The cyprinodont fish Rivulus (Aplocheiloidei: Rivulidae) in Trinidad and Tobago: molecular evidence for marine dispersal, genetic isolation and local differentiation

Biological background

Rivulus hartii (Boulenger, 1890), the jumping guabine, is a euryhaline fish known from eastern Colombia, the northern coasts of Venezuela, and the Caribbean islands of Trinidad, Tobago, Margarita and Grenada (Boeseman 1960; Robins et al. 1991; Huber 1996). It is the most widespread freshwater fish in Trinidad (Kenny 1995), where it occurs in estuaries, rivers both below and above rapids and waterfalls, in mountain streams, lowland swamps as well as in temporary, shallow rain pools. It grows up to 10 cm in length (Price 1955; Kenny 1995; Bührnheim and Fernandes 2003). Previous studies in Trinidad have shown that the highest population densities of R. hartii are found in habitats that are inaccessible to other fish (Liley and Seghers 1975; Fraser and Gilliam 1992; Gilliam et al. 1993; Fraser et al. 1995, 1999; Reznick et al. 1997; Gilliam and Fraser 2001). In Trinidad, it also is the only fish that can ascend or bypass waterfalls and rapids, and is known to move from rivers to steep rocky mountain streams (Gilliam et al. 1993; Fraser et al. 1995, 1999; Fraser et al. 2001; Gilliam and Fraser 2001). Notably, individuals are capable of jumping out of the water and may survive on, and travel substantial distances across, damp leaf-litter (Seghers 1978), so that many populations are found some distance away from rivers, in nearby pools (Reznick 1982; Costa 1987).

Taxon sampling

Rivulus specimens were captured using handnets during July to August 2002 and 2003. Sites sampled are shown in Fig. 1 and listed in Table 1. After collection, they were killed in 0.01% benzocaine and preserved in 95% ethanol. Freshwater fish surveys and reviews (Price 1955; Boeseman 1960; Kenny 1995) have shown that Rivulus hartii is the only rivulidine species present in Trinidad and thus made identification of individuals simple (specimen museum numbers are listed in Table 1). Despite careful search the only other fish found in the sampled localities was the Guppy, Poecilia reticulata. These were present only at Blue Basin (in the river below the waterfall, not in the steep stream where Rivulus was captured), and at Lopinot.

Laboratory protocols

DNA extraction protocols were similar to those described by Sambrook et al. (1989). Mitochondrial DNA (mtDNA) was extracted from caudal tail muscle and purified using standard phenol/chloroform protocols. The primers L14724 (Kocher et al. 1989) and H15557 (Hillis et al. 1996) were used to amplify a fragment of approximately 830 bp under standard PCR conditions: 1× (94°C 2 min), 30× (93°C 1 min; 52°C 1 min and 72°C 1 min), 1× (72°C 10 min, 30°C 1 min and 4°C hold) using commercial reagents and protocols similar to those recommended by the manufacturer (Promega, UK). Amplified cyt b fragments were purified by gel electrophoresis and recovered by silica/chaotrope spin column (Qiagen, Crawley, UK). These cyt b sequences are approximately twice as long as those previously reported (Collier et al. 1998). Moreover, 12S rDNA sequences were also obtained from a few animals, but showed no variation.

Sequencing of both strands was performed by the in-house sequencing unit with the amplification primers and dideoxy chain terminators (BigDyes, PE Biosystems, Warrington, UK), and analysed in an ABI 377 apparatus (Applied Biosystems, Perkin Elmer, Warrington, UK). After removing PCR primers and incomplete terminal sequences, 756 base pairs (252 codons) were available for analysis. For DNAs T528 and T531 poor quality internal reads (respectively 62 and 81 nt long) were replaced by Ns. All nucleotide sequences could be aligned without gaps and, when translated into amino acids using the vertebrate mitochondrial code, stop codons were absent.

Phylogenetic analysis

Analyses were performed with paup*4.b.10 (Swofford 2002) and MrBayes 3.0 (Huelsenbeck and Ronquist 2001). The PTP test (Faith and Cranston 1991) was used to assess the presence of non-random structure in the data, based on a heuristic search of 100 randomisations. Trees were constructed by maximum parsimony (MP; heuristic search with TBR branch exchange), maximum likelihood (ML) and Bayesian maximum likelihood (BML) optimality criteria but only ML and BML results are shown because MP did not contribute useful information. The Akaike Information Criterion (Posada and Buckley 2004) best-fit ML model was identified using Modeltest 3.06 (Posada and Crandall 1998) for the ML analysis, and MrModelTest 2.0 (Nylander 2004) for the Bayesian analyses.

Clade support in MP and ML was inferred by bootstrapping and in BML by the frequency with which a clade appeared in the saved trees. BML analyses were performed with default priors and Markov chain settings, and with random starting trees. The gamma shape parameter and proportion of invariant sites were estimated from the data and a codon-based model of site-specific rate heterogeneity was implemented. Trees were sampled every 100 generations for 1 000 000 generations. The log-likelihood scores plateau were reached at about 5000 and 1000 generations (2, 3, respectively) and a consensus tree was constructed from the last 1000 (Fig. 2) and 100 (Fig. 3) trees (100 000 and 10 000 generations, respectively). The likelihood ratio test (Huelsenbeck and Crandall 1997) was used to decide whether ML-clock and ML + clock trees differed significantly. Relative rate tests were performed with RRtree 1.1.13 (Robinson et al. 1998).

Saturation of substitutions was evaluated by plotting (in Excel) transition against transversion ‘p’ distances and fitting the linear or power regression that gave the highest r² value. TCS 1.18 (Clement et al. 2000) was used to reduce the sequences to haplotypes, and because of the observed low variability and the large number of identical sequences within and between localities, all phylogenetic analyses were performed on the resulting eight haplotypes.

Statistical analysis of population diversity and structure

The geographical structuring of genetic variation was evaluated with Φ-statistics, using the analysis of molecular variance (amova) in ARLEQUIN (Excoffier et al. 1992; Schneider et al. 2000). The significance of variance components and Φ-statistics were assessed with 10 000 data permutations. Haplotype frequencies and a minimum spanning network (MSN) were estimated using TCS 1.18 (Clement et al. 2000), following the method of Templeton et al. (1992).

Sequence characteristics

As in other fish, base composition shows an anti-G bias (A = 27%; C = 27%; T = 32%; G = 14%) (Cantatore et al. 1994; Zhu et al. 1994; Martin and Bermingham 1998; Lee et al. 2001; Doadrio and Dominguez 2003; Peng et al. 2004). Base composition heterogeneity was absent (Chi-squared test, p = 0.96). The scatter-plot of transition versus transversion ‘p’ distances showed a linear relationship (r² = 0.96), indicating that there was no detectable saturation. There were 77 variable nucleotides (69 parsimony-uninformative and eight informative); 63 variable sites were third codon positions, five were second codon positions and one was a first codon position, but none resulted in an amino acid substitution.

Among the 36 R. hartii individuals sampled, eight haplotypes were found (Table 1, Fig. 3). Six haplotypes occurred in samples from the Northern Range: Blue Basin (H1 and H4); Paria River (H3); El Tucuche and Maracas Waterfall (H5); Mount Saint Benedict and Lopinot (H7); and Tucker Valley (H8). One haplotype (H7) occurred in both of the two Northern Range localities (Mount Saint Benedict, Lopinot) and in the Central (Mount Tamana, Mount Harris) and southern (La Lune River) ranges. A single haplotype was recovered from every sample site except Blue Basin and Mount Harris, at each of them two haplotypes were found (H1 +  H4 and H6 +  H7, respectively). In Tobago, only one haplotype (H2) was found.

Phylogenetic analyses

The eight haplotype sequences were aligned with the previous cyt b sequences (each of 360 nt, obtained from Genbank) derived from specimens from western Brazil and Venezuelan or Guianese locations. The 360 nt region present in all specimens was analysed by BML to identify a suitable local outgroup for analysis of the 756 nt ingroup alignment. This analysis (Fig. 2) clearly identifies the existence of a derived R. hartii clade containing sister sub-clades, one clade comprising the specimen from south-eastern Caracas, the haplotype H1 and the sequence from the mainland Paria peninsula locality, a specimen from the island of Margarita and yet another comprising the remaining island haplotypes. This result indicates that H1 is an appropriate local outgroup for the main analyses that follow. The levels of divergence associated in this alignment with the various named species of Rivulus are given in Table 2.

The 360 and 756 nt alignments have strong non-random structure (PTP test, p = 0.01, respectively). From the 756 nt alignment ML and BML analyses both recovered two moderately well-supported clades, shown in Fig. 3. One clade comprises haplotypes H6, H7 and H8, from low-altitude localities in the Northern, Central and Southern Ranges, while the second comprises two sub-clades, one uniting the Tobago haplotype H2 with H3 from the nearest Trinidad locality (Paria River), the other uniting the adjacent Northern Range localities Blue Basin (H4) and El Tucuche/Maracas Waterfall (H5). Thus this analysis provides evidence for both geographical differentiation between Rivulus lineages in different watersheds and for communication between now-isolated sites.

Although there are no direct calibrations of molecular evolution rates in Rivulidae, the customary estimate of approximately 1.5% divergence per million years (Bermingham and Avise 1986; Bernatchez et al. 1991; Bernatchez and Dobson 1991; Zardoya and Doadrio 1999; Machordom and Doadrio 2001; Doadrio et al. 2002; Doadrio and Dominguez 2003) was used to obtain a rough idea of the likely ages of the divergences observed, given that relative rate tests found no significant differences between lineages (p > 0.05) in the rate of change in sequences from all the localities. Divergence estimates used for the molecular clock analysis are given in Table 3.

Analysis of variance

Because of the high sequence divergence observed between H1 and the ingroup, and the large contribution that this haplotype must make to variability, it was excluded from all amova analyses. These located 72% of the variation between and 28% within populations (Table 4). amova also revealed a significant subdivision between the northern (Tobago and Paria River) and southern drainages (all other localities) and between low- and high-altitude populations (<300 versus >300 m above sea level). Other variables, such as mountain ranges, faunal groups (Kenny 1995) or river drainages, showed no significant differentiation (p > 0.05).