A Supplement to Science, produced by the AAAS Office of Publishing and Member Services

Supported by

Sequencing3.indd 1 9/7/06 3:16:11 PM

 www.roche-applied-science.com

Genome Sequencer 20 System

Accelerate Amplicon Analysis

Clonal sequencing of amplicons enables ultra-deep
genetic analysis at levels of sensitivity not previously
possible.

Cancer Research

I Discover somatic mutations with unprecedented

speed, accuracy, and sensitivity.*

I Study DNA methylation patterns with high precision.

Genome Sequencer 20 Amplicon
Software screen. The histogram shows Resequencing for Medical Research
the deviations between read consensus
sequences and reference sequences. I Uncover and screen SNPs at the population and
In addition, the software displays the individual level.
corresponding alignment of reads and
reference sequences for comparison I Deconvolute complex populations of sequence data.
purposes.
I Eliminate tedious and expensive manual analysis of
Visit www.genome-sequencing.com mixed chromatograms.
for more information and published
references, call 800 262 4911 (U.S.),
First to the finish with ultra-deep analysis of
or contact your local sales representative.
amplicons — new for the Genome Sequencer 20
System.
* Thomas, R. et al., Nature Medicine advance online publication,
25 June 2006 (doi:10.1038/nm1437)

Roche Diagnostics
Roche Applied Science
© 2006 Roche Diagnostics. All rights reserved. Indianapolis, Indiana

Sequencing3.indd 2 9/7/06 3:17:40 PM

 Contents
2 Introductions:
Pushing the Envelope
S. Sanders

New-Generation Sequencing Technologies:

Faster Results and New Applications
T. Harkins

4 A N C I E NT DN A
New Methods Yield Mammoth Samples
A. Gibbons

5 Metagenomics to Paleogenomics:
Large-Scale Sequencing of Mammoth DNA
H. N. Poinar, et al.

11 The Race for the $1000 Genome

Fast, cheap genetic analyses will soon become a
reality, and the consequences—good and bad—
will affect everybody
R. F. Service

16 No Longer De-Identified
A. L. McGuire and R. A. Gibbs

20 Finding Criminals Through DNA of Their

Relatives
F. R. Bieber, C. H. Brenner, D. Lazer

24 Genomic Sequencing of Pleistocene Cave

Bears
J. P. Noonan, et al.

29 Whole Genome Assembly Using Paired End

Reads in E. coli, B. licheniformis, and S.
cerevisiae
T. Jarvie

Copyright © 2006 by the American Association for

the Advancement of Science. All rights reserved.

Sequencing3.indd 3 9/7/06 3:17:56 PM

 Introductions

Pushing the Envelope

W
hen the human genome sequencing project was initiated in 1990, the
time required to complete the project was predicted to be at least 15 years.
However, determination on the part of researchers coupled with technological
advances that allowed more sequence data to be squeezed out of each sequencing run
led to a reduction of two years in that target. A remarkable achievement, considering
the ambitious timeline. The fundamental principle on which this technology was based
has essentially not changed since its development by Fred Sanger in 1977. Although a
number of significant improvements have been made, the methodology still has a number
of inherent technical and physical limitations, such as the limited throughput of samples
and the requirement for cloning and amplification of the DNA prior to sequencing. These
are believed to, for all intents and purposes, preclude further significant improvements in
speed, cost, and throughput.
This circumstance has prompted the development of a number of new technologies
through which researchers have looked at the problem afresh and proposed interesting
solutions, or presented novel twists on the original Sanger technique. Not all have
achieved implementation yet, but the parties involved are plainly focused on making their
technologies available to users as soon as possible. The different approaches include single
molecule sequencing (VisiGen, LI-COR), sequencing by ligation (Agencourt), and so-
called sequencing-by-synthesis (Roche/454 Life Sciences, Microchip Biotechnologies,
NimbleGen). An interesting new methodology utilizing nanopore sequencing, which
benefits from not requiring amplification or cloning of target DNA, is being researched
by a number of laboratories. Although still in the initial stages of development, it perhaps
gives a glimpse of where sequencing may be going in the future. The field as a whole has
been invigorated not only by increased industry spending, but also by the offer of prize
monies from different groups (such as the Genome X Prize offered by Craig Venter), plus
the availability of some sizable NIH grants.
Each new technology is not without its limitations and drawbacks. In particular, limited
sequence reads may reduce the throughput of certain systems; these weaknesses are being
aggressively addressed, with good success. Additionally, important ethical and privacy
concerns that frequently come hand in hand with such technological advances need to be
urgently addressed. These issues aside, it is abundantly clear that the $1,000 genome is
tantalizingly close to being realized. Its benefits to “omics” research will be many, from
the ability to perform SNP comparisons over entire genomes to the sequencing of the full
complement of siRNAs in individual transcriptomes, or the capacity for genomewide
comparisons of genetic aberrations in tumor samples. Appealingly, it could also make the
Holy Grail of personalized medicine a reality sooner rather than later.
Sean Sanders, Ph.D.
Commercial Editor
Science Office of Publishing and Member Services

Sequencing3.indd 4 9/7/06 3:18:11 PM

 New-Generation Sequencing Technologies:
Faster Results and New Applications

I
n the last decade, DNA sequencing has been a cornerstone for nearly every aspect of
biological research and drug discovery. Past advances in Sanger sequencing, coupled
with the shotgun method of sample preparation, enabled the completion of the whole
genome, accelerating discoveries in other fields such as gene expression and genotyping.
The capillary electrophoresis–based Sanger method is currently the most popular DNA
sequencing technology and was the foundation of the human genome project. In addition
to sequencing complete genomes, the Sanger technique has been used for multiple
applications; for example, fragment sequencing for clone verification or the detection
of variations in the genome for use in population studies or disease-focused research.
As this method has been steadily improved over the past ten years, costs have fallen by
approximately 90% and throughput has increased tenfold. Nevertheless, this standard
method has reached its physical limits.
For the first time in nearly a decade, a new sequencing technology is commercially
available. Based upon pyrosequencing chemistries, the Genome Sequencer 20 System
has energized the DNA sequencing market, providing more data faster and creating new
applications to address existing scientific questions. The Genome Sequencer 20 System,
developed by 454 Life SciencesTM Corporation and distributed by Roche Diagnostics,
uses a sequencing approach called “Sequencing by Synthesis.” The technique is based on
the clonal amplification of single molecules on beads (microparticles), which are isolated
in an emulsion, and the subsequent massively parallel sequencing of the amplified DNA
on the beads placed in wells of a PicoTiterPlate™ device. The Genome Sequencer 20
System achieves a sequencing output of at least 20 Mb (200,000 fragments) per 5.5-hour
run, with an average read length of 100 base pairs. The system’s detection method is
based on the conversion of pyrophosphate—released during the polymerase-catalyzed
attachment of a nucleotide to the complementary strand—into light via an enzyme
cascade.
As the throughput and sensitivity represent a breakthrough relative to the capabilities
of the Sanger method, the new Genome Sequencer 20 System is very much suited not
only for the sequencing of whole genomes of microorganisms, but also for resequencing
of amplicons and small RNA fragments, supporting transcriptome analysis. The ability
to sequence whole genomes in a rapid manner, combined with the ability to identify
and quantify low-abundance DNA subpopulations within a sample, has enabled new
applications. Some recent developments include studies on drug resistance and the
pathogenicity of bacteria; identification of human DNA variations, including low-
abundance mutations in cancer studies; establishing the onset of drug resistance in HIV
or HCV; and differentiating the modes of action of antibiotics.
New-generation sequencing technologies are gradually opening up new applications,
particularly in the medical field and in the discovery and development of drugs.
If sequencing is to become a standard component of individualized diagnosis and
preventive medicine—the next step to bridge basic research tools to patient care—the
costs per sequencing run must be reduced still further. The international planning target
is US$1,000 for a complete human genome. A large number of research institutes
and companies are now working on technologies designed to lower costs and increase
throughput. Roche Diagnostics and 454 Life Sciences Corporation continue working
together to further develop the Genome Sequencer System to enable researchers to open
more doors to new applications and solve previously unaddressable problems.
Timothy Harkins, Ph.D.
Marketing Manager
Roche Diagnostics
Roche Applied Science
3

Sequencing3.indd 5 9/7/06 3:18:25 PM

 ANCIENT DNA

New Methods Yield Mammoth Samples

A
ncient DNA has always held the jawbone. This specimen had been recov-
promise of a visit to a long-vanished ered from the shore of Lake Taimyr, where
world of extinct animals, plants, and very cold winters and short, cool, and dry
even humans. But although researchers summers turned out to be ideal conditions
have sequenced short bits of ancient DNA for preserving DNA.
from organisms including potatoes, cave Poinar sent the DNA-rich sample
bears, and even Neandertals, most samples to genomicist Stephan C. Schuster at
have been too damaged or contaminated Pennsylvania State University, University
for meaningful results. Park, who is working with a new genome
Now in a paper published online by sequencer developed by a team at Stanford
Science this week*, an international team University and 454 Life Sciences Corp.
reports using new technology to sequence of Branford, Connecticut (Nature, 15
a staggering 13 million basepairs of both September, p. 376). This rapid, large-scale
nuclear and mitochondrial DNA from a sequencing technology sidesteps the need
27,000-year-old Siberian mammoth. Also to insert DNA into bacteria before ampli-
this week, a Nature paper reports using a fying and sequencing it. Instead, scientists
souped-up version of more conventional break DNA into small fragments, each at-
methods to sequence a mammoth’s entire tached to a tiny bead and encapsulated by a
mitochondrial genome. lipid bubble where the DNA is multiplied
Besides helping reveal the origins of into many copies for sequencing. Because
mammoths, the new nuclear data serve as a each fragment is isolated before copying,
dramatic demonstration of the power of the the method avoids bias from copying large
new technique to reliably sequence large amounts of contaminant DNA from bacte-
amounts of ancient DNA, other research- ria or humans.
ers say. “The ‘next generation’ sequencer The researchers were stunned by how
that was used [in the Science paper] will well the method worked on ancient DNA,
revolutionize the field of ancient DNA,” which is notoriously difficult to extract
predicts evolutionary biologist Blair and sequence: “I would have been happy if
Hedges of Pennsylvania State University we got 10,000 bases of mammoth DNA,”
in University Park. Ancient DNA pioneer said Poinar. Instead, they got 28 million
Svante Pääbo of the Max Planck Institute basepairs, 13 million from the mammoth
for Evolutionary Anthropology in Leipzig, itself. Their preliminary analysis shows
Germany, who co-led the independent mi- that the mammoth was a female who
tochondrial study, calls the nuclear DNA shared 98.55% of her DNA with modern
work “really great—the way forward in African elephants. But mammoths were
ancient DNA is to go for the nuclear ge- apparently closest kin to Asian elephants,
nome with technologies like this.” as shown by Pääbo’s mitochondrial study,
To get mammoth samples for the new which retrieved about 17,000 basepairs.
method, molecular evolutionary geneticist Poinar’s team also found sequences
Hendrik Poinar of McMaster University in from bacteria, fungi, viruses, soil micro-
Hamilton, Canada, took bone cores from organisms, and plants, which the research-
woolly mammoths found in permafrost and ers say will help reconstruct the mam-
stored in a frigid Siberian ice cave. When moth’s ancient world. The technique was
Poinar returned the samples to his lab, he so productive that the authors predict it
was surprised by the amount of DNA that will be used soon to sequence entire ge-
emerged, particularly from one mammoth nomes of extinct animals.
–ANN GIBBONS
*www.sciencemag.org/cgi/content/abstract/1123360 With reporting by Michael Balter.

Sequencing3.indd 6 9/7/06 3:18:44 PM

 Metagenomics to Paleogenomics:
Large-Scale Sequencing of Mammoth
DNA
Hendrik N. Poinar,1,2,3* Carsten Schwarz,1,2 Ji Qi,4 Beth Shapiro,5 Ross D. E. MacPhee,6
Bernard Buigues,7 Alexei Tikhonov,8 Daniel H. Huson,9 Lynn P. Tomsho,4 Alexander
Auch,9 Markus Rampp,10 Webb Miller,4 Stephan C. Schuster4*
We sequenced 28 million base pairs of DNA in a metagenomics approach, using
a woolly mammoth (Mammuthus primigenius) sample from Siberia. As a result
of exceptional sample preservation and the use of a recently developed emulsion
polymerase chain reaction and pyrosequencing technique, 13 million base pairs
(45.4%) of the sequencing reads were identified as mammoth DNA. Sequence
identity between our data and African elephant (Loxodonta africana) was 98.55%,
consistent with a paleontologically based divergence date of 5 to 6 million years.
The sample includes a surprisingly small diversity of environmental DNAs. The high
percentage of endogenous DNA recoverable from this single mammoth would allow
for completion of its genome, unleashing the field of paleogenomics.

C
omplete genome sequences of ex- ability to retrieve nuclear DNA (nDNA).
tinct species will answer long-stand- Most DNA extracted from fossil remains
ing questions in molecular evolution is truncated into fragments of very short
and allow us to tackle the molecular basis length [<300 base pairs (bp)] from hydro-
of speciation, temporal stages of gene evo- lysis of the DNA backbone, cross-linking
lution, and intermediates of selection dur- due to condensation (1, 2), and oxidation
ing domestication. To date, fossil remains of pyrimidines (3), which prevents exten-
have yielded little genetic insight into sion by Taq DNA polymerase during poly-
evolutionary processes because of poor merase chain reaction (PCR). In addition,
preservation of their DNA and our limited DNA extracts are a mixture of bacterial,
fungal, and often human contaminants,
1
McMaster Ancient DNA Center, 2Department complicating the isolation of endogenous
of Anthropology, 3Department of Pathology DNA. In the past, these problems could
and Molecular Medicine, McMaster University,
only be indirectly overcome by concentrat-
1280 Main Street West, Hamilton, ON L8S 4L9
Canada. 4Pennsylvania State University, Center ing on the small number of genes present
for Comparative Genomics and Bioinformatics, on the maternally inherited mitochondrial
310 Wartik Building, University Park, PA 16802, genome, which is present in high copy
USA. 5Henry Wellcome Ancient Biomolecules
Centre, Department of Zoology, Oxford University,
number in animal cells. This approach se-
South Parks Road, Oxford, OX1 3PS, UK. verely limits access to the storehouses of
6
Division of Vertebrate Zoology/Mammalogy, genetic information potentially available
American Museumof Natural History, 79th Street in fossils of now-extinct species. In a few
and Central Park West, New York, NY 10024,
USA. 7#2 Avenue de la Pelouse, F-94160 St. rare cases, investigators have managed
Mandé, France. 8Zoological Institute, Russian to isolate and characterize nuclear DNA
Academy of Sciences, Universitetskaya nab.1, from fossil remains preserved in arid cave
Saint Petersburg 199034, Russia. 9Center for
Bioinformatics (ZBIT), Institute for Computer
deposits (4–6) or, more commonly, perma-
Science, Tübingen University, 72076 Tübingen, frost-dominated environments (7, 8) and
Germany. 10 Garching Computing Center ice (9), where the average burial tempera-
(RZG), Boltzmannstrasse 2, D-85748 Garching, ture can be as low as –10˚C (10). Under
Germany.
*To whom correspondence should be addressed. these conditions, preservation is enhanced
E-mail: [email protected] (H.N.P.); scs@bx. by reduced reaction rates: In permafrost
psu.edu (S.C.S.)

Sequencing3.indd 7 9/7/06 3:19:02 PM

 settings, theoretical calculations predict extremely cold winters, combined with
DNA fragment survival up to 1 million short, cool summers and little annual pre-
years (11, 12). cipitation, have ensured that conditions
Although more nuclear DNA is pres- optimal for the preservation of bones and
ent in cells than mitochondrial DNA, ac- teeth prevailed there for most or all of the
cess to the nuclear genome even in well- late Pleistocene.
preserved fossil material remains difficult To obtain a better perspective on the
with PCR-based approaches, which must preservational integrity of this sample,
target known sequences from specific we quantitated the number of amplifiable
genes. To find the ideal sample and ana- mtDNA cytochrome b gene fragments
lytical approach for paleogenomics, we of increasing length up to 1 Kb (fig. S1)
screened eight of the morphologically (13). The DNA extract (from 58 mg bone)
best preserved mammoth remains in the contained ~7 × 106 copies of the 84-bp
collections of the Mammoth Museum, a mtDNA fragment in 100 μl or 121 × 106
dedicated permafrost “ice cave” facility copies per gram of bone (13). On the basis
in the town of Khatanga in the southeast- of this, we estimated that the concentration
ern part of the Taimyr Peninsula, Russian of total amplifiable DNA in our sample
Federation, and maintained by Cerpolex/ (assuming a copynumber ratio of 1000:1,
Mammuthus Expeditions (CME). We mtDNA:nDNA) was on the order of 0.73
extracted DNA from these samples, us- μg of mammoth DNA per gram of bone.
ing ancient DNA methodology, to avoid We extracted 1 g of bone and concentrated
to the greatest extent possible exogenous the DNA to 100 μl, which was subse-
DNA (13), because we are well aware of quently used for library construction and
the problems and pitfalls associated with sequencing technology that recently be-
potential contamination. These samples came available (13, 19). This in vitro tech-
were screened with a quantitative PCR nique circumvents amplification or clon-
(qPCR) assay designed for the mam- ing biases by compartmentalizing single
moth mitochondrial cytochrome b gene DNA molecules before the amplification
(14), with primers designed to amplify step in a lipid vesicle, thereby maintaining
mitochondrial DNA molecules from both the original DNA template distribution.
African and Asian elephants (Loxodonta The lipid-enclosed single DNA molecule,
africana and Elephas maximus). We quan- attached to a sepharose bead 28 μm in di-
titated the number of amplifiable mito- ameter, undergoes a PCR reaction, yield-
chondrial DNA (mtDNA) molecules of ing sufficient DNA copies for sequencing.
84 bp in length. The eight samples ranged Sequencing is performed with a pyrose-
from 1 × 106 copies per gram to 96 × 106 quencing methodology (19).
copies per gram, all excellent samples as We obtained 302,692 sequence reads
judged by ancient DNA standards (15). averaging 95 bp, with read length being
However, one sample in particular, CME limited by the sequencing approach rather
2005/915, was exceptional. The specimen, than by the 630-bp average DNA fragment
an edentulous mandible dated to 27,740 ± size obtained from the extract after shear-
220 14C years before the present (uncor- ing. The total sequence data produced for
rected; Beta 210777), was recovered on the bone metagenome was 28 × 106 bp.
the shore of Baikura-turku, a large bay on We aligned the sequencing reads with cur-
the southeastern side of Lake Taimyr, the rent (November 2005) assemblies of the
largest freshwater body in Eurasia north genome sequences of African elephant (L.
of the Arctic Circle. The large numbers africana), human, and dog (Canis familia-
and high quality of late Quaternary fos- ris), downloaded from www.genome.ucsc.
sils recovered from the Taimyr Peninsula edu (Table 1 and Fig. 1). Alignments were
have prompted several major investiga- computed by the program BLASTZ (20),
tions in recent years (16), including stud- with parameters chosen to identify only
ies of ancient DNA (14, 17, 18). Taimyr’s regions of about 90% identity or higher

Sequencing3.indd 8 9/7/06 3:19:17 PM

 Table 1. Total percent of aligned reads and their relative identities to African elephant, human,
and dog.
Elephant Human Dog
Total no. reads 302,692 (100%) 302,692 (100%) 302,692 (100%)
Aligned reads 137,527 (45.4%) 4,237 (1.4%) 3,775 (1.2%)
Uniquely aligning reads 44,442 (14.7%) 3,901 (1.3%) 3,548 (1.2%)
Multiply aligned reads 93,085 (30.8%) 336 (0.1%) 227 (0.1%)
Reads with at least 95% identity 90,507 (30.0%) 1,184 (0.4%) 1,140 (0.4%)
Reads with 100% identity 21,952 (7.3%) 116 (0.04%) 142 (0.05%)
Uniquely aligning base pairs 4,332,350 318,966 291,714
Identity in unique alignments 98.55% 92.68% 92.91%
Mitochondrial reads 209 – –
Identity in mitochondrial reads 95.93% – –
Mitochondrial base pairs 16,419 – –

Fig. 1. Characterization of Mammoth

the mammoth metagenomic (predicted to align)
9.09%
library, including percent-
age of read distributions to Mammoth
various taxa. Host organism (alignable
prediction based on BLASTZ Unidentified to elephant)
comparison against GenBank sequences 45.43%
18.42%
and environmental sequences
database.

Environmental
sequences
14.15% Virus
0.09% Alignable to human
Other eukaroyta 1.40%
4.15% Archaea Bacteria Alignable to dog
0.24% 5.76% 1.25%

(13). As we have not detected any con- read length of 132 bp. To test whether the
vincing mappings of a read to the human Y observed hits were more likely to be de-
chromosome, despite random distribution rived from endogenous mammoth DNA,
across the genome, we conclude that our as opposed to potential contaminants such
mammoth was a female. as human DNA, we repeated the BLASTZ
A total of 137,527, or 45.4% of all analyses as above, this time comparing our
reads, aligned to the African elephant sequence reads to the currently available
genome (Fig. 1, Table 1) (13), currently versions of the human and dog genomes.
available at 2.2-fold coverage, with an Only 4237 reads (1.4%) aligned to human
estimated number of base pairs in the ge- and 3775 (1.2%) to dog (at our threshold
nome of 2.3 × 109 bp (www. broad.mit. of approximately 90% identity). Between
edu/mammals/#chart). A twofold coverage 1% and 5% of any two distantly related
approximates only 80% of the total ge- mammalian genomes should align at 90%
nome, so a conservative estimate is that half identity or greater, because roughly 0.5%
of our reads would align to a completed el- of these genomes consist of protein-coding
ephant sequence. Among all reads, 44,442 segments conserved at that level (21), and
(14.7%) aligned to only one position in noncoding DNA contributes a somewhat
the elephant genome, and 21,952 (7.3%) larger fraction (22). Thus, the fraction of
exhibited a perfect (100%) match, up to a our reads that show at least 90% identity

Sequencing3.indd 9 9/7/06 3:19:36 PM

 with human and dog is what is expected if fest as an excess of elephant C aligned to
only mammoth DNA were sequenced. mammoth T over elephant T aligned to
To further assess the possibility of mammoth C. Indeed, the ratio of the rates
contamination in our DNA sample, we to and from mammoth relative to elephant
explicitly considered what would be ex- is 1.91 (15297/7990) for C-to-T and 1.15
pected if contaminating human DNA (9530/8252) for G-to-A [C-to-T human
sequences were present. If some of our 0.96 (3849/4027); dog 1.11 (3622/3276);
reads were human DNA, then these reads G-to-A human 0.87 (3541/4060); and
should align with nearly 100% identity to dog 0.97 (3321/3440)]. Thus, we find no-
human, and at most, 5% of these could be ticeable deamination of cytosines in our
expected to align at 90% identity or higher extract. In addition, we analyzed the fre-
to any nonprimate mammalian genome quencies with which substitutions likely
(only about 5 to 6% of the genome ap- to be attributable to postmortem damage
pears to be under negative selection (23), occurred in the amplified fragments of
and our 90% threshold is far above the mtDNA by comparison with the publicly
neutral level). If our data contained hu- available mammoth mitochondrion ge-
man reads, we could be essentially certain nome (GenBank accession #NC_007596,
that a large fraction of them would align DQ188829) (13). We found 222 reads that
to human at or above 97% identity over at aligned to the public GenBank mammoth
least 80% of their length and not align to mitochondrial genome (Fig. 2). Two hun-
either elephant or dog. Only 14 reads sat- dred nine reads gave a total of 18,581 bp,
isfy these criteria. Thus, there are, at most, 7617 bp of which were overlapping, re-
trace levels of human contamination in the sulting in a total coverage of 10,964 bp of
sample or the process, and the 14 poten- a possible 16,770 bp (65%). One hundred
tial examples may simply be regions that fourteen of the 209 reads (55%) matched
are well conserved between mammoth and the previously published sequence ex-
all other mammals but that happen to be actly. One hundred nine base differences
missing from the current assemblies of the were observed between the reads and the
dog and elephant genomes. We applied the published sequence, 49 of which were not
same approach to look for potential con- supported by overlapping reads. The ma-
tamination of dog DNA in our sample and jority of these substitutions (84%) were
found none. either C-to-T or G-to-A transitions, as is
To determine substitution patterns expected if the substitutions were due to
between mammoth and other species (ta- postmortem DNA deamination. The re-
ble S2), we used the subset of reads that maining 13 reads differed significantly
aligned to only one position. We observed from the published data and may be evi-
98.55% identity between mammoth and dence of potential nuclear inserts of DNA
elephant. As this number does not cor- from the mitochondrion (13), which have
rect for alterations of the sequence due to been reported previously to be com-mon
damage caused by base decomposition, we in elephants and mammoths (26).
are likely underestimating the amount of These findings are well within the pre-
sequence similarity. Base damage in fos- dicted levels of damage for ancient DNA
sil matrices can result in a myriad of base and demonstrate the feasibility and ben-
changes; however, the most commonly ob- efits of ancient whole-genome sequencing
served change has been deamination of cy- without previous amplification, as over-
tosine to uracil (24, 25) resulting in C-to-T lapping reads from a multifold coverage
and G-to-A substitutions. We therefore would easily correct for decompositional
looked for asymmetries in base composi- base changes accrued during the sample’s
tion between modern and ancient genomes depositional history and serve as a DNA
(table S2). We reasoned that C-to-T tran- damage correction filter. The ratio of
sitions due to DNA damage would mani- mtDNA to nuclear DNA for our sample

Sequencing3.indd 10 9/7/06 3:33:12 PM

 Fig. 2. Distribution of 209
reads along the mam-
moth mitochondrial ge-
nome (GenBank accession
#NC_007596, DQ188829).
Average fragment length
was 89 bp; not shown to
scale. To determine whether
the reads were randomly
distributed along the ge-
nome, we compared the
distribution of fragment
lengths that would result
from cutting the genome
at the 5¶ end of each read
against that resulting from
100 million randomly gen-
erated distributions of 209
reads. Despite multiple
overlapping regions, the
real distribution was not
significantly different from
the empirical distribution
of fragment lengths (P 0
0.069).

was 1:658, which agrees with what one reads (100%) against the nonredundant
would expect given a 1:1000 copy-number (nr) and the environmental database (env_
ratio for nDNA versus mtDNA. nr). Using an adjustable factor for bitscore
Despite the presence in our sample of (13), we classified the reads simultane-
an exceptionally high percentage (54.5%, ously to the individual kingdom, phylum,
including reads predicted to align to el- class, order, family, and genus down to the
ephant) of mammoth DNA, relative to species level wherever possible (fig. S2
environmental contaminants, 45.5% of and table S3), excluding all hits matching
the total DNA derives from endogenous Gnathostomata (jawed vertebrates).
bacteria and nonelephantid environmental The remaining 12,563 hits within the
contaminants. In addition to ubiquitous Eukaryota (4.15%) were only surpassed
contaminants resulting from handling or by the number of bacterial hits, 17,425
conditions of storage, these exogenous (5.76%), and hits against the environ-
species are likely to represent taxa pres- mental database, 42,816 (14.15%). The
ent at or immediately after the time of the kingdom Archaea was hit infrequently
mammoth’s death, thereby contributing to with only 736 hits (0.24%). Within this
the decomposition of the remains. To ac- group, the Euryarchaeota dominated the
quire a glimpse of the biodiversity of these Crenarchaeota by a ratio of 16:1. In the
communities, we have devised software bacterial superkingdom, the most preva-
(GenomeTaxonomyBrowser) (27, 28) that lent species were found to be proteobacte-
allows for the taxonomic identification of ria, 5282 (1.75%); Firmicutes (gram-posi-
various species on the basis of sequence tives), 940 (0.31%), mostly Bacilli and
comparison and current phylogenetic Chlostridia; Actinobacteria, 2740 (0.91%);
classification at the National Center for Bacteroidetes, 497 (0.16%); and the group
Biotechnology Information (NCBI) tax- of the Chlorobi bacteria, 248 (0.08%).
onomy browser as of November 2005 Other identified microorganisms included
(www.ncbi.nlm.nih.gov/Taxonomy/tax- the fungal taxa Ashbya, Aspergillus, and
onomyhome.html). We compared 302,692 Neurospora/Magnaporte with 440 hits

Sequencing3.indd 11 9/7/06 3:33:24 PM

 (0.14%). We also found 278 hits against References and Notes
1. S. Pääbo et al., Annu. Rev. Genet. 38, 645 (2004).
viral sequences, which could be assigned 2. H. N. Poinar et al., Science 281, 402 (1998).
to dsDNA viruses, 193 hits (0.06%); ret- 3. M. Höss, P. Jaruga, T. H. Zastawny, M. Dizdaroglu, S. Pääbo,
rotranscribing viruses, 20 hits (0.01%); Nucleic Acids Res. 24, 1304 (1996).
4. H. N. Poinar, M. Kuch, G. McDonald, P. S. Martin, S. Pääbo,
and ssRNA viruses, 46 hits (0.02%). Curr. Biol. 13, 1150 (2003).
The soil-inhabiting eukaryotic species, 5. V. Jaenicke-Despres et al., Science 302, 1206 (2003).
Dictyosteliida, with 127 hits (0.04%), and 6. M. Bunce et al., Nature 425, 172 (2003).
7. M. Höss, S. Pääbo, N. K. Vereshchagin, Nature 370, 333
Entamoeba, with 64 hits (0.02%), were (1994).
found to be underrepresented, as were 8. A. D. Greenwood, C. Capelli, G. Possnert, S. Pääbo,
Mol. Biol. Evol. 16, 1466 (1999).
hits against the two nematode genomes, 9. E. Willerslev, A. J. Hansen, B. Christensen, J. P.
277 hits (0.09%). A detailed identification Steffensen, P. Arctander, Proc. Natl. Acad. Sci. U.S.A. 96,
of plant species is handicapped, because 8017 (1999).
10. E. Willerslev, A. J. Hansen, H. N. Poinar, Trends Ecol. Evol.
presently only the two plant genomes, 19, 140 (2003).
Arabidopsis thaliana (mouse-ear cress) 11. C. I. Smith, A. T. Chamberlain, M. S. Riley, C. Stringer,
and Oryza sativa (rice), are publicly availa- M. J. Collins, J. Hum. Evol. 45, 203 (2003).
12. H. N. Poinar, M. Hoss, J. L. Bada, S. Pääbo, Science 272,
ble. We found the hits against grass species 864 (1996).
to outnumber the ones from Brassicales by 13. Materials and methods are available as supporting
material on Science Online.
a ratio of 3:1, which could be indicative of 14. R. Debruyne, V. Barriel, P. Tassy, Mol. Phylogenet. Evol.
ancient pastures on which the mammoth is 26, 421 (2003).
believed to have grazed. 15. O. Handt, M. Krings, R. H. Ward, S. Pääbo, Am. J. Hum.
Genet. 59, 368 (1996).
From this classification, it is evident 16. R. D. E. MacPhee et al., J. Arch. Sci. 29, 1017 (2002).
that nonvertebrate eukaryotic and pro- 17. R. MacPhee, A. Tikhonov, D. Mol, A. D. Greenwood,
BMC Evol. Biol. 5, 49 (2005).
karyotic species occur at approximately
18. A. D. Greenwood, C. Capelli, G. Possnert, S. Pääbo, Mol.
equal ratios, with the mammalian fraction Biol. Evol. 16, 1466 (1999).
dominating the identifiable fraction of the 19. M. Margulies et al., Nature 437, 376 (2005).
20. S. Schwartz et al., Genome Res. 13, 103 (2003).
metagenome. The paucity of fungal spe- 21. W. Makalowski, J. H. Zhang, M. S. Boguski, Genome Res.
cies is surprising, as is the low number of 6, 846 (1996).
reads from nematodes. 22. A. Siepel et al., Genome Res. 15, 1034 (2005).
23. R. H. Waterston et al., Nature 420, 520 (2002).
Recently, a whole-genome approach 24. T. Lindahl, Nature 362, 709 (1993).
was attempted from DNA of the ex- 25. M. Hofreiter, V. Jaenicke, D. Serre, A. von Haeseler, S. Pääbo,
Nucleic Acids Res. 29, 4793 (2001).
tinct cave bear Ursus spelaeus, yielding 26. A. D. Greenwood, S. Pääbo, Mol. Ecol. 8, 133 (1999).
~27,000 bp of endogenous genetic mate- 27. D. H. Huson, A. Auch, J. Qi, S. C. Schuster, in preparation.
rial from 1.1 to 5.8% of all DNA reads 28. GenomeTaxonomyBrowser will be made available to
readers upon request.
(29). We have produced 13 million bp of 29. J. P. Noonan et al., Science 309, 597 (2005).
endogenous genetic material from 45% of 30. We thank D. Poinar, C. Flemming, and E. Willerslev
for help in mammoth sampling; N. E. Wittekindt and
all DNA reads, some 480 times as much A. Rambaut for help with the manuscript; and two anonymous
DNA sequence and 15 times the percent- reviewers. We also thank the Natural Sciences and
age. The ability to obtain this level of Environmental Research Council of Canada (299103-2004)
for a grant to H.N.P. and McMaster University for financial
genetic information from extinct species support. R.D.E.M. was supported by NSF OPP 0117400,
makes it possible to consider detailed B.S. was supported by the Wellcome Trust, and W.M. was
supported by NIH grant HG02238. S.C.S. thanks The
analysis of functional genes and fine- Pennsylvania State University for initial funding.
scale refinement of mutation rates. A rapid
identification assay of single nucleotide
polymorphisms (SNPs) would be of great Supporting Online Material
value for studying population genetics of www.sciencemag.org/cgi/content/full/1123360/DC1
Materials and Methods
Pleistocene mammals and plants, which in Figs. S1 and S2
turn could help elucidate their responses Tables S1 to S4
to climate changes during late glacial and References

early postglacial time and ultimately shed 2 December 2005; accepted 15 December 2005
Published online 20 December 2005;
new light on the cause and consequences 10.1126/science.1123360
of late Quaternary extinctions. Include this information when citing this paper.

10

Sequencing3.indd 12 9/7/06 3:23:41 PM

 The Race for the $1000 Genome
Fast, cheap genetic analyses will soon become a reality, and
the consequences—good and bad—will affect everybody

MARCO ISLAND, FLORIDA—Computers are getting more confident that it’s a real
aren’t the only things getting better and possibility. “From what I’ve listened to
cheaper every time you turn around. the last few days, there is no physical
Genome-sequencing prices are in free fall, principle that says we shouldn’t be able to do
too. The initial draft of the first human a $1000 genome,” says Harvard University
genome sequence, finished just 5 years sequencing pioneer George Church.
ago, cost an estimated $300 million. (The Even today, the declining cost of
final draft and all the technology that genome sequencing is triggering
made it possible came in near $3 billion.) a flowering of basic research, looking
Last month, genome scientists completed at broad-ranging topics such as how
a draft of the genome sequence of the the activation of genes is regulated and
second nonhuman primate—the rhesus understanding genetic links to cancer. And
macaque—for $22 million. And by the end as prices continue to drop, sequencing will
of the year, at least one company expects revolutionize both the way biologists hunt
to turn out a full mammalian genome for disease genes and the way medical
sequence for about $100,000, a 3000-fold professionals diagnose and treat diseases.
cost reduction in just 6 years. In fact, some researchers say cheap
It’s not likely to stop there. Researchers sequencing technology could finally usher
are closing in on a new generation of in personalized medicine in a major way.
technology that they hope will slash the “The promise of cheap sequencing
cost of a genome sequence to $1000. is in the understanding of disease and
“Advances in this field are happening biology, such as cancer, where the genome
fast,” says Kevin McKernan, co–chief changes over time,” says Dennis Gilbert,
scientist at Agencourt Bioscience in chief scientist of Applied Biosystems,
Beverly, Massachusetts. “And they are the leading gene-sequencing-technology
coming more quickly than I think anyone company based in Foster City, California.
was anticipating.” Jeffrey Schloss, who “It will enable different kinds of science to
heads the sequencing-technologies grant be done.” Of course, as with other forms
program at the National Human Genome of high technology, that promise brings
Research Institute (NHGRI) in Bethesda, new risks as well. Researchers expect
Maryland, agrees. “People are roundly cheap sequencing to raise concerns about
encouraged and nervous,” Schloss says— the proliferation of bioterrorism agents as
encouraged because their own technologies well as patient privacy.
are working, and nervous because their
competitors’are too. The race is on
A host of these novel sequencing The first group to produce a technology
technologies were on display last month capable of sequencing a human genome
at a meeting here.* Although no one at sequence for $1000 will get instant
the meeting claimed to have cracked the gratification, as well as potential future
$1000 genome sequence yet, researchers profits: In September 2003, the J. Craig
Venter Science Foundation promised
$500,000 for the achievement. That
*Advances in Genome Biology and Technology challenge has since been picked up by
Conference, Marco Island, Florida, 8-11 February the Santa Monica, California–based
2006.
X Prize Foundation, which is expected
This News Focus article was published in the 7 to up the ante to between $5 million
March 2006 issue of Science. and $20 million. But the competition

11

Sequencing3.indd 13 9/7/06 3:24:02 PM

 A; ADAPTED FROM GRAPH PROVIDED BY JEFFREY SCHLOSS/NHGRI
Decrease in the Cost of Finished DNA Sequencing
10.00

Cost per finished bp

1.00

0.10

0.01
1990 1995 2000 2005

Free fall. As with computer technology, the plunging cost of DNA

sequencing has opened new applications in science and medicine.

really began in earnest in 2004, when the the mix of reagents used to perform the
National Institutes of Health launched a synthesis—stop the process when one of
$70 million grant program to support them is tacked onto the end of the growing
researchers working to sequence a complete DNA strand. The result is a soup of newly
mammal-sized genome initially for synthesized DNA fragments, each of
$100,000 and ultimately for $1000. That which started at the same point but ends at
program has had an “amazing” effect a different base along the chain.
on the field, encouraging researchers to Today’s sequencers separate these
pursue a wide variety of new ideas, says fragments by passing the soup through tiny
Church. That boost in turn has led to a capillaries containing a gel; the shorter
miniexplosion of start-up companies, each the fragment, the faster it moves through
pursuing its own angle on the technology the gel. The process, known as capillary
(see table, p. 15). electrophoresis, is so effective that each
All are racing to improve or replace a fragment that emerges from the capillary
technology first developed by Fred Sanger is just one base longer than the one that
of the U.K. Medical Research Council in preceded it. As each fragment emerges, it
the mid-1970s that is the basis of today’s is hit by a laser, which causes the altered
sequencing machines. The technique base at the fragment’s tip to fluoresce. A
involves making multiple copies of the computer records the identity of these bases
DNA to be sequenced, chopping it up and the sequence in which they appear.
into small pieces, and using those pieces Eventually, the process generates billions
as templates to synthesize short strands of stretches of sequence that are fed into
of DNA that will be exact complements pattern-recognition software running on a
of stretches of the original sequence. The supercomputer, which picks out overlaps
synthesis essentially mimics the cell’s and stitches the pieces together into a
processes for copying DNA. complete genome sequence.
The technology relies on the use of A long list of refinements in capillary
modified versions of the four bases that electrophoresis systems, coupled with
make up DNA, each of which is tagged increased automation and software
with a different fluorescent marker. A short improvements, has driven down the
DNA snippet called a primer initiates the costs of sequencing 13-fold since these
synthesis at a specific point on the template machines were introduced in the 1990s.
DNA, and the altered bases—which are Most of the new technologies aim to
vastly outnumbered by normal bases in miniaturize, multiplex, and automate the

12

Sequencing3.indd 14 9/7/06 3:24:26 PM

 process even further. They fall into three quality,” Egholm says. Higher accuracy
main camps. The first, called sequencing is critical because two genomes being
by synthesis, tracks bases as they are compared, such as those of normal cells
added to a growing DNA strand. Second and cancer cells, could differ in only one
is a group of techniques that sequence part per million.
single DNA molecules. Finally, nanopore- David Bentley, chief scientist for
sequencing technologies coax DNA to Solexa in Little Chesterford, U.K., also
wriggle through a tiny pore and read the reported heady progress. Like 454’s
bases either electronically or optically as approach, Solexa’s turns separate snippets
they go by. into roughly 1000 exact copies. Instead
Sequencing-by-synthesis strategies of attaching individual DNA strands to a
have a head start. Indeed, one company, separate bead, Solexa researchers fix each
454 Life Sciences Corp. in Branford, strand to a different spot on a glass slide,
Connecticut, already has a commercial much as they do in standard microarrays.
instrument; it sold 20 of them last They then duplicate those strands,
year. The company’s technique, called creating myriad tiny DNA islands. Finally,
pyrosequencing, first chops a genome into in a step akin to Sanger sequencing,
stretches 300 to 500 base pairs long, unzips they use nucleotides with four different
the double strands, discards one strand, colors and standard microarray optics
and links the other to compounds tethered to simultaneously track the growth of
to a plastic bead—each bead gets just one strands complementary to those attached
strand. These snippets are then copied by to the slide. Bentley reported that his team
the polymerase chain reaction (PCR) until had sequenced a 162-kilobase stretch
the copies cover each bead. The beads are of human DNA and compared it to the
separated on a plate containing as many as standard reference sequence worked out
1.6 million wells and dosed with a series by the Human Genome Project. Their
of sequencing reagents and nucleotides. sequencing turned out to be more than
Every time a nucleotide is tacked onto 99.99% accurate and spotted all 162
a growing DNA chain, the reaction common mutation sites known as single-
triggers the release of a compound called nucleotide polymorphisms known to exist
pyrophosphate, which in turn prompts a in that stretch of DNA.
firefly enzyme called luciferase in the well Church has developed a slightly
to give off a flash of light. By correlating different sequencing approach, part of
the recorded flashes from each cell with which Harvard has licensed to Agencourt.
the nucleotides present at the time, a In this approach, called sequencing by
computer tracks the systematic sequence ligation, researchers start with a short
growth of hundreds of thousands of DNA snippet of DNA bound to a bead or a
snippets simultaneously. surface. They then add a short stretch of
In August 2005, 454 Life Sciences DNA called an anchor primer that binds
researchers reported that they had a known starter sequence on the DNA
sequenced the nearly 600,000-base genome snippet. Additional nine-base primers,
of a bacterium known as Mycoplasma known as query primers, are then added
genitalium with an accuracy of 99.4%, as to the mix. These primers come in each
well as the larger 2.1-megabase genome possible sequence combination, and each
of Streptococcus pneumoniae (Science, has a labeled A, G, T, or C at just one
5 August 2005, p. 862). At the Florida position. If a short primer with a correct
meeting, Michael Egholm, 454’s vice complementary sequence binds to the
president for molecular biology, reported DNA, an enzyme called ligase stitches it to
that they had since sequenced four different the anchor primer to hold it to the surface,
microbial genomes, each with greater than and the other primers, which bind less
99.99% accuracy. “In a 6-month period, tightly, are washed off. The mix is then hit
we have dramatically improved the data with a blast of laser light to reveal the color

13

Sequencing3.indd 15 9/7/06 3:24:51 PM

 of fluorescence that gives away the identity nucleotide and the time when it cuts it off
of the newly bound base. Finally, the query the nanoparticle, the researchers reverse
and anchor primers are stripped away, and the electric field, driving away nucleotide-
another anchor primer is added as the first nanoparticle pairs not bound to the DNA.
step to identifying the next base in the Then they snap a picture to see the color
template strand. Agencourt’s McKernan of the fluorescent particles still bound
said their version of the technology could to the polymerase. Once the nucleotide
currently sequence some 200 million is cut free, the nanoparticle drifts away,
bases a day and may reach 3 billion a day and the process is repeated to identify the
by August. next base. At the meeting LI-COR’s John
Williams predicted that this technique
Slow start, strong finish? could produce read lengths of up to 20,000
Despite these advances, sequencing by bases.
synthesis has its drawbacks. One is that But another technology altogether may
the techniques read relatively short DNA hold the most revolutionary potential,
snippets—usually several hundred base Church says. Called nanopore sequencing,
pairs in length or less, compared with the this family of techniques aims to sequence
1000 or so in current capillary systems. DNA strands as they thread their way
That can make it hard to reassemble all the through tiny synthetic or natural pores,
pieces into a continuous genome sequence. each just 1.5 nanometers or so across.
Another drawback is that they rely on Numerous groups are pursuing nanopore
PCR, which is expensive and can introduce synthesis techniques, but researchers
copying errors. Greater experience with acknowledge that they have far to go.
the new sequencing technologies may “We’re still learning about the science of
improve matters. 454’s Egholm, for nanopores,” Schloss says.
example, says his team has developed a No group has yet reported using such a
prototype version of their technology that setup to sequence DNA one base at a time.
increases read lengths from 100 base pairs But in a series of papers beginning in 1996,
to 400. Several groups are developing ways researchers led by John Kasianowicz and
to sequence a single copy of a long DNA Daniel Branton at the National Institute of
strand, thereby achieving longer reads and Standards and Technology in Gaithersburg,
avoiding PCR. Maryland, reported that they could use
One approach being pursued by protein-based pores embedded in a lipid
VisiGen Biotechnologies in Houston, membrane first to detect snippets of DNA
Texas, anchors a polymerase—the enzyme and then to differentiate snippets with
that tacks new nucleotides to a growing all A’s from those made up of C’s. But
DNA chain—onto a surface and feeds because proteins and lipids are fragile,
it a template strand. As the polymerase other groups have begun making their
then adds fluorescently labeled bases to pores out of silicon and other electronic
a complementary strand, an advanced materials in hopes of developing a more
optical system detects the tiny flashes from robust technology that can also integrate
the single molecule, allowing a continuous transistors and other electronic devices.
sequence to be read. A variation of this In most versions of nanopore technology,
approach by LI-COR Biosciences in researchers use tiny transistors to control
Lincoln,Nebraska,anchors single-stranded a current passing across the pore. As the
DNA and polymerase molecules to an four different DNA bases go through, they
electrode, and then uses an electric field perturb that electric signal in different
to drive nucleotides linked to fluorescent ways, causing the voltage to spike upward
nanoparticles in solution toward the or drop in a way that identifies the
polymerase. In the instant between the nucleotide passing through.
time when the polymerase latches onto the At the meeting, for example, chemist

14

Sequencing3.indd 16 9/7/06 3:25:08 PM

 Generation next. Companies racing for the $1000 genome sequence strive simultaneously for low cost, high
accuracy, the ability to read long stretches of DNA, and high throughput.

Gregory Timp of the University of Illinois, strains in patients at the earliest stages. In
Urbana-Champaign, reported that his team another study, they quickly analyzed the
has generated electrical readings of DNA sequence of non–small cell lung cancer
moving through nanopores. Unfortunately, cells and identified the specific mutations
the DNA wriggled back and forth so much that give rise to drug resistance.
that the researchers had trouble teasing In similar studies, Thomas Albert
out the sequence of bases in the chain. and colleagues at NimbleGen Systems,
But Timp says he and his colleagues are a biotechnology firm in Madison,
finishing a second-generation device that Wisconsin, used their version of
uses electric fields to keep the movement sequencing-by-synthesis technology to
of the DNA under control. If it works, the identify the mutations in Helicobacter
technology promises to read long stretches pylori—the microbe responsible for
of DNA without the need for expensive ulcers—that cause resistance to a drug
optical detectors. known as metronidazole, as well as the
mutations in the tuberculosis-causing
“We have to worry now” bacterium that trigger resistance to a new
No matter which technology or tech- TB drug. The power of such studies is
nologies make it to market, the scien- “unbelievable,” Snyder says, because they
tific consequences of lower sequencing hold out the hope of enabling doctors to
costs are bound to be enormous. “I think tailor medicines to battle diseases most
it’s going to have a profound impact on effectively. Some personalized-treatment
biology,” says Yale University molecular strategies are already in use: Herceptin,
biologist Michael Snyder. for example, is targeted to patients with
Some early progress is already on a specific genetic form of breast cancer.
display. At the Florida meeting, for example, But cheap sequencing should make them
454’s Egholm reported that he and his far more widespread, Church says. Basic
colleagues used their technology to researchers are looking at the early benefits
identify as many as four genetic variants of cheap sequencing as well. At the
of HIV in single blood samples, in contrast meeting, for example, Snyder talked about
to today’s technology, which identifies his team’s use of gene chips to map the sites
just the dominant strain. The technique, where transcription factors—proteins that
Egholm says, could eventually help control when genes are turned on—bind to
doctors see the rise of drug-resistant HIV the genome. The technology is effective,

15

Sequencing3.indd 17 9/7/06 3:25:25 PM

 but gene chips are expensive. So Snyder to screen out those at high risk for disease.
is turning to cheap sequencing technology Finally, the same sequencing technology
to rapidly sequence the millions of DNA that could potentially help create beneficial
fragments needed to identify transcription new microbes, such as ones tailored to
factors. Church says he is also using turn out large amounts of hydrogen gas to
cheap sequencing techniques to propel his power fuel cells, could also make it easier
group’s synthetic-biology efforts to create to create new bioterrorist pathogens.
an extensive tool kit of microbial “parts” “We have to worry about these issues
that can be mixed and matched to enable now, because we will be sequencing with
microbes to perform new functions. very high throughput in 10 years,” Timp
Like most new technologies, ultracheap says. Schloss notes that NHGRI has long
sequencing casts shadows as well. For supported research on ethical, legal, and
starters, Church says, it’s hard to imagine social concerns. However, he adds, “it’s
what privacy will mean once anyone with very hard to do it in the abstract.” With
a sample of your DNA can determine who technology advancing at a rapid clip,
you are, your heritage, and what diseases neither the benefits nor the concerns of
you’re likely to inherit. Celebrities and ultracheap sequencing are likely to remain
politicians may soon face a world hungry abstract for long.
to scrutinize their genes. Among ordinary
people, many analysts worry that insurers
and employers will use genetic information —ROBERT F. SERVICE

No Longer De-Identified
Amy L. McGuire1* and Richard A. Gibbs2

A
s DNA sequencing becomes more the privacy of subjects whose sequenced
afford able and less time-consuming, DNA is publicly released have largely
scientists are adding DNA banking been addressed by ensuring that the data
and analysis to research protocols, resulting are “de-identified” and that confidentiality
in new disease specific DNA databases. A is maintained (1–2). There is a large
major ethical and policy question will be literature on the various data-management
whether and how much information about models and computer algorithms that
a particular individual’s DNA sequence can be used to provide access to genetic
ought to be publicly accessible. data while purportedly protecting privacy
Without privacy protection, public trust (3–6). We believe that minimizing risks
will be compromised, and the scientific and to subjects through new developments in
medical potential of the technology will data and database structures is crucial and
not be realized. However, scientific utility should continue to be explored, but that
grows with increased access to sequenced additional safeguards are required.
DNA. At present, ethical concerns about Scientists have been aware for years of
the possibility that coded or “anonymized”
sequenced DNA may be more readily
1 linked to an individual as genetic databases
Center for Medical Ethics and Health Policy,
Baylor College of Medicine, 2Human Genome proliferate (1, 3, 7, 8). In 2004, Lin and
Sequencing Center, Baylor College of Medicine,
colleagues demonstrated that an individual
One Baylor Plaza, Suite 310D, Houston, TX
77030, USA. *Author for correspondence. E-mail: can be uniquely identified with access to
[email protected] just 75 single-nucleotide polymorphisms

16

Sequencing3.indd 18 9/7/06 3:25:52 PM

 (SNPs) from that person (9). Genomewide may choose to impose more stringent
association studies routinely use more than requirements on investigators, but there is
100,000 SNPs to genotype individuals. no federal mandate to do so now.
Although individual identification from To resolve the tension between privacy
the public release of these data would protection and access to DNA data
currently require a reference sample, the needed for progress in medical research,
privacy risk associated with public data Lin and colleagues recommend a tiered
release is fueled by the extraordinary pace data-access approach, with sensitive
of technological developments and the data layers masked from full public view
rapid proliferation of electronic databases. (9). This approach has the advantage
If protective measures are not adopted of minimizing privacy risks without
now, public trust will be compromised, unduly sacrificing progress, but suffers
and genomic research will suffer. from a lack of flexibility to respond to
Genetic sequencing typically involves individual preferences and judgments. It
three phases of investigation: (i) subject also threatens to slow the pace of research,
recruitment and sample collection because current policy calls for sequenced
(primary clinical investigation), (ii) DNA data to be released publicly within 24
sequencing and data broadcast (genomic hours of their generation (17), whereas
sequencing study), and (iii) data retrieval obtaining approval to access restricted
and analysis (secondary-use research). databases could take weeks, if not months.
Institutional Review Board (IRB) oversight We believe that restricted access should be
and informed consent are unambiguously offered as an option to subjects, but that it
required for the first phase of sample not be adopted as a general approach for
collection, because it clearly involves all genomic sequencing studies.
human subjects research. There are also Kohane and Altman propose that
detailed consent requirements for some researchers specifically seek out volunteers
large-scale sequencing studies, such as the who are most willing to have their health
HapMap project, that cover the second and data publicly shared and that these subjects
third phases. However, it is our experience have explicit control over who has access
that, in general, the consent process for to their data (18). Relying only on such
most disease-specific genetic research is “information altruists” to participate in
not protective for these phases and that the genetic studies would potentially create
privacy risks associated with public data- subject bias, influencing the ability of
sharing are not stated. Consent for these investigators to identify disease alleles
studies is highly variable, and in most relevant to the population at large.
cases, subjects are simply told that genetic We propose that general safeguards be
analysis will be performed, without any put in place to encourage understanding
explanation of what that means or with of and trust in genomic sequencing
whom the resulting data will be shared. studies. As an essential first step, genomic
Further, participants are typically not sequencing studies should be recognized
offered the opportunity to participate in as human subjects research and brought
the research if they do not want their data unambiguously under the protection of
publicly broadcast (10). existing federal regulation. This would
In the United States, there are now two have the effect of mandating informed
federal regulations that could potentially consent for public release of potentially
apply to such studies—the Common identifiable sequenced data and requiring
Rule, which regulates all federally funded IRB oversight to ensure that risks to
research and sets forth the federal policy for subjects are minimized and that informed
the protection of human research subjects consent, or a waiver of the requirement to
(11) and the Health Insurance Portability obtain informed consent, is obtained (11).
and Accountability Act (HIPAA) Privacy Specifically, we recommend a stratified
Rule, which restricts certain unauthorized consent process in which all subjects who

17

Sequencing3.indd 19 9/7/06 3:26:08 PM

 SUGGESTED DATA RELEASE OPTIONS
Data released Identifiers Privacy risk Benefit
Option 1 Multiple gene loci >75 SNPs Higher Ability to study interactions
among different genes

Option 2 Single gene loci Typically <20 SNPs Intermediate Ability to study individual genes

Option 3 No data released None Low Ability to include subjects with

low risk tolerance, which increases
generalizability and applicability
to specific subsets of the population

uses and disclosures of patients’ participate in future genomic sequencing

identifiable protected health information studies are fully informed about how their
by covered entities (12). Neither one DNA data may be broadcast and have
specifically mandates IRB oversight or the authority to decide with whom they
subject consent for the public release want their data shared (19). A number of
of sequenced data. The Common Rule options could be presented; we propose
would not apply if genomic sequencing three levels of confidentiality (see table
studies were not considered to constitute above). A more rigorous assessment of
human subjects research. Human subjects the risks and benefits associated with each
research is defined under the Common of these options will be necessary.
Rule as research involving “an individual Some of the practical challenges
about whom the investigator … obtains include providing adequate disclosure
data through intervention or interaction and education about a complex risk
with the individual, or identifiable private calculus, ensuring subject comprehension,
information” (11).According to a guidance coordinating a system of restricted
document published in 2004 by the Office access, and managing a complex database
for Human Research Protections (OHRP), that accounts for subjects’ informed
because the data are collected and coded disclosure preferences. Although it may
by the primary clinical investigator, and represent a substantial departure from the
the sequencing investigator is prohibited traditional model of informed consent in
from deciphering the code, the data research, stratified consent procedures
are not considered identifiable, and the are commonly used in clinical medicine,
sequencing study is not subject to federal where patients frequently make informed
regulation (13). Brief IRB review may be choices about treatment options on the
necessary to confirm that the research does basis of individual values and judgments.
not involve human subjects, but once that Stratified consent procedures are also
determination is made, no IRB oversight being considered in other areas of research
or informed consent is mandated. where subjects have to make complicated
Similarly, HIPAA does not provide decisions, such as what type of future
unambiguous protection because it research they are willing to participate
has not been clear that genomic data in and to what extent they want research-
constitute readily identifiable protected related incidental findings reported back
health information (14–16), and even if to them.
they do, institutions vary in whether the There may be concern about the
scientists who conduct sequencing studies added burdens on IRBs. McWilliams
are considered “covered entities” who and colleagues have shown that there is
must comply with HIPAA. Institutions currently considerable variability among

18

Sequencing3.indd 20 9/7/06 3:26:26 PM

 local IRBs, particularly in how they deal 2. The International HapMap Consortium, Nat. Genet. 5,
467 (2004).
with DNA banking, risk-benefit analysis, 3. L. Sweeney, J. Law Med. Ethics 25, 98 (1997).
and consent for ge netic research (20). 4. B. A. Malin, J. Am. Med. Inform. Assoc. 12, 28 (2005).
They recommend centralized IRB 5. J. E. Wylie, G. P. Mineau, Trends Biotechnol. 21, 113 (2003).
6. L. Sweeney, Ed., “Navigating computer science research
oversight for multicenter research. through waves of privacy concerns: Discussions among
Although some might fear a negative computer scientists at Carnegie Mellon University” ACM
impact on subject participation in genomic Comput. Soc. 34 (April 2004); (http://privacy.cs.cmu.edu/
dataprivacy/projects/csresearch1.pdf).
research, stratified consent merely 7. B. Malin, L. Sweeney, J. Biomed. Inform. 37, 179 (2004).
restricts the ability to release sequenced 8. National Cancer Institute (NCI), NIH, “Confidentiality,
data security, and cancer research: Perspectives from the
data publicly. If anything, it may boost
National Cancer Institute” (NCI, Bethesda, MD, 23 March
enrollment by providing an opportunity 1999); (www3.cancer.gov/confidentiality.html).
for even the most risk-averse members of 9. Z. Lin, A. B. Owen, R. B. Altman, Science 305, 183 (2004).
10. The International HapMap Consortium, “Template
society to participate in research, while consent form” (www.hapmap.org/consent.html.en).
ensuring optimal privacy protection. 11. “Protection of human subjects,” 45 Code of Federal
Empirical study of these and Regulations (C.F.R.) § 46 (2005).
12. “Security and privacy,” 45 C.F.R § 164 (2002).
other challenges associated with the 13. Office for Human Research Protections, U.S. Department
implementation of a stratified consent of Health and Human Services (HHS), “Guidance on
model in research is essential for research involving coded private information or biologi-
cal specimens” (HHS, Washington, DC, 10 August 2004);
future policy development. In addition, (www.hhs.gov/ohrp/humansubjects/guidance/cdebiol.pdf).
federal legislation prohibiting genetic 14. HHS, “Research repositories, databases, and the HIPAA
discrimination would significantly alter Privacy Rule” (HHS, Washington, DC, January 2004);
(http://privacyruleandresearch.nih.gov/pdf/research_
the risk-benefit calculus associated with repositories_final.pdf).
public data release and should be enacted 15. P. A. Roche, Science 307, 1200 (2005).
without delay (21). Although it would not 16. M. A. Rothstein, J. Law Med. Ethics 33, 89 (2005).
17. NHGRI, “Reaffirmation and extension of NHGRI rapid
obviate the ethical obligation to obtain data release policies: Large-scale sequencing and other
subject consent, it may foster public trust community resource projects” (NIH, Bethesda, MD,
and positively affect the willingness of February 2003); (www.genome.gov/10506537).
18. I. S. Kohane, R. B. Altman, N. Engl. J. Med. 353, 2074
subjects to participate in genomic research (2005).
and to share their genetic data publicly. 19. Human Genome Organization (HUGO) Ethics Committee,
“Statement on human genomic databases” (HUGO,
London, December 2002);
(www.gene.ucl.ac.uk/hugo/HEC_Dec02.html).
References and Notes 20. R. McWilliams et al., JAMA 290, 360 (2003).
1. National Human Genome Research Institute (NHGRI), U.S. 21. Genetic Information Nondiscrimination Act of 2005, S.
National Institutes of Health (NIH), NHGRI-DOE [U. S. 306 (109th Congress).
Department of Energy] Guidance on Human Subjects 22. We thank S. E. McGuire, B. A. Brody, L. B. McCullough,
Issues in Large-Scale DNA Sequencing (17 August 1996, M. A. Majumder, and R. R. Sharp. R.A.G. is supported by
updated 27 April 1998) (www.genome.gov/10000921). grants from the NHGRI.

19

Sequencing3.indd 21 9/7/06 3:27:47 PM

 Finding Criminals Through DNA
of Their Relatives
Frederick R. Bieber,1* Charles H. Brenner,2 David Lazer3
Analyses of the DNA databases maintained by criminal justice systems might
enable criminals to be caught by recognizing their kin, but this raises civil liberties
issues.

D
NA methods are now widely used for 2, 9). Such methods could potentially
many forensic purposes, including be applied to searches of the convicted
routine investigation of serious offender/arrestee DNA databases. When
crimes and for identification of persons crime scene samples do not match anyone
killed in mass disasters or wars (1–4). in a search of forensic databases, the
DNA databases of convicted offenders are application of indirect methods could
maintained by every U.S. state and nearly identify individuals in the database who
every industrialized country, allowing are close relatives of the potential suspects.
comparison of crime scene DNA profiles This raises compelling policy questions
to one another and to known offenders about the balance between collective
(5). The policy in the United Kingdom security and individual privacy (10).
stipulates that almost any collision with To date, searching DNA databases
law enforcement results in the collection to identify close relatives of potential
of DNA (6). Following the U.K. lead, the suspects has been used in only a small
United States has shifted steadily toward number of cases, if sometimes to dramatic
inclusion of all felons, and federal and effect. For example, the brutal 1988
six U.S. state laws now include some murder of 16-year-old Lynette White,
provision for those arrested or indicted. At in Cardiff, Wales, was finally solved in
present, there are over 3 million samples in 2003. A search of the U.K. National DNA
the U.S. offender/arrestee state and federal Database for individuals with a specific
DNA databases (7). Statutes governing the single rare allele found in the crime scene
use of such samples and protection against evidence identified a 14-year-old boy with
misuse vary from state to state (8). a similar overall DNA profile. This led
Although direct comparisons of DNA police to his paternal uncle, Jeffrey Gafoor
profiles of known individuals and unknown (11). Investigation of the 1984 murder
biological evidence are most common, of Deborah Sykes revealed a close, but
indirect genetic kinship analyses, using not perfect, match to a man in the North
the DNA of biological relatives, are often Carolina DNA offender database, which led
necessary for humanitarian mass disaster investigators to his brother, Willard Brown
and missing person identifications (1, (12). Both Gafoor and Brown matched the
DNA from the respective crime scenes,
1 confessed, and were convicted.
Department of Pathology, Brigham and Women’s
Hospital and Harvard Medical School, Boston, MA Although all individuals have some
02115, USA. 2DNAVIEW, Oakland, CA 94611, and genetic similarity, close relatives have very
School of Public Health, University of California, similar DNA profiles because of shared
Berkeley, CA 94720, USA. 3John F. Kennedy School
of Government, Harvard University, Cambridge, MA
ancestry. We demonstrate the potential
02138, USA. value of kinship analysis for identifying
Authors are alphabetized to reflect equal promising leads in forensic investigations
contributions. Comments and ideas expressed on a much wider scale than has been used
herein are their own.
*Author for correspondence. to date.
E-mail: [email protected] Let us assume that a sample from a

20

Sequencing3.indd 22 9/7/06 3:28:06 PM

 Probability to find this relative if one exists
1

0.8

0.6

0.4

0.2

0
0 10 100 1000
Number of leads investigated (k)

Finding the genetic needle in a large haystack. The probability of

identifying a close relative (i.e., parent/child) of a known offender by
kinship searching is shown. Crime scene evidence would be searched
against each profile in a simulated offender DNA database. A parent/
child would be identified 62% of the time as the very first lead, and
99% of the time among the first 100 leads. Although these familial
searching methods do not invariably distinguish parent/child from sib-
lings, they have a high chance of identifying close relatives, if they exist,
among the database samples with the highest LRs.

crime scene has been obtained that is not Our simulations demonstrate that
an exact match to the profile of anyone kinship analysis would be valuable now
in current DNA databases. Using Monte for detecting potential suspects who are
Carlo simulations (13, 14), we investigated the parents, children, or siblings of those
the chances of successfully identifying whose profiles are in forensic databases.
a biological relative of someone whose For example, assume that the unknown
profile is in the DNA database as a possible sample is from the biological child of
source of crime scene evidence (15). Each one of the 50,000 offenders in a typical-
Monte Carlo trial simulates a database sized state database. Of the 50,000 LRs
of known offenders, a sample found at comparing the “unknown” sample to
a crime scene, and a search. The search each registered offender in the database,
compares the crime sample with each the child corresponds to the largest LR
catalogued offender in turn by computing about half the time, and has a 99% chance
likelihood ratios (LRs) that assess the of appearing among the 100 largest LRs
likelihood of parent-child or of sibling (see chart). An analysis of potential sibling
relationships (1, 16). We used published relationships produced a similar curve
data on allele frequencies of the 13 short (13).
tandem repeat (STR) loci on which U.S. These results could be ref ined
offender databases are based and basic by additional data—for example,
genetic principles (17–19). A high LR is large numbers of single nucleotide
characteristic of related individuals and polymorphisms (SNPs). Better and
is an unusual but possible coincidence immediately practical, a seven-locus
for unrelated dindividuals. The analysis Y-STR haplo type analysis on the crime
of each simulation therefore assumes that scene and the list of database leads would
investigators would follow these leads in eliminate 99% of those not related by male
priority order, starting with those in the lineage (20). Data mining (vital records,
offender database with the highest LR for genealogical and geographical data) for
being closely related to the owner of the the existence of suitable suspects related
crime scene DNA sample. to the leads can also help to refine the list.

21

Sequencing3.indd 23 9/7/06 3:28:21 PM

 The potential for improving effec- light of recent trends to expand collections
tiveness of DNA database searches is to arrestees and those convicted of
large. Consider a hypothetical state in minor crimes and misdemeanors (25).
which the “cold-hit” rate—the chance of Although use of retained samples for
finding a match between a crime scene other purposes is prohibited by federal and
sample and someone in the offender several state laws, sample retention also
database—is 10%. Suppose that among has been a controversial practice.
criminals who are not (yet) in the database Debates on the expansion of the scope
themselves, even 5% of them have a close of DNA collections for offender and
(parent/child or sibling) relative who arrestee databases, as well as collections
is. From our projections that up to 80% of volunteer samples, e.g., through DNA
(counting the 10 best leads) of those 5% dragnets, have concentrated on the balance
could be indirectly identified, it follows between society’s interests in security and
that the kinship analyses we describe could privacy interests of individuals who might
increase a 10% cold-hit rate to 14%—that be included in the database and on the
is, by 40%. There have been 30,000 cold fairness and equity of including some
hits in the United States up to now (5). in the databases but not others (26, 27).
Kinship searching has the potential for Privacy interests include genetic privacy
thousands more. [as DNA samples can yield medical and
Success of kinship searching depends other information (28)] and locational
most saliently on a close relative of the privacy (where the contributor has been
perpetrator actually being in the offender and left DNA). As with any investigative
database. Studies clearly indicate a strong technique, these DNA matching strategies
probabilistic dependency between the will lead to investigation of the innocent.
chances of conviction of parents and their Existing state and federal statutes do
children, as well as among siblings (21). not specifically address familial searches,
Consistent with these studies, in a U.S. and it is unlikely such search strategies
Department of Justice survey, 46% of jail were even considered at the time original
inmates indicated that they had at least one statutes were written. Use of familial
close relative who had been incarcerated searching methods described herein
(22). Such observations do not define or could raise new legal challenges, as a new
delineate the possible complex roles of category of people effectively would be
genetics, environment, and society in placed under lifetime genetic surveillance.
criminal behavior. Its composition would reflect existing
The widespread implementation demographic disparities in the criminal
of genetic kinship analysis for indirect justice system, in which arrests and
database searches would represent a convictions differ widely based on race,
critical shift in the use of government ethnicity, geographic location, and social
forensic data banks, as they could identify, class. Familial searching potentially
as potential suspects, not just those in the amplifies these existing disparities. These
database, but their close relatives as well. issues need to be confronted, as widespread
Genetic surveillance would thus shift from use of various familial searching tools,
the individual to the family. including formal kinship analysis, is
Challenges to forensic DNA data foreseeable. The de facto inclusion of
banking have been based largely on claims kin into DNA data banks may lead some
of U.S. constitutional protections from to oppose familial searching. It may lead
unreasonable search and seizure. Such others to support calls for a universal
challenges have not prevailed, as the courts DNA database (29), which to date have
have ruled that the interests of public safety been rejected. Other options include
outweigh individual privacy interests (23, limiting familial searching methods to
24). These DNA collections have sparked investigation of the most serious crimes
considerable controversy, especially in and defining statistical thresholds that

22

Sequencing3.indd 24 9/7/06 3:28:54 PM

 minimize intrusions on innocent parties 15. In the simulations, we made a variety of simplifying
assumptions (e.g., regarding random mating, mutation
(30). rates). These results are thus, of course, approximations
The rapid proliferation and expansion that will need experimental validation.
16. C. C. Li, L. Sacks, Biometrics 10, 347 (1954).
of DNA collections along with the
17. B. Budowle et al., J. Forensic Sci..44, 1277 (1999).
results of our analyses require careful 18. J. Butler, Forensic DNA Typing (Elsevier Academic Press,
consideration of the implications of Burlington, MA, ed. 2, 2005).
19. A. J. F. Griffiths et al., An Introduction to Genetic Analysis
familial searching methods. Every agency (Freeman, New York, ed. 2, 2004).
or country considering such methods 20. Data from Y-Chromosome Haplotype Reference STR
should evaluate attendant policy, ethical, database (YHRD), see (www.yhrd.org).
21. C. Smith, D. Farrington, J. Child Psychol. Psychiatr. 45,
and legal implications, in addition to their 230 (2004).
valuable investigatory potential. 22. U.S. Bureau of Justice Statistics, Correctional Populations
in the United States, 1996 (NJC 170013, U.S.
References and Notes Department of Justice, Washington, DC, April 1999),
1. C. H. Brenner, B. S. Weir, Theor. Popul. Biol. 63, 173 p. 62 (www.ojp.usdoj.gov/bjs/pub/pdf/cpius964.pdf).
(2003). 23. See United States v. Kincade, 379 F. 3d 813 (9th Cir.
2. C. H. Brenner, Forensic Sci. Int. 157, 172 2004) (en banc).
(2006). 24. State v. Raines, 875 A. 2d 19 (Md. 2004) (collecting
3. F. R. Bieber, in DNA and the Criminal Justice System, cases).
D. Lazer, Ed. (MIT Press, Cambridge, MA, 2004), pp. 25. D. Cardwell, New York Times, 4 May 2006.
23–72. 26. A. Etzioni, in DNA and the Criminal Justice System,
4. L. Biesecker et al., Science 310, 1122 (2005). D. Lazer, Ed. (MIT Press, Cambridge, MA, 2004), pp.
5. F. Bieber, J. Law Med. Ethics 34, 222 (2006). 197–224.
6. U.K. Criminal Justice Act, 2003 27. D. Lazer, V. Mayer-Schoenberger, J. Law Med. Ethics 34,
(www.opsi.gov.uk/acts/en2003/2003en44.htm). 366 (2006).
7. See (www.fbi.gov/hq/lab/codis/index1.htm). 28. D. Altshuler, A. G. Clark, Science 307, 1052 (2005).
8. For a summary of DNA database legislation in the 29. D. H. Kaye, M. S. Smith, in DNA and the Criminal Justice
United States, see (www.aslme.org). System, D. Lazer, Ed. (MIT Press, Cambridge, MA, 2004),
9. B. Budowle, F. R. Bieber, A. Eisenberg, Legal Med. 7, 230 pp. 247–284.
(2005). 30. R. Williams, P. Johnson, J. Law Med. Ethics 33, 545 (2005).
10. D. Lazer, M. Meyer, in DNA and the Criminal Justice 31. We gratefully acknowledge the American Society of Law,
System, D. Lazer, Ed. (MIT Press, Cambridge, MA, 2004), Medicine and Ethics, through NIH grant 1R01 HG2836-
pp. 357–390. 01, and the NSF (grant 0131923 to D.L.) for partial sup-
11. BBC News, “How police found Gafoor,” 4 July 2003 port. We thank E. Smith for research assistance.
(http://news.bbc.co.uk/1/hi/wales/3038138.stm).
12. R. Willing, USA Today, 7 June 2005, p. 1 (www.usatoday. Supporting Online Material
com/news/nation/2005-06-07-dna-cover_x.htm). www.sciencemag.org/cgi/content/full/1122655/DC1
13. Materials and methods are available as supporting
material on Science Online.
14. N. Metropolis, S. Ulam, J. Am. Stat. Assoc. 44, 335 Published online 11 May 2006; 10.1126/science.1122655
(1949). Include this information when citing this paper.

23

Sequencing3.indd 25 9/7/06 3:29:11 PM

 Genomic Sequencing
of Pleistocene Cave Bears
James P. Noonan,1,2 Michael Hofreiter,3 Doug Smith,1 James R. Priest,2 Nadin
Rohland,3 Gernot Rabeder,4 Johannes Krause,3 J. Chris Detter,1,5 Svante Pääbo,3
Edward M. Rubin1,2*

Despite the greater information content of genomic DNA, ancient DNA studies have
largely been limited to the amplification of mitochondrial sequences. Here we de-
scribe metagenomic libraries constructed with unamplified DNA extracted from
skeletal remains of two 40,000-year-old extinct cave bears. Analysis of ~1 mega-
base of sequence from each library showed that despite significant microbial con-
tamination, 5.8 and 1.1% of clones contained cave bear inserts, yielding 26,861 base
pairs of cave bear genome sequence. Comparison of cave bear and modern bear se-
quences revealed the evolutionary relationship of these lineages. The metagenomic
approach used here establishes the feasibility of ancient DNA genome sequencing
programs.

G
enomic DNA sequences from ex- ert environments, which are well suited to
tinct species can help reveal the preserving ancient DNA (9–12). However,
process of molecular evolution that the remains of most ancient animals, in-
produced modern genomes. However, the cluding hominids, have not been found in
recovery of ancient DNA is technologi- such environments.
cally challenging, because the molecules To circumvent these challenges, we
are degraded and mixed with microbial developed an amplification-independent
contaminants, and individual nucleotides direct cloning approach to constructing
are often chemically damaged (1, 2). In metagenomic libraries from ancient DNA
addition, ancient remains are invariably (Fig. 1). Ancient remains are obtained from
contaminated with modern DNA, which natural environments in which they have
amplifies efficiently compared with an- resided for thousands of years, and their
cient DNA, and therefore inhibits the de- extracted DNA is a mixture of genome
tection of ancient genomic sequences (1, fragments from the ancient organism and
2). These factors have limited most previ- sequences derived from other organisms
ous studies of ancient DNA sequences to in the environment. A metagenomic ap-
polymerase chain reaction (PCR) amplifi- proach, in which all genome sequences in
cation of mitochondrial DNA (3–8). In ex- an environment are anonymously cloned
ceptional cases, small amounts of single- into a single library, may therefore be a
copy nuclear DNA have been recovered powerful alternative to the targeted PCR
from ancient remains less than 20,000 approaches that have been used to re-
years old obtained from permafrost or des- cover ancient DNA molecules. We chose
to explore this strategy with the extinct
1
United States Department of Energy Joint
cave bear instead of an extinct hominid, to
Genome Institute, Walnut Creek, CA 94598, USA. unambiguously assess the issue of mod-
2
Genomics Division, Lawrence Berkeley National ern human contamination (1, 2). In ad-
Laboratory,Berkeley,CA94720,USA. 3MaxPlanck dition, because of the close evolutionary
Institute for Evolutionary Anthropology, Leipzig, relationship of bears and dogs, cave bear
D-04103, Germany. 4Institute of Paleontology, sequences in these libraries can be identi-
University of Vienna, Vienna, A-1010 Austria. fied and classified by comparing them to
5
Biosciences Directorate, Lawrence Livermore
the available annotated dog genome. The
National Laboratory, Livermore, CA 94550, USA.
*To whom correspondence should be addressed. phylogenetic relationship of cave bears
E-mail: [email protected] and modern bear species has also been

24

Sequencing3.indd 26 9/7/06 3:29:29 PM

 Fig. 1. Schematic illustration of the ancient DNA extraction and library construction pro-
cess. Extracts prepared from a cave bear tooth (library CB1) and a bone (library CB2)
contain cave bear DNA (red) and mixture of DNA from other organisms (gray).

inferred from mitochondrial sequences, hits to carnivore sequence was 68 bp long

providing the opportunity to compare the and covered 58% of the insert; whereas
phylogenetic information content of cave the average hit for the 289 clones in li-
bear mitochondrial and genomic DNA brary CB2 with significant carnivore hits
(13). was 70 bp and covered 49% of the insert.
We extracted DNA from a cave bear None of these clones showed homology to
tooth recovered from Ochsenhalt Cave, cave bear mitochondrial DNA sequences,
Austria, and a cave bear bone from which is consistent with the expected ratio
Gamssulzen Cave, Austria, dated at 42,290 of nuclear versus mitochondrial clones,
(error +970/–870) and 44,160 (+1400/ given the much greater size of the nuclear
–1190) years before the present, respec- genome (14). Based on the amount of cave
tively, by accelerator mass spectrometry bear genomic DNA used to construct li-
radiocarbon dating (table S1). We used brary CB2, we estimate that it could yield
these ancient DNA molecules to construct >10-fold coverage of the cave bear genome
two metagenomic libraries, designated if sequenced completely (14).
CB1 and CB2 (Fig. 1) (14). These librar- BLAST hits to the dog genome from
ies were constructed in a laboratory into libraries CB1 and CB2 were on average
which modern carnivore DNA has never 92.4 and 92.3% identical to dog, respec-
been introduced. Ancient DNA molecules tively. To confirm that these sequences
were blunt end–repaired before ligation were indeed those of the cave bear, we de-
but were otherwise neither enzymati- signed primers against 124 putative cave
cally treated nor amplified. We sequenced bear sequences and successfully amplified
9035 clones [1.06 megabases (Mb)] from and sequenced 116 orthologous sequenc-
library CB1 and 4992 clones (1.03 Mb) es from the modern brown bear. All 116
from library CB2. The average insert sizes of these modern bear sequences were at
for each library were 118 base pairs (bp) least 97% identical to their cave bear or-
and 207 bp, respectively. thologs, verifying that all or nearly all of
We compared each insert in these li- the carnivore sequences in both libraries
braries to GenBank nucleotide, protein, are genuine cave bear genomic sequences.
and environmental sequences, and the Only 6 of 14,027 sequenced clones had
July 2004 dog whole genome shotgun as- an insert that was identical to modern hu-
sembly, by using Basic Local Alignment man genomic DNA (Fig. 2, A and B). The
Search Tool (BLAST) software with an average BLAST hit length to the human
expect value cutoff of 0.001 and a mini- genome for these clones was 116 bp, and
mum hit size of 30 bp (14–16). 1.1% of the average hit covered 76% of the insert.
clones in library CB1 (Fig. 2A) and 5.8% Although we cannot establish a formal
of clones in library CB2 (Fig. 2B) had sig- insert length or insert coverage threshold
nificant hits to dog genome or modern bear that differentiates between ancient and
sequences. Our direct-cloning approach modern inserts because of the limited
produces chimeric inserts, so we defined number of modern sequences we obtained,
as candidate cave bear sequence only that these values are significantly greater than
part of the insert that had a hit to dog or the corresponding values for clones with
bear sequence. The average hit for the cave bear sequences (P < 0.05 for the dif-
100 clones in library CB1 with significant ference in both average clone length and

25

Sequencing3.indd 27 9/7/06 3:34:41 PM

 Fig. 2. Characterization of two independent cave bear genomic libraries. Predicted origin of 9035
clones from library CB1 (A) and 4992 clones from library CB2 (B) are shown, as determined by
BLAST comparison to GenBank and environmental sequence databases. Other refers to viral or
plasmid-derived DNAs. Distribution of sequence annotation features in 6,775 nucleotides of
carnivore sequence from library CB1 (C) and 20,086 nucleotides of carnivore sequence from library
CB2 (D) are shown as determined by alignment to the July 2004 dog genome assembly.

average insert coverage calculated by two- likely biased toward less-damaged ancient
tailed t test). This result suggests that it DNA fragments and modern DNA, which
may be possible to discriminate between could contribute to the abundance of pro-
inserts derived from short ancient DNA karyotic sequences relative to cave bear
molecules and inserts containing mod- sequences in these libraries. The represen-
ern undamaged DNA in ancient DNA tation we observe in the libraries thus re-
libraries. This may have relevance to the flects the proportion of clonable sequences
application of these methods to ancient from each source, not the true abundances
hominids, in which the ability to distin- of such sequences in the original extracts.
guish ancient hominid DNA from modern However, the results from both libraries
contamination will be essential. demonstrate that substantial quantities of
The remaining inserts with BLAST genuine cave bear genomic DNA are ef-
hits to sequences from known taxa were ficiently end-repaired and cloned, despite
derived from other eukaryotic sources, this possible bias. A considerable fraction
such as plants or fungi, or from prokary- of inserts in each library (17.3% in library
otic sources (bacteria and archaea), which CB1 and 11.2% in library CB2; Fig. 2) had
provided the majority of known sequences hits only to uncharacterized environmental
in each library. The endrepair reaction sequences. The majority of these clones
performed on each ancient DNA extract is had BLAST hits to GenBank sequences

26

Sequencing3.indd 28 9/7/06 3:30:05 PM

 Fig. 3. (A) Phylogenetic relationship of cave bear and modern bear sequences obtained by
maximum-likelihood estimation using 3201 aligned sites. (B) Phylogeny obtained using all sites
from (A), excluding cave bear and orthologous modern bear sites corresponding to two heavily
damaged cave bear clones (table S3). Substitution rates, total substitutions (black), and GC–AT
transitions (red) for each branch are shown. Orthologous dog sequence was used to root the trees.
Percent support for internal nodes is also shown.

derived from a single soil sample (17), dog genome assembly were annotated as
consistent with the contamination of each RefSeq exons, and ~10% of the cave bear
cave bear bone with soil bacteria from the sequences we obtained appear to be con-
recovery site. As in other metagenomic strained overall, whereas 5 to 8% of posi-
sequencing studies, most inserts in each tions in sequenced mammalian genomes
library had no similarity to any sequences are estimated to be under constraint (19).
in the public databases. This discrepancy may be caused by our
To annotate cave bear genomic se- use of BLAST sequence similarity to
quences, we aligned each cave bear se- identify cave bear sequences, an approach
quence to the dog genome assembly us- that is biased in favor of more-constrained
ing BLAST-like alignment tool (BLAT) sequences. Nevertheless, coding sequenc-
(18). 6.1% of 6775 cave bear nucleotides es, conserved noncoding sequences, and
from library CB1 and 4.1% of 20,086 repeats appear in both cave bear genomic
cave bear nucleotides from library CB2 libraries at frequencies roughly propor-
aligned to predicted dog RefSeq exons, tional to what has been observed in mod-
in a total of 21 genes distributed through- ern mammalian genomes.
out the dog genome (Fig. 2, C and D, and To determine whether the cave bear
table S2). 4.1% and 6.2% of cave bear nu- sequences we obtained contain sufficient
cleotides, respectively, from library CB1 information to reconstruct the phylogeny
and library CB2 aligned to constrained of cave bears and modern bears, we gener-
nonexonic positions in the dog genome ated and aligned 3201 bp of orthologous
with phastCons conservation scores < 0.8 sequences from cave bears and modern
(conserved noncoding, Fig. 2, C and D) black, polar, and brown bears and esti-
(14). The majority of cave bear sequence mated their phylogeny by maximum like-
in each library, however, aligned to dog lihood (Fig. 3A) (20). This phylogeny is
repeats or regions of the dog genome with topologically equivalent to phylogenies
no annotated sequence features. These lat- previously obtained using cave bear and
ter sequences are likely fragments of neu- modern bear mitochondrial DNA (13).
trally evolving, nonrepetitive sequence This result further indicates that our li-
from the cave bear genome. Constrained braries contain genuine cave bear se-
sequences are slightly overrepresented quences and demonstrates that we can
in our set: Only 1.7% of bases in the obtain sufficient ancient sequences from

27

Sequencing3.indd 29 9/7/06 3:30:26 PM

 those libraries to estimate the evolutionary approach. By revealing the phylogenomic
relationships between ancient and modern terrain of recent mammalian evolution,
lineages. these efforts should help identify the mo-
The substitution rate we estimated for lecular events underlying adaptive differ-
cave bears is higher than that in any other ences among modern species.
bear lineage. On the basis of results from
PCR-amplified ancient mitochondrial References and Notes
DNA, cytosines in ancient DNA can un- 1. S. Pääbo et al., Annu. Rev. Genet. 38, 645 (2004).
2. M. Hofreiter, D. Serre, H. N. Poinar, M. Kuch, S.
dergo deamination to uracil, which results Pääbo, Nat. Rev. Genet. 2, 353 (2001).
in an excess of G-to-A and C-to-T (GC– 3. C. Hänni, V. Laudet, D. Stehelin, P. Taberlet, Proc.
AT) transitions (21). The inflated substitu- Natl. Acad. Sci. U.S.A. 91, 12336 (1994).
4. M. Krings et al., Cell 90, 19 (1997).
tion rate in cave bears is likely caused by 5. M. Krings, H. Geisert, R. W. Schmitz, H. Krainitzki, S.
an excess of such events, because many Pääbo, Proc. Natl. Acad. Sci. U.S.A. 96, 5581 (1999).
6. M. Hofreiter et al., Mol. Biol. Evol. 19, 1244 (2002).
of the substitutions assigned to the cave 7. D. Serre et al., PLoS Biol. 2, 313 (2004).
bear lineage are GC–AT transitions (Fig. 8. M. Hofreiter et al., Proc. Natl. Acad. Sci. U.S.A. 101,
3A). These presumably damage-induced 12963 (2004).
9. P. Goloubinoff, S. Pääbo, A. C. Wilson, Proc. Natl.
substitutions complicate phylogenetic Acad. Sci. U.S.A. 90, 1997 (1993).
reconstruction and the identification of 10. A. D. Greenwood, C. Capelli, G. Possnert, S. Pääbo,
Mol. Biol. Evol. 16, 1466 (1999).
functional sequence differences between 11. V. Jaenicke-Després et al., Science 302, 1206 (2003).
extinct and modern species. However, 12. H. Poinar, M. Kuch, G. McDonald, P. Martin, S. Pääbo,
these substitutions are not randomly dis- Curr. Biol. 13, 1150 (2003).
13. O. Loreille et al., Curr. Biol. 11, 200 (2001).
tributed among all cave bear sequences but 14. Materials and methods are available as supporting
are clustered on a few clones from library material on Science Online.
CB2. Thus, two clones from library CB2 15. The dog genome sequence generated by the Broad
Institute and Agencourt Biosciences was obtained
have three and four GC–AT transitions from the University of California at Santa Cruz Dog
specific to cave bears, an observation that Genome Browser (http://genome.ucsc.edu).
16. S. F. Altschul et al., Nucleic Acids Res. 25, 3389 (1997).
is extremely unlikely given that the occur- 17. S. G. Tringe et al., Science 308, 554 (2005).
rence of true randomly arising substitu- 18. W. J. Kent, Genome Res. 12, 656 (2002).
tions on cave bear clones should follow 19. G. M. Cooper et al., Genome Res. 14, 539 (2004).
20. We cannot formally exclude alternative topologies,
a Poisson distribution (table S3). When given the small number of nucleotide sites in the
these two clones are excluded from the analysis and the limited divergence among these
analysis (Fig. 3B), the apparent substitu- bear species, which also accounts for the low percent
support for the internal nodes. Damage-induced
tion rate and the excess of GC–AT transi- substitutions in cave bears also perturb the phylog-
tions in cave bears are reduced, with little eny. The topology with the next-best likelihood score
has cave bears as the outgroup to the modern bears
impact on estimates of substitution rates due to an excess of GC–AT substitutions in several
in the modern bear lineages. Although we library CB2 sequences.
cannot distinguish individual GC–AT sub- 21. M. Hofreiter, V. Jaenicke, D. Serre, A. von Haeseler, S.
Pääbo, Nucleic Acids Res. 29, 4793 (2001).
stitutions from deamination-induced dam- 22. Data have been deposited into GenBank with accession
age, these observations provide a quanti- numbers CZ551658 to CZ552046. We thank members
of the Rubin and Pääbo laboratories for insightful
tative means to identify cloned ancient discussions and support. This work was performed
DNA fragments with an excess of cytosine under the auspices of the U.S. Department of Energy’s
deamination events. It is also likely that Office of Science Biological and Environmental Re-
search Program and by the University of California;
heavily degraded ancient DNA fragments Lawrence Berkeley National Laboratory; Lawrence
are poorly end-repaired and will therefore Livermore National Laboratory; and Los Alamos Na-
appear at reduced frequency in ancient tional Laboratory under contract numbers DE-AC03-
76SF00098, W-7405-Eng-48, and W-7405-ENG-36,
DNA genomic libraries. respectively, with support from NIH grants U1
Although small amounts of genomic HL66681B and T32 HL07279 and at the Max Planck
sequence have previously been obtained Institute for Evolutionary Anthropology.
by amplification from <20,000-year-old Supporting Online Material
remains (12), the direct cloning strategy www.sciencemag.org/cgi/content/full/1113485/DC1
employed here has yielded considerably Materials and Methods
Tables S1 to S3
more genomic DNA sequence from much References
older samples. Based on our results, an-
cient DNA sequencing programs for ex- 12 April 2005; accepted 26 May 2005
Published online 2 June 2005;
tinct Pleistocene species, including homi- 10.1126/science.1113485
nids, are feasible using a metagenomic Include this information when citing this paper.

28

Sequencing3.indd 30 9/7/06 3:31:19 PM

 Whole Genome Assembly using
Paired End Reads in E. coli, B. licheniformis,
and S. cerevisiae
Corresponding author: Tom Jarvie, 454 Life Sciences Corporation, Branford, CT, USA, Email: [email protected]

Introduction fungi. The whole genome sequencing application

employs shotgun sequencing of whole genomic
The Genome Sequencer System is the first- DNA, typically to a 20-fold level of oversampling
generation instrument employing the 454 for a given genome. The Genome Sequencer 20
sequencing technology [1]. This system enables Instrument is capable of generating 20 million
researchers to perform rapid and comprehensive bases, or 200,000 reads of 100 bp, of sequence data
“whole genome shotgun sequencing” – the per 5.5-hour run, enabling fast and cost-effective
determination of the nucleotide sequence of entire generation of the required level of data. The
genomes. Using this technology platform, scientists standard whole genome de novo sequencing
have been able to generate whole genome sequences application utilizes the Newbler Assembler software
for a wide variety of organisms, from viruses to to assemble the reads into a number of unordered
and unoriented contigs that typically cover >99% of
the non-repeat regions of the genome. In contrast
to the Sanger method of sequencing that employs
Paired End reads in the original assembly process,
454 has developed a new protocol to generate a
library of Paired End reads that are used to
determine the orientation and relative positions of
contigs produced by the de novo shotgun
sequencing and assembly.

These Paired End reads are approximately 84-

nucleotide DNA fragments that have a 44-mer
adaptor sequence in the middle flanked by a 20-mer
Figure 1: An illustration of the Paired End assembly
process. Paired End reads are used to order and orient the sequence on each side. The two flanking 20-mers
contigs derived from the Newbler assembly. The large are segments of DNA that were originally located
blue lines represent contigs generated from the whole approximately 2.5 kb apart in the genome of
genome shotgun sequencing and assembly. The multiple
interest. The ordering and orienting of contigs
blue and grey lines represent Paired End information. The
blue segments represent the two 20 nucleotide regions generates scaffolds which provide a high-quality
that were sequenced while the dotted grey line represents draft sequence of the genome. The 454 method, as
the distance between those two sequenced regions. compared to the Sanger method, dramatically
reduces the time involved to generate a high-quality
draft from weeks (or months) to days.
Materials and Methods

Materials Methods

Equipment:
For complete details on the Paired End DNA library
Genome Sequencer 20 Instrument (Software 1.0.53)
preparation procedure and sequencing, please refer to
Reagents: the GS Guide to Paired End Sequencing, the GS
Sample Preparation: GS Paired End Adaptor Kit; emPCR Kit User's Manual, and the Genome Sequencer 20
GS emPCR Kit II (Amplicon A, Paired End) Operator’s Manual.
Sequencing: GS 20 Sequencing Kit; GS PicoTiterPlate Kit

A detailed list of all required equipment,

reagents and methods is provided in the GS
Guide to Paired End Sequencing.

29

Sequencing3.indd 31 9/7/06 3:31:41 PM

 Procedure
Library Generation
Generation of a Paired End library:
Intact genomic DNA is fragmented to
yield an average length of 2.5 kb. The
fragmented genomic DNA is
methylated with EcoRI methylase to
protect the EcoRI restriction sites. The
ends of the fragments are blunt-ended,
polished, and an adaptor DNA oligo is
blunt-end ligated onto both ends of the
digested DNA fragments. Subsequent
digestion with EcoRI cleaves a portion Figure 2: Illustration of the steps required to generate a
Paired End library for sequencing.
of the adaptor DNA, leaving sticky
ends. The fragments are circularized
and ligated, resulting in 2.5 kb circular copies of clonally amplified DNA. This bead-bound
fragments. The adaptor DNA library is then deposited onto the PicoTiterPlate for
contains two MmeI restriction sites, and after sequencing with the standard protocol.
treatment with MmeI the circularized DNA is
The Paired End data is combined with standard
cleaved 20 nucleotides away from the restriction
whole genome shotgun sequencing data at an
sites in the adaptor DNA. This digestion generates
oversampling level of approximately 20-fold.
small DNA fragments that have the adaptor DNA in
Shotgun sequencing is performed using the
the middle and 20 nucleotides of genomic DNA
standard whole genome preparation protocols for
that were once approximately 2.5 kb apart on each
the Genome Sequencer System. An updated version
end. These small, biotinylated DNA fragments are
of the Newbler Assembler, which incorporates both
purified from the rest of the genomic DNA by using
standard 100 bp reads and Paired End reads into the
streptavidin beads. The purified Paired End
assembly, is used to generate contigs and scaffolds
fragments are processed through the standard
for the genome of interest.
library preparation protocol for the Genome
Sequencer System. Bioinformatics
Post-Run Analysis: Flow data was assembled using
emPCR and Sequencing
the Assembly Software (Newbler Assembler
The Paired End library is processed through the
Software) of the Genome Sequencer 20 Software
standard emPCR process to generate hundreds of
Version 1.0.53, as described in the GS 20 Data
thousands of beads, each containing millions of
Processing Software Manual.

A) B) C)

Figure 3:
A) The Paired End library is amplified onto beads using the emPCR process.
B) The bead-bound library is deposited onto the PicoTiterPlate.
C) The Paired End reads are sequenced on the Genome Sequencer 20 Instrument.

Sequencing3.indd 32 9/7/06 3:31:54 PM

 Results 98 unoriented contigs. One additional sequencing
run of a Paired End library yielded approximately
Assembly of Escherichia coli K12 255,000 reads. The Paired End data reduced the
The 4.6 Mb genome of E. coli K12 was shotgun assembly to 9 scaffolds covering 99.2% of the
sequenced in three sequencing runs, yielding genome.
approximately 22-fold oversampling. Assembly Assembly of Saccharomyces cerevisiae
performed with the Newbler Assembler software The 12.2 Mb genome of Saccharomyces cerevisiae
resulted in 140 unoriented contigs. One additional S288C (containing 16 haploid chromosomes and
sequencing run of a Paired End library yielded one 86 kb mitochondrion) was shotgun sequenced
approximately 112,000 reads. The Paired End data in nine sequencing runs, yielding approximately
reduced the assembly to 24 scaffolds covering 23-fold oversampling. The assembly performed
98.6% of the genome. with the Newbler Assembler software resulted in
Assembly of Bacillus licheniformis 821 unoriented contigs. Two additional sequencing
The 4.2 Mb genome of Bacillus licheniformis runs of a Paired End library yielded approximately
ATCC14580 (DSM 13) was shotgun sequenced in 395,000 reads. The Paired End data reduced the
three sequencing runs, yielding approximately assembly to 153 scaffolds covering 93.2% of the
27-fold oversampling. Assembly performed with genome.
the Newbler Assembler software resulted in

Figure 4: De novo assembly results for E. coli K12 aligned Spaces between the pink bars are typically a result of repeat
against a reference genome. The reference genome is regions that cannot be uniquely assigned to a region in the
represented by the top black line. The standard whole genome. With the addition of one sequencing run of Paired
genome shotgun sequence and assembly is represented End reads (represented by the purple bars) using the
by the pink bars. Repeat regions of the genome are Genome Sequencer 20 Instrument, the genome sequence
represented by the green bars at the bottom of the figure. becomes much more complete.

NOTICE TO PURCHASER
RESTRICTION ON USE: Purchaser is only authorized to use the
References
Genome Sequencer 20 Instrument with PicoTiterPlate devices [1] Margulies, M (2005) Nature 437:376-380
supplied by 454 Life Sciences Corporation and in conformity with
the procedures contained in the Operator’s Manual. For more information, visit
Trademarks
www.genome-sequencing.com
PICOTITERPLATE is a trademark of 454 Life Sciences Corporation,
Branford, CT, USA.
Other brands or product names are trademarks of their respective
holders.

Roche Diagnostics GmbH

Roche Applied Science
68298 Mannheim
Germany
www.roche-applied-science.com

Sequencing3.indd 33 9/7/06 3:32:08 PM

 The future of

is here.
www.aaas.org/future

Sequencing3.indd 34 9/7/06 3:32:22 PM

 www.roche-applied-science.com

Genome Sequencer 20 System

First to the Finish

Revolutionize your research with the newest innovation in
rapid whole genome sequencing and assembly — the new
Genome Sequencer 20 from Roche Applied Science.
Perform genomic analyses not currently possible due to
the cost and time constraints of current methods.
Generate data fast
Sequence at least 20 million bases in one 5.5-hour run.
Spend less per base
Reduce cost per raw base with rapid, efficient
amplification and sequencing through picoliter format
with massive parallelization.
Simplify sample preparation
Eliminate cloning and colony picking with single-tube
library preparation per genome.
Expand your versatility
Perform a wide variety of projects with de novo assembly
and mapping software.
Be first to the finish! Explore the power of the newest
revolution in sequencing today — visit
www.genome-sequencing.com, call 800 262 4911
(U.S.), or contact your local sales representative.

Roche Diagnostics
Roche Applied Science
© 2006 Roche Diagnostics. All rights reserved. Indianapolis, Indiana

Sequencing3.indd 35 9/7/06 3:32:35 PM

 www.roche-applied-science.com

Genome Sequencer 20 System

First to the Finish

Analyze transcriptomes like never before

I Identify miRNA expression

I Perform expression profiling
I Uncover unknown or altered transcripts
I Discover transcription factor binding sites
Flowgram of a GS 20 read
with unprecedented speed and sensitivity

Harness the horsepower of the newest

revolution in sequencing today —
visit www.genome-sequencing.com,
call 800 262 4911 (U.S.), or contact your
local sales representative.

Roche Diagnostics
Roche Applied Science
© 2006 Roche Diagnostics. All rights reserved. 04926447001 Indianapolis, Indiana

Sequencing3.indd 36 9/7/06 3:32:52 PM