Uncovering the roles of rare variants in common disease through whole-genome sequencing.

Abstract

Although genome-wide association (GWA) studies for common variants have thus far succeeded in explaining only a modest fraction of the genetic components of human common diseases, recent advances in next-generation sequencing technologies could rapidly facilitate substantial progress. This outcome is expected if much of the missing genetic control is due to gene variants that are too rare to be picked up by GWA studies and have relatively large effects on risk. Here, we evaluate the evidence for an important role of rare gene variants of major effect in common diseases and outline discovery strategies for their identification.

Full text

Page 1

Genome-wide association (GWA) studies, which have
thus far focused on very common SNPs, have been
completed for most common human diseases and
many related traits. These studies have found most of
the very common gene variants (minor allele frequency
(MAF) > ~5%) in the human genome and have iden-
tified over 500 independent strong SNP associations
(p < 1 × 10–8) (see the National Human Genome
Research Institute (NHGRI) Catalog of Published
Genome-Wide Association Studies). However, most of
these associated SNPs have very small effect sizes, and
the proportion of heritability explained is at best mod-
est for most traits1. Furthermore, GWA signals have
rarely been tracked to causal polymorphisms, which
has led many to assume that the causal variants must
have subtle regulatory effects. Partially in response
to this assumption, genome-wide patterns of gene
expression have been characterized in multiple tissue
types, and many common variants that influence gene
expression have been identified2,3. Surprisingly, vari-
ants associated with gene expression show little overlap
with those associated with disease4. Although the sys-
tematic identification of rare variants associated with
common diseases has not yet been feasible, several rare
variants have nevertheless been identified that confer a
substantial risk of disease. For example, autism, men-
tal retardation, epilepsy and schizophrenia have been
shown to be influenced by rare structural variants that
affect genes5. Additionally, it seems possible that some,
perhaps even many, of the current GWA signals could
reflect the effect of multiple rare variants that have been
credited to common variants6. Whatever underlies
the GWA signals for common diseases, it is clear that
GWA studies of very common variants rarely succeed
in securely implicating specific genes in specific com-
mon diseases, which greatly limits their importance in
applications such as drug development.
Taken together, these observations support the
long-established idea that rare variants could be
the primary drivers of common diseases7–9. Over the
past several years, this view has become less popular,
at least with some in the research community, than
the theory that common diseases are more strongly
influenced by many common gene variants with small
effects on risk10–13. Although the relative impact of com-
mon and rare variants on common diseases remains
an unanswered empirical question, we think that the
results of the past several years make a strong case
for common diseases being more similar to Mendelian
diseases than is postulated by the common disease–
common variant model14. In particular, it seems pos-
sible that much of the genetic control of common
diseases is due to rare and generally deleterious variants
that have a strong impact on the risk of disease in indi-
vidual patients. It is also likely that the variants with
the largest effect sizes will be those that have obvious
functional consequences. This rare, functional variant
model is not inconsistent with the absence of secure
linkage evidence for most common diseases (for exam-
ple, nearly every chromosome arm has been ‘linked’
to schizophrenia and bipolar disorder15–17). Variants
that increase the risk of disease by less than about
Center for Human Genome
Variation, Duke University
Medical School, Durham,
North Carolina 27708, USA.
Correspondence to D.B.G.
e-mail: d.goldstein@duke.edu
doi:10.1038/nrg2779
Minor allele frequency
Ranging from 0 to 50%, this
is the proportion of alleles
at a locus that consists of
the less frequent allele.
This number does not take
genotype into account.
Effect size
The increase in risk (or
proportion of population
variation) that is conferred
by a given causal variant.
Heritability
The proportion of phenotypic
variation in a trait that is due
to underlying genetic variation.
In studies of humans, this value
is usually calculated by
comparing trait correlations in
individuals of varying degrees
of relatedness.
Uncovering the roles of rare
variants in common disease through
whole-genome sequencing
Elizabeth T. Cirulli and David B. Goldstein
Abstract | Although genome-wide association (GWA) studies for common variants have
thus far succeeded in explaining only a modest fraction of the genetic components of
human common diseases, recent advances in next-generation sequencing technologies
could rapidly facilitate substantial progress. This outcome is expected if much of the
missing genetic control is due to gene variants that are too rare to be picked up by GWA
studies and have relatively large effects on risk. Here, we evaluate the evidence for an
important role of rare gene variants of major effect in common diseases and outline
discovery strategies for their identification.
ApplicAtions of next-generAtion sequencing
REVIEWS
NATuRe ReVIeWS | Genetics
VOluMe 11 | juNe 2010 | 415
© 20 Macmillan Publishers Limited. All rights reserved10

Page 2

Mendelian disease
A disease that is carried in
families in either a dominant or
recessive manner and that is
typically controlled by variants
of large effect in a single gene.
Imputation
Based on the known linkage
disequilibrium structure in fully
genotyped individuals, the
genotype of untyped variants
can be inferred in individuals
who are genotyped for a
smaller number of variants.
Exome
The exome is the collection of
known exons in our genome:
this is the portion of the
genome that is translated into
proteins. As exons comprise
only 1% of the genome
and contain the most easily
understood, functionally
relevant information,
sequencing of only the exome is
a cheaper method of identifying
most of the variants that
are most likely to affect a trait.
Linkage disequilibrium
A nonrandom association
between alleles at different loci.
fourfold are expected to generate inconsistent linkage
evidence18, leaving scope for the presence of many rare
variants with a strong impact on disease risk. The availa-
ble linkage evidence for many common diseases securely
confirms high locus heterogeneity; otherwise, the
weak linkage evidence would tend to point towards
the same genome regions in different families, which
is generally not the case. This is not to say that common
variants play no part in common diseases. Indeed, stud-
ies of some Mendelian diseases indicate that common
variants may often have a key role as modifiers of the
effects of rarer, more highly penetrant contributors to
disease risk19,20, and it seems reasonable to expect that
this also holds true for common diseases. This view
could have fundamental implications for the discov-
ery of the genetic bases of common diseases and for
therapeutic applications.
In this Review, we first discuss the insights provided
by GWA studies into the genetic architecture under-
lying common diseases. We then present evidence
supporting the potentially important role of rare vari-
ants in common diseases. Next, we outline strategies,
namely whole-genome and whole-exome sequencing,
for the discovery of rare causal variants, with a focus on
family-based designs and extreme-trait designs. Finally,
we suggest that the underlying causes of Mendelian and
common diseases may be more similar than previously
supposed, and we offer a perspective of what the future
holds for these types of studies.
the genetic architecture of common diseases
Although most of the complex traits that have been
studied using the GWA approach have reasonable levels
of heritability, the proportion of that heritability that
has been explained by very common variants is surpris-
ingly low1. This limited impact is well-illustrated by two
diseases that have been subject to exceptionally large
studies. Despite a discovery sample size of 10,128 and
a replication sample size of 53,975, the 18 common
variants that were found to be significantly associated
with type 2 diabetes seem to explain only about 6% of
the increased risk of disease among relatives21,22 (BOX 1).
likewise, a meta-analysis of schizophrenia GWA stud-
ies that included a total of 8,008 cases and 19,077 con-
trols identified only 7 significant SNPs, some in high
linkage disequilibrium with each other and each with an
odds ratio below 1.3, despite a heritability of 80–85%
for schizophrenia23. Pharmacogenetic traits may be an
exception: they have not been extensively studied, but
associated variants of large effect have been identified
by GWA studies24–28. This difference in effect size may
reflect the lack of long-term selection on many such
traits (BOX 2).
Several explanations for the missing heritability,
and solutions for finding it, have been offered. One
popular view had been that because disease status is
determined by many underlying factors, intermedi-
ate phenotypes related to the condition would pro-
vide a more tractable target in GWA studies than the
Box 1 | Variants identified through gWA for an example of a common human disease
Type 2 diabetes has been studied using some of the largest
genome-wide association (GWA) sample sizes to date.
Three large case–control studies were recently subjected
to a meta-analysis with a total sample size of 4,549 cases
and 5,579 controls22: 1,924 cases and 2,938 controls came
from the Wellcome Trust Case Control Consortium78,
1,464 cases and 1,467 controls came from the Diabetes
Genetics Initiative79, and 1,161 cases and 1,174 controls
came from the Finland–United States Investigation of
Non-Insulin-Dependent Diabetes Mellitus Genetics80. The
first two studies were performed using a different chip
(Affymetrix 500k) from that used in the last study (Illumina
317k), but the imputation of missing genotypes allowed for
a combined analysis of 2.2 million SNPs for all 10,128
participants. Twelve SNPs had already been established as
associated with type 2 diabetes in these earlier studies,
and the meta-analysis confirmed six additional associated
SNPs. In total, these 18 variants only explained 6% of
the heritability of type 2 diabetes21. Furthermore, the
underlying causal variant has not definitively been found
for any of these significant associations, although
potassium inwardly-rectifying channel, subfamily J,
member 11 (KCNJ11) and Wolfram syndrome 1 (WFS1)
were already known to contain rare variants that
influence the disease, and there is evidence supporting
potentially causal common variants in transcription
factor 7-like 2 (TCF7L2), solute carrier family 30, member 8
(SLC30A8) and KCNJ11 (Refs 81–83). The table shows
type 2 diabetes-associated variants reported by Zeggini
et al.22; the maximum odds ratio reported was 1.37.
snP
rs7903146
Odds ratio
1.37
snP location
Intron of TCF7L2
rs130711681.35 Between SYN2
and PPARG
rs70209961.26 Near CDKN2A and
CDKN2B
rs69315141.25 Intron of CDKAL1
rs18012821.18 Exon of PPARG
rs44029601.17Intron of IGF2BP2
rs50154801.17 Near HHEX
rs5215 1.16 Exon of KCNJ11
rs102829401.15 Exon of SLC30A8
rs80501361.15Intron of FTO
rs75785971.15 Exon of THADA
rs109239311.13 Intron of NOTCH2
rs45807221.11Near WFS1
rs127797901.11Between CDC123
and CAMK1D
rs177051771.1Near TCF2
rs864745 1.1Intron of JAZF1
rs7961581 1.09Between TSPAN8
and LGR5
rs46071031.09Near ADAMTS9
REVIEWS
416 | juNe 2010 | VOluMe 11
www.nature.com/reviews/genetics
© 20 Macmillan Publishers Limited. All rights reserved10

Page 3

Endophenotype
An intermediate phenotype
that is heritable and
associated with a disease but
is not itself a symptom of the
disease. Although there is little
evidence to support the
theory, it has been argued that
endophenotypes would be a
more tractable target for
genetic analysis than the
relevant disease state itself.
Haploinsufficiency
This occurs when a diploid
organism only has one copy
of a gene and both copies are
required for correct function.
This is one way that a
protein-truncating mutation
can influence predisposition
to a disease.
disease itself. Available data, however, provide little
support for this view. For example, cognitive traits
are perhaps the leading endophenotype in schizophre-
nia, but GWA studies have proved no more success-
ful for uncovering the genetic basis of cognitive traits
in normal individuals than for studying the causes of
schizophrenia itself29,30.
Another popular theory is that the variation con-
tributing to common diseases has subtle effects and
is context-dependent. For example, some have argued
that specific combinations of common variants may
lead to substantially greater effects on risk than is
apparent for any individual common variant. Although
this explanation is theoretically possible, there are cur-
rently few clear examples of common variants identi-
fied in GWA studies that interact strongly with each
other or with the environment to affect a complex
trait. Nevertheless, methods for detecting interac-
tions among the many common variants in the human
genome are still immature31, and it remains possible
that novel analytic and experimental approaches
will identify new risk profiles that emerge from
multi-way interactions.
Finally, a recent study suggests that parent-of-origin
effects may be important for common disease32, and
that the incorporation of such information into GWA
studies may increase the estimated contribution of
associated common variants to disease risk.
rare variants and common diseases
For common variants that have been definitively shown
to be associated with a trait, the underlying causal vari-
ant has rarely been found, although there is supporting
evidence for certain traits; examples include a prostate-
cancer-associated SNP that is upstream of a gene and is
associated with expression level in cell lines, and breast-
cancer-associated intronic SNPs that are associated
with expression and binding of transcription factors33,34.
Because pinning down common causal variants with
subtle effects is difficult, many could eventually be
confirmed over time. However, even for apparently
clear examples, such as age-related macular degen-
eration (AMD) — a now classic example of a disease-
associated common variant — there is still no general
agreement on whether the identified coding variant
explains the observed associations35–37. The common
disease–common variant hypothesis14 led to the gen-
eral assumption that some as-yet-unidentified common
variant is responsible for the associations seen in most
GWA studies7,22,38,39. This assumption has controlled
much of the follow-up research, in which the focus has
generally been on resequencing in the linkage disequi-
librium block defined by common variants40–43. even
when rare variants in the same regions as GWA signals
have been identified44,45, supporting the view that they
could influence disease risk, these seem to be viewed as
independent of the original association.
What are rare variants? Although the GWA studies
performed so far have been designed to fully character-
ize direct effects of very common variants, they have
provided little information about whether the rest of
the genetic control for each trait is only slightly more
rare than the general detection threshold in these GWA
studies (about 5%) or substantially more rare. This
question is crucial because the answer determines the
optimal research strategies. In TABLe 1, we outline a divi-
sion of potential causal variant frequencies and their
implications for analysis methods.
Rare copy-number variants. The same gene chips used
to evaluate common SNPs can also be used to analyse
the roles of common and rare large copy-number vari-
ants (CNVs), and this approach, along with others, has
identified secure connections between CNVs and com-
mon diseases5. Although systematic work on large CNVs
in disease is in its infancy, specific CNVs are associated
with an effect on the risk of disease that dramatically
exceeds the effects of most common variants associated
with a disease. For example, three rare deletions were
recently shown to be associated with schizophrenia with
odds ratios estimated at 2.7, 11.5 and 14.8 (Ref. 46), which
are substantially higher than the typical values for com-
mon variants that are associated with a disease: of those
GWA study case–control associations with p-values
below 1 × 10–8, only 13% had odds ratios above 2, and only
1% had odds ratios above 10 (see the NHGRI Catalog of
Published Genome-Wide Association Studies). There are
two obvious possibilities for the pathogenicity of these
deletions: haploinsufficiency and the revealing of recessive

Box 2 | pharmacogenetic traits
Although the effect sizes of SNPs identified by genome-wide association (GWA)
studies have been very modest for most complex traits, this has not been true for
the small number of pharmacogenetic traits that have been studied in this
manner. Of the nearly 450 (not screened for independence) GWA signals from
case–control studies that were reported to have a p-value below 1 × 10–8 in the
National Human Genome Research Institute (NHGRI) Catalog of Published
Genome-Wide Association Studies, only five were studies of pharmacogenetic
traits (two of these are replications of one of the others). Despite this imbalance,
these traits have some of the strongest associations found: all five have an
odds ratio above 1.95, which is true of only 13.8% of associated SNPs for
non-pharmacogenetic traits, and two of them have odds ratios above 25.0, an
effect size seen for only one non-pharmacogenetic association. These results may
simply reflect the low-hanging fruit, and as more pharmacogenetic GWA studies
are performed, it may be found that most pharmacogenetic variants do not have
effects this large. However, thus far they support the long-standing theory that
the genetic variants that contribute to pharmacogenetic traits will often have
little phenotypic consequence before the administration of a drug84 (note that this
is not to say that variants that influence drug response have never been under
selection for other reasons — for example, it has been postulated that genetic
variants that affect cytochrome P450, family 2, subfamily D, polypeptide 6
(CYP2D6)85 may have been influenced by alkaloid toxins in the diet). Because
humans have spent most of their evolutionary history unexposed to drugs, many
pharmacogenetic traits have largely escaped the attention of natural selection.
This is in contrast to other complex traits, which have been under comparatively
heavy selection unless their effects on fitness come very late in life. Because of
this lack of constraint on the evolution of variants that affect pharmacogenetic
traits, they have had the opportunity to rise in frequency through genetic drift.
Although this does not mean that such variants will always be of a sufficiently
high frequency to be detected by GWA studies, they are more likely to have a high
frequency than the variants that have been under selection due to their effect on
common diseases and other complex traits.
REVIEWS
NATuRe ReVIeWS | Genetics
VOluMe 11 | juNe 2010 | 417
© 20 Macmillan Publishers Limited. All rights reserved10

Page 4

Haplotype
A combination of alleles
that are inherited together.
mutations on the homologous chromosome (fIG. 1).
Rare CNVs are simply one class of rare variants that
may contribute to common diseases, and the example
of Mendelian diseases (see below) strongly suggests that
rare CNVs will be a minority of all the rare variants
that contribute to disease risk.
Synthetic associations. It has long been recognized
that a common variant could record a diluted signal
of risk due to a rarer causal variant of large effect. An
early paper from the HapMap consortium made this
point in 2003, noting: “...even a relatively uncommon
disease-associated variant can potentially be discovered
using this approach. Reflecting its historical origins, the
uncommon variant will be travelling on a chromosome
that carries a characteristic pattern of nearby sequence
variants. In a group of people affected by a disease, the
rare variant will be enriched in frequency compared with
its frequency in a group of unaffected controls.”10
Nevertheless, the properties of such associations had
not been systematically explored until a recent study by
Dickson et al.6. They considered a model in which one
or more ‘rare’ variants (defined as below the threshold
of reliable representation in GWA studies) were the only
contributors to disease risk. They then asked how often a
signal of risk conferred by such variants would be picked
up and credited to common variants in a typical GWA
study (fIG. 2). They found that in typical GWA sample
sizes, rare variants can easily create genome-wide sig-
nificant associations, and the properties of such asso-
ciations can be dramatically different from what occurs
when the causal variant is common. For example, the
causal variants can be megabases away from the com-
mon variants that carry the signal of the association,
and the real risk effects can be several-fold stronger
than what is credited to a common variant. In addition,
and counter-intuitively, the likelihood of a genome-wide
significant signal being credited to common variants can
increase with the number of rare causal variants in
a region (because multiple rare variants can load up
onto the same haplotype defined by a common variant).
Dickson et al. referred to these associations as ‘synthetic
associations’, a special case of indirect association. This
term was not meant to imply that such associations are
spurious but was intended to emphasize the differences
between the properties of these associations and what is
expected when the causal variant is common.
It is unknown how many GWA signals may be due
to synthetic associations, but the ease with which they
could occur does question some of the inferences that
have been drawn about the results of GWA studies and
also mandates a more open-minded approach to follow-
ing up GWA signals. For example, when associated SNPs
are far from the nearest gene, it is often assumed that
a regulatory variant of subtle effect must be the cause,
such as for the SNPs in 8q24 that are associated with
colo rectal cancer47–49 and located hundreds of kilobases
from the oncogene MYC. Although there is evidence
supporting MYC regulation by this region, there has
been no definitive link between the associated variants
and this regulation50,51. However, as a synthetic associa-
tion can be driven by rare variants that are megabases
from the GWA signal6 (fIG. 2), the ‘locus’ implicated by
a GWA study might be much larger than has typically
been assumed. The effect sizes of the causal variants may
also be substantially larger than that of the common var-
iant that tags them. Accepting the possibility of synthetic
associations means that follow-up sequencing efforts
for GWA study results should extend well beyond the
block of linkage disequilibrium surrounding a discovery
variant that is defined by common variants.
Although there has been little systematic effort
devoted to finding synthetic associations, it has been
shown that rare or less common variants in some of the
same genomic regions as GWA signals influence dis-
ease risk. For example, one region that has been asso-
ciated with height through GWA studies contains the
gene growth differentiation factor 5 (GDF5), which was
already known to contain rare variants that result in
skeletal disorders45,52. In a study of type 1 diabetes that
sequenced the area around a GWA signal in the gene
interferon induced with helicase C domain 1 (IFIH1),
four rare or less common functional variants were iden-
tified that were protective in addition to the originally
associated common variant44. More recently, a com-
mon variant (MAF of 0.194 in european-Americans) in
C20orf194 was found in a GWA study to be associated
with haemolytic anaemia in response to interferon-α and
ribavirin treatment (for chronic hepatitis C infection)
with a p-value of 1.1 × 10–45 (Ref. 27). However, follow-
up genotyping of two functional variants with lower
frequencies (MAFs of 0.076 and 0.123) in the nearby
gene inosine triphosphatase (ITPA) showed that these
variants entirely accounted for the original GWA signal.
Table 1 | potential frequencies of causal variants in complex traits
Variant class
Very common
Minor allele frequency
Between 5 and 50%
implications for analysis
Amenable to association analysis using current genome-wide
association methods
Less common Between 1 and 5%
Amenable to association analysis using variants catalogued
in the 1000 Genomes Project
Rare (but not
private)
Less than 1% but still polymorphic in
one or more major human populations
Amenable to framework of extreme phenotype
resequencing, as well as co-segregation in families
PrivateRestricted to probands and immediate
relatives
Difficult to analyse except through co-segregation in families.
As linkage evidence will (by definition) be modest, discovery
would be limited to the most recognizable of variants
REVIEWS
418 | juNe 2010 | VOluMe 11
www.nature.com/reviews/genetics
© 20 Macmillan Publishers Limited. All rights reserved10

Page 5

Normal copy of gene with no mutations
a Haploinsufficiency
b Revealing of recessive mutations
Patient 1
One copy of gene contains a recessive
mutation for epilepsy
Deletion of gene on homologous chromosome
reveals recessive mutation for epilepsy
Deletion of gene on homologous chromosome
results in disease-causing haploinsufficiency
Patient 2
One copy of gene contains a recessive
mutation for schizophrenia
Deletion of gene on homologous chromosome
reveals recessive mutation for schizophrenia
1000 Genomes Project
An international research
consortium that will sequence
the genomes of 1,200
individuals of various
ethnicities. Most individuals
will be sequenced to low
coverage, or in exons only. The
goals are to catalogue human
variation with minor allele
frequencies of ~1% or greater
and to refine and optimize
strategies for sequencing
large numbers of genomes.
Coverage
The number of sequence reads
that have alignments that
overlap a certain position.
Because current sequencing
strategies produce random
reads, resulting in an uneven
distribution of reads across
the genome, a high average
coverage is required to assure
that most bases in the genome
are covered by multiple reads.
Furthermore, using a model in which each minor allele,
for either lower frequency variant, was counted once,
the p-value dropped to 2.2 × 10–92. This is one of the
few examples in which the causal variants from a GWA
signal have been definitively identified, and in some
ways the findings are consistent with the characteristics
of a synthetic association, in which multiple functional
variants load up in the same direction on a haplotype
defined by a more common variant included on the gene
chip. In this case, the causal variants explaining the asso-
ciation are common, perhaps because the trait relates to
a drug response53 (BOX 2).
strategies for identifying disease-causing variants
The basic paradigm used in the recently completed
phase of GWA studies was to catalogue very common
variants and genotype them either directly or indirectly
(through linkage disequilibrium) in cases and controls.
These GWA studies successfully represented variants
with a frequency of ~5% or higher in the general popu-
lation. If variants that are slightly less common than this,
in the range of 1–5%, contribute significantly to com-
mon diseases, then a simple extension of this paradigm is
appropriate. This line of thought implicitly underlies the
1000 Genomes Project, which is extending the catalogue
of known human variants down to a frequency near 1%
(see the 1000 Genomes website). Chips for a new wave
of GWA studies, which will account for these less com-
mon variants, are already being designed by companies
such as Illumina and Affymetrix. A second analysis of
phenotypes using the millions of genotypes available
on these chips is likely to identify some new associa-
tions. If, however, much of the causation is due to rare
variants, such an approach is unlikely to uncover much
more of the genetic control of diseases than has already
been revealed by the current GWA efforts. The CNVs
that have recently been implicated in many common
diseases generally have population frequencies that are
considerably below this threshold46,54,55.
To capture such rare variants, it will be necessary to
sequence entire genomes of cases instead of genotyp-
ing a catalogue of variants. up until recently, this was a
daunting task. However, new machines56 can currently
sequence 55 billion bases in about 10 days; although
this rate has been increasing with remarkable rapidity,
the cost per accurate base call is still the most impor-
tant parameter and needs improving. This impressive
amount of data provides sufficient ‘coverage’ to iden-
tify most variants present in a given genome, and the
increase in sequencing capacity is allowing more labo-
ratories to sequence whole genomes. Although sample
sizes would have to be large to identify rare variants of
small effect, there are many examples of variants (or
genotypes) with large enough effects to be strongly
enriched in specific populations. For example, indi-
viduals who are homozygous for the C-C chemokine
receptor 5 Δ32 (CCR5Δ32) deletion, who comprise
around 1% of the general european population, are
virtually immune to HIV infection57–59. This protection
increases the frequency of such homozygotes by up to
20-fold in individuals who have been highly exposed
to HIV and yet remain uninfected60. In another exam-
ple, the frequency of the risk factor human leukocyte
antigen-B*5701 (HLA‑B*5701) is increased as much
as 30-fold in individuals who are hypersensitive to the
antiretroviral drug abacavir compared with the general
population61–63. Although neither of these variants is
rare, they show that the frequency of variants influ-
encing complex traits can be so enriched in carefully
defined populations that they could be readily identified
even in a data set as large as would be generated in a
whole-genome sequencing study.
Whole-genome and whole-exome sequencing. The most
comprehensive study of the role of inherited variation
in diseases will involve whole-genome sequencing of all
enrolled subjects. Such studies will eventually be carried
out in a manner similar to GWA studies, with very large
sample sizes that will provide sufficient statistical evi-
dence to implicate variants on the basis of association
evidence alone. In the interim, it will be important to
focus on designs that are optimized to detect the role of
causal variants in smaller samples.
Proof-of-concept examples of the identification
of rare, disease-causing variants are now available for
whole-genome and whole-exome sequencing strategies.
For example, Ng et al.64 sequenced the exomes of four
unrelated cases with Freeman–Sheldon syndrome and
eight controls. Although the cause of this disease was
already known, the authors showed that the causal
Figure 1 | Role of deletions in disease predisposition. The figure shows a
heterozygous deletion (brackets) and five genes (blue boxes) on the homologous
chromosome, with lightning bolts representing mutations. Deletions can be linked
to a disease either by causing a deleterious haploinsufficiency (a) or by revealing
deleterious recessive mutations (b). In b, the recessive mutations that are revealed
could predispose to different conditions. Because individuals bearing a recessive
mutation predisposing to one condition or another would effectively become
homozygotes for those mutations in the presence of a deletion, the same deletion
can become associated with multiple conditions. For example, deletion of a region at
2p16.3 has been shown to be associated with schizophrenia, autism and mental
retardation54,101–103. The easiest way to resolve these possibilities will be through
sequencing of the homologous chromosome in individuals with the relevant deletions.
REVIEWS
NATuRe ReVIeWS | Genetics
VOluMe 11 | juNe 2010 | 419
© 20 Macmillan Publishers Limited. All rights reserved 10

Page 6

Nature Reviews | Genetics
Common variant:
disease-associated allele
Rare variant: causes disease
Common variant:
alternative allele
Gene 1 Gene 2LD block
GWA ‘locus’ assuming
common causal variant
GWA ‘locus’ if rare variants
create ‘synthetic’ associations
Indel
A small insertion or deletion of
nucleotides. If it occurs in an
exon and is not a multiple of
three in length, it results in a
frameshift and usually the loss
of gene function.
Splice-site variant
A variant, usually found at the
intron–exon boundary, that
alters the splicing of an exon
to its surrounding exons.
Non-synonymous variant
A genetic variant that changes
a codon for one amino acid to
another amino acid. Many
non-synonymous variants are
well-tolerated, but others can
cause a disease.
Co-segregation
In the pedigree of a family with
a condition, the segregation
pattern shows how often the
putative causal variant is found
to coincide with the condition.
When a variant coincides with
the condition in a family, the
condition and the variant are
said to co-segregate.
variants were evident after whole-exome sequencing,
as the causal gene was the only one that contained at
least one coding indel or non-synonymous or splice-site
variant in all four cases that was not present in any of the
controls nor in dbSNP.
Choi et al.65 used whole-exome sequencing to discover
the cause of disease in an individual with an unclear diag-
nosis. They identified a small number of homozygous
missense mutations in positions that were highly con-
served from invertebrates to humans. One such variant
was in a gene known to cause congenital chloride-losing
diarrhoea, consistent with the patient’s symptoms.
Most importantly, whole-genome and whole-exome
sequencing have been used to identify the previously
unknown causes of diseases. For example, Ng et al.66
discovered the unknown cause of Miller syndrome
by performing whole-exome sequencing in four cases
from three kindreds and eight controls. They located the
causal gene by identifying genes that had coding indels,
non-synonymous variants or splice-site variants in all of the
cases but in none of the controls. Although the causes of
such Mendelian diseases will be easier to discover than
those of more complex diseases, these successes indicate
that whole-genome sequencing of even a small number
of individuals can identify causal variants.
until complete genomic sequencing is inexpensive
enough to use in the large sample sizes that would be
needed to perform whole-genome association studies
with no a priori weighting of putative functional vari-
ants, two designs are likely to be the primary engines
of discovery. The first design involves selecting families
that have multiple affected individuals (family-based
sequencing), and the second approach involves select-
ing individuals that are at the extreme ends of a trait
distribution (extreme-trait designs) (fIG. 3). extreme-
trait designs will be particularly important for identify-
ing variants that are rare but not private and that have
modest to high effect sizes.
Sequencing affected individuals in families. under this
design, a family with multiple individuals affected with
a common disease would be selected. One economical
design could involve the initial sequencing of the most
distantly related, co-affected family members and the
identification of overlapping variants, as the more dis-
tantly related the co-affected individuals, the fewer genetic
variants they will share. However, even distant relatives
will share too many variants to allow easy identifica-
tion of the causal variants, even when relatives share the
same rare causal variants (which will not always be
the case). The list of variants will therefore need to
be further screened by function, zygosity, population fre-
quency and/or the type of gene affected to focus on the
most likely candidates. Promising variants would then
be checked for co-segregation in the family. However, for
most common diseases, the evidence emerging from co-
segregation analyses will not be sufficient to distinguish
among a large set of candidate pathogenic variants. For
example, consider a pedigree involving 16 individuals
who are affected with bipolar disorder and 35 who are
unaffected67; the probability of co-segregation by chance
would be 0.0003. This statistic means that co-segregation
analysis would be a possible approach for prioritizing
among a small set of key functional variants, such as
protein-truncating variants, but would be insufficient
for distinguishing among more general classes, such
as all coding variants, let alone all identified variants.
With such a small discovery sample size, the success-
ful narrowing down of possible causal variants will be
heavily dependent on the causal variant being rare or
private and of obvious functional consequence. Most
likely, the discovery paradigm will require combining
co-segregation evidence with evidence from additional
Figure 2 | synthetic associations. Shown is a set of
chromosomes (grey boxes) containing a common variant
(green and yellow boxes) and four rare variants (red Xs).
At the top, the area of linkage disequilibrium (LD)
surrounding the common variant is shown, as are nearby
genes. Each rare variant shown can cause the disease,
and collectively they are more common on the
haplotype containing the green version of the common
variant. In a genome-wide association (GWA) study for
this disease, the rare variants will not be directly
assessed but will often create a signal that is credited
to one or more common variants. In most cases (as
illustrated), the signal credited to the common variant
would be far weaker than the real effects of the causal
variants. Furthermore, because the LD block containing
the common variants does not extend to the gene
containing the causal variants, one might assume that
the causal variant must be regulatory. However, in
synthetic associations, causal variants can lie within a
much larger region than the LD block surrounding the
associated, common variant.
REVIEWS
420 | juNe 2010 | VOluMe 11
www.nature.com/reviews/genetics
© 20 Macmillan Publishers Limited. All rights reserved10

Page 7

Variant of
large effect
at low
frequency
in general
population
Sequencing affected relatives
Identify shared variants
that are likely to
cause disease
Genotype candidate variants
Look for co-segregation
with disease in family
Whole-genome sequencing
Rare variant enriched in
population with extreme
phenotype
Targeted genotype
Rare variant associates
with phenotype in larger
population
a Sequencing affected individuals in families
b Extreme-trait sequencing
families and association evidence that is based on the
typing of candidate variants in large case–control
cohorts. The development of appropriate test statistics
that combine these different lines of evidence is a current
priority for the field.
This suggests that modest linkage evidence may be a
promising way to narrow down the region (and variants)
of interest. For example, Sobreira et al. discovered the
cause of metachondromatosis, a dominant Mendelian
disease with incomplete penetrance, by first running a
linkage analysis on a small pedigree with DNA for only
12 family members68. The analysis implicated six regions,
with logarithm of the odds (lOD) scores of 2.5, 1.8 and
1.07, that comprised a total of 42 Mb. The authors then
sequenced the entire genome of one affected patient
from this pedigree and eight unrelated controls, and
focused analysis on private, good-quality variants with
functional consequences for protein-coding genes in
the 42 linked megabases. This allowed them to iden-
tify the causative protein-truncating variant, which was
found in a linkage peak with the second-highest lOD
score; this variant was confirmed by locating a similar
variant in a second, unrelated family with the disease.
Although these results are for a Mendelian disease, and
similar studies for common diseases are unlikely to be
this simple, they show that a combination of weak link-
age evidence and functional prioritization can be used
to identify the causative variant by sequencing only one
patient genome.
Extreme-trait sequencing. A basic extreme-trait design
would be to sequence a small, carefully selected pop-
ulation at one or both ends of the extremes of a phe-
notype (this idea has been discussed previously69 and
an exploration of power in this context is available70).
Obvious examples include individuals who are known
to be highly exposed to HIV but remain uninfected or
individuals at the extremes of the distributions for blood
pressure. Because variants that contribute to the trait
will be enriched in frequency in such a population, even
small sample sizes may suggest many candidate variants
that can then be genotyped for confirmation in a much
larger group of samples. Consider a case in which a vari-
ant is enriched 30-fold in such an extreme population, as
with hypersensitivity to abacavir. If this variant were rare
in the general population, say with a MAF of 0.1%, one
would need to sequence about 30 extreme individuals
to see it once. Although identifying enrichment for this
particular variant would probably involve the sequenc-
ing of at least 100 extreme individuals, this would not
Figure 3 | strategies for identifying disease-causing
variants. Two discovery strategies using next-generation
sequencing are likely to be of particular importance
while sequencing costs remain high: sequencing in families
with multiple affected individuals (a; shaded individuals
are affected), and sequencing individuals at one or both
ends of a trait distribution (b; the size of the individual
represents the severity of the phenotype). In the case of
family-based sequencing, it may often prove economical
to first sequence the most distantly related co-affected
individuals. Under either scenario, it is likely that follow-up
genotyping in additional families or cohorts will be of
particular importance to confirm the role of candidate
variants. Red stars represent the causal variant. In a, stars
of other colours represent variants that are shared by the
sequenced individuals and do not segregate in the family.
REVIEWS
NATuRe ReVIeWS | Genetics
VOluMe 11 | juNe 2010 | 421
© 20 Macmillan Publishers Limited. All rights reserved10

Page 8

be the case if multiple rare variants in the same gene
affect the phenotype. Searching for genes with an
enrichment of rare variants in extreme individuals will
implicate this gene even at a low number of sequenced
genomes, although again this will be heavily dependent
on the variants being of obvious function. Although cur-
rently the only examples of the identification of causal
variants by whole-genome or whole-exome sequencing
come from Mendelian diseases, Ng et al.64,66 used this
technique to identify genes that were enriched for rare
variants in a very small number of cases versus controls:
the same variant did not need to be present in all cases.
For the follow-up of variants identified in extreme-trait
sequencing, family members of the extreme individuals
will be invaluable for confirming potentially causal vari-
ants through co-segregation analysis. This strategy will
be an effective complement to extreme-trait sequenc-
ing in nearly every situation in which family members
who show variability in the trait are available for such
a follow-up.
These two strategies will be some of the most effec-
tive methods for using whole-genome sequence data
from a small number of individuals. It seems likely that
for many diseases that affect only a small proportion of
the population (for example, schizophrenia, epilepsy and
amyotrophic lateral sclerosis), both designs will be used
and will complement one another. However, such studies
will also face many challenges. The success of family-
based sequencing will depend on the rarity of causal
variants in the population at large, and the genetic bases
of conditions that affect a relatively small proportion of
the population are likely to be more easily identified than
conditions, such as type 2 diabetes, that affect a much
higher proportion of the population. In the case of such
common diseases, even families that carry relatively high
impact rare variants will show imperfect co-segregation,
and the enrichment of specific causal variants among
affected, unrelated individuals will be much lower. For
extreme-trait sequencing, accurate phenotyping will be
of vital importance; as only a small number of individu-
als will be sequenced, the inclusion of even a small pro-
portion of misclassified individuals because of factors
such as measurement error could affect the analysis. At
small sample sizes, both methods will also depend on the
causal variants being of obvious functional consequence.
Furthermore, the accurate calling of variants and suf-
ficient coverage to identify the presence of causal vari-
ants in all individuals will be crucial. Finally, the sheer
number of rare and private variants that will be identi-
fied will make identification of the causal variants dif-
ficult; this problem will be compounded by both locus
and allelic heterogeneity.
the importance of recognizability
In both family-based and extreme-trait discovery
paradigms, a key determinant of statistical power will
be whether causal sites tend to be ‘recognizable’ —
that is, whether the variants have an obvious function.
examples of recognizable variants would be ones that
delete some or all of a gene, introduce a premature stop
site into a protein or result in a non-conservative amino
acid substitution. The general failure of GWA studies of
very common variants to identify causal sites has led to
the widespread belief that most signals must reflect sub-
tle regulatory variants that are not easily recognized69–71.
However, this view is based on assumption, not observa-
tion, and it is possible that the variants that contribute to
common diseases behave more like those that contribute
to Mendelian diseases, in which the variants tend to have
obvious effects72,73. In contrast to this lack of information
about the underlying causes of complex diseases, tens
of thousands of variants that cause Mendelian diseases
have been catalogued and characterized. Most of these
variants are readily recognizable72,73. If common diseases
have genetic causes that are, in this way, similar to those
of Mendelian diseases, we can expect most of them to
be missense, nonsense or splicing variants, or obviously
functional deletions or insertions.
To illustrate the importance of recognizability, con-
sider the number of candidate variants in different
functional groups. For example, Shianna et al. com-
pared the frequencies of different classes of variations
in ten case and ten control genomes (K. V. Shianna et al.,
unpublished data). There were 383,913 variants (single-
nucleotide variants and indels) present in at least two
cases and no controls. Testing such a large number of
variants would require a p-value of 1.3 × 10–7 to declare
a significant association. However, if testing is restricted
to only variants that affect the coding sequence, this
number drops to 2,354, which requires a p-value of only
2.2 × 10–5. If testing is restricted to only protein-truncating
variants, the number drops further to 152, which requires
a p-value of 3.3 × 10–4. It will therefore be essential to
develop methods that prioritize variants based on the
likelihood that they contribute to disease (BOX 3).
future directions
Both the professional and lay communities have adjusted
to the apparently limited impact of genetic differences
on common diseases. For example, there seems to have
been a dampening of enthusiasm in industry for the use
of genetic association data in prioritizing drug targets,
whereas the provision of genetic risk profiles has become
commonplace and even allows social engagement at
‘spit parties’, at which party-goers produce saliva sam-
ples for genotyping. This attitude is in stark contrast to
the more than 25 million Americans with rare genetic
diseases (see the uS National Institutes of Health Office
of Rare Diseases Research website), many of which can
be diagnosed with highly accurate genetic tests (see
the GeneTests website). However, these reactions to the
characteristics of common genetic variation may be at
odds with rare genetic contributions to common dis-
eases. The identification of rare variants that influence
diseases is suddenly feasible thanks to next-generation
sequencing, and it is possible that discovery will move
faster than generally anticipated. If rare, obviously func-
tional variants have a big influence on common diseases,
whole-genome sequencing will allow definitive connec-
tions to be rapidly established between specific genes
and many important common diseases. Definitive con-
nections — for example, a clearly functional mutation
REVIEWS
422 | juNe 2010 | VOluMe 11
www.nature.com/reviews/genetics
© 20 Macmillan Publishers Limited. All rights reserved10

Page 9

Compound heterozygote
When an individual inherits two
different recessive mutations,
one from each parent, in
the same gene that cause the
same phenotype. An example
would be a single-nucleotide
variant causing a codon for an
amino acid to be changed into
a stop codon in one allele
and a 4-bp deletion in the
other allele: each of these
variants knock out their
respective allele, resulting in
neither copy functioning.
in a single gene conferring a strongly elevated risk of a
disease — would provide validated therapeutic targets
for the pharmaceutical industry. Indeed, it may not be
an overstatement to suggest that genetic discovery could
be the most likely avenue for ameliorating the ongoing
crisis in global drug development74. However, this pre-
diction assumes that rare variants will be found that have
large influences on common diseases, that their biologi-
cal functions will be obvious and that locus and allelic
heterogeneity will not obfuscate insights into the mecha-
nisms of disease. How often these assumptions will hold
is currently unknown and will largely determine the rate
of discovery in the coming years.
Before discovery genetics focused on rare variants
can be systematized in a manner analogous to what
was successfully accomplished in the study of common
variants, many technical challenges will need to be over-
come. The development of analysis techniques to cope
with the millions of variants called per genome will be
a high priority, as will the development of techniques
that can combine data about different rare variants into
one analysis. New software for accurately identifying
CNVs and inversions will also be important, as there
is currently no highly accurate method for identifying
even a high proportion of all such variants present in
a genome. even methods for calling simpler variants,
such as those affecting a single nucleotide, will need to
be improved, as currently there are tens of thousands of
false-positive variant calls per genome75, which further
complicates the search for causal variants. even more
effort must be applied to accurately calling small inser-
tions and deletions, which are currently called much
less accurately than single site changes. Furthermore, it
will be essential for researchers to carefully choose cases
and controls for each study: as costs restrict sample size,
the success of each project will be highly dependent
on the careful selection of individuals to sequence. If rare
variants of large effect are found, a transition towards
the study of particular pathogenic variants that strongly
influence diseases will greatly increase the importance of
designs that allow for the evaluation of family members
carrying putative causal variants. The availability of large
control cohorts who can be recalled and phenotypically
evaluated will also be crucial, especially if unscreened
controls from the general population are used for mul-
tiple studies involving different phenotypes. unlike
variants with weak influences, which could be expected
to appear in the population at large without much phe-
notypic effect, a variant of strong effect is less likely to
appear in an individual without any phenotypic conse-
quence. Confirmation of such potentially causal variants
will therefore often require the careful evaluation of the
phenotype in any controls who are carriers.
until the cost of whole-genome sequencing drops to a
level that is similar to whole-exome sequencing (includ-
ing the cost of the exome capture step), whole-exome
sequencing will be a good alternative strategy for many
studies. Because the most obvious disease-influencing
variants will be the clearly functional ones, whole-exome
sequencing should pack most of the variation of interest
into a much smaller, more cost-effective and more easily
interpreted bundle. However, there are drawbacks to this
technique: most importantly, it almost entirely misses
structural variation. Whole-exome sequencing is also
restricted to a certain set of exons: if causal variants lie
in exons that are not targeted, they will not be identified.

Box 3 | Bioinformatic approaches and challenges for whole-genome sequencing studies
The analysis of sequence data presents a fundamentally different challenge from the analysis of a targeted set of
polymorphisms, as was the paradigm for genome-wide association (GWA) studies. The most fundamental distinction
is that in a sequencing study, it is necessary to assess all variants that have been identified, and many of these variants
will not have yet been included in polymorphism databases. To deal with such data, it is necessary to have an
integrated bioinformatic environment that begins with the accurate calling of variants based on the sequence data
and that includes the categorization of all identified variants, novel or known, into functional categories.
The challenges begin with variant calling. Even after the sequence reads for a whole genome have been aligned,
which takes a few days using programs such as BWA86 or Bowtie87, single-nucleotide variants and indels must be
identified, a process that takes programs such as SAMTools88 or ssahaSNP a few more days. The identification of other
classes of variants, such as copy-number variants or inversions, requires the use of variant calling methods that are
based on a combination of read depth and read pair methods89–91. Following these variant calling procedures, it is
crucial that the called variants are annotated, distinguished in terms of their most likely functional consequences and
presented to researchers in a format that can be used for analysis. Many existing visualization tools can conveniently
display the aligned short reads from sequence data in the context of annotation features (for example, genes
and known variants); however, they were not designed to actually annotate the variants92–95. To perform annotation and
analysis, many groups have constructed local annotation servers64, used their own in-house annotation pipelines96–99
or sometimes developed new genome browsers100. However, there is still a need for an efficient and publicly available
software tool that can annotate, organize and interpret the millions of variants that are called in each genome.
Our group has developed a unique software environment called SequenceVariantAnalyzer that takes as input the
variants called for each genome sequenced, as well as the quality scores and coverage statistics, and annotates and
stores them for subsequent analysis. Individual genomes are classified as either cases or controls, and variants can be
screened by function, location, zygosity and quality. The software comes with a number of functions for prioritizing
genes or variants, such as identifying variants that are found in the homozygous state only in cases, searching for genes
that are enriched with truncating variants only in cases and not in controls, and identifying compound heterozygotes.
Software that performs these functions makes the analysis of whole-genome sequence data possible, and will be
essential for human geneticists in the coming years.
REVIEWS
NATuRe ReVIeWS | Genetics
VOluMe 11 | juNe 2010 | 423
© 20 Macmillan Publishers Limited. All rights reserved10

Page 10

Additionally, the capture methods currently used are
nonspecific enough that they require the sequencing of a
far greater number of bases than expected based on the
size of the exome, which makes whole-exome sequenc-
ing prices comparable to those of low-coverage whole-
genome sequencing. However, low-coverage sequencing
will miss many of the variants present in only a single
individual. For these reasons, high-coverage whole-
genome sequencing will be the method of choice once it
becomes more affordable, and the rapid increase in the
sequencing capacity of existing platforms, as well as
the development of new, less-expensive platforms76, sug-
gests this will not be more than a few years away.
We predict that, within the next 5 years, rare and
private variants with moderate to large effects on
many complex traits will be discovered. With regard to
personal genomics, it is possible that genome sequenc-
ing could provide specific enough information about the
risk of common diseases to influence lifestyle choices
and the use of relatively non-invasive monitoring pro-
grams (for example, imaging); it might even lead to an
expansion of interest in couples-based and fetal screen-
ing. This outlook is tempered by past experience that
has shown that risk factors for disease, even when well-
characterized, are sometimes not particularly useful in
the clinical setting77; this may well continue to be true
even if rare variants of large effect are found to influ-
ence common diseases. However, we may be cautiously
optimistic that if a large enough set of definitive con-
nections can be established between specific genes and
diseases, a subset of these discoveries may lead to clear
clinical deliverables.
1.
Maher, B. Personal genomes: the case of the missing
heritability. Nature 456, 18–21 (2008).
One of the first articles to explicitly recognize that
GWA studies explain a small part of the genetic
components of many diseases.
Kasowski, M. et al. Variation in transcription factor
binding among humans. Science 328, 232–235
(2010).
Pickrell, J. K. et al. Understanding mechanisms
underlying human gene expression variation with RNA
sequencing. Nature 464, 768–772 (2010).
Heinzen, E. L. et al. Tissue-specific genetic control of
splicing: implications for the study of complex traits.
PLoS Biol. 6, e1 (2008).
Stankiewicz, P. & Lupski, J. R. Structural variation in
the human genome and its role in disease. Annu. Rev.
Med. 61, 437–455 (2010).
Dickson, S. P., Wang, K., Krantz, I., Hakonarson, H. &
Goldstein, D. B. Rare variants create synthetic
genome-wide associations. PLoS Biol. 8, e1000294
(2010).
Bodmer, W. & Bonilla, C. Common and rare variants in
multifactorial susceptibility to common diseases.
Nature Genet. 40, 695–701 (2008).
Schork, N. J., Murray, S. S., Frazer, K. A. & Topol, E. J.
Common vs. rare allele hypotheses for complex
diseases. Curr. Opin. Genet. Dev. 19, 212–219 (2009).
Pritchard, J. K. Are rare variants responsible for
susceptibility to complex diseases? Am. J. Hum. Genet.
69, 124–137 (2001).
10. International HapMap Consortium. The International
HapMap Project. Nature 426, 789–796 (2003).
11. Pritchard, J. K. & Cox, N. J. The allelic architecture of
human disease genes: common disease-common
variant.or not? Hum. Mol. Genet. 11, 2417–2423
(2002).
12. Stephens, J. W. & Humphries, S. E. The molecular
genetics of cardiovascular disease: clinical
implications. J. Intern. Med. 253, 120–127 (2003).
13. Plomin, R., Haworth, C. M. & Davis, O. S.
Common disorders are quantitative traits. Nature Rev.
Genet. 10, 872–878 (2009).
14. Reich, D. E. & Lander, E. S. On the allelic spectrum of
human disease. Trends Genet. 17, 502–510 (2001).
15. Tandon, R., Keshavan, M. S. & Nasrallah, H. A.
Schizophrenia, ‘just the facts’ what we know in 2008.
2. Epidemiology and etiology. Schizophr. Res. 102,
1–18 (2008).
16. Crow, T. J. How and why genetic linkage has not solved
the problem of psychosis: review and hypothesis.
Am. J. Psychiatry 164, 13–21 (2007).
17. Serretti, A. & Mandelli, L. The genetics of bipolar
disorder: genome ‘hot regions’, genes, new potential
candidates and future directions. Mol. Psychiatry 13,
742–771 (2008).
18. Risch, N. & Merikangas, K. The future of genetic
studies of complex human diseases. Science 273,
1516–1517 (1996).
19. Steinberg, M. H. & Adewoye, A. H. Modifier genes and
sickle cell anemia. Curr. Opin. Hematol. 13, 131–136
(2006).
20. Thein, S. L. & Menzel, S. Discovering the genetics
underlying foetal haemoglobin production in adults.
Br. J. Haematol. 145, 455–467 (2009).
2.
3.
4.
5.
6.
7.
8.
9.
21. Manolio, T. A. et al. Finding the missing heritability
of complex diseases. Nature 461, 747–753
(2009).
22. Zeggini, E. et al. Meta-analysis of genome-wide
association data and large-scale replication identifies
additional susceptibility loci for type 2 diabetes.
Nature Genet. 40, 638–645 (2008).
23. Shi, J. et al. Common variants on chromosome
6p22.1 are associated with schizophrenia. Nature
460, 753–757 (2009).
24. SEARCH Collaborative Group et al. SLCO1B1 variants
and statin-induced myopathy — a genomewide study.
N. Engl. J. Med. 359, 789–799 (2008).
25. Tanaka, Y. et al. Genome-wide association of IL28B
with response to pegylated interferon-α and ribavirin
therapy for chronic hepatitis C. Nature Genet. 41,
1105–1109 (2009).
26. Daly, A. K. et al. HLA-B*5701 genotype is a major
determinant of drug-induced liver injury due to
flucloxacillin. Nature Genet. 41, 816–819 (2009).
27. Fellay, J. et al. ITPA gene variants protect against
anemia in patients treated for chronic hepatitis C.
Nature 464, 405–408 (2010).
28. Ge, D. et al. Genetic variation in IL28B predicts
hepatitis C treatment-induced viral clearance.
Nature 461, 399–401 (2009).
29. Need, A. C. et al. A genome-wide study of common
SNPs and CNVs in cognitive performance in the
CANTAB. Hum. Mol. Genet. 18, 4650–4661 (2009).
30. Cirulli, E. T. et al. Common genetic variation and
performance on standardized cognitive tests. Eur.
J. Hum. Genet. 3 Feb 2010 (doi:10.1038/
ejhg.2010.2).
31. Bhattacharjee, S. et al. Using principal components
of genetic variation for robust and powerful
detection of gene–gene interactions in case–control
and case-only studies. Am. J. Hum. Genet. 86,
331–342 (2010).
32. Kong, A. et al. Parental origin of sequence variants
associated with complex diseases. Nature 462,
868–874 (2009).
33. Meyer, K. B. et al. Allele-specific up-regulation of
FGFR2 increases susceptibility to breast cancer.
PLoS Biol. 6, e108 (2008).
34. Chang, B. L. et al. Fine mapping association study
and functional analysis implicate a SNP in MSMB at
10q11 as a causal variant for prostate cancer risk.
Hum. Mol. Genet. 18, 1368–1375 (2009).
35. Hughes, A. E. et al. A common CFH haplotype, with
deletion of CFHR1 and CFHR3, is associated
with lower risk of age-related macular degeneration.
Nature Genet. 38, 1173–1177 (2006).
36. Hageman, G. S. et al. Extended haplotypes in the
complement factor H (CFH) and CFH-related (CFHR)
family of genes protect against age-related macular
degeneration: characterization, ethnic distribution and
evolutionary implications. Ann. Med. 38, 592–604
(2006).
37. Spencer, K. L. et al. Deletion of CFHR3 and CFHR1
genes in age-related macular degeneration. Hum. Mol.
Genet. 17, 971–977 (2008).
38. Frayling, T. M. Genome-wide association studies
provide new insights into type 2 diabetes aetiology.
Nature Rev. Genet. 8, 657–662 (2007).
39. McCarthy, M. I. & Hirschhorn, J. N. Genome-wide
association studies: potential next steps on a genetic
journey. Hum. Mol. Genet. 17, R156–R165 (2008).
40. Bouatia-Naji, N. et al. A variant near MTNR1B is
associated with increased fasting plasma glucose
levels and type 2 diabetes risk. Nature Genet. 41,
89–94 (2009).
41. Frayling, T. M. et al. A common variant in the
FTO gene is associated with body mass index
and predisposes to childhood and adult obesity.
Science 316, 889–894 (2007).
42. Todd, J. A. et al. Robust associations of four new
chromosome regions from genome-wide analyses
of type 1 diabetes. Nature Genet. 39, 857–864
(2007).
43. Pillai, S. G. et al. A genome-wide association study in
chronic obstructive pulmonary disease (COPD):
identification of two major susceptibility loci.
PLoS Genet. 5, e1000421 (2009).
44. Nejentsev, S., Walker, N., Riches, D., Egholm, M. &
Todd, J. A. Rare variants of IFIH1, a gene implicated in
antiviral responses, protect against type 1 diabetes.
Science 324, 387–389 (2009).
This study showed that rare variants in the same
region as a GWA signal for diabetes were
associated with the disease.
45. Sanna, S. et al. Common variants in the GDF5–
UQCC region are associated with variation in
human height. Nature Genet. 40, 198–203 (2008).
46. Stefansson, H. et al. Large recurrent microdeletions
associated with schizophrenia. Nature 455, 232–236
(2008).
One of the first studies to identify rare CNVs
associated with a common disease.
47. Gruber, S. B. et al. Genetic variation in 8q24
associated with risk of colorectal cancer. Cancer Biol.
Ther. 6, 1143–1147 (2007).
48. Tomlinson, I. et al. A genome-wide association scan of
tag SNPs identifies a susceptibility variant for
colorectal cancer at 8q24.21. Nature Genet. 39,
984–988 (2007).
49. Zanke, B. W. et al. Genome-wide association scan
identifies a colorectal cancer susceptibility locus on
chromosome 8q24. Nature Genet. 39, 989–994
(2007).
50. Prokunina-Olsson, L. & Hall, J. L. No effect of
cancer-associated SNP rs6983267 in the 8q24
region on co-expression of MYC and TCF7L2 in normal
colon tissue. Mol. Cancer 8, 96 (2009).
51. Sotelo, J. et al. Long-range enhancers on 8q24
regulate c-Myc. Proc. Natl Acad. Sci. USA (2010).
52. Weedon, M. N. et al. Genome-wide association
analysis identifies 20 loci that influence adult height.
Nature Genet. 40, 575–583 (2008).
53. Goldstein, D. B. Common genetic variation and
human traits. N. Engl. J. Med. 360, 1696–1698
(2009).
54. Need, A. C. et al. A genome-wide investigation of
SNPs and CNVs in schizophrenia. PLoS Genet. 5,
e1000373 (2009).
55. Kumar, R. A. et al. Recurrent 16p11.2 microdeletions
in autism. Hum. Mol. Genet. 17, 628–638 (2008).
56. Metzker, M. L. Sequencing technologies — the next
generation. Nature Rev. Genet. 11, 31–46 (2010).
REVIEWS
424 | juNe 2010 | VOluMe 11
www.nature.com/reviews/genetics
© 20 Macmillan Publishers Limited. All rights reserved10

Page 11

57. Dean, M. et al. Genetic restriction of HIV-1 infection and
progression to AIDS by a deletion allele of the CKR5
structural gene. Hemophilia Growth and Development
Study, Multicenter AIDS Cohort Study, Multicenter
Hemophilia Cohort Study, San Francisco City Cohort,
ALIVE Study. Science 273, 1856–1862 (1996).
58. Liu, R. et al. Homozygous defect in HIV-1 coreceptor
accounts for resistance of some multiply-exposed
individuals to HIV-1 infection. Cell 86, 367–377 (1996).
59. Samson, M. et al. Resistance to HIV-1 infection in
caucasian individuals bearing mutant alleles of
the CCR-5 chemokine receptor gene. Nature 382,
722–725 (1996).
60. Huang, Y. et al. The role of a mutant CCR5 allele
in HIV-1 transmission and disease progression.
Nature Med. 2, 1240–1243 (1996).
61. Mallal, S. et al. Association between presence
of HLA-B*5701, HLA-DR7, and HLA-DQ3 and
hypersensitivity to HIV-1 reverse-transcriptase
inhibitor abacavir. Lancet 359, 727–732 (2002).
62. Martin, A. M. et al. Predisposition to abacavir
hypersensitivity conferred by HLA-B*5701 and a
haplotypic Hsp70-Hom variant. Proc. Natl Acad. Sci.
USA 101, 4180–4185 (2004).
63. Young, B. et al. First large, multicenter, open-label
study utilizing HLA-B*5701 screening for abacavir
hypersensitivity in North America. AIDS 22,
1673–1675 (2008).
64. Ng, S. B. et al. Targeted capture and massively
parallel sequencing of 12 human exomes. Nature 461,
272–276 (2009).
The first study to show that next-generation
sequencing can be used to identify
disease-causing variants.
65. Choi, M. et al. Genetic diagnosis by whole exome
capture and massively parallel DNA sequencing.
Proc. Natl Acad. Sci. USA 106, 19096–19101 (2009).
The first study to diagnose a disease using
next-generation sequencing.
66. Ng, S. B. et al. Exome sequencing identifies the cause
of a mendelian disorder. Nature Genet. 42, 30–35
(2010).
67. Yang, S. et al. Genomic landscape of a three-
generation pedigree segregating affective disorder.
PLoS ONE 4, e4474 (2009).
68. Sobreira, N. L. M. et al. Whole genome sequencing of a
single proband together with linkage analysis identifies
a Mendelian disease gene. PLoS Genet. (in the press).
69. Manolio, T. A., Brooks, L. D. & Collins, F. S. A HapMap
harvest of insights into the genetics of common
disease. J. Clin. Invest. 118, 1590–1605 (2008).
70. Verlaan, D. J. et al. Targeted screening of
cis-regulatory variation in human haplotypes.
Genome Res. 19, 118–127 (2009).
71. Barrett, J. C. et al. Genome-wide association defines
more than 30 distinct susceptibility loci for Crohn’s
disease. Nature Genet. 40, 955–962 (2008).
72. Stenson, P. D. et al. The Human Gene Mutation
Database: 2008 update. Genome Med. 1, 13 (2009).
73. Botstein, D. & Risch, N. Discovering genotypes
underlying human phenotypes: past successes for
Mendelian disease, future approaches for complex
disease. Nature Genet. 33, 228–237 (2003).
A thoughtful overview of the kinds of mutations
responsible for Mendelian disease that provides
many insights about appropriate designs for
studying common disease.
74. Caskey, C. T. The drug development crisis: efficiency
and safety. Annu. Rev. Med. 58, 1–16 (2007).
75. Roach, J. C. et al. Analysis of genetic inheritance in a
family quartet by whole-genome sequencing. Science
10 Mar 2010 (doi:10.1126/science.1186802).
76. Drmanac, R. et al. Human genome sequencing using
unchained base reads on self-assembling DNA
nanoarrays. Science 327, 78–81 (2010).
77. Clayton, D. G. Prediction and interaction in complex
disease genetics: experience in type 1 diabetes.
PLoS Genet. 5, e1000540 (2009).
78. Wellcome Trust Case Control Consortium.
Genome-wide association study of 14,000 cases
of seven common diseases and 3,000 shared
controls. Nature 447, 661–678 (2007).
A technically important early study providing
well-powered GWA tests for multiple conditions.
79. Diabetes Genetics Initiative. Genome-wide association
analysis identifies loci for type 2 diabetes and
triglyceride levels. Science 316, 1331–1336 (2007).
80. Scott, L. J. et al. A genome-wide association study of
type 2 diabetes in Finns detects multiple susceptibility
variants. Science 316, 1341–1345 (2007).
81. Hamming, K. S. et al. Coexpression of the type 2
diabetes susceptibility gene variants KCNJ11 E23K
and ABCC8 S1369A alter the ATP and sulfonylurea
sensitivities of the ATP-sensitive K+ channel.
Diabetes 58, 2419–2424 (2009).
82. Nicolson, T. J. et al. Insulin storage and glucose
homeostasis in mice null for the granule zinc
transporter ZnT8 and studies of the type 2 diabetes-
associated variants. Diabetes 58, 2070–2083 (2009).
83. Gaulton, K. J. et al. A map of open chromatin in human
pancreatic islets. Nature Genet. 42, 255–259 (2010).
84. Motulsky, A. G. Drug reactions enzymes, and
biochemical genetics. JAMA 165, 835–837 (1957).
85. Ingelman-Sundberg, M. Genetic polymorphisms
of cytochrome P450 2D6 (CYP2D6): clinical
consequences, evolutionary aspects and functional
diversity. Pharmacogenomics J. 5, 6–13 (2005).
86. Li, H. & Durbin, R. Fast and accurate short read
alignment with Burrows–Wheeler transform.
Bioinformatics 25, 1754–1760 (2009).
87. Langmead, B., Trapnell, C., Pop, M. & Salzberg, S. L.
Ultrafast and memory-efficient alignment of short
DNA sequences to the human genome. Genome Biol.
10, R25 (2009).
88. Li, H. et al. The Sequence Alignment/Map format and
SAMtools. Bioinformatics 25, 2078–2079 (2009).
89. Hormozdiari, F., Alkan, C., Eichler, E. E. & Sahinalp, S. C.
Combinatorial algorithms for structural variation
detection in high-throughput sequenced genomes.
Genome Res. 19, 1270–1278 (2009).
90. Yoon, S., Xuan, Z., Makarov, V., Ye, K. & Sebat, J.
Sensitive and accurate detection of copy number
variants using read depth of coverage. Genome Res.
19, 1586–1592 (2009).
91. Simpson, J. T., McIntyre, R. E., Adams, D. J. &
Durbin, R. Copy number variant detection in inbred
strains from short read sequence data. Bioinformatics
26, 565–567 (2010).
92. Milne, I. et al. Tablet — next generation sequence
assembly visualization. Bioinformatics 26, 401–402
(2010).
93. Bao, H. et al. MapView: visualization of short reads
alignment on a desktop computer. Bioinformatics 25,
1554–1555 (2009).
94. Manske, H. M. & Kwiatkowski, D. P. LookSeq: a
browser-based viewer for deep sequencing data.
Genome Res. 19, 2125–2132 (2009).
95. Arner, E., Hayashizaki, Y. & Daub, C. O.
NGSView: an extensible open source editor for
next-generation sequencing data. Bioinformatics 26,
125–126 (2010).
96. Schuster, S. C. et al. Complete Khoisan and Bantu
genomes from southern Africa. Nature 463,
943–947 (2010).
97. Bentley, D. R. et al. Accurate whole human genome
sequencing using reversible terminator chemistry.
Nature 456, 53–59 (2008).
One of the first studies to sequence an entire
human genome using next-generation sequencing.
98. Wang, J. et al. The diploid genome sequence of an
Asian individual. Nature 456, 60–65 (2008).
99. Ng, P. C. et al. Genetic variation in an individual
human exome. PLoS Genet. 4, e1000160
(2008).
100. Axelrod, N. et al. The HuRef Browser: a web resource
for individual human genomics. Nucleic Acids Res. 37,
D1018–D1024 (2009).
101. Kirov, G. et al. Comparative genome hybridization
suggests a role for NRXN1 and APBA2 in
schizophrenia. Hum. Mol. Genet. 17, 458–465
(2008).
102. Friedman, J. M. et al. Oligonucleotide microarray
analysis of genomic imbalance in children with
mental retardation. Am. J. Hum. Genet. 79,
500–513 (2006).
103. Autism Genome Project Consortium et al. Mapping
autism risk loci using genetic linkage and
chromosomal rearrangements. Nature Genet. 39,
319–328 (2007).
Acknowledgements
We thank D. Ge, E. L. Heinzen, A. C. Need, J. C. Fellay,
J. M. Maia, E. K. Ruzzo and H. F. Willard for helpful comments
on the manuscript.
Competing interests statement
The authors declare no competing financial interests.
DAtABAses
Entrez Gene: http://www.ncbi.nlm.nih.gov/gene
CYP2D6 | GDF5 | IFIH1 | ITPA | KCNJ11 | SLC30A8 | TCF7L2 |
WFS1
OMIM: http://www.ncbi.nlm.nih.gov/omim
Freeman–Sheldon syndrome | metachondromatosis
furtHer inforMAtion
Authors’ homepage: http://humangenome.duke.edu
1000 Genomes: http://www.1000genomes.org
dbSNP: http://www.ncbi.nlm.nih.gov/projects/SNP
GeneTests: http://genetests.org
NHGRI Catalog of Published Genome-Wide Association
Studies: http://www.genome.gov/26525384
SequenceVariantAnalyzer: http://www.svaproject.org
ssahaSNP: http://www.sanger.ac.uk/resources/software/
ssahasnp
US National Institutes of Health Office of Rare Diseases
Research: http://rarediseases.info.nih.gov/AboutUs.
aspx#GARD
All links ARe ActiVe in the Online Pdf
REVIEWS
NATuRe ReVIeWS | Genetics
VOluMe 11 | juNe 2010 | 425
© 20 Macmillan Publishers Limited. All rights reserved10