EDITORIALS
Genomic medicine: SNPs
on chips?
Vicki Whitehall and Barbara Leggett
Royal Brisbane and Women’s Hospital Research Foundation, Clinical
Research Centre, Brisbane, Queensland and Queensland Institute of
Medical Research, Conjoint Gastroenterology Laboratory, Brisbane,
Queensland, Australia
See article in J. Gastroenterol. Hepatol. 2008; 23: 948–953.
Understanding the molecular events underlying the initiation and
progression of colorectal cancer (CRC) is essential for implement-
ing appropriate screening strategies for early detection and treat-
ment regimes for improved outcome. Highly penetrant, inherited
gene mutations account for only 3–5% of CRC cases, whilst a
family history is reported by approximately one-third of all
patients. The environment has been the presumed culprit in the
remaining ‘sporadic’ cases and indeed associations have been
observed between multiple environmental risk factors and
increased disease incidence.
1
However, despite intense efforts to
prove causality, many associations with environmental factors are
modest and remain controversial. The reason for this may in part
be due to the genetic heterogeneity of sample populations,
whereby subtle molecular differences may alter the magnitude of
risk conferred by an environmental insult.
In recent years, attention has turned to low-penetrance,
common genetic variants (single nucleotide polymorphisms,
SNPs) as modifiers of cancer risk and disease course. It is con-
ceivable that low-penetrance variants will influence the disease
onset and phenotype of a significant proportion of ‘sporadic’
CRCs. Furthermore, the impact of such variants could be sub-
stantial since multiple alleles are likely to act in an additive or
multiplicative fashion.
2,3
This effect is unlikely to be linear due to
interaction with other SNPs and the environment, both of which
may increase risk or be protective. Variant alleles may be caus-
ally associated with disease risk or phenotype, or may be in
linkage disequilibrium with the causative allele variant. Identify-
ing SNPs associated with disease is currently the subject of
intense investigation. For this field to transition from scientific
interest to clinical utility, we will need to determine which SNPs
contribute to CRC risk and phenotype, better understand gene–
gene and gene–environment interactions and collaborate with
bioinformaticians to devise the complex algorithm necessary to
translate this genetic information into clinical recommendations
for screening and disease management.
Two major approaches for identifying and validating clinically
relevant SNPs are genome-wide scans and hypothesis-driven
candidate SNP studies, the latter based on physical location near a
gene or region of potential importance in carcinogenesis. In recent
months, a number of major discoveries have highlighted the
advantages of large scale, genome-wide approaches for identify-
ing low-penetrance alleles. A SNP on chromosome 8q24.21
(rs6983267) was identified in two independent studies
4,5
as con-
ferring between 1.17- and 1.27-fold increased risk of CRC. These
odds ratios were highly statistically significant and reproducible in
multiple validation cohorts consisting of many thousands of
patients and control subjects, including a third independent study
of the SNP.
6
A variant lying within the third intron of the SMAD7
gene (rs4939827) was also identified as a CRC susceptibility locus
using a microarray SNP chip containing over 550 000 loci, and
was estimated to contribute to approximately 15% of CRC cases.
7
An alternative to genome-wide association is the candidate gene
approach. For example, based on initial linkage data to families
with hereditary mixed polyposis syndrome, the CRAC1 locus on
chromosome 15q13.3 was investigated as a susceptibility locus for
sporadic CRC.
8
This revealed statistically significant risk estimates
of 1.23 (heterozygotes) and 1.70 (homozygotes) for the rs4779584
SNP, and is estimated to account for approximately 15% of CRC
cases. It may also be necessary to examine associations in de-
fined CRC subgroups. For example, an MLH1 promoter SNP
(rs1800734) was not associated with increased risk of CRC in an
unselected cohort. However, when case patients were stratified for
level of microsatellite instability (MSI), there was a highly signifi-
cant association observed between the minor allele and a high level
of MSI.
9
Large-scale candidate gene approaches have also been
successful, and may involve the construction of a ‘hypothesis-
driven SNP chip’. A recent study using this approach resulted in
the identification of 44 SNPs associated with CRC.
10
An important consideration when designing association studies
is adequate powering based on allele frequencies. It is estimated
that for low-penetrance variants, where an odds ratio of 1.1–1.5
would be expected, a cohort of over 1000 subjects would be
necessary to achieve statistical significance.
11
When assessing
multiple SNPs, stringent statistical tests must be applied to correct
for multiple comparisons, which further necessitates large
numbers of subjects. Cohort ethnicity is also an important factor as
SNP frequencies may vary widely in different populations. Meta-
analyses have been performed in an effort to achieve sufficient
cohort numbers; however, ethnic heterogeneity must be avoided
unless allele frequencies are known to be stable. Control popula-
tions must also be matched to case samples for age, gender, and
ethnicity. Validation studies in secondary cohorts are also essential
to confirm initial findings.
In addition to estimation of disease risk, SNP profiling may also
provide insights into the expected clinical course of a tumor. Ulti-
mately this may be important for determining treatment modalities
and predicting outcome. In the present issue of the Journal,
Yoshiya and colleagues have adopted a candidate gene approach
to investigate whether defined clinicopathologic features
co-segregate with any of five common SNPs (CCND1, p21
cip1
,
DCC, MTHFR, and Exo1).
12
These were chosen based on their
genomic location either within or near the coding regions of genes
with putative roles in carcinogenesis. The study cohort consisted
of 114 patients with colorectal cancer. This is a relatively small
cohort for an association study, and the authors emphasize that the
data is preliminary and requires further validation. Associations
were observed between DCC and depth of invasion (P = 0.04), and
MTHFR and tumor size (P = 0.02). After correction for multiple
Correspondence
Professor Barbara Leggett, Department of Gastroenterology and
Hepatology, Royal Brisbane and Women’s Hospital, Level 9, Ned
Hanlon Building, Herston, QLD 4029, Australia. Email:
Barbara_Leggett@health.qld.gov.au
Author’s contributions: VW and BL contributed equally to the drafting
of this editorial.
doi:10.1111/j.1440-1746.2008.05422.x
823 Journal of Gastroenterology and Hepatology 23 (2008) 823–832 © 2008 The Authors
Journal compilation © 2008 Journal of Gastroenterology and Hepatology Foundation and Blackwell Publishing Asia Pty Ltd