New acoustic and molecular data shed light on the poorly known Amazonian frog Adenomera simonstuarti (Leptodactylidae): implications for distribution and conservation

Adenomera simonstuarti is a poorly known species complex inhabiting western Amazonia. Here we reevaluate the species diversity within this complex based on previously documented and newly acquired molecular and phenotypic data. We also redescribe the calling pattern of the nominal species based on the original recording (Peru) and a new recording (Brazil). Our results indicate eight geographically structured genetic lineages and the nominal species with a multi-note call pattern. This is the fi rst association of calls and DNA sequence from a voucher specimen, thereby enabling the assignment of A. simonstuarti to one specifi c lineage within the complex. The multi-note call was not previously reported and represents an important additional diagnostic character within Adenomera. The geographic distribution of A. simonstuarti is substantially narrowed down to the southwestern portion of the entire geographic range recognized for the complex. The lack of taxonomic resolution in the complex is a major conservation concern by preventing us from evaluating the potential threats and extinction risks of each of the lineages. Future research should follow the protocol of combining calls and DNA sequences associated with voucher specimens as a means to address the taxonomic status of genetic lineages within the A. simonstuarti complex. European Journal of Taxonomy 682: 1–18 (2020) 2


Introduction
Morphologically indistinguishable species (cryptic species) have challenged taxonomists and systematists across several taxonomic groups (Bickford et al. 2007). This evolutionary trend occurs when speciation generates distinction in one or more biological traits, for instance, molecular, acoustic or ecological characters, but there is no or very subtle morphological differentiation (Cherty et al. 1978). High levels of cryptic species have been documented for many Neotropical frog groups, especially over the past decades (e.g., Fouquet et al. 2007Fouquet et al. , 2016Padial & De la Riva 2009;Simões et al. 2010;Jungfer et al. 2013). Understanding and conserving biodiversity levels, partly hidden in complexes of cryptic species, in such a megadiverse region strongly depends on the continued investigation of multiple sources of information to be compared with morphological variation (Padial et al. 2010).
A striking example of a Neotropical frog group with predominance of cryptic species is the genus Adenomera Steindachner, 1867 (e.g., Angulo & Reichle 2008;Carvalho & Giaretta 2013a;Carvalho et al. 2019aCarvalho et al. , 2019b. These small-sized leptodactylids (snout-vent length up to 34 mm; Kok et al. 2007) are widely distributed in South America east of the Andes, currently comprising 21 described species (Carvalho et al. 2019b). A phylogenetic study of the genus based on a comprehensive geographic sampling revealed many putative new species, reported as candidate species (Fouquet et al. 2014). Some of these nominal and candidate species of Adenomera exhibit marked genetic divergence among populations and some of them are also known to have distinct call patterns, suggesting extensive cryptic diversity within the genus (Fouquet et al. 2014;Carvalho et al. 2019cCarvalho et al. , 2019d. Fouquet et al. (2014) classifi ed the species diversity of Adenomera into eight major clades, one of them being the Amazonian endemic A. andreae clade. This clade contains three described species, A. andreae (Müller, 1923), A. chicomendesi Carvalho, Angulo, Kokubum, Barrera, Souza, Haddad & Giaretta, 2019, and A. simonstuarti (Angulo & Icochea, 2010), plus three candidate species reported as Adenomera sp. C, Adenomera sp. D, and Adenomera sp. T (Fouquet et al. 2014;Carvalho et al. 2019b).
Adenomera simonstuarti was described from Camisea (Province of La Convención, District of Echarate, Region of Cusco), a region of lowland forest in southwestern Peruvian Amazonia, based on a series of four specimens, and two referred specimens from Pando, in northern Bolivia (Angulo & Icochea 2010). A few years later, Fouquet et al. (2014) showed, based on molecular evidence, that the species could actually be more widely distributed throughout lowland forests of western Amazonia and Andean montane forests, even though those authors also mentioned in their taxonomic considerations (see Fouquet et al. 2014: appendix S2a) that the deep genetic subdivisions within A. simonstuarti could suggest the existence of more than one species under the nominal species. A major limitation that holds back researchers to advance in the taxonomic resolution of this species complex is the lack of associated phenotypic and molecular data for the nominal species. Specimens have to date been identifi ed as A. simonstuarti based on morphological and geographical data. Moreover, the only call description available is from the holotype in the original description, from which tissue samples were not collected (and neither were they collected from paratypes). It is important to highlight that species identifi cation within Adenomera should be treated with caution in such cases which acoustic and/or molecular data are not available in a frog genus having a notably high number of undescribed and/or cryptic species (Carvalho & Giaretta 2013a;Fouquet et al. 2014;Carvalho et al. 2019b;Cassini et al. 2020).
Here, we reevaluate the species diversity within the Adenomera simonstuarti complex by combining novel acoustic and molecular data, enabling for the fi rst time that the nominal species could be linked to a specifi c genetic lineage within the complex. We also reinterpret the calling pattern of A. simonstuarti based on the original recording from the type locality in Peru and a new recording from the Brazilian Amazonia. Lastly, we discuss on the implications for distribution and conservation status of the genetic lineages subsumed under A. simonstuarti across their entire geographic range, resulting from the circumscription of the nominal species to one specifi c lineage.

Taxon sampling and identifi cation
We collected fi ve individuals of Adenomera that we associated with A. simonstuarti based on morphology, color patterns, and/or call characteristics, as follows: (1)

Molecular analysis
Genomic DNA was extracted from muscle and liver tissues preserved in 100% ethanol from three specimens (INPA-H 39792, 39814 and 40967) using standard protocols of a commercial kit (Wizard ® , Promega, Madison, USA). We sequenced a fragment of the mitochondrial gene cytochrome c oxidase subunit I (COI; 657 bp), a widely used molecular marker for this frog group (Fouquet et al. 2014;Lyra et al. 2017). The primers CHmL4 (5-TYTCWACWAAYCAYAAAGAYATCGG-3) and CHmR4 (5-ACYTCRGGRTGRCCRAARAATCA-3) (Che et al. 2012) were used to perform amplifi cation of the selected fragment via Polymerase Chain Reaction (PCR). The amplifi cation reactions used a mix with 1.2 μL of 10 mM dNTPs, 3 μL of a 5X amplifi cation buffer, 1.2 μL of 25 mM MgCl 2 , 1.0 μL of DNA in a concentration of 50 ng/μL, 0.5 μL of each primer at 10 mM, 0.15 μL of Taq DNA polymerase and 7.45 μL of ddH 2 O. Reaction conditions began with an initial heating step at 94°C for 60 s, followed by 35 cycles of denaturation at 94°C for 20 s, annealing at 50°C for 50 s and extension at 72°C for 90 s, followed by a fi nal extension at 72°C for 10 min. PCR products were purifi ed with polyethylene glycol 8000, submitted to a sequencing reaction following BigDye Terminator Cycle Sequencing Kit (Applied Biosystems, Waltham, USA) protocols, and sequenced with an ABI 3130 XL automated sequencer (Applied Biosystems, Waltham, USA). These laboratory procedures were conducted in the Thematic Laboratory of Molecular Biology of INPA. We used Geneious 7 (Kearse et al. 2012) for sequence editing.
Taxon sampling for the molecular analysis included each candidate new species and operational taxonomic units of nominal species from the eight major clades delimited by Fouquet et al. (2014), including all those identifi ed therein as A. simonstuarti, as well as sequences from related genera to be used as outgroups (Lithodytes Fitzinger, 1843, Hydrolaetare Gallardo, 1963and Leptodactylus Fitzinger, 1826. Besides the mtDNA gene COI, we downloaded additional sequences from another mtDNA gene (cytochrome b -cytb; 607 bp) and two nuclear genes (proopiomelanocortin A -POMC and recombination activating gene 1 -RAG1; 547 and 1422 bp, respectively) from the online repository GenBank (Clark et al. 2016). Accession numbers and other information of sequences included in the molecular analysis are provided in the Supplementary File SM.01. New sequences produced for this study were deposited in GenBank under the accession numbers MT472180-MT472182. We used MAFFT 7 online (Katoh & Standley 2013) to independently align the sequences of each gene under the G-INS-i strategy, more suited for protein coding genes (Katoh & Standley 2013). All genes were posteriorly concatenated, leading to a fi nal aligned database containing 105 sequences and 3233 bp. We divided the dataset considering fi rst, second, and third positions of the codon of each gene, and conducted the search for the best-fi tting substitution models and partition schemes with PartitionFinder 2.1.1 (Lanfear et al. 2017) under the corrected Akaike information criterion (AICc; Hurvich & Tsai 1989). Best scheme indicated fi ve partitions, with the general time-reversible model (GTR; Tavaré 1986) with a gamma distribution of rates across sites (+G) as the best-fi tting nucleotide substitution model for the fi rst and third position of cytb, and third position of POMC and RAG1, whereas the GTR+G with a proportion of invariant sites (+I) was indicated as the best-fi tting nucleotide substitution model to remaining codon positions (all from COI, fi rst and second of POMC and RAG1, and second and third of cytb).
We reconstructed phylogenetic trees using both Bayesian inference (BI) and maximum likelihood (ML) optimality criteria. For the Bayesian analysis we used two independent runs of 5.0 × 10 7 generations, starting with random trees and four Markov chains (one cold), sampled every 10000 generations in MrBayes 3.2.6 (Ronquist et al. 2012), discarding 25% of generations and trees as burn-in. We used the standard deviation of split frequencies (< 0.01) and estimated sample size (> 200) to assess run convergence with Tracer 1.7 (Rambaut et al. 2018). We conducted maximum likelihood analysis using RaxML 8.2.10 (Stamatakis 2014), searching the most likely tree 100 times and with 1000 non-parametric bootstrap replicates to assess support.
We used Mega 7 (Kumar et al. 2016) to compute the uncorrected and corrected (Jukes-Cantor model) pairwise genetic distances of the COI fragment among specimens of Adenomera simonstuartimissing data removed using pairwise deletion option. Both uncorrected and corrected genetic distances were considered for our study in order to increase accuracy and comparability of results. With the Approximate Barcode Gap Discovery method (ABGD, Puillandre et al. 2012), we conducted an analysis to delimit lineages of A. simonstuarti based on comparisons of uncorrected intra-versus interspecifi c genetic distances in COI. The analysis were run at the ABGD online server using a prior of intraspecifi c divergence (P) between 0.001 and 0.1, a proxy for minimum relative gap width (X) of 0.5, and a number of bins (n) of 30. Based on an intraspecifi c divergence of 1%, a recognized threshold in delineation analysis among vertebrate groups and the end of a plateau for lineage number (Puillandre et al. 2012), we considered the 16 th partition to delineate lineages.

Acoustic analysis
We recorded the advertisement call of one male A. simonstuarti from the upper Juruá River (see locality #1 earlier) using a Sony PCM-DC50 digital recorder (sampling rate = 44.1 kHz; bit depth = 16) and built-in microphones. The recording was stored as stereo-channel wave fi le (left channel was kept for the acoustic analysis). The sound recording was deposited in Fonoteca Neotropical Jacques Vielliard (Unicamp, Brazil) under the accession number FNJV 45412. Information on the recording is as follows: individual recorded at 09.40 h in the morning; air temperature around 25°C. We also reanalyzed some of the original calls recorded from the type locality (Angulo & Icochea 2010) in order to allow direct and standardized comparisons (FNJV 45409-11). We analyzed calls using an interface built between an expanded version (0.9.6.1) of Soundruler (Gridi-Papp 2007) and Matlab 6.5.2 (Matlab 2004). Note rate was quantifi ed manually in Audacity 2.1.1 (Audacity Team 2017). Acoustic defi nitions and terminology follow those of Carvalho et al. (2019b). Acoustic traits were quantifi ed through automated analysis, for which we developed settings in the software to recognize and delimit the acoustic units both in the time and frequency domains. Data are presented as range (mean ± standard deviation). Ranges include the span of values from the raw dataset. In the case of pulse duration, given that acoustic signals analyzed had more than one pulse, we fi rst averaged the duration of each pulse of a given note (call mean) and then obtained the averaged mean for each male analyzed from the mean duration of call pulses (individual mean), and lastly, we obtained the grand means and associated standard deviations by averaging individual means. We applied two bandpass (500-Hz high-pass and/or 5000-Hz low-pass) fi lters to some of the sound fi les in Soundruler prior to conducting the acoustic analysis to reduce background noise caused by wind and/or rain. Spectrogram parameters were set as follows: FFT size = 1024 points, FFT overlap = 90%, window type = Hanning, contrast = 70%; those for the automated analysis were (in sample sizes): detection (smoothing = 500, resolution = 1), delineation (smooth factor = 1, smoothing = 15 or 100, and resolution = 1); critical amplitude ratio = 0.8 or 1.0. We produced sound fi gures using seewave 2.1.0 (Sueur et al. 2008) and tuneR 1.3.2 (Ligges et al. 2017), in R 3.5.0 (R Core Team 2018). Spectrogram settings were: window Hanning, FFT size = 256 points, and FFT overlap = 90%; the level of frequency components was indicated by a relative 30-dB color scale (red = maximum energy).

Phylogenetic relationships and genetic diversity
Both BI and ML phylogenetic reconstructions (Fig. 1) yielded similar results with regard to relationships in the Adenomera andreae clade and the monophyly of A. simonstuarti. All three new sequences from southwestern Brazilian Amazonia were recovered nested within A. simonstuarti (Fig. 1). The ABGD delimitation analysis recovered eight genetic lineages within A. simonstuarti (Fig. 2) with noticeable geographic structure (Figs 1-2). Mean genetic distances in COI among the lineages of A. simonstuarti (Table 1) range from 3.2−7.6% (uncorrected) and from 3.3−8.0% (corrected), whereas within-lineage genetic distances reach a maximum value of 1.7% (uncorrected and corrected).
Our genetic voucher INPA-H 40967 (Fig. 3) is the only specimen of Adenomera simonstuarti with associated acoustic data. Morphological and color features of the specimen fully agree with those presented in the original description of A. simonstuarti (Angulo & Icochea 2010). This voucher specimen from the upper Juruá River constitutes the lineage 3 together with other specimens from the upper Amazon basin in southwestern Amazonia (Figs 1-2). The lineage 3 is regarded hereinafter as conspecifi c with the nominal species. The other two new COI sequences (lower Juruá River) fell within the lineage 2. These two vouchers also have the recognized morphotype of A. simonstuarti (Fig. 4), but acoustic data for this lineage remain unknown. Mean genetic distances between the COI lineages 2 and 3 are noticeable, ranging from 5.0% (uncorrected) to 5.3% (corrected).

Habitat and natural history
The call voucher of Adenomera simonstuarti from the upper Juruá River (INPA-H 40967; Fig. 3), corresponding to the lineage 3, was collected from an open bamboo forest, approximately 2 km from BR-364 road. This individual and other two were heard calling from an old clearing surrounded by decomposing fallen logs. The three individuals called hidden underneath dense leaf litter, and only one of them (the call voucher) were found while surveying the area. Adenomera simonstuarti and A. andreae were found syntopically in this area.
The four specimens from the lower Juruá River (INPA-H 39792, 39796 and 39813-14; Fig. 4), corresponding to lineage 2, were collected in a non-fl ooded lowland forest (terra fi rme forest) with dense understory layer. Three specimens  were found in a forest affected by anthropogenic activities (i.e., logging), located close to Comunidade Cumaru village. This could indicate a certain degree of tolerance of the lower Juruá population to habitat disturbance, given that human occupation and activities in this region have begun during the late 1980s (ICMBio 2009). The specimen INPA-H 39796 (Fig. 4) was collected from a preserved forest, distant from that village. Specimens in the lower Juruá River were sympatric with A. andreae and A. hylaedactyla. Adenomera simonstuarti and A. andreae were found syntopically inside the forest, whereas A. hylaedactyla was only found along riverbanks.

Distribution patterns
Adenomera simonstuarti (= lineage 3) is distributed in the upper Amazon Basin of southwestern Brazilian and Peruvian Amazonia, and two locations in the eastern slopes of the Andes in south central Peru. Populations linked to the other seven lineages are in most cases allopatric among each other. Some lineages are widely distributed, such as lineage 1, from lowland and montane forests in the upper Amazon Basin of Peru and Brazil. Other lineages, such as lineage 2, may be narrowly distributed on the east bank of the lower Juruá River. Other distribution patterns include: lineage 4 in lowland Amazonia of northeastern Ecuador and extreme northern Peru; lineage 5 in Venezuelan Andes montane forests; lineage 6 in the Marañón-Ucayali interfl uve; lineage 7 in the upper Amazon River; and lineage 8 in the upper Negro River. Based on the geographic patterns of each of the lineages, we could expect that some of them may have distributions associated with interfl uve regions, such as lineages 4, 6 and 7 (Fig. 2). Another interesting pattern is that the nominal species (lineage 3) and other lineages (e.g., lineage 8) are distributed in the upper Amazon Basin, while some others are distributed in the middle-lower portions of major southern tributaries of the Amazon River (e.g., lineage 2; Fig. 2).

Discussion
We did not examine for the morphological analysis most of the genetic vouchers linked to the Adenomera simonstuarti lineages of Fouquet et al. (2014); see their appendix S1a. The only exception is the specimen INPA-H 5337 (see Material examined) belonging to the lineage 2 from the lower Juruá River (previously reported as QU5337 by Fouquet et al. 2014). Likewise, acoustic data for the lineages other than the one containing the nominal species remain unknown. Nevertheless, by reinterpreting the calling pattern of A. simonstuarti (i.e., multi-note advertisement call) and linking the nominal species to a specifi c genetic lineage, our study contributes to the potential discrimination between the nominal species and closely related, putative new species within the A. simonstuarti complex. Due to the lack of acoustic and morphological data that could help to corroborate the existence of multiple, unnamed lineages within this species complex, a plausible alternative hypothesis would be the one of A. simonstuarti as a single species containing deep conspecifi c lineages across its geographic range in western Amazonia. Future studies should follow the protocol of combining calls and DNA sequences associated with voucher specimens as a means to fully address the taxonomic status of the other seven genetic lineages within the A. simonstuarti complex, regarded herein as putative new species.
Of special relevance is the acquisition of DNA sequences for the Amazonian Adenomera population from Camisea, in southeastern Peru, sympatric with A. simonstuarti in the type locality region. That population was originally reported as Adenomera cf. andreae by Angulo & Icochea (2003), but referred hereinafter to as Adenomera sp. from Camisea. Based on the few calls available (recording FNJV 45413; see Fig. 5C-D), we briefl y and qualitatively characterized the advertisement call of Adenomera sp. from Camisea as nonpulsed and given as single notes, with the dominant frequency at the fundamental harmonic, and with negligible frequency modulation. These acoustic traits distinguish this taxon from the sympatric A. simonstuarti and all other Amazonian species of Adenomera (for acoustic comparisons in Adenomera, see Carvalho et al. 2019bCarvalho et al. , 2019cCarvalho et al. , 2019d. We also examined the recorded male of Adenomera sp. from Camisea (accession number: MUSM 18219), which differs from nominal A. simonstuarti by lacking the nearly solid, black-colored stripe along the underside of the forearm. In fact, the specimen is morphologically more similar to other members of the A. andreae clade, especially by the presence of toe tips fully expanded into small discs (Carvalho et al. 2019b(Carvalho et al. , 2019d. Based on its distinctive call, however, Adenomera sp. from Camisea cannot be conspecifi c with nominal A. andreae or any other described species and candidate new species of the A. andreae clade with described calls ( Carvalho et al. 2019bCarvalho et al. , 2019d, except for two Peruvian lineages (i.e., Adenomera sp. D and Adenomera sp. T; Fouquet et al. 2014) whose calls remain unknown (for further discussion, see Fouquet et al. 2014: appendix S2a). Adenomera sp. from Camisea should therefore correspond to an unnamed species that may be conspecifi c with one of the two candidate new species of the A. andreae clade with unknown calls (Adenomera sp. D and Adenomera sp. T), with one of the lineages under A. simonstuarti for which morphological data have not yet been assessed, or might even have not been genetically sampled to date.
The age of initial diversifi cation of the Adenomera simonstuarti complex is unknown but likely during the Miocene, given that the divergence time between this species complex and its sister clade (A. chicomendesi + Adenomera sp. D) was estimated to have occurred during this geological period (8.0-8.5 Ma; Fouquet et al. 2014). The A. simonstuarti complex inhabits an Amazonian region that has been affected by several landscape changes and hydrological instability during the Miocene (Albert et al. 2018), which might have generated the deep genetic divergence that mirrors the allopatric distributions of lineages within the complex (Figs 1-2). Furthermore, the result of intense climatic variation in the region during the Pleistocene, which modifi ed the extension and structure of the forest habitats in the region (Arruda et al. 2018), might also have contributed to the more recent divergences within this complex.
The southwestern lowland Amazonian forests consist of a heterogeneous mosaic of habitats. We surveyed for Adenomera in fl ooded forests (várzeas) and other riparian environments, but individuals of the A. simonstuarti complex were only found in non-fl ooded (terra fi rme) forest. These non-fl ooded forests are patchily distributed within the heterogeneous landscape, in some cases distantly located from the main course of rivers, which are the primary access route by researchers (Oliveira et al. 2016). For this reason, the limited access to the apparently preferred habitat may bias the understanding of the geographic range and patterns of distribution of lineages within the A. simonstuarti complex. The conservation status of A. simonstuarti was originally assessed as Data Defi cient (DD), pending future surveys to confi rm that it could also occur in between known localities of Peru and Bolivia (Angulo & Icochea 2010). With the inclusion of the new occurrence record in the Brazilian state of Acre (Fig. 2) Barlow et al. 2020).
Despite the recommendation of extinction risk of nominal Adenomera simonstuarti in the Least Concern category, it is important to highlight the potential threats and extinction risk to the other seven lineages subsumed within the A. simonstuarti complex, regarded as putative new species. This is especially relevant because conservation strategies are in many cases not feasible as long as the taxonomic status of unnamed lineages is not fully resolved; see Angulo & Icochea (2010) for a proposition on the impacts of cryptic species complexes on biodiversity and conservation assessments. By circumscribing A. simonstuarti to one of eight genetic lineages within the complex, its distribution is dramatically narrowed down to the southernmost portion of the entire range of the species complex (~2 million km² EOO; Fig. 2), corresponding to a decrease of 90% in EOO. Our current knowledge on species richness and distribution is still insuffi cient for accurate evaluations of conservation status and distribution patterns of the A. simonstuarti complex. In the opposite scenario, by lumping the other seven lineages back into A. simonstuarti and considering the deep genetic divergence as intraspecifi c variation, the conservation of metapopulations displaying high genetic variability should also be taken into consideration as a signifi cant component to safeguard the biological heritage (Crandall et al. 2000).