Genetic diversity of the mantellid frog Blommersia blommersae, and description of a new anuran species from south-eastern Madagascar

ABSTRACT A range-wide assessment of genetic variation in one mitochondrial (16S rRNA) and one nuclear-encoded (Rag-1) gene fragment of the widespread Madagascar frog Blommersia blommersae revealed the presence of up to 12 deep genetic lineages. Many of these differed by genetic distances >3% in the 16S gene. In the Ranomafana area in the southern central east of Madagascar, two mitochondrial lineages differing by 1.2–1.8% in 16S occurred in close syntopy across multiple sites. A phylogeny of representative samples based on multiple mitochondrial genes supported three main mitochondrial clades within B. blommersae, with lineages from the (i) north, (ii) north and northern central east, and (iii) southern central east, respectively. In addition, one lineage from Sainte Luce in the south-east was sister to all other species of Blommersia, and thus clearly not belonging to B. blommersae, to which it has been tentatively assigned previously. Except for the Sainte Luce population, most of the mitochondrial lineages did not show a concordant and consistent differentiation in Rag-1 and no obvious morphological differences were detected among the lineages. We therefore refrain from taxonomic conclusions for them at this time, although indications exist that the populations from the northern central east and southern central east may show certain differences in relative note duration of advertisement calls, which require further study. However, the Sainte Luce individuals were highly divergent in 16S (>10%) and Rag-1, phylogenetically isolated, and characterised by differences in colour pattern. This lineage represents the most evolutionarily divergent species of Blommersia known to date and is herein formally named and described as Blommersia dupreezi sp. nov.


Introduction
Among the anuran family Mantellidae, endemic to Madagascar and the Comoran island Mayotte, the genus Blommersia contains 12 species of small-sized and semi-arboreal frogs (AmphibiaWeb 2022) that typically breed in swamps, but with some species specialised to stagnant side pools of rainforest streams or phytotelmic habitats in fallen palm leaves or water-filled tree holes (Andreone et al. 2010;Vieites et al. 2020). While some species of Blommersia are microendemic to small ranges (e.g. Vences et al. 2010;Pabijan et al. 2011), others are widespread across Madagascar, such as B. wittei in the north-west and north, a still scientifically unnamed candidate species provisionally named B. sp. Ca5 (Vieites et al. 2009;Perl et al. 2014) in the west, and B. blommersae which is known from the entire eastern rainforest band of Madagascar, from the southeast to the north .
The eastern widespread B. blommersae is a species living in and near swamps, in or at the edges of rainforest, primarily at mid-elevations, where its loud chirp calls can be commonly heard during the rainy season. Similar to several other scansorial mantellids, B. blommersae deposits its eggs on leaves overhanging the water, and the exotrophic tadpoles drop into the water where they complete their development (Blommers-Schlösser 1979). As with many other widespread amphibians in Madagascar, molecular work has shown a substantial genetic differentiation among populations, with several deep conspecific lineages (e.g. Vieites et al. 2009;Rodríguez et al. 2015), but these have not yet been compared in depth. Given that the various mitochondrial lineages assigned to B. blommersae have sometimes been paraphyletically arranged in single-marker trees (e.g. Vieites et al. 2009), a taxonomic revision is needed to understand if B. blommersae might be a species complex.
Here, we undertake a first analysis of the genetic variation of B. blommersae across its range. Based on DNA sequences of one mitochondrial and one nuclear-encoded gene, plus a multi-gene mitochondrial phylogeny, we aim to test whether (i) the genetically divergent lineages assigned to B. blommersae represent a monophyletic group, (ii) these lineages are allopatrically distributed in geographically coherent groups of populations or co-occur at the same sites, and (iii) one or several of these lineages may represent new, distinct species.

Materials and Methods
This study is based on voucher specimens collected during field expeditions in Madagascar between 2000 and 2013, and tissue samples taken from the preserved specimens. Frogs were caught during nocturnal searches, typically by locating calling males in swamps. We anesthetised specimens by immersion in MS222 or chlorobutanol solution, and subsequently euthanised them with an overdose of the same substances. We removed tissue samples for molecular analysis and stored them separately in 1.5 ml vials with pure ethanol. Voucher specimens were subsequently fixed with 4% formalin or 95% ethanol and then preserved in 70% ethanol. Vouchers were deposited in the following collections: Zoologische Staatssammlung München (ZSM), Zoological Museum Amsterdam (ZMA; collections now in Naturalis, Leiden), and the Université d'Antananarivo, Département de Biologie Animale (UADBA). Additionally, type material from the Muséum National d'Histoire Naturelle, Paris (MNHN) was studied. We herein use FGZC, FGMV and ZCMV to refer to field numbers of F. Glaw and M. Vences. A full list of all field numbers and museum catalogue numbers, as well as sequences and sequence accession numbers, is available as Microsoft Excel and tab-delimited table from the Zenodo repository under the DOI 10.5281/zenodo.7416008 (note that most vouchers deposited in UADBA have not been catalogued in that collection and therefore have no final catalogue numbers yet, and numerous sequences from GenBank are from unvouchered tissue samples).
A manual caliper was used to take the following measurements (all by MV) at an accuracy of 0.1 mm: snout-vent length (SVL); maximum head width (HW); head length from tip of snout to posterior edge of snout opening (HL); horizontal tympanum diameter (HTD); horizontal eye diameter (HED); distance between anterior edge of eye and nostril (END); distance between nostril and tip of snout (NSD); distance between both nostrils (NND); fore-limb length, from limb insertion to tip of longest finger (FORL); hand length, to the tip of the longest finger (HAL); hind limb length, from the cloaca to the tip of the longest toe (HIL); foot length (FOL); foot length including tarsus (FOTL); and tibia length (TIBL). Geographical regions within Madagascar were named according to Boumans et al. (2007).
Vocalisations were recorded in the field using different types of tape recorders (Tensai RCR-3222, Sony WM-D6C) with external microphones (Sennheiser Me-80, Vivanco EM 238), and with a digital recorder with built-in microphones (Edirol R-09), and saved as uncompressed wave format. Recordings were sampled or re-sampled at 22.05 kHz and 32-bit resolution and computer-analysed using the software Cool Edit Pro 2.0. Most call recordings were taken from Vences et al. (2006). We obtained frequency information through Fast Fourier Transformation (FFT; width 1024 points) at Hanning window function. Spectrograms were produced with Blackman window function at 256 bands resolution. Sensitive filtering was applied to remove background sounds, only to frequencies outside the prevalent bandwidths of calls. Temporal measurements were summarised as range with mean ± standard deviation in parentheses. We followed the recommendations of Köhler et al. (2017) for analysis, description and terminology, using the note-centred terminological scheme.
For molecular study we extracted DNA from ethanol-preserved tissue samples using a standard salt extraction protocol (Bruford et al. 1992). A list of all primers and polymerase chain reaction (PCR) protocols used is included in Supplementary Table S1. We DNA barcoded available and not previously sequenced tissue samples assigned to B. blommersae for a fragment of the mitochondrial 16S rRNA gene (16S) that spans about half of the gene at its 3' terminal portion and that has regularly been used for molecular taxonomy of Malagasy amphibians (e.g. Vieites et al. 2009), including Blommersia (e.g. Glaw and Vences 2002;Vences et al. 2010, Pabijan et al. 2011Glaw et al. 2019;Vieites et al. 2020). The dataset was complemented with sequences available for GenBank.
We furthermore sequenced for most samples a fragment of the nuclear-encoded recombination activating gene 1 (Rag-1).
Lastly, to reconstruct reliably the (mitochondrial) phylogeny of lineages within the genus Blommersia, we assembled a concatenated data set of five fragments of four mitochondrial genes, i.e. two adjacent fragments of the 16S rRNA gene (corresponding roughly to the fragments at the 3' and 5' terminus, together making up almost the entire gene), a fragment of the 12S rRNA gene, and a fragment each of the cytochrome b (cob) and cytochrome oxidase subunit 1 (cox1) genes. For this concatenated data set we combined sequences for representatives of the main mitochondrial lineages of B. blommersae with sequences of all other species of Blommersia plus a series of outgroups.
We used MEGA7 (Kumar et al. 2016) to align sequences with the Clustal algorithm. From the sequences of the 3' fragment of the 16S gene, three different alignments were prepared. Firstly, for an initial assessment of genetic variation, one alignment (16SALL) contained all available sequences of specimens assigned to B. blommersae plus one outgroup (Guibemantis pulcher). Secondly, to avoid possible artefacts caused by incomplete or only partially overlapping sequences in species delimitation, we prepared two trimmed alignments (16STRIM1 and 16STRIM2) without outgroups, where all sequences had approximately similar lengths (differing by a maximum of 10 bp among each other): 16STRIM1 was prepared to keep the maximum number of sequences, including some rather short ones; it contained 183 16S sequences for an alignment length of 353 bp. 16STRIM2 was prepared to keep the maximum alignment length while removing all shorter sequences, but keeping at least one sequence per collecting site; it contained 137 sequences for an alignment length of 507 bp. All alignments were submitted to the Zenodo repository (DOI 10.5281/zenodo.7416008).
A phylogeny from the 16SFULL alignment was inferred under the Maximum Likelihood (ML) optimality criterion in MEGA7, after selecting the most appropriate substitution model based on the Bayesian Information Criterion. Node support was assessed with 500 bootstrap replicates.
For the multigene analysis, we first concatenated sequences using Concatenator (Vences et al. 2022) and then ran a partitioned ML analysis (Chernomor et al. 2016) using IQtree (Nguyen et al. 2015) with 500 bootstrap replicates. Nine character sets were defined, one each for 12S rRNA and the two 16S rRNA fragments as well as a separate set each for the three codon positions of cob and cox1, and substitution models inferred using the ModelFinder function of IQtree (Kalyaanamoorthy et al. 2017).
For graphically representing the relationship among alleles (haplotypes) of the Rag-1 gene fragment, we used a network approach. Due to the presence of heterozygous sites, haplotypes were inferred with the PHASE algorithm (Stephens et al. 2001) implemented in the DnaSP software (Version 5.10.3; Librado and Rozas 2009). We inferred a ML tree inferred under the Jukes-Cantor substitution model in MEGA7 (choosing this simple model to avoid over-parametrisation), and used this tree together with the respective alignment used as input for Haploviewer (written by G.B. Ewing; http://www. cibiv.at/~greg/haploviewer), a software that implements the methodological approach of Salzburger et al. (2011).
Pairwise distances between groups were calculated based on the TRIM2 dataset of 16S sequences, using TaxI2 (Vences et al. 2021b). For species delimitation, we used ASAP (Puillandre et al. 2021) on the TRIM1 and TRIM2 alignments to carry out an initial assessment of species partitions based on the 16S (3' fragment) data and subsequently compared the favoured ASAP partition for concordance with evidence from Rag-1 differentiation.

Molecular phylogenetics and differentiation
The 16SALL alignment of the 3' fragment of the 16S gene consisted of 532 bp for 195 ingroup sequences plus one outgroup. The ML tree inferred from this data set ( Figure 1) under a T92 + G substitution model revealed a highly structured topology, with sequences from many localities forming separate mitochondrial clades. Sequences from the locality Sainte Luce in the south-east were placed sister to all other sequences assigned to B. blommersae. Most of the deeper nodes in the tree received only negligible bootstrap support, suggesting that this short 16S fragment alone is unable to reliably reconstruct the relationships among the main clades.
The ten alternative species partitions suggested by ASAP for the alignment 16STRIM1 contained 2-38 subsets, with ASAP scores between 3.5-9.5. The most conservative partition with only two subsets separated the sequences from the Sainte Luce population from all others. The lowest (best) ASAP score of 3.5 characterised a partition consisting of 12 subsets that were distributed largely in allopatry. The analysis based on 16STRIM2 suggested partitions with 2-16 subsets, with ASAP scores ranging from 2.0-9.0. The preferred partition with a score of 2.0 had 13 subsets, in agreement with those suggested from the 16STRIM1 alignment except placing a single sequence from Ambohitantely (ZCMV 5683) into a separate subset. For downstream analysis we used the 12-subset partition from 16STRIM1, considering that the deviant Ambohitantely haplotype is in need of confirmation, and named the subsets as (mitochondrial) lineages according to their geographic occurrence ( Figure 2) in the North (NOR1, NOR2), Northern Central East (NCE1, NCE2, NCE3), Southern Central East (SCE1, SCE2, SCE3, SCE4, SCE5, SCE6) and South East (SE). Distribution of these lineages was particularly convoluted in the southern central east, where many nearby localities (e.g. Vevembe, Manombo, Beronono) had their own lineages (but with very few, sometimes only single, samples available), and where the only case of syntopic co-occurrence of two lineages was observed (lineages SCE1 and SCE2) in the Ranomafana area. The type locality of B. blommersae (Andasibe) is in the range of the lineage NCE1.
The Rag-1 haplotype network ( Figure 3) reconstructed from 371 bp sequences of 116 individuals separated the three haplotypes of the SE specimens by 13 mutational steps from all others. The remaining network (containing sequences of all lineages except SCE6) revealed a complex structure with a certain geographic pattern where most alleles of the mitochondrial lineages SCE1 and SCE2 from Ranomafana cluster close to each other, and samples of NCE1 forming two distinct, coherent clusters. However, haplotype sharing was extensive, for instance between SCE1 and SCE2, but also between several other lineages, with one allele shared by individuals of as many as six mitochondrial lineages. The two main Rag-1 clusters containing individuals of the mitochondrial NCE1 lineage were placed in opposite parts of the network, with as many as 11 mutational steps separating them.
The concatenated alignment for multigene phylogeny inference consisted of 3200 bp of five fragments of four mitochondrial genes, for which ModelFinder selected the following substitution models: 12S rRNA: GTR + F + I + G4, 16S rRNA 3': GTR + F + I + G4, 16S rRNA 5': TIM2 + F + I + G4, cox1 (1st): TNe + I + G4, cox1 (2nd): F81 + F, cox1  Table 1. Minimum and maximum uncorrected pairwise distances among main lineages of Blommersia blommersae in percent, calculated from a 507 bp alignment of 137 sequences. Grey cells contain minimum and maximum distances found within the respective lineages. NA, not applicable (in cases of only a single sequence per lineage).   All other mitochondrial lineages of B. blommersae formed a clade sister to the species B. galani, B. dejongi and B. variabilis that are microendemics from low elevations along the eastern coast of Madagascar. This B. blommersae clade however received no significant bootstrap support (43%); within it, the Sorata specimen (lineage NOR1) was sister to a clade of all other B. blommersae which received 100% bootstrap support. This suggests that the taxonomic identity and species assignment of the NOR1 lineage will require additional scrutiny in future studies. Within B. blommersae, not considering the highly divergent SE lineage and the NOR1 Figure 4. Maximum Likelihood tree calculated from a multigene dataset (3200 bp) of the mitochondrial genes for 12S rRNA, 16S rRNA (two fragments), cytochrome b (cob) and cytochrome oxidase subunit 1 (cox1). The tree was inferred using a partitioned approach with IQtree, and numbers at nodes are support values from a bootstrap analysis (500 replicates).

Morphological differentiation
Non-molecular information on the various populations of B. blommersae is scarce. Morphologically, specimens from all mitochondrial lineages agree with the original description of B. blommersae (Guibé 1975) in their comparatively small size, lateral metatarsalia closely connected by muscular tissue (not by only webbing), and only rudimentary webbing on feet. Colouration is inconspicuous, typically with a light brown dorsum and sometimes with a thin light vertebral line or a broader vertebral stripe ( Figure 5). The tympanic region is dark brown in most or all specimens (often not clearly visible in active specimens at night whose colour often appears paler and less contrasted). Specimens from most populations have typically a white lateral stripe that starts as an incomplete frenal stripe (beginning underneath the eye) that is very contrasted underneath the dark brown tympanic region and runs along the flank where it often fades in the posterior third of the body ( Figure 5). In specimens from the Sainte Luce population (SE lineage) this incomplete white frenal stripe, however, stops already before the fore-limb insertion and instead, a dark brown/blackish stripe runs along the flanks as a continuation of the dark brown tympanic patch ( Figure 6).

Bioacoustics of Blommersia blommersae
Only few recordings of vocalisation from populations referred to B. blommersae are available, and none of these comes from specific, genotyped individuals. Also, in some cases it is unclear whether calls were emitted from highly motivated males, and thus the recordings from different sites may not be fully comparable. All recorded calls are loud and conspicuous chirps.
All recordings contain calls composed of single notes and calls composed of multiple notes, mostly two-note calls (Figures 7-12). These multinote calls are herein considered to represent regular advertisement calls, whereas single-note calls are considered to be barely motivated calls or calls with territorial function. In the analyses below we therefore focus on the multinote-calls and largely ignore the single-note calls.
Calls recorded at Andasibe on 2 February 1994 (20:15 h) at 21.4°C air temperature (Figure 7) consist of two pulsed notes repeated at fast succession. In most calls, both notes of the call have almost the same duration, with few exceptional cases where the first note is slightly shorter than the second. Pulses are clearly separated and countable. Call energy is distributed in a broad frequency band. Numerical parameters of 16 analysed calls from at least four individuals are as follows: call duration 250-300 ms (268.1 ± 16.5 ms); note duration 96-150 ms (109.3 ± 15.5 ms); inter-note interval 48-61 ms (53.0 ± 4.3 ms); pulses/note 20-33 (28.8 ± 4.0); dominant frequency 5318-5956 Hz (5623 ± 272 Hz); prevalent bandwidth 3500-7800 Hz. Calls were repeated at very irregular intervals, not allowing for a reliable calculation of call rate.
Calls recorded at Ranomafana on 29 February 1996 (21:30 h) at 22.0°C air temperature ( Figure 10) consist of two short pulsatile notes, with the first note being significantly shorter than the second note (first note about ⅗ the duration of second note). Although pulsatile in character (not tonal), no clearly separated pulses are recognisable within notes and to the human ear calls are rather 'noisy' in character. Numerical parameters of four analysed calls from three individuals are as follows: call duration 154-164 ms (159.8 ± 4.6 ms); note duration 45-89 ms (65.1 ± 20.2 ms); inter-note interval 30-32 ms (31.5 ± 1.0 ms); dominant frequency 6010-6279 Hz (6098 ± 101 Hz); prevalent bandwidth 3000-8500 Hz.
Calls recorded at Vevembe on 9 February 2004 (18:20 h) at 23.0°C air temperature ( Figure 12) consist of short single-note calls only. These calls may therefore not represent regular advertisement calls, but the single notes in the short recording available are repeated at rather regular succession. The notes are pulsed, with pulses being partly fused but countable. Numerical parameters of five analysed calls from one individual are as follows: call duration (= note duration) 42-60 ms (54.3 ± 8.3 ms); inter-call interval 453-571 ms (498.7 ± 63.4 ms); pulses/note 14-21 (17.7 ± 3.5); dominant frequency 6352-6589 Hz (6468 ± 87 Hz); prevalent bandwidth 4000-9500 Hz. Although available data on calls of certain clades of B. blommersae are too sparse to allow for conclusive interpretations, our bioacoustic analysis at least provides the following indications. Calls from Andasibe and Ankeniheny are almost identical and are here both considered to represent subclade NCE1. Calls from Ranomafana and Vohiparara (SCE1/SCE2) have a tendency to differ from NCE1 calls by having a slightly shorter first note, while the two notes in NCE1 calls are of similar duration. Calls from Anjanaharibe (from a population not studied with molecular methods, but possibly belonging to NOR2) are rather similar to those of NCE1 when considering note structure, but differ by containing a higher maximum number of notes per call. With reference to the multigene tree (Figure 4), this could indicate that the difference in relative note duration separates lineages in the SCE clade from those in the NCE/ NOR2 clade.
Most interesting are the differences among calls from Ranomafana and Vohiparara, where both calls exhibit a similar duration and overall structure, with the first note being shorter than the second note, but differ significantly by pulsatile 'noisy' (Ranomafana) versus clearly pulsed (Vohiparara) amplitude structure in notes. This difference is also clearly evident to the human ear. Given that one call may correspond to subclade SCE1 and the other call to SCE2, confirmation of the herein identified differences in advertisement calls of the two, partly syntopic, subclades in B. blommersae, would indicate reproductive isolation (possibly driven by character displacement in calls)which however is not supported by the widespread haplotype sharing in Rag-1 (Figure 3).

Taxonomic conclusion
Taken together, the molecular data provide evidence for substantial genetic variation among populations that have traditionally been assigned to B. blommersae. For populations from the north, northern central east, and southern central east, the data draw a rather complex picture, with multiple deeply divergent lineages forming three main mitochondrial clades but extensive haplotype sharing among these lineages in the nuclear-encoded gene fragment studied. From the limited material available for study, we also did not detect substantial morphological or bioacoustic differences between these lineages, and their status therefore remains in need of further revision. However, there are multiple lines of evidence suggesting that the individuals studied from Sainte Luce are distinct at the species level. They are very highly divergent in mitochondrial and nuclear DNA, represent a phylogenetically isolated lineage that is sister to all other Blommersia, and differ in colour pattern from the other lineages of B. blommersae studied, including NCE1 whose range includes the type locality of the nominal species, Andasibe. We consequently consider the Sainte Luce specimens representing a new, scientifically unnamed species, which we formally name and describe below.

Diagnosis
A species of the genus Blommersia in the subfamily Mantellinae based on presence of intercalary elements between penultimate and ultimate phalanges of fingers and toes (verified by external examination), occurrence in Madagascar, small body size (female SVL < 25 mm), absence of femoral gland rudiments in females, head distinctly longer than wide, and molecular phylogenetic relationships. The presence of femoral glands in males is probable but could not be verified due to the lack of adult male specimens. From other species of Blommersia, the new species is distinguished as follows: from B. angolafa, B. galani, B. dejongi, B. grandisonae, B. nataliae, B. transmarina, and B. wittei, by having lateral metatarsalia connected by dense tissue (vs. separated by webbing), only traces of foot webbing (vs. weakly developed but distinct webbing), and additionally, from B. angolafa, B. galani and B. grandisonae by the presence of vomerine teeth (vs. absence); from B. variabilis, by showing only traces of foot webbing (vs. weakly developed but distinct webbing), and by lateral metatarsalia completely connected by dense tissue (vs. separated at least partly by webbing in many individuals); from B. domerguei, B. kely, and B. sarotra by the presence of vomerine teeth (vs. absence), and in addition from B. domerguei and B. kely by largely different colour pattern. Morphologically the new species is most similar to B. blommersae but differs by the presence of vomerine teeth (vs. absence) and the colour pattern with distinct dark brown stripe running along the flanks (vs. white lateral stripe).
The taxonomic status and distinctive character states of Blommersia grandisonae (Guibé, 1974) require a short comment. This species was described from Ambana in the Anosy mountains, geographically relatively close to Sainte Luce, and the type series is also characterised by dark stripes running along flanks. Molecular data from topotypical B. grandisonae have so far not become available. It would be appealing to hypothesise the new species described herein corresponds to B. grandisonae, but several characters of the type series (including the holotype), examined by us in 2021, conclusively contradict this hypothesis, in particular the presence of webbing, separated lateral metatarsalia, and absence of vomerine teeth. The identity and status of B. grandisonae will be the subject of another forthcoming study.

Description of the holotype
Specimen in a good state of preservation ( Figure 13). Tongue removed as tissue samples for molecular analysis, a small ventral cut on the left body side for gonad examination. SVL 23.2 mm, for further measurements see Table 2. Body slender; head longer than wide, slightly narrower than body (which is rather wide due to the presence of mature oocytes in the body cavity); snout rounded in lateral view, obtusely pointed in dorsal and ventral views; nostrils directed laterally, protuberant, nearer to snout tip than to eye; canthus rostralis indistinct, straight; loreal region slightly concave; tympanum distinct, round, its diameter 72% of eye diameter; supratympanic fold distinct, curved above tympanum where it follows the tympanum outline, straight from tympanum to almost the point of fore-limb insertion where again it curves downwards; tongue absent, its shape therefore not ascertainable; vomerine teeth present, distinct, forming an ovoid agglomeration posteriomedial to each choana; choanae small, round, located toward the front of the palate; maxillary teeth present. Arms slender, subarticular tubercles present, distinctly protuberant, single; fingers without webbing; relative length of fingers 1 < 2 < 4 < 3 on both hands, first finger quite long and only slightly shorter than second finger; finger discs distinctly enlarged, nuptial pads absent. Hind-limbs slender; tibiotarsal articulation reaches tympanum when the hind-limb is adpressed along the body; lateral metatarsalia closely connected by tissue; inner and outer metatarsal tubercles distinct; only traces of webbing between toes; relative length of toes 1 < 2 < 5 < 3 < 4. Skin on the upper surface smooth, without folds or ridges. Ventral skin smooth. Body cavity filled with oocytes containing a light and a dark pole (ascertained by partial dissection).
After 17 years in preservative (Figure 13), the dorsum is uniformly brown, with few scattered irregular dark brown flecks. A distinct, sharply bordered dark brown stripe runs from the tip of the snout to the eyes, and from the eye along the flanks to the inguinal region, covering the entire tympanum and dissolving into a series of small brown spots in the posterior third of the body. An indistinct light frenal stripe is visible, running from underneath the eye towards the fore-limb insertion. The ventral side is uniformly light, with dark colour corresponding to the liver and oocytes shining through the belly skin. Some fine brownish spotting is present on throat and chest. Ventral side of fore-and hind-limbs with more intense brown colour. In life ( Figure  6A), colouration was similar, with indistinct and poorly contrasting dark brown markings visible on the dorsum, including a V-shaped dark brown marking in the interorbital region, and colour on the flanks gradually changing from a brown dorsal to a beige ventral colouration; the dark brown lateral stripe therefore stands out highly contrasted. The frenal stripe is white and also strongly contrasting. Dorsal side of hind-limbs with a pattern of small dark brown spots.

Variation
The holotype and paratype ZSM 210/2005 have well-developed oocytes in their body cavity, recognisable by external view (verified through partial dissection in the holotype); paratype ZSM 214/2005 has smaller, developing oocytes as verified by dissection. The other paratypes are considered subadult; no femoral glands are visible in external or internal view (according to Glaw et al. 2000; verified in three specimens through dissection), and gonad examination in one specimen did not yield any indication of maturity.

Natural history and distribution
Specimens were sitting in the leaf litter on the ground of forest fragments or between low vegetation in more open habitats. The species is currently only known from the type locality, the Sainte Luce littoral forest in south-eastern Madagascar, although it might be regionally more widespread in littoral and low-elevation forest habitats. However, no Blommersia species similar to B. dupreezi has so far been recorded from protected areas near Sainte Luce, like Tsitongambarika or Andohahela (Birdlife International 2011; Raselimanana et al. 2018).

Etymology
The species epithet is a patronym honouring Louis du Preez (North-West University, Potchefstroom, South Africa), parasitologist and herpetologist, in recognition of his substantial contributions to understanding the polystome parasites of Malagasy anurans.

Discussion
This study provides additional evidence for deep phylogeographic structure in supposedly widespread species of Madagascar's herpetofauna. Other, similar examples include the frogs Gephyromantis luteus and Guibemantis liber (Vences et al. 2021a), and reptiles such as the chameleons Brookesia superciliaris, Furcifer lateralis and F. pardalis, or the geckos Paroedura gracilis, Phelsuma guttata, Uroplatus phantasticus, and many others (Boumans et al. 2007;Ratsoavina et al. 2010Ratsoavina et al. , 2012Florio et al. 2012;Grbic et al. 2015;Florio and Raxworthy 2016;Mohan et al. 2019). In some cases, the distinct phylogroups have been recognised and named as different species (e.g. Florio et al. 2012) while in other cases, for now, they are considered provisionally as deep conspecific lineages. Leaving aside the highly divergent Blommersia dupreezi, described herein, which simply was a morphologically cryptic species unrelated to B. blommersae, we currently find the evidence in this cluster of lineages inconclusive for taxonomic decisions. It is possible that the three main mitochondrial clades, NOR1 vs. NOR2/NCE vs. SCE (Figure 1) represent distinct species, given their high mitochondrial divergences. On the other hand, the fact that the Rag-1 sequences from the NCE1 lineage are found in two strongly different clusters of the haplotype network, suggest that perhaps the mitochondrial data may be misleading due to phenomena of mitochondrial capture via introgressive hybridisation. Unfortunately, only a limited number of genotyped voucher specimens and no vouchered call recordings of B. blommersae are available because this species has usually been considered as widespread and well known, and thus not of priority for taxonomic study.
The phylogenetic position of B. dupreezi is surprising. Its isolated position was already suggested by the 16S tree of Vieites et al. (2009) but since that tree was based on an alignment of hundreds of sequences for a rather short DNA fragment, we suspected that this placement was artefactual, given the limited phylogenetic resolving power of this fragment (Chan et al. 2022). The multi-gene phylogeny presented here revealed this species indeed is sister of all other Blommersia, and is separated by a rather long branch from all other species. This demonstrates that small and morphologically extremely inconspicuous species of frogs can be of high importance to understand the diversity and evolution of tropical anurans. A similar situation was found with Wakea madinika, the sole representative of its genus , which morphologically is similar to Blommersia but in the mantellid tree forms the sister group of the bright-coloured Madagascar poison frogs of the genus Mantella. The discoveries of B. dupreezi and W. madinika also exemplify the usefulness of performing collections of tissues and voucher specimens even of species that are considered widespread and without immediate scientific interest, given that they might turn out to be cryptic lineages representing unexpectedly deep evolutionary histories.
Blommersia dupreezi represents a large amount of evolutionary history worth conserving (e.g. Sechrest et al. 2002), but it is only known from a single site, Sainte Luce, a small patch of littoral forest in the south-east of Madagascar. While we consider it to be likely that the species may also be present in other littoral forest fragments in the region, it remains true that this is one of the most threatened ecosystems in Madagascar (Ingram and Dawson 2006), which hosts a large number of endemic species of herpetofauna (e.g. Nussbaum 1993, 1994;Lehtinen et al. 2011;Scherz et al. 2019). Along with other endemic species of Madagascar's south-eastern forests such as the frogs Guibemantis annulatus and G. wattersoni, an IUCN Red List classification of Critically Endangered may be adequate for B. dupreezi based on IUCN criterion B1ab(iii) (IUCN 2012(IUCN , 2017, given the probably small extent of occurrence (EOO), and an ongoing decline in the quality and extent of its habitat. A reliable estimate of the species' EOO is not possible with current data, given that it is known only from a single site; however, the entire remaining littoral forest of the south-east of Madagascar has an extent of only 28.4 km 2 , and the largest parcels of Sainte Luce Forest together sum up to just 10.9 km 2 (Bollen and Donati 2006). Hence, if the species is indeed restricted to this habitat type, its EOO is exceedingly small.
Alternatively, given its cryptic appearance and the poor knowledge about its distribution range, a classification as Data Deficient could also be advocated. However, given the relatively Madagascar-wide provenance of samples of B. blommersae in our study, we consider it as relatively unlikely that B. dupreezi would be substantially more widespread across the island. Independent from the threat status it might be assigned, the presence of this evolutionarily highly distinct species in one of Madagascar's south-eastern littoral forests supports the hypothesis that this habitat hosts numerous additional rare or scientifically unnamed species, many of which might be microendemic to this ecosystem (Funnell et al. 2012;Hyde Roberts et al. 2016). The conservation, but also the continued biological exploration of this ecosystem should therefore be given emphasis. These permits include authorisation for euthanising and collecting the respective specimens; additional ethics clearance is not necessary for these activities under Malagasy law. We thank the following colleagues, students and guides who assisted during fieldwork: P Bora, P-S Gehring, M Pabijan, JL and C Patton, E Rajeriarison, A Rakotoarison, T Rajoafiarison and FM Ratsoavina. We are grateful to M Kondermann and G Keunecke for help with laboratory work. This study took place in the framework of collaboration accords among the authors' institutions, the Department of Animal Biology of the University of Antananarivo, and the Ministry of the Environment of the Republic of Madagascar. Financial support for field work was granted by the Volkswagen Foundation and Deutsche Forschungsgemeinschaft.