Admixture mapping identifies 8q24 as a prostate
cancer risk locus in African-American men
Matthew L. Freedmana,b,c, Christopher A. Haimanc,d, Nick Pattersonb,c, Gavin J. McDonaldb,e, Arti Tandonb,e,
Alicja Waliszewskab,e,f, Kathryn Penneyb, Robert G. Steene,g, Kristin Ardlieb,h, Esther M. Johni,j,
Ingrid Oakley-Girvani,j, Alice S. Whittemorej, Kathleen A. Cooneyk,l, Sue A. Inglesd, David Altshulerb,e,m,n,
Brian E. Hendersond, and David Reichb,e,o
aDepartment of Medical Oncology, Dana–Farber Cancer Institute, Boston, MA 02115;bProgram in Medical and Population Genetics, Broad Institute of
Harvard and Massachusetts Institute of Technology, Cambridge, MA 02142;dDepartment of Preventive Medicine, Keck School of Medicine, University of
Southern California, Los Angeles, CA 90089; Departments ofeGenetics andmMedicine andgBiopolymers Facility, Harvard Medical School, Boston, MA 02115;
fLaboratory of Molecular Immunology, Center for Neurologic Disease, Brigham and Women’s Hospital, Boston, MA 02115;hGenomics Collaborative, Division
of SeraCare Life Sciences, Inc., Cambridge, MA 02139;iNorthern California Cancer Center, Fremont, CA 94538;jDepartment of Health Research and Policy,
Stanford University School of Medicine, Stanford, CA 94305;kDepartments of Medicine and Urology andlComprehensive Cancer Center, University of
Michigan, Ann Arbor, MI 48109; andnCenter for Human Genetic Research and Department of Molecular Biology, Massachusetts General Hospital,
Boston, MA 02114
Communicated by Eric S. Lander, Broad Institute, Cambridge, MA, July 12, 2006 (received for review May 24, 2006)
A whole-genome admixture scan in 1,597 African Americans identi-
The increased risk because of inheriting African ancestry is greater in
men diagnosed before 72 years of age (P < 0.00032) and may
contribute to the epidemiological observation that the higher risk for
prostate cancer in African Americans is greatest in younger men (and
attenuates with older age). The same region was recently identified
through linkage analysis of prostate cancer, followed by fine-map-
ping. We strongly replicated this association (P < 4.2 ? 10?9) but find
that the previously described alleles do not explain more than a
fraction of the admixture signal. Thus, admixture mapping indicates
a major, still-unidentified risk gene for prostate cancer at 8q24,
motivating intense work to find it.
association ? human genetics
and 27,350 deaths in 2006 (1). African Americans have the highest
incidence of prostate cancer in the United States, ?1.6-fold
higher than European Americans (http:??jncicancerspectrum.
oxfordjournals.org?cgi?statContent?cspectfstat;18). The higher
risk (2–4) prompted the hypothesis that genetic factors in part
account for this difference. If there are genetic risk variants that
differ substantially in frequency across populations, admixture
mapping should have power to detect them.
The idea of admixture mapping is to screen through the
genome of populations of mixed ancestry such as African
Americans (5), searching for regions where the proportion of
DNA inherited from either the ancestral European or African
population is unusual compared with the genome-wide average.
Admixture mapping requires a relatively small number of mark-
ers for a whole-genome scan: a couple of thousand, rather than
the hundreds of thousands estimated to be necessary in nonad-
mixed populations (5, 6). Because the mixture between Euro-
pean and West-African populations occurred within the past 15
generations (5), stretches of DNA with contiguous European
and African ancestry have not had much time to break up
because of recombination and typically extend millions of base
every few million base pairs (Mb), rather than every few
thousand as with linkage disequilibrium mapping.
Although admixture mapping was first proposed ?50 years
ago (7) and has good power to detect risk variants that are
strikingly different in frequency across populations (6, 8), it has
not been practical until recently. Appropriate panels of markers
(5), combined with analytical methods (8–10), made possible the
rostate cancer is the most common noncutaneous malig-
nancy among U.S. men, with an estimated 234,460 new cases
first admixture scans (11, 12) in 2005. Here, we describe a
whole-genome admixture scan focusing on prostate cancer, a
disease that has long been considered a test case for admixture
mapping because of its marked difference in incidence rates
across populations. We identify a highly significant association at
8q24. The same broad region has recently been implicated in
prostate cancer by Amundadottir et al. (13). In addition to
providing independent evidence of a locus at 8q24, the present
study provides two pieces of information. First, we show an
association with earlier age of diagnosis. Second, we show that
the alleles identified in the previous study are insufficient to
explain more than a small fraction of the admixture signal. Thus,
the causative alleles remain to be identified.
We studied 1,597 prostate cancer cases and 873 controls, the
majority of which were participants in the Multiethnic Cohort
study (14) (810 cases and 730 controls) (Table 5, which is
published as supporting information on the PNAS web site). The
other samples came from six studies, including studies that
specifically ascertained cases with high-grade tumors, advanced-
stage disease, diagnosis at a young age, or occurrence in a family
with multiple affected individuals (15–17) (Table 1). The present
study was designed to include more cases than controls, because
admixture mapping works by comparing the proportion of
ancestry in cases to the rest of their own genomes. In principle,
controls are not needed (6, 8); however, we included controls
because they are useful for follow-up analyses (8).
All 2,470 samples (1,597 cases and 873 controls) were geno-
typed by using one of two panels of markers chosen to be highly
different in frequency between West Africans and European
Americans (5). A total of 1,792 samples were genotyped in the
‘‘phase 1’’ panel [previously used in a scan for multiple sclerosis
genes (12)] and 1,266 SNPs passed quality filters and were used
on the PNAS web site). The remaining 678 samples were typed
in a second-generation ‘‘phase 2’’ panel that extracts more
information per SNP; 1,365 SNPs passed quality filters and were
used in analysis. The analysis combines information from both
panels into a single logarithm of odds (LOD) score statistic at
Conflict of interest statement: No conflicts declared.
Abbreviations: LOD, logarithm of odds; OR, odds ratio.
cM.L.F., C.A.H., and N.P. contributed equally to this work.
oTo whom correspondence should be addressed at: Department of Genetics, Harvard
Medical School, New Research Building, 77 Avenue Louis Pasteur, Boston, MA 02115.
© 2006 by The National Academy of Sciences of the USA
September 19, 2006 ?
vol. 103 ?
no. 38 www.pnas.org?cgi?doi?10.1073?pnas.0605832103
each locus; observations ?5 are considered strongly indicative of
a disease locus (8). Formal significance is assessed by Bayesian
methods. We take 10 to the power of the LOD score and average
genome average is ?100, then the Bayesian odds in favor of a
disease locus is 100:1, and we interpret the data as showing
significant evidence of a disease gene (8).
An initial admixture scan of 1,303 African-American prostate
cancer cases produced a peak LOD score of 2.2 at 8q24. The
signal was higher in a secondary analysis of individuals with a
younger age at diagnosis, with the peak LOD score rising to 3.8
in the individuals who were ?68 years of age (the threshold
giving the strongest evidence of association). After genotyping
294 additional cases and 15 additional SNPs at 8q24 to obtain
better local information about ancestry (see Materials and
Methods), the peak LOD score increased to 4.1 in all cases and
as high as 8.4 in the 1,176 who were diagnosed at ?72 years of
age. To correct for inflation of the score because of choosing the
age threshold that gave the strongest significance, we integrated
the evidence for association over an evenly spaced range of
cutoffs (see Materials and Methods and Table 7, which is pub-
lished as supporting information on the PNAS web site). This
analysis yielded a peak LOD score of 7.1 (Fig. 1). Averaging 10
to the power of the LOD scores at equally spaced points
genome-wide, we obtained a genome-wide average score of
?19,000, exceeding the threshold of 100 for significance (8).
After correcting for multiple hypothesis testing [by dividing by 4,
because we tested four phenotypes (age, grade, stage, and
familial disease) and focused on the one giving the strongest
evidence], the odds in favor of a disease locus still greatly
exceeded the threshold of 100 for significance.
The analysis in the previous paragraph used age of diagnosis as
a covariate but did not directly test whether men with younger age
of diagnosis have higher risk at 8q24 than older men. To formally
test this hypothesis, we exploited the fact that ANCESTRYMAP
assigns scores for association to each individual separately (e.g.,
individual factors such as ?0.02, 0.12) and then sums over all
individuals to produce the total LOD score (Table 2). We rank-
ordered the 1,588 cases in the scan for whom we had age informa-
tion from youngest to oldest (Fig. 2). If the locus is not associated
with age of diagnosis, the cumulative LOD should increase steadily
to reach the total as additional samples are added. In fact, it rises
to 5.4 LOD points above expectation at 71 years of age. To test
whether this rise is significant, we permuted the data, reassigning
ages of onset to different individuals (so that, in the randomized
variation). In 1,000,000 permutations, only 318 showed a change in
LOD score compared with the expectation exceeding the observed
5.4 (P ? 0.00032). Repeating the analysis with a subset of samples
obtained from a single prospective cohort [804 cases of African
Americans with prostate cancer from the Multiethnic Cohort
(MEC) Study], the association to age was also significant (P ?
0.0011). These results indicate that there is a formally significant
any associations when a similar analysis was applied to other
subphenotypes: stage, grade, or family history (Supplemental Note
Table 2. Admixture scan summaries
All prostate cancer cases
High grade (Gleason score ?7)
Advanced stage (regional or metastatic cancer)
Family history (prostate cancer in a first-degree relative)
Age of diagnosis of ?72 years and high-density genotyping at 8q24
Drop out every even marker from run no. 5 to demonstrate independence of markers used
Drop out every odd marker from run no. 5 to demonstrate independence of markers used
Diagnosis at ?72 years of age, high density and best model of 1.54-increased risk because of
Integrating over age-of-diagnosis cutoffs as a formal test for statistical significance
*Indicates a scan that meets formal criteria for genome-wide statistical significance.
Table 1. Characteristics of cases and controls from seven sources
Source LocationCases Controls
? 1 SE
? 1 SE
in a first-
peak LOD if
L.A. County Men’s
Study Early Onset
Flint Men’s Health
Bay Area Men’s
CA & HI
23.57 ? 0.50
22.34 ? 0.83
25.42 ? 0.57
26.37 ? 2.13
CA104—20.89 ? 1.37—60 (45–65)3149141.01
19.50 ? 1.01
18.05 ? 1.21
CA823619.06 ? 1.5220.13 ? 2.1564 (44–78)2594281.16
All U.S.47—16.16 ? 1.51—62 (39–81)14 3828 0.57
1,597873 22.11 ? 0.3625.32 ? 0.55 65 (39–88)212918 7.14
PCGP, Prostate Cancer Genetics Project; Euro., European.
*To assess how much each of the seven cohorts contributes to the signal of association, we removed each from the main admixture scan (run no. 9 in Table 2)
and assessed how the peak LOD score at 8q24 changes. All seven cohorts contribute positively.
Freedman et al. PNAS ?
September 19, 2006 ?
vol. 103 ?
no. 38 ?
1 in Supporting Text, which is published as supporting information
on the PNAS web site).
To explore how much of the increased incidence of prostate
cancer in African-American men might be explained by African
(as compared with European) ancestry at 8q24, we evaluated the
risk for individuals carrying zero, one, and two chromosomes
with African ancestry at the locus. Each African-derived chro-
mosome is associated with ?1.54-fold increased risk in younger
individuals (90% credible interval 1.38–1.74) (Supplemental
Note 2). We also estimated the proportion of control samples
with zero, one, and two African-derived chromosomes, respec-
tively (6.4%, 37.8%, and 55.8%, respectively). Extrapolating to
the broader African-American population, the prostate cancer
incidence in all African Americans IALL is higher than the
chromosomes at the locus by a factor of [(0.064) (1) ?
(0.378)(1.54) ? 0.558(1.542)] ? 1.969. Thus, the fraction of all
prostate cancer incidence for African Americans ?72 years of
age that could be explained by ancestry at this locus is (IALL?
IEE)?IALL? 1 ? (1?1.969) ? 49% (with a 90% credible interval
of 39–59%). Thus, if it were possible to develop a treatment that
reduced prostate cancer risk in the African-American popula-
tion to the level that is seen in men who carry two copies of 8q24
inherited from recent European ancestors, the rate of prostate
cancer would decrease by ?49%. The total risk for prostate
cancer that can be attributed to 8q24 in African-American men
of the peak. It aligns with the microsatellite and SNP recently associated with prostate cancer by Amundadottir et al. (13) (dashed line). (c) The 95% credible
taken from http:??genome.ucsc.edu) (data from the May 2004 genome assembly).
Summary of results for the whole-genome admixture scan and characteristics of the 8q24 peak of association. (a) We present the LOD score at equally
bution to the chromosome 8 locus, we rank-ordered the individuals by age of
onset and then calculated a score for increasing age cutoffs. The score rises to
5.40 above the expectation for 1,176 individuals diagnosed at ?72 years of
age. To evaluate whether this rise is unexpected, we permuted the data
1,000,000 times, randomizing scores with respect to individuals’ ages of onset
(guaranteeing that there is no relationship between age of diagnosis and
contribution to the evidence of association). In only 318 of 1,000,000 permu-
tations did we see a rise as high as in our data (P ? 0.00032).
To formally test for a relationship between age of onset and contri-
www.pnas.org?cgi?doi?10.1073?pnas.0605832103Freedman et al.
?72 years of age is still greater, because alleles at 8q24 increase
prostate cancer risk even in chromosomes of entirely European
origin (13). Thus, 8q24 has a major effect on population risk of
prostate cancer, especially in younger African Americans.
Using the LOD scores at 8q24, we also calculated a posterior
probability distribution to estimate the position of the disease-
causing variants (see Fig. 1b and Materials and Methods). The
95% credible interval spans 3.80 Mb, from 125.68–129.48 Mb in
build 35 of the human genome reference sequence (13.9 cM) and
contains nine known genes (Fig. 1c). However, the admixture
scan does not provide information about which gene or alleles
within the locus confer risk.
Independently of this study, Amundadottir et al. (13) reported
an SNP allele [A at rs1447295; odds ratio (OR) ? 1.51; P ? 1.0 ?
10?11] and a microsatellite allele (?8 at DG8S737; OR ? 1.62;
P ? 2.7 ? 10?11) that map to the same region as the admixture
peak (at 128.546 Mb and 128.554 Mb, respectively) and are
was observed in European and African Americans, whereas the
A allele effect was detected only in European-derived popula-
tions. The authors did not show, however, that either allele was
causally involved in disease but instead suggested that they were
both in linkage disequilibrium with an as-yet-unidentified causal
variant. They also did not identify which gene in the region might
be responsible for prostate cancer risk.
To directly compare the results of the two studies, we tested
the previously associated alleles in the African-American cases
and controls [excluding samples from Michigan, because they
overlap those studied by Amundadottir et al. (13); see Materials
and Methods]. The goal was to test whether the ?8 allele at
DG8S737 contributes to disease risk in African Americans
beyond the risk that can be accounted for by the admixture
signal. (Supplemental Note 3). We were concerned that the
previously detected association in African Americans by
Amundadottir et al. (13) (P ? 0.0022, estimated OR of 1.60)
might simply reflect an admixture signal across a large region
(because of systematic differences in ancestry between cases and
controls across several million base pairs of 8q24) and thus might
not provide fine-mapping information in African Americans.
Although Amundadottir et al. (13) tested for mismatching of
cases and controls in overall proportion of ancestry, they did not
control for a local rise in African ancestry throughout 8q24 in
cases but not controls. Such a rise would be expected to cause
thousands of alleles in the region that just happen to be more
frequent in African Americans (including the microsatellite ?8
allele) to show association with prostate cancer. When we
correct for this effect in the African-American samples from the
present study (Supplemental Note 3), we find that the contribu-
tion of the ?8 allele to risk is nonsignificant (P ? 0.22) (Table
3). The OR of 0.93–1.17 (95% credible interval) also rules out
the OR ? 1.60 reported in African-Americans (13).
We next expanded the replication analysis to the four eth-
nicities in the MEC other than African Americans, by genotyp-
ing rs1447295 in 1,614 prostate cancer cases and 1,547 controls
from these populations. The evidence for association is signifi-
cant overall (P ? 4.2 ? 10?9), as well as separately in each group:
Japanese Americans (P ? 0.00034), Native Hawaiians (P ?
0.00015), Latino Americans (P ? 0.0014), and European Amer-
icans (P ? 0.022) (Table 4). This analysis replicates the associ-
ation identified by Amundadottir et al. (13), although we did not
test for the possible confounding factor of population stratifi-
cation. Interestingly, we do not replicate the association to tumor
grade (Gleason ?8 vs. Gleason ?8; P ? 0.47) reported by
Amundadottir et al. (13).
These results confirm the finding of Amundadottir et al. (13)
that the 8q24 locus is important in prostate cancer. However, the
alleles they reported do not explain the admixture signal (Sup-
plemental Note 4 and Table 3). The specific variants causing
increased risk for prostate cancer in African American because
of 8q24 thus remain to be identified.
We have used admixture mapping to identify a locus at 8q24 that
substantially affects risk for prostate cancer. We highlight four
First, this study shows that admixture mapping can be a
powerful and practical way to map genetic variants for complex
disease (5, 18). The results motivate the application of admixture
mapping to other disorders, especially those like prostate cancer
in which incidence varies across populations. These results also
highlight the scientific value of studies to find disease genes in
specific ethnic groups, such as African Americans.
Second, we show that the 8q24 locus contributes to a major
increased risk for prostate cancer in African Americans with
African ancestry at 8q24. The difference between these individ-
uals and African Americans with European ancestry at 8q24
Americans. If one could intervene medically to reduce the risk
for prostate cancer in African Americans ?72 years of age to
what would be expected if all African Americans had European
ancestry at the locus, the incidence in men ?72 years of age
would decrease by approximately 49%. We also show that the
admixture signal at 8q24 cannot be explained by the alleles
identified by Amundadottir et al. (13); instead, there must be
major, unmapped risk alleles at the locus.
Third, we detect a highly significant association of 8q24 with
age. This finding is intriguing because it is known epidemiolog-
ically that the differential incidence of prostate cancer in African
versus European Americans is greater at younger ages and is
attenuated with older age (ref. 19; http:??jncicancerspectrum.
lance, Epidemiology, and End Results (SEER) Program registry
data indicate that, for men diagnosed at ?55 years of age,
African Americans have a 2.27-fold higher rate than European
Americans, but the ratio decreases to 1.48-fold for men diag-
nosed at ?75 years of age (19). Genetic variation at 8q24 may be
responsible for part of this effect.
Fourth, we identify a 3.8-Mb interval containing nine known
genes that is likely to harbor variant(s) explaining the admix-
ture peak. This is a tractable region for follow-up analysis.
Somatic genetic data independently highlight the 8q24 region
as one of the most frequently amplified regions in prostate
cancer tumors (20, 21). The c-MYC oncogene, a key regulator
in cellular proliferation, lies within the peak. Overexpression
of c-MYC has been shown to induce tumors in mice and to
create a cancer phenotype in benign prostatic epithelium (22,
23). It is possible that c-MYC could be the gene responsible for
the prostate cancer risk, but no structural or regulatory variant
has yet been identified.
Follow-up work will be necessary to identify the as-yet-
undiscovered causal risk variant(s) at 8q24. Ultimately, discov-
ering the causal gene(s) at 8q24 may translate into better
Table 3. Allelic association tests in African Americans adjusting
for local rise in African ancestry
P valueOR (95% CI)
A allele at rs1447295
?8 allele at DG8S737
Haplotype of A and ?8*
Cases diagnosed at ?72 years of age and all controls. P values are one-
tailed, testing for the previously associated allele (14) being more common
(Supplemental Note 3). CI, confidence interval.
*For the haplotype test, we phased the cases and controls together, before
carrying out the association analysis.
Freedman et al. PNAS ?
September 19, 2006 ?
vol. 103 ?
no. 38 ?
understanding of prostate cancer and may play a role in strat-
egies for screening of the population and identifying new targets
for treatment and prevention.
Materials and Methods
largest number came from the Multiethnic Cohort (MEC), a
prospective cohort that began in 1993 and is still ongoing, which
ascertains prostate cancer cases and controls by linking to
databases from the California Cancer Registry, the Los Angeles
County Cancer Surveillance Program, and the Hawaii Cancer
Registry (14). The samples used in the admixture scan were all
African-American cases and controls; however, for the valida-
tion genotyping of the rs1447295 SNP, we also genotyped
prostate cancer cases and controls from four other ethnicities in
the MEC: European Americans, Latino Americans, Japanese
Americans, and Native Hawaiians. The second largest number of
samples came from the Los Angeles County Men’s Health Study
(1999–2002), which was enriched for individuals with advanced-
stage or high-grade prostate cancer, as identified through hos-
pitals and private histopathology laboratories in Los Angeles
County. The Bay Area Men’s Health Study (15) (1997–2000) was
enriched for individuals with regional- or distant-stage disease.
The Study of Early Onset Prostate Cancer (1993–1995) was
based in the San Francisco–Oakland Bay Area and included only
individuals with histologically confirmed prostate cancer who
were ?66 years of age at diagnosis. The Genomics Collaborative,
Ltd. samples were obtained from consenting individuals under-
going surgery for prostate cancer throughout the U.S. and were
provided to this study at no cost by means of an academic
collaboration. The Flint Men’s Health Study samples (1996–
2002) were obtained through a case-control study of prostate
cancer in Genesee County, Michigan. The University of Mich-
igan Prostate Cancer Genetics Project (PCGP) samples were
obtained from an ongoing family-based study of prostate cancer
susceptibility. PCGP cases have a family history of prostate
cancer or early age at diagnosis defined as ?55 years of age (we
analyzed data only from the man with the youngest age of
diagnosis in each family). We note that both the Flint Men’s
Health Study and PCGP samples (16, 17) overlap with those
studied by Amundadottir et al. (13). The samples were provided
by K.A.C. for replication purposes blinded to the locus under
study. The results reported here, which also use a different type
of information to localize disease genes (admixture linkage
disequilibrium), are thus fully independent.
Genotyping. The phase 1 and phase 2 panels of SNPs were both
genotyped by using the Illumina BeadLab genotyping platform
(24) [supplemented for phase 1 by Sequenom MassARRAY
genotyping (25)]. At the 8q24 peak, we genotyped an additional
15 SNPs using Sequenom technology to extract maximal infor-
mation about ancestry [these SNPs were chosen to have high
frequency differentiation between the European and West-
African populations (5) based on data from the Human Hap-
lotype Map (26)]. We used previously described protocols to
remove SNPs that did not perform well in genotyping, that were
in linkage disequilibrium with each other in the ancestral
European and West-African populations, or that did not seem to
have appropriate intermediate frequencies in the African Amer-
icans compared with the ancestral populations (12). The
rs1447295 genotyping was carried out by using the Applied
Biosystems Inc. (ABI, Foster City, CA) Assay-on-Demand tech-
nology following the manufacturer’s recommended protocol,
and all of the African Americans were also genotyped at
rs1447295 by using Sequenom technology. The DG8S737 geno-
typing was carried out by using ABI True Allele PCR Premix,
with 5-pmol forward (5?-6FAM-TGATGCACCACAGAAAC-
CTG-3?) and 5-pmol reverse (5?-GTTTCAAGGATGCAGCT-
CACAACA-3?) primers, and 60 ng of DNA per reaction.
Reactions were analyzed on an ABI3730xl DNA Analyzer.
Samples were scored by the ABI GeneMapper V3.7 software,
with all genotypes confirmed by an experienced technician. To
check the microsatellite genotyping results, we compared 168
samples that overlapped between this study and that of Amunda-
Admixture Analysis. We used the ANCESTRYMAP software (8)
to carry out the screens for association with prostate cancer.
ANCESTRYMAP calculates a statistic for association at every
position in the genome, under a prespecified family of risk
models, calculating the likelihood of the data at the locus under
an average of disease models versus the likelihood of the data if
the locus has nothing to do with disease (the log base 10 of this
is the LOD score). For most runs, we assume equally likely
models of 0.3-, 0.4-, 0.6-, 0.7-, 0.8-, 1.2-, 1.5-, and 2-fold increased
risk because of each copy of a European allele. This family of
models reflects the hypothesis that African-derived alleles are
more likely to confer risk but also tests for the alternative
possibility. To obtain an overall assessment of the evidence for
a disease locus anywhere in the genome, we average the factors
for association at each point separately, providing a genome-
wide assessment of whether there is a locus in the genome
Admixture Scan Accounting for Age of Diagnosis. We carried out an
admixture scan taking into account the possibility that individuals
with a younger age of diagnosis contribute a more powerful
admixture signal, while not inappropriately inflating the signal of
association by picking the cutoff giving the strongest signal. We ran
22 independent scans for all individuals in the data set with
diagnosis at ?50, ?53, ?56, ?57, ?59, ?60, ?61, ?62, ?63, ?64,
?65, ?66, ?67, ?69, ?70, ?71, ?73, ?74, ?75, ?76, and ?78
years of age, as well as all cases (Table 7). Approximately 73 new
samples were added in for each consecutive run. We then averaged
Table 4. rs1447295 association in the Multiethnic Cohort
Group within the
No. of samples
A allele, %
(95% confidence interval)Cases ControlsCases Controls
All samples together
4.2 ? 10?9
*OR estimated by using logistic regression adjusted for age.
†OR estimated by using logistic regression adjusted for age as well as ethnicity.
www.pnas.org?cgi?doi?10.1073?pnas.0605832103 Freedman et al.
the genome scores for association, which gives a statistically ap-
propriate assessment of the evidence for association.
Permutation Analysis to Test Whether Some Phenotypes Contribute
Unduly to the Signal of Association at 8q24. To test whether the
correlation of the 8q24 admixture association with a phenotype
is significant, we carried out permutation analyses, considering
separately the effect of stage of disease, grade of tumor, family
history, and age of diagnosis (Fig. 2 and Supplemental Note 1).
For each phenotype, we rank-ordered individuals by their values
of the phenotype. We then calculated a cumulative LOD score
each cutoff. We recorded the greatest excess or shortfall of the
cumulative LOD score compared with the expectation if it
increased linearly. We then wrote a PERL script to randomly
permute the values of the phenotype over the samples, elimi-
nating any relationship between the phenotype and score. A P
value was calculated as the fraction of 1,000,000 permutations
that produced a score for association as extreme as the data.
Inferring the Position of the Disease Locus. To infer the position of
the disease locus, we note that the LOD scores at each point of
the genome can be taken to the power of 10 to give the relative
probability of that locus containing the disease allele. After
normalization, this calculation provides a probability distribu-
tion for the position of the locus. A 95% credible interval is
obtained from the central area under the peak (Fig. 1c).
We thank the men with and without prostate cancer who participated in
this study, Eric Lander and two reviewers for comments and criticism,
Loreall Pooler and David Wong from the University of Southern
California Genomics Laboratory for help with sample handling and
genotyping, Courtney Montague at Harvard Medical School for assis-
tance with genotyping, and the National Center for Research Resources
Center for Genotyping and Analysis at the Broad Institute, without
which this work would not have been possible. The genotyping for this
work was supported by National Institutes of Health (NIH) Grant
CA63464 (to B.E.H., C.A.H., D.A., and D.R.). M.L.F. was supported by
a Department of Defense Health Disparity Training-Prostate Scholar
Award (DAMD 17-02-1-0246), by a Howard Hughes Medical Institute
physician postdoctoral fellowship, and by Dana–Farber?Harvard Part-
ners Cancer Care Prostate Specialized Programs of Research Excellence
(SPORE). N.P. was supported by NIH Career Transition Award
HG02758. E.M.J. and S.A.I. were supported by California Cancer
Research Program Grants 99-00527V-10182 and 99-00524V-10258, re-
of Michigan SPORE in Prostate Cancer (CA69568), the University of
Michigan Department of Urology, and the University of Michigan
Comprehensive Cancer Center. K.A.C. was supported by NIH Awards
CA69568 and CA79596, and I.O.-G. and A.S.W. were supported by NIH
Award CA67044. D.A. is a Charles E. Culpeper Scholar of the Rock-
efeller Brothers Fund and a Burroughs Wellcome Fund Clinical Scholar
in Translational Research. D.R. is the recipient of a Burroughs Well-
come Career Development Award in the Biomedical Sciences.
1. Jemal A, Siegel R, Ward E, Murray T, Xu J, Smigal C, Thun MJ (2006) CA
Cancer J Clin 56:106–130.
2. Bunker CH, Patrick AL, Konety BR, Dhir R, Brufsky AM, Vivas CA, Becich MJ,
Trump DL, Kuller LH (2002) Cancer Epidemiol Biomarkers Prev 11:726–729.
3. Freedland SJ, Isaacs WB (2005) Prostate 62:243–252.
4. Simard J, Dumont M, Labuda D, Sinnett D, Meloche C, El-Alfy M, Berger L,
Lees E, Labrie F, Tavtigian SV (2003) Endocr Relat Cancer 10:225–259.
5. Smith MW, Patterson N, Lautenberger JA, Truelove AL, McDonald GJ,
Waliszewska A, Kessing BD, Malasky MJ, Scafe C, De Jager PL, et al. (2004)
Am J Hum Genet 74:1001–1013.
6. McKeigue PM (1997) Am J Hum Genet 60:188–196.
7. Rife DC (1954) Am J Hum Genet 6:26–33.
8. Patterson N, Hattangadi N, Lane B, Lohmueller KE, Hafler DA, Oksenberg
JR, Hauser SL, Smith MW, O’Brien SJ, Altshuler D, et al. (2004) Am J Hum
9. Hoggart CJ, Shriver MD, Kittles RA, Clayton DG, McKeigue PM (2004) Am
J Hum Genet 74:965–978.
10. Montana G, Pritchard JK (2004) Am J Hum Genet 75:771–789.
11. Zhu X, Luke A, Cooper RS, Quertermous T, Hanis C, Mosley T, Gu CC, Tang
H, Rao DC, Risch N, et al. (2005) Nat Genet 37:177–181.
12. Reich D, Patterson N, De Jager PL, McDonald GJ, Waliszewska A, Tandon
A, Lincoln RR, DeLoa C, Fruhan SA, Cabre P, et al. (2005) Nat Genet
13. Amundadottir LT, Sulem P, Gudmundsson J, Helgason A, Baker A, Agnarsson
BA, Sigurdsson A, Benediktsdottir KR, Cazier JB, Sainz J, et al. (2006) Nat
DO, Monroe KR, Earle ME, Nagamine FS (2000) Am J Epidemiol 151:346–357.
15. John EM, Schwartz GG, Koo J, Van Den Berg D, Ingles S (2005) Cancer Res
16. Cooney KA, Strawderman MS, Wojno KJ, Doerr KM, Taylor A, Alcser KH,
Heeringa SG, Taylor JM, Wei JT, Montie JE, et al. (2001) Urology 57:
17. Cooney KA, McCarthy JD, Lange E, Huang L, Miesfeldt S, Montie JE,
Oesterling JE, Sandler HM, Lange K (1997) J Natl Cancer Inst 89:955–959.
18. Kittles RA, Chen W, Panguluri RK, Ahaghotu C, Jackson A, Adebamowo CA,
Griffin R, Williams T, Ukoli F, Adams-Campbell L, et al. (2002) Hum Genet
19. Whittemore AS, Keller JB, Betensky R (1991) J Natl Cancer Inst 83:1231–
20. van Duin M, van Marion R, Vissers K, Watson JE, van Weerden WM, Schroder
FH, Hop WC, van der Kwast TH, Collins C, van Dekken H (2005) Genes
Chromosomes Cancer 44:438–449.
21. Visakorpi T, Kallioniemi AH, Syvanen AC, Hyytinen ER, Karhu R, Tammela
T, Isola JJ, Kallioniemi OP (1995) Cancer Res 55:342–347.
22. Ellwood-Yen K, Graeber TG, Wongvipat J, Iruela-Arispe ML, Zhang J,
Matusik R, Thomas GV, Sawyers CL (2003) Cancer Cell 4:223–238.
23. Williams K, Fernandez S, Stien X, Ishii K, Love HD, Lau YF, Roberts RL,
Hayward SW (2005) Prostate 63:369–384.
24. Fan JB, Oliphant A, Shen R, Kermani BG, Garcia F, Gunderson KL, Hansen
M, Steemers F, Butler SL, Deloukas P, et al. (2003) Cold Spring Harbor Symp
Quant Biol 68:69–78.
25. Tang K, Fu DJ, Julien D, Braun A, Cantor CR, Koster H (1999) Proc Natl Acad
Sci USA 96:10016–10020.
26. The International HapMap Consortium (2005) Nature 437:1299–1320.
Freedman et al. PNAS ?
September 19, 2006 ?
vol. 103 ?
no. 38 ?