Historical range of tiger distribution is shown in light tan and current range is shown in dark tan, while green dots indicate individual historical recordings of tigers outside of normal distribution [1]. Green ‘’ indicate records from the Middle Ages [4]. Black lines demarcate presumed subspecies boundaries [3]. Abbreviations correspond to traditionally named tiger subspecies, arranged chronologically by date of naming. 1) tigris Linnaeus, 1758[37]; 2) virgata Illiger, 1815[38]; 3) altaica Temminck, 1844[39]; 4) sondaica Temminck, 1844[39]; 5) amoyensis Hilzheimer, 1905[40]; 6) balica Schwarz, 1912[41]; 7) sumatrae Pocock, 1929[42]; 8) corbetti Mazak, 1968[43]; 9) jacksoni Luo et al., 2004[11]. Lettered arrows indicate postulated dispersal avenues: (A) Indian, southern route; (B) Siberian, northern route; and (C) Silk road/ Gansu route with (D) secondary eastward dispersal. See text for details. Redrawn from Figures 19 and 20 in Mazak [1] and Figure 1 in Kitchener and Dugmore [3].

Haplotype designations are color coded by subspecies of the tigers that carried them. PTV-2 is a Caspian tiger (Panthera tigris virgata) specimen for which all gene segments attempted (1257 bp) in Caspian tigers were successfully sequenced (see Table S2). Other Caspian tigers produced partial sequences identical to PTV-2. The only exceptions were three individuals, each displaying a single derived nucleotide difference when compared to PTV-2 (found only in that individual and in no other tigers of any subspecies). Likewise, the only mtDNA haplotype carried by Amur or “Siberian” tigers (P. t. altaica) proved to be a single derived step away from the haplotype of PTV-2, suggesting a close relationship between the Amur and Caspian tiger subspecies. Tiger haplotypes carried by all but the Caspian subspecies are from a previously published dataset, [11] while a clouded leopard (Neofelis nebulosa) sequence (GenBank DQ257669) [14] was used to root the tree. The tree depicted was inferred using maximum parsimony, with the number of steps/homoplasies listed above the branches, while (for major clades) bootstrap percentages are listed below branches for maximum parsimony, maximum likelihood and Neighbour Joining methods. We used full length mtDNA sequences of clouded leopard, leopard and snow leopard to root the tree; all combinations of 1, 2 or 3 outgroups yielded trees with similar topology to the one depicted, with the same basal position for the P. t. amoyensis AMO1 haplotype, and a close relationship between P.t. virgata and P. t. altaica haplotypes.