SNPs in the kallikrein gene region associated with prostate cancer risk: true cause or association by design?

We recently conducted a genome-wide association study (GWAS) using data from 1,854 PrCa cases with clinically detected (not PSA screened) PrCa diagnosed at <60 years or with a family history of the disease, and 1,894 population screened controls with a prostate specific antigen (PSA) of <0.5ng/ml (Eeles et al. 2008). These were analysed for 541,129 SNPs using the Illumina Infinium platform. Putative associations were evaluated using a further 3,268 cases and 3,366 controls. After these two stages, associations at seven loci, on chromosomes 2,3,6,7,10,11,19 and X, reached genome-wide levels of significance (p=2.7×10⁻⁸ to p=8.7×10⁻²⁹). The SNP rs2735839 on chromosome 19 lies between two kallikreins, PSA (KLK3) and hK2 (KLK2). It was associated with a per allele OR for PrCa of 0.83 (95%CI 0.75-0.91; ptrend in stage 2 = 0.0002; ptrend overall = 2×10⁻¹⁸). We also showed that rs2735839 was strongly associated with PSA level, in the direction consistent with the disease association (per allele rise in geometric mean PSA 1.12., p=6×10⁻⁸).

Ahn et al (this issue of Nature Genetics) analysed 24 tagSNPs in the kallikrein region on chromosome 19 (to include KLK1, KLK2, KLK3 and KLK15) in five studies and found that none showed a significant association with PrCa risk. They also confirm the association between several SNPs, including rs2735839, and PrCa risk. They raise the possibility that the association found with PrCa risk in our study may reflect the selection of subjects based on PSA levels rather than a causal relationship with PrCa risk.

It is clear that the selection of controls in stage 1 of our study for low PSA levels did influence the association in stage 1. This is reflected in the minor allele frequency for rs2735839, which is 21.1% in the stage 1 controls, compared with 14%-15% in the UK 1958 birth cohort and the CGEMS study (males and females). However, the controls in stage 2 were not highly selected for PSA level. The only selection was to exclude controls with PSA levels of >10, and to require a negative prostatic biopsy if the PSA was >4. The MAF in the stage 2 controls (15.2%) is similar to that in other control populations and indicates that any selection bias at this stage was minimal.

To further evaluate the evidence for this association, we have undertaken an analysis of rs2735839 (together with SNPs at the other loci identified in our GWAS) in 13 further case-control studies as part of The PRACTICAL consortium. These studies comprise 7,370 PrCa cases and 5,342 controls. The estimated per allele OR for PrCa associated with rs2735839 was 0.89 (95% CI 0.83-0.95; p=.0007), very close to our original estimate (Kote-Jarai, CEBP in press, 2008, cited with permission). There was no evidence of heterogeneity in the OR estimates among studies (see figure 1). We also note that when data from the five CGEMS studies are combined, the per allele OR is also remarkably similar (per allele OR 0.90, 0.83-0.90; P=.01), although this was not formally significant using the 4 degree freedom test given by Ahn et al (2008). If the results from our stage 2, PRACTICAL and CGEMS are combined, the overall evidence of association reaches genome-wide levels of significance (p<10⁻⁸), demonstrating that, even disregarding our stage 1 result, the association is unlikely to be due to chance. The overall effect size, while modest, is comparable to that seen for other cancer associated loci.

figure 1.

None of the control series used in PRACTICAL, nor in CGEMS, involved selection for PSA level, and for this reason and those given above, the association appears unlikely to be driven purely by control selection. Selection bias related to case ascertainment is an alternative possible explanation. We excluded from our GWAS any cases identified through PSA screening, and several of the studies included in PRACTICAL are drawn from populations where PSA screening has not been used (e.g. the study from Finland). Thus, the association is unlikely to be due to PSA screening for asymptomatic disease. PSA testing is, however, also used in the process of diagnosis of symptomatic disease. This raises the possibility of a more subtle bias, in that some cases may have raised PSA related to the genotype but not related to their disease. Whether or not this potential bias is significant could be resolved using genotyping in studies based on biopsy of whole populations not driven by the PSA level (e.g. the Prostate Cancer Prevention Trial; Thompson et al., 2007), or studies where mortality is the endpoint.

Conversely, there are plausible biological grounds for believing that the association with KLK polymorphisms may be causal. For example, polymorphisms in the promoter of KLK3 are associated with alterations in androgen receptor binding (e.g. Lai et al., 2007). Moreover, it is known that PSA level is a long-term predictor of prostate cancer risk (Lilja et al, 2007), and it is plausible that determinants of PSA level, including genetic determinants, may influence prostate cancer risk.

The PRACTICAL Consortium

UK Genetic Prostate Cancer Study.

The Institute of Cancer Research & The Royal Marsden NHS Foundation Trust, Sutton UK.

Rosalind A. Eeles

Michelle Guy

Zsofia Kote-Jarai

Steve Edwards

Audrey Ardern-Jones

Rosemary Wilkinson

Amanda Hall

Rosemary Wilkinson

Lynne O'Brien

Daniel Leongamornlert

Malgorzata Tymrakiewitz

Sameer Jhavar

David P. Dearnaley

Alan Horwich

Robert A. Huddart

Vincent S. Khoo

Christopher C. Parker

Christopher J. Woodhouse

Alan Thompson

Tim Christmas

Chris Ogden

Cyril Fisher

Charles Jamieson

Colin S. Cooper

The UK Genetic Prostate Cancer Study Collaborators

The British Association of Urological Surgeons' Section of Oncology

Cancer Research UK Genetic Epidemiology Group, Cambridge, UK:

Douglas Easton

Ali Amin Al Olama

Jonathan Morrison

The ProtecT Study, UK:

David Neal

Jenny Donovan

Freddie Hamdy

Angela Cox

Sarah Lewis

Paul M. Brown

Gemma Marsden

The UK ProtecT Study Collaborators

University of Nottingham, UK:

Kenneth Muir

Artitaya Lophatananon

Chulabhorn Cancer Research Centre, Thailand:

Jo-fen Liu

The Melbourne Studies, Australia:

Graham Giles

John Hopper

Gianluca Severi

Melissa Southey

Dallas English

John Pedersen

Fred Hutchinson Cancer Research Center, Seattle, Washington, USA:

Janet L. Stanford

Claudia A. Salinas

Joseph S. Koopmeiners

National Human Genome Research Institute, National Institutes of Health, Bethesda MD, USA:

Elaine A. Ostrander

Danielle M. Karyadi

Bo Johanneson

University of Southern California Keck School of Medicine, Los Angeles CA, USA:

Sue A. Ingles

Mariana C. Stern

Roman Corral

Northern California Cancer Center, Fremont, California, USA and Stanford University School of Medicine, Stanford, California, USA:

Esther M. John

Amit D. Joshi

The Montreal Study:

William D. Foulkes

Nancy Hamel

Kimberley Kotar

Ulm, Germany:

Walter Vogel

Christiane Maier

Rainer Kuefer

Manuel Luedeke

Harald Surowy

Bärbel Weber

Kathleen Herkommer

Thomas Paiss

Tampere, Finland:

Johanna Schleutker

Tiina Wahlfors

Henna Mattila

Sanna Siltanen

Sanna Pakkanen

Jarkko Isotalo

Teuvo L. Tammela

Mika Matikainen

The Hannover Prostate Cancer Study:

Thilo Dörk

Peter Schürmann

Andreas Meyer

Stefan Machtens

Jörn Hagemanns

Peter Hillemanns

Michael Bremer

Johann H. Karstens

Prostate Cancer Study in Valais, Switzerland:

Pierre O. Chappuis

Pierre Hutter

Cédric Biedermann

Hans-Ulrich Peter

Nicolas Defabiani

Sabine Bieri

Christophe Girardet

Isabelle Konzelmann

Michéle Stalder

Marie-Mathilde Meier

Queensland, Australia:

Mary-Anne Kedda

Kimberly Hinze

Amanda Spurdle

Judith Clements

Beth Newman

Suzanne Steginga

Tracy O'Mara

John Yaxley

David Nicol

Megan Ferguson

David Nicol

R.A. (Frank) Gardiner

Joanne Aitken

Mayo Clinic:

Daniel Schaid

Stephen Thibodeau

Liang Wang

Julie Cunningham

Shannon K. McDonnell