Phylogeny and historical biogeography of trans-Andean cichlid fishes ( Teleostei : Cichlidae )

We reconstruct the historical biogeography of cichlid fishes endemic to the trans-Andean region of NW South America. DNA sequences were used to study historical biogeography of the cichlid genera Andinoacara (Cichlasomatini) and Mesoheros (Heroini). Two eventbased methodological approaches, parsimony-based Statistical Dispersal-Vicariance Analysis (S-DIVA) and likelihood-based DispersalExtinction Cladogenesis (DEC in Lagrange) were used for ancestral-area reconstructions. Molecular clock analysis of the whole group of Neotropical Cichlidae (using mtDNA and nucDNA markers) was calibrated using BEAST by six known cichlid fossils. The historical biogeography of both studied trans-Andean cichlid genera is best explained by a series of vicariance events that fragmented an ancestrally wider distribution. Both genera have a highly congruent vicariant historical biogeography in their shared distribution in the ColombianEcuadorian Choco. The Andean uplift and formation of the Central American isthmus strongly impacted the distribution patterns of the freshwater ichtyofauna in the NW Neotropics as suggested by the historical biogeography of the two studied cichlid groups. Despite strong congruence in their historical biogeography the two studied cichlid lineages (part of the tribe Cichlasomatini and Heroini, respectively) have highly different evolutionary substitution rates in the studied mtDNA cytb marker.


Introduction
Cichlid fish diversity in trans-Andean South America The core of trans-Andean South America (the Pacific slope river drainages of South America plus the Mag dalena and Maracaibo river regions in the Caribbean drainage) is formed by the Tumbes-Choco-Magdalena (TCM) region.This is one of the 25 biodiversity hotspots (Reid, 1998;MyeRs et al., 2000), and Choco further represents one the wettest regions on the planet.The TCM region, extending from eastern Panamá to north-western Perú, is bordered by two additional biodiversity hotspots, Mesoamerica in the northwest, and the Tropical Andes (in terms of global biodiversity levels the richest region) in the east.
Compared with the high biodiversity of terrestrial animals and plants, and with the great watersheds on the Atlantic (cis-Andean) side of South America (SA), the freshwater fish fauna of trans-Andean SA is relatively sparse.The TCM hotspot has about 250 species of freshwater fish (nearly half of which are endemic) in 54 families (MitteRMeieR et al., 2005).In total, representatives of four cichlid groups (Andinoacara Mu si lová,Říčan& novák2009, Mesoheros McMahan&cha krabarty2015 in McMahan et al. 2015, Caquetaia Fowler 1945, and Geo phagus heckel1840) are distributed in the TCM region and only two of these genera are distributed throughout the whole TCM region (Andinoacara and Meso heros).Mesoheros is endemic to the TCM region (bordered in the south by the Tumbes River drainage in NW Perú), while Andinoacara distribution includes also the Magdalena basin and reaches even further in trans-Andean SA (Lake Maracaibo) and also into cis-Andean northern SA (in the Orinoco River drainage and on Trinidad), and in the south almost to Lima in central Peru.
Andinoacara belongs to the tribe Cich laso matini and presently includes eight valid species (Mu silová et al., 2009a, b;one of them described recent ly -wijkMark et al., 2012).Due to its distribution range it represents one of the most suitable cichlid genera for historical biogeography studies of trans-Andean SA related to the uplift of the Andes and formation of the Central American isthmus.The present study includes all of the valid Andinoacara species and three additional lineages with uncertain taxonomic status.
Mesoheros is a representative of the tribe Heroini (sister taxonomic group to Cichlasomatini) and is related to a subgroup of Middle American cichlid fishes (Říčan et al., 2013).Samples from the whole distribution area of the group except for south-central Pacific Colombia (putative M. gephyrus) are included in this study.

Historical Biogeography of trans-Andean South America
NW South America represents a region with a high level of geological activity where large changes have occurred in the geological past.The uplift of Andes initiated already in the Late Cretaceous (lundberg et al., 1998) but was most prominent in the second half of the Cenozoic and has heavily impacted the geological and biological history of South America (lundberg et al., 1998; antonelli et al., 2009).It significantly reshaped the river systems on the whole continent including the large proto-Orinoco-Amazon system (rodriguez-olarte et al., 2011), and the uplift itself formed an important barrier for species distributions in the region of Northern South America and between South and Middle America (lundberg et al., 1998).The major changes in configuration of the major river basins occurred quite recently, between 10 and 3 Mya, when e.g.trans-Andean SA was separated from the Amazon, and when Orinoco became independent from the paleo-Magdalena and the Amazon river (except for two adjacent connections still persistent to this day).Lake Maracaibo, the former delta of the proto-Orinoco-Amazon system, was formed by rise of the Merida Andes in Late Pliocene (rodriguez-olarte et al., 2011).Such intense watershed reshaping can strongly impact the species distribution as well as promote the spe ciation process (Joyce et al., 2005).Additionally, species distribution patterns were also affected by concurrent periodical marine incursions into the continent (lund berg et al., 1998;Montoya-burgos, 2003).Lastly, the incipient formation of the Panama isthmus in the Miocene/Pliocene (followed by its break up and renewed formation around 3 Mya;coates et al., 2004) played a major role in the colonization of Central America by freshwater fishes (berMinghaM & Martin, 1998;sMith &berMinghaM,2005;Říčan et al., 2013).
Prior to ca. 10 Mya contacts between cis-and trans-Andean regions would have been enabled by at least two connections.First, the uplift of the Northern Andes (Western, Central and Eastern Cordilleras in Colombia and Merida Andes in Venezuela) was incomplete and included also marine incursions until 11.8 Mya (lundberg et al., 1998).Second, the Western Andean portal, a connection and also putative marine incursion into the continent between the Northern and Southern Andes at the current Ecuadorian-Peruvian border, probably persisted up to 13 -11 Mya (antonelli et al., 2009).

Distribution patterns in Neotropical fishes
Generally, allopatric speciation promoted by vicariance is the most common way of fish speciation in the Neo tropical region and this is surprisingly true even for the Amazon, despite the presence of only few obvious hard barriers (albert & reis, 2011).The mountainous Northern South America is a typical example of a region with distribution barriers represented by several mountains ranges.The uplift of the four northern Andean mountain chains (Western, Central and Eastern Cordillera in Colombia, and the Merida Andes in Venezuela) during last 10 My strongly impacted the biota inhabiting the original continuous area of the recent Magdalena, Maracaibo and trans-Andean drainages.Many species of freshwater fishes in this region went extinct here but have (or their close relatives have) survived in the larger river systems of neighbouring South America, such as the Amazon or Orinoco (rodriguez-olarte et al., 2011).There are only a few genera shared across more river systems in Northern South America, and both dispersal and vicariance theories were used to explain these distribution patterns (rodriguez-olarte et al., 2011).Cichlids (i.e. genus An di noa cara) represent one of the lineages widely distributed throughout virtually all of NW South America.Most of the cichlid species in this region have allopatric (i.e.non-overlapping) distribution areas, except for the trans-Andean Choco region where basal species of two Andinoacara lineages live in sympatry (and syn topy; ZM and OG, pers.obs.).

The main goal of the present study
This is the first detailed study of cichlids from trans-Andean and NW South America focusing on their species diversity and historical biogeography.Previous studies considered the whole tribes Cichlasomatini and Heroini and have provided the biogeographical overview on a broader scale (Musilová et al., 2008;Říčan et al., 2013).The historical biogeography approach is used herein to reconstruct ancestral areas and the evolutionary histories by distinguishing between vicariance and dispersal events during evolution of the two cichlid genera.Further, the historical biogeography of the two cichlid genera is dated using molecular clocks calibrated by fossils in a wider all-Neotropical cichlid context.Lastly, we interpret the historical biogeography of the cichlid fishes in comparison with the geological history of NW South America.

Material used in this study
The material of more than 150 specimens of eight An dinoacara species and three Mesoheros species was collected through museum collections, university collections, via aquarium trade importers and several private collections of aquarium hobbyists including our own for the specific task of the present study.Samples were only obtained from trusted aquarium hobbyists or importers who carry pure lineages with known localities of origin and these were verified independently with multiple samples.Finally, samples from previous studies from GenBank were used to complement the data sets (Table 1).
Our sampling covers virtually the whole distribution areas of both cichlid groups.The areas not represented in our data sets are the Colombian basins Dagua, Patia and Mira from departments Valle de Cauca, Cauca and Nariño with a remaining complicated geopolitical situation.

DNA sequencing and phylogenetic analyses
The mitochondrial cytochrome b gene was amplified using the primers FishCytB-F and TrucCytB-R from sevilla et al. (2007).The process of DNA isolation, amplification and sequencing, as well as sequence editing and alignment was identical to the protocol used in Musilová et al. (2009a) andconcheiro-Pérez et al. (2007).All obtained sequences in this study were submit-ted to GenBank (KJ127312 -KJ127457 for Andinoacara and KJ127458 -KJ127474 for Mesoheros).
Phylogenetic analyses of mitochondrial sequence data (cytochrome b) were performed using Bayesian Inference as implemented in MrBayes 3.0 (huelsenbeck &ronquist, 2001).The best-fit model for genetic evolution was selected with jModeltest (Posada, 2008) using the Akaike information criterion.Bayesian analyses were performed using two independent runs of four Metropolis-coupled chains (MCMC) of 10 million generations each, to estimate the posterior probability distribution.Tree topologies were sampled every 100 generations, and majority-rule consensus trees were estimated after discarding the first 25% generations.Bayesian analyses were performed for the "full data set", where all samples were included and also for the "haplotype data set" for unique haplotypes.

Molecular clock and fossil calibration of Neotropical cichlids
Since no fossil is known for Andinoacara and only one fossil is known for a related genus of Mesoheros ( †Nan dopsis woodringi), we have used an external dating approach.First we have reconstructed the phylogenetic tree for all neotropical cichlids based on published data (see Appendix S1 and Figure S1).This "Neotropical cichlid tree" reconstructed by Bayesian Evolutionary Analyses by Sampling Trees (BEAST, druMMond & raMbaut, 2007) was then calibrated by all known cichlid fossils, for which both the phylogenetic position and a non-ambiguous age estimate are available (Appendix S1).Based on this "Neotropical cichlid tree" we estimated the age of both Andinoacara and Mesoheros root nodes and we used these estimates as secondary calibration points for the molecular clock analyses of both studied groups.The detailed genus-level ultrametric phylogenetic trees were obtained by BEAST analyses under the relaxed molecular clock models.The run in BEAST was performed for 10 million generation under the evolutionary model as suggested by jModeltest (for detailed parameters see Appendix S1).
Additionally, divergence rates of cytochrome b were calculated as the uncorrected p-distances based on haplotypes for each node in PAUP software (swoffoRd, 2003).

Historical biogeography analyses
We performed historical biogeography analyses by two approaches, the parsimony-based Statistical Dispersal -Vicariance Analyses (S-DiVA, yu et al., 2010) and the maximum-likelihood based Dispersal-Extintion-Cla dogenesis model (DEC) in Lagrange (ree&sMith, 2008).The goal of the analyses was to reconstruct vicariance/ dispersal scenarios for both Andinoacara and Mesoheros, as well as putative ancestral areas of both genera and all  nodes.We used S-DiVA for the datasets of both studied cichlid groups providing detailed biogeographic information (up to 10 distribution regions were allowed in order to allow for complete vicariance, see below).Such data set was however too large for the DEC analysis and it did not converge.We further applied both S-DiVA and DEC analyses on the genus Andinoacara with larger-scale biogeographic units (four "super regions") to specifically address the question about the role of trans-Andean region in the evolution of this genus.For both analyses, we used only unique haplotypes to reconstruct the ultrametric trees to approximate the assumption of fully resolved (bifurcated) tree required by biogeographic softwares.Haplotypes were defined in the FABOX software (villesen, 2007) and the trees were reconstructed by both MrBayes and BEAST methods.The fully bifurcated trees served then as inputs for the biogeo graphy analyses.
The S-DiVA represents an event-based approach for the reconstruction of historical biogeography and it summarizes the reconstructions across a sampled pool of trees from the Bayesian analysis runs.In S-DiVA we analysed the data sets from four different runs of BEAST to find the ancestral areas and to reconstruct the biogeographic scenario of the genera Andinoacara and Mesoheros.The last 5000 of sampled trees from each BEAST run were used (from the total of 10000 sampled trees) for the S-DiVA analyses.Similarly, the MrBayes consensus tree as well as its 5000 trees randomly sampled from the MrBayes run .tfiles served as an input for S-DiVA.
The DEC model analysis in Lagrange was performed using the haplotype MrBayes and BEAST trees, similar as for the S-DiVA analysis.The input file for Lagrange software (ree&sMith, 2008) was generated on the online web configurator (http://www.reelab.net/lagrange/configurator/index) requiring the ultrametric tree and the distribution matrix.We performed the analysis for different settings considering the maximum number of areas allowed in the ancestral areas (from 2 to 5 having four bigeographic regions + outgroup).
The following distribution regions were used for coding of biogeographical analyses in Andinoacara and in Mesoheros: A -Orinoco, B -Maracaibo, C -Magda le na, D -Tuira (eastern Panamá -Pacific slope), E -Chi ri qui (SW Costa Rica plus W Panama -Pacific slope), F -Choco (Colombia), G -Cayapas/Santiago system (Ecua dor, Pacific slope), H -Ecuador + Rio Tumbes (Pa ci fic slope), I -Peru (Pacific slope), J -Chagres (eastern Pa na má -Caribbean slope) (see also maps in Fig 2 and 3).Fishes from the genus Andinoacara are found in nine (A -I) from these ten regions and Mesoheros distribution covers four regions (F, G, H, J).
In the more widely distributed genus Andinoacara we additionally performed a biogeographic analysis with larger "super regions" (more akin to the wider freshwater ecoregions sensu abell et al. (2008)).These "super regions" were: TR -trans-Andean region of South America (corresponds to the aforementioned regions F, G, H and I), MG -Magdalena basin (= region C), OR -Orinoco & Maracaibo basin (= regions A and B) and CA -Central America (= regions D and E).
All biogeographic analyses were run as unconstrained (connections between all areas were allowed) and the maximum of the areas allowed in the ancestral area corresponding to the number of regions + 1 (outgroup), i.e. 10 and 5, respectively, not to introduce any bias in a pri ori constraining the parameters.Further, analyses with different set of allowed maxareas was performed to test the data under different assumptions.

Phylogeny of trans-Andean cichlids
The genus Andinoacara is divided into two well supported clades (the A. pulcher group = the northern clade, and A. rivulatus group = the southern clade; Fig. 1a, 2).Within the northern clade (A.pulcher group), three species and two undescribed lineages were found as monophyletic clades.In Mesoheros all three studied species were found as monophyletic with strong support (Fig. 1b) and M. atro maculatus (regan1912) from Atrato, San Juan, and neigh bouring river drainages represents the northern sister group to the well supported southern clade of M. orna tus (regan1905) from northern Ecuador and southern Colombia and M. festae (boulenger1899)from the rest of Ecuador and northernmost Peru.

Distribution patterns of trans-Andean cichlids
The observed distribution patterns in both cichlid genera (Andinoacara and Mesoheros) are highly congruent.In both genera, two sister clades are present, one with the distribution in the southern and one in the northern part of the area.The three species of Mesoheros can be considered as corresponding to the three southern Andinoacara rivulatus group species when excluding A. stalsbergi (the only Neotropical cichlid in its southernmost trans-Andean area).The northern clades of both genera have made it into Central America, where they, however, are not sympatric (unlike in Choco), with M. atromaculatus limited to the southernmost isthmus on the Caribbean slope while A. coeruleopunctatus to the Pacific slope up to southern Costa Rica (Figs 2 and 3).
In both genera there is an old divergence between the Santiago river system (Western Ecuador) and the rest of Ecuador (+ Peru) (Fig. 1, 2 and 3).The Santiago river system in Western Ecuador (together with neighbouring southern Colombia) represents therefore an important

Historical biogeography and molecular clock of trans-Andean cichlid fishes
The relaxed molecular-clock analysis based on the Neotropical cichlids data set estimated the age of An di noa cara to be 19.33 million years (My), and of Me so heros 14.17 My (Appendix S1).These time estimates were then used as secondary calibration points for the individual Andinoacara and Mesoheros molecular clock estimates.The estimated ages within the studied groups are shown in Fig. 2 for Andinoacara and in Fig. 3 for Mesoheros.
Both biogeographic methods (i.e.S-DiVA and DEC-Lagrange) applied within this study were to some extent sensitive to the set parameters, namely to the maximum of areas allowed to be present in the ancestral area ("maxareas").We ran the analyses with various possible settings, ranging from a large ancestral area (allowing all tested biogeographic regions in maxareas; i.e. nine regions in Andinoacara and four regions in Mesoheros), through a wide-scale dataset (four "super" regions for Andinoacara) to only two areas allowed.See Fig. 4 and ancestral nodes in Fig. 2 and 3.The DEC-Lagrange method was performed only on the wide-scale ("super regions") analysis, because the more complex data sets with fine-scale analysis of nine regions were not converging.There is a justification for the maxareas constraints, such as the non-existence of connection of developing Central America with South America until the Great American Biotic Interchange (and thus its impossible presence in the ancestral area) in the time of lineage origin, or the fact that the maximum of areas occupied by any of recent species is low, e.g. in Andinoacara any species is present in just two "super regions" or three biogeographic regions.Such information has been used as a constraining estimation for the maxareas set in several studies, e.g. in Amazonian frogs (santos et al., 2009) or birds (nylander et al., 2008).On the other hand, vicariance is often considered as a null hypothesis in biogeography and any restriction in the number of maxareas may cause the a priori rejection of this hypothesis.
Based on our results we thus prefer the unconstrained analyses showing that vicariant events and allopatric speciation were the most important drivers of diversification in both studied groups (Andinoacara and Mesoheros).The ancestral area of Andinoacara likely consists of the whole of its recent distribution except for the Maracaibo region and probably Central America.This latter point represents a potential conflict between our analyses (where Central America is within the ancestral area in virtually all analyses) and the geological development of the Central American isthmus (which was probably not connected to South America until the Pliocene; but see Montes et al. 2015 and Discussion).
The following scenario describes a series of predominantly vicariant events that have shaped the present diversity of Andinoacara.The first event separated the northern and southern Andinoacara clades within the region of the Choco (F) at .Recently, both lineages live sympatrically in Choco, and therefore secondary dispersal within the region occurred later.The following three vicariant events were reconstructed in the southern clade, the first of which occurred between the Chocoan A. biseriatus (F) and the remaining clade (G, H, I) around 13.74 Mya (19.7 -8.7 Mya), followed by a vicariance separating A. sapayensis in the Santiago river (G) from the rest of Ecuador and Peru (G, H) at 11.97 Mya (17.1 -7.1 Mya), and the third vicariant event occurred between A. rivulatus and A. stalsbergi and sep arated the Ecuadorian (H) and Peruvian (I) areas at 2.9 Mya (4.0 -1.2 Mya; Fig. 2).In the northern clade there are two alternative scenarios for the basal node which either shows a vicariance of the Choco (F) from the rest (D, E, C, A; MrBayes topology), or a vicariance between Choco (F) + Central America (D, E) from Magdalena (C) + Orinoco (A; BEAST topology).The BEAST topology is the one that provides dating of the biogeographical events and based on it the basal vicariance in the Northern clade was dated at 5.28 Mya (3.1 -8.0 Mya) followed by the separation of Choco (F) from Tuira (D) + Chiriqui (E) at 4.39 Mya (2.3 -7.0 Mya), and by vicariance between the two Central American provinces, i.e.Tuira (D) and Chiriqui (E) at 2.85 Mya (1.4 -4.5 Mya).In the Magdalena-Maracaibo-Orinoco clade (A.latifrons -A.pulcher -A.sp."Orinoco" complex) vicariance between Magdalena (C) and Orinoco (A) occurred at around 3.36 Mya (5.3 -1.7 Mya; Fig. 2).
Only two dispersals were reconstructed in the An di noacara biogeographic scenario, both within the same lineage of A. pulcher.First is the colonization of Ma racaibo (B) and the second is the recolonization of Mag dalena (C), both from the Orinoco region (A).Both dispersals were found to be recent colonisations and the fishes from Maracaibo, Magdalena and Orinoco basins even share haplotypes in this case (Fig 2).
The biogeographic scenario of Mesoheros also has some ambiguity in its resolution (Fig. 3).It requires either four vicariances, or three vicariances and one dispersal event to explain the present distribution pattern (Fig. 3).The difference is in the presence or absence of Chagres (eastern Panama) in the ancestral area, when the first scenario would suggest a completely vicariant explanation, whereas the alternative scenario would assume a later dispersal from the Choco to the Chagres (both reconstructions have a similar probability; 53% vs. 47%).The basal vicariance occurred therefore between the Choco (F) (+ possibly Chagres [J]) and the Santiago (G) + E cua dor (F) around 14.17 .The second vicariance was between M. ornatus in San- The biogeographic areas used here are biogeographic "super" regions (because of limitations of the DEC analysis; map in B) and they are mapped on the phylogenetic trees from MrBayes (C) and BEAST analyses (D).A) Estimated possible ancetral areas shown with their relative probability in the brackets.Only combinations over 5% (0.05) of probability are shown.If more than five combinations were reconstructed, results are presented by the column bars, showing the percentual presence of each region in the estimates of ancestral areas (100% means the region is present in all estimated combinations).E) + F) Different results based on different assumptions -E) a fully vicariant scenario with all regions present in the ancestral area (TR+OR+CA+MG), when the maximum number of areas is not constrained (= 5 areas allowed).F) the ancestral area in trans-Andean region followed by the colonization to the other regions was supported under the constrained settings (= 2 areas allowed); A = ancestor of the whole genus Andinoacara.P = ancestor of the Andinoacara pulcher group.TR = trans-Andean region, CA = Central America, OR = Orinoco + Maracaibo, MG = Magdalena tiago (G) and M. festae in Ecuador (H) at 9.3 Mya (14.2 -4.9 Mya).Within the Northern M. atromaculatus clade the oldest node is equivocally reconstructed as a vicariance between the Choco (F) and Chagres (J) or as only including the Choco (see above and Fig. 3), and the node is dated at 3.4 Mya (5.6 -1.6 Mya).If not at 3.4 Ma then a definite vicariance between the Choco (F) and Chagres (J) occurred at 2.3 Ma (3.8 -1.1 Ma; Fig. 3) Despite the larger distribution area of Andinoacara compared to Mesoheros, both groups show congruent biogeography as well as age estimation of events.The geogra phically and phylogenetically corresponding south ern Andinoacara rivulatus group and the southern Me so heros festae group show a much more similar age (12 Mya vs. 9.3 Mya) than uncorrected p-distances (12.5% vs. 6.0%).
The molecular clock estimates in both studied genera are thus in better correspondence than the uncorrected molecular divergences, which show a two-fold difference.In the cichlasomatine Andinoacara virtually all nodes show a divergence rate of slightly more than 1% per million years (1.03 -1.42%, with outliers starting at 0.92%).In the heroine Mesoheros on the other hand all nodes show a divergence rate about half that of the cichlasomatine An di noa cara, with an average around 0.62% per million years (0.53 -0.73%).

Discussion
Phylogeny and notes on the systematics of the two trans-Andean cichlid genera The phylogenetic reconstruction of the genus Andinoacara is consistent with previous studies (kullander, 1998;Musilová et al., 2008;Musilová et. al., 2009a;Musilová et al., 2009b).In case of Mesoheros our results are also in agreement with the previous studies (Říčan et al., 2013;McMahanet al., 2015).
The genus Andinoacara contains eight valid species and three undescribed lineages (Fig. 1).The status of a recently described species, A. blombergi (wijkMark et al., 2012), is questionable based on our results.In this study, we cover all species including also samples from the Esmeraldas drainage in Ecuador from where the putative new species (A.blombergi) was described.In our analyses based on the cytochrome b gene the A. blombergi samples do not form a unique clade but instead are nested among the other phylogenetically very recent A. rivulatus populations from other parts of Ecuador (Figs 1 and 2).Andinoacara blombergi was described (wijkMark et al., 2012) solely based on shape differences (head length and interorbital width), but such characters are known as plastic in response to different ecological conditions (e.g. bouton et al., 2002).For example we have found similar variation in the sympatric M. festae, where also the lowland populations are more deep-bodied whereas the highland populations are more stream-lined (OŘ pers.obs.).
In order to further test our hypothesis of conspecifity between A. blombergi and A. rivulatus we have additionally sequenced the nuclear intron S7 (intron 1) for the four samples from the Esmeraldas (A. blombergi) as well as for the 23 samples of A. rivulatus.This nuclear gene is variable enough to distinguish between most cichlid species including all other Andinoacara species (Musilová et al., 2008(Musilová et al., , 2009b;;Říčan et al., 2008, 2013).However, we found that the Esmeraldas samples (putative A. blombergi) share the same haplotype with most of the A. rivulatus populations in this nuclear gene (data not shown).Contrarily, difference between sister species A. rivulatus and A. stalsbergi in this marker amounts to a divergence of 1.6% (Musilová et al., 2009b).Based on our mitochondrial results and this nuclear confirmation, we suggest considering A. blombergi as a species of questionable status, which should be tested in future by more complex genomic approach based on a substantial sampling of both species.
In both cichlid genera there is also substantial genetic divergence in their northern groups that suggests the potential existence of more species in NW South America than is currently assumed.In Andinoacara the only undisputed species in the northern group is Central American A. coeruleopunctatus, whose species status is well supported by our results.The remaining northern Andinoacara species have been variously classified as A. pulcher, A. latifrons, or both, however lacking clear differential diagnosis.Our analyses suggest that there are possibly four rather than two species involved.One lineage (A.sp."Choco") is clear-cut since it is a unique clade in mtDNA endemic to the Choco and allopatric from the rest of the clade, or possibly the sister group of the Central American A. coeruleopunctatus (Fig. 2).Three additional lineages appear in the region, with mostly allopatric distribution.One lineage is A. latifrons from the Magdalena, Sinu and Cauca, a well-supported, unique and also allopatric clade in mtDNA.The rest are populations traditionally most often classified as A. pul cher, which are however not monophyletic, but still share the same distribution in the cis-Andean Orinoco drainage plus Maracaibo and also are found in the (upper!)Magdalena.Whether A. pulcher is one or more species remains to be studied with additional (genomic) markers as well as with larger sampling effort covering more populations.

Rates of evolution, molecular clock calibration and the age of the cichlids
Since there are no fossils known for the studied genera, we made use of a fossil-based calibration of the whole Neotropical Cichlidae, from which we took secondary calibration points for the detailed molecular clock analyses within both studied groups (see Appendix and Fig S1).
Recently, several studies considering the dating of cichlids were published (e.g.McMahan et al., 2013;Říčan et al., 2013;lóPez-Fernández et al., 2013;Fried-Man et al., 2013) that have utilized fossils of Neotropical cichlids (for example the three Lumbrera formation fossils, i.e.Gymnogeophagus eocenicus Malabarba et al., 2010, Plesioheros chauliodus Perez et al., 2010and Pro terocara argentina Malabarba et al., 2006; see Fig. S1) and/or other non-cichlid fossils.The age estimates resulting out of the aforementioned studies are in strong dis agree ment, some of them assuming unlikely interpretations, e.g. in one extreme a huge gap in the fossil record (lóPez-Fernández et al., 2013), while in the other extreme most of the known cichlid fossils pre-dating the suggested first cichlid split (FriedMan et al., 2013).See Appendix S1 for more detailed discussion about cichlid dating.Further, none of the recently or previously published molecular-clock studies provides a reliable estimate of the time frame of evolution of the cichlasomatine Andinoacara.Because of the mentioned reasons we have reconstructed our own molecular calibration tree.

Historical biogeography of the trans-Andean cichlids in the context of Andean orogeny
Vicariant events and allopatric speciation were identified as the most important drivers of diversification in both studied cichlid genera (Andinoacara and Mesoheros).The general biogeographic pattern and molecular clock dating in both genera are highly consistent and include: (1) The basal evolutionary lineages are distributed in the Colombian Pacific-slope Choco; (2) there is a homo logous basal division into corresponding southern and northern groups; (3) both genera have very similar ages of the northern and southern groups; (4) both genera have highly corresponding ages and patterns of diversification within the southern lineage (vicariant events between the Choco (F) and the Santiago river system (G), and between the Santiago (G) and rest of Ecuador (H)).
Mountain ranges are the textbook example of a barrier splitting a previously continuous distribution area and thus forcing vicariant evolution and allopatric speciation (cRisci et al., 2003).The Andes, and especially their northern part, are geologically extremely complex and yet relatively young mountains, mostly having risen to substantial elevations during the last ~ 20 My (though their uplift started ~ 80 Mya).The Andes represent the most important distributional barrier in the Neotropics and they have caused diversification in most terrestrial groups of animals (reviewed in antonelli et al., 2010) including frogs (antonelli et al., 2010), geckos (Go na dotes;gaMble et al., 2008), birds (wrens of the genus Campylorhynchus; Piciformes; MooRe et al., 2006), spider and howler monkeys (Ateles, Alouatta;collins & dubach, 2000;cortes-ortiz et al., 2003), rodents (Dac ty lomys, Isothrix; Patterson &velazco, 2008) as well as plants (antonelli et al., 2009).Andinoacara clearly represents the best cichlid model group for the study of diversification triggered by the Andean orogeny in the whole of tropical South America, especially together with its sister group, i.e. the cis-Andean genera Bujurquina kullander 1986 and Tahuantinsuyoa kullander 1986 (Musilová et al., 2009).The distribution and diversity of these cis-Andean cichlid genera is centred mostly along the Andes from Venezuela to Argentina (in the Andean lowland piedmont up to ca 1000 m elevation) and it complements the distribution of its northern counterpart, i.e. genus Andinoacara.
Prior to the substantial elevation of the Andes the Andinoacara-Bujurquina-Tahuantinsuyoa clade (Mu silo vá et al., 2008, 2009; Fig. S1) and the herichthyine clade of the heroine cichlids, to which the Mesoheros genus belongs (Říčan et al., 2013), probably shared a large ancestral area in NW South America (around 30 Mya) which today includes both cis-and trans-Andean South America) or the large area was at least connected through two areas within the developing Andes.One likely connection was the Western Andean portal, originally a marine incursion from the west coast through the Central / Northern Andes low altitude divide (present Ecuador-Peru boundary) into present western Amazonia that persisted till 13 -11 Mya (antonelli et al., 2009;santos et al., 2009).The better known connection between present cis-and trans-Andean South America was around the northern terminus of the developing Andes (through the mouth of the northward flowing paleo-Amazon before the separation of the Maracaibo, Magdalena and Orinoco) and it terminated at around the same time as the Western Andean portal (12 -10 Ma) with the final elevation of the Eastern Venezuelan Andes (lundberg et al., 1998;rodríguez-olarte et al., 2011;schaeFFer, 2011).The ancestral evolution of both studied cichlid genera (Figs 2 and 3) and of their respective evolutionary lineages thus occurred prior to the final closing of both of these cis-and trans-Andean connections.
The oldest (i.e.basal) nodes of both genera differ in both their age and in their ancestral area reconstruction, which is older (ca 20 vs. 14 Ma) and wider (cf.Figs 2 and  3) in Andinoacara than in Mesoheros.Both differences can be explained with the help of their sister-groups.Based on Říčan et al. (2013) the Mesoheros genus is either a non-native immigrant from Middle America into Choco or a native group whose ancestral area has (or its sister clades have) been restricted by extinction events.On the other hand, Andinoacara clearly is a native NW South American genus both based on its wide reconstructed ancestral distribution and the biogeographical analysis including its sister group, i.e. the Bujurquina-Tahuantinsuyoa clade (Musilová et al., 2008).
The basal divergence within Andinoacara (around 20 Mya) occurred within the area of the Colombian Choco (Fig. 2) while in Mesoheros (around 14 Ma) occurred south of the Choco (Fig. 3) and the basal splits are thus not attributable to the same biogeographical event.
In both studied cichlid genera the oldest clear vicariant events occurred in their southern group, possibly in agreement with the older substantial elevation of this central section of the Andes than of the Andes to the north (lundberg et al., 1998; rodriguez-olarte et al., 2011).
In both cichlid genera the events are homologous having occurred at the same time (around 12 to 9.3 Ma) and in the same area within present Ecuador between the northern Santiago river basin species (A.sapayensis and M. or na tus) and the rest of Ecuador and Peru to the south (Fig. 2, 3).
The remaining diversification within both An di noa cara and Mesoheros occurred after the isolation of trans-Andean South America from cis-Andean South America.In both genera the diversification in this period was virtually limited to the northern part of their distribution.In the south there is only the nominotypical species of Me so heros (M.festae) with a very shallow molecular di versity (Fig. 3) and only two species of Andinoacara (A. ri vu latus and A. stalsbergi; Fig. 2).The northern Andean orogeny has thus generated a higher diversity and also more complicated biogeographical patterns in both genera.
In both cichlid genera their northern groups show complex biogeographical patterns with complex implications for species-level classification (see above).At least four species of Andinoacara have evolved here as a direct consequence of the Andean orogeny with each of the Andean chains isolating an endemic species of Andinoacara and both studied genera have also colonized lowermost Central America (see below).The Western Andes isolate A. sp.Choco, the Central American A. coe ruleopunctatus and M. atromaculatus, the Western and Eastern Andes restrict the distribution of the endemic A. latifrons, and the Eastern Andes isolate the cis-Andean A. pulcher from the trans-Andean species (Figs 2 and 3).The eastern branch of the Eastern Andes (i.e.Merida Andes) in Venezuela which separates the Maracaibo basin from the Orinoco basin is however not an evident barrier in Andinoacara or (more likely) the barrier has been compromised by dispersal (native or artificial) since A. pulcher (clade II) is found in both areas without corresponding molecular divergences (Fig. 2).In many other fish groups like stingrays (Potamotrygon;loveJoy et al., 1998) or catfishes (Hypostomus;Montoya-burgos, 2003), the lake Maracaibo lineages represent distinct and permanently separated populations after a putative single vicariance event (loveJoy et al., 2010).However, in An di noacara, the Maracaibo population (also referred to as Andinoacara sp."Maracaibo") from the A. pulcher group is very likely a result of recent colonization event from the (upper) Magdalena region (based on present sampling effort) or the Orinoco basin, however the exact colonization mechanism and timing is not known and requires further study.
There is one peculiar record of Andinoacara in the Amazon basin, namely in the southern Andean region of Ecuador in the Zamora river basin (OŘ, pers. obs.;Alf Stalsberg, pers. comm.;wijkMark et al., 2012).Based on our phylogenetic analysis, the Zamora populations are nested within A. rivulatus (Fig. 1).They do not form a monophyletic lineage within A. rivulatus because their clade also includes several samples from the adjacent but trans-Andean Ecuadorian Guyaquil basin.The presence of A. rivulatus in the cis-Andean region is thus most likely result of a recent colonization (or more likely introduction similar as in A. pulcher to Maracaibo; see above) and it is likely not linked to the Andean orogeny.
Finally, both studied genera have made it into Lower Central America even though their distributions there are non-overlapping.Mesoheros is found only in eastern Panama on the narrow Caribbean slope, while Andinoacara is found only in the Pacific-slope rivers but in a larger area from eastern Panama to SE Costa Rica.In Andinoacara Central America is in virtually all biogeographical analyses (both unconstrained and constrained, except for S-DiVA maxareas = 2; Figs 2 and 4) found in the ancestral area of the genus, which suggests an old colonization prior to the final Pliocene consolidation of the Panamanian Isthmus and the coterminal Great American Interchange (GAI; Marshall et al., 1982;cody et al., 2010, berMinghaM&Martin, 1998).Several studies have recently shown that e.g.fossil mammals or Middle American cichlids have reached central and eastern Panama already in middle Miocene between 19.5 and 14 Ma (kirby & MacFadden, 2005;MacFadden, 2006;kirby et al., 2008;Říčan et al., 2013) but this dispersal was in the opposite direction, i.e. from North and Middle America to South America.Accordingly to this early dispersal there are also recent geological discoveries (Montes et al., 2015) suggesting the potential presence of purely freshwater connection between South America and Panama arc in middle Miocene.Contrary to this, most of the organismal studies (see Říčan et al., 2013 for a summary) as well as previous geological studies (e.g.iturralde-vinent & MacPhee, 1999;coates et al., 2004;kirby et al., 2008) demonstrate that such early dispersal into Lower Central America from South America was unlikely, because of the eastern Panama -Colombia sea gap (the Darien gap) between Central and South America.Our results for Andinoacara suggest that it might be an old lineage within Lower Central America.Contrarily, Mesoheros did not colonize Central America in the strict geological sense because it is found only in eastern Panama.The reconstruction of its colonization of this westernmost area is also equivocal, with the inclusion or exclusion of eastern Pacific-slope Panama at the root of the group being equally likely.In this case however both options (3.4 and 2.3 Ma; Fig. 3) fit the known evolution of the Panamian isthmus.
In both cichlid lineages there is an agreement in the vicariance between Panama and Colombia at the Serranía del Darién (2.3 Ma in Mesoheros and 4.4 Ma in Andinoacara; Figs. 2 and 3), which corresponds to the final completion and recent configuration of the Panamian isthmus and to the age of the Great American Interchange in the Pliocene (berMinghaM&Martin, 1998).
To conclude, our study provides the first evolutionary-biogeographic reconstruction of two cichlid groups in trans-Andean South America.We have shown similar evolutionary patterns in these two genera, probably shaped by geological conditions in the past, and we have identified the Colombian Choco as an important region present in the ancestral area of both genera.We have further found that one cichlid genus (Andinoacara) became possibly established in Central America prior to the closure of the Panamian Isthmus.Finally we provide an updated fossil-based molecular-clock dating for all Neotropical cichlids.

Fig. 1 .Fig. 2 .Fig. 3 .
Fig. 1.Phylogenetic relationships of trans-Andean cichlids with distributions and localities.A) Bayesian tree of the genus Andinoacara.Result of Bayesian analyses in MrBayes with 10 million generation based on cytochrome b "full data set" including samples from 7 valid Andinoacara species and three putative undescribed species.Position of the recently described questionable species A. blombergi within the A. rivulatus marked by a green star symbol.B) Bayesian phylogeny of the Mesoheros genus based on cytochrome b "full data set".MrBayes run of 10 million generations; numbers in the nodes represent the Bayesian posterior probability values for the nodes.Asterisks mark nodes with posterior probability support equal to 1.00.C+D) distribution maps of both cichlid genera with their schematic phylogenies, c) Andinoacara and d) Mesoheros.Blue symbols on the map represent the Southern clade and red symbols the Northern clade in both genera.

Fig. 4 .
Fig. 4. Comparison of ancestral area reconstruction for the genus Andinoacara based on Statistical Dispersal-Vicariance Analysis (S-DiVA) and Dispersal-Extinction-Cladogenesis model (DEC) in Lagrange.The biogeographic areas used here are biogeographic "super" regions (because of limitations of the DEC analysis; map in B) and they are mapped on the phylogenetic trees from MrBayes (C) and BEAST analyses (D).A) Estimated possible ancetral areas shown with their relative probability in the brackets.Only combinations over 5% (0.05) of probability are shown.If more than five combinations were reconstructed, results are presented by the column bars, showing the percentual presence of each region in the estimates of ancestral areas (100% means the region is present in all estimated combinations).E) +

Table 1 .
Samples used in the phylogenetic study with the collection code of the DNA voucher.For the sequences downloaded from GenBank, the accession number (GB no.) is mentioned.
The lineage A. sp."Choco" distributed in the pacific-coast basins is either basal clade of whole northern clade (MrBayes analysis) or sister group to the Central American A. coeruleopunctatus (kner 1863); (BEAST analysis).Andinoacara pulcher (gill 1858), A. latifrons (steindachner 1878) and A. sp."Orinoco" represent the crown group of the northern clade.The southern clade (A.rivulatus group) is represented by four well defined species.The basalmost is A. bise riatus (regan1913) from the pacific basins of Colombia (Atrato, San Juan), followed by A. sapayensis (regan 1903) from northern Ecuador and then the sister species A. rivulatus (günther1860) and A. stalsbergi Musilová, schindler & staeck 2009 from Ecuador and Peru, respectively (Fig. 1). A. blombergi wijkMark, kullander & salazar 2012 clusters non-monophyletically within the A. rivulatus populations.