A fantastic new species of secretive forest frog discovered from forest fragments near Andasibe, Madagascar

We describe a fantastic new species of forest frog (Mantellidae: Gephyromantis: subgenus: Laurentomantis) from moderately high elevations in the vicinity of Andasibe, Madagascar. This region has been surveyed extensively and has a remarkably high anuran diversity with many undocumented species still being discovered. Surprisingly, by exploring areas around Andasibe that lacked biodiversity surveys, we discovered a spectacular and clearly morphologically distinct species, previously unknown to science, Gephyromantis marokoroko sp. nov., documented for the first time in 2015. The new species is well characterised by a very rugose and granular dorsum, dark brown skin with bright red mottling, sparse light orange to white spots on the ventre, vibrant red eyes and femoral glands present only in males that consist of eight medium-sized granules. Bioacoustically, the new species has a quiet advertisement call that differs from related species by having a moderate call duration, 2–4 strongly pulsed notes and a slow note repetition rate. Furthermore, it has substantial differentiation in mitochondrial DNA, with pairwise distances of 7–9% to all other related species in sequences of the mitochondrial 16S rRNA marker. Additional evidence is given through a combined four mitochondrial markers and four nuclear exons concatenated species tree, strongly supporting G. striatus as the sister species of the new species in both analyses. The discovery of this new species highlights the need for continued inventory work in high elevation rainforests of Madagascar, even in relatively well-studied regions.


Introduction
Madagascar hosts an impressively diverse and almost entirely endemic amphibian fauna, diversified into a multitude of different habitats and micro-habitats . Amongst the small, brown leaf litter frogs, members of the genus Gephyromantis Methuen, Recently, Madagascar's unique biota has been the focus of intensive field surveys (e.g. Goodman and Benstead 2003;D'Cruze et al. 2009;Hutter et al. 2015;Scherz et al. 2017a), with many previously unknown species discovered, typically through extensive molecular identification of candidate species lineages (Vieites et al. 2009;Perl et al. 2014). This is true for Gephyromantis, which has several candidate lineages identified (Scherz et al. 2017a(Scherz et al. , 2017bVences et al. 2017), some of which have recently been described (Scherz et al. 2017b. Despite these barcoding efforts, entirely new species, not previously identified from molecular barcoding, are still being discovered and described (Scherz et al. 2017a;, suggesting that many new species to science remain elusive. Herein, we describe another Gephyromantis species new to science from the subgenus Laurentomantis that has not been previously identified through molecular barcoding studies. This new species is not morphologically cryptic and was immediately recognisable as a new species upon discovery in recent expeditions to remote high-elevation forests surrounding the village of Andasibe that lack biodiversity surveys. Despite the Andasibe region being one of the most well-studied areas in Madagascar (Vieites et al. 2009), this study shows that clearly recognisable new species that have not been barcoded are still being discovered by recent surveys. As a result of these discoveries, we emphasise that continued exploration and surveys in Madagascar are needed, even in relatively well-studied regions. Conservation of small forest fragments is especially important, as many undiscovered species may remain undetected.

Materials and methods Terminology
We follow the unified concept of species (i.e. general lineage concept), which defines a species as a separately evolving lineage (Simpson 1961;Wiley 1978;de Queiroz 1998de Queiroz , 2005de Queiroz , 2007. We use multiple lines of evidence (i.e. secondary criteria) in assessing species boundaries, combining data from morphology, phylogenetics, bioacoustics and biogeography (de Queiroz 2007;Padial et al. 2010;Vences et al. 2013). This evidence is then considered equally and used as support for the hypothesis that a given population is an independently evolving lineage and, thus, a distinct species. Family and generic names follow the taxonomy proposed by Glaw and Vences (2006). Geographic regions for biogeographic analyses are defined according to Boumans et al. (2007). According to this zonation, the Andasibe area is within a region named the "Northern Central East" of Madagascar.

Specimen collection and morphological measurement
Specimens were collected at night through targeted searches of the new species' microhabitat. Specimens were euthanised using Tricaine, fixed in ~ 10% formalin (buffered with sodium phosphate to ~ 7.0 pH) for 24 hours and then stored in 70% ethanol for long-term preservation. We deposited and examined alcohol-preserved specimens from the amphibian collections at the Biodiversity Institute of the University of Kansas (KU) and Département de Biologie Animale, Antananarivo (UADBA) (Appendix I). Additional collection acronyms used herein are FAZC, ZCMV, FGZC and LR (field number series of F. Andreone, M. Vences, F. Glaw and L. Raharivololoniaina, respectively), FGMV (field number series shared between M. Vences and F. Glaw) and ZSM (Zoologische Staatssammlung München, Germany). All photographs were taken by CRH, unless otherwise noted.

DNA sequencing and phylogenetics
Following euthanasia, we extracted whole livers and left hind limb muscles and stored the tissues in 95% ethanol. We obtained new genetic data for four specimens of the new species and one specimen from five other species in Laurentomantis from the 3' fragment of the 16S rRNA mitochondrial marker widely used for molecular comparisons and species barcoding in Mantellidae (e.g. Vieites et al. 2009). The methods for DNA extraction, primers used, PCR amplification and sequencing are described in Hutter et al. (2018). Finally, additional mitochondrial and nuclear markers from one specimen of the new species (KU 343230) were acquired by extracting the target markers from samples sequenced using the FrogCap Ranoidea-V1 probe-set (Hutter et al. 2021; available at: https://github.com/chutter/FrogCap-Sequence-Capture). Probe design, sequencing and analytical methods are described in Hutter et al. (2021) in detail. After sequencing, DNA data were manually edited for quality in Geneious R9 (Biomatters 2016). Sequences were deposited in Gen-Bank and their associated voucher specimens and accession numbers are provided in Appendix I.
We aligned the new sequences with 16S sequences from Kaffenberger et al. (2012) to confirm the subgeneric relationship of the new species (tree not shown). We next chose sequences for 16S from all Laurentomantis and several representatives from other species in Gephyromantis and the distantly-related Mantella madagascariensis as outgroups. In total, we supplemented these new data with 182 published sequences of Gephyromantis specimens from GenBank. The distantly-related Mantella madagascariensis was used as an outgroup to root the phylogeny. GenBank accession numbers and their associated specimen data are included in Appendix I.
The 16S rRNA sequence data were first aligned with MAFFT v.7.3 using the RNA alignment algorithm Q-ins-I (Katoh and Stanley 2013). We used Maximum Likelihood (ML) in IQ-Tree v.1.5.5 (Nguyen et al. 2015) to conduct phylogenetic tree reconstruction with default options selected. We used ModelFinder (Kalyaanamoorthy et al. 2017) to find a best-fit partitioning scheme and selected models of molecular evolution for each partition considering all models. We assessed support using 1000 ultrafast bootstrap replicates (Minh et al. 2013). Strongly supported nodes are those with 95 or higher bootstrap (BS).
For Bayesian Inference (BI), we used MrBayes 3.2 (Ronquist et al. 2012) and the best partitions and models selected above. We used reversible jump Markov Chain Monte Carlo to accommodate uncertainty in model selection (parameter set: nst=mixed). The analysis was run for two independent runs of 50 million generations sampling every 1000 generations. Chain mixing and stationarity were assessed by examining the standard deviation of split frequencies and by plotting the -lnL per generation using Tracer 1.5 software (Rambaut and Drummond 2007), where we discarded 25% of the generations as burn-in. Finally, results were combined using logCombiner 1.10 software (Rambaut and Drummond 2007) to obtain a 50% majority rule consensus tree and node posterior probabilities. Strongly supported nodes are those with a posterior probability (PP) of 0.95 or higher.

Bioacoustics
Advertisement calls were recorded in the field with a Marantz PMD 661 MKII Field Recorder and a Sennheiser MKH 8060 shotgun microphone. The calls were recorded in WAV format with a sampling rate of 44.1 kHz/s with 16 bits/sample. Advertisement calls analysed here have been deposited on FigShare (10.6084/m9.figshare.16728994). Calling males were recorded while inside plastic collecting bags at ~ 100 cm because we could not approach them close enough to record them in the field (we did not perceive a difference between the captive and in situ advertisement calls). We measured call parameters using RavenPro 1.5 (K. Lisa Yang Center for Conservation Bioacoustics 2014). Frequency information was obtained through Fast Fourier Transformation (FFT; width 1012 points). A Hanning window (512 bands) was used to create the spectrogram. Measures are reported as the range followed by the mean ± two standard deviations from the mean. Terminology generally follows Köhler et al. (2017), with a call defined as the entire assemblage of acoustic signals emitted in sequence and notes are subunits separated by temporally distinct segments of background noise between each note.
Finally, we evaluated the amount of bioacoustic differences between species following Vieites et al. (2009). We considered differences in general call structure (e.g. pulsed/tonal notes, consistent note arrangements, amplitude envelope shape; Ryan and Rand 1990) and such temporal variables that are putatively less influenced by temperature, body size and behaviour (e.g. note duration, pulse rate; Gerhardt et al. 2000) to be important traits for distinguishing species.

Results
We discovered a morphologically distinct new species belonging to the subgenus Laurentomantis from Gephyromantis in the Andasibe area of Northern Central East Madagascar ( Fig. 1), present at high elevations in several small forest fragments. The new species can be readily identified morphologically through its rugose and granular dorsal texture with prominent ridge elements, red dorsal colouration on a dark brown background, bright red eyes, the relatively large number of eight granules within each femoral gland and absence (or indistinction) of vertebral stripe (Fig. 2). Finally, comparisons of the uncorrected raw genetic distances give a minimum distance of 6-9% with G. ventrimaculatus in the mitochondrial marker 16S rRNA and greater distances with other species in Laurentomantis (Fig. 3). Phylogenetically, the new species position is poorly supported in 16S rRNA, but strongly supported sister to G. striatus in the combined nine marker mitochondrial and nuclear phylogeny (BS = 98; PP = 1.00; Fig. 4). Furthermore, the genetic distances between G. striatus and the new species are 7-9%. The new species also has an advertisement call similar to that of other Laurentomantis, but can be distinguished through the combination of a moderate call duration, differing note structure with 2-4 clearly defined pulses and slower note repetition rate when compared to related species. Given the strong evidence, we describe the new species as follows: Common English name: The Rugose Forest Frog. Available names. There are no junior synonyms available that could be assigned to the new species from the subgenus Laurentomantis.
Etymology. The specific epithet marokoroko is a Malagasy word meaning "rugose" or "rugged". The name was chosen to describe the rugose skin texture of this species. The name is to be treated as an invariable noun in apposition.
Diagnosis. Gephyromantis marokoroko (Fig. 2) is a member of the family Mantellidae, subfamily Mantellinae, as diagnosed by Glaw and Vences (2006). The new species can be diagnosed to the genus Gephyromantis morphologically through its granular dorsum, moderately enlarged fingertips, absence of foot webbing, bifid tongue and small femoral glands present only in males as a small number of large granules (type 2; Glaw et al. 2000). Within Gephyromantis, the new species can be diagnosed to the subgenus Laurentomantis through its irregular and rough granular dorsum, single subgular vocal sac in males, completely connected lateral metatarsalia, inner and outer metatarsal tubercle present and tympanum is the same size in male and female.
Gephyromantis marokoroko is characterised by bright red eyes, prominent ridge elements on dorsum, life colouration with a dark brown ground colour with mottled red and grey, hind-limbs dark brown containing red crossbands, absence of red colouration on the sides of thighs and ventre, white spots on grey-coloured ventre and males with bulbous type 2 femoral glands with eight granules in two rows of four on each thigh. Furthermore, the new species is characterised by an advertisement call with a moderately long call duration (1095-1431 ms), 22-28 notes/call, 2-4 strong amplitude-modulated pulses per note and a dominant frequency of 2250-2812 Hz. Finally, Gephyromantis marokoroko has a large genetic distance of 6% or greater amongst related species in the 16S rRNA marker and has strongly supported reciprocal monophyly to all other species in Laurentomantis (Fig. 3).
Gephyromantis marokoroko can be distinguished from other members of Laurentomantis morphologically (Table 1; Fig. 5). The rugose and granular dorsal texture with prominent ridge elements and red mottled colouration and the larger number of eight prominent femoral gland granules per femur readily characterise this species from other Laurentomantis (Figs 3 and 5). The new species is easily distinguished from G. horridus (Boettger 1880), G. malagasius (Methuen and Hewitt 1913) and G. ranjomavo  by lacking tibial glands, its larger number of femoral gland granules and its rugose and granular dorsal texture with prominent ridge elements. Furthermore, the new species is easily distinguished from G. ventrimaculatus, where G. marokoroko has eight distinct femoral gland granules on each thigh (eight irregularly-shaped femoral gland granules in G. ventrimaculatus), by the dark grey and red dorsal colouration (light brown in G. ventrimaculatus) and by lacking blue marbling on the ventral surfaces (Fig. 5).
The most similar species morphologically is G. striatus (Vences et al. 2002), but the new species differs from G. striatus through its larger number of femoral gland granules (8 vs. 3-6), the vertebral stripe is absent or indistinct and short (always distinct in G. striatus), bright red eye (orange-brown in G. striatus) and its prominent and strong ridge elements, as well as the dark grey and red colouration on the dorsum (weak ridge elements and brown and orange colouration on the dorsum in G. striatus).
Bioacoustically, the advertisement call of Gephyromantis marokoroko is similar to other species in Laurentomantis and can be distinguished from all other species in this subgenus through the following combination of continuous call characters: (1) moderately long call duration (1095-1431 ms); (2) 2-4 strongly amplitude-modulated pulses per note; and (3) a note repetition rate of 14-20 notes/s. Gephyromantis striatus, G. malagasius and G. horridus have overlapping call durations with the new    Figure 4. Results of phylogenetic analyses of the concatenated alignment of five mitochondrial and four nuclear markers for Maximum Likelihood (ML) and Bayesian Inference (BI). Topology is a consensus tree from IQ-Tree. On the right, the dots represent markers that were present in blue and absent in orange for each sample. The notes marked with a circle are those that did not receive perfect support (Bootstrap = 100; Posterior Probability = 1.00) from ML and BI, with the support values as BS on top and PP on the bottom. Note that Gephyromantis marokoroko sp. nov. has strong support in both analyses for a sister relationship to G. striatus. species and overlapping note repetition rates, except for G. striatus, which has the fastest note repetition rate (Table 1). Despite these similarities, the clearly pulsed notes alone distinguish the new species from all other Laurentomantis, except G. ventrimaculatus (Angel 1935), which has ca. 5-6 pulses/note; however, G. ventrimaculatus differs by having the shortest call duration non-overlapping with other Laurentomantis species at 407-455 ms and a slightly faster note repetition rate of 21-24 notes/s. Temperature is not likely to be an important factor in the characteristic differences described here, as structural characters, such as clearly defined pulses, would not be affected by temperature (Schneider 1974). Motivation might affect number of notes emitted and, thus, call duration; however, the recording of G. ventrimaculatus is of a highly motivated male (i.e. many calls emitted in a short time) while the call of the new species was recorded from males which did not appear to be very motivated, emitting only 1-2 calls within an hour. Finally, comparisons could not be made to G. ranjomavo as calls were not available; however, the new species is clearly morphologically distinct (see above).
Description of the holotype. Fixed in 10% buffered formalin solution, preserved in 70% ethanol, in good state of preservation, except for skin loss near the anterior dorsum, with left thigh muscle removed for tissue sample. Adult male, SVL 26.0 mm. Body very slender; head longer than wide HL 33.4% of SVL; slightly wider than body, HW 33.7% of SVL; snout of moderate length, ESD 16.2% of SVL; snout rounded in dorsal and later-al view; nostrils directed laterally, slightly protuberant, nearer to snout tip than eye; ED larger than END; canthus rostralis indistinct, concave; loreal region slightly concave; single subgular vocal sac; gular glands absent. Tympanic annulus distinct and round, small, TD 64.5% of ED; supratympanic fold indistinct and irregular, tympanic membrane lighter than ground colouration. Vomerine teeth not visible on the buccal roof, present under mucosal skin; choanae small, rounded. Tongue longer than wide; ovoid in shape, posteriorly bifid. Dermal fold along lower jaw absent. Arms slender, subarticular tubercles single; outer and inner metacarpal tubercles present, indistinct. Fingers without webbing; nuptial pads absent; relative finger length 2 < 1 < 4 < 3; second finger distinctly shorter than fourth finger, only slightly shorter than finger one; finger discs distinctly enlarged, larger on third and fourth finger. Hind limbs slender; lateral metatarsalia connected; inner metatarsal tubercle distinct, outer metatarsal tubercle small, but recognisable; TIBL 55.2% of SVL; FOL 45.2% of SVL. Tibial glands absent. Toes without webbing; relative toe length 1 < 2 < 5 < 3 < 4; toe three distinctly longer and broader than toe five; toe discs distinctly enlarged. Femoral glands large, well delimited, having eight distinct clusters on each femur of almost the same size, in two rows of four. Skin coarsely granular and heavily rugose on dorsal surfaces; large and sharply elevated tubercles and ridges irregularly distributed across dorsal surfaces, with less distinct ridges on the lower back; some ridges are symmetrical, larger tubercles and short ridges present on head and anterior dorsal region. Ventral skin granular on stomach, throat and limbs.
After four years in preservative, dorsal ground colouration is a uniform dull brown including forelimbs and hind-limbs. The red colouration has faded to become light brown. Lighter coloured spots on ventral surfaces are still present.
In life (Fig. 2), dorsal colouration is a dark grey ground colour with thick, bright red mottling distributed on the dorsum. Many of the raised ridges are dark grey with bright red edges. Lighter red stripe present short distance up the dorsum. Lateral head the same as dorsum, tympanum a lighter brown. Flanks are also dark grey, but have less bright red colouration, typically only found on ridges. Forelimbs have same colouration as dorsum, except bright red colouration is more spotted, with a few lighter red spots. Hind-limbs have same colouration as forelimbs, except with red crossbands present on the dorsal surface. A whitish annulus is present before the terminal disc on fingers and toes, fingers and toes light brown. Ventral surfaces brown, with no red present. White and light-yellow spots are present and scattered moderately along the ventre. Ventral sides of arms and hind-limbs brownish-grey, with light red spotting. Femoral glands lighter brown than surrounding limb surfaces. Single subgular vocal sac is light grey, with some light-yellow spotting down the centre. Jaw has scattered light-red spots along the lip. The pupil is black with a bright red iris, with black reticulations around the outer margin of the iris.
Variation. All paratypes resemble the holotype in morphology and colouration. In life, dorsal colouration varies slightly in the amount and intensity of red present. Spotting on the ventral surfaces varies in the colouration of the spots being white, light-yellow, light-orange or light-red. The vertebral stripe varies from being absent in some individuals to indistinct in others. Females lack femoral glands and have a granular texture on the femur.
Morphometry of type series. Measurements of the holotype and paratypes are shown in Table 2.
Bioacoustics. We recorded three calls from two males at Vohidrazana Forest after collection at ca. 02:00 hr on 6 January 2016. Males call infrequently with extremely quiet calls from the upper surfaces of leaves up to 50 cm above the ground. The recorded male was captured and placed in a separate plastic collecting bag. Males would not call when we were within recording distance, so we placed the microphone 100 cm away from the bag near where it was captured and moved several metres away. Calls were recorded during light rain at a temperature of 20.4 °C.
Phylogenetics. The phylogenetic results support the morphological diagnosis by placing Gephyromantis marokoroko within the Laurentomantis subgenus with strong support. At the species level, G. marokoroko is monophyletic with strong support in ML and BI analyses (BS = 100, PP = 1.00; Fig. 3). Uncorrected p-distances, using the 16S fragment, indicate that G. ventrimaculatus has the lowest distance to the new species, at ~ 6-9%. The combined nine marker multi-locus dataset places the new species sister to G. striatus with strong support (BS = 98; PP = 1.00) in both BI and ML analyses (Fig. 4). Overall, these results provide strong evidence that the species is a separately evolving lineage and strong evidence for the new species phylogenetic placement.
Distribution. Gephyromantis marokoroko is known from several sites in the forests in the vicinity of Andasibe, but has only been found at high elevation sites (~ Natural history. Gephyromantis marokoroko is apparently locally rare and, thus far, only found within undisturbed, primary forests at highland elevations (ca. 1000-1200 m). Individuals of the species were perched on the surfaces of vegetation less than 50 cm in height (Fig. 7). The species was infrequently encountered, always after moderate to heavy rain, with multiple individuals occasionally grouped in small clusters (~ 20 m 2 ). The species' call is very quiet and irregular and is barely audible to a human observer, even within three metres of a calling individual. Individuals of the new species were often found syntopically with another Laurentomantis, G. sp. Ca13, which is a candidate species identified in Vieites et al. (2009). Other syntopic Gephyromantis include G. eiselti, G. salegy, G. sp. aff. plicifer (not yet assessed for a candidate species number) and G. cornutus.
Conservation status. The new species is known from Andasibe-Mantadia National Park and several other managed areas (e.g. Vohimana, the community managed Vohidrazana Forest and Tavalobe). However, as currently understood, the distribution of this species is severely fragmented and restricted to only four known high-elevation localities (~ 1000-1200 m), which are very small patches with no connectivity (Fig. 1). Many other high elevation sites in the region have been surveyed by the authors over three field seasons. Furthermore, Vohidra-zana Forest and Tavalobe face ongoing threats that result in the reduction of quality and extent of habitat. For example, slash-and-burn agriculture and forest products are frequently extracted directly from this species' habitats that are outside protected areas. Given this information, we categorise this species as "Endangered" [B1ab(iii-iv)] following IUCN Criteria (IUCN 2001).

Discussion
Gephyromantis (Laurentomantis) marokoroko sp. nov. is a clearly distinct species, as evidenced through morphology, bioacoustics and molecular phylogenetics. The new species can be readily distinguished from other members in Laurentomantis by its heavily rugose granular skin, vibrant red eyes, bright red body colouration and distinctive femoral glands. The call of G. marokoroko also differs from all other Laurentomantis through its moderately long call duration, clearly pulsed notes and slower note repetition rate. Phylogenetic analyses strongly support the new species as monophyletic in the 16S rRNA mitochondrial marker multi-sample dataset. Additionally, the single-sample per species dataset of nine-markers (five mitochondrial and four nuclear) and both phylogenetic analyses strongly supported G. striatus and G. marokoroko as sister species (Fig. 4). In addition, morphological similarity in the number and shape of femoral glands and the occasional presence of vertebral stripe support this relationship.
Gephyromantis marokoroko is a remarkable discovery that was immediately obvious as a new species in the field as its general appearance is very distinct and spectacular, with several clear morphological differences from related species. Distinctive new species are typically discovered in poorly unexplored areas and G. marokoroko was discovered in the well-explored vicinity of Andasibe. In addition, the species had never been barcoded before, eluding past herpetological surveys. This new species highlights the importance of continued fieldwork in Madagascar, as the discovery of previously undocumented new species is occurring frequently (Lambert et al. 2017;Scherz et al. 2017a;, despite the extensive past barcoding efforts for Malagasy frogs (e.g. Vieites et al. 2009;Perl et al. 2014). Typically, such species have low population densities, small geographic or elevational ranges and/ or are in areas that have not been extensively surveyed. Many recent species descriptions of Malagasy frogs are from previously-known candidate species and/or are cryptic lineages that required molecular evidence to diagnose (e.g. Hutter et al. 2015;Scherz et al. 2017c;Vences et al. 2017) and it is uncommon to find new species that have not already been documented through barcoding efforts (e.g. Vieites et al. 2009;Perl et al. 2014). These discoveries are rare and there are only a few recent examples of new discoveries that include G. lomorina  and Boophis masoala ). The continuation of basic field inventories is, therefore, clearly necessary to fully understand the patterns of species richness and complete evolutionary histories of frogs in Madagascar and other tropical regions.
The discovery and conservation of these new and unique species is critically important as habitat loss con-tinues, especially in the study area. The distribution of G. marokoroko is severely fragmented and restricted to only four locations and occurs in small habitat patches (Fig. 1). While the species is protected within Andasibe-Mantadia National Park and Vohimana Special Reserve, it has only been found in low abundance in single, very small habitat patches. The localities Vohidrazana Forest and Tavalobe face ongoing threats from slash-and-burn agriculture and forest products are frequently extracted directly from this species' habitat that are outside protected areas. Furthermore, climate change could exacerbate these risks reducing further the suitable habitat for this already micro-endemic species.
are especially grateful to Centre ValBio, John Cadle and Eileen Larney for their logistical help and hospitality. We also thank various friends and colleagues for logistic, field and laboratory help: James Herrera, Russ Mittermeier, Molly Bletz, Asa Conover, Barbara Korten, Vincent Premel, Devin Edmonds, Gaga and Regis.
Author contributions CRH and SML conceived the study. CRH wrote the first draft of the manuscript and the other authors provided input. ZFA collected phenotypic data. All the co-authors were involved with fieldwork, data collection and logistics.