Morphometry and DNA barcoding reveal cryptic diversity in the genus Enteromius (Cypriniformes: Cyprinidae) from the Congo basin, Africa

1 Department of Biology, Systemic Physiological and Ecotoxicological Research, University of Antwerp, Groenenborgerlaan 171, 2020 Antwerpen, Belgium. 2,4 Royal Museum for Central Africa, Section Vertebrates, Ichthyology, Leuvensesteenweg 13, 3080 Tervuren, Belgium. 2,4 Department of Biology, Laboratory of Biodiversity and Evolutionary Genomics, KU Leuven, Charles Deberiotstraat 32, 3000 Leuven, Belgium. 3 Royal Belgian Institute of Natural Sciences, OD Taxonomy and Phylogeny, Vautierstraat 29, 1000 Brussels, Belgium; Department of Biology, Evolutionary Ecology Group, University of Antwerpen, Campus Drie Eiken, building D, room D.150 Universiteitsplein 1, 2610 Antwerpen, Belgium. † Equally contributing authors


Introduction
With some 3060 species in about 375 genera already described (Froese & Pauly 2017), Cyprinidae is the largest family of freshwater fi shes in the world.Within this family, the former genus Barbus sensu lato (s.l.) Cuvier & Cloquet, 1816, was a very extensive paraphyletic aggregation of over 800 species occurring in Europe, Africa and Asia.Based on ploidy level, Barbus s.l. has been divided into three groups: diploids (2n = 48 or 50), tetraploids (2n = 100) and hexaploids (2n = 148-150) (Agnèse et al. 1990;Oellerman & Skelton 1990;Rab et al. 1995;Berrebi & Valiushok 1998).The phylogeny of some of these groups has been examined (e.g., Agnèse et al. 1990;Tsigenopoulos et al. 2010), and a number of studies suggested to split Barbus s.l.into several genera at least based on the level of ploidy (Berrebi et al. 1996).Due to their uncertain taxonomic position, the African small-sized diploid species were, until recently, referred to as 'Barbus' (between quotation marks) as proposed by Berrebi et al. (1996).A recent study explored the classifi cation of the subfamily Cyprininae that includes Barbus s.l., using mitochondrial and nuclear genes (Yang et al. 2015).The results of this study supported a close relationship within the diploid species between the Asian Puntius Hamilton, 1822 species and the African small 'Barbus'.As such, they confi rmed earlier results based on karyology (Golbutsov & Krysanov 1993;Rab et al. 1995).Yang et al. (2015) proposed the revalidation of the genus Enteromius Cope, 1867 to accommodate all African diploid 'Barbus' species, as it is the oldest available genus group name for these fi shes.Recently, this revalidation has been criticized (Schmidt & Bart Jr. 2015); we prefer, however, the use of Enteromius over 'Barbus'.
The taxonomy of species of Enteromius is to date insuffi ciently known, resulting in diffi culties in identifi cations and incomplete inventories of the species.It also hampers further studies on phylogenetic relationships, which are still obscure among species of Enteromius (Berrebi et al. 2014;Yang et al. 2015;Ren & Mayden 2016).While these problems have been tackled in various Asian taxa (Berrebi & Tsigenopoulos 2003;Pethiyagoda et al. 2012), the need for such revisions remains for the African species.
This study focuses on some species of Enteromius from the Congo basin.The Congo basin comprises the second largest catchment area in the world after the Amazon (Snoeks et al. 2011).It harbours a very high diversity of fi shes with at least 957 valid species currently listed, excluding lakes Kivu and Tanganyika as well as the Malagarazi system (Froese & Pauly 2017;Musschoot, RMCA, pers. comm.).The Congo basin is generally divided into three main sections: the Upper Congo (or Lualaba), which runs from the basin's source in Zambia down to the Wagenia Falls near the city of Kisangani in north-eastern DR Congo; the Middle Congo, which starts at the Wagenia Falls, and fl ows in a large arc, fi rst northwestwards, then southwards, before widening into Pool Malebo; and the Lower Congo, which starts at the outlet of Pool Malebo and runs down to the mouth of the basin in the Atlantic Ocean (Runge 2007).Despite the fact that the Congo basin is recognized as a hotspot for fi sh diversity (Snoeks et al. 2011), the ichthyofauna in large parts of the basin remains poorly studied.To date, 90 morphologically often very similar species of Enteromius are recognized from the Congo basin (Vreven, RMCA, pers. comm.).Identifi cation keys for the region are lacking and the designation of so-called species complexes is the result of the identifi cation diffi culties.One of the specifi c taxonomic problems within the Congolese species of Enteromius concerns the E. miolepis/eutaenia complex, which comprises species characterized by a strongly ossifi ed fi rst dorsal ray that is serrated along its posterior margin and a blackish mid-lateral band that extends from the tip of the snout to the caudal fi n base or onto the mid-central part of the caudal fi n (Tweddle & Skelton 2008;Banyankimbona et al. 2012).Due to the similarity of the species within this complex (Poll 1976), the available information on morphological characteristics is often insuffi cient to arrive at a correct species identifi cation.In addition, E. miolepis (Boulenger, 1902) and E. eutaenia (Boulenger, 1904) are both characterized by high intraspecifi c morphological variation, which suggests that these taxa could be polyspecifi c.Some Congolese specimens were identifi ed as E. holotaenia (Boulenger, 1904) or E. kerstenii (Peters, 1868), although both species were described from outside the Congo basin, from the Ogowe River in Gabon and from a coastal river in Zanzibar respectively.With their ossifi ed, serrated dorsal spine and blackish mid-lateral band, they can be considered members of the E. miolepis/eutaenia complex.Enteromius holotaenia can only be distinguished from other members of the E. miolepis/eutaenia complex by the presence of a black dorsal fi n tip, which is not always clearly visible, and E. kerstenii by the presence of a red opercular spot and the absence of large scales on the dorsal fi n base.However, in the Congolese specimens identifi ed as E. kerstenii, some enlarged scales seem to be present at the dorsal fi n base, although not very pronounced.
In order to unravel the taxonomy of species of Enteromius from the Congo basin and delineate the various species, a morphologic approach alone may be inadequate.The present study ascertains whether a multidisciplinary approach might facilitate the differentiation of taxa within the genus Enteromius from the Congo basin.For this, some species of the north-eastern part of the basin, a particularly poorly explored region, were selected as a case study.Additionally, because of the taxonomic problems in the E. miolepis/eutaenia complex discussed above, we have examined specimens from this complex from other regions within the Congo basin as well.We explored whether this multidisciplinary approach would result in the discovery of cryptic species, and the detection of (additional) morphological characteristics to discriminate between these species.
To attain these objectives, we used DNA barcoding (Hebert et al. 2003) together with traditional morphometrical techniques.Molecular techniques like DNA barcoding have already proven to be powerful for the identifi cation of both marine and freshwater fi sh taxa (Becker et al. 2011;Pereira et al. 2013), and to detect cryptic species diversity or taxonomic inconsistencies (Goodier et al. 2011;Collins & Cruickshank 2012;Decru et al. 2016).However, the uncritical use of molecular tools to identify species, especially using only a single molecular locus (mt genome), is unwarranted (DeSalle et al. 2011), and traditional morphology has proven its value as a suitable technique to assess the diversity in many taxa.Therefore, we combined both methodologies in the present study, an approach that has already proven to be successful in several taxonomic studies (e.g., Olayemi et al. 2012;Stiassny et al. 2013;Lavoué & Sullivan 2014).

Material & methods
Specimen selection (Fig. 2B, Table 1) In total, we used 181 specimens of Enteromius from the Congo Basin for morphological and/or genetic analyses.To investigate the taxonomic diversity of Enteromius in the north-eastern part of the Congo basin (Middle Congo), we examined specimens of all species of Enteromius for which tissue samples had been recently collected from rivers in this region, i.e., the Ituri, its tributary the Epulu, the Lobilo, and the Lomami/Lobaye system.To unravel the E. miolepis/eutaenia complex, we also examined specimens from various sampling sites in the Upper, Middle and Lower Congo, identifi ed as E. miolepis, E. eutaenia, E. holotaenia or E. kerstenii.For the Middle Congo, we included specimens from the Léfi ni, the Itimbiri, the Ituri/Epulu, the Lobilo, and the Lomami/Lobaye.For the Upper Congo we included samples from the Loboya (a tributary of the Maiko), and the Luapula.Finally, for the Lower Congo we used samples from the rivers Luki and Inkisi.These specimens were collected during ten expeditions that were carried out between 2005 and 2012.
Relevant types were selected based on the literature-based identifi cations and include the holotype of E. miolepis described from the Yembe River at Banzyville (Ubangi river system); 15 syntypes of E. holotaenia described from the Ogowe River in Gaboon; one lectotype and fi ve paralectotypes of E. eutaenia described from Huila (Mossamedes) in Angola; the lectotype of E. kerstenii described from the coast opposite Zanzibar; the holotype of E. brazzai (Pellegrin, 1901) described from the Sangha River in Mobaka; the holotype and fi ve paratypes of E. tshopoensis (De Vos, 1991), a junior synonym of E. brazzai, which was described from the Tshopo River; the lectotype and three paralectotypes of E. pellegrini (Poll, 1939) described from Lake Kivu; and two paratypes of E. atromaculatus (Nichols & Griscom, 1917) described from the Yakuluku River (Lévêque & Daget 1984) (see also Appendix 1).
We attempted to use the same specimens for the morphological and genetic analyses, but this was not always possible.As some specimens were lost or too damaged, 40 specimens that were genetically examined could not be used for morphological analyses.In 19 of these cases, literature-based identifi cations could not even be done, and the specimens were identifi ed as Enteromius sp.As such, we morphologically examined 177 specimens (Table 1, Appendix 1), including several type specimens.Similarly, not all specimens that we examined morphologically could be successfully DNA barcoded, and genetic samples are unavailable for the type specimens.

Specimen identifi cations
Based on published characteristics, we assigned the non-type specimens to four 'a priori' groups: 60 to E. cf.miolepis (which represents the E. miolepis/eutaenia complex); 15 to E. cf.brazzai; 26 to E. cf.pellegrini; and 40 to E. cf.atromaculatus.We used 'cf.' to indicate the uncertainty connected to their identifi cations.The four groups mainly differ in the morphology of their dorsal spine (ossifi ed or fl exible; serrated or not) and colour pattern (Fig. 1).Enteromius cf.miolepis has an ossifi ed serrated dorsal spine and a mid-lateral band on the fl ank and through the eye; E. cf.brazzai has a fl exible unserrated dorsal spine and no melanin markings; E. cf.pellegrini has an ossifi ed serrated dorsal spine and large dark spots that can fuse into a mid-lateral band; and E. cf.atromaculatus has a fl exible unserrated dorsal spine and small spots that can fuse into a mid-lateral band.

Morphometry
For the 177 examined specimens, we obtained 17 measurements with a vernier caliper (Helios, 0.05 mm), as well as 10 meristics.Counts and measurements were based on Bamba et al. (2011), with some small modifi cations: the pre-occipital distance, post-dorsal distance I, post-anal distance I, and body-depth I were not included; the standard length, post-dorsal distance II, and post-anal distance II were measured up to the insertion of the fi rst caudal fi n ray of the upper lobe instead of to the middle of the caudal peduncle base.Barbel lengths were also coded according to Bamba et al. (2011).Head measurements were expressed as percentages of head length (HL) and body measurements as percentages of standard length (SL).

Data analyses
We used principal component analysis to explore the multivariate data matrix and to reduce the large number of variables into a few meaningful axes (Snoeks 2004;Decru et al. 2012).PCAs were executed in Past 3.15c (Hammer et al. 2001).Raw data were used for meristics, log-transformations for the measurements.By using log-transformations, the fi rst component can be interpreted as a proxy for size, since it is characterized by loadings with the same sign and same order of magnitude for all variables (Bookstein et al. 1985).Because the barbel lengths were coded (Bamba et al. 2011) as categorical variables, they were not included in the PCAs.PC loadings of the illustrated PCAs are given in the Appendices 2-8.

DNA Extraction, PCR amplifi cation and sequencing
We successfully barcoded 114 specimens using DNA extracted from fi nclips (Table 1).The method used is based on the FISH-BOL (Fish Barcode of Life) protocol (Steinke & Hanner 2011).DNA was extracted using a 'NucleoSpin® Tissue Kit' following the instructions of the manufacturer (Macherey-Nagel, Germany).Subsequently, polymerase chain reactions (PCR) were used to amplify the mitochondrial cytochrome c oxidase I (COI) gene.A part of the samples was amplifi ed with 'Fish Cocktail', an M13 tailed primer combination (Ivanova et al. 2007) (Table 2).A standard 25 μL PCR mix consisted of 2.5 μL PCR buffer (10x); 2.5 μL dNTP (2 mM); 1.25 μL 'Fish Cocktail' (2 μM); 0.2 Taq DNA Polymerase (5 units per μL); 16.75 μL mQ-H2O and 2.0 μL of the extracted DNA sample.The PCR profi le was 3 min at 94°C, followed by 35 cycles of 40 s at 94°C, 40 s at 52°C, and 1 min at 72°C, plus a fi nal extension of 10 min at 72°C.The success rate of this PCR-mix was relatively low, which led to the development of specifi c primers, Bbus F and Bbus R, for the amplifi cation of samples that weren't successfully sequenced with 'Fish Cocktail' (Table 2).In this PCR mix, instead of 'Fish Cocktail', 2.5 μL of each specifi c primer was used, and an annealing temperature of 54°C instead of 52°C.Amplifi ed products were verifi ed on 1.2% agarose gels.Afterwards the PCR products were purifi ed using a Nucleofast® 96 PCR kit (Macherey-Nagel, Germany), and sequenced bidirectionally using BigDye Terminator v.3.1 and an ABI 3130 capillary sequencer.

DNA analyses
The DNA sequences were assembled and visually checked in CodonCode Aligner 5.1.4(CodonCode Corporation) and aligned in MEGA 5.2 using Muscle Alignment (Edgar 2004;Tamura et al. 2011).The longer M13-sequences were shortened to the length of the Bbus-sequences (558 bp).Next, MEGA 5.2 was used to execute a model test and to create a phylogenetic tree.The model test indicated the GTR+G+I as the optimal model using the AIC criterion.This model was used to construct a Maximum-Likelihood (ML) tree with 1000 bootstrap replications.As both the ML tree and NJ (Neighbour Joining) tree gave similar results, statistical node support of both methods are visualised on the ML tree.

Evaluation of the literature-based identifi cations
In the ML tree obtained (Fig. 2), lineages with less than 2% sequence divergence were collapsed and named after the river system from which the specimens were collected.Specimens from the Lomami/ Lobaye system and the Lobilo were grouped under the label 'Kisangani region', since there appeared to be little or no genetic difference between samples of these nearby affl uents.
The ML tree shows 23 lineages within the assayed Enteromius samples, representing the four 'a priori' morphospecies in the following quantities/properties: 13 of which belong to the E.  group.Enteromius cf.miolepis, E. cf.atromaculatus and E. cf.pellegrini each form clearly supported clades, while this is not the case for E. cf.brazzai.There is often a substantial genetic distance between lineages occurring in different rivers, but also between some lineages detected in the same river.For example, within the E. cf.miolepis group, we observed a considerable difference between specimens of the Luapula (three lineages; 14.6% sequence divergence between Luapula 1 and 2; 15.3% between Luapula 1 and 3; 19.0% between Luapula 2 and 3), and the Luki (two lineages; 7.7% sequence divergence).Interestingly, the sequences obtained for the Upper, Middle and Lower Congo populations did not group according to these zoogeographic regions.However, two out of the three lineages of the Luapula, a river system geographically remote from the other rivers included in this study, also show the largest genetic distances within the E. cf.miolepis clade.Enteromius cf.brazzai is not resolved as a single clade, and consists of three lineages, two of which were detected in the 'Kisangani region' (19.8% sequence divergence).In contrast to the other groups, E. cf.pellegrini consists of a single lineage that is composed of specimens from a single river, the Ituri.The E. cf.atromaculatus group contains an important amount of genetic variation, with fi ve of its six lineages occurring in the Ituri River.Remarkably, one of those lineages also contains one sample from the 'Kisangani region'.

Morphometric groupings
Firstly, PCAs were executed on all the specimens examined, after which more detailed PCAs were performed on each literature-based 'a priori' group separately.

Overall analyses
The highest loadings on PC2 for a PCA on 17 log-transformed measurements (n = 177) are for the eye diameter (ED), the post-anal distance (PoAD) and the dorsal fi n length (DoFL); PC1 is a proxy for size (see above).On a scatterplot of PC2 against PC1, the four 'a priori' groups cannot all be distinguished from each other (Fig. 3).However, specimens of E. cf.pellegrini and E. cf.brazzai are completely separated from each other on PC2.This is mainly because E. cf.pellegrini has a smaller ED, a smaller dorsal fi n base length (DoFBL) and a smaller PoAD than E. cf.brazzai.Clearly E. cf.miolepis occupies the largest morphospace on PC2, which comprises most of the type specimens included, except for two paralectotypes of E. eutaenia and one syntype of E. holotaenia.The E. cf.atromaculatus polygon comprises the paratypes of E. atromaculatus.The specimens of E. cf.pellegrini, however, only overlap slightly with the type specimens of E. pellegrini and the specimens of E. cf.brazzai are separated from the holotype of E. brazzai and types of E. tshopoensis.Furthermore, the type specimens of E. tshopoensis are separated from the other groups, mainly on PC2, due to a smaller PoAD and a larger DoFL.
The highest loadings on PC1 for a PCA on 10 meristics (n = 177) are for the number of scales between the lateral line and the belly (L-B Sc), the number of scales between the dorsal fi n and the lateral line (D-L Sc) and the number of scales between the lateral line and the pelvic fi n (L-P Sc); on PC2 for the number of scales on the lateral line (LL Sc), the number of scales between the occiput and the base of the fi rst dorsal fi n ray (PD Sc) and the number of pelvic fi n rays (PelFR).Similar to the analysis for the measurements, a scatterplot of PC2 against PC1 does not allow the separation of the four 'a priori' groups (Fig. 4).Again, E. cf.miolepis is the group showing the largest variation.Also, E. cf.pellegrini and E. cf.brazzai are again completely separated on PC1, mainly due to E. cf.pellegrini having a higher D-L Sc (4.5-5.5 vs 3.5) and a higher L-B Sc (5-6 vs 4-5).Furthermore, the four 'a priori' groups all overlap with their respective type specimens.
To examine whether morphological differences could be detected between the different genetic lineages within each group, we performed PCAs on E. cf.miolepis, E. cf.brazzai and E. cf.atromaculatus separately.To investigate whether one or more lineages represent a currently valid species, relevant type specimens were also included in these analyses.This resulted in a multitude of analyses for which the most important outcomes are presented below.

Morphometric comparisons among E. cf. miolepis lineages
Because of the high number of lineages and specimens for E. cf.miolepis, we analysed the major geographical regions, the Upper, the Middle and the Lower Congo, separately (see also Table 1), but still included the relevant type specimens to check whether some groups could be allocated to these species.Only the results of the latter are discussed as an example.
For the Lower Congo, we detected three genetic lineages within E. cf.miolepis (Fig. 2), i.e., one group containing specimens from the Inkisi and two groups with specimens from the Luki (Luki 1 and Luki 2).The highest loadings on PC1 for a PCA on 10 meristics (n = 36) are for LL Sc, CP Sc and PecFR; on PC2 again for PecFR, AFR and PD Sc.The specimens from the Inkisi are well separated from the Luki lineages on both PC1 and PC2; and there is only one specimen overlap between Luki 1 and Luki 2 (Fig. 5).In addition, specimens from these latter two lineages noticeably differ in the length of their barbels.Lineages Inkisi and Luki 1 can be clearly distinguished from all type specimens (Fig. 5).Luki 2 overlaps with the type specimens of E. eutaenia and E. holotaenia, but Luki 2 differs from the E. eutaenia types in barbel length.A PCA of the log-transformed measurements did not separate the genetic lineages from each other or from the type specimens (not illustrated).For the Middle Congo and Upper Congo, we detected respectively six and four genetic lineages of E. cf.miolepis (Fig. 2).Similar exploratory morphometric analyses of these genetic lineages, as for the Lower Congo discussed above, indicated that almost all could be separated from each other as well as from the included type specimens.However, specimens from 'Kisangani region' 1, Itimbiri and the syntypes of E. holotaenia clustered on the PCAs.Yet, the Itimbiri specimens had a different colour pattern (a black anal fi n tip), which is absent in the 'Kisangani region' 1 specimens as well as in the syntypes of E. holotaenia.
When comparing all lineages of E. cf.miolepis (Lower, Middle and Upper Congo), each could be morphologically distinguished from the others based on meristics (Fig. 6), measurements and/or barbel length (not illustrated).Only specimens from Kisangani region' 1 and Itimbiri could not be separated from each other, but differed in colour pattern (see above).

Morphometric comparisons among E. cf. brazzai lineages
We observed three genetic lineages within E. cf.brazzai (Fig. 2).The highest loadings on PC1 for a PCA on 10 meristics (n = 22) are for LL Sc, PecFR and PelFR; on PC2 for CP Sc, L-B Sc and again PecFR.Specimens from the 'Kisangani region' 2 are well separated from the Ituri 3 lineage on PC2 and both genetic lineages differ from the 'Kisangani region' 3 lineage and type specimens on PC1 (Fig. 7).The polygon of specimens from the 'Kisangani region' 3 lineage comprises the holotype of E. brazzai and overlaps with the types of E. tshopoensis.However, the eight specimens from the 'Kisangani region' 3 lineage as well as the holotype of E. brazzai differ from the types of E. tshopoensis by the absence of barbels.

Morphometric comparisons among E. cf. atromaculatus lineages
We detected six genetic lineages within E. cf.atromaculatus (Fig. 2).Although there was only 1.75% sequence divergence between the lineages of Epulu 2 and Ituri 8, we interpreted these lineages as separate Operational Taxonomic Units (OTUs, i.e., clusters of similar DNA sequences) because of the observed differences in colour pattern (specimens from the Epulu 2 lineage have mid-lateral dots, while specimens from Ituri 8 display a vague mid-lateral band).The single specimen from Ituri 7 was lost and could not be measured.The highest loadings on PC1 for a PCA of 10 meristics (n = 42) are for LL Sc, PecFR and D-L Sc; on PC2 for CP Sc, again PecFR and D-L Sc.A plot of PC2 vs PC1, separates the two specimens from the Ituri 6 lineage from all other lineages based on PC2 (Fig. 8).The specimens from the Ituri/'Kisangani region' lineage are separated from all other lineages mainly on PC1.The Ituri 5 lineage is separated from the Ituri 6 and Epulu 2 lineages along PC2, from the E. atromaculatus type specimens on PC1, and from the Ituri/'Kisangani region' and Ituri 8 lineages on a combination of PC1 and PC2.
As the initial PCA resulted in a great overlap between the Epulu 2 and Ituri 8 lineages, we carried out a second PCA on 10 meristics, only including specimens from these lineages and the types of E. atromaculatus.The highest loadings on PC1 are for LL Sc, D-L Sc and CP Sc; on PC2 for PecFR, PD Sc and the number of pelvic fi n rays (PelFR).On a plot of PC2 versus PC1, specimens from Epulu 2 and Ituri 8 still overlap and the types of E. atromaculatus overlap with specimens from Epulu 2 (Fig. 9).These groups also overlap on a PCA of 17 log-transformed measurements (not illustrated).

Discussion
Multidisciplinary approach to detect cryptic species Specimens from the genus Enteromius are notoriously diffi cult to identify.As, apart from some regional reviews (e.g., Bamba 2012), no extensive continent-wide reviews have been carried out for this genus, the available literature for species identifi cations is often limited to the accounts of the original species descriptions.Recent molecular studies already revealed Enteromius to be non-monophyletic (Yang et al. 2015;Ren & Mayden 2016).
Our study examined whether a multidisciplinary approach could provide more insight into the taxonomy of some species of Enteromius from the Congo basin and whether it might reveal cryptic diversity.For the examined species in this study, a literature-based identifi cation led to the delineation of four so-called 'a priori' species.The reliability of these 'a priori' identifi cations was tested using a combined approach that consisted of DNA barcoding and a detailed morphometric approach that allowed the comparison of the specimens from our collection with relevant type specimens.
This approach revealed a high number of potentially new species of Enteromius within the examined samples/specimens.DNA barcoding uncovered the existence of 23 genetic lineages within the four 'a priori' species obtained by literature-based identifi cations.The genetic distances between many of the lineages, even within the 'a priori' species, were substantial.Sometimes they even reached almost 20%, which is considerably larger compared to most other African freshwater fi sh taxa, that usually have interspecifi c divergences lower than or around 10% for the COI gene (see, e.g., Lowenstein et al. 2011;Decru et al. 2016).Only three of the four 'a priori' species agreed with supported clades in our ML tree, i.e., E. cf.miolepis, E. cf.pellegrini and E. cf.atromaculatus, while the lineages of E. cf.brazzai do not form a supported clade.Subsequent exploratory morphometric analyses revealed that, despite the small sample sizes, most lineages could be separated on the basis of their morphology.The fact that specimens from separate genetic lineages are morphologically distinguishable, suggests that they may represent different species.Therefore, our hypothesis that at least the E. miolepis/eutaenia complex represents a polyspecifi c complex appears to be confi rmed by our results, as within E. cf.miolepis we detected no less than 13 putative species.Also in the ML tree of Ren & Mayden (2016), samples from the E. miolepis/ eutaenia complex formed fi ve distinct lineages, three identifi ed as E. kerstenii, one as E. miolepis, and one as E. holotaenia.Furthermore, respectively three and six putative species were detected in the 'a priori' species E. cf.brazzai and E. cf.atromaculatus.Interestingly, one of the important differences between the E. cf.brazzai lineages was the number of branched pelvic fi n rays (7 vs 8).As far as we are aware, variation in the number of pelvic fi n rays is uncommon in Enteromius.Enteromius cf.pellegrini is the only 'a priori' species that appears to be monospecifi c within the study area.
Providing formal descriptions with diagnoses of the undescribed species of Enteromius lies outside the scope of the present study, as it requires a detailed morphological examination of more specimens of each OTU and from other areas to unequivocally delineate species-specifi c diagnostic characteristics.In some analyses, type specimens ended up within or very close to the polygon of a particular OTU: the types of E. holotaenia partially overlapped with two OTUs of E. cf.miolepis, i.e., Luki 2 and 'Kisangani region' 1; the holotype of E. brazzai fell within an OTU of E. cf.brazzai from the Kisangani region ('Kisangani region' 3); and the types of E. atromaculatus overlapped with an OTU of E. cf.atromaculatus from the Epulu (Epulu 2).However, this does not necessarily imply that the OTUs are conspecifi c with the respective types.Our results indicated that in almost every studied river stretch at least one distinct species of Enteromius occurs, which makes it very unlikely that one of our studied OTUs from the Congo basin indeed represents E. holotaenia, a species originally described from the Ogowe River in the Lower Guinea ichthyofaunal province.Most of the specimens examined lacked the black dorsal tip present in E. holotaenia.However, it was present in most of the specimens from Luki 2, in two of the nine specimens examined from 'Kisangani region 1', in both specimens from Itimbiri 1 and in one of the fi ve specimens from the Léfi ni.This demonstrates that this feature is not consistently present in the lineages.Hence, it does not seem a good criterion for species to distinguish these taxa, also because this characteristic is paraphyletic in the ML tree (Fig. 2).
For E. atromaculatus and E. brazzai the situation appears to be different.Enteromius atromaculatus is described from the Yakuluku River (Uele River), which is in the vicinity of the Epulu River; and Enteromius brazzai is described from the Sangha River, which is also part of the Middle Congo.It is therefore plausible that the Epulu 2 population is conspecifi c with E. atromaculatus, and that 'Kisangani region' 3 is conspecifi c with E. brazzai.This implies that at least 21 of our 23 OTUs may represent new species that will need to be formally described in the future.

Distribution patterns and speciation processes
All species to which the specimens were originally assigned are presumed to have large distribution ranges that, except for E. atromaculatus, extend beyond the Congo basin.These wide distribution ranges of the four 'a priori' species on the one hand, and the often much narrower distribution ranges of the putative new species detected through DNA barcoding on the other, may be a reason why up to now, these OTUs were not detected as different species.Indeed, as the majority of earlier studies on ichthyofaunal diversity concern collections from small areas and even individual rivers, the studied specimens are often identifi ed as one of the morphologically similar species presumed to occur in the studied region.Unfortunately, these identifi cations rarely imply detailed comparisons with specimens from other regions, let alone type specimens.In our study, most of the sympatric lineages were morphologically clearly distinct, and were as such already assigned to different 'a priori' species.In addition, the allopatric lineages within each of these 'a priori' species were morphologically suffi ciently similar to remain grouped for as long as they were not subjected to comparisons based on multivariate morphometric methods.Although minor morphological differences can be interpreted as intraspecifi c geographic variation among different populations, the detected morphological and genetic differences among the examined specimens of Enteromius, suggests that these populations probably represent separate, undescribed species.Although DNA barcodes allowed for the clustering of putative conspecifi cs, the obtained tree does not necessarily refl ect the phylogenetic relationships within the genus Enteromius.A review of the phylogeny would not only require a comprehensive sampling of all species of Enteromius (some 200 species), but also the inclusion of outgroups.Nevertheless, some scenarios can be proposed concerning their evolutionary history.As in each of the geographically isolated river stretches studied distinct putative species appear to occur, allopatric speciation may well have been the main mode of speciation in this genus.However, the phylogenetic relationships among the species remain largely unresolved, as there is a lack of resolution in several parts of the ML tree.While it is possible that the lack of resolution is due to incomplete taxon sampling and/or to the small number of characters, we suggest a plausible cause to be rapid radiation, which often results in diffi culties in resolving phylogenetic patterns (Fernández & Vrba 2005;Koblmüller et al. 2010).The unresolved phylogenies would in that case be the result of an almost concurrent differentiation between populations from different river stretches on a short evolutionary timescale.The fact that in this scenario adaptation to different niches has not been necessary, could thus be the cause that only minor morphological differences exist, which were only detected after multivariate exploratory techniques.
For E. cf.atromaculatus and E. cf.pellegrini, the populations from the Ituri River and its most important right-bank affl uent, the Epulu River, have a slightly different colour pattern, which seems to support the scenario of allopatric speciation.However, for E. cf.atromaculatus, there was only a genetic divergence of 1.75% between these populations (Epulu 2 and Ituri 8 on Fig. 2A), and even no divergence for E. cf.pellegrini (Ituri/Epulu on Fig. 2A).Also based on counts and measurements, both populations of E. cf.atromaculatus (e.g., Figs 8-9) and E. cf.pellegrini (not illustrated) overlap.These two instances could be examples of intraspecifi c geographical colour variation rather than of two separate species in each of the two groups.This conclusion is unexpected, since even morphologically very similar populations were found to represent separate putative species in almost every individual river stretch.The observation that the populations from the Ituri and the Epulu cannot be distinguished based on the multivariate analyses and do not appear to be genetically distinct, is even more surprising since these two river stretches are isolated.Indeed, the discussed populations are separated from each other by the presence of two waterfalls, one on the Epulu, just upstream of its confl uence with the Ituri, and one on the Ituri itself (Vreven, RMCA, pers. comm.).These falls could form important barriers for fi sh species and populations, at least for upstream dispersal.However, this hypothesis does not appear to be supported by our results.
Interestingly, in some rivers multiple lineages of the same 'a priori' species appear to co-exist.This implies that in some instances morphologically similar putative (a posteriori) species may occur in sympatry; this is, e.g., the case for three lineages of E. cf.miolepis in the Luapula and two in the Luki River.Why morphologically very similar but separate species have evolved in the same rivers cannot be inferred from the present data set, and would require a phylogenetic analysis on multiple genetic markers and, e.g., additional information on ecology.
The distribution patterns, in combination with morphological (dis)similarities and the unresolved phylogenetic relationships, indicate that multiple allopatric speciation events on a short evolutionary timescale is a plausible mode of speciation for the examined species of Enteromius.Such allopatric divergences can occur in river systems when hydrological changes cause (simultaneous) disconnections of river stretches.Remarkably, some morphologically similar putative species occur however in sympatry.

Impact on documented species richness
By applying DNA barcoding in combination with multivariate analyses of morphometric data to several Enteromius populations from the Congo basin and the relevant type specimens, the number of species identifi ed has increased from four based on literature to putatively 23.Considering the fact that only a part of the Congo basin and only some species have been studied, our results could imply a considerable increase of the number of species within this genus when extrapolated to the entire Congo basin or even Africa as a whole.A similar approach applied to other tropical vertebrates has yielded similar results.For instance, in a study on African giant pouched rats (Cricetomys, Nesomyidae), the combined use of DNA barcoding and cranial measurements lead to the discovery of at least three new species (Olayemi et al. 2012).Also for the fi sh genus Pseudobarbus (Cyprinidae) occurring in southern Africa, 15 separate lineages were identifi ed, using cytb and 16S data, within the (at that time) seven valid species within the genus, most of which were confi rmed by morphological results (Swartz et al. 2009).Also in Neotropical frogs, a study based on 16S rDNA indicated a huge underestimation of the current species richness, as up to 115% additional species were discovered (Fouquet et al. 2007).In the latter study however, no morphological analyses were performed to support the validity of the putative new species.These different studies indicate that not only for African fi shes, but for several other vertebrate taxa in different regions on Earth, the current species richness could be severely underestimated.However, every case has its specifi c patterns, which indicates that a taxon-specifi c approach is needed for species detection and delineation.Systematically using a multidisciplinary approach could therefore result in an enormous increase of the overall documented species richness.
Correct insight in the number of species within a certain taxon is urgently needed to address conservation issues.When the number of species is highly underestimated, the decline of species may also be far worse than initially detected (Fouquet et al. 2007).Therefore, using molecular techniques can be of key importance for traditional taxonomy in accelerating the pace of species detection and description.
Abbreviations of the measurements and counts: AFR = Anal fi n rays BD = Body depth CP Sc = Caudal peduncle scales DFR = Dorsal fi n rays D-L Sc = Scales dorsal fi n-lateral line

Fig. 2 .
Fig. 2. A. ML tree based on 558-bp-long Enteromius COI sequences with 1000 bootstrap replications, with node support shown as NJ/ML bootstrap (bootstrap values > 95% are shown; lineages < 2% sequence divergence were collapsed), the label 'Kisangani region' contains samples from the Lomami/ Lobaye system and the Lobilo.B. Map of the Congo basin with the sampled river stretches indicated according to the phylogenetic lineages.

Table 1 .
Number of examined non-type specimens of Enteromius Cope, 1867 per river (system) studied.A. Specimens other than those belonging to the E. miolepis/eutaniea complex.B. Specimens belonging to the E. miolepis/eutaenia complex.Numbers represent morphometrically analysed specimens vs genetic samples, then the number of specimens analysed morphologically as well as genetically.

Table 2 .
Sequences 5'-3' of the primers used for the PCR reactions.
cf. miolepis group; three to the E. cf.brazzai group; one to the E. cf.pellegrini group; and six to the E. cf.atromaculatus