Introduction

Adaptive radiation—the rapid generation of exceptional adaptive diversification within a lineage—is considered one of the central explanations for the diversity of life on earth (Schluter 2000; Gavrilets and Losos 2009). Ecological opportunities play an important role in adaptive radiation, especially when an ancestral species invaded an unutilized niche early in the radiation. The species then diversifies, as new ecological niches provide evolutionary opportunities with relaxed selection or strong disruptive or directional selection (Schluter 2000; Eldredge et al. 2005; Kapralov and Filatov 2006). It is well accepted that divergent natural selection is the primary mechanism of adaptive evolution, and this can be evidenced at the molecular level (Schluter 2001, 2009; Rogers and Bernatchez 2007). Well-known empirical examples of adaptive radiation include: Darwin’s finches, Hawaiian honeycreepers (Schluter 2000), cichlid fishes in East African great lakes (Salzburger et al. 2007) and cichlid fishes in the Nicaraguan crater lakes (Elmer et al. 2010b). The genetic basis of adaptive radiation greatly attracts the interest of biologists (Barrier et al. 2001; Kapralov and Filatov 2006; Jeukens et al. 2009) because, among other reasons, it can give insight to the genetic basis of evolution with reduced phylogenetic and geographical noise.

Cichlid fishes are an ideal model system to study the genetic basis of adaptive radiation. First, species richness is extremely high compared with other adaptive radiations; in the great lakes of East Africa almost 2000 species have evolved from a common ancestor within the past few million years (Meyer et al. 1990; Salzburger et al. 2002; Verheyen et al. 2003; Elmer et al. 2009). Second, the age of the adaptive species flocks range among lakes, allowing for comparisons across different temporal scales. The oldest extant radiation began more than 2 million years ago in Lake Tanganyika (Meyer et al. 1990; Meyer 1993; Salzburger et al. 2005; Seehausen 2006). In Lake Victoria, more than 500 endemic species, the renowned ‘superflock’ of cichlid fishes, evolved within the past 100,000 years (Meyer et al. 1990; Verheyen et al. 2003; Abila et al. 2004; Elmer et al. 2009). Third, a great amount of phenotypic and morphological diversity has arisen with adaptation to new ecological niches. Specifically, cichlids have repeatedly evolved parallel or convergent body shapes and colorations suited to similar but often independent environments (Kocher et al. 1993; Ruber et al. 1999; Koblmuller et al. 2004; Hulsey et al. 2008; Elmer et al. 2010b). Yet many cichlid species remain extremely genetically similar (Sturmbauer and Meyer 1992). Thus, they can be regarded as “natural mutants” (Meyer et al. 1993; Kuraku and Meyer 2008). Recent studies have identified multiple genes that contribute to the adaptive radiation of cichlids by using non-candidate gene (Gerrard and Meyer 2007; Elmer et al. 2010a) and candidate gene approaches to seek the genetic basis of traits related to the visual sensory system (Sugawara et al. 2002; Spady et al. 2005; Seehausen et al. 2008; Hofmann et al. 2009), parental care behavior (Summers and Zhu 2008), reproductive evolution (Gerrard and Meyer 2007), coloration (Salzburger et al. 2007) and jaw morphology (Terai et al. 2002b; Kijimoto et al. 2005; Albertson and Kocher 2006). Fourth, a growing amount of molecular data from African cichlids (Watanabe et al. 2004; Kobayashi et al. 2009; Salzburger et al. 2008; Lee et al. 2010) and neotropical cichlids (Elmer et al. 2010a) are available by which to investigate and examine the molecular evolution of genes that contribute to the adaptive radiation of cichlid fishes. The recent availability of molecular data from neotropical cichlids (Fan, Elmer and Meyer, in prep.) provides reliable outgroup information (Zardoya et al. 1996; Farias et al. 1999) for the tests of positive Darwinian selection detection in African cichlids, since the short divergence time between African and neotropical cichlids is relatively short (85.1 and 40.5 million years, based on gondwana fragmentation and fossil record) (Genner et al. 2007) compared to the evolutionary divergence (>100 million years) between cichlid fishes and other model system fishes (Steinke et al. 2006).

The African cichlid adaptive radiation is famous for the extremely variable morphologies that have arisen and diversified (Meyer 1993). Candidate genes inferred from zebrafish have proven successful at elucidating patterns of molecular evolution and adaptive diversification in cichlid fishes (e.g. Terai et al. 2003; Sugie et al. 2004; Salzburger et al. 2007). In the present study, we analyze the evolution of a candidate gene related to morphological changes and skin development in fishes: the epithelial cell adhesion molecule (EPCAM or CD326). EPCAM belongs to the cell adhesion molecular (CAM) family (Baeuerle and Gires 2007; Trzpis et al. 2007), which not only play a role in cell adhesion, but also in cell proliferation, migration, and differentiation. These processes are known to be fundamental in morphogenesis (Trzpis et al. 2008a). For example, a recent study indicates that the mutant in this gene display defects in both epithelial morphologies and integrity in zebrafish embryo development (Slanchev et al. 2009) though its role in ocular epithelial cells remains unknown (Forrester et al. 2010). We tested the role of Darwinian selection in the molecular evolution of presumably functional changes in the EPCAM gene across diverse lineages of cichlid fishes, based on the ratio of non-synonymous to synonymous mutations (K a/K s ratio, or ω). The ratio of K a/K s is a strong signal of positive, or directional, Darwinian evolution and has been used extensively to identify and analyze the role of selection in gene evolution in fishes (e.g. Dann et al. 2004; Gerrard and Meyer 2007; Elmer et al. 2010a).

Using a candidate gene approach based on EPCAM sequences from neotropical and African cichlid lineages, we considered the following hierarchy of hypotheses. First, given that African great lake cichlids show extremely high morphological diversity, we tested for evidence of positive selection on the molecular evolution of EPCAM in the African clade as a whole. Given that this analysis indicated a strong positive signal of positive selection on EPCAM, we then sought in more detail the timing of this selection and the nucleotide regions under selection. Thus, second, we tested the non-neutral molecular evolution in each of the derived lineages of African cichlids (recent evolution) and the ancestral lineages (earlier evolutionary pressures). If EPCAM was only recently under positive selection then we should identify patterns in derived lineages. However, if EPCAM was involved in earlier diversification, then we predicted to find positive selection on the ancestral sequences.

Materials and Methods

The full length protein-coding nucleotide sequence of the EPCAM gene from the stickleback (Gasterosteus aculeatus) genome (transcript ID: ENSGACT00000003469; 921 bp) was downloaded from the Ensembl database (Hubbard et al. 2002) and used as a query sequence to search the NCBI EST database (Boguski et al. 1993), with the BLAST (Altschul et al. 1990) search restricted to the family Cichlidae. The EPCAM gene is composed of nine exons according to the Ensembl database (Hubbard et al. 2002). The EST sequence of EPCAM was downloaded for each cichlid species for which it was available and a single EST sequence per species was assembled by CAP3 (Huang and Madan 1999) using default parameters when multiple ESTs were available for one species. EPCAM sequences from two neotropical crater lake cichlid species Amphilophus amarillo and Amphilophus sagittae (unpublished data) were included as the outgroup sequences.

The assembled EST sequences were aligned in ClustalX using default parameters (Thompson et al. 2002). After alignment, we identified the open reading frame (ORF) in the assembled cichlid ESTs by comparing with EPCAM protein-coding sequence from stickleback. Non-coding regions were excluded from further analysis. To exclude estimation bias caused by partial sequences, we used only full length protein-coding nucleotide sequences of EPCAM in the further analysis. Based on incomplete tilapia genome sequences that are publicly available (Kocher, unpublished data http://cichlid.umd.edu/blast/, database: RRS5KB-SCAFF.e60.c0.p60), only one copy of the EPCAM gene can be found in cichlids so our analyses will not be complicated by the asymmetrical evolution of paralogues.

The best molecular substitution model for EPCAM was selected by a hierarchical likelihood ratio test after testing in Modeltest version 3.7 (Posada and Crandall 1998). We reconstructed a phylogenetic tree using a maximum likelihood approach with PhyML version 3.0 (Guindon et al. 2009) given an HKY model (Hasegawa et al. 1985) of molecular evolution and gamma-distributed rate variation among sites. The robustness of the topology was tested with the highest likelihood using 100 bootstrap cycles. Trees were visualized in Figtree version 1.3.1 (http://tree.bio.ed.ac.uk/software/figtree/).

The improved branch-site model (Zhang et al. 2005) by codeml in the PAML package Version 4.4 (Yang 2007) was employed for the positive selection analysis at broad groupings of lineages. The codon-based model is more sensitive to detect positive Darwinian selection on particular lineage comparing with other average sites methods (Yang and Nielsen 2002). We first compared the signal of positive selection between different lineages: (i) African cichlids versus neotropical cichlids and (ii) Lake Victoria ‘superflock’ Haplochromis species versus all other cichlids. The branches of interest were assigned as foreground branches in the branch-site model.

We conducted further analyses to detect whether there is a signal of positive selection in different stages of cichlid evolution: in the derived or ancestral lineages. We used the branch-site model to separate the branches in a phylogenetic tree into two classes: foreground branches (i.e., branches of interest) and background branches. The user-specified foreground branch is tested for a signal of positive selection in comparison with a null model that assumes that signals of negative (class 0) and neutral selection (class 1) on codons. The alternative model is that the foreground branches will show a signal of positive selection relative to the background branches showing negative or neutral selection. To avoid local optima in the maximum likelihood integrations, three different initial omega values (ω = 0.5, ω = 1, ω = 2) were used to evaluate the parameter estimation. The improved branch-site model assumes positive natural selection can only involve a small number of codons, which is a robust method to detect positive selection in recent diverged species (Zhang et al. 2005). The likelihood ratio test (LRT) was used to determine the statistical significance of the signal of positive selection by comparing the likelihood difference of a gene sequence under the null and alternate models (Yang and Nielsen 2002; Zhang et al. 2005). The LRT method has been shown to be both accurate and robust in simulation studies (Anisimova et al. 2001).

Given that the signal of positive selection can be affected by only a small portion of the codons in a gene (Golding and Dean 1998), positively selected sites were inferred using a Bayes empirical Bayes (BEB) method, which reduces sampling error in small data sets (Yang et al. 2005) and therefore improves accuracy and robustness for our analysis. BEB probabilities greater than 0.75 at single codon positions were considered significant values.

Results

Cichlid EMCAP Genes and Phylogeny Construction

The dataset of cichlid ESTs currently publicly available on NCBI is mainly composed of sequences from African cichlid species. Using stickleback full length cDNA of the EPCAM gene as query sequence, we assembled EPCAM sequences from the African basal haplochromine Astatotilapia burtoni, which is a riverine and lacustrine species (Salzburger et al. 2008), the Nile tilapia Oreochromis niloticus, which is an African riverine habitat species that has recently invaded many lake habitats (Lee et al. 2010), and three African Lake Victoria ‘superflock’ cichlids Haplochromis chilotes, Haplochromis sp. ‘Matumbi hunter’, and Haplochromis sp. ‘red tail sheller’ (Watanabe et al. 2004; Kobayashi et al. 2009) (Supplementary Table 1). Neotropical representatives were two species of Midas cichlid, Amphilophus amarillo and Amphilophus sagittae from Nicaraguan crater lake Xiloá (accession number: JN391522, JN391523). The assembled sequence from A. burtoni was excluded from further analyses since it did not reach our length criterion (451 bp in the coding region) and would therefore bias the selection analyses.

Maximum likelihood phylogenetic reconstruction of the EPCAM gene produced a topology with high supporting values at branch nodes (Fig. 1). The topology was in agreement with current understanding about the evolutionary relationships and geographical distribution of cichlids (Meyer et al. 1990; Genner et al. 2007). Two divergent clades were identified: one for the neotropical cichlids and the other for the African cichlids. Within the African clade, tilapia is basal to the monophyletic tribe Haplochromini.

Fig. 1
figure 1

The phylogeny of African and neotropical cichlids based on the EPCAM gene. The symbol ** and * indicate the bootstrap values equal to 100 or at least 84. The numbers of non-synonymous/synonymous mutations are listed in bold (not italic) under the branches. The ancestral sequences of all internal nodes were build under the M8 (beta & w) model in PAML

Full size image

Test of Positive Selection in Major Cichlid Clades

First we tested for positive selection on the EPCAM gene between African (Oreochromis + Haplochromis) and neotropical (Amphilophus) cichlids using the modified branch-site model (Zhang et al. 2005). Significant positive selection was detected in the evolution of the EPCAM gene sequence in the African cichlid clade (ωF = 6.08, P < 0.001) (Table 1). EPCAM sequences were identical in both species of Amphilophus and therefore showed no sign of selection within the neotropical clade.

Table 1 Parameter estimations of the branch-site model for three evolutionary hierarchies of cichlid fishes: total clades, derived lineages, and ancestral lineages
Full size table

Second, we tested for a signal of positive selection on the EPCAM sequences exclusively from the clade of three Haplochromis cichlids from Lake Victoria and identified a strong signal of positive selection (ωF = 37.03, P < 0.001, Table 1) on the EPCAM gene in the haplochromine lineage.

Test of Positive Selection in Derived Cichlid Lineages

Each cichlid species was specified as a foreground branch to test for positive selection separately in each derived lineage. We found significant positive selection on the Haplochromis sp. ‘red tail sheller’ branch (ωF = 999, P < 0.001, Table 1). No signal of positive selection was identified in Haplochromis sp. ‘Matumbi hunter’ or Haplochromis chilotes (Table 1). The extreme value of ωF = 999 indicates rare synonymous substitutions in the foreground branch.

The branch-site model predicted positively selected amino acid sites in Haplochromis sp. ‘red tail sheller’ at positions 72 and 198 (BEB probability > 0.75) which resulted the substitutions from Lysine (K) to Arginine (R) in position 72 and from Proline (P) to Glutamic acid (E) in position 192. To exclude the possibility of an inflated prediction of positive selection that can be caused by polymorphisms in very recently diverged populations (Peterson and Masel 2009), we checked the SNP information for the positively selected site. The novel alleles showing a sign of selection are unique to Haplochromis sp. ‘red tail sheller’ (data not show), which indicates it is not due to shared polymorphism.

Test of Positive Selection on Ancestral African Lineages

We tested for a signal of positive selection in the ancestral sequences of each hierarchical clade of the African cichlids, i.e., (i) ancestor to H. chilotes + Haplochromis sp. ‘Matumbi hunter’, (ii) ancestor to all Haplochromis, and (iii) ancestor to all African cichlids (Oreochromis + Haplochromis) (Fig. 1). All three ancestral sequences showed a statistically significant signal of positive selection, though the strength of selection differs across hierarchies (Table 1). The selection pressure in the ancestor sequence of all Haplochromis is stronger than that of the evolutionarily deeper grouping of all African cichlids (ωF = 75.84 vs. 16.34). An ωF value of 999 (e.g. at the ancestral sequence to Haplochromis sp. ‘Matumbi hunter’ and H. chilotes) is caused by sequence divergence comprised only of non-synonymous mutations (i.e., K s = 0).

Interestingly, different, non-overlapping codon sites showed significant signals of positive selection across the three hierarchies of ancestral sequences (Fig. 2). This may reflect that the nucleotide substitutions were driven by temporally and molecularly different selection pressures and, therefore, show different evolutionarily independent signals.

Fig. 2
figure 2

Sites showing a signal of positive selection in the ancestral branch of all African cichlids (a), all Haplochromine cichlids (b), and Haplochromis sp ‘Matumbi hunter’ and H. chilotes (c). The position stands for the codon position on the cDNA sequence. The horizontal line indicates the Bayes empirical Bayes probability of positively selected sites equal to 0.75

Full size image

Unfortunately, there is no structural information for the EPCAM gene available in the current protein data bank (Berman et al. 2000). Therefore, we predicted the protein structure prediction using the interproscan software (Quevillon et al. 2005) online tool (http://www.ebi.ac.uk/Tools/InterProScan/). The thyroglobulin 1 domain was identified (E value = 5.1e−14) and predicted to span from codon position 89–137, a common domain in the EPCAM family (Baeuerle and Gires 2007). However, all of the positively selected nucleotide sites that were determined in our analysis are located outside of this domain and were without significant domain search hits.

Discussion

In this study, we analyzed the evolution of the EPCAM gene across diverse cichlid lineages. Because of its importance in fish morphogenesis (Slanchev et al. 2009), we considered EPCAM to be a relevant candidate for being involved in the exceptional morphological and phenotypic diversifications of cichlid fishes. We identified a strong signal of positive direction selection in the evolution of the African cichlid lineage, especially that of the Lake Victoria ‘superflock’ of cichlid fishes, the Haplochromis (Table 1; Fig. 1). Selection was less strong in the more species depauperate tilapia lineage. The fact that EPCAM was not always found to be under selection in all lineages indicates it is not obligatory that the gene show such a signal through its evolution.

Local adaptation is one of the driving forces in the evolution of cichlid species (Kocher 2004). One prediction that stems from this is that different, lineage-specific directional selection pressures should be identified on the genes involved in local adaptation. The more than 2,000 cichlid species found in East Africa display a great variation in phenotypic attributes such as body shape, jaw shape, and coloration. Thus, we hypothesized that a gene known to play an important role in morphological variation in other fishes (Slanchev et al. 2009) may a target of positive natural selection during cichlid evolution.

To test this hypothesis, we first compared the evolution of the EPCAM gene between representative African and neotropical cichlids and identified that the EPCAM gene is under significant and strong positive selection in the African cichlids as a whole (Fig. 1; Table 1). Then we examined in more detail the Lake Victoria ‘superflock’ of Haplochromis cichlids, that are renowned for their spectacular adaptive radiation (Meyer et al. 1990; Verheyen et al. 2003; Salzburger and Meyer 2004) and, indeed, identified a significant signature of positive selection in the EPCAM gene in this group (ωF = 37.03, P < 0.001).

Second, we teased apart the hierarchical level at which Darwinian selection is evidenced in the evolution of the African cichlids. In the derived lineages of four species, that are representative of extremely rapidly evolved Lake Victoria cichlid “super flock” only the species Haplochromis sp. ‘red tail sheller’ showed significant signatures of positive selection in the evolution on EPCAM sequence (ωF = 999, P < 0.001). The extremely high signal of selection pressure (undefined or “999”) indicates that only non-synonymous mutations and no synonymous mutations were found in this branch. This small genetic difference could result in considerable phenotypic changes, for example by a single amino acid replacement between closely related species (Hoekstra et al. 2006) or may be a signal of relaxed purifying selection during adaptive evolution (Hughes 2007). We propose that positive selection, rather than relaxed purifying selection, more likely drove the fixation that is only found at these non-synonymous sites based on two reasons. First, the function and amino acid sequences of the EPCAM gene are conserved across vertebrate evolution (Trzpis et al. 2008b; Slanchev et al. 2009). Second, if relaxed purifying selection is caused by small population size (Ohta 1995), we would expect to observe an increased genetic diversity in the genome wide scale. However, other researches show that ESTs and multiple genomic nucleotide markers show very limited genetic differentiations among Lake Victoria cichlid fishes (Elmer et al. 2009; Kobayashi et al. 2009).

We did not detect a signature of positive selection in the other two Lake Victoria cichlids, H. chilotes and Haplochromis sp. ‘Matumbi hunter’ though we did detect selection in the common ancestral branch. Two possible explanations for this could be either the insufficient genetic variability between these recently diverged species, or a signal of purifying selection on derived branches. A similar change of selection regime, from positive selection on the ancestral branch to purifying selection on the derived branch was, for example, shown in the evolution of gut lysozymes in the colobine lineage of primates (Messier and Stewart 1997).

In the processes of speciation in cichlids, it is proposed that the most major morphological changes occur at the initial stages of habitat adaptation, often followed by subsequent subtler evolution into more finely partitioned ecological niches and color divergence based on sexual selection (Kocher 2004). To test whether this model is correct, it is important to determine whether positive selection influenced the sequences earlier than in the derived, extant lineages. Three waves of positive selection were detected in the ancestral sequences of the African cichlids (ωF = 16.34, P < 0.001), Haplochromis cichlids family in Lake Victoria (ωF = 75.84, P < 0.001) and Haplochromis sp. ‘Matumbi hunter’ and H. chilotes (ωF = 999, P < 0.01). These results suggest that functional changes in the EPCAM gene happened at different and various stages during African cichlid evolution and thus the signal of selection pressures differs along the phylogeny. The strongest positive selection was identified in the ancestral lineage leading to the extraordinarily diverse lineage of haplochromine cichlids, which could be the result of adaptation to the vast set of new open lake niche once the colonization of Lakes Victoria and Malawi from riverine haplochromine lineages took place (Salzburger et al. 2005; Elmer et al. 2009). The positively selected sites among three ancestral sequences were not overlapping, which may reflect the divergent selection caused by adaptations to different environments (Fig. 2a vs. Fig. 2b vs. Fig. 2c).

Cichlids from the African Great Lakes have been found to have accelerated rates of non-synonymous substitution in a number of candidate genes relative to riverine sister lineages. Comparative analyses and functional analyses tended to indicate that this molecular pattern is the result of stronger natural selection, including sometimes sexual selection pressures, in the lacustrine habitat. For example, a high rate of non-synonymous substitution was also identified in the molecular evolution of the hagoromo gene in haplochromine cichlids from the endemic adaptive radiations of the Great Lake African cichlids. This gene—also a candidate selected based on its role in zebrafish development—might play an important role in pigmentation, and its evolution might be driven by sexual selection (Terai et al. 2002a). This was further confirmed with analyses that showed that cichlid lineages differ with respect to the complexity of hagomoro gene alternative splicing, and that this splicing complexity correlates with lineage speciation rates (Terai et al. 2003). A similar pattern was found with the pigmentation candidate gene mitf (Santini and Meyer, unpublished data; Sugie et al. 2004). Genes related to vision are also known to be under positive selection in lacustrine compared to riverine East African haplochromine cichlid species (e.g., Sugawara et al. 2002). Thus, our inferences from the analyses of EPCAM are in agreement with findings from other genes relevant to adaptive phenotypes in the adaptive radiation of cichlids. It is unclear why the signals of natural selection differ among the three extant haplochromines in our study (Table 1), but it may be that their different ecological niches play a role. Further research on the evolution of EPCAM sequence in species from different lineages and habitats will be required to clarify this pattern.

Theoretical analyses have suggested that K a/K s estimates may be elevated in recently diverged species due to incomplete sorting of ancestral polymorphism (Peterson and Masel 2009). That most of the derived African lineages in our study do not show significant positive selection, while the inferred ancestral sequence does, suggests this phenomenon apparently does not play a strong role in our analyses. Furthermore, we found that positively selected alleles in extant species apparently species-specific.

Our current research is limited to EPCAM sequences only from Lake Victoria, the youngest adaptive radiation of the eastern Africa great lakes and rooted with recently determined neotropical cichlid EPCAM sequences. Adding more sequences of this candidate gene from the African lakes that harbor older cichlid species flocks, such as Lake Malawi and Lake Tanganyika, would provide a more comprehensive view of the role of EPCAM during the formation of the adaptive radiations of cichlids. Since cichlid fishes have evolved parallel phenotypes presumably under similar selective regimes due to ecological speciation, one could test the prediction that natural selection drives the same advantageous alleles to fixation independently in closely related species that share similar environments (Schluter 2009). The fixation of different positively selected nucleotide sites within the EPCAM gene suggests that it could be employed as candidate gene to test whether parallel morphologies in East Africa great lake cichlids (Kocher et al. 1993) are caused by parallel, independent substitutions in morphologically relevant genes.