Ancient DNA

Abstract

In the past two decades, ancient DNA research has progressed from the retrieval of small fragments of mitochondrial DNA from a few late Holocene specimens, to large-scale studies of ancient populations, phenotypically important nuclear loci, and even whole mitochondrial genome sequences of extinct species. However, the field is still regularly marred by erroneous reports, which underestimate the extent of contamination within laboratories and samples themselves. An improved understanding of these processes and the effects of damage on ancient DNA templates has started to provide a more robust basis for research. Recent methodological advances have included the characterization of Pleistocene mammal populations and discoveries of DNA preserved in ancient sediments. Increasingly, ancient genetic information is providing a unique means to test assumptions used in evolutionary and population genetics studies to reconstruct the past. Initial results have revealed surprisingly complex population histories, and indicate that modern phylogeographic studies may give misleading impressions about even the recent evolutionary past. With the advent and uptake of appropriate methodologies, ancient DNA is now positioned to become a powerful tool in biological research and is also evolving new and unexpected uses, such as in the search for extinct or extant life in the deep biosphere and on other planets.

1. Introduction

The evolutionary processes that generated modern species and populations are commonly inferred through the analysis of morphological and genetic markers, along with studies of abundance and distribution patterns (e.g. Avise 2000). However, the analysis of contemporary organisms can provide only indirect evidence of this history, and reconstructions remain tentative if they cannot be checked against the fossil record. While this has long been possible for morphological characters, ancient DNA (aDNA) methods are now providing a means to also record genetic changes in real time, at least over short geological time-scales.

The initial aDNA publications are just two decades old, and the methods and standards of this new field are still evolving rapidly. To review the current position and future directions of aDNA research, we discuss the colourful history of early publications, DNA degradation and long-term survival, contamination and authentication criteria, and the implications of recent advances.

2. History

The first aDNA studies used bacterial cloning to amplify small sequences retrieved from skins of animal and human mummies, and revealed the inefficient reaction kinetics of this technique (Higuchi et al. 1984; Pääbo 1985, 1989). These studies demonstrated that the genetic material surviving in ancient specimens was often principally microbial or fungal in origin, and that endogenous DNA was generally limited to very low concentrations of short, damaged fragments of multi-copy loci such as mitochondrial DNA (mtDNA). The invention of the polymerase chain reaction (PCR) made it possible to routinely amplify and study even single surviving molecules, allowing the number and range of aDNA studies to diversify rapidly (Pääbo 1989; Pääbo & Wilson 1988; Pääbo et al. 1989; Thomas et al. 1989). However, the enormous amplifying power of PCR also created an increased sensitivity to contamination from modern DNA, and simultaneously, a major potential source of such contamination through the extraordinary concentrations of previously amplified PCR products. As a consequence, false positives resulting from intra-laboratory contamination remain a major problem in aDNA research. In fact, the large number of PCR cycles used to amplify aDNA means that it is actually quite difficult not to obtain positive (although not authentic) results. Perhaps unsurprisingly, many of the most extravagant aDNA reports have since been either disproved or effectively disregarded. This includes early spectacular claims of DNA sequences surviving for millions of years (Myr) in plants (Golenberg et al. 1990; Soltis et al. 1992, although see Kim et al. 2004), dinosaur bones (Woodward et al. 1994) and amber inclusions (Cano et al. 1992a,b, 1993; DeSalle et al. 1992, 1993; Poinar et al. 1993; DeSalle 1994). Some of these sequences originated from obvious human or microbial contamination (Zischler et al. 1995a; Gutierréz & Marn 1998), whereas it has not proved possible to repeat others independently (Sidow et al. 1991; Austin et al. 1997a,b). Many other claims remain in limbo, where a lack of appropriate methods or replication renders them effectively meaningless, e.g. human sequences from ‘Mungo man’ in Australia, or Cheddar Gorge in the UK (Adcock et al. 2001; Cooper et al. 2001a) and the recovery of bacterial DNA and cells from amber and halite that are claimed to be many millions of years old (Cano & Borucki 1995; Vreeland et al. 2000; Fish et al. 2002; Willerslev et al. 2004a).

Despite this somewhat tarnished history, recent advances in knowledge about the tempo and mode of DNA template damage, sample contamination and biochemical diagenesis have improved standards, and aDNA is now emerging as a viable scientific discipline. A series of large-scale studies have begun to reveal the true potential of aDNA to record the methods and processes of evolution, providing a unique way to test models and assumptions commonly used to reconstruct patterns of evolution, population genetics and palaeoecological change.

3. DNA degradation

3.1 Damage and repair

The post-mortem instability of nucleic acids is central to the methodological problems inherent in aDNA research. In metabolically active tissues, damage to the DNA molecules is rapidly and efficiently repaired via a host of repair pathways (Lindahl 1993). By contrast, DNA damage accrues over time in inactive (dead or dormant) cells owing to factors such as spontaneous hydrolysis and oxidation, and consequently most ancient specimens do not contain any amplifiable endogenous DNA, while those that do possess only fragments in the 100–500 bp size range (Pääbo 1989; Handt et al. 1994; Höss et al. 1996). Early research showed that post-mortem DNA decay was characterized by strand breaks, baseless sites, miscoding lesions and crosslinks, and that these caused both sequencing artefacts and the preferential amplification of undamaged contaminant DNA (Pääbo 1989; Lindahl 1993; Höss et al. 1996; figure 1). However, the relative rates of various types of DNA damage and their mode of accumulation remain poorly characterized (Gilbert et al. 2003a,b).

Figure 1.

Post-mortem DNA modification in fossil remains, with the structures altered by damage shown in red. (a) Formation of strand breaks (single-stranded nicks) by hydrolytic damage. (i) Direct cleavage of the phosphodiester backbone (A). (ii) Depurination resulting in a baseless site (AP-site) (B) followed by breakage of the sugar backbone through β-elimination (C). Strand breaks are believed to be largely responsible for the short amplification length and the high rate of DNA loss in fossil remains. (b) Different types of crosslink formation. (i) Inter-strand crosslinks by alkylation. (ii) Intermolecular crosslinks by Maillard reaction. Crosslinks may prevent the amplification of endogenous template molecules, increasing the risk of contamination. (c) Oxidative and hydrolytic modification of bases resulting in (i) blocking lesions or (ii) miscoding lesions. Some oxidative damage results in lesions blocking the polymerase enzyme, and promoting chimeric sequences through ‘jumping PCR’. Hydrolytic damage of bases may result in miscoding lesions, for example, deamination of cytosine and adenine to uracil and hypoxathine, respectively. These lesions will result in the incorporation of erroneous bases during amplification.

aDNA amplification products are commonly characterized by sequencing artefacts, with a general bias toward CG→TA transitions and to a lesser extent AT→GC (Hansen et al. 2001; Hofreiter et al. 2001a; Gilbert et al. 2003a,b). The high number of CG→TA events appears to be owing to a high rate of hydrolytic deamination of cytosine (and its homologue 5-methyl cytosine) to uracil and thymine (figure 1c). The nomenclature is complicated by the complementary nature of DNA since an observed sequence change may have originated on either strand and so both possible paths are shown (e.g. C→T, or G→A on the opposing strand). However, post-mortem biochemical reaction pathways may be sufficiently limited that the original mutation event and strand can be identified, providing a means for sophisticated studies of post-mortem DNA decay and the detection of ‘jumping PCR’ artefacts (Gilbert et al. 2003b; figure 1c). A further important consideration is that miscoding lesions are not randomly distributed across the mtDNA genome as previously thought, but are concentrated in ‘hot spots’ where repeated hits occur. Intriguingly, the distribution of these regularly damaged sites in humans and bovids is very similar to those observed in regular evolutionary substitutions, meaning that aDNA damage can generate sequence artefacts that mimic expected evolutionary changes (Gilbert et al. 2003a; M. T. P. Gilbert, unpublished results). Recent studies of permafrost-preserved DNA indicate that inter-strand crosslinks, which prevent amplification, may accrue more quickly post-mortem than the single-stranded nicks that are largely responsible for fragmentation (figure 1a,b). Therefore, DNA sequences may be present in fossil remains long after negative amplification results are obtained (Geigl 2002; Willerslev et al. 2004a).

Several methods have been used to overcome post-mortem modifications and increase the quantity and quality of amplifiable DNA templates and sequence reliability. Uracil-N-glycosylase (UNG) removes deamination products of cytosine and is an important means to test the origins of sequence variation (Hofreiter et al. 2001a; Gilbert et al. 2003a; figure 1c). N-phenacylthiazolium bromide (PTB) appears to break intermolecular cross-links caused by advanced glycosylation end products, although the exact mode of operation remains unclear (Poinar et al. 1998; figure 1b). Similarly, high fidelity polymerase enzymes such as Pfu and Taq HiFi minimize sequence error rates and increase amplification efficiency (Willerslev et al. 1999; Cooper et al. 2001b). However, the tempo and mode of post-mortem damage pathways remain too poorly understood to design effective in vitro enzymatic repair treatments for aDNA (Di Bernardo et al. 2002).

3.2 Long-term survival

It is clear that constant low temperatures play a central role in the longevity of aDNA molecules (Lindahl 1993; Hofreiter et al. 2001b; Smith et al. 2001; Willerslev et al. 2004b). The oldest authenticated aDNA reports are all from permafrost settings, including greater than 50 kyr (thousand year)-old mammoth mtDNA (Höss et al. 1994), a greater than 65 kyr-old bison mtDNA (Gilbert et al. 2004a; Shapiro et al. 2004) and 300–400 kyr-old plant chloroplast DNA (cpDNA) and 400–600 kyr-old bacterial sequences (Willerslev et al. 2003a, b, 2004a; figure 2). In addition, Holocene and Pleistocene permafrost-preserved bones have permitted amplification products in the 900–1000 bp size range (Barnes et al. 2002; Lambert et al. 2001). Other features, such as rapid desiccation and high salt concentrations, may also prolong DNA survival (Lindahl 1993). However, kinetic calculations predict that small fragments of DNA (100–500 bp) will survive for no more than 10 kyr in temperate regions and for a maximum of 100 kyr at colder latitudes owing to hydrolytic damage (Poinar et al. 1996; Smith et al. 2001). Even under ideal conditions, amplifiable DNA is not thought to survive for longer than 1 Myr.

Figure 2.

Key aDNA studies through time. Reports are shown with claimed age, date of publication, type of sequence and whether the results were authenticated by independent replication.

The oldest DNA sequences may well exist in polar icecaps, where constant temperatures of as low as −50°C and samples of more than 800 kyr old (e.g. Dom C, Antarctica) hold much promise. There have been several reports about ancient plant, microbial and viral DNA from ice core samples of up to 100 kyr old (Ma et al. 1999; Priscu et al. 1999; Willerslev et al. 1999; DePriest et al. 2000; Christner et al. 2001; Hansen & Willerslev 2002), although authentication standards, such as the independent replication of results, have varied considerably.

4. Contamination and authentication of results

The criteria for authenticating aDNA results are continually evolving as new materials (e.g. microbial DNA and cultures) are studied (Handt et al. 1994; Rollo & Marota 1999; Austin et al. 1997b; Cooper & Poinar 2001; Hofreiter et al. 2001b; Willerslev et al. 2004b). Although critical steps such as the cloning and independent replication of results have been widely accepted, reports are still being published in high-profile journals without these basic authentication procedures. It is a matter of concern that this includes studies of both ancient human and Neandertal sequences where contamination risks are pronounced (Adcock et al. 2001; Serre et al. 2004a). Significantly, this is also the case especially for all reports of DNA older than the theoretical limit of survival (0.1–1 Myr; figure 2). Table 1 and figure 3.) list the rigorous criteria and null hypotheses, in methodological order, required to authenticate aDNA results. However, it is important to note that even when all of these criteria have been passed the results may potentially still not be authentic.

Table 1.

The analysis of nucleic acids (and cultures) from fossil remains should address the following criteria (see § 4 for details).

(Appropriate laboratory facilities and controls, independent replication and cloning of amplification products are essential and should not be compromised (after Handt et al. 1994; Cooper & Poinar 2001; Hofreiter et al. 2001b; Gilbert et al. 2003a,b; Willerslev et al. 2004b). numts, nuclear insertions.)

criteria of authenticity

physically isolated pre-PCR facility

nightly UV irradiation of surfaces, isolated ventilation, flow hoods or glove boxes, daily cleaning with bleach and movement only from pre-PCR to post-PCR buildings/areas

blank controls

extraction and PCR controls should be performed at a ratio of at least 1:5 and 1:1 with samples, respectively; similar ratios are required for culturing experiments

independent reproducibility

a subset of results should be verified by independent replication in another laboratory to rule out intra-laboratory contamination; critical for key results

cloning and sequencing

a subset of PCR products should be cloned to assess damage, contamination, and to detect nuclear insertions (numts); repeat extractions and amplifications should also be assessed

time-dependent pattern of damage and diversity

provides strong support for antiquity in microbial studies

decontamination of reagents and specimens

bleach, acid, ultra-filtration, baking and UV irradiation of reagents and tools should be routine; particularly important in ancient human and microbial studies. The surface of samples should be removed.

DNA from associated remains

can provide strong circumstantial support in ancient human and microbial studies

appropriate molecular behaviour

an inverse relationship between amplification length and strength support authenticity; expected copy number relationship (e.g. between mtDNA, chloroplast DNA (cpDNA) and nuclear DNA (nuDNA)

uracil-N-glycosylase (UNG) treatment

UNG removes deamination products of cytosine, and is particularly important in studies where results involve few substitution differences

quantification of starting templates

particularly important in studies where interpretation is based on few substitution differences, or few specimens

biochemical preservation

measures of diagenesis can indicate whether DNA survival is likely

Figure 3.

The sequential series of null hypotheses necessary for authenticating aDNA sequences, starting from the PCR reaction. Green boxes show the procedures, blue boxes represent an evaluation of results. Yellow boxes are considerations that need to be taken into account, and grey boxes are suggestions for additional procedures. Green and red arrows indicate a positive and negative result respectively. Blue arrows are suggestions for further action.

4.1 Contamination

A major difficulty in detecting and preventing contamination is that the scale of the problems involved is not easily appreciated. For example, a successful PCR reaction can contain some 10¹²–10¹⁵ amplified molecules in a volume of less than 50μl, which is too large a concentration to comprehend effectively (Kwok & Higuchi 1989). Air movement created when opening PCR tubes or transferring liquids will create and disperse microscopic aerosol droplets, which can easily contain over a million copies of the template per 0.005μl. As a consequence, PCR products can quickly become widely distributed across laboratory surfaces, corridors and through entire buildings via personnel movement and air-handling systems. Since one aerosol droplet can easily contain a thousand times the amount of amplifiable mtDNA found in 1 g of many ancient human specimens (10⁵–10⁶ copies; Handt et al. 1996; Cooper et al. 2001b), aDNA laboratories must be completely isolated both physically and logistically, preferably in buildings free from all molecular biological research. Furthermore, daily personnel movement should only proceed from ancient to modern laboratories, i.e. up the concentration gradient. Such simple precautions can prove as effective as high-tech positive air-pressure and UV irradiation systems, if rigorously enforced.

Human and microbial DNA and cells are ubiquitous in all laboratory settings. It is prudent to assume that all laboratory reagents and tools are contaminated with human and microbial DNA when arriving from the manufacturer. Extensive cleaning of reagents (e.g. ultrafiltration) and tools is essential, with complete decontamination requiring prolonged exposure (e.g. UV irradiation (45 W, 72 h), baking (more than 180°C, 12 h), acid (2.5 M HCl, 48 h) and/or sodium hypochlorite (5%, 48 h)). Laboratory reagents and commercial equipment marked ‘sterile’ are not guaranteed to be free of either viable cells or nucleic acids. Similarly, autoclaving does not prevent the amplification of short DNA fragments (less than or equal to 150 bp), and often contaminates material with bacterial DNA. These issues are critically important in ancient human and microbial studies, but contamination often cannot be completely ruled out (Willerslev et al. 2004b; M. T. P. Gilbert, unpublished data).

The most intractable problems occur when the sample itself has been contaminated prior to analysis. This issue is of major significance with archaeological material, where specimens have been handled, and often washed, during excavation by a variety of individuals who may have DNA markers close to or even identical to the specimen DNA (Cooper 1997; Serre et al. 2004a). Similarly, the passive or active movement of cells in ancient microbial studies makes it difficult or impossible to completely exclude recolonization of ancient materials over time.

4.2 Detecting contamination

Over the past two decades a simple set of criteria has been developed to detect laboratory-based contamination (Pääbo 1989; Cooper & Poinar 2001). This includes blank DNA extraction and PCR controls, which need to be performed at high ratios in aDNA experiments (e.g. 1:5 and 1:1 with normal extracts, respectively). It is important to note that negative blank controls may easily conceal low-level laboratory contamination owing to poorly understood carrier effects (Cooper 1993; Handt et al. 1994). This commonly overlooked problem and the high risk of PCR-based contamination demonstrate why independent replication in a separate laboratory is so crucial (Cooper & Poinar 2001; Willerslev et al. 2004b). In general, the use of positive controls should be avoided owing to the contamination risk that they pose.

The independent replication of results by another laboratory is currently the strongest argument against laboratory-based contamination (whether from tools, reagents, researchers or PCR products) because it is highly unlikely that the same erroneous sequence will be independently obtained twice. The routine independent replication of results also maintains the highest standards of experimental practice in participating laboratories, and this psychological factor should not be undervalued in such contamination-prone research. In our opinion, the history of erroneous aDNA publications illustrates the need for all ongoing work to feature the independent replication of a subset of samples (e.g. 10%, or all key results), even when unsurprising results are obtained. Importantly, it is necessary not just to perform, but also to clearly demonstrate that the research is of the highest possible standards.

Sample contamination is often more difficult to detect than laboratory-based contamination, and is a pervasive problem in studies of ancient human material (Serre et al. 2004b). As many museum specimens have been handled for decades, contaminant molecules may also appear ancient and exhibit appropriate molecular behaviour, i.e. an inverse relationship between amplification efficiency and fragment length (Handt et al. 1994). For archaeological remains, reproducible amplification of the same sequence by independent laboratories using different parts of a specimen, for example, tooth and femur, would provide good support. Standard tests, such as phylogenetic sense (Cooper & Poinar 2001; Hofreiter et al. 2001b), can be weak in such situations (Cooper 1997) and are generally not useful in studies of ancient microbes where novel sequences are the norm, owing to the lack of knowledge about extant microbial diversity (Willerslev et al. 2004b).

Quantification of template copy number by real time PCR is useful when a single sequence is amplified (Handt et al. 1996), but less so when ‘universal’ primers are used to obtain sequences from a variety of taxa (e.g. many microbial or plant studies). Results obtained from amplifications starting from less than 1000 template molecules have proved difficult to reproduce and can be biased by damage-induced misincorporation of bases (Handt et al. 1996; Gilbert et al. 2003a). Quantification also does not reveal whether some or all of the starting molecules are contaminants.

4.3 Nuclear insertions

Insertions of mtDNA and cpDNA sequences in nuclear chromosomes are of special concern in aDNA studies. It has not been possible to purify organellular DNA from fossil remains, and nuclear insertions (numts) are generally not easily distinguishable from damaged or even authentic mtDNA sequences. The chances of detecting numts can be maximized by amplifying overlapping fragments and cloning PCR products. If mistaken for bona fide organelle sequences, numts can seriously confound phylogenetic and population genetic studies. For instance, there is an example of a reported dinosaur DNA sequence that was later found to be a human numt (Zischler et al. 1995a). This problem is intensified as all parts of the human mtDNA are found in the nuclear genome, sometimes as fragments up to 14654 bp in length, and numbering more than 40 copies (Mourier et al. 2001). Numts have been reported from mammoth (Greenwood et al. 1999), moa (Cooper et al. 2001b) and rhino (Orlando et al. 2003) and should be taken into account when amplifying mtDNA and cpDNA sequences (figure 3; table 1). If fully characterized, numts may also be used as ‘molecular fossils’ to improve phylogenetic resolution (Zischler et al. 1995b).

4.4 Predicting DNA survival

The exaggerated risk of false positive results in aDNA studies has stimulated attempts to predict DNA survival in fossil remains prior to genetic analysis. For example, the extent of amino acid racemization of the l to d-enantiomers of aspartic amino residues has been related to DNA depurination (Poinar et al. 1996; figure 1a). Flash pyrolysis with gas chromatography/mass spectrometry (GC/MS) suggests that samples containing amplifiable DNA sequences have abundant pyrolysis products assigned to 2,5-diketopiperazines of proline-containing dipeptides (Poinar & Stankiewicz 1999). Studies of ‘thermal age’, i.e. the number of years required for a certain degree of DNA degradation (by depurination) at a constant temperature of 10°C, indicate that sites with thermal ages of below 17 kyr may contain specimens with amplifiable DNA (Smith et al. 2001).

Although non-PCR-based methods of addressing DNA survival are needed, especially when dealing with ancient human DNA and micro-organisms, they are currently not reliable. Recent reports imply that depurination kinetics will not be similar to racemization in all tissues (Collins et al. 2002). For example, the presence of aspartyl and asparagine residues (Geiger & Clarke 1987), actual amino acid sequence and the rate of denaturation (Collins et al. 1999) all influence aspartic acid racemization without affecting depurination. Furthermore, the exact thershold values for amino acid racemization associated with DNA survival continues to evolve (Poinar et al. 1996; Schmitz et al. 2002; Serre et al. 2004a), and the original Neanderthal-DNA-containing specimen (Krings et al. 1997) falls outside this guideline, confirming that the method has limitations. Additionally, depurination is not the only significant form of damage determining the survival of amplifiable DNA sequences. Recent studies have shown that intermolecular and inter-strand cross-links may be more significant (Poinar et al. 1998; Willerslev et al. 2004a; figure 1b). Additionally, thermal age prediction has not proved useful in dry conditions (Poinar et al. 2003), although the development of approaches to include rate kinetics of damage processes other than hydrolytic deamination promise to make the method useful for predicting DNA survival across a range of sites. Lastly, there is evidence to suggest that microenvironments may offer very different preservation potentials within samples, limiting the predictive power of these methods. As a consequence, simple methods such as the amplification of DNA from associated remains (e.g. plants, animals) offer an alternative means to predict the survival of DNA in ancient human and microbial specimens.

5. Ancient animal and plant genetics

5.1 Population genetics

aDNA holds tremendous potential for the study of ancient animal and plant populations. Studies on brown bears, penguins, cave bears, horses, dogs and bison (Leonard et al. 2002; Loreille et al. 2001; Vila et al. 2001; Barnes et al. 2002; Hofreiter et al. 2002, 2004; Leonard et al. 2002; Ritchie et al. 2004; Shapiro et al. 2004) have shown that aDNA can reveal population movements and local extinctions back into the Late Pleistocene. Such studies have considerable power to examine the effects of climate change (e.g. around the Last Glacial Maximum) and to test theories and develop methods used in population genetics and palaeobiology. For example, aDNA studies of Beringian brown bears revealed surprisingly large amounts of haplogroup extinction and replacement during the Late Pleistocene and Holocene, and very little interchange of females between populations (Barnes et al. 2002). Similarly, recent large-scale studies of North American and Eurasian bison mtDNA sequences reveal how fluctuations in genetic diversity over the past 150 kyr are closely linked to climate change (Shapiro et al. 2004). Interestingly, the initial decline in steppe bison genetic diversity started more than 20 kyr earlier than signs of significant early human presence in North America and therefore cannot be linked to human overkill as previously suggested (Martin 1984).

Studies of ancient populations have repeatedly shown that modern patterns of distribution and diversity are often owing to relatively recent events, and can give very misleading views of past evolutionary processes (Cooper et al. 1996; Barnes et al. 2002; Goldstein & DeSalle 2003; Shapiro et al. 2004). This can have major conservation implications, such as the legal ability to reintroduce extirpated taxa (e.g. Laysan duck, Anas laysanensis), or the apparent creation of new conservation units or species when formerly clinal distributions are disrupted (e.g. northeastern beach tiger beetle, Cicindela dorsalis dorsalis). Conversely, studies of several ancient populations of large vertebrates have shown extensive gene flow between taxa recognized as different species or subspecies, suggesting that the morphological plasticity in some fossil groups may have been considerably underestimated (Bunce et al. 2003; Huynen et al. 2003; Shapiro et al. 2004).

The combination of ancient sequences and coalescent methods has considerable power to reconstruct detailed demographic histories, test models of population genetics and reveal much novel information about microevolutionary processes. These methods can even recover demographic data for taxa that have been through recent population bottlenecks, which would normally remove genetic signals (Shapiro et al. 2004). Such methods also provide an opportunity to directly estimate evolutionary rates of nucleotide substitution and directly date phylogenetic events without using an external palaeontological calibration for a molecular clock (Lambert et al. 2001; Ritchie et al. 2004; Shapiro et al. 2004). The results show that palaeontologically calibrated rate estimates are often significantly slower than those calibrated from aDNA population studies and may reflect differences in sequence substitution processes operating at short and long time-frames.

5.2 Nuclear DNA

Sequences from multicopy loci are easier to recover from fossil remains than single-copy sequences, probably owing to the presence of larger numbers per cell. Consequently, there has been an almost complete reliance on mtDNA and cpDNA sequences in aDNA research. However, these markers may not reflect the overall history of the genome when closely related species or population genetic questions are addressed (Hofreiter et al. 2001b). Recent results show that multi- and single-copy nuclear DNA (nuDNA) sequences can also be obtained from fossil remains, such as Pleistocene sloth coprolites from Nevada (Poinar et al. 2003). This study used single-copy nuDNA sequences to resolve the phylogenetic position of the extinct Shasta ground sloth (Nothrotheriops shastensis), and demonstrated that Pleistocene DNA survival was also possible in specimens preserved in warm arid climates and not just permafrost-preserved fossils (Greenwood et al. 1999, 2001).

Plant aDNA sequences have been retrieved from samples such as seeds (Rollo & Amici 1991; O’Donoghue et al. 1996; Brown 1999; Banerjee & Brown 2002) and ancient maize cobs (Jaenicke-Després et al. 2003). Interestingly, the maize study recorded how variants were selected at three nuDNA genes that effect architecture, storage protein synthesis and starch production. The results show that these genes were largely homogenized as early as 4.4 kyr ago by Mexican farmers, although one of the loci was not fixed until ca. 2 kyr ago. Encouragingly, this study indicates that aDNA studies of selection will become increasingly possible as phenotypically important loci are identified by genome projects.

5.3 Organellular genomics

The majority of aDNA research has relied on relatively short sequences (ca. 1 kb) obtained through the amplification of numerous small overlapping fragments (ca. 100–500 bp). However, the field was recently brought into the genome era when the sequences of two complete mtDNA genomes (ca. 16.5 kb) were determined for extinct species of New Zealand moa (Cooper et al. 2001b). Partial mitochondrial genome sequences of two other moa were also generated (Haddrath & Baker 2001), demonstrating that it is possible to reconstruct long genetic sequences from fossil remains using numerous short amplifications. Such large amounts of sequence data are desirable as they permit detailed phylogenetic analyses and statistically precise estimates of molecular divergence dates.

It is important to note that even if it becomes possible to reconstruct ancient nuDNA sequences, it would still be impossible to bring extinct organisms back to life. Among many other requirements, the cloning of complex organisms needs a complete and undamaged nuDNA genome, packaged correctly, and a compatible maternal host. The characteristic damage and extensive fragmentation of aDNA make this unfeasible.

5.4 Dirty DNA

The discovery that diverse vertebrate mtDNA and plant cpDNA sequences may be preserved in permafrost and cave sediments has opened the way for aDNA analyses of biological change at the ecosystem level (Hofreiter et al. 2003; Willerslev et al. 2003a, b).

It has been possible to recover DNA from small samples (ca. 2 g) of sediments up to 300–400 kyr old, including sequences of herbs, shrubs, trees, mosses, and megafauna such as mammoth, bison and horse (Willerslev et al. 2003a, b). Sedimentary DNA offers the potential to link fauna and flora directly in space and time over broad areas, and could provide an important complement to pollen records which are complicated by variations in influx rates, long-distance dispersal and vegetative plant growth (Guthrie 1990; Anderson et al. 1994). Angiosperm pollen generally does not contain cpDNA (Shi-yi 1997), so the majority of the plant cpDNA sequences in sediment must originate from seeds and/or somatic tissues such as fine rootlets (Willerslev et al. 2003a, b). The origin of the animal DNA sequences in sediments seems to be waste products and potentially nails and skin cells. (Lydolph et al. 2004). Although the potential of sedimentary DNA is remarkable, important concerns include the effects of water movement on DNA mobility in non-frozen areas, and possible biases in the deposition and survival rates of different taxa. Encouragingly, free DNA molecules in the soil have been shown to bind tightly to the edges of humic acids, clays, feldspar and quartz, which limits their movement and degradation by endonucleases (Lorenz & Wackernagel 1994).

6. Ancient human DNA: a contentious issue

While animal and plant aDNA studies have seen rapid progress, contamination issues have undermined promises that such research would also revolutionize bioarchaeology (Herrmann & Hummel 1996). Early successful studies using material from hot environments such as Egypt (Pääbo 1985) and Florida (Hauswirth et al. 1994) are now recognized as probable contaminants. For example, the Egyptian mummy sequence was a very large (3.4 kb) fragment of nuDNA, which is highly unusual, and was recovered from a region where temperatures make DNA survival very unlikely (Marota et al. 2002; Gilbert et al. 2004b).

Several reports show that despite rigorous protocols (Cooper & Poinar 2001; Hofreiter et al. 2001b), modern human contamination is widespread in amplification products from ancient extracts (Richards et al. 1995; Handt et al. 1996; Krings et al. 1997; Kolman & Tuross 2000; Hofreiter et al. 2001b). It also appears impossible to clean contemporary human DNA from human bones and teeth despite extensive treatment with UV irradiation and bleach (M. T. P. Gilbert, unpublished data). It appears that the porosity of bone and dentine in teeth are the main entry routes for DNA generated from sweat, skin fragments and exhaled cells, reinforcing the urgent need for disposable gloves and face-masks during excavation and handling of archaeological specimens. Interestingly, hair may be a more reliable source for ancient human DNA studies, as it appears less susceptible to contamination than bone and teeth (Gilbert et al. 2004a).

Many excavated archaeological remains appear to contain DNA from multiple individuals (Gilbert et al. 2003a,b), raising the issue of how to authenticate ancient human DNA when ‘unique’ sequences, such as the Neanderthal (Krings et al. 1997) or distinct modern human groups like the Andaman Islanders (Endicott et al. 2003), are not reproducibly obtained. A good example is the analysis of Italian Cro-Magnon specimens (Caramelli et al. 2003), where comprehensive protocols of authentication (Cooper & Poinar 2001; Hofreiter et al. 2001b) were followed. However, because the resulting sequences were indistinguishable from modern Europeans, sample contamination must remain the null hypothesis. By contrast, Serre et al. (2004a) assume that it is impossible to authenticate any modern human sequence obtained from archaeological specimens and instead confirm the absence of Neanderthal-specific mtDNA sequences from five European early modern human (EMH) specimens. Since coalescence theory indicates that the (inferred) modern human mtDNA sequences of the five EMH specimens are unlikely to exactly match the 5–7 ancestral lineages of modern populations, this effectively doubles the number of modern human mtDNA lineages known to exist in the Late Pleistocene. This value was used with population genetic models to calculate that the maximum Neanderthal genetic contribution to EMH is likely to have been less than 25% (Serre et al. 2004a). Although not independently replicated, this study demonstrates how aDNA can dramatically increase the resolving power of population genetics studies (Cooper et al. 2004).

The retrieval of putative Neanderthal mtDNA sequences (Krings et al. 1997, 1999, 2000; Ovchinnikov et al. 2000; Schmitz et al. 2002) is the major highlight in ancient human DNA studies because it allowed direct testing of hypotheses about the origin of the modern human gene pool. Importantly, recent suggestions that ancient sequences such as the Neanderthal results might be due to PCR artefacts (Pusch & Bachmann 2004) appear unjustified, and may result from poor experimental design and methodology (Serre et al. 2004b). Until recently there seemed little hope for obtaining DNA sequences from other extinct hominids (e.g. Homo erectus). However, the recent discovery of the ‘Hobbit’-sized Homo floresiensis on the island of Flores (Indonesia) dated to be just 18 kyr old (Brown et al. 2004; Morwood et al. 2004) will potentially allow for DNA characterization. Otherwise, advances in protein-sequencing techniques and the stability of certain proteins (Nielsen-Marsh et al. 2002) may also provide a means for such comparisons, although the limited phylogenetic utility of short amino acid sequences constrains the resolving power possible.

A final complicating issue in ancient human mtDNA studies is the authentication of haplogroup designations. Mutational hot spots may generate erroneous, but potentially credible results when PCR reactions are initiated from small numbers of DNA molecules (Gilbert et al. 2003a,b). This is exacerbated in studies of human control region sequences, where haplogroup identifications are routinely categorized using fewer than five site changes, and in extreme cases (e.g. some European groups) by as few as one. In such situations, the use of real-time PCR and UNG treatment is highly advisable (table 1). A far more reliable approach is to characterize multiple variable positions around the mitochondrial genome to define a haplotype, as shown by Maca-Meyer et al. (2004) in a study of the colonization of the Canary Islands.

7. The microbial problem

Many studies have reported the survival of ancient microbial DNA from archaeological remains (e.g. Reid et al. 1999; Rollo et al. 1995, 2002; Rollo & Marota 1999; Cano et al. 2000; Raoult et al. 2000; Rothschild et al. 2001; Zink et al. 2001; Fletcher et al. 2003) but unfortunately very few have used independent replication to authenticate results. As a consequence there is some doubt over the validity of many studies, especially when research has been done in microbiology departments where the microbial DNA sequences might be expected as a local contaminant. It is a matter of concern that many of the ancient microbial sequences are similar or identical to modern strains, or come from environments unlikely to preserve DNA (Gilbert et al. 2004b). This is unfortunate, as the rapid evolutionary rates of many pathogens provide a unique opportunity to use phylogenetic analyses to verify the sequences as ancient, i.e. ancestral to modern diversity. Where independent replication has been attempted, such as studies of Yersinia pestis sequences in European plague victims (Raoult et al. 2000), the results have been negative (Gilbert et al. 2004c).

Reports of viable microbial cells and DNA from materials hundreds of millions of years old, such as fossil plant and animal remains, deep sediments, amber and halite (e.g. Kennedy et al. 1994; Cano & Borucki 1995; Shi et al. 1997; Vorobyova et al. 1997; Vreeland et al. 2000; Fish et al. 2002; Vreeland & Rosenzweig 2002; Willerslev et al. 2004a) also seem surprising. By contrast, a recent study of bacterial DNA under ideal deep-frozen conditions suggests a maximum survival time for amplifiable bacterial DNA of between 400 kyr and 1.5 Myr (Willerslev et al. 2004a), which is close to theoretical predictions (figure 2).

Claims regarding ancient micro-organisms are extremely difficult to authenticate because a novel sequence cannot be used as a criterion for authentication since only ca. 1–5% of the potential modern contaminants (i.e. extant microbial diversity) are thought to be known (Ward et al. 1992). Furthermore, like human DNA, microbial contamination is everywhere (Willerslev et al. 2004b) and recent results suggest that contemporary microbes have a global distribution and that endemics are rare, i.e. everything is everywhere (e.g. Fenchel et al. 1997; Finlay 2002). As a consequence, the independent replication of results will be a less rigorous criterion of authenticity in ancient microbial studies, and clear time-dependent patterns of diversity and DNA degradation may be needed to demonstrate authenticity (Willerslev et al. 2004b).

Despite these problems, nearly all of the geologically ancient microbial claims have not been checked by either the analysis of DNA preservation states, or more disconcertingly, by independent replication. For example, reports of bacterial DNA and viable cells surviving for hundreds of millions of years in halite were published in Nature (Vreeland et al. 2000; Fish et al. 2002) without verification of the result by real-time PCR, DNA damage data or independent replication. Several papers have commented on the exceptional similarity of these microbial DNA sequences to contemporary ones (Graur & Pupko 2001; Nickle et al. 2002) including gram-negative proteobacteria with fragile cell walls and no known adaptations to long-term DNA preservation. It is also notable that members of nearly all major microbial groups have been recovered from multi-million-year-old materials, despite possessing huge differences in hardiness and resistance to DNA degradation (Kennedy et al. 1994), strongly suggesting that contamination may be involved.

8. Mars: a possible new research area

During the coming decades, the search for evidence of extant or extinct life on other planets, especially Mars, will be a central focus of scientific research. Martian samples will provide an opportunity to test ideas such as Panspermia, which suggests that life may have travelled between planets. While contentious (Willerslev et al. 2004c), it is clear that cells such as bacterial endospores can survive considerable amounts of time in space (at least 6 years; Horneck et al. 1994) and the heat and pressure developed during meteorite escape and entry (Nicholson et al. 2000).

Any nucleic acids present on Mars have high chances of long-term survival owing to the extremely low temperatures found in areas such as the Martian permafrost and polar icecaps, which are believed to be millions or billions of years old. Calculations have shown that the Martian icecaps are likely to be in the −50°C to −110°C temperature range (Fisher 2000), which would theoretically allow a 100 bp bacterial DNA fragment to survive depurination for between ca. 3.4×10⁹ and 3.1×10²¹years (Willerslev et al. 2004b), which is well beyond what can be expected on Earth. Of course, any microbial-like DNA sequences obtained from Martian samples will face the same problems of authentication as those from fossil remains on Earth, and may indeed represent contamination carried by spacecraft. Approaches similar to those suggested for authenticating ancient bacterial claims (table 1) may be a starting point, but it will be extremely difficult because even the strictest protocols for authentication cannot completely rule out contamination, and novel sequences would not necessarily imply an endogenous origin.

For any form of new exploration, whether extraterrestrial, deep sea or icecap, it is critical that the severe contamination problems revealed by previous aDNA studies are taken into account when designing instruments and methods for the collection and analysis of nucleic acids (Hansen et al. 2002). In this way, the hard lessons learned over the past two decades will be useful for a completely different set of environments, and potentially even more valuable specimens.