Population structure in two sympatric species of the Lake Tanganyika cichlid tribe Eretmodini: evidence for introgression

Sample collection

A total of 123 specimens of Eretmoduscyanostictus were collected from five localities covering 138 km of mostly uninterrupted rocky shoreline, and 75 specimens from Tanganicodus cf. irsacae were collected from four localities in the Democratic Republic of Congo (Fig. 1), covering its entire known distribution range of 35 km. This study includes two populations of Eretmodus (KOR and KAP; for locality abbreviations see Fig. 1) that were not studied in a previous phylogenetic analysis (Rüber et al. 1999). For the microsatellite analyses two populations (Eretmodus KAP and Tanganicodus KIK) were omitted due to small sample sizes, therefore, a total of 117 specimens of E.cyanostictus and 69 specimens of T. cf. irsacae were used. Samples were preserved in 70% ethanol and voucher specimens have been deposited in the Africa Museum in Tervuren, Belgium.

Characterization of microsatellite DNA

DNA was extracted using a proteinase K, phenol–chloroform protocol (Kocher et al. 1989). A total of 10 microsatellite loci, which were developed for other East African cichlid species, were initially tested by sequencing to assess their utility for population studies in eretmodine cichlids. For sequencing, polymerase chain reactions (PCR) were carried out in 25 µL volumes [Tris 10 mm, pH 9.0, 1.5 mm MgCl₂, 0.2 mm of each dNTP, 0.2 µm of each primer, and 0.65 units of Taq DNA polymerase (Pharmacia)] using the PCR conditions given in Kellog et al. (1995) for UNH001 and UNH002, Zardoya et al. (1996) for TmoM5, TmoM7, TmoM11, TmoM13, TmoM25, and TmoM27, and van Oppen et al. (1997b) for Pzeb1 and Pzeb3, with adjustments of annealing temperatures (Table 1). PCR fragments were either ligated into pGEM-T plasmid vectors (Promega) or using the TopoTA cloning kit (Invitrogene). Clones were sequenced with M13 universal (−40) and reverse primers with an Applied Biosystem 373 A DNA sequencer using the Taq Dye Deoxy Terminator Cycle Sequencing Kit FS (PE Biosystems).

Microsatellite markers

Once suitable microsatellite loci were selected (see Results) amplified products were electrophoresed on 6% denaturating polyacrylamide gels using an ALF Express DNA sequencer (Pharmacia) with short glass plates. Gels were run between 2–4 h, depending on the allele size, at 55 °C (15 W, sampling interval 1 s). Alleles were sized with the program Allele Links^TM (Pharmacia) by using internal standards run in every lane. Internal size standards were obtained by PCR from M13mp18 + template DNA as described in van Oppen et al. (1997b). For locus TmoM25, a 453-bp sizer was designed for this study, using the reverse primer 5′-ATTTCGGAACCA CCATCAAA-3′ in combination with the universal M13mp18 + forward primer and an annealing temperature of 54 °C. The following size standards were used: 200 bp + 353 bp; 100 bp + 200 bp; 312 bp + 453 bp; 100 bp + 200 bp; 259 bp + 353 bp; and 100 bp + 200 bp for the loci TmoM5, TmoM11, TmoM25, Pzeb1, Pzeb3, and UNH002, respectively. Allelic corrections were performed for each locus using the sequenced alleles as references (Table 1) and, in addition, for each locus 2–3 randomly chosen individuals were run on all gels to ensure consistency of fragment sizing.

mtDNA

The proline tRNA with a segment of the mitochondrial control region was amplified using conditions and primers described in Rüber et al. (1999). Samples were sequenced using the AutoCyle^TM Sequencing kit (Pharmacia) and run on an ALF express DNA sequencer (Pharmacia). Some samples were sequenced on an Applied Biosystem 373 A DNA sequencer using the Taq Dye Deoxy Terminator Cycle Sequencing Kit FS (PE Biosystems). The nucleotide sequence data were deposited in EMBL/GenBank under accession numbers AJ407050–AJ407077 for the sequenced microsatellite alleles and AJ407078–AJ407147 for the control region sequences.

Population genetic analyses

Genetic diversity based on microsatellite DNA and mtDNA was assessed for all populations. For the microsatellite data the genetic polymorphism was estimated for each population with genepop 3.1d (Raymond & Rousset 1995) as the number of alleles per locus (N_A), the observed (H_O) and the expected heterozygosity (H_E). Samples were then tested for departures from Hardy–Weinberg equilibrium using probability tests with significance determined by the Markov chain method (Guo & Thompson 1992). Genotypic linkage disequilibrium was evaluated with genepop 3.1d for each pair of loci in each population and significance was determined through a log-likelihood based exact test (Goudet et al. 1996). mtDNA polymorphism was assessed by estimating both haplotype (H; Nei 1987) and nucleotide (π; Tajima 1983) diversity within populations. Differences in genetic diversity between the two species were assessed using the Mann–Whitney U-test (Sokal & Rohlf 1995).

Population structure

We tested for differences between populations in their allelic (microsatellites) or haplotypic (mtDNA sequences) distributions with an exact test of population differentiation (Raymond & Rousset 1995) as implemented in arlequin 2.000 (Schneider et al. 2000). The significance of the results was determined by the Markov chain method. For the microsatellite data, multilocus probabilities were produced using Fisher’s (1954) method of combining probabilities. We then quantified the extent of genetic differentiation among samples by calculating Weir & Cockerham’s (1984) estimators of F-statistics (θ_ST) with arlequin 2.000. Significance of results was obtained through 5000 permutations. Table-wide rejection levels for multiple tests were calculated with sequential Bonferroni adjusted P-values throughout the analyses (Rice 1989).

We performed an analysis of molecular variance (amova), as described in Michalakis & Excoffier (1996), in order to partition the genetic diversity (based either on allele or haplotype frequency) into: (i) variance between species; (ii) variance among populations within species; and (iii) variance within populations. The significance of the variance components associated with the different levels of genetic structure was tested using nonparametric permutation procedures as implemented in arlequin 2.000.

For the microsatellite data, we assessed relationships among populations using chord distances (D_CE; Cavalli-Sforza & Edwards 1967), computed for all pairwise comparisons with phylip version 3.572c (Felsenstein 1995). Pairwise distances were used to build a neighbour-joining (NJ) tree (Saitou & Nei 1987) and robustness of the tree topology was assessed through 1000 bootstrap replications (Felsenstein 1985). Phylogenetic relationships among mtDNA haplotypes were estimated using the NJ method based on Kimura 2-parameter corrected distances (Kimura 1980). Robustness of the inferred NJ tree was tested with paup* version 4.0b4a (Swofford 1997) using the bootstrap method with 1000 resamplings.

Finally, we explored the demographic history of the two species. To detect recent effective population size reductions from the microsatellite allele frequencies we used the program bottleneck 1.2.02 (Cornuet & Luikart 1996). This program uses the Sign and Wilcoxon tests to test for an excess of the observed heterozygosity which, for loci evolving under the infinite allele model (IAM), should be larger in populations having experienced a recent bottleneck than the heterozygosity expected for the observed number of alleles in populations under mutation-drift equilibrium. Another bottleneck signature tested with bottleneck 1.2.02 is a deviation of the allele frequency distribution from the approximately L-shaped distribution expected under mutation-drift equilibrium (Luikart et al. 1998). To test for a population expansion we applied Fu’s (1997)F_s-test of neutrality as implemented in arlequin 2.000, which is based on the comparison of rare and frequent mutations. After an episode of population growth, the coalescent theory predicts an excess of low frequency alleles and recent mutations in nonrecombining sequences compared to populations at equilibrium (Slatkin & Hudson 1991).

Microsatellite DNA

Of 10 microsatellite loci tested six were chosen for population analyses on the basis of the repeat number in the sequenced alleles (Table 1). The selected loci showed at least 10 uninterrupted dinucleotide repeats in one of the individuals studied and were, therefore, considered to exhibit sufficient potential for polymorphism for population analyses. The six selected loci were: TmoM5, TmoM11, TmoM25, Pzeb1, Pzeb3, and UNH002.

The core sequences of four of the 10 microsatellite loci sequenced differed from those reported in the original studies. One out of two sequenced eretmodine specimens showed a perfect (AC)₃₃ repeat compared to a compound-perfect (GC)_n(AC)_n repeat reported for locus TmoM5 in Tropheus moorii (Zardoya et al. 1996). In eretmodine cichlids locus TmoM25 had imperfect (CA)_nCGT(CA)_n alleles, compared to perfect (CA)_n alleles found in T. moorii (Zardoya et al. 1996). The two eretmodine species sequenced for TmoM13 showed a perfect (CA)_n repeat motif compared to a (CA)_n(CG)_n repeat motif found in T. moorii (Zardoya et al. 1996). The sequences available for locus Pzeb1 showed differences in the repeat in all species examined so far, and further indicate the presence of a poly T region following the microsatellite that also varies in length among alleles (Table 1). In eretmodines the repeat motives (GT)_nGA(T)_n and (GT)_nTTGC(T)_n were found.

Estimates of variability at the six microsatellite DNA loci within populations are shown in Table 2, and allele frequencies are presented in Appendix I. The total number of detected alleles per locus ranged from eight in Pzeb3 to 41 in TmoM5. The mean number of alleles (A) and expected heterozygosity (H_E) across populations are significantly smaller (Mann–Whitney U-test; P = 0.05) in the Tanganicodus populations than in the Eretmodus populations (e.g. A, 4.7–5.7 compared to 8.5–13; H_E, 0.41–0.49 compared to 0.74–0.82). At locus Pzeb3 all Tanganicodus populations are fixed for allele 313, which only occurs in a high frequency (> 0.500) in Eretmodus KOR. At locus UNH002 the Tanganicodus populations show either one or two alleles (160 and 162). Allele 160 is absent in the Eretmodus populations from KAM and KYE, and allele 162 is absent in Eretmodus from KOR (Appendix I).

Six of 42 single-locus tests for deviation from Hardy–Weinberg equilibrium gave significant results, but only three [TmoM11 in Tanganicodus (KAM) and Pzeb3 in Eretmodus (KAM and KIK)] remained significant when adjusted for table-wide significance by a sequential Bonferroni procedure. Linkage disequilibrium was observed in three out of 105 pairwise comparisons [E. cyanostictus KAM, TmoM5 and TmoM25 (P = 0.050), TmoM25 and Pzeb1 (P = 0.015); E. cyanostictus KYE, TmoM25 and Pzeb3 (P = 0.007)] but none of them remained significant after sequential Bonferroni correction.

mtDNA

The 198 DNA sequences comprised 70 mitochondrial haplotypes (Table 3 and Appendix II). The overall haplotype and nucleotide diversity were 0.947 and 0.040, respectively. Haplotype diversity was significantly smaller (Mann–Whitney U-test; P = 0.05) for the Tanganicodus populations than for the Eretmodus populations but not so for the nucleotide diversity (see Discussion). For these comparisons, two populations (Eretmodus KAP and Tanganicodus KIK) were left out due to small sample sizes. Two haplotypes (37 and 66) were shared among individuals classified as Eretmodus and Tanganicodus (Appendix II).

Population structure

Across all populations, differentiation tests based on microsatellites showed highly significant heterogeneity of allele frequencies (P << 0.001). Single locus pairwise tests were all significant except for eight comparisons [E. cyanostictus KAM and KIK (TmoM11, TmoM25, Pzeb1); E. cyanostictus KIK and KYE (TmoM25); T. cf. irsacae KOR and KAP (TmoM5, TmoM25, UNH002); T. cf. irsacae KAM and KAP (UNH002)]. Multilocus probabilities of heterogeneity between population pairs were all highly significant (P < 0.001; for T. cf. irsacae KOR and KAP, P < 0.01).

Based on the microsatellite data, estimates of overall θ_ST indicated significant levels of genetic variance among all populations (θ_ST = 0.250, P << 0.001). For the four Eretmodus populations the value was θ_ST = 0.127 (P << 0.001) and for the three Tanganicodus populations it was θ_ST = 0.240 (P << 0.001). All population pairwise estimates of multilocus θ_ST were significantly different from zero (Table 4). Our estimates of fixation indices are not substantially affected and remain significantly different from zero when loci TmoM11 and Pzeb3, which showed significant deviation from Hardy–Weinberg equilibrium in a few populations (see Table 2), are excluded.

Population differentiation tests based on mtDNA haplotype frequencies were highly significant for all pairwise comparisons (P << 0.001; E. cyanostictus KAM and KIK, P = 0.0015). Overall, estimates of θ_ST indicated significant levels of genetic variance among all populations (θ_ST = 0.263, P << 0.001). For the four Eretmodus populations the value was θ_ST = 0.089 (P << 0.001) and for the three Tanganicodus populations it was θ_ST = 0.526 (P << 0.001). Population pairwise estimates of θ_ST were all significantly different from zero (Table 4).

The amova revealed that 18.4% of the total variance in microsatellite allele frequency was due to between species heterogeneity (F_CT = 0.184) and 12.9% was caused by differences among populations within species (F_SC = 0.158). The remainder (68.7%) was due to variation within populations (F_ST = 0.312). For the mtDNA data, the corresponding values were 7.8% (F_CT = 0.078), 21.1% (F_SC = 0.228) and 71.1% (F_ST = 0.289, Table 5).

Population clustering and phylogenetic analysis

A majority-rule consensus NJ tree (based on 1000 bootstrap-replicated chord distance matrices) grouped most populations by species rather than geographical location (Fig. 2). All except one internal node were well supported (i.e. bootstrap values > 70%). For example, the node grouping all Eretmodus was supported by a bootstrap value of 91% (Fig. 2).

The NJ phylogeny of the 70 identified mtDNA haplotypes shows a basal split into two genetically distinct lineages (bootstrap values 94% and 86%; Fig. 1). These lineages were previously defined as the E. cyanostictus lineage C and the T. cf. irsacae lineage D (Rüber et al. 1999). Within lineage C two major clades were observed, one comprising all the E. cyanostictus from KYE (clade C-1, bootstrap value 100%) and another that includes all individuals of E. cyanostictus from the localities KAM and KIK (clade C-2, bootstrap value 75%). Clade C-2 also includes a haplotype shared by one individual of E. cyanostictus from KIK and one of T. cf. irsacae from KIK (37), and three haplotypes found in a few individuals of T. cf. irsacae from either KAM or KIK (38–40, Appendix II).

Within lineage D three major clades were found (bootstrap values 62%, 68% and 97%). Clade D-3 comprises four haplotypes (53–56) representing 24 individuals of T. cf. irsacae from KAM and KIK. Clade D-2 containes haplotypes 68 and 70 from three T. cf. irsacae (localities KAP and KOR). And finally, clade D-1 contains 12 haplotypes. In this clade, haplotypes 57–65 are found in E. cyanostictus from KAP and KOR; haplotype 66 is found in four E. cyanostictus from KOR, in nine T. cf. irsacae from KOR, and in 22 T. cf. irsacae from KAP; the remaining two haplotypes (67 and 69) are found in eight T. cf. irsacae from KOR and in one T. cf. irsacae from KAP. The average sequence divergence between individuals from lineage C and D was 6.20% ± 0.12% (net sequence divergence corrected for intrapopulation polymorphism was 4.45% ± 0.15%). Within lineage C and D the average sequence divergence between individuals was 2.56% ± 0.10 and 1.36% ± 0.03, respectively. Applying a molecular clock calibration based on cichlid control region sequences (5.6% per site per 10⁶ years; Nagl et al. 2000) on the net sequence divergence suggests that the two lineages (C and D) separated roughly 0.8 million years ago.

Demographic history

Both the Sign and the Wilcoxon tests gave no indication for heterozygote excess, and the allele frequency distributions indicated no deviation from an approximate L-shaped distribution in any of the populations. Thus, these results indicate that, according to the microsatellite data, none of the observed populations underwent a recent bottleneck. Fu’s test supported a demographic expansion of the Eretmodus populations with the exception of Eretmodus KOR (F_S-values ranged from −6.91 to −13.26; P ≤ 0.0001; Eretmodus KOR F_S = −2.79, P = 0.0718. All Tanganicodus populations showed positive F_S-values and are therefore compatible with population stasis (F_S-values ranged from 1.16 to 5.54; P > 0.750).

Microsatellite sequences

In eretmodine cichlids four out of 10 loci have undergone mutational events in the microsatellite repeats, compared to the species from which the microsatellites were obtained (Table 1). Differences in the repeat motif as found here underscore the need for more DNA sequence data to adequately assess genetic variability within and among species, and to understand the evolutionary processes of these markers (e.g. Zardoya et al. 1996). Such information may also prove essential to correct for size-homoplasies in estimating demographic population parameters such as structure, gene flow, and effective population size (e.g. Angers & Bernatchez 1997). Another illustration of this issue was given in a recent study that revealed extensive size homoplasy at locus Pzeb4 within and among 11 Lake Malawi cichlids species (van Oppen et al. 2000). These authors further cautioned against using population parameters based on the stepwise mutation model (SMM) if conformance to SMM for the loci used were not demonstrated. Our results indicate that loci TmoM11, Pzeb3, and UNH002, which exhibit perfect dinucleotide repeats in all the cichlid species studied so far, might be less affected by size homoplasy than other loci that show more complex mutations at the intra- and interspecific level (Table 1). Further comparative data of other cichlid species might help to distinguish loci that are more affected by size homoplasy than others.

Intra- and interspecific genetic variation

Genetic variation inferred from mtDNA sequences and microsatellites was consistently lower in the Tanganicodus than in the Eretmodus populations (Tables 2 and 3). Only the nucleotide diversity did not differ significantly between the two species due to haplotypes 39 and 40 found in T. cf. irsacae at KAM (Table 3) that clustered with Eretmodus haplotypes in clade C-2 (Fig. 1). Excluding these haplotypes from the analysis reduces the nucleotide diversity of the Tanganicodus KAM population to 0.0005 ± 0.001 resulting in a significantly lower nucleotide diversity in all the Tanganicodus populations in comparison to the Eretmodus populations (Mann–Whitney U-test; P = 0.05). The relatively high nucleotide diversity observed in Tanganicodus KIK, which was not included in the analyses (see Results), can also be explained by the presence of two haplotypes (37 and 38, found in three individuals) that clustered with clade C-2 (Fig. 1).

The observed species-specific differences in genetic diversity suggest that the two species might have had different evolutionary histories in terms of random genetic drift, variation of effective population size, or natural selection. We estimated N_E for the Eretmodus and Tanganicodus populations using the formula π = 4N_Eµ, where π is the nucleotide diversity and µ is the mutation rate assumed to be 2.8 × 10⁻⁸ (substitutions × base pair⁻¹ × year⁻¹) for the control region in cichlids (Nagl et al. 2000). For the Eretmodus populations the estimated N_E is in the order of 2–4 × 10⁵, whereas for the Tanganicodus populations it is in the order of 2–10 × 10⁴. Parker & Kornfield (1997) and Nagl et al. (1998) obtained estimates of either N_f or N_E of similar magnitude for several species of lakes Malawi and Victoria cichlids. The difference in the estimated effective population size between the two eretmodine species agrees with field observations by Hori et al. (1983), who studied the abundance and microdistribution of cichlids on a rocky shore in the northern part of the lake and recorded higher abundance of E. cf. cyanostictus (lineage A) than of T. irsacae (lineage A).

The mtDNA control region sequences revealed a relatively large amount of polymorphism in the Eretmodus populations in comparison to other studies on lacustrine cichlids. For example, Meyer et al. (1996) found an overall nucleotide diversity in the mtDNA control region of 0.0026 and 0.0037 in two Simochromis species (25 and 28 individuals studied) from Lake Tanganyika collected over a shoreline of 440 km. Bowers et al. (1994) found an overall nucleotide diversity of 0.0008 and 0.0050 in two species of the mbuna genus Melanochromis from the southern part of Lake Malawi (68 and 42 individuals studied). They also found that 75% of the populations were fixed for a single haplotype. A collection of all available Lake Malawi mbuna control region sequences revealed 58 control region haplotypes with a total of 46 segregating sites among 180 individuals from 34 species representing several mbuna genera [Parker & Kornfield (1997); see Table 3 for a comparison with the two species studied in this paper]. The low levels of mtDNA variation found in many Lake Malawi mbunas may be due to recent colonization of habitat patches from refugia populations in the recent past due to lake level fluctuations (van Oppen et al. 1997a; Arnegard et al. 1999; Markert et al. 1999; Sturmbauer et al. 2001). The level of expected heterozygosity based on microsatellite data found in the Eretmodus populations (0.72 average across all loci and all populations), is comparable with data for Melanochromis auratus (0.67; Markert et al. 1999), Labeotropheus fulleborni (0.83; Arnegard et al. 1999) and for four Pseudotropheus species (0.72 average over species; van Oppen et al. 1997a) from Lake Malawi. In the light of the findings from this and other studies, T. cf. irsacae is the genetically least polymorphic lacustrine cichlid species found so far. There is no evidence for a recent population bottleneck that may account for the lower genetic variation found in the Tanganicodus populations. On the other hand, we found evidence of a population expansion in Eretmodus, the species that shows higher genetic diversity. With the information at hand it is not possible to determine whether the higher genetic diversity in Eretmodus is only associated with the hypothesized expansion or if there are additional species-specific differences in life history traits and/or environmental factors that might affect their population sizes.

Population structure

Both microsatellite markers and mtDNA sequences revealed significant levels of genetic structure within the two sympatric eretmodine species — along 138 km of coastline for the Eretmodus populations and over a distance of 35 km for the Tanganicodus populations. The higher degree of population structuring observed in Tanganicodus may be attributed to the lower levels of genetic diversity, in both microsatellites and mtDNA markers within populations, that influence measures of population differentiation (e.g. Nagylaki 1998). Other factors such as dispersal ability and phylopatry may contribute to differences in population structure between the two species.

Among the 70 identified haplotypes, shared haplotypes were never found between more than two neighbouring populations. For example the Eretmodus populations at KAM and KIK with 15 and 24 haplotypes, respectively, separated by a distance of 7 km of rocky shore, shared only two haplotypes. In the Tanganicodus populations (Appendix II) haplotype 67 was observed in over 40% of the individuals from KOR, but not at KAP which is situated only 7 km away. Such a low frequency of shared mtDNA haplotypes among localities is likely to be an indicator for the expressed site fidelity of eretmodine cichlids, which may rapidly lead to genetic isolation among populations. Although the amount of shared mtDNA haplotypes may change with increasing sample size, our microsatellite results support a high degree of genetic isolation between populations.

Philopatry coupled with the patchy distribution of rocky habitats along the lake shores have been identified as important factors for population divergence among Lake Malawi mbuna cichlids (van Oppen et al. 1997a; Arnegard et al. 1999; Markert et al. 1999). The availability of suitable habitat patches as well as their distribution is strongly affected by lake level fluctuations. A recent study confirmed that lake level fluctuations in Lake Tanganyika, induced by tectonic and/or climatic events, have strongly affected the distribution and fragmentation of rocky habitat patches in the littoral (Cohen et al. 1997). Cohen et al. (1997) identified seven lake lowstand periods in the Northern Tanganyika basin that have been named and dated as follows: (f) 1.1 Ma; (d) 393–363 Ka (1 Ka = 1000 years); (c) 295–262 Ka; (b) 193–169 Ka; (a) 40–35 Ka; (x) 23 Ka; (y) 18 Ka. The rise of the lake to its present level was completed 12 Ka (A. Cohen, personal communication; see also Sturmbauer et al. 2001). The magnitudes of the lake lowstands below present lake level were given by these authors as 650–700 m for (f); 350 m for (d); 350 m for (c), 250 m for (b) and 160 m for (a). Gasse et al. (1989) give estimates of 300–400 m for the magnitude of the late Pleistocene lowstands (x) and (y). Other authors suggest them to be more likely in the order of 150–200 m (A. Cohen, personal communication). Lake level fluctuations in the Holocene were less severe in Lake Tanganyika (maximum 75 m; see Coulter 1991) compared to Lake Malawi where it has been reported that a lake level drop of 120 m occurred 200 years ago (Owen et al. 1990).

As they may cause fusion or isolation of populations, water level changes in shallow areas of the lake are likely to have severe effects on population structure. At steeply sloping shores, lake level fluctuations will mostly cause a vertical dislocation of the populations and, as in these zones rocky habitats often extend far into deep water, are less likely to promote cycles of fusions and isolations of populations (Sturmbauer et al. 1997; Sturmbauer 1998). This is the case for all localities analysed in this study. We provide evidence for significant population differentiation along a mostly continuous rocky shoreline that is only rarely interrupted by few small sandy bays and influent rivers, with areas of sand deposition that might act as barriers to gene flow, and that may not have been much affected by lake level fluctuations. A question that needs to be studied in more detail is whether the small sandy beaches observed in the study area act as effective dispersal barriers and significantly contribute to population differentiation in the rock-dwelling eretmodine cichlids, or whether divergence also occurs under parapatric conditions —, i.e. along a continuous habitat (Taylor et al. 2001).

Evidence for introgression

Throughout its narrow distribution range T. cf. irsacae occurs in sympatry with E. cyanostictus. Chord distances grouped the Eretmodus populations in a separate cluster (Fig. 2). The mitochondrial gene phylogeny (Fig. 1), on the other hand, shows that some individuals of T. cf. irsacae from KAM and KIK were placed within the E. cyanostictus lineage (clade C-2), whereas all individuals of E. cyanostictus from KAP and KOR were resolved within the T. cf. irsacae lineage (clade D-1).

The observed distribution of mitochondrial haplotypes in these two closely related species is incongruent with the morphology-based species assignment which clearly indicates their morphological distinctness. Shared ancient polymorphism as a result of incomplete lineage sorting or introgression during secondary contact may be responsible for this incongruence. At the present state of knowledge we favour the introgression scenario over incomplete lineage sorting. An argument against incomplete lineage sorting is the local occurrence of the mtDNA haplotypes found in lineages C and D that conflict with the species tree. In addition, given the average sequence divergence of 6.20% separating the two lineages, we consider it unlikely that retained ancestral haplotypes in Eretmodus at KOR and KAP would be identical or nearly identical to the Tanganicodus haplotypes at those localities. There are five fixed substitutions between lineage C and D haplotypes. Lineage D haplotypes show the following diagnostic substitutions: 40T, 123 T, 125 C, 248 A, and 294 C. Two of these sites show transversions (125 A/C and 248 T/A) which are diagnostic for lineage D haplotypes in regard to all other eretmodine lineages [based on a total of 288 eretmodine control region sequences; this study and Rüber et al. (1999)]. Therefore, we consider it more likely that the unique transversional mutations at sites 125 and 248 occurred along the branch grouping all lineage D haplotypes rather than it being present as a polymorphism in the shared ancestral gene pool from which E. cyanostictus, T. cf. irsacae and other eretmodine cichlids may have originated.

The possible mtDNA introgression cases we observe in our data are spatially restricted and asymmetric: complete introgression from Tanganicodus to Eretmodus at KOR and KAP, and partial introgression from Eretmodus to Tanganicodus at KIK and KAM. In contrast, the microsatellite allele frequency distributions indicate a limited degree of nuclear introgression, although allele 313 at locus Pzeb3 which is fixed in the Tanganicodus populations appears in a relatively high frequency in Eretmodus at KOR (see Results and Appendix I). Moreover, introgressed populations do not differ phenotypically from nonintrogressed ones. These observations, together with the genetical differentiation observed between sympatric Tanganicodus and Eretmodus populations (Table 4), suggests that hybridization might have taken place locally for a short period of time during secondary contact, and that there is no ongoing gene flow. Reinforcement may have contributed to the rapid evolution of reproductive isolation of the two sister species upon secondary contact, despite continuing sympatry (Butlin 1989).

The fixation of Tanganicodus mtDNA in the introgressed Eretmodus populations at KAP and KOR may have occurred either by chance (via founder effects or drift) or by selection. Although we have no indication that selection is responsible for the fixation of Tanganicodus mtDNA in the two introgressed Eretmodus populations, selective advantages of heterospecific mtDNA have been suggested as explanations of similar cases among salmonid species (Wilson & Bernatchez 1998 and references therein). The proposed direction of mtDNA introgression from Tanganicodus to Eretmodus at KAP and KOR seems at odds with the observation of higher genetic diversity and hence larger population sizes found in the latter populations. It is possible, however, that small Eretmodus founder populations colonized this area of the lake and established secondary contact with Tanganicodus populations, and that the presumed expansion of population size in Eretmodus happened after the interspecific replacement of mtDNA due to introgression.

There is increasing evidence, particularly from fishes, that mtDNA introgression between allopatrically diverged taxa in zones of secondary intergradation may be more common than previously thought (e.g. Dowling & deMarais 1993; Bernatchez et al. 1995; Wilson & Bernatchez 1998). However, it remains to be tested to what extent historic events of introgressive hybridization can explain patterns of shared haplotypes among species, or even genera, in other cichlid species flocks. Incomplete lineage sorting has been postulated for the lakes Malawi and Victoria cichlid flocks (Moran & Kornfield 1993, 1995; Parker & Kornfield 1997; Nagl et al. 1998, 2000). As in Lake Tanganyika, intermediate phenotypes are rare or absent in these lakes, suggesting that little or no hybridization is going on at the present time, possibly due to strong assortative mating (Seehausen et al. 1997; Knight et al. 1998; van Oppen et al. 1998). In contrast to Lake Tanganyika, the ages of these radiations are much younger, which makes incomplete lineage sorting more likely. Nevertheless, introgression caused by, for example, changes in water transparency, which may have occasionally broken down reproductive barriers among multiple species (see Seehausen et al. 1997), may in some cases be an alternative hypothesis.