Diagenesis was studied in DNA obtained from Siberian permafrost (permanently frozen soil) ranging from 10,000 to 400,000 years in age. Despite optimal preservation conditions, we found the sedimentary DNA to be severely modified by interstrand crosslinks; single- and double-stranded breaks; and freely exposed sugar, phosphate, and hydroxyl groups. Intriguingly, interstrand crosslinks were found to accumulate ∼100 times faster than single-stranded breaks, suggesting that crosslinking rather than depurination is the primary limiting factor for ancient DNA amplification under frozen conditions. The results question the reliability of the commonly used models relying on depurination kinetics for predicting the long-term survival of DNA under permafrost conditions and suggest that new strategies for repair of ancient DNA must be considered if the yield of amplifiable DNA from permafrost sediments is to be significantly increased. Using the obtained rate constant for interstrand crosslinks the maximal survival time of amplifiable 120-bp fragments of bacterial 16S ribosomal DNA was estimated to be ∼400,000 years. Additionally, a clear relationship was found between DNA damage and sample age, contradicting previously raised concerns about the possible leaching of free DNA molecules between permafrost layers.
THE successful retrieval of ancient DNA (aDNA) sequences from fossil remains is limited by the postmortem instability of nucleic acids. It is generally assumed that DNA fragmentation caused by depurination (followed by cleavage of the phosphodiester backbone by β-elimination) is the fastest of the spontaneous chemical reactions limiting the half-life of amplifiable aDNA (e.g., Pääbo and Wilson 1991; Poinar et al. 1996, 2002; Marota et al. 2002). However, the notion behind this assumption remains untested, and some recent studies suggest that crosslinking preventing amplification due to reduced denaturation might be a significant contributor to DNA diagenesis in some environments (Poinar et al. 1996; Willerslev et al. 2004a; Willerslev and Cooper 2005). Here we address the role of strand breaks vs. crosslinks in limiting the amplification of aDNA molecules under permafrost conditions. This was done by comparing the relative rates of single-stranded breaks (SSBs) and interstrand crosslinks (ICLs) in newly obtained DNA and previously described DNA (Mitchell et al. 2005) extracted directly from five Siberian permafrost cores ranging from 10,400 to 400,000–600,000 years in age (for experimental details see supplemental materials at http://www.genetics.org/supplemental/).
Depurination and crosslinking:
The presence of depurination-generated SSBs was investigated by treating parts of the extracted permafrost DNA with alkali or shrimp alkaline phosphatase (SAP). These treatments remove sugar residues and phosphate groups, respectively, leaving 5′-termini OH groups that are available for a radioactive phosphate group (32P), using the T4 polynucleotide kinase (PNK) enzyme (Sambrook and Russel 2001). Additionally, radioactive nucleotides were incorporated into the recessed or intrinsic 3′-OH ends of the permafrost DNA, using the Klenow fragment of Escherichia coli DNA pol I (Telford et al. 1979). These treatments revealed the presence of a substantial number of 5′-OH termini, 5′-termini free sugar radicals, and 5′-PO4 groups as well as 3′-OH termini (Wilcoxon matched pairs, P < 0.05), corresponding to the end products expected by SSBs generated after depurination by β-elimination (Lindahl and Nyberg 1972). Additionally, the treatments showed higher yields of PO4 groups relative to free sugar residues, agreeing with the pattern previously observed in depurination studies of fully hydrated DNA from phage PM2 (Lindahl and Andersson 1972). We therefore concluded that SSBs generated by depurination are present in the permafrost-preserved DNA.
The presence of ICLs was investigated by the refractivity of the DNA extracts to heat denaturation, using a very sensitive microplate fluorescence assay and PicoGreen (Molecular Probes, Eugene, OR) as described in Willerslev et al. (2004a). The fluorescent DNA probe has 10-fold greater affinity for double-stranded than for single-stranded DNA, and thus a loss of duplex DNA induced by denaturation causes a significant reduction in fluorescence signal. By comparing the extent of signal reduction in the permafrost DNA with that of an undamaged control (E. coli) measured at comparable concentrations we found that all DNA extracts are significantly affected by ICLs (Table 1 ).
Lesion frequencies were obtained as described previously (Lindahl and Nyberg 1972) and were used to calculate the rate constants for SSBs and ICLs by assuming that the DNA degradation follows first-order kinetics (Table 1). Thus, the amount of undamaged bases (Bt) at time t is given by the equationwhere B0 is the initial concentration of bases and k is the rate constant. The amount of damaged bases Bd = B0 − B0e−kt and the lesion frequency (flesion = Bd/B0) is given by
Since e−kt = 1 − flesion, it follows that k can be estimated as the slope of the line −ln(1 − flesion) plotted against time. This gave a rate constant for SSBs of 1.4 × 10−15 sec−1 (resulting in a half-life of 1.7 × 107 years; Table 1), which is somewhat similar to the depurination rate constant determined by Lindahl and Nyberg (1972) of 2.5–3.1 × 10−15 sec−1 (extrapolated to −12°, the temperature of the permafrost). The rate constant for ICLs was calculated to be 1.0 × 10−13 sec−1, which is ∼40 times faster than the preliminary estimates reported by Mitchell et al. (2005), and results in a half-life of 2.2 × 105 years. This suggests that ICLs are accumulating ∼100 times faster than SSBs under frozen conditions (Table 1). As crosslinks hinder DNA amplification by preventing template denaturation, the result contradicts the general assumption that fragmentation through depurination is the most important type of damage limiting the amount of amplifiable DNA in fossil remains, including those in permafrost. This finding is in agreement with previous claims of measurable putative DNA sequences being present in permafrost even though amplification products cannot be obtained (Willerslev et al. 2004a).
Predicting DNA survival:
Some of the most commonly used models for predicting the long-term survival of DNA in fossil remains are those based on amino acid racemization (Poinar et al. 1996) and thermal age (Smith et al. 2001). Implicit in these models is a belief that depurination is the only significant type of damage limiting the half-life of amplifiable aDNA. Recent studies of DNA from fossil coprolites that are preserved under warm and arid conditions have questioned the general validity of this assumption by showing that intermolecular crosslinks, rather than fragmentation, determine the access to template molecules for PCR (Poinar et al. 1998). Our results suggest that also under frozen conditions the key factor in determining the half-life of amplifiable DNA is crosslinks rather than SSBs. Thus, the obtained results seriously question the reliability of models purely relying on depurination kinetics to predict the long-term survival of permafrost-preserved DNA. Importantly, Arctic permafrost is a fairly dry environment with only 3–8% of the total water being in an unfrozen state (Willerslev et al. 2004b) and limited hydrolytic damage could possibly account for the fairly low levels of observed SSBs relative to crosslinks. Additionally, the close to neutral pH in the Siberian permafrost (Willerslev et al. 2004a) favors long-term DNA preservation. Thus, similar studies are needed on various types of materials to determine to what extent these results can be generalized. In our opinion, it is likely that the relative rates of various damage types are highly variable across environments, depending on their exact composition of free water, pH, salt, oxidative free radicals, etc. Therefore, it is most likely impossible to use general models, such as those of amino acid racemization (Poinar et al. 1996) and thermal age (Smith et al. 2001), to predict the long-term survival of DNA across all environments. Instead, damage rates for each unique environment need to be investigated before more realistic predictions on DNA survival, such as those shown below, can be made.
Bacterial DNA survival:
The estimated rate constants for the degradation of DNA in permafrost samples permit us to roughly calculate, for a given starting number of template molecules, the expected number remaining after a given time. Assuming that the DNA degradation follows first-order kinetics, the amount of undamaged bases at time t is Bt = B0e−kt, and the proportion of undamaged bases is Bt/B0 = e−kt. Under the assumption that all bases have the same chance of being damaged, this corresponds to the probability that any given base is still undamaged at time t. The chance that two neighboring bases are both undamaged at time t is then (e−kt)2 assuming independence, and the chance that all bases in a fragment of length n are undamaged is (e−kt)n. From this we first estimated the probability of survival of 120-bp fragments, assuming that the only process that affects DNA survival is single-stranded breaks (Figure 1). The survival probability decreases rapidly and after 2 million years the chance of a 120-bp fragment surviving would be 2.5 × 10−5. Although this probability is low, in light of the high initial number of templates that would have been present when the soil entered the frozen state, this still would result in the survival of a considerable number of template molecules. For example, direct counts of bacteria in surface soils usually reveal ∼109 bacterial cells per gram of soil, and a specific study from Siberian permafrost found 108 bacteria in the active layer directly above the permanently frozen soil (Kobabe et al. 2004). In a previous study, we attempted to amplify bacterial DNA fragments from the same permafrost samples as used in this study, using DNA extracted from 2 g of soil (Willerslev et al. 2004a). If we assume that each of the bacterial cells had, on average, 3.8 copies of the 16S rRNA gene (Fogel et al. 1999), then one PCR reaction from the 1.5- to 2-million-year-old sample would contain between 2.0 × 103 and 2.6 × 104 templates [3.8 copies cell−1 × 2 g × 108 cells g−1 (5 μl/50 μl)(2.5 × 10−5) and 3.8 copies cell−1 × 2 g × 108 cells g−1 (5 μl/50 μl)(3.5 × 10−4)] (Figure 1). This number should be sufficient to reproducibly allow for amplification of template DNA (Handt et al. 1996). However, in the previous study we were unable to amplify bacterial 16S rDNA fragments of 120 bp from DNA extracts older than 400,000–600,000 years (Willerslev et al. 2004a), suggesting that processes other than SSBs are involved.
It is more complicated to use the rate constants for interstrand crosslinks to make similar calculations. The PicoGreen assay does not provide exact information on how single base pairs behave. The affinity for PicoGreen of double-stranded DNA is probably increased some distance from a crosslink. Even if we assume that each crosslink results in increased affinity up to 5 bp during denaturation, crosslinks would reduce amplifiability much more rapidly than single-stranded breaks. Accordingly, amplifiable 120-bp bacterial 16S rDNA should not be present after ∼400,000 years (Figure 1). This estimate is in agreement with our previous results showing independent and reproducible amplification of 120-bp 16S bacterial rDNA from the permafrost samples dated to be up to 400,000–600,000 years old (the same samples as used in this study), but not from samples dated to be 1.5–2 million years old (Willerslev et al. 2004a). These simple calculations additionally support the view that ICLs rather than SSBs are the most important damage type prohibiting amplification of DNA extracted from the permafrost samples.
Damage/age relationships and DNA leaching:
Although theory predicts an increase in DNA damage with time (Hebsgaard et al. 2005), previous studies of diagenesis in fossil remains have not been able to confirm this (Pääbo 1989; Höss et al. 1996; Gilbert et al. 2003). A likely explanation is that preservation conditions rather than the age of the specimens determine the degree of DNA degradation (Willerslev et al. 2004b). Interestingly, this is the first time that DNA damage can be clearly linked to sample age. This is true for strand breaks and ICLs (Table 1). Additionally, statistical analysis showed a significant increase (ANOVA, P = 0.025) in incorporation of radioactive phosphates (i.e., the PNK end-labeling assays after alkaline or SAP treatments) with sample age, as did the Klenow fill-in of radioactive nucleotides (Wilcoxon matched pairs, P < 0.05). These observations are in agreement with the geological interpretation that Siberian permafrost preservation conditions must have remained constant over time, making permafrost unique for comparative studies of DNA damage rates. Furthermore, these observations provide further support of the authenticity of the results, as contaminated samples would not be expected to produce such an age-specific relationship. Such a relationship between damage and age is unlikely to be found in most other environments due to variable preservation conditions. Importantly, the observed damage/age relationship reveals that leaching of DNA between sediment layers must be minor at most. The sections under study do not constitute a continuous layering of sediments due to, among other things, tectonic uplift and erosion. Therefore, leaching of DNA between layers would disturb any clear pattern between the sediment age and damage of the DNA. The results therefore contradict previously raised concerns about the leaching of free DNA molecules between frozen layers that would severely complicate any paleoenvironmental reconstructions based on permafrost-preserved sedimentary DNA (Pääbo et al. 2004).
Ancient DNA repair:
Limited success of aDNA repair has been achieved using DNA polymerase I (E. coli), and T4 DNA ligase (Pääbo 1989; Pusch et al. 1998). Our results confirm that this approach will probably repair some of the gaps and nicks in the ancient sediment DNA. However, as a major part of the 5′ termini do not have PO4 groups, and as this method requires the presence of 3′-OH and 5′-PO4 termini, the amount of repairable ancient templates must be limited. Furthermore, because SSBs in the aDNA most likely result from depurination, which consequently results in 3′ termini with aldehyde groups, gap filling by DNA polymerase will be inhibited. To repair the majority of SSBs in aDNA we propose a pretreatment with PNK and AP endonuclease I, which will assure that most 5′ and 3′ ends have PO4 and OH groups, respectively. Subsequent to this initial enzymatic treatment, gap filling by a polymerase with a low strand displacement capability and 3′ → 5′ exonuclease activity (e.g., E. coli Pol II), which can remove carbon radicals from the 5′ end, and gap filling by T4 DNA ligase should greatly increase the yield of amplifiable aDNA fragments. A note of caution, however, is that applying enzymatic assays like the ones described to ancient DNA is very difficult. Many of the repair techniques need to be optimized to very low amounts of DNA. Issues like contamination with modern templates during the experimental procedure need to be dealt with efficiently if the end product of the repaired amplified sequence is to be reliable and credible. Contaminating DNA molecules can enter the procedure at different levels, so appropriate controls and criteria need to be developed (Willerslev and Cooper 2005).
We thank Per A. Pedersen and Simon Y. W. Ho from The August Krogh Institute for ideas and use of the laboratory and Karen Vasquez, M. D. Anderson Cancer Center, for control sequences. E.W., J.B., and T.B. were supported by the The Danish Natural Science Research Council, The Carlsberg Foundation, and The Wellcome Trust. R.R. was funded by a grant from The Danish National Science Research Council. L.P. and D.M. were supported by the U.S. National Science Foundation grant no. OPP-0085589. D.A.G. was supported by the Russian Foundation for Basic Research, grant 01-05-65043.
↵1 These authors contributed equally to this work.
Communicating editor: N. Takahata
- Received February 16, 2006.
- Accepted March 25, 2006.
- Copyright © 2006 by the Genetics Society of America