not shown). Among the few excep-tions, half involve migration between the sympatricsocial forms in Corrientes and Formosa or betweenthe geographically adjacent populations of CampoGrande and São Gabriel do Oeste (these populationpairs also do not comprise distinct clusters in theSTRUCTURE analyses). The remaining exceptions inthe BAYESASS analyses are the two forms in Corri-entes, which appear to experience significant nucleargene influx from neighbouring populations in Argen-tina and across the Mesopotamia floodplain in Brazil(this admixture is evident as well in the STRUC-TURE results).
A corollary of the marked population genetic struc-ture in native S. invicta is that it tends to be morepronounced in Brazil than Argentina, as indicated bysignificantly elevated values of FST for the nuclear
markers and FST for the mtDNA in Brazil. Thisregional effect on differentiation is apparent as wellfrom the STRUCTURE analysis using all the geneticdata, which depicts the Argentine samples as consti-tuting a single cluster whereas the five remaininginferred clusters largely comprise single populationsor pairs of adjacent populations in Brazil. Thisregional effect is mirrored by lower estimates of bothhistorical and contemporary nuclear gene flow amongBrazilian than among Argentine populations, asinferred from MIGRATE and BAYESASS, respec-tively. (Surprisingly, MIGRATE analyses of themtDNA provide no evidence for relatively lower his-torical gene flow in Brazil.) The present study thusextends to the nuclear genome the conclusion ofAhrens et al. (2005) that Brazilian populations gen-erally are more strongly differentiated than those inArgentina. These authors speculated that differencesin the distribution of suitable habitat (open, disturbedareas) may be involved, with the more patchy distri-bution of such habitat in southern Brazil than north-ern Argentina creating greater barriers to gene flow.Expansion of the range of S. invicta into Brazil fromArgentina could also be involved, if long distancecolonizations coupled with ensuing reductions ineffective population sizes created enhanced opportu-nities for stochastic divergence in population genefrequencies (see also below).
The remarkable degree of divergence measured forsome pairs of Brazilian populations raises the ques-tion of whether any are sufficiently genetically iso-lated to have embarked on independent evolutionarypaths. Six pairs of Brazilian populations jointlyexhibit values of FST > 0.35 and FST > 0.925, whichcorrespond to biparental and maternal evolutionarilyeffective gene flow levels of less than 0.47 and 0.04,respectively [using FST = 1/(4 Nem + 1) for the nuclearmarkers and FST = 1/(2 Nem + 1) for the mtDNA;Slatkin, 1987; Neigel, 1997; Whitlock & McCauley,1999]. Although the meaning of such gene flow esti-mates can be controversial (Whitlock & McCauley,1999), these maximal values are well below thethreshold values of 1.0 and 0.5 that yield an equilib-rium between gene flow and drift for neutral nuclearand mtDNA markers, respectively, in a simple islandmodel (Slatkin, 1987). Direct estimates of levels ofadmixture from STRUCTURE confirm that geneexchange between these populations is rare, with theproportion of foreign nuclear genes assignable toeach paired population estimated at only 0.4–1.3%(mean = 0.8%) (analyses for each population pair wereconducted as described in the Material and Methodsfor the general analyses, with no prior informationand K = 2 clusters). Thus, at least these severalpaired populations are likely to be genetically andevolutionarily independent of one another. We note
that the average values of FST and FST for these sixpopulation pairs, 0.67 for the allozymes, 0.26 for themicrosatellites, and 0.95 for the mtDNA, are in the93rd, 84th, and 97th percentiles, respectively, of suchestimates obtained in large-scale surveys of popula-tion differentiation in numerous nominal animalspecies (Morjan & Rieseberg, 2004).
The Arroio dos Ratos X form of S. invicta appears tobe completely reproductively isolated from the sym-patric Y form, judging from the observations that thetwo share no alleles at the allozyme gene Est-2, thatthey display dramatic allele frequency differences atseveral other nuclear loci, and that they possess non-overlapping sets of haplotypes belonging to differentmajor clades (Fig. 7; Ross & Shoemaker, 2005). Evi-dently, the X form also has participated in littlerecent gene exchange with the other populations (it isa member of three of the six highly divergent pairsconsidered above; see also Fig. 6). The clade of SouthAmerican fire ants including S. invicta and its closestrelatives is thought to be in a phase of active radiationof species, based on the generally slight nucleargenetic differentiation between species, frequent para-phyly of their mtDNA, and subtle or inconsistentmorphological differentiation between them (Ross &Trager, 1990; Ross & Shoemaker, 2005; Shoemakeret al., 2006a; Pitts et al., 2007). In this view, somenominal species with sizeable ranges, such as S. in-victa, may be expected to comprise entities that occupyvarious points on the continuum from freely inter-breeding, genetically indistinguishable populations tofully reproductively isolated, genetically divergedpopulations. The Arroio dos Ratos X form appears tofall at the latter end of this spectrum and, as such,probably should be considered a cryptic species.
HIGHER-LEVEL GENETIC STRUCTURE
Higher-level genetic affinities of sampled S. invictaare evident at scales of hundreds to thousands ofkilometers across the native range. One importantexample is the IBD patterns observed for bothgenomes (marginally significant for the mtDNA).Such patterns show that neighbouring populationsare genetically more similar to one another than aremore distant populations, presumably owing to morerecent shared ancestry and/or higher contemporarygene flow (Malécot, 1991; Slatkin, 1993; Hardy &Vekemans, 1999; Yang, 2004). The observed strongnuclear IBD pattern apparently is not the result of along-term equilibrium between genetic drift withinpopulations and gene flow among them because thereis no general trend for increased variation in FST
values with distance between populations (Hutchison& Templeton, 1999). Non-equilibrium processes suchas geographical range expansion from a single point
via serial founding of peripheral populations also canproduce IBD patterns (Ramachandran et al., 2005).The evidence that we found for expansion of S. invictafrom an original source population in northern Argen-tina suggests that such a process may have contrib-uted to the emergence of the observed IBD patterns.
Higher-level genetic structure is further evidentfrom the fact that NMDS analyses of both the nuclearand mtDNA data recognize genetically distinctivegroups of adjacent populations, even though the com-position of the groups so recognized does not corre-spond closely between the two genomes. For themtDNA, the tree of haplotype relationships providesadditional evidence for a higher-level geographicalcomponent to the distribution of mtDNA variation. Asexamples, one haplotype clade arising from the basalsplit in the mtDNA tree (clade 2) is confined to twoadjacent Argentine populations, whereas a large cladeof more recently derived lineages (clades 5–7) is con-fined almost exclusively to Brazil.
These particular examples of geographical restric-tion of mtDNA lineages hint at the presence of adiscontinuity in the distribution of sequence variationbetween Argentina and Brazil. Indeed, a ‘phylogeo-graphical break’ coincident with a landform on theArgentina/Brazil border, the Mesopotamia floodplain,was formally demonstrated by Ahrens et al. (2005) bymeans of nested clade phylogeographical analysis.The situation is made more complex with our additionof new sequences, because the predominantly Brazil-ian haplotypes no longer constitute a monophyleticgroup (Fig. 7); thus, the existence of such a breakcannot be upheld if this term implies the geographicalsegregation of a single major haplotype clade (forvarying uses of this and related terms, Comes &Abbott, 2000; Riginos & Nachman, 2001; Manel et al.,2003; Avise, 2004: 288). Nonetheless, the results ofthe AMOVA analyses clearly demonstrate mtDNAdifferentiation between populations from the tworegions that transcends differentiation within eachregion, and estimates of mtDNA gene flow rates fromMIGRATE implicate lower rates of historical geneflow across Mesopotamia than within either region.Thus, this floodplain appears to have acted as animportant historical barrier to mitochondrial geneflow in fire ants. The Río Paraná, which forms thewestern boundary of the floodplain, has been hypoth-esized to be an important biogeographical barrier inseveral plant groups as well (dos Santos, 1995;Romaniuc-Neto, 1998).
In contrast to the mtDNA, there is no evidence thatnuclear gene flow across Mesopotamia is substan-tially impeded. It is unlikely that homoplasy of themicrosatellites masks some signal of ancient regionaldivergence because the allozymes similarly reveal nodifferentiation at this level in the AMOVA analyses
(Table 2). Rather, demographic differences betweenthe sexes may explain the differential effect of thisbarrier on gene flow at the two genomes. Any coloni-zation of new areas across a geographical barriernecessarily involves females (mated queens), yet sub-sequent longer-distance gene flow between estab-lished populations, including migration across such abarrier, is likely to occur predominantly by males ifthey possess superior dispersal abilities (Shoemakeret al., 2006b). A scenario in which trans-Mesopotamiarange expansion occurred via queens from Argentinepopulations dispersing intermittently across thefloodplain, but in which males subsequently disperseback at much higher rates than queens, is consistentwith the different directionalities in gene flow acrossMesopotamia that we observed for the nuclear andmtDNA markers.
OUT OF ARGENTINA
Multiple lines of evidence from both genomes suggestthat the ancestors of the native S. invicta we studiedmost closely resembled extant northern Argentinaants, and that other sampled populations were estab-lished more recently by long-distance colonizationand/or range expansion starting from this source.First, estimates of FK values from STRUCTUREsuggest that the northern Argentine populations,especially Corrientes, most closely resemble such ahypothetical ancestral population and that thegenetic similarity of the study populations to thisancestor falls off with distance from Corrientes. Also,there is greater among-population differentiation andlower within-population diversity in Brazil thanArgentina, as expected if relatively recent foundingof, and subsequent lack of gene flow among, popula-tions east and north of Mesopotamia reduced theirlong-term effective sizes (McCauley, Raveill &Antonovics, 1995; Pannell & Charlesworth, 2000).Our added observation that diversity at both genomesconsistently decreases with distance from Corrientesis an important signal of such range extensionoutward from Corrientes by means of sequentialfounder events (Ramachandran et al., 2005), a sce-nario that could also explain the observed IBD pat-terns in the absence of migration–drift equilibrium.Finally, our MIGRATE analyses of the mtDNA impli-cate historical queen-mediated gene flow across Meso-potamia primarily in the direction from Argentina toBrazil. Ahrens et al. (2005) suggested earlier on thebasis of nested clade analysis of the mtDNA thatnorthern Argentine S. invicta populations are charac-terized by long-term persistence and uninterruptedgene flow, whereas Brazilian populations are theproducts of more recent range extension across thefloodplain.
This Brazilian population is notable for its relativelyhigh genetic diversity, substantial single-locus andlinkage disequilibrium, and presence of multiplegenetic clusters inferred by STRUCTURE, all of whichappear to be attributable to immigration from astrongly differentiated neighbouring population (theArroio Y form). This immigration clearly has not led tofull introgression, suggesting that it is ongoing or,perhaps more likely, that there is strong positiveassortative mating or selection against recombinantgenotypes in Rinco. Thus, the situation here mayresemble that in the neighbouring Arroio population,where the X and Y forms were recognized and sepa-rated for analysis a priori based on the apparent lackof introgression between them (Ross & Shoemaker,2005). Although the northern Argentine populations ofCorrientes and Formosa were shown to be more highlyadmixed than Rinco based on the STRUCTURE andBAYESASS analyses, the absence of disequilibrium inthese populations hints that the admixture is rela-tively old and that introgression is comprehensive.This is consistent with the view that ‘admixture’ inthese Argentine populations in fact largely representsretention of ancestral polymorphisms, as expected ifthey represent relatively ancient populations withhistorically large effective sizes.
GENE FLOW BETWEEN CO-OCCURRING SOCIAL FORMS
The co-occurring social forms that we studied tend tobe little differentiated from one another in compari-son to geographically separated populations (Figs 6,7; Ross et al., 1997). Indeed, exact tests for Formosarevealed only modestly significant nuclear differentia-tion (FST = 0.012) and insignificant mtDNA differen-tiation (FST = 0.034) between the forms. The socialforms in Corrientes display somewhat greater, highlysignificant, divergence at both genomes (FST = 0.020,FST = 0.201), and the BAYESASS analysis suggestsvirtually no nuclear gene exchange between the formsthere (data not shown). Differentiation is expected todevelop between sympatric social forms over timebecause most routes of gene exchange seem to beprecluded due to social incompatibilities or otherfactors (Ross et al., 1997, 1999). Shallower between-form divergence in Formosa than Corrientes maysignal a more limited period of co-occurrence of theforms at the former site, as suggested by the negativeassociation between such divergence and age of popu-lations seen in S. invicta in the USA (Shoemakeret al., 2006b).
CONCLUSION
Solenopsis invicta displays pronounced regional popu-lation differentiation at the nuclear and mtDNA
genomes in its native range. This differentiation issufficiently developed between some Brazilian popu-lations that they are likely to be completely geneti-cally and evolutionarily independent; indeed, theoccurrence in sympatry of one pair of clearly repro-ductively isolated entities demonstrates the potentialfor the processes driving fire ant population differ-entiation to culminate in speciation. Pronouncedregional genetic differentiation, including the pres-ence of cryptic species, evidently is common also inother groups of ants and can be caused by limiteddispersal capabilities of sexuals, patchiness of suit-able habitats, or even social barriers to gene flow(Liautard & Keller, 2001; Van der Hammen, Pedersen& Boomsma, 2002; Sanetra & Crozier, 2003; Clemen-cet et al., 2005; Goodisman & Hahn, 2005). A majorchallenge in such groups is to decide which entitieswarrant recognition as species taxa and to appreciatethe practical and heuristic consequences of such deci-sions (Porter, 1990; Hey et al., 2003; Sites & Marshall,2004). Resolution of these issues in nominal S. invictawill help us gain a clearer picture of the evolution ofthe species group of South American fire ants towhich it belongs and thus facilitate research on thisimportant group of ants. The presence of geneticallyunique geographical populations of native S. invicta,whether ultimately regarded as conspecifics or not,has great practical relevance in that tracing thesource of recently established invasive populationsaround the globe should be feasible.
Our analyses point to the importance of serial long-distance colonizations and range expansions from anorthern Argentina source as major features of thehistorical phylogeography of S. invicta responsible forthe observed patterns of differentiation. The rela-tively deep sequence divergence reported between themajor mtDNA clades (up to 5.1%; Shoemaker et al.,2006a) coupled with the geographical partitioning ofmany of these clades suggests a substantial period ofoccupation of South America by S. invicta in more-or-less isolated regional populations (Pinceel, Jordaens& Backeljau, 2005). Drift and/or local selection appar-ently has been sufficiently strong that, in some cases,the resulting diverged entities have come into second-ary contact with minimal or no restoration of geneflow, thus illustrating the concept that reproductiveisolation can develop in association with rangeextension (Tregenza, Pritchard & Butlin, 2000). Thedynamic nature of gene flow patterns and populationstructure in these ants is reflected in evidence forboth an initial expansion into Brazil from Argentinaas well as recent nuclear gene flow primarily in thereverse direction. Future work will focus on generat-ing data for very large numbers of nuclear markers ofvarious classes from an expanded set of populationschosen to completely cover the native range of
nominal S. invicta. Such data may provide a wellresolved timeline for the major demographic eventsthat have shaped the evolution of these ants.
ACKNOWLEDGEMENTS
We thank Dietrich Gotzek for comments on an earlierdraft of this manuscript. This research was supportedby grants from the United States Department ofAgriculture NRICGP (2003-35302-13497 and 2006-35302-16561) and the Swiss NSF. The use of trade,firm, or corporation names in this publication is forthe information and convenience of the reader. Suchuse does not constitute an official endorsement orapproval by the United States Department of Agri-culture or the Agricultural Research Service of anyproduct or service to the exclusion of others that maybe suitable.
REFERENCES
Ahrens M, Ross KG, Shoemaker DD. 2005. Phylogeo-graphic structure of the fire ant Solenopsis invicta in itsnative South American range: roles of natural barriers andhabitat connectivity. Evolution 59: 1733–1743.
Ahrens M, Shoemaker DD. 2005. Evolutionary history ofWolbachia infections in the fire ant Solenopsis invicta. BMCEvolutionary Biology 5: 35.
Avise JC. 2000. Phylogeographyraphy: the history and forma-tion of species. Cambridge, MA: Harvard University Press.
Ballard JWO, Whitlock MC. 2004. The incomplete naturalhistory of mitochondria. Molecular Ecology 13: 729–744.
Balloux F, Brunner H, Lugon-Moulin N, Hausser J,Goudet J. 2000. Microsatellites can be misleading: anempirical and simulation study. Evolution 54: 1414–1422.
Beaumont MA. 2004. Recent developments in genetic dataanalysis: what can they tell us about human demographichistory? Heredity 92: 365–379.
Beerli P. 2004. MIGRATE, version 1.6. Seattle, WA: Depart-ment of Genome Sciences, University of Washington.
Beerli P, Felsenstein J. 1999. Maximum-likelihood estima-tion of migration rates and effective population numbers intwo populations using a coalescent approach. Genetics 152:763–773.
Beerli P, Felsenstein J. 2001. Maximum likelihood estima-tion of a migration matrix and effective population sizes inn subpopulations by using a coalescent approach. Proceed-ings of the National Academy of Sciences of the UnitedStates of America 98: 4563–4568.
Bohonak AJ, Davies N, Villablanca FX, Roderick GK.2001. Invasion genetics of new world medflies: testing alter-native colonization scenarios. Biological Invasions 3: 103–111.
Buonaccorsi VP, McDowell JR, Graves JE. 2001. Recon-ciling patterns of inter-ocean molecular variance from four
classes of molecular markers in blue marlin (Makaira nig-ricans). Molecular Ecology 10: 1179–1196.
Callcott AMA, Collins HL. 1996. Invasion and range expan-sion of imported fire ants (Hymenoptera: Formicidae) inNorth America from 1918–1995. Florida Entomologist 79:240–251.
Clemencet J, Viginier B, Doums C. 2005. Hierarchicalanalysis of population genetic structure in the monogynousant Cataglyphis cursor using microsatellite and mitochon-drial DNA markers. Molecular Ecology 14: 3735–3744.
Comes HP, Abbott RJ. 2000. Random amplified polymor-phic DNA (RAPD) and quantitative trait analyses acrossa major phylogeographical break in the Mediterraneanragwort Senecio gallicus Vill. (Asteraceae). MolecularEcology 9: 61–76.
Estoup A, Jarne P, Cornuet J-M. 2002. Homoplasy andmutation model at microsatellite loci and their conse-quences for population genetics analysis. Molecular Ecology11: 1591–1604.
Evanno G, Regnaut S, Goudet J. 2005. Detecting thenumber of clusters of individuals using the softwareSTRUCTURE: a simulation study. Molecular Ecology 14:2611–2620.
Excoffier L, Smouse PE, Quattro JM. 1992. Analysis ofmolecular variance inferred from metric distances amongDNA haplotypes: application to human mitochondrial DNArestriction data. Genetics 131: 479–491.
Falush D, Stephens M, Pritchard JK. 2003. Inference ofpopulation structure using multilocus genotype data: linkedloci and correlated allele frequencies. Genetics 164: 1567–1587.
Felsenstein J. 2004. PHYLIP (phylogeny inference package).Seattle, WA: Department of Genome Sciences, University ofWashington.
Gaudeul M, Till-Bottraud I, Barjon F, Manel S. 2004.Genetic diversity and differentiation in Eryngium alpinumL. (Apiaceae): comparison of AFLP and microsatellitemarkers. Heredity 92: 508–518.
Goodisman MAD, Hahn DA. 2005. Breeding system, colonystructure, and genetic differentiation in the Camponotusfestinatus species complex of carpenter ants. Evolution 59:2185–2199.
Guiller A, Bellido A, Madec L. 1998. Genetic distances andordination: the land snail Helix aspersa in North Africa asa test case. Systematic Biology 47: 208–227.
Gyllenstrand N, Seppa P. 2003. Conservation genetics ofthe wood ant, Formica lugubris, in a fragmented landscape.Molecular Ecology 12: 2931–2940.
Hardy OJ, Vekemans X. 1999. Isolation by distance in acontinuous population: reconciliation between spatial auto-correlation analysis and population genetics models.Heredity 83: 145–154.
Hedrick PW. 1999. Perspective: highly variable loci and theirinterpretation in evolution and conservation. Evolution 53:313–318.
Hedrick PW. 2005. A standardized genetic differentiationmeasure. Evolution 59: 1633–1638.
Henshaw MT, Kunzmann N, Vanderwoude C, SanetraM, Crozier RH. 2005. Population genetics and historyof the introduced fire ant, Solenopsis invicta Buren(Hymenoptera: Formicidae), in Australia. AustralianJournal of Entomology 44: 37–44.
Hey J, Machado CA. 2003. The study of structuredpopulations - new hope for a difficult and divided science.Nature Reviews Genetics 4: 535–543.
Hey J, Waples RS, Arnold ML, Butlin RK, Harrison RG.2003. Understanding and confronting species uncertainty inbiology and conservation. Trends in Ecology and Evolution18: 597–603.
Hurst GDD, Jiggins FM. 2005. Problems with mitochondrialDNA as a marker in population, phylogeographic and phy-logenetic studies: the effects of inherited symbionts. Pro-ceedings of the Royal Society of London Series B, BiologicalSciences 272: 1525–1534.
Hutchison DW, Templeton AR. 1999. Correlation of pair-wise genetic and geographic distance measures: inferringthe relative influences of gene flow and drift on the distri-bution of genetic variability. Evolution 53: 1898–1914.
Krieger MJB, Keller L. 1997. Polymorphism at dinucleotidemicrosatellite loci in fire ant (Solenopsis invicta) popula-tions. Molecular Ecology 6: 997–999.
Kruskal JB. 1964. Multidimensional scaling by optimizinggoodness of fit to a nonmetric hypothesis. Psychometrika 29:1–17.
Kruskal JB, Wish M. 1978. Multidimensional scaling.Beverly Hills, CA: Sage Publications.
Leberg PL. 2002. Estimating allelic richness: effects ofsample size and bottlenecks. Molecular Ecology 11: 2445–2449.
Lee CE. 2002. Evolutionary genetics of invasive species.Trends in Ecology and Evolution 17: 386–391.
Lessa EP. 1990. Multidimensional-analysis of geographicgenetic-structure. Systematic Zoology 39: 242–252.
Lewis PO, Zaykin D. 2002. GDA (Genetic Data Analysis).Available at http://lewis.eeb.uconn.edu/lewishome/
Liautard C, Keller L. 2001. Restricted effective queen dis-persal at a microgeographic scale in polygynous populationsof the ant Formica exsecta. Evolution 55: 2484–2492.
Lofgren CS. 1986. History of imported fire ants in the UnitedStates. In: Lofgren CS, Vander Meer RK, eds. Fire ants andleaf cutting ants: biology and management. Boulder, CO:Westview Press, 36–49.
Malécot G. 1991. The mathematics of heredity. San Fran-cisco, CA: Freeman.
Manel S, Schwartz MK, Luikart G, Taberlet P. 2003.Landscape genetics: combining landscape ecology and popu-lation genetics. Trends in Ecology and Evolution 18: 189–197.
McCauley DE, Raveill J, Antonovics J. 1995. Local found-ing events as determinants of genetic structure in a plantmetapopulation. Heredity 75: 630–636.
tion of spotted sunfish (Lepomis punctatus) in the FloridaEverglades. Molecular Ecology 12: 355–368.
Mescher MC, Ross KG, Shoemaker DD, Keller L,Krieger MJB. 2003. Distribution of the two social forms ofthe fire ant Solenopsis invicta (Hymenoptera: Formicidae)in the native South American range. Annals of the Entomo-logical Society of America 96: 810–817.
Mitton JB. 1997. Selection in natural populations. New York:Oxford University Press.
Morjan CL, Rieseberg LH. 2004. How species evolve collec-tively: implications of gene flow and selection for the spreadof advantageous alleles. Molecular Ecology 13: 1341–1356.
Neigel JE. 1997. A comparison of alternative strategies forestimating gene flow from genetic markers. Annual Reviewof Ecology and Systematics 28: 105–128.
O’Reilly PT, Canino MF, Bailey KM, Bentzen P. 2004.Inverse relationship between FST and microsatellite poly-morphism in the marine fish, walleye pollock (Theragrachalcogramma): implications for resolving weak populationstructure. Molecular Ecology 13: 1799–1814.
Pannell JR, Charlesworth B. 2000. Effects of metapopula-tion processes on measures of genetic diversity. Philosophi-cal Transactions of the Royal Society of London Series B,Biological Sciences 355: 1851–1864.
Pinceel J, Jordaens K, Backeljau T. 2005. ExtrememtDNA divergences in a terrestrial slug (Gastropoda, Pul-monata, Arionidae): accelerated evolution, allopatric diver-gence and secondary contact. Journal of EvolutionaryBiology 18: 1264–1280.
Pitts JP, McHugh JV, Ross KG. 2007. Revision of thefire ants of the. Solenopsis saevissima species-group(Hymenoptera: Formicidae). Zootaxa in press.
Porter AH. 1990. Testing nominal species boundaries usinggene flow statistics: the taxonomy of two hybridizingadmiral butterflies (Limenitis, Nymphalidae). SystematicZoology 39: 131–147.
Pritchard JK, Stephens M, Donnelly P. 2000. Inferenceof population structure using multilocus genotype data.Genetics 155: 945–959.
Ramachandran S, Deshpande O, Roseman CC, Rosen-berg NA, Feldman MW, Cavalli-Sforza LL. 2005.Support from the relationship of genetic and geographicdistance in human populations for a serial founder effectoriginating in Africa. Proceedings of the National Academyof Sciences of the United States of America 102: 15942–15947.
Raymond M, Rousset F. 1995a. GENEPOP (version 1.2):population genetics software for exact tests and ecumeni-cism. Journal of Heredity 86: 248–249.
Raymond M, Rousset F. 1995b. An exact test for populationdifferentiation. Evolution 49: 1280–1283.
Riginos C, Nachman MW. 2001. Population subdivision inmarine environments: the contributions of biogeography,geographical distance and discontinuous habitat to geneticdifferentiation in a blennioid fish, Axoclinus nigricaudus.Molecular Ecology 10: 1439–1453.
Romaniuc-Neto S. 1998. Biodiversity and speciation in thesouth of Brazil and the basin of the Paraná river: influ-
ences in the Sorocea A. St-Hil. (Moraceae) genus speciescomplex. Comptes Rendus de L’Academie des SciencesSerie II Fascicule A - Sciences de la Terre et des Planetes327: 669–675.
Ross KG, Keller L. 1995. Ecology and evolution of socialorganization: insights from fire ants and other highly euso-cial insects. Annual Review of Ecology and Systematics 26:631–656.
Ross KG, Krieger MJB, Shoemaker DD, Vargo EL,Keller L. 1997. Hierarchical analysis of genetic structure innative fire ant populations: results from three classes ofmolecular markers. Genetics 147: 643–655.
Ross KG, Shoemaker DD. 2005. Species delimitation innative South American fire ants. Molecular Ecology 14:3419–3438.
Ross KG, Shoemaker DD, Krieger MJB, DeHeer CJ,Keller L. 1999. Assessing genetic structure with multipleclasses of molecular markers: a case study involving theintroduced fire ant Solenopsis invicta. Molecular Biologyand Evolution 16: 525–543.
Ross KG, Trager JC. 1990. Systematics and populationgenetics of fire ants (Solenopsis saevissima complex) fromArgentina. Evolution 44: 2113–2134.
Rousset F. 1996. Equilibrium values of measures of popula-tion subdivision for stepwise mutation processes. Genetics142: 1357–1362.
Rousset F. 1997. Genetic differentiation and estimation ofgene flow from F-statistics under isolation by distance.Genetics 145: 1219–1228.
Sanetra M, Crozier RH. 2003. Patterns of population sub-division and gene flow in the ant Nothomyrmecia macropsreflected in microsatellite and mitochondrial DNA markers.Molecular Ecology 12: 2281–2295.
dos Santos EP. 1995. Phylogeny and phytogeography of thegenus Salvia L. section Rudes (Benth.) EpI. (Lamiacees).Biogeographyraphica (Paris) 71: 15–32.
Schneider S, Roessli D, Excoffier L. 2000. Arlequin: asoftware for population genetics data analysis, version2.000. Geneva: Genetics and Biometry Laboratory, Depart-ment of Anthropology, University of Geneva.
Shimizu KK. 2002. Ecology meets molecular genetics inArabidopsis. Population Ecology 44: 221–233.
Shoemaker DD, Ahrens ME, Ross KG. 2006a. Molecularphylogeny of fire ants of the Solenopsis saevissima species-group based on mtDNA sequences. Molecular Phylogeneticsand Evolution 38: 200–215.
Shoemaker DD, Costa JT, Ross KG. 1992. Estimates ofheterozygosity in two social insects using a large number ofelectrophoretic markers. Heredity 69: 573–582.
Shoemaker DD, DeHeer CJ, Krieger MJB, Ross KG.2006b. Population genetics of the invasive fire ant Solenop-sis invicta in the USA. Annals of the Entomological Societyof America 99: 1213–1233.
Sites JW, Marshall JC. 2004. Operational criteria for delim-iting species. Annual Review of Ecology, Evolution andSystematics 35: 199–227.
Sivasundara A, Hey J. 2003. Population genetics ofCaenorhabditis elegans: the paradox of low polymorphism ina widespread species. Genetics 163: 147–157.
Slatkin M. 1987. Gene flow and the geographic structure ofnatural populations. Science 236: 787–792.
Slatkin M. 1993. Isolation by distance in equilibrium andnon-equilibrium populations. Evolution 47: 264–279.
Sundström L, Keller L, Chapuisat M. 2003. Inbreedingand sex-biased gene flow in the ant Formica exsecta.Evolution 57: 1552–1561.
Trager JC. 1991. A revision of the fire ants, Solenopsisgeminata group (Hymenoptera: Formicidae: Myrmicinae).Journal of the New York Entomological Society 99: 141–198.
Tschinkel WR. 2006. The fire ants. Cambridge, MA: HarvardUniversity Press.
Tsutsui ND, Suarez AV, Holway DA, Case TJ. 2001.Relationships among native and introduced populations ofthe Argentine ant (Linepithema humile) and the source ofintroduced populations. Molecular Ecology 10: 2151–2161.
Van der Hammen T, Pedersen JS, Boomsma JJ. 2002.Convergent development of low-relatedness supercolonies inMyrmica ants. Heredity 89: 83–89.
Van Driesche R, Bellows TS. 1996. Biological control.New York, NY: Chapman & Hall.
Veuille M, Baudry E, Cobb M, Derome N, Gravot E. 2004.Historicity and the population genetics of Drosophila mela-nogaster and D. simulans. Genetica 120: 61–70.
Weir BS, Cockerham CC. 1984. Estimating F-statistics forthe analysis of population structure. Evolution 38: 1358–1370.
Whitlock MC. 2005. Combining probability from indepen-dent tests: the weighted Z-method is superior to Fisher’sapproach. Journal of Evolutionary Biology 18: 1368–1373.
Whitlock MC, McCauley DE. 1999. Indirect measures ofgene flow and migration: FST � 1/(4Nm+1). Heredity 82:117–125.
Wilson GA, Rannala B. 2003. Bayesian inference of recentmigration rates using multilocus genotypes. Genetics 163:1177–1191.
Yang RC. 2004. A likelihood-based approach to estimatingand testing for isolation by distance. Evolution 58: 1839–1845.
Young FW. 1996. Vista: the visual statistics system. ResearchMemorandum 94-1(b), 2nd edn. Chapel Hill, NC: LLThursone Psychometric Laboratory, University of NorthCarolina.
Zaykin D, Zhivotovsky L, Weir BS. 1995. Exact tests forassociation between alleles at arbitrary numbers of loci.Genetica 96: 169–178.
Zhang DX, Hewitt GM. 2003. Nuclear DNA analyses ingenetic studies of populations: practice, problems and pros-pects. Molecular Ecology 12: 563–584.
