Table 1 summarizes indices of genetic diversity. With the exception of cyt b for Peru, haplotype and nucleotide diversity for the cyt b and control region was high. The lower diversity of cyt b in Peru was first described by Cassens et al. 14 and was attributed to high levels of mortality along the Peruvian coastline.

The dusky dolphin d-loop and cyt b haplotype networks recovered a Peru-specific lineage divergent from other geographic regions by several missing intermediate haplotypes (Figures 2 and 3). Similarly, the Bayesian analysis recovered a monophyletic clade of Peruvian haplotypes with 100% posterior probability, and generally supported the phylogenetic pattern of the statistical parsimony networks (Figure 2). The Bayesian analysis indicated with 96% posterior probability that the root of the dusky dolphin phylogeny was not the Peruvian clade, but rather occurred along a lineage of South American and New Zealand haplotypes (Figure 2).

For each of the mitochondrial and nuclear genetic markers, both an AMOVA and F_st values revealed significant partitioning of variation among Peru, Argentina, New Zealand, and South Africa (Table 2). The high levels of divergence reflect the very low number of haplotypes found in more than one geographical region. For example, only 3 cyt b (A1.2, A1.1, and DLPH; Figure 2) and d-loop (A20, AR10, A18; Figure 3) haplotypes occurred in >1 geographic region. In no instance were Peruvian mitochondrial haplotypes shared with any other locality. Thus, the highest F_st values were between Peru and other localities, with the level of divergence between South Africa and Argentina considerably lower than either was to New Zealand (Table 2). Despite the high degree of divergence between geographic regions, no apparent pattern of a linear, easterly increase in divergence was evident.

The geographic distribution of haplotypes within the statistical parsimony networks revealed a genetic signature inconsistent with a linear dispersal route. In both cyt b (Figure 2) and control region (Figure 3) networks, New Zealand lineages were notably separated from the Peruvian clade by a mixture of haplotypes from South Africa, Argentina, New Zealand, and several missing haplotypes. With the exception of Peru, at least one haplotype at the distal tip of each central haplotype group was from New Zealand. Both mitochondrial networks lacked a central haplotype in high frequency, with the majority of most frequent haplotypes clustered at the tips of the networks. The actin network (Figure 4) was dominated by three common alleles (a, b, l) at the center of the network present. These alleles varied in frequency in all four biogeographic regions and in the sister taxon, L. obliquidens. In 3 of the 5 tip clades, alleles were common to South Africa and New Zealand. The tips of the network contained alleles present in lower frequency and with more restricted geographic distributions. In one instance an allele from Argentina was placed at the periphery of the network (haplotype 'e'), and this allele was also sampled in New Zealand.

The MIGRATE analysis produced estimates of the amount of migration between geographical regions that suggests multiple east to west migrations dominated by dispersal from New Zealand to South Africa and Argentina and an absence of migration from Peru to other regions (Figure 5). The relative temporal shift in migration rates is evident in the comparison of the independent analyses of cyt b and the d-loop. The analysis of the cyt b gene suggests early high levels of migration from New Zealand to South Africa and Argentina with weak or no migration from South Africa or Argentina to Peru. The d-loop suggests a reduction in the rate of migration from New Zealand to South Africa and Argentina when compared to cyt b. In the estimates from the d-loop a weak, but significant, migration is observed from Argentina to South Africa and New Zealand. This was not represented in the cyt b results. Overall, these results support: 1) early and continued isolation of Peru from other regions; 2) at least 1, and possibly 2, periods of migration from New Zealand to South Africa, with the first migration period represented by cyt b suggesting a high, unidirectional migration rate; 3) the later onset of migration from Argentina to South Africa representing a possible 3^rd period of migration into South Africa; 4) a high rate of unidirectional migration from New Zealand to Argentina correlated with that of migration to South Africa, suggesting an eastward dispersal event from New Zealand, and 5) followed by a reduced migration rate between New Zealand, Argentina, and South Africa, as would be expected of populations that gradually became genetically isolated.

Divergence times calculated from cyt b data indicated a divergence between L. obscurus and L. obliquidens of 1.9 [95% CI: 1.3–2.9] million years ago (mya), and the Peruvian lineage separated from other geographic regions 0.63 [95% CI: 0.38–0.98] mya.

The Bayesian analysis recovered a phylogeny (not shown) that supported previously established relationships between New World species of Engraulis with two major lineages, a "New World" clade with California (E. mordax) and a monophyletic South American (E. ringens/E. anchoita) lineage in which E. ringens and E. anchoita are reciprocally monophyletic, and an "Old World" clade comprised of a polyphyletic assemblage of E. australis, E. encrasicolus, E. capensis, and E. japonicus 650. Our results suggest a divergence between E. mordax and South American species at 9.0 [95% CI: 6.9–10.0] mya, with E. ringens and E. anchoita diverging at 1.6 [95% CI: 1.0–2.5] mya.

The parsimony network of anchovy cyt b sequences displayed similar geographical patterns in the distribution of genetic diversity as that of the dusky dolphin (Figure 6).

The overall pattern of genetic variation observed for the dusky dolphin in the Southern Hemisphere is not congruent with the west-wind drift hypothesis. Bayesian analysis of mitochondrial DNA sequences suggests the dusky dolphin diverged from L. obliquidens via a north to south equatorial transgression approximately 2 million years ago, with confidence intervals overlapping previously published estimates. However, the root of the dusky dolphin dispersal in the Southern Hemisphere was not Peru as previously proposed 1319, but rather somewhere in the southern South Pacific/Indian Ocean basins. This result is further supported by the divergence of Peru from other regional populations relatively recently, and considerably later than the L. obliquidens/L. obscurus divergence 2.0 million years ago (Figure 7). There are several lines of evidence from the distribution of genetic variation that support a larger, South Pacific/Atlantic Ocean population as the source of dusky dolphin lineages. First, the presence of haplotypes from New Zealand at both the center and the periphery of the actin network suggest an old New Zealand population. Second, the central cluster of haplotypes from New Zealand, South Africa, and Argentina in low frequencies at the center of the mitochondrial DNA haplotype networks (Figures 2 and 3) is indicative of the remnants of a historical population with a large geographical distribution that did not include Peru. Third, there are greater levels of genetic divergence between New Zealand and other regions, and the lowest levels of divergence between South Africa and Argentina (Table 2). This pattern of genetic divergence is consistent with an early transgression to the South Pacific/Indian Ocean followed by expansion into the Atlantic.

Given that Peru is not the probable root of dusky dolphin origin, there are two alternative scenarios to explain the marked isolation of Peru from other geographical regions, both of which support a South Pacific/Indian Ocean origin. First, the north to south transgression during the Plio-Pleistocene resulted in the formation of a large population of dusky dolphins that eventually extended from New Zealand to Australia, South Africa, Argentina, and dispersed through Drake's passage between the Antarctic Peninsula and Tierra del Fuego into Peru. There is evidence that glacial cycles had profound effects on the availability of a temperate dispersal corridor for the dusky dolphin from Peru into the Atlantic, both through the Strait of Magellan and Drake's Passage 20212223. These alterations in the availability of a dispersal route may have resulted in the isolation of a Peruvian population from other regions, and may explain the contemporary gap in the distribution of the dusky dolphin along the Peruvian coast from the Strait of Magellan to southern Tierra del Fuego (Figure 1). Second, equatorial transgression led to a large, cohesive population of dusky dolphins that extended from Peru across the southern Pacific and Indian Oceans. Changes in migration patterns across the southern Pacific, coupled with restricted dispersal through Drake's passage and the Strait of Magellan resulted in the early isolation of Peru from other geographic regions. Although our data are insufficient to discriminate between these alternative hypotheses, neither is consistent with a Peruvian origin.

Our analysis of the phylogeography of the dusky dolphin does not support a linear, eastward dispersal route predicted by the west-wind drift hypothesis. Our data do not show a pattern of isolation-by-distance from Peru to New Zealand. To the contrary, historical migration patterns suggest an alternative hypothesis of a founding population in the South Pacific/Indian Ocean that expanded its geographic range to the Atlantic via a temperate dispersal corridor (Figure 5). Evidence from analysis of cyt b and d-loop sequences suggests that the migration rate per generation from New Zealand westward was highest relatively early, followed by continued, but reduced, westward migration later in history. This reduction in westward migration was accompanied by weak, but potentially significant, migration from Argentina eastward to South Africa and New Zealand (Table 3, Figure 5). This hypothesis receives further support from the occurrence of very few haplotypes in more than one geographic region, and the relatively large number of missing intermediates between haplotypes typical of an old population subject to the effects of genetic drift. In the Southern Hemisphere, eastern South America, South Africa, and New Zealand are loosely connected by a series of small islands within a band of temperate sea surface temperatures (Figure 1). These islands are used at least periodically by the dusky dolphin 924 and likely served as stepping-stones for reciprocal genetic exchange across the South Pacific and Indian Oceans. Although the dusky dolphin may have at one time dispersed via oceanic islands, a recent study of microsatellite variation between South Africa and Argentina found no evidence for contemporary dispersal of dusky dolphins between these regions 14. The large cluster of New Zealand-specific haplotypes at the periphery of the cyt b and control region networks (Figures 2 and 3), with large, mixed clusters of Argentina/South Africa haplotypes is indicative of the slow decay of contact between geographical regions over time with increased genetic distance. Overall, this suggests that the observed pattern of genetic diversity in contemporary dusky dolphin populations is the relic of a historically large population that has since become isolated into 4 regional populations.

The parsimony network of anchovy cyt b sequences displayed similar geographical patterns in the distribution of genetic diversity as that of the dusky dolphin (Figure 6). Most notable similarities were: (1) marked divergence between South American haplotypes from those in the Atlantic and Indian Oceans, and (2) north-south equatorial transgressions in the Pacific Ocean with (3) a putative ancestor in the North Pacific. The correlation between the patterns of divergence between the dusky dolphin and anchovy species are likely related to paleoceanographic conditions during the Pleistocene (Figure 7). In particular, the date of L. obliquidens/L. obscurus divergence is correlated with the cooling and subsequent warming of sea surface temperatures in the EEP, with the point estimate of equatorial transgression occurring toward the end of the cooling period when primary productivity was well beyond its peak (Figure 7). The extension of nutrient rich waters in the EEP likely offered an arena for the expansion of anchovy in the EEP northward towards the equator into the tropics, and coincidentally, the cooling of tropical waters, combined with increased prey distribution and abundance, may have provided a natural conduit for the transgression of L. obliquidens from the Atlantic into the Pacific. The warming of sea surface temperatures and declining productivity in the EEP coincides not only with the divergence of L. obliquidens/L. obscurus, but also with the divergence of E. ringens and E. anchoita (Figure 7). This suggests the warming of the EEP and subsequent decline in productivity narrowed the distribution of E. ringens, which isolated eastern and western South American populations, and consequently resulted in an anti-tropical division of dolphin populations. Based on our study, this founder dolphin population took refuge in the cooler South Pacific/Indian Ocean.

Approximately 1 million years after the equatorial transgression into the Southern Hemisphere, there is evidence for a relatively early founding of L. obscurus in New Zealand followed by migration westward to Argentina and South Africa (Table 3, Figure 5). In addition to the evidence from the distribution of genetic diversity, these migration events support a diffusion of the dusky dolphin through the Indian Ocean into the Atlantic, followed by colonization and expansion of populations in coastal upwelling regions. There are several inferences from the history of anchovy 6 that are correlated with the phylogeographic pattern observed in the dusky dolphin. First, anchovy species are extremely sensitive to variation in sea surface temperatures and current systems, and this has been especially true in South Africa and Australia where climate change has had a very dynamic history 6. As a result, South African and Australian species have suffered multiple extinctions, bottlenecks, and large-scale introgressions from other regions during Plio-Pleistocene climate change. The most recent colonization of Australia was most likely via equatorial transgression from Japan. Second, there is evidence for two Pleistocene dispersals across the Indian Ocean, one at 1.5 and the other 1.2 million years ago, between Europe (Atlantic) and Japan (west Pacific) via Australia and South Africa (Atlantic/Indian Oceans) "stepping stones" 6. Similar to anchovy, our data suggest that South Africa experienced at least 2, if not 3, independent dusky dolphin colonization events. If we assume that the cyt b and d-loop provide some insight into the chronology of dusky dolphin dispersal events, our data suggest dusky dolphin migration events from New Zealand and Argentina to South Africa (Figure 5) potentially coincide with the two anchovy dispersal events. The first was from New Zealand (Figure 5), followed by a second weaker migration event from Argentina and perhaps a third from New Zealand (Figure 5). These colonization events are further supported by a lack of, or weak, migration to or from South Africa during multiple periods (Table 3, Figure 5), suggesting that dusky dolphins, similar to anchovy, were extirpated from South Africa perhaps due to similar response to environmental change.

DNA sequence data were derived from tissues of L. obscurus obtained from either skin swabs of living animals or post-mortem biopsies of beach-cast animals, or from GenBank accessions (Table 4). For tissue samples, genomic DNA was isolated with either a standard phenol-chloroform techniques or with a DNeasy Kit (Qiagen, Valencia, CA). DNA from the same individual was used to amplify and sequence all gene regions when possible. Tissues from Peru were not available, but sequences were obtained from GenBank accessions (Table 4). The complete mitochondrial cytochrome b (cyt b) gene (1040 nucleotides), 474 base pairs (bp) of the mitochondrial control region, and 995 bp of nuclear DNA (nDNA) intron I of the muscle actin gene (actin) were amplified with PCR. External primer sets included: (1) cyt b-5'-GAAAAACCAYCGTTGTWATTCAACT-3' and 5'GTTTAATTAGAATYTYAGCTTTGGG3'; (2) control region – tPro and Dlp5 25; (3) Actin- 5'-GATTTGGTCCCTCTATGTCTCT-3' and 5'-TACTTTTGAACTTGCCACCTAC-3'. PCR reactions included the following: 5 μl of PCR buffer, 25 mM MgCl, 10 mM dNTP's (2.5 mM each), 1 μl 10 mg/ml bovine serum albumin (BSA), 1 μl 10 uM primer, and 5 U of Taq. Conditions of the PCR were 94°C, 2 min followed by 35 cycles at 92°C, 30 s, annealing 30 s, and extension at 72°C, 30 s. Published annealing temperatures were used except: 765F/766R, 50°C; LagActin1/2, 58°C. PCR products were purified with QIAquick PCR purification kits (Qiagen, Valencia, CA). Amplicons were sequenced with ABI (Applied Biosystems, Foster City, CA) BigDye Terminator chemistry, and products were run on an ABI 377. Internal primers cyt b (560, 5'-GCAACCCTAACACGATTCTTCG-3'; 610, 5'-CCAGTTTCGTGTAGGAATAATAGG-3') and actin (Act5-L, 5'-CCACTACTTTAGGCAG-3'; M13Act5R-H, 5'-TGTAAAACGACGGCCAGTCTGCCTAAACTAGTGG-3' were used to sequence fragments to resolve ambiguities.

Sequenced fragments were edited and compiled with Sequencher v. 4.1 (Gene Codes Corporation, Ann Arbor, Michigan). A consensus of sense and antisense strands for each individual sequence and data partition were compiled and exported to MacClade vs. 4.05 26. Amino acid translations of the open reading frames of cyt b were examined for stop codons to verify sequence orthology. Neither actin nor cyt b contained any insertions or deletions. The control region had a single indel that was accommodated in the alignment with a minor alignment adjustment. Actin sequences were subjected to a BLAST search to verify sequence orthology. In all BLAST searches, amplicons retrieved sequences from other delphinid or mammalian taxa as the closest match, providing evidence for successful amplification of the target locus.

Heterozygous nucleotide positions of the actin intron were determined objectively with the program Mutation Surveyor (SoftGenetics, State College, Pennsylvania). This software identifies with 99% accuracy heterozygous nucleotide positions, by employing an algorithm that incorporates measures of peak height, peak intensity, and background noise from ABI electrophereograms. Putative heterozygous positions identified with Mutation Surveyor were cross-verified by examination of raw electrophereogram data in Sequencher. Only cross-verified sites were used in subsequent determination of allelic variation at the locus. Allele and genotype frequencies of actin were determined statistically with the program PHASE version 2.1 2728. Three independent chains of 1000 interactions each were performed with other parameters as default. Only those alleles and genotypes resolved with > 90% posterior probabilities were used in subsequent analyses.

Arlequin version 2.00 29 was used to estimate haplotype diversity (i.e., gene diversity) (h) and nucleotide diversity (π) 30. Standard error of these estimates was determined from a null distribution generated from 10000 random permutations of the data. DnaSP version 4.0 31 was used to determine the number of polymorphic sites (S) from aligned sequence data.

Modeltest version 3.06 32 was used to test 54 nested hypotheses of nucleotide substitution for each data partition. The Hasegawa, Kishino, and Yano 33 (HKY85) model with a correction for rate variation among sites (α = 0.66) and a proportion of invariant sites (I = 0.39) was selected as the best fit model for the mtDNA control region. For the cyt b gene, the model selected was HKY85 with a correction for rate variation among sites (α = 0.003). The actin intron also had HKY85 as the best-fit model, but with equal rates among sites. These best-fit models were used in all analyses that employed a model of evolution.

A Bayesian analysis of cyt b sequences from L. obliquidens and L. obscurus (Table 4), and Lissodelphis borealis (GenBank Accessions EF093025–EF093028) as an outgroup, was performed in Mr. Bayes version 3.1 34. The purpose of this analysis was to: (1) determine the geographical population at the root of the dusky dolphin origin, (2) provide support for the monophyly of Peruvian population haplotypes, and (3) provide a topology for subsequent estimation of divergence dates. The Bayesian Metropolis-coupled Markov Chain Monte Carlo analysis consisted of two independent searches of 25 × 10⁶ steps (1 cold, 3 heated chains to insure adequate mixing) with the best-fit model of substitution for cyt b determined a priori with Modeltest. The program Tracer version 1.1.1 35 was used to determine convergence and the burn-in point of chains. Trees prior to convergence were eliminated and a majority-rule consensus tree of the remaining topologies was constructed in PAUP* version 4.0b10 36.

We tested the hypothesis of a unidirectional, linear dispersal route by examining the geographic partitioning of genetic variation. The method of Templeton et al. 37 as implemented in the program TCS version 1.18 38 was used to construct separate statistical parsimony networks for each of the three data partitions. The statistical parsimony method was chosen because of its ability to infer missing intermediates within the network, which is more appropriate for examining intra-specific relationships than standard phylogenetic methods that assume bifurcating lineages 3940. Furthermore, reticulations, or alternative (homoplasious) parsimonious connections, are permitted with this method, thus allowing one to test hypotheses regarding the complex historical processes that shape populations. The final network gives a visual representation of the number of mutations separating any two haplotypes (or alleles). Mitochondrial DNA haplotypes were joined to the network within 95% parsimony confidence limits 38. Because of its slower rate of coalescence, parsimony confidence limits were expanded to 90% for actin alleles. Actin alleles of L. obliquidens were included in the statistical parsimony network to examine the pattern of allelic diversity at the base of the L. obliquidens/L. obscurus divergence. Reticulations in the actin parsimony network were observed and attributed either to lack of lineage coalescence, homoplasious substitutions, or recombination among sites. Of these, recombination is the most problematic for inferring evolutionary history from nuclear loci. Therefore, the likelihood of recombination among alleles was evaluated with the program Lamarc version 1.2.2 41.

An analysis of molecular variance (AMOVA) 42 and F-statistics 43 were calculated with Arlequin, version 3.1 29 to determine amount of genetic variation within and among geographic regions for cyt b, control region, and actin data. The significance of these statistics was determined by comparison of observed values to those derived from 10000 random permutations of the data. The "case-control" option in PHASE 2.1 was used to group actin alleles by geographic region and to subsequently test the null hypothesis of random association of alleles among localities. Test for Hardy-Weinberg equilibrium of actin genotypes was performed in Arlequin, version 3.1.

The program MIGRATE 4445 was used to examine the pattern of dispersal (migration) between geographic regions. Although cyt b and d-loop are linked, the rate of evolution of each data partition is different with d-loop being the fastest evolving region of the mammalian mitochondrion. Therefore, the cyt b and d-loop data were analyzed independently in order to examine dispersal events over different relative time scales, under the assumption that the d-loop will provide evidence for the most contemporary dispersal/migration events. Each data partition was subjected to a maximum likelihood MCMC search strategy with initial theta values generated from F_st estimates, and evolutionary model parameters determined from MODELTEST. Each search was comprised of a combination of 10 short and 3 long chains with an adaptive heating scheme of 4 chains (temperatures 1.0, 1.2, 1.5, 3.0) and a swap interval of 1 in order to insure adequate mixing among chains. Ten short chains of 200,000 steps were run with a sampling increment of 20, for a total of 10,000 sampled trees per chain. A total of 3 long chains were run for 20,000,000 steps, with a sampling increment of 20 for a total of 100,000 trees sampled.

We estimated dates of divergence between L. obliquidens/L. obscurus lineages and the isolation of the L. obscurus Peruvian population from other geographical regions to determine if historical changes in sea surface temperature, primary productivity, current systems, and the phylogeography of the dusky dolphin are correlated. The Bayesian cyt b phylogeny with branch lengths was subsequently used in RHINO version 1.2 46 to estimate via maximum likelihood the date of divergence and 95% confidence intervals. A previous study of Lagenorhynchus cyt b 12 suggests the evolution of this locus is clock-like; therefore dates of divergence were calculated under the assumption of a molecular clock with substitution model parameters predetermined by Modeltest. Two calibration points were used, both strongly corroborated from the fossil record and molecular divergence estimates 47: (1) Phocoenidae/Monodontidae, 13.0 my; (2) Phocoenida/Platanistidae, 28 myr.

Published cyt b sequences (n = 58) for Old World (Engraulis capensis, E. japonicus, E. encrasicolus) and New World (E. mordax, E. ringens, E. anchoita) anchovy species were assembled from GenBank (Accessions AY923766–AY923823) 48. A Bayesian analysis was performed on a matrix of 465 base pairs, and included Gonorhynchus greyi (NC004702), Chanos chanos (NC004693), Sardiops sagax (AF472586), and S. caeruleus (AF472585) to root the phylogeny. The Bayesian Metropolis-coupled Markov Chain Monte Carlo analysis consisted of two independent searches of 2.0 × 10⁶ steps (1 cold, 3 heated chains to insure adequate mixing) with a model of substitution referenced in Grant (reference). In order to compare dates of divergence of anchovy species and the dusky dolphin populations, the resulting topology and the program RHINO 1.2 46 were used to estimate times of divergence of Northern Hemisphere E. mordax from Southern Hemisphere E. ringens/E. anchoita clade, and between western (E. ringens) and eastern (E. anchoita) South America. The oldest European anchovy fossil is 10 million years old 49, thus this date was used as a calibration point. In addition, a previously published cyt b haplotype network 50 was compared to that of the dusky dolphin in order to assess similarities in the geographic distribution of genetic variation. This network, plus additional, previously published phylogeographic inferences for anchovy (i.e., 650) were summarized and compared to the phylogeographic history of the dusky dolphin.