Technical Comments

Response to Comment on “Late Pleistocene human skeleton and mtDNA link Paleoamericans and modern Native Americans”

See allHide authors and affiliations

Science  20 Feb 2015:
Vol. 347, Issue 6224, pp. 835
DOI: 10.1126/science.1261188


Prüfer and Meyer raise concerns over the mitochondrial DNA (mtDNA) results we reported for the Hoyo Negro individual, citing failure of a portion of these data to conform to their expectations of ancient DNA (aDNA). Because damage patterns in aDNA vary, outright rejection of our findings on this basis is unwarranted, especially in light of our other observations.

Using an analysis of nucleotide damage patterns and fragment sizes in Illumina sequences from the Hoyo Negro remains, Prüfer and Meyer (1) suggest that the mitochondrial haplogroup D1 sequences we obtained were not endogenous to the sample. Although they raise important concerns, we note that (i) the assignment of this individual to haplogroup D1 was based on analyses of three DNA extracts from two laboratories and (ii) expectations for the amount of damage observed in an ancient sample of this age and context are not firmly established.

We took precautions to minimize contamination from exogenous DNA [see the supplementary materials (SM) in (2)] and conducted several analyses to assess whether the extracted DNA exhibited expected characteristics of ancient DNA (aDNA) and to confirm haplogroup D1. We made the following observations: (i) this haplogroup was not detected in negative controls; (ii) amplification of mitochondrial DNA (mtDNA) fragments <200 base pairs (bp) was not uniformly successful (i.e., sporadic amplification was observed); (iii) failure to generate X- and Y-chromosome amplicons; (iv) AluI site loss at nucleotide position (np) 5176, diagnostic of haplogroup D, was confirmed through intra- and interlaboratory replication using three DNA extractions; and (v) hypervariable region sequences from all three extracts yielded differences from the revised Cambridge Reference Sequence (3) consistent with membership in haplogroup D1. Collectively, these observations are consistent with a low copy number and a highly degraded DNA sample, characteristics expected from aDNA (4). Moreover, our results make phylogenetic sense (5).

As disclosed [SM in (2)], an attempt to reconstruct a complete mitogenome from the first extract (HN-WSU-1) was compromised by cross-contamination from a sample (of haplogroup C) processed in parallel at the University of Illinois. Consequently, we used the high-throughput data to confirm the presence of haplogroup D1 definitive single-nucleotide polymorphism (SNPs) [SM (2)], similar to the approach of screening SNPs in deeply sequenced amplicons compromised by contamination (6). We did not include an assessment of postmortem damage in the SM for two reasons. First, the DNA library was prepared using AT-overhang adapter ligation and a proofreading enzyme (Kapa HiFi DNA Polymerase), which limits our ability to detect nucleotide misincorporation damage patterns (7). Second, because most sequence reads did not contain SNPs specific to haplogroup D, it was not possible to present a damage profile specific to the complete D1 mitogenome.

Prüfer and Meyer’s (1) efforts to disentangle sequence reads from multiple mitogenomes and assess them for signs of DNA damage are informative. However, they considered only two diagnostic SNPs for haplogroup D1. We followed their example and examined reads containing any of the eight diagnostic SNPs for subhaplogroup D1 (nps 2092, 3010, 4883, 5178, 8414, 14668, 16325, and 16362). Through this analysis, we identified previously undetected polymerase chain reaction (PCR) duplicates (i.e., reads that had been trimmed to different lengths based on sequence quality). Our analysis of the Illumina data set reveals only 27 unique reads with diagnostic haplogroup D SNPs, sequenced to a 3.86× depth. Fragment length ranges from 35 to 100 bp, with an average of 85.8 bp. In these 27 reads, we see no clear indication of G-to-A misincorporations near the 3′ end of reads or C-to-T misincorporations near the 5′ end, confirming the general observations of Prüfer and Meyer (1).

Prüfer and Meyer (1) argue for a specific DNA damage pattern as a reliable means of authenticating aDNA sequences, but that expectation makes assumptions about the rate of deamination over time and the frequency of resulting nucleotide misincorporations. Because this form of damage is tremendously variable (8, 9) and not time dependent (8, 10), it is unclear whether postmortem damage at the 3′ and 5′ ends should be observable among so few reads from Hoyo Negro. If damage observed in reads from the ~30,000-year-old Kostenki 14 human (~38%) (9) is an appropriate model for Hoyo Negro, the binomial probability of observing 0/27 reads with damage is 2.4 × 10−6. However, if we accept a model of 1 to 9% damage (from Arctic samples ≤ 5400 years old) (8), the probability would be 0.078 or higher. It is probable that neither model is appropriate for the Hoyo Negro remains. Thus, although the absence of damage-induced nucleotide misincorporations in the Illumina haplogroup D sequence reads does raise concerns about contamination, it is also possible that the 27 reads showing diagnostic D SNPs are derived from endogenous aDNA and that we cannot detect postmortem damage due to chance and/or our library preparation methods.

If the Hoyo Negro mtDNA results represent contamination, its source is unknown. Because DNA analysis was a key objective of the study, extreme caution was used in sample recovery and transport. Haplogroup D1 is also absent among the collection and laboratory teams. Finally, all samples were submerged in 6% sodium hypochlorite before extraction, a reliable means of surface decontamination (11). The putative contamination could have originated from laboratory reagents, but the reagents used by each laboratory originated from separate lot numbers. Although contamination could affect multiple lots, the probability of observing haplogroup D1 contamination in three independent extracts at two laboratories is low, though not zero.

Although it might never be possible to authenticate aDNA from humans with absolute certainty (1), investigators have offered a variety of recommendations for properly conducting this type of research [e.g., (4, 5, 12)]. Prüfer and Meyer’s (1) suggestion is that the authentication of human aDNA should hinge on the presence of hallmark damage signatures (9, 13). However, it is premature to rely so heavily on such damage patterns without a greater appreciation of the variance expected across samples from different environments. Furthermore, although decades of research have suggested that contaminant DNA will be more intact than aDNA, recent work brings this assumption into question (9) and demonstrates that contamination can also take on forms of damage expected of aDNA (14). We, therefore, still regard independent replication as an important criterion for authentication in human ancient DNA studies, despite the recent trend of ignoring this criterion. Nevertheless, with ongoing analysis of DNA from the Hoyo Negro remains, we are paying closer attention to patterns of DNA damage as an additional means of evaluating the authenticity of the sequence data.


View Abstract

Navigate This Article