None of the letters in response to Thomson et al. (1) undermine our conclusions. However, several issues have been raised, which we address in this reply. Beavan (2) dismisses some of the concerns that have been raised about the accuracy of the radiocarbon dates of the El Arenal-1 chicken bones, which are immediately pre-Columbian. Although procedures, such as ultrafiltration of amino acids, are common practice for suboptimal bone samples, such as the oldest El Arenal-1 sample, further complex issues, including dietary sources and the potential for indirect marine carbon input (3), mean that a detailed assessment of the site through multiple further dates would be required to exclude the possibility that the specimens might actually be post-Columbian. This approach seems particularly advisable given that analogous issues were raised about a surprisingly early date for New Zealand colonization based on Pacific rat bone dates generated at the same laboratory using similar procedures (4), which were subsequently shown to be erroneously old (5).

Storey and Matisoo-Smith (6) raise the issue of potential contamination of laboratory components with modern DNA, including domestic animal DNA, which is well recognized in ancient DNA research (7, 8). Multiple control reactions are acknowledged to be necessary but insufficient to detect low levels or sporadic contamination, and issues such as the “carrier effect” (7, 8) can mask the true frequency of contamination. Storey and Matisoo-Smith (6) do not dispute that we have demonstrated that at least a proportion of the results reported in Storey et al. (9) were contaminated with the common chicken haplogroup E. Unfortunately, it is impossible to demonstrate the full extent of this problem within the study, as it was not possible to reanalyze most of the samples, the details of the reported independent replication were not given, and the results are not available. However, it is important to note that in tropical sites only recent samples (500–1,400 y old) are predicted to yield DNA via PCR (10), and all of the chicken specimens older than this [>1410 cal B.P. (9, 11)] yielded a likely laboratory contaminant (haplogroup E). This result is perhaps more consistent with increased rates of laboratory contamination because authentic DNA levels diminish below detectability, than with two waves of chicken introductions across the Pacific (9).

Bryant (12) questions two aspects of the statistical analyses conducted in Thomson et al. (1). The first issue concerns the Bayesian Serial Simcoal and approximate Bayesian computation analysis. We thank the author for identifying an error in the figure legend of figure S11 in ref. 1. This legend should read: “Eight alternate scenarios were also tested: route from Europe–South America with or without migration, H3 or H1 respectively, with each model having two variations…and a route from Pacific-South America with or without migration, H4 or H2 respectively” (1). The figure itself is correct and shows that the Europe–South America model without the migration matrix was the more likely model.

We applied the suggested two-sample proportion test (Fisher’s exact test) (12), and found that the detection of the two different reported levels of E haplogroups in the Pacific remains unlikely (0/22 and 15/31; P = 8.67 × 10⁻⁵). Although it is true that the statistical tests assume random sampling from a naturally occurring distribution, the temporal and geographic mixing in the island archaeological sites analyzed seem likely to reflect an approximate sample of the chicken diversity present on any island. Although it is possible that combining chicken samples from different islands in an analysis may obscure interisland variability in the proportion of “D”s vs. “E”s, there is only a single site common to both studies (Anakena on Rapa Nui), and the limited number of samples prevents statistical analysis. However, given that haplogroup E chickens are found worldwide, whereas the ancient D haplotypes we identified are specific to the Pacific, we maintain that the latter are likely to be authentic, but at least some—if not all—of the reported E haplotypes (9, 11) represent contamination.