The Desert Pupfish Cyprinodon macularius is listed as an endangered species by the International Union for Conservation of Nature and Natural Resources (Miller, 1979) and by the United States government (U.S. Department of the Interior, 1986). In this paper, we use mitochondrial DNA (mtDNA) variation to describe the genetic structure of populations traditionally classified as C. macularius and to provide a basis for recommendations regarding conservation genetics of the species. The results, together with geological history and color patterns in nuptial males, indicate that populations grouped as C. macularius represent two evolutionarily divergent entities that should be recognized as separate species. We hereafter refer to the two species as the “desert pupfish complex.”

The historic range of the desert pupfish complex once extended from Gila River tributaries in southeastern Arizona and northern Sonora, westward to the Salton Sea area of southern California and southward into the Colorado River Delta region in Sonora and Baja California (Miller, 1943). Across this region, Miller and Fuiman (1987) recognized one wide-ranging subspecies, C. m. macularius, and a local endemic, C. m. eremus, restricted to Quitobaquito Springs, Organ Pipe Cactus National Monument, southern Arizona. The wide-ranging subspecies has disappeared from the Gila and lower Colorado River areas of Arizona/California (Minckley, 1973; Moyle, 1976), and the remaining natural populations are much smaller and more fragmented than in earlier times (Hendrickson and Varela, 1989; Dunham and Minckley, 1998). This decline is a result of interactions with introduced species and habitat losses resulting from reservoir construction, surface-water diversion projects, groundwater depletion, and other factors (Miller and Fuiman, 1987; Shoenherr, 1988; Hendrickson and Varela, 1989).

Previous studies of genetic variation in the desert pupfish complex have been restricted to populations in the United States (Turner, 1984), or they included only one or two samples of transplanted Mexican populations (Turner, 1983; Dunham and Minckley, 1998). Other genetic studies that included desert pupfish were aimed primarily at resolving evolutionary relationships among western pupfishes and included samples from only one or two localities (Turner, 1974; Echelle and Dowling, 1992; Echelle and Echelle, 1993).

Materials and Methods

In 1997–1998, collections of the desert pupfish complex were made at 11 sites (n = 19–25) representing all major populations (Fig. 1; Appendix). We also included 33 specimens from a captive population at Dexter National Fish Hatchery and Technology Center (DNFH) in New Mexico. Specimens were initially frozen in liquid nitrogen or on dry ice and stored in the lab at −75 C until processed for analysis.

View Figure

• Download as Powerpoint
• Download Figure

DNA was extracted from muscle tissue following the method of Longmire et al. (1997), and the polymerase chain reaction (PCR) and single-stranded conformational polymorphism (SSCP) analysis (Orita et al., 1989; Fan et al., 1993) were used to screen all specimens for variation in portions of the mitochondrial D-loop and ND2 gene. For ND2, we used primers ND2B-L and ND2E-H (T. E. Dowling, pers. comm.) to amplify and sequence the entire gene. Using the ND2 sequence, we developed internal primers for a 333-bp segment (5′-CCTTCCTTTGCTAATGAACC-3′ and 5′-CCAA-TTTTAAGTGCCAGGG-3′) at the 5′ end (positions 8 to 340) and a 325-bp segment (5′-ATACCTCGCCACCTCTTG-3′ and 5′-AGCCAGATTGTTGCGGAG-3′) at the 3′ end (positions 661 to 986). For SSCP analysis of the D-loop, we used primers L15926 (Kocher et al., 1989) and H16498 (Meyer et al., 1990) to amplify a portion of mtDNA extending from the threonine tRNA gene through 337 bp of the D-loop; these are primers E and K in Lee et al. (1995). To screen for SSCP variation, the three mtDNA segments from each specimen were PCR-amplified from genomic DNA in separate reactions and in the presence of α-³²P-dCTP, 2.0 µl of each diluted PCR product was electrophoresed in a 5% polyacrylamide gel (40:1 acrylamide:bis acrylamide; 10% glycerol) at 300 V (3 W) for 22 h, and bands were visualized by autoradiography.

For phylogenetic analysis, we PCR-amplified the entire ND2 gene (1047 bases; primers ND2B-L and ND2E-H) and the 337-bp fragment of the mtDNA D-loop (primers L15926 and H16498) in 35 specimens representing all composite haplotypes detected by SSCP analysis. The amplicons were sequenced directly using either an ABI 373 or 377 automated sequencer. All haplotypes occurring at more than one locality were sequenced in representatives from two to six localities, depending on distribution of the haplotype. One haplotype occurred in nine of the 11 samples and was sequenced in nine specimens from six different localities. To test the monophyly of haplotypes in desert pupfish, we included GenBank sequences provided by Duvernell and Turner (1998) for three other pupfishes, two species grouped with the desert pupfish complex in the “western pupfish clade” (Echelle and Dowling, 1992), the Amargosa Pupfish, C. nevadensis (GenBank accession numbers: ND2, AF028308; D-loop, AF028290), and Owens Pupfish, C. radiosus (ND2, AF028311; D-loop AF028293), and a rather distantly related (Echelle and Echelle, 1998) outgroup species, the Sheepshead Minnow, C. variegatus (ND2, AF028313; D-loop, AF028299). The mtDNA of C. radiosus is the sister group to the mtDNA lineage in the desert pupfish complex (Echelle and Dowling, 1992). Duvernell and Turner's (1998) sequences included the entire ND2 gene, and their D-loop sequences are based on the primers we used.

Sequences for each composite haplotype identified by SSCP and those extracted from GenBank were aligned with CLUSTAL W (Thompson et al., 1994) and subjected to phylogenetic analysis with PAUP (vers. 3.05, D. L. Swofford, Champaign, IL, 1991, unpubl.). Sequence data were coded as discrete, unordered characters, and we used PAUP's heuristic search algorithm with 10 iterations of random input order of taxa to find the most parsimonious trees. To evaluate robustness of the results, we performed a heuristic bootstrap analysis (Felsenstein, 1985) of 100 iterations with 10 replications of random input order and tree-bisection-reconnection for each iteration. Gaps representing four indels in the control region were treated as missing data.

Haplotype frequencies and sequences for the three mtDNA segments assayed for SSCP variation were used in analyses of population genetic structure. We used Arlequin (vers. 1.1 S. Schneider, J.-M. Kueffer, D. Roessli, and L. Excoffier, Geneva, Switzerland, 1997, unpubl.) to compute within-sample estimates of haplotype diversity (h) and nucleotide diversity (π; from Tajima-Nei distances, no gamma correction), pairwise F_ST and Φ_ST statistics, exact tests (Raymond and Rousset, 1995) of difference in haplotype frequencies, and an analysis of molecular variance (AMOVA; Excoffier et al., 1992). The AMOVA partitioned molecular variance into proportions resulting from within-sample variation, differences among samples within regions, and differences among regions. We designated three regions for this analysis: the Salton Sea area (sites 1–3, Fig. 1), the Colorado River Delta (sites 4–9), and the Río Sonoyta/Quitobaquito area (sites 10–11). The AMOVA was performed on haplotype frequencies alone (producing F-statistics) and on frequencies plus sequence divergences (producing Φ-statistics; Excoffier et al., 1992). Also using Arlequin, we obtained significance tests for individual, pairwise F_ST and Φ_ST values and the variance components in the AMOVAs [nonparametric permutation method described by Excoffier et al. (1992)]. For multiple pairwise tests, we used the stepwise Bonferroni correction (Rice, 1989) to obtain a Type I error less than 0.05.

Results

SSCP analysis of haplotype variation among 258 specimens of desert pupfish identified 10 variants for the 5′ end of the ND2 gene and 5 for the 3′ end of the gene, producing 13 composite ND2 haplotypes; nine variants were detected for the D-loop, giving 18 composite haplotypes across the three segments (Table 1). All SSCP-identified haplotypes for each of the three segments had at least one unique base substitution. In three instances, haplotypes identical on the basis of SSCP differed by one base substitution in the midportion of the ND2 gene that was not surveyed for SSCP variation; these variants were ignored in the analysis because additional sequencing would have revealed other base substitutions for other portions of the mtDNA. All ND2 and D-loop sequences have been deposited in GenBank (accession numbers AF198967–AF199002).

Table 1. Number of Occurrences for 18 Haplotypes in the Desert Pupfish Complex, with Haplotype Diversity (h) and Nucleotide Diversity (π) for the ND2 Gene, the D-loop, and Both Combined. Locality numbers correspond with those in Figure 1 and the Appendix

View Fullscreen

Genetic structure

The 18 haplotypes detected by SSCP analysis varied at a total of 18 sites (15 transitions, three transversions) across the two ND2 segments surveyed: 11 in the 333-bp segment at the 5′ end and seven in the 325-bp segment at the 3′ end. Five variable sites (all with transitions) were detected in the ND2 segments not included in the SSCP analysis (389 bp). For the 337-bp D-loop fragment, there were 12 variable sites and 13 substitutions (eight transitions, five transversions). Individual haplotypes differed at one to 10 sites. The measures of haplotype and nucleotide diversities showed no geographic pattern of variation and no consistent pattern of variation between ND2 and the D-loop (Table 1).

All populations had two to five SSCP variants except the DNFH sample, which was monomorphic for haplotype A. The Río Sonoyta and Quitobaquito populations exhibited no significant difference in haplotype frequencies (Tables 1–2). These populations shared no haplotypes with samples from the Salton Sea and the Colorado River Delta areas. Thus, all pairwise tests of F_ST and Φ_ST and the corresponding exact tests of differences in haplotype frequency were highly significant (P < 0.0001) in comparisons of Río Sonoyta/Quitobaquito samples with Salton Sea and Colorado River Delta samples. All Salton Sea and Colorado River Delta samples had high frequencies (60–89%) of haplotype A, and, with the Bonferroni correction, there were no significant differences among paired samples within or between the two regions. There was, however, evidence of significant divergence between the captive stock (DNFH) we examined and its parent population (site 6, Fig. 1; F_ST = 0.31, P < 0.00001; Φ_ST = 0.21, P < 0.00001).

The AMOVA based on F-statistics indicated that 34.3% of the variation in haplotype frequencies is attributable to differences among the three regions (Río Sonoyta/Quitobaquito, Salton Sea, and Colorado River Delta), 64.8% to variation within samples, and only 0.9% to differences between samples within regions. The corresponding values for Φ-statistics (“total diversity”; i.e., haplotype frequencies plus sequence divergences) were, respectively, 75.8%, 23.6%, and 0.6%, indicating that a large proportion of the among-region genetic variance reflects sequence differences and not just haplotype frequencies. For both analyses, there were highly significant differences among regions (P < 0.00001) but not among samples within regions (P = 0.11–0.15). More than 70% of the total diversity reflects differences between samples from the Río Sonoyta/Quitobaquito region and those from the Salton Sea and Colorado River Delta regions. When the Río Sonoyta/Quitobaquito populations were excluded from the analysis, only 3.7% of total diversity was attributable to differences between the Salton Sea and Colorado River Delta regions (Table 2). The corresponding value for differences between the samples from Río Sonoyta and Quitobaquito Springs was 2.9%.

Table 2. Percentage of Genetic Variation Explained by Different Levels of Geographic Structure in Desert Pupfish. Results of hierarchical analyses of genetic diversity for different segments of mtDNA. Asterisks signify probability levels: *<0.05, **<0.01, ***<0.0001

View Fullscreen

Considering only haplotype frequencies, roughly equivalent proportions of gene diversity were attributable to differences among regions for the ND2 gene and the D-loop sequences (Table 2). However, when sequence divergence was included in the AMOVA, the among-region component of diversity was much greater for ND2 than for the D-loop (87% vs 11%; Table 2). Within regions, the two sequences are roughly equivalent in proportion of genetic diversity attributable to differences among samples.

A small (3.7%), but statistically significant, among-region component of variation was detected in the combined ND2/D-loop data for the Salton Sea and Colorado River Delta when sequence data were included in the analysis but not when sequence data were excluded (Table 2). The two regions share the common haplotype (A), and the difference in sequence information is attributable to the remaining, relatively uncommon, haplotypes in the two regions (Table 1). The geographically intermediate population at Cerro Prieto (Site 4, Fig. 1) shared one uncommon haplotype (E) with Salton Sea populations and none with the other populations on the delta.

Phylogenetic analysis

The 1384 sites available for phylogenetic analysis of the ND2/D-loop sequences included 191 variable sites, 54 of which were parsimony-informative (40 ND2 sites, 14 D-loop sites); that is, two or more (but not all) haplotypes in the ingroup differed from the outgroup (C. variegatus) haplotype by a shared base substitution. Only 13 sites (12 ND2, 1 D-loop) were informative regarding relationships among haplotypes of the desert pupfish complex. The heuristic search for the shortest tree with the combined ND2/D-loop data produced two equally parsimonious topologies (230 steps; CI = 0.89) after collapsing all nodes with no synapomorphies. The strict consensus tree was identical to the 50% majority-rule consensus tree from the bootstrap analysis (Fig. 2). These trees, and similar trees produced by analysis of ND2 separately, supported monophyly of the desert pupfish complex and the previously reported (Echelle and Dowling, 1992; Duvernell and Turner, 1998) close relationship between haplotypes of the complex and those of C. radiosus (Fig. 2). D-loop data also supported monophyly for the desert pupfish complex (bootstrap support = 99%).

View Figure

• Download as Powerpoint
• Download Figure

Haplotypes of the desert pupfish complex formed two monophyletic clades, one comprising four haplotypes restricted to the Río Sonoyta/Quitobaquito area, and a second comprising 14 haplotypes restricted to the Salton Sea/Colorado River Delta area (Fig. 2). The Salton Sea/Colorado River Delta clade included one subclade containing haplotypes B (widespread in Salton Sea area) and F (widespread in the Colorado River Delta) and one containing all other Salton Sea/Colorado River Delta haplotypes. In the latter subclade of 12 haplotypes, the most widespread member (Haplotype A; Table 1) was also the most plesiomorphic, with no evidence of divergence from the hypothetical common ancestor of the group, whereas the other members of the group showed 1–3 derived mutations (Fig. 2). These 12 haplotypes formed a large polytomy with no phylogenetic resolution except that haplotypes C and D from the Salton Sea area formed a monophyletic pair, as did haplotypes I and N from the Colorado River Delta.

Bootstrap analysis of the ND2 data alone produced a topology (180 steps; CI = 0.90) identical to that for the combined ND2/D-loop data except that it did not support the clade containing haplotypes B and F; instead these haplotypes clustered as separate parts of the large polytomy in Figure 2 that includes the remaining Salton Sea/Colorado River Delta haplotypes. Bootstrap support for most internal nodes of the tree were greater with the ND2 data alone than with the combined ND2/D-loop data (Fig. 2).

Discussion

Reciprocal monophyly between the mtDNA haplotypes of the Río Sonoyta/Quitobaquito populations and those of the Salton Sea/Colorado River Delta suggests long, mutually exclusive evolutionary histories (Neigel and Avise, 1986) for the two groups, a hypothesis that is consistent with geological history. Isolation of the ancestral Río Sonoyta/Quitobaquito drainage from the Colorado River Delta probably occurred sometime in the last 100,000 years when lava beds resulting from eruptions of the Sierra Pinacate Volcanic Field diverted the flow of the Río Sonoyta southward to the Gulf of California and away from its previously westward course toward the Colorado River Delta (Ives, 1964; Donnelly, 1974; Turner, 1983). More recent separation of the Río Sonoyta and Quitobaquito Springs populations would explain the lack of a significant difference in haplotype frequencies between our samples from these areas.

Geological history and the mtDNA genealogy indicate that the desert pupfish complex comprises two separate entities that have been diverging for perhaps 100,000 years. They are diagnosable on the basis of two independent characters, mtDNA and male breeding colors, and we believe they should be treated as separate species, the Quitobaquito pupfish, C. eremus Miller and Fuiman, in Río Sonoyta and the Quitobaquito Springs area, and the desert pupfish C. macularius in the remainder of the range of the desert pupfish complex. The two groups clearly qualify as species under both the evolutionary species concept (Wiley, 1978; Frost and Hillis, 1990) and the various forms (Mayden and Wood, 1995) of the phylogenetic species concept. Analyses of morphometrics, meristics, and protein electrophoretic variation show no fixed or otherwise marked differences between the two groups (Miller, 1943; Turner, 1983; Miller and Fuiman, 1987). At the same time, those studies reveal no patterns of geographic variation that are discordant with the grouping based on mtDNA and male breeding colors. Ranking C. eremus and C. macularius as species under the phylogenetic species concept avoids ambiguity regarding their ontological status (Cracraft, 1983, 1989). Treating them as subspecies would connote the possibility that, like many recognized subspecies, they are arbitrarily defined parts of a taxon rather than discrete entities existing independently of human concepts.

Recent speciation probably explains the lack of fixed allelic differences for eight polymorphic proteins (Turner, 1983) and the low RFLP estimate of overall mtDNA sequence-divergence (0.34%; Echelle and Dowling, 1992) between C. macularius and C. eremus. The level of mtDNA divergence is similar to that (0.16–0.64%, x̄ = 0.42) within a monophyletic group of pupfish haplotypes in which those of C. nevadensis are paraphyletic (Echelle and Dowling, 1992; Duvernell and Turner, 1998) with respect to the Devils Hole Pupfish, C. diabolis, a morphologically divergent species, which, as Miller (1981) suggested, may have evolved within the last 20,000 years. The species flock of five pupfishes in Lake Chichancanab on Mexico's Yucatan Peninsula provides another example of rapid morphological divergence with negligible divergence in allele frequencies for protein-encoding genes (Humphries, 1984) and a lack of reciprocal mtDNA monophyly (Strecker et al., 1996). In this example, speciation and striking morphological divergence (Humphries and Miller, 1981) might have occurred in as little as 8000 years (Strecker et al., 1996).

The relatively low level of morphological divergence between C. macularius and C. eremus despite levels of molecular divergence similar to, or higher than, that in the other two pupfish examples just described probably reflects mode of speciation. Vicariant speciation involving large populations is not necessarily associated with divergence in morphology or other outward expressions of adaptational differences (Smith and Todd, 1984; Dimmick et al., 1999). Rapid morphological change in C. diabolis, an example of speciation by peripheral isolation, probably reflects the unique, cavernous habitat in Devils Hole and the consistently small size of the resident population (Miller, 1950). Rapid morphological evolution in the Lake Chichancanab species flock, which represents intralacustrine (possibly sympatric) speciation, has been attributed to ecological divergence driven by competition (Humphries, 1984).

Recent gene flow probably explains the lack of significant genetic divergence among pupfish populations within the Salton Sea area and in the Colorado River Delta. Dunham and Minckley (1998) reported homogeneous allele frequencies for four polymorphic allozyme loci in their samples of C. macularius from four sites in the Salton Sea area. Samples from two captive stocks, including the DNFH stock we examined, gave indirect evidence of divergence among populations on the delta (Dunham and Minckley, 1998), but this might reflect genetic drift in captivity. The DNFH stock began with 280 founders captured in 1983 from our Santa Clara Slough site (Dunham and Minckley, 1998; their “Canal Sanchez Toboada”). In our survey, this stock had only the common haplotype of C. macularius, whereas all samples from the wild had two to five haplotypes. This suggests founder effect or subsequent genetic drift during the 15 years of captivity. Correspondingly, the population also showed reduced allozyme variation relative to other samples of the species (Dunham and Minckley, 1998; 1 polymorphic locus vs 3 to 4). In contrast with our results for the Salton Sea area and the allozyme homogeneity reported by Dunham and Minckley (1998), Turner (1983) found significant heterogeneity for four of eight polymorphic allozyme loci in samples from the area. Dunham and Minckley (1998:12) suggested that the present homogeneity may reflect “a recent dramatic increase in pupfish abundance and gene flow.”

A significant, but rather small, among-region component of mtDNA genetic diversity in C. macularius occurs between the Salton Sea area and the Colorado River Delta. Similarly, Turner (1983) and Dunham and Minckley (1998) reported allozyme divergence between wild populations in the Salton Sea area and captive stocks derived from the delta, although this might reflect genetic drift in the captive stocks. The low level of sequence divergence detected for mtDNA probably reflects the dynamic nature of the regional hydrology. Although extant populations are separated by large expanses of desert, there has been ample historic opportunity for gene flow. For example, two of the more isolated areas, the Salton Sea and Laguna Salada, are in topographic sinks that, even in recorded history, have experienced repeated cycles of flooding and desiccation as the river meandered across the deltaic plain (Carpelan, 1961). The last flow of the river into the Salton Sink occurred from 1905 to 1907 when the Colorado River breached an irrigation-canal system and created the present Salton Sea (Carpelan, 1961). Laguna Salada last received flow from the river during high flows in 1983 and 1984 that apparently were associated with dispersal and increased abundance of pupfish populations on the delta (Hendrickson and Varela, 1989).

The dynamics of populations in the Salton Sea/Colorado River Delta area include periods of “boom and bust” (Dunham and Minckley, 1998:12), with population expansion and dispersal alternating with population declines, isolation, and local extinctions. This effect is particularly pronounced for fishes in the arid regions of western North America (Minckley et al., 1986). For C. macularius, decline, isolation, and extinction of populations have been greatly exaggerated by anthropogenic environmental alterations. In such instances, populations should be managed to simulate historic patterns of dispersal (Meffe and Vrijenhoek, 1988), with periodic intermixing and, when conditions are favorable, transfers into areas of extirpation. Dunham and Minckley (1998) also recommended manipulation of local habitats to provide the harsh, unstable conditions that would favor pupfish over the usually less tolerant nonnative species.

The significant among-region component of variation in C. macularius suggests that the Salton Sea and the Colorado River Delta regions should be managed as separate units. Although the same mtDNA haplotype was common throughout both regions, the distribution of less common haplotypes indicates a lack of wholesale intermixing. Under natural conditions in the past, brief periods of surface-water connection between Salton Sea and the delta were separated by decades or longer with no connection (Sykes, 1937). A conservative management approach would avoid intermixing pupfish between the two regions beyond what occurs when the present human-regulated hydrology fails under the forces of nature. Within these regions, neither the available genetic data nor the morphological analysis by Miller and Fuiman (1987) provide a basis for partitioning populations into separate conservation units. A similar conclusion might be reached for C. eremus based on mtDNA variation alone, but the potentially long history of isolation between these two populations and the evidence of some degree of morphological divergence (Miller and Fuiman, 1987) indicate a need for conservative management with no artificial intermixing of the populations in Quitobaquito Springs and Río Sonoyta.