Reticulate evolution became first apparent in phylogenetic reconstructions based on 1047 bp of the mitochondrial ND2 gene. The phylogeny of the ossified-group lamprologines (Fig. 2A) is largely congruent with previous reconstructions based on different taxon sampling 58, subdividing the "ossified group" into 4 – 5 clades depending on the tree-building algorithm used. A linearized tree analysis suggests that the majority of clades arose at a divergence level of about 7% (TrN+Γ distances; Fig. 3), which corresponds to major cladogenesis events in other Lake Tanganyika cichlid lineages 91011 and platythelphusid crabs 12. Their diversification may thus have been induced by the same environmental factor, most likely by a substantial drop of the lake level during a period of aridification in eastern Africa about 2.5 to 3 million years ago 13.

Despite the strength of the phylogenetic signal and the high resolution of reconstructed species relationships, the mtDNA phylogeny contains several groupings in striking conflict with morphological, ecological and behavioral similarities. Remarkably, all inconsistencies involve gastropod shell breeders. The first pair of eco-morphologically and behaviorally highly similar species resolved in different clades on the mitochondrial tree is Neolamprologus similis and N. multifasciatus (Fig. 2A). Males and females of both species are sufficiently small to enter gastropod shells – N. multifasciatus is the smallest Tanganyikan cichlid known to date (3.5 cm)- and both species are facultative shell-breeders, which arrange their nests by excavating sand craters around small accumulations of empty shells and live in family groups. The second group of morphologically and behaviorally similar species separated in the mitochondrial phylogeny includes Lamprologus ocellatus, L. meleagris, L. speciosus and N. wauthioni. These species are sexually monomorphic (see Table 1 for more information), small-bodied obligatory shell-breeders with a maximum size of about 6 cm that prefer areas with low densities of empty gastropod shells. In each species, the males defend territories and hold harems of two to five females for whom they bury shells in a way that only the opening of the shell remains accessible to a resident female 14. L. meleagris occurs sympatrically with N. wauthioni in the north, and with L. speciosus in the south of its distribution range. Despite their morphological and behavioral similarity, each of these four species clusters with different, eco-morphologically very dissimilar, species (Fig. 2A). Although contrasting with phenotypic data, the mitochondrial tree topology receives high bootstrap support, and a monophyletic clade of L. speciosus, L. meleagris, L. ocellatus and N. wauthioni requires at least 56 additional steps in a MP phylogeny (P < 0.001; Δ -ln L = 206.84, P < 0.001). Forcing N. similis and N. multifasciatus into a monophyletic clade did not significantly increase tree length (P > 0.05; Δ -ln L = 24.43, P > 0.05), but bootstrap support for the non-sister group relationship between N. multifasciatus and N. similis was high.

In contrast, species relationships inferred from a set of 199 polymorphic AFLP markers were consistent with predictions based on phenotypic trait similarities, as each of the two groups of eco-morphologically similar species was resolved as a monophyletic clade (Fig. 2B). Disparate mitochondrial and nuclear reconstructions were also obtained for the phylogenetic position of another shell-breeding species: Neolamprologus fasciatus clustered with L. callipterus and N. wauthioni in the mtDNA-phylogeny, but was placed ancestral to N. multifasciatus and N. similis by AFLP analyses. A constrained mitochondrial tree congruent with the nuclear placement of N. fasciatus would require significantly more evolutionary steps under both parsimony and likelihood criteria (at least 22 steps in MP, P < 0.001; Δ -ln L = 97.08, P < 0.001 in ML). Although the current data allow only limited inference of the phylogenetic history of N. fasciatus, we can definitely rule out a close affinity to the genus Altolamprologus, which has been suggested due to superficial morphological similarities 14.

Ancient hybridization and ancient incomplete lineage sorting are alternative explanations for the observed discrepancy between the mitochondrial phylogeny and relationships predicted by nuclear markers, phenotypes and behavior. Mismatches among gene trees can result from the differential assortment of ancestral polymorphisms, when intervals between successive branching events are too short for lineage sorting to be completed within each branch prior to the next split 1516. This phenomenon occurs, for example, when taxa speciate rapidly during adaptive radiation, and has been repeatedly reported from the cichlid species flocks in the East African Great Lakes 1517181920. In the mitochondrial phylogeny of the shell-breeding Lamprologini, the taxa affected by inconsistent mitochondrial and nuclear tree topologies (L. speciosus, L. meleagris, L. ocellatus and N. wauthioni; and N. multifasciatus and N. similis) have their most recent common ancestor near or at the base of the tree. Long interior branches and bootstrap support for many interior nodes in the mitochondrial phylogeny do not indicate a rapid burst of cladogenesis, such that there may be no major effect of ancient incomplete lineage sorting on the phylogenetic reconstruction.

Nuclear loci coalesce more slowly 21, and are therefore subjected to lineage sorting incongruence over longer periods. The combination of multiple independent loci, as implemented by the AFLP technique, is expected to overcome the problem of idiosyncratic lineage sorting at individual loci 222324. Additionally, the species relationships suggested by our AFLP data are supported by phenotypic and behavioral similarities, which makes it unlikely that the AFLP clades result from stochastic lineage sorting rather than reflecting actual relationships.

We contend that the incongruence between the mitochondrial phylogeny and multiple nuclear markers, phenotypes and behavior is best explained by ancient introgressive hybridization events. Furthermore, it is possible that hybridization was directly associated with the speciation of L. meleagris, N. wauthioni, L. speciosus, and N. multifasciatus (Fig. 4), and perhaps also N. fasciatus. Species status according to the biological species concept 25 of the presumed hybrid taxa is supported by sympatric occurrences of L. meleagris with L. speciosus in the south and N. wauthioni in the north of its distribution range, and the lake-wide, continuous distribution of N. fasciatus.

Although published mitochondrial phylogenies of the mouthbrooding tribes encompass representative species samples [910112627, and unpublished data), striking incongruencies with behavioral, ecological and morphological data were detected only in the tribes Tropheini 262829 and Eretmodini 30. In the tropheine genus Petrochromis, the incongruence was explained by parallel evolution of distinct eco-morphotypes 26, whereas disparate mitochondrial placements of color morphs and species in the tropheine genus Tropheus 2829 and in the tribe Eretmodini 20 were attributed to introgressive hybridization after secondary admixis. In contrast, diversifying hybridization has been proposed in studies of the substrate breeding Lamprologini, including a role of introgressive hybridization in the speciation of Neolamprologus marunguensis 31 and of Lepidiolamprologus nkambae, a non-shell-breeding member of the "ossified group" of lamprologines 8. Our results with the shell-breeding lamprologines provide a further example of potential evolutionary consequences of introgressive hybridization in another guild of lamprologine cichlids.

In two instances, species were resolved non-monophyletically. Altolamprologus calvus was paraphyletic in relation to A. compressiceps in the AFLP tree, and Lepidiolamprologus sp. "meeli-boulengeri" was paraphyletic with respect to L. meeli in the AFLP tree, while the two taxa were placed in different clades by the mitochondrial sequences. The taxonomy of these species is not well resolved, in that populations with intermediate phenotypes and phenotypic divergence across geographic distances have not yet been adequately addressed (Schelly, pers. comm.). The here identified incongruence between the phylogenetic and current taxonomic resolution must not be dismissed as a possible consequence of persistent ancestral polymorphism in the genetic data, but pinpoints a need for more detailed taxonomic and molecular genetic work on these taxa.

Evidence for past hybridization affecting the evolutionary history of the "ossified group" of lamprologines obviously raises the question whether there is also evidence for ongoing hybridization among shell-breeding cichlids. Indeed, within a small area of suitable habitat containing several L. callipterus nests in southern Lake Tanganyika (near Wonzye, Zambia), we collected four specimens that could not be identified to species level but were clearly lamprologine cichlids. Although lamprologines hybridize readily in captivity, this is the first proof of viable natural hybrids in Lake Tanganyika. Mitochondrial sequences determined that the mothers of the hybrids were members of the N. brevis/N. calliurus – clade (Fig. 2A). Hybrid phenotypes and data from six microsatellite loci suggested that two of the hybrids were sired by L. callipterus and the other two by N. fasciatus (Fig. 5A). Morphological similarity indices (Fig. 5B), and a principal component analysis of morphometric and meristic measurements (Fig. 5C) further supported the phenotypically and genetically inferred identity of the parental species. Furthermore, territorial males of the two putative paternal species L. callipterus and N. fasciatus are too large to enter empty shells (see Fig. 1), and release their sperm over the entrance of their mate's shell to fertilize her eggs. Dispersion of sperm into adjacent shells by water currents and wave action can bypass prezygotic isolation between species sharing a shell-bed, and result in trans-specific fertilization and hybridization, when post-zygotic isolation is incomplete.

Within the lamprologines, the inferred hybridization partners are only distantly related, as the split between the mitochondrial lineages including N. brevis/calliurus and the lineage containing L. callipterus and N. fasciatus is estimated as 3.80 (± 1.33) – 4.58 (± 1.60) million years before presence (TrN+Γ distance, 9.08 ± 1.33%; gamma-corrected amino acid distance, 4.44 ± 1.55%) based on a molecular clock for the ND2 gene of cichlid fish 11. The two N. fasciatus hybrids and one of the L. callipterus hybrids shared a mitochondrial haplotype, whereas a different haplotype was detected in the second L. callipterus hybrid (Fig. 1). Hence, the four hybrid individuals represent at least three independent hybridization events between the genetically and morphologically highly divergent species, suggesting that hybridization among shell-breeding species is not uncommon. Introgressive hybridization is commonly associated with loss of diversity through extinction or homogenization of species 323334, and the maintenance of species integrity in the face of hybridization suggests that introgression between lamprologine species is constrained by low hybrid fitness. Accidental hybridization among shell-breeders is independent of prezygotic species barriers and unresponsive to reinforcement of pre-mating isolation, such that hybridization rates may not decrease over long divergence times despite possible costs through reduced hybrid fitness.

Representatives of 31 lamprologine cichlid species, including the outgroup taxa Variablichromis moorii, Telmatochromis vittatus and Julidochromis ornatus, were collected during several sampling trips to Lake Tanganyika from 1992 to 2004 or obtained via the aquarium trade. Fin clips were taken from fresh specimens and preserved in 96% ethanol. Voucher specimens are available from the authors.

We analyzed 1047 bp of the entire mitochondrial ND2 gene for 69 individuals, representing 30 sampled species and four putative hybrids. When available, previously published sequences were used 18. Total DNA-extraction, polymerase chain reaction (PCR) and chain termination sequencing followed standard protocols 942. For both, PCR and chain termination reaction sequencing, we used the primers MET, ND2.2A, TRP 43 and ND2.T-R 11. Sequences were visualized on an ABI 3100 Sequencer (Applied Biosystems). All sequences are available from GenBank under the accession numbers listed in Additional File 1.

Alignment of DNA sequences was performed using the Sequence Navigator software (Applied Biosystems). Throughout the presented phylogenetic analyses, hierarchical likelihood ratio test statistics were calculated using the program Modeltest 3.06 44 to evaluate appropriate models of molecular evolution for model based tree reconstructions, and phylogenetic reconstruction based on NJ, ML and MP criteria were calculated with the PAUP* program package (version 4.0b5) 45. In a first step, a NJ tree including all 69 taxa {substitution model TrN+I+Δ, 46; proportion of invariable sites (I), 0.4147; gamma shape parameter (α), 0.9162} was used to select a representative subset of 48 taxa to minimize computation time for subsequent ML and maximum parsimony (MP) analysis. The reduced data set was then used for NJ and ML analyses (substitution model TrN+I+ Δ; I, 0.4083; α, 0.8659; base frequencies: A, 0.2530; C, 0.3531; G, 0.1308; T, 0.2630), and for MP tree searches with transitions at third codon positions weighted 3:1 with respect to transversions, based on the estimated transition/transversion (TI/TV) ratio of 2.9627. Transversions at third codon positions of fourfold degenerate amino acids were weighted 2:1 with respect to transitions, according to the estimated TI/TV ratio of 2.1140. Synonymous transitions at first positions of leucine codons were treated like transitions at third codon positions. Bootstrapping (1000 pseudo-replicates for NJ and MP; 100 pseudo-replicates for ML) and quartet puzzling 47 (25 000 random quartets) were applied to estimate support of the obtained topologies. Phylogenetic relationships were also estimated by a Bayesian method of phylogenetic inference using MRBAYES 3.0b4 48. Posterior probabilities were obtained from a 1,000,000 generation Metropolis-coupled Markov chain Monte Carlo simulation (four chains; chain temperature, 0.2) with parameters estimated from the data set. Trees were sampled every hundred generations and the first 10% of all trees were excluded as burn-in to allow likelihood values to reach stationary.

To derive a relative dating of diversification events in the "ossified group" of lamprologines, we computed a linearized tree, using the program LINTRE 49. To test for constancy of the rate of base substitution among all taxa we performed a branch length test implemented in the program LINTRE based on TrN+ Δ distances (α = 0.2841). Neolamprologus leloupi showed a significantly deviating substitution rate at the 1% level and was therefore excluded from the calculation of a clock-constrained tree. Mean and standard deviation of average pairwise TrN+ Δ distances were calculated to achieve a relative dating of diversification events. Since no fossil record is available as calibration point for the "ossified group" of lamprologines, a molecular clock for the cichlid ND2 gene based upon gamma corrected amino acid distances 11, calculated with the program TREE-PUZZLE 5.0 50, was applied to obtain absolute datings for cladogenesis events.

The significance of differences between alternative tree topologies was evaluated by Kishino-Hasegawa tests 51 and Shimodeira-Hasegawa tests 52 under maximum parsimony and likelihood criteria, respectively.

For AFLP fingerprinting, whole genomic DNA was extracted from 47 specimens representing 25 species of the "ossified group" of lamprologines and the three outgroup species Julidochromis ornatus, Telmatochromis vittatus and Variabilichromis moorii applying proteinase K digestion followed by protein precipitation with ammonium acetate and ethanol precipitation of DNA. Restriction digestion of 100 ng of genomic DNA was performed in a total volume of 50 μl using 0.5 μl MseI (10 units/μl, New England Biolabs), 0.25 μl EcoRI (200 units/μl, New England Biolabs), 5 μl enzyme buffer (10×), 0.5 μl BSA (100×) and high performance liquid chromatography (HPLC) water, and incubation for three hours at 37°C. For ligation of adaptors, 1 μl EcoRI-adaptor (50 pmol/μl), 1 μl MseI-adaptor (5 pmol/μl), 1 μl T4 ligase buffer, 0.2 μl T4 DNA ligase and 6.8 μl HPLC water were added to the product of the restriction digestion and incubated over night at 37°C. The ligation products were subsequently diluted to 180 μl using HPLC water. PCR was carried out in two steps as recommended by 53. Preselective amplifications were performed with 3 μl of the diluted ligation product, 0.4 μl each of EcoRI and MseI preselective primers (10 μM), 2 μl 10 × MgCl₂ buffer, 2 μl 10 × dNTP mix (10 μM) and 0.6 μl Taq DNA polymerase (5 units/μl, BioTherm™) in a final volume of 20 μl. The preselective primers consisted of the adaptor primer sequence with a single selective nucleotide at the 3' end (EcoRI-pre: A, MseI-pre: C). The preselective PCR used the following temperature profile: 2 min at 72°C followed by 20 cycles of 20 sec at 94°C, 30 sec at 56°C, and 2 min at 72°C, then a holding step at 60°C for 30 min. PCR products were diluted 1:10 for selective amplification. The following recipe was used for selective amplifications with primers extending 3 bp beyond the adaptor sequence: 1 μl of the diluted preselective PCR product, 6.3 μl HPLC water, 0.8 μl 10 × dNTP mix (10 μM), 1 μl 10 × MgCl₂ buffer, 0.4 μl Taq DNA polymerase (5 units/μl; BioTherm™), 1 μl selective MseI primer (10 μM) and 1 μl selective EcoRI primer (1 μM) labeled with the fluorescent dye FAM. The primer combinations used for selective amplification were: EcoRI-ACA/MseI-CAT, EcoRI-ACT/MseI-CAT, EcoRI-ACT/MseI-CAA, EcoRI-ACT/MseI-CAC. The temperature profile for the selective PCR was as follows: 2 min at 94°C followed by 10 cycles with 20 sec at 94°C, 30 sec at annealing temperature, which decreased in each cycle by 1°C from 65°C to 56°C, and 2 min at 72°C. The PCR continued for 25 cycles with 20 sec at 94°C, 30 sec at 56°C, and 2 min at 72°C, followed by a holding step at 60°C for 30 min. All amplifications were performed on a GeneAmp PCR system 9700 (Applied Biosystems). Selective amplification products were visualized on an ABI 3100 automated sequencer (Applied Biosystems) along with an internal size standard (GeneScan-500 ROX, Applied Biosystems).

Raw fragment data were analyzed using GENEMAPPER (version 3.7, Applied Biosystems). Presence or absence of peaks (presumed to represent homologous fragments) was scored by eye (in order to avoid misinterpretations inherent to automated fragment scoring) within a range of 100–500 bp and assembled as a binary (1/0) matrix by GENEMAPPER. In a few cases, fragments were scored as missing data when character states could not be determined unambiguously. Matrices from the different primer combinations were assembled into one data set.

A neighbor joining tree based on pairwise genetic distances 54 obtained from the presence/absence matrix was calculated with the program TREECON 1.3b 55. Bootstrap values from 1000 pseudo-replicates were used as a standard measure of confidence in the reconstructed tree topology.

Similarity-indices between hybrid individuals and eight possible parental species were calculated based upon the presence or absence of 13 qualitative characters (Additional File 2). Furthermore, a principal component analysis based upon 13 morphometric (Additional File 3) and 8 meristic (Additional File 4) characters 5657 in the four hybrids and the presumed parental species was performed (n = 5 individuals per species). Since the first principal component of the morphometric dataset proofed to be mainly influenced by the standard length, it was not used for analysis. Thus, we plotted PC1 of the meristic data against PC2 of the morphometric data. Loading factors are depicted in Additional Files 5 and 6.

Microsatellite markers were used to support morphological inferences of species identity of the parents of the collected hybrid individuals. The four hybrid individuals, one to three representatives of 15 species of obligatory and facultative shell-breeding lamprologines and populations of the three candidate parent species (as identified by morphological criteria) were genotyped at six microsatellite loci with low intraspecific variation in lamprologine cichlids (Pzeb3 58, TmoM25, TmoM27 59, UNH154 60, UNH855 and UNH952 61; see Additional Files 7 and 8). Populations of the three candidate parental species, Lamprologus callipterus (n = 19), Neolamprologus brevis/calliurus (n = 24) and N. fasciatus (n = 21) were sampled at the same locality as the hybrid individuals (Wonzye, S 08°43', E 31°08'). Mitochondrial sequences placed all four hybrid individuals in the clade containing N. brevis and N. calliurus, which are very closely related and often synonymized since sub-adult individuals of these two species cannot be distinguished in the field (Figs. 2, 3, 4).

Total DNA-extraction and amplification of microsatellites followed standard protocols 62. Fragments were visualized on an ABI 377 automated sequencer (Applied Biosystems) using forward primers labeled with fluorescence dyes FAM, TET or HEX, and the internal size standard Genescan-500 TAMRA (Applied Biosystems).