Identification of low penetrance alleles for lung cancer: The GEnetic Lung CAncer Predisposition Study (GELCAPS)

Lung cancer is a major cause of cancer mortality worldwide [1]. In the United Kingdom, it accounts for more than 33,000 cancer deaths each year (Cancer Research UK). The disease is frequently cited as a malignancy solely attributable to environmental exposure, principally tobacco smoking. It has, however, long been postulated that individuals may differ in their susceptibility and there is strong evidence from epidemiological studies for a familial risk [reviewed in [2]]. Direct evidence for a genetic predisposition is provided by the increased risk of lung cancer associated with a number of rare Mendelian cancer syndromes, such as in carriers of germline TP53 [3]and RB [4, 5] mutations, as well as in patients with Bloom's [6] and Werner's [7] syndromes.

The two major types of lung cancer, non-small cell lung cancer (NSCLC), and small cell lung cancer (SCLC) account for 75% and 25% of cases respectively. Although the histological features are different between these (reflected in differences in patterns of gene expression), there are similarities in the spectrum of underlying somatic genetic alterations suggesting commonality in pathogenesis. Moreover, the observation that the familial risks are not subtype dependent [8‐13] and that histological concordance between affected family members is poor [9] is consistent with the hypothesis of a "generic" inherited susceptibility to lung cancer.

The genetic basis of inherited susceptibility to lung cancer outside the context of the rare Mendelian cancer predisposition syndromes is at present undefined, but a model in which major gene loci account for the excess familial risk seems unlikely. One hypothesis about the allelic architecture of susceptibility proposes that part of the genetic risk is caused by disease loci, which include common, low penetrance alleles. This "common-disease common-variant" hypothesis implies that conducting association analyses based on scans of Single Nucleotide Polymorphisms (SNPs) should be a powerful strategy for identifying low-penetrance variants [14, 15].

Previous studies aimed at identifying low penetrance alleles for lung cancer susceptibility have largely been based on a candidate gene approach formulated on preconceptions as to the role of specific genes in the development of the disease. Perhaps not surprisingly most studies have to date only evaluated a restricted number of polymorphisms, primarily in genes implicated in the metabolism of tobacco-associated carcinogens and the protection of DNA from carcinogen-induced damage. However, without a clear understanding of the biology of lung cancer predisposition the definition of suitable genes for the disease is inherently problematic making an unbiased approach to loci selection highly desirable.

Despite much research, few definitive low penetrance susceptibility alleles for lung cancer have been to date unequivocally been identified through candidate-based association studies. As with many other diseases, positive associations have been reported for various polymorphisms of genes such as GSTT1 [16], GSTM1 [17], ERCC2 [18], CYP1A1 [19], and TP53 [20] from small studies, but few of the initial positive results have been replicated in subsequent studies. The inherent statistical uncertainty of case-control studies involving just a few hundred cases and controls seriously limits the power of such studies to reliably identify genetic variants conferring modest but potentially important risks.

In addition to genetic variation affecting the risk of developing cancer it is increasingly being recognised that genetic variation, not necessarily in the same genes, may also affect clinical outcome. As with case-control association studies aiming to identify novel susceptibility alleles the same issues of study power pertain to the search for prognostic markers and such studies are again contingent on access to large case-series.

Following the sequencing of the human genome, large-scale harvests of SNPs have been conducted and > 10 million documented. Patterns of linkage disequilibrium (LD) between SNPs have been characterised allowing subsets of SNPs (tagging SNPs) to be selected that capture a large proportion of the common sequence variation in the human genome. This coupled with the advent of highly efficient analytical platforms allow whole genome-wide studies (GWAS) for disease associations to be conducted cost effectively. The relationship between patients' genotype and risk of lung cancer is now open for exploration.

The identification of genes associated with cancer predisposition and determination of their contribution to disease incidence are contingent on having DNA samples from large, systematic series of cancer patients. The resulting genetic epidemiological data provides the information on which to base the identification, counselling and management of at-risk individuals. The National Cancer Research Network (NCRN) was established to provide support for clinical cancer research in England and is one of the most substantial and constructive developments in the area of cancer research to be made in recent years in the United Kingdom. In England, serving a population of 50 million people, the NCRN is made up of 34 geographically distinct Networks covering the entire country. Within each Network there are clinical research support staff and infrastructure to promote accrual of patients into trials and studies, and the collection of high quality clinico-pathological data and appropriate biological samples. Hence the NCRN presents a major scientific initiative not only in the field of clinical trials but also in the field of genetic epidemiology.

To create a resource for identifying low penetrance alleles for lung cancer we established GELCAPS (GEnetic Lung CAncer Predisposition Study) in March 1999 to collect DNA and clinico-pathological data from a large series of lung cancer cases. Within 5-years of setting up the initiative by linkage with the NCRN it has been possible to create a world-class resource of biological and clinico-pathological data from over 5,000 individuals with lung cancer.

After securing ethical permissions at a national level through the Multi Research Ethics Committee we started recruitment to GELCAPS in March 1999. Ascertainment of cases was restricted to 28 centres and accrual was maximally 10–20 patients per month. After GELCAPS was incorporated into the NCRN (National Cancer Research Network) portfolio in March 2002 it was subsequently rolled out across England after individual centers had obtained local ethical permissions. Adoption by the NCRN was associated with a significant increase in patient and control accrual (Figure 1). Eventually 140 oncology centers (Figure 2) became active participants in GELCAPS with patient ascertainment averaging ~100 cases per month. The remit and operational procedure by which patients are accrued to NCRN adopted studies does not allow collection of compliance data within each participating center. However, we estimate based on our intimate knowledge of the clinical activities of three centers that patient accrual to GELCAPS is ~70% of those invited to participate.

Bild vergrößern

Bild vergrößern

The original target of GELCAPS was to assemble a series of 2,000 lung cancer cases. Given the efficiency by which samples were being accrued following adoption of GELCAPS by the NCRN a new target of at least 4,000 cases was deemed to be eminently feasible within the time frame for which funding had been secured.

We terminated accrual to GELCAPS in July 2004 by which time samples from 5,269 cases with primary lung cancer and 2,094 controls had been recruited. The majority of cases were male (64%) reflecting the sex preponderance of disease. Whilst the mean age of controls was comparable to cases (62.9 years, SD = 10.6) not surprisingly 69% were female (Table 1). Similarly, the prevalence of smoking was significantly higher amongst cases compared to controls. A high proportion of the cases ascertained had been diagnosed with lung cancer at a young age (Table 1); specifically 1,617 (~31%) of the cases were aged less than 60 years old at diagnosis, compared with < 10% in the general population. The frequency of the various forms of lung cancer was, however, in keeping with that observed in UK general population – ~23% being affected with SCLC and ~73% with NSCLC (Table 1).

To date we have acquired follow up data on 1,187 patients; specifically, information on the staging, management and clinical outcome permitting comparison patients randomly drawn from the general population. Stage at presentation for each of the different subtypes of lung cancer was similar to that observed in the general population; specifically, for patients with SCLC, somewhat less than half (43%) presented with limited disease and of the patients with NSCLC, 13% had stage I, 15% had stage II, 43% had stage III, and 29% had stage IV disease. The majority of patients with limited stage SCLC had been treated with a combination of radical radiotherapy and chemotherapy, whilst all patients received chemotherapy. The main treatment modality for SCLC patients with extensive disease was chemotherapy. Patients with early stage NSCLC (stage I and II disease) were mainly treated with surgical resection of the primary tumor whilst about one third received chemotherapy and radical radiotherapy. The mainstay treatment modality of patients with stage III and IV NSCLC was chemotherapy. Overall the median survival time (MST) for the subset of 1,187 GELCAPS patients was 18.6 months. Prognosis was significantly correlated with stage at presentation, with those presenting early have a far better survival (Figure 3). Patients with SCLC had a MST of 26.0 and 10.5 months if diagnosed with limited and extensive disease respectively. For those with NSCLC, MSTs ranged from 12.1 months for stage IV patients to 32.3 months for stage I disease.

Bild vergrößern

Recent data from GWASs of breast [21, 22], prostate [23‐27] and colorectal (CRC) cancer [28‐31] provides strong evidence for the involvement of common disease-causing alleles and suggests that a relatively large number of genes influence the aetiology in most cancers in the patient population as a whole.

To exploit the advances brought about by the human genome projects, future work in cancer genetics will be dependent upon the acquisition of large well-characterised cohorts of cancer cases. Here we have demonstrated that the centralisation of cancer services in the UK offers an opportunity to establish large, well-characterised cohorts by targeting collection to the largest centres. Moreover mobilising NCRN networks provides a means of delivering consistently the data and sample collection to complete genetic epidemiology studies, relating to the detection of main effects on the required scale.

Because ascertainment of cases through GELCAPS has been based on clinical centres specialising in the treatment of lung cancer a high proportion of cases have been diagnosed young. While this means cases are not fully representative of disease in the general population the distribution of age at diagnosis serves to empower GELCAPS for identifying disease-causing alleles by virtue of genetic enrichment.

Given that constitutional genotypes may well influence patient prognosis it is highly desirable that survivorship is not confounding influence on sample collection. As survival rates in patients recruited to GELCAPS were not significantly different to those documented in previously published audits of lung cancer in the UK there is no evidence that "healthy study participant" selection will have genetically biased ascertainment. For all participants, sex, ethnicity and age at sampling have been documented. The geographical area of birth and area of residence within the UK is known for all of the individuals and this information can be used to allow analyses stratified by region of residence, reducing any effects of population stratification. The possibility of population stratification leading to false inference of disease-genotype association can readily be addressed by adjusting for known region/ethnicity or by using information on unlinked genetic markers.

We acknowledge the potential problem of differential bias in genotyping samples accrued from different sources. Although the samples collected through GELCAPs have been ascertained from many clinical centres we have no evidence that this has affected sample quality as we have previously documented call rates of 99.8% in samples genotyped for 1,500 SNPs [32‐35] and Quantile-Quantile plots of test association statistics provide no evidence for differential bias

The NCRN research networks are established within cancer care networks where access to partners is readily available and direct. They are not designed to collect samples from the general population so our choice of collecting samples from partners was a pragmatic one appropriate for the NCRN. Inevitably in studies such as GELCAPS a smaller number of samples from controls will be collected than from cases since in addition to lack of compliance many patients do not have a current partner. The sex of controls ascertained through initiatives such as GELCAPS will usually be of the opposite gender to cases, and controls are potentially over-matched with respect to many lifestyle risk factors. Theses limitations can be offset to a large degree by using samples collected from the healthy spouses/partners of one cancer as a source of controls for a different cancer. This is something we are currently pursuing with respect to a similar NRCN sponsored initiative the National Study of Colorectal Cancer Genetics (NSCCG)

Because of the difficulty of obtaining sufficiently detailed data on environmental exposure in studies such as GELCAPS, and because there are issues to do with comparability of exposure data from controls assembled from different studies, it is acknowledged that studies of environmental risk factors including gene-environment interaction will be limited in resources such as GELCAPS. The main value of collections such as GELCAPS will be in studies of genetic risk factors and gene-gene interactions; hypotheses regarding gene-environment interaction require alternative datasets, such as the European Prospective Investigation into Cancer and nutrition (EPIC) study [36], which are centred around population based-cohorts. Accepting such limitations our experience in developing GELCAPS serves to illustrate how large DNA databases for genetic analyses can rapidly be developed in the UK. At present we have only collected outcome data on around 20% of cases recruited to GELCAPS. By completing the collection of follow up data on all cases we shall be able to assemble a unique series for examining the influence of constitutional genotype on clinical outcome in the population setting.

Finally, it is noteworthy that the value of GELCAPS has been demonstrated in a recent GWAS of lung cancer we have conducted in which we have been able to robustly identify a susceptibility variant for the disease mapping to 15q [37].

The authors declare that they have no competing interests.

Cancer Research UK: http://www.cancerresearchuk.org

GELCAPS: http://public.ukcrn.org.uk/Search/StudyDetail.aspx?StudyID=781

National Cancer Research Network: http://www.ncrn.org.uk

NHS Cancer Plan: http://www.dh.gov.uk/assetRoot/04/01/45/13/04014513.pdf

National Study of Colorectal Cancer Genetics: http://www.icr.ac.uk/research/research_sections/cancer_genetics/cancer_genetics_teams/molecular_and_population_genetics/nsccg/index.shtml

European Prospective Investigation into Cancer and Nutrition: http://epic.iarc.fr/