Multigene phylogeny of cyprinodontiform fishes suggests continental radiations and a rogue taxon position of Pantanodon

We studied phylogenetic relationships among major clades in the tooth carps (Cyprinodontiformes) based on a concatenated DNA sequence alignment of two mitochondrial and three nuclear gene segments, totalling 2553 bp, in 66 ingroup terminals. The inferred tree supports monophyly of the major tooth carp subgroups, aplocheiloids and cyprinodontoids, and of several aplocheiloid subclades corresponding to the well-established families (Aplocheilidae, Nothobranchiidae, Rivulidae), each of which is restricted to major continental settings (India-Madagascar, Africa, South America). Contrary to previous molecular studies, our tree supports a sister-group relationship of the aplocheilids and nothobranchiids, rather than a nothobranchiid-rivulid clade. Within cyprinodontoids, the phylogeny matched more closely continent-scale distribution than current classification, suggesting that the delimitation of the families Cyprinodontidae, Poeciliidae, and Valenciidae is in need of revision. The East African Pantanodon stuhlmanni did not show close relationships with any other taxon in our analysis, suggesting that the phylogenetic position and classification of this rogue taxon is in need of further study.

Understanding cyprinodontiform phylogeny has the potential to inform studies on the evolution of annualism and live bearing, and on the biogeographic origins of these ishes.Traditionally the origins of tooth carps, especially those in the Aplocheiloidei, are interpreted as being a consequence of ancient vicariance (e.g., MurPHy & Collier, 1997;sParks & sMitH, 2005;saMonDs et al., 2012;Costa, 2013), in particular because their cladogenesis largely relects the breakup of the Gondwana supercontinent in deep Mesozoic times, with the Indian Aplocheilus considered being sister to the Malagasy-Seychellean Pachypanchax, and the South Amercan Rivu lidae sister to the African Nothobranchiidae (MurPHy & Collier, 1997;sParks & sMitH, 2005).The vicariance hypothesis for aplocheiloid origins however requires conirmation as it conlicts with clade ages recovered in several studies (e.g., Crottini et al., 2012;near et al., 2012, 2013;brougHton et al., 2013) that place the origin of the entire cyprinodontiform clade into the latest Mesozoic or early Cenozoic, similar to that of cichlids (VenCes et al., 2001;FrieDMan et al., 2013).Particularly relevant for this aspect of cyprinodontiform biogeography are the endemic tooth carps occurring on Madagascar, the fourth largest island of the world.This island has been separated from all other landmasses since the Mesozoic and is characterized by a unique and highly endemic biota, yet many of its radiations appear to have originated after its isolation (yoDer & noWak, 2006;saMonDs et al., 2012).Madagascar is inhabited by two native genera of cyprinodontiforms: the genus Pachypanchax with currently six Malagasy and one Seychellean species; and the genus Pantanodon, with one described and one undescribed species known from Madagascar, and one species occurring in Eastern Africa (sParks, 2003;loiselle, 2006).So far, no molecular data are available for Pantanodon, and only one Malagasy species of Pachypanchax has been included in molecular phylogenies (MurPHy & Collier, 1997;Crottini et al., 2012).
As a irst step to improve the understanding of higher-level cyprinodontiform relationships, we newly determined a data set of two mitochondrial and three nuclear genes for a set of 66 cyprinodontiform terminals.Our data set spans seven of the ten currently accepted families and includes the enigmatic Pantanodon.By highlighting various unsolved questions and taxa that merit furter phylogenetic study, we anticipate our study inform and facilitate future systematic revisions of tooth carps.

Materials and Methods
If not indicated as wild caught by precise collecting locations (Table 1), samples were from aquarium strains.Voucher specimens of the majority of specimens were preserved, labeled with provisional numbers (ZCMV -Miguel Vences Zoological Collection) and will be deposited in the Zoologische Staatssammlung München, Germany.Tissue samples were preserved in pure ethanol.DNA was sampled from in clips of the preserved vouchers, or from eggs.Total genomic DNA was extracted from tissue or swab samples using Proteinase K (10mg/ml) di gestion followed by a standard salt-extraction protocol (bru ForD et al., 1992).Two markers of the mitochondrial and three markers of the nuclear genome were targeted: Segments of the mito chondrial genes cytochrome oxidase sub unit I (COX1) and 16S rRNA were ampliied, respectively, Polymerase chain reactions were performed in a inal volume of 10 μl using 0.3 μM of each primer, 0.25 mM of dNTPs, 0.4 U GoTaq and 1.25 × Reaction Buffer (Promega).
PCR products were puriied using Exonuclease I and Shrimp Alkaline Phosphatase (SAP) or Antarctic Phosphatase (AP) according to the manufacturer's instructions (NEB).Puriied PCR templates were directly sequenced using dye-labeled dideoxy terminator chemistry on an ABI 3130 automated DNA sequencer (Applied Biosystems).We checked chromatograms and corrected errors manually in CodonCode Aligner 3.5.6(CodonCode Corporation).Newly obtained sequences were submitted to GenBank (accession numbers: KJ844613-KJ844868).Sequences of outgroup taxa were taken from Genbank, and largely correspond to those determined by near et al. (2012).We used a representative of the Gobiiformes (Perccottus) as outgroup and included a series of atheriniform, beloniform and perciform species as hierarchical outgroups.
We used MEGA 5 (taMura et al., 2011) to align protein-coding sequences (COX1, RAG1, GLYT, SREB2) manually and the non-coding 16S using the MUSCLE algorithm.The 16S alignment was subsequently processed with Gblocks 0.91b software (Castresana, 2000) with a less stringent 50% threshold for the deinition of reliable lanking positions and the remaining parameters at default settings, to exclude ambiguously aligned sections.
The AICc criterion implemented in PartitionFinder

Results and Discussion
The concatenated alignment of 16S COX1, RAG1, SREB2 and GLYT sequences, after exclusion of hypervariable 16S sections suggested by Gblocks, consisted of 2553 bp and 76 terminals (996 bp mtDNA and 1557 bp nDNA).Bayesian Inference analysis excluding Pantanodon stuhlmanni (Fig. 1) resulted in high posterior probabilities of almost all deep nodes in the phylogeny: cyprinodontiforms, aplocheiloids, cyprinodontoids, as well as the families Aplocheilidae, Nothobranchiidae, and Rivulidae all received maximum support (PP = 100%).A few relationships within more shallow clades are worth mentioning as well: (i) In the aplocheiloid clade, the included Malagasy terminals together formed the sister group of the Seychellean P. playfairii, and the differentiation among the two included Malagasy species was rather low.According to an unpublished molecular data set of R. D. Randrianiaina with full taxon coverage (pers.comm. in 2014), the two populations included here are among the most divergent within Madagascar, so that monophyly of the entire Malagasy clade vs. the single Seychellean species is likely.(ii) Within nothobranchiids, representatives of Diapteron and Chromaphyosemion were placed within Aphyosemion; if conirmed by future studies, then the taxonomy of these genera is in need of revision.

Within aplocheiloids, the
Inclusion of Pantanodon stuhlmanni in the analysis led to an identical topology, with Pantanodon in an unsupported position sister to the cyprinodontoids (red in Fig. 1).However inclusion of this taxon reduced posterior probabilities of several nodes, of which certainly the basal node of cyprinodontiforms is the most relevant: monophyly of tooth carps was not signiicantly supported any more (87%) after inclusion of Pantanodon.In all single gene trees (Supplementary Materials) Pantanodon is placed outside of cyprinodontiform clade, although typically without strong support.We therefore conclude that the phylogenetic position of this enigmatic taxon remains largely unresolved, but certainly it is not a close relative to the African lampeyes classiied within the poeciliids (subfamilies Aplocheilichthyidae and Procatopodinae).This taxon appears to it well in the deinition of a "rogue" taxon which in various single gene analyses is placed in varying positions of the phylogeny and blurs the phylogenetic signal in a combined analysis of all markers (e.g., sanDerson & sHaFFer, 2002).Costa (2012) based on morphological characters placed Pantanodon, including two fossil species from the Oligocene and Miocene of Europe, within the Poeciliidae.Unfortunately we have not been able to obtain tissue samples of Malagasy species of Pantanodon.Of the two Malagasy species, one (P.madagascariensis) is probably extinct, and a second, undescribed species appears to be restricted to a very small swamp area on the southern east coast of the island (sParks, 2003).Whether these ishes are really related to the African P. stuhlmanni requires conirmation, and their molecular study is all the more a priority given the intriguingly isolated position of the African species.
We are aware that the phylogenetic hypothesis presented herein needs to be seen with some caution.Many samples in our analysis were supplied from aquarium strains, thus without reliable locality information and with unveriied species-level identiication.This for instance might be relevant, in our tree, for the topology within nothobranchiids such as non-monophyly of Aphyosemion relative to Diapteron and Chromaphyosemion.However, we here focus on relationships among genera and major clades and at this level the identiication of all of our samples is reliable.Our discussion of biogeography and higher classiication therefore remains valid.Taxa from the families Goodeidae, Profundulidae and Anablepidae (all three from suborder Cyprinodontoidei) are missing from our data set, but at least the position of goodeids is not expected to yield large surprises as setiaMarga et al. (2008) found Xenotoca eiseni, a Central American representative of this family, highly supported within Cyprinodontoidei (and sister to a Neotropical poeciliid).
Because the nuclear genes could not be sequenced for all taxa due to ampliication failure, our inal align-ment contained a proportion of ca.22% missing data.However, all higher taxa (families and subfamilies) had the two mitochondrial genes sequenced from at least one representative (Table 1), and all but one (Valenciidae) had at least two of the three nuclear genes sequenced.Given the overall congruence of single-gene trees we are coni dent that our preferred phylogeny (Fig. 1) has not been inl uenced by missing-data artefacts.
We have refrained from reconstructing a timetree of tooth carps from our data because comprehensively assessing the precise placement of available fossils at particular nodes was beyond the scope of this i rst study.However, despite the absence of reliable dates of origin for the major clades, a number of biogeographical conclusions can be drawn.With aplocheiloids, our tree strongly supports a sister group relationship of the Indo-Malagasy and the African clade (i.e., Aplocheilidae and Nothobranchiidae).This contrasts with previous molecular reconstructions (MurPHy & Collier, 1997) which placed the African and South American radiations sister to each other.Within cyprinodontoids it is remarkable that the recovered topology (Fig. 1), despite low support of many nodes, more closely matches continentscale geographic distribution than current classii cation.
American taxa from both North and South America, including Caribbean islands, are placed paraphyletically at the basis of cyprinodontoids, while an Old World clade (Palearctis plus Africa) is strongly supported.
Our preliminary study highlights a series of promising i elds for further research.First, the family-level classii cation of cyprinodontoids clearly is in need of revision; this requires including additional genera, especially those assigned to the Goodeidae, Profundulidae and Anablepidae which are missing from our data set.Second, inclusion of sequences from Malagasy Pantanodon, and a wider outgroup sampling will be helpful to determine the true phylogenetic position of P. stuhlmanni, a species that might merit inclusion in a distinct family.
Indo-Malagasy aplocheilids and the African nothobranchiids were sister groups with maximum support.Wihin the cyprinodontoids, families and interfamilial relationships were less unambiguously resolved.The Palearctic genus Aphanius was placed outside the family Cyprinodontidae and instead was sister to the equally Palearctic genus Valencia (the sole representative of Valenciidae) with marginal support (95%), and this Palearctic clade was sister to egg-laying African poeciliids (lampeyes).Nearctic and Neotropical cyprinodontids, fundulids and poeciliids instead were placed successively as sister groups of the Palearctic/African clade.