Analysis of the phylogenetic relationship among clouded leopards based on a concatenated fragment of 900 bp of mitochondrial DNA comprising sequences from three mitochondrial gene segments (185 bp ATPase-8, 286 bp Cyt b and 426 bp control region) revealed 54 variable sites among clouded leopards. Of those 39 were fixed nucleotide differences between specimens from the islands of Borneo/Sumatra (N. diardi) and specimens from the mainland (N. nebulosa nebulosa). There were 12 nucleotide differences in the Cyt-b gene (4.2 %), 10 in the ATPase-8 gene (5.4 %) and 17 in the control region (3.9 %). In comparison, three Panthera species (P. pardus, P.tigris and P. onca) were separated by 38–52 nucleotide differences in the same fragment. Four fixed nucleotide differences distinguished the individuals from Sumatra and Borneo. Altogether we found 13 haplotypes among all the clouded leopards examined. N. nebulosa nebulosa (N = 59 GenBank accession no. in Table 2. see Figure 1) had six haplotypes (NEB 1–6), N. diardi (Borneo, N = 7) had 5 haplotypes (DIB 1–5) and in N. diardi (Sumatra, N = 3) we found 2 haplotypes (DIS 1 & 2).

Phylogenetic analysis of mtDNA haplotypes using minimum evolution (ME), maximum parsimony (MP) and maximum likelihood (ML) approaches produced congruent topologies that strongly support the reciprocally monophyletic status of N.nebulosa and N. diardi with high bootstrap values (100 % ME/MP, 98 % ML) (Figure 2A). Furthermore, Sumatran individuals were separated from Bornean clouded leopards supported by high bootstrap values; except for the ME approach (46% ME, 85 % MP, and 81 % ML).

A nuclear pseudogene insertion of cytoplasmic mtDNA often referred as numt have been reported in numerous organisms [reviewed in 18] including the Panthera genus and the domestic cat 192021. This can complicate the analysis of mtDNA, because numt can coamplify with the mitochondrial genes. We cannot exclude the presence of numts in clouded leopards. However, it is very unlikely that we sequenced numt instead of mtDNA, because all our protein coding sequences had open reading frames 22. Furthermore other nuclear genes did not amplify with our ancient DNA, even if we tried to sequence very short fragments. Therefore, we conclude that most likely we did not encounter any evidence of numt.

Composite genotypes from 18 felid-specific microsatellite loci 2324 were obtained from 11 clouded leopard samples (four mainland specimens NNE 1 – 4, four Bornean specimens NDB 1 – 4 and three Sumatran specimens NDS 1 – 3), two domestic cats (Felis catus; FCA 1 – 2) and one leopard (PPA 1). Of the 10 loci, which were used before in clouded leopards, Buckley-Beason et al. 15 showed that six microsatellite loci (FCA 82, FCA 105, FCA 132, FCA 144, FCA 261, FCA 310) had non-overlapping allele sizes between N. nebulosa (mainland) and N. diardi (Borneo). For only one of these six loci, FCA 261, non-overlapping allele-sizes could not be confirmed by wider sampling of Bornean and Sumatran specimens. Out of eight microsatellite loci, which were not tested in clouded leopards before, three (FCA 23, FCA 43 and HDZ 859) did not overlap in allele sizes. Overall expected heterozygosity H_E in clouded leopards ranged from 0.488 in the Bornean population to 0.652 in N. nebulosa and exceeds the observed heterozygosity H₀ in all three populations, with Bornean specimens having the lowest observed heterozygosity H₀ with 0.361 (Table 3).

Neighbor-joining analysis of individual clouded leopard genotypes based on two microsatellite genetic distance estimators (Dps & Dkf) produced concordant topologies; both trees support the species distinction among clouded leopards (Figure 3). Individuals from Borneo and Sumatra form a monophyletic clade with 100 % (Dps & Dkf) bootstrap support (BS) distinguishing them from mainland specimens and the outgroups. The microsatellite analysis lends further support to the phylogeographic subdivision observed in mtDNA analysis between Borneo and Sumatra individuals, however with lower bootstrap support values (Sumatra clade with 69 % Dps/61%Dkf BS, and Borneo clade with 50/46 % BS).

To evaluate population distinctiveness we tested the microsatellite data using a Bayesian algorithm as implemented in BAPS version 4.14 25. To estimate hidden substructures, within clouded leopards BAPS suggested to partition clouded leopard into five populations (p > 0.99) three different mainland populations, Sumatra and Borneo. However mtDNA results and previous analysis based on a broader sampling, including more microsatellite loci found no substructure among mainland populations 15. Therefore we assume that the population partition within our four mainland individuals is a result of the small sampling size and thus set a maximum of three populations. BAPS then grouped all mainland, Sumatra, and Borneo individuals together, respectively (p = 1). We confirmed the individual assignments to each population with admixture analysis in BAPS 2627. In this scenario all individuals were assigned to three unique clusters N. nebulosa,N. diardi (Borneo) and N. diardi (Sumatra) with very high admixture coefficients (all q > 0.98, except for one Sumatran individual (NDS1) q > 0.95) (Figure 2B). No individual had a Bayesian p-value less than 0.05 (all p > 0.91) revealing no evidence of admixed background and none or very low degrees of past and present gene flow between all three populations.

MtDNA sequence differences were used to estimate the divergence time of N.nebulosa and N. diardi and the time of origin of Sumatran clouded leopards. Using a calibration of 6.37 MYA for the divergence of clouded leopards from the Panthera lineage 14, N. diardi diverged from N. nebulosa about 2.86 MYA (95 % CI of 1.71–4.02 MYA) and Sumatran and Bornean clouded leopards diverged about 437,000 years ago (95 % CI of 30,000 – 845,000 years ago).

We sampled four wild-born specimens from the islands of Borneo, three ancient samples from Sumatra and four recent animals from known geographic origin on the mainland (Table 1) in addition to 58 clouded leopard individuals described previously 15. Panthera species and domestic cat individuals as outgroups for mitochondrial DNA (mtDNA) analysis were specified before 15 and for outgroup comparison of the microsatellite analysis one leopard (Panthera pardus) and two domestic cats (Felis catus) were sampled (Table 1).

DNA was extracted from fecal samples using QiAmp Stool Mini Kit (Qiagen, Hilden, Germany) and from serum and whole blood using QiAmp DNeasy (Qiagen, Hilden, Germany). For the extraction of DNA from historical museum samples hide and dry tissue samples were cut into small pieces using a sterile scalpel followed by a standard proteinase K digestion with an extended incubation interval at 56°C for up to 72 hours 40 until only little solid tissue remained. A standard phenol/chloroform extraction procedure was then used for DNA extraction 52. DNA was suspended in 50 μl of double distilled H₂O.

We used a 426 bp portion of a central conserved region within the D-loop of the control region 53 in addition to two mtDNA (ATPase-8 and Cyt-b) genes 15. The control region primers were modified by Janecka JE (pers. comm.) [(F) CTC AAC TAT CCG AAA GAG CTT] and [(R) CCT GTG GAA CAT TAG GAA TT]. In total we amplified 900 bp of mtDNA.

PCR reactions were performed in a final volume of 25 μl containing 2.5 μl MolTaq 10x PCR Buffer, 2 mM MgCl₂, 0.2 mM dNTPs, 1 μM of each primer, 2 units of MolTaq polymerase (Molzym GmbH, Bremen, Germany) and 2 μl of genomic DNA. PCR reactions were performed in an Eppendorf Mastercycler (Eppendorf GmbH, Wesseling-Berzdorf, Germany), with an initial denaturation step at 95°C for 3 min, followed by 35 cycles of denaturation at 94°C for 30 s, annealing at 56°C for 45 s, elongation at 72°C for 45 s, and were completed with a final elongation step at 72°C for 10 min. PCR products were purified by ultra filtration through Montage TM filter devices (Millipore GmbH, Schwalbach, Germany) and sent to Seqlab (Seqlab Laboratories, Göttingen, Germany) for sequencing. Sequences were edited, assembled and aligned using ClustalW 54 implemented in BioEdit (Version 7.0.5.2) 55, before being exported to PAUP (Version 4.0b10) 56 for phylogenetic analysis. Sequences from each of the mtDNA fragments were concatenated into a 900 bp sequences, because results from separate analysis of each gene fragment showed identical topologies 57 and mitochondrial genes usually do not recombine 58. Phylogenetic relationships among haplotypes were estimated using minimum evolution (ME), maximum likelihood (ML) and maximum parsimony (MP) 5659. We used Kimura 2-parameter distance with neighbor-joining (NJ) algorithm followed by tree-bisection reconnection branch-swapping procedure (TBR) for the ME analysis. MP trees were conducted using a heuristic search, with 10 random taxon addition replicates and TBR branch swapping. The ML approach was performed using the HKY85 model 60. Each phylogenetic tree was rooted with the domestic cat sequence. Reliability of all trees was tested with bootstrap values by 1000 replicates of heuristic search and TBR branch swapping.

The approximate age of separation between N. nebulosa and N. diardi was estimated using LINTREE 61. A neighbor-joining tree 59 was generated with Kimura 2-parameter γ-corrected distances 62 using the combined 900 bp mtDNA sequence. The molecular clock test implemented in LINTREE 61 showed that the sequences did not deviate significantly from the rate constancy test (p > 0.05). The coalescence point between clouded leopards and the Panthera genus, being 6.37 MYA based upon a comprehensive analysis of nuclear gene sequences and multiple fossil dates 14, was chosen to be the calibration point for this study. We used a range of two standard errors to calculate a 95 % confidence interval.

We used 10 felid dinucleotide microsatellite primers (FCA 8, FCA 45, FCA 77, FCA82, FCA 105, FCA 126, FCA 132, FCA 144, FCA 261, FCA 310) 23, which were already used in a previous study on clouded leopards 15. Those microsatellites are located on 8 felid autosomes and all of them are at least 5 centimorgans apart from each other 23. In addition to those ten microsatellite loci of known allele size ranges for clouded leopards, we applied 8 microsatellites of unknown allele sizes for clouded leopards, FCA 23 and FCA 43 23, and HDZ 3, HDZ 57, HDZ 64, HDZ 89, HDZ817, HDZ 859 24. PCR amplifications were performed in a final reaction volume of 10 μl utilizing described methods 2324. The IR-dye-labeled PCR products were diluted and analyzed on a LI-COR 4300 DNA-Analyser (LI-COR Bioscience GmbH, Bad Homburg, Germany). Data were collected and analyzed using Saga Generation 2 (Version 3.2.1).

To test their performance as population genetic markers, all microsatellites were tested for deviations from linkage disequilibrium (LD) using GENEPOP on the web version 3.4 63. Measures of microsatellite genetic variation in terms of observed and expected heterozygosities were estimated with Arlequin 3.1 64.

Pairwise genetic distance among clouded leopards and the two outgroup species was estimated with two microsatellite genetic distance estimators: the proportion of shared alleles (Dps) and the kinship coefficient (Dkf) with the [1 – ps/kf] option in MICROSAT 65. Phylogenetic NJ-trees were constructed from the Dps and Dkf distance matrixes using NEIGHBOR (included in PHYLIP version 3.66) 66. Bootstrap values for 1000 bootstrap replicates in MICROSAT were calculated using CONSENSE TREE (included in PHYLIP version 3.66) 66. Trees were drawn using the program TREEVIEW (version 1.6.6) 67.

A Bayesian clustering method as implemented in BAPS 252627 was used to infer population structure based on multilocus microsatellite genotype data. This program estimates the hidden population substructure by testing whether the allele frequencies between populations are significantly different. A major advantage compared to most other methods is that the number of populations is treated here as an unknown parameter that can be estimated from the dataset. We performed 10 independent runs of clustering of individuals with the microsatellite genotypes to ensure homogenous results. In all ten runs we obtained similar results (data not shown). After the clustering of individuals by their allele frequencies the results were used to perform an admixture analysis. We used 500 iterations and a number of 1000 reference individuals per population each with 20 iterations. The estimated admixture coefficient for an individual in each cluster q (maximum = 1) was used as a measure of correct assignments. The Bayesian p-value tells the proportion of reference individuals simulated from the population in which the individual was originally clustered having the admixture coefficient to the cluster smaller than or equal to the individual 27. Individuals having p-values larger than 0.05 are by default considered as having "non-significant" evidence for admixture 27.