Abstract

Lung adenocarcinoma and squamous cell carcinoma are the deadliest cancers worldwide. Smoking and ionizing radiation are potent carcinogens affecting strongly both lung cancer subtypes. Several biological analyses have been performed to characterise the genetic mutations leading to adenocarcinoma and squamous cell carcinoma, and different genomic spectra have been observed. Biological markers of smoking related damage could be found, leading to a deep knowledge of cellular smoking effects. Less is known about the biological effects of radiation in human carcinogenesis. Risks have been quantified with epidemiological studies of these carcinogens. Based on the biologically substantiated assumption that the number of mutations is linearly related to the dose, in radiation epidemiology it is standard to model effects linearly. These models do however not have a biological interpretation and are disconnected from general statistical methods. Here we fill both gaps. First we apply statistical generalised additive models to examine the functional relation between risk and smoking and radiation effects. Secondly, with mechanistic multi-scale models we integrate molecular biology and epidemiology to describe the carcinogenesis of lung adenocarcinoma and squamous cell carcinoma. To investigate the incidence of lung adenocarcinoma and lung squamous cell carcinoma we analysed two cohorts: first the Life Span Study cohort of atomic bomb survivors of Hiroshima and Nagasaki, and second the Eldorado cohort of Canadian Uranium miners. Exposures differed strongly between cohorts. Residents of Hiroshima and Nagasaki were exposed to a relative high dose of gamma radiation for a short time, while the miners were exposed to a protracted and lower exposure to alpha and gamma radiation. Information about smoking habits is available only for the former cohort. Three types of models were applied to analyse the effects of radiation and smoking: state-of-the-art statistical risk models of radiation protection, statistical generalized additive models and mechanistic risk models. Although there were quantitative differences in effect size and significance, each result is presented below only for a single model. For lung adenocarcinoma the best mechanistic model was a two pathway model. Smoking and radiation effects showed markedly different patterns: both acted on the apoptosis rate of precancerous cells but on different pathways without any interaction. A linear radiation effect was found in one pathway and a linear-exponential smoking effect in the other pathway. Independently of these results we analysed genomic data of American patients. It is known that the genetic damage of people with adenocarcinoma can be grouped into three pathways: the receptor mutant (RMUT ) pathway, the transducer mutant pathway (TMUT ), and other signatures (OWT ). We could show that signatures of TMUT and the OWT pathways do differ much less from each other than both differed to the RMUT pathway. Therefore, there is also genetic evidence that adenocarcinoma fall into two main classes. The two pathways of the mechanistic model could be associated to the RMUT and RMUT+OWT pathways by their risk patterns in age and smoking. On the other hand, for squamous cell carcinoma one pathway was sufficient to describe the incidence data. Although effects of radiation appeared to be highly significant, they could be traced back to arise only from the first five years of follow up (33 cases therein). When the first five years were excluded, no significant radiation effect could be found. Interestingly, for lung squamous cell carcinoma the mechanistic models could fit the effects of cigarette smoking in initiation and promotion. This was different for lung adenocarcinoma, where the main effect of smoking was a promotion of already existing pre-cancerous clones. For both, lung adenocarcinoma and squamous cell carcinoma, no interaction between radiation and smoking could be fitted for the Life Span Study cohort. Results from analysis of the Eldorado cohort were in line with the results presented above. For lung adenocarcinoma both, the state-of-the-art statistical risk models and the generalised additive models, could find only a significant effect of radiation exposure. For lung squamous cell carcinoma, vice versa, both models could find only a significant effect of gamma radiation exposure. Concluding, we showed that lung cancer cannot be investigated as a single endpoint but the different subtypes have to be analysed separately. Different radiation qualities act differently to the different subtypes, indicating different biological processes. Analogously, although smoking is an important risk factor for all subtypes, its effects were different and with different magnitudes.