Massive Genetic Diversity Loss Within Asian Elephant Subspecies: New Insights for Determining Conservation Priorities

Abstract

In this study, we provide a comprehensive genomic analysis of population structure, genetic diversity, and load across the four Asian elephant (Elephas maximus) subspecies using whole-genome sequencing data of 81 wild-born individuals, with most of them currently living under human care. Our results reveal striking contrast in inbreeding: the Bornean subspecies (E.m. borneensis) exhibits the highest level of inbreeding, with ∼65% of its genome in runs of homozygosity (ROH), and the lowest genome-wide heterozygosity (∼46 heterozygous sites per 100 kb). The Sumatran subspecies (E.m. sumatranus) shows intermediate levels of inbreeding (35% ROH) and heterozygosity (∼101 sites per 100 kb), while the mainland and Sri Lankan subspecies (E.m indicus and E.m. maximus) display similarly low inbreeding (∼10% to 11% ROH) and high heterozygosity (∼130 to 134 sites per 100 kb). Notably, when correcting for ROH coverage, heterozygosity levels are comparable across all four subspecies (∼136 to 155 sites per 100 kb), indicating that differences in genetic diversity are primarily driven by loss of heterozygosity within ROHs due to inbreeding. Genetic load remains relatively low in all subspecies, especially in highly conserved genomic regions. The high genetic diversity observed in the island-dwelling E.m. maximus subspecies likely reflects historical gene flow linked to intensive human-mediated elephant trade, raising questions about its taxonomic status. Given its high inbreeding, E.m. borneensis represents the highest conservation priority, with genetic rescue as a potential management strategy. This study highlights the urgent need for focused conservation efforts and further genomic research to safeguard the unique genetic diversity and long-term viability of Asian elephants.

Significance.

Understanding how demographic history shapes genetic diversity and load is central to predicting the persistence of endangered species. Using whole-genome resequencing across all four Asian elephant subspecies, we show that extreme variation in inbreeding, rather than subspecies-specific divergence, primarily drives differences in genome-wide diversity. The Bornean elephant exhibits exceptionally high levels of homozygosity, while mainland populations retain high diversity, yet all subspecies share similar levels of heterozygosity once inbreeding is accounted for, indicating largely shared ancestral genetic variation. Despite strong differences in inbreeding, we find limited evidence for the accumulation of highly deleterious mutations, suggesting effective purging even in small populations. Instead, moderate-effect variation persists, highlighting the importance of nearly neutral processes in shaping genetic load. Our results demonstrate how integrating runs of homozygosity with genome-wide load estimates refines conservation assessments and identifies populations where genetic rescue may be most effective.

Introduction

Genetic load reflects the combined effects of demographic processes and evolutionary forces on mutations within populations (Loewe and Hill 2010; Charlesworth and Willis 2009). Selection and genetic drift jointly determine the fate of deleterious, neutral, and beneficial variants, but their relative importance depends on population size (Kimura 1983; Ohta 1992). In small populations, drift weakens the efficiency of selection, allowing mildly deleterious alleles to persist and beneficial alleles to be lost (Whitlock 2000; Kyriazis et al. 2021). Under these conditions of nearly neutral evolution, genetic load is expected to increase through the accumulation of moderately deleterious mutations (Ohta 1992; Eyre-Walker and Keightley 2007). Small populations also experience inbreeding, which increases homozygosity and exposes recessive deleterious alleles (Charlesworth and Charlesworth 1987; Hedrick and Garcia-Dorado 2016). While inbreeding can sometimes facilitate purging of strongly deleterious mutations, it often comes at the cost of reduced effective population size and fitness (Xue et al. 2015; Robinson et al. 2019). Advances in genomic sequencing now make it possible to empirically test these predictions in natural populations of endangered species, providing critical insights into how demographic history shapes genetic load and long-term viability (Dussex et al. 2021; Robinson et al. 2023).

Owing to their long generation times and large home ranges typical for large mammals, elephantids are particularly vulnerable to evolutionary forces acting in small populations. The Asian elephant (Elephas maximus L.), a keystone species, plays a crucial role in maintaining the ecological balance of the forests in South and Southeast Asia. However, populations have declined sharply due to anthropogenic habitat fragmentation, human-elephant conflict, and poaching (Menon and Tiwari 2019; de Silva et al. 2023; Thant et al. 2023; Wang et al. 2025). Currently, the Asian elephant species is listed as Endangered on the IUCN Red List status, with current total population estimates suggesting only 30,000 to 40,000 individuals remain in the wild, roughly 5-fold fewer than wild populations of African savannah elephants (Loxodonta africana) (Chase et al. 2016; Thouless et al. 2016; Gobush et al. 2022) and comparable to current population estimates of African forest elephants (Loxodonta cyclotis: Maisels et al. 2013; Thouless et al. 2016; Gobush et al. 2021). Numbers for two of the four currently recognized Asian elephant subspecies (Williams et al. 2020) are even worse: the wild population of the Bornean elephant subspecies (E. maximus borneensis) (Alfred et al. 2010; McLean et al. 2024) is estimated to be around 2,000 individuals. Similar population estimates are found for the Sumatran elephant (E. maximus sumatranus), which is listed as Critically Endangered (Hedges et al. 2005; Williams et al. 2020). Habitat fragmentation has further isolated Asian elephant populations, limiting gene flow and increasing the risk of inbreeding and loss of genetic diversity (Fernando and Lande 2000; Frankham 2005). These processes reduce adaptive potential and population resilience, threatening long-term survival. Genetic diversity is a key factor in the long-term survival of species, as it influences adaptability to environmental change, disease resistance, and reproductive success (Loewe and Hill 2010), making its erosion a critical concern. Although inbreeding effects may be less immediate than habitat loss or poaching, they pose a significant long-term risk to population viability (Cook et al. 2023).

Prior to the advent of next generation sequencing, Asian elephant genetics has traditionally been studied using microsatellite and mitochondrial markers, which revealed considerable variation in genetic diversity across different populations (Vidya et al. 2005; Fickel et al. 2007; Ahlering et al. 2011; Goossens et al. 2016; Budd et al. 2021; De et al. 2021; Maurer et al. 2024; Sharma et al. 2018). These studies consistently found that mainland populations, such as those in India, Myanmar, Vietnam, and Thailand, exhibit higher genetic diversity compared to isolated island populations in Sumatra and Borneo. The Bornean elephant population, in particular, harbors the lowest levels of genetic diversity, likely due to historical isolation and restricted gene flow (Fernando et al. 2003; Sharma et al. 2018). Surprisingly, the Sri Lankan population maintains the highest genetic diversity among Asian elephant populations, despite being an island population with a relatively small size (Fernando et al. 2011). Many populations also show signs of incipient divergence due to genetic drift (Goossens et al. 2016; De et al. 2024), indicating the need for higher-resolution genome-wide data. Comprehensive genome-wide assessments are essential to inform conservation efforts alongside evolutionary analyses to discern the role of nearly neutral evolution in shaping genetic variation in these endangered populations.

Advances in molecular techniques, particularly whole-genome sequencing (WGS), offer new opportunities to investigate population structure, admixture, genome-wide genetic diversity including runs of homozygosity, and genetic load in nonmodel species (Bertorelle et al. 2022; Theissinger et al. 2023). Recent genomic studies on elephants, including both Asian and the two African species, have analyzed WGS data to reveal the impact of habitat fragmentation on population connectivity, demonstrating reduced contemporary gene flow between populations, and to reconstruct demographic history, identifying ancient and recent population bottlenecks (Palkopoulou et al. 2018; Tollis et al. 2021; Sun et al. 2023; Khan et al. 2024). Most importantly, a recent study incorporating WGS data from all recognized Asian elephant subspecies uncovered clear population structure, with relatively recent splits (between 50,000 and 170,000 years) separating Bornean, Sumatran, and mainland Asia populations, and modest barriers to gene flow in Sri Lanka (Kappelhof et al. 2025). These results are in line with the four currently recognized Asian elephant subspecies: E. maximus indicus on the mainland, E. maximus maximus in Sri Lanka, E. maximus sumatranus on Sumatra, and E. maximus borneensis on Borneo (Williams et al. 2020; McLean et al. 2024). Although these initial studies included limited sample sizes, they suggest that effective population sizes within subspecies may be even smaller than overall census numbers, potentially accelerating genetic declines, especially in island populations.

Here, we integrate whole-genome resequencing data across the geographic range of the Asian elephant to estimate genome-wide diversity and compare patterns of genetic diversity, inbreeding, and genetic load within and among all four recognized subspecies. While previous genomic studies have provided important insights into individual populations or subspecies, they have largely been limited in their sample sizes and in their ability to directly compare genomic metrics across the full species range within a single analytical framework. By applying a standardized comparative population genomic approach, we enable consistent estimation of key evolutionary parameters across subspecies.

Specifically, we (i) quantify and compare genome-wide genetic diversity across subspecies, (ii) assess variation in inbreeding levels using the genome-wide runs of homozygosity (ROH) distribution pattern, (iii) estimate and compare genetic load among subspecies, and (iv) evaluate how demographic history and geographic isolation have shaped present-day patterns of genetic variation across the species range.

Together, these analyses provide a unified empirical framework to examine how nearly neutral evolutionary processes and variation in inbreeding shape genome-wide variation in a long-lived, endangered megafaunal species. More broadly, this study addresses how demographic history and population structure interact to shape genetic health in small and declining populations, a central question in conservation genomics.

Results

We used a whole-genome population genomic approach to assess genetic diversity, inbreeding, and genetic load across the four Asian elephant subspecies. Our dataset included 81 wild-born Asian elephant individuals (48 individuals newly sequenced here) with known provenance and traceable records of origin, with most of them currently living in North American or European zoological institutions (Table S1). Their geographic origins and sampling distribution across the four subspecies are illustrated in Fig. 1. Of these, 54 individuals belonged to E. maximus indicus (hereinafter indicus) from India (n = 18), Myanmar (n = 9), Thailand (n = 11), Vietnam (n = 9), and Peninsular Malaysia (n = 7); 10 individuals represented E. maximus maximus (hereinafter maximus) from Sri Lanka, six individuals were E. maximus sumatranus (hereinafter sumatranus) from Sumatra, and 11 individuals were E. maximus borneensis (hereinafter borneensis) from Borneo.

Fig. 1.

Current distribution of wild Asian elephant populations and the currently recognized subspecies designations across Asia. Sampling locations represent the country of origin and sample sizes of the elephant populations included in this study; exact geographic sampling locations within countries are unknown.

Whole-genome sequences of these 81 Asian elephants were aligned to the E. maximus reference genome mEleMax1 (De Panis et al. 2026), resulting in an average coverage of ∼22× (6.0× to 42.6×) per individual (details in Table S1). After population variant calling and filtering, we retained and analyzed 4,947,714 high-quality single-nucleotide polymorphisms (SNPs).

Clear Population Structure Across Asian Elephant Subspecies

Principal component analysis (PCA) of SNP data shows three distinct clusters corresponding to Asian elephant subspecies indicus, sumatranus, and borneensis (Fig. 2a and b), while the fourth recognized subspecies, maximus, does not form a clearly distinct genetic cluster. For visualization and interpretation of fine-scale structure within mainland Asia, individuals were grouped by country of origin rather than by subspecies. PC1 explains 14.4% of the total variance and separates the Bornean cluster from the others, with the Mainland and Sri Lankan cluster on one extreme and the Sumatran cluster intermediate, leaning more toward the Bornean cluster. PC2 explains 7.9% of the total variance and separates the Sumatran cluster from the other clusters, placing it at one extreme, while the Mainland/Sri Lankan cluster and Bornean cluster occupy the opposite end, separated from each other. Within the Mainland and Sri Lanka cluster, PC2 reveals a geographic gradient: individuals from India at the top and individuals from Peninsular Malaysia at the bottom, with Myanmar, Thailand, and Vietnam in between. However, some individuals deviate from this pattern: one Sri Lankan elephant (in yellow) clusters with Indian individuals; two Thai elephants (in lighter green) cluster with Indian individuals; one Sumatran elephant (in blue) clusters between Indian and Thai/Myanmar individuals. Lastly, three Indian elephants cluster with Vietnamese and Peninsular Malaysian individuals. PC3 (7.3% variance) further emphasizes this gradient and unexpected clustering. Notably, Sri Lankan individuals align most closely with Myanmar individuals on PC3, contrasting with their position on PC2.

Fig. 2.

Population structure of Asian elephants based on whole-genome SNP data from eight geographic populations: Sri Lanka (LKA), India (IND), Myanmar (MMR), Thailand (THA), Vietnam (VNM), Peninsular Malaysia (MAL), Sumatra (SUM), and Borneo (BR). a and b) PCA reveal three major genetic clusters corresponding broadly to Bornean, Sumatran, and Mainland/Sri Lankan elephants, while mainland populations form a geographic gradient from India to Peninsular Malaysia. PC1 separates Bornean elephants from all other populations, whereas PC2 distinguishes Sumatran elephants from mainland and Bornean individuals. Several individuals show atypical clustering patterns, including one Sri Lankan and two Thai individuals clustering with Indian elephants, one Sumatran individual intermediate between mainland populations, and three Indian individuals clustering closer to Vietnamese and Peninsular Malaysian elephants. c) Admixture analyses support the PCA results. At K = 2, Bornean elephants form a distinct ancestry cluster, with Sumatran elephants showing mixed ancestry. At K = 3, Sumatran elephants form a separate cluster, while higher K values progressively differentiate peripheral populations, particularly Vietnam and Peninsular Malaysia. Shared ancestry patterns among mainland populations and several atypical individuals are consistent with the PCA results. Individuals showing genomic clustering inconsistent with their reported geographic origin are highlighted in the admixture plots. Individuals outlined in black were excluded from downstream analyses due to low sequencing coverage (<10×), whereas individuals outlined in red were retained and assigned to subspecies based on their genomic clustering rather than their reported origin.

Admixture Analyses Corroborate PCA-based Population Structure

Admixture analyses corroborate the population structure observed in the PCA results (Fig. 2c), with individuals again grouped by country of origin to resolve structure within mainland populations. Seven individuals that appeared differently than expected according to their documented country of origin are highlighted in red and black. Cross-validation (CV) analyses indicated the lowest error at K = 2, while differences among higher K values were small, suggesting limited support for a single optimal K beyond broad-scale structure. At K = 2, Bornean elephants form a distinct cluster separate from the other Asian elephants, while Sumatran individuals exhibit genomic signatures of mixed ancestry from both clusters. At K = 3, Sumatran individuals also form a distinct cluster, although one individual exhibits atypical ancestry and stands out compared to the other Sumatran individuals, consistent with the PCA results. At K = 4, Vietnamese and Peninsular Malaysian individuals separate from the Mainland/Sri Lanka cluster but retain shared genomic signatures. Moreover, at K = 4, one Sri Lankan and one Thai individual show significant shared ancestry with Indian elephants. At K = 5, most Indian elephants form a distinct cluster, while three Indian individuals show significant shared ancestry with Vietnamese and Peninsular Malaysian individuals that became separate clusters. Sri Lankan elephants also show admixture with Sumatran individuals at K = 5, though they consistently share ancestry with mainland populations across all K values. At the higher K values, Sri Lankan elephants primarily share ancestry with Peninsular Malaysian and Indian elephants. Increasing K primarily led to the separation of geographically distant populations (eg Borneo, Sumatra, Malaysian Peninsula/Vietnam), a pattern consistent with a geographic effect such as isolation-by-distance rather than discrete population clusters.

For downstream population genomic analyses, we retained only individuals with a minimum sequencing coverage of 10× to ensure sufficient coverage essential for reliable analysis of genetic load and genetic diversity. After excluding 16 individuals with coverage less than 10×, 65 individuals remained for population genomic analyses, 42 from the indicus subspecies (India, n = 14; Peninsular Malaysia, n = 6; Myanmar, n = 9; Thailand, n = 4; Vietnam, n = 9), 7 from maximus (Sri Lanka), 11 from borneensis (Borneo), and 5 from sumatranus (Sumatra). Of the seven individuals that appeared differently in the PCA and admixture than expected according to their reported country of origin, five (highlighted in black in the admixture plot) were excluded for downstream analyses due to a coverage of less than 10×, and two individuals (highlighted in red in the admixture plot) were included in downstream analyses but were assigned to the Asian elephant subspecies that corroborated with their genomic clustering in PCA and admixture plot in Fig. 2. PCA and admixture plots for the dataset with the 16 individuals with coverage less than 10× excluded can be found in Fig. S2. As this study focuses on the population genomics of the four currently recognized Asian elephant subspecies, all subsequent analyses are conducted using this subspecies-level grouping. A TreeMix maximum-likelihood phylogenetic reconstruction based on genome-wide SNP data further supported this grouping scheme (Fig. S3).

Significant Differences in Diversity and Inbreeding Among Asian Elephant Subspecies

In small populations, inbreeding becomes more common as mating among closer relatives becomes inevitable under short coalescent times. This leads to ROH as regions on independently inherited haplotypes identical by descent accumulate. As shown in Fig. 3a, borneensis shows the highest inbreeding levels of all four subspecies, with 65% of its genome in ROH. This is significantly higher than the inbreeding levels for the other subspecies (W, P < 0.05 for all comparisons; Table 1). The sumatranus subspecies ranks second highest, with 35% of the genome in ROH, significantly lower than for borneensis (W, P < 0.05) but significantly higher than for indicus and maximus (W, P < 0.05 for both comparisons). The latter two subspecies show similar inbreeding levels (11% vs. 10% in ROH) with no significant difference (W, P > 0.18).

Fig. 3.

Dotplots showing comparisons between the four Asian elephant subspecies for a) inbreeding in terms of fraction of the genome in ROH (fROH). b) Heterozygosity in terms of the number of heterozygous sites per 100 kb (nucleotide diversity (π)). c) Heterozygosity in terms of the number of heterozygous sites per 100 kb corrected for ROH coverage. d) Average length of ROH per individual in base pairs.

Table 1.

Mean values per subspecies for four genomic metrics: (i) inbreeding coefficient based on runs of homozygosity (fROH), (ii) genome-wide heterozygosity measured as the number of heterozygous sites per 100 kb, (iii) heterozygosity corrected for ROH coverage, and (iv) average length of ROHs per individual in number of base pairs

Subspecies	Mean fROH	Mean heterozygosity	Mean heterozygosity corrected for ROH coverage	Mean average length of ROHs (max)
borneensis (b)	0.65	46.86	136.83	469,192(852,663)
indicus (i)	0.11	130.27	149.25	297,807 (525,711)
maximus (m)	0.10	134.89	149.82	288,657 (356,110)
sumatranus (s)	0.35	101.09	155.83	383,039 (415,803)

Pairwise comparison	fROH	Heterozygosity	Heterozygosity corrected for ROH coverage	Average length of ROHs
b vs. i	1.6e-10*	1.6e-10*	0.26	1.07e-5*
b vs. m	0.00013*	0.00013*	0.45	3.78e-4*
b vs. s	0.00069*	0.00069*	0.21	0.01*
i vs. m	0.18826	0.07871	0.66	0.9
i vs. s	3.9e-06*	3.9e-06*	0.21	0.008*
m vs. s	0.00303*	0.00303*	0.26	0.008*

Maximum average length of ROHs per individual within brackets. fROH and heterozygosity values were derived from whole-genome SNP data. Pairwise comparisons among subspecies for each metric are shown in the lower section, with corresponding P-values indicating statistical significance from Wilcoxon rank-sum tests. Asterisks indicate statistically significant comparisons (P < 0.05).

Consistent with evolutionary theory, heterozygosity declines in small populations due to inbreeding and reduced population sizes, which lower coalescent timelines. Mean heterozygosity (heterozygous sites per 100 kb window) is lowest for the borneensis subspecies (∼46 heterozygote sites per 100 kb), significantly lower than heterozygosity for any other subspecies (W, P < 0.05 for all comparisons; Fig. 3b; Table 1). The sumatranus subspecies shows intermediate mean heterozygosity (∼101 heterozygous sites per 100 kb), significantly higher than borneensis (W, P < 0.05) but significantly lower than indicus and maximus (W, P < 0.05 for both comparisons). Again, indicus and maximus show similar heterozygosity levels (∼130 heterozygote sites per 100 kb vs. ∼134 heterozygote sites per 100 kb) with no significant difference (W, P > 0.05).

To determine whether reduced heterozygosity in Bornean and Sumatran elephants reflects genome-wide loss of diversity or is confined to inbred regions, we estimated genetic diversity in noninbred genomic regions by correcting for ROH coverage (Fig. 3c). Interestingly, when correcting for ROH coverage, heterozygosity levels are similar across all four subspecies, ranging from ∼136 heterozygous sites per 100 kb in borneensis to ∼155 heterozygote sites per 100 kb in sumatranus, with no significant differences for all pairwise comparisons of the four subspecies (W, P > 0.2 for all comparisons; Table 1).

To evaluate whether the observed differences among subspecies were influenced by unequal sample sizes, we repeated all pairwise comparisons using 10 subsampled datasets with equal representation per subspecies (n = 5 individuals; Tables S4 to S6). The main patterns observed in the full dataset were highly consistent across subsampling replicates. In particular, differences in fROH and genome-wide heterozygosity remained robust, with all comparisons involving borneensis and sumatranus consistently significant and associated with large effect sizes. In contrast, comparisons between indicus and maximus were less consistent, reflecting smaller and more variable effect sizes.

Notably, heterozygosity corrected for ROH coverage did not show significant differences in any of the subsampling replicates, despite consistent effect directions. This supports the observation from the full dataset that baseline levels of genetic diversity are largely shared among subspecies and indicates that the observed differences in genome-wide heterozygosity are primarily driven by variation in the extent of homozygosity rather than differences in the underlying pool of neutral genetic variation.

The borneensis subspecies shows the longest stretches of ROHs (Fig. 3d and Fig. 4a to d; mean ROH length of 469,192 bp; Table 1, column 5), and the sumatranus subspecies ranks second highest (mean length of ROH = 383,039 bp), while the indicus and maximus subspecies show similar shorter ROH lengths (mean length of ROH = 297,807 vs. 288,657 bp). These differences in average ROH length were largely consistent across subsampling replicates, with effect directions conserved for all pairwise comparisons despite some variation in statistical significance, particularly for comparisons involving indicus (Table S6). The long ROH stretches in borneensis and sumatranus suggest recent inbreeding events. ROH segments among the five borneensis individuals show near-complete overlap (Fig. 4e), indicating highly shared ancestry and reduced genomic heterogeneity. This pattern was confirmed quantitatively, with borneensis showing a significantly higher pairwise ROH overlap (mean Jaccard index = 0.809 ± 0.011 SD) compared to indicus (0.115 ± 0.028), maximus (0.121 ± 0.024), and sumatranus (0.358 ± 0.052; Kruskal–Wallis P = 3.33 × 10⁻⁷; all pairwise comparisons with borneensis P < 0.001). Overlap is also present, though less extensive, in sumatranus, whereas indicus and maximus show little concordance in ROH segment positions among individuals, consistent with greater genomic diversity.

Fig. 4.

Genome-wide nucleotide diversity and ROH distribution across Asian elephant subspecies. a to d) Nucleotide diversity (π) window plots for one representative individual per Asian elephant subspecies: borneensis (a), indicus (b), maximus (c), and sumatranus (d). Diversity is shown across 27 chromosomes using a 100 kb sliding window. Orange bars indicate ROH. Values depicted in the middle of the window plots represent population mean fROH values for the respective subspecies. e) Segment plot with the distribution of ROHs on chromosome 1 (SUPER_1) for five individuals per subspecies: sumatranus (blue), maximus (yellow), indicus (green), and borneensis (red). The x axis displays the position on chromosome 1. The y axis displays the individual ID numbers (1 to 5: sumatranus; 6 to 10: maximus; 11 to 15: indicus; 16 to 20: borneensis).

Genetic Load in Conserved Genomic Regions Remains Low and Not Fully Fixed

Although genetic diversity declines in small populations due to inbreeding and drift, nearly neutral theory predicts that moderately detrimental variants (with selection coefficients below the threshold of 4N_es < 1.0) can persist as effectively neutral among segregating variants. To assess genetic load in the four Asian elephant subspecies, we predicted variant effects using SnpEff and calculated genomic evolutionary rate profiling (GERP) scores for each SNP position, merging both metrics into a single dataset (Davydov et al. 2010; Cingolani et al. 2012). Variants with GERP > 2 were classified as highly conserved, while those with <2 were considered low to moderately conserved. Here, we report results for the highly conserved regions (GERP score > 2). Results for genetic load in the low to moderately conserved regions (GERP score < 2) are provided in Table S8.

The total number of fixed sites for the highly conserved regions are highest for the borneensis subspecies and lowest for the indicus subspecies (zero in this case). In all four Asian elephant subspecies, the number of fixed high-effect SNPs (column 2, Table 2) is low, in line with the expectation that strongly deleterious mutations exert effects too large for genetic drift to override the action of selection. The number of SNPs with moderate effect that are fixed in these populations is higher (column 3), consistent with nearly neutral expectations.

Table 2.

Number of fixed SNPs with predicted functional effects (high or moderate) in highly conserved regions (GERP > 2) across the four Asian elephant subspecies

GERP > 2
Subspecies	High effect	Moderate effect	Low effect	Total	Percentage high	Percentage moderate
borneensis	6	1637	27,872	29,515	0.020	5.546
indicus	0	0	0	0	0	0
maximus	0	293	2893	3186	0	9.196
sumatranus	4	905	13,744	14,653	0.027	6.176

Within highly conserved genomic regions, the borneensis subspecies exhibits the highest mean proportion of realized genetic load for both high-impact effect (Fig. 5a, 0.72) and moderate-impact effect variants (Fig. 5b, 0.77), with most deleterious mutations expressed rather than masked. In line with other results in this study, we find an intermediate mean fraction of realized load for the sumatranus subspecies for both the high and moderate impact effects (Fig. 5a, 0.58 and Fig. 5b, 0.64) and the lowest fraction and similar values for the indicus and maximus subspecies, with most impactful variants being masked (Fig. 5a, 0.29 vs. 0.16 and Fig. 5b, 0.42 vs. 0.41). Patterns are consistent when comparing the results for the high-effect and moderate-effect variants. Across all subspecies, the mean proportion of realized load vs. masked load is lower for the high impact effects than for the moderate impact effects. Violin plots with the proportions of realized load for the low- to moderately conserved regions can be found in Fig. S9.

Fig. 5.

Violin plots showing proportion of realized load in the four Asian elephant subspecies for genomic regions with GERP score higher than two for variant high effects (a) and moderate effects (b). Realized load was defined as high or moderate impact effects in homozygous state, whereas masked load was defined as high or moderate impact effects in heterozygous state.

The distribution of high- and moderate-effect variants within highly conserved regions largely mirrors genome-wide patterns (Table 2). To facilitate visualization, Fig. 6 shows a representative subset of the data, restricted to chromosome 1 and five individuals per subspecies. Within this subset, most variants are shared among subspecies, suggesting they originate from ancestral variation predating subspecies divergence. Nevertheless, a number of subspecies-specific variants are also present, indicating that potentially deleterious mutations have arisen or become fixed more recently within individual lineages.

Fig. 6.

Same segment plot as in Fig. 3e, but now the ROH fragments in gray and the high (red dot) and moderate (blue triangle) effects in homozygous state in the highly conserved regions of the genome (GERP score > 2). The x axis displays the position on chromosome 1. The y axis displays the individual number; 1 to 5: sumatranus; 6 to 10: maximus; 11 to 15: indicus; 16 to 20: borneensis.

Discussion

In this study, we present a comprehensive genomic assessment of genetic diversity, inbreeding, and genetic load across the four recognized Asian elephant subspecies, highlighting how demographic history and geographic structure shape genetic health in long-lived megafauna. Population structure analyses revealed genetic differentiation among subspecies, broadly consistent with our previous findings (Kappelhof et al. 2025), which identified three major genetic clusters (borneensis, sumatranus, and indicus) with indicus and maximus showing close genetic affinity.

Consistent with earlier studies (Fernando et al. 2003; Sharma et al. 2018), borneensis exhibits high levels of genomic inbreeding and markedly reduced genetic diversity, while sumatranus shows intermediate levels. In contrast, indicus and maximus maintain similar, relatively low levels of inbreeding and high levels of genetic diversity, reflecting larger and more connected populations. Despite these contrasts, genetic load is broadly shared among subspecies. We observe a low number of high-impact variants in highly conserved genomic regions, although our analyses do not directly assess fixation dynamics or the evolutionary processes underlying their frequency. This pattern may be consistent with strong purifying selection acting on constrained genomic regions and/or a limited accumulation of strongly deleterious variants across subspecies but does not on its own provide evidence for effective purging of deleterious mutations. Together, these results highlight patterns of purifying selection acting on conserved genomic regions and the differential impact of demographic history across subspecies (Hussin et al. 2015; Dussex et al. 2023). While we do not directly quantify rates of accumulation or removal of deleterious mutations, the observed distribution of potentially deleterious variation is broadly consistent with expectations under nearly neutral theory, where the efficiency of selection and drift depends strongly on effective population size (Kimura 1983; Ohta 1992; Kyriazis et al. 2021). Our findings further underscore the vulnerability of island subspecies to inbreeding-driven loss of heterozygosity (Frankham 1997, 2008; Frankham et al. 2010) and highlight the value of whole-genome data for comparative conservation genomics (Allendorf et al. 2010; Kardos et al. 2021).

Baseline of Genetic Diversity in Asian Elephants Has Remained Stable

Genetic diversity is expected to decline in small populations compared to large populations as lineages share a common ancestor in fewer generations (Watterson 1975; Kimura 1983; Tajima 1983). Additionally, in small populations, effective inbreeding even across distant relatives may increase (Wright 1951), and higher chances of consanguineous mating among close relatives can still reduce diversity further (Charlesworth and Charlesworth 1987). These multiple evolutionary forces are important for endangered taxa as they may cause genetic diversity to erode with potentially severe effects (Franklin 1980). With 48 newly sequenced genomes for Asian elephants, we can assess the relative impact of these competing evolutionary forces on natural diversity at the genetic level. We observe signatures of inbreeding with identity by descent in the genomes of Asian elephants from Borneo and Sumatra, a warning sign for small populations. In our studied animals, for genome-wide heterozygosity, we observe the pattern where borneensis has the lowest level of diversity, with 65% of the genome captured in ROHs, and sumatranus displays an intermediate level, with 35% in ROHs. Asian elephants from indicus/maximus harbor the highest level of genome-wide heterozygosity with minimal ROHs. Among the Asian elephant subspecies, borneensis has the highest fROH, as well as the longest mean ROH length across individuals. However, their mean length of ROH is below 2 Mb (∼200 kb), which is consistent with ROH arising from inbreeding over multiple past generations, under standard mammalian recombination rate assumptions (∼1 cM/Mb) (Jensen-Seaman et al. 2004).

Inbreeding causes the loss of heterozygosity in a single generation, where alleles become identical by descent because of shared ancestry (Wright 1951). These effects could be partially reversed, even in a single generation, if mating occurs with nonrelatives. However, genetic decline in small populations can also occur when population sizes are small and genetic variation is lost under normal coalescent processes that result in loss of alleles across the population as a whole (Watterson 1975; Kimura 1983; Tajima 1983). This type of genetic erosion is far more difficult to correct, as new genetic diversity is expected to require long timescales to emerge and spread through natural populations. To assess genetic diversity while accounting for recent inbreeding, we applied a correction to genome-wide heterozygosity based on the proportion of the genome covered by ROH. While overall heterozygosity differs among the four Asian elephant subspecies, heterozygosity corrected for ROH coverage shows similar levels across subspecies, with no significant differences. The greater variability in heterozygosity observed for the borneensis subspecies when correcting for ROH coverage likely stems from its extensive ROH coverage, which restricts the portion of the genome contributing to estimates outside ROH and may thus exaggerate apparent differences among individuals. However, consistency analyses based on repeated random subsampling of five individuals per subspecies showed that pairwise comparisons remained in the same direction across all runs, suggesting that the observed patterns are robust despite differences in sample size and within-subspecies variability. The similar levels of heterozygosity corrected for ROH coverage suggest that baseline neutral genetic diversity is largely shared among subspecies, consistent with their relatively shallow divergence times and long generation time (Kappelhof et al. 2025). This indicates that differences between the four subspecies are primarily driven by variation in genome-wide heterozygosity and the extent of ROH, particularly the elevated ROH burden observed in borneensis and sumatranus, a pattern also reported in other megafauna species (Lorenzen et al. 2011; Guðjónsdóttir et al. 2026). This is also consistent with predictions from nearly neutral theory, where in recently bottlenecked populations, patterns of genetic variation largely reflect shared ancestral demography rather than subspecies-specific processes (Ohta 1992; Kyriazis et al. 2021). While ROH-corrected heterozygosity remains substantial in some subspecies, this reflects retained background genetic variation rather than the absence of inbreeding effects.

Populations such as borneensis, which combine extensive long ROH tracts indicative of recent inbreeding with remaining background diversity, may still retain a substantial reservoir of segregating deleterious variation. Under continued or future demographic decline, such variation could become more strongly affected by drift and more likely to be exposed to selection, consistent with theoretical expectations on the dynamics of genetic load (Kyriazis 2025). This pattern is consistent with expectations for populations that have undergone recent demographic contraction following historically larger effective population sizes, in contrast to long-term small populations such as the vaquita, which exhibit persistently low genome-wide diversity (Robinson et al. 2023).

These findings suggest that monitoring of genetic diversity and inbreeding patterns may be important for conservation assessments, alongside continued attention to habitat protection and population management strategies.

Relatively High Genetic Diversity Within Island-Inhabiting Maximus Subspecies

Species inhabiting an island are expected to show limited genetic diversity due to high genetic differentiation and the absence of gene flow with other populations (Frankham 1997; Groombridge 2007; Frankham 2008). Emerging genetic studies of island populations have revealed unique changes at the genetic level consistent in Isle Royale wolves (Hedrick et al. 2019), Channel Island foxes (Robinson et al. 2016), and Wrangel Island mammoths (Rogers and Slatkin 2017; Dehasque et al. 2024). These new observations are consistent with theoretical expectations that smaller population sizes in island populations would result in higher effective inbreeding, lower genetic diversity, and a higher load of detrimental mutations (Wright 1951, Ohta 1973, Kimura 1983).

Among Asian elephants, there are three cases of island populations on Sri Lanka (maximus), Sumatra (sumatranus), and Borneo (borneensis) where we might expect to observe such patterns of relatively low genetic diversity (Fernando et al. 2003; Vidya et al. 2009). However, we found low inbreeding (Fig. 3a; Table 1) and high genetic diversity (Fig. 3b;  Table 1) for the maximus subspecies. Moreover, genetic diversity for the maximus subspecies was even higher than for the indicus subspecies (Fig. 3b;  Table 1). This is in line with previous studies based on mtDNA haplotypes and microsatellites, where the maximus subspecies was also found to have the highest diversity of haplotypes and heterozygosity, exhibiting incomplete geographical partitioning with several nonadjacent mainland countries (Fleischer et al. 2001; Fernando et al. 2003; Vidya et al. 2009). In addition to this, we observed a weak genetic separation of the maximus subspecies from the indicus subspecies in our PCA and admixture plots, an observation we also reported in our previous study (Kappelhof et al. 2025). This, in combination with the surprisingly low levels of inbreeding and high genetic diversity for an island-inhabiting subspecies, raises the question whether the recognition of the Asian elephant population on Sri Lanka as a subspecies is valid from a genetic perspective.

There has been a debate on the origin of the maximus subspecies in the conservation literature for the last three decades (Fernando et al. 2000; Fleischer et al. 2001). Studies of the mtDNA of Asian elephants have shown that the maximus subspecies shares many haplotypes with the indicus mainland subspecies (Fernando et al. 2000). The haplotype composition of the maximus subspecies shows the highest similarity with distant elephant populations in Thailand and Myanmar, as well as with neighboring southern Indian populations (Hartl et al. 1996; Fernando et al. 2000). We here confirm this pattern (Fig. 2a and b), consistent with previous nuclear genomic analyses showing that the maximus subspecies clusters most closely with individuals from Myanmar (Kappelhof et al. 2025). At the same time, our population structure analyses reveal varying degrees of admixture among Sri Lankan individuals, which is not fully consistent with simple geographic expectations of strongest affinity with Indian populations based on proximity. In the admixture results, several Sri Lankan individuals show ancestry components shared with both Indian and Southeast Asian populations, including Malaysia and Myanmar, suggesting a more complex demographic history than simple geographic isolation would predict.

Intensive human-induced elephant trade has taken place between the Sri Lankan and mainland populations for various purposes (Fleischer et al. 2001; Fernando et al. 2011). Sri Lanka served as a major trade hub for over 1,000 years, importing elephants for warfare, breeding, and training (Sukumar 1989). The Pegu region of southern Myanmar was a major source of elephants for Sri Lanka and Southern India (Digby 1971). Although many of these traded elephants were held in captivity, frequent escapes allowed for gene flow between the populations from geographically distinct countries (Fleischer et al. 2001; Vidya et al. 2009). This could explain the current clustering of Sri Lanka and Myanmar individuals. However, due to limited data, it remains hard to conclude whether the putative Sri Lankan genomic signature is still present today. Importantly, this study primarily relies on genomic data from Asian elephant individuals that are currently living in North American and European zoos but are wild-born. Since these individuals could have at some point been traded, it raises questions about how likely it is that they represent the original Sri Lankan elephant population as well as the current elephant population on Sri Lanka. Therefore, more genomic research with additional data from Sri Lankan elephants is warranted to draw definite conclusions on the validity of the subspecies designation for the Sri Lankan elephant population.

Moderately Detrimental Variation Accumulated Within Asian Elephant Subspecies

Small populations are at risk for detrimental variation that accumulates under nearly neutral dynamics but also from unmasking of recessive mutations that might cause severely detrimental phenotypes, especially in inbred genomic regions (Franklin 1980). Severely detrimental variants may be purged in small populations, as lethal phenotypes are unmasked (Garcia-Dorado 2012), although timelines to fully extinguish recessive mutations may be long. Even though we observed widespread inbreeding in populations from Borneo and Sumatra, evidence for inbreeding depression in the borneensis and sumatranus subspecies has not yet been reported in the scientific literature. Such a disparity between genotypic signals and phenotypic observations could indicate that the majority of the strongly deleterious alleles have already been purged and the population is beyond the critical first phase of extinction risk—however, this remains speculative, especially because it could be understudied. To determine the extent of potential purging and detrimental variation, we assessed genetic load in Asian elephant populations.

From our analyses of genetic load, we observe few (or zero) fixed high-effect SNPs within subspecies across the highly conserved regions (Table 2, column 2). High-effect SNPs in conserved regions can have significant consequences, as genes in these regions are more likely to regulate important processes and functions (Hussin et al. 2015; Dussex et al. 2023). The low number of fixed high-effect SNPs in these regions is consistent with either limited accumulation of highly deleterious variants or their removal through purifying selection in highly conserved regions of the genome across Asian elephant subspecies. Given the high inbreeding levels that are observed in the borneensis subspecies, purging during the initial phase of the bottleneck could have removed the alleles that contributed strongest to inbreeding depression, a pattern observed in mammals with small population sizes, such as the saola, as well (Garcia-Erill et al. 2025). In line with these results, the borneensis subspecies exhibits a greater fraction of realized load compared to masked load than the other Asian elephant subspecies (Fig. 5).

While we observe only a few cases of fixed highly deleterious variants in the population, the number of within-subspecies fixed moderate-effect variants is substantially higher than for high-effect variants, particularly in the borneensis subspecies (Table 2, column 3). Unlike high-effect variants, which are efficiently removed by selection, moderate-effect variants often persist and can accumulate over time if the selective disadvantage is beneath the nearly neutral threshold (Ohta 1992). This accumulation of moderately detrimental variation is potentially insidious: while individually their effects are too subtle to be removed efficiently via selection, collectively they may reduce fitness and compromise population viability. Whether the many fixed moderate-effect alleles in Bornean elephants escaped purging or phenotypic effects have not been recorded yet but are apparent remains to be investigated. These findings highlight the importance of considering moderate-effect variants in genetic load assessments, rather than focusing solely on high-effect mutations (Robinson et al. 2023, Chen et al. 2025). Greater use of genetic data and population genetic-informed management policies may help preserve unique island species alongside habitat preservation and protective measures that maintain living populations.

Conservation Implications of Genetic Diversity

From a management perspective, the borneensis subspecies emerges as the most vulnerable subspecies, indicating that genetic rescue could be explored as a potential conservation intervention. During genetic rescue, genetic diversity from one population is introduced to another, thereby increasing overall genetic diversity and minimizing inbreeding in the receiving population (Whiteley et al. 2015; Quinn et al. 2024; Sun et al. 2024; Wilder et al. 2025). A risk of genetic rescue is outbreeding depression, caused by introducing genetic diversity from a population that is too distant, which can lead to genomic incompatibilities or maladaptation to the local environment (Whiteley et al. 2015). Taking the results from this study and previous studies into account, the borneensis subspecies is a candidate for genetic rescue due to its high levels of inbreeding (Fernando et al. 2003; Goossens et al. 2016; Sharma et al. 2018). In contrast to previous work, which was largely based on microsatellite or mtDNA data and did not allow direct genome-wide comparisons across Asian elephant subspecies, our whole-genome dataset provides a unified framework to directly compare levels of inbreeding, genetic diversity, and genetic load across all four subspecies. Since we observe similar levels of heterozygosity corrected for ROH coverage, this may suggest comparable levels of standing genetic variation across subspecies. While this pattern could be consistent with a reduced risk of outbreeding depression, this inference remains tentative; similar heterozygosity does not necessarily reflect shared polymorphisms or the absence of phylogenetic divergence, and alternative explanations, including independent evolutionary histories, cannot be excluded. This observation corresponds with the relatively recent divergence times found for Asian elephants, potentially limiting the accumulation of unique and incompatible alleles (Kappelhof et al. 2025). Moreover, the absence of fixed high-effect SNPs in highly conserved regions across Asian elephant subspecies reduces the risk of introducing highly deleterious mutations by genetic rescue, an issue that has been observed previously in Isle Royale wolves (Robinson et al. 2019). Thus, genetic rescue using one of the other three Asian elephant subspecies may represent a potential option from a genomic perspective for the borneensis subspecies, with the sumatranus subspecies being the best candidate due to the most recent divergence time from the borneensis subspecies (Kappelhof et al. 2025). This parallels the successful introduction of Texas cougars into the Florida panther population, which alleviated inbreeding depression with minimal maladaptation (Johnson et al. 2010).

However, genetic matching is not the sole consideration for potential avenues of intervention. As the sumatranus subspecies is critically endangered (Williams et al. 2020), its use to genetically rescue the borneensis could harm the survival of the sumatranus subspecies. Moreover, introducing Sumatran elephants in wild populations of the borneensis subspecies will come with numerous practical difficulties, particularly with respect to logistics, animal welfare, and post-release survival. However, for captive populations of Asian elephants, genetic rescue of the borneensis subspecies with the sumatranus subspecies or other Asian elephant subspecies could be considered. Genetic rescue from within the borneensis subspecies does not seem realistic, considering the ROH fragments on chromosome 1 (Fig. 4e), where we observed that these ROH fragments almost completely overlap among the five borneensis individuals. However, our data also emphasize limits to genetic rescue between Asian elephant subspecies, since much of the genetic load on chromosome 1 is shared (Fig. 6) and this may constrain the potential effectiveness of genetic rescue. This pattern is consistent with the possibility that part of the genetic load in Asian elephants reflects past, shared demographic events, although a more quantitative assessment would be required to disentangle the relative contributions of shared vs. lineage-specific processes. Thus, rescue strategies may alleviate inbreeding but will not erase the burden of historical load, while the effects of this historical load may be mitigated if purging has been effective within Asian elephant subspecies. Demonstrating purging would require explicit analyses of the frequency and distribution of deleterious alleles across different ROH classes or fitness-related data, which are beyond the scope of this study. Efforts to preserve genetic diversity outside inbreeding tracts may be key to maintaining genetic variation in the borneensis subspecies, for which genetic rescue could be one of the tools. More broadly, the Asian elephant case illustrates how integrating whole-genome data can refine conservation priorities by distinguishing between risks that stem from recent demography and those embedded in deeper evolutionary history.

Massive Genetic Diversity Loss in Asian Elephants?

In conclusion, “massive genetic diversity loss” among Asian elephant subspecies applies primarily to the island subspecies borneensis and sumatranus, both of which exhibit higher levels of inbreeding and markedly reduced genetic variation, consistent with historical bottlenecks and prolonged isolation. In contrast, the indicus and maximus subspecies retain high genetic diversity, suggesting that diversity loss has not been as severe in these populations. For maximus, historical human-mediated trade and translocation may have contributed to maintaining genetic variation by mixing previously isolated lineages.

By comparing genetic diversity and load across subspecies, we demonstrate that demographic history, geographic isolation, and potentially human-mediated movement jointly shape the balance of diversity, inbreeding, and load. These patterns mirror those observed in other endangered vertebrates and demonstrate the value of placing conservation genomics within a broader evolutionary framework. Genomic data not only sharpen extinction risk assessments but also identify when interventions such as genetic rescue are most likely to succeed. For Asian elephants, urgent action is warranted for the borneensis, while further research is needed to clarify the taxonomic status of maximus. More broadly, our findings underscore that endangered species retain both vulnerabilities and evolutionary resilience—and that genomics offers a powerful lens to disentangle the two.

Materials and Methods

Sample and Data Collection

At Wageningen University and Research, blood samples were obtained from 46 individual Asian elephants (E. maximus) housed in zoological institutions affiliated with the European Association of Zoos and Aquaria (EAZA). These samples were provided by the EAZA Biobank (Table S1). All samples were collected during routine health assessments or regular training procedures to ensure minimal invasiveness. Individuals were selected based on the availability and reliability of data regarding their wild origin. Provenance data were obtained through the EAZA Ex situ Program for Asian elephants and are maintained within the Species360 Zoological Information Management System (ZIMS; zims.species360.org).

Sampling aimed to represent each of the four recognized Asian elephant subspecies while maximizing geographical coverage across the species' native range. Genomic DNA was extracted using the Qiagen Puregene Blood Kit, following the manufacturer's protocol “DNA purification from whole blood.” Extracted DNA was sent to BGI Tech Solutions and Novogene for sequencing. WGS was performed on the Illumina platform, yielding 150 bp paired-end reads with an average sequencing depth of approximately 25×.

At the Smithsonian's National Zoo and Conservation Biology Institute, blood samples were obtained from 17 wild-caught Asian elephants of known provenance housed in North American zoos. Provenance data were obtained from the Elephant Studbook. DNA was extracted from whole blood using the BioSprint 96 DNA Blood Kit (Qiagen, Germantown, MD, USA) on a BioSprint 96 robot. DNA extracts were built into double-indexed Illumina libraries and sequenced on Illumina Novaseq 6000 platforms at Admera Health (South Plainfield, NJ, USA) and the Duke University GCB Genome Sequencing and Genomic Technologies (Durham, NC, USA).

At the University of Potsdam, DNA from eight wild Bornean elephants (Sabah Wildlife Department export permit: JKM/MBS.1000-2/3 JLD.4(89); CITES export permit number [from Sabah, Malaysia]: 1097; CITES import permit number [into Germany]: DE-E-02035/22) was extracted and purified with the Qiagen DNeasy Blood & Tissue Kit. The DNA was then sheared using a Covaris S220 sonicator aiming for a fragment size of 500 bp. These fragments were then built into double-stranded libraries. Libraries were sent for PE150 sequencing on the Illumina NovaSeq6000 system to Novogene (Cambridge, UK).

In addition to newly sequenced individuals, raw short-read data from ten Asian elephants were retrieved from publicly available repositories (Tollis et al. 2021; Prado et al. 2023). The final dataset comprised WGS data for 81 Asian elephant individuals. This included 54 individuals of the indicus subspecies (India, n = 18; Myanmar, n = 9; Thailand, n = 11; Vietnam, n = 9; Peninsular Malaysia, n = 7), 10 individuals of the maximus subspecies, 6 of the sumatranus subspecies, and 11 of the borneensis subspecies. A complete overview of the individuals and associated genomic data is provided in Table S1.

For phylogenetic analyses, WGS data from one African savanna elephant (L. africana; sample ERR2260497) were additionally retrieved from publicly available repositories (Palkopoulou et al. 2018) and used exclusively as an outgroup taxon. Details of this genome are provided in Table S1.

Mapping of Whole-Genome Sequence Data

Short-read whole-genome sequences from all 81 Asian elephant individuals were mapped to the E. maximus reference genome (mEleMax1), which was generated by the G10K-VGP project (NCBI assembly accession: GCF_024166365.1). The reference genome was indexed prior to alignment. Reads were aligned to the reference using BWA-MEM2 (v2.2.1) with default parameters (Vasimuddin et al. 2019). PCR duplicates were marked and removed using samblaster (v0.1.26) (Faust and Hall 2014). Read files originating from the same individual were merged using samtools merge, and all resulting BAM files were sorted and indexed using samtools (Danecek et al. 2021). Mapping quality and coverage statistics were assessed with Qualimap (v2.2.2-dev) (Okonechnikov et al. 2016). Final genome coverage across individuals ranged from approximately 6× to 42×.

Variant Calling of SNPs

To identify SNPs, population-scale variant calling was performed on the 81 sequenced individuals using FreeBayes (v1.3.1) (Garrison and Marth 2012), with the following parameters: –use-best-n-alleles 4 –min-base-quality 10 –min-alternate-fraction 0.2 –haplotype-length 0 –ploidy 2 –min-alternate-count 2. Only high-confidence variants were retained, filtering for a minimum QUAL score of 20, corresponding to a 99% probability of a true variant at a given site. Filtering was conducted using vcftools (v0.1.16), and the resulting VCF file was indexed with tabix, yielding an unfiltered set of 26,377,688 variant sites (Danecek et al. 2011). The quality of the variant calls was further evaluated using bcftools (v1.9) (Danecek et al. 2021).

Subsequent variant filtering was performed using VCFtools (v0.1.16). The raw VCF file was filtered based on genotype quality and sequencing depth using the following parameters: –minGQ 30 –minDP 6 –maxDP 100 –max-meanDP 36. Variants with low depth of coverage were excluded due to their higher likelihood of representing false positives, while variants with excessively high depth were removed to avoid potential mapping artifacts or SNPs located in repetitive regions of the genome.

In addition, variants located on the sex chromosomes (SUPER_X & SUPER_Y) and all indels were removed using the –not-chr and –remove-indels options (Danecek et al. 2015). To prepare the dataset for population structure analysis using ADMIXTURE, SNPs with more than 20% missing genotypes across individuals were excluded (−max-missing 0.8).

Finally, to reduce redundancy due to linkage disequilibrium (LD), the dataset was pruned using PLINK (v1.90b3.38) with the –indep-pairwise 50 10 0.1 setting (Purcell et al. 2007). After all filtering steps, the final dataset contained 4,947,714 high-quality SNPs suitable for downstream analyses.

PCA and admixture

Population genetic structure was investigated using PCA and admixture analysis. PCA was conducted with PLINK (v1.90b3.38) using the parameters –pca –chr-set 28 –allow-extra-chr (Purcell et al. 2007) to obtain a broad overview of the genetic variation and clustering among individuals.

Admixture analysis was performed using ADMIXTURE (v1.3.0) with default settings (Alexander et al. 2009) to infer individual ancestries and identify genetic structure within the dataset. The analysis was carried out without assuming any prior population structure. The number of clusters (K) was explored in a range from two to five to evaluate potential substructure within the sample. For each K value, CV was used to assess model fit and identify the most likely number of clusters, based on minimization of the CV error.

Seven individuals (EleMaxF525, EleMaxM160, EleMaxM263, EleMaxF290, EleMaxF84, EKDN240012687, and EKDN230016702) showed discordance between their documented geographic origin and their clustering in principal component and admixture analyses. Because most samples originated from European and North American zoological collections, historical documentation errors cannot be excluded, particularly for individuals transferred from the wild several decades ago.

For population genomic analyses, sixteen individuals (EleMaxF26, EleMaxF234, EleMaxF525, EleMaxF127, EleMaxM160, EleMaxM270, EleMaxM263, EleMaxF290, EleMaxF288, EleMaxF273, EleMaxF84, EleMaxM126, EleMaxF77, EleMaxF28, EleMaxF245, and EleMaxM139) with a mean sequencing coverage below 10× were excluded using VCFtools (v0.1.16; –remove-indv; Danecek et al. 2011). After filtering, a total of 65 Asian elephant individuals were retained for population genomic analyses. A detailed overview of sample inclusion per analysis is provided in Table S1.

For the 65 Asian elephant individuals with sufficient coverage (>10×), subspecies assignment was based on genetic clustering rather than original documentation, as the present study focuses on subspecies-level patterns of genetic diversity, inbreeding, and genetic load rather than fine-scale geographic structure within the Asian mainland. Consequently, two individuals originally assigned to Sri Lanka (EKDN240012687) and Thailand (EKDN230016702), respectively, were classified as belonging to the indicus subspecies based on their genomic affinity. This reassignment resulted in one individual (EKDN240012687) originally labeled as Sri Lankan being treated as non-maximus in downstream analyses.

Phylogenetic Analysis

To provide an independent assessment of the subspecies-level grouping used throughout downstream analyses, we performed a population-level phylogenetic reconstruction using TreeMix v1.13 (Pickrell and Pritchard 2012). Because the primary variant-calling dataset was generated exclusively for Asian elephant population genomic analyses, a separate population variant-calling workflow was performed for phylogenetic reconstruction that additionally included one African savanna elephant (L. africana; sample ERR2260497) as an outgroup taxon.

Analyses were conducted using a representative subset of the genomic dataset consisting of four individuals per Asian elephant subpopulation (LKA, IND, MMR, THA, VNM, MAL, SUM, and BR), together with the African savanna elephant outgroup. Individuals were selected to provide balanced representation across populations while minimizing differences in sample size among groups.

Short-read sequences were mapped to the E. maximus reference genome (mEleMax1) following the same alignment and preprocessing pipeline described above. SNP discovery and filtering for the phylogenetic dataset were conducted using FreeBayes v1.3.1, VCFtools v0.1.16, and PLINK v1.90 following the same general filtering framework applied in the primary population genomic analyses. Variants were filtered to retain high-quality biallelic SNPs only. Indels were removed, and sites were filtered using a minimum site quality threshold of 30, minimum sequencing depth of 5, maximum sequencing depth of 50, maximum missingness of 5%, and a minimum minor allele count of 2.

To minimize the effects of linkage disequilibrium, SNPs were LD-pruned in PLINK v1.90 using the command –indep-pairwise 50 10 0.2, corresponding to a 50 kb sliding window, a step size of 10 SNPs, and an (r²) threshold of 0.2. Population allele frequencies were subsequently calculated for each subspecies grouping and converted into TreeMix input format using plink2treemix.py.

TreeMix analyses were performed with zero migration edges (m = 0) using blocks of 500 SNPs (-k 500) and the African savannah elephant sample as the outgroup. The resulting maximum-likelihood tree was used solely to evaluate whether sampled individuals clustered according to the four currently recognized Asian elephant subspecies.

Nucleotide Diversity and Inbreeding Analysis

To assess overall genetic diversity and individual inbreeding levels, heterozygosity analyses were performed using the original VCF file filtered solely for variant quality with VCFtools (v0.1.16, –minGQ 30) and removing the sex chromosomes and scaffolds (−not-chr) (Danecek et al. 2011). Nucleotide diversity (π) was calculated for each individual using a sliding window approach with a window size of 100 kb (−window-pi 100000).

ROH, defined as continuous homozygous regions within the genome indicative of consanguinity, were identified using bcftools roh (v1.9; Narasimhan et al. 2016), which implements a hidden Markov model (HMM) framework. ROH were called from hard genotype calls using the -G 30 option, which assigns a phred-scaled probability of 30 to genotypes other than the observed call. Alternate allele frequencies were obtained directly from the INFO/AF field of the input VCF.

In contrast to sliding-window approaches (eg PLINK), the HMM framework used by bcftools assigns each marker probabilistically to either an autozygous or non-autozygous state. ROH are therefore defined as contiguous stretches of markers inferred to be in the autozygous state, and no user-defined thresholds for minimum ROH length, marker density, or maximum inter-marker gaps are applied.

From the set of inferred ROH segments, we retained only those with an average phred-scaled quality score greater than 80 for downstream analyses. Individual inbreeding coefficients (fROH) were calculated as the proportion of the genome contained within ROH segments. Genome-wide nucleotide diversity (π) was estimated for each individual as the mean pairwise sequence divergence across nonoverlapping 100 kb windows. Although π is typically defined at the population level, in this study, it was used as an individual-level proxy for observed heterozygosity by calculating within-individual heterozygosity across genomic windows.

To account for the bias introduced by long ROH, we calculated a corrected measure of heterozygosity by scaling genome-wide π by the proportion of the genome not covered by ROHs. Specifically, heterozygosity corrected for ROH coverage was computed as π/(1 − fROH), where fROH represents the fraction of the genome contained in ROHs. This approach does not physically exclude ROH segments from the data but provides a normalization for differences in autozygosity among individuals, allowing comparison of heterozygosity in a non-ROH genomic background across individuals.

Differences in inbreeding levels (fROH), overall nucleotide diversity (with and without correction for ROH coverage), and average length of ROH per individual between the four Asian elephant subspecies were evaluated for statistical significance using a Kruskal–Wallis test, followed by pairwise Wilcoxon rank-sum tests for post hoc comparisons in R (v3.6.2). To assess the robustness of pairwise comparisons to unequal sample sizes, subsampling analyses were performed by randomly selecting equal numbers of individuals per subspecies (n = 5) across 10 replicates. For each replicate, pairwise Wilcoxon rank-sum tests were conducted, and consistency was evaluated based on the proportion of significant results and the direction and magnitude of effect sizes across replicates.

Nucleotide diversity across the genome was visualized using the R script ROHcircle.R (https://git.wur.nl/NinoMenger/vcf-population-genomics-pipeline). Dot plots were generated in R using the ggplot2 package. To visualize the distribution and overlap of ROH segments among individuals, segment plots were generated in R using the segments function. For this purpose, chromosome 1 (SUPER_1) was selected, and five representative high-coverage individuals per subspecies were included (for indicus: IND (n = 1), MMR n = 1), THA (n = 1), VNM (n = 1), MAL (n = 1)), chosen to represent each subpopulation. These plots were used for qualitative comparison of ROH segment sharing patterns among individuals within and between subspecies. In addition, pairwise ROH overlap among individuals within each subspecies was quantified using a Jaccard index based on the proportion of overlapping ROH base pairs relative to the total union of ROH regions, calculated using GenomicRanges in R.

Variant Annotation

To predict the functional effects of genetic variants, we used SnpEff v5.1 (Cingolani et al. 2012). SnpEff offers databases for a selection of supported reference genomes. As E. maximus is unsupported, we constructed a custom SnpEff database using the species-specific reference genome (mEleMax1) and gene annotation files. The genome FASTA and GFF3 annotation files were obtained from Ensembl and configured according to the SnpEff database build instructions.

The custom database was built using the SnpEff build -gff3 command, and the resulting index was used to annotate the original VCF file, which was filtered for gene quality, and sex chromosomes and scaffolds were removed with VCFtools (v0.1.16, –minGQ 30 –not-chr). SnpEff produces an annotated version of the input VCF, with features comparable to the Ensembl variant effect predictor (McLaren et al. 2016). Annotations include variant consequences (eg synonymous, missense, nonsense, intronic, splice sites, UTRs) as well as categories based on a variant's predicted functional impact (high, moderate, low, modifier), which was then used in downstream analyses to assess the genomic distribution and potential biological relevance of the variants.

Computation of Conservation Score Evolutionary Constraint (GERP Scores)

To investigate conservation across the Asian elephant genome, we used GERP analysis (Davydov et al. 2010). GERP is a package for analyzing evolutionary rates and finding constrained elements in a multispecies alignment. GERP analysis consists of two parts: gerpcol, which analyzes a multispecies alignment and quantifies constraint at individual positions, and gerpelem, which finds constrained multiposition elements based on the results from gerpcol. In this study, we only used gerpcol, and the per-position scores. Briefly, gerpcol takes an evolutionary (phylogenetic) tree and a multispecies alignment and computes “rejected substitution” scores, defined as “the difference between the expected (neutral rate) and the observed (maximum likelihood estimate)” (Davydov et al. 2010). In other words, it quantifies the reduction in the number of substitutions compared to the neutral expectation (Huber et al. 2020). GERP uses the HKY85 model of evolution (unequal nucleotide frequencies and unequal transition/transversion rates) (Hasegawa et al. 1985), and nucleotide frequencies are estimated from the multiple alignment. If a site has a GERP score of 4, it thus means that there are four (4) fewer substitutions at that position than would be expected from the neutral rate of evolution.

For the GERP analysis, we followed the protocol from von Seth et al. (2021) with the adaptation by Taylor et al. (2024). First, we downloaded the 43 mammal species that are included in the 43-way Enredo-Pecan-Ortheus multiple alignments (Table S10) corresponding to Release 112 of Ensembl (http://www.ensembl.org). Then, we generated a TimeTree phylogeny (http://www.timetree.org) with the same mammal species as in the alignment, adding the local Asian elephant (E. maximus) reference genome. For each of the outgroup species (all but E. maximus), we converted the reference genomes to fastq format with a fake read length of 150 using BBmap (Bushnell et al. 2017) and aligned them to the Asian elephant reference genome using bwa-mem (Li, 2013). The resulting alignment files were converted to BAM format and were filtered for reads aligning to more than one genomic location and for supplementary reads. The BAM files were then sorted and converted to FASTA format using htsbox (https://github.com/lh3/htsbox). Lastly, we split the files per chromosome scaffold (27 autosomes, X and Y), excluding unplaced scaffolds, and concatenated these together, resulting in one FASTA alignment file per chromosome, including all 43 mammal species.

The resulting FASTA alignments were used to calculate the conservation scores with gerpcol, together with the time tree. A tree scaling factor (–s) of 0.0022, representing the average mammalian mutation rate of 2.2 × 10−9⁶ (Kumar and Subramanian 2002), was used, as TimeTree outputs branch lengths in million years instead of substitutions per site, which gerpcol requires. Additionally, gerpcol does not output genomic coordinates together with the scores. To accommodate this, we used the script gerp_to_position.py adapted from the GenErode (Kutschera et al. 2022) workflow to add genomic coordinates based on the E. maximus reference genome to the respective GERP output.

Based on the distribution of scores in our dataset, we defined higher evolutionary constraint as GERP > 2 and lower as GERP < 2. We assume that mutations happening in more conserved regions are more likely to be deleterious. However, it should be noted that GERP assumes constant selection pressures across all phylogenetic branches (Pollard et al. 2010), meaning that lineage-specific selection or functional turnover could be missed. In general, more deleterious mutations tend to have higher scores; however, caution should be taken as scores are not particularly predictive of strength of selection (Huber et al. 2020).

To calculate and quantify genetic load, we used the annotations from SnpEff. In this study, we define “realised load” as the number of sites with a homozygous carrier status of alleles with high or moderate predicted impact and “masked load” as the number of sites with a heterozygous carrier status. Additionally, we used the GERP scores to further investigate the distribution of variants with high and moderate effect, ie the pattern of genetic load across the genome. As sites in highly conserved genomic regions are expected to be less tolerant to mutations (Sullivan et al. 2023), we expect that a variant with high effect that is also located in a highly conserved region is more likely to be deleterious and thus contributes more to genetic load.

Visualizations of the data in the case of the violin plots were generated in R using the ggplot2 and patchwork packages.

Supplementary Material

Supplementary material is available at Genome Biology and Evolution online.

Funding

This study was funded by the NWO (Netherlands Organisation for Scientific Research), National Institute of General Medical Science R35 GM133376 to RL Rogers, University of North Carolina at Charlotte Internal funding to RL Rogers, the Smithsonian Institution, the Smithsonian Women’s Committee (2015, grant #10; 2020, grant #22), the Friends of the National Zoo Conservation (2016, grant #30), and Dr. Janice Sanders. We also acknowledge the financial support provided by Rotterdam Zoo, which made an additional contribution to the project.

Data Availability

Raw sequence data from Asian elephants sampled at North American zoos are available in the NCBI Sequence Read Archive under BioProject PRJNA1306073. Raw sequence data from Asian elephants sampled at European zoos and wild Bornean elephants are available in the European Nucleotide Archive (ENA) under Projects PRJEB98188 and PRJEB120944, respectively.