2.1 Samples

Dried tissue remains from the skull or tissue from the nasal cavity of 29 archival sun bear specimens were collected from several natural history museums (Table S1), providing a broad geographic distribution of samples.

We also collected saliva samples from 21 sun bears, which originated from 12 provinces in Cambodia and which were housed at the ‘Free the Bears’ sanctuary in Phnom Tamao Zoo and Wildlife Rescue Centre. We included up to two samples per province in order to obtain a balanced geographic coverage of Cambodia.

2.2 Archival and non-invasively collected DNA

DNA from archival samples was extracted following Wilting et al. (2016), in a laboratory designed for and limited to the use of archival material. DNA extractions were conducted in batches consisting of three samples and one negative control and verified to be of sun bear origin by amplification of a 146 bp long portion of the mitochondrial cytochrome b gene using species specific primers (Table S2).

DNA extractions from non-invasively collected samples (salivary mucosal cells) were carried out using the GEN-IAL First DNA All-tissue DNA extraction Kit (GEN-IAL GmbH) following manufacturer's instructions. This work was carried out in a separate facility, dedicated to molecular work with non-invasively collected material.

2.3 Library preparation and hybridization capture

Illumina sequencing libraries were constructed following Fortes and Paijmans (2015), using double-indexing with 8-bp indexes. Prior to indexing we conducted qPCRs to determine the optimal PCR cycle number for each sample.

To reduce effects by target DNA degradation and external DNA contaminations, we carried out hybridisation capture for both sample types (archival, non-invasively collected).

Baits for hybridization capture were generated using long range (LR) PCR products that spanned three large overlapping regions of the mitogenome (using a fresh Helarctos malayanus sample as template). The LR-PCR primers were designed using the genbank reference sequence NC_009968: HMA1-F (5′-ACGACCTCGATGTTGGATCAGG-3′) and HMA1-R (5′-AGGGCTACAGCGAACTCGAGA-3′), yielding a 6152 bp long product, HMA2-F (5′-GCCACACTCATTCACACCTACCA-3′) and HMA2-R (5′-AGTCCTTTCTGGTTGGAGACTGTG-3′), yielding a 5299 bp long product, and HMA3-F (5′-ACCAACGCCTGAGCCCTACT-3′) and HMA3-R (5′-GCGCTTTAGTGAGGGAGGCC-3′), yielding a 6416 bp long product. Amplifications were carried out in 50 μL reaction volumes: 18 μL dH₂O, 25 μL Bioline Myfi mix (Bioline GmbH), 2 μL of forward and reverse primers (10 μM), 3 μL of template DNA (100 ng/μL). PCR conditions were as follows: initial denaturation at 94°C for 3 min, followed by 40 cycles of denaturation at 94°C for 30 s; annealing at 60°C for 30 s; extension at 68°C for 7 min with the final extension at 68°C for 10 min.

LR-PCR products were sheared (target size 300 bp), pooled equimolarly and then converted into biotinylated baits following Maricic et al. (2010). Targeted capture was carried out using hybridization temperatures appropriate for sample type (Paijmans et al., 2016). Paired-end sequencing was carried out on the Illumina MiSeq platform (Illumina) using v3 150-cycle kits.

2.4 Bioinformatic workflow

We de-multiplexed paired-end reads using bcl2fastq v2.17.1.14 (Illumina, Inc.) and removed adapter sequences using Cutadapt v1.3 (Martin, 2011). Using trimmomatic (Bolger et al., 2014), we applied a sliding window approach for quality trimming with the phred quality threshold set at Q = 20. Next, we merged the adapter-clipped and quality-trimmed sequences using flash v1.2.8 (Magoc & Salzberg, 2011). Merged sequences were then mapped to the reference sun bear mitogenome sequence (GenBank accession no. NC_009968) with the BWA aln algorithm v0.7.10 (Li & Durbin, 2009). Mapped sequences were then de-duplicated using Markduplicates from Picard-Tools v1.106 (https://github.com/broadinstitute/picard), followed by variant calling using Samtools v1.1 (Li et al., 2009) and Bcftools v1.2 (http://github.com/samtools/bcftools). Positions along the mitogenome were N-masked if their sequence depth was below 5×, or when they did not conform to an 80% majority rule for base calling.

2.5 Summary statistics

The final data set of complete or near-complete sun bear mitogenomes consisted of 15 archival and 17 non-invasively collected samples, analyzed together with three published sequences (GenBank: MN807949, FM177765 and EF196664; see Table 1 for details).

Mitogenome sequences were aligned and manually curated in Geneious v. 8.1.9 (Kearse et al., 2012). Summary statistics for the complete dataset (N = 35), as well as subdivisions based on phylogenetic analyses (below) were generated using DnaSP v.6.12.03 (Rozas et al., 2017). For these estimates we excluded sites with missing data, ambiguous data, or gaps.

2.6 Phylogenetic analyses

To generate a time-calibrated mitochondrial phylogeny of sun bears, we first estimated the root age in a species level analysis calibrated based on the fossil ages of other representatives of the Ursidae. We then applied this root age estimate to population level analyses of sun bear mitochondrial sequences. This two step strategy was chosen to best accommodate the assumptions of the tree priors used in these respective analyses.

For the species level analysis, we aligned two divergent sun bear mitogenomes (HMA_26_TH, and HMA_35_CAMB) and one mitogenome from each of the following nine species: Ursus maritimus (Polar bear; acc. no. AF303111), Ursus arctos (Brown bear; acc. no. HQ685956), Ursus thibetanus (Asian black bear; acc. no. EF196661), Ursus americanus (American black bear; acc. no. AF303109), Melursus ursinus (Sloth bear; acc. no. FM177763), Tremarctos ornatus (Spectacled bear; acc. no. EF196665), Ailuropoda melanoleuca (Giant Panda; acc. no. EF196663), Arctodus simus (extinct Short-faced bear; acc. no. FM177762) and Ursus ingressus (extinct Cave bear; acc. no. KX641331). We then used PartitionFinder v2.1.1 (Lanfear et al., 2016) to identify an optimal set of partitions and substitution models among all combinations of tRNAs, rRNAs and the three codon positions of the protein coding genes, using the greedy search algorithm and considering all substitution models available in BEAST, using the Bayesian Information Criterion. The Bayesian phylogenetic analysis package BEAST v.1.8.2 (Drummond et al., 2012) was then used to estimate species phylogeny and divergence times (Figure S1). A birth-death tree model was used, with a lognormal relaxed clock model, with a uniform prior on the mean substitution rate of 0 to 20% per million years (My). Time-calibration was achieved using informative uniform priors based on fossil and other evidence, following the approach of a previous study of the Ursidae (Kumar et al., 2017). Monophyly was enforced for all calibrated nodes. These were: (1) a prior based on the divergence of the Tremarctinae and Ursinae clades between 7 and 14 My, based on a fossil of Tremarctine bear Plionarctos (Tedford, 2001); (2) a prior for the basal divergence of the Ursidae with an upper limit of 12 My, based on a fossil of the Ailuropodinae (Abella et al., 2012) and a lower limit of 20 My, determined by a molecular dating study (Wu et al., 2015); (3) a prior on divergence of the Ursinae between 4.3 and 6 My based on the age of Ursus minimus (Gustafson, 1978); and (4) a prior on the divergence of U. arctos and U. maritimus between 0.48 and 1.1 My based on previous studies (Cahill et al., 2015; Hailer et al., 2012; Li et al., 2011). All other priors were left at default values. Initial runs showed a lack of convergence for some substitution model parameters, and so we substituted them with simpler models (Table S3). The final BEAST run involved 20 million generations, sampling the MCMC chain every 1000 generations, in order to achieve adequate burn in and posterior sampling of all parameters (ESS >200), assessed using the program Tracer 1.7.

The divergence estimate recovered for the two sun bear mitogenomes (mean 0.3054 My, standard deviation 0.0371 My) was then applied as normal prior on the root height for the intraspecific analyses, and were run with a Bayesian Skyline coalescent population model with eight groups. Data partitions and substitutions were estimated using PartitionFinder as described previously. As we did not expect variation in mitochondrial substitution rate within sun bears, we used a strict clock model with a uniform prior on the per-lineage substitution rate of 0 to 20% per My. Details of the MCMC runs were as described above. The maximum clade credibility tree was then selected from the posterior sample, and node heights scaled to the median posterior age, using TreeAnnotator 1.8.2.

Maximum-likelihood phylogenetic analyses for both mitogenome datasets were performed using RAxML 8.2.12 (Stamatakis, 2014). Clade support was assessed using 100 rapid bootstrap replicates assuming the GTR + CAT substitution model, with a subsequent thorough maximum likelihood search for the best tree assuming the GTR + G substitution model. For the species level analysis, the tree was rooted using the Giant Panda (acc. no. EF196663) as an outgroup (Figure S2). For the intraspecific analysis, the Asian black bear (acc. no. EF196661) was used as an outgroup (Figure S3).

We also generated a dataset combining portions of our mitochondrial sequences with a set of recently published ~1800 bp long mtDNA sequences from sun bears from Thailand, Peninsular Malaysia and Borneo (GenBank acc. no. MW316324–MW316405; Lai, Chew, et al., 2021). This combined dataset of shorter sequences consisted of 117 mtDNA sequences: 32 from this study, 82 from Lai, Chew, et al. (2021), one from Lai, Ratnayeke, et al. (2021), one from Yu et al. (2007), and one from Krause et al. (2008; Table S4). The alignment had a length of 1466 bp after removal of the repetitive portion of the control region. We generated a time-calibrated phylogeny of these sequences using the root age calibration described above, assuming a constant population size through time and a HKY + G substitution model. All other parameters of the analysis were kept as described above.

3.1 Mitogenome phylogeny

Phylogenetic analysis of the sun bear mitogenome dataset (35 sequences) revealed two well-supported clades (posterior clade credibility 1 for each clade), henceforth referred to as “Mainland clade” and “Sunda clade” following Kunde (2017), that diverged ~295 thousand years ago (kya; CI₉₅: 371–221 kya; Figure 2). These clades are also present in the phylogeny reconstructed using a maximum likelihood (ML) approach (Figure S3).

The Mainland clade includes sequences (N = 20) from Cambodia, China, and Thailand, as well as a sequence from a sun bear of unknown provenance (Genbank acc. no. FM177765; Krause et al., 2008). Within this clade, two lineages diverged ~171 kya (CI₉₅: 224–121 kya), of which one is currently only represented by a single sequence from Cambodia; support for this as a distinct lineage from the remaining mainland sequences is lower in the ML analysis (Figure S3). The remaining 19 sequences are in the second lineage, with a coalescence-time estimated at ~47 kya (CI₉₅: 64–31 kya). These sequences are mostly derived from Cambodian samples (16 of 19), and no obvious geographic structuring is apparent.

The Sunda clade includes sequences (N = 15) from Borneo, Sumatra and Peninsular Malaysia (all part of Sundaland), but also from Thailand, and from three archival samples of unknown provenance (sample details in Table S1). This clade also gave rise to two major lineages, which diverged ~128 kya (CI₉₅: 170–93 kya), that correspond to mtDNA lineages previously identified (Lai, Chew, et al., 2021): the Peninsular lineage and Sabah lineage (indicated in Figure 2). These two lineages have a large geographic overlap: both include sun bears from Sumatra, Peninsular Malaysia and Thailand. Notably, we were only able to retrieve the mitogenome for one sample from Borneo, so we are unable to determine if both or only one of these lineages occurs on this island. Furthermore, as mentioned above, the sample from Borneo shared its putative haplotype with a sample from Thailand.

It is evident that both the Mainland and Sunda clades are present in Thailand (Figure 2; Figure S3). Unfortunately, it is not known whether these Thai sun bear samples originated from North or South of the Isthmus of Kra (5–13°N), an important zoogeographic transition zone between Sundaland and continental mainland.

3.2 Shorter mtDNA sequences

We also considered the 35 mitogenome sequences in the context of a recent phylogenetic study on shorter mtDNA sequences of sun bears from Borneo, Peninsular Malaysia and Thailand (Lai, Chew, et al., 2021). Phylogenetic analysis of this dataset (N = 117 sequences; alignment length 1466 bp; Figure 3) also recovered the two major clades with high support (posterior clade credibility 1 for each clade; Figure 2), but both were much younger than the previous estimate of ~1.5 Mya (Lai, Chew, et al., 2021). This discrepancy in coalescence time estimates between the two studies likely reflects (i) our removal of the repetitive portion of the control region (which is difficult to resolve using short-read data), and (ii) the differences in choice of calibration points and dating methods between the two studies (see Section 2 for details).

We also enlarged the Mainland clade by Thailand samples T7, T8, T9, T10, T12 and T14 (Lai, Chew, et al., 2021), bringing the total to 26 sequences. Consistent with the mitogenome phylogeny, there are two lineages in this clade, of which one is represented by a single sequence (HMA_65_CAMB), albeit with a more recent coalescence time ~121 kya (CI₉₅: 181–66 kya).

Within the Sunda clade, we recovered the two previously identified major lineages: the Peninsula and the Sabah lineages (Lai, Chew, et al., 2021; indicated in Figure 3). Our estimate for the coalescence time of these lineages was ~180 kya (CI₉₅: 260–112 kya), again, much later than the previous estimate of ~600 kya (Lai, Chew, et al., 2021), but earlier than the estimate of ~128 kya based on mitogenome data (Figure 2).

The Peninsula lineage (N = 40) consisted mostly of samples from Peninsular Malaysia (N = 27), but also included samples originating from Thailand (N = 7) and Sumatra (N = 3), as well as three samples of unknown origin (Figure 3). The coalescence time estimate for this lineage was ~127 kya (CI₉₅: 193–73 kya), earlier than the estimate of ~82 kya based on mitogenome data (Figure 2). Consistent with the mitogenome phylogeny, this analysis revealed that the Peninsula lineage was also present in Sumatra. Our study also increased the number of sequences from Thailand and Peninsular Malaysia in this lineage.

The Sabah lineage (N = 51) consisted mostly of samples from Borneo (N = 37), but also included samples originating from Thailand (N = 8), Peninsular Malaysia (N = 4), and Sumatra (N = 2). The coalescence time estimate for this lineage was ~125 kya (CI₉₅: 186–73 kya), older than the estimate of ~99 kya based on mitogenome data (Figure 2). Consistent with the mitogenome phylogeny, this analysis revealed that the Sabah lineage was also present in Sumatra. Our study also increased the number of sequences from Borneo, Thailand, and Peninsular Malaysia in this lineage.

Thus, both the Peninsula and Sabah lineages were found in Peninsular Malaysia, Thailand and Sumatra, while the Sabah lineage was also present on Borneo. Inclusion of our samples in this analysis reduced the geographic structuring suggested by the results of Lai, Chew, et al. (2021); that is to say, some sub-lineages that were previously restricted to one geographic region, now include sequences from one or two additional geographic regions.

There is geographic origin information for some of the Thai sun bear samples from Lai, Chew, et al. (2021). Among the seven Thailand samples in the Mainland clade, there is geographic origin information for two (T7, T12), both of which come from North of the Isthmus of Kra (‘Thailand East’ in Lai, Chew, et al., 2021; indicated by ‘#’ in Figure 3). Among the 15 Thailand samples in the Sunda clade, there is geographic origin information for two (T1 in Sabah lineage, T11 in Peninsula lineage), both of which come from South of the Isthmus of Kra (‘Thailand West’ in Lai, Chew, et al., 2021; indicated by ‘*’ in Figure 3).

4.1 Secondary contact

The geographic distribution of mitogenome clades reflects vicariance between sun bears of mainland Indochina and Sundaland, a biogeographic pattern that is observed in many other SE Asian species that are widely distributed (Woodruff & Turner, 2009). It is unknown if the Mainland clade was lost in sun bears before, during, or after colonization of the Sundaland. If this loss occurred before or during colonization, this suggests that sun bears from the Sunda clade have persisted in the Thai-Malay Peninsula since the Middle Pleistocene. If the loss of the Mainland clade occurred after colonization, this suggests secondary contact of the two lineages in the Thai-Malay Peninsula, as both major clades are found in Thailand. Such migration was possible when the exposed shelf connected the mainland and the major islands of the Sundaland during the low sea levels of the late Pleistocene (Bird et al., 2005), and has been inferred for sun bears (Lai, Chew, et al., 2021) and other large mammals, such as leopard Panthera pardus (Wilting et al., 2016), leopard cat Prionailurus bengalensis (Patel et al., 2017), and muntjac Muntiacus muntjak (Martins, Fickel, et al., 2017).

If there was secondary contact of these two lineages following a northward migration of sun bears of the Sunda clade, we can consider two scenarios: (1) the replacement of sun bears of the Mainland clade by bears of the Sunda clade in portions of the Thai-Malay Peninsula, or (2) the recolonization of the Thai-Malay Peninsula by bears from the Sunda clade following a local extinction of Mainland clade sun bears, possibly resulting from the Toba volcano super eruption on Sumatra ~74 kya (Costa et al., 2014). The first scenario suggests that Sundaic sun bears would have had some competitive advantage over individuals of the local mainland population during the period(s) when the Sunda Shelf was exposed, and migration of Sundaic bears to the Thai-Malay Peninsula was possible. In the second scenario, we would need to consider that the volcanic eruption would have most likely severely affected the Sumatran sun bear population (as suggested for other species, e.g. leopard: Wilting et al., 2016). In this context, it is of note that the Peninsula lineage has thus far only been detected in Sumatra, Peninsular Malaysia and Thailand, but not on Borneo, raising the question where this lineage may have endured following the Toba eruption. A candidate region is Kalimantan, the Indonesian portion of Borneo, for which no mtDNA data are currently available.

Clearly, the available molecular data are not adequate to address these open questions, and further sampling is needed. Analyses should ideally also include nuclear genomic data, which would also be extremely valuable to address the taxonomic uncertainty regarding the sun bears of Borneo.

4.2 Bornean sun bears

There are clear phenotypic differences between the sun bears on Borneo and those from elsewhere in the species' range, and currently two subspecies are recognized: the broadly distributed H. m. malayanus and the Bornean H. m. euryspilus (Meijaard, 2004). However, several authors have suggested that further morphological and molecular data are needed to conclusively resolve the subspecific status of Bornean sun bears (Kitchener, 2010; Lai, Chew, et al., 2021).

Mitochondrial data can not disclose the mechanisms responsible for the observed phenotypic differences. However, it does show that Bornean sun bears harbor one Sundaic lineage (Sabah lineage) that is present throughout the region, including Sumatra, Peninsular Malaysia and Thailand, while a second widespread Sundaic lineage (Peninsula lineage) has thus far not been observed on Borneo. If the absence of the Peninsula lineage on Borneo is real (and not the result of sampling bias), then the question arises why the two lineages do not coexist on Borneo. It is also unknown how far their divergence time (~128 kya) relates to the divergence of Bornean sun bears. In this context, it can also not be discounted that historical or modern human-mediated translocation of sun bears may have contributed to the observed distribution of mitochondrial lineages in the Sundaic region, further complicating matters.

4.3 Human-mediated translocations

Sun bears are hunted for the illegal pet trade (Foley et al., 2011; Gomez et al., 2020; Krishnasamy & Shepherd, 2014; Lee et al., 2015; Shepherd & Shepherd, 2010), and are traded on and among the Sundaic islands (e.g., Gomez et al., 2019). Rescue and release of such trafficked bears may result in the introduction of non-endemic mtDNA haplotypes into local populations. It is worth noting that such translocations do not need to be recent. There is a history of human-mediated introduction or translocation of mammals (e.g., Long, 2003) for various reasons (e.g., cultural, commercial, pest-control, accidental). Such translocations included large mammals and carnivores in SE Asia, among others cervids (Rusa timorensis and R. unicolor, Martins, Schmidt, et al., 2017), pigs (Groves, 1984), the leopard cat (Prionailurus bengalensis, Patel et al., 2017), the Malay civet (Viverra tangalunga, Veron et al., 2014), and the Asian palm civet (Paradoxurus hermaphroditus, Flannery et al., 1995). Because some of these human-mediated translocations had commercial reasons (e.g., Groves, 1984), it is conceivable that there was also historical trade in bears and their derivatives (e.g., Hose, 1893), which may have included the translocation and subsequent (intentional or unintentional) release of sun bears. Investigation of Early Holocene samples using ancient DNA techniques could address the impact of human-mediated translocations. Unfortunately, sun bear fossils are rare, and the species inhabits an environment that is not conducive to ancient DNA preservation (e.g., Paijmans et al., 2013).