An integrative framework to reevaluate the Neotropical catfish genus Guyanancistrus (Siluriformes: Loricariidae) with particular emphasis on the Guyanancistrus brevispinis complex

Characterizing and naming species becomes more and more challenging due to the increasing difficulty of accurately delineating specific bounderies. In this context, integrative taxonomy aims to delimit taxonomic units by leveraging the complementarity of multiple data sources (geography, morphology, genetics, etc.). However, while the theoretical framework of integrative taxonomy has been explicitly stated, methods for the simultaneous analysis of multiple data sets are poorly developed and in many cases different information sources are still explored successively. Multi-table methods developed in the field of community ecology provide such an intregrative framework. In particular, multiple co-inertia analysis is flexible enough to allow the integration of morphological, distributional, and genetic data in the same analysis. We have applied this powerfull approach to delimit species boundaries in a group of poorly differentiated catfishes belonging to the genus Guyanancistrus from the Guianas region of northeastern South America. Because the species G. brevispinis has been claimed to be a species complex consisting of five species, particular attention was paid to taxon. Separate analyses indicated the presence of eight distinct species of Guyanancistrus, including five new species and one new genus. However, none of the preliminary analyses revealed different lineages within G. brevispinis, and the multi-table analysis revealed three intraspecific lineages. After taxonomic clarifications and description of the new genus, species and subspecies, a reappraisal of the biogeography of Guyanancistrus members was performed. This analysis revealed three distinct dispersals from the Upper reaches of Amazonian tributaries toward coastal rivers of the Eastern Guianas Ecoregion. The central role played by the Maroni River, as gateway from the Amazon basin, was confirmed. The Maroni River was also found to be a center of speciation for Guyanancistrus (with three species and two subspecies), as well as a source of dispersal of G. brevispinis toward the other main basins of the Eastern Guianas.

Introduction Species identification, characterization, and naming remain fundamental and critical steps in biological science. Since the establishment of the Linnean system [1], species have been described mainly on the basis of morphological and phenotypic characteristics. Through time, however, morphology alone has been shown to be limited in its ability to delineate species boundaries, and led to a proliferation of names and nomenclatural instability [2]. In addition, cryptic diversity (reviewed in [3]) remained hidden from traditional morphological approaches (see e.g. [4,5]). Modern technological developments, including DNA sequencing, provide new tools allowing the detection of hidden diversity. In particular, the DNA barcoding approach [6], quickly appeared to be an efficient methodology for detecting cryptic biodiversity (e.g. [7][8][9][10][11][12][13][14][15]), even though in certain cases, such as recent divergence [16,17] or mitochondrial introgression [18], barcoding may fail to discriminate between species (e.g. [19,20]). Morphological data (used in systematics) and molecular data such as DNA barcodes (used in biodiversity studies) are not mutually exlusive, and often are complementary means of delineating species. Indeed, combining multiple data sources is the most efficient way to support robust species hypotheses [21][22][23]. Formalized under the designation "integrative taxonomy" [2,24] (reviewed in [25,26]), this approach tries to use the complementarity of the different fields of study (e.g. morphology, genetics, biogeography, ecology, ethology, etc.) to delineate, describe and name species. Various protocols have been proposed to integrate these different data sources [2,[25][26][27], but most of them correspond to guidelines that explore the different datasets successively to corroborate taxonomic hypotheses, or only focus on a given type of data (e.g. [28]). The way that results of the different analyses are interpreted, i.e. in a cumulative or a congruent way [25], also has an impact on the results, leading to an over-estimation of the number of species and lower confidence in species identity in the former case, and to an underestimation of the number of species and higher confidendence in species identity in the latter. Moreover, comparing results of different analyses, which can be based on qualitatively different data (e.g. linear measurements for morphometric analyses, sequence alignments for phylogenetic trees or distances matrices, GPS coordinates for distributional data, etc.), in the same descriptive framework remains a challenge.
Community ecologists, confronted by the same issue of combined analysis of various data types, developed multi-table methods (e.g. [29][30][31][32][33][34]). Based on the co-inertia criterion [29], multi-table analyses look for common structures present in different data sets, and include them in a common analysis. The link between all tables is defined by row, since all different observations (e.g. the abundances, the distributions, the life traits, etc.) rely on the same statistical units (e.g. the specimens, the stations, etc.). These analyses are particularly flexible and allow the inclusion of multiple data types, and have already been used in different fields including e.g. ecology, medical research, agronomy, evolutionary biology, and genomics [19,29,31,[35][36][37][38][39][40]. In addition, these methods allow evaluation of the statistical significance of the congruence between data types and the amount of common information present in the different tables. We consider the integrative approach of multi-table methods highly appropriate for the resolution of species delineation in a group of poorly-differentiated catfishes from the Guianese Region.
The northeastern part of the Guiana Shield, including Suriname and French Guiana, overlaps the Eastern Guianas Neotropical freshwater ecoregion [41]. This region ranges between the Demerara River in the west and the Oyapock River in the east and probably supports more than 500 described species, of which 169 are considered endemic (freshwater ecoregions of the world: http://www.feow.org/ecoregions/details/311, accessed 31th Jan. 2017), making the Eastern Guianas a region of high biodiversity importance [42,43]. The Eastern Guianas' river considered as a species complex, with the true G. brevispinis possibly restricted to the Nickerie River system (see [55]). However, additional sources of information from genetic markers were deemed necessary to confirm their taxonomic status. The five Atlantic coastal G. brevispinis lineages of the Guianas were found to form a monophyletic group that originated from an ancestral colonization event from the Amazonian Basin, hypothesized to have been through river capture between northern Amazon tributaries and the upper Maroni River Basin. In the Guianas, subsequent dispersal would mainly have resulted from temporary connections between adjacent rivers when sea levels were low, and subsequent diversification of isolated populations during periods with high sea levels [60].
Considering the recent genus revalidation and questions about the type species, and the potential existence of new and/or endangered species, the present work uses an integrative approach combining morphology, genetics and spatial data to reappraise Guyanancistrus, focusing on the enigmatic Guyanancistrus brevispinis species complex. Most known populations were included in this analysis, principally based on material collected by the authors and their collaborators in the past 15 years. After a comparative diagnosis of the genus, the type species is redefined and its morphological and genetic variation delineated. Several new species revealed by the study are also described. Detailed descriptions and morphological comparisons of Guyanancistrus niger and G. longispinis are already available [59,61,62] and will not be repeated, but a practical key to all Guyanancistrus species is provided. After this taxonomic clarification, the biogeography of all Guyanancistrus members is re-evaluated to investigate dispersal processes, putative local extinctions, and speciation events.

Ethics statement
No protected species (local restrictions, IUCN or CITES listed species) were examined in the study. Most specimens and tissue samples were obtained from museum collections and/or by local populations or fishermen. No experimentation was conducted on live specimens. For specimens and associated tissue samples obtained from the field, specimens were collected and exported with appropriate permits: Préfecture de la Région Guyane, Arrété 03/17/PN/EN to collect in the Réserve Naturelle des Nouragues in 2003; Ministry of Agriculture, Animal Husbandry and Fisheries to export fishes from Suriname in 2005, 2007, 2008, 2012, and 2014. Material obtained from the Parc Amazonien de Guyane in 2014 was collected under the direct supervision of PAG authorities. When collecting occurred in non protected areas of French Guiana, sampled specimens were declared to the French DEAL (French environmental protection ministry) before export. Immediately after collection, fish were anesthetized and sacrificed using water containing a lethal dose of eugenol (clove oil). Fin clips were taken after death and specimens fixed for long term preservation in museum collections. All work was conducted in accordance with relevant national and international guidelines, and conforms to the legal requirements (Directive 2010/63/EU of the European Parliament and of the Council on the protection of animals used for scientific purposes, the Swiss ordinance OPAn 455.1 of OSAV, and recommendations and regulations of DETA-DGNP permit number 20160422/01 AS).
(NZCS); National Museum of Natural History-Naturalis, Leiden (RMNH), presently holding the former Zoological Museum Amsterdam (ZMA) collection; Museum für Naturkunde, Berlin (ZMB); Zoologisches Staatssammlung, Munich (ZSM). Specimens included in morphometric analyses are indicated by an asterisk in specific lists of materials in main text and supplementary file (S1 Text), followed by number when needed.

Morphology
Measurements and counts (S1 Table) were obtained from a total of 269 specimens, and were only carried out on one side in cases of paired characters. Specimens were measured with a digital calliper to the nearest 0.01 mm following Fisch-Muller et al. [5]. Measurements are presented in tabular form as percentages of standard length (SL) except for subunits of the head, which are expressed as percentages of head length (HL). Counts are the following: premaxillary and mandibular teeth were counted for the emergent row, adding obviously missing teeth shown by gaps in tooth rows. Dermal plate counts included: 1) lateral plates in the median series of rows, according to Schaefer [63], 2) plates bordering the supraoccipital, 3) predorsal plates, counted dorsally along a median line between supraoccipital and nuchal plate, 4) lateral plates of dorsal series along the dorsal-fin base, 5) lateral plates in dorsal series between end of dorsal-fin base and adipose-fin insertion, 6) lateral plates in dorsal series between adipose-fin insertion and caudal fin, 7) lateral plates in ventral series between the anal and the caudal fins, 8) lateral plates in dorsal series between end of dorsal fin when adpressed and adipose-fin spine insertion, and 9) lateral plates in dorsal series along unpaired median plate(s) preceeding adipose fin. All plate counts are whole numbers except (8) and (9) that were counted to the nearest half-plate. Dorsal-fin and anal-fin branched rays were counted; other fin-ray counts do not vary among Hypostominae species and were only obtained for type specimens and part of the non-type material.

Morphometry
Morphometric and meristic data were subjected to multivariate analyses to reveal the morphological structure of the different species and populations of Guyanancistrus under study, with the addition of the possibly congeneric Pseudancistrus megacephalus to resolve taxonomic uncertainties. Prior to the analyses, all specimens smaller than 20 mm were excluded to minimize the bias introduced by allometric growth. Missing data, due to broken fin rays, were estimated for specimens belonging to a given population using the least squares method with the standard length (SL) used as the explanatory variable. Then, all morphometric data were standardized by SL and log transformed to control for size effect, to preserve and linearize allometric growth, and to prevent spurious correlations in the use of simple ratios [64]. Meristic data were used raw. The final table included data from 269 specimens belonging to 27 different morphs and populations, and contained 38 variables (24 morphometric and 14 meristic). This table was centered and reduced to allow comparison of variables expressed in different units, and submitted to a principal component analysis (PCA) using the correlation matrix to reveal its structuring. Because Guyanancistrus members appeared morphologically very close, and given the large number of variables in relation to the number of groups, a between group analysis (BGA) was secondarily performed on PCA results. To prevent artificial groupings, the different populations and morphs collected in different places for a given species were considered independently and used as a grouping factor (n = 27 groups). Prior to the BGA, a Monte Carlo permutation test on the value of between-group inertia was conducted using 9,999 random permutations to test against the absence of group effect. Multivariate analyses were performed using the ade4 1.7-4 [65] and ade4TkGUI 0.2-9 [66] packages in R 3.3.2 [67].

Genetics
To estimate the genetic diversity and species boundaries of Guyanancistrus members, the 5' region of the cytochrome c oxidase I (COI) mitochondrial gene was amplified for a DNA barcode analysis. In addition, a molecular phylogeny was reconstructed for 77 putative Guyanancistrus members and 12 outgroup species based on the analysis of mitochondrial and nuclear gene fragments (Table 1). Outgroup representatives were chosen from other genera, clades, and subfamilies of the Loricariidae following results of Covain & Fisch-Muller [50] and Lujan et al. [49]. The samples analyzed came from the tissue collection of MHNG. Markers were selected for their ability to resolve between-and within-species relationships, as well as deeper relationships at the intra-familial rank. For this we selected fast evolving markers such as the mtCOI, and the intronic regions of the nuclear Fish Reticulon-4 receptor (f-rtn4r) gene, whereas more conserved exonic regions of f-rtn4r and recombination activating gene 1 (rag1) provided information for deeper relationships. Total genomic DNA was extracted with the DNeasy Tissue Kit (Qiagen) following the instructions of the manufacturer. The PCR amplifications were carried out using the Taq PCR Core Kit (Qiagen). To amplify and sequence in a single run the standard 650 bp barcode region with high quality, fragment size was increased to 900 bp using two newly designed primers: 5COI-F (5'-CTC GGC CAT CCT ACC TGT G-3') and 5COI-R2 (5'-CGG GTG TCT ACG TCC ATT CCA ACT G-3'). The amplifications were performed in a total volume of 50 μl, containing 5 μl of 10x reaction buffer, 1 μl of dNTP mix at 10mM each, 1 μl of each primer at 10 μM, 0.2 μl of Taq DNA Polymerase equivalent to 1 unit of Polymerase per tube, and 1 μl of DNA. Cycles of amplification were programmed with the following profile: (1) 3 min. at 94˚C (initial denaturing), (2) 35 sec. at 94˚C, (3) 30 sec. at 53˚C, (4) 55 sec. at 72˚C, and (5) 5 min. at 72˚C (final elongation). Steps 2 to 4 were repeated 39 times. Amplifications of the nuclear rag1 and f-rtn4r genes followed Sullivan et al. [68] and Covain et al. [69] respectively. PCR products were purified with the High Pure PCR Product Purification Kit (Roche). Sequencing reactions were performed with the Big Dye Terminator Cycle Sequencing Ready Reaction 3.1 Kit (Applied Biosystems) following instructions of the manufacturer, and were loaded on an automatic sequencer 3100-Avant Genetic Analyzer (Applied Biosystems, Perkin-Elmer). Newly generated sequences were deposited in GenBank and BOLD with accession numbers provided in Table 1, while complementary sequences from previously published studies were obtained from GenBank (accession numbers and corresponding references in Table 1). The DNA sequences were edited and assembled using BioEdit 7.0.1 [70], aligned using ClustalW [71], and final alignment was optimized by eye.
For the barcode analysis, the aligned COI sequences were converted into a distance matrix to evaluate sequence divergences using the Kimura 2 Parameter (K2P) metrics [72] with pairwise deletion for missing data as implemented in spider 1.3.0 [73] in R. This K2P matrix was used to compute between-and within-species distances to allow threshold optimization and evaluate the existence of a barcoding gap for correct species identification. A levelplot graph allowing a graphical representation of the distance matrix was also computed using the lattice 0.20-34 [74] and colorspace 1.2.7 [75] packages in R.
For the phylogenetic reconstruction, four partitions were created corresponding to the COI, rag1, exonic regions of f-rtn4r, and intronic regions of f-rtn4r genes. Two phylogenetic reconstruction methods allowing the analysis of partitioned data were used. First, a maximum likelihood (ML) reconstruction was performed with RAxML 7.2.6 [76] and raxmlGUI 1.0 [77] using the GTRGAMMA model [78,79] with each partition assigned its own parameters. Robustness of the results was estimated by rapid bootstrapping [80,81] with 1,000 pseudoreplicates. Second, a Bayesian inference analysis was conducted in MrBayes 3.2.6 [82,83]. Two runs of four chains (one cold, three heated) were conducted simultaneously for 2 x 10 7 generations using the same model as the ML analysis (nst = 6, rates = gamma, and each partition assigned its own parameters), with the tree space sampled each 1000th generation. After a visual representation of the evolution of the likelihood scores, and checking for the stationarity of all model parameters using Tracer 1.5 [84] (i.e. potential scale reduction factor (PSRF), uncorrected roughly approached 1 as runs converged [85], and Effective Sample Size (ESS) of all parameters above 200), the 2 x 10 6 first generations were discarded as burn-in. The remaining trees were used to compute the consensus tree. Bayesian inference was performed using the CIPRES Science Gateway 3.3 [86].

Distribution
To explore the distributional patterns of the different species of Guyanancistrus, georeferenced data were recorded for the localities of specimens deposited in official collections. These collections were selected for housing a large sampling of Guianese species, i.e. MHNG, MNHN, RMNH, and ZMA (the two latter now grouped in the Naturalis Biodiversity Center, Leiden). In addition, relative abundances were computed according to the number of specimens in each batch for each species and locality. Relative abundances per species and per locality were then ploted as pie charts onto a geographic map. Because of the heterogeneity of samples (i.e. comprising a collection of small and large numbers of specimens), a constant was added for each abundance estimate for readability (i.e. occurrence + abundance). The map was reconstructed using raster images and shapefiles obtained from the HydroSHEDS [87] project website (http://www.worldwildlife.org/pages/hydrosheds) in conjunction with the shapefiles 0.7 [88], mapplots 1.5 [89], raster 2.5.8 [90], and colorRamps 2.3 [91] packages in R. Pie charts were computed with the mapplots package.

Multi-table analysis
Because Guyanancistrus brevispinis has been claimed to contain a species complex comprising at least five species [60], and that morphological characteristics were ambiguous, an integrative approach appeared necessary to clarify species boundaries. Because preliminary analyses provided information about the morphometric, phylogenetic, and spatial structures of G. brevispinis, the three types of information were united in a multiple co-inertia analysis (MCOA) [30] to identify the possible common structures of all datasets. For this, the three datasets were restricted to the subset of individuals and populations (n = 51 individuals distributed in 15 populations) for which all three types of information were available. Each of the three reduced  [93] using Cailliez [94] correction for non-Euclidian distance matrices was performed, to reveal its structuring. This analysis provided a tree-free representation of the distance matrix, where the pairwise distances between individuals on the axes are equal to the genetic pairwise distances of the matrix. The spatial structure was revealed by a PCoA performed on pairwise geographic distances computed from GPS coordinates using the geosphere 1.5.5 [90] package in R. A first assessment of a possible link between the three tables was obtained by performing pairwise Monte-Carlo permutation tests on the value of the RV coefficient [95] using 9,999 random permutations. Preliminary analyses, permutation tests, and multi-tables analyses were performed using the ade4 package in R.

Biogeography
Finally, after clarification of all systematic issues, a reappraisal of the phylogeography of Guyanancistrus members was performed to elucidate the different routes of dispersal, and extinction and speciation events which occurred in the Eastern Guianas (sensu [41]). For this a Dispersal Extinction Cladogenesis (DEC) analysis [96,97] was performed using the BioGeo-BEARS 0.2.1 [98] package in R. The DEC model possesses two free parameters (d = dispersal and e = extinction) and allows maximum likelihood estimates of ancestral areas along branches and nodes of a phylogenetic tree. The DEC model implemented in BioGeoBEARS is equivalent to the one implemented in LAGRANGE [97] but with the possibility of adding an additional parameter j for founder effect events as an additional explanation for cladogenesis. For ancestral area reconstructions, the phylogenetic tree was reduced to the ingroup, polytomies resolved using ape in R, and 11 areas were defined that corresponded to the present catchment areas of the coastal rivers of the Guianas where Guyanancistrus members were collected, i.e. from west to east: the Corantijn, Nickerie, Saramacca, Suriname, Maroni, Mana, Sinnamary, Comté-Orapu, Approuague, and Oyapock rivers, with the addition of the Amazon Basin (including Paru de Oeste and Jari rivers) for species distributed outside of the Guianas. For the calculation, the maximum number of ancestral areas allowed to be reconstructed to a given node was set to four. The best model for the maximum likelihood reconstruction was evaluated by likelihood ratio test (LRT) since DEC and DEC + j were nested models differing in a single parameter (j).

Nomenclatural acts
The electronic edition of this article conforms to the requirements of the amended International Code of Zoological Nomenclature, and hence the new names contained herein are available under that Code from the electronic edition of this article. This published work and the nomenclatural acts it contains have been registered in ZooBank, the online registration system for the ICZN. The ZooBank LSIDs (Life Science Identifiers) can be resolved and the associated information viewed through any standard web browser by appending the LSID to the prefix "http://zoobank.org/". The LSID for this publication is: urn:lsid:zoobank.org:pub:72F36C36-69D7-4E59-AF6A-8A88C78CFD86. The electronic edition of this work was published in a journal with an ISSN, and has been archived and is available from the following digital repositories: PubMed Central, LOCKSS.

Morphometry
The between-group inertia recorded by the BGA represented 48.13% of the total inertia of the preliminary PCA (sum of eigenvalues of the BGA / sum of eigenvalues of the PCA: 19.73155 / 41 = 0.4812573). The permutation test (Fig 1D) was highly significant, with none of the null hypothesis sampling distribution of randomized values greater than the observed value of between-group inertia (X obs = 0.4812573; p Xrand ! p Xobs = 0.0001). A significant group effect was thus present in the data, and observed between-group differences were not due to chance. Morphometric data were mainly structured on the first two axes of BGA ( Fig 1C) which explained 57.52% of the total between-class inertia (30.78% for axis 1 and 26.74% for axis 2). The first morphometric plane (axes 1 and 2) of individuals split the different populations and species of Guyanancistrus into four main groups ( Fig 1A). On the negative side of the first axis, the first group corresponded to a population originating from the northern slope of the Mitaraka Mountains (Gsp3; Jari River slope), followed by a mix of mostly G. brevispinis populations, thereafter called brevispinis group. In decreasing order of negative scores of variables (Fig 1B), species and population were rather characterized by higher values for number of branched anal-fin rays, interdorsal distance, number of plates bordering the supraoccipital, number of predorsal plates, caudal peduncle length, and upper caudal spine length. On the positive side, different species were aligned along the axis including G. niger, G. longispinis, the holotype of Pseudancistrus megacephalus, a population of Guyanancistrus from the Potaro River in Guyana  also identified as P. megacephalus by Eigenmann in 1908 (Gpot), and a population from the Nassau Mountains (GBM). These species and populations were rather characterized by higher values of cleithral width, supracleithral width, caudal peduncle depth, and head depth at supraoccipital. Along the second axis, on the positive side, G. niger, G. longispinis, P. megacephalus, and the population from Potaro River were split from all other populations and species and constituted the longispinis group. Higher values of pectoral spine length, dorsal spine length, and dorsal-fin base length split the longispinis group from other Guyanancistrus members. On the negative side of Axis 2, the population from the Nassau Mountains was separated by higher values of premaxillary tooth cup length, interbranchial distance, and dentary tooth cup length. The brevispinis group remained poorly characterized morphologically, with all of its members grouped at the center of principal axes. The morphometric approach was thus insufficient to delimit species boundaries of the brevispinis group members, and only six species could be characterized: G. niger, G. longispinis, P. megacephalus, a putatively three new species Gpot from Potaro River (= P. megacephalus sensu Eigenmann, 1908), Gsp3 from Mitaraka Mountains, and GBM from Nassau Mountains.

DNA barcodes
The sequence alignment of 77 COI barcodes reached a total length of 889 positions. No insertions, deletions, or stop codons were observed in any sequence. Five lineages and three levels of variations were highlighted by the K2P distances heatmap (Fig 2A). The first lineage comprised all populations of G. brevispinis and the one from the Nassau Mountains. Within-group variations ranged between 0 and 0.023 (mean = 0.011) whereas between-group ones ranged between 0.174 and 0.003 (mean = 0.08). The second group was constituted by populations from the Mitaraka Mountains (Jari and Maroni sides), a population from Paru de Oeste River, and one from the Brownsberg Mountains. In this group, within-group distances ranged between 0 and 0.027 (mean = 0.012), and between-group distances between 0.055 and 0.156 (mean = 0.075). The third group was constituted by G. niger members (within-group K2P distances 0 < d < 0.003, mean = 0.002; between-group K2P distances 0.120 < d < 0.193, mean = 0.136), and the fourth one by G. longispinis members (within-group K2P distances 0 < d < 0.0016, mean = 0.0008; between-group K2P distances 0.120 < d < 0.193, mean = 0.136). The last group comprised a single representative from the Paru de Oeste River. This specimen displayed high between-group variations ranging between 0.13 and 0.17 (mean = 0.14) K2P distances. One level of variation was revealed in global within-group distances, whereas three levels were present in between-group distances ( Fig 2B). Both global between-and within-group variations showed strong overlap with global within-group distances ranging between 0 and 0.023 (mean = 0.011), and global between-group distances ranging between 0.004 and 0.193 (mean = 0.088). The barcoding gap analysis revealed the absence of positive differences between the furthest intra-group distances and the closest non-specific for 51 individuals (Fig 2C). These individuals consisted mainly of representatives of G. brevispinis, the individuals from the Nassau Mountains, those from both sides of the Mitaraka Mountains, two specimens from Paru de Oeste and one from the Brownsberg Mountains, indicating that the barcoding approach was insufficient to discriminate all of the species of Guyanancistrus without ambiguity. The threshold optimization ( Fig 2D) delivered a minimum of false positive matches and minimum cumulative error with a K2P distance of 0.004, much below the usually accepted threshold (around 1-2%).
The barcoding approach was thus unable to distinguish between intra and inter specific variations for several species, including members of the putative G. brevispinis complex, and only seven mitochondrial lineages could be identified (1: G. brevispinis including GBM from Nassau Mountains; 2: a mix of Gsp3 and Gsp4 from Mitaraka Mountains; 3: the weakly diferenciated Gsp1 from Paru de Oeste and 4: Gkum from Brownsberg Mountains; 5: G. niger, 6: G. longispinis, and 7: Gsp2 from Paru de Oeste).

Molecular phylogeny
Given the poor results of the barcoding approach, nuclear marker sequences were added to the data set, and a molecular phylogeny reconstructed. In addition to the 77 mt COI gene fragments of Guyanancistrus members, 2 COI sequences of outgroup species, 56 sequences of the partial nuclear gene f-rtn4r, and 69 sequences of rag1 were sequenced (Table 1) The phylogenetic analysis revealed the presence of seven species of Guyanancistrus and of a new genus and species in the data. No species complex was present within G. brevispinis but three lineages of infraspecific rank emerged.

Distribution
Different distribution patterns were present in the data (Fig 4). (1) Species could be widespread. This pattern characterized G. brevispinis which dominated the other species in terms of both occurrences and abundances. The species was distributed in all important drainages including the Corantijn, Nickerie, Saramacca, Suriname, Maroni, Mana, Comté-Orapu, Approuague, and Oyapock rivers, and represented 86.5% of all specimens collected. When G. brevispinis was co-distributed with other Guyanancistrus members, such as in the Oyapock Basin, it appeared less frequent and abundant. (2) Species could be restricted to a region including a few basins or a single basin. This pattern was observed for G. niger and G. longispinis, both endemic to the Oyapock River, but distributed throughout the Oyapock drainage. (3) Species could be hyperendemic and restricted to a single place. This pattern concerned species restricted to mountainous areas of the Nassau, Brownsberg, Four Brothers, and Mitaraka mountains. Three basins had different distribution patterns for different species; patterns 1 and 2 were present in the Oyapock, whereas patterns 1 and 3 were present in the Maroni and Saramacca rivers.

Multi-tables analysis
Because only phylogenetic information was able to discriminate G. brevispinis among all other Guyanancistrus members ( Table 2) and none of the five lineages claimed as putative new species [60] could be clearly delineated by the different analyses, the multi-table approach was Phylogenetic tree (ln L = -15182.6) inferred from the combined analysis of sequences of partial COI mitochondrial gene, and partial f-rtn4r and rag1 nuclear genes. Numbers above branches correspond to bootstrap supports above 50 followed by posterior probabilities above 0.7. Background colors provide species boundries. When two colors are provided at the same level, left color refers to species limit whereas right color refers to subspecies limit; colors derived from applied. Prior to the analysis, the DNA barcode table was removed to minimize redundancy and avoid over weighting this information since the COI gene had been used to reconstruct  Table 2. Ability of the different methods to delineate without ambiguity species limits in Guyanancistrus spp. At least two congruent sources of information are expected to support taxonomic decisions. the phylogeny. A first assessment of the relationships between genetics (i.e. the phylogeny), morphology, and geography was performed using pairwise RV tests between preliminary analyses of the reduced datasets (PCoA for genetics and geography, and PCA for morphometric data). All pairwise tests (Table 3) showed strong and significant vector correlations between tables (p-value = 0.0001), with the pairwise correlations indicating that the genetic data were slightly more correlated to the geography (RV = 0.566, p-value = 0.0001) and morphological data (RV = 0.532, p-value = 0.0001) than the latter were to the geography (RV = 0.491, pvalue = 0.0001). However, all RV coefficients were globally comparable among pairwise comparisons with around 50% of common signal between tables. The first plane of MCOA accounted for 69.31% of the total co-structure (52.98% for axis 1 and 16.33% for axis 2) ( Fig  5C). The amount of variation explained by MCOA axes was similar to those obtained in the separate analyses, but with a lesser contribution from morphology. Indeed, 99.86% ((0.396 + 0.338)/(0.457 + 0.278) = 0.734/0.735) of the genetic data structure, 69.2% of the morphological data structure, and 100% of the geographic data structure were represented on the first two axes of the MCOA (Co-Inertia/Inertia in Table 3). The contribution of each table to the quantity maximized by MCOA (i.e. sum of squared covariance between the linear combinations of the variables of each table and the compromise = Cov 2 in Table 3) highlighted the relative importance of geography for the first axis, and of genetics for the second. Morphology contributed least to the compromise for both axes. The associated correlations (Cos 2 in Table 3) showed that the first two axes of the compromise were strongly linked to each separated table except for the second axis derived from geographic data (0.890 and 0.962 for the genetic data, 0.865 and 0.758 for the morphometric data, and 0.965 and 0.157 for the geographic data). The   Three groups of infra-specific rank showing significantly structured data concerning their distribution, genetics, and morphology were consequently recognized.  Table 4  Diagnosis. Guyanancistrus was shown to be monophyletic based on mitochondrial and nuclear DNA sequences. No unique morphological character was found to diagnose the genus which belongs to the Ancistrini tribe of the Hypostominae subfamily. The following combination of characters distinguishes Guyanancistrus from all other Hypostominae genera: head and body dorsoventrally depressed; head and body plates not forming prominent ridge or crest; snout rounded and flattened; snout covered with plates except tip region and, sometimes, a small area on each side of tip of snout; plates on all parts of snout forming a rigid armor covered with numerous short odontodes; presence of odontodes over a broad area on the opercle; presence of enlarged cheek odontodes supported by evertible plates; these odontodes straight with tips slightly curved, not strongly hook-shaped; absence of whisker-like cheek odontodes; absence of enlarged odontodes along snout margin; presence of a dorsal iris operculum; lips forming an oval disk; dentary and premaxillary with numerous viliform and bicuspid teeth; presence of a small buccal papilla, no enlarged dentary papilla; seven branched dorsal-fin rays; presence of an adipose fin; no membranous extension between end of dorsal fin and adipose fin; five series of lateral plates extending to caudal fin; lateral plates not keeled and not bearing enlarged odontodes; lateral plates of ventral series on caudal peduncle angular but not keeled; abdominal region entirely naked. Guyanancistrus is mostly similar to Cryptancistrus n. gen in external appearance. It is distinguished from Cryptancistrus primarily by the uniformity of its snout plates and odontodes (in Cryptancistrus. posterior part of lateral margin of snout do not form a rigid armor but rather a soft fleshy border, and bears slightly enlarged odontodes with small tentacules sensu Sabaj et al. [102]). It can additionally be distinguished from Cryptancistrus by the presence of a skin region bordering the exposed portion of opercle that is distinctly narrower than the latter (vs roughly as large as the latter).
Distribution. Endemic to the Atlantic coastal rivers and upper Amazonian tributaries of the north-eastern Guiana Shield.
Guyanancistrus brevispinis (  Diagnosis. Guyanancistrus brevispinis is discriminated from all congeners except G. nassauensis n. sp. by specific barcode sequences (see BOLD numbers in Table 1) and by much shorter evertible cheek odontodes, longest ones usually only reaching the first half of the opercle, or, in some large specimens measuring more than 70 mm SL (probably adult males), surpassing the middle of the opercle, but not reaching its last quarter (vs reaching between last quarter up to far beyond posterior end of opercle except in very small specimens). Evertible cheek odontodes are shorter in G. brevispinis than in G. nassauensis with regard to size of specimens; in the latter, large specimens (likely adult males) that have odontodes reaching beyond the middle of the opercle measure only 40 mm. Guyanancistrus brevispinis is a larger species than G. nassauensis (maximum known SL: 152 mm vs 61 mm). It is further discriminated from G. nassauensis by smaller dentary and premaxillary tooth cusps (in % of head length, respectively: 15. Guyanancistrus brevispinis is readily distinguished from G. longispinis and G. niger by color pattern (body and fins medium brown with paler yellow to orange medium-sized spots to transverse bands, vs brown-black with either small roundish yellow spots for G. longispinis    Integrative systematics of Guyanancistrus with description of a new genus and six new species Integrative systematics of Guyanancistrus with description of a new genus and six new species  ( Fig 6A and 6B), or white dots for G. niger (Fig 6D)). Guyanancistrus brevispinis can also be distinguished from G. brownsbergensis and G. tenuis by a smaller number of plates between adpressed dorsal fin and adipose-fin spine (0.5-3, mean 2, vs respectively 3-3.5, mean 3, and 3-4, mean 3.5), from G. brownsbergensis by a lower peduncle depth (8.6-11.3, mean 10.0% of SL, vs 11.4-11.6, mean 11.5), and from G. megastictus by the lower number of plates bordering the supraoccipital (2-3, mean 3 vs 4) and by a color pattern with smaller roundish spots (not covering four plates) or straighter bands on posterior part of body and fins. Description. Morphometric and meristic data in Table 4. Relatively large-sized species for Guyanancistrus (up to 152 mm SL). Head and body dorsoventrally depressed. Dorsal profile Integrative systematics of Guyanancistrus with description of a new genus and six new species gently convex from snout tip to dorsal-fin origin, usually more flattened posterior to orbit, slightly convex and sloped ventrally from dorsal-fin origin to adipose fin, then slightly concave to procurrent caudal-fin rays, and rising to caudal fin. Ventral profile flat from snout to base of caudal fin.
Dorsal contour of head smooth, no ridge or keel, inconspicuous rounded elevations on the midline of the snout and anterior to orbits, supraoccipital nearly flat. Dorsal margin gently flattened from base of first branched dorsal-fin ray to base of adipose fin between very slight Head and body plated dorsally, plates generally covered by short and uniformly distributed odontodes. Tip of snout always naked, and, except in some large specimens, small naked area meeting the dorsolateral edge of upper lip on each side of tip of snout; in large specimens, dorsolateral margin of the upper lip supporting from few odontodes up to several plates covered with small odontodes (Fig 7A, 7B and 7C). Lateral margin of snout covered with plates forming a rigid armor with short odontodes. Opercle supporting odontodes. A narrow unplated area bordering posterodorsal margin of opercle. Evertible cheek plates with enlarged odontodes in highly variable number, from less than ten up to approximately 40 in some large specimens. These cheek odontodes straight with tips curved, longest reaching from the first quarter (in small specimens) up to the third quarter (in large specimens) of the opercle. Usually three rows of plates and a curved nuchal plate between supraoccipital plate and dorsal-fin spinelet. Five series of lateral plates extending to caudal fin. Odontodes on lateral series of plates not Integrative systematics of Guyanancistrus with description of a new genus and six new species forming keels. Odontodes on posterior part of pectoral-fin spine only slightly enlarged, except in large specimens (presumably males). Abdominal region totally naked. No platelike structure before the anal fin. Ventral part of caudal peduncle plated; presence of a large smooth area devoid of odontodes around anal fin.
Dorsal-fin origin slightly anterior to pelvic-fin origin. Dorsal fin relatively short; when adpressed, never reaching adipose fin, often very distant from it. Adipose fin roughly triangular, preceded by one (or two fused into one) median unpaired raised plate. Adipose spine straight or slightly convex dorsally, membrane posteriorly convex. Pectoral-spine tip reaching to approximately one-third of pelvic spine in most specimens, exceptionally extending over its middle in large specimens (presumably males). Anal fin with weak spine, its margin convex. Caudal fin concave, ventral lobe longer than dorsal lobe. Fin-ray formulae: dorsal II,7; pectoral I,6; pelvic i,5; anal i,5 (except 3 specimens with I,4); caudal i,14, i.
Coloration. Dorsal coloration pattern highly variable according to size of specimen and collection locality (see Fig 8 and discussion). In life, base color light grey-brown or orangebrown to dark brown, except whitish ventral region. Dorsal spotting pattern varies from presence of indistinct paler spots, to presence of numerous small to medium-sized, roundish to elongated, faded to brilliant yellow-orange spots on the whole body, or on its anterior part only; in that case spots form similarly colored transverse bands on posterior part of body. Such bands are observed on juveniles of all populations, and remain on larger specimens in some populations.
Spots of similar color to those of body are usually present on dorsal and caudal fins, centred on fin rays and often combined to form regular or irregular transverse bands, more generally on caudal fin. Pectoral and pelvic fins with less-distinct spots, sometimes with large pale areas. Fin margins can be of same color as spots, especially in small specimens. Fin membranes usually grayish-brown.
Distribution and habitat. Guyanancistrus brevispinis was found in the main Guianese river systems of Suriname and French Guiana, including the Corantijn at Guyanese-Surinamese border, the Nickerie, Coppename, Saramacca, Suriname, Maroni, Mana, Sinnamary, Mahury (Comté-Orapu), Kaw, Approuague and the Oyapock at the French-Brazilian border. These rivers all have a south-north orientation and flow into the Atlantic Ocean. It inhabits most clear-water rivers, streams and creeks, with running waters and rocky or sandy bottoms, over which specimens blend remarkably well. The species may be locally abundant. Its habitat is usually shadowed by primary rainforest, with measured water temperature varying from 23.7 to 28.4˚C, conductivity from 13 to 42 us and pH from 5.96 to 7.06.
Etymology. The specific name brevispinis, derived from Latin and meaning short thorns, was originally given in reference to the short evertible cheek odontodes.
Distribution. The nominate subspecies occurs from the Corantijn River basin in western Suriname to the upper Tapanahony River basin, a Marowijne River tributary in eastern Suriname. It was not found in French Maroni River tributaries (Fig 4).
Etymology. As for nominal species. Non type material. See S1 Text. Table 1), which are however not unique as they are partly shared by Guyanancistrus nassauensis. It is distinguished from the latter by the diagnostic characters listed under species diagnosis. No morphometric variable unambiguously distinguishes bifax from other subspecies (see Table 4), but several have significantly different mean values. Compared to both brevispinis and orientalis, on average, bifax has a smaller interbranchial distance (in % of SL: 21.91 ± 0.65 vs 22.73 ± 0.97 for brevispinis and 22.22 ± 0.95 for orientalis; p-value 1.624e -8 ), a more depressed caudal peduncle (in % of SL: 9.45 ± 0.47vs 10.50 ± 0.40 for brevispinis and 9.97 ± 0.40 for orientalis; p-value 2.2e -8 ), and a longer caudal fin (lower caudal-fin spine in % of SL: 33.63 ± 2.38 vs 32.79 ± 2.36 for brevispinis and 32.53 ± 2.11 for orientalis; p-value 0.00684). Additional differences of morphometric variables from either brevispinis or orientalis are statistically significant, some being listed under the respective diagnoses of these subspecies. On average, the number of plates between dorsal base and adipose fin is smaller for bifax (6. ± 0.5 vs 6 ± 1 for brevispinis and 6 ± 0.5 for orientalis; p-value 0.01807).

Diagnosis. Guyanancistrus brevispinis bifax is differentiated from other subspecies and species of Guyanancistrus by its barcode sequences (see BOLD numbers in
Remark: In several populations of the subspecies throughout its area of distribution, the head of some specimens (MHNG 2683.043; MHNG 2699.060; MHNG 2723.008) has a distinctive appearance, with an enlarged forehead part (Fig 10), usually coupled with a slightly longer snout, head and/or predorsal length, as well as an enlarged mouth. This difference between individuals appears not linked to sex, and apparently independant of the genetic data examined.
Distribution. Guyanancistrus b. bifax occurs from the Maroni River and its eastern tributaries up to the Mana and Sinnamary rivers basins in French Guiana (Fig 4).
Etymology. Named bifax, a noun in apposition, meaning two faces, for the different appearances of the head observed within the subspecies (see Remark).
Etymology. The name orientalis, from the Latin name oriens, is given because of the eastern distribution of the subspecies.  , and by color pattern (body and fins uniformly brown or with indistinct medium sized paler spots, vs brown-black with either small roundish yellow spots for G. longispinis, or white dots for G. niger). It is further separated from all G. brevispinis group species by having, on average, the widest body, the deepest and longest head, the largest interbranchial distance, the shortest fins, and the highest number of teeth (see Table 5).

Guyanancistrus nassauensis
Description. Morphometric and meristic data in Table 5. Small-sized species (largest specimen observed 61.0 mm SL; holotype, 42.0 mm SL, likely a breeding male). Head and body dorsoventrally depressed and wide. Dorsal profile gently convex from snout tip to dorsal-fin origin, usually more flattened posterior to orbit, slightly convex and sloped ventrally from dorsal-fin origin to adipose fin, then slightly concave to procurrent caudal-fin rays, and rising to caudal fin. Ventral profile flat from snout to base of caudal fin.
A low median ridge from tip of snout to nostrils, sometimes bordered by lateral depression, a slight elevation anterior to orbits, supraoccipital slightly convex to flat. Dorsal margin gently flattened from base of first branched dorsal-fin ray to base of adipose fin between very slight ridges formed with lateral plates of dorsal series. First lateral plates of mid-ventral series forming low lateral ridge. Caudal peduncle roughly ovoid in cross section, flattened ventrally, and more compressed posteriorly.
Large, rounded and laterally flattened snout. Eye relatively small. Large oval mouth, lower lip wide, not or just reaching pectoral girdle, upper lip narrower. Lips forming an oval disk, covered with short papillae. Presence of a single narrow buccal papilla. Very short maxillary barbel. Teeth short and strong with a relatively long bicuspid crown, lateral lobe about half size of medial lobe.
Head and body plated dorsally, plates generally covered by short and uniformely distributed odontodes.Tip of snout largely naked. Lateral margin of snout covered with plates forming a rigid armor with short odontodes. Opercle supporting odontodes. A narrow unplated area bordering posterodorsal margin of opercle. Evertible cheek plates with enlarged odontodes in highly variable number, from fewer than ten up to approximately 35 in some large specimens. These cheek odontodes straight with tips curved, the longest usually reaching middle of the opercle, or beyond in large specimens. Two to four rows of plates between supraoccipital plate and dorsal-fin spinelet, nuchal plate often covered by skin. Five series of lateral plates extending to caudal fin. Odontodes on lateral series of plates not forming keels. Odontodes on posterior part of pectoral-fin spine enlarged, only slightly in small specimens, much more significantly in large specimens (presumably males). Abdominal region totally naked. No platelike structure before the anal fin. Ventral part of caudal peduncle covered with plates showing a highly reduced number of odontodes.
Dorsal-fin origin slightly anterior to pelvic-fin origin. Dorsal fin short; when adpressed, far from reaching preadipose unpaired plate. Adipose fin roughly triangular, preceded by one, or two fused into one, median unpaired raised plate. Adipose spine straight or slightly convex dorsally, membrane posteriorly convex. Pectoral-spine short, tip usually reaching the first quarter of pelvic spine, exceptionally extending up to the first third in large specimens (presumably males). Anal fin short with weak spine, its margin convex. Caudal fin slightly concave, ventral lobe longer than dorsal lobe. Fin-ray formulae: dorsal II,7; pectoral I,6; pelvic i,5; anal i,4 (except 1 specimen, 49.6 mm SL, with i,5); caudal i,14, i.
Coloration. In alcohol, dorsal part of body uniformly grey-brown, ventral part yellowish except usually patches of melanophores on lateral parts and in the anal region, and abdomen whitish. Fin-rays brownish, with medium-sized spots by some specimens, these spots forming or not forming bands; margin of caudal fin often orange-or red-brown; fin-membranes usually not pigmented, or pigment restricted to areas bordering rays. In life (based on a photograph of one specimen), dorsal coloration of body brown with some lighter ill-defined orangebrown spots; fins orange-brown, fin-membranes hardly pigmented (Fig 7E).
Distribution and habitat. Guyanancistrus nassauensis is known solely from Paramaka Creek and some of its tributaries, Marowijne River Basin, in the Surinamese Nassau Mountains (an area of approximately 20x20 km2) (Fig 4). At an elevation of 277 m, the type locality is located in a northern branch of Paramaka Creek, a medium-sized and shallow stream (3-7 m. width; less that 50 cm depth) with pools and some riffle habitat, a rocky substrate, and bordered by terra firme rainforest. Water was transparent, with a mean pH of 6.26, conductivity 24.2 μS/cm and temperature 23.2˚C. Specimens were collected there by electrofishing with set seine, along with several Harttiella crassicauda, another species endemic to streams in the Nassau Mountains.
Etymology. The name nassauensis is a reference to the distribution of the new species which is only known in streams in the Nassau Mountains, an area now under threat of a proposed bauxite mine and illegal gold mining. Diagnosis. Guyanancistrus brownsbergensis is differentiated from all congeners by its specific barcode sequences (GBOL697-14, GBOL689-14, GBOL690-14, GBOL691-14, GBOL726-14, GBOL725-14, and GBOL724-14). It is morphologically distinguished from G. longispinis and G. niger by a much shorter pectoral-fin spine (in % of SL: 27.7-29.7, mean 28.7, in SL vs 31.9-45.5, mean 40.2, and 33.3-48.0, mean 42.8, respectively), having shorter odontodes, and by color pattern (body and fins medium grey-brown with yellowish beige to light brown medium to large-sized spots, vs brown-black with either small roundish yellow spots for G. longispinis, or white dots for G. niger). Guyanancistrus brownsbergensis is distinguished from all species of the Guyanancistrus brevispinis group except G. nassauensis by a deeper caudal peduncle (11.4-11.6, mean 11.5, vs 11.3 or less % of SL). It is differentiated from G. nassauensis by smaller dentary and premaxillary tooth cusps (18.5 and 17.6% of head length, vs 24.2-31.9, mean 27.6, and 25.4-31.4, mean 28.1) by an anal fin with 5 branched rays (vs 4).
The caudal peduncle of G. brownsbergensis is not only deep, but it is also short compared to G. tenuis and G. megastictus Description. Morphometric and meristic data in Table 5. Head and body strongly dorsoventrally depressed. Dorsal profile gently convex from snout tip to dorsal-fin origin, flattened posterior to orbit, slightly convex and sloped ventrally from dorsal-fin origin to end of adipose fin, then slightly concave and rising to caudal fin. Ventral profile flat from snout to base of caudal fin.
Very low median ridge from tip of snout to nostrils present, parallel to this, similar inconspicuous ridges from snout border to nostrils, then somewhat more elevated to orbits, supraoccipital nearly flat. Dorsal margin gently flattened along dorsal-fin base between very slight ridges formed with lateral plates of dorsal series. First lateral plates of mid-ventral series forming low lateral ridge. Caudal peduncle high, roughly ovoid in cross section, flattened ventrally, and more compressed posteriorly.
Snout rounded anteriorly. Eye relatively small. Lips forming an oval disk, covered with short papillae. Presence of a single small and narrow buccal papilla. Lower lip wide, not reaching pectoral girdle, upper lip narrower. Very short maxillary barbel. Teeth slender, bicuspid, lateral lobe about half size of medial lobe.
Head and body plated dorsally, plates generally covered by short and uniformely distributed odontodes. Tip of snout naked; small (small specimens) to minute (holotype, which is the largest specimen) naked area on each side of the latter; dorsolateral margin of the upper lip supporting patches of odontodes (very small in the smallest specimen). Lateral margin of snout covered with plates forming a rigid armor with short odontodes. Opercle supporting odontodes. A narrow unplated area bordering posterodorsal margin of opercle. Evertible cheek plates with approximately 25 to 50 enlarged odontodes. These cheek odontodes straight with tips curved, the longest reaching beyond end of opercle. Three rows of plates and a curved nuchal plate between supraoccipital plate and dorsal-fin spinelet. Five series of lateral plates extending to caudal fin. Odontodes on lateral series of plates not forming keels. Odontodes on posterior part of pectoral-fin spine moderately enlarged. Abdominal region totally naked. No platelike structure before the anal fin. Ventral part of caudal peduncle plated; presence of a large smooth area devoid of odontodes around anal fin.
Dorsal-fin origin slightly anterior to pelvic-fin origin. Dorsal fin relatively short; when adpressed, distant by one (holotype) or two plates from median unpaired plate preceding adipose fin. Adipose fin roughly triangular; spine slightly convex dorsally, membrane straight or slightly concave posteriorly. Pectoral-spine tip reaching first quarter of pelvic spine. Anal fin with weak spine, its margin convex. Caudal fin concave, ventral lobe longer than dorsal lobe. Fin-ray formulae: dorsal II,7; pectoral I,6; pelvic i,5; anal i,5; caudal i,14, i.
Coloration. Dorsal coloration pattern grey-brown in life, lighter in alcohol. Medium to large-sized spots irregularly distributed on dorsal part of body. These spots are yellowish-beige to light brown. Spots on fins, at least on caudal, forming bands. Border of dorsal and caudal fins orangish colored in life. Ventral coloration pale yellow and unspotted (Fig 7F). Etymology. Species named for the Brownsberg Nature Park in Brownsberg Mountains, in which it was found, and which is presently under threat from illegal gold mining [109].

Diagnosis. Guyanancistrus teretirostris is distinguished from all congeners by its specific barcode sequences
Description. Morphometric and meristic data in Table 5. Head and body dorsoventrally depressed. Dorsal profile gently convex from snout tip to orbit level, then nearly flat, slightly convex and sloped ventrally from dorsal-fin origin to adipose fin, then slightly concave to procurrent caudal-fin rays, and rising to caudal fin. Ventral profile flat from snout to base of caudal fin.
Dorsal contour of head smooth, no ridge or keel, inconspicuous rounded elevations on the midline of the snout and anterior to orbits, supraoccipital nearly flat. Dorsal margin gently flattened from base of first branched dorsal-fin ray to base of adipose fin between very slight ridges formed with lateral plates of dorsal series. First lateral plates of mid-ventral series forming low lateral ridge. Caudal peduncle roughly ovoid in cross section, flattened ventrally, and more compressed posteriorly.
Snout fully rounded anteriorly. Eye moderately large. Lips forming an oval disk, covered with short papillae. Presence of a single narrow buccal papilla. Lower lip wide, not reaching pectoral girdle, upper lip narrower. Short maxillary barbel. Teeth bicuspid, lateral lobe about half size of medial lobe.
Head and body plated dorsally, plates generally covered by short and uniformely distributed odontodes. Tip of snout naked; very small area on each side of the latter is also naked in the smaller paratypes; dorsolateral margin of the upper lip supporting several platelets with short odontodes, or naked (smallest paratype). Lateral margin of snout covered with plates forming a rigid armor with short odontodes. Opercle supporting odontodes. A narrow unplated area bordering posterodorsal margin of opercle. Evertible cheek plates with approximaterly 25-35 enlarged odontodes. These cheek odontodes straight with tips curved, longest nearly reaching posterior end of opercle (smallest paratype) or beyond (holotype and second paratype). Three rows of plates and a curved nuchal plate between supraoccipital plate and dorsal-fin spinelet. Five series of lateral plates extending to caudal fin. Odontodes on lateral series of plates not forming keels. Odontodes on posterior part of pectoral-fin spine only slightly enlarged. Abdomen totally naked. No platelike structure before the anal fin. Ventral part of caudal peduncle plated; presence of a moderately large smooth area devoid of odontodes around anal fin.
Dorsal-fin origin slightly anterior to pelvic-fin origin. Dorsal fin relatively short; when adpressed, distant by at least one plate from median unpaired plate preceding adipose fin. Adipose fin roughly triangular; spine slightly convex dorsally, membrane posteriorly convex. Pectoral-spine tip nearly reaching third of pelvic spine (holotype), or less. Anal fin with weak spine, its margin convex. Caudal fin concave, ventral lobe longer than dorsal lobe. Fin-ray formulae: dorsal II,7; pectoral I,6; pelvic i,5; anal i,5 (i,4 in paratype 87 mm SL); caudal i,14,I.
Coloration. In alcohol, dorsal ground color of body medium grey-brown, somewhat darker on head and lighter on lower part of caudal peduncle. Body dorsally covered with yellow-beige to light-brown small to medium-sized spots, usually rounded anteriorly and more irregular in shape posteriorly. Body ventrally yellow-beige, with abdomen whitish, unspotted.
All fins of similar color to body, and spotted, anal fin excepted. Spots of dorsal-fin medium sized and rounded. Spots of paired fins similar but less distinct; anterior part of these fins clearly darker that posterior part. Spots of caudal fin forming three to four large light and irregular transverse bands; margin of fin apparently orangish colored.
Distribution. Known from the Upper Parú de Oeste River (Fig 4). Etymology. The species name teretirostris, is derived from the Latin words teres, meaning rounded and smooth, and rostris, meaning snout; an allusion to the snout shape of this species.
Guyanancistrus tenuis, new species. urn:lsid:zoobank.org:act:225451B6-1C40-4DC0-BC5A-553A4B2530D4  Diagnosis. Guyanancistrus tenuis is distinguished from all congeners by its specific barcode sequences (GBOL739-14, GBOL738-14, and GBOL737-14). Morphologically, it is distinguished from G. longispinis and G. niger by a much shorter pectoral-fin spine (in % of SL: 26.0-28.1, mean 27.0, vs respectively 31.9-45.5, mean 40.2, and 33.3-48.0, mean 42.8) supporting shorter odontodes, and by color pattern (yellow-beige medium to large-sized spots and bands on body and fins, vs small roundish yellow spots for G. longispinis, or white dots for G. niger). Guyanancistrus tenuis is a particularly slender species, and, with G. megastictus, it is the most depressed of all brevispinis group species. Guyanancistrus tenuis can be separated from G. nassauensis and G. brownsbergensis by lower head depth values (13.5-14.8 mean 14.0% of SL vs 15.2 or more), and from all species but G. brevispinis by a lower caudal peduncle depth (8.9-9.6 mean 9.3% of SL vs 10.4 or more), but only mean head depth discriminates G. tenuis from G. brevispinis (14.1-19.6, mean 16.2% of SL in the latter). In the brevispinis species group, G. tenuis additionally shows the narrowest body, distinguishing it from G. nassauensis and G. megastictus Description. Morphometric and meristic data in Table 5. Head and body up to caudal peduncle very dorsoventrally depressed and narrow, resulting in a slender aspect. Dorsal profile gently convex from snout tip to orbit level, then nearly flat to dorsal-fin origin, slightly convex and sloped ventrally from that point to adipose fin, then slightly concave to procurrent caudal-fin rays, and rising to caudal fin. Ventral profile flat from snout to base of caudal fin.
Dorsal contour of head smooth, usually a very low median ridge from tip of snout to nostrils, slight elevation anterior to orbits, sometimes (including holotype) bordered by a shallow lateral depression, supraoccipital nearly flat. Dorsal margin gently flattened from base of first branched dorsal-fin ray to base of adipose fin between very slight ridges formed with lateral plates of dorsal series. First lateral plates of mid-ventral series forming low lateral ridge. Caudal peduncle roughly ovoid in cross section, flattened ventrally, and more compressed posteriorly.
Snout rounded anteriorly. Eye moderately large. Lips forming an oval disk, covered with short papillae. Presence of a single narrow buccal papilla. Lower lip wide, not reaching pectoral girdle, upper lip narrower. Short maxillary barbel. Teeth slender, bicuspid, lateral lobe about half size of medial lobe.
Head and body plated dorsally, plates generally covered by short and uniformely distributed odontodes. Tip of snout naked, and often (particularly in small specimens) also a very small naked area on each side of the latter, separated by a plated area which continues for a short distance on dorsolateral margin of upper lip. Lateral margin of snout covered with plates forming a rigid armor with short odontodes. Opercle supporting odontodes; on its ventral margin, odontodes usually slightly enlarged. A relatively large unplated area bordering posterodorsal margin of opercle. Evertible cheek plates with enlarged odontodes in highly variable number, from approximately 10 up to approximately 35 in large specimens. These cheek odontodes straight with tips curved, longest not reaching middle of opercle in smallest specimens, but reaching it or beyond in specimens of approximately 50 mm SL, and reaching last quarter to well beyond end of opercle in specimens of approximately 60 mm SL (beyond in holotype). Usually three rows of plates and a curved nuchal plate between supraoccipital plate and dorsalfin spinelet. Five series of lateral plates extending to caudal fin. Odontodes on lateral series of plates not forming keels. Odontodes on posterior part of pectoral-fin spine very slightly enlarged. Abdominal region totally naked. No platelike structure before the anal fin. Ventral part of caudal peduncle plated; a moderately large smooth area devoid of odontodes around anal fin.
Dorsal-fin origin slightly anterior to pelvic-fin origin. Dorsal fin short; when adpressed, tip of fin very distant from adipose fin, and even far from reaching preadipose unpaired plate. Adipose fin roughly triangular, preceded by one, or two fused into one, median unpaired raised plate. Adipose spine straight or slightly convex dorsally, membrane posteriorly straight or slightly convex. Pectoral-spine tip reaching slightly over pelvic-fin origin. Anal fin with weak spine, its margin convex. Caudal fin concave, ventral lobe longer than dorsal lobe. Finray formulae: dorsal II,7; pectoral I,6; pelvic i,5; anal i,5; caudal i,14, i.
Coloration. In alcohol, dorsal ground color of body brown, covered with yellow-beige medium-sized spots on head, becoming gradually much larger spots up to end of caudal peduncle. In small specimens, some spots are roundish and large (but covering fewer than four plates) but spots usually coalesce to form large and highly contrasted stripes on posterior part of body (Fig 16). Ventrally, color of body more or less uniformly light brown apart from the abdomen, which is mainly whitish, sometimes with diffuse brown pigmentation.
All fins colored similarly to dorsum, and spotted. Spots of dorsal fin medium sized, forming transverse bars or not; usually a dark spot on membrane between origin of spine and first branched ray. Spots of other fins less distinct. Spots of caudal fin forming two to three highly constrasted, large and irregular light-colored transverse bands; tips of fin light-colored.
Distribution. Guyanancistrus tenuis is known solely from a small forest tributary of the Mapaoni River, Upper Jari River Basin, in the Massif du Mitakara, a mountain range in the far southwest of French Guiana (Fig 4). This north-south oriented mountain creek was essentially rocky, shallow (20-60 cm depth), with medium to strong currents, and some pools.
Etymology. The name tenuis is a Latin word meaning thin, in reference to the slender body of the species.
Guyanancistrus megastictus, new species. urn:lsid:zoobank.org:act:AAB3CE9E-055D-4DD6-8B9C-682FC8F5E50B (Figs 6C and 17; Table 5  Description. Morphometric and meristic data in Table 5. Dorsal profile gently convex from snout tip to orbit level, then nearly flat to dorsal-fin origin, slightly convex and sloped ventrally from that point to adipose fin, then slightly concave to procurrent caudal-fin rays, and rising to caudal fin. Ventral profile flat from snout to base of caudal fin. Dorsal contour of head smooth, a very low median ridge from tip of snout to nostrils, slight elevation anterior to orbits, bordered (paratype) or not (holotype) by a shallow lateral depression, supraoccipital nearly flat. Dorsal margin gently flattened from base of first branched dorsal-fin ray to base of adipose fin between very slight ridges formed with lateral plates of dorsal series. First lateral plates of mid-ventral series forming low lateral ridge. Caudal peduncle roughly ovoid in cross section, flattened ventrally, and more compressed posteriorly.
Snout rounded anteriorly. Eye relatively small. Lips forming an oval disk, covered with short papillae. Presence of a single triangular buccal papilla. Lower lip wide, not reaching pectoral girdle, upper lip narrower. Short maxillary barbel. Teeth slender, bicuspid, lateral lobe about half the size of medial lobe.
Head and body plated dorsally, plates generally covered by short and uniformely distributed odontodes. Tip of snout naked, and a minute naked area on each side of the latter, separated by a plated area which continues for a short distance on dorsolateral margin of upper lip. Lateral margin of snout covered with plates forming a rigid armor with short odontodes. Opercle supporting odontodes. A straight unplated area bordering posterodorsal margin of opercle. Evertible cheek plates with approximately 25-30 enlarged odontodes, straight with tips curved, longest nearly reaching posterior end of opercle or beyond. Three rows of plates and a curved nuchal plate between supraoccipital plate and dorsal-fin spinelet. Five series of lateral plates extending to caudal fin. Odontodes on lateral series of plates not forming keels. Odontodes on posterior part of pectoral-fin spine very slightly enlarged. Abdominal region totally naked. No platelike structure before the anal fin. Ventral part of caudal peduncle plated; a moderately large smooth area devoid of odontodes around anal fin.
Dorsal-fin origin slightly anterior to pelvic-fin origin. Dorsal fin short; when adpressed, tip of fin very distant from adipose fin, and even far from reaching preadipose unpaired plate. Adipose fin roughly triangular, preceded by one, or two fused into one, median unpaired raised plate. Adipose spine relatively long compared to other Guyanancistrus species, slightly convex dorsally, membrane posteriorly straight. Pectoral-spine tip reaching slightly beyond pelvic-fin origin or nearly one fifth of fin spine (holotype). Anal fin with weak spine, its margin convex. Caudal fin concave, ventral lobe longer than dorsal lobe. Fin-ray formulae: dorsal II,7; pectoral I,6; pelvic i,5; anal i,5; caudal i,14, i.
Coloration. In alcohol, dorsal ground color of body a bleached brown, covered with yellow-beige medium-sized spots on head, then large spots, and very large roundish spots (covering at least six lateral plates) and bars posterior to dorsal-fin origin level and up to end of caudal peduncle. Ventrally, color of body more or less uniformly yellowish apart from mainly whitish abdomen.
All fins except anal similarly colored to dorsum, and lightly spotted. Spots of dorsal fin large, forming one or two large transverse bars. Spots on paired fins less distinct. Anal fin yellowish. Caudal fin mainly light colored: narrow brownish base, followed by a very large lightly colored transverse bar, then a narrower brownish transverse bar, and tips of fin light.
In life (Fig 7C), background color greenish-brown, with darker areas surounding the light spots, and caudal-fin base also darker.
Distribution. Guyanancistrus megastictus is known from a small forest tributary of the Upper Maroni River Basin in the Massif du Mitaraka, a mountain range in the far south-west of French Guiana (Fig 4). The only two known specimens were caught poison fishing in a shallow (20-60 cm depth) and mainly sandy portion of this river named Crique Alama.
Etymology. The Latin word megastictus is derived from the Ancient Greek mega, meaning large, and stictos, meaning spotted, in reference to the presence of very large size spots on body and fins.
Key to species of Guyanancistrus.  -14). No unique morphological character was found to diagnose the genus which belongs to the Ancistrini tribe of the Hypstominae subfamily. The following combination of characters distinguishes Cryptancistrus from all other Hypostominae genera: head and body dorsoventrally depressed; head and body plates not forming prominent ridge or crest; snout rounded, and covered with contiguous plates except tip region, and posterior part of lateral margin of snout; latter area forming a soft fleshy border, and bearing slightly enlarged odontodes associated with small fleshy tentacules sensu Sabaj et al. [102]); presence of odontodes over a broad area on the opercle; presence of numerous enlarged cheek odontodes supported by evertible plates; these odontodes straight with tips slightly curved, as opposed to strongly hook-shaped; absence of whisker-like cheek odontodes; absence of enlarged odontodes along snout margin; presence of a dorsal iris operculum; lips forming an oval disk; dentary and premaxillary with numerous viliform and bicuspid teeth; presence of a small buccal papilla, no enlarged dentary papilla; seven branched dorsal-fin rays; presence of an adipose fin; no membranous extension between end of dorsal fin and adipose fin; five series of lateral plates extending to caudal fin; lateral plates not keeled and not bearing enlarged odontodes; lateral plates of ventral series on caudal peduncle angular but not keeled; abdominal region entirely naked. Cryptancistrus is externally mostly similar to Guyanancistrus. It is distinguished from Guyanancistrus primarily by the Table 6. Comparision of DEC and DEC+j models using Likelihood Ratio Test (LRT). Max nb of areas: maximum number of areas allowed in ancestral geographic range. LnL: log likelihood of the ancestral reconstruction. Numparams: number of parameters included in the model. d: rate of range expansion ("dispersal"). e: rate of range contraction ("extinction"). j: weight of jump dispersal event ("founder effect"). D statistic: 2*ΔLnL. DF: number of free parameters. Integrative systematics of Guyanancistrus with description of a new genus and six new species fleshy posterior part of lateral margin of snout bearing slightly enlarged odontodes associated with small fleshy tentacules (vs plates along margin of snout forming a rigid armour covered with minute odontodes, absence of tentacules) (Fig 18E). It can additionally be distinguished from Gruyanancistrus by a skin region bordering the exposed portion of opercle roughly as large as the latter (vs distinctly narrower than the latter).

Parameter estimates LRT
Etymology Diagnosis. As given for genus. Description. Morphometric and meristic data of the holotype (only known specimen) in Table 5. Head and body dorsoventrally depressed. Dorsal profile gently convex from snout tip to orbit level, then nearly flat, slightly convex and sloped ventrally from dorsal-fin origin to adipose fin, then slightly concave to procurrent caudal-fin rays, and rising to caudal fin. Ventral profile flat from snout to base of caudal fin.
Dorsal contour of head smooth, no ridge or keel, inconspicuous rounded elevations on the midline of the snout and anterior to orbits, supraoccipital nearly flat. Dorsal margin gently flattened from base of first branched dorsal-fin ray to base of adipose fin between very slight ridges formed with lateral plates of dorsal series. First lateral plates of mid-ventral series forming low lateral ridge. Caudal peduncle roughly ovoid in cross section, flattened ventrally, and more compressed posteriorly.
Snout rounded anteriorly. Eye relatively large. Lips forming an oval disk, covered with short papillae. Presence of a single narrow buccal papilla. Lower lip wide, not reaching pectoral girdle, upper lip narrower. Very short maxillary barbel. Teeth bicuspid, lateral lobe about half size of medial lobe.
Head and body plated dorsally, plates generally covered by short and uniformly distributed odontodes. Snout plated except tip naked. Anterior margin of snout carrying slightly enlarged odontodes; meeting the latter, dorsolateral margin of upper lip supporting plates and short odontodes. Posterior part of lateral margin of snout forming a soft fleshy border bearing slightly enlarged odontodes with small tentacules sensu Sabaj et al. [102], cutaneous sheath surrounding base of odontodes being enlarged and partially detached from odontodes. Opercle supporting odontodes, those on inferior margin slightly enlarged. A large unplated area bordering posterodorsal margin of opercle. Evertible cheek plates with approximately 40 https://doi.org/10.1371/journal.pone.0189789.g020 enlarged odontodes, straight with tips curved, longest reaching beyond the end of opercle (on right side of the holotype; longest odontodes missing on left side). Usually three rows of plates and a curved nuchal plate between supraoccipital plate and dorsal-fin spinelet. Five series of lateral plates extending to caudal fin. Odontodes on lateral series of plates not forming keels. Odontodes on posterior part of pectoral-fin spine enlarged. Abdominal region totally naked. No platelike structure before the anal fin. Ventral part of caudal peduncle plated; presence of a small smooth area devoid of odontodes around anal fin.
Dorsal-fin origin slightly anterior to pelvic-fin origin. Dorsal fin relatively large; when adpressed, nearly reaching adipose fin. Adipose fin roughly triangular, preceded by a median unpaired raised plate. Adipose spine nearly straight, membrane posteriorly convex. Pectoralspine tip reaching approximately one-fifth of pelvic spine. Anal fin with weak spine, its margin convex. Caudal fin obliquely truncate, very slightly concave, inferior part longer (part of upper lobe damaged by holotype). Fin-ray formulae: dorsal II,7; pectoral I,6; pelvic i,5; anal i,5; caudal i,14, i.
Coloration. In alcohol, dorsal ground color of body medium brown, covered with yellowish-beige spots. Spots roundish, medium to large-sized, larger on posterior part than on anterior part, and non-coalescent. Ventrally, plated parts of body yellow-beige, abdomen whitish with some yellow-beige areas.
All fins of slightly darker color than body, and similarly spotted apart from anal. Spots of dorsal fin and paired fins medium sized; a dark spot on membrane between origin of spine and first branched ray. Spots of caudal fin larger, some of them coalescent, more or less forming two large light and irregular transverse bands; margin of fin light.
Distribution. Known from a single specimen from the Upper Parú de Oeste River (Fig 4), and collected along with Guyanancistrus teretirostris n. sp. and an unidentified Hypostomus species, and with two recently described Loricariinae, Cteniloricaria napova and Harttia tuna.
Etymology. The Latin name similis, meaning similar, refers to the strong morphological resemblance between the new species of Cryptancistrus and the type species of Guyanancistrus, G. brevispinis.

Biogeography of Guyanancistrus members
Comparison between likelihoods of ancestral area reconstructions along the phylogenetic tree using DEC and DEC + j models showed that the latter had significantly better fit ( Table 6). Resolutions of the different polytomies of the phylogenetic tree placed the population of Guyanancistrus brevispinis brevispinis of Tapanahony River in sister position to G. b. bifax and G. b. orientalis members, and split the population of G. b. brevispinis of Saramacca River in two subpopulations. Even though apparently contradictory to previous results, these resolutions did not impact ancestral area reconstructions, and reinforced the power of the analysis. The biogeographic analysis of Guyancistrus members under DEC + j model reconstructed a broad ancestral area comprising Amazonian headwaters including Upper Jari and Paru de Oeste rivers, the Oyapock, and Maroni rivers (Fig 20) at the root of the phylogenetic tree, even though this reconstruction was ambiguous (see pie charts of states probabilities in Fig 20). From this ancestral area, the G. niger and G. longispinis lineages split from all other ancestral Guyanancistrus by vicariance of the Oyapock Basin. Then a second vicariant event occurred between Amazonian headwaters and the Maroni Basin, splitting the G. megastictus, G. tenuis, G. teretirostris and G. brownsbergensis lineages from that of G. nassauensis and G. brevispinis. In the Amazonian group, two dispersals followed by speciation occurred. From an Amazonian ancestor (likely from Jari River), the ancestors of G. megastictus dispersed toward the Maroni Basin, whereas ancestors of G. brownsbergensis likely dispersed from the Paru de Oeste River toward the Saramacca River. In the Maroni group, two speciation events occurred, leading on one side to G. nassauensis and on the other to G. brevispinis. Dispersal patterns of G. brevispinis members appeared more complex, with multiple dispersals among Guianese rivers. From the central Maroni River, a first dispersal occurred to the west toward the Suriname River. Then, a second dispersal from headwaters of the Surinamese Maroni (i.e. Marowijne River) took place toward the headwaters of the Corantijn River, whereas part of the ancestral population of the Maroni River stayed in this basin (now present in the Tapanahony River). From the Upper Corantijn River, ancestral populations spread toward the Lower Corantijn, and then dispersed again to the East toward the Nickerie River, and from the latter to the Saramacca River. All these movements lead to the differentiation of the western form G. brevispinis brevispinis. A second dispersal, again from the Maroni, occurred to the east toward the Oyapock River. From there, ancestral populations successively dispersed to the west toward the Approuague and Comté-Orapu rivers, leading to the establishment of the present eastern form G. brevispinis orientalis. Finally, once more from the Maroni, ancestral populations dispersed toward the Upper Sinnamary River, and then toward the Mana River, and from the latter toward the Sinnamary leading to the establishment of the present Central form G. brevispinis bifax.

The genus Guyanancistrus
A general similarity of head and body shape, hardly elevated and describing smooth contours, seems to unite all Guyanancistrus species. However no shared characteristic was found to be unique for the group. Guyanancistrus longispinis and G. niger appear morphologically quite distinct compared with species of the brevispinis group. Even so, mitochondrial and nuclear sequences of the various species unambiguously showed Guyanancistrus to be monophyletic ( [49,50,110], present results), confirming the validity of the genus as it was originally separated from Lasiancistrus by its author (Isbrücker in [52]). Nevertheless, while the latter is well diagnosed [111], it is not very closely related to Guyanancistrus [50] (see also Armbruster [54]: 12, based on G. brevispinis). The sister clade of Guyanancistrus contained three genera: Hopliancistrus confirming results of Covain & Fisch-Muller [50], Corymbophanes as in Lujan et al. [49], and the new genus Cryptancistrus.
Hopliancistrus is globally similar to Guyanancistrus but clearly distinguished by the diagnostic presence of up to three strongly hook-shaped cheek odontodes and of enlarged odontodes on the sides of the snout. Corymbophanes is also easily distinguished from Guyanancistrus by the absence of evertible cheek odontodes, the absence of an adipose fin, and by the presence of an elongate postdorsal ridge of 13-17 raised unpaired platelets [112]. In contrast, Cryptancistrus is only distinguished from Guyanancistrus by the posterior part of lateral margin of its snout forming a soft fleshy border and bearing slightly enlarged odontodes with small tentacules ( [102]; see Fig 18). It is interesting to note that this organisation of odontodes on sides of snout is reminiscent of the condition observed in Hopliancistrus, one of its sister genera. These odontodes grow larger in large specimens of Hopliancistrus (even becoming stout at the corner of the snout in males), a condition that might be hypothesized for Cryptancistrus in the absence of material apart from the single holotype.
Despite the absence of obvious morphological characteristics for Guyanancistrus, a unique combination of external characters allows genus recognition with respect to its sister genera, and species assignation within the genus, was found for the diagnosis presented. Armbruster ([54]: 12) suggested that Guyanancistrus (which he placed in synonymy of Pseudancistrus) was not likely to be a monophyletic entity because of divergence of external characters between species. He particularly cited the development of « at least small hypertrophied odontodes on the snout » of G. niger (vs lack in the others), however these enlarged odontodes are present solely in two lateral tufts on the dorsolateral edges of the upper lip (Fig 7D), not on the plates outlining the snout contour as in the case of Pseudancistrus. Hopliancistrus tricornis also has two lateral regions of the snout with enlarged odontodes, that suggest similarity with G. niger [50]; again, however, they are not on the lip, but on the dorsolateral margin of the snout. Odontodes are present to a variable extent on the dorsal margin of the upper lip in several Hypostomines, including other Guyanancistrus species (Fig 7A, 7B and 7C) and Hopliancistrus tricornis, but they are usually small. Tufts of enlarged odontodes on the dorsolateral edge of the upper lip are thus characteristic of G. niger. The degree of evertibility of the cheek plates in G. brevispinis and G. niger was also found to be divergent by Armbruster (loc. cit.), but we saw no significant difference in this character between Guyanancistrus species.
Chaetostomus megacephalus, described by Günther in 1868 [113], is an additional but currently insufficiently known species that might also belong to Guyanancistrus (see Fig 1). Long considered as an Hemiancistrus species, it was moved to Pseudancistrus by Armbruster [53,54], who recently stated that this taxon needs further work [114] [115]). Indeed, the holotype has evertible cheek odontodes and no enlarged odontodes along the snout margin despite its large size (122.8 mm SL). Morphologically it is similar to G. longispinis group members, but is quite distinct from the nominal species (Fig 1). The type locality of C. megacephalus was indicated as "Surinam" in the original description, but Günther [116] later added that it was obtained from the collection of Dr. van Lidth de Jeude specifying that it was "probably from Surinam". Unfortunately we can only refer to the holotype. Specimens collected in the Essequibo River Basin in British Guiana described by Eigenmann ([117]: 231) as Hemiancistrus megacephalus appear more likely a distinct and probably new species [20] morphologically very close to G. longispinis (Fig 1). Despite extensive field collecting in Suriname, we have been unable to find any additional specimens (as well as other teams in Guyana; J. W. Armbruster, pers. com.).
The case is similar for Chaetostomus macrops Lütken, 1874, also known from a single specimen from "aquis Surinamensibus", and considered a synonym of megacephalus from the early 20 th century [117] until recently [114,118]. It has a particularly wide and elevated orbital rim reminiscent of that observed in Hemiancistrus medians, from which, however, it is easily distinguished by the absence of keeled and rough-toothed trunk plates [20] and by the presence of odontodes over a broad area on the opercle [114]. Morphologically, C. macrops is most similar to the species collected in the Potaro River by Eigenmann than to P. megacephalus (see S1 Fig) and could correspond to a distinct species. The collection of fresh material is essential before making taxonomic decisions concerning these species.

The G. brevispinis complex
The integrative taxonomy methodology reviewed in Padial et al. [25] was sufficient to congruently discriminate eight species of Guyanancistrus, including five new species, and a new genus of the Loricariidae (see Table 2). However, G. brevispinis could not be significantly distinguished from all other species using the different approaches except phylogeny alone, and its different subspecies could not be clearly delineated regardless of the method employed (morphometry, DNA barcodes, phylogeny, or distribution). In this context, use of the multitable approach integrating all available information was particularly suitable. This method allowed the evaluation of the amount of common information present in the different datasets and its significance through RV tests, and demonstrated that half of the variation recorded in the different tables was significantly linked. Indeed, the unifying structure provided by the MCOA simultaneously revealed significant covariations between morphometric characteristics, phylogenetic structure, and distributional patterns in all available populations of G. brevispinis, and clearly highlighted three groups of infraspecific rank. Surprisingly, the presence of two lineages of G. brevispinis within the Maroni Basin was revealed. One lineage included all populations of the eastern Maroni grouped with other populations of G. b. bifax, and the second was located in the Upper Tapanohony River, a western tributary of the Maroni River, grouped with the populations of G. b. brevispinis close to populations from Upper Suriname and Corantijn rivers. This unexpected result highlights the central role played by the Maroni Basin in the distributional pattern of G. brevispinis members, with an east-west partition of this drainage. This central role was also confirmed by the biogeographic reconstruction, which resolved no fewer than five successive dispersal events originating from the Maroni Basin toward other drainages of the Eastern Guianas. Two of them concerned ancestral populations of G. brevispinis which dispersed to the west toward the Upper Suriname and Corantijn rivers respectively (leading to establishment of the future G. b. brevispinis), a third eastward toward the Upper Oyapock (future G.b. orientalis), and a fourth and fifth toward the Sinnamary and Mana respectively (future G.b. bifax), but with the persistence of two distinct lineages within the Maroni Basin.
These results only partially corroborate the findings of Cardoso and Montoya-Burgos [60] who recovered five lineages (instead of three) among G. brevispinis including distinct lineages from: (1) Oyapock-Comté-Approuague basins, (2) Maroni-Mana-Sinnamary basins, (3) Suriname River, (4) Corantijn River, and (5) Nickerie River. However, genetic distances between the Nickerie and Suriname rivers' representatives in their phylogenetic tree were not markedly greater than those within their Maroni-Mana-Sinnamary lineage (our G. b. bifax), such as the representatives of the Sinnamary and Mana basins. These authors also highlighted the central role played by the Maroni Basin as the gateway of ancestors of G. brevispinis from the Amazon Basin. Using phylogenetic topological tests, they hypothesized a single entrance from headwaters of the Maroni River followed by a first westward dispersal, assuming a stepping-stone pattern of dispersal. Then dispersal strategies were evaluated based on haplotypic diversity and genetic-geographic structure comparisons between populations here described as G. b. bifax, revealing favoured dispersal routes through coalescing river mouths during low sea level periods. If the entrance of ancestral forms of Guyanancistrus originating from the Amazon Basin in the Maroni River is confirmed by the present study, the reconstructed dispersal pattern of G. brevispinis members is much more complex than a simple stepping-stone process, probably related to river capture events (direct dispersal from the Maroni to the Corantijn and Oyapock rivers). In addition, Cardoso and Montoya-Burgos [60] reported a single Amazonas lineage which has been shown in the present study to contain three distinct species: G. teretirostris from Paru de Oeste River (Pb.BR652, Pb.BR653, and Pb.BR654 in Cardoso and Montoya-Burgos, 2009 [60]), G. tenuis from Jari River (Pb.MIT03, and Pb.MIT04 in Cardoso and Montoya-Burgos, 2009 [60]), and G. megastictus from Maroni River (Pb.MIT02 in Cardoso and Montoya-Burgos, 2009 [60]), leading to two dispersal events from the Amazonian tributaries toward the Maroni Basin for their study. The present study also revealed a third dispersal between Paru de Oeste and Saramacca rivers. All these dispersals resulted in speciation within the Eastern Guianas, with the particularity of the Maroni River hosting three newly formed species; two hyperendemics restricted to montaneous areas (G. nassauensis and G. megastictus), and one widely distributed (G. brevispinis) and comprising two subspecies (G. b. brevispinis and G. b. bifax). The Maroni Basin was thus a center of speciation for Guyanancistrus members resulting in increased local endemicity, as well as a source of dispersal to other drainages of the Eastern Guianas.
Cardoso and Montoya-Burgos [60] tentavely provided diagnostic characters to distinguish their different lineages, but most of them relyed on global estimates of shape and color patterns. Conversely, the MCOA used here, by unifying different variables contained in different data sets within the same analysis, allowed the extraction of diagnostic characteristics for each group. Moreover, the ability to include phylogeny with the other data sets allowed the interpretation of covariations between morphometric, phylogenetic, and distributional variables in an evolutionary perspective. Evolution of shape of G. brevispinis members was thereby linked to genetic and geographic divergences. Guyanancistrus brevispinis bifax evolved a mean shape intermediary between G. b. brevispinis to the west and G. b. orientalis to the east, characterized by more numerous plates on the caudal peduncle, which was also less deep in this subspecies. This result contrasted with that of Cardoso and Montoya-Burgos [60], who characterized the same group as having the highest body shape. Guyanancistrus b. brevispinis evolved a broader head and anterior body in the west whereas G. b. orientalis evolved a slender appearance with a longer caudal peduncle and more numerous plates along the body in the east, a result in general agreement with Cardoso and Montoya-Burgos [60]. It is interesting to note that G. b. bifax possesses two morphs (see Fig 10), both present in the whole area of distribution of the subspecies. Even though not statistically supported, morphotypes with broad mouth (identified by the letters BM in Fig 1) were all placed closer to G. nassauensis in the morphometric analysis, and appeared clearly distinct from other morphotypes with a normal mouth. Given that G. nassauensis was introgressed by G. brevispinis (see Fig 2), implying hybridization between these two co-occuring (at least in the Nassau Mountains) sister species, this morphological characteristic may result from retention of genes from G. nassauensis in the genome of G. b. bifax.
Color patterns also appeared highly variable in G. brevispinis. Polychromatism is not rare in fish and can be related to sexual selection driving the appearance of strong sexual dimorphism ( [119][120][121][122], reviewed in [123]). Numerous genera of the Loricariidae exibit strong sexual dimorphism through the development of hypertrophied odontodes (e.g. in Peckoltia, Panaque, Neblinichthys, Sturisoma, Farlowella, Spatuloricaria, Rineloricaria), development of fleshy tentacles on snout (e.g. in Ancistrus), lip enlargement (e.g. in Loricariichthys, Hemiodontichthys), or teeth characteristics (e.g. in Loricaria), but dichromatism has not previously been reported, even though several genera display colorful patterns (e.g. Pseudacanthicus, Leporacanthicus, Hypancistrus, Scobinancistrus, Peckoltia, Panaqolus). Color variations in Loricariidae appear sex-independent, and rather related to natural selection (e.g. for camouflage over the substrate) or to random drift. However, such variations in G. brevispinis, with the appearance of very diverse patterns ranging from spots to marbling and reticulations imply rather relaxed selective constraints acting on phenotypes since different patterns can be observed within the same basin. Alternatively similar patterns can be observed between distant basins implying multiple convergent evolution and/or retention of ancestral patterns among populations. Cardoso and Montoya-Burgos [60] tentatively classified their different lineages using pattern characteristics, but if this criterion applied for a given population, it often failed to characterize other populations of the same basin or equally applied to other populations of a distinct lineage (see Fig 8). For example, Cardoso and Montoya-Burgos [60] distinguished their lineage from the Corantijn River (Sipaliwini River; Fig 8A and 8B), from all other populations by the head having small light vermiform marks and the body faint parallel light bands becoming highly visible on the caudal peduncle. However, Fig 8C, corresponding to another population from the Corantijn Basin (Kabalebo River), shows that this population was particularly dark, without such obvious markings. The same situation occurred with the Comté-Approuague-Oyapock lineage (G. b. orientalis), which was supposed to be distinguished from all other lineages by the head having small light dots and the body faint, parallel, light bands becoming highly visible on the caudal peduncle; the specimen from the Comté-Orapu Basin (Fig 8I) and the one from the Oyapock River (Fig 8J), are clearly distinct from each other. Moreover, the characteristics of the Oyapock population better reflected the definition provided for the Maroni-Mana-Sinnamary lineage (G. b. bifax), theoretically distinguished by large light dots, irregular in shape [60]. Given the high variability of the species, color patterns do not appear to be relevant for identification purposes.

Notes on the ecology of Guyanancistrus species
Guyanancistrus brevispinis occurs in the lowland rivers and large tributaries in the interior of Suriname and French Guiana (i.e. upstream of the most downstream rapids), mainly in strong currents in or immediately downstream of rapids. During the day adult G. brevispinis were observed foraging on the algal biofilm on boulders and bedrock in the Middle Suriname River together with Cteniloricaria platystoma and Harttia surinamensis. Adult G. brevispinis are well camouflaged when feeding on these substrates. Juveniles of G. brevispinis were collected in a mountain stream (400 m above mean sea level) in Lely Mountains [56] with cool (23.3˚C), clear (Secchi disc visibility 150 cm) water with low conductivity (24 μS cm -1 ) and neutral pH of 7.5. Postlarvae of G. brevispinis (15 mm TL) were collected in a headwater tributary of the Upper Palumeu River in a deep (> 1 m) pool under a 60-m high waterfall (Fig 8.1. in Mol & Wan Tong You [124]); the water was cool (23.5˚C), clear (turbidity 5 NTU), slightly acidic (pH 5.9) with low conductivity (20 μS cm -1 ), low alkalinity (4.75 mg CaCO 3 L -1 ) and some tannins (2.6 mg L -1 ) [125].
G. brevispinis has the largest distribution within Eastern Guianas, with an area of distribution ranging from the Corantijn River in western Suriname to the Oyapock River in eastern French Guiana (Fig 4). Alternatively, most of the other Guianese species (the Amazonian species are insufficiently known) appear highly restricted to mountains, having a similar distributional pattern to Harttiella [19,55,56], a group of hyperendemic dwarf loricariids restricted to mountainous forest creeks. At least two species of Guyanancistrus (G. nassauensis and G. brownsbergensis) have developed adaptations to this kind of biotope (small streams, cool water temperature, low productivity. . .) including dwarfism.
Guyanancistrus nassauensis and G. brownsbergensis are each known from a single mountain stream, in the Nassau Mountains (Paramaka Creek) and Brownsberg Mountains (Kumbu Creek), respectively. With this very restricted distribution (< 20x20 km 2 ) both species can be considered hyperendemics and currently the two species are threatened with extinction by proposed and ongoing mining activities.
In Paramaka Creek, Guyanancistrus nassauensis occurs syntopically with juvenile Guyanancistrus brevispinis and with Harttiella crassicauda, a second endemic species from the Nassau Mountains. However, G. nassauensis occurs both on the plateau in perennial flowing headwaters and in the upper mainstem of Paramaka Creek (lower slopes of the plateau; altitude range 120-530 m amsl), whereas H. crassicauda only occurs on the plateau proper (230-530 m amsl; [126]). In the IJs Creek tributary of Paramaka Creek on the Nassau plateau (467 m amsl) both G. nassauensis and H. crassicauda occur in cool (22.6˚C), shallow (40 cm water depth), clear (Secchi transparency > 40 cm) water with low conductivity (28 μS cm -1 ), neutral pH of 7, low inorganic N (0.067-0.120 mg L -1 ), relatively high organic N (0.307-0.592 mg L -1 ), low total P (0.002-0.010 mg L -1 ) and high organic C (2.916-4.972 mg L -1 ) [56]. The bottom substrate is gravel with boulders and bedrock (with the red filamentous algae Batrachospermum sp. attached to it) and near the edge of the plateau in slightly deeper water (approximately 50 cm) stands of the emergent macrophyte Thurnia sphaerocephala occur [56]. In the upper mainstem of Paramaka Creek, as well as in some upstream branches on the plateau, G. nassauensis occurs syntopically with G. brevispinis [126].
Guyanancistrus brownsbergensis was collected only in the upper reaches of Kumbu Creek on the Brownsberg bauxite plateau at an altitude of 200-430 m amsl. The habitat of G. brownsbergensis seems very similar to that of G. nassauensis in the Upper Paramaka Creek on the Nassau bauxite plateau. The upper Kumbu Creek is a small (2.5-3.7 m wide, 26-52 cm deep) mountain stream with moderate to strong flow (0.3-0.56 m s -1 ) and cool (23.1-23.2˚C), mostly clear water with high dissolved oxygen content (7.1-7.7. mg L -1 ), neutral pH of 7-7.5 and low electrolyte content (30.8-31.6 μS cm -1 ). The bottom substrate is mainly gravel, boulders and bedrock. We observed no aquatic vegetation in the stream, but overhanging terrestrial vegetation, submersed root masses, woody debris, leaf litter and rock crevices offered ample hiding places for G. brownsbergensis. During the day, adult G. brownsbergensis were observed on several occasions throughout the year resting in moderate current in front of a rock crevice in a relatively deep (50 cm) pool upstream of the 50-m high Kumbu Falls.
We have no information on the ecology of G. teretirostris, C. similis, G. tenuis and G. megastictus, although the latter two species apparently also occur in small mountain streams, perhaps comparable to the habitat of G. nassauensis and G. brownsbergensis.
Other species are larger, particularly when they inhabit the main stream of rivers as does G. niger, the largest species of Guyanancistrus, which lives in the rapids of the Oyapock River along with Pseudancistrus barbatus. Guyanancistrus niger is much less abundant than other species, and we collected only adult specimens, suggesting that adults and juveniles may not be syntopic. Guyanancistrus brevispinis and G. longispinis were collected together in the Oyapock drainage, but while G. brevispinis seems to prefer small forest streams, G. longispinis was more often found in the main channel, on the rocky bottom of riffles, where it can be relatively abundant ( [59]; personal observations).
Supporting information S1 Text. Supplementary material examined for this study. Species are listed in alphabetical order, followed by country, river basin, catalog number, number of specimens examined in the lot, locality, collector, and date of sampling. Specimens included in morphometric analyses are indicated by an asterisk followed by number when needed. (DOCX) S1 Table. Raw morphometric and meristic data for the 269 specimens analysed in the morphometric study. Measurements are in mm. Variables labelled as in Tables 4 and 5, and spe Tables 4 and 5. c: eigenvalues of the BGA. C. macrops appeared closer to members of the G. longispinis group, and particularly the species of Guyanancistrus from Potaro River (GspPotaro) identified as 'Pseudancistrus' megacephalus by Eigenmann in 1912. 'P.' megacephalus (positive score on axis 2) also appeared distinct from C. macrops (negative score on axis 2), and both of them distinct from H. medians, type species of Hemiancistrus. (TIF)