A new cryptic species of Pithecopus (Anura, Phyllomedusidae) in north-eastern Brazil

The genus of Neotropical frogs Pithecopus includes 11 species occurring east of the Andes from southern Venezuela to northern Argentina. Recent genetic approaches pointed out an unusual genetic diversity among populations from localities in north-eastern Brazil recognized as P. nordestinus. In fact, one of these studies confirmed the hypothesis that the São Francisco River acted as an effective geographical barrier during vicariant events in the evolutionary history of P. nordestinus, resulting in two principal, highly divergent clades. Herein we formally describe this divergent clade as a new cryptic species of Pithecopus from north-eastern Brazil, the sister clade of P. nordestinus. It differs from other species of Pithecopus, except for P. azureus and P. nordestinus, by its small body size, lack of the reticulate pattern on flanks, smaller head width, and advertisement calls generally composed of a threepulsed core. ANDRADE F.S. et al., A new cryptic species of Pithecopus 109

For comparisons with the most closely related species (the lowland species of Pithecopus), we used the same dataset recently used to describe P. araguaius (Haga et al. 2017a): 21 types (holotype + 20 paratopotypes) and 31 non-type specimens of P. nordestinus from Alagoinhas (Bahia), Areia Branca (Sergipe) and Laranjeiras (Sergipe); 18 types (holotype + 17 paratopotypes) and 18 non-type specimens of P. araguaius from Chapada dos Guimarães (Mato Grosso) and Santa Terezinha (Mato Grosso); 13 topotypes from Asunción, Paraguay, one non-type specimen from Corrientes, Argentina, and 11 nontype specimens from Bela Vista (Mato Grosso do Sul) (the nearest acoustic sample from its type-locality) of P. azureus. Further details on examined specimens are in the Appendix.

Bioacoustics
Seventy-three advertisement calls of six males of the new species were recorded between 19:11 and 22:07 on four field trips in May 2011 (days 7, 17, 22 and 24; temperature of air 28-29°C and water ANDRADE F.S. et al., A new cryptic species of Pithecopus 111 25-26°C). Recordings were obtained with a Marantz PMD 222 cassette recorder, coupled to an Audio-Technica AT835b directional microphone. Recordings were digitised at 48 kHz and 24 bit resolution.
For acoustic comparisons, we also used the dataset of Haga et al. (2017a), supplemented with three recordings of P. nordestinus (FNJV 31155, FNJV 45456, FNJV 45468) that were deposited in the Fonoteca Neotropical Jacques Vielliard (FNJV). Therefore, the complete comparative dataset includes advertisement calls from four males of P. azureus (Haga et al. 2017b); 105 calls from six males of P. nordestinus (from the municipalities of Areia Branca, Itabaiana and Maruim, Sergipe; Gandu and Igrapiúna, Bahia); 24 calls from seven males of the type-series of P. araguaius; 212 calls of 41 males of P. hypochondrialis (from the municipalities of Serra do Navio, Amapá; Barra do Garças, Mato Grosso; Brasília, Distrito Federal; Araguari and Uberlândia, Minas Gerais; Padre Bernardo, Pirenópolis and Uruaçu, Goiás). Call vouchers are listed in Supplementary file 1. We reanalysed two males and 34 calls of P. nordestinus (FNJV 45456,FNJV 45468) that were used in Vilaça et al. (2011). As different authors may analyze calls differently and in order to ensure the reliability of our results, we have based the acoustic comparisons only on our own data.
Sound files are deposited in the Arquivo Sonoro da Coleção de Anuros,UFU, and FNJV, UNICAMP. Call descriptions and the acoustic terminology follow Köhler et al. (2017) and Haga et al. (2017a: appendix B -S1 table: Acoustic terminology employed for the species of Pithecopus). According to Haga et al. (2017a), the calls of some species of Pithecopus often have a few low-amplitude isolated pulses at the end (more common) or at the beginning (unusual). We used the main (high-amplitude) group of pulses of each call (hereafter called ʻcore portionʼ) of the new species, P. araguaius, P. hypochondrialis, P. azureus and P. nordestinus to compare call traits equivalently among these species. Analysed sound files are listed in Supplementary file 1.
We calculated means and standard deviations (SD) for each individual and then the overall mean and SD was calculated based on those values, whereas the range encompassed the minimum and maximum values for the whole sample. We analysed calls using Raven Pro 1.5, 64 bit version (Bioacoustics Research Program 2014) with the following settings: window type = Hann, window size = 256 samples, 3 dB filter bandwidth = 270 Hz, brightness = 50%, contrast = 50%, overlap = 85% (locked), colour map = "Cool", DFT size = 1024 samples (locked), grid spacing (spectral resolution) = 46.9 Hz. We analysed temporal traits in oscillograms and spectral traits in spectrograms (Köhler et al. 2017); we used the 'Peak Frequency' function to determine the peak of the dominant frequency. We generated sound figures through Seewave ver. 1.6 package (Sueur et al. 2008) and tuneR package (Ligges et al. 2018), R environment (ver. 3.3.1) (R Core Team 2017). The Seewave settings for the spectrograms were: Hanning window, 85% overlap, and 256 points resolution (FFT).

Phylogenetic tree and species-delimitation tests
The phylogenetic analyses were based on the Bayesian inference (BI) method. The dataset was composed of 149 terminals with 1045 characters from mitochondrial 16S ribosomal sequences, 892 characters from NADH dehydrogenase subunit 2 (ND2) sequences and 422 characters from Seven in Absentia Homolog 1 (Siah) sequences. We included the individuals assigned to P. nordestinus from the north and the south clades according to the dataset and topology of Bruschi et al. (2019) (see Supplementary file 2). To best evaluate the phylogenetic relationship status of the new species in comparison with sister species / taxa from the genus, we included sequences of at least three individuals (when available) of each of the species of Pithecopus deposited in GenBank (https://www.ncbi.nlm.nih.gov/genbank -see list of specimens in Supplementary file 2). Callimedusa tomopterna (Cope, 1868) was used as an outgroup because, according to Duellman et al. (2016), it represents a sister group of the clade which comprises the Pithecopus genus.
Bayesian inference was based on a Markov Monte Carlo (MCMC) analysis conducted on MrBayes 3:2.7 (Huelsenbeck & Ronquist 2001) with two independent runs, each run with four chains and sampling every 1000 generations for 20 million generations. MEGA X 10.1.8 (Kumar et al. 2018) selected the best evolutionary model through Akaike Information Criteria (AIC) (16S: GTR + G; ND2: HKY + I + G; Siah: K2P). The first 25% of the trees were excluded as burn-in. To confirm the quality of the parameters of the BI, Tracer software ver. 1.7.1 (Rambaut et al. 2018) was used, and only Effective Sample Sizes (ESS) values over 200 were considered acceptable.
We performed the Generalized Mixed Yule-Coalescent (GMYC) (Pons et al. 2006;Fontaneto et al. 2007) and the Poisson Tree Process (bPTP) (Zhang et al. 2013) methods to test species boundaries. The GMYC analysis requires an ultrametric tree that was estimated using a lognormal uncorrelated relaxed molecular clock model and coalescent constant size process, performed using BEAST ver. 1.10.4 (Drummond et al. 2012), with 20 million generations and sampling every 10 000 generations. The trees were estimated based on the HKY substitution model. We used the software Tracer ver. 1.7.1 (Rambaut et al. 2018) to check log files and all model parameters reached an ESS > 200. After removing 10% of the trees as part of the burn-in procedure, the remaining trees were summarized in a single tree with the Tree Annotator ver. 1.10.4 (Drummond et al. 2012). The GMYC analysis was carried out in R 3.6.1 software environment and performed with the packages Species Limits by Threshold Statistics (ʻsplitsʼ) (Ezard et al. 2014) and Analyses of Phylogenetics and Evolution in R Language (ʻapeʼ) (Paradis et al. 2018) using a single threshold and standard parameters [interval = c(0, 10)].
The bPTP does not require an ultrametric tree and uses the number of substitutions to model the speciation and coalescent events. For the analysis, the best maximum-likelihood (ML) tree reconstructed with RAxML (Kozlov et al. 2019) using an HKY model of nucleotide substitution selected with MEGA X 10.1.8 was used as input. The bPTP was performed on the web server (http://species.h-its.org/) for 5 × 105 generations, with a thinning value of 1000 and a burn-in of 10%.

Statistical analysis
Considering the morphological and acoustic (multivariate) datasets separately, we sought for discrimination between species by applying two functions in R environment: (1) randomForest (RF) (radomForest ver. 4.6-12 package) (Liaw & Wiener 2002) and (2) (Haga et al. 2017a). RandomForest algorithm (machine learning) constructs many (e.g., 1000) classification trees using bootstrap samples of the data (each split using the best predictors among those randomly chosen at each node) then generating classifiers and aggregating results by voting to classes (Liaw & Wiener 2002). The classic Discriminant Analysis (DA) depends on multivariate normality (Pohar et al. 2004) and on a larger number of objects than variables. The multivariate normality of the original data was evaluated through the function mardiaTest (MVN package; Korkmaz et al. 2014). The DAPC performs analyses on the Principal Component scores (Jombart 2008;Jombart & Ahmed 2011). The application of a DA on a few axes (preserving about 95% of the variance) of a Principal Component Analysis, as performed by DAPC, reduces the sample-variable imbalance problems between objects and traits (Jombart et al. 2010).
For the multivariate analysis and statistical tests, we used all the morphometric features detailed above. For the acoustic analyses, we used: call duration (entire call), number of pulses, pulse duration, inter-pulse interval within core, core duration (main stronger group of pulses), pulses per core, pulses per second, isolated pulse and peak of dominant frequency. Because both multivariate analyses, to both datasets, were concordant in species discrimination (see Results), we present the results of RF classification only in tables and the results of DAPC only in scatter plots.
The acoustic and morphometric traits were tested for statistical significance of the differences among species through the Exact Wilcoxon-Mann-Whitney Rank Sum Test, function wilcox_test of the package Coin (resampling statistics model; Hothorn et al. 2008) in R environment. As these tests were done between species / populations pairs, the significance levels (ʻPʼ) were adjusted considering the number of pairings through the P correction method of Holm (p.adjust function of the base package in R environment; see Chen et al. 2017). We considered significance when P ≤ 0.01. (4) fingers and toes long and slender with terminal discs poorly developed; and (5) grasping (opposable to the others) finger I and toe I. Pithecopus gonzagai sp. nov. differs from the highland species of Pithecopus by the (6) lack of the reticulate pattern on flanks, and (7) head width smaller than 11.2 mm.

Etymology
The specific name honours Luiz Gonzaga do Nascimento, better known as Luiz Gonzaga. He was a Brazilian singer, songwriter, musician, poet and one of the most influential figures of Brazilian popular music in the twentieth century. Luiz Gonzaga has been credited for presenting the rich universe of northeastern musical genres to the rest of the country. He was born and raised in the municipality of Exu, state of Pernambuco, Brazil. Pithecopus gonzagai sp. nov. also occurs in the state of Pernambuco, which is equally its type locality. Paratopotypes BRAZIL • 23 ♂♂; same collection data as for holotype ;ZUEC 19664 to 19668, 19670 to 19684, 19686 to 19688 • 4 ♀♀; same collection data as for holotype ;ZUEC 19661 to 19663, 19669. Other material examined BRAZIL -Pernambuco • Limoeiro;ZUEC 19661 to 19688 • Bom Conselho;ZUEC 19617, 19619, 19622 to 19625, 19628 to 19633 • Caruaru;ZUEC 19610 to 19616 • Poção;ZUEC 19638, 19640 to 19644, 19646, 19649 to 19650, 19654 • Recife;ZUEC 19655 to 19660. -Alagoas • Pilar; ZUEC 19573,

Description Holotype
General aspect slender ( Fig. 2A-B); snout truncate in dorsal and lateral views ( Fig. 2C-D). Head wider than long; loreal region slightly concave; canthus rostralis rounded, smooth; nostrils small, subcanthal, placed latero-frontally, closer to snout tip than to eyes; internarial distance longer than eye-nostril distance and tympanum diameter, but smaller than eye diameter; eyes latero-frontally positioned; tympanum nearly circular, with annuli undefined at superior border; tympanum diameter less than half of eye diameter; supratympanic dermal fold present, beginning on right side of tympanum and ending near insertion of arm; dorsolateral macrogland indistinct; no external vocal sac; tongue nearly ovoid, free posteriorly, longer than wide, without pigmentation on base; vomerine teeth absent; choanae small, located laterally, slightly rounded. Upper arm thin and forearm robust; no finger webbing; comparative finger length when adpressed I < II < IV < III (Fig. 2E); finger discs poorly developed; finger I enlarged at base; nuptial asperity covering most of dorsal surface of finger I, except tip; palmar tubercles poorly developed, subarticular tubercles developed, barely distinguishable from adjacent supernumerary tubercles (Fig. 2E); inner and outer metacarpal tubercles poorly developed; comparative toe length when adpressed II < III < I < V < IV (Fig. 2F); toe webbing absent; plantar callosities poorly developed, inner and outer metatarsal tubercles poorly developed; subarticular tubercle developed, single and rounded; supernumerary tubercles rounded and poorly developed (Fig. 2F); legs slender; dorsal skin smooth; ventral skin granulated on belly, throat and thigh, smooth on tibia, tarsus and foot. Chest and right thigh slightly damaged ventrally due to tissue sampling. Cloacal region moderately granulated.

Colouration in life
Dorsum of head and body green. Vertebral line on back present in some individuals (Fig. 3C), but absent in all individuals of the type series. Loreal region and eyelids green. Eyes outlined with thin white line. Edge of jaw bordered by black line. Green dorsal region delimited by white dorsolateral line extending from mouth end until around middle of axilla-groin length. Same white line also observed in distal parts of forelimbs, extending from elbow to end of finger IV. Black dorsolateral line (below white line) delimits ventral region; same line also found on forearms, tibia and foot portion. Anterior and posterior surfaces of upper arm, thigh and tibia, extending partially onto foot, orange with well-defined vertical black stripes. Forearm anterior surface extending to hand coloured orange with well-defined vertical black stripes. Orange inguinal region with well-defined vertical black stripes. Green dorsal surface extending onto all limbs: arm, forearm, thigh, knee and tibia; green dorsolateral surface of tarsus and foot ( Fig. 3A-C).
The new species is distinguished from P. rusticus by the absence of the slightly reticulated pattern on the palpebral membrane and the throat region (pattern unique to P. rusticus; Bruschi et al. 2014).
From its closer relatives (the lowland species), Pithecopus gonzagai sp. nov. is distinguished by being smaller than P. nordestinus, P. azureus and P. hypochondrialis in SVL, head width and length of axillagroin, internarial distance, and upper arm, hand, thigh, tibia and foot lengths (Exact Wilcoxon-Mann-Whitney Test: P < 0.01; see Table 3). In addition, based on P values obtained from the Exact Wilcoxon-Mann-Whitney Test (Table 3), P. gonzagai sp. nov. is smaller than P. nordestinus in eye-nostril distance (1.8-2.8 mm in the new species vs 2.0-3.1 mm in P. nordestinus), forearm length (6.3-8.8 mm in the new species vs 6.9-9.3 mm in P. nordestinus) and tarsus length (7.9-10.5 mm in the new species vs 8.8-11.1 mm in P. nordestinus). The new species is larger than P. araguaius in SVL, head length, head width, axilla-groin and eye-nostril lengths, and upper arm, hand, thigh and foot lengths (see Table 3). The randomForest model on morphometric traits classified 90% of the males of the new species correctly (Table 4). It was the species with the highest percentage of males classified correctly in comparison with the other four taxa. DAPC based on morphological traits yielded no noticeable discrimination among the new species and P. azureus, P. nordestinus and P. araguaius (Fig. 4A). However, it is possible to notice slight discrimination between the new species and P. hypochondrialis (Fig. 4A), with a greater separation along axis 1 (LD1 = 58%; axis x), but the axis 2 (LD2 = 25%; axis y) also contributes to the separation. Tarsus length (17%), head width (12%), tibia (12%) and foot (11%) lengths mainly accounted for species separation along LD1 (Fig. 4A); while tarsus (25%) and hand (16%) lengths, eye-  (Caramaschi, 2006). Values presented as mean ± SD (range). * = mean of two isolated pulses; ** = mean of two inter-pulse intervals.
Morphometric measurements of all examined specimens and the eigenvectors of the DAPC based on morphometric traits are in Supplementary files 3 and 4, respectively.
In comparison with the lowland species, the new species differs from P. palliatus by having the advertisement call with a higher dominant frequency (above 1900 Hz in P. gonzagai sp. nov. vs 1580 Hz in P. palliatus) and with a single note (= core portion) (double notes in P. palliatus; Köhler & Lötters 1999 We were unable to find qualitative or quantitative diagnostic acoustic characters (absence of overlaps) between the new species, P. nordestinus and P. azureus. The randomForest model on acoustic traits was unable to distinguish one male of the new species from those of P. hypochondrialis (  other five males were correctly classified. All six males of P. nordestinus were correctly classified. The univariate analyses did not recover differences between calls of the new species, P. azureus and P. nordestinus. In addition, there are overlaps in the acoustic traits of these three species. Among the lowland species, P. araguaius is the more acoustically distinct species (Fig. 4B). The greater separation among lowland species was along axis 1 (LD1 = 91%), while axis 2 contributed much less to the separation (LD2 = 7%). Pulses per core (30%), core duration (22%), isolated pulse (21%) and number of pulses per call (18%) mainly accounted for species separation along axis 1 (Fig. 4B). Eigenvectors of the DAPC based on acoustic traits are in Supplementary file 5.

Phylogenetic inferences and species delimitation test
Bayesian inference was congruent with the phylogenetic intrageneric relations recovered by Faivovich et al. (2010), Duellman et al. (2016) and Haga et al. (2017a). In our topologies, we recovered two wellsupported divergent clades (North and South) traditionally assigned to P. nordestinus, coincident with the phylogeographic breaks reported by Bruschi et al. (2019). The paratopotypes of P. gonzagai sp. nov. were recovered nested all with populations from north of SFR while the specimens of P. nordestinus were assembled with populations from south of SFR (Fig. 5, Supplementary file 6). The P. gonzagai sp. nov. + P. nordestinus are sister-group of the P. azureus (Fig. 5, Supplementary file 6).
The single-locus species discovery strategies by GMYC and bPTP approaches were performed in a dataset composed of 148 sequences consisting of 1049 nucleotides from mitochondrial DNA 16S gene, 134 sequences consisting of 892 nucleotides from ND2 and 73 sequences consisting of 422 nucleotides from Siah. Both the GMYC and bPTP identified within the stringent threshold two taxonomic entities, congruent with two main lineages recovered by Bayesian inference (see Fig. 5, Supplementary files 6 , 7, 8, 9).
Our results recovered same species clusters as shown by Bruschi et al. (2019) in a multi-locus study. . Because the differentiation among samples of P. nordestinus and P. gonzagai sp. nov. based on mitochondrial markers revealed high genetic distances level, the GMYC model suggested 52 entities composed of 32 distinct clusters for 16S, three entities composed of three clusters for ND2 and 11 entities composed of ten clusters for Siah, however, with the exception of two individuals from Areia Branca, no mixture between samples of P. nordestinus and P. gonzagai sp. nov. was recovered in our inferences.  Table 4. Confusion matrix for five of the lowland species of Pithecopus Cope, 1866 based on morphometric and acoustic (values in bold) datasets by means of a randomForest model. Settings: number of tree permutations = 1000; number of variables tried at each split = 3; error rates = 25.5 % | 10.9 %. Fig. 4. Scatterplot of the discriminant analyses of principal components (DAPC) on the morphometric and acoustic datasets of Pithecopus araguaius Haga et al., 2017, P. azureus (Cope, 1862, P. gonzagai sp. nov., P. hypochondrialis (Daudin, 1800) and P. nordestinus (Caramaschi, 2006). A. The two first axes on the morphometric data (11 first PCs, 95% retained variance). Variance explained by the axes: LD1 = 58% (F-statistic = 131.7) and LD2 = 25% (F-statistic = 57.3). B. The two first axes on the acoustic data (5 first PCs, 96% retained variance). LD1 = 91% (F-statistic = 182.0) and LD2 = 7% (F-statistic = 13.5). Fig. 5. Phylogenetic relationships of Pithecopus Cope, 1866 based on 16S rDNA mitochondrial fragment. Topology inferred from the Bayesian inference based on the GTR+G model. Posterior probabilities are shown at each node. Scale bar represents the number of substitutions per site. Expanded topology for P. gonzagai sp. nov. and P. nordestinus (Caramaschi, 2006) is in Supplementary file 6.
The tree resulting from the bPTP analysis recovered two species from the best ML search, completely separating individuals of the south (P. nordestinus) from individuals of the north (Pithecopus gonzagai sp. nov.). The highest Bayesian solution was similar to the ML tree; however, a total of 118 (16S), 85 (ND2) and 54 (Siah) putative entities were recovered by a simple heuristic search. In contrast, a mass sampling can contain individuals from real isolated clades and individuals in different states of structured populations, which may be the cause of the high number of entities found in the results from GMYC and bPTP analyses.

Advertisement call
Six males and 73 advertisement calls of the new species were recorded and analysed. Quantitative call traits are summarized in Table 2. The advertisement call of P. gonzagai sp. nov. consists of a single pulsed note emitted sporadically (Fig. 6A-B). The calls are generally composed of a main group of pulses (= core portion) (Fig. 6A), which may be followed by one low-amplitude final pulse (Fig. 6B).
When present (37 % of analysed calls), the isolated pulses were limited to one and lasted 4-12 ms, separated from the core portion by a long interval of 8-22 ms. The pattern of three-pulsed core without the isolated pulse was more common (63% of analysed calls; Fig. 6A), and lasted 17-35 ms. Taking into account the isolated pulses, the call lasted 17-49 ms and has 3-4 pulses. The pulse duration varied from 5-12 ms, emitted at rates of 86-177 pulses / second. The inter-pulse interval (or no interval) within the core lasted 0-4 ms. The peak of dominant frequency varied from 1969 to 2391 Hz ( Fig. 6A-B).
Six males and 105 advertisement calls of P. nordestinus (sister species) were recorded and analysed. The advertisement call of the sister species of the new taxon also contains a single pulsed note, with sporadic emission (quantitative call traits are summarized in Table 2; Fig. 6C-D). Core duration varied from 17 to 33 ms and the intervals between the core and the isolated pulses varied from 6 to 48 ms. The number of pulses per core varied from 3 to 4 pulses, with pulse intervals (or no interval) within the core from 0 to 6 ms. Pulses were arranged in: (1) a three-pulsed core with one isolated pulse (62.9%); (2) a threepulsed core with no isolated pulse (16.2%); (3) a three-pulsed core followed by two isolated pulses, with long inter-pulse interval between them (17.1%; Fig. 6D); (4) a four-pulsed core followed by an isolated pulse (2.9%); and (5) a four-pulsed core followed by two isolated pulses, with long inter-pulse interval between them (1.0%). The entire call duration lasted 19-85 ms and has 3-6 pulses, the duration of which varied from 3 to 16 ms, emitted at rates of 91-177 pulses / second. The peak of dominant frequency varied from 1969-2250 Hz (Fig. 6D).

Distribution
Pithecopus gonzagai sp. nov. is known from the type locality and from the 16 municipalities in four north-eastern Brazilian states (Rio Grande do Norte, Paraíba, Pernambuco and Alagoas), based on molecular, acoustic and morphological evidence. It is possible that all populations occurring north of the SFR can be attributed to this new species, such as the populations from the states of Ceará and Piauí (Haddad et al. 2013;Roberto & Loebmann 2016); whereas all populations from south of the SFR (states of Sergipe and Bahia) are attributed to P. nordestinus (Fig. 1).

Natural history
Adult males of P. gonzagai sp. nov. call in the open areas, by the margins of lentic environments (mostly ponds) during the rainy season of the year (which could vary across its distribution). Females lay eggs on leaves over the water bodies (Fig. 3D), from where exotrophic tadpoles hatch and drop into the water. Therefore, the reproductive mode is the number 24 (sensu Haddad & Prado 2005). Different call types (advertisement, distress, warning and fighting calls) were described and previously attributed to P. nordestinus (see Toledo et al. 2015). However, part of those calls must now be attributed to P. gonzagai sp. nov.: the distress and warning calls described by Toledo et al. (2015) were recorded from individuals sampled in the state of Rio Grande do Norte. Therefore, this population falls within the distribution of P. gonzagai sp. nov., not P. nordestinus. Thus, it is clear that both species (P. nordestinus and P. gonzagai sp. nov.) have a complex vocal repertoire that should be further explored.

Conservation remarks
Besides corroborating previous observations, we draw attention to a recent governmental act to solve the water supply in semi-arid regions of north-eastern Brazil, the so-called São Francisco River Transposition Project. Already in progress, the transposition project is diverting water from SFR to temporary rivers and reservoirs of the polygon of droughts (called in Portuguese ʻpolígono das secasʼ) (Lee 2009). This region is north of the SFR and suffers historically with the effects of prolonged droughts. In this project, there are two canals called the North and the East Axes (Fig. 1), which pump water from SFR to the driest regions when necessary (RIMA 2004). We may predict two possible impacts to the regional native fauna, including the present pair of species, P. nordestinus and P. gonzagai sp. nov.: a reduction of the geographic barrier effectiveness and / or even creation of new artificial barriers elsewhere. Therefore, the whole historical evolutionary dynamics of the species that inhabit this region would be modified. In the case of P. gonzagai sp. nov., its type locality and some others may be affected because they are between the new permanent and artificial rivers and the reduced SFR (Fig. 1). Therefore, we suggest a long-term genetic survey in this area to monitor the possible effects of the SFR Transposition Project.

Discussion
Our new species hypothesis was written under the ʻGeneral Lineage Concept of Speciesʼ (de Queiroz 1998(de Queiroz , 2007, which treats species as separately evolving metapopulation lineages. We found significant differences in 12 morphometric traits when we compared P. gonzagai sp. nov. with P. nordestinus, the former being smaller in all traits. These traits corroborate phylogenetic evidence (Bruschi et al. 2019) and species delimitation tests showed here. As P. gonzagai sp. nov. is a phenotypically cryptic species, we are unable to indicate a reliable phenotypic diagnostic character to distinguish it from P. nordestinus. That is why P. gonzagai sp. nov. has been historically masked by its phenotypic similarity with P. nordestinus (Caramaschi 2006;Loebmann & Haddad 2010;Brand et al. 2013;Pinto et al. 2013;Neiva et al. 2013). The ancient interruption of the gene flow due to a historical shift in the course of SFR is the best explanation for the P. nordestinus and P. gonzagai sp. nov. split from their most recent common ancestor during the Plio-Pleistocene transition (Bruschi et al. 2019). Despite the clear split between P. nordestinus and P. gonzagai sp. nov., samples from Catieté and Aurelino Leal (Western Bahia state) could represent a putative secondary contact zone between the two species (Bruschi et al. 2019) or the signature of incomplete lineage sorting. The cyto-nuclear noise detected in Western Bahia  Indeed, the routine morphological identification of the species of Pithecopus is not easy mostly due to the high level of conservative morphology observed between P. hypochondrialis, P. azureus and P. nordestinus (Bruschi et al. 2013;Haga et al. 2017a). It is noteworthy that P. azureus and P. nordestinus are also morphometrically indistinguishable from one another according to the present analysis and those carried out by Haga et al. (2017a). Moreover, acoustic traits used in our analysis did not discriminate the new species from P. nordestinus and P. azureus. However, these three allopatric species are currently isolated by geographical barriers (Bruschi et al. 2013(Bruschi et al. , 2019Haga et al. 2017a). If we consider the allopatry and that the phenotypic divergence is fuelled by the joint influence of character displacement and sexual selection (see Pfennig & Pfennig 2010), the selective pressures acting on call traits could be very weak or simply non-existent.
In addition, Haga et al. (2017b) pointed out that the similarities observed among the advertisement calls of P. hypochondrialis, P. azureus and P. nordestinus (the lowland species), and the resulting difficulty in discriminating them, indicate that acoustic traits are uninformative for these phylogenetically closely related species. All six males of P. nordestinus were correctly classified in the randomForest model. This discrepancy can probably be explained by the fact that calls with two isolated pulses were only reported for P. nordestinus. The other five lowland species have, at most, one isolated pulse following the core portion. Although we have found significant differences in some acoustic traits, there are overlaps in their amplitudes, which may affect the specific recognition by females of these lowland species. Moreover, we were unable to find clear acoustic diagnosis between the two sister highland species, P. oreades and P. centralis, as there are overlaps in all their acoustic traits ). Therefore, as well as for P. gonzagai sp. nov. and P. nordestinus, the acoustic traits are not useful to distinguish other related species of Pithecopus. Also, populations of P. nordestinus and P. gonzagai sp. nov. from the Atlantic Rainforest and from the Caatinga showed differences in the interval between the pulses of the advertisement call, revealing putative intraspecific variation (Röhr et al. 2020). The authors pointed out that their results indicate how multiple evolutionary forces may act simultaneously on the advertisement calls of frogs (Röhr et al. 2020).
A characterisation of tadpoles of P. nordestinus was recently provided (Dubeux et al. 2020). However, this study analysed individuals collected north of the SFR (states of Alagoas and Rio Grande do Norte), therefore, P. gonzagai sp. nov. is the most suitable taxonomic identification for these tadpoles characterised by Dubeux et al. (2020), not P. nordestinus. The advertisement call of P. nordestinus described from the municipality of Igrapiúna, state of Bahia (about 145 km northwest of the type locality) is similar to those described here in call duration, pulse duration and dominant frequency but slightly different in number of notes per call and inter-pulse interval (see Vilaça et al. 2011). Possibly, these differences can be explained by an intraspecific variation or even an applied methodology, since the molecular data indicate that the populations south of the SFR correspond to P. nordestinus. Using species distribution modelling algorithms and based on direct field sampling, scientific literature, museum collections, and available online databases, Silva et al. (2020) provided a new distribution area for P. nordestinus which was broader than previously known. In addition, these same authors suggested a new polygon for conservation purposes for this species. In contrast, the distribution suggested by them overlaps with the geographic distributions of P. nordestinus and P. gonzagai sp. nov. reported by Bruschi et al. (2019) and the present study.
Pithecopus gonzagai sp. nov. and P. nordestinus are examples that fulfil the definition for cryptic species of Bickford et al. (2007): "two or more distinct species that are erroneously classified (and hidden) under one species name". The identification of cryptic species complexes allows us to better plan biodiversity conservation strategies (Bickford et al. 2007). Thus, the description of P. gonzagai sp. nov. as a new species has a direct impact for conservation planning on a regional scale. Future projects evaluating north-eastern diversity are needed to identify cryptic species that may be divided by the natural course of the São Francisco River.