Lineages to landscapes: mitogenomic insights and climate refugia informing proactive conservation of the endangered Tricarinate Hill Turtle (Melanochelys tricarinata) in the Indian subcontinent Authors: Imon Abedin, Angkasa Putra, Hye-Eun Kang, Arunima Singh, Shailendra Singh, Won-Kyo Jung, Hyun-Woo Kim, Shantanu Kundu Published: 2025 ABSTRACT The Tricarinate Hill Turtle (Melanochelys tricarinata), endemic to the Indian subcontinent, faces significant threats from intensifying anthropogenic pressures and severe habitat degradation. Thus, to facilitate conservation efforts for this endangered species, a comprehensive assessment of its genomics and ecology is imperative. The present study adopts two independent approaches by characterizing the complete mitogenome and evaluating current and future habitat suitability. The mitogenome (16,745 bp) comprises 37 genes, with most protein-coding genes beginning with the canonical ATG start codon. The codon usage analysis revealed that arginine, leucine, and serine are the most frequently used amino acids. The control region contains a termination-associated sequence, four conserved sequence blocks, and two distinct tandem repeat motifs. The phylogenetic assessment using both Bayesian inference and Maximum-likelihood methods consistently placed M. tricarinata in a distinct clade, separate from other geoemydids. The mitogenome-based TimeTree analysis revealed that M. tricarinata diverged from its closest relatives during the Oligocene. The comparative analyses of partial mitochondrial genes revealed substantial divergence between Melanochelys congeners (3.71–5.20% in cox1, 7.99–9.29% in cytb), with low haplotypic diversity in M. tricarinata and high in M. trijuga. The ensemble model identified suitable habitats under both current and future climate scenarios. Under present scenario, approximately 374,657 km2 of ideal habitat was delineated within the training extent with a mean corridor connectivity of 0.377, which was reduced to 238,039 km2 and mean connectivity of 0.518 when restricted to existing forest cover. Moreover, only 31,450 km2 of the suitable habitat for this species falls within Protected Areas, representing just 11.912% of the total identified favorable area. The future projections suggest a potential reduction of up to 40% in suitable habitat area due to climatic changes, accompanied by increased fragmentation, smaller patch sizes, and over 80% decline in connectivity across future scenarios. Thus, this study provides comprehensive insights into the systematic position, evolutionary history and ecological requirements of M. tricarinata, offering a critical scientific foundation to guide effective conservation and management strategies for this imperiled species across its native range.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-26890-5. Introduction The past century has witnessed an unprecedented acceleration in global biodiversity loss, driven by escalating anthropogenic pressures, climate change, and large-scale ecological disruption1–3. Within this unfolding collapse, Testudines, one of the most ancient and evolutionarily distinct vertebrate lineages, are experiencing dramatic population declines and heightened extinction risk4. The family Geoemydidae represents one of the most taxonomically diverse clades within the order Testudines, encompassing three recognized subfamilies, 19 genera, and 75 valid species5. Globally, this lineage exhibits a broad biogeographic distribution, extending across Asia, Europe, North Africa, and parts of the Americas. They inhabit a wide range of ecological settings, from strictly aquatic to predominantly semi-aquatic environments6. Notably, a substantial proportion of species within this family are experiencing severe population declines, with many classified among the most threatened turtles globally, as highlighted by the IUCN Tortoise and Freshwater Turtle Specialist Group (TFTSG)5,7. The Tricarinate Hill Turtle Melanochelys tricarinata is a small to medium-sized freshwater turtle endemic to the Indian subcontinent. The species was initially classified under the genus Geoemyda, but was later reassigned to the genus Melanochelys based on subsequent taxonomic revisions5,8. This species is characterized by a highly domed carapace (lengths of up to 174 mm), bearing three distinct longitudinal keels extending from the nuchal to the posterior margin. The carapace is dark brown to black, contrasting with a pale-yellow plastron marked by irregular dark blotches; the head is small, slightly pointed, and olive to dark in color with a narrow snout, while the robust, scaly limbs are adapted for terrestrial movement8. This species is primarily terrestrial, inhabiting forested areas and grasslands of the Ganges and Brahmaputra plains, as well as moist deciduous to wet evergreen forests in the Himalayan foothills, typically at elevations up to 300 meters9. It is regarded as semi-fossorial and exhibits peak activity during the monsoon, showing a marked preference for stream-edge habitats. Consequently, the most recent IUCN assessment of 2020 has classified the species as ‘Endangered’ under the threatened category10. The primary driver of population decline in M. tricarinata is attributed to the widespread loss and degradation of forest habitats across its distribution range9,11. In addition to habitat-related threats, anthropogenic pressures pose significant challenges to the survival of this species. This taxon is exploited for local subsistence consumption and has, in recent years, gained popularity in the Asian pet trade10. The market surveys and seizure reports have documented substantial numbers of this species in illegal trade channels. This species is also reportedly exploited for the production of decorative curio masks, which are commonly sold in tourist and handicraft markets across South and parts of Southeast Asia11,12. In response to these mounting threats, the IUCN TFTSG has recommended comprehensive research on this species, emphasizing the need for an integrated approach to inform and guide species- and site-specific conservation planning. In addition to morphological investigations, genetic studies have also been conducted on M. tricarinata, although most of these have relied primarily on partial or fragmented gene sequences13–15. These preliminary studies have provided initial insights into the phylogenetic relationships and systematic placement of this species within the family Geoemydidae. However, in-depth molecular data for the species have neither been generated nor incorporated into recent comprehensive studies or global initiatives aimed to clarify its precise phylogenetic position16. The recent advances in mitogenomic research have demonstrated the utility of complete mitochondrial genomes in elucidating genetic structure and interspecific relationships17–19. Globally, herpetologists have sequenced complete mitogenomes for numerous freshwater turtle species to refine their phylogenetic positioning and evolutionary relationships20–26. However, the complete mitochondrial genome of Melanochelys species remains unavailable, thereby limiting the understanding of its matrilineal lineage, systematic position, and evolutionary history within geoemydids. Beyond morphological and molecular studies, understanding habitat dynamics also plays a crucial role in formulating effective conservation strategies for threatened vertebrate species27. The IUCN TFTSG has advocated for integrated, multidisciplinary approaches that combine ecological and genetic data to support robust, evidence-based conservation planning25,26,28,29. Despite this, research on habitat suitability and the impacts of climate change on freshwater turtles remains sparse across its native range. In this context, assessing habitat suitability at high spatial and temporal resolutions not only improves the understanding of ecological and biogeographical patterns but also plays a critical role in guiding targeted conservation interventions30–33. Hence, to fill in existing research gaps and support the proactive conservation strategies of the endangered M. tricarinata, the present study adopts two independent approaches. The strategies that apply molecular and ecological approaches individually are crucial for gaining comprehensive insights into the conservation of this threatened species, as highlighted by international commitments under the Convention on Biological Diversity (CBD), including the Aichi Biodiversity Targets and the Kunming–Montreal Global Biodiversity Framework34,35. Therefore, the present study aimed to (i) sequencing the complete mitochondrial genome of M. tricarinata using next-generation sequencing (NGS) technologies; (ii) analyzing Testudines mitochondrial genomes to uncover genetic variation and structure; (iii) performing cladistic and TimeTree analyses to elucidate the systematic position and matrilineal evolutionary relationships of this species within broader Testudines lineages; and (iv) identifying climate refugia and corridor connectivity of M. tricarinata within its distribution range using ensemble species distribution model (SDM) in present and future climatic scenarios. This study seeks to fill critical knowledge gaps in understanding the taxonomic placement and matrilineal evolutionary history of the species, while also assessing its habitat dynamics under present and projected climate scenarios (Supplementary Fig. S1). The synchronous approach adopted in this research not only provides a comprehensive framework for guiding species-specific conservation strategies but also offers broader insights applicable to the conservation of other freshwater turtles worldwide. Materials and methods Sample collection and ethics statement A single specimen of M. tricarinata was obtained from the species colony maintained by Turtle Survival Alliance Foundation India (TSAFI) in Uttar Pradesh, India. The species identification was confirmed based on morphological characteristics following standard traits described in the previous literatures8,9. Necessary permission was obtained from the office of the Principal Chief Conservator of Forests (Wildlife), Lucknow, India issued by the Environment, Forest and Climate Change Department, Government of Uttar Pradesh (Permit number 1451/23-2-12 G dated January 05 2021) and all methods were performed in accordance with the relevant guidelines and regulations approved by the TSAFI. A blood sample (100 μL) was collected aseptically from the hind limb after sedation with Alfaxalone at a dosage of 20–30 mg/kg. The biological sample was conserved in a 1.5 mL EDTA-containing microcentrifuge tube and stored at 4 °C for subsequent molecular analyses. No animals were captured or vouchered during this study, and the sampled individual was released back into the colony immediately following biological sampling. All procedures adhered to the ARRIVE 2.0 guidelines for the ethical use of animals in research (https://arriveguidelines.org)36. Mitogenome sequencing through next-generation approach The total mitochondrial DNA was isolated using the Alexgen DNA Kit (Alexius Biosciences, Ahmedabad, Gujarat, India), strictly adhering to the manufacturer’s guidelines37. The DNA concentrations were quantified using a Qubit 4.0 fluorometer (Thermo Fisher Scientific, USA) to ensure accuracy in downstream applications. The mitochondrial genome sequencing was conducted at Unipath Specialty Laboratory Ltd. (http://www.unipath.in/) in Ahmedabad, India. For sequencing, a paired-end library was prepared utilizing the QIAseq FX DNA Library Kit (Cat. No. 180479, Qiagen). The fragmentation of genomic DNA was performed using the Covaris M220 Focused Ultrasonicator (Covaris Inc., San Diego, CA, USA), generating fragment sizes ideal for library construction. The adapter ligation was then carried out on both ends of the fragmented DNA to facilitate Illumina sequencing compatibility. Further, to enhance coverage from low DNA input, high-fidelity amplification was performed using the HiFi PCR Master Mix (Takara Bio Inc., Kusatsu, Shiga, Japan). The quality and fragment size distribution of the final libraries were assessed using the TapeStation 4150 system (Agilent Technologies, Santa Clara, CA, USA) with High Sensitivity D1000 ScreenTape, following the manufacturer’s protocols. Libraries passing quality control were subsequently sequenced on the Illumina NovaSeq 6000 platform (Illumina, San Diego, CA, USA), enabling high-throughput, paired-end sequencing. Mitogenome assembly, annotation and submission Approximately 25 million high-quality paired-end reads were utilized for mitochondrial genome assembly and annotation using Geneious Prime v2023.0.138. The gene regions and their transcriptional orientations were further validated using two independent annotation platforms, including MITOS as implemented within the Galaxy web server (https://usegalaxy.eu) and MitoAnnotator (http://mitofish.aori.u-tokyo.ac.jp/annotation/input/)39,40. Further, to validate the protein-coding genes (PCGs) sequences, their corresponding amino acid translations were assessed using the vertebrate mitochondrial genetic code via the ORF Finder tool (https://www.ncbi.nlm.nih.gov/orffinder/). The finalized mitochondrial genome sequence was submitted to GenBank (Accession No. PP341306) using the Sequin submission platform, along with a detailed gene feature file containing annotations and strand information. Structural analysis of mitogenome The circular figure of the M. tricarinata mitochondrial genome was constructed using the MitoAnnotator web server, with manual annotation of intergenic spacer regions and gene overlaps. The lengths and nucleotide compositions of PCGs, ribosomal RNAs (rRNAs), and transfer RNAs (tRNAs) were analyzed using MEGA version 1241. Further, to evaluate nucleotide composition bias, AT-skew and GC-skew values were calculated using the formulas AT-skew = (A − T)/(A + T) and GC-skew = (G − C)/(G + C)42. The relative synonymous codon usage (RSCU) for the studies species was examined using DnaSP version 6.043. The secondary structure of tRNAs were determined through tRNA and tmRNA prediction (Aragorn) in Galaxy version 0.6 web server44. The control region (CR) was analyzed to delineate its structural domains based on previously published literature, and a comparative analysis was conducted using the mitogenomes of 33 representative species within the subfamily Geoemydinae26,45. In addition, tandem repeat motifs within the CR were recognized using the Tandem Repeats Finder program (https://tandem.bu.edu/trf/trf.html)46. Due to the absence of CR annotations in the mitogenomes of three species (Geoemyda japonica: LC574978, Mauremys japonica: AP019397, and Mauremys reevesii: AP019398), these taxa were excluded from the comparative analyses. Data assembly and phylogenetic analyses To investigate the cladistics relationships of M. tricarinata within the broader lineage of Testudines, the complete mitochondrial genomes of 60 species were retrieved from GenBank (Supplementary Table S1). Altogether, the phylogeny comprises 58 species representing 10 families within the suborder Cryptodira, along with mitogenomes of three species from three families under the suborder Pleurodira, which were included as outgroup taxa. Further, to establish a comprehensive phylogenetic framework within the suborder, a concatenated alignment of 13 mitochondrial PCGs was generated using iTaxoTools v0.147. The best-fit nucleotide substitution model (GTR + G + I) was identified using PartitionFinder 248. The phylogenetic analyses were executed using both Bayesian Inference (BI) and Maximum-likelihood (ML) methods. The BI was conducted in MrBayes v3.1.2 using nst = 6, employing one cold and three heated Metropolis-coupled MCMC chains for 10 million generations49. The trees were sampled every 100 generations, with the first 25% discarded as burn-in. The MCMC analysis was performed until convergence was achieved, indicated by the standard deviation of split frequencies reaching 0.01 and the potential scale reduction factor (PSRF) for all parameters approaching 1.0. The ML tree was inferred using PhyML v3.0 with 1000 bootstrap replicates under default settings50. Both the phylogenies were further visualized and annotated using the Interactive Tree of Life (iTOL) v4 platform, facilitating clearer interpretation51. Divergence time estimation Due to the computational advantages of adaptive and green computing, the MEGA platform enables faster analyses compared to the BI approach52–54. Accordingly, divergence times for the present dataset were estimated using the RelTime method implemented in MEGA v12. The assembled PCGs sequences were used to construct a ML topology, which was subsequently exported in .nwk format as the initial tree. The outgroup taxa representing the suborder Pleurodira (Peltocephalus dumerilianus, Pelomedusa subrufa, and Chelus fimbriata) were designated prior to divergence time estimation. The TimeTree analysis was conducted with two calibration constraints specified in the calibration editor, incorporating fossilized birth–death priors: (i) the divergence between Trionychia (Carettochelyidae + Trionychidae) and the remaining Cryptodira, constrained to 108–152 Ma with a 95% highest posterior density (HPD); and (ii) the divergence of Geoemydidae, constrained to 39–57 Ma with a 95% HPD, as supported by earlier studies11,16,55. Genetic distance and inter-specific haplotype network The partial mitochondrial sequences of cytochrome c oxidase subunit I (cox1) and cytochrome b (cytb) were compiled for both Melanochelysspecies. A total of eight cox1 and five cytb sequences of M. tricarinata, and 10 cox1 and 16 cytb sequences of M. trijuga were retrieved from GenBank (Supplementary Table S2). Two datasets were constructed (cox1: 585 bp and cytb: 1140 bp) after alignment in CLUSTAL X56 and the species-specific nucleotide variations were identified using MEGA v12. The pairwise genetic distances were calculated using the Kimura 2-parameter (K2P) model in MEGA v12. The genetic diversity indices, including the number of inter-specific haplotypes, haplotype diversity (Hd), nucleotide diversity (π), number of polymorphic sites (S), and Wright’s fixation index (Fst) were estimated using DnaSP version 6.0 (Rozas et al. 2017). The inter-specific haplotype relationships were further visualized through a Templeton, Crandall, and Sing (TCS) network constructed in POPART version 1.757,58. Occurrence records and training extent A total of 17 occurrence records of M. tricarinata were recorded by the TSAFI team across the Brahmaputra and Ganges River basins (Fig. 1). Additionally, to enhance the spatial representation of occurrence data, secondary records were also compiled from the IUCN GeoCAT platform, which aggregates species locations from multiple open-access biodiversity repositories59. Additionally, occurrence data were extracted from published literature and reports5,14,15,60. In total, 114 presence records were aggregated utilizing primary and secondary sources, covering locations within and beyond the IUCN-delineated range of the species. Moreover, to ensure a robust and unbiased dataset, records based on preserved specimens or captive individuals were excluded during the aggregation of secondary data. Notably, to minimize spatial autocorrelation and reduce model overfitting, presence points were spatially rarefied at a 1 km2 resolution using the spatial rarefaction tool in SDM Toolbox v2.461. This resolution was specifically selected to align with the pixel size of the environmental raster layers. Following rarefaction, 78 unique occurrence records were retained for modeling within the training extent. This limited occurrence record can still provide a reliable basis for developing ensemble SDMs for this species, as demonstrated in earlier studies62–66.Fig. 1The geographic distribution of M. tricarinata across the Indian subcontinent and its mitochondrial genome architecture are illustrated in the figure. The map highlights both the IUCN-defined range and the extended training extent, incorporating occurrence records obtained from primary field surveys and verified secondary sources. The map was generated by using ArcGIS version 10.6. Inset earth map was captured from Google Earth. The circular representation of the complete mitogenome, generated using the MitoAnnotator web server (https://mitofish.aori.u-tokyo.ac.jp/annotation/input/), depicts genes encoded on the heavy and light strands. The gene groups including PCGs, rRNAs, tRNAs, and the non-coding control region are distinguished by color arc, with clear demarcation of their boundaries. The species photograph was taken by the author (S.S.) and edited manually in Adobe Photoshop CS 8.0 The geographic distribution of M. tricarinata across the Indian subcontinent and its mitochondrial genome architecture are illustrated in the figure. The map highlights both the IUCN-defined range and the extended training extent, incorporating occurrence records obtained from primary field surveys and verified secondary sources. The map was generated by using ArcGIS version 10.6. Inset earth map was captured from Google Earth. The circular representation of the complete mitogenome, generated using the MitoAnnotator web server (https://mitofish.aori.u-tokyo.ac.jp/annotation/input/), depicts genes encoded on the heavy and light strands. The gene groups including PCGs, rRNAs, tRNAs, and the non-coding control region are distinguished by color arc, with clear demarcation of their boundaries. The species photograph was taken by the author (S.S.) and edited manually in Adobe Photoshop CS 8.0 Furthermore, the study employed a bounding box around the IUCN-delineated range as the training extent to evaluate the habitat suitability of M. tricarinata both within and beyond its currently inferred range. This approach was adopted to account for regions outside the IUCN range where the species has been reported and to assist conservation planners in identifying potential refugia beyond the known range67. Utilizing this approach, the study aims to support proactive conservation efforts by highlighting additional areas that may be optimal for protection. Moreover, this broader extent facilitates the identification of environmental envelopes analogous to those within the species’ known habitat, thereby enhancing the understanding of its ecological niche beyond the formally recognized boundary. Predictors for ensemble modelling The 19 standard bioclimatic variables commonly used in the SDM were obtained from the WorldClim database (https://www.worldclim.org/)68. In addition, topographic variables, viz., elevation, slope, and aspect, were derived from 90-m resolution Shuttle Radar Topography Mission (SRTM) data (http://srtm.csi.cgiar.org/srtmdata/). All spatial layers were standardized to a 30 arcsecond (~ 1 km2) resolution using the Spatial Analyst extension in ArcGIS 10.6 (Supplementary Table S3). The correlation between the variables was assessed using the SAHM (Software for Assisted Habitat Modeling) package in VisTrails69. The variables displaying Pearson, Spearman, or Kendall correlation coefficients (r) greater than 0.7, were omitted from modeling70 (Supplementary Fig. S2). After addressing the multicollinearity, eight variables were chosen for the final modelling purpose. Further, to evaluate the potential impacts of climate change, future habitat suitability was modeled under two Shared Socioeconomic Pathways (SSPs): SSP245 (moderate emissions) and SSP585 (high emissions), for two-time horizons: 2041–2060 and 2061–2080. These climate projections were sourced from the HadGEM3-GC31 LL model, a part of the Coupled Model Intercomparison Project Phase 6 (CMIP6)71,72. This specific model was selected for its reliability and is considered one of the most robust models for South and Southeast Asia73,74. Moreover, to provide a more comprehensive and realistic assessment of habitat suitability across the extended range, the modeled suitable areas were also overlaid with current forest cover based on ESRI Sentinel-2 Land Cover, following established methodologies to support effective conservation planning67. Model setup and performance evaluation An ensemble modeling framework was employed to predict habitat suitability of this species utilizing five distinct algorithms: Boosted Regression Trees (BRT), Multivariate Adaptive Regression Splines (MARS), Generalized Linear Models (GLM), Maximum Entropy (MaxEnt), and Random Forests (RF)31,75,76. This integrative approach capitalized on the complementary strengths of each algorithm to enhance model robustness and predictive performance77. The ensemble model was executed using the SAHM (Software for Assisted Habitat Modeling) package within the VisTrails platform69,78. The model outputs consisted of continuous habitat suitability maps with values ranging from 0 (unsuitable) to 1 (highly suitable). These were further binarized into presence–absence maps using the sensitivity equals specificity (SES) threshold criterion. Only models achieving an area under the receiver operating characteristic curve (AUC) value ≥ 0.75 were retained for ensemble generation79. A consensus ensemble map was then produced, with pixel values ranging from 0 to 5, indicating the number of models forecasting preferred habitat at each location. A score of 5 represented full agreement between all five algorithms, allowing for an assessment of model consensus and habitat configuration. Additionally, a variance map was generated from the continuous habitat suitability outputs for spatial uncertainty across model predictions (Supplementary Fig. S3). Further, to evaluate the reliability of model predictions, several performance metrics were calculated, including AUC, True Skill Statistic (TSS), Cohen’s Kappa, Proportion Correctly Classified (PCC), specificity, and sensitivity. These metrics were computed for both training datasets and across 10 cross-validation replicates80–83. The assessment of corridor connectivity in present and future scenarios was conducted using a circuit theory-based approach widely applied in ecological connectivity analyses84. The Circuitscape toolbox in ArcGIS 10.6 facilitated the analysis, with occurrence points defined as network nodes85. The conductance surface was generated from probability outputs produced by the ensemble model. This approach enabled simulation of movement flows across heterogeneous landscapes. The pairwise source/ground mode was applied, utilizing the probability surface as the conductance raster and assigning occurrence points as focal nodes in the analytical setup. Landscape geometry assessment The configuration and shape geometry of suitable habitat patches for M. tricarinata within the training extent were evaluated under both current and projected future climatic scenarios. To achieve this, several class-level landscape metrics within an 8-cell neighborhood were evaluated using FRAGSTATS version 4.2.186, a widely adopted tool in landscape ecology and environmental management. This software provides comprehensive metrics for analyzing spatial patterns that offer valuable insights into landscape configuration87. The key class metrics employed in this study were number of patches (NP), largest patch index (LPI), patch density (PD), total edge (TE), aggregation index (AI), and landscape shape index (LSI). The metrics such as NP, PD, TE, and LPI offered detailed information on patch quantity, size, edge length, and density. In contrast, LSI was used to evaluate shape complexity, while AI quantified spatial aggregation and proximity of patches, thereby reflecting the degree of habitat clustering within the landscape. Results Mitogenomic characterization and codon usage patterns The complete mitochondrial genome of M. tricarinata was successfully sequenced and annotated, providing comprehensive insights into its genetic composition and structural organization. The circular mitogenome spanned 16,745 bp and was deposited in GenBank under accession number PP341306. It comprised the typical set of 37 genes, including 13 PCG, 22 tRNA, two rRNA genes, and a major non-coding CR (Fig. 1). Of these, 28 genes were encoded on the heavy (H) strand, while the remaining nine were located on the light (L) strand. The L-strand genes included one PCG, NADH dehydrogenase subunit 6 (nad6), and eight tRNA genes: tRNA-glutamine (trnQ), tRNA-alanine (trnA), tRNA-asparagine (trnN), tRNA-cysteine (trnC), tRNA-tyrosine (trnY), tRNA-serine 2 (trnS2), tRNA-glutamate (trnE), and tRNA-proline (trnP) (Table 1). The comparative mitogenomic analysis among members of the subfamily Geoemydinae revealed that mitogenome lengths vary substantially, ranging from 16,023 bp in Cuora yunnanensis to 17,448 bp in Geoemyda spengleri (Supplementary Tables S1, S4). Within the mitogenome of M. tricarinata, the PCGs represented 11,417 bp, accounting for 68.18% of the total length, while tRNA and rRNA genes comprised 1,576 bp (9.41%) and 2573 bp (15.37%), respectively, and the CR covered 1175 bp (7.02%). A total of 10 intergenic spacer regions were identified, cumulatively spanning 49 bp, with the longest spacer (25 bp) located between trnN and trnC, followed by a 9 bp spacer between NADH dehydrogenase subunit 4 (nad4) and tRNA-histidine (trnH). Additionally, 14 overlapping regions were found, totaling 45 bp, including the longest overlaps of 10 bp each occurring between cox1 and trnS2, as well as between ATP synthase subunit 8 (atp8) and ATP synthase subunit 6 (atp6). Most PCGs initiated with the standard ATG start codon, with exceptions including cox1 and cytb, which began with GTG. The termination codons were diverse with five PCGs (cox2, atp8, nad4L, nad4, and nad5) ended with the canonical TAA codon, two genes (cox1 and nad6) utilized with the alternative AGG codon, and six exhibited incomplete stop codons (TA- or T–). The nucleotide composition of the M. tricarinata mitogenome was biased toward adenine and thymine, with an overall AT content of 60.95% and a GC content of 39.05%. Among the four nucleotides, adenine (35.06%) and cytosine (26.87%) were the most prevalent, while thymine (25.89%) and guanine (12.18%) were less abundant (Supplementary Table S4). The region-specific nucleotide formation demonstrated varying degrees of AT richness, viz., PCGs (62.69%), tRNAs (60.18%), rRNAs (60.44%), and CR (67.74%), with corresponding GC contents of 37.31%, 39.82%, 39.56%, and 32.26%, respectively (Fig. 2A–B). Among Geoemydinae species, the AT bias ranged from 60.15% in Mauremys sinensis to 62.53% in Heosemys depressa, with AT skew values varying from 0.098 in G. spengleri to 0.150 in M. tricarinata and GC skew ranging from − 0.376 in M. tricarinata to − 0.320 in Cyclemys dentata (Supplementary Table S4). Furthermore, the RSCU analysis of M. tricarinata illustrated comparable codon preferences for those ending in A/U and G/C, with each group comprising 32 codons. Among the 20 standard amino acids, arginine, leucine, and serine displayed the highest codon usage frequencies, while methionine and tryptophan, each encoded by a single codon (AUG and UGG, respectively), were the least frequent (Fig. 2C).Table 1List of annotated mitochondrial genes, boundaries, size, and intergenic nucleotides of M. tricarinata.GeneDirectionStartEndSize (bp)Anti-codonStart codonStop codonIntergenic nucleotidetrnF + 17171GAA− 1rrnS + 711034964− 1trnV + 1034110471TAC− 2rrnL + 1103271116090trnL2 + 2712278877TAA1nad1 + 27903760971ATGTA-− 1trnI + 3760383071GAT1trnQ–3832390372TTG− 2trnM + 3902397170CAT0nad2 + 397250101039ATGT–− 1trnW + 5010508778TCA0trnA–5088515770TGC0trnN–5158523174GTT25trnC–5257532367GCA6trnY–5330540172GTA1cox1 + 540369501548GTGAGG− 10trnS2–6941701272TGA1trnD + 7014708471GTC1cox2 + 70867787702ATGTAA0trnK + 7788786174TTT1atp8 + 78638030168ATGTAA− 10atp6 + 80218703683ATGTA-0cox3 + 87049487784ATGT–− 1trnG + 9487955569TCC0nad3 + 95569904349ATGT–0trnR + 9905997571TCG− 1nad4L + 997510,274300ATGTAA− 7nad4 + 10,26811,6441377ATGTAA9trnH + 11,65411,72471GTG0trnS1 + 11,72511,79167GCT− 2trnL1 + 11,79011,86273TAG− 1nad5 + 11,86213,6821824ATGTAA− 5nad6–13,67814,205528ATGAGG0trnE–14,20614,27671TTC3cytb + 14,28015,4261147GTGT–0trnT + 15,42715,49973TGT0trnP–15,50015,57071TGG0CR15,57116,7451175Fig. 2(A) The bar chart visualizing the nucleotide composition percentages (AT and GC content) across the complete mitogenome of M. tricarinata and its genomic components, including PCGs, tRNAs, rRNAs, and CR. (B) The bar graph illustrating the relative proportions of individual nucleotides (A, T, G, and C) within the entire mitogenome and each specific genomic region. (C) The RSCU values for each amino acid and their respective codons, demonstrating codon usage bias. The codons are grouped by amino acid along the x-axis, with RSCU values plotted on the y-axis. The color-coded bars and codon boxes differentiate synonymous codons. The bar diagrams were generated by using Microsoft Excel and edited manually in Adobe Photoshop CS 8.0 Table 1List of annotated mitochondrial genes, boundaries, size, and intergenic nucleotides of M. tricarinata.GeneDirectionStartEndSize (bp)Anti-codonStart codonStop codonIntergenic nucleotidetrnF + 17171GAA− 1rrnS + 711034964− 1trnV + 1034110471TAC− 2rrnL + 1103271116090trnL2 + 2712278877TAA1nad1 + 27903760971ATGTA-− 1trnI + 3760383071GAT1trnQ–3832390372TTG− 2trnM + 3902397170CAT0nad2 + 397250101039ATGT–− 1trnW + 5010508778TCA0trnA–5088515770TGC0trnN–5158523174GTT25trnC–5257532367GCA6trnY–5330540172GTA1cox1 + 540369501548GTGAGG− 10trnS2–6941701272TGA1trnD + 7014708471GTC1cox2 + 70867787702ATGTAA0trnK + 7788786174TTT1atp8 + 78638030168ATGTAA− 10atp6 + 80218703683ATGTA-0cox3 + 87049487784ATGT–− 1trnG + 9487955569TCC0nad3 + 95569904349ATGT–0trnR + 9905997571TCG− 1nad4L + 997510,274300ATGTAA− 7nad4 + 10,26811,6441377ATGTAA9trnH + 11,65411,72471GTG0trnS1 + 11,72511,79167GCT− 2trnL1 + 11,79011,86273TAG− 1nad5 + 11,86213,6821824ATGTAA− 5nad6–13,67814,205528ATGAGG0trnE–14,20614,27671TTC3cytb + 14,28015,4261147GTGT–0trnT + 15,42715,49973TGT0trnP–15,50015,57071TGG0CR15,57116,7451175 List of annotated mitochondrial genes, boundaries, size, and intergenic nucleotides of M. tricarinata. (A) The bar chart visualizing the nucleotide composition percentages (AT and GC content) across the complete mitogenome of M. tricarinata and its genomic components, including PCGs, tRNAs, rRNAs, and CR. (B) The bar graph illustrating the relative proportions of individual nucleotides (A, T, G, and C) within the entire mitogenome and each specific genomic region. (C) The RSCU values for each amino acid and their respective codons, demonstrating codon usage bias. The codons are grouped by amino acid along the x-axis, with RSCU values plotted on the y-axis. The color-coded bars and codon boxes differentiate synonymous codons. The bar diagrams were generated by using Microsoft Excel and edited manually in Adobe Photoshop CS 8.0 Structural features of tRNAs and control region In the present study, the characteristic cloverleaf configuration was adopted by all 22 tRNA molecules, as revealed by the secondary structure prediction of M. tricarinata. The canonical structural elements, including the dihydrouridine arm, acceptor stem, anticodon loop, and TΨC arm, were exhibited by each tRNA. Owing to the absence of base pairing within the dihydrouridine arm, a simplified secondary structure was observed specifically in tRNA-serine 1 (trnS1). Both structural flexibility and functional stability of the tRNAs were provided by the presence of canonical Watson–Crick base pairs (A–T, G–C) together with non-canonical wobble base pairs (G–U). Notably, ten tRNAs of M. tricarinata viz., tRNA-leucine 2 (trnL2), tRNA-lysine (trnK), tRNA-glycine (trnG), trnQ, trnA, trnN, trnC, trnY, trnS2, and trnP were found to contain G–U wobble base pairs. These non-canonical base pairs were distributed among various structural domains, such as the acceptor stem, dihydrouridine arm, anticodon arm, and TΨC arm (Supplementary Fig. S4). The CR of M. tricarinata indicated a well-defined structural organization, consisting of multiple functional elements, including the termination-associated sequence (TAS) and four conserved sequence block (CSB) domains: CSB-F, CSB-1, CSB-2, and CSB-3 (Fig. 3). Among these domains, the CSB-3 was the longest at 23 bp, followed by CSB-2 (18 bp), CSB-F (15 bp), and CSB-1 (8 bp). Structurally, the conserved unique motifs were detected in M. tricarinata and other Geoemydinae species, including the ‘TACAT’ in the TAS, ‘ATAAGCAACCCTTGT’ in CSB-F, and ‘GACATATA’ in CSB-1. In CSB-2, a one-base-pair deletion was observed in M. tricarinata and 30 other Geoemydinae species, with the exception of Heosemys annandalii, Sacalia bealei, and Sacalia quadriocellata. These two Sacalia species exhibited a two-base-pair deletion within this domain, whereas H. annandalii retained the full 18-bp sequence, with the deletion site filled by a thymine nucleotide. Within the CSB-3, both M. tricarinata and H. annandalii presented nucleotide composition variability, in contrast to the other 31 Geoemydinae species analyzed, which showed highly conserved sequences. However, C. yunnanensis completely lacked the CSB-3 domain, as evidenced by the absence of the corresponding nucleotide sequence. Among the 34 Geoemydinae species with identifiable CR, 32 reflected a characteristic pattern of variable number tandem repeats (VNTRs) (Fig. 3). However, two species, C. yunnanensis and M. sinensis, did not show any detectable tandem repeat sequences in their CRs. Conversely, the focal species M. tricarinata harbored two distinct consensus tandem repeat motifs within its CR, consisting of a 35-bp motif repeated twice and a 5-bp motif repeated approximately 13.6 times. Additionally, Cuora galbinifrons showed the highest number of distinct consensus tandem repeat types in this dataset, with a total of five types, whereas the other Geoemydinae species possessed between one and four varying types of consensus tandem repeats.Fig. 3The linearized representations and comparative analyses of the CR sequences of M. tricarinata and 32 other Geoemydinae species. The upper panel shows aligned nucleotide sequences of conserved sequence blocks (CSB-F, TAS, CSB-1, CSB-2, CSB-3) and variable number tandem repeats (VNTRs), with nucleotide positions annotated in parentheses. The star symbol indicates nucleotides conserved among all species. The lower panels reveal species-specific tandem repeat structures within the CR, detailing repeat unit lengths (bp) and copy numbers represented by colored ovals, while gray boxes denote the absence of repeats in two Geoemydinae species. The linearized representations and comparative analyses of the CR sequences of M. tricarinata and 32 other Geoemydinae species. The upper panel shows aligned nucleotide sequences of conserved sequence blocks (CSB-F, TAS, CSB-1, CSB-2, CSB-3) and variable number tandem repeats (VNTRs), with nucleotide positions annotated in parentheses. The star symbol indicates nucleotides conserved among all species. The lower panels reveal species-specific tandem repeat structures within the CR, detailing repeat unit lengths (bp) and copy numbers represented by colored ovals, while gray boxes denote the absence of repeats in two Geoemydinae species. Phylogenetic placement The Bayesian phylogenetic construction successfully resolved all Testudines species into their particular family and subfamily classifications with strong posterior probability support in each node (Fig. 4). A close cladistic relationship was revealed within Cryptodira between the families Geoemydidae and Testudinidae (land tortoises), distinct from other marine and freshwater groups. Within Geoemydidae, the subfamily Batagurinae formed a monophyletic lineage, whereas the subfamily Geoemydinae exhibited a paraphyletic configuration. Specifically, the two Geoemyda species (G. japonica and G. spengleri) clustered independently from the remaining Geoemydinae taxa, resulting in the observed paraphyly within the constructed phylogenies. The mitogenomic analysis further delineated three internal clades within Geoemydinae subfamily, first clade comprising species from the genera Melanochelys, Sacalia, Notochelys, Heosemys, and Cyclemys; the second clade exclusively containing species of the genus Cuora; and the third clade including species of the genus Mauremys. Notably, M. tricarinata was clearly distinct from other geoemydid taxa, identified as an ancestor lineage to species of first internal clade of Geoemydinae. The Maximum-likelihood analysis produced a tree topology congruent with the Bayesian inference, providing high bootstrap support for all nodes within the Geoemydidae clade (Supplementary Fig. S5).Fig. 4The Bayesian phylogenetic tree, inferred from concatenated sequences of 13 protein-coding genes across representative taxa within the order Testudines, robustly resolves the phylogenetic placement of M. tricarinata (indicated by a red star) within the family Geoemydidae, specifically clustering within the subfamily Geoemydinae. Posterior probability values, displayed in black at each node, indicate strong statistical support for the inferred relationships. The phylogeny is built by using MrBayes version 3.1.2 and illustrated through Interactive Tree of Life (iTOL) platform, version 4 (https://itol.embl.de/). The species photograph was taken by Sushmita Kar and edited manually in Adobe Photoshop CS 8.0 The Bayesian phylogenetic tree, inferred from concatenated sequences of 13 protein-coding genes across representative taxa within the order Testudines, robustly resolves the phylogenetic placement of M. tricarinata (indicated by a red star) within the family Geoemydidae, specifically clustering within the subfamily Geoemydinae. Posterior probability values, displayed in black at each node, indicate strong statistical support for the inferred relationships. The phylogeny is built by using MrBayes version 3.1.2 and illustrated through Interactive Tree of Life (iTOL) platform, version 4 (https://itol.embl.de/). The species photograph was taken by Sushmita Kar and edited manually in Adobe Photoshop CS 8.0 Divergence time and evolution of Melanochelys The ML RelTime method produced precise estimates of divergence times for all internal nodes within the major clades of Testudines, based on the current mitogenome-derived phylogenetic topology. Within Geoemydidae, Rhinoclemmys punctularia (subfamily Rhinoclemmydinae) was identified as the earliest diverging lineage, with an estimated split dating to the middle Eocene (~ 55 Ma). Subsequently, the upper geoemydid clade comprising Mauremys and Cuora was inferred to have diversified during the Oligocene (~ 33 Ma). In contrast, the lower geoemydid clade including Geoemyda, Orlitia, Geoclemys, Pangshura, Hardella, and Batagur, was estimated to have diverged from their common ancestor in the late Eocene (~ 45 Ma). The focal species, M. tricarinata, was found to have diverged from its close relatives (Sacalia, Heosemys, Notochelys, and Cyclemys) during the Oligocene (~ 42 Ma) (Fig. 5). Collectively, the present TimeTree analyses suggest that the three major geoemydid clades originated from a common ancestor during the Eocene (~ 52–44 Ma).Fig. 5Maximum Likelihood–based TimeTree depicting divergence times of M. tricarinata and other geoemydids. Divergence estimates are shown in black at each node, with green bars representing 95% highest posterior density intervals. Calibration points (red rhombus) are based on fossilized birth–death priors16,55,108. The chronogram was generated in MEGA v12 and manually edited in Adobe Photoshop CS 8.0. The species photograph was captured by the author (A.S.) Maximum Likelihood–based TimeTree depicting divergence times of M. tricarinata and other geoemydids. Divergence estimates are shown in black at each node, with green bars representing 95% highest posterior density intervals. Calibration points (red rhombus) are based on fossilized birth–death priors16,55,108. The chronogram was generated in MEGA v12 and manually edited in Adobe Photoshop CS 8.0. The species photograph was captured by the author (A.S.) Genetic diversity and inter-specific haplotypes The present study identified 20 nucleotide differences in the cox1 gene and 72 differences in the cytb gene between M. tricarinata and M. trijuga (Fig. 6A). The pairwise genetic distance analysis of cox1 sequences revealed low intraspecific variation in M. tricarinata (0–0.34%), whereas M. trijuga exhibited greater intraspecific variability (0–1.38%). In contrast, the two species displayed substantial interspecific divergence in cox1 (3.71–5.20%) (Fig. 6B). Similarly, cytb sequences indicated low intraspecific divergence in M. tricarinata (0–0.35%) but higher variability in M. trijuga (0–3.05%), with pronounced interspecific divergence ranging from 7.99–9.29% (Fig. 6C). The interspecific haplotype network analysis further demonstrated clear genetic separation between the two species, with 21 mutational steps in cox1 and 82 mutational steps in cytb. In the combined cox1 dataset, the diversity indices were Hd = 0.8758, π = 0.02330, and S = 33, yielding three haplotypes in M. tricarinata and seven in M. trijuga (Fig. 6D, Supplementary Table S5). For cytb, Hd = 0.9571, π = 0.03862, and S = 139, recovering three haplotypes in M. tricarinata and 11 in M. trijuga (Fig. 6E). The population genetic analysis strongly supported the genetic distinctiveness of the two sister species, with high and significant differentiation indices (cox1: Fst = 0.90026, Nm = 0.03; cytb: Fst = 0.87617, Nm = 0.04).Fig. 6(A) Unique nucleotide sites in the mitochondrial cox1 and cytb genes of two Melanochelys congeners. (B) Genetic distances between M. tricarinata and M. trijuga based on the cox1 gene. (C) Genetic distances between M. tricarinata and M. trijuga based on the cytb gene. (D) Interspecific haplotype network based on cox1. (E) Interspecific haplotype network based on cytb. Nucleotide site variations and genetic distances were estimated using MEGA v12, and TCS haplotype networks were constructed in POPART v1.7. Circle sizes correspond to haplotype frequencies, vertex colors indicate M. tricarinata (red) and M. trijuga (green), and black circles represent median vectors (hypothetical haplotypes). All panels were combined and manually edited in Adobe Photoshop CS 8.0. The species photographs were taken by Sushmita Kar (M. tricarinata) and author S.S. (M. trijuga). (A) Unique nucleotide sites in the mitochondrial cox1 and cytb genes of two Melanochelys congeners. (B) Genetic distances between M. tricarinata and M. trijuga based on the cox1 gene. (C) Genetic distances between M. tricarinata and M. trijuga based on the cytb gene. (D) Interspecific haplotype network based on cox1. (E) Interspecific haplotype network based on cytb. Nucleotide site variations and genetic distances were estimated using MEGA v12, and TCS haplotype networks were constructed in POPART v1.7. Circle sizes correspond to haplotype frequencies, vertex colors indicate M. tricarinata (red) and M. trijuga (green), and black circles represent median vectors (hypothetical haplotypes). All panels were combined and manually edited in Adobe Photoshop CS 8.0. The species photographs were taken by Sushmita Kar (M. tricarinata) and author S.S. (M. trijuga). Model evaluation and predictor importance The ensemble model demonstrated strong predictive performance, as all constituent models surpassed the AUC threshold of 0.75 in both training and cross-validation datasets (Supplementary Fig. S6, Table 2). Among the models, MARS yielded the highest training AUC (0.963), while Random Forest (RF) had the lowest (0.933). In contrast, during cross-validation, RF achieved the highest AUC (0.953), whereas MARS recorded the lowest (0.841). The difference between the training and cross-validation AUC values (ΔAUC) was largest for the MARS model (0.122), while the RF model had the smallest ΔAUC (0.020). Moreover, all performance metrics such as TSS, PCC, Kappa, sensitivity, and specificity, indicated strong model performance across both the training and cross-validation datasets. Furthermore, the variance map indicated generally low variability across predictions, demonstrating high agreement among algorithms and confirming the robustness of the ensemble model in identifying core suitable habitats. Among the bioclimatic and topographic predictors, the study identified precipitation of the wettest quarter (bio_16) as the most influential variable, contributing 38.344% to the model (Table 3). This was followed by elevation and precipitation of the driest quarter (bio_17), with a contribution of 22.555% and 13.270%, respectively.Table 2Performance evaluation of each modeling technique and the ensemble model for habitat suitability predictions of M. tricarinata. BRT, Boosted Regression Trees; MARS, Multivariate Adaptive Regression Splines; GLM, generalized linear model; MaxEnt, maximum entropy; RF, random forest; CV, cross validation; AUC, area under the curve; PCC, proportion correctly classified; TSS, true skill statistic.ModelDatasetAUCΔAUCPCCTSSKappaSpecificitySensitivityBRTTrain0.9520.08684.70.6960.6830.8540.842CV0.86683.10.6660.6510.820.846GLMTrain0.9340.08187.10.7430.7320.8750.868CV0.853760.5150.5060.720.795MARSTrain0.9630.12289.50.7910.7820.8960.895CV0.84179.90.6070.5860.80.807MAXENTTrain0.9580.06489.40.7890.780.8960.893CV0.89479.80.5820.5740.740.842RFTrain0.9330.02088.70.770.7640.8750.895CV0.95386.30.7080.7080.7950.912Table 3Mean (μ) percentage contribution of environmental covariates in the final habitat suitability model for M. tricarinata. Aspect (aspect), Mean Diurnal Range (bio_2), Temperature Seasonality (bio_4), Precipitation Seasonality (bio_15), Precipitation of Wettest Quarter (bio_16), Precipitation of Driest Quarter (bio_17), Elevation (elevation), Slope (slope), BRT, boosted regression trees; MARS, multivariate adaptive regression splines; GLM, generalized linear model; MaxEnt, maximum entropy; RF, random forest.VariableBRTGLMMARSMaxEntRFMean (μ)Mean (μ) (%)aspect0.0000.0000.0000.0000.0000.0000.009bio_150.0000.2160.0630.0040.0010.05710.964bio_160.2970.2700.1820.1280.1160.19938.344bio_170.0000.2130.0860.0290.0170.06913.270bio_20.0000.0000.0000.0020.0000.0000.093bio_40.0000.1840.0000.0070.0000.0387.370elevation0.0970.2940.0000.1630.0290.11722.555slope0.0000.1340.0540.0030.0000.0387.394 Table 2Performance evaluation of each modeling technique and the ensemble model for habitat suitability predictions of M. tricarinata. BRT, Boosted Regression Trees; MARS, Multivariate Adaptive Regression Splines; GLM, generalized linear model; MaxEnt, maximum entropy; RF, random forest; CV, cross validation; AUC, area under the curve; PCC, proportion correctly classified; TSS, true skill statistic.ModelDatasetAUCΔAUCPCCTSSKappaSpecificitySensitivityBRTTrain0.9520.08684.70.6960.6830.8540.842CV0.86683.10.6660.6510.820.846GLMTrain0.9340.08187.10.7430.7320.8750.868CV0.853760.5150.5060.720.795MARSTrain0.9630.12289.50.7910.7820.8960.895CV0.84179.90.6070.5860.80.807MAXENTTrain0.9580.06489.40.7890.780.8960.893CV0.89479.80.5820.5740.740.842RFTrain0.9330.02088.70.770.7640.8750.895CV0.95386.30.7080.7080.7950.912 Performance evaluation of each modeling technique and the ensemble model for habitat suitability predictions of M. tricarinata. BRT, Boosted Regression Trees; MARS, Multivariate Adaptive Regression Splines; GLM, generalized linear model; MaxEnt, maximum entropy; RF, random forest; CV, cross validation; AUC, area under the curve; PCC, proportion correctly classified; TSS, true skill statistic. Table 3Mean (μ) percentage contribution of environmental covariates in the final habitat suitability model for M. tricarinata. Aspect (aspect), Mean Diurnal Range (bio_2), Temperature Seasonality (bio_4), Precipitation Seasonality (bio_15), Precipitation of Wettest Quarter (bio_16), Precipitation of Driest Quarter (bio_17), Elevation (elevation), Slope (slope), BRT, boosted regression trees; MARS, multivariate adaptive regression splines; GLM, generalized linear model; MaxEnt, maximum entropy; RF, random forest.VariableBRTGLMMARSMaxEntRFMean (μ)Mean (μ) (%)aspect0.0000.0000.0000.0000.0000.0000.009bio_150.0000.2160.0630.0040.0010.05710.964bio_160.2970.2700.1820.1280.1160.19938.344bio_170.0000.2130.0860.0290.0170.06913.270bio_20.0000.0000.0000.0020.0000.0000.093bio_40.0000.1840.0000.0070.0000.0387.370elevation0.0970.2940.0000.1630.0290.11722.555slope0.0000.1340.0540.0030.0000.0387.394 Mean (μ) percentage contribution of environmental covariates in the final habitat suitability model for M. tricarinata. Aspect (aspect), Mean Diurnal Range (bio_2), Temperature Seasonality (bio_4), Precipitation Seasonality (bio_15), Precipitation of Wettest Quarter (bio_16), Precipitation of Driest Quarter (bio_17), Elevation (elevation), Slope (slope), BRT, boosted regression trees; MARS, multivariate adaptive regression splines; GLM, generalized linear model; MaxEnt, maximum entropy; RF, random forest. Suitable area and corridor connectivity The model identified approximately 374,657 km2 as suitable habitat within the extended training extent in the present scenario (Fig. 7A, Supplementary Table S6). However, of this suitable area, only 238,039 km2 currently overlaps with the current forest cover (Fig. 7B). Moreover, only 31,450 km2 of the favorable habitat for this species falls within Protected Areas (PAs), representing just 11.912% of the total identified suitable area (Supplementary Table S7). Furthermore, within the IUCN-defined range, only 176,117 km2 was predicted as suitable habitat, representing just 38.19% of the total IUCN-delineated distribution under present conditions. Additionally, the study revealed overall low habitat connectivity within the training extent. The mean corridor connectivity across the range of the species was estimated to be 0.377 (training range) and 0.518 (within tree cover) under the current scenario (Fig. 7C, 7D). However, the future projections indicate a worsening trend, with a substantial decline in suitable habitat areas and corridor connectivity due to climate change. Specifically, under the SSP245 scenario, the suitable habitat within the training extent is projected to decline by 18.301–21.077% across both future time periods compared to the present (Fig. 8). The decline is more pronounced under the SSP585 scenario, with a reduction ranging from 31.580–37.106% (Fig. 8). In addition, the suitable areas overlapping with current forest cover are also projected to decrease significantly, with losses ranging between 9.953% and 27.414% across both timeframes under both scenarios. Notably, suitable habitat within the IUCN-delineated range is expected to decline steeply, with reductions between 31.159% and 39.565% under future climate projections (Fig. 9A, Supplementary Table S6). The most severe loss is observed in the 2061–2080 period under both scenarios, with the SSP585 scenario exhibiting the highest overall decline compared to SSP245 for the same timeframe (Fig. 8, Supplementary Table S6). This decline in suitable habitat is also projected within PAs, with losses ranging from 36.149–43.192% across the two time periods under SSP245 and from 37.993–50.368% under SSP585 compared to the present scenario. Moreover, the mean corridor connectivity for the species is projected to decline substantially under future climatic scenarios, with losses exceeding 80% across all projections in both the training range as well as within forest cover (Fig. 9B, Supplementary Table S8). Under the SSP245 scenario, the reduction in mean corridor connectivity ranges from 91.692–91.915% within the training range, while within forest cover the decline is 82.278% and 82.465% across the two future periods, respectively (Fig. 10). Similarly, under the SSP585 scenario, the projected losses reach 92.314% and 92.984% within the training range, whereas within forest cover, losses are estimated at 82.359% and 82.498% for the respective time periods relative to the present scenario (Fig. 10).Fig. 7Predicted suitable habitat and corridor connectivity for M. tricarinata in present scenario. (A) Suitable habitat within the training extent and (B) suitable habitat constrained within existing forest cover. The five habitat suitability classes (1–5) represent levels of model agreement, with higher classes indicating greater consensus among models. Class ‘0’ (unsuitable habitat with zero model agreement) is excluded from the maps. Panels (C) and (D) depict habitat connectivity gradients (low to high) within the training extent and within forest cover, respectively. The map was generated using ArcGIS version 10.6 and manually framed in Adobe Photoshop CS 8.0Fig. 8Predicted suitable habitat extent for M. tricarinata under two future time frames of the SSP245 and SSP585 scenarios, shown within the training extent and the current forest cover. The map was generated using ArcGIS version 10.6 and manually formatted in Adobe Photoshop CS 8.0Fig. 9(A) Bar plot depicting the suitable habitat areas (in km2) for M. tricarinata across three spatial extents: the training extent, within current forest cover, and the IUCN-defined range, under present and future climate scenarios. (B) Bar plot showing the mean corridor connectivity of M. tricarinata across the training extent and within forest cover under present and future climatic scenarios. The species photograph was taken by author (A.S.)Fig. 10Predicted corridor connectivity for M. tricarinata under two future time frames of the SSP245 and SSP585 scenarios, shown within the training extent and the current forest cover. The map was generated using ArcGIS version 10.6 and manually formatted in Adobe Photoshop CS 8.0 Predicted suitable habitat and corridor connectivity for M. tricarinata in present scenario. (A) Suitable habitat within the training extent and (B) suitable habitat constrained within existing forest cover. The five habitat suitability classes (1–5) represent levels of model agreement, with higher classes indicating greater consensus among models. Class ‘0’ (unsuitable habitat with zero model agreement) is excluded from the maps. Panels (C) and (D) depict habitat connectivity gradients (low to high) within the training extent and within forest cover, respectively. The map was generated using ArcGIS version 10.6 and manually framed in Adobe Photoshop CS 8.0 Predicted suitable habitat extent for M. tricarinata under two future time frames of the SSP245 and SSP585 scenarios, shown within the training extent and the current forest cover. The map was generated using ArcGIS version 10.6 and manually formatted in Adobe Photoshop CS 8.0 (A) Bar plot depicting the suitable habitat areas (in km2) for M. tricarinata across three spatial extents: the training extent, within current forest cover, and the IUCN-defined range, under present and future climate scenarios. (B) Bar plot showing the mean corridor connectivity of M. tricarinata across the training extent and within forest cover under present and future climatic scenarios. The species photograph was taken by author (A.S.) Predicted corridor connectivity for M. tricarinata under two future time frames of the SSP245 and SSP585 scenarios, shown within the training extent and the current forest cover. The map was generated using ArcGIS version 10.6 and manually formatted in Adobe Photoshop CS 8.0 Landscape shape dynamics of suitable extent The suitable areas within the training extent also exhibited changes in spatial configuration and the shape geometry in future projections, reflecting the impacts of shifting climatic conditions (Table 4). In addition to the overall decline in suitable areas, the future scenarios also indicate a significant reduction and, in some cases, complete loss of suitable habitat patches. This is reflected by a decrease in the number of patches (NP), ranging from 46.340–52.487%. The loss of patches also led to a notable decrease in patch density (PD), with a reduction of up to 52.497% in the future scenarios. Moreover, patch size also decreased substantially, as shown by a decline in the LPI by more than 20.272%. This shrinkage also led to a reduction in total edge (TE) length by over 25.420%. As patches became smaller and more isolated, their geometrical complexity diminished, as indicated by a decline in LSI of up to 29.146%. Furthermore, the decline in spatial proximity between patches is indicated by a reduction in the aggregation index (AI) of up to 13.97%, which highlighted the increasing fragmentation and isolation of suitable habitats under future climate scenarios.Table 4Landscape geometry metrics of suitable habitat patches for M. tricarinata within the training extent under present and future climatic scenarios. SSP, shared socioeconomic pathways; NP, number of patches; PD, patch density; LPI, largest patch index; TE, total edge; LSI, landscape shape index; AI, aggregation index.ScenarioNPPDLPITELSIAIPresent392114,697,064.5992.7941391.76071.00888.536SSP245 (2041–2060)21047,884,6182.2281037.96858.60285.564SSP245 (2061–2080)20657,738,4672.1361020.06455.89580.960SSP585 (2041–2060)19237,206,3312.074958.91252.16278.484SSP585 (2061–2080)18636,981,4842.000905.93650.31276.167 Table 4Landscape geometry metrics of suitable habitat patches for M. tricarinata within the training extent under present and future climatic scenarios. SSP, shared socioeconomic pathways; NP, number of patches; PD, patch density; LPI, largest patch index; TE, total edge; LSI, landscape shape index; AI, aggregation index.ScenarioNPPDLPITELSIAIPresent392114,697,064.5992.7941391.76071.00888.536SSP245 (2041–2060)21047,884,6182.2281037.96858.60285.564SSP245 (2061–2080)20657,738,4672.1361020.06455.89580.960SSP585 (2041–2060)19237,206,3312.074958.91252.16278.484SSP585 (2061–2080)18636,981,4842.000905.93650.31276.167 Landscape geometry metrics of suitable habitat patches for M. tricarinata within the training extent under present and future climatic scenarios. SSP, shared socioeconomic pathways; NP, number of patches; PD, patch density; LPI, largest patch index; TE, total edge; LSI, landscape shape index; AI, aggregation index. Discussion The reptiles represent nearly one-third of all tetrapod species globally, yet approximately one-fifth of them are currently threatened with extinction88. However, despite this alarming status, these cold-blooded vertebrates continue to be one of the most overlooked groups in both scientific research and conservation initiatives89. Therefore, there is a pressing need to deepen the scientific understanding of their biology and ecological requirements to better inform conservation strategies90. This imperative is further underscored by international commitments under the CBD, which emphasize integrated research, particularly for threatened taxa34,35. In alignment with these global priorities, the present study adopts both genetic and ecological approach by unveiling the complete mitochondrial genome of M. tricarinata to elucidate its genomic structure and variation, phylogenetic relationship, genetic diversity, and evolutionary pattern, while simultaneously evaluating its habitat dynamics under present and projected future scenarios. The insights outlined here are intended to support more effective and informed conservation planning for M. tricarinata and other testudines species. Mitogenomic insights Overall, the mitogenomic characterization of M. tricarinata in this study significantly advances the understanding of the systematics and molecular evolution within the Geoemydidae family. The notable nucleotide variability observed within the PCGs, rRNAs, and CR highlights the potential of these genomic elements as informative molecular markers for assessing genetic diversity in Geoemydidae91,92. The observed sequence variation among different lineages of Testudines mitogenomes likely results from the gradual accumulation of mutations over time, with natural selection predominantly preserving mitochondrial function93. The mitogenomic analysis of M. tricarinata revealed that 28 genes are encoded on the H-strand and nine on the L-strand, aligning with the conserved mitogenome organization typical of Testudines24,94. The intergenic region of this species also identified the longest non-coding spacers between trnN–trnC and nad4–trnH, alongside prominent overlapping regions between cox1–trnS2 and atp8–atp6, as observed in other geoemydids25. The predominance of the canonical ATG start codon across most of the PCGs in M. tricarinata reflects a conserved trait in Testudines mitogenomes22,26. Conversely, the incomplete stop codons found in several genes are presumed to be completed via post-transcriptional polyadenylation, a well-documented mechanism observed in vertebrate mitochondrial transcripts95. In the present study, eight codons of M. tricarinata were significantly overrepresented (RSCU > 1.6), whereas twelve were underrepresented (RSCU < 0.6). This codon usage pattern may reflect the combined influence of mutational pressure and natural selection in terrestrial ecosystems, potentially selecting for codons that align with mitochondrial translational efficiency in reptiles96. The leucine codon was the most prevalent in M. tricarinata, consistent with patterns reported in other geoemydids mitogenomes, indicating a conserved codon bias for this amino acid22,96. Conversely, methionine and tryptophan, each encoded by a single codon (AUG and UGG, respectively), exhibited low RSCU values, correlating with their limited synonymous codons in M. tricarinata as evidenced in turtle mitogenomes with modified genetic code97. Interestingly, the presence of non-canonical G–U wobble base pairs in the tRNAs of M. tricarinata may contributes to structural diversity and enhance versatility in recognizing codons98. Thus, the characterization of this novel M. tricarinata mitogenome provides a framework for investigating the influence of genetic variation and selective pressures. The CR of M. tricarinata exhibits a well-defined organization comprising the TAS and CSB-F, CSB-1, CSB-2, and CSB-3 domains, with species-specific nucleotide variations and two distinct tandem repeat motifs. Such structural components align with the functional significance of the CR in regulating mitochondrial DNA replication and transcription, as demonstrated for Testudines99. Furthermore, lineage-specific tandem repeats in M. tricarinata contribute to variations in CR length, with the highly variable sites serving as effective molecular markers for assessing intraspecific variation and population differentiation100. Matrilineal phylogenetic relationships DNA sequences of extant Testudines from diverse geographic regions have become pivotal in systematics and conservation biology, particularly through their application in phylogenetic analyses101–105. The phylogenetic studies are essential for biodiversity research and conservation, as they refine taxonomic classifications, elucidate evolutionary histories, and reveal species unique roles, ecosystem functions, and resilience for guiding more informed conservation decisions106,107. A recent multi-marker study revealed that past climatic changes played a crucial role in shaping the diversification of modern turtles, encompassing all 19 geoemydid genera16. In addition, previous studies have employed both mitochondrial and nuclear markers to resolve species-level identification among Melanochelys congeners13–15. However, earlier studies often relied on limited taxon sampling or distinct sets of genetic markers, restricting their ability to reconstruct relationships across the entire lineage. Thus, the present mitogenome-based framework provides a higher-resolution perspective on the phylogenetic relationships of Geoemydidae within the broader Testudines lineage. The clustering pattern of M. tricarinata inferred from the complete mitogenome is congruent with earlier investigations, supporting its close affinity with Sacalia, Notochelys, Heosemys, and Cyclemys16. At the same time, the mitogenomic analyses also revealed a paraphyletic clustering of Geoemydinae members, consistent with prior findings23–26,101. Moreover, the mitogenome based cladistic patterns recovered here contrast with the most extensive nuclear gene-based phylogenetic attempt, which included 63 species (84% of all geoemydids)16. Despite these advances, previous analyses either focused on partial genetic markers or excluded M. tricarinata, leaving a substantial gap in understanding its precise systematic position. Although the present investigation incorporates M. tricarinata, the dataset encompasses only 64% of recognized geoemydid taxa. Therefore, to achieve a more comprehensive understanding of phylogenetic relationships within Geoemydidae, additional mitogenomic data are required for the remaining 27 species. Evolutionary pattern of Melanochelys Beyond providing phylogenetic resolution, divergence time analyses are essential for reconstructing the evolutionary history of Testudines. Although divergence time estimation in these ancient lineages has been widely attempted, the results often remain inconsistent, largely due to uneven sampling strategies or differential use of molecular markers16,55,108,109. Notably, the most recent phylogenomic study, which included nine geoemydid genera but excluded Melanochelys, placed the diversification of the family Geoemydidae during the Eocene55. A comparable timeframe was further inferred from multiple nuclear markers, although only a single Melanochelys congener (M. trijuga) was included within the broader Testudines lineage16. However, both recent studies omitted M. tricarinata from their datasets, leaving the divergence time of this species unresolved. In contrast, the present study uniquely integrates mitogenomes from 48 geoemydid species under 14 genera with explicit inclusion of M. tricarinata, thereby offering an additional and more targeted perspective on the evolutionary history of this species. Thus, by applying known fossilized birth–death priors, our timetree reveals the divergence of M. tricarinata from its closest relatives during the Oligocene. This timeline contrasts with earlier results, which dated the divergence of its congener M. trijuga to the early Miocene16. Thus, while prior studies placed the origin of M. trijuga in the Miocene, the present mitogenome-based timetree suggests an older, Oligocene origin for M. tricarinata. This discrepancy underscores the need for further comprehensive analyses that incorporate both congeners of Melanochelys (M. tricarinata and M. trijuga), employ broader genomic datasets, and utilize multiple fossil calibrations to refine divergence time estimates for this lineage. Genetic distinctiveness of Melanochelys congeners This study also provides molecular evidence of genetic differentiation between the two congeners of Melanochelys (M. tricarinata and M. trijuga), as revealed by analyses of two partial mitochondrial loci (cox1 and cytb). Notably, the observed nucleotide differences in cox1 and cytb, along with substantial interspecific genetic distances, confirm the distinctiveness of these two species. The focal species M. tricarinata exhibited low intraspecific variation for both markers, whereas the M. trijuga displayed higher genetic diversity. The haplotype network analyses further highlighted non-overlapping haplotypes, and the population genetic indices with high Fst ​corresponding to Nm values below ‘1’ indicate strong genetic isolation and limited gene flow between two Melenochelys species110. Specifically, the greater number of haplotypes and higher nucleotide diversity identified in M. trijuga propose a more complex demographic history or greater population structuring compared to M. tricarinata. Consistent with this, the haplotype network of M. trijuga illustrates multiple sub-clusters, which likely demonstrate regional population differentiation and may correspond to recognized putative subspecies (M. trijuga trijuga, M. trijuga coronata, M. trijuga edeniana, M. trijuga indopeninsularis, M. trijuga parkeri, and M. trijuga thermalis)5,111. Nonetheless, further genomic and geographically extensive analyses are required to determine whether these lineages warrant formal taxonomic recognition. Ecological perspective The SDM identified potential climate refugia for this species across its range under both current and future climatic scenarios. Specifically, under present conditions, the delineated favorable habitat within the training extent was estimated at approximately 374,657 km2, with a mean corridor connectivity of 0.377 (training range) and 0.518 (within forest cover). Notably, the coverage of suitable areas within PAs corresponds to only 11% in the present scenario, highlighting that the majority lies outside protection and emphasizing the need to look beyond existing PAs also and to strengthen protection efforts in areas beyond the currently designated boundaries112. However, considering that this species is primarily terrestrial and occupies forested and riverine grassland habitats, the suitable area within the existing forest cover was found to be 238,039 km2. This highlights the need to prioritize the protection of these forested landscape patches within and beyond its known IUCN range. Furthermore, within the IUCN range, only about 38.19% of the area was identified as ideal habitat. Therefore, suitable areas located beyond the designated IUCN range, where species presence has also been documented, warrant consideration for inclusion in future conservation assessments and range revisions113. Furthermore, future projections indicate a substantial loss of suitable habitat within the training extent, with reductions of up to 40% under future scenarios driven by climate change. This habitat loss is not limited to the species overall range but is also evident within PAs, where reductions of up to 50.368% are projected in future scenarios compared to the present. This finding aligns with earlier studies and underscores the need for PA management to integrate climate change monitoring into planning to improve conservation effectiveness, especially in biodiversity hotspots facing rapid climatic shifts114,115. This concerning trend of habitat declines caused by the climatic shifts is also well-documented in recent ecological studies116. The present study similarly projects substantial habitat loss for M. tricarinata, with future scenarios indicating marked reductions in suitability across the foothills of Central Himalayas, parts of its current range, similar to other testudines species such as Hardella thurjii and Geoclemys hamiltonii25,26. This anticipated loss is largely attributed to climatic shifts, highlights regions where environmental conditions may no longer remain viable for the persistence of this geoemydid species117,118. The areas projected to become unsuitable in the future, despite current species presence, could potentially lead to local extinctions and consequent range shrinkages. Such contractions not only reduce the distribution of M. tricarinata but can also disrupt broader biological and ecological dynamics119,120. The habitat decline is further compounded by severe fragmentation and altered landscape configuration, reducing corridor connectivity under future scenarios. This is evident from the loss of suitable patches, as indicated by decreasing NP and PD, with remaining patches becoming smaller, more dispersed, and increasingly isolated. This is evident from reductions in LPI, TE, and AI, collectively highlighting the intensifying fragmentation of suitable habitats within the study area. Such fragmentation is likely to undermine landscape connectivity, restricting dispersal and gene flow among subpopulations121. These extensive habitat losses and fragmentation align with the corridor connectivity assessment, with mean connectivity projected to decline by over 80% in the future across both the training range and forested areas. This further underscores the impact of climatic shifts in disrupting functional connectivity between suitable habitat patches122–125. The increased isolation of these patches may elevate risks of genetic drift and inbreeding, ultimately reduce genetic diversity and limit the capacity to adapt to ongoing environmental changes126,127. Additionally, the emergence of bioclimatic variable such as precipitation of the wettest quarter (bio_16) and precipitation of the driest quarter (bio_17) as important predictor with contribution of 38.344% and 13.270%, respectively highlights their importance for this ectothermic species. These bioclimatic variables can significantly influence nest incubation environments, which are crucial for hatchling viability, embryonic development, and sex determination in temperature-dependent species128. The suboptimal thermal conditions may lead to lower hatching success, skewed sex ratios, and reduced offspring fitness128. Additionally, shifts in precipitation can alter hydrological regimes, affecting the availability and stability of aquatic habitats essential for juvenile survival and growth129. Such fluctuations may also cause habitat dryness or flooding, impacting resource availability and shelter from predators128. Overall, these climatic pressures may reduce reproductive success, hinder juvenile development, and disrupt population connectivity, thereby threatening long-term persistence. The findings highlight the significance of identified regions as potential climate refugia that could help sustain M. tricarinata populations under current and future climatic conditions129,130. Limitations and recommendations The present study provides essential baseline information for genetic and ecological investigations on M. tricarinata. It reports the novel mitogenome of M. tricarinata, detailing its structure and variation relative to other geoemydids. However, as the mitogenome of its congener M. trijuga remains unexplored, generating and analyzing it would provide a more comprehensive understanding of mitogenomic evolution within Melanochelys. While the phylogenetic placement of M. tricarinata is consistent with previous studies, the TimeTree analyses revealed notable discrepancies in divergence time estimates, likely attributable to uneven taxon sampling and the use of limited calibration points. Therefore, expanding mitogenomic data across additional Testudines families would improve divergence time estimation and yield a more robust evolutionary framework. The genetic diversity assessments based on two partial mitochondrial genes indicated high intraspecific variation in M. trijuga, which may reflect the recognized subspecies diversity. Thus, the broader sampling and inclusion of complete mitogenomes from multiple populations of M. trijuga would better capture genetic diversity and demographic patterns across its range. Further, to refine phylogenetic and evolutionary inferences, large-scale phylogenomics or whole-genome sequencing of all 75 extant geoemydid species is recommended. Furthermore, this study has certain inherent limitations regarding the SDM that also highlight directions for future research. The reliance on a single GCM may not fully capture the breadth of uncertainties associated with climate projections and hence multiple GCMs need to be integrated to provide a more comprehensive representation of potential climatic shifts. Similarly, the present analysis considered only climatic and topographic factors, which capture just part of the ecological complexity. Incorporating additional biotic and environmental variables such as habitat characteristics, species interactions, and land-use dynamics would enhance the ecological relevance and predictive power of future models. It is therefore recommended to conduct assessments within the identified habitat envelope to gather additional presence records across the species range. Further research on aspects such as home range and related ecological parameters is also needed to strengthen future conservation efforts. Building on the foundation established here, subsequent research can generate more nuanced insights into species responses to changing environmental conditions. Conservation implications Nevertheless, as more genetic data for this species become available across its range, these can be integrated with habitat suitability models to enable a more robust landscape genetics assessment, thereby strengthening conservation planning. Such integration could help identify potential barriers to gene flow as well as critical habitat corridors that sustain connectivity among different populations. These efforts will help elucidate the genetic structure within and between populations, guiding more precise conservation strategies. Such an approach is particularly relevant for informing conservation breeding programs, especially in mitigating risks associated with inbreeding effects, genetic bottleneck, and random demographic fluctuations. Additionally, the habitat suitability models identified specific areas under current and future scenarios that warrant prioritization for conservation. Hence, focused field surveys and dedicated research in these regions are crucial for proper assessment and the formulation of effective conservation plans. Based on these findings, it is advisable to consider revising the IUCN range of M. tricarinata to include areas that match the predicted environmental envelope and where species presence has been recorded. Further, managing land cover change is also imperative, particularly the protection of existing forests and timber plantations such as Sal Forest (Shorea robusta), which constitute primary habitats for this species. In addition to this, the monitoring of illegal wildlife trade routes should be strengthened through regular surveillance in collaboration with enforcement agencies and conservation organizations. Moreover, community outreach and awareness programs should be initiated at the local level, especially in areas surrounding identified suitable habitats. These initiatives would benefit from collaborative efforts between international agencies, NGOs, and government departments to ensure broad participation and sustained impact. Collectively, these conservation measures will not only help safeguard M. tricarinata populations but also contribute to the overall health and resilience of ecosystems across South Asia. Conclusion This study adopts two independent scientific approaches of mitogenomic data and habitat dynamics to support the development of informed conservation strategies for the threatened M. tricarinata in the Indian subcontinent. Specifically, it presents the first complete mitogenome analysis of M. tricarinata, contributing novel insights into its mitogenomic architecture, sequence variation, phylogenetic interpretation, and divergence time estimation within the broader Testudines lineage, thereby enhancing understanding of its systematic position and matrilineal evolutionary history. Complementing the genetic analysis, an ensemble SDM approach provides a detailed assessment of habitat dynamics and corridor connectivity under current and projected future climate scenarios. Together, this multidisciplinary framework establishes a robust scientific foundation for designing evidence-based management and conservation strategies for M. tricarinata. Moreover, these findings hold broader significance for the conservation of other freshwater turtle species globally, particularly in the face of accelerating climate change. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1. Supplementary Material 1.