REVIEWS

A P P L I C AT I O N S O F N E X T- G E N E R AT I O N S E Q U E N C I N G

Reconstructing ancient genomes

and epigenomes
Ludovic Orlando1,2, M. Thomas P. Gilbert1,3 and Eske Willerslev1
Research involving ancient DNA (aDNA) has experienced a true technological revolution
in recent years through advances in the recovery of aDNA and, particularly, through
applications of high-throughput sequencing. Formerly restricted to the analysis of only
limited amounts of genetic information, aDNA studies have now progressed to
whole-genome sequencing for an increasing number of ancient individuals and extinct
species, as well as to epigenomic characterization. Such advances have enabled the
sequencing of specimens of up to 1 million years old, which, owing to their extensive DNA
damage and contamination, were previously not amenable to genetic analyses. In this
Review, we discuss these varied technical challenges and solutions for sequencing
ancient genomes and epigenomes.

Osseous materials
Calcified animal tissues,
such as bones and teeth.
1
Centre for GeoGenetics,
Natural History Museum
of Denmark, University of
Copenhagen, Øster Voldgade
5–7, Copenhagen 1350C,
Denmark.
2
Université de Toulouse,
University Paul Sabatier
(UPS), Laboratoire AMIS,
CNRS UMR 5288, 37 allées
Jules Guesde, 31000
Toulouse, France.
3
Trace and Environmental
DNA Laboratory, Department
of Environment and
Agriculture, Curtin University,
Perth, Western Australia
6102, Australia.
Correspondence to L.O.
e‑mail: lorlando@snm.ku.dk
doi:10.1038/nrg3935
Published online 9 June 2015
The 1984 publication of a short mitochondrial DNA
sequence from the quagga, a zebra-like equid that has
been extinct since the 1880s, initiated the field of ancient
DNA (aDNA) research1. Following concomitant devel-
opment of PCR and realization that DNA survived in
osseous materials2, the future of aDNA research looked
bright. However, the degraded nature of aDNA3 coupled
with the sensitivity of PCR to contamination — whether
derived from environmental microorganisms or human
handling, and thus embedded in the samples, or in the
form of laboratory and/or reagent contamination — con-
tributed to a series of publications based on false-positive
results. Given that these problems seriously undermined
the field’s broader scientific interest and reliability
until the mid‑2000s, few would have expected that, by
the field’s twenty-fifth birthday, the genome of an ancient
human4 and draft genomes of the extinct mammoth5
and Neanderthals6 would have been sequenced. Today,
many tens of ancient genomes, ranging from microbial
pathogens7–13 to vertebrate genomes14–29 (including the
quagga19), have been sequenced.
Paleogenomics is driven by high-throughput sequenc-
ing (HTS) platforms, some of which generate data from
billions of short DNA fragments per run30. In most pale-
ogenomic studies, DNA libraries are generated by ligat-
ing the genomic extract to generic adaptors, amplified
using PCR and then subjected to HTS using so‑called
second-generation sequencing platforms. This contrasts
with traditional PCR-based approaches, in which loci
are individually targeted and sub-amplicon-sized DNA
is unexploitable. In addition to enabling whole-genome
sequencing, HTS revealed how a diverse range of fos-
sil specimens that were previously ignored owing to an
inability to yield PCR amplicons nevertheless contained
ultrashort aDNA fragments (~30–50 bp). Combining
HTS with extraction methods tailored to the short,
damaged aDNA molecules increased the time win-
dow for aDNA sequencing by an order of magnitude
to at least 1 million years in permafrozen regions17 and
500,000 years in temperate caves31,32. Beyond genomes,
the profiling of the epi­genetic landscape (that is, epig-
enomes) of these ancient samples has recently become
feasible33,34, conferring the potential to characterize
regulatory changes throughout evolutionary timescales.
However, there are also difficulties in paleogenomic stud-
ies. Indeed, HTS has enhanced some of the challenges,
including data authentication and contaminant iden-
tification, as well as accounting for inflated error rates
caused by damaged nucleotides.
In this Review, we discuss key technological devel-
opments underpinning the paleogenomic revolution
(FIG. 1) and describe post-mortem damage types com-
mon to aDNA and how they can be accounted for (and
even exploited). Furthermore, we discuss how aDNA
targets can be enriched relative to other DNA, how the
resulting sequences can be analysed, and recent progress
in characterizing ancient epigenomes. Throughout, we
highlight current limitations and provide perspectives for
future developments. As most advances relate to human
calcified tissues (bones and teeth), we principally focus
on these. Some of the key findings addressing long-
standing debates in our own global population history

NATURE REVIEWS | GENETICS VOLUME 16 | JULY 2015 | 395

© 2015 Macmillan Publishers Limited. All rights reserved

REVIEWS

Date range of methods

2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016

Whole‑genome in‑solution capture78

13‑Mb‑long mammoth
DNA (454)52
0.7× mammoth genome (454)5
16× Paleo−Eskimo genome (Illumina)4
1.3× Neanderthal genome (454 and Illumina)6
30× Denisovan genome (Illumina)16
Whole‑chromosome target enrichment77
Microarray‑based target enrichment73
In‑solution target enrichment; PCR probes70
Primer extension capture69
Selective uracil enrichment68
ssDNA library16,66
Extraction of ultrashort DNA fragments31
True single‑molecule DNA sequencing (Helicos)45
High‑throughput DNA sequencing
Mammoth Methylomes33
proteome115
Methylome and nucleosome maps34
6.8× Iceman 52× Neanderthal genome (Illumina)22 aDNA studies
genome and date of
400,000‑year‑old mitochondrial genomes31,32
(SOLiD)15 publication
1.1× 700,000‑year‑old horse genome (Illumina and Helicos)17
Maize kernel transcriptomes113

Figure 1 | Major advances in ancient genomics. The major methodological advances described in this Review are
Nature
presented with respect to milestones in paleogenomics, including whole-genome sequencing and Reviews | Genetics
the characterization of
transcriptomes, epigenomes and proteomes. Average genome fold-coverage (×) and sequencing platforms are indicated
where applicable. aDNA, ancient DNA; ssDNA, single-stranded DNA.

Second-generation
sequencing
High-throughput short-read
DNA sequencing platforms
that require library
construction and thus
modification of the DNA before
sequencing. Most commonly
represented by the Illumina,
GS‑FLX (454), ABI SOLiD and
Ion Torrent series.
Resonance structures
Dynamic, alternative forms of
molecular groups, such as
nucleotide bases, that result
from electron delocalization
within the molecule.
are summarized in BOX 1 to illustrate the diversity of
information that can be gathered, and recent literature
describing other key evolutionary insights revealed by
ancient genomics have been reviewed elsewhere35–38.
aDNA damage and tailored extractions
aDNA damage. aDNA damage accumulates over time
and was originally characterized using enzymatic reac-
tions to reveal the presence of particular types of DNA
damage (such as abasic sites and crosslinks3) or gas chro-
matography experiments coupled with mass spectrome-
try 39. Later approaches inferred damage types on the basis
of mutational patterns in sequence data40–43. Specifically,
an excess of C→T mutations, and their significant reduc-
tion following treatment with uracil DNA glycosylase41,
revealed cytosine deamination to uracil (a thymine
analogue) as the most prominent base modification.
HTS data subsequently refined our understanding of
such damage, demonstrating that deamination increases
towards read termini44, consistent with expectations of
faster rates in the overhanging single strands at the frag-
ment termini16,44,45 (FIG. 2). HTS data also revealed that
depurination drives post-mortem DNA fragmentation, as
genomic positions preceding read starts (corresponding
to breaks or abasic sites in aDNA molecules) often con-
sist of purines44. This bias appears towards adenines for
younger samples but guanines for older samples, possibly
reflecting differences in fragmentation dynamics46 and/
or base-specific resonance structures47. Statistical models
exploiting nucleotide misincorporation patterns in HTS
data sets revealed single-strand breaks in aDNA44,48, most
likely at nicks or abasic sites. Finally, whereas indirect
detection methods indicated that polymerase-blocking
lesions such as interstrand crosslinks could be prominent
in aDNA49, direct experimental assays based on HTS data
suggested a more minor contribution50. Therefore, their
general importance may be context dependent.
Targeting ultrashort fragments. Extensive aDNA frag-
mentation was documented early in the field’s his-
tory, with later quantitative PCR assays revealing up to
100‑fold decreases in the abundance of PCR templates
for each doubling of target size51. As HTS generally allows
most aDNA molecules to be sequenced over their full
length, the resulting distribution represents a size-decay
curve52 that enables direct quantitative comparisons of
fragmentation across specimens through space, time
and environmental conditions53. Although random DNA
fragmentation should decrease molecule numbers expo-
nentially as size increases, aDNA templates often peak at
40–80 bp before this decay is observed. The exact median
length observed reflects the overall fragmentation levels
experienced after death, which generally increase with
the depositional temperature53,54. However, the deviation
from the expected exponential decay curve for ultrashort
sizes suggests that common extraction protocols do not
recover, and thus do not optimally exploit, this fraction
of molecules.
This challenge was met by introducing improved
silica-based extraction protocols that modify volume
and composition of the DNA-binding buffer 31. These
methodological improvements increased recovery
rates of 35–50-bp molecules by twofold to fivefold, and
greatly contributed towards the sequencing of even very

396 | JULY 2015 | VOLUME 16 www.nature.com/reviews/genetics

© 2015 Macmillan Publishers Limited. All rights reserved

 REVIEWS

Box 1 | Human evolution insights: one of the principal achievements of ancient genomics
An area of great interest in the study of human evolution is clarifying the admixture history and the migration routes
followed by our ancestors to create contemporary patterns of genetic variation112. Study of the historical hair of an
Aboriginal Australian revealed the existence of a migration from Africa or the Middle East that reached Australia and that
took place 20,000–30,000 years earlier than the migration that gave rise to present-day Europeans and Asians14. The
36,200‑year-old bone remains from an Upper Paleolithic man from Kostenki, Russia, were also found to be genetically
closer to contemporary Europeans than to contemporary Asians, suggesting an earlier date for the separation between
these populations27. The 24,000‑year-old remains of a child from Mal’ta, south-central Siberia, Russia, showed strong
genetic affinities not only with Europeans but also with Native Americans, indicating a mixed population ancestry for the
first Americans24. The Solutrean theory, which assumed a European origin across the Atlantic for the Paleo-Indian Clovis
culture in North America, could be ruled out because the 12,600‑year-old cranial remains of the Anzick individual
belonging to this culture shows greater genetic affinities to Native Americans than to Europeans25.
The peopling of Europe and the effect of the agricultural revolution have also received great attention15,18,21,27,71,89,103,105,121.
The main genetic components present in modern Europeans seem to have already differentiated by 36,200 years ago27,
and their later dispersal involved several migration waves71,122. The expansion of the first Neolithic farmers resulted in
mixing hunter-gatherer Mesolithic and near-eastern population backgrounds within western Europe ~7,500 years
ago18,21,71,121. A later extensive migration took place ~4,500 years ago from the steppes and was associated with the spread
of Indo-European languages into Europe71. The possibility to gather genome-wide data at population scales from ancient
individuals now provides an opportunity for a fine reconstruction of population migration and admixture patterns from
classical antiquity to modern times.
In some cases, ancient genomes have revealed direct genetic continuity across different archaeological cultures,
questioning theories assuming that culture only changes through the migration of peoples and not simply though the
spread of ideas. The first example is provided by the Paleo-Eskimos from the New World Arctic, who represent distinct
cultural units but were found to represent a single population, first replaced by Inuit <1,000 years ago4,23.
Ancient genomic data have also enhanced our understanding of deeper human evolution, showing that modern
humans admixed with Neanderthals6,22 ~50,000–60,000 years ago27,123 while expanding out of Africa. Current data
suggest that more-recent admixture events might even have taken place122, but this needs further investigation. The
existence of the Denisovans, who may represent a distinct population of Neanderthals from the Altai mountains that
substantially contributed to the genomic diversity of present-day Melanesians, was revealed based on genomic
data14,16,105,123. The sequencing of the mitochondrial genome of ~400,000‑year-old hominins from Atapuerca, Spain, also
revealed genetic affinities between these individuals and the Denisovans, although genome-wide information is needed
to understand the underlying population history32,124.
Finally, when the age of a specimen can be determined using radiocarbon dating, it can be used as a calibration point
to help in the estimation of genome-wide mutation rates. When applied to the remains of a 45,000‑year-old human
from Siberia, this technique confirmed that the autosomal mutation rate was about half of that estimated from the
human–chimpanzee divergence122, which is consistent with recent estimates from human pedigrees125. Similarly, when
the age of a sample falls outside the range of radiocarbon dating, the deficit of mutations observed along the
phylogenetic branch leading to the specimen can be used to provide an estimate for when it lived16,22.

Pre-digestion
Exposure of ancient calcified
materials to a short initial
digestion aimed at removing
substantial fractions of
exogenous contaminants.
454
The initial generation of GS‑FLX
sequencing platforms based
on pyrosequencing, before
their acquisition and renaming
by Roche.
old (for example, ~400,000‑year-old) specimens31,32.
Furthermore, light pre-digestion of bone or tooth pow-
der before full extraction on the remaining undigested
matter significantly increases the relative proportion of
endogenous DNA recovered45,55–58, probably by washing
away microbial contaminants57 or fully liberating DNA
from the matrix 59. Finally, the specific tissue sampled
(for example, petrosal bone versus other bones18,25 and
cementum versus dentine58,60) and sampling procedures
(for example, drilling at low versus high speed60) affect
the quality of extracted aDNA.
DNA library construction and amplification
General recommendations. Second-generation sequenc-
ing requires template molecule modification through
adaptor ligation30. Both library construction and subse-
quent PCR amplification represent sources of error 61,62.
The parts of a genome sequenced can be affected by
adaptor binding biases and/or the relative efficacy of PCR
enzymes to amplify the constructs. Which and where
nucleotide misincorporations occur during these amplifi-
cations also confer errors in resulting sequences16,44,61. For
example, the Phusion polymerase, which was originally
part of the Illumina library building procedure, preferen-
tially amplifies short and relatively GC‑rich templates62.
The same is true for related polymerases, such as Phusion
Hot Start I and II, even when high-fidelity buffers are
used. This bias is reduced, or even disappears, when other
polymerases are used, and Accuprime Pfx, Herculase II
Fusion and Pfu Turbo Cx Hotstart currently seem to
be better alternatives than the most commonly used
polymerases, AmpliTaq Gold and Platinum Taq High-
Fidelity 62. Increasing PCR cycle number often reduces
the molecular complexity of DNA libraries63; thus, poly-
merases should be carefully selected, PCR amplification
cycles minimized and/or independent PCR reactions
undertaken in parallel to limit such biases. This has
important consequences for authenticating aDNA data
and quantifying post-mortem DNA damage, as expected
misincorporation models require tailoring to the exact
experimental procedure followed45,64.
Double-stranded DNA libraries. Different DNA library
construction methods also show clear differences in
efficiency. Early aDNA libraries were based around
454‑compatible blunt-end approaches42–44,52 (FIG. 3a).

NATURE REVIEWS | GENETICS VOLUME 16 | JULY 2015 | 397

© 2015 Macmillan Publishers Limited. All rights reserved

REVIEWS

T/A ligation
A common DNA ligation
technology that relies on
complementary pairing of
thymine and adenine
overhangs at the 3ʹ ends of the
adaptors and inserts to be
ligated, respectively.
Shotgun sequencing
The sequencing of fragmented
DNA in the absence of any
selection strategy.
However, as adaptor ligation is random, a fraction of
the constructs do not contain both of the different adap-
tors and thus cannot be sequenced using this method.
Another possible limitation is adaptor dimer formation
during ligation; if amplified and sequenced, these waste
sequencing capacity. Illumina introduced T/A ligation to
overcome this in their original library construction pro-
cedure, in which aDNA fragments have an overhanging
adenine added (known as A‑tailing) to facilitate liga-
tion to T‑tailed adaptors (FIG. 3b). However, this strategy
seems to be suboptimal for aDNA, mostly because tem-
plates starting with thymines are less efficiently processed
during ligation61. Thus the (often substantial) fraction
of templates containing deaminated cytosine residues
(thymine analogues) at their temini44 fails to incorpo-
rate into libraries61. TruSeq libraries, which also rely on
T/A ligation, have also been shown to introduce signifi-
cant amounts of palindromic artefacts, whereby short
sequence segments at read starts are copied towards
read ends65.
Single-stranded DNA libraries. A subsequent devel-
opment was library construction directly on single-
stranded DNA (ssDNA) templates16,66. In this method,
DNA is denatured using heat into single strands and
then ligated to a first adaptor, before extension with
Bst polymerase generates the complementary strand. A
second adaptor is ligated at the 3ʹ end of the comple-
mentary strand, and the full construct is then amplified
by PCR (FIG. 3c). Inclusion of biotin in the first adaptor
allows minimal DNA loss during purification using
streptavidin-coated paramagnetic beads. The develop-
ment of this method enabled characterization of the
Denisovan genome at ~30× coverage using DNA extracts
generated from 40 mg of bone material16. Although the
method is sometimes beneficial on highly degraded
osseous materials31,32 (as both strands and every single-
strand break of endogenous DNA molecules have 3ʹ ter-
mini that are compatible with their incorporation into
libraries), its benefit on less-degraded and non-osseous
materials remains unverified.
Enriching for aDNA
aDNA extracts are metagenomic mixtures. The endog-
enous DNA within most ancient specimens is usually
embedded within high levels of environmental microbial
DNA. Although there are notable exceptions (including
some keratinized materials4,67, particularly dense bones
such as the petrosal bone18,25, and intentionally preserved
materials from museums or herbaria8), it is unusual for
the endogenous DNA content in most calcified remains
to account for more than a few percent of the total DNA
content. DNA preservation and environmental microbial
contamination levels can show extreme variation within
a single bone. For example, extracts and libraries con-
structed from a single 36,000‑year-old European human
bone yielded 0.1–8.0% of human DNA27, and even greater
variation (0.5–27.8%) was seen using the early Native
American ‘Anzick’ cranial bone25.
High microbial contaminant DNA levels render
shotgun sequencing of genomes uneconomical. Thus,
several methods have been developed that improve
accessibility to endogenous aDNA. These enrichment
strategies are used either during library construction,
by preferentially incorporating damaged aDNA frag-
ments68, or after library construction, by separating

R CpG Y R Y R Y
Y R Y R CpG Y R
m

Post‑mortem DNA decay

Single‑strand break
5′ UpG 3′ 5′ 3′ 5′ 3′
3′ 5′ 3′ 5′ T pG 3′ 5′

Overhang
Post‑mortem base modification Post‑mortem base modification Abasic site

O O
CH3
NH NH
O O O
N N OH
CH2 O CH2 O CH2
5′ O 5′ O 5′ O

3′ 3′ 3′
O O O O O O
P P P
O O O O O O

Figure 2 | Typical ancient DNA molecules. A diverse range of thymines (blue) when cytosines were methylated (mC). Such deaminations
degradation reactions affect DNA post-mortem and result in extensive occur much faster at overhanging ends. Other modifications include
Nature Reviews abasic
| Genetics
fragmentation (preferentially at purine nucleotides) and base modifications. sites (green) and single-strand breaks (vertical lines). The chemical
The most common base modification identified in high-throughput structures of three damage by‑products (uracils, thymines and abasic sites)
sequencing data sets is deamination of cytosines into uracils (red), or are shown. R, purine; Y, pyrimidine.

398 | JULY 2015 | VOLUME 16 www.nature.com/reviews/genetics

© 2015 Macmillan Publishers Limited. All rights reserved

 REVIEWS

endogenous and exogenous fractions through anneal-
ing to pre-defined sets of probes (in solution69–72 or on
microarrays7,73). Intended capture targets range from
whole mitochondrial genomes (~16 kb31,32,69,72,74,75) or
ancient commensal and pathogenic bacterial genomes
(~4 Mb7,10–13) to large sets of single-nucleotide poly-
morphisms (SNPs) (~400,000 SNPs71), whole exomes
(~30 Mb73,76), chromosomes (~30 Mb77) and even whole
nuclear genomes (~3 Gb29,78–80). Other approaches that
have been demonstrated, although not used in the
most recent relevant studies, include targeted diges-
tion of environmental microbial DNA using restriction
enzymes6 and primer extension capture (PEC)69. Before
discussing enrichment strategies further, we highlight
that currently none is able to recover 100% of the tar-
get molecules, and thus they come at a cost of reduced

a dsDNA library c ssDNA library

U U (Strand 1)
(Strand 2)

End repair Denaturation

U (From strand 1)
U
A (From strand 1)
(From strand 2)
Adapter ligation
ssDNA adapter ligation
U
A
U
Fill in

U
A
Extension

U
A

b A-tailed library
U

Adapter ligation
End repair

U
U A
A

Extension and A-tailing

U A
A A Heat denaturation

A
Adapter ligation
Supernatant
T U A
A A T

Figure 3 | Constructing ancient DNA libraries. The three most common
types of ancient DNA (aDNA) libraries are shown. 5ʹ-phosphate groups are
indicated with black circles, single-strand DNA breaks are shown as vertical
lines, biotinylated adaptor groups are shown in red, and streptavidin-
coated beads are shown in grey. a | To construct a double-stranded DNA
(dsDNA) library, aDNA is first end-repaired. It is then ligated
to double-stranded adaptors (blue), and the resultant nicks are filled in to
construct library templates devoid of single-strand breaks. b | To construct
an A‑tailed DNA library, aDNA is end-repaired and then A‑tailed (that is, an
adenine is added to the 3ʹ ends of the strands) to facilitate subsequent
ligation to T‑tailed adaptors while disfavouring ligation between adaptor
pairs. The adaptors are typically Y‑shaped (that is, they are complementary
at the T‑tailed end but have non-complementary arms at the other end).
Nature Reviews | Genetics
The use of such adaptors results in aDNA strands being flanked by distinct
non-complementary adaptor sequences at each end to enable subsequent
unidirectional sequencing through the aDNA fragment. Nicks resulting
from ligation are filled‑in through PCR post-ligation. c | To construct a
single-stranded DNA (ssDNA) library, aDNA is first denatured into single
strands using heat and then ligated to biotinylated single-stranded
adaptors. The original DNA strand is then copied using DNA polymerase
extension, and a second adaptor is ligated to enable further PCR
amplification and sequencing. Purification steps are performed using
streptavidin-coated paramagnetic beads. Part c adapted with permission
from REF. 16, American Association for the Advancement of Science.

NATURE REVIEWS | GENETICS VOLUME 16 | JULY 2015 | 399

© 2015 Macmillan Publishers Limited. All rights reserved

REVIEWS
Primer extension capture
(PEC). An enrichment
technology based on the
ligation of short 5ʹ-biotinylated
oligonucleotides (including a
12‑nucleotide-long spacer
followed by a primer of 18–25
nucleotides that is designed to
match a particular region of
interest) to single-stranded
target molecules. This is
followed by a single round of
polymerase-based extension
so as to increase the length
over which the molecules are
hybridized.
Tiled probes
Probes that overlap in their
positioning on the target so
as to ensure that every target
position is covered by more
than one different probe.
Chimeric DNA libraries
Recombination between
libraries containing different
template molecules during
library PCR amplification,
resulting in hybrid (chimeric)
sequences that do not
represent true biological
sequences.
Double-indexed DNA
libraries
DNA libraries in which short
(for example, 8 bp long)
unique nucleotide indexes
are incorporated within both
adaptors used during library
construction. Indexes are
bordered by known sequences
that serve to prime index
sequencing reactions and also
enable library attachment to
the surface of the sequencing
flow cell.
400 | JULY 2015 | VOLUME 16
library complexity. Therefore, the upper threshold on
the maximum sequencing depth attainable from a given
library is reduced, and users must consider the end goal
of their analyses before determining whether capture is
a sensible strategy over direct shotgun sequencing. If
the goal is to sequence to high coverage, highly complex
libraries showing relatively high endogenous content
can be shotgun-sequenced4,6,18,22,26, but enrichment of
multiple libraries is advisable in other cases71,72.
Damaged template enrichment. One approach selec-
tively targets damaged DNA molecules68 during ssDNA
library preparation16,66. After the DNA strand comple-
mentary to the original template is generated, con-
structs are 5ʹ-phosphorylated, which enables ligation
to a non-phosphorylated adaptor (FIG. 4a). Following
extension with Bst polymerase to fill the nick located
5ʹ of this adaptor, treatment with uracil DNA glycosy-
lase and endonuclease VIII (USER mix) is implemented
to first replace deaminated cytosines with abasic sites
and then to cleave out these abasic sites81. The new 3ʹ
end is then dephosphorylated and used for priming a
new extension. Thus, all library strands that originally
harboured deaminated cytosines are reconstructed
over their full length and are available in the reaction
supernatant for further amplification and sequenc-
ing. The undamaged DNA template fraction remains
attached to streptavidin-coated paramagnetic beads
and can be retained for other uses. This method has
shown great specificity when applied to samples from
Late Pleistocene Neanderthals showing extreme levels
of deamination68. Importantly, in all extracts tested,
the relative contamination from modern human DNA
decreased by ~1.6‑fold following selective enrichment,
suggesting that undamaged templates resulting from
recent manipulations of the specimen could readily
be filtered. Furthermore, the endogenous content of
one sample increased by 3.7–5‑fold, which markedly
reduced the genome sequencing cost. Future experi-
ments will no doubt explore the wider potential of this
method. For now, users should bear in mind that any
endogenous undamaged molecules will not be retained
and will thus be lost, making the method only appro-
priate for the most damaged samples. Additionally, any
DNA carrying damage will be enriched, potentially
providing access to the genomes of associated ancient
microorganisms (although these can show reduced
DNA damage levels compared to their human hosts9).
Extension-free target enrichment in solution. Target
enrichment approaches based on target–probe hybrid-
ization are currently widely used. These require heat
denaturation of DNA libraries to enable annealing of
library inserts to overlapping tiled probes along target
regions. Probes can be economically generated using
long-range PCR, if fresh DNA material from closely
related species can be extracted70, through PCR ampli-
con shearing and then ligation to a biotinylated adaptor.
This probe library can be amplified (with biotinylated
primers) and used in an unlimited number of enrich-
ment reactions. Following annealing at stringencies that
© 2015 Macmillan Publishers Limited. All rights reserved
can be adapted depending on the phylogenetic distance
between targets and probes, streptavidin-coated beads
are washed to eliminate library constructs with inserts
showing no genetic proximity to the targeted regions,
and the final fraction is amplified and sequenced.
This strategy has predominantly been used for
sequencing mitochondrial genomes72,74,75,82–85, bacte-
rial plasmids86 and short nuclear loci84. Hybridization
is even successful when probes diverge from targets
by 10–13%82, which is useful if no close living relative
and/or reference genome is available. This can also be
exploited to detect probe carry-over post-sequencing
if the DNA from a distantly related organism was
used for preparing probes (for example, if DNA from
a European bison was used when enriching for aDNA
from aurochs83). Alternatively, potential probe carry-over
can be eliminated before sequencing using dedicated
molecular tools. For example, replacement of deoxythy-
midine triphosphate (dTTP) by deoxyuridine triphos-
phate (dUTP) in probes enables subsequent digestion
with uracil DNA glycosylase before amplification
and sequencing 83.
Biotinylated probes can also be custom designed
and synthesized, which enables specific probe til-
ing and in silico assessment for secondary structures,
homogeneous GC content and annealing tempera-
tures. Different manufacturers can now deliver such
probes, with related procedures apparently achieving
similar efficiency 80. Depending on the overall size of
the genomic regions targeted, multiple libraries can, in
theory, be enriched as pools to achieve faster hands‑on
times. However, owing to the probable formation of
chimeric DNA libraries during post-capture PCRs, pooling
of libraries before capture should ideally be avoided, or
if pooling is used then the constituent libraries should
at least be double-indexed DNA libraries87 to enable chi-
maera identification and elimination from subsequent
analyses. Increasing probe tiling densities (11 bp ver-
sus 24 bp) did not consistently improve enrichment for
~670 nuclear loci in archaeological maize, suggesting
that even relatively reduced probe densities can be used
to efficiently recover the full molecular complexity of
DNA libraries88.
In general, custom-synthesized biotinylated probes
are most economical when targeting fairly small regions
(hundreds of kilobases to a few megabases) owing to
probe synthesis costs. However, microarrays can achieve
extremely high probe numbers (approximately 1 million
each) and, if manufacturers consent, can be chemically
treated to cleave the probes from the microarray sur-
face, thus recovering large sets of probes at relatively
reasonable costs71,76,77. Synthetic DNA probes are built
into biotinylated probe libraries using biotinylated adap-
tors of minimal size (~20 bp) to limit interference during
probe–target annealing. The known adaptor sequence
allows further amplification, thereby immortalizing the
probe set at low cost. In this way, Fu et al.77 used 8.7 mil-
lion probes to recover most of the non-repetitive frac-
tion of chromosome 21 from a 40,000‑year-old human
specimen from Tianyuan cave, China. In addition, they
targeted ~3,500 200‑bp‑long regions around positions
www.nature.com/reviews/genetics
 REVIEWS

a Selective uracil enrichment b WISC

U (Strand 1) Endogenous fraction
(Strand 2) U
A
Denaturation
Exogenous fraction

U (From strand 1)
Heat denaturation
(From strand 1)
(From strand 2) Hybridization

ssDNA adapter ligation

U

Target-enriched fraction
U
Washing U
and elution

Extension and
phosphorylation

U
A Hybridization

Biotinylated
RNA probes

Adapter ligation In vitro transcription

U
A

Probe DNA library

Fresh DNA extract

USER treatment
Figure 4 | Enriching DNA libraries for ancient inserts. a | Selective uracil
enrichment is shown. 5ʹ-phosphate groups are indicated with black circles,
A single-strand DNA breaks are shown as vertical lines, biotinylated adaptor groups are
shown in red, and streptavidin-coated beads are shown in grey. A single-stranded
DNA (ssDNA) library is built until the polymerase extension step. DNA is then
phosphorylated to enable the ligation of the second adaptor. This contrasts with
the ssDNA library procedure, in which the ligation occurs between the 5ʹ end of the
second adaptor and the 3ʹ end of the newly synthesized strand (FIG. 3c). DNA is then
Extension
treated with uracil DNA glycosylase and endonuclease VIII (USER mix) to generate
and then cleave out abasic sites at cytosines that were deaminated into uracils
post-mortem. The 3ʹ-phosphate groups at these new termini are then removed
A (not shown). The resulting 3ʹ‑OH ends now serve to prime an extension with a DNA
polymerase, which copies throughout the whole length of the strand complementary
to where the damage was. As a result, the supernatant now contains double-stranded
DNA (dsDNA) library templates corresponding to the original deaminated strands.
Other library templates remain unaffected and can be separated, as they remain
bound to streptavidin-coated paramagnetic beads. b | In whole-genome in-solution
Extension
capture (WISC), ssDNA templates from an ancient DNA (aDNA) library are prepared.
The target, endogenous aDNA is shown as thin black lines, whereas the exogenous
contaminating DNA is shown as thin green lines; adaptors are shown as thick blue
A Supernatant
lines. In parallel, a probe DNA library is prepared from fresh modern DNA extracts (thin
red lines) and used to generate biotinylated RNA probes through in vitro transcription.
T7 adaptors to enable in vitro transcription are shown in thick purple lines. The
aDNA library is annealed to the RNA probes, low-complexity DNA and adaptor
blockers (the latter two are not shown for simplicity). The library fraction of interest
is then recovered following elution from streptavidin-coated paramagnetic beads.
Part a adapted with permission from REF. 68, Cold Spring Harbor Laboratory Press.
Part b adapted with permission from REF. 78, The American Society of Human Genetics.

Nature Reviews | Genetics

NATURE REVIEWS | GENETICS VOLUME 16 | JULY 2015 | 401

© 2015 Macmillan Publishers Limited. All rights reserved

REVIEWS
Mate pairs
Pairs of sequences derived from
both ends of a DNA library.
Edit distance
The number of sequence
mismatch counts between
reads and targets.
402 | JULY 2015 | VOLUME 16
known to carry allelic variants in archaic and modern
humans, thereby enabling direct estimates of archaic
hominin ancestry within the Tianyuan specimen. The
method was also used to obtain the exome sequence of
two Neanderthals from Spain and Croatia76 and, more
recently, sequence data from ~400,000 loci within a sin-
gle reaction71. This target enrichment procedure reduced
the genotyping costs by at least 45‑fold per ancient speci-
men71 and enabled genome-wide analyses of ancient
individuals at population scales. In this analysis, two
52‑nucleotide-long probes were selected to be located on
each side of a polymorphic site, and two were centred
on the polymorphic site, each representing one of the
two possible alleles.
Solid-phase target enrichment. Direct application
of microarrays can also enrich large sets of targets,
using approaches originally described for modern
DNA89. First used in the aDNA context to characterize
exome sequences from a 49,000‑year-old Neanderthal
specimen73, microarrays have subsequently enabled
whole-genome sequencing from bacterial strains
responsible for major historical epidemiological out-
breaks7,9–13, including the Black Death7. Microarrays
also provide interesting alternatives to real-time PCR
and shotgun sequencing for parallel screening of >100
pathogens12,90. This is particularly appropriate for iden-
tifying ancient pathogens, which often leave no physical
skeletal evidence and are generally found only as trace
material. Possible drawbacks are poor detection of the
most divergent genomic regions and omission of regions
with important genomic rearrangements (such as
insertions) or unknown additional plasmids that do not
segregate in modern strains.
Whole-genome enrichment. There is a growing interest
in characterizing the entire genome sequence of ancient
individuals at population scales. However, none of the
methods presented above is appropriate for pulling
down whole human genomes, as this requires synthe-
sizing gigabases of probes. Whole-genome in‑solution
capture (WISC)78 and a commercial alternative with
similar performance79,80 fill this niche, enabling eco-
nomical whole-genome enrichment. WISC starts with
the preparation of a genome-wide RNA probe library
from a species with a genome that is closely related to
the target genome in the aDNA sample (FIG. 4b). These
RNA probes are generated from a genomic DNA library
flanked by adaptors containing T7 promoters that
enable a relatively inexpensive reaction, in vitro tran-
scription. This in vitro transcription step is carried out
in the presence of biotin 16–UTP, so that the resultant
RNA probes are biotinylated. The biotinylated RNA
probes are annealed to the ssDNA of a heat-denatured
aDNA library, while low-complexity DNA and adaptor-
blocking RNA oligonucleotides improve stringency and
reduce enrichment for highly repetitive regions. Non-
hybridized DNA is washed away, whereas the bound,
enriched library fraction is finally released following
RNase treatment (which precludes probe carry-over)
and amplified before sequencing.
© 2015 Macmillan Publishers Limited. All rights reserved
WISC-like approaches consistently improve the pro-
portion of sequences that can be mapped to the human
reference genome compared to shotgun sequencing
(6–159‑fold), at least when based on double-stranded
DNA libraries 29,78–80. As hybridization efficiency
increases with target length79, its efficacy may be reduced
when analysing libraries built using single-strand meth-
ods16,66, which routinely exhibit smaller mean target mol-
ecule sizes. The fraction of reads that align to repetitive
regions also generally increases with WISC, despite the
use of an excess of low-complexity DNA. Unsurprisingly,
WISC-enriched libraries show reduced complexity, so
that almost every unique insert can be sequenced with
minimal sequencing efforts78. As an example, 5–10 mil-
lion sequencing reads generated using WISC-enriched
libraries of a Bronze Age Danish human hair sample and
a pre-Columbian Peruvian human bone were found to
cover 7,000–21,000 ancestry-informative markers, which
proved to be sufficient for inferring the continental
groups that are the closest to these ancient individuals78.
Analysing aDNA
From reads to genome alignments. Most available
paleo­genomes were generated using Illumina technolo-
gies, although there are exceptions5,15,17. Analysis of the
underlying sequence data mainly relies on computa-
tional approaches developed for handling HTS data
from modern DNA material, with some additional par-
ticularities. Most procedures are implemented within the
open-source PALEOMIX package91, in which reads are
trimmed of adaptor sequences using AdapterRemoval92
and collapsed when mate pairs are available and over-
lap significantly, filtered for a minimal size of 25–30 bp
and aligned against reference genomes of interest using
Burrows–Wheeler Aligner (BWA)93 or Bowtie 2 (REF. 94).
Alignments showing low-quality scores and PCR dupli-
cates are further removed using the MarkDuplicates pro-
gram from Picard tools, and reads are locally realigned
around small insertions and deletions (indels) to improve
overall genome quality using the IndelRealigner tool from
the Genome Analysis Toolkit (GATK)95. PALEOMIX can
also quantify DNA damage levels using mapDamage2
(REF. 48) and perform phylogenomic and metagenomic
analyses using modules mostly based on inferences deriv-
ing from ExaML (Exascale Maximum Likelihood)96 and
MetaPhlAn (Metagenomic Phylogenetic Analysis)97,
respectively.
Unlike sequences in other re‑sequencing genome
projects, in which mismatches relative to the refer-
ence genome generally are derived from sequencing
errors and polymorphisms, aDNA sequences exhibit
substantial fractions of nucleotide misincorporations
that result from sequencing damaged bases. As these
misincorporations cluster towards read termini, seed-
ing approaches, whereby only the most upstream part
of the sequence is used for speeding up identification
of possible alignments along the genome, should be
avoided98. Parameters controlling acceptance thresholds
for read‑to‑reference edit distance should be adapted to
the phylogenetic distance to the reference genome, as
overly conservative procedures will under-represent the
www.nature.com/reviews/genetics

Probabilistic aligners
Mapping algorithms that can
accommodate non-uniform
distributions of sequencing
errors along reads, generally
leading to improved alignments
between reads and reference
genomes.
Thermal age
The predicted time that
it would have taken an
archaeological sample to
produce the observed degree
of DNA degradation were the
sample exposed to a constant
temperature of 10 °C since
deposition. Thermal age has
been proposed to adjust the
chronological age of a sample
to its thermal history and to
help in predicting the likelihood
of DNA surviving in
archaeological remains.
Haplotypes
The DNA sequences of
haploid chromosomes.
Derived alleles
Alleles that are evolutionarily
derived in a lineage of interest
and that are not represented
in an ancestral population
or species.
Ancestral alleles
Alleles in the ancestral state
before a mutation took place
in a descending population,
species or lineage.
Nearly fixed
Fixed alleles are those that
are derived and present in
all individuals in a descendent
population or species. Nearly
fixed alleles therefore
represent those that are
present in nearly all individuals
(thus close to fixation, for
example, showing allelic
frequencies of 99% in the
population).
NATURE REVIEWS | GENETICS
most polymorphic regions and under-estimate heterozy-
gosity levels. Conversely, overly permissive procedures
will inflate the alignment false-positive rate, resulting
in regions with many reads from different organisms,
which is a particular challenge for aDNA data, given
its complex mixture of endogenous and exogenous
reads52,57.
Owing to the accumulation of nucleotide misincor-
poration towards read ends, probabilistic aligners based
on position-scoring matrices have been developed to
embed aDNA features from the aligning step. Available
aligners include Mapping Iterative Assembler (MIA)69,
ANFO Short Read Aligner/Mapper 6 and BWA-PSSM
(position-specific scoring matrix)99, and these generally
show good performance for short reads and/or low-
quality data, although some show running times that
are compatible only with alignments against relatively
small reference genomes (for example, mitochondrial
genomes). Importantly, such probabilistic approaches
handle platform-specific error profiles in a sound
statistical framework.
Authenticating aDNA data. Following read alignment,
analyses often focus on authenticating whether sequenc-
ing data are ancient. Software such as mapDamage45,100
or pmdtools101 can test the presence of typical nucleotide
misincorporation patterns that result from inflated cyto-
sine deamination rates at overhangs. Such patterns can
be first obtained by preparing libraries on an aliquot of
the DNA extract, while saving the remaining fraction for
preparing almost damage-free libraries following USER
treatment 81. This will limit nucleotide misincorpora-
tion effects on downstream analyses. Alternatively, mild
USER treatment, which removes most, but not all, of the
damage signature, has been proposed to enable sequence
authentication and population analyses using the same
sample aliquot 72.
Nucleotide misincorporation patterns can be
exploited to fit statistical models of post-mortem DNA
damage and estimate cytosine deamination rates and
nick frequencies44,48. Even though deamination rates at
overhangs were reported to increase linearly with time
across a wide range of archaeological sites and preserva-
tion conditions46, this pattern has not been confirmed
within archaeological sites in permafrost17 or temperate
environments72. Additionally, different remains from the
same specimen and/or extracts from the same remain
can show variable levels of DNA damage27,32. This sug-
gests complex relationships in which both global condi-
tions, as reflected in the thermal age of a given specimen,
and microenvironmental factors (within and between
remains) drive the amount of DNA damage ultimately
measured. In our opinion, these complex relationships,
and the dependency of damage quantification on the
library preparation and amplification procedures, pre-
clude the use of strict minimal thresholds of expected
DNA damage levels as authentication criteria. Thus,
quantitative comparison with the levels observed for
samples excavated at the same or similar archaeological
sites, and processed with the same experimental tools, is
recommended.
© 2015 Macmillan Publishers Limited. All rights reserved
REVIEWS
Statistical damage models also allow correction of
base quality scores depending on their probability
of being the result of nucleotide misincorporations at
damage sites48, thus limiting their possible effect on
downstream analyses. However, we emphasize that for
low-coverage data — in which mismatches are observed
on a few reads at best and penalized when close to read
termini — this procedure can potentially inflate the
genetic proximity to the reference genome. SNP calling
can also benefit from genotype callers, such as SNPest4,102,
that explicitly model post-mortem DNA damage as a
possible source of error. Furthermore, nucleotide mis-
incorporation patterns can be used by computational
tools to sort the fraction of reads that show evidence
of post-mortem damage101, which is useful when there
is substantial modern DNA contamination. Although
extremely conservative and not cost-effective (as not
all aDNA molecules carry post-mortem DNA damage
and many true aDNA reads will be discarded), damage-
based filtering approaches have shown great success in
characterizing whole-mitochondrial sequences from
extensively contaminated Neanderthal specimens101
and an ~400,000‑year-old hominin32. Finally, com-
paring analytical outcomes when considering the full
population of reads or only the most damaged frac-
tion (and disregarding mutations, such as transitions,
that derive from post-mortem damage40–44) can provide
evidence that the results are not driven by damage and
contamination artefacts103.
In addition to revealing nucleotide misincorporation
patterns, mapDamage also delivers the base composition
of the genomic regions directly flanking DNA inserts
and therefore tests depurination as the main driver for
DNA fragmentation44,48,100, which can also help authen-
tication. This pattern is substantially affected following
USER treatment, which mainly cleaves DNA down-
stream of unmethylated cytosine residues, therefore
resulting in an excess of cytosines at genomic positions
just preceding read starts16,72.
Estimating contamination levels. Nucleotide mis­
incorporation and base compositional patterns can be
detected in even substantially contaminated samples.
This can happen when treating the outer sample surface
with bleach before DNA extraction, which can help to
remove a fraction of fresh DNA contaminants but also
introduces signatures of DNA damage within the remain-
ing contaminants104. This can also happen when a mix-
ture of highly degraded aDNA templates and undamaged
DNA contaminants is incorporated into libraries. A suite
of tools has thus been developed for further authenticat-
ing aDNA data (in particular for human aDNA). The
current methods available exploit the sequence infor-
mation at sites and/or haplotypes with known variation
across species and/or populations. For example, modern
human contamination in Neanderthal HTS data has
been estimated using the relative proportion of derived
alleles and ancestral alleles observed at mitochondrial sites
showing nearly fixed derived alleles in modern humans6.
A similar rationale was used to estimate the possible con-
tribution of different human population backgrounds105
VOLUME 16 | JULY 2015 | 403
REVIEWS
Epialleles
Allelic variants showing
identical genetic sequences
but different epigenetic
marks, such as different
methylation patterns.
404 | JULY 2015 | VOLUME 16
or species83 to final mitochondrial consensus sequences.
A statistically more powerful contamination estimator for
mitochondrial reads that uses linkage information at the
read level has been developed74.
As the cellular mitochondrial number is variable
across cell types and tissues, contamination estimates
based on mitochondrial sequence data do not directly
reflect the true contamination levels of the nuclear
genome106. Heterozygosity levels observed on male
X chromosomes can be used as a nuclear contamina-
tion proxy. As males are haploid for most X chromo-
some loci, base discordance between overlapping reads
should result only from sequencing errors and should be
distributed randomly along the chromosome. However,
if modern human DNA contamination is present, dis-
cordance rates should inflate at sites that are polymor-
phic within contemporary populations14. For archaic
hominin specimens, nuclear contamination rates can
be calculated from fixed alleles that are derived in mod-
ern humans6. For female ancient human samples, the
presence of sequences that are known to be unique to
the Y chromosome can also reflect the presence of con-
tamination from male-derived sources106. Triallelic sites
at autosomes could potentially be used in the future to
estimate levels of nuclear contamination with modern
human DNA, irrespective of the sample gender.
Genome completion and error rates. Reliable contami-
nation estimates can generally be recovered from the
data aligning to the X chromosome using even low-
depth information, as long as each single genomic posi-
tion is covered once on average (that is, ~1× coverage).
Ultimately, the exact fraction of the genome that is cov-
ered depends on the sequencing effort and the sequence
length. For aDNA sequence reads of 60 nucleotides,
~87% of the human genome is non-repetitive, and there-
fore reads of similar size (or shorter) cannot be uniquely
aligned to the remaining ~13% of the genome4. For
example, the genome of a Paleo-Eskimo Greenlander
of the Saqqaq culture was sequenced to ~16× coverage,
with ~20% of the genome missing. This achieved ~20×
coverage at positions covered at least once, although
some variation was observed along the chromosomes, as
half of the positions showed a depth of coverage of ≤7×.
Using DNA polymerases that reduce size and base com-
positional biases during library amplification62 can help
to limit such variation, although nucleosome protection
can also lead to specific patterns of depth‑of‑coverage
variation along the genome (see below).
Overall sequence accuracy of ancient genomes is
another parameter that is worth considering, as sequenc-
ing errors will have an impact on downstream analyses.
Genome-wide error rates are generally estimated using
three‑way alignments that include the genome of a
closely related outgroup (for example, the chimpanzee)
and a high-quality genome from a living conspecific
individual107. The excess of derived alleles observed in
the genome of the ancient individual provides an esti-
mate for its error rate relative to the high-quality modern
genome. Unsurprisingly, this rate is highly dependent on
DNA damage levels and the molecular tools used before
© 2015 Macmillan Publishers Limited. All rights reserved
and during sequencing. The best alternative developed
so far involves USER treatment followed by paired-end
sequencing 81, which can generally reduce error rates to
<0.1–0.2% per base16,27, although with the caveat that
the error rate seems to be only marginally reduced in
CpG dinucleotide contexts owing to the high degree of
methylation at such sites16,33.
When authenticated on the basis of low contamina-
tion levels and evidence of post-mortem DNA dam-
age, the genome-wide data gathered can be used for
various analyses, including phylogenomic reconstruc-
tion 9–11,16,17,26,91 and inference of population history
(BOX 2). In addition, sequence data can be exploited to
reconstruct genome-wide epigenetic maps (see below),
thus paving the way for evaluating the extent to which
epigenetic reprogramming has influenced recent human
evolution38 (FIG. 5).
Reconstructing ancient epigenomes
Genome-wide epigenetic maps. Cytosine methylation,
which predominantly occurs in vertebrates at CpG
dinucleotides, is the most commonly studied epigenetic
modification. Classically, methylomes (that is, genome-
wide maps of methylated cytosines) are reconstructed
using bisulfite sequencing that converts unmethylated
cytosines into uracils, which are sequenced as thymines,
leading to CpG→TpG mutations. Methylated cytosines
are not converted and are sequenced as regular CpG
sites. Although it was successfully used to reveal fine-
scale methylation patterns at four nuclear loci on the
DNA extracted from a Late Pleistocene bison bone108,
this approach generally requires large amounts of DNA,
as it inflicts extensive damage to DNA and is therefore
generally inappropriate for aDNA. However, similar
CpG→UpG conversions naturally occur post-mortem.
In contrast to bisulfite treatment, methylated epialleles
(mCpGs) are deaminated into TpGs, whereas unmeth-
ylated epialleles (CpGs) are deaminated into UpGs. As
cytosine deamination is elevated in methylated con-
texts109, a substantial fraction of mCpGs is expected to be
converted into TpGs. Molecular tools that prevent UpG
sequencing can therefore reveal methylated epialleles by
tracking CpG→TpG mutations33,34.
Briggs et al.81 first exploited this feature of aDNA,
using HTS data generated from library inserts in
which uracil residues were removed following USER
treatment. Gokhman et al.33 subsequently generated
high-resolution aDNA methylation maps using high-
quality Neanderthal and Denisovan genomes, identify-
ing ~2,000 regions that show differential methylation
patterns in the bones of modern versus archaic homi-
nins. Homeobox D10 (HOXD10), which encodes a
key regulator of limb development, is found in one
such region, and its epigenetic reprograming is pro-
posed to have participated in the shaping of specific
morpho-anatomical features along the human lineage33.
Pedersen et al.34 used sequence data from the high-
coverage Paleo-Eskimo Saqqaq genome4 to also extract
genome-wide methylation information, as DNA libraries
were amplified using a polymerase that does not bypass
uracils. Furthermore, CpG→TpG substitution rates were
www.nature.com/reviews/genetics

Box 2 | Reconstructing population histories
One of the most common first steps in the analysis of genome-wide data from ancient
humans is the characterization of their closest relatives among modern populations.
Such inferences are generally based on principal component analysis (PCA) or
statistical clustering, using software such as Admixture126. A benefit of statistical
clustering is that it also enables documentation of contamination levels through
determining whether the ancient samples exhibit a genetic contribution that could be
derived from the research team4. With shotgun sequencing at low depth of coverage
(for example, ≤8×), genotypes cannot be reliably determined, and analyses are
generally performed using pseudo-haploid data in which sequence reads from many
loci consist of a random sampling of only one of the two constituent alleles, and thus
individuals are considered to be homozygous for the unique allele sampled at a given
locus. The genomic regions covered across multiple individuals are then also limited,
which reduces the number of orthologous loci overlapping known genetic variation in
modern populations. In such cases, the ancestry of each ancient individual can be
determined using multidimensional scaling (MDS), which exploits pairwise measures
of genetic distances in a panel of individuals, calculated by normalizing the sum of all
instances where two individuals show different alleles by the total number of loci with
no missing data in each pair. This procedure is implemented in the bammds package127.
Additionally, Procrustes transformation of individual PCA projections based on the
particular vector of single-nucleotide polymorphisms covered in each specimen and
the same reference panel can help to visualize the population affinities of a group of
ancient individuals within a single analysis103.
However, PCA-based approaches reflect not only population ancestries but also the
temporal sampling between ancient and modern individuals128. Thus, at best, MDS, PCA
and clustering analyses should be viewed as formulating evolutionary hypotheses,
which subsequently require testing using approaches such as model-based inference,
as well as coalescence simulations14, D‑statistics129 and population f‑statistics130.
Population f‑statistics methods, such as the f3‑statistics, have been developed for
detecting populations with mixed ancestries and identifying populations that are
closest to ancient individuals130. D‑statistics has received particular attention because it
originally supported the theory that admixture occurred between Neanderthals and
non-African modern humans6. D-statistics is based on four-way alignments that include
one outgroup (O) and three populations (H1, H2 and H3), of which two (H1 and H2) are
more closely related. For example, in the case of Neanderthals, with the following
configuration (O = Chimpanzee, H3 = Neanderthals; H2 = Eurasians, H2 = Africans),
positive D‑statistics indicate an excess of shared polymorphisms and possible
admixture between Neanderthals and Eurasians6,16,22. However, this observation is also
compatible with gene flow into Africans from a currently unsampled and divergent
ghost population129, as well as with population subdivision in Africa, with Neanderthal
and Eurasian ancestors leaving Africa from related population backgrounds131.
Admixture events can be further dated from the distribution of introgressive block
lengths in modern and ancient individuals27,122,132, as recombination reduces their size
over time. The resulting date seemed to be too recent to be compatible with a scenario
involving population subdivision in Africa, which confirmed admixture with
Neanderthals outside Africa.
Ghost population found to correlate with known methylation levels at pro-
An unsampled population that moters, exons, introns and CpG islands. This was not
exchanges migrants with other observed for other CpN dinucleotides, confirming that
sampled populations and that methylation drives the signal. The authors also used
can be identified based on
admixture signatures left in
CpG→TpG substitutions at read starts (where deami-
descending populations. nation rates are maximal) to infer ancient methylation
levels for genomic regions overlapping the loci from the
Introgressive block lengths Illumina Infinium Human Methylation450BeadChip
Population admixture
array. The inferred methylation profile from the Saqqaq
introduces a mosaic of ancestry
blocks along the genome, the sample was found to cluster with hair follicle meth-
lengths of which decrease with ylation profiles, which is in agreement with the tissue
each subsequent generation originally used for DNA extraction.
owing to recombination.
Introgressive block lengths
can therefore be exploited
Genome-wide nucleosome maps. Library inserts
to determine the date of derived from endogenous aDNA generally show uni-
admixture events. modal size distributions that are typically centred
NATURE REVIEWS | GENETICS
© 2015 Macmillan Publishers Limited. All rights reserved
REVIEWS
around 40–80 bp. However, several aDNA sequence
data sets exhibit striking 10‑bp periodicity in their size
distribution4,21,25,26,34,79. Pedersen et al.34 proposed that
this results from nucleosome protection, with DNA
fragmentation preferentially occurring at nucleotides
facing away from nucleosomes. Assuming that nucle-
osomes are strongly positioned and phased along DNA
scaffolds, and recalling that the turn of the DNA helix
is 10 bp long, only 1 nucleotide per 10 bp would be
fully exposed to hydrolysis. If this is true, then nucleo-
some protection should also drive additional patterns.
For example, DNA fragmentation should occur pref-
erentially within spacers, which are nucleosome-free
regions of ~50 bp separating successive ~150-bp DNA
blocks covered by nucleosomes. Fewer endogenous
reads should therefore map to spacer regions, lead-
ing to depth‑of‑coverage periodicities of ~200 bp,
with peaks of coverage corresponding to nucleosome
centres and correlating with both in silico predicted
and experimentally derived nucleosome maps. These
predicted periodicities were confirmed in the Saqqaq
sample data, even following correction for base com-
positional effects, which can substantially affect
depth‑of‑coverage variation during library amplifica-
tion62. This finding, together with expected patterns
of methylation and depth of coverage within CTCF
regions and splicing sites, confirmed the nucleosome
protection hypothesis.
Nucleosomes might protect DNA from cleavage that
occurs during cellular apoptosis and/or post-mortem34.
As similar periodicity patterns have been found not
only in ancient hair follicles4,34, which have undergone
extensive apoptosis, but also in other ancient tissues
that are not particularly affected by apoptosis, such as
teeth21 and bones25,26,79, we expect that ancient nucleo-
some maps could, in the future, be reconstructed across
a wide range of samples. Recalling that such patterns
are also absent from many of the samples analysed so
far, further work is needed to understand which fac-
tors drive the preservation of signatures of nucleosome
protection after death.
Assessing ancient gene expression levels. Post-mortem
DNA damage enables the reconstruction of ancient
methylome and nucleosome maps. Given the central
role of epigenetic states in regulating chromatin acces-
sibility to transcription factors, this information can be
tentatively used to infer ancient gene expression levels.
Encouragingly, methylation ratios between gene bod-
ies and promoter regions (a proxy for gene expression)
showed strong correlation with hair follicle expression
levels measured using high-throughput RNA sequenc-
ing (RNA-seq)34. However, further work is needed to
develop genuine proxies that accurately measure ancient
gene expression levels. The epigenome of each cell type
is complex, and ancient samples will necessarily span a
range of tissues, with unbalanced contributions from
different cell types, which will possibly result in vari-
able validity of expression predictions across samples,
age, sex and health conditions. As one example, genome
hypermethylation is a known response to viral infection
VOLUME 16 | JULY 2015 | 405
REVIEWS
Nucleosome Spacer DNA
mCpG
CpG
Depth of
coverage
CpG→TpG
Figure 5 | Tracking ancient nucleosome and methylation maps. DNA wrapped
Nature Reviewsin| Genetics
around nucleosomes can be protected post-mortem and over-represented
high-throughput sequencing (HTS) data. Therefore, depth‑of‑coverage patterns
along the genome can be exploited to position the location of nucleosomes on
ancient genomes. Similarly, post-mortem deamination at CpG sites transforms
methylated CpG (mCpG) sites into TpG sites but transforms unmethylated CpG sites
into UpG sites. With molecular tools disabling the sequencing of the UpGs, CpG→TpG
mutations in HTS data provides an opportunity to detect ancient mCpGs,
with hypomethylated regions showing low CpG→TpG conversion rates and
hypermethylated regions showing high CpG→TpG conversion rates. Adapted with
permission from REF. 38, American Association for the Advancement of Science.
in plants, and methylation assays for ancient plant mate-
rial can therefore be used to monitor viral exposure in
ancient populations109.
Conclusions
Recent technical developments have enhanced our
understanding of the properties of aDNA molecules and
how we should best proceed to maximize their retrieval.
In some environments, this enables genomic characteri-
zation throughout much of the past million years17,31,32.
Ongoing research and the increasing wealth of sequenc-
ing data generated will undoubtedly further improve
current approaches in the near future. DNA extraction
represents an area with great potential for improve-
ment, especially if tailored to the molecular structures,
niches and microenvironmental parameters that best
preserve DNA.
The discovery that post-mortem cytosine deami-
nation preferentially occurs at overhangs was impor-
tant for the development of authentication criteria44.
CTCF regions However, other base modifications, including pyrimi-
Genomic regions targeted by dine derivatives, have been identified 39. Improved
CCCTC-binding factor (CTCF)
and involved in regulating the
characterization of the chemical features of aDNA mol-
three-dimensional structure ecules, as well as their methylation and nucleosome pro-
of chromatin and transcription tection patterns, could therefore open new avenues for
by mediating long-range data authentication. This will also improve our ability
interactions between genomic
to correct sequence analyses from as-yet-unidentified
sequences.
biases and provide opportunities for targeting damaged
Admixture templates before sequencing. The development of engi-
Interbreeding of individuals neered DNA polymerases that can bypass specific DNA
from multiple population lesions introduced post-mortem110 could also facilitate
origins, resulting in the
introduction of DNA from one
library construction and amplification.
population into the genomes Importantly, although the approaches outlined
of a second population. here improve aDNA retrieval and analyses, the HTS
406 | JULY 2015 | VOLUME 16
© 2015 Macmillan Publishers Limited. All rights reserved
technologies themselves had the greatest impact on the
field. Although not originally designed for aDNA, their
massive throughput coupled with their ability to sequence
short molecules rendered them ideal for aDNA applica-
tions. Therefore, it is likely that future HTS platforms that
directly sequence DNA bases and their modifications with
minimal (if any) library preparation will drive the future
of aDNA research. The results of the initial application of
true single-molecule DNA sequencing are encouraging,
having demonstrated substantial improvement in relative
amounts of accessible endogenous sequences17,45,56.
Although most paleogenomic studies have focused
on a limited number of individuals, current approaches
allow the characterization of genome-wide SNP vari-
ation at ancient population scales71,111. Future studies
can be expected to investigate genetic variation in large
population samples on the high-density SNP or even
whole-genome scale, thus improving our understanding
of past demographic, adaptive and admixture trajectories
with greater detail112.
Besides delivering ancient genomes and epig-
enomes, new methodological developments have also
provided access to ancient transcriptomes113,114 and
proteomes17,115,116. Owing to the biochemical processes
inherent in animal cell death, animal tissues are unlikely
to represent good reservoirs for long-term RNA sur-
vival. Materials still exist in other organisms that do not
undergo autolysis. One example is plant seed, a tissue that
requires RNA survival for germination and that has dem-
onstrated ancient RNA survival going back hundreds to
thousands of years113,114. Such materials may contribute
to our understanding of how gene expression path-
ways have been remodelled during domestication.
Additionally, a wide range of ancient proteins have been
sequenced from Late115 and Middle17 Pleistocene speci-
mens. With half-lives exceeding that of DNA, ancient
peptides might be the only way to retrieve genetic
information from the early Pleistocene and even earlier
time periods. Within a much more recent time range,
namely the past few thousand years, studies of proteins
have already delivered information that is not obtainable
from DNA, such as whether milk products were already
consumed in particular ancient societies117. Molecular
analyses of dental plaque, which offers a rich reservoir
entrapping biomolecules derived not only from the host
but also from its diet and the oral microbiome118,119, may
also hold great promises, especially now that computa-
tional approaches have been developed to compare the
diversity of past and present microbiomes57.
A final question worth considering is whether the
technological breakthroughs in ancient genomics may
offer pathways towards de-extinction120. Bringing back
lost species is of growing interest, and although it is a
topic fraught with challenges ranging from the ethical to
the technological, for many extinct species a key starting
requisite will be a well-characterized reference genome.
As new extraction and computational methods expand
the age range and quality of specimens from which such
data can reliably be obtained, so too will the range of
species that could be considered as possible targets for
de-extinction attempts.
www.nature.com/reviews/genetics

1. Higuchi, R., Bowman, B., Freiberger, M., Ryder, O. A.
& Wilson, A. C. DNA sequences from the quagga, an
extinct member of the horse family. Nature 312,
282–284 (1984).
2. Hagelberg, E., Sykes, B. & Hedges, R. Ancient bone
DNA amplified. Nature 342, 485 (1989).
3. Pääbo, S. Ancient DNA: extraction, characterization,
molecular cloning, and enzymatic amplification.
Proc. Natl Acad. Sci. USA 86, 1939–1943 (1989).
4. Rasmussen, M. et al. Ancient human genome
sequence of an extinct Paleo-Eskimo. Nature 463,
757–762 (2010).
This study takes advantage of the relative absence
of environmental microorganisms within ancient
hairs to characterize the first high-quality ancient
human genome.
5. Miller, W. et al. Sequencing the nuclear genome of the
extinct woolly mammoth. Nature 456, 387–390 (2008).
6. Green, R. E. et al. A draft sequence of the Neandertal
genome. Science 328, 710–722 (2010).
This paper reports the first draft genome of an
archaic hominin and many methodological
developments that are still commonly used for
characterizing and analysing ancient genomes.
7. Bos, K. I. et al. A draft genome of Yersinia pestis from
victims of the Black Death. Nature 478, 506–510
(2011).
This paper reports the first genome isolated from
an ancient pathogenic bacterium, confirming the
Black Death as a plague epidemic. It revealed that
no derived variant is unique to the medieval strain,
suggesting that non-genetic factors enhanced the
virulence of the pathogen.
8. Martin, M. D. et al. Reconstructing genome evolution
in historic samples of the Irish potato famine
pathogen. Nat. Commun. 4, 2172 (2013).
9. Schuenemann, V. J. et al. Genome-wide comparison of
medieval and modern Mycobacterium leprae. Science
341, 179–183 (2013).
10. Bos, K. I. et al. Pre-Columbian mycobacterial genomes
reveal seals as a source of New World human
tuberculosis. Nature 514, 494–497 (2014).
11. Devault, A. M. et al. Second-pandemic strain of Vibrio
cholera from the Philadelphia cholera outbreak.
N. Engl. J. Med. 370, 334–340 (2014).
12. Devault, A. M. et al. Ancient pathogen DNA in
archaeological samples detected with a microbial
detection array. Sci. Rep. 4, 4245 (2014).
13. Wagner, D. M. et al. Yersinia pestis and the Plague of
Justinian 541–543 AD: a genomic analysis. Lancet
Infect. Dis. 14, 319–326 (2014).
14. Rasmussen, M. et al. An Aboriginal Australian genome
reveals separate human dispersals into Asia. Science
334, 94–98 (2011).
15. Keller, A. et al. New insights into the Tyrolean Iceman’s
origin and phenotype as inferred by whole-genome
sequencing. Nat. Commun. 3, 698 (2012).
16. Meyer, M. et al. A high-coverage genome sequence
from an archaic Denisovan individual. Science 338,
222–226 (2012).
This paper describes a novel method for
constructing aDNA libraries using ssDNA
templates, which enabled the characterization of
the Denisovan genome at a quality rivalling that
of modern genomes, starting from only minute
amounts of DNA extracts.
17. Orlando, L. et al. Recalibrating Equus evolution using
the genome sequence of an early Middle Pleistocene
horse. Nature 499, 74–78 (2013).
This study takes advantage of both
second-generation (high-throughput, and library-
and amplification-dependent) and third-generation
(high-throughput, and library- and amplification-
independent) sequencing technologies to present
the oldest genome sequence hitherto
characterized: that of an ~700,000‑year-old horse.
18. Gamba, C. et al. Genome flux and stasis in a five
millennium transect of European prehistory. Nat.
Commun. 5, 5257 (2014).
19. Jónsson, H. et al. Speciation with gene flow in equids
despite extensive chromosomal plasticity. Proc. Natl
Acad. Sci. USA 111, 18655–18660 (2014).
20. Malaspinas, A. S. et al. Two ancient human genomes
reveal Polynesian ancestry among the indigenous
Botocudos of Brazil. Curr. Biol. 24, R1035–R1037
(2014).
21. Olalde, I. et al. Derived immune and ancestral
pigmentation alleles in a 7,000‑year-old Mesolithic
European. Nature 507, 225–228 (2014).
22. Prüfer, K. et al. The complete genome sequence of a
Neanderthal from the Altai Mountains. Nature 505,
43–49 (2014).
NATURE REVIEWS | GENETICS
23. Raghavan, M. et al. The genetic prehistory of the New
World Arctic. Science 345, 1255832 (2014).
24. Raghavan, M. et al. Upper Paleolithic Siberian
genome reveals dual ancestry of Native Americans.
Nature 505, 87–91 (2014).
25. Rasmussen, M. et al. The genome of a Late
Pleistocene human from a Clovis burial site in western
Montana. Nature 506, 225–229 (2014).
26. Schubert, M. et al. Prehistoric genomes reveal the
genetic foundation and cost of horse domestication.
Proc. Natl Acad. Sci. USA 111, E5661–E5669 (2014).
27. Seguin-Orlando, A. et al. Genomic structure in
Europeans dating back at least 36,200 years. Science
346, 1113–1118 (2014).
28. Ramirez, O. et al. Genome data from a sixteenth
century pig illuminate modern breed relationships.
Heredity 114, 175–184 (2015).
29. Schroeder, H. et al. Genome-wide ancestry of 17th
century enslaved Africans from the Caribbean.
Proc. Natl Acad. Sci. USA 112, 3669–3673 (2015).
30. Metzker, M. L. Sequencing technologies — the next
generation. Nat. Rev. Genet. 11, 31–46 (2010).
31. Dabney, J. et al. Complete mitochondrial genome
sequence of a Middle Pleistocene cave bear
reconstructed from ultrashort DNA fragments. Proc.
Natl Acad. Sci. USA 110, 15758–15763 (2013).
32. Meyer, M. et al. A mitochondrial genome sequence of
a hominin from Sima de los Huesos. Nature 505,
403–406 (2014).
33. Gokhman, D. et al. Reconstructing the DNA
methylation maps of the Neandertal and the
Denisovan. Science 344, 523–527 (2014).
34. Pedersen, J. S. et al. Genome-wide nucleosome map
and cytosine methylation levels of an ancient human
genome. Genome Res. 24, 454–466 (2014).
This study exploits DNA degradation patterns in
HTS data sets to characterize, for the first time,
genome-wide nucleosome and methylation maps
from an ancient human and infer ancient gene
expression levels and the age at death of the
individual.
35. Ermini, L., Der Sarkissian, C., Willerslev, E. &
Orlando, L. Major transitions in human evolution
revisited: a tribute to ancient DNA. J. Hum. Evol. 79,
4–20 (2015).
36. Shapiro, B. & Hofreiter, M. A paleogenomic
perspective on evolution and gene function: new
insights from ancient DNA. Science 343, 1236573
(2014).
37. Orlando, L. & Cooper, A. Using ancient DNA to
understand evolutionary and ecological processes.
Ann. Rev. Ecol. Evol. Syst. 45, 573–598 (2014).
38. Orlando, L. & Willerslev, E. An epigenetic window into
the past? Science 345, 511–512 (2014).
39. Höss, M., Jaruga, P., Zastawny, T. H., Dizdaroglu, M. &
Pääbo, S. DNA damage and DNA sequence retrieval
from ancient tissues. Nucleic Acids Res. 24,
1304–1307 (1996).
40. Hansen, A. J., Willerslev, E., Wiuf, C., Mourier, T. &
Arctander, P. Statistical evidence for miscoding lesions
in ancient DNA templates. Mol. Biol. Evol. 18,
262–265 (2001).
41. Hofreiter, M., Jaenicke, V., Serre, D., von Haeseler, A.
& Pääbo, S. DNA sequences from multiple
amplifications reveal artifacts induced by cytosine
deamination in ancient DNA. Nucleic Acids Res. 29,
4793–4799 (2001).
42. Stiller, M. et al. Patterns of nucleotide
misincorporations during enzymatic amplification and
direct large-scale sequencing of ancient DNA. Proc.
Natl Acad. Sci. USA 103, 13578–13584 (2006).
43. Gilbert, M. T. et al. Recharacterization of ancient DNA
miscoding lesions: insights in the era of sequencing-
by‑synthesis. Nucleic Acids Res. 35, 1–10 (2007).
44. Briggs, A. et al. Patterns of damage in genomic DNA
sequences from a Neandertal. Proc. Natl Acad. Sci.
USA 104, 14616–14621 (2007).
This study characterizes typical nucleotide
misincorporation and fragmentation patterns using
HTS data from aDNA extracts, which have been
subsequently used as essential authentication
criteria.
45. Orlando, L. et al. True single-molecule DNA
sequencing of a Pleistocene horse bone. Genome Res.
21, 1705–1719 (2011).
46. Sawyer, S. et al. Temporal patterns of nucleotide
misincorporations and DNA fragmentation in ancient
DNA. PLoS ONE 7, e34131 (2012).
47. Overballe-Petersen, S., Orlando, L. & Willerslev, E.
Next-generation sequencing offers new insights into
DNA degradation. Trends Biotechnol. 30, 364–368
(2012).
© 2015 Macmillan Publishers Limited. All rights reserved
REVIEWS
48. Jónsson, H. et al. mapDamage2.0: fast approximate
Bayesian estimates of ancient DNA damage
parameters. Bioinformatics 29, 1682–1684 (2013).
49. Hansen, A. J. et al. Crosslinks rather than strand
breaks determine access to ancient DNA sequences
from frozen sediments. Genet. 173, 1175–1179
(2006).
50. Heyn, P. et al. Road blocks on paleogenomes —
polymerase extension profiling reveals the frequency
of blocking lesions in ancient DNA. Nucleic Acids Res.
38, e161 (2010).
51. Poinar, H. N., Kuch, M., McDonald, G., Martin, P. &
Pääbo, S. Nuclear gene sequences from a late
Pleistocene sloth coprolithe. Curr. Biol. 13,
1150–1152 (2003).
52. Poinar, H. N. et al. Metagenomics to paleogenomics:
large-scale sequencing of mammoth DNA. Science
311, 393–394 (2006).
This study reports the first genetic analysis of
ancient specimens based on a HTS technology,
paving the way for whole-genome sequencing from
ancient specimens.
53. Allentoft, M. E. et al. The half-life of DNA in bone:
measuring decay kinetics in 158 dated fossils.
Proc. Biol. Sci. 279, 4724–4733 (2012).
54. Smith, C. I., Chamberlain, A. T., Riley, M. S.,
Stringer, C. & Collins, M. J. The thermal history of
human fossils and the likelihood of successful DNA
amplification. J. Hum. Evol. 45, 203–217 (2003).
55. Schwarz, C. et al. New insights from old bones:
DNA preservation and degradation in permafrost
preserved mammoth remains. Nucleic Acids Res. 37,
3215–2129 (2009).
56. Ginolhac, A. et al. Improving the performance of true
single molecule sequencing for ancient DNA. BMC
Genomics 13, 177 (2012).
57. Der Sarkissian, C. et al. Shotgun microbial profiling of
fossil remains. Mol. Ecol. 23, 1780–1798 (2014).
58. Damgaard, P. et al. Improving access to endogenous
DNA in ancient bones and teeth. BioRxiv http://dx.doi.
org/10.1101/014985 (2015).
59. Salamon, M., Tuross, N., Arensburg, B. & Weiner, S.
Relatively well preserved DNA is present in the crystal
aggregates of fossil bones. Proc. Natl Acad. Sci. USA
102, 13783–13788 (2005).
60. Adler, C. J., Haak, W., Donlon, D. & Cooper, A.
Survival and recovery of DNA from ancient teeth and
bones. J. Archaeol. Sci. 38, 956–964 (2011).
61. Seguin-Orlando, A. et al. Ligation bias in illumina
next-generation DNA libraries: implications for
sequencing ancient genomes. PLoS ONE 8, e78575
(2013).
62. Dabney, J. & Meyer, M. Length and GC‑biases during
sequencing library amplification: a comparison of
various polymerase-buffer systems with ancient and
modern DNA sequencing libraries. Biotechniques
87–94 (2012).
63. Young, A. L. et al. A new strategy for genome
assembly using short sequence reads and reduced
representation libraries. Genome Res. 20, 249–256
(2010).
64. Seguin-Orlando, A. et al. Amplification of TruSeq
ancient DNA libraries with AccuPrime Pfx:
consequences on nucleotide misincorporation and
methylation patterns. STAR 1, STAR2015112054892
315Y.0000000005 (2015).
65. Star, B. et al. Palindromic sequence artifacts generated
during next generation sequencing library preparation
from historic and ancient DNA. PLoS ONE 9, e89676
(2014).
66. Gansauge, M. T. & Meyer, M. Single-stranded DNA
library preparation for the sequencing of ancient or
damaged DNA. Nat. Protoc. 8, 737–748 (2013).
67. Gilbert, M. T. et al. Whole-genome shotgun sequencing
of mitochondria from ancient hair shafts. Science 317,
1927–1930 (2007).
68. Gansauge, M. T. & Meyer, M. Selective enrichment of
damaged DNA molecules for ancient genome
sequencing. Genome Res. 24, 1543–1549 (2014).
69. Briggs, A. et al. Targeted retrieval and analysis of five
Neandertal mtDNA genomes. Science 325, 318–321
(2009).
70. Maricic, T., Whitten, M. & Pääbo, S. Multiplexed DNA
sequence capture of mitochondrial genomes using
PCR products. PLoS ONE 5, e14004 (2010).
71. Haak, W. et al. Massive migration from the steppe was
a source for Indo-European languages in Europe. Nature
http://dx.doi.org/10.1038/nature14317 (2015).
72. Rohland, N., Harney, E., Mallick, S., Nordenfelt, S. &
Reich, D. Partial uracil–DNA–glycosylase treatment for
screening of ancient DNA. Phil. Trans. R. Soc. B 370,
20130624 (2015).
VOLUME 16 | JULY 2015 | 407
REVIEWS
73. Burbano, H. A. et al. Targeted investigation of the
Neandertal genome by array-based sequence capture.
Science 328, 723–725 (2010).
This paper reports the first characterization of an
ancient exome using target enrichment approaches
on microarrays.
74. Fu, Q. et al. A revised timescale for human evolution
based on ancient mitochondrial genomes. Curr. Biol.
23, 553–559 (2013).
75. Vilstrup, J. T. et al. Mitochondrial phylogenomics of
modern and ancient equids. PLoS ONE 8, e55950
(2013).
76. Castellano, S. et al. Patterns of coding variation in the
complete exomes of three Neandertals. Proc. Natl
Acad. Sci. USA 111, 6666–6671 (2014).
77. Fu, Q. et al. DNA analysis of an early modern human
form Tianyuan Cave, China. Proc. Natl Acad. Sci. USA
110, 2223–2227 (2013).
This paper describes a target enrichment
procedure exploiting millions of DNA probes
cleaved from user-designed DNA microarrays to
characterize the almost complete sequence of the
non-repetitive fraction of chromosome 21 for an
~40,000‑year-old human.
78. Carpenter, M. L. et al. Pulling out the 1%: whole-
genome capture for the targeted enrichment of
ancient DNA sequencing libraries. Am. J. Hum. Genet.
93, 852–864 (2013).
This paper reports the first whole-genome target
enrichment method, which makes use of
self-generated RNA probes. The method
substantially reduces the operational cost of target
enrichment and allows genetic analyses of specimens
with only minute amounts of aDNA templates.
79. Enk, J. M. et al. Ancient whole genome enrichment
using baits built from modern DNA. Mol. Biol. Evol.
31, 1292–1295 (2014).
80. Avila-Arcos, C. et al. Comparative performance of two
whole-genome capture methodologies on ancient DNA
Illumina libraries. Methods Ecol. Evol. http://dx.doi.
org/10.1111/2041-210X.12353 (2015).
81. Briggs, A. et al. Removal of deaminated cytosines and
detection of in vivo methylation in ancient DNA.
Nucleic Acids Res. 38, e87 (2010).
This paper presents an enzymatic procedure based
on the treatment of DNA extracts with USER mix,
which can considerably reduce the sequencing error
rate of ancient genomes by limiting the effect of
nucleotide misincorporations at damaged sites.
82. Mason, V. C., Li, G., Helgen, K. M. & Murphy, W. J.
Efficient cross-species capture hybridization and
next-generation sequencing of mitochondrial genomes
from noninvasively sampled museum specimens.
Genome Res. 21, 1695–1704 (2011).
83. Zhang, H. et al. Morphological and genetic evidence
for early Holocene cattle management in northeastern
China. Nat. Commun. 4, 2755 (2013).
84. Fabre, P. H. et al. Rodents of the Caribbean: origin and
diversification of hutias unraveled by next-generation
museomics. Biol. Lett. http://dx.doi.org/10.1098/
rsbl.2014.0266 (2014).
85. Foote, A. D. et al. Tracking niche variation over
millennial timescales in sympatric killer whale
lineages. Proc. Biol. Sci. 280, 20131481 (2013).
86. Schuenemann, V. J. et al. Targeted enrichment of
ancient pathogens yielding the pPCP1 plasmid
of Yersinia pestis from victims of the Black Death.
Proc. Natl Acad. Sci. USA 108, E746–E452 (2011).
87. Kircher, M., Sawyer, S. & Meyer, M. Double indexing
overcomes inaccuracies in multiplex sequencing on the
Illumina platform. Nucleic Acids Res. 40, e3 (2012).
88. Avila-Arcos, M. C. et al. Application and comparison of
large-scale solution-based DNA capture-enrichment
methods on ancient DNA. Sci. Rep. 1, 73 (2011).
89. Hodges, E. et al. Hybrid selection of discrete
genomic intervals on custom-designed microarrays
for massively parallel sequencing. Nat. Protoc. 4,
960–974 (2009).
90. Bos, K. I. et al. Parallel detection of ancient pathogens
via array-based DNA capture. Phil. Trans. R. Soc. B
370, 20130375 (2015).
91. Schubert, M. et al. Characterization of ancient and
modern genomes by SNP detection and phylogenomic
and metagenomic analysis using PALEOMIX. Nat.
Protoc. 9, 1056–1082 (2013).
This paper presents a fully automated pipeline
performing all sequence analyses associated with
re‑sequencing genomic projects, phylogenomic
inference and metagenomic profiling. It is applicable
to both modern and ancient sequence data sets.
408 | JULY 2015 | VOLUME 16
92. Lindgreen, S. AdapterRemoval: easy cleaning of next-
generation sequencing reads. BMC Res. Notes 5, 337
(2012).
93. Li, H. & Durbin, R. Fast and accurate short read
alignment with Burrows–Wheeler transform.
Bioinformatics 25, 1754–1760 (2009).
94. Langmead, B. & Salzberg, S. L. Fast gapped-read
alignment with Bowtie 2. Nat. Methods 9, 357–359
(2012).
95. McKenna, A. et al. The genome analysis toolkit: a
MapReduce framework for analyzing next-generation
DNA sequencing data. Genome Res. 20, 1297–1303
(2010).
96. Kozlov, A. M., Aberer, A. J. & Stamatakis, A. ExaML
version 3: a tool for phylogenomic analyses on
supercomputers. Bioinformatics http://dx.doi.
org/10.1093/bioinformatics/btv184 (2015).
97. Segata, N. et al. Metagenomic microbial community
profiling using unique clade-specific marker genes.
Nat. Methods 9, 811–814 (2012).
98. Schubert, M. et al. Improving ancient DNA read
mapping against modern reference genomes. BMC
Genomics13, 178 (2012).
99. Kerpedjev, P., Frellsen, J., Lindgreen, S. & Krogh, A.
Adaptable probabilistic mapping of short reads using
position specific scoring matrices. BMC Bioinformatics
15, 100 (2014).
100. Ginolhac, A., Rasmussen, M., Gilbert, M. T.,
Willerslev, E. & Orlando, L. mapDamage: testing for
damage patterns in ancient DNA sequences.
Bioinformatics 27, 2153–2155 (2011).
101. Skoglund, P. et al. Separating endogenous ancient
DNA from modern day contamination in a Siberian
Neandertal. Proc. Natl Acad. Sci. USA 111,
2229–2234 (2014).
102. Lindgreen, S., Krogh, A. & Pedersen, J. S.
SNPest: a probabilistic graphical model for estimating
genotypes. BMC Res. Notes 7, 698 (2014).
103. Skoglund, P. et al. Origins and genetic legacy of
Neolithic farmers and hunter-gatherers in Europe.
Science 336, 466–469 (2012).
104. García-Garcerà, M. et al. Fragmentation of
contaminant and endogenous DNA in ancient
samples determined by shotgun sequencing;
prospects for human paleogenomics. PLoS ONE 6,
e24161 (2011).
105. Sánchez-Quinto, F. et al. Genomic affinities of two
7,000‑year-old Iberian hunter-gatherers. Curr. Biol.
22, 1494–1499 (2012).
106. Green, R. E. et al. The Neandertal genome
and ancient DNA authenticity. EMBO J. 28,
2494–2502 (2009).
107. Reich, D. et al. Genetic history of an archaic hominin
group from Denisova Cave in Siberia. Nature 468,
1053–1060 (2010).
108. Llamas, B. et al. High-resolution analysis of cytosine
methylation in ancient DNA. PLoS ONE 7, e30226
(2012).
109. Smith, O. et al. Genomic methylation patterns in
archaeological barley show de‑methylation as a time-
dependent diagenetic process. Sci. Rep. 4, 5559
(2014).
110. d’Abbadie, M. et al. Molecular breeding of
polymerases for amplification of ancient DNA.
Nat. Biotech. 25, 939–943 (2007).
111. da Fonseca, R. R. et al. The origin and evolution of
maize in the American Southwest. Nat. Plants
1, 14003 (2015).
112. Pickrell, J. K. & Reich, D. Toward a new history and
geography of human genes informed by ancient DNA.
Trends Genet. 30, 377–389 (2014).
113. Fordyce, S. L. et al. Deep sequencing of RNA from
ancient maize kernels. PLoS ONE 8, e50961 (2013).
114. Allaby, R. G. et al. Using archaeogenomic and
computational approaches to unravel the history of
local adaptation in crops. Phil. Trans. R. Soc. B 370,
20130377 (2015).
115. Cappellini, E. et al. Proteomic analysis of a Pleistocene
mammoth femur reveals more than one hundred
ancient bone proteins. 11, 917–926 (2012).
116. Cappellini, E., Collins, M. J. & Gilbert, M. T.
Unlocking ancient protein palimpsests. Science 343,
1320–1322 (2014).
117. Warinner, C. et al. Direct evidence of milk
consumption from ancient human dental calculus.
Sci. Rep. 4, 7104 (2014).
118. Adler, C. J. et al. Sequencing ancient calcified dental
plaque shows changes in oral microbiota with dietary
shifts of the Neolithic and Industrial revolutions.
Nat. Genet. 45, 450–455 (2013).
© 2015 Macmillan Publishers Limited. All rights reserved
119. Warinner, C. et al. Pathogens and host immunity in the
ancient human oral cavity. Nat. Genet. 46, 336–344
(2014).
120. Shapiro, B. How to Clone a Mammoth — the Science
of De‑extinction (Princeton University Press, 2014).
121. Lazaridis, I. et al. Ancient human genomes suggest
three ancestral populations for present-day
Europeans. Nature 513, 409–413 (2014).
122. Fu, Q. et al. Genome sequence of a 45,000‑year-old
modern human from western Siberia. Nature 514,
445–449 (2014).
123. Krause, J. et al. The complete mitochondrial DNA
genome of an unknown hominin from southern
Siberia. Nature 464, 894–897 (2010).
124. Orlando, L. A. 400,000‑year-old mitochondrial
genome questions phylogenetic relationships
amongst archaic hominins. Bioessays 36,
598–605 (2014).
125. Scally, A. & Durbin, R. Revising the human mutation
rate: implications for understanding human evolution.
Nat. Rev. Genet. 13, 745–753 (2012).
126. Alexander, D. H., Novembre, J. & Lange, K.
Fast model-based estimation of ancestry in unrelated
individuals. Genome Res. 19, 1655–1664 (2009).
127. Malaspinas, A. S. et al. bammds: a tools for assessing
the ancestry of low-depth whole-genome data using
multidimensional scaling (MDS). Bioinformatics 30,
2962–2964 (2014).
128. Skoglund, P., Sjodin, P., Skoglund, T., Lascoux, M. &
Jakobsson, M. Investigating population history using
temporal genetic differentiation. Mol. Biol. Evol. 31,
2516–2527 (2014).
129. Durand, E. Y., Patterson, N., Reich, D. & Slatkin, M.
Testing for ancient admixture between closely related
populations. Mol. Biol. Evol. 28, 2239–2252 (2011).
130. Patterson, N. et al. Ancient admixture in human
history. Genetics. 192, 1065–1093 (2012).
131. Eriksson, A. & Manica, A. Effect of ancient population
structure on the degree of polymorphism shared
between modern human populations and ancient
hominins. Proc. Natl Acad. Sci. USA 109,
13956–13960 (2012).
132. Sankararaman, S., Patterson, N., Heng, L., Pääbo, S.
& Reich, D. The date of interbreeding between
Neandertals and modern humans. PLoS Genet. 8,
e1002947 (2012).
Acknowledgements
This work was supported by the Danish Council for
Independent Research, Natural Sciences (FNU,
4002‑00152B and 0602‑02383B); the Danish National
Research Foundation (DNFR94); the Lundbeck Foundation
(R52‑5062); a Marie Curie Career Integration Grant (FP7
CIG‑293845); and the “Chaires d’Attractivité 2014” IDEX,
University of Toulouse, France.
Competing interests statement
The authors declare no competing interests.
FURTHER INFORMATION
AdapterRemoval: https://github.com/slindgreen/
AdapterRemoval
Admixture: http://www.genetics.ucla.edu/software/admixture/
ANFO Short Read Aligner/Mapper: https://bioinf.eva.mpg.
de/anfo/
bammds: http://dna.ku.dk/~sapfo/bammds.html
Bayesian reconstruction of ancient DNA fragments:
https://github.com/grenaud/leeHom
Bowtie 2: http://bowtie-bio.sourceforge.net/bowtie2/index.
shtml
Burrows–Wheeler Aligner: http://bio-bwa.sourceforge.net/
BWA-PSSM: http://bwa-pssm.binf.ku.dk/
ExaML: http://sco.h-its.org/exelixis/web/software/examl/
index.html
Genome Analysis Toolkit: https://www.broadinstitute.org/
gatk/
mapDamage and mapDamage2: http://ginolhac.github.io/
mapDamage/
Mapping Iterative Assembler: http://mia-assembler.
sourceforge.net/
MetaPhlAn: http://huttenhower.sph.harvard.edu/metaphlan
PALEOMIX: https://github.com/MikkelSchubert/paleomix
Picard: http://broadinstitute.github.io/picard
pmdtools: https://code.google.com/p/pmdtools/
SNPest: https://github.com/slindgreen/SNPest
Thermal Age Web Tool: http://thermal-age.eu/
ALL LINKS ARE ACTIVE IN THE ONLINE PDF
www.nature.com/reviews/genetics