Available online at www.sciencedirect.com

12 (2009) 210–214www.elsevier.com/locate/ygyno

Gynecologic Oncology 1

Evaluating new candidate SNPs as low penetrance risk factors in sporadicbreast cancer: A two-stage Spanish case–control study

Ana Vega a,1, Antonio Salas b,⁎,1, Roger L. Milne c, Begoña Carracedo a, Gloria Ribas d,Álvaro Ruibal e, Antonio Cabrera de León f, Ana González-Hernández g,

Javier Benítez c,d, Ángel Carracedo a,b

a Fundación Pública Galega de Medicina Xenómica (FPGMX-SERGAS), CIBERER, Santiago de Compostela, Galicia, Spainb Unidad de Genética, Instituto de Medicina Legal, Facultad de Medicina, Universidad de Santiago de Compostela,Galicia, Spain

c Centro Nacional de Genotipado (CeGen), Centro Nacional de Investigaciones Oncológicas (CNIO), Madrid, Spaind Programa de Genética del Cáncer Humano, CIBERER, Centro Nacional de Investigaciones Oncológicas (CNIO), Madrid, Spain

e Medicina Nuclear, Hospital Clínico Universitario,Santiago de Compostela, Galicia, Spainf Universidad de La Laguna and Hospital San Juan de Dios,Tenerife, Spain

g Instituto Canario de Investigación del Cáncer and Unidad de Investigación del Hospital de La Candelaria, Tenerife, Spain

Received 1 August 2008Available online 23 October 2008

Abstract

Objectives. A polygenic model has been proposed in order to explain the genetic susceptibility to sporadic breast cancer. According to thismodel, common population variants would be responsible for low to modest effects on the risk of developing the disease. We have carried out ahigh-throughput SNP genotyping project in order to shed some light on the complex genetic aetiology of non-familial breast cancer.

Methods. Ninety-one genes have been selected because of their implications in several candidate cell pathways for breast cancer. A total of 640SNPs in these genes were genotyped in a series of 450 consecutive cases and 448 controls from mainland Spain. Promising SNPs were thenstudied in an independent series of 294 cases and 299 controls from the Canary Islands.

Results. In the first case–control series we identified 25 SNPs with P-values below 0.05 (under a 1 df Chi-square test), five of them with P-values below 0.01 (best=0.0008). In the stage 2 Canary Islands series, odd ratios (OR) for two SNPs in HUS1 were in a consistent direction.

Conclusions. SNPs located at the gene HUS1 are good candidates for further investigation in independent association studies and functionalassays.© 2008 Elsevier Inc. All rights reserved.

Keywords: Breast cancer; Association study; Case–control; Candidate genes; SNPs; Gene–gene interaction

Introduction

Little is known about the complex aetiology of breast cancer.BRCA1 and BRCA2 are the most common high penetrancegenes described but only account for around 25% of familieswith hereditary breast cancer and a negligible number of

⁎ Corresponding author. Unidad de Genética Forense, Universidad deSantiago de Compostela, A Coruña, Galicia, Spain. Fax: +34 981 580336.

E-mail address: antonio.salas@usc.es (A. Salas).1 Contributed equally to this work.

0090-8258/$ - see front matter © 2008 Elsevier Inc. All rights reserved.doi:10.1016/j.ygyno.2008.09.012

sporadic breast cancers [1]. This fact, together with the failure oflinkage studies to identify further high susceptibility genes,points to a polygenic model. According to this model [2]susceptibility to breast cancer is conferred by a large number ofalleles, each contributing a small effect. Interest in population-based studies is continuously growing as a consequence of theaccessibility to genomic databases, with many million SNPsidentified. The HapMap International SNP genotyping project,conceptually based on the ‘common variant-common disease'hypothesis [3], has substantially favored the strategy known asthe ‘indirect' approach; according to this method, prior

211A. Vega et al. / Gynecologic Oncology 112 (2009) 210–214

knowledge of the SNP functionality is not necessary, since it ispossible to test for association by exploring SNP markers inlinkage disequilibrium (LD) with potential causal variants.

During the last decade, the analysis of one or a fewpolymorphisms in some isolated breast cancer candidate geneshas been quite popular [4–17]. On the other hand, high-throughput genotyping studies and genome-wide scans havebegun to emerge, identifying various potential low penetranceloci involved in sporadic breast cancer [18–25].

Here we aim to contribute to the knowledge of the complexgenetic aetiology of sporadic breast cancer by performing a two-stage, large-scale genotyping study that covers 91 breast cancercandidate genes.

Materials and methods

Study subjects

Cases in the Spanish mainland series were 450 women withbreast cancer and the mean age at diagnosis was 58 years (range25 to 85 years), recruited between 2000 and 2003. All caseswere collected from consecutive series recruited via threepublic Spanish hospitals: Hospital La Paz (∼15%), FundaciónJiménez Díaz (∼48%) and Hospital Monte Naranco (37%).Controls in the Spanish mainland series were 448 Spanishwomen free of breast cancer at ages ranging from 24 to 85 years(mean=52 years), and recruited between 2000 and 2004, viathe Menopause Research Centre at the Instituto Palacios (50%),the Colegio de Abogados (30%) and the Centro Nacional deTransfusiones (20%). The cases and controls genotyped in thepresent study are a subset of the samples genotyped in Milne etal. [18]; however, the SNPs genotyped were different (seebelow). The Canary Island series consisted of 294 sporadicbreast cancer cases from public hospitals and 299 controls fromthe cohort study “CDC of the Canary Islands” that enrolledparticipants from the general population across the archipelago.Mean ages for cases and controls were 55 years (range 33 to73 years) and 49 years (range 35 to 73 years) respectively.

Informed consent was obtained from all participants andthe study was approved by the institutional review boards ofHospital Clínico Universitario (Santiago de Compostela, Galicia,Spain) and Hospital La Paz, Madrid.

DNA extraction

Genomic DNA was isolated from peripheral blood lympho-cytes using automatic DNA extraction (Magnapure, Roche)according to the manufacturer's recommended protocols. DNAwas quantified using picogreen and diluted to a finalconcentration of 50 ng/ul for genotyping.

Candidate gene choice, SNP selection

Most of the 91 candidate genes analyzed (see Supplementarydata in Table S1) were selected because they belong to variousBRCA1/2-linked pathways. SNP selection across each of thesegenes was carried out using a combined direct/indirect approach

and using information from HapMap Phase I. Density was theprimary criterion with 1 SNP spaced every ∼10 kb on average.We gave priority to tagSNPs defining common haplotypes(indirect) but also to those SNPs with potentially functionaleffects wherever possible (direct). An additional set of 28unlinked SNPs across the genome were independently selectedto be used as markers to assess population stratification. All 28were chosen to be at least 100 kb from genes.

Genotyping

Genotyping in the Spanish mainland series was carried outusing the high-throughput SNPlex™ (Applied Biosystem, Foster-City, USA) platform (located at theCentroNacional deGenotipado[CeGen] in Santiago de Compostela). Genotyping, both in theCanary Island series and for the 28 marker SNPs to assesspopulation stratification in the mainland series, was carried out atthe CeGen in Santiago de Compostela, using the MassARRAYgenotyping system (Sequenom Inc., San Diego, USA). TheGenotyping Data Filter (GDF) software (http://bioinformatics.cesga.es/gdf/) was use to pre-process the genotyping data.

Association tests

STATA v.8. was used to test for departure from HardyWeinberg equilibrium (HWE) of candidate SNPs among controlsamples. A modified Bonferroni-corrected nominal P-valuethreshold of 0.05/N1⁎ was used, where N1⁎ is the “effectivenumber of independent marker loci” after consideration of LDbetween SNPs (marker loci) on the same chromosome. N1⁎ wascalculated by applying the web-based program SNPSpD [26,27]to SNPs on individual chromosomes and summing estimatesacross chromosomes. This approach approximates adjustmentfor multiple testing using permutation methods [26,28].

Associations were assessed for individual SNPs by comparingallele frequencies between cases and controls using a one degreeof freedom Chi-square test, or Fisher's exact test for cell countsbelow five. Age in years was adjusted for by including it as acategorical variable with categories: b35; 35–39; 40–44; 45–49;50–54; 55–59; N60 in unconditional logistic regression models.Three methods were considered to address the issue of multipletesting. The Bonferroni correction was applied based on N2⁎

independent tests. N2⁎ was calculated in the same way as N1⁎

(described above), but after first excluding SNPs found to violateHWE. This method appropriately accounts for the non-independence of SNPs on the same chromosome due to LD[26,27]. The distribution-free method of controlling the FDR ofBenjamini and Liu, which is robust to the presence of non-independent explanatory variables [29], was also applied. Finally,Haploview v3.11 [30] was used to carry out a permutation test.

The cocaphased software [31] was used to test for haplotypeassociations. Preliminary results were shown in a congresspresentation [32]. The nonparametric and model-free MDRmethod was used to test for gene–gene interactions [33–35].Finally, we ran Structure v.2.0 [36] to investigate the potentialstructure of our case–control data using 28 unlinked neutralbialelic loci.

212 A. Vega et al. / Gynecologic Oncology 112 (2009) 210–214

Results

SNP characteristics

A total of 640 SNPs belonging to 91 genes were genotypedin the Spanish mainland series. Allele frequencies in controlswere highly consistent with those reported for CEPH indivi-duals of the HapMap project as reported in Ribas et al. [37].About 68% of the SNPs were located in introns, 5% in codingregions, 6% 5′-upstream, 9% 3′-downstream, 11% 3′-utr, and1% 5′-utr. Fourteen SNPs were monomorphic (see Table S1)and were therefore excluded from all analyses. All the SNPwere in Hardy–Weinberg equilibrium under the criteriaspecified above (N1⁎=363.5) with the exception of five SNPs(rs1972958, rs2448345, rs4858767, rs661713, and rs747443)which were also excluded from further analyses.

Individual SNP association test in the Spanish mainland series

Table S1 gives unadjusted P-values for the comparison ofallele frequencies for individual SNPs. A total of 25 SNPs showedP-values below the standard 0.05 threshold, while five SNPsyielded P-values below 0.01 (Table 1). These latter candidateSNPs belong to genes GRIP1 (rs2172881; P=0.0063), HUS1(rs1056663 and rs2708861, P=0.0019 and P=0.0017, respec-tively), NRIP1 (rs926184; P=0.0043), and PPARGC1A(hcv8264045; P=0.0097). None of these P-values remainedbelow 0.05 after multiple test correction under any of the threemethods used. Note in addition that five of SNPs with P-valuesbelow 0.05 had MAF below 10%. Table 1 summarizes theestimated ORs with the corresponding 95% confidence interval(CI) for the five SNPs with P-values below 0.01 (P-values for the640 SNPs are available in Table S1). All had MAF higher than40% except one (rs926184) with a MAF of just 2%. An intronicSNP (rs2172881) in GRIP1 and an intergenic SNP (rs926184)upstream of NRIP1 appeared associated with protection frombreast cancer (OR=0.76 and 0.15, respectively). The other three,an intronic SNP (hCV8264045) in PPARGC1A and two SNPs inHUS1, one (rs2708861) on intron 3 and the other (rs1056663) onexon 8, were all associated with increased risk. The two SNPs inHUS1 were in high LD (r2=0.99). Odds ratios did not changesubstantially after adjustment for age. These associations didnot remain statistically significant after correction for multipletesting by the Bonferroni correction (N2⁎=359.5), control of theFDR nor the permutation test.

Table 1Data on the five candidate SNPs identified under the uni-variate Chi-square test com

SNP_ID Gene Region Spanish mainland C

OR (95% CI) P-value O

rs2172881 GRIP1 Intronic 0.76 (0.63–0.93) 0.0092 0rs1056663 HUS1 Coding 1.36 (1.12–1.64) 0.0010 1rs2708861 HUS1 Intronic 1.37 (1.13–1.67) 0.0008 1rs926184 ⁎ NRIP1 Intergenic 0.15 (0.02–0.66) 0.0050 0hCV8264045 PPARGC1A Intronic 1.29 (1.06–1.57) 0.0092 0

* Fisher's exact test was applied for this particular SNP; MAF: minor allele frequ

Analysis of the 28 unlinked SNPs in stage 1 using Structurefound that the posterior probability was very close to one forK=1, a result which was consistent across independent runs.Also consistent was the finding that, for K=2, the minimumand maximum ancestry coefficients were 0.45 and 0.55, res-pectively (see also Milne et al. [18]). In other words, at thislevel of resolution, we did not detect evidence of populationstratification.

Haplotype analysis in the spanish mainland series

We observed a total of three two-SNP haplotypes, one three-SNP, and one four-SNP haplotypes showing a signal of positiveassociation, with P-values below 0.001 (Table 2). All two-SNPhaplotypes were formed by SNPs in relatively high LD(r2N0.75); however, one of the SNPs forming the three-SNPhaplotype (rs289916) was in low LD with the other two(r2 =0.39). The three and four-SNPs haplotypes are extensionsof a LD block located in the gene NMI (as observed by the highr2 values characterizing the four-SNPs haplotype). Thespectrum of haplotype frequencies varied from 0.01 to 0.25.While these associations were not statistically significant aftercorrection for multiple testing, they identify three additionalcandidate genes: YWHAH, NMI and ESR1.

Gene–gene interactions in the spanish mainland series

Table S2 shows the best MDR models. The results arepartially consistent in all the MDR runs and using the threefiltering methods. The SNP rs1268891 in the NCOR2 geneappears in all two- and three-factor combinations, while thers1168308 (GRIP1) and the rs1050540 (TP53) SNPs alsoappear in some combinations. The statistical significance for thebest gene–gene models disappeared after correction for multi-ple testing using the permutation test.

Evaluation of candidate SNPs in the canary island series

The 25 SNPs identified as candidates from the stage 1 studyin Spanish mainland series were genotyped in the Canary Islandseries. While none of the associations from stage 1 wereindependently replicated, odds ratios for the association with thetwo SNPs in HUS1 were consistent in direction. The Mantel–Hanzel combined odds ratio for rs1056663 was 1.28 (95% CI:1.10–1.48) with corresponding, unadjusted P-value 0.0013,

paring allele frequencies

anary Islands Combined MAFcontrols

R (95% CI) P-value OR (95% CI) P-value

.99 (0.78–1.27) 0.9516 0.85 (0.73–0.99) 0.0412 0.45

.13 (0.90–1.44) 0.2904 1.28 (1.10–1.48) 0.0013 0.47

.11 (0.87–1.40) 0.4119 1.27 (1.09–1.47) 0.0019 0.47

.81 (0.50–1.31) 0.4231 0.65 (0.43–0.99) 0.0422 0.02

.96 (0.76–1.22) 0.7252 1.14 (0.98–0.33) 0.0801 0.44

ency.

Table 2Haplotype analysis of the Spanish mainland series data (stage 1)

SNP_ID Gene Haplotypealleles

LR GlobalP-value (df)

HaplotypeP-value

Freq.cases

Freqcontrols

r2

Two-SNP haplotypesrs2246704–rs2858753 YWHAH A–A 14.87 0.0005 (2) 0.0006 0.25 0.18 0.96rs2271811–rs3771882 NMI C–A 14.96 0.0018 (3) 0.0003 0.02 0.04 0.89rs3778082–rs3778090 ESR1 T–A 13.34 0.0012 (2) 0.0003 0.11 0.06 0.75

Three-SNP haplotypesrs289916–rs2271811–rs3771882 NMI G–C–A 14.16 0.0068 (4) 0.0005 0.01 0.03 0.39–0.39–0.89

Four-SNP haplotypesrs2271811–rs3771882–rs3771886–rs6433327 NMI G–A–G–C 12.17 0.0068 (3) 0.0008 0.01 0.04 0.89–0.60–0.90–0.61

LR= maximum likelihood ratio for the region interrogated by the corresponding haplotype, Global P-value refers to the Chi-square value for the region interrogated bythe haplotype; df=degrees of freedom; r2=correlation coefficient between SNPs; for the three-SNP haplotype, r2 is given for the three pairwise combinationsrs289916-rs2271811, rs289916-rs3771882, and rs2271811-rs3771882, respectively, whereas in the four-SNPs haplotype refers to the combinations rs2271811–rs3771882, rs2271811–rs3771886, rs2271811–rs6433327, rs3771882–rs3771886, rs3771882–rs6433327, and rs3771886–rs6433327, respectively.

213A. Vega et al. / Gynecologic Oncology 112 (2009) 210–214

while for rs2708861 it was 1.27 (95% CI: 1.09–1.49, P-value=0.0019). Results from haplotype and interaction ana-lyses were not consistent with those from stage 1. The SNPrs3824486 (one of the SNPs in a candidate three-wayinteraction, see Table S2) showed some evidence of associationindividually (OR=1.37; 95% CI = 1.08–1.74; unadjusted P-value=0.0092), but in the opposite direction to that estimated instage 1 (OR=0.88; 95% CI=0.70–1.10; P-value=0.2424).

Discussion

In our two-stage case–control study, we first aimed to test forpotential cryptic population structure in stage 1 in order to rule itout as a potential cause of some spurious associations. Noevidence of strong population stratification was observed in themainland Spanish series, though we are aware that the use of 28unlinked loci may not completely exclude the existence of a moresubtle level of population substructure. In the first stage study, wedetected nominal P-values below 0.05 for 25 SNPs with involvedin either individual or haplotype associations, or interactions. Ofthese, only GRIP1, HUS1, NRIP1 and PPARGC1A harbor SNPswith associated P-values below 0.01 (Table 1); the best valuebeing 0.0008 for rs2708861 in HUS1. While none of theseputative associations were independently confirmed in the stage 2Canary Islands series, the ORs for two SNPs in HUS1 wereconsistent in direction. These SNPs are in complete LD and theP-value for analysis of the pooled data for the exonic SNPrs1056663 was 0.0013. The protein product of HUS1 (hidrox-yurea sensitive 1) [38] plays a key role in the pathway “Role ofBRCA1, BRCA2 and ATR in cancer susceptibility”. This SNPoccurs at high frequencies in the Spanish population(MAFN45%). Note that there is independent evidence thatrelates HUS1 with cancer [38,39]; therefore, our results addsupport to the potential role of this gene in sporadic breast cancer.

As in previous published attempts [19], we detected somenovel SNPs with weak evidence for association with non-familiar breast cancer in stage 1. Our two-stage designhighlighted SNPs in HUS1 as the most promising candidates.We here advance that further studies are needed in order to

replicate this potential association, since power was limited inour second, Canary Island series. Ideally, in order to avoidundesirable type I errors that commonly arise in population-based studies [40], weak statistical signals for associationmirrored by P-values below a certain threshold should not betaken on face value as true biological associations. Furtherresearch, including replication studies in independent samplesand functional assays, would therefore be desirable.

Conflict of interest statementThe authors declare that there are no conflicts of interest.

Acknowledgments

Thanks to Pilar Zaragoza and José Ignacio Arias for providingpart of the samples. This work was supported by grants from theXunta de Galicia, PGIDIT06PXIB208079PR and PGI-DIT06BTF910101PR, given to AS and AV respectively, twogrants from the Fundación de Investigación Médica MutuaMadrileña awarded to AS and AV, and two grants from theSpanish Ministerio de Sanidad y Consumo, PI052275 andPI061712, given to AV. and AG. respectively. Genoma Españaalso supported the present study.

Appendix A. Supplementary data

Supplementary data associated with this article can be found,in the online version, at doi:10.1016/j.ygyno.2008.09.012.

References

[1] Thompson D, Easton D. The genetic epidemiology of breast cancer genes.J Mammary Gland Biol Neoplasia 2004;9:221–36.

[2] Antoniou AC, Pharoah PD, McMullan G, Day NE, Stratton MR, Peto J,et al. A comprehensive model for familial breast cancer incorporatingBRCA1, BRCA2 and other genes. Br J Cancer 2002;86:76–83.

[3] Altshuler D, Brooks LD, Chakravarti A, Collins FS, Daly MJ, Donnelly P.A haplotype map of the human genome. Nature 2005;437:1299–320.

[4] Dunning AM, Healey CS, Pharoah PD, Teare MD, Ponder BA, Easton DF.A systematic review of genetic polymorphisms and breast cancer risk.Cancer Epidemiol Biomarkers Prev 1999;8:843–54.

214 A. Vega et al. / Gynecologic Oncology 112 (2009) 210–214

[5] Lee KM, Choi JY, Park SK, Chung HW, Ahn B, Yoo KY, et al. Geneticpolymorphisms of ataxia telangiectasia mutated and breast cancer risk.Cancer Epidemiol Biomarkers Prev 2005;14:821–5.

[6] Angele S, Romestaing P, Moullan N, Vuillaume M, Chapot B, Friesen M,et al. ATM haplotypes and cellular response to DNA damage: associationwith breast cancer risk and clinical radiosensitivity. Cancer Res 2003;63:8717–25.

[7] Szabo CI, Schutte M, Broeks A, Houwing-Duistermaat JJ, ThorstensonYR, Durocher F, et al. Are ATM mutations 7271T→G and IVS10-6T→Greally high-risk breast cancer-susceptibility alleles? Cancer Res 2004;64:840–3.

[8] Lindeman GJ, Hiew M, Visvader JE, Leary J, Field M, Gaff CL, et al.Frequency of the ATM IVS10-6T→G variant in Australian multiple-casebreast cancer families. Breast Cancer Res 2004;6:R401–407.

[9] Bernstein L, Teal CR, Joslyn S, Wilson J. Ethnicity-related variation inbreast cancer risk factors. Cancer 2003;97:222–9.

[10] Maguire P, Holmberg K, Kost-Alimova M, Imreh S, Skoog L, LindblomA. CGH analysis of familial non-BRCA1/BRCA2 breast tumors andmutation screening of a candidate locus on chromosome 17q11.2-12. Int JMol Med 2005;16:135–41.

[11] Gold B, Kalush F, Bergeron J, Scott K, Mitra N, Wilson K, et al. Estrogenreceptor genotypes and haplotypes associated with breast cancer risk.Cancer Res 2004;64:8891–900.

[12] Wedren S, Lovmar L, Humphreys K, Magnusson C, Melhus H, SyvanenAC, et al. Oestrogen receptor alpha gene haplotype and postmenopausalbreast cancer risk: a case control study. Breast Cancer Res 2004;6:R437–449.

[13] Tamimi RM, Hankinson SE, Spiegelman D, Kraft P, Colditz GA, HunterDJ. Common ataxia telangiectasia mutated haplotypes and risk of breastcancer: a nested case–control study. Breast Cancer Res 2004;6:R416–422.

[14] Cox DG, Kraft P, Hankinson SE, Hunter DJ. Haplotype analysis ofcommon variants in the BRCA1 gene and risk of sporadic breast cancer.Breast Cancer Res 2005;7:R171–175.

[15] Benusiglio PR, Lesueur F, Luccarini C, Conroy DM, Shah M, Easton DF,et al. Common ERBB2 polymorphisms and risk of breast cancer in awhite British population: a case–control study. Breast Cancer Res 2005;7:R204–209.

[16] Lesueur F, Pharoah PD, Laing S, Ahmed S, Jordan C, Smith PL, et al.Allelic association of the human homologue of the mouse modifier Ptprjwith breast cancer. Hum Mol Genet 2005;14:2349–56.

[17] Barroso E, Milne RL, Fernández LP, Zamora P, Arias JI, Benitez J, et al.FANCD2 associated with sporadic breast cancer risk. Carcinogenesis2006;27:1930–7.

[18] Milne RL, Ribas G, González-Neira A, Fagerholm R, Salas A, González E,et al. ERCC4 associated with breast cancer risk. A two-stage case–controlstudy using high-throughput genotyping. Cancer Res 2006;66:9420–7.

[19] Kammerer S, Roth RB, Reneland R, Marnellos G, Hoyal CR, MarkwardNJ, et al. Large-scale association study identifies ICAM gene region asbreast and prostate cancer susceptibility locus. Cancer Res 2004;64:8906–10.

[20] Kammerer S, Roth RB, Hoyal CR, Reneland R, Marnellos G, Kiechle M,et al. Association of the NuMA region on chromosome 11q13 with breastcancer susceptibility. Proc Natl Acad Sci U S A 2005;102:2004–9.

[21] Hoyal CR, Kammerer S, Roth RB, Reneland R, Marnellos G, Kiechle M,et al. Genetic polymorphisms in DPF3 associated with risk of breast cancerand lymph node metastases. J Carcinog 2005;4:13.

[22] Stacey SN,ManolescuA, SulemP, Thorlacius S, Gudjonsson SA, JonssonGF,et al. Common variants on chromosome 5p12 confer susceptibility to estrogenreceptor-positive breast cancer. Nat Genet 2008;40:703–6.

[23] Stacey SN,ManolescuA, SulemP,Rafnar T,Gudmundsson J,Gudjonsson SA,et al. Common variants on chromosomes 2q35 and 16q12 confer susceptibilityto estrogen receptor-positive breast cancer. Nat Genet 2007;39:865–9.

[24] Hunter DJ, Kraft P, Jacobs KB, Cox DG, Yeager M, Hankinson SE, et al. Agenome-wide association study identifies alleles in FGFR2 associated withrisk of sporadic postmenopausal breast cancer. Nat Genet 2007;39:870–4.

[25] Easton DF, Pooley KA, Dunning AM, Pharoah PD, Thompson D,Ballinger DG, et al. Genome-wide association study identifies novel breastcancer susceptibility loci. Nature 2007;447:1087–93.

[26] Nyholt DR. A simple correction for multiple testing for single-nucleotidepolymorphisms in linkage disequilibrium with each other. Am J HumGenet 2004;74:765–9.

[27] Nyholt DR. Evaluation of Nyholt's procedure for multiple testingcorrection — author's reply. Hum Hered 2005;60:61–2.

[28] Li J, Ji L. Adjusting multiple testing in multilocus analyses using theeigenvalues of a correlation matrix. Heredity 2005;95:221–7.

[29] Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practicaland powerful approach to multiple testing. J R Stat Soc B 1995;57:289–300.

[30] Barrett JC, Fry B, Maller J, Daly MJ. Haploview: analysis andvisualization of LD and haplotype maps. Bioinformatics 2005;21:263–5.

[31] Dudbridge F. Pedigree disequilibrium tests for multilocus haplotypes.Genet Epidemiol 2003;25:115–21.

[32] Vega A, Salas A, Phillips C, Sobrino B, Carracedo B, Ruiz-Ponte C, et al.Large-scale single nucleotide polymorphism analysis of candidates forlow-penetrance breast cancer genes. Breast Cancer Res 2005;7(Suppl 22):S22–22.

[33] Ritchie MD, Hahn LW, Moore JH. Power of multifactor dimensionalityreduction for detecting gene–gene interactions in the presence ofgenotyping error, missing data, phenocopy, and genetic heterogeneity.Genet Epidemiol 2003;24:150–7.

[34] Ritchie MD, Hahn LW, Roodi N, Bailey LR, Dupont WD, Parl FF, et al.Multifactor-dimensionality reduction reveals high-order interactionsamong estrogen-metabolism genes in sporadic breast cancer. Am J HumGenet 2001;69:138–47.

[35] Hahn LW, Ritchie MD, Moore JH. Multifactor dimensionality reductionsoftware for detecting gene–gene and gene–environment interactions.Bioinformatics 2003;19:376–82.

[36] Pritchard JK, Stephens M, Donnelly P. Inference of population structureusing multilocus genotype data. Genetics 2000;155:945–59.

[37] Ribas G, Gonzalez-Neira A, Salas A, Milne RL, Vega A, Carracedo B,et al. Evaluating HapMap SNP data transferability in a large-scalegenotyping project involving 175 cancer-associated genes. Hum Genet2006;118:669–79.

[38] Zimmerman ES, Chen J, Andersen JL, Ardon O, Dehart JL, Blackett J,et al. Human immunodeficiency virus type 1 Vpr-mediated G2 arrestrequires Rad17 and Hus1 and induces nuclear BRCA1 and gamma-H2AXfocus formation. Mol Cell Biol 2004;24:9286–94.

[39] Huang J, Yuan H, Lu C, Liu X, Cao X, Wan M. Jab1 mediates proteindegradation of the Rad9-Rad1-Hus1 checkpoint complex. J Mol Biol2007;371:514–27.

[40] Ioannidis JP, Ntzani EE, Trikalinos TA. 'Racial' differences in geneticeffects for complex diseases. Nat Genet 2004;36:1312–8.