Population genetic analysis of microsatellite variation of guppies (Poecilia reticulata) in Trinidad and Tobago: evidence for a dynamic source–sink metapopulation structure, founder events and population bottlenecks


  • Present address: N. J. Barson, Centre for Ecological and Evolutionary Synthesis (CEES), Department of Biology, University of Oslo, PO Box 1066, Blindern, N-0316 Oslo, Norway.

Dr Cock van Oosterhout, Evolutionary Biology Group, Biological Sciences, University of Hull, Hull HU6 7RX, UK.
Tel.: +44 1482  465505; fax: +44 1482 465458; e-mail: c.van-oosterhout@hull.ac.uk


Riverine fish populations are traditionally considered to be highly structured and subject to strong genetic drift. Here, we use microsatellites to analyse the population structure of the guppy (Poecilia reticulata), focussing on the headwater floodplain area of the Caroni drainage in Trinidad. We also analyse the population genetics of guppies in the Northern Drainage in Trinidad, a habitat characterized by rivers flowing directly into the sea, and a small isolated population in Tobago. Upland Caroni populations are highly differentiated and display low levels of genetic diversity. However, we found no evidence to suggest that these upland populations experienced recent population crashes and the populations appear to approach mutation–drift equilibrium. Dominant downstream migration over both short- and long-time frames has a strong impact on the population genetics of lowland Caroni populations. This drainage system could be considered a source–sink metapopulation, with the tributary furthest downstream representing a ‘super sink’, receiving immigrants from rivers upstream in the drainage. Moreover, the effective population size in the lowlands is surprisingly low in comparison with the apparently large census population sizes.


Due to their dendritic structure, river populations are predicted to be particularly vulnerable to fragmentation which may be exacerbated by unidirectional migration (Fagan et al., 2002, 2005). Many riverine populations are also fragmented by natural barriers (waterfalls and rapids) and man-made structures such as dams, weirs and culverts. These barriers can strongly influence the dispersal rate and migration pattern, even of fishes with strong swimming abilities such as salmonids (Wofford et al., 2005).

Seasonal fluctuations in river flow rates are known to have a particularly severe impact on the ecology and genetic composition of riverine organisms in headwater habitats (Grether et al., 2001; Crispo et al., 2006). Such perturbations may be particularly severe in the tropics because of heavy rains in the wet season (e.g. van Oosterhout et al., 2007). Flush avoidance and retention behaviour have evolved to compensate for involuntary downstream movement. The effects of flooding could be ameliorated by active upstream migration (Speirs & Gurney, 2001; Croft et al., 2003) and utilization of refugia in streams (Lancaster, 2000). Nevertheless, flooding can result in local population extinction (e.g. Grether et al., 2001), and Müller’s (1954)‘drift paradox’ describes the inevitability of extinction in a closed population subject to dominant downstream migration.

Headwater riverine fish populations are often characterized by high levels of local inbreeding due to small population size, geographic isolation (Fagan et al., 2002, 2005) and dominant unidirectional gene flow with the direction of water flow (Hänfling & Weetman, 2006). Even larger lowland populations can show reduced levels of diversity as a result of pronounced population substructuring (Ward et al., 1994; Jacobsen et al., 2005). However, the effect of restricted migration is expected to be particularly severe in upland populations that have been founded after repeated colonization events. Serial founder events (Clegg et al., 2002; Lambert et al., 2005) and extinction–recolonization cycles in upland populations are expected to result in severe genetic drift that reduces genetic variation. The population genetics of riverine populations can have strong effects on the potential for local adaptation, especially in small and isolated upland populations. On the one hand, genetic isolation reduces the homogenizing influence of gene flow and allows for local adaptation. However, it also deprives these upland populations from novel genetic variation contributed by migrants that may become the focus for selection (Garant et al., 2007). In addition, their small effective population size renders these populations prone to the stochastic effects of random genetic drift and the detrimental effects of inbreeding, and it may limit the adaptive evolutionary potential.

The guppy, Poecilia reticulata, is a tropical freshwater fish that has become a classic model animal for ecology and evolution studies due to marked phenotypic and genetic heterogeneity that occurs over a confined geographic area (Magurran, 2005). It is an ideal organism to study the population genetic effects of geographic isolation and small population size in the wild as it occupies a range of habitats from small fragmented headwater streams to deep lowland pools. Traditionally, the lowland and upland guppy populations have been classified as high- and low-predation habitats, reflecting the differences in prevalence of fish predators. More recently, research highlighted other differences between these habitats, including primary productivity (Reznick et al., 2001) and parasite faunas (Cable & van Oosterhout, 2007). Previous studies indicate that guppy populations are highly structured and that headwater populations show signs of local inbreeding (Carvalho et al., 1991; Shaw et al., 1994; Crispo et al., 2006). Despite the evidence for reduced variation, little differences in the effective population size (Ne) were found to exist between upland and lowland populations and, hence, drift was thought to play an equally important role in both habitats (Shaw et al., 1994). Recent advances in population genetic data analysis, including Bayesian and coalescent approaches, allow for a more effective utilization of genetic data and greater inference about current and historical population processes. In addition, these new methods require fewer assumptions about the population structure (Neigel, 2002; Pearse & Crandall, 2004).

Despite the important role of the guppy as an evolutionary model organism, previous studies of their population genetics have used mtDNA or allozymes, or focussed on a single river only (e.g. Crispo et al., 2006; van Oosterhout et al., 2006b,c). Understanding the inter-relationships among the wild Trinidadian guppy populations, particularly the extent of historical and contemporary gene flow, population structuring and population stability, is essential to the interpretation of evolutionary studies. This study aims to address these issues. First, we contrasted contemporary and long-term gene flow, hypothesizing that contemporary migration would be heavily downstream biased as a result of flushing during spate conditions in the wet season. We predict that contemporary downstream migration outweighs the effect of compensatory upstream migration, whereas long-term gene flow is more balanced due to the additional upstream movement associated with the historical colonization. Second, we hypothesized that the estimates of the long-term Ne in upland and lowland populations are relatively similar. We predict that drift plays an important role in both habitats because the upland populations are small and genetically isolated, whereas lowland populations are highly structured. Third, we hypothesized that upland populations would show evidence for population bottlenecks resulting from extinction–recolonization dynamics. The final hypothesis concerns a guppy population on Tobago that resembles those of the western Trinidadian mtDNA lineage (Fajen & Breden, 1992). The authors suggested that such resemblance might be explained by a recent introduction of fish from Trinidad into the Hillsborough River in Tobago. We therefore tested whether the relatively small guppy population in Tobago showed evidence of a recent founder event.

Materials and methods

Fish collection and DNA extraction

Guppies (P. reticulata) were collected from four rivers within the Caroni Drainage (Caura, Lopinot, Guanapo and Aripo) and two within the Northern Drainage (Yarra and Marianne) in the Northern Range of Trinidad, June to July 2003 (Fig. 1). Fish were sampled from both the upper and lower reaches of rivers in the Caroni Drainage, and the mid- and lower reaches of the Marianne River and the lower Yarra. A sample was also collected from the Lambeau River, Tobago (Grid Ref. 11°10.552 60°45.076). This broad scale sampling design allowed comparisons of population structure between the extreme upper and lower reaches of rivers within the Caroni drainage with populations from the Northern Drainage and a hypothesized recently founded population on Tobago. Fish were captured as partial or whole shoals by identifying cohesive groups of individuals swimming together in the stream and isolating these fish into a confined area using a fine mesh net. Fish were then captured from within the net boundary using small plastic buckets. Between one and nine shoals were collected from each site, the mean length of river sampled at each site was 100 m for the upland populations and 88 m for the lowland, mean distance between shoals 24.4 m in the upland and 25.6 m in the lowland. All fish were preserved in 90% ethanol and DNA was extracted from the caudal fin using the HotSHot protocol (Truett et al., 2000).

Figure 1.

 Map of Northern Trinidad showing the rivers sampled in the present study; inset shows location of the detailed map within Trinidad. UC, Upper Caura; LC, Lower Caura; LL, Lower Lopinot; UL, Upper Lopinot; LG, Lower Guanapo; UG, Upper Guanapo; LA, Lower Aripo; UA, Upper Aripo; LY, Yarra; LM, Lower Marianne; MM, Mid-Marianne.

Microsatellite genotyping

Extracted DNA was amplified at eight microsatellite loci; four interrupted repeats Pr36, Pr39, Pr80 and Pr92 (Becher et al., 2002), three perfect dinucleotide repeats Pret77, Pret69 (Watanabe et al., 2003) and Hull9-1 (CA)n and a perfect tetranucleotide repeat Hull70-2 (GTGC)n (van Oosterhout et al., 2006b), except for the Caura which was not genotyped at Pr39 or Pr80 and Tobago samples which were not genotyped at Pr80. Only those loci genotyped for all populations were used in the analysis, although all available loci were used for the calculation of summary statistics. Forward primers were labelled with the TET/HEX/FAM dye set. Microsatellites were amplified using a Qiagen multiplex PCR kit following the manufacturer’s instructions with 30 PCR cycles and annealing temperatures of 56 °C for H9-1, Pret77 and Pret69; 53 °C for Pr80 and Pr39 and 52 °C for Pr92, Pr36 and H70-2. Loci with the same annealing temperature were amplified in a single 10-μL PCR. The PCR products were resolved on an ABI377 sequencer (Applied Biosystems, Foster City, CA, USA) using TAMRA 350 (Applied Biosystems, Foster City, CA, USA) internal size standard. Sample standards were run on each gel to ensure consistent allele size scoring within and among gels. Microsatellites were sized using genescan and genotyper (Applied Biosystems). Microsatellites were chosen which had intermediate levels of polymorphism and relatively high heterozygosity so that they would be informative for population genetics. However, there is increasing data to suggest that the selection of microsatellites can lead to an ascertainment bias (e.g. Pardi et al., 2005).

Null alleles and linkage disequilibrium

Genotype frequencies were checked for the presence of null alleles, large allele drop out and stuttering using Micro-Checker 2.2.3 (van Oosterhout et al., 2004, 2006a). Due to the occurrence of common alleles with frequencies in excess of 50%, the presence of null alleles was also checked by calculating Fis using fstat (Goudet, 1995). A Bernoulli equation, = [N!/(N − K)!K!]αK(1 − α)N − K, where N is the number of tests and K is the number below the designated type I error rate (α), was applied to the results for all populations to account for multiple tests (Moran, 2003). The Bernoulli procedure accounts for multiple tests but removes problems associated with over conservatism of the Bonferroni procedure by reducing the likelihood of type II errors. We considered the Bernoulli procedure as more appropriate where the multiple tests evaluate an overarching hypothesis, such as in this case, but use the Bonferroni correction when the evaluation concerns which individual test of a group are significant (see FST below) (Kinnison et al., 2002). A locus was considered to contain null alleles if the number of populations with a significant Fis was greater than that expected by chance alone (α = 0.05). Of the eight loci tested, only Hull 70-2 showed evidence of null alleles using the Fis procedure. Micro-Checker 2.2.3 also showed null alleles at this locus in three of the six populations for which the test could be performed. This locus was excluded from subsequent analysis. Linkage and Hardy–Weinberg equilibria were tested in genepop v. 3.4 (Raymond & Rousset, 2000).

Genetic diversity and population structuring

Allelic richness was calculated in agarst (Harley, 2001). Populations were jackknifed to account for differences in sample size. Weir & Cockerham’s (1984)FST was calculated as a measure of population differentiation (microsatellite analyser, msa 4.0; Dieringer & Schlötterer, 2002). Significance levels were calculated using 10 000 permutations and strict Bonferroni correction (Rice, 1989). Group-level Bayesian analysis in baps v. 3.1 (Corander et al., 2003, 2004) was used to test for population clustering.

Effective population size and migration rate – long term

The effective population sizes and rates of gene flow were estimated simultaneously using a maximum likelihood coalescent approach implemented in migrate (Beerli & Felsenstein, 1999, 2001). This software estimates theta (Θ) which equals four times the effective population size, Ne, times the mutation rate, μ (4Neμ), and a migration rate parameter M, which is the immigration rate m divided by the mutation rate μ. This measure quantifies the number of new variants introduced into the population by immigration relative to mutation.

migrate was run four times; the first run used FST-based estimates as the start point. Subsequent runs used the results of the previous run as start values. The migration rate (M) was converted into the number of migrants per generation (Nem) by multiplying M by Θ and then dividing by four. This method of estimating migration rates assumes that the populations are in equilibrium, these conditions appear to be largely met by our data (see bottleneck analysis below). The program was run until Ne and Nem estimates were consistent between runs, either reaching an asymptote or having overlapping 95% confidence intervals. Five hundred short chains and 5000 long chains were run; the burn-in was set to 10 000 and we used a heating scheme with the following temperatures 1.0, 1.2, 1.5, 3.0 for all runs. Ninety-five per cent confidence intervals of the migration rate Nem were compared to determine whether downstream migration was significantly greater than upstream migration. The 95% CI accurately reflect the variation in the predicted parameter estimates among loci for both Ne and Nem.

Guppy populations from the Northern and Caroni Drainages and Tobago were analysed separately. Based on mtDNA data, the gene pools in both drainages have been separated for more than 6000 guppy generations (Fajen & Breden, 1992). Homoplasy of microsatellite alleles is likely to significantly bias the results obtained from the coalescence procedure when populations have been separated for more than 6000 generations (Estoup et al., 2002). The drainages were therefore split, which avoids the confounding effects of size homoplasy. The analysis with migrate were conducted twice for the Caroni drainage populations. First, we ignored possible migration between rivers in a ‘river-by-river’ analysis, and only accounted for migration between upland and lowland sites within rivers. We then repeated the analysis, but allowing for migration between rivers in the ‘whole drainage analysis’ and estimating Θ for all populations simultaneously. The Θ values were then translated into Ne estimates using a microsatellite mutation rate of 5 × 10−4, which is the most frequently employed rate in fish (Estoup & Angers, 1998; Lippe et al., 2006).

Migration rate – contemporary

The levels of admixture within each individual was analysed using baps (Corander et al., 2003, 2004). Initially a partial Bayesian cluster analysis was performed at the group level. During this phase, the program tests whether the sampled populations are better described by a smaller number of populations. An admixture analysis was then carried out using the populations defined by the cluster analysis. This mixture analysis estimates the proportion and source of admixture within each individual of a population. We estimated the amount of gene flow among populations by summing the within-individual admixture proportions across all individuals in a population for each source population and then scaling by the number of individuals sampled in the receiving population. This allowed us to compare the amount of downstream vs. upstream gene flow (Hänfling & Weetman, 2006). The status of each population as a source/sink for migrants was also assessed using the total immigration minus the total emigration. The total immigration was calculated by summing the mean proportion of admixture received from all the other populations and, similarly, the total emigration was computed using the mean proportion of admixture donated to all the other populations.

Population bottlenecks

Evidence for population bottlenecks was tested using bottleneck (Cornuet & Luikart, 1996; Piry et al., 1999) and by calculating Garza and Williamson’s M and the critical M value below which a population bottleneck is likely (Garza & Williamson, 2001). These two measures detect different effects of a bottleneck. bottleneck tests for a relative heterozygosity excess that is apparent for a few generations after a bottleneck. The excess in heterozygosity develops because initially the allelic diversity declines faster than heterozygosity due to the loss of rare alleles. The observed allele frequency distribution was compared with that of a population in mutation–drift equilibrium assuming the TPM (two phase model; Di Rienzo et al., 1994) with 70% stepwise mutation model (SMM) and 30% infinite allele model (IAM). Deviations between the observed and expected frequency distributions were tested using the Wilcoxon’s signed rank test. bottleneck was run for 10 000 iterations.

Garza and Williamson’s M is the mean ratio between the number of alleles and the allelic range, and can be considered a measure of the ‘fullness’ of the allelic distribution (Garza & Williamson, 2001). A population in mutation–drift equilibrium will have a relatively full allelic range, whereas a population that has suffered a bottleneck will have lost rare alleles, and so will have gaps in the allelic range. Garza and Williamson’s M was calculated using M_P_Val (http://swfsc.noaa.gov/textblock.aspx?Division=FED&id=3298) which also generates a simulated distribution of M for each population. The mean number of nonstepwise mutations was set as 0.12 and the mean size of larger mutation as 2.8, as recommended by Garza & Williamson (2001), theta was set to the value estimated by migrate for each population. There is evidence of a population decline if less than 5% of the replicates in the simulated distribution are below the critical M value. The critical value for M (Mcrit) for each population was calculated using Critical_m (http://swfsc.noaa.gov/textblock.aspx?Division=FED&id=3298).


No loci showed evidence of linkage disequilibrium. Tests for Hardy–Weinberg equilibrium (HWE) revealed significant departures in 12 of 120 comparisons (10 heterozygote deficits and two heterozygote excess, α = 0.05). After Bonferroni correction only Pr36 and 9-1 from the Lower Aripo (LA) showed departure from HWE (corrected α = 0.00083). Using the Bernoulli procedure, again only the LA population showed evidence of deviation from HWE, having four of 10 tests with significant deviations at α = 0.05 (Pr36, Pr92, Pret69 and H9-1 all heterozygote deficits) suggesting that this population may deviate from HWE.

Genetic diversity and population structuring

Upland guppy populations of the Caroni drainage showed very low levels of genetic diversity (alleles per locus 1.5–2.53) and heterozygosity (0.023–0.252) compared with lowland populations (alleles per locus 8.24–11.26; heterozygosity 0.6105–0.719). The populations of the Northern Drainage had intermediate allelic diversity (5.32–5.66), but the levels of heterozygosity (0.591–0.638) were equivalent to those of the lowland Caroni populations. The Tobago population had low allelic diversity (3.74 alleles per locus) and low heterozygosity (0.34).

Weir & Cockerham’s (1984)FST was significant for all pairwise comparisons between populations. The FST ranged from 0.169 to 0.806 for between-drainage comparisons and from 0.013 to 0.927 for within-drainage comparisons. The FST between upland populations of the Caroni drainage were particularly high (0.443–0.927), reflecting a high level of genetic differentiation and population isolation. The level of genetic differentiation between lowland populations was considerably lower (0.028–0.288). Group-level analysis in baps (Corander et al., 2003, 2004) assigned the 12 population samples to 11 clusters. The mid- and lower Marianne (sample sites only 3 km apart) were the only two populations that were not supported and clustered together in a single population. The mid-Marianne is not as high upstream as the upper populations sampled in the rivers of the Caroni Drainage. Consequently, this population displays genetic characteristics different from that of the other upland populations, which explains its relative weak genetic differentiation.

Effective population size and migration rate – long term

Based on the ‘river-by-river’ analysis (i.e. assuming no migration exists between rivers), the mean Θ was substantially lower in the uplands than in the lowlands (Fig. 2a). Assuming a microsatellite mutation rate of 5 × 10−4 (Lippe et al., 2006), the mean (±SE) effective population size (Ne) of upland populations is Ne = 244 (± 42) and lowland populations is Ne = 910 (± 84). When allowing for migration between Caroni drainage rivers in the ‘whole drainage analysis’, the Θ values of upland and lowland populations were more similar (Fig. 2b, Table 1). The Θ values in this analysis correspond to a mean effective population size of Ne = 163 (± 23) and Ne = 368 (± 103) for the upland and lowland populations respectively. Note that depending on the value of the mutation rate assumed, these Ne estimates will vary; estimated Ne for 5 × 10−4 and 1 × 10−4 are compared in Table 1. However, the lowland populations appear to be on average at least twice as large as those in the uplands. The estimated Θ for the mid-Marianne was approximately half that of the lower Marianne (Θ = 0.415 and 0.955 respectively), which corresponds to an effective population size of Ne = 208 and 478. The Yarra and Tobago both yielded surprisingly high estimates of theta, Θ = 1.684 and 2.329; Ne = 842 and 1165 respectively.

Figure 2.

 Estimates of theta (a) using migrate software with rivers analysed separately. Theta in lowland populations (open symbols) is much higher than that in the upland populations (solid symbols); and (b) using migrate with all Caroni drainage populations analysed simultaneously. Theta in lowland populations (open symbols) is much higher than that in the upland populations (solid symbols) only for the Lopinot and Caura Rivers.

Table 1.   Observed and expected heterozygosity and the number of alleles per loci in 12 guppy populations from Trinidad and Tobago.
PopulationNNumber of shoalsHet (obs)Het (exp)Alleles per locusAlleles per locus jackknifeNe (μ = 5 × 10−4)Ne (μ = 1 × 10−4)
  1. Upland populations are shown in bold. All estimates are based on seven loci.

  2. *Estimated from five loci.

  3. †Estimated from six loci.

Lower Yarra8850.64940.63816.145.66842 (793–898)4210 (3966–4492)
Mid-Marianne12960.62110.593265.39207 (198–217)1037 (990–1086)
Lower Marianne6040.59810.59095.575.32477 (427–536)2387 (2136–2682)
Upper Caura*6710.01790.02331.81.64185 (173–199)926 (864–995)
Lower Caura*13190.65950.691513.210.06486 (458–516)2430 (2291–2580)
Upper Lopinot5960.04840.05211.571.5132 (123–141)659 (614–707)
Lower Lopinot6960.72460.71912.8611.26588 (532–653)2940 (2658–3264)
Upper Guanapo5040.20570.25222.572.53118 (109–128)592 (546–642)
Lower Guanapo5250.62640.61058.578.24257 (235–282)1285 (1173–1411)
Upper Aripo4470.13310.12252.292.29218 (200–239)1091 (999–1193)
Lower Aripo5880.51720.62149.438.52140 (129–152)700 (644–762)
Tobago†5040.33000.33974.173.741164 (990–1325)5821 (4949–6624)

Downstream migration within tributaries in the Caroni drainage was moderate to high, with more than one migrant per generation (range of Nem: 1.009–1.736), except for the Aripo River (Nem = 0.135). Estimates of upstream migration were low, with Nem ranging between 0.037 and 0.144, and with 95% CIs approaching zero. In all upland–lowland comparisons except in the Aripo River, the 95% CI did not overlap between the up- and downstream migration estimates. The very low estimates of both up and downstream migrations in the Aripo River may be a result of restricted movement due to separation of these populations by a large waterfall. Restricting the analysis to within river comparisons in the Caroni drainage increased the up- and downstream migration estimates substantially (range of Nem downstream: 2.51–8.78 and upstream: 0.337–2.202). Migration was also downstream biased in the Marianne River (Nem downstream: 15.482 (95% CI 12.901–18.836) and upstream: 1.058 (95% CI 0.914–1.216)). Approximately one migrant per generation was exchanged between the lower Yarra and lower Marianne (1.181; 95% CI 0.823–1.676), with one migrant every two generations exchanged in the opposite direction (0.580; 95% CI 0.441–0.759).

Migration rates between lowland populations were highly variable and sometimes substantial (range of Nem 0.093–6.686). There was also evidence of unidirectional migration between lowland populations. In most cases, there was greater migration from eastern to western tributaries (i.e. downstream in the Caroni River) than vice versa (Fig. 3). There was one noteworthy exception in that long-term migration between the Lopinot and Caura River was upstream rather than downstream biased (Fig. 3). This suggests that guppies are able to re-enter tributaries from the Caura River despite the upstream migration that would be required.

Figure 3.

 Long-term upstream (solid symbols) and downstream (open symbols) migrations between the lowland populations in the Caroni drainage. The Nem is estimated using the coalescent-based approach in Migrate. With the notable exception of the Lopinot–Caura comparison, migration between rivers appears to be downstream biased.

Migration rate – contemporary

All lowland populations show considerable levels of admixture (Fig. 4). Guppies of the Lower Aripo population showed significantly less admixture than fish from any of the other lowland populations of the Caroni Drainage (Mann–Whitney U-test: W < 4650.0, P < 0.0078 in all three pairwise comparisons). The level of admixture in upland populations was considerably less than in the lowlands (Mann–Whitney U-test: W < 1912.0, P < 0.001), again emphasizing that these populations are genetically isolated. Downstream gene flow was greater than upstream migration in all comparisons (mean downstream migration 0.171 ± 0.03; mean upstream migration 0.006 ± 0.004). Upland populations appeared to contribute migrants (source populations), whereas the large lowland populations were net receivers of migrants, i.e. sink populations (Fig. 5). The analysis on the contemporary within-river migration (upstream less than downstream) thus corroborates the conclusions based on long-term migration patterns.

Figure 4.

 Admixture analysis in baps on upland and lowland guppy populations within rivers of the Caroni drainage, Trinidad. Guppies of the lowland populations (open symbols) have a significantly higher proportion of their genomes admixed than fish from the upland populations (black symbols).

Figure 5.

 Difference between average proportions of immigrant genetic material compared with emigrant gene flow contributed to other populations. Net negative values represent source populations with a relative excess of migrants leaving the population and positive values are sink populations characterized by immigration. Black bars indicate upland, white bars lowland and grey bars Northern Drainage and Tobago.

Guppies tend to have more of their genome admixed with conspecifics from tributaries further east (upstream) than with those to the west (downstream) (Fig. 6). This pattern is strong in comparisons with the Aripo River (the most eastwards river), but less pronounced for other comparisons. This suggests that contemporary gene flow is largely mediated by downstream migration, from the Aripo River into other (more westwards) rivers. The pattern of contemporary migration is in general congruent with that of long-term migration showing an overall downstream bias. The exception of this pattern is the Lopinot and the Caura where there is a weak downstream bias in the contemporary estimate but an upstream in the long-term estimate.

Figure 6.

 Migration estimates between lowland rivers of the Caroni drainage showing the effects of river order. For all comparisons, guppies have more of their genome admixed with fish from a river further east than from a river from the west. For example, the Guanapo guppies show a higher level of admixture with fish from the Aripo (open symbol) than vice versa (solid symbol).

Population bottlenecks

Analysis with the software bottleneck did not show evidence for a recent bottleneck in any of the lowland (Wilcoxon test: P > 0.953) or upland (Wilcoxon test: P > 0.922 all test results) populations of the Caroni Drainage. In fact, lowland Caroni populations and the Tobago population appeared to have a relative heterozygote deficit, which is opposite to the effect expected after a bottleneck (LA: P = 0.020, LG: P = 0.039, LL: P = 0.027, LC: P = 0.078 n.s. and Tobago: P = 0.045). By contrast, populations from the Northern Drainage showed a heterozygote excess which is evidence for a population bottleneck (Yarra: P = 0.055 n.s. and Marianne: P = 0.020). None of these results are highly significant and so would be rendered nonsignificant after a Bonferroni correction. However, it is highly unlikely that this many significant results (at α = 0.05) would be obtained by chance alone (P = 0.000106).

Garza and Williamson’s ‘M’ statistic exceeded the critical value of M (Mcrit) in almost all upland populations (= 0.83–0.881; Mcrit = 0.789–0.82; P > 0.064 for all tests) except in the Upper Aripo (= 0.81; P = 0.048). This suggests that the Upper Aripo population may have experienced a recent population decline. The statistical power of Garza and Williamson’s M might be compromised in the upland populations given the dominance of a single allele for most loci. However, the results do largely corroborate the analysis with bottleneck which showed that the upland populations exhibited no signs of recent population bottlenecks. Surprisingly, however, all lowland populations from the Caroni Drainage showed evidence for recent bottlenecks or founder effects (= 0.525–0.671; Mcrit = 0.78–0.821, P < 0.003 for all tests). Guppy populations of the Northern Drainage also showed evidence of recent populations declines (= 0.612–0.675; Mcrit = 0.785–0.819; P < 0.001 for all tests) and the Tobago population showed clear signs of a bottleneck or founder event (= 0.324; Mcrit = 0.761; P < 0.001). It is highly unlikely to obtain this many significant results at the number of tests performed (α = 0.05; P < 0.000001). In addition, most of the results are also significant after the strict Bonferroni correction (adjusted P = 0.004), except for the Upper Aripo which is marginally significant.


This study of wild populations of guppies (P. reticulata) used modern coalescent and Bayesian approaches to infer population demography and migration patterns over both contemporary and long-term timescales. Freshwater fish populations have long been regarded as having low genetic diversity and being highly structured with restricted gene flow (Ward et al., 1994). In the present study, headwater populations of the Caroni drainage showed extremely low levels of diversity for the microsatellite markers analysed (mean alleles per locus 1.5–2.53) with remarkably high FST values (> 0.443), (Table 2). However, these populations had little or no detectable signals of population bottlenecks, and the long-term persistence of guppy populations in headwater habitats does not appear to be jeopardized by involuntary downstream flushing as suggested by Müller’s (1954) drift paradox.

Table 2. Weir & Cockerham’s (1984)FST.
FST valuesLower YarraMid-MarianneLower MarianneTobagoUpper GuanapoLower GuanapoUpper LopinotLower LopinotUpper AripoLower AripoUpper CauraLower Caura
  1. All values are significant at P < 0.01 except those at *P = 0.0396. Bold values represent within-drainage pairwise comparison.

Lower Yarra            
Lower Marianne0.2820.013*          
Upper Guanapo0.5310.5440.5810.680        
Lower Guanapo0.3430.3510.3290.4910.233       
Upper Lopinot0.5040.6120.6550.7630.8640.605      
Lower Lopinot0.1870.2880.2620.3700.3760.1620.323     
Upper Aripo0.4810.5750.6190.7300.8600.5990.8800.453    
Lower Aripo0.2850.3870.3790.4830.5460.2880.5940.2020.357   
Upper Caura0.4780.6280.6840.8060.9100.6690.9260.4430.9270.705  
Lower Caura0.1690.2930.2750.3840.3730.2020.3360.0280.4200.2180.313 

Lowland populations showed high levels of diversity and lower differentiation, corroborating findings of earlier allozyme studies (Carvalho et al., 1991; Shaw et al., 1991, 1994). As suggested by Shaw et al. (1994), these populations do appear to be characterized by ongoing migration both within and among rivers. Long-term and contemporary genetic analyses indicated that lowland populations receive gene flow from the lowland populations of other rivers in the catchment, particularly in a downstream direction. A recent study of mtDNA variation throughout the range of P. reticulata suggested that over the long term (the last 200 000 years) effective gene flow occurred from the headwater tributaries to the Orinoco River delta, and between rivers (Alexander et al., 2006). The present study provides evidence for ongoing gene flow between the headwater and lower tributaries of the Caroni River, Trinidad. Both studies suggest that large fast-flowing rivers are not a barrier to gene flow in guppies. Magurran & Phillip (2001) showed that guppies were almost ubiquitous in Trinidadian rivers and that they are extremely good dispersers. This dispersal capability along with a capacity to utilize a wide range of habitats means that most guppies that disperse will find suitable habitat and conspecifics with whom to reproduce. Evidence of river capture between the Caroni and Northern Drainages is suggestive of appreciable gene flow. This has important implications for the use of these populations as replicates of adaptive divergence in evolutionary studies.

Fraser et al. (1999) found that killifish used corridor habitat, which had been assumed would be hostile, for both feeding and reproduction and that they showed no detectable stress in this habitat. This suggests that fish populations may be more continuously distributed than is often perceived which may have important implications for the applicability of metapopulation theory to fish populations (Schtickzelle & Quinn, 2007). The dispersal capacity and cosmopolitan habitat exploitation of the guppy have important implications for species cohesion despite diversifying local selection. Crispo et al. (2006) suggested that physical processes dominated patterns of gene flow in guppies despite divergent selection regimes. It also illustrates an important aspect of the biology of organisms occupying dendritic habitats; frequently the connecting branches regarded as movement corridors can actually also be used as habitat (Campbell Grant et al., 2007).

Gene flow and admixture

The highest level of admixture in lowland populations of the Caroni Drainage is found in the Caura River, the furthest downstream population of those sampled in the drainage. Although probably also used as habitat, the fast-flowing, turbid waters of the Caroni River force a strong downstream directionality in the gene flow. Interestingly, the rivers further downstream in the drainage appear to receive more migration in the long term and had greater levels of admixture from other lowland populations than those higher in the catchment (rivers more eastwards). This suggests that entry into the lower tributaries is more likely than those higher up; however, there are also more potential source populations upstream for these lower tributaries. It is also possible that the guppies that enter or live in the strong-flowing Caroni River are likely to be transported downstream, especially during the high flow of the rainy season, and as a result of this current-mediated transport are more likely to transfer genes to more lowland tributaries. This hypothetical migration pattern could explain the increasing level of admixture of rivers further downstream (Fig. 6). We propose that the Caura constitutes a ‘super sink’ population, a large catchment area for guppies from other Caroni Rivers that are washed down during heavy rains.

The estimated migration rates among populations appear high in comparison with the FST estimates. Two populations separated by significant FST estimates were collapsed into a single population by using the baps analysis. However, the FST was low (0.013) between these two populations and could be below the detection limits of this type of analysis (Latch et al., 2006). This inconsistency may reflect a lack of fit to the assumptions that underlie the inference of gene flow from FST estimates (Whitlock & McCauley, 1999). For example, gene flow is assumed to be symmetrical, which is clearly not an appropriate assumption for guppy populations. Additionally, we have confidence in the coalescent-based estimates of migration rates and patterns as they appear consistent with our knowledge of guppy biology in this seasonally fluctuating environment (see, e.g. van Oosterhout et al., 2007).

We found a consistent downstream bias in migration within tributaries in all contemporary and four of five long-term migration estimates. This suggests that ongoing downstream migration during wet season rains contributes significantly to the lowland gene pools. We found no evidence suggesting that active compensatory movement of fish upstream (Croft et al., 2003) affected the pattern of gene flow in either the contemporary or long term. Both Haskins et al. (1961) and Becher & Magurran (2000) also found evidence for downstream-biased migration and this appears to have strong effects on genetic diversity irrespective of barrier waterfalls (Shaw et al., 1994). This study thus corroborates other studies, confirming that downstream-biased gene flow is a common feature of fish inhabiting upland river systems (Hänfling & Weetman, 2006).

Population bottlenecks or admixture?

In the lowland populations of the Caroni drainage, bottleneck revealed a relative heterozygote deficit. This signal is the opposite of that expected after a population bottleneck and could be explained by: (i) population growth in a closed system, (ii) population structuring or (iii) admixture (G. Luikart, personal communication). By contrast, Garza & Williamson’s (2001)M suggested that bottlenecks had occurred in these lowland populations. Recent simulation studies indicated that both methods may erroneously detect a population bottleneck and that they have different optimal conditions (Williamson-Natesan, 2005). Based on this simulation work, we may expect Garza and Williamson’s method to perform best in the lowland populations and bottleneck to perform better in the uplands. However, Garza and Williamson’s M may additionally be sensitive to the inflow of novel genetic variants and the result of bottleneck could also be interpreted as indicative of population expansion resulting from immigration. The guppy populations in the lowlands appear to be open to migration, a conclusion supported both by contemporary (baps) and long-term (migrate) migration estimates. We suggest that neither population growth in a closed system nor population structuring are responsible for the observed excess of rare alleles, but rather that immigration from genetically differentiated populations into the lowland populations created an apparent excess of novel alleles and an incomplete allelic distribution. Although the contribution of this mechanism is difficult to assess as we have only sampled one upland population for each river, there is concordance between the dominant upland allele and the lowland allele frequency distribution. For example, the bimodality of the lowland allele frequency distribution of Pr39 (in the LA) and Pret69 and Pr 36 (in the LC) is consistent with the input of the dominant upland allele (see Supporting information). The admixed lowland populations may constitute genetically diverse ‘sink’ populations with small isolated upland populations acting as the source populations, contributing gene flow into the much larger downstream populations. This pattern may be common in riverine populations that are dominated by unidirectional currents (Hänfling & Weetman, 2006).

The present study suggests that the upland populations in the Caroni drainage have small effective population sizes and are genetically highly differentiated (FST > 0.85). Typically, the alleles in the upland gene pools differ by one or two repeat units (see Supporting information), suggesting that most of the genetic diversity in the uplands has resulted from in situ mutation (assuming a predominantly stepwise mutation model). The populations may have experienced severe drift and loss of alleles during a founder event in the past, but new genetic variation is generated (and lost) in a mutation–drift balance. We hypothesized that the upland populations would be vulnerable to extinction and recolonization as a result of their low census size, lack of genetic variability and vulnerability to downstream flushing of guppies during spate conditions in wet season floods (see also van Oosterhout et al., 2007). However, with the possible exception of the Upper Aripo, our analysis suggests that the upland populations have not faced recent population crashes, but appear to be in mutation–drift equilibrium. van Oosterhout et al. (2006b,c) showed that the Upper Aripo population maintained a high level of genetic diversity at the major histocompatibility complex (MHC), similar to that observed in the Aripo population in the lowlands; this suggests that the Upper Aripo populations are not subject to an extinction–recolonization dynamics. The marginal significant M-statistic of the Upper Aripo population may reflect seasonal fluctuations in population size. This conclusion is supported by a mark–recapture experiment in the upper Aripo, which showed a dramatic decline in recapture rate of guppies after torrential wet season rains (van Oosterhout et al., 2007).

In contrast to Crispo et al. (2006), our analysis suggests that the lowland populations of the Northern Drainage are not in mutation–drift equilibrium, showing evidence of recent population bottlenecks. These rivers drain directly into the Caribbean Sea and, therefore, migration between rivers would require the crossing of a marine barrier. The large downstream migration into the lower Marianne (downstream Nem = 15.482, 95% CI: 12.901–18.836) suggests that guppies might be flushed out to sea when the river is in spate. The coalescent-based approach suggested that 0.5–1.0 migrant per generation are exchanged between the Lower Marianne and the Lower Yarra, implying that migration is possible between these rivers in both directions. During the rainy season, the salinity of the water off the northern coast of Trinidad can be reduced to 10–15 ppt (Kenny, 2000), which may facilitate occasional migration. However, the low sampling density in the Northern Drainage may have resulted in upwards bias in our estimates of migration rates (see discussion below).

The population of the Lambeau River in Tobago also showed low allelic diversity (alleles per locus 3.74) but, unlike the headwater populations of the Caroni drainage, this population also had a very low value for Garza and Williamson’s M (0.324) indicating a recent bottleneck. Analysis with the software bottleneck showed that this population is not in mutation–drift equilibrium; similarly to the lowland populations of the Caroni drainage they displayed a significant heterozygote deficiency, rather than the excess expected following a bottleneck. Altogether, these results are consistent with the hypothesis that the Tobago population is a bottlenecked founder population that has undergone recent rapid population expansion.

Effective population size

Guppy populations attain very large size in the lowland which is thought to result from reduced forest shading and increased nutrients arising from anthropogenic inputs (Reznick et al., 2001). Long-term effective population size in the lowlands appears to be many orders of magnitude lower than the apparent contemporary census size. Surprisingly, the effective population sizes in the lowlands are only double or quadruple that of upland sites. This convergence of effective population sizes could be a result of the long-term nature of the coalescence estimate (see Palstra et al., 2007). A previous study on guppies also found surprising similarities between the effective population sizes of upland and lowland populations (Shaw et al., 1994). This study did not use a coalescent approach, but explained these results by suggesting that upland and lowland populations have comparable levels of population stability (or instability). Our data reveal apparent stability and mutation–drift equilibrium in the uplands despite dramatically low levels of neutral genetic diversity. We suggest that upland populations are stable but small, whereas the lowland populations are relatively large (in census size) but fluctuating.

The effective size in the lowlands appears to be overestimated when between-river migration is not taken into account because all the genetic diversity needs to be explained by mutation (as well as by migration from the low-variability upland populations). Large differences in the Ne of upland vs. lowland populations were found in a study of the Aripo River by van Oosterhout et al. (2006b) using the nonequilibrium program IM which is limited to the analysis of population pairs (Hey & Nielsen, 2004). Beerli (2004) showed that the estimated population sizes become more inflated with an increased number of immigrants from unknown populations (the so-called ‘ghost populations’). The potential bias caused by ghost populations is more pronounced when the migration rate among populations is higher (Slatkin, 2005). The effect of ghost populations is apparently reduced by taking between-river migration into account. Due to the long computational time associated with coalescent analysis programs, it can be temping to split the data set for analysis based on preconceptions about the most likely sources of migrants. However, the analyses conducted here and those of Hänfling & Weetman (2006) suggest that such partitioning of data sets can bias Nem and Θ estimates.


Long-term and contemporary estimates of gene flow both showed evidence for downstream-biased dispersal despite evidence for flushing avoidance and compensatory upstream migration. Contrary to our expectations, the small upland populations appeared to be stable and in migration–drift equilibrium, whereas the large lowland populations appeared to be out of migration–drift equilibrium as a result of high levels of immigration. The Caroni drainage can be considered as a source–sink metapopulation, with the river furthest downstream representing a ‘super sink’, receiving immigrants from rivers upstream in the drainage. The effective population size in the lowlands appears to be many orders of magnitude lower than their census size and remarkably similar to that of the upland populations. This result is remarkable, given that the genetic variation in these lowland populations is considerably higher than that in the uplands. Consistent with our hypothesis, the guppy population in Tobago showed evidence of a recent founder event.


We thank Ryan S. Mohammed for assistance with fieldwork and the ASA Wright Nature Centre for access to sites. Additionally, we thank David Weetman and Mike Bruford for commenting on an earlier version of this manuscript. We also thank four anonymous reviewers whose comments on an earlier version greatly improved this manuscript. The project was funded by the Natural Environment Research Council, UK (research grant NER/B/S/2002/00410; and research Fellowships NER/J/S/2002/00706 and NER/I/S/2000/00885 to JC and CvO respectively).