Fig. 4. The ancestry of dogs.

a, PCA on post-LGM and present-day wolves (X), based on f₄-statistics only of the form f₄(X,A;B,C), where A, B and C are any of 21 wolves predating 28 ka. Dogs are projected, and coloured by f₄(AndeanFox,X;Zhokhov dog 9.5 ka,Tel Hreiz dog 7.2 ka). b, For dogs (X), f₄(AndeanFox,X;Zhokhov dog 9.5 ka,Tel Hreiz dog 7.2 ka) horizontally against f₄(AndeanFox,X;Belaya Gora wolf 18 ka,Hohle Fels wolf 13 ka) vertically (Pearson’s r = 0.86, P = 3 × 10^–23). Bars denote ±1 standard error estimated from a block jackknife. Silhouettes from phylopic.org. c, log₁₀(P values) for qpAdm models fit to dog targets, where a low P value means the model can be rejected. An ancient dhole was used to represent unsampled, divergent ancestry; models including this source have black outlines. Points are jittered horizontally to avoid overlap. d, qpAdm ancestry proportions for dogs, using the Zhokhov (9.5 ka) dog and a present-day Syrian wolf as proxies for eastern and western dog progenitor ancestry, respectively. Bars denote ±1 standard error estimated from a block jackknife. e, Map of early and relevant later dogs and their ancestry proportions as in d. Black crosses indicate the locations of wolves from 25–10 ka that can be rejected as dog progenitors. Base map from the mapdata R package. k, thousand years. f, Admixture graph model of major dog lineage relationships, fit by qpGraph with no outlier f-statistics. Edge lengths are in units of F_ST (×1,000).