A Major Lung Cancer Susceptibility Locus Maps to Chromosome 6q23–25

Introduction

There are more individuals who die from lung cancer than from breast, colon, and prostate cancer combined, with an estimated 173,700 new cases and 160,440 deaths expected in the United States in 2004 (Jemal et al. 2004). The overall 5-year survival rate of lung cancer is 15% (Jemal et al. 2004). Advances in the detection and treatment of this disease have only resulted in a marginal improvement in mortality rates. In the last 50 years, lung cancer incidence increased by 249%, and mortality increased by 259% (Welch et al. 2000).

Cancer of the lung has frequently been cited as an example of a malignancy that is solely determined by the environment (Doll and Peto 1981; Blot and Fraumeni 1996). The risks associated with cigarette smoking (Burch 1980; Doll and Peto 1981; Carbone 1992; Doll et al. 1994; Blot and Fraumeni 1996) and certain occupations, such as mining (Seaton 1984), shipbuilding, and petroleum refining (Blot and Fraumeni 1976; Blot et al. 1979; Gottlieb and Steadman 1979) are well established. There is little doubt that the majority of lung cancer cases are attributable to cigarette smoking and other behavioral and environmental risk factors (Beckett 1993). However, numerous studies also suggest the involvement of genetic risk factors.

Investigators have long hypothesized that individuals differ in their susceptibility to environmental insults (Motulsky 1957; Heath 1958; Friberg 1959). More than 40 years ago, Tokuhata and Lilienfeld (1963) provided epidemiologic evidence for familial aggregation of lung cancer after accounting for personal smoking, which suggested the possible interaction of genes, shared environment, and common lifestyle factors in the etiology of lung cancer. Fraumeni et al. (1975) reported an increased risk of lung cancer mortality in siblings of probands with lung cancer, and positive family history has consistently been found to be a risk factor for lung cancer in a number of case-control studies (reviewed in Bailey-Wilson et al. 1998; Sellers and Bailey-Wilson 1998; Etzel et al. 2003).

Genetic modeling studies have also suggested that familial aggregation of lung cancer may be due to inheritance of only a few genetic factors. Segregation analyses of Louisiana families gave evidence for inheritance of a rare major autosomal gene that acts in conjunction with cigarette smoking to produce an earlier age at onset of the cancer (Sellers et al. 1990, 1991, 1992). Under this model, heterozygotes with the susceptibility allele and with average levels of cigarette smoking had relative risks of 14, 11.8, and 6.2 at age 50 years, 60 years, and 70 years, respectively, compared with noncarriers who had average levels of smoking. Additional segregation analyses of these data examining cohort and polygenic effects continued to indicate the effects of an allele on inherited susceptibility to lung cancer and also on risk for a broader group of smoking-related cancers (lung, oral cavity, esophagus, nasopharynx, larynx, pancreas, bladder, kidney, and uterine cervix) after smoking was allowed for (Chen et al. 1991; Bailey-Wilson et al. 1992). Gauderman et al. (1997) applied a Gibbs sampling method to the same lung cancer data set and found evidence for a dominant major locus with significant effects of smoking, but no evidence of gene-environment statistical interaction on the logistic scale.

Similar findings were reported by Yang et al. (1999), who noted evidence for Mendelian codominant inheritance with modifying effects of smoking and chronic bronchitis, using families of nonsmoking cases diagnosed at ages 40–59 years. Daw et al. (in press) performed oligogenic segregation analysis of time to onset of lung cancer on >2,000 families. The results of their work indicate the likely segregation of risk-conferring alleles at three or four genetic loci. The low number of estimated loci strongly suggests that genetic linkage studies could identify genetic factors for familial lung cancer (FLC).

On the basis of the evidence from these studies of the existence of cancer susceptibility genes that may act in conjunction with cigarette smoking to increase risk of lung and other throat cancers, we performed a genomewide-scan linkage study in families selected for aggregation of lung cancer. We considered lung, laryngeal (ICD-9 161.0–161.9), oropharyngeal (ICD-9 146.3–146.9), or hypopharyngeal (ICD 148.0–148.9) cancer, henceforth grouped together under the term “lung and throat (LT) cancer,” to be pleiotropic effects of the same gene. The high case-fatality rate (15% 5-year survival) and low resection rate (25%) makes the study of families with lung cancer particularly challenging, because it is difficult to collect adequate numbers of biospecimens for DNA analysis. A multidisciplinary collaborative effort was necessary to identify and accrue large numbers of families with FLC to test the hypothesis that there are genetic variants that greatly increase the risk of developing lung cancer. Recognizing the complexity of modeling the impact of environmental factors for this complex disease, we made an a priori decision that a simple autosomal dominant affected-only model would constitute our primary analytic strategy. Further analyses in which we also evaluated the evidence for linkage, allowing for the effects that smoking behavior, age, and sex have on the risk for lung cancer, were performed and are described.

Material and Methods

Results

Table 1 summarizes characteristics of the 52 pedigrees. There were 223 persons affected with lung cancer; 13 persons affected with laryngeal (ICD-9 161.0–161.9), oropharyngeal (ICD-9 146.3–146.9), or hypopharyngeal (ICD 148.0–148.9) cancer; and 5 persons affected with both primary lung cancer and one of the other three primary cancers. The median number of affected persons per family was four. There were 36 families with affected persons in two or more generations. Six pedigrees contained affected family members in three generations, and one other pedigree contained affected individuals in four generations. Many families also contained affected persons with onset of lung cancer at an early age. The median age at onset for lung, laryngeal, and pharyngeal cancer was 60 years, less than the median value of 70 years for age at diagnosis in the general white population (Ries et al. 2000). Minimum age at onset within the family may be a better indicator of the potential for existence of a susceptibility gene, since it is not influenced by older sporadic cases. In our families, the minimum age at onset within the family ranged from 29 years to 68 years with a median value of 50.5 years.

For the 241 affected individuals in the 52 families with FLC (table 1), 86 blood samples were available for global genotyping. Of 149 individuals in the 52 families who were deceased at the time of pedigree development, we were able to globally genotype 50 (34%). DNA from archival tissue specimens was used to globally genotype 11 affected family members. We reconstructed the genotypes of 39 persons, using genotypes of spouses and offspring, including genotypes from archival tissue of five spouses. Genotyping was also performed for markers on chromosomes 6 and 12 for five deceased affected persons whose archival specimens yielded limited DNA. Without the ability to globally genotype archival specimens, we would not have genotypes from 15% of the affected family members needed for linkage analysis.

Figure 1 shows the maximum (recombination fractions of ⩽40%) two-point homogeneity LOD scores across the genome, under the simple dominant low-penetrance model without inclusion of age and smoking exposure. Positive HLOD scores that approached or exceeded 1.0 were calculated for markers on chromosomes 1, 4, 6, 9, 12, 20, and 21. Two-point analysis gave an HLOD of 0.94 at D6S2436 in all families and an HLOD of 1.5 in the 38 families with four affected relatives, whereas for the 23 highest risk families, the two-point HLOD score was 2.1. Figure 2 shows that inclusion of age and smoking in the LODLINK models gives maximum two-point HLOD scores of 1.48 on chromosome 6 at C6S1848 in all families, but, for marker D6S2436, the LODLINK results including smoking and sex effects yielded lower HLOD scores than the simple model did. On chromosome 12, LODLINK yielded two-point HLODs of 1.54 at marker D12S372, 1.1 at D12S375, and 0.4 at D12S2070. The HLOD scores that were obtained using the model that incorporated the covariates were higher in some regions than the maximum two-point HLODs that were obtained under the simple dominant low-penetrance affected-only model with no covariates.

Multipoint parametric linkage under the simple dominant low-penetrance affected-only model yielded a maximum HLOD of 2.79 at 155 cM (marker D6S2436) on chromosome 6q in the 52 families (fig. 2), with 67% of families estimated to be linked. Multipoint analysis of the 38 families with four affected relatives gives an HLOD of 3.47 at this same location, with 78% of families estimated to be linked, whereas for the 23 highest risk families, the multipoint HLOD score is 4.26, with 94% estimated to be linked (fig. 3 and table 2). Conversely, linkage analysis of the 14 families with three affected relatives yielded negative LOD scores. A predivided samples test for heterogeneity (Ott 1999) that compared the evidence for linkage in multigenerational families with five affected relatives (23 families) to the other familes was significant (P=.007). The 1-HLOD support interval in the 23 multigenerational families extends from C6S1848 at 146 cM to 164 cM near marker D6S1035.

The estimate of the proportion of linked families from heterogeneity linkage analysis is known to be imprecise and may be inaccurate in studying complex traits, although the test for linkage in the presence of heterogeneity is robust and powerful (see, e.g., Greenberg and Abreu 2001; Whittemore and Halpern 2001; Hodge et al. 2002; Vieland and Logue 2002). However, the relative increase in the estimated proportion of linked families in subsets of families with an increasing number of affected individuals suggests a decrease in heterogeneity in the more informative families. In the subset of 23 highly informative families, 14 (61%) have LOD scores >0.0 at D6S2436, 11 (48%) have LOD scores >0.2, and 8 (35%) have LOD scores >0.4. These data demonstrate that more than half of these 23 pedigrees contribute toward the positive LOD score on 6q.

Further multipoint analysis that includes smoking behavior and that uses a liability class for each subject yielded somewhat lower HLOD scores. The maximal HLOD score from analysis of all 52 families was 2.63, with the HLOD score of 3.04 for the white families. For the 38 families with ⩾4 cases, the HLOD score was 1.96, and for the 23 multigenerational pedigrees, the HLOD score was 2.90.

Multipoint linkage analysis under the simple autosomal dominant decreased-penetrance model yielded HLODs close to 1.0 on chromosomes 12, 14, and 20 (table 2). Among the 19 families that showed positive LOD scores on chromosome 6, we also found positive multipoint LOD scores on chromosome 12, with a maximum HLOD score of 1.6 at D12S2070, and 72% of families were estimated to show linkage.

Nonparametric multipoint analyses of LT cancer with the use of the variance-components approach (binary option) and the mixed-effects Cox models (with multipoint IBD sharing calculated by SIMWALK2) also gave support for linkage of LT cancer in the 6q region at C6S1848 at 146 cM (LOD=2.46 and 2.01, respectively) in the 23 highest risk families with sex and smoking status (yes/no) as covariates (fig. 4). On chromosome 12, these nonparametric methods also gave evidence of linkage for LT cancer (LOD=0.36 [coxme] and 0.63 [binary]) in the 23 highest risk families with sex and smoking status (yes/no) as covariates.

The results of our multipoint analyses indicated much stronger evidence for linkage from multipoint analysis, compared with the two-point analysis. To evaluate the impact that sparseness in availability of subjects in the extended pedigrees has on LOD score calculations, we used a simulation approach implemented with SLINK (Ott 1989; Weeks et al. 1990). We simulated data for the 23 multigenerational families, assuming penetrance of 40% in carriers and 4% in noncarriers, and then analyzed the data, assuming penetrance of 10% in carriers and 1% in noncarriers. We then performed linkage analysis of the disease susceptibility and 1 or 2 markers (the order was D6S2436–0.02 cM–disease–10 cM–D6S1035, and we used D6S2436 alone for two-point analyses). We also simulated data under the assumption that we could only obtain samples from those individuals who actually gave samples (sparse case), versus the situation in which all individuals give samples (dense case). In sparse pedigrees, the increase in LOD scores from two-point to three-point analyses was 34%, whereas, for dense pedigrees, it was only 19%. We anticipate that simulations using C6S1848 might further indicate a gain in LOD scores from multipoint analysis, but four-point analysis using SLINK was prohibitively time consuming. These studies suggest that, for sparse pedigrees, multipoint analysis is relatively more informative than it is for dense pedigrees with no missing genotypes of affected persons.

To explore the effect that smoking has on risk for lung cancer among carriers of the susceptibility haplotype, we first obtained from SIMWALK2 the most likely haplotypes from 21 of the 23 multigenerational pedigrees (in 2 pedigrees, we were unable to clearly identify a susceptibility haplotype). Then, we scored all individuals as either carriers (n=223, of which 87 were affected, and 90.4% of the affected individuals smoked) or noncarriers (n=344, of which 19 were affected, and 83% of affected individuals smoked). We then performed a Cox proportional hazards survival analysis using SAS 8.0 to assess the effects of smoking and sex separately among carriers and noncarriers of the susceptibility haplotype, treating time to onset of LT cancer as an endpoint and sex and cigarette pack-years as predictors. The results of this analysis showed a mild effect of increasing smoking on risk for lung cancer among carriers (hazards ratio [HR] per pack-year = 1.004, P=.13), whereas, among noncarriers, there was a significant effect of smoking on risk for lung cancer (HR per pack-year = 1.018, P=.0023). When smoking was treated as a categorical yes/no variable, we found that smoking increased risk for both carriers (HR = 3.31, P=.002) and noncarriers (HR = 2.95, P=.10). These sets of observations suggest that smoking at any level increases risk in carriers of inherited susceptibility from a locus on chromosome 6q, whereas increasing smoking increases risk for noncarriers.

Discussion

These results provide clear evidence for a major susceptibility locus on chromosome 6q influencing LT cancer risk—particularly in the multigenerational densely affected families—with characteristics consistent with an autosomal dominant or codominant major locus. As expected, the multipoint analyses that test for linkage in the presence of heterogeneity (multipoint HLODs) gave the most power in our analyses. In addition to the compelling evidence for linkage to chromosome 6q, we also found suggestive evidence for linkages to some other regions (Lander and Kruglyak 1995). There is also some indication of a possible epistatic interaction between the putative loci on chromosomes 6 and 12, but more data are required to confirm this interaction.

The 1-HLOD support interval on 6q obtained from the multipoint analysis of the multigenerational families extends from C6S1848 at 146 cM (147.95 Mb) to 164 cM (159.94 Mb) near D6S1035. We have also provided the location of the markers on 6q in million base pairs (Mb), since the chromosomal regions of allelic loss are reported in the literature in Mb (see fig. 5 legend). The interval supporting linkage overlaps a genomic region on 6q that exhibits allelic loss in non–small-cell lung carcinoma (Petersen et al. 1997; Virmani et al. 1998; Luk et al. 2001; Goeze et al. 2002). These studies used cytogenetic techniques of comparative genomic hybridization (CGH) and/or fluorescent in situ hybridization to detect large regions of chromosomal imbalance on 6q. Berrieman et al. (2004) detected allelic loss between 6q25-qter for 50% of the samples examined, and Goeze et al. (2002) detected deletion between 6q14–24 in at least 60% of the 59 primary lung tumors examined. The study by Petersen et al. (1997) examined 50 non–small-cell lung carcinomas by CGH, and 46% exhibited chromosomal imbalance in the interval supporting linkage to 6q. Luk et al. (2001) examined 23 lung tumors and detected 6q loss in 30% of the tumors. There are no published studies that examined lung tumors to define a minimum region of deletion on 6q by loss-of-heterozygosity (LOH) analysis. However, two studies did detect LOH with several markers on 6q (located between 152 Mb and 167 Mb) in 50% of the lung tumor/lung cancer cell lines examined (Merlo et al. 1994; Virmani et al. 1998). These analyses of sporadic non–small-cell lung carcinomas clearly demonstrate frequent allelic loss on regions of 6q that overlap our linkage interval.

Numerous other tumor types also have allelic loss on regions of 6q which overlap the interval supporting linkage, including breast tumor (Noviello et al. 1996; Utada et al. 2000; Cesari et al. 2003; Zeller et al. 2003), ovarian tumor (Hansen et al. 2002; Cesari et al. 2003), mesothelioma (Jensen et al. 2003), pancreatic tumor (Abe et al. 1999; Barghorn et al. 2001), squamous cell carcinoma (SCC) of the oral cavity (Tong et al. 2004), melanoma (Millikin et al. 1991), and Hodgkin lymphoma (Re et al. 2003). These studies utilized LOH analyses to determine regions of minimum chromosomal deletion. Figure 5 illustrates the overlap of these minimum regions of deletion within our interval supporting linkage on 6q. The horizontal lines denote regions of deletion, and the numbers above each line denote the reference from which the data were obtained (see fig. 5 legend). Since some studies identified two or more distinct minimum regions of deletion, a number may appear above more than one horizontal line. The majority of the detected regions of allelic loss on 6q are located in our linkage region (fig. 5), although deletions are observed in two other locations. For example, allelic loss has been detected in prostate tumors, non-Hodgkin lymphomas, and acute lymphoblastic leukemias in a region proximal (90–105 Mb) to our linkage region (Merup et al. 1998; Srikantan et al. 1999; Zhang et al. 2000; Hyytinen et al. 2002; Verhagen et al. 2002; Konishi et al. 2003). Tumor-suppressor genes may also reside on 6q in a region telomeric to our linkage region (Tibiletti et al. 2000).

There is considerable overlap between these regions of minimum deletion shown in figure 5, including the following: (1) overlap of regions denoted by horizontal lines 1, 2, 4, 6, and 8, with the minimum region of overlap around markers D6S305 and D6S1599; (2) overlap of lines 12, 11, 5, 3, and 1, with the minimum region of overlap between D6S978 and D6S1637; (3) overlap of lines 11, 9, 7, 3, and 1, with the minimum region of overlap between D6S1648 and D6S1055; and (4) overlap of lines 11, 10, 8, 7, and 1, with the minimum region of overlap between D6S270 and D6S308. Also, some of the overlap occurs between minimum regions of deletion detected in different tumor types (e.g., horizontal lines 1 and 3 depict regions of deletion detected from breast tumors, line 5 from ovarian tumors, line 11 from melanomas, and line 12 from Hodgkin lymphomas). These data are consistent with the existence of one or more tumor-suppressor genes in the linkage region that we have identified for LT cancer in this study. There are several interesting candidate genes in this region, including four putative tumor-suppressor genes: SASH1 (148.8 Mb), LATS1 (150 Mb), IGF2R (160.45 Mb), and PARK2 (161.7 Mb). Also, other genes in this region are involved in some aspects of the regulation of cellular proliferation or the prevention of DNA damage. The sequencing of exons in these candidate genes is underway in families that show evidence of linkage of lung, laryngeal, oropharyngeal, or hypopharyngeal cancer to 6q markers.

We have detected compelling evidence of linkage on 6q using a simple dominant low-penetrance affected-only model. This linkage model was chosen as our primary analytical approach because of uncertainty about the strength of the relationship between smoking behavior and lung cancer risk in the high-risk families we are studying. As indicated in our prior studies of Li-Fraumeni syndrome (Hwang et al. 2003), smoking could have a much less important role in the context of genetic susceptibility than in the general population, and parametric models that rely heavily on smoking behavior might lead to a deflation of LOD scores for detection of linkage. Furthermore, possible deflation of LOD scores could occur if there is a common genetic susceptibility between smoking behavior and LT cancers. In addition, since ∼90% of the affected family members in our studies smoked, a weighting of only the affected individuals in this low-penetrance model has the effect of allowing for smoking status while ignoring information from unaffected subjects. The nonzero sporatic rate (1%) allows for individuals who are not gene carriers to be affected as a result solely of their smoking exposure, and the relatively low penetrance (10%) for gene carriers allows nonsmoking unaffected persons to have a high probability of being a gene carrier. Finally, simple models such as this one are known to recover a large portion of the linkage information available for many types of complex traits, as long as the mode of inheritance (dominance) at the trait locus is correctly specified (Hodge et al. 1997; Greenberg et al. 1998; Abreu et al. 1999, 2002; Durner et al. 1999; Greenberg and Abreu 2001).

The detection of significant linkage on 6q by the simple model, without explicitly including smoking history as a covariate in the model, does not imply that smoking exposure is an unimportant risk factor for lung cancer in family members who are carriers of the genetic risk factor but, rather, it implies that the genetic factor on 6q is strong enough to be detected by the low-penetrance affected-only linkage model with our pedigrees. Smoking exposure is also an important risk factor, since 90% of the affected individuals in the 52 pedigrees were smokers, and 89% of individuals in the 23 multigenerational families with 5 or more affected members smoked. The Cox modeling that we performed suggested that the inferred carriers of a 6q mutation may be sensitive to any level of smoking, rather than suggesting an increasing gradient of risk with increasing smoking behavior, as is usually observed in the general population. In the segregation analysis approach used by Sellers et al. (1990), the difference in the effect of smoking behavior on risk could not be effectively modeled because of the difficulties in fitting a covariate by unmeasured genotype interactions.

Because it is possible that better modeling of the joint effect of genetic factors and smoking exposure might improve power to detect linkage, we also included smoking behavior in two-point and multipoint parametric analyses and in multipoint nonparametric linkage analyses. Our two-point parametric analyses with the use of LODLINK gave stronger evidence of linkage to some parts of the 6q region than did the simple model in two-point analyses (fig. 2). However, the results from the multipoint analysis showed decreased (but still strong) evidence for linkage when smoking behavior was included in the analysis. It is possible that the model previously derived from segregation analysis of a population-based case series of lung cancer is not appropriate for the highly selected families we are studying in this linkage study, as discussed above. If the Sellers et al. (1990) model inaccurately models the differential effect of smoking behavior on carriers and noncarriers, then the use of it could yield lower evidence for linkage, particularly in the multipoint analysis, which is known to be more sensitive to model misspecification than the two-point linkage analysis (Risch and Giuffra 1992). Work by both Dizier et al. (1993) and Durner et al. (1999) has shown that the use of models that are the result of population-based segregation analysis in the linkage analysis of a very complex trait can reduce or conceal evidence of linkage.

Nonparametric multipoint analysis incorporating age and smoking status as covariates and using the variance-components and mixed-effects Cox models also support the same genomic region found by the parametric models. An advantage of the nonparametric linkage analysis is nonspecification of a model. Thus, when similar evidence for linkage is obtained with nonparametric analysis, then the evidence of linkage from the parametric model is substantiated. It is also known that multipoint parametric analysis can be more powerful than the nonparametric approach, especially under dominant inheritance (Durner et al. 1999; Greenberg and Abreu 2001). In future studies, and particularly once the gene predisposing to lung cancer in these families is identified, an evaluation of the effect that smoking has on lung cancer risk will be of great value in understanding carcinogenesis for carriers.

In the study of complex traits such as LT cancer, genetic heterogeneity is expected and has been observed for other cancers with hereditary components, such as breast cancer, colorectal cancer, malignant melanoma, and prostate cancer. Our results are consistent with this expectation. In other complex diseases, subdivision by clinical characteristics, such as age at onset and pattern of inheritance in the pedigrees, has often resulted in reduction of heterogeneity. As we subdivided our pedigrees into those that were more consistent with a strong genetic susceptibility (38 families with four or more affected relatives) and further subdivided them into those that were most consistent with an autosomal dominant susceptibility locus (23 families with five or more affected individuals in two or more generations), the evidence for linkage at 6q increased substantially, and the estimate of the proportion of linked families increased, approaching 1.0 for the subset of 23 multigenerational pedigrees. In fact, in this subset, the homogeneity LOD score and the HLOD were virtually identical, suggesting that these clinical familial characteristics had substantially reduced the heterogeneity.

Elsewhere, Hodge et al. (1997), Greenberg and Abreu (2001), and Abreu et al. (2002) have shown that there is some inflation of the type I error rate (false-positive rate) when one uses multiple penetrance models and when one computes HLOD scores as opposed to LOD scores. These authors have shown that the critical LOD threshold of 3.0 should be increased by 0.3 for maximizing over penetrance only (Hodge et al. 1997), by 0.47 when two-point HLODs are used (Abreu et al. 2002), and by 0.7 when multipoint HLODs are used (Greenberg and Abreu 2001). Thus, these authors suggest that adding 1 LOD unit to the significance threshold of LOD=3.0 (corresponding to a P value of .0001), which results in a corresponding critical threshold of HLOD=4.0, is a somewhat conservative adjustment for an analysis in which multipoint HLODs are calculated and multiple penetrance models are used. Our observed multipoint HLOD of 4.26 in the multigenerational families would satisfy these criterion.

We intend to follow up the results of this study by fine mapping the most significant regions. We also will attempt to replicate these linkage findings in an independent set of families with FLC. As a result of limited biospecimen availability, effective performance of a linkage study for a rapidly fatal disease such as lung cancer is difficult. However, the ability to globally genotype archival specimens greatly improved the power of our linkage study. In other cancers, discovery of susceptibility genes has led to greater understanding of the biological processes that cause these diseases and ultimately will lead to better methods for prevention and treatment. Likewise, we believe that discovery of lung cancer susceptibility genes will also be important in improving our understanding of this devastating disease.