*2.3. Evolutionary Relationships*

*Taxon and gene sampling—*we combined genetic sequences available from previous work (mostly [2,3,19]) and new sequences generated during this study (Table S2). Genomic extraction, amplification, and sequencing are as described in Guayasamin et al. [2] and Castroviejo-Fisher et al. [3]. The final dataset contains complete or partial sequences of 10 genes representing 6513 bp of data (mitochondrial: 12S rRNA, 16S rRNA, ND1; nuclear: BDNF, C-MYC exon 2, CXCR4, NCX1, POMC, RAG1, SLC8A3). We sampled 49 outgroup taxa from a large range of families within Hyloidea (Table S2), and include all three species from Allophrynidae, the sister group to Centrolenidae [2,12]. Finally, ingroup taxon sampling includes 251 terminals, representing 113 named species, and 24 putative new species. The percentage of named species sampled (i.e., matrix completeness) for each marker is as follows: 12S = 90%; 16S = 94%; BDNF = 55%; C-MYC exon 2 61%; CXCR4 = 61%; NCX1 = 59%; ND1 = 81%; POMC = 61%; RAG1 = 62%; and SLC8A3 = 60% (complete marker statistics can be found in Table S3). For most species of Ecuadorian glassfrogs, we obtained a total of ~2733 bp from the following mitochondrial markers: 12S rRNA (~907 bp), 16S rRNA (~864 bp), and *ND1* (~960 bp). See Table S2 for genes sequenced for each species and GenBank Accession Numbers.

*Candidate species*—to assist in identifying putative new species for future study, we employed the candidate species designation system of Vieites et al. [29]. Prior to this study, divergent centrolenid lineages have been designated in inconsistent ways (using sp. or <sup>a</sup>ff. or museum numbers), which can result in confusion in identifying and studying tentative new species. Given the increasing number of putative new species in Centrolenidae, establishing candidate species aims to better organize and maintain the growing discovery of undescribed lineages. Importantly, designating a divergent lineage as a putative new species does not necessarily mean it is a new species; rather, the system is meant to identify these lineages for future study, incorporating multiple lines of evidence (i.e., morphology, bioacoustics, biogeography, or nuclear genes) in order to determine their species status.

Candidate species are typically identified through evidence from genetic distances or phylogenetic lineage divergence through widely used genetic markers that provide a basis for comparison (i.e., 12S, 16S, CO1; [29,53]). In this study, we used the mitochondrial genetic markers 12S, 16S, and ND1 and integrated evidence from genetic distances and phylogenetic relationships to identify candidate species. We first used a threshold of 3% in identifying divergent lineages, which is a threshold that most named centrolenid species exceed (this study). Putative new species were then numbered using a scheme of "sp\_CaXX", with numbering beginning at 01 for each genus. Finally, we note that Vieites et al. [29] also uses "unconfirmed" and "confirmed" designations, whereas "confirmed" candidate species have additional evidence (morphology or calls) for their status as species and are simply awaiting detailed analyses and subsequent description. We did not use these designations, as most putative new centrolenid species are supported only through mitochondrial genetic divergence.

*Phylogenetics*—the analyzed dataset contains complete or partial sequences of 10 genes representing 6513 bp of data (mitochondrial: 12S rRNA, 16S rRNA, ND1; nuclear: BDNF, C-MYC exon 2, CXCR4, NCX1, POMC, RAG1, SLC8A3). See Tables S2 and S3 for details. Analyses were conducted using maximum likelihood (ML) and Bayesian (BA) criteria. We did not perform searches under the parsimony criterion because of its limitations under certain conditions (long branch attraction; [54,55]) and lack of theoretical support for transition/transversion bias and variation in substitution rates among di fferent nucleotide sites [56]. The 12S and 16S rRNA sequence data were aligned using MAFFT 7.4 [57], using the Q-INS-i algorithm that takes RNA secondary structure into consideration. Protein coding genes were also aligned in MAFFT using the AUTO function and manually inspected for accuracy and open reading frames.

Maximum likelihood was run in the IQ-TREE v1.5.5 software [58]. The data were automatically partitioned, and the best model was implemented using ModelFinder within IQ-TREE [59], which groups partitions with the same model and similar rates and simultaneously searches model and tree space. Node support was assessed via 100 ultra-fast bootstrap replicates [60].

For Bayesian phylogenetic analyses and divergence dating, we used BEAST v2.4.5 [61]. We used a single secondary calibration point using ages estimated from Hutter et al. [62], which estimates divergence dates for Hyloidea using 18 genetic markers and 8 fossil/geographic calibrations. We used median age of Centrolenidae with a normal distribution to capture the 95% confidence interval from this study (Mean = 33.4 Myr; Sigma = 6). We used a single uncorrelated lognormal relaxed clock prior linked across all partitions. We partitioned data by codon position for protein-coding genes and used the Bayesian bModelTest package within BEAST to select the best model for each partition. We estimated the relaxed clock rate using an initial value of 1e-9 and a broad prior (in our case, a gamma distribution with a shape parameter of 0.001 and scale parameter of 1000). We used a Yule speciation process for the tree prior. We ran Markov chain Monte Carlo (MCMC) searches for a total of 100 million generations, sampling every 10,000 generations. Stationarity was assessed by examining the standard deviation of the split frequencies and by plotting the -ln*L* per generation, using Tracer v1.5 [63]; trees generated before stationarity were discarded as "burn-in", which was 20% of trees.

*Species conservation status*—global and local (Ecuador) conservation status of glassfrogs follow the categorization and criteria established by the International Union for Conservation of Nature (IUCN) [64] (Figure 17), including rate of decline, population size, area of geographic distribution, and degree of distribution fragmentation. The following categories were used: (i) *Not Evaluated:* The species has not ye<sup>t</sup> been evaluated against the criteria, (ii) *Data Deficient:* There is inadequate information to make a direct or indirect assessment of its risk of extinction based on its distribution and/or population status, (iii) *Least Concern:* The species is widespread and abundant and not under immediate risk of extinction, (iv) *Near Threatened:* For species that are not currently threatened, but are close to qualifying for or are likely to qualify for a threatened category in the near future, (v) *Vulnerable:* A taxon is Vulnerable when it is considered to be facing a high risk of extinction in the wild (Criteria A to E for Vulnerable), (vi) *Endangered:* When the species is considered to be facing a very high risk of extinction in the wild (Criteria A to E for Endangered), (vii) *Critically Endangered:* When the species is considered to be facing an extremely high risk of extinction in the wild (Criteria A to E for Critically Endangered), (viii) *Extinct in the Wild:* When the species is known only to survive in cultivation, in captivity, or as a naturalized population, outside its historical range. A taxon is presumed Extinct in the Wild when exhaustive surveys in known and/or expected habitats, at appropriate times, throughout its historic range have failed to record an individual, (ix) *Extinct:* When there is no reasonable doubt that the last individual of the species has died. A taxon is presumed Extinct when exhaustive surveys in known and/or expected habitats, at appropriate times (diurnal, seasonal, annual), throughout its historic range have failed to record an individual.

**Figure 17.** International Union for Conservation of Nature (IUCN) categories of species conservation status.

## *2.4. Biogeographic Regions of Ecuador*

In this study, we applied biogeographic regions of continental Ecuador as a simplification of vegetation types [65,66] (Figure 18). This system has the virtue of combining the characteristics of the vegetation and the historic isolation between the western and eastern slopes of the Andes and the eastern and western lowlands. Mean annual precipitation and mean annual temperature for each of the recognized regions are shown in Table 2. We also summarized the land cover of Ecuador, illustrating which ecosystems have suffered intensive deforestation and which are still preserved (Table 3, Figure 18).

*Dry Shrub—*characterized by a combination of warm and extremely dry conditions. Annual precipitation can be as low as 60 mm in the westernmost locality (Salinas, Santa Elena Province). This region covers an area of 8033 km<sup>2</sup> and is restricted to the coastal margin of central Ecuador (Figure 18). In some areas, grasses introduced for raising livestock have replaced native plants. In the drier habitats, xerophytic plants are dominant, especially cacti and other thorny plants [67].

*Western Deciduous Forest—*this forest occurs at an elevation of 50–300 m in central and northern Ecuador (100–400 m in southern Ecuador) and covers 25,673 km<sup>2</sup> (10.3% of Ecuadorian territory, Figure 18). Conditions are drier and the terrain has lower tree densities than in evergreen forests. The trees are generally shorter than 20 m with an understory that can be dense with abundant herbaceous plants. Some tree species lose their leaves during the dry season [67]. More than half of the land cover of this habitat type has been converted for agriculture and grazing cattle.

*Chocoan Tropical Rainforest—*this rainforest is the second largest biogeographic region in Ecuador, with 31,732 km<sup>2</sup> at elevations ranging from sea level to 300 m (Figure 18). It has a closed canopy with trees that can reach 30 m in height and with an understory dominated by ferns and Araceae [67]. Tree diversity is high, with more than 100 species/ha with diameter at breast height >10 cm, but lower than in the Amazonian Tropical Rainforest [68]. Habitat destruction rate in this region is the highest in Ecuador and only 18.3% of its natural vegetation remains.

*Western Foothill Forest—*this forest covers 15,305 km<sup>2</sup> on the western Andean slopes with an elevational range of 300–1300 m (400–1000 m in southern Ecuador). Plant endemism is high, especially between latitudes 0◦ and 3◦ S [67]. This forest is structurally similar to its counterpart from the eastern Andean slope, although the amphibian communities are highly differentiated.

**Figure 18.** Land use, biogeographic regions, and provinces of Ecuador. Land cover: Modified from Ron et al. [66]; "Mosaics" are mixtures of natural vegetation and either agricultural land or pastures. "Other" includes urban areas, shrimp farms, lakes, rivers, glaciers, and sand banks.Biogeographic regions: Shown as a simplification of vegetation types (Sierra et al. [65], as modified by Ron et al. [66]). Provinces: Provinces are divided into three broad geographic regions: Coast, Andes, and Amazon.


**Table 2.** Glassfrog diversity and climate in the Ecuadorian biogeographic regions, as defined in Figure 14 [65,66]. Note that any given species can occur in more than one biogeographic region.

**Table 3.** Land cover (as percentage) in Ecuadorian biogeographic regions [66] (Figure 15). Mosaics: Mixtures of natural vegetation and either agricultural land or pastures. Other: Includes urban areas, shrimp farms, lakes, rivers, glaciers, and sand banks.


*Western Montane Forest—*this evergreen forest covers 21,576 km<sup>2</sup> at an elevational range between 1300–3400 m (1000–3000 m in southern Ecuador; Figure 18). The canopy is generally below 25 m with a high abundance of epiphytic plants (especially mosses, ferns, orchids, and bromeliads). At intermediate elevations, especially during the afternoon, the forests become misty and receive horizontal precipitation from low, overhanging clouds. Western Montane Forest is restricted to narrow stretches between the basin of the Mira River (close to the Colombian border) and the basins of the Chanchán and Chimbo rivers (2◦ S). It is replaced by drier habitats (principally Andean Shrub) south of 4◦ S, close to the border with Peru. Only 35% of its natural vegetation remains unaltered (Table 3).

*Páramo—*this is the vegetation type that reaches the highest elevation and covers 15,976 ha (6.1% of the territory; Figure 18). Depending on the region, its lower limit lies between 3000 and 3600 m. Short herbaceous plants generally forming tight clumps dominate the vegetation. The plants are adapted to cold temperatures and to low availability of water. Open grassy areas predominate but are

mixed with small patches of forest or shrubs [69]. At higher elevations, the vegetation is restricted to sparse clumps on otherwise bare land. Because of the occurrence of frequent freezes, agriculture is limited, and this has ameliorated anthropogenic habitat degradation. In this region, only 21.1% of the natural vegetation has been cleared or severely fragmented, the lowest proportion for any region (Table 3). However, the Páramo is the region with the highest proportion of endangered amphibians.

*Andean Shrub—*this biogeographic region lies between 1400 and 3000 m and has an area of 11,266 km2; it is found in the inter-Andean basins between the Cordillera Occidental and Cordillera Oriental (Figure 18). As a result of rain shadow e ffects from both mountain chains, the Andean Shrub has a relatively low precipitation (Table 2). Although originally dominated by shrubs, most of the vegetation has been replaced by crops, pastures, or forests of exotic trees (*Eucalyptus* and *Pinus*; [69]). In dry valleys (e.g., Chota, Guayllabamba, and Patate) the native vegetation is spiny. Andean Shrub is almost unrepresented in the Ecuadorian National System of Protected Areas. Habitat degradation is severe; more than half the land cover is devoted to agriculture or to raising cattle (Table 3).

*Eastern Montane Forest—*this forest covers 31,555 km<sup>2</sup> between 1300 m and 3600 m on the eastern Andean slopes (Figure 18). The vegetation is structurally similar to that from the Western Montane Forest. Above 2900 m, the soil of the forest is covered by moss and the trees are irregularly shaped [69].

*Eastern Foothill Forest—*this evergreen forest covers 13,133 km<sup>2</sup> between elevations of 600 m and 1300 m (Figure 18), and is a mixture of tree species from the Andes and the lowlands of the Amazon Basin [68]. The canopy reaches up to 30 m in height and encloses a dense sub-canopy and understory. Tree diversity is lower (130 species/ha, >10 cm DBH) [68] than in the Amazonian Tropical Forest. Average annual precipitation is the second highest of all regions (2833 mm).

*Amazonian Tropical Rainforest—*the Amazonian Tropical Rainforest is the most extensive biogeographic region in Ecuador with a total area of 73,909 km<sup>2</sup> (29.8% of the Ecuadorian continental territory; Figure 18). It is restricted to elevations below 600 m and has the highest average annual precipitation of any region (3349 mm). The dominant forest type, *Terra Firme*, is characterized by well-drained soils. The canopy is 10–30 m high, punctuated with emergen<sup>t</sup> trees up to 40 m (and rarely 50 m); small gaps created by fallen trees [68,70] are common. Tree diversity is high with 200–300 species of trees/ha (>10 cm DBH) [68,70]. Other vegetation types in this region include *Várzea* forest (flooded with white water), *Igapó* forest (flooded with black water), riparian woodland forest, river island scrub, and *Mauritia flexuosa* palm swamps [68,71]. At the local scale (≤100 km2), amphibian diversity reaches its global peak in the Amazonian Tropical Rainforest of Ecuador [72,73].

## *2.5. Potential Distribution*

Estimating the distribution of a species is challenging, since biotic, abiotic, and historic factors come into play. As an approximation to species distributions and being aware of the associated caveats [74,75], we modelled the ecological niche of all species for which we had at least 10 independent localities (>1 km apart). Given that our ecological models did not include variables such as biotic interactions, random extinction, vagility, influence of diseases or introduced species, we used our results only as proxy, and are fully aware that predicted areas might be, in most cases, overestimated; nevertheless, models are useful for conservations assessments. Locality data were obtained from the following museums: AMNH, BMNH, FHGO, KU, QCAZ, MECN, MZUTI, ZSFQ, and USNM. All specimens were directly examined by Juan M. Guayasamin or Diego F. Cisneros-Heredia.

We used all 19 climatic variables from the WorldClim database (www.worldclim.org). These bioclimatic variables are derived from monthly temperature and rainfall values [76]. A correlation analysis was used to determine independent variables for the distribution of each species. The potential distributions of centrolenid species were estimated using default settings in Maxent v. 3.3.3k [77,78]. We used 70% of the presence records as training data, and the other 30% was used to evaluate the model. Maxent was run 10 times to obtain the ecological model for each species. In order to evaluate models, we employed the area under the curve (AUC), a value that is an indicator of model performance [77,79]. The AUC value was calculated in Maxent and ranges between 0.5 (random classification) and 1

(perfect fit). We discarded models that had AUC values below 0.7 [80]. Maxent continuous models were reclassified in order to obtain a binary map of potential presence or absence, using the maximum training sensitivity plus specificity threshold (MTS + S), a method that has been shown to produce highly accurate predictions [81,82]. The potential distribution models were edited using ArcMap v.10 to obtain more realistic estimates of species distributions by removing areas predicted as present that are located in inaccessible biogeographic regions (e.g., species restricted to the Amazon basin predicted as present in the Chocó and vice versa; species from the Pacific slope of the Andes predicted as present in the Amazonian slope and vice versa).

*Impacts of human activities to species*—in order to build a *layer of human impact (LHI)* for Ecuador, we used shape files containing information on land use, roads, human settlements, human population density, mining, and oil exploration and concession shape files (ESRI, 2003). The original shape files were converted to raster files in ArcMap v. 10; we then calculated Euclidean distances in each raster file to compute buffer areas around specific features (i.e., roads, oil fields), giving values according to the intensity of human impact and distance from the specific human activity (Table 4). In the case of population density, the intensity value was equal to the logarithmic scale of the population density value for each grid cell. We obtained raster layers with buffer zones and intensity values for each of the impacts, which were added using the map calculator tool (ArcMap v. 10) to obtain a single map summarizing all considered threats.


**Table 4.** Intensity values and distance of influence of human activities.

## *2.6. Registration of New Nomenclatural Acts*

According to the International Commission on Zoological Nomenclature (ICZN), which produces the International Code of Zoological Nomenclature, the electronic publication of this article in portable document format (PDF) represents a published work. Therefore, the new species name contained in the PDF is e ffectively published under the International Code of Zoological Nomenclature from the electronic edition alone. This publication and the nomenclatural acts contained in it have been registered in ZooBank, the online o fficial register for the ICZN. The ZooBank Life Science Identifiers (LSIDs) can be accessed and viewed through standard web browsers by appending the LSID to the prefix http://zoobank.org/. The online version of this work is archived and available from the following digital repositories: Diversity, CLOCKSS, and e-Helvetica.
