Cytonuclear genetic architecture in mosquitofish populations and the possible roles of introgressive hybridization

Spatial genetic structure in populations of mosquitofish (Gambusia) sampled throughout the south‐eastern United States was characterized using mitochondrial (mt) DNA and allozyme markers. Both sets of data revealed a pronounced genetic discontinuity (along a broad path extending from south‐eastern Mississippi to north‐eastern Georgia) that corresponds to a recently recognized distinction between the nominal forms G. affinis to the west and G. holbrooki to the east. However, several populations from the general contact region exhibited unusual allelic associations in high frequency, suggestive of evolutionary processes within a zone of introgressive hybridization. These involve: (i) cytonuclear profiles representing combinations of nuclear and mitochondrial genotypes that tended to be more nearly species‐specific and concordant elsewhere; and (ii) significant nuclear gametic disequilibria, perhaps attributable to positive assortative mating and/or differential fitnesses of homospecific vs. recombinant genotypes. However, outside this suspected hybrid region, ‘heterospecific’ genetic markers also appeared in low frequency, thus complicating interpretations. These discordant alleles on a broader geographic scale may reflect: (a) the retention of polymorphisms from an ancestral gene pool; (b) occasional evolutionary convergence (especially with respect to electrophoretic mobility of allozyme alleles); (c) the ‘footprints’ of a moving hybrid zone; or (d) differential introgressive penetrance across the current hybrid region.


Introduction
Considerable attention has been focused recently on ecological, demographic and evolutionary factors that mould observed associations between genotypes and geographic distributions (Avise 1989). Kenewed interest in this area has been motivated in part by refinement of molecular techniques capable of resolving microevolutionary variation in genetic markers with contrasting modes of inheritance. Concordance in spatial distributions and depths of genealogical topologies across multiple genetic markers, and among evolutionarily independent taxa, has provided strong evidence linking contemporary population 'Present address: Alaska Fish and Wildlife Research Center, US Fish & Wildlife Service, 101 1 E. Tudor Rd, Anchorage, AL 99503,

USA.
Correspondence: J.C. Avise. structure to historical processes (Wilson ef al. 1985;Avise et al. 1987). Joint examinations of nuclear and cytoplasmic (e.g. mitochondrial) genes allow further assessment of how sexual asymmetries in gene flow, mating preferences, or other demographic factors can influence population genetic structure, particularly in hybrid settings (Arnold et al. 1988;Asmussen et al. 1989;Asmussen & Arnold 1992).
Secondary hybrid zones offer favourable settings to study evolutionary forces that may have affected genetic distributions (Hewitt 1988). Taxa involved are likely to exhibit numerous genetic differences that evolved in allopahy and that can serve as convenient markers in the secondary contact region. Genotype-specific differences in mating preference, fertility, or viability are often more readily identified in hybrid settings, making their effects on population genetic structure easier to evaluate. Finally, the shape and width of a hybrid zone, and the ability of genetic material to penetrate it, may be functions of the magnitude and direction of gene flow relative to the inten-

M t D N A characterization
sity-and direction of selection (Barton & Hewitt 1985).
Mosquitofish (Gambusia) occur over a large area of the south-eastern USA (Lee ef al. 1980) and exhibit considerable spatial variation in gene frequencies (Wooten et al. 1988). Mate choice experiments (Scribner, in press), studies of species-specific differences in life-history traits (Reznick 1981;Scribner 1992Scribner , 1993 and preliminary analyses within a putative hybrid zone in nature (Wooten & Lydeard 1990; Lydeard et a/. 19911, all suggest that mating between two nominal species -G. affinis (Baird and Girard) and G. holbrooki (Girard) -is not random. Furthermore, fertility and viability differences among parental and hybrid genotypes have been shown to lead to changes in population rates of recruitment and mortality Mitochondria1 DNA (mtDNA) was isolated from fresh tissues using standard procedures of CsCl density gradient centrifugation (Lansman et al. 1987). Because of low mtDNA yields, only large (> 30 mm standard length) females were used. Eleven restriction enzymes (listed in Table 1) were scored in n = 4 individuals for most localities. Restriction digests were conducted overnight under specifications recommended by the enzyme supplier (Boerhinger, Mannheim). Digestion fragments were endlabelled with appropriate 35S-radionuclides, separated on 1 .O% agarose gels, and revealed by autoradiography (Brown 1980;Maniatis et al. 1982). Fragment sizes were compared against a I-kb ladder standard (Gibco).
within experimental mixed populations (Scribner 1993), and concordantly to changes in population gene frequen-

Allozyme characterization
cies (Scribner & Avise, in press). Because of initial reports indicating extensive sympatric-parapatric associationbased on both genetic (Wooten et al. 1988;Lydeard et al. 1991 ) and morphological (R. Angus, personal communication) criteriaand hybridization between these species (Wooten & Lydeard 1990), and because predictions about the nature of selection against specific parental and hybrid genotypes are available, Gambusia offer an excellent opportunity for studying processes governing introgressive hy-  Wooten & Lydeard 1990). Alleles were designated by gel mobilities relative to a reference electromorph from G. affinis ('100').
backdrop against which to interpret results for Gambusia.
The objectives of this study were: Statistical analyses 1 to estimate levels and spatial patterns of mtDNA and allozyme divergence within and among populations of G. affinis and G. holbrooki throughout the Southeastern USA; 2 to assess distributional patterns of uni-and biparentally inherited markers within and outside a suspected hybrid zone; and 3 to interpret results in the context of the contemporary natural histories of the species and the historical zoogeography of the region.

Materials and methods
Fish were collected using dip nets from 45 locations in seven states (SC, GA, FL, AL, MS, LA.and TX) during the spring and summer of 1989, and the specimens were immediately frozen in liquid nitrogen for later protein analysis. Additional females were randomly chosen from field collections obtained from each site and returned live to the laboratory for mtDNA isolations.
All differences among mtDNA gel-fragment patterns could be interpreted as gains or losses of particular restriction sites. Multienzyme mtDNA genotypes were designated by lower-case letters, and estimates of nucleotide sequence divergence among composite genotypes were calculated using the 'site method' (Nei & Li 1979). Distance estimates were used to construct a UPGMA phenogram (Sneath & Sokal 1973). A data matrix of restriction site presence/absence was also used to generate Wagner parsimony networks (by the branch-swapping option) and a consensus network (using the majority-rule option) of PAUP (version 3.0; Swofford 1990). Bootstrapping (100 replicates; Felsenstein 1985) was conducted to allow confidence statements about particular branches in the Wagner network. Frequencies of allozyme alleles normally characteristic of G. affrnis and G. holbrooki were calculated for all locations, and a 'species index score' (number of G. affinis allelesrange 0-10) was computed for each individual. Multilocus genotypes were used to estimate composite measures of gametic phase disequilibrium (AAB; Weir 1979;Weir & Cockerham 1989), and chi-square tests (Weir & Cockerham 1989) were used to evaluate whether these disequilibrium measures for each locale differed significantly from zero. Although comparisons of disequilibrium measures may be biased somewhat by differences in allele frequency across locations, suggested adjustments (Hill 1974;Hedrick 1987) typically assume random union of gametes, which is unlikely in our hybrid populations. Therefore only major qualitative points will be emphasized from the disequilibrium analyses.
Allelic frequencies were used to calculate genetic distances between population samples (Rogers 1972), and a UPGMA phenogram (Sneath & Sokall973) was constructed from the distance matrix. Because only taxonomically informative loci were assayed, differences among localities should not be interpreted as estimates of overall genetic relatedness.

Spatial structuring of mtDNA haplotypes
A total of 173 individuals from 45 localities was screened for mtDNA variation. Nine of 11 enzymes produced poly-morphic fragment profiles, revealing 65 restriction sites and 22 distinct haplotypes ( Table 1). Differences among these haplotypes were attributable to changes at 30 restriction sites.
Phenetic analyses revealed two distinct clusters that differ by an estimated sequence divergence of P = 0.038 (Fig. la). This fundamental distinction in the UPGMA phenogram was also evident in the qualitative parsimony analysis, where statistical support based on the bootstrap replicates conducted was at the 100% level (Fig. lb). Five enzymes (BumHI, BgZII, DraII, HincII, and PvuII) contributed to the fundamental genetic distinction by providing diagnostic and mutually concordant support for these two haplotype groups. These mtDNA groups also showed a strong geographic orientation, with the genetic break following a south-west-north-east orientation from the Pascagoula River in south-eastern Mississippi to the headwaters of the Savannah River in north-eastern Georgia ( Fig.2). At no locale were representatives of both mtDNA groups observed, but this may be due to the small sample sizes per locality.
Within either mtDNA assemblage, strong statistical support for additional mtDNA groupings was generally lacking (Fig. lb). Nonetheless, geographic population structure was evident in the distributions of particular  mtDNA clones. For example, mtDNA haplotype (a) was nearly confined to the northern Atlantic coast drainages surveyed, and haplotypes (v) and (u) were confined to samples from Texas and eastern Louisiana, respectively (Fig. 2). Beyond these cases, only two mtDNA haplotype distributions warrant mention. In the UPGMA analysis, two genotypic clusters within the G. holbrooki group differed by B an estimated P = 0.01, and this distinction also received moderate (67%) bootstrap support in the parsimony analysis (Fig. 1). One of these putative lineages (clones c, e, f, p, q and r) was confined to Gulf coast locales, whereas the other was widespread throughout the G. holbrooki range and was fixed in the samples from Atlantic coast drainages (Fig. 2). Fig. 2 Geographic distribution of the mtDNA haplotypes described in Table 1. Closed and open circles represent the two major clonal assemblages ( Fig. 1)   Regional differences in measures of mtDNA diversity were also observed. Lowest genetic diversity occurred in populations from northern Atlantic coast drainages of Georgia and South Carolina where only one mtDNA haplotype (a) was observed (Fig. 2). Highest mtDNA diversities were found in G. holbrooki populations collected from southern regions of drainages flowing into the Gulf of Mexico in Mississippi, Alabama, and Florida (due partly to the occurrence there of haplotypes representing both of the two major holbrooki mtDNA haplotype groups). This latter region was also characterized by a number of unique and localized mtDNA haplotypes that differ at one or two restriction sites from the common holbrooki genotype (h).

Spatial variation in allozyrne frequencies
A total of 1259 individuals from the same 45 localities was screened for allozyme variation. At the five polymorphic nuclear loci monitored, allelic distributions revealed a macrogeographic population structure similar to that observed among mtDNA haplotypes (Figs 3 and 4), with a major genetic discontinuity in western Georgia and eastern Alabama registered in allele frequency differences and species index scores (Table 2). At most locations, relatively few holbrooki alleles were observed in fish that on the basis of mtDNA genotype would be characterized as G. ufinis. Thus the western locales 1-5 exhibited no holbrooki alleles, whereas locales 9-15 and 27 had holbrooki alleles in frequencies less than 0.25. However, an exception involved    (Table 2).
At the nuclear loci assayed, many populations along the mtDNA contact area between G. holbrooki and G. affinis were characterized by significantly higher non-zero gametic disequilibria than described for most other localities (Fig. 5). Estimates of nuclear gametic disequilibrium were significant in populations 7,8,20,23 and 28, which exhibited holbrooki mtDNA, and populations 9, 12, 14 and 21, which exhibited uffinis mtDNA ( Table 2). All of thesepopulations occur in the general region of contact between G.
nffinis and G. holbrooki in Alabama and western Georgia.

Discussion
Whereas a spatial genetic partitioning of G. uffinis and G. holbrooki would appear straightforward based on mtDNA designation alone (Figs 1 and 2), lack of perfect concordance with the allozyme allele frequencies, and the presence of nonparental genotypic combinations and intermediate species index scores for nuclear loci in the general contact region, raise the likelihood that historical introgression has occurred over a broad central area in the south-eastern USA. On the other hand, observations of nonrandom associations among nuclear alleles in this same area suggests that there also has been a strong tendency for speciesspecific assortative mating (and /or selection against specific nuclear genotypes). The picture is further complicated because nuclear alleles of both species also segregate in low frequency in some populations outside the current geographic region of the major genetic discontinuity. The spatial distributions of distinct mtDNA haplotypes and nuclear allozymes documented here are generally consistent with expectations based on the recently described ranges of G. holbrooki and G. uffinis (Wooten et ul. 1988).
[This outcome was not a foregone conclusion, because Gumbusiu relationships and taxonomy have been highly controversial, and all populations in the south-eastern USA are often treated as belonging to a single species with geographic and genetic affinities uncertain (e.g. Lee et al. 1980)l. Furthermore, the boundary region between the holbrooki and nffinis genetic types generally is concordant with mtDNA phylogeographic discontinuities reported within several other fish species in the region (Bermingham et ul. 19861, as well as with an area of concentration for species' distributional limits within numerous taxonomic assemblages of freshwater fishes (Swift et al. 1985;Wiley & Mayden 1985). These similarities in distinction between closely related eastern and western fish faunas of the south-eastern USA argue strongly for an overriding effect of historical biogeographic phenomena in shaping macrogeographic population genetic structure (Avise 1992). A second genetic subdivision of lesser magnitude, recognized in mtDNA analyses (Figs 1 and 2) as well as by overall allozyme frequency differences (Table 21, distinguishes many (but not all) populations of G. holbrooki from the Gulf coastal plain from those to the east in peninsular Florida, Georgia, and the Carolinas. Thus, for the southeastern Gnmbusia complex as a whole two important historical separations are implicated: an early, well-supported genetic separation leading to the current differences between G. uffinis and G. holbrooki, and a later, less well-supported genetic separation leading to eastern vs. Gulf coast forms of G. holbrooki.
One possibility is that ancestors of G. holbrooki may have been isolated in an eastern refugium, perhaps the Florida peninsula, which appears to be the stronghold for the eastern genetic form of many freshwater fish species (Bermingham & Avise 1986). A second dispersal or vicariant event may then have initiated the separation between eastern and western holbrooki types. If so, there may also have been recent gene flow between these latter areas, because the mtDNA genotype (h) (as well as others within the same haplotype group; Fig. 1) are shared by Gulf coast and Atlantic coast populations (Fig. 2). Alternatively, such inter-regional sharing of genotypes might be attributable to the retention of unmodified ancestral haplotypes.
Additional population substructure within G. uffinis and G. holbrooki provides further clues to historical patterns of gene flow and population separation. For example, G. holbrooki in the northern Atlantic drainages is depauperate in mtDNA variation and carries in high frequency an otherwise uncommon mtDNA haplotype, (a), which is only one mutation step removed from (h) which predominates to the south (Fig. 2). These northern drainages were probably colonized fairly recently from a larger and more variable source population in Florida or southern Georgia. Within G. ufinis, a distinctive mtDNA genotype (u) (Fig. 1) was confined to and fixed in the sample from one locale in eastern Louisiana (Fig. 2), indicat-. ing that geographic structure occurs within this species as well.

Historical and contemporary hybridization
Recent attention has been focused on concordance and discordance of different molecular markers in hybrid zones. Use of multiple markers is particularly important for the study of introgression because the complexity of hybridization processes precludes a universal marker for understanding these phenomena ( Avise et al. 1984;Szymura et ul. 1985;Baker et a/. 1989;Dowling et ul. 1989).
The availability of several allozyme and mtDNA markers in the current study allows examination of the degree of geographic concordance across independent loci (Figs 2 and 4). Populations within the general contact region between G. uffinis and G. holbrooki were not characterized solely by parental and F, individuals (species index scores, Table 2), but also included recombinant nuclear and cytonuclear genotypic associations strongly suggestive of historical introgression. Observations of significant gametic disequilibria among nucIear genotypes indicate that opposing forces are also at work in the hybrid zone that have prevented a randomization of genotypes that otherwise might be expected under extensive hybridization alone. High disequilibrium could be maintained either by gene flow of parental types into the hybrid zone or by positive assortative mating and/or selection against certain recombinant genotypes.
Additional evidence from controlled laboratory crosses (Scribner 1992,19931, and temporal genetic studies of 'artificial' hybrid populations between G. uffinis and G. holbrooki (Scribner & Avise, in press), argue strongly for the possibility of rapid genetic changes within natural Gumbusia hybrid zones and for movements in hybrid zone position. A lower population carrying capacity for G. uffinis, and lower fitness of juveniles of ujfinis maternal parentage in mixed populations with G. holbrooki (due to differences in size at birth, growth rates, and size and age at sexual maturity; Scribner 1992Scribner , 1993 suggest a directionality to changes in genotypic composition. In addition, mixed-species experimental populations exhibited a rapid and directional change in genotypic composition toward elimination of affinis genotypes over a period of 2 years (about four generations). These experimental studies were conducted in the absence of gene flow, and under relatively uniform environmental conditions, but the results clearly showed that genotypic composition in hybrid zones might be quite dynamic in nature as well. Existing theory predicts that hybrid zones will tend to move from areas of high fitness and population density to regions where these properties are low (Nagylaki 1976;Barton & Hewitt 1985).
Nuclear alleles normally characteristic of G. holbrooki were observed in low or moderate frequency particularly west and north of the zone of major mtDNA discontinuity (Figs 2 and 4). The presence of occasional electromorphs 'in advance' of the mtDNA break is suggestive of differential introgression of nuclear and mtDNA markers. This interpretation is consistent with behavioraI observations of species-specific differences in degree of assortative mating: in laboratory and field experiments, female G. uffinis mate randomly with regard to species whereas G. holbrooki females mate almost exclusively with males of their own species (Scribner, in press). Thus, the presence of nuclear holbrooki alleles in hybridized populations that carry afinis POPULATION GENETIC STRUCTURE IN Garnbusiu 147 mtDNA might be attributable to male-mediated introgression. It has further been reported that F, females from crosses between G. holbrooki females and G. affinis males are infertile, whereas no decline in fertility was observed among F, females from the reciprocal cross (Reznick 1981). This result too is consistent with retardation of holbrooki cytoplasmic penetrance through the hybrid zone. Although the direction of both pre-and post-zygotic isolating mechanisms suggests that holbrooki nuclear genes may precede maternally inherited mitochondria1 genes through a hybrid region via male-biased gene flow, the possibility that these electromorphs represent retention of ancestral states (or the products of convergent evolution) cannot be eliminated entirely.
Conversely, nuclear alleles that normally predominate in G. affinis were observed occasionally in G. holbrooki populations from Florida and Atlantic and Gulf coastal populations (Fig. 3). The presence of affinis alleles could again reflect simply the retention of ancestral polymorphisms predating the species separation. Alternatively, affinis alleles in at least some G. holbrooki populations might represent evolutionary footprints from an earlier broadly distributed G. affinis population that historically was assimilated by ancestral G. holbrooki dispersing out of Florida. However, arguing rather strongly against the footprint scenario are the facts that: (a) affinis mtDNA genotypes were not similarly retained in G. holbrooki populations, despite the previously mentioned asymmetries in mating preference and hybrid fertility that would appear to favour such maintenance; and (b) affinis-type electromorphs were observed in moderate frequency even in Florida locations where G. uffinis had presumably never been present.
Previous studies of introgressive hybridization in Gurnbusia species have stressed additional ecological factors as important in maintaining species integrity in sympatry, such as habitat preferences (Hubbs 1959;Hubbs & Delco 1962;Hubbs & Peden 1969). These studies involved G. uffinis and several localized endemics in Texas (G. geiseri, G. georgei and G. heterochin). The authors suggested that ecological separation based on water chemistry, temperature, and distribution of vegetation contributed to positive assortative mating and prezygotic isolation. Similar factors may also be at work in the holbrooki-uffinis contact region. For example, in Atlantic and Gulf coast drainages that contain both species, G. uffinis predominates in the headwaters whereas G. holbrooki generally predominates along the coastal plain. This suggests that even in the absence of firm geographic impediments to dispersal, ecological preferences can have an important influence on species distributions. Whether ecological conditions partially segregate G. affinis and G. holbrooki on finer, syntopic spatial scales remains to be determined.

Synopsis
There are two general but somewhat opposing messages to be derived from these analyses of cytonuclear population structure in mosquitofish. First is the impressive amount of historical information apparently retained in a relatively small number of nuclear and cytoplasmic loci. Not only is a major distinction between two otherwise cryptic Cambusiu forms in the south-eastern USA blatantly obvious from the genetic data, but also several finer details about geographic relationships within these groups, and probable patterns of gene flow and introgressive hybridization between them have been revealed. However, a second and competing message is that there are also definite limits to inferences that can be drawn from purely descriptive surveys of contemporary population structure. As is evident from the discussion above, alternative scenarios can often be advanced to account for observed genetic architectures (in Garnbusia o r elsewhere), such that historical reconstructions based solely on static descriptions of contemporary population structure will inevitably have elements of uncertainly that may never be overcome entirely. However, for hybridizing species that can be manipulated experimentally, additional insight into the forces governing population genetic architecture may be gained through detailed behavioral observations on ecological and mating preferences, as well as from 'population cage' experiments designed to monitor temporal genetic changes following secondary contact and introgressive hybridization in controlled settings. Results of such experiments with Garnbusia will be presented elsewhere.