Historical biogeography of aplocheiloid killifishes ( Teleostei : Cyprinodontiformes )

Aplocheiloid killifishes, a diversified group of primary freshwater fishes occurring in tropical and subtropical regions of the Americas, Africa and south-eastern Asia, have been the focus of debates among biogeographers using dispersal and vicariance approaches. The aim of the present paper is to infer biogeographical events responsible for the present distribution of aplocheiloid killifishes using an event-based methodology (DIVA) in conjunction to a phylogeny combining mitochondrial DNA sequences and morphology. Optimal ancestral reconstructions support vicariance events chronologically congruent to northern Gondwana break-up, including separation of Madagascar, India, South America and Africa plates (about 121 – 84 Ma), as well as congruent to paleogeographical events within the Africa plate, such as the widening of the Benue Trough (about 90 – 80 Ma) and the start of activity of the East African Rift System (about 30 Ma), and within the South American plate, as the formation of Gaarland (about 35 – 33 Ma), uplift of the Andean Eastern Cordillera (11.8 Ma) and the interruption of the paleo-Amazonas river basin by the uplift of the Vaupés Swell (about 11 – 7 Ma). The reconstructions also support geodispersal events related to the colonization of the Greater Antilles (about 35 – 33 Ma) and Central America areas (3.7 – 3.4 Ma) by aplocheiloids through land connections, besides some dispersal events through the Zaire, East Africa, Amazon and Eastern Brazil areas.


Introduction
The killifish suborder Aplocheiloidei comprises a diversified clade of small teleosts, usually about 25 -50 mm of total length as maximum adult size, with over 620 species occurring in shallow freshwater or rarely brackish environments.Aplocheiloids are mostly known by many included species being popular aquarium fishes, but some of them are also often used as experimental animals in laboratories (e.g., Harrington & Kallman, 1968;Wourms, 1972;ParK & Kim, 1984).However, the suggestive geographical distribution of the three families contained in the suborder, with Aplocheilidae endemic to Madagascar, Seychelles, and south-eastern Asia, Nothobranchiidae endemic to sub-Saharan Africa, and Rivulidae endemic to Middle and South America (Costa, 2004), has brought the Aplocheiloidei to the focus of debates involving conflicting explanations about patterns of historical biogeography in the tropics (lundberg, 1993;murPHy & Collier, 1997;briggs, 2003;sParKs & smitH, 2005).
Biogeographical relationships among aplocheiloid lineages were first discussed just after myers (1931) formally recognising the killifish group presently known as the suborder Aplocheiloidei.myers (1938), when noticing the possible closely relationships among some South American and African aplocheiloid taxa, claimed that trans-continental distribution of aplocheiloids could not be viewed as evidence of an ancient connection between South America and Africa, since those fish lineages "... belong to the secondary division of fresh-water fishes, and that a fortuitous marine dispersal of one or more species at some time in the Tertiary is not impossible, particularly if a Brazil-Guinea ridge remained for a time as an island chain" (myers, 1938: 353).The term 'secondary division of fresh-water fishes' had been first proposed in that same paper, referring to fish groups, including cyprinodontiforms and cichlids, in which members were often tolerant to salt water.Later, myers (1958) considered the aplocheiloid Asiatic genus Aplocheilus as a basal lineage, suggesting its present geographic distribution as a result of recent dispersal from Africa.
A half century after myers recognising aplocheiloids as a group, Parenti (1981) analysed aplocheiloid relationships, considering Asiatic taxa more closely related to African taxa than to American taxa, at same time defending a Gondwanan origin for the group followed by vicariance events derived from the ancestral Mesozoic continent break-up.This proposal was rejected by lundberg (1993) based both on the absence of cyprinodontiforms and closely related teleost orders in Mesozoic fossil deposits and the incongruence between timing of phylogenetic events (i.e., African taxa more closely related to Asiatic taxa than South American taxa) and the drift-vicariance model (India-Madagascar separating earlier from Africa-South America plate, South America separating later from Africa).In addition, lundberg (1993) tentatively explained the present distribution pattern of aplocheiloids by marine dispersal favoured by the occurrence of some taxa in brackish water habitats, electing the Neotropical region as a centre of origin for aplocheiloid dispersal.
murPHy & Collier (1997), based on a molecular phylogeny for 25 aplocheiloid species and two cyprinodontiform out-groups, in which sister group relationships between taxa from south-eastern Asia and Madagascar/ Seychelles and between a clade of sub-Saharan Africa and another from South America were supported, considered again aplocheiloid distribution as a result of vicariance by fitting well the Gondwana drift model.This hypothesis was not accepted by briggs (2003), who insistently argued again that the group is recent, a conclusion derived from cyprinodontiforms being represented in fossil record only after mid-Tertiary, besides belonging to the myers' secondary freshwater fish category, citing as example Rivulus marmoratus Poey (= Kryptolebias marmoratus), which has been found in numerous coastal localities between Florida and Brazil.A subsequent contrary view was provided by sParKs & smitH (2005), who emphatically criticise the use of unmeaningful statements about dispersal ability and lack of fossils against phylogenetic evidence.Increasing controversy, HertWig's (2008) analysis of morphological characters refuted aplocheiloid monophyly and supported the Asiatic genus Aplocheilus as the sister group to a clade comprising all other cyprinodontiforms.
The break-up of Gondwana, a sequence of geologic events between Jurassic and Cretaceous responsible for delimitating the present continents of the southern hemi-sphere, has been an important paleogeographical scenario for our understanding about timing of biotic evolution in the tropics (e.g., sanmartín & ronquist, 2004).However, evidence extracted from aplocheiloid distribution patterns have some limitations, including reduced number of out-group and in-group taxa in phylogenetic analyses, not representing most lineages, and omission of analytical methods to infer chronological congruence among phylogenetic diversity and paleogeographical events after break-up of Gondwana.The objective of the present study is to analyse the historical biogeography of aplocheiloids using an event-based methodology in conjunction to a phylogeny based on a representative sample of aplocheiloid lineages and outgroups, combining DNA sequences and morphology, in order of checking the temporal congruence between diversification of aplocheiloids and our present knowledge about paleogeographical events in the region inhabited by them.

Material and methods
Phylogenetic analysis.Terminal taxa were select to encompass all the main aplocheiloid lineages, in order of to combine the greatest possible sample of different gene sequences available in GenBank with representative material available for morphological analysis (see Costa, 2012 for a list of material used in morphological studies).Generic classification followed Costa (1998aCosta ( , 2010Costa ( , 2011a) ) (Costa, 2007(Costa, , 2011b)).Sequences were aligned using Clustal-W (CHenna et al., 2003) and subsequently optimized manually.
Morphological characters were extracted from phylogenetic studies involving aplocheiloid lineages and outgroups (Parenti, 1981(Parenti, , 2008;;rosen & Parenti, 1981;Costa, 1998bCosta, , 2004Costa, , 2006Costa, , 2008Costa, , 2009aCosta, , 2009bCosta, , 2011cCosta, , 2012)), where character and character states are discussed and illustrated.Characters were reformatted to reduce ambiguity following sereno (2007) for character statement formulation.Morphological characters included osteology, laterosensory system and other cephalic structures, fin morphology and colour patterns.Colour patterns when consistently identified (i.e.all specimens of two or more terminal taxa sharing a similar colour pattern, composed of one or more colours, at the same position of a morphological structure) were treated as independent characters.Characters showing high levels of subjectivity in character state delimitation among the numerous terminal taxa (e.g.dorsal and anal fin morphology, flank colour patterns in males) were excluded from the analysis.All characters were treated as unordered.A total of 199 morphological characters were examined (Appendix 1), which were checked both in terminal taxa and in species closely related to terminal taxa, in a total of 280 species of aplocheiloids and 28 species of outgroups (see complete list in Costa, 2012).Osteological preparations (c&s) were made according to taylor & Van dyKe (1985).Terminology for frontal squamation followed Hoedeman (1958) and for cephalic neuromast series Costa (2001).Distribution of character states of morphological characters among terminal taxa appears in Appendix 2.
The search for most parsimonious trees (using 'traditional' search and setting random taxon-addition replicates to 10, tree bisection-reconnection branch swapping, multitrees in effect, collapsing branches of zerolength, characters equally weighted, and a maximum of 1,000 trees saved in each replicate) and bootstrap analysis (1,000 replicates) were performed with TNT 1. 1 (goloboff et al., 2008).Character states of all morphological characters were treated as unordered and genes were analysed giving equal weight to all sites.
Biogeography.Reconstruction of the biogeographical history of aplocheiloid lineages was inferred using the dispersal-vicariance analysis (DIVA) (ronquist, 1996, 1997), an event-based parsimony method with an explicit treatment of multiple identified processes (vicariance, dispersal, extinction, and sympatric speciation), with relative costs previously assumed (vicariance and sympatric speciation events with a cost of zero; dispersal and extinction events with cost of one per unit area added or deleted from the distribution) (ronquist, 1996).This method is recommended in studies aiming to reconstruct the biogeographical history of a single lineage in the absence of a general area cladogram (e.g., ronquist, 1997; sanmartín, 2007).Although recent developments in geology make possible identification of several paleogeographical events, not much has been known from drainage evolution in the South American plate, as well as high level of controversy is still found in the literature about the drainages of the African plate (e.g., goudie, 2005).Thus, DIVA methodology is herein chosen by having the advantage of not being constrained by particular modes of areas relationships (e.g., Van boCxlaer et al., 2006).The search for the optimal historical reconstruction, in which the total cost is minimized under a parsimony criterion, was performed using the computer program DIVA 1.2 (ronquist, 1996).The exact search of DIVA was used without restricting the number of areas in which the ancestor occurred.
Areas of endemism were delimited on the basis of congruence among distributional ranges exhibited by monophyletic groups of aplocheiloid fishes.The 11 areas of endemism herein analyzed, with the respective taxonomic units supporting them, were: 1) Greater Antilles (GA) -including Cuba, Pinos and Hispaniola islands, the geographic distribution of the genus Rivulus.2) Central America (CA) -all extension of continental Central America, including southern Mexico in north, and the adjacent trans-Andean portion of South America containing the Magdalena River basin in the south, the geographic distribution of Cynodonichthys.3) Orinoco (Or) -comprising the Orinoco river basin and adjacent coastal areas, including the Maracaibo region, the geographic distribution of the genera Austrofundulus, Gnatholebias, Rachovia, Terranatos, Renova, Micromoema, and one monotypic genus, Llanolebias that is closely related to Gnatholebias (HrbeK & taPHorn, 2008) and was not included in the analysis.4) Amazon (Am) -including the whole Amazonas river basin and adjacent river basins in Guianas and northeastern Brazil, the geographic distribution of Anablepsoides, Aphyolebias, Laimosemion, Maratecoara, Moema, Neofundulus, Pituna, Plesio le bias, Pterolebias, Spectrolebias, Trigonectes, and two rare plesiolebiasine genera, Papiliolebias and Steno le bias, closely related to Maratecoara and Pituna (Costa, 2011c).5) Brazilian Shield (BS) -including the Paraná, São Francisco, upper Tocantins, upper Araguaia, middle Je quitin honha and Jaguaribe rivers basins, besides adjacent smaller basins in northeastern and southern Brazil, the geographic distribution of Austrolebias, Cynolebias and Hypsolebias, and another cynolebiasine genus Simp so nichthys, not included in the analysis.6) Eastern Brazil (EB) -comprising the coastal rivers drainages between Para guaçu river and Patos lagoon basins, the geographic distribution of the genera Atlantirivulus, Cam pel lo lebias, Leptolebias, Nematolebias, Notho le bias, and Krypto le bias with exception of an estuarine species widespread along northeastern Brazil and Caribbean region (K.marmo ratus Poey); also endemic to the area, but not included in the analysis, are Cynopoecilus, the sister group of Cam pel lo lebias (e.g.Costa, 1998a), Ophthal mo lebias and Xenu ro lebias, two basal cynolebiasine genera (e.g.Costa, 2010), and Prorivulus, a basal rivulid genus of uncertain phylogenetic position.7) West Africa (WA) -includ ing river basins between Gambia and western Ghana, the geographic distribution of Archiaphyosemion, Cal lo pan chax and Scriptaphyosemion.8) Zaire (Za) -including the Zaire and Niger river basins, besides smaller adjacent coastal drainages between eastern Ghana and Cabinda (Angola), the geographic distribution of the genera Aphyo semion and Fundulopanchax, besides two small genera not included in the analysis, the rare Episemion and the monotypic Foerschichthys.9) East Africa (EA)including East African river basins between Somalia and  Summary of the optimal reconstructions of ancestral distribution using dispersal -vicariance analysis.Broad vertical lines in the tree are hypothesized vicariance events associated to dated paleogeographical events indicated by broad arrows, whereas narrow vertical lines have arbitrary position relative to time scale; black squares on nodes are vicariance events not attributed to particular paleogeographical events and black circles are duplications; narrow arrows indicate dispersal (+) events.Nodes are sequentially numbered; ancestral distributions appear near nodes when unambiguously assessed.Areas of endemism: (Am) Amazon, (BS) Brazilian Shield, (CA) Central America, (EA) East Africa, (EB) Eastern Brazil, (GA) Greater Antilles, (In) India, (Md) Madagascar, (Or) Orinoco, (WA) West Africa, (Za) Zaire.
South Africa, Rift Lakes basins, White Nile River drainage, and Lake Tchad basin, the geographic distribution of Nothobranchius.10) Madagascar (Md) -comprising Madagascar and Seychelles, the geographic distribution of Pachypanchax.11) India (In) -comprising India and Sri Lanka, the geographic distribution of Aplocheilus, except for A. panchax, a salt tolerant species occurring since southern Pakistan to East Timor.
Terminal taxa for the biogeographical analysis were monophyletic units with consistent geographical distribution, usually genera.As a consequence, species were substituted by their respective genera.In order of to reduce the number of terminal taxa in the biogeographical analysis, some monophyletic groups of genera sharing the same geographical range were clustered into a single terminal taxa.Although two salt water tolerant species of the genera Aplocheilus and Kryptolebias being geographically widespread to extend their geographic range much beyond the areas of endemism herein delimited, since freshwater basal species of those genera are geographically restricted to India and Eastern Brazil areas, respectively, their distribution ranges were assigned to the latter areas.Since the South American Paraguay river basin has been considered as a recent composite biogeographical area (e.g., Costa, 2010, 2011c), genera occurring both in the Amazon area and Paraguay basin were assigned as endemic to the former area, whereas the only genus simultaneously occurring in Paraguay basin and Brazilian Shield area, Austrolebias, was assigned to the latter one.

Results
The single most parsimonious phylogenetic tree for 61 aplocheiloid taxa and six outgroups is illustrated in Fig. 1.It corroborates monophyly of the Aplocheiloidei and included families.The biogeographical analysis indicates a series of vicariance events responsible for family distribution, whereas distribution of intra-familial groups is shaped both by vicariance and dispersal events (Fig. 2), which are below discussed ant tentatively associated to paleogeographical scenarios.
The first three most basal vicariance events, represented by nodes 1, 2 and 3 in Fig. 2, are congruent to the separation of Madagascar and India areas from northern South America plus Africa, Madagascar area from India area (break-up of Mascarene basin, 84 Ma), and South America from Africa (opening of the South Atlantic Ocean, about 110 Ma), respectively.Within the African lineage, two vicariance events delimited by DIVA reconstruction may also be associated to paleogeographical events.The first one (node 4 in Fig. 2 The second vicariance event in Africa (node 6 in Fig. 2) indicates a fragmentation in central Africa, isolating the eastern area today inhabited by Nothobranchius (East Africa area), from the mid-western area inhabited by the aphyosemiine clade comprising Aphyosemion and Fundulopanchax (Zaire area).The limit between both areas is congruent with multiple paleogeographical events associated with the start of activity of the East African Rift System in the Oligocene, about 30 Ma (burKe, 1996).These events comprise a series of expressive transformations in the long Trans-African paleo-rivers, a large drainage network including at different times portions of the present Niger, Tchad, Zaire, Nile and Zambezi courses, formerly draining to east and northeast (e.g., goudie, 2005; stanKieWiCz & de Wit, 2006).Sections of the system were successively disrupted by the uplift associated with the East African Rifting, imposing new barriers for aplocheiloids between Zaire and Eastern Africa areas.
In the Neotropical region, DIVA analysis indicates some possibilities for ancient isolation of basal lineages in Eastern Brazilian area, which may be considered as a consequence of the formation of the passive continental margin of South America just after the break-up of South America and Africa (Potter, 1997;ribeiro, 2006).The basal divergence between the subfamilies Cynolebiasinae and Rivulinae is parsimoniously hypothesized to be related to a vicariance event separating the former group in Eastern Brazil area from the latter in Amazon area (node 8).The evolution of each subfamily is predominantly characterized by a series of duplication events with in those areas, with geodispersal events associated to vi ca riance events, some which are chronologically congruent with some major key paleogeographical events in South America.Distribution of the Cynolebiasinae implicates an ancestral distribution in East Brazil and Brazilian Shield areas, besides one dispersal event to the Amazon area.A recent biogeographical analysis of species of the tribe Cynolebiasini provides an alternative scenario (Costa, 2010), in which the gradual formation of taphrogenic basins due to vertical movements between rifted blocks caused river capture from the adjacent upland (the Brazilian Shield area), and consequent multiple biotic dispersals from the Brazilian Shield area to the Eastern Brazil area (almeida & Carneiro, 1998;ribeiro, 2006).This paleogeographical model is congruent with the occurrence of successive basal lineages of the cynolebiasine clade in the Eastern Brazil area (Costa, 2010).
DIVA indicates the most recent ancestor of the Rivulinae as endemic to the Amazon area, which is inhabited by representatives of most Rivulinae lineages, including Laimosemion that is supported as the sister group to the remaining rivulines.The second most basal split in the Rivulinae clade indicates a dispersal event from the Amazon area to the Greater Antilles area, followed by vicariance isolating a Greater Antilles area taxon, which may me attributed to the formation and later rupture of a long land connection between central Cuba and northern South America, known as Gaarland, during the transition Eocene -Oligocene, about 35 -33 Ma (iturralde-Vinent & maCPHee, 1999; iturralde-Vinent, 2005).During the Upper Oligocene (about 27 -25 Ma), as a consequence of a higher sea level, this connection was interrupted, a paleogeographical scenario implicating in isolation of that rivuline lineage, the ancestor of Rivulus.
Duplication events within the Amazon area are supported by DIVA analysis, thus indicating that all the ancestors of each of the three most diversified clades of the Rivulinae were endemic to that area.Among lineages of the melanorivuline clade, a dispersal event within the genus Melanorivulus is proposed to be occurred from the Amazonian area to the Brazilian Shield area.This dispersal event is supported by a recent taxonomical study, in which M. schuncki (Costa & de luCa), the only species of the genus endemic to the lower Amazon, is considered as the probable sister-group to the clade comprising the remaining about 35 species of the genus (Costa & de luCa, 2011).Another dispersal event is hypothesized to have occurred from the Amazon area to the south-eastern Brazil area, followed by vicariance separating the latter area inhabited by the ancestor of Atlantirivulus (node 18).
More interestingly, the two latest events assigned to the melanorivuline clade (node 19) are more parsimoniously explained as dispersal-vicariance events involv-ing the Central America, which has been connected to South America just recently.The uplift of the Andean Eastern Cordillera at 11.8 Ma (Hoorn et al., 1995), with the formation of the present Magdalena Valley and isolation of the plain areas of north-western corner of South America, is herein hypothesized as a first causal factor for the vicariance event originating the present genera Anablepsoides from Cis-Andean northern South America and Cynodonichthys endemic to the region encompassing Central America and Trans-Andean northern South American (Magdalena river basin).This paleogeographical scenario was chronologically followed by the formation of the connection between that northern corner of Trans-Andean South America and Central America through the closure of the Panama isthmus between 3.7 and 3.4 Ma (duque-Caro, 1990; Hoorn et al., 1995), permitting geodispersal of the ancestor of Cynodonichthys through the newly formed Central America.
DIVA supports a later dispersal event to the Orinoco area, followed by vicariance separating both areas (node 22).A remarkable evolutionary radiation would then follow the initial colonization of the Orinoco area by a rivuline ancestor, thus originating the highly morphologically diverse genera (e.g.Austrofundulus, Gnatholebias, Terranatos) today found in the area, composing the rachoviine clade.These events are congruent with recent paleogeographical reconstructions, which have consistently shown that the paleo-Amazonas river flowed to north, thus continuous to the present Orinoco river basin (Hoorn et al., 1995(Hoorn et al., , 2010;;mora et al., 2010).This river basin configuration was broken with the uplift of the Vaupés Swell occurred between the Late Miocene and Pliocene, about 11 -7 Ma, as a result of a sequence of processes related to the Andean Orogen (Hoorn et al., 2010;mora et al., 2010), separating the Orinoco and Amazonas river basins, and forming the present transcontinental Amazonas basin flowing to east.Therefore, biogeographical pattern exhibited by the rachoviine clade is congruent with the paleo-Amazonas-Orinoco basin, whereas the rupture of that basin is herein considered as a causal factor promoting the vicariance event assigned to the node 22.
Much insight about evolution, biogeography and historical ecology of aplocheiloids came during collecting trips; I am especially grateful to Claudia Bove and Bruno Costa for accompanying me in numerous trips in South and Middle America, Africa and south-eastern Asia localities.I am deeply indebted to all those that have provided material to study, particularly Douglas Nelson for the exchange of large important collections and Maurice Kottelat for the gift of rare material.This study was funded by CNPq (Conselho Nacional de Desenvolvimento Científico e Tecnológico -Ministério de Ciência e Tecnologia).Thanks are due to the Willi Hennig Society for making TNT available.

Fig. 1 .
Fig. 1.Most parsimonious cladogram of phylogenetic relationships between 61 representatives of the Aplocheiloidei and six outgroups, based on a combined set of segments of three mitochondrial genes (cytochrome b, 12S rRNA and 16S rRNA; 1145 sites, of which 655 were parsimony-informative, 333 were constant and 158 were variable but not parsimony-informative) and 199 morphological characters (tree length: 7447).Numbers above branches are bootstrap values.Dotted rectangle delimits the Aplocheiloidei.

Fig. 2 .
Fig.2.Summary of the optimal reconstructions of ancestral distribution using dispersal -vicariance analysis.Broad vertical lines in the tree are hypothesized vicariance events associated to dated paleogeographical events indicated by broad arrows, whereas narrow vertical lines have arbitrary position relative to time scale; black squares on nodes are vicariance events not attributed to particular paleogeographical events and black circles are duplications; narrow arrows indicate dispersal (+) events.Nodes are sequentially numbered; ancestral distributions appear near nodes when unambiguously assessed.Areas of endemism: (Am) Amazon, (BS) Brazilian Shield, (CA) Central America, (EA) East Africa, (EB) Eastern Brazil, (GA) Greater Antilles, (In) India, (Md) Madagascar, (Or) Orinoco, (WA) West Africa, (Za) Zaire.
) separates West Africa from the remaining parts of Africa, thus isolating the ancestor of the clade comprising Archiaphyosemion, Callopanchax, Scriptaphyosemion and Epiplatys, an event also detected by murPHy & Collier (1997).This event is here tentatively related to the widening of the Benue Trough, occurred still in the Cretaceous, about 90 -80 Ma (e.g., fairHead, 1988; fairHead & binKs, 1991; stanKieWiCz & de Wit, 2006), when Africa still moved as two plates separated by rifting in West Africa, erecting a potential geographical barrier for aplocheiloids.The reconstruction model supported by DIVA also demands a dispersal of Epiplatys lineages from West Africa to Zaire and East Africa areas, events already in part postulated in the recent phylogeographical analysis of the genus Epiplatys (Collier et al., 2009).Epiplatys is a species-rich and widespread genus, with lineages endemic to both West Africa and Zaire areas (Collier et al., 2009), besides some species inhabiting the White Nile and Tchad regions of the East area herein delimited.