The genus Pethia
Following the recognition of the genus by Pethiyagoda et al. , several studies confirmed the monophyly of Pethia [e.g., 10, 11, 18, 25, 26]. Nevertheless, about half of the 43 species included in the genus since 2012 are yet to be represented in a molecular phylogenetic framework. Doubt remains as to the generic placement of several of these. While Pethiyagoda et al.  allocated 23 species to the genus, they drew attention to others for which insufficient information was available, upon which to make a generic placement. ‘Pethia’ narayani, ‘P.’ sharmai, ‘P.’ aurea, ‘P.’ cania, ‘P.’ gelius, ‘P.’ guganio, ‘P. castor’, and ‘P. pollux’ are morphologically so distinct from Pethia sensu stricto and other Smiliogastrini that they may warrant placement in different genera [16, 18, 19]. The original description by Hora  of ‘Pethia’ narayani does not mention a serrated last unbranched dorsal-fin ray, which is a synapomorphy in Pethia. Hora  mentions only that “The dorsal fin… possesses a feeble and articulated spine which is considerably longer than the head; its free border is slightly concave though rounded at the top.” The holotype of ‘P.’ narayani, illustrated in Hora , exhibits three vertical bands on the side of the body. ‘Pethia’ narayani in fact may belong to the recently described smiliogastrine genus Waikhomia . ‘Pethia’ sharmai differs from all congeners in having 40 or more scales along the lateral-line row (vs. 30 or less in Pethia sensu stricto: see Additional file 1: Table S1). It superficially resembles ‘P.’ guganio. ‘Pethia’ aurea, ‘P.’ cania and ‘P.’ gelius appear to form a closely related group united by a striking colour pattern, distinguishing the group from Pethia s.s. . In fact, even in our phylogenies, the GenBank sequences identified as “P. gelius” formed a distinct lineage basal to Pethia s.s. (Fig. 4). It seems possible also that ‘P.’ castor, and ‘P.’ pollux , from Myanmar, may belong to a lineage distinct from Pethia. The phylogenetic relationships of the mentioned Indian and Myanmarese species remain to be explored.
The type species of Pethia, P. nigrofasciata, a Sri Lankan endemic, is included in the present study. Consistent with this, the 36 species we treat as belonging to Pethia s.s. (Additional file 1: Table S1) are characterized by having the following suite of characters: 8 branched dorsal-fin rays; 5 branched anal-fin rays; the last unbranched dorsal-fin ray strong, serrated on its posterior margin; 2 or 3 black spots, blotches or bands laterally, including one on the humeral-cleithral region and another above the anal fin or on the caudal peduncle; the lateral line more often incomplete (26 species) than complete (10 species); barbels usually absent (32 species); and 19–30 scales in the lateral-line series. The maximum size for the genus is usually < 50 mm SL . Most species (perhaps all) are sexually dichromatic. All five species of Pethia in Sri Lanka are consistent with the above conception of the genus.
Our taxon sampling includes all five Sri Lankan species of Pethia, sampled from the island’s principal rivers, including cytb sequences derived from reliably identified specimens for P. bandula as well as almost all the species reported from South India, to which the Sri Lankan species would be expected to have their closest relationships. This dense sampling adds confidence to the relationships inferred from our phylogenetic analysis.
The concatenated phylogeny confirms our hypothesis that the Sri Lankan representatives of the genus do not form a monophyletic group. This relationship was hypothesized on the basis of four of the five species (P. bandula, P. nigrofasciata, P. cumingii and P. reval) exhibiting a similar morphology and being confined to the island’s south-western wet and intermediate zones (rainfall > 2.5 m/y and 1.8–2.5 m/y, respectively). The wet-zone diversifications of several other cypriniform genera have been shown to be monophyletic [e.g., Systomus: 11, Devario: 14, and Rasbora: 15], whereas such diversifications are rare in the dry zone [15, 39]. Of the four species of Pethia endemic to the island’s south-western wet zone, only one, P. reval, has a range extending into the western intermediate zone, as far north as the Deduru basin. Pethia melanomaculata, in contrast, is confined largely to the dry zone, though extending also to the intermediate zone in the east-draining Mahaweli and Gal basins, and the west-draining Deduru basin. It differs from the other four Sri Lankan species of Pethia also in morphology (Figs. 3b–c, 8a, 9a). The sister-group relationship of P. melanomaculata is not clearly resolved in our phylogenies (Fig. 4). We suspect this species may have a closer phylogenetic relationship to a lineage from peninsular India that is not represented in our dataset. Similar relationships have been observed for other freshwater fishes widespread in the dry zone of Sri Lanka [40,41,42] with few exceptions .
While our results recover Sri Lankan Pethia as polyphyletic, the four southwestern wet zone species are not recovered as monophyletic, as hypothesized (Fig. 4). The phylogenetic relationship between P. nigrofasciata and P. bandula is not clearly resolved. Similarly, the relationship between P. cumingii and P. reval too, is ambiguous. Further, the sister-group of the clade that includes P. reval and P. cumingii in all the analyses is a well-supported clade that includes several Indian species. The clade that includes Pethia nigrofasciata and P. bandula is recovered as the sister group of the clade that includes P. cumingii, P. reval, P. ticto, P. longicauda and P. pookodensis, with strong node support in all the analyses (Fig. 4, Additional file 1: Figs. S1, S2). Thus, our phylogeny suggests that the five species of Sri Lankan Pethia derive from two or three discrete colonization events from the Indian mainland. Multiple colonization events have been recovered also in the case of other freshwater-fish diversifications in Sri Lanka, such as Laubuka, Rasbora, Devario and Systomus [11, 14, 15, 39]. In those cases, the diversifications in the island’s wet zone within each genus were shown to stem from a single colonization from India. The present findings support two equally parsimonious scenarios for colonization of Pethia in Sri Lanka. One is that the common ancestors of P. melanomaculata, P. nigrofasciata and P. bandula, and P. cumingii and P. reval derive from three independent colonization events from the Indian mainland. If this scenario is confirmed, then Pethia would be the first freshwater fish genus in which a wet-zone diversification deriving from multiple independent colonization events has been detected in the island. An alternative scenario would be two colonization events from mainland India, being the common ancestors of P. melanomaculata and P. nigrofasciata, P. bandula, P. cumingii and P. reval, followed by a back-migration to India.
This is noteworthy because, despite having been connected by a broad isthmus during episodes of low sea level, post-Miocene biotic exchanges of forest-adapted taxa between India and Sri Lanka have been infrequent [5, 10, 11]. Though subaerial for most of the Plio-Pleistocene, the Palk Isthmus appears, because it was too arid, to have acted more of a filter than a conduit for the dispersal of forest-adapted taxa [5, 8, 10]. While P. nigrofasciata is a rainforest-adapted species, P. cumingii and P. reval are not obligatory rainforest associates (discussed below). We hypothesize, based on our results, that the common ancestor of P. cumingii and P. reval was a generalist. In such a scenario, a back migration to India by the common ancestor of P. cumingii and P. reval through the arid Palk Isthmus or the colonization of the rainforests of the island’s wet zone, are both plausible.
Pethia bandula—P. nigrofasciata
In our cytb and concatenated rag1 + cytb trees, Pethia bandula renders P. nigrofasciata paraphyletic (Fig. 4, Additional file 1: Fig. S2). Pethia bandula is a Critically Endangered species confined to a single localized population in a ~ 3-km stretch of a small stream within the Kelani basin . It therefore enjoys strict protection, and sampling is not permitted. As such, we were limited to using the 540–552 bp sequences from the cytb locus available on GenBank, much shorter than the 1082 bp contained in the cytb sequences newly generated in this study. This may have led to a weakening of the phylogenetic signal represented by P. bandula.
It is also possible that P. bandula is the result of a recent speciation event. Its range lies at the northern extremity of that of P. nigrofasciata (Fig. 1b). If P. bandula is an incipient species, it could be that the lineages are as yet incompletely sorted or even introgressed, leading to it and P. nigrofasciata not being recovered as reciprocally monophyletic based on the genetic markers used in the present study. The observation of mixed morphology in some populations of P. nigrofasciata in the headwaters of the Attanagalu basin proximal to the type locality of P. bandula appear consistent with such a hypothesis (Fig. 1g–n, Additional file 1: Table S9). Despite all three molecular species-delimitation methods we applied (PTP, mPTP and ABGD) grouping them as a single species, P. bandula and P. nigrofasciata are easily distinguished on morphological criteria alone . A genomic approach may reveal clearer structuring and genetic differentiation between the two species, while also revealing evidence of incomplete lineage sorting or hybridization [44,45,46,47].
Pethia cumingii—P. reval
The phylogenetic relationship between P. reval and P. cumingii too, is not clearly resolved in our concatenated phylogeny (Fig. 4), with all three molecular species-delimitation methods recovering them as a single species. Nevertheless, P. reval and P. cumingii are easily distinguished by the red and yellow, respectively, of their fins, in addition to a suite of morphological characters . There is also a clear geographical signal in the phenotypes of the two species (Fig. 2). The red-finned populations occur exclusively in the northern Deduru, Ma, Attanagalu and Kelani basins, whereas yellow-finned populations occur exclusively in the more southerly Bentara and Gin basins. The Kalu basin lies between the northern and southern watersheds that host exclusively the red or yellow-finned populations assigned the names P. reval and P. cumingii, respectively. While populations in the Kalu usually have yellow fins, individuals with a mix of red and yellow or orange fins occur in some localities (Fig. 2d). It could be that speciation in these two lineages too, is recent, with as yet incomplete lineage sorting or introgression. Both these species are assessed as Endangered , and accurate recognition of the species’ taxonomic status is important in conservation management. While our single-locus species-delimitation and the phylogenetic analyses based on cytb and rag1 failed to separate P. reval and P. cumingii, we adopt a conservative taxonomic approach and retain them as valid species based on their distinct morphology and allopatric distribution. Similar to the case of P. bandula and P. nigrofasciata, we expect a genome-wide analysis to recover clear structuring and genetic differentiation between these species while also revealing whether the mentioned phenotypic discrepancies in the Kalu basin are the result of hybridization [28, 44,45,46,47].
Geographic ranges and habitats
Pethia nigrofasciata is confined to Sri Lanka’s wet-zone basins, from the Attanagalu in the north to the Walawe in the south. It occurs in clear-water streams and rivers with gravel or pebble substrates. The habitats of P. nigrofasciata resemble those of the other widespread endemic species in the rainforests of the wet zone, such as Devario micronema, Laubuka varuna, Rasbora wilpita and Systomus pleurotaenia [11, 14, 15, 39]. In contrast, P. reval and P. cumingii occupy broader ecological niches. While the former occurs close to banks in rivers associated with rainforest habitats, it is encountered also in pools in the lowland floodplains and in streams traversing rice paddies, with substrates of silt or debris. Thus, although the extension of its range as far north as the Deduru basin in the intermediate zone is unsurprising, P. reval is among the two endemic freshwater fishes that occur in both the wet zone and the intermediate zone, the other such species being the silurid catfish Ompok argestes .
Pethia cumingii, however, is more associated with rainforests, occurring in both shaded streams and rivers. While our samples of this species derive only from the Kalu, Bentara and Gin basins, we have observed it also in the Nilwala and Walawe basins further south. Meanwhile, Pethia bandula is confined to a single small stream at Galapitamada, its type locality, which traverses a rice-paddy landscape. This region was likely occupied by rainforest prior to anthropogenic modification. In contrast, compared to its four wet-zone congeners, P. melanomaculata occupies a broad ecological niche in the dry and intermediate zones. It occurs in lotic habitats such as rivers, streams, canals, as well as lentic habitats such as seasonal pools and reservoirs and, unlike the other Sri Lankan species of Pethia, does not appear to be associated with shade or riparian vegetation.
In some rainforest habitats in the Attanagalu and the Kelani basins, P. nigrofasciata and P. reval occur in syntopy. Similarly, in such habitats in the Kalu, Bentara, Gin, and Nilwala basins, P. nigrofasciata and P. cumingii occur in syntopy. While we have not encountered P. reval and P. melanomaculata in syntopy in the intermediate zone, this would be expected [22, 32]. At the type locality of P. bandula, no other species of Pethia occurs. However, both P. nigrofasciata and P. reval occur in nearby streams [22; present study].
Several recent studies have explored the phylogeographic structure of cyprinid species confined to Sri Lanka’s southwestern wet zone, such as Devario micronema, Laubuka varuna, Rasbora wilpita, and Systomus pleurotaenia [11, 14, 15, 39]. All these rainforest associates show strong within-basin genetic structure, with limited gene flow between even adjacent basins. In these cases, it appears that inter-basin dispersal is inhibited by the concerned species being restricted to shaded clearwater streams draining the foothills of the island’s central mountains. They are thus absent from the lowland floodplain, across which there is potentially hydrological connectivity between basins when flooding follows episodes of heavy rainfall. In cases where inter-basin geneflow had in fact occurred, it was inferred that this was the result of headwater river-capture events rather than via the lowland floodplain [5, 11].
Pethia nigrofasciata too, shows within-basin phylogeographic structure, with no cytb haplotypes shared between basins (Fig. 5). As mentioned above, our concatenated phylogeny (Fig. 4) recovered P. nigrofasciata as two well-supported, sympatric subclades, one spanning the distribution of the species in Sri Lanka, from the Attanagalu to the Walawe basins, and the other confined to the region between the Kalu and Gin basins, inclusive. Such a pattern has not been observed in the other phylogeographic studies of Sri Lankan cyprinids published so far [5, 11, 14, 15, 39]. Given that our genetic dataset is limited, it is difficult to offer an explanation for this observation. However, in our rag1 nuclear dataset (Additional file 1: Fig. S2), these two subclades are not apparent. It is possible that these lineages underwent secondary admixture between allopatrically evolved populations . The complex topography of the southwestern wet zone may have imposed historical biogeographic barriers to gene flow between the two populations. The cytb haplotype network of P. nigrofasciata too (Fig. 5b), does not suggest inter-basin gene flow through headwater capture between the adjacent basins. The star-like pattern of the rag1 haplotype network of P. nigrofasciata (Fig. 5c), however, may suggest a recent range expansion, even though the neutrality tests were not significant. Broader sampling within each river basin and genome-wide data may reveal a clearer picture of the evolutionary history of these two mitochondrial lineages in P. nigrofasciata. While no haplotypes are shared between these two subclades, samples from the same locality may belong to both subclades. For example, the cytb haplotypes N10 and N11 occur in members of subclades 1 and 2, respectively, at the same locality in the Bentara basin. The occurrence of distinct mitochondrial lineages in syntopy has also been observed in the southwestern basins of Sri Lanka for the cyprinid Garra ceylonensis . This further supports our hypothesis that these samples may have derived from two historically separate matrilineal evolutionary lineages.
While P. cumingii and P. reval are not obligatory rainforest associates, three of the four subclades of this species-pair exhibit distinct phylogeographic structure (Fig. 4). Subclades A and B contain haplotypes unique to P. reval, and subclade D contains haplotypes unique to P. cf. cumingii from the Kalu basin. Subclade C, however, includes haplotypes shared between both P. reval and P. cumingii. One cytb haplotype, R5, is shared between the adjacent Ma and Attanagalu basins, while another (C5) is shared between the Bentara and Gin. Haplotype C9, however, is disjunct between the Attanagalu (in P. reval) and Gin (in P. cumingii) basins. It may represent a shared ancestral haplotype which is now fixed in P. cumingii. Interestingly, three cytb (C1–C3) and three rag1 (R2–R4) haplotypes are unique to the Kalu basin, that shares no haplotypes with any other basin.
The species of Pethia confined to the island’s south-western wet and intermediate zones show evidence of strong phylogeographic structure. Unlike in other endemic cyprinids studied so far, there is little evidence of gene flow between adjacent basins [11, 14, 15, 39]. This could be because the diversification of these lineages has been recent, with ancestral polymorphism retained and lineage sorting as yet incomplete. A genome-wide analysis could provide a clearer understanding of the evolutionary history of these species.
The phylogeographic structure observed in Pethia melanomaculata resembles that in Laubuka lankensis, which too, has a similar distribution, being confined to the dry and intermediates zones . In both these species, three regional haplogroups can be identified: northwest, Mahaweli, and eastern. Only a single cytb haplotype is shared between rivers: the adjacent Kala and Malwathu basins in the northwest haplogroup. While the remaining haplotypes are unique, they are separated by relatively few mutational steps, in contrast to the condition observed in P. nigrofasciata, P. cumingii, and P. reval.
Most of the dry zone’s fishes derive from recent (Pleistocene) dispersants from India, adapted to an arid, strongly seasonal climate . These exhibit little phylogeographic structure [10, 11, 14, 15, 39, 41]. Within the widespread species in the dry zone, the populations from the eastern basins (principally the Gal, Kumbukkan, Menik and Kirindi, which drain the eastern slopes of the central hills) appear to show greater genetic diversity compared with populations in the northwest and the Mahaweli basins. This region lies within the intermediate zone and benefits from higher annual—though less markedly seasonal—rainfall than the northwest and Mahaweli dry zone. It has two endemics confined to it: Rasbora adisi and Laubuka hema [15, 39]. As P. melanomaculata too, demonstrates, the eastern basins show substantial isolation from their neighbours. The region has until now not attracted attention as a focus for conservation, but clearly warrants such consideration.
Nevertheless, perhaps owing to the wet zone’s greater topographic complexity , and despite its extent being only about a quarter that of the dry zone, nucleotide and haplotype diversity in the wet-zone endemics P. nigrofasciata, P. cumingii and P. reval are greater than in P. melanomaculata. This phenomenon has been observed previously in species pairs in which one is confined to the wet zone while the other is distributed across the dry zone, such as Devario micronema vs D. malabaricus, Laubuka varuna vs. L. lankensis, Systomus pleurotaenia vs S. sarana, and Rasbora wilpita vs R. dandia [11, 14, 15, 39]. As Potter et al.  show, genetic diversity in low-dispersal vertebrate species tends to be higher in mesic, topographically complex biomes, compared to that of species inhabiting dry and topographically less complex biomes.
The haplotype networks of both P. nigrofasciata and P. reval indicate shared mitochondrial haplotypes (N5 and R8, respectively) between the west-draining Kelani and east-draining Mahaweli basins. These basins share a common boundary along a 40 km long, 600–2000-m high ridge that extends from Ginigathena to the Horton Plains. Sudasinghe et al.  reported a shared haplotype between populations of the dwarf snakehead Channa orientalis between the two basins, suggesting that gene flow between them is possible. In the case of Pethia, however, Wikramanayake  recorded a translocation experiment in which both P. nigrofasciata and P. reval (which he referred to as Puntius cumingii) were introduced to a stream near Ginigathena (6.987°N, 80.499°E). Whether stemming from this introduction or other undocumented ones [see: 32], both P. nigrofasciata and P. reval now occur as far as 40 km downstream, at Peradeniya.
Wikramanayake  reported the stocks of P. reval and P. nigrofasciata introduced to the Mahaweli in 1981 to have come from the Kelani and Kalu basins, respectively. He was not, however, associated with the original translocation experiment and based his report on information from secondary sources. Sudasinghe et al.  showed that the stock of Rasboroides pallidus translocated in this experiment originated not from the Kalu, as reported by Wikramanayake , but likely from the Bentara basin. In the present study, we show the populations of Pethia introduced to Mahaweli derive from multiple sources. The Mahaweli populations of P. reval contain haplotypes belonging to both subclade A (native to the Deduru, Ma and Attanagalu basins) and subclade B (native to the Kelani). Similarly, the Mahaweli populations of P. nigrofasciata contain haplotypes otherwise unique to the Kelani and Kalu basins, in addition to several unique haplotypes, all within the subclade 1 of native P. nigrofasciata. In both species, the multiple unique haplotypes in the Mahaweli (R9–R11 in P. reval, N20–N23 in P. nigrofasciata) suggest that our sampling density underrepresents the haplotype diversity of their native populations.
Deraniyagala  reported P. reval (as Puntius cumingii) from Peradeniya, which suggests that a translocation occurred even before that reported by Wikramanayake . It is also possible that populations of both P. reval and P. nigrofasciata may have escaped from the fisheries station at Ginigathena, and perhaps also from Peradeniya University, both on the Mahaweli River [54, 55]. The populations of both these fishes in the Mahaweli may thus result from independent founder events spanning several decades, a scenario consistent with our results. This is unsurprising in the light of both species having been popular in the ornamental fish trade for almost a century now.