Using DNA sequences of the most variable section of the control region (358–360 bp following the tRNA Pro; Lee et al. 1995), we searched for genetic traces of lake level fluctuations in populations of cichlid fishes which were caused by a temporary fusion of several previously isolated populations and/or incipient species and their subsequent split. To assess the consequences of the most recent minimum lake level over 400–600 m in Lake Tanganyika and 250–500 m in Lake Malawi, we analyzed populations at steeply sloping shores close to the lake basin(s) that are likely to have remained unaffected by lake level fluctuations, as well as populations from shallow regions of the lakes which would have been most severely affected by lake level fluctuations. To investigate the consequences of the dry-up in Lake Victoria, we used all available sequence data of individuals from distant localities in the lake and in surrounding water bodies representing potential refuge areas for colonizers after the dry-up (Nagl et al. 2000).
For each lake, we plotted all presently available mitochondrial haplotypes on their location of origin and related them to the inferred paleoshorelines during the last major lowstand or to potential refuge areas in the case of Lake Victoria, to compare the pattern with the predictions. The species assignments were disregarded, and the phylogenies were taken as mitochondrial gene trees bearing phylogeographic information. For Lake Tanganyika, we selected taxa of the genera Tropheus, Eretmodus, and Ophthalmotilapia that exhibited strong phylogeographic structuring pointing to a limited ability for dispersal across sand or mud coasts and open water (Brichard 1978; Sturmbauer and Meyer 1992; Sturmbauer and Dallinger 1995; Verheyen et al. 1996; Sturmbauer et al. 1997). Of 243 DNA sequences available for the genus Tropheus, 29 haplotypes (60 individuals) from 15 localities relevant to assessment the consequences of the latest dramatic lake level fluctuation were selected, some of which were published earlier (Sturmbauer and Meyer 1992; Sturmbauer et al. 1997; see map in fig. 1A).
The Tropheus samples represent four mitochondrial lineages from three regions of the lake and are presently assigned to three nominal species (Tropheus moorii, Tropheus kasabae, and Tropheus brichardi). Of 50 published DNA sequences ofEretmoduscf.cyanostictus(lineage A as defined in Ru¨ber, Verheyen, and Meyer 1999), three mitochondrial haplotypes from four localities formed a monophyletic cluster that was used for our analyses of divergence levels. Of 39 DNA sequences (50 individuals; Hanssens et al., unpublished data) available for the genus Ophthalmotilapia, five sequences of Ophthalmotilapia nasutafrom six localities formed a phylogeographically informative cluster that was used in the analysis. The EMBL accession numbers of the DNA sequences ofEretmoduscf.cyanostictusrelevant to this paper are Z97412 (Bemba), Z97411 (Luhanga), X90612 (Ngombe 92/1), and X90632 (Cape Kabogo 92/ 40). Those of Ophthalmotilapia boopsare Z95983— Z95995, those of Ophthalmotilapia heterodontaare Z95996—Z96001, those ofOphthalmotilapia nasutaare Z96002—Z96015, and those ofOphthalmotilapia ventralis are Z96016—Z96020. The accession numbers for Tropheus are AJ295902–AJ295924, in addition to those already published.
The analyzed DNA sequences of Lake Malawi rock-dwelling cichlids, termed mbuna,were published previously (Bowers, Stauffer, and Kocher 1994; Parker and Kornfield 1997) and comprise 26 haplotypes (50 individuals of 24 nominal species of 13 locations) from both the central steeply sloping core region of the lake and the southernmost shallow sections of the lake (see map in fig. 2A). Due to incomplete lineage sorting among mbunaspecies, identical genotypes were frequently found among individuals of different species, even if hybridization did not occur among sympatric species (Albertson et al. 1999). The DNA sequences available for Lake Victoria haplochromines (Meyer et al. 1990; Nagl et al. 2000) include 56 specimens of 24 described plus 9 undescribed species. All analyzed taxa belong to a monophyletic assemblage (called lineage VC in Nagl et al. 2000) in which the majority of individuals are Lake Victoria endemics, and a few others stem from surrounding water bodies.
Of the localities analyzed from Lake Victoria, three were situated at the northeastern shore, four at the southern edge of the lake basin near Mwanza, and four additional localities situated in surrounding rivers and lakes (see map in fig. 3A). All three data sets used in this analysis were tested for their relative rates of base substitutions using the computer program PHYLTEST (Kumar 1996). Therefore, we defined three monophyletic lineages, the first comprising Tropheus (29 taxa), the second comprising Lake Malawimbuna(45 taxa), and the third comprising Lake Victoria cichlids (15 taxa). Three consecutive analyses were performed so that each of the three lineages was once used as the reference group. For each data set, a mitochondrial phylogeny was constructed using the parsimony and neighbor-joining algorithms of PAUP*, version 4.64d (Swofford 1998). We then selected one of the most-parsimonious trees that in all three analyses was most similar to the neighbor-joining tree to plot genotypes and their locations of origin in the form of minimum spanning trees, since they best illustrate all genotypes descending from the same ancestral lineage at the time of their most recent admixis (see figs. 1–3). Genotype clusters relevant to assess the extent of lake level fluctuations had to contain identical or very closely related genotypes found both in habitats with steep slopes that are most likely also available during periods of low lake level and in habitats that are only colonizable during high-water-level periods. In some cases, habitats were situated at opposite shores of a lake, separated from each other by deep water. For each haplotype cluster, the geographic extension of the latest sec ondary admixis and dispersal event was derived from the haplotype distribution. To compare the relative ages of haplotype clusters within and among the lakes, average mutation differences among haplotypes belonging to a phylogeographically informative cluster were calculated. This was done by defining an ancestral haplotype for each cluster and subsequently calculating arithmetic means and standard deviations of all mutation counts between the ancestral and the derived haplotypes. Identical haplotypes were counted according to their frequencies.
An absolute age estimate was attempted by calibrating the molecular divergences by a new age estimate for Lake Malawi using average genetic divergences of two basal lineages of the Lake Malawi cichlid species flock. The geological history of the Lake Malawi basin is complex, so the age of the lacustrine habitat can only be dated in the form of a minimum-maximum estimate. While the geological age is assumed to be 4–5 Myr, the lake was most likely almost or completely dry for several thousands of years in the late Pleistocene from 1.6 MYA until maximally 1 MYA and minimally 570,000 years ago. We thus estimated the age of its species flock to be between 570,000 years and 1 Myr (Delvaux 1995), assuming that the emergence of the ecological diversity of Lake Malawi cichlids did not predate this period, be cause most suitable lacustrine habitats were lacking in a shallow and swampy lake.
To calibrate the mutation rate of the 359-bp segment of the control region, we calculated the average genetic distance among two ancient Lake Malawi cichlid lineages (Meyer et al. 1990). We compared published DNA sequences of 26mbuna(Kocher et al. 1993; Bowers, Stauffer, and Kocher 1994; Parker and Kornfield 1997), 13utaka(Meyer et al. 1990; Lee et al. 1995; Parker and Kornfield 1997), and five additional species sequenced by us for this study: Aulonocara jacobfreibergi, Fossorchromis rostratus, Melanochromis caeruleus ‘yellow’, Nimbochromis livingstoni,and Nimbochromis venustus(table 1). After performing a relativerate test (Takezaki, Rzhetsky, and Nei 1995), average pairwise Kimura distances were calculated among all taxa of each lineage. Two mbunasequences and three utakasequences were excluded because of different substitution rates.