There was a total of 43 specimens that were collected throughout our biodiversity surveys in Guizhou, Guangxi, Yunnan and Chongqing, China, from 2017 to 2024. Six specimens were identified as L. yunyangensis from Qiyaoshan Nature Reserve and Lianhua Village, Renhe Town, Yunyang County, Chongqing; six were L. alpina from Wulian Mountain, Jingdong County, Yunnan; one was L. ventripunctata from Jiangguin Village, Lingyun County, Guangxi; eight were L. suiyangensis from Huoqiuba Nature Reserve, Suiyang County, Guizhou; 12 were Leptobrachella sp1 and Leptobrachella sp2 from Wuliang Mountain, Jindong County, Yunnan; three were Leptobrachella sp3 from Shibing County, Guizhou; and seven were an undescribed species from Xianyuan Town, Xishui County, Guizhou. All of the specimens used for morphological studies were fixed in a 10% formalin buffer and then transferred to 75% ethanol and stored at the Animal Ecology Laboratory of Guizhou Normal University (GZNU), Guiyang City, Guizhou Province, China.

The catalogue and distribution of Leptobrachella were reorganised through field surveys, literature reviews and database searches using Amphibian Species of the World (https://amphibiansoftheworld.amnh.org/) (Frost 2024) and GBIF (https://www.gbif.org/). Both databases were accessed on 13 August 2024. After that, the distribution sites for every species were rasterised to a 0.5° × 0.5° resolution (~ 111 × 111 km²) grid system using ArcGIS v.10.4 and species richness was obtained by summing the total number of species occurring in the grid cells.

Genomic DNA was extracted from muscle tissues using a DNA extraction kit (Tiangen Biotech Co., Ltd., Beijing, China). We referred to prior studies (Chen et al. 2018; Luo et al. 2022b) that amplified and sequenced the mitochondrial 16S ribosomal RNA gene (16S rRNA) and six nuclear gene fragments, i.e., brain-derived neurotrophic factor (BDNF), sodium/calcium exchanger 1 (NCX1), neurotrophin 3 (NTF3), recombination activating gene 1 (RAG1), rhodopsin (RHOD) and solute carrier family 8 member 3 (SLC8A3) (Suppl. material 6). PCR amplifications were performed in a 20 μl reaction volume with the following cycling conditions: an initial denaturing step at 95 °C for 5 min, 35 cycles of denaturing at 95 °C for 1 min, annealing at 50–57 °C for 1 min and extending at 72 °C for 1 min, followed by a final extension at 72 °C for 10 min. The products were sequenced on an ABI Prism 3730 automated DNA sequencer at Chengdu TSING KE Biological Technology Co. Ltd (Chengdu, China). All of the newly-obtained sequences have been submitted to GenBank (Suppl. material 7).

We used a total of 296 sequences and constructed two sequence matrices for phylogenetic analyses, i.e. mitochondrial 16S rRNA (dataset 1) and the combined sequences of six nuclear genes (dataset 2). Multiple sequence alignment was performed using MAFFT v.7.4 (Katoh and Standley 2013) within PhyloSuite v.1.2.3 (Zhang et al. 2020) and checked using MEGA v.7.0 (Kumar et al. 2016) to rule out possible errors. Partitionfinder v.2.1.1 (Lanfear et al. 2017) was used to select the best-fit partitioning and nucleotide substitution model for the two datasets, based on the Bayesian information criterion. In dataset 2, every gene fragment was pre-set as an independent partition. We referred to Chen et al. (2018) to select Leptobrachium boringii, Leptobrachium huashen, and Megophrys glandulosa as outgroups (Suppl. material 7).

Phylogenetic trees were reconstructed using Bayesian Inference (BI) and Maximum Likelihood (ML) methods, based on best-fit partitioning and nucleotide substitution models. The BI analysis was performed using MrBayes v.3.2.1 (Ronquist et al. 2012). Each BI analysis was run independently using four Markov Chain Monte Carlo chains (three heated chains and one cold chain) starting with a random tree; each chain was run for 2 × 10⁷ generations and sampled every 1000 generations. Convergence of the data runs was confirmed when the average standard deviation of split frequencies was less than 0.01. The ML analysis was performed using IQ-tree v.2.0.4 (Nguyen et al. 2015), based on the best-fit model with 10000 ultrafast bootstrap (UFB) replicates (Hoang et al. 2018). The ML analysis was performed until a correlation coefficient of at least 0.99 was achieved. Nodes were considered to be highly supported when the Bayesian Posterior Probability (BPP) value for the BI analysis was > 0.95 (Ronquist et al. 2012) and the UFB value for the ML analysis was > 95% (Hoang et al. 2018). Genetic distances were calculated based on dataset 1 using the uncorrected p-distance model and 1000 bootstrap replications in MEGA v.7.0 (Kumar et al. 2016).

The sampling of species, molecular dating and phylogenetic reconstruction were performed in BEAST v.1.8.2 (Drummond et al. 2012) using dataset 1. Two calibration nodes were employed for the estimation of divergence times with reference to recent studies: (1) the crown age for the family Megophryidae was ca. 51.5–57.7 million years ago (Ma) (Mahony et al. 2017); the most recent common ancestor of the genus Leptobrachella occurred at 31.55 Ma (95% highest posterior density (HPD): 22.74–43.65 Ma) (Chen et al. 2018). The BEAST analysis used an uncorrelated lognormal relaxation clock and a Yule tree prior. The BEAST analysis was run for 200 million generations and sampled every 1000 generations. All of the calibrations used a normal prior, monophyly and standard deviation values of 2.2. Convergence of the run parameters was checked using Tracer v.1.7.1 (Rambaut et al. 2018) to ensure that the effective sample size of all of the parameters was greater than 200. Three runs were made in total and the trees were finally merged using LogCombiner v.2.4.7. A maximum clade credibility tree was generated using TreeAnnotator v.2.4.1 (implemented in BEAST) by applying a burn-in of 25%.

Due to the limitations of our molecular markers, dispersal pathways were not the focus of this study; instead, we were more interested in whether or not lineages had radiated and diversified. Therefore, we assessed the dispersal and in situ diversification history of Leptobrachella referring to the biogeographical meta-analyses utilised in prior studies (Klaus et al. 2016; Jiang et al. 2019; Xu et al. 2021; Li et al. 2022), i.e., the maximum number of observed dispersal events (MDisE) per million years and the maximum number of in situ diversification events (MDivE) per million years.

The ancestral ranges of Leptobrachella were estimated using the R package BioGeoBEARS (Matzke 2013). We divided the distribution ranges into five major regions: (A) Borneo, (B) Southern Thai-Malay Peninsula, (C) Northern Thai-Malay Peninsula, (D) Indo-Burma and (E) Southern China (Chen et al. 2018). Three models, i.e., dispersal-extinction-cladogenesis (DEC) (Ree and Smith 2008), dispersal-vicariance analysis (DIVALIKE) (Ronquist 1997) and the Bayesian biogeographical inference model (BAYAREALIKE) (Landis et al. 2013), as well as the +J model (founder-event), were employed in the analyses. The maximum range was set to two, based on actual surveys and the consensus (Chen et al. 2018) in literature that current Leptobrachella are narrowly distributed.

To infer the patterns of speciation in Leptobrachella based on the results of the biogeographic analyses, we referred to the criteria of Xu et al. (2021) concerning the phylogenetic tree to categorise these patterns into two types: dispersal and in situ diversification. We divided the observed dispersal events and in situ diversification events defined as the maximum number of dispersal events per 0.1 million years (MDisE) (Klaus et al. 2016) and the maximum number of in situ diversification events per 0.1 million years (MDivE) (Xu et al. 2021), respectively, to show the trends of these two types of speciation events over time. The MDisE/MDivE rates were calculated by summing potential dispersal or in situ diversification events over time in 0.1-Mya slices, based on the obtained confidence intervals for divergence times. In this analysis, we used the cpt.mean function (method = “AMOC,” Q = 0.1) of the R package changepoint (Killick and Eckley 2014) to identify significant change points in the MDisE/MDivE curve.

Another concern was whether rapid radiation had occurred in Leptobrachella. Lineage-Through-Time (LTT) plots and Bayesian Analysis of Macroevolution Mixtures (BAMM) were used to estimate the rates of speciation and the accumulation of lineages over time. The LTT plot was drawn from the total set of tree files from the BEAST analysis using Tracer v.1.7.2. BAMM v.2.5.0 (Rabosky et al. 2013) and BAMMtools v.2.1.10 (Rabosky et al. 2014) were used to estimate the diversification rates of Leptobrachella over time. The R package BAMMtools v.2.1.10 (Rabosky et al. 2014) evaluated the a priori parameters of BAMM. Time trees generated using BEAST were used for species-extinction analysis. Due to the small number of endpoints in the time tree (< 500 tips), we assigned an a priori value of 1.0 to the variable “expectedNumberOfShifts.” One million generations were run and the trees were sampled every 1,000 generations; the first 10% of these were burn-in within BAMMtools v.2.1.10 (Rabosky et al. 2014). We also performed a macroevolutionary cohort analysis using BAMMtools to summarise the probability of species sharing common macroevolutionary rate dynamics (Rabosky et al. 2014).

Morphological data for the putative new species followed Fei et al. (2009) and Rowley et al. (2013). Morphological data were measured for description and comparison and measured to the nearest 0.1 mm utilising digital calipers. The measurements were as follows: SVL = snout-vent length (from tip of snout to vent); HDL = head length (from tip of snout to rear of jaws); HDW = head width (head width at the commissure of the jaws); SNT = snout length (from tip of snout to the anterior corner of the eye); EYE = eye diameter (diameter of the exposed portion of the eyeballs); IOD = interorbital distance (minimum distance between upper eyelids); IND = internasal distance (distance between nares); UEW = upper eyelid width (measured as the greatest width of the upper eyelid); NEL = nostril-eyelid length (distance from nostril to eyelid); TMP = tympanum diameter (horizontal diameter of the tympanum); TEY = tympanum-eye distance (distance from anterior edge of the tympanum to posterior corner of the eye); HND = hand length (distance from distal end of radioulna to tip of phalanx of finger III); LAHL = length of the lower arm and hand (distance from tip of the third finger to elbow); LW = lower arm width (maximum width of the lower arm); HLL = hindlimb length (distance from tip of fourth toe to vent); THL = thigh length (distance from vent to knee); TIB = tibia length (distance from knee to heel); TW = maximal tibia width; TFL = length of foot and tarsus (distance from the tibiotarsal articulation to the distal end of toe IV) and FOT = foot length (from proximal edge of the inner metatarsal tubercle to the tip of the fourth toe) (Suppl. material 8). Comparative morphological data for species of the genus Leptobrachella were obtained from literature (Suppl. material 9). Due to the high likelihood of undiagnosed diversity within the genus (Rowley et al. 2016; Yang et al. 2016), where available, we relied on examination of topotype material and original species descriptions. Sex was determined by the presence of internal vocal sac openings and the presence of eggs in the abdomen through external inspection.

We performed a statistical analysis of the morphometric data for both species because of the morphological similarity between the new species and similar species. A size-corrected value defined as the ratio of every character to SVL was calculated and, then, the data were log-transformed for subsequent morphometric analyses in order to reduce the impact of allometry. We performed a principal components (PCs) analysis of the morphology values of these measurements, based on eigenvalues greater than one. The maximum variance method and simple bivariate scatter plots were used to explore and characterise the morphometric differences between the new species and closely-related species. One-way analysis of variance (ANOVA) was conducted to determine the significance of differences in morphometric characters between the new species and the aforementioned similar species. All of the statistical analyses were performed using SPSS 21.0 (SPSS, Inc., Chicago, IL, USA) and differences were considered statistically significant at a p-value < 0.05.

We collected a total of 388 distribution sites for 104 species using field surveys and published data (Suppl. material 10). The distribution of these species was mapped in 1° × 1° (approximately 111 × 111 km²) grid cells. The most abundant cells were located in Borneo, Indo-Burma and southern China, in that order. For Indo-Burma, the hotspot cells were concentrated in central and northern Vietnam-southern Yunnan, China and northern Laos. For southern China, the hotspot cells were concentrated in the Wuyi Mountains in the east and the Nanling and Wuling Mountains in the west (Fig. 1).

Figure 1. 

Distribution of Leptobrachella hotspots and collection sites for the new species. A–D. Denote four record-gap regions: A. Eastern China, comprising the Luoxiao Mountains, the Lianhua Mountains, the Xuefeng Mountains and the Wuyi Mountains; B. The Hengduan Mountains; C. Myanmar D. Cambodia and southern Vietnam. The base maps are from the Standard Map Service website (http://bzdt.ch.mnr.gov.cn/index.html; Map Approval No. GS(2020)4619). This illustration was created and provided by Tao Luo.

The BI phylogenetic tree was based on the mitochondrial 16S rRNA gene with a length of 532 base pairs (bp). This tree had poorly resolved phylogenetic relationships between species associated with short marker lengths; however, the analysis was still able to delineate two large clades. Following Chen et al. (2018), clades I and II may be further divided into five and two subclades, respectively, where subclades A1 and A2 were highly resolved (Fig. 2). Three samples from Xishui, Guizhou, were clustered together as the sister clade of L. suiyangensis. Three putative undescribed species, namely, Leptobrachella sp1, Leptobrachella sp2, and Leptobrachella sp3, from Yunnan and Guizhou, China, were also identified (Fig. 2).

Figure 2. 

Bayesian Inference (BI) tree based on mitochondrial 16S rRNA (532 bp). The scale bar represents 0.05 nucleotide substitutions per site. This illustration was created and provided by Tao Luo.

The combined sequence data from six nuclear genes (3594 bp in total length) also failed to resolve the phylogeny of Leptobrachella (Fig. 3A). This was possibly due to the lack of a large number of species being sampled, generating a topology inconsistent with the mitochondrial DNA data; however, the three samples from Xishui, Guizhou formed an independent lineage as a basal clade of ((L. bijie + L. jinshaensis) + (L. chishuiensis + L. purpuraventra)) (Fig. 3A). Haplotype networks, based on RAG1 and NTF3, showed that unique, non-shared haplotypes were present in the Xishui samples and multiple linked mutations occurred within closely-related species (Fig. 3B, D); however, in the nuclear gene NCX1, the Xishui samples shared haplotypes with two undescribed species, Leptobrachella sp1 and Leptobrachella sp2 (Fig. 3C).

Figure 3. 

Bayesian Inference (BI) tree based on six nuclear genes (A) and nuclear haplotypes (B–D). BPP from BI analyses/UFB from ML analyses are listed next to the nodes. Asterisks and “-” indicate support values of 1.00/100 and below 0.60/60. The scale bar represents 0.01 nucleotide substitutions per site. B. RAG1; C. NCX1; D. NTF3. This illustration was created and provided by Tao Luo.

The smallest pairwise genetic divergence between the Xishui samples and 96 species of the genus Leptobrachella was 1.7% (vs. L. suiyangensis) to 17.9% (vs. L. gracilis). These levels were similar to or higher than the divergence levels amongst recognised sister species; for example, 1.2% between L. bijie and L. jinyunensis, 0.2% between L. dong and L. bourreti and 1.9% between L. liuig and L. mangshanensis (Suppl. material 11). In conclusion, the genetic divergence, matrilineal tree and nuclear gene tree were all highly supportive of the Leptobrachella populations in Xianyuan Town, Xishui County, Guizhou, as an independent phylogenetic species.

The time tree, estimated using BEAST, was inconsistent with the BI tree obtained using the 16S rRNA (Fig. 4A and Suppl. material 1). The most recent common ancestor of the genus Leptobrachella was estimated as occurring in the Early Oligocene, ca. 32.18 Ma (95% HPD = 29.39–35.33 Ma). Subsequently, divergence between sister clades for Clade F was estimated to have occurred in the Oligocene to Early Miocene, ca. 20.97–28.90 Ma. The Xishui samples diverged from L. suiyangensis at 1.52 Ma (95% HPD = 0.93–2.79 Ma). According to the best model DEC+J (Suppl. material 12), the ancestors of Leptobrachella inhabited Borneo and Indo-Burma (Suppl. material 1). Based on this time tree, Indo-Burma was identified as the source of outward dispersal, but there was also the reverse, i.e. southern China to Indo-Burma (Suppl. material 2).

Figure 4. 

Mitochondrial 16S rRNA-based time tree and diversification dynamics. A. Speciation rates along the phylogeny (the rates are reflected by the branch colours; cold colours denote slower rates and warm colours denote faster rates); B–E. Speciation rates through time from the analysis of five datasets. Lines denote means and lighter shadows denote the 95% highest posterior density; F. LTT plot showing mean lineage accumulation over time in millions of years (Ma); G. Diversification dynamics of the genus Leptobrachella based on the maximum number of dispersal events (MDisE) and the maximum number of in situ diversification events (MDivE) per Ma; H. Geographical distribution and tectonic zone range, modified from Chen et al. (2018). ARSZ: Ailao Shan–Red River shear zone. This illustration was created and provided by Tao Luo.

The phylorate plot generated from the BAMM analysis showed a distinct increase in the speciation rate of Leptobrachella in the basal and apical branches at around 32.18 Ma and 8.89 Ma (Fig. 4B, nodes 1 and 2). In the LTT plot, acceleration of the lineage accumulation rate occurred around 10 Ma, 5 Ma, and 1 Ma (Fig. 4F). Plots of the speciation rate through time for the whole 16S rRNA dataset showed that the speciation rate and net diversification rate of Leptobrachella increased from 30 to 25 Ma, decreased from 25 to 10 Ma and continued to increase after 10 Ma (Fig. 4C, F). Extinction rates were somewhat similar (Fig. 4D). The differences in speciation rates amongst Leptobrachella may also be observed in the macro-evolutionary cohort display (Suppl. material 3). Taken together, the results suggest that Leptobrachella has historically undergone several rapid radiations.

We summarised the events associated with Leptobrachella diversification, including 178 in situ diversification events and 21 dispersal events (Suppl. material 13). In situ diversification was the primary pattern of speciation. Biogeographic meta-analyses showed similar trends for in situ diversification and dispersal over time. Both dispersal and in situ diversification events began in the Early Oligocene (~ 30 Ma). The initial speciation rates were slow, then increased rapidly at around 15.0 Ma and peaked around 10.52 Ma [95% HPD = 6.56–14.42] and 8.70 Ma [95% HPD = 7.78–9.57] (Table 1). These events were followed by rapid declines at 4 Ma and 12 Ma. For Borneo, Indo-Burma and southern China, the inflection points of the rapid increase in the in situ diversification rates occurred at 7.07 Ma, 17.02 Ma, and 11.95 Ma, peaked at 9.85 Ma, 10.75 Ma, and 4.74 Ma, and decreased significantly at 2.11 Ma, 3.20 Ma, and 1.30 Ma (Table 1). Additionally, the in situ diversification peaks for Indo-Burma and southern China were staggered, suggesting that diversification was not synchronised. Taken together, the results suggest that the Miocene (5–15 Ma) was a key period for the diversification of Leptobrachella.

Principal component analysis applied to the Xishui samples and L. suiyangensis using 15 measured characters showed that a total of five principal components were extracted. The first three components explained 63.42% of the total variance, with PC1 accounting for 28.31%, PC2 for 19.73% and PC3 for 15.38% (Suppl. material 14). In the scatterplot of PC1 versus PC2, Xishui samples and L. suiyangensis formed distinct clusters and were separated on the PC1 axis (Fig. 5). HDW, EYE, IOD, UEW, HND, LW, and HLL were the mainly loaded PC1 (Suppl. material 14). The results of the ANOVA tests indicated that the Xishui samples differed significantly from L. suiyangensis, based on several morphometric characters, including EYE, IOD, HND, LW, and HLL (all p-values < 0.05; Table 2). Combining morphological and genetic differences, we describe the Leptobrachella population from Xianyuan Town, Xishui County, Guizhou, China, as the new species Leptobrachella xishuiensis sp. nov.

Figure 5. 

Scatter plots of the first and third principal components for Leptobrachella xishuiensis sp. nov. and L. suiyangensis. This illustration was created and provided by Tao Luo.

The classification of Asian leaf-litter frogs has been an enigma and several taxonomic schemes have been proposed. The genus Leptolalax Dubois, 1980 was widely accepted and assigned two subgenera, Lalos Dubois, Grosjean, Ohler, Adler & Zhao, 2010 and Leptolalax Dubois, 1980 (Dubois 1980; Delorme et al. 2006; Dubois et al. 2010). Despite phylogenetic support for both subgenera (Poyarkov et al. 2015; Yuan et al. 2017), controversy remains (Ohler et al. 2011; Matsui et al. 2017). Based on large-scale sampling, Chen et al. (2018) revealed that Leptobrachella nested within Leptolalax, thus placing it in the synonymy of Leptobrachella and they rejected the hypothesis that two subgenera were contained within Leptolalax, ending a long-standing controversy. Several genera have also been proposed by Chinese taxonomists, for example, Paramegophrys and Carpophrys. Thus, Paramegophrys was also utilised as a valid genus name for Asian leaf-litter frogs in China for some time (Jiang et al. 2008; Mo et al. 2008; Fei et al. 2009). The genus Paramegophrys was rejected by Dubois et al. (2010), based on Article 9.9.9 of the International Code of Zoological Nomenclature (4^th ed.); however, considering the morphological differences, Leptobrachella within China was placed in Paramegophrys by Fei and Ye (2016). For Carpophrys, the genus name is invalid as a nomen nudum, based on Article 14 of the International Code of Nomenclature (4^th ed.), as the type species was not specified (Dubois 1981; Fei and Ye 1992; Dubois et al. 2010). As it is a taxon with many new species descriptions in recent years, the complete phylogeny of this genus remains to be constructed.

Through amplification and analysis of published data, we have provided additional insights into the phylogeny of Leptobrachella. Our results clearly show that Leptobrachella is divided into two major clades (BPP = 1.00), Clade I and Clade II, that are highly resolved in the BI tree for 16S rRNA. This result differs from those of prior studies (Chen et al. 2018; Luo et al. 2022b; Chen et al. 2024). We support Leptolalax as a synonym of Leptobrachella (Chen et al. 2018) and we also propose three hypotheses for the classification of the genus: (1) place all of the species in Leptobrachella (the single genus option) comprising Clades I and II; (2) retain either Leptobrachella or Leptolalax in clade I only, assigning either Leptolalax or Leptobrachella to Clade II (a two-genera option); (3) retain Leptobrachella, Clade I only, with Leptolalax as a synonym of Leptobrachella and create a new genus for Clade II that may restore the validity of Paramegophrys (a two-genera option). Geographically and morphologically, the majority of Clade I is from south of Indo-Burma, where the clade is characterised by the absence of supra-axillary and ventrolateral glands and small body size; Clade II is largely from central and northern Indo-Burma and southern China and it is characterised by the presence of supra-axillary and ventrolateral glands and medium to large body size (Suppl. material 4). While the two clades differ morphologically and geographically to some extent, a detailed morphological examination of Clade I is needed for possible revisions. In addition, the phylogeny, based on multiple nuclear gene loci, was more consistent with the results of Chen et al. (2018). Therefore, we prefer the most conservative “single-genus option” until additional data are available.

Based on field surveys and published data, we identified three hotspots in the distribution of Leptobrachella. Only the Borneo and Indo-Burma regions corresponded to prior global biodiversity hotspots (Myers et al. 2000). The remaining area is located in the mountainous regions of southern China, including the Wuyi, Nanling and Wuling Mountains and is the habitat of 55.6% of Leptobrachella species in China. This overlaps with new hotspots for amphibians recently identified in China (Xu et al. 2024). Southern China has experienced rapid economic growth and severe loss of protected areas in recent years (Xu et al. 2017; Ma et al. 2019, and the fauna has experienced pressure from human activities (Xu et al. 2024). Therefore, these three mountainous regions need urgent protection. We note that there are four recorded gap areas and that additional surveys of these areas are necessary.

To estimate the origin and dynamic history of diversification in Leptobrachella, we assessed the divergence times and MDEs of 101 species using mitochondrial 16S rRNA sequences. Although the topology was not well resolved, utilising 95% HPD overcomes the effects of this uncertainty to some extent and this approach is widely utilised (Klaus et al. 2016; Jiang et al. 2019; Xu et al. 2021; Li et al. 2022). Therefore, the results of the MDivE analysis are utilised in the discussion. The distribution of Leptobrachella spans nearly 30 degrees of latitude and the diversity hotspots are concentrated near orogenic zones, i.e. the QTP and the Ailao Shan–Red River shear zone. Thus, the diversification of Leptobrachella was most likely driven by geology and the Asian monsoon climate and less by temperature (Luo et al. 2023).

Molecular dating and MDivE analyses revealed that Leptobrachella originated at ~ 32 Ma. The MDivE has two major (35–14 Ma, 13–6 Ma) or three minor acceleration phases (35–20 Ma, 19–14 Ma and 13–6 Ma) and a deceleration phase (6 Ma–present), increasing rapidly at ~ 15 Ma, with the highest peak occurring at ~ 8.7 Ma (Fig. 4G, Table 2). The origin of the genus is consistent with geological and phylogenetic evidence hypothesising that 35 Ma was the initiation of the left-lateral extrusion in Indochina (Schärer et al. 1994; Leloup et al. 2001; Gilley et al. 2003; Li et al. 2020b) or was in the midst of an accelerated process (Li et al. 2024b), but predates the rapid QTP uplift (25–15 Ma; Ding et al. (2017); Ding et al. (2022)). The first small peak of the MDivE is at ~ 20.5 Ma, which is consistent with the dramatic change in the India-Asia convergence rate and angle (Li et al. 2024b). The MDivE first decreased at ~ 20 Ma, implying a slowing of the Indochina left-lateral extrusion and coinciding with the end of the geological left-lateral offset (Chen et al. 1992; Leloup et al. 1995; Leloup et al. 2001; Wang et al. 2020b). However, the rapid uplift of the QTP at 20 Ma (Ding et al. 2017) may have led to the end of a short-lived slowdown in diversification in the MDivE.

From 19 to 16 Ma, the MDivE continued to increase, likely driven by the continued sharp uplift of the QTP (Ding et al. 2017), although the left-lateral extrusion in Indochina may have been slowing (Li et al. 2024b). A second decrease in the MDivE occurred at ~ 15 Ma, a period that is often regarded as the end of the left-lateral offset (Leloup et al. 1995, 2001; Wang et al. 2020b). The MDivE reached a second valley at 13.5 Ma (Fig. 4G) and the Red River Fault began to be dislocated at the same time (~ 13–12 Ma) (Zhang et al. 2006; Zhang et al. 2009). Phylogenetic evidence supports the hypothesis that the left-lateral extrusion in Indochina may also have continued until this time (Li et al. 2024b); that is, the shift in the left-to-right lateral movements of the ARSZ (~ 13 Ma) and the slowing of the rapid uplift of the QTP (~ 15 Ma) may have occurred in the mid-Miocene (Ding et al. 2017, 2022; Li et al. 2024b).

From 13 to 6 Ma, the MDivE underwent a final dramatic increase, peaking at 8.70 Ma (95% CI: 9.57–7.78 Ma; Table 2), suggesting that more intense orogenic and climatic shifts occurred during this period. Geology and phylogeny have shown that a second, more rapid outward extrusion of Indochina occurred at this time and that the rate and angle of Indo-Asian convergence again changed rapidly (Li et al. 2024b). Geological studies also suggest that the right-lateral extrusion of Indochina occurred during the late Miocene (11–5 Ma) (Zhang et al. 2009; Fyhn and Phach 2015). The most distinct peak of the MDivE occurred at 8.70 Ma, consistent with the Ailao Shan thermochronologic age (8.4–6.8 Ma) (Zhang et al. 2009) and the age of the Asian monsoon intensification (Farnsworth et al. 2019). From 6 Ma to the present, MDivE has continued to decline as orogeny and monsoon climate intensification slowed. A similar three-stage diversification pattern was also reflected in the speciation rate-through-time plots (Fig. 4D). The dynamic patterns of MDivE suggest that the Indochina extrusion (left and right) and QTP uplift went through multiple phases. Global temperatures continued to decline over the same period (Westerhold et al. 2020), while the MDivE continued to rise and, thus, the contribution of temperature to diversification may have been limited.