Samples.—Most plagopterin taxa now occupy limited ranges (Fig. 1) in the lower Colorado River system. Lepidomeda albivallis and Lepidomeda mollispinis pratensis are restricted to single localities and L. m. mollispinis, L. vittata, and Plagopterus argentissimus are found only in single drainages. Previous studies of variation in L. vittata indicated limited divergence among localities (Tibbets, 1998; Tibbets et al., 2001); therefore, single individuals of Lepidomeda mollispinis mollispinis, L. vittata, and Plagopterus argentissimus were used to represent each. Previous studies of Meda fulgida (Anderson and Hendrickson, 1994; Tibbets and Dowling, 1996) identified divergent populations in the Gila and Verde Rivers; therefore each was represented. Because of the widespread distribution and potential paraphyly of Snyderichthys to the plagopterins, multiple samples were also examined (Bonneville basin, Bear and Snake River drainages).

Representatives of the three major clades (western, shiner, and chub) of Coburn and Cavender (1992) were also examined, with special emphasis on hypothesized sister taxa to plagopterins. These names will be used throughout manuscript. Chubs were Campostoma anomalum, Couesius plumbeus, Hybognathus hankinsoni, Margariscus margarita, Nocomis micropogon, and Semotilus atromaculatus. Shiners included Cyprinella spiloptera, Luxilus chrysocephalus, and Pimephales notatus. Representatives of the western clade were Acrocheilus alutaceus, Agosia chrysogaster, Gila cypha, Moapa coriacea, Orthodon microlepidotum, and Rhinichthys atratulus. Based on relationships inferred from morphological characters (Cavender and Coburn, 1992; Coburn and Cavender, 1992), sequences from the leuciscins Abramis brama (Briolay et al., 1998; GenBank accession nunber Y10441) and Notemigonus crysoleucas (Schmidt and Gold, 1995; GenBank accession number U01318) were used as outgroups.

Sequencing.—Sequences were obtained from PCR products by either standard dideoxy sequencing or cycle-sequencing and deposited in GenBank (accession numbers in Material Examined). In the standard method, templates were obtained through two rounds of amplification. In the first, a double-stranded product was generated as follows: 1–5 μl (approximately 1–5 ng) of genomic DNA; 0.25 units of Taq DNA polymerase [from Promega Corp. or purified from a clone as described by Pluthero (1993)]; 4 μl of dNTP mix (final concentration of 200 μM of each of the four dNTPs); 2.5 μl MgCl2 stock (final concentration of 2.5 mM); 2.5 μl of 10× buffer stock [provided by the enzyme supplier or as described in Pluthero (1993)]; 1.25 μl of each primer (final concentration of 0.5 μM of each); sterile-distilled water to a final volume of 25 μl. Amplification was performed by 20 cycles of denaturation at 94 C for 1 min, annealing at 48 C for 1 min, and extension at 72 C for 2 min.

Single-stranded DNA for sequencing was generated by using 10 μl of 1:100 dilution of this double-stranded PCR product. Conditions for amplification were identical except that one of the primers was diluted (1:100) and annealed at 52 C. Single-stranded products were cleaned by centrifugation (Millipore Corp.), as indicated by the supplier, and sequenced using the Sequenase 2.0 kit (U.S. Biochemical) as described in Dowling and Naylor (1997). Alternatively sequences were obtained from cleaned, double-stranded PCR products by cycle sequencing, following recommendations of the manufacturer (Perkin-Elmer), or by automated sequencing of dye-labeled products (using an ABI377).

Most primers (Table 1) were designed by T. R. Schmidt from a series of sequences from eastern North American cyprinids (described in Schmidt and Gold, 1993); however, it was necessary to redesign some internal primers from sequences of several other minnows obtained with external primers. External primers (LA and HA) were positioned in conserved flanking tRNA genes. Because of its large size, sequences for the entire gene were obtained from two separate sets of primers (LA-HD and LD, LD_IB, or LD_LUX-HA, respectively, Table 1), producing fragments approximately 600 bp in length. The complete sequence of each fragment was obtained using two sets of primers, the limiting external primer in the asymmetric amplification reaction and an internal primer from the same strand (Table 1).

Data analysis.—Sequences were aligned by eye using the homologous sequence of Cyprinus carpio (Chang et al., 1994) as reference. Although this can be a problem for noncoding sequences, the alignment attained for the protein-coding sequence was unambiguous. Phylogenetic analysis was performed using PAUP4.0b8w (D. L. Swofford, Sinauer Associates, Sunderland, MA, 1998, unpubl.). Topologies were obtained by heuristic search using the tree bisection-reconnection (TBR) method with 50 random-addition sequences and uninformative characters excluded. Relative strength of nodes was determined by completion of 1000 bootstrap replicates (Felsenstein, 1985; Hillis and Bull, 1993) in PAUP, using the fast-search option. Neighbor-joining trees were also generated with PAUP, using Tamura-Nei distances corrected for rate variation using the alpha value (alpha = 0.3) estimated by maximum likelihood from a larger study of cytb variation in western cyprinids (Smith et al., in press). Relative strength of nodes was determined by completion of 1000 bootstrap replicates in PAUP.

Estimation of divergence times.—Rates of molecular evolution were estimated from match of fossil to molecular data in the context of the molecular phylogeny (Table 2). This method permits calculation of estimated rates of divergence, which then enable estimates of divergence times for biogeographic units. Numerators in the rate equation are pairwise Tamura-Nei distances expressed as percentage sequence divergence of cytb. Denominators in the rate equation are estimated times (in millions of years = Ma) of divergence for those taxa. The fossil record provides estimates of times of divergence as inferred from oldest fossil occurrences of diagnostic apomorphies of lineages (Smith et al., in press). Regression of pairwise percentage divergence against time of first appearance in the fossil record yields an estimated rate of divergence for cytb in cyprinids (Fig. 2). When the intercept is constrained to pass through the origin (Hillis et al., 1996), this estimate is 1.31% sequence divergence per pairwise comparison per Ma (or 0.66% per lineage per Ma), with an error of 10% to 25% (Smith et al., in press). This assumes no genetic divergence at the time of branching and unbiased estimates of the denominator. This method probably overestimates the rates of molecular evolution because the first appearance of fossil taxa are almost surely underestimated but has the advantage of being based on a sample of values for numerators and denominators, rather than a single assumption. If, however, estimates of ages of ancestors are biased toward underestimation, the regression should not be constrained, producing a line with a positive intercept and shallower slope of 1.05% per million years (or 0.53% per lineage per million years).

Constancy of evolutionary rates was examined using the relative rates test as implemented in the program PHYLTEST (vers. 2.0, S. Kumar, Pennsylvania State University, University Park, PA, 1996). Estimates of divergence time were corrected for variation in rates of evolution among lineages by calculating branch lengths for each lineage using MEGA2 (S. Kumar, K. Tamura, I. B. Jakobsen, and M. Nei, Arizona State University, Tempe, AZ, 2001). Because it is difficult to determine whether constrained or unconstrained estimates are more appropriate, minimum and maximum divergence estimates were derived using the constrained (0.66% per Ma) and unconstrained calibrations (0.53% per Ma).

Phylogenetic position of the Plagopterini.—Sequences of the cytb gene indicate that Snyderichthys is a member of Plagopterini as suggested by DeMarais (1992) and Simons and Mayden (1997). Monophyly was strongly supported, with bootstrap values ≥ 80% in all analyses.

These results are also consistent with the conclusion of Simons and Mayden (1997) that plagopterins are related to Margariscus, Couesius, and perhaps Semotilus. When using unweighted and weighted parsimony, plagopterins fall within the monophyletic creek chub clade, with Margariscus their closest relative. In the neighbor-joining topology, a group consisting of Couesius and Margariscus was the sister lineage to plagopterins; however, the creek chub clade (sensu Simons and Mayden) was not monophyletic as Semotilus was the most divergent North American cyprinid. The three results are similar, however, and consistent with results of Simons and Mayden (1997) using 12S rDNA. Unfortunately, neither study provided strong bootstrap support, indicating that additional data are necessary to further test these relationships. A close relationship of the plagopterins with Margariscus and Couesius is supported by biogeography. The present distribution of these northern genera complements that of the plagopterins near the Continental Divide in western Wyoming (Fig. 1).

Relationships within the Plagopterini.—Studies of allozyme variation (DeMarais, 1992) supported a close relationship of Snyderichthys to plagopterins, rendering them polyphyletic. The examination by Simons and Mayden (1997) indicated Snyderichthys was more closely related to Lepidomeda vittata than to Meda fulgida. These relationships, however, were not strongly supported by bootstrap analysis of weighted characters as the Snyderichthys + L. vittata node occurred in 64% of bootstrap replicates while monophyly of plagopterins was indicated in 58% of replicates.

In this analysis, sequence variation at cytb produced highly resolved relationships supported by high bootstrap values. In all analyses, Lepidomeda [L. m. mollispinis and L. m. pratensis from the Virgin River drainage and L. albivallis from the adjacent pluvial White River, a Pleistocene-Holocene tributary to Virgin River (Fig. 1)] formed a trichotomy most closely related to Snake River samples of Snyderichthys. The sister species to this group is L. vittata of the Little Colorado River. Snyderichthys from the central and southern Bonneville basin are the most divergent.

Although previous studies found Snyderichthys closely affiliated with Lepidomeda, none had sufficient geographic samples to identify polyphyly of the species and existence of two forms of Snyderichthys. As reported here, analyses of several additional populations from several drainages (Johnson and Jordan, 2000) showed differentiation of northern and southern Snyderichthys. Previous studies of life-history variation (Johnson et al., 1995, and references therein) identified differences in spawning time between Main Creek (Bonneville basin) and Sulphur Creek, Wyoming (Bear River), providing further evidence for differentiation. Given this, Snyderichthys should be considered a possible synonym of the genus Lepidomeda and possibly divided into two species. As yet, however, we have not found diagnostic morphological characters to differentiate S. copei of the upper Snake and Bear Rivers (type locality “tributary of Bear River at Evanston, Wyoming,” Jordan and Gilbert, 1881) and what may be S. aliciae (Jouy, 1881; type locality Provo River at Utah Lake).

Biogeographic implications.—The broader relationships hypothesized here corroborate patterns derived from allozymes, morphology, and karyology (DeMarais, 1992; Miller and Hubbs, 1960; Uyeno and Miller, 1973). Divergence times were estimated with the method of Smith et al. (in press) using the neighbor-joining tree (Fig. 4), allowing comparison of biogeographic hypotheses and geologic events. Tests for variation in rates of evolution among nodes of interest were conducted, with only the deepest node within the plagopterin lineage (Lepidomeda + Snyderichthys and Plagopterus + Meda) exhibiting significant deviation from equal rates. Although variation in rates among lineages may cause estimates of divergence from the mean and standard errors to be misleading, the shortest and longest branches from the common ancestor were used to provide minimum and maximum estimates of divergence time, respectively (S. Kumar, pers. comm.).

The Meda + Plagopterus, Lepidomeda, and Snyderichthys clades last shared a common ancestor 17 to 40 Ma (Table 4), corresponding to or predating the initial formation of the southern Colorado Plateau margin 22 to 25 Ma (Potochnik, 1989) and the beginning of extensional faulting of adjacent terrains > 20 Ma (Dokka and Ross, 1995) to form the Basin-and-Range Physiographic Province. Divergence of Meda and Plagopterus also was early, 19 to 34 Ma (Table 4), and may have occurred when the Gila and Colorado paleorivers (Howard, 1996) were isolated from one another, or associated with any one of the numerous closed basins that were formed south of the Plateau (Nations et al., 1982, 1985; Peirce, 1987). Divergence of Meda populations in the Verde and upper Gila basins occurred 0.9 to 3.4 Ma. This separation might be attributable to habitat isolation because of a proclivity of the genus for hard-bottomed streams, separated by the sand-bottomed, erosive nature of the mainstem lower Gila River. Alternatively, it might also have resulted from isolation in endorheic drainage basins (e.g., Douglas et al., 1999).

Differentiation of the southern and central Bonneville basin Snyderichthys and L. vittata (presently in only the Little Colorado River basin) from a common progenitor, 7.8 to 11.6 Ma and 6.3 to 8.6 Ma, respectively, was also likely related to isolation through Plateau uplift and extensional collapse of Basin-and-Range. These time estimates are bracketed by estimated plateau-margin development as recent as approximately 6.0 Ma on its extreme southwest (Luchitta, 1979, 1984) and other Bonneville Basin-rim structures like the Wasatch Front (30 Ma or before; Anderson and Mehnert, 1979).

A limited time of differentiation (0.0–1.3 Ma) exists for Lepidomeda taxa of the Virgin River basin [L. m. mollispinis of the Virgin River, L. m. pratensis from Meadow Valley Wash (= pluvial Carpenter River, a recent tributary of the Virgin), and L. albivallis of pluvial White River]. Lepidomeda altivelis, an extinct White River species was in this clade as well. Similar divergence times between our middle Snake and Bear river samples (0.0–0.4 Ma) and Central and Southern Bonneville samples (0.6–0.9 Ma) of Snyderichthys from north and south of the “Old River Bed” are consistent with geological history (Oviatt et al., 1992).

Affinities of the Bear and Snake River populations to each other rather than to the central and southern Bonneville populations (Bear River is now tributary to the Bonneville basin) must reflect Middle to Late Pleistocene divergence, before Bear River's most recent connection (∼34,000 years ago; Bright, 1963) with the Bonneville system. It will be of interest to examine Snyderichthys from Wood River, Idaho. Miller (1945) found morphological differences; since Wood River enters the middle Snake, molecular characters are predicted to resemble the middle Snake River form.

Highlands dating to Early Miocene or before (Minckley et al., 1986; Drummond et al., 1993) contain the Bear River watershed that now flows into the Bonneville basin from Wyoming, Idaho, and Utah. Bear Lake basin also hosted a succession of Miocene to Recent lakes (Peale 1879; Hubbs and Miller, 1948; Miller, 1965). Three lake-adapted, endemic whitefishes (Snyder, 1919) and an endemic Cottus attest to both its antiquity and drainage relations. Exchange of Bear River between the Snake and Bonneville basins caused by block faulting and volcanism (Morrison, 1965; Stokes, 1979) are also corroborated by fossil fishes. A relative of the Bear Lake whitefishes occurs in Mio-Pliocene deposits of the Snake River Plain, southern Idaho (Smith, 1975), and two others were in the Bonneville system in Pleistocene (Smith, 1981); an endemic Cottus in Utah Lake (also Bonneville) is a possible Pleistocene relative of the Bear Lake endemic (Smith et al., 1968); and Bright (1967) reported a fossil lakesucker (Chasmistes) related to taxa in the upper Snake or Bonneville (Utah Lake) from Thatcher Basin, along Bear River's path during the most recent, Late Pleistocene diversion from the (Bear River into Lake Bonneville).

The sister relationship (1.3–7.4 Ma; Table 4) between Snyderichthys of the Bear River-Snake complex and L. mollispinis (plus L. albivallis and presumably L. altivelis) of the Virgin-pluvial White system is more difficult to rationalize. It corresponds in part with a past hypothesis of Snake-Bonneville-Virgin River drainage connections by Keyes (1917, 1918), who proposed the upper Snake River as a former Bonneville-system tributary that ultimately flowed into the lower Colorado River. Geologists did not take Keyes seriously (Stokes, 1979). Hubbs and Miller (1948:31) rejected his proposal and provided biogeographic evidence against it in a later monograph (Hubbs et al., 1974) on fishes of western Nevada. Ives (1948) identified a potential connection from the Escalante Basin of Lake Bonneville southward through Meadow Valley Wash (hence to Virgin River). Smith (1966) and Smith and Koehn (1971) found morphologic and genetic evidence for such communication. Taylor (1985) proposed a Late Miocene or Pliocene connection based on molluscan distributions from upper Snake River through the Bonneville area and pluvial White River to the lower Colorado basin.

Molecular similarity between Virgin River Lepidomeda and Bear-Upper Snake River Snyderichthys suggests connections between these drainages about 5 to 6 Ma. At this time, the Salt Lake Formation in northern Utah and southern Idaho was separate from the Snake River drainage on the Western Snake River Plain (Taylor, 1985). The Snake River was connected to the Sacramento (Wheeler and Cook, 1954; K. R. Fecht, S. P. Reidel, and A. M. Talklman, 1982, unpubl.) and Klamath drainages (Taylor, 1960; Smith et al., 2000). Based on molecular data linking northern Snyderichthys to Virgin River Lepidomeda, we hypothesize connections following structural alignments of basin-and-range topography in the vicinity of the Bonneville Basin, which was not lacustrine at the time (G. K. Gilbert, Lake Bonneville, U.S. Geological Survey V, 1890, unpubl.).

If this pattern is the consequence of vicariance, the progenitor of Lepidomeda and Snyderichthys was distributed throughout the Lower Colorado and Upper Snake Rivers around 12 Ma and survived to differentiate as relicts of tectonism disrupting a continuous distribution. The populations now represented in the Bonneville (Snyderichthys) and Little Colorado (L. vittata) were connected until 6.3 to 8.6 Ma. More recently diverged forms (e.g., L. m. mollispinus, L. m. pratensis, L. albivallis, L. altivelis) likely resulted from disruption of water connections through recent aridification of the region (Axelrod, 1979; Winograd et al., 1992). These temporal relationships demonstrate the importance of tectonism and reliction in speciation of western fishes.

If the vicariance hypothesis is correct, Snyderichthys will become a synonym of Lepidomeda and the southern form will be recognized as a separate species, possibly L. aliciae (Jouy). At this time, however, we must also consider two other possibilities. Polyphyly of Snyderichthys mtDNA lineages could also result from stochastic processes associated with the sorting of different lineages (Neigel and Avise, 1986); however, the high levels of sequence divergence among them would make this alternative less likely. A possible alternative could be introgressive transfer of mtDNA among lineages. Extensive transfer of mtDNA among species is common in cyprinids (Dowling et al., 1989; Dowling and Hoeh, 1991) including species in these basins (Dowling and DeMarais, 1993; Gerber et al., 2001). Broader sampling and analysis of additional independent loci is necessary to evaluate these alternatives.

Material examined.—General locality information for ingroup sequences used in this study. Precise information available upon request to TED. Sequences generated for this study were deposited in Genbank under the accession numbers AF452072–AF452094. Genbank numbers for sequences from other studies are provided in parentheses. Acrocheilus alutaceus, OR, Maries River drainage, Benton County; Agosia chrysogaster, AZ, Gila River drainage, Greenlee County; Campostoma anomalum, IN, Illinois River drainage, LaPorte County; Couesius plumbeus, AK, Yukon River drainage, North Star Borough; Cyprinella spiloptera, NY, Susquehanna River drainage, Tioga County (U66605); Gila cypha, AZ, Colorado River drainage, Coconino County; Hybognathus hankinsoni, MI, Lake Michigan drainage, Charlevoix County; Lepidomeda albivallis, NV, White River drainage, Nye County; Lepidomeda mollispinis mollispinis, AZ, Virgin River drainage, Mohave County; Lepidomeda mollispinis pratensis, NV, Meadow Valley Wash, Lincoln County; Lepidomeda vittata, AZ, Chevelon Creek, Little Colorado River drainage, Coconino County; Luxilus chrysocephalus, TN, Tennessee River drainage, Blount or Hancock Counties; (U66595); Margariscus margarita, MI, Charlevoix County, Lake Michigan drainage; Meda fulgida, AZ, Gila River drainage, Graham Co., and AZ, Verde River drainage, Yavapai County; Moapa coriacea, NV, Moapa River drainage, Clark County; Nocomis micropogon, MI, Raisin River drainage, Washtenaw County; Orthodon microlepidotum, CA, King's River drainage, Fresno County; Pimephales notatus, MI, Raisin River drainage, Washtenaw County (U66606); Plagopterus argentissimus, UT, Virgin River drainage, Washington County; Rhinichthys atratulus, MI, Rouge River drainage, Macomb County; Semotilus atromaculatus, MI, Raisin River drainage, Washtenaw Co.; Snyderichthys copei, ID, Goose Creek drainage, Cassia County; WY, Bear River drainage, Uinta County; UT, Sevier River drainage, Garfield County; and UT, Spanish Fork River drainage, Utah County.