No evidence for parallel sympatric speciation in cichlid species of the genus Pseudotropheus from north-western Lake Malawi


  • Present address : M. J. H. Van Oppen, Australian Institute of Marine Science, PMB No. 3, Townsville MC, Queensland 4810, Australia.

 G.F. Turner, Department of Biological Sciences, University of Hull, Hull HU6 7RX, UK. e-mail:


Abstract To test the hypothesis of parallel speciation by sexual selection, we examined length variation at six microsatellite loci of samples from four sites of four to six putative species belonging to two subgenera of rocky shore mbuna cichlids from Lake Malawi. Almost all fixation indices were significantly different from zero, suggesting that there is presently little or no gene flow among allopatric populations or sympatric species. Analysis of variance indicated that genetic distances among allopatric populations of putative conspecifics were significantly lower than among sympatric populations of heterospecifics. The topology of trees based on distance matrices was also largely consistent with the hypothesis that the putative species are monophyletic and have thus not evolved in parallel in their present locations. If parallel speciation does occur in Malawi cichlids, it may be on a larger spatial scale than investigated in our study.


When discussing the explosion of new forms in the fossil record of the early Cambrian Burgess Shale, Gould (1989) coined a memorable metaphor to describe his view of the primacy of contingency in biological evolution: if one could rewind the tape of life, each re-run would produce a dramatically different outcome. The implications of this view of history are profound, suggesting that the organisms that dominate the world today are largely the beneficiaries of good fortune, rather than good design. Parallel and convergent evolution of ecological adaptations has long been known. However, there is now evidence that similar environmental conditions not only produce similar forms but essentially the same new species can arise repeatedly in different places (reviewed by Johannesson, 2001), as may have been the case with sticklebacks in British Columbia, Canada (Rundle et al., 2000). It might seem even more surprising that parallel speciation may be driven by sexual selection, which is often considered to be more capricious and less directly influenced by the environment than ecological selection. However, we propose that it may be worth investigating this possibility in the cichlid fishes of the East African lakes.

In Lake Victoria, it is common to find two or more ecologically and morphologically similar sympatric species that differ clearly in male courtship colour (Seehausen, 1996). Strikingly, the same colour combinations re-occur in many different genera, and even in different lakes (Seehausen et al., 1999). It seems that there have been repeated switches between a small number of different colour combinations during the radiation of these fishes.

Several studies have provided evidence for sympatric speciation among cichlid fishes in African Lakes (Schliewen et al., 1994; Seehausen & van Alphen, 1999; Shaw et al., 2000). Seehausen & van Alphen (1999) observed that the males of many Victorian species have a largely blue breeding dress, while the males of other anatomically very similar and thus probably closely related species have red and yellow courting males. Many of these pairs of species have largely sympatric ranges, and it was suggested that these might have speciated sympatrically in situ (Seehausen & van Alphen, 1999). These findings have been interpreted as evidence for repeated parallel evolution of similar male breeding colours during sympatric speciation by sexual selection (Seehausen, 2000). It seems that in some of these species pairs, male breeding colour may be entirely responsible for reproductive isolation (Seehausen et al., 1997; Seehausen & van Alphen, 1998). Thus, it would not seem implausible that some of these populations may freely interbreed with other similarly coloured but independently derived populations in the event of secondary contact. This would be parallel speciation by sexual selection.

Population structure among Lake Victoria cichlids has not been studied in detail. However, the rocky shore mbuna cichlids of Lake Malawi are split by habitat discontinuities into many genetically isolated populations (van Oppen et al., 1997a; Arnegard et al., 1999; Markert et al., 1999). This would provide opportunity for repeated parallel evolution of species. In Lake Malawi, there are many sympatric taxa that differ in male breeding colour and little else, and in many cases it has been shown that these are reproductively isolated species (van Oppen et al., 1998). As in Lake Victoria, many of the same colour combinations reappear in different genera or species complexes (Ribbink et al., 1983; Konings, 2001). Reinthal & Meyer (1997) suggested that mitochondrial sequences indicated that four populations of Pseudotropheus (Tropheops) spp. in the southern part of Lake Malawi formed a clade with one northern population. The other three northern populations formed a separate clade. This suggests parallel evolution of morphological and ecological traits, although not of colour patterns. However, analysis of a larger data set has cast doubt on this conclusion (Parker & Kornfield, 1997).

In this study, we attempted to test the hypothesis that colour variation among allopatrically and sympatrically occurring populations and putative species reflects parallel speciation at four different sites separated by large habitat discontinuities along a 50-km stretch of the north-western shore of Lake Malawi.

Materials and methods


Fishes were sampled from four sites on the western shore of Lake Malawi: Ruarwe, Mara Rocks, Cape Manulo and Nkhata Bay (Fig. 1). Nkhata Bay and Cape Manulo are near to the opposite ends of a 33-km long region of rocky shore broken by a number of relatively small sandy beaches. Ruarwe and Cape Manulo are about 25 km apart: most of this distance is made up of the 20-km wide Usisya Bay which is sandy, apart from the small shallow rocky promontary in the middle. The Mara Rocks lie a few kilometres off this headland, isolated from the mainland by a deep trench. As our observations have shown that the P. zebra at the promontary resemble those of Ruarwe in colour, and not those of Mara Rocks, it seems that this trench is a major barrier to mbuna dispersal.

Figure 1.

Study area, showing the four sample areas. Mara Rocks is a group of islands about 1 km off the shore. The coasts are largely rocky and suitable habitats for the study species, apart from the large sandy beaches (shaded areas) to the north and south of Mara Rocks. The broken lines are 300 and 500 m depth contours. Scale bars are approximate.

We sampled species belonging to two subgenera of the genus Pseudotropheus. The classification of these fishes is rather confused (see for example Stauffer et al., 1997; Condé & Géry, 1999; Konings, 2001), and we follow Eschmeyer's on-line catalogue of fishes ( The subgenus Maylandia, which is sometimes treated as a full genus, contains the species of the P. zebra complex, also known by the junior synonym of Metriaclima. These species have terminal mouths and long slender bicuspid and tricuspid teeth, which they use to comb loose material from among filamentous algae growing on rock surfaces, although they also feed on plankton. The species of the subgenus Tropheops are also known as the P. tropheops complex or the genus Tropheops. The outer teeth of P. (Tropheops) are set together in the underslung mouth to form a cutting blade, which is used to shear off filamentous algae from the rock surface. By contrast to the clearly adaptive morphological differences between mbuna genera or subgenera, species within a genus or subgenus vary only slightly in morphology but are usually clearly different in colour. We sampled three species of each subgenus (Table 1): several of these show geographical variation, particularly in male colour and in presence or absence of rarer female colour morphs (Table 2).

Table 1.  Sample size and collection site for P. (Maylandia) and P. (Tropheops) species. Abbreviations for each site and species are used hereafter in tables, figures and text. See ‘Methods’ for details.
SiteSample size
P. (M.) zebra (Mz) P. (M.) ‘gold’ (Mg) P. (M.) callainos (Mc) P. (T.) ‘olive’ (To) P. (T.) ‘band’ (Tb)P. (T.) ‘ mauve’ (Tm)
Nkhata Bay (N)1045012111072119
Cape Manulo (C) 53Not collected 65 30Not collected 37
Mara Rocks (M) 80Not collected 61 53Not collected 53
Ruarwe (R) 5940 57 6124Not analysed
Table 2.  Geographical variation in colour for putative species of P. (Maylandia) and P. (Tropheops) . Abbreviations are given in Table 1 . Absence of a morph from a sample site does not necessarily mean that it does not occur there, rather that no positive record exists.
SitePutative species
P. zebra (Mz) P. ‘gold’ (Mg) P. callainos (Mc) P. ‘olive’ (To) P. ‘band’ (Tb) P. ‘ mauve’ (Tm)
  1. * Populations considered as heterospecific by Ribbink et al. (1983 : P. ‘zebra chilumba’) and Konings (2001 : Metriaclima fainzilberi ).

  2. † Population considered as heterospecific ( P. ‘zebra pearly’) by Ribbink et al. (1983 ).

  3. ‡ Population considered as heterospecific by Ribbink et al. (1983 : P. ‘tropheops weed’) and Konings (2001 : Tropheops ‘weed’).

Nkhata BayMale : blue with dark bars,  blue dorsal fin, black  chin (BB), occasionally  OB or O morphs Male : yellow with  dark bars Male : blue (B),  occasionally  W morph Males : yellow-  brown Males : yellow and  blue Males : blue
Female : brown with dark  bars (BB); orange with  dark blotches (OB);  bright orange (O) Female : brown with  dark bars (BB);  orange with dark  blotches (OB) Female : blue (B),  white (W) or  blotched (OB) Females : silvery  grey with rows  of black spots Females : sandy/ grey with rows  of dark spots Females : yellow
Cape ManuloBB, OB as Nkhata BayNot collectedB, W as Nkhata BayAs Nkhata BayNot collectedAs Nkhata Bay
Mara RocksMale : blue with faint bars,  blue dorsal fin, yellow  chin (BB) *Not collectedB, W morphs  recordedAs Nkhata BayNot collectedMales : blue  with yellow  chest
Female : BB as Nkhata  Bay; O dull  orange     Females : yellow
RuarweMale : blue with darkbars,  dark horizontal band in  dorsal fin, yellow chin  (BB) *As Nkhata BayBoth sexes all whiteAs Nkhata Bay,  although males  recorded as  slightly more  intensely  colouredAs Nkhata Bay, but  males recorded  as less intensely  colouredNot analysed
Female : BB, OB  as Nkhata Bay      

In July 1995 and 1996, we collected 24–120 individuals of each species from each site (Table 1), as described by van Oppen et al. (1997a, 1998). We were unable to collect enough P.‘gold’ and P.‘band’ from the two least accessible sites, Mara Rocks and Cape Manulo. We did not include P.‘mauve’ from Ruarwe in the analysis, because we discovered during the second collecting trip that this taxon actually represented a cryptic species pair at this site, which we had not distinguished initially. In the present study, the samples from Nkhata Bay are the same as those reported by van Oppen et al. (1998).

Data analysis

Samples were screened for variation at six polymorphic microsatellite loci as described by van Oppen et al. (1997b). The six loci (Kellogg et al., 1995; van Oppen et al., 1997b) include four perfect dinucleotide repeats (Pzeb1–3 and UNH002) and two imperfect dinucleotide repeats, Pzeb4 and Pzeb5. Allele frequencies, expected (He) and observed (Ho) heterozygosities were calculated using genepop 3.1 (Raymond & Rousset, 1995). Samples were tested for linkage disequilibrium and departure from Hardy–Weinberg equilibrium by the Markov chain method. Heterogeneity in genotype distribution for all loci and all pairwise comparisons, was tested based on an assumption of no differentiation. Nonamplifying ‘null’ alleles were found from pedigree analysis of six pairs of P. zebra adults from Nkhata Bay (van Oppen et al., 1998). Null alleles were found in four of the six loci. We therefore assumed that null alleles were present in other species as well. The frequency of null alleles was estimated following the approach of Markert et al. (1999). Further, to assess the effects of the loci that have null alleles on the significance of population differentiation estimates, weighted FST statistics over the two loci with no null alleles were obtained and tested as described below.

As both drift and mutation probably influence differences between putative species pairs, and those between populations are probably mostly influenced by drift (Slatkin, 1995), we followed two approaches in investigating the amplitude of genetic differentiation between populations and taxa. First, the computer program Arlequin 1.1 (Schneider et al., 1997) was used to calculate pairwise fixation indices, based on allele frequency variation using an amova framework to estimate weighted FST statistics (θ) over all loci (Weir & Cockerham, 1984; Excoffier et al., 1992). Secondly, we calculated ρ, which is based on the stepwise mutation model (SMM) and takes into account the differences in sample size between populations and differences in variance between loci (Goodman, 1997). Indeed, it has been empirically shown that calculating ρ using this approach is unaffected by the differences in sample size and variability as long as samples are moderately large (i.e. n ≥ 50) (Ruzzante, 1998). Thus, we only calculated pairwise ρ estimates for comparisons where each sample comprised at least 50 individuals.

The significance of genetic subdivision was assessed using 1000 permutations in both Arlequin and RSTCALC. To correct for multiple simultaneous comparisons, sequential Bonferroni corrections were applied to all pairwise tests using a global significance level of 0.05 (Rice, 1989).

Genetic divergence among populations and taxa was estimated by two approaches because of recent debate on model-specific distance estimators for microsatellite loci (e.g. Takezaki & Nei, 1996; Goodman, 1997; Angers & Bernatchez, 1998). Therefore, we first quantified Cavalli-Sforza & Edwards's (1967) chord distance (DCE) and Nei's distance using GENEDIST. Secondly, we calculated Goldstein et al.'s (1995) delta-mu squared (δµ2) pairwise distances using RSTCALC (Goodman, 1997). Phenograms based on Cavalli-Sforza and Edwards' distance (DCE), Nei's distance and δµ2 were constructed using the NEIGHBOUR program with the neighbour-joining algorithm (Saitou & Nei, 1987). A further phenogram was estimated using a maximum likelihood algorithm (CONTML). Support for the tree nodes was assessed by 1000 bootstraps of gene frequencies using the SEQBOOT program and compiled using the CONSENSE program. All these programs are included in the phylip computer package, version 3.5c (Felsenstein, 1993).

From the matrix of pairwise genetic distances, comparisons were classed according to whether they were between conspecifics in allopatry, and heterospecifics in sympatry or allopatry. Heterospecific comparisons were further divided into comparisons among members of the same subgenus and species of different subgenera. These five classes of comparisons were then tested for differences in mean genetic distances, using one-way analysis of variance (anova) (SPSS). For this analysis, we used Nei's distance and also Slatkin's (1995) linearization transformation of FST applied to θ, as this is believed to render its behaviour more clock-like.


Genetic diversity and tests of disequilibrium

All species and populations compared were highly genetically diverse showing high heterozygosities at most loci (Table 3). A total of 60, 44, 43, 22, 17, and 10 alleles were observed at Pzeb1, UNH002, Pzeb2, Pzeb3, Pzeb4, and Pzeb5, respectively. Significant deviations from Hardy–Weinberg equilibrium in the form of heterozygote deficits were present in some populations for most loci. Heterozygote deficiencies were assumed to be mostly as a result of nonamplifying alleles, because in a pedigree test using P. zebra, true-breeding null alleles were found in four of the six loci (van Oppen et al., 1998). Few significant heterozygote deficits were found in any species at the loci that had no null alleles in the pedigree test, whereas the four loci where null alleles were detected exhibited frequent heterozygote deficiencies. The analysis carried out only with the loci that had no null alleles (Pzeb4 and 5) showed that only 11 θ estimates of 171 were nonsignificant: Cape Manulo P. zebra vs. P. callainos from both Nkhata Bay and Cape Manulo and P. zebra from Ruarwe; Nkhata Bay P. callainos and P. zebra vs. P. zebra from Ruarwe; Ruarwe P.‘gold’ vs. Mara Rocks P. zebra and both populations of P.‘band’; Cape Manulo P.‘olive’ vs. conspecifics at Nkhata Bay and also P.‘band’ at Ruarwe; and finally the two populations of P.‘band’. As 93% of significant θ estimates remained significant when loci with null alleles were excluded from the analysis, the presence of null alleles in four of the loci clearly had a negligible effect on the estimates of population differentiation. Exact tests for genotypic linkage disequilibrium confirmed the absence of physical linkage at these loci, as previously reported (van Oppen et al., 1997a).

Table 3.  Number of alleles (NA), observed ( Ho ) and expected ( He ) heterozygosity. Significant deviations from Hardy–Weinberg equilibrium are under H0  = HWE and H1  = heterozygote deficit.
Site/speciesPzeb 1 Pzeb 2 Pzeb 3 Pzeb 4 Pzeb 5 UNH002
  1. * Sequential Bonferroni adjusted P for FIS significantly different from zero. Null = estimated frequency of null alleles calculated for cases with significantly nonzero FIS . For other abbreviations see Table 1 .


At locus Pzeb5, allele 129 occurred at high frequencies in several populations or was fixed. Statistically significant differences (P < 0.05) in genotype frequency distributions were demonstrated in most pairwise comparisons for all other loci (data not shown), whether comparing sympatric putative species or allopatric putative conspecific populations. Allele frequency distributions could be very different. Take, for example, T.‘mauve’ at two different sites for locus UNH002. The three most abundant alleles in the Mara Rocks sample, 199, 215 and 217, had a combined frequency of almost 45% in that site while they were amplified at a frequency of <3% from the same putative species at Cape Manulo.

Microsatellite genetic differentiation

All comparisons between pairs of allopatric populations of putative conspecifics gave fixation indices significantly different from zero (Table 4). This was irrespective of whether the index was based on the infinite alleles model (θ) or the SMM (ρ). Not surprisingly comparisons of allopatric populations of different species were all different as well. These figures indicate that there is little or no gene flow between these allopatric populations.

Table 4.  Pairwise fixation indices for all putative species pairs and conspecific populations in allopatry. Fixation indices were computed following the infinite allele model ( θ ) lower diagonal, and the stepwise mutation model standardized Slatkin RST ( ρ ) ( Goodman, 1997 ) upper diagonal. When information from all six loci were used in the analysis, all indices were significantly different from zero ( P  < 0.05, after Bonferroni correction), apart from the comparison of RST for P. ( Tropheops ) ‘band’ and ‘mauve’ at Nkhata Bay (asterisked). Figures in bold typeface are comparisons of allopatric conspecifics. For abbreviations see Table 1.
McN 0.0370.2680.1050.1410.1100.2920.0460.190NC0.5850.221NC0.4160.3340.619NC0.273NC
McC0.077 0.1610.0290.0580.0860.1930.0100.122NC0.5300.161NC0.3370.2600.548NC0.191NC
McM0.1370.070 0.1230.0190.1280.0100.0990.045NC0.3410.066NC0.0940.0710.390NC0.044NC
McR0.0770.0660.063 0.0500.0900.1640.0190.129NC0.4830.145NC0.2780.2260.541NC0.181NC
MzN0.0870.0740.0810.057 0.0650.0290.0230.028NC0.3020.020NC0.1170.0550.346NC0.045NC
MzC0.0670.0270.0890.0580.024 0.1460.0390.051NC0.3870.114NC0.2530.1510.505NC0.092NC
MzM0.0880.0380.0360.0950.0280.036 0.1210.035NC0.2490.042NC0.0590.0140.306NC0.020NC
MzR0.0680.0480.0870.0320.0250.0220.042 0.075NC0.4100.069NC0.2250.1430.458NC0.119NC
TmN0.1450.0850.0680.1510.0850.0830.0480.096 NC0.2550.048NC0.1340.0420.359NC0.004*NC
TmC0.1490.1110.0670.1680.0910.1000.0480.1050.030 NCNCNCNCNCNCNCNCNC
TmM0.1090.1230.1290.1890.1310.1120.0920.1370.1060.098 0.379NC0.2190.2580.283NC0.230NC
ToN0.1610.1940.2060.1680.1670.1670.1660.1370.2070.1740.206 NC0.1300.0450.438NC0.042NC
ToC0.1510.2020.2030.1860.1660.1760.1650.1540.2020.2010.1830.030 NCNCNCNCNCNC
ToM0.1510.1550.1520.1600.1360.1390.1210.1180.1550.1240.1540.0530.030 0.0650.284NC0.110NC
ToR0.1200.1540.1480.1660.1320.1380.1150.1240.1460.1290.1270.0670.0590.031 0.363NC0.022NC
MgN0.1540.1160.0800.1770.0880.1130.0510.1190.0810.0860.1300.2320.2180.1590.165 NC0.351NC
MgR0.0640.0930.0910.1190.0610.0730.0360.0740.1130.1140.1050.1780.1760.1450.1210.103 NCNC
TbN0.0970.0730.0360.1290.0660.0710.0660.0720.0380.0300.0750.1520.1380.1020.0820.0490.040 NC

All comparisons of between sympatric species also differed significantly from zero using θ, whereas for ρ, only one comparison was nonsignificant, that between T.‘band’ and T.‘mauve’ at Nkhata Bay (ρ = 0.005, P > 0.05). This supports previous findings (van Oppen et al., 1998) that sympatrically occurring putative species of Pseudotropheus are reproductively isolated, even when distinguished on little more than the colour of mature males and in some cases females. In addition, fixation indices were significantly different from zero for all comparisons between allopatric populations of different putative species, indicating that gene flow among them is very low or does not occur at all.

Population relationships

Contrary to the predictions of the parallel speciation model, comparisons of pairs of allopatric populations of putative conspecifics showed significantly lower mean genetic distances than pairs of sympatric species of the same subgenus (Fig. 2; Table 5). This pattern was evident irrespective of whether the analysis was carried out with Slatkin's linearized FST or Nei's distance. Also, there was no indication that sympatric congeneric populations were more similar than allopatric populations of different species of the same subgenus, which would be expected if there was any significant gene flow among sympatric species now or in the recent past. Likewise, sympatric populations of different subgenera showed similar pairwise distances to allopatric populations of different subgenera.

Figure 2.

Contrary to the predictions of the parallel speciation model, comparisons of allopatric populations of putative conspecifics (group 1) show lower genetic distances than sympatric populations of species of the same subgenus (group 2). Also, there is no indication that sympatric consubgeneric populations (group 2) show lower distances than allopatric populations of different species of the same subgenus (group 3), which would be expected if there was any significant gene flow among sympatric species now or in the recent past. Likewise, sympatric populations of different subgenera (group 4) or allopatric populations of different subgenera (group 5) show very similar distances. Similar results are found for comparisons of Slatkin's linearized FST (a) or Nei's genetic distance (b). Shown are means and 95% confidence intervals. For Slatkin's FST , means and confidence intervals were calculated on square-root transformed data. Statistical analyses are presented on Table 5 .

Table 5.  Statistical comparisons among categories of genetic distance measures, expressed as probabilities of accepting the null hypothesis given by post-hoc pairwise comparisons using Tukey's HSD. Means and confidence intervals of these data are presented in Fig. 2 . Numerical designations of columns indicate the same categories as are given in full on row titles. Below diagonal : square-root transformed values of Slatkin's linearized FST . This transformation was required to eliminate significant heteroscedasticity (transformed data: Levene's test: F4,166  = 1.70; P  = 0.152, n.s.). Testing the null hypothesis of no difference in FST among categories of comparisons was carried out by one-way analysis of variance on the transformed data ( F4,166  = 11.85; P  < 0.001). Above diagonal : Nei's genetic distances. Overall anova : F4,166  = 9.05; P  < 0.001. Data not significantly heteroscedastic: Levene's test: F4,166  = 1.51; P  = 0.202, n.s. *: P  < 0.05; ***: P  < 0.001.
Population comparison12345
1. Allopatric, same species0.049*0.014*<0.001***<0.001***
2. Sympatric, different species, same subgenus0.012*0.9960.6600.605
3. Allopatric, different species, same subgenus<0.001***>0.9990.110.045*
4. Sympatric, different subgenus<0.001***0.5760.295>0.999
5. Allopatric, different subgenus<0.001***0.4320.076>0.999

A similar analysis carried out on RST values found no significant differences among any of the categories (anova: F4,100 = 1.86, P = 0.123). The analysis of RST figures had lower statistical power than the analyses of FST and Nei's D, as the four population samples with 40 or fewer individuals were omitted from this analysis, reducing the number of comparisons by almost 40%. However, the rank order of the means of the categories was roughly the same, with allopatric conspecifics having the lowest distance values and sympatric heterospecifics of the same subgenus the second lowest.

Although poorly resolved, the phenograms constructed with the neighbour-joining algorithm using chord distance (DCE) or Nei's distance (Fig. 3a,b) or the maximum likelihood phenogram (Fig. 3c) depicted several clusters of conspecific populations, within the subgenus Tropheops, supported by nodes with relatively high (>75%) bootstrap values. Only the tree based on Nei's distance (Fig. 3b) exhibited clusters of sympatric heterospecific populations: two nodes clustering pairs of sympatric taxa, and two further nodes joining groups including sympatric heterospecific taxa. In all cases, these clusters comprised populations of P. zebra and P. callainos. However, bootstrap support for these clusters was weak, ranging from 18 to 45%. If each of the species had a single origin, a fully resolved tree would contain 13 nodes uniting only conspecific populations. The neighbour-joining phenogram based on DCE recovered 10 such nodes, and the maximum likelihood phenogram 9. On the other hand, if all speciation had taken place in situ in the present habitats, a maximum of 15 nodes should unite exclusively sympatric populations. More conservatively, if it is assumed that parallel speciation had occurred exclusively within subgenera, up to 11 nodes could be expected to unite clusters of all-sympatric populations. The neighbour-joining phenogram based on Nei's distance recovered two such nodes but with low bootstrap support, whereas neither of the other methods produced any at all.

Figure 3.

Neighbour-joining and maximum-likelihood unrooted phenograms estimating relationships among 19 putative species and/or conspecific populations of Lake Malawi mbuna cichlids based on (a) Cavalli-Sforza and Edwards' chord distance ( DCE ), (b) Nei's distance and (c) maximum-likelihood algorithm. See Table 1 for key to site and species names. Percentages of a thousand bootstrap values are given along branches for nodes with at least 75% support.

Trees constructed from the δμ2 distance matrix depicted clusters that were supported by very low bootstrap values and generally did not support either the parallel or monophyletic model. In fact, they resolved several nodes that made little biological sense, clustering allopatric populations of P. (Maylandia) with P. (Tropheops) species that are morphologically and ecologically distinct. Exclusion of Pzeb4, which was known a priori not to conform to SMM (van Oppen et al., 2000), did not improve the resolution of the trees (data not shown).


Our results provide no support for the parallel sympatric speciation hypothesis for the surveyed taxa and at the geographical scale under consideration. Although the tree nodes are mostly poorly resolved, probably as a result of the small number of loci used in this study (Takezaki & Nei, 1996), we believe the study has provided evidence for the monophyletic origin for all of the P.‘olive’ populations studied and for the P.‘mauve’ populations south of Usisya Bay. The overall support for monophyly of species emerges from the anova on linearized FST and on Nei's distances, and also from the neighbour-joining phenogram based on chord distances and the maximum likelihood tree. Only the neighbour-joining tree based on Nei's distances is more ambiguous, but even here, the strongly supported nodes are consistent with monophyletic species origins among P. (Tropheops), while nodes suggesting parallel in situ speciation among P. (Maylandia) are much more weakly supported. These results therefore do not suggest that parallel sympatric speciation has been a common event in the diversification of the cichlid populations we have investigated, at least not within the habitat patches they occupy in north-western Lake Malawi.

Can our results rule out parallel sympatric speciation in these taxa? One possible interpretation might be that parallel speciation has occurred in situ, but that gene flow among allopatric populations of conspecifics has been sufficiently high that all traces of their independent origin have been erased. This is impossible to refute, but it seems less parsimonious than the alternative, particularly considering the high fixation indices among allopatric conspecific populations. Rocky shore mbuna are generally sedentary, both as adults and juveniles, avoid open water and only reluctantly cross open sand (see Ribbink et al., 1983; Rico & Turner, 2002; references therein). Previous studies have indicated that significant population structuring may be caused by sandy bays of a few hundred (van Oppen et al., 1997a) or even, when a cold inflowing stream is present, tens of metres wide (Rico & Turner, 2002). In the present study, most populations were separated by sandy bays many kilometres wide. The Nkhata Bay and Cape Manulo populations, although more than 25 km apart, lie at opposite ends of largely unbroken rocky coast. Gene flow is more likely between these two sites, which are the only ones where we have no records of differences in colour among the sampled taxa (Table 2).

We cannot rule out the possibility that a few of the taxa in our study could have diverged in situ, or at least have exchanged genes with sympatric heterospecific populations relatively recently. The best candidates for this might be among the Maylandia species, where there are hints of similarities among sympatric P. zebra and P. callainos in the tree based on Nei's distances. Furthermore, investigation of relationships on a large spatial scale may yet reveal cases of parallel speciation in Lakes Malawi and Victoria.

Had we found evidence of apparent monophyly of sympatric species pairs, could we have made a strong case for parallel speciation? One alternative explanation might be introgression rather than common ancestry. Contradictory claims have been made for the geographical mode of origin of sympatric species pairs of sticklebacks (Gasterosteus spp.) in recently deglaciated lakes in British Columbia, Canada (Taylor & McPhail, 2000). A mitochondrial DNA (mtDNA)-based phylogeny provided what appeared to be evidence for monophyly of some species pairs within lakes and thus for sympatric speciation (Taylor & McPhail, 1999). However, a microsatellite phylogeny appears to suggest an allopatric ‘double invasion’ mode of speciation. The apparent monophyly suggested by mtDNA is now believed to be the result of introgression among these young taxa, which can often leave a strong imprint on mtDNA which is a single locus with a four-fold smaller effective population size than the nuclear genome. It is therefore conceivable that other studies that have reported evidence for sympatric speciation based on mtDNA data alone (e.g. Schliewen et al., 1994; Shaw et al., 2000) may give a different picture if based on multilocus nuclear markers. There have been many reports of fertile and viable cichlid hybrids in laboratory studies (e.g. Crapon de Caprona & Fritzsch, 1984; Seehausen et al., 1997) and there is morphological evidence for hybridization among wild Lake Victoria cichlid species (Seehausen et al., 1997) and molecular evidence for ancient introgression among Lake Tanganyika cichlids (Rüber et al., 2001). Incomplete lineage sorting of mitochondrial and other markers has been well-documented in Lake Malawi cichlids (e.g. Parker & Kornfield, 1997; Takahashi et al., 2001), as would be expected if introgression had been common among divergent lineages.

What if introgression had led to apparent monophyly of sympatric taxa when assessed from many unlinked nuclear loci? This could only mean that introgression was so frequent as to cause massive gene flow across the entire genome. Might this mean that these taxa could be considered conspecific irrespective of their prior history? Given enough time, the gene pool should become homogenized, unless there was strong disruptive selection against intermediates (Clarke et al., 1996). However, nearly neutral alleles might persist for a considerable time at high frequencies in a large population, whereas they might be less likely rise to high frequency de novo. Such populations could retain polymorphisms at both male courtship trait loci and female preference loci. Under conditions of poor water clarity, differences in male colour might not be easily perceived either by conspecifics or predators, and they might be nearly neutral. Under appropriate conditions, such as increased water clarity, disruptive sexual selection could be restored (Seehausen et al., 1997). Re-establishment of the same reproductive isolating barriers following a period of introgression has been termed ‘respeciation’ (Turner, in press). This would mimic the conditions assumed by several models of speciation by sexual selection, where populations initially contain several common alleles for male courtship traits and for female preferences for these traits in linkage equilibrium (Wu, 1985; Higashi et al., 1999). Although these assumptions have been criticized as unrealistic and more stringent assumptions suggested (e.g. Turner & Burrows, 1995), this may have been unduly pessimistic.

Finally, it is appropriate to mention that several genetic measures, which take allele size into account and are linear with time, have been developed for phylogenetic inferences based on microsatellite-generated distances. However, Angers & Bernatchez (1998) empirically showed that, as suggested by Takezaki & Nei (1996), the use of distances that are independent of mutation models and that have lower sampling errors should be more efficient in depicting tree topologies among closely related taxa. Our results provide further empirical support that the use of methods such as the chord distance is more reliable than δμ2 distance in resolving population relationships, providing higher confidence values in tree topology, better clustering concordance and higher components of genetic variance among population groups. This is particularly relevant when a small number of loci have been used and the divergence among taxa is recent.


This study of genetic differentiation among sympatric putative species and allopatric populations of mbuna cichlids from Lake Malawi provides evidence for the following conclusions. (1) Heterogeneity in gene frequencies and highly significant fixation indices were observed among almost all populations tested, suggesting minimal or no gene flow. (2) Colour similarities of allopatric taxa in the study area generally reflect recent-shared ancestry, rather than in situ parallel origin. (3) The use of a genetic distance that makes no assumption of mutational process (such as DCE) is more appropriate in depicting tree topologies when the number of loci used is small, and their mutational properties are unknown. (4) The fact that only 7% of the significant θ estimates became nonsignificant when loci with null alleles were excluded from the analysis suggests that null alleles have negligible effects on population structure estimates.


This work was funded by the NERC UK grant GR3/9443. We thank the government of Malawi for permission to carry out this research, the Department of Fisheries for providing facilities, and S. Mapila, O. Kachinjika, M. Banda, S. Chiotha, O. Msiska, I. Côté, M. Chiumia and H. Ngulube for their assistance, and O. Seehausen, J. Turgeon, B. Angers, J. Blais and M. Taylor for valuable comments on earlier versions of this manuscript.