Lung cancer remains a leading cause of cancer death throughout the western world with five- year survival rates as low as 15%. While tobacco remains the leading cause of lung cancer, there is increasing recognition that other environmental, occupational and genetic factors including air pollution, asbestos, radon and silica are potential causes. Asbestos remains the second leading cause of lung cancer accounting for approximately 4-12% of all cases. Australian workers have been occupationally exposed to respirable asbestos fibres in construction, mining, manufacturing and other industries. In Australia, asbestos was mined in Western Australia (Wittenoom 1938-1966) and New South Wales (Baryulgil 1940-1979) as well as being imported from South Africa and Canada. Asbestos products have been used extensively in construction - approximately one in three Australian homes built prior to 1987 and most public buildings contain some form of asbestos. The long latency period from exposure to disease onset, means that we are now starting to see peak incidences, prompting the need for better diagnostic tools.
Asbestos-related lung cancer (ARLC) is hard to distinguish from other lung cancers on the basis of clinical and pathological criteria with most cases exposed to asbestos also exposed to tobacco. Persisting uncertainty surrounding the quantitative and biological interaction between asbestos and tobacco and lack of available biomarkers, makes attribution of the cause of lung cancer in people exposed to both asbestos and tobacco difficult for insurance groups, employer organizations, worker advocates and representatives, and compensation review boards. This doctoral thesis addresses the hypothesis that asbestos-related lung cancers will have distinct patterns of DNA damage, epigenetic regulation, gene expression and structural variation compared with cancers induced purely by tobacco. It is expected that integration of epigenetic, gene expression and structural information will provide a complete molecular profile of ARLC and that candidate biomarkers identified through this approach may provide potential targets for drug development.
The thesis aims were to:
I. Identify aberrant regions of copy number gain/loss and high level amplifications/deletions that discriminate asbestos-related and non-asbestos related lung cancers and verify these changes in an independent test set of cases.
II. Identify whether mutations in common oncogenes are associated with asbestos-related lung cancers using the OncoCarta Mutation Profiling platform.
III. To determine whether FGFR1 amplification is an early event in lung carcinogenesis by investigating copy number changes in pre-neoplastic lesions and primary lung cancers, then identifying whether it is associated with prior asbestos exposure.
IV. Identify methylation signatures capable of predicting lung cancer from normal lung by profiling differences in gene methylation, then testing the prediction capabilities in independent test sets of tumours.
V. To compare methylation profiles of asbestos-related and non-asbestos related lung tumours and identify methylated tumour suppressor genes that discriminate these two phenotypes.
VI. To verify the biological importance of six candidate asbestos-related lung adenocarcinoma (ARLC-AC) genes identified from prior supervised analysis of gene expression data comparing asbestos-related and non-asbestos related phenotypes, by testing expression in three independent test sets including ‘in-house’ and public datasets.
VII. To select candidate genes from supervised analysis of gene expression data comparing asbestos-related and non-asbestos related phenotypes in SCC samples based on observational criteria (P-value and fold change) and biological data. Then, to validate gene expression levels using an independent method (qRT-PCR) in the training set (original microarray samples) and an independent test set of phenotypically matched samples.
Copy number aberration (CNA; DNA gain/loss) for 63 lung cancers (18 ARLC, 45 NARLC) was assessed using ‘in-house’ copy number data from Agilent CGH arrays (G4410B) to determine whether asbestos specific changes were identifiable in lung. Significant regions of CNA were identified using the Genomic Identification of Significant Targets in Cancer (GISTIC) algorithm developed by the Broad Institute. A combined analysis of all 63 lung cancers identified no significant differences in aberration frequency between ARLC and non-asbestos related lung cancer (NARLC), with identified regions of CNA being similar in the two groups. Verification in an independent ARLC cohort failed to replicate the findings of The Prince Charles Hospital (TPCH) dataset finding minimal overlap between the two studies. Mutation analysis of 19 common oncogenes using Sequenom’s OncoCarta technology also failed to demonstrate differences in mutation frequency between ARLC and NARLC although the small sample size may have limited power to detect a difference.
Next, whole-genome methylation profiling was performed to identify somatic methylation changes in lung cancer compared to normal “non-malignant” lung, and in ARLC compared to NARLC to identify asbestos-specific methylation changes. Methylation profiling was performed by hybridising bisulfite-converted DNA from either tumour or normal tissue to 27K Illumina Infinium Methylation27 V1.0 microarrays, containing ~14,000 known genes. Class prediction modelling using BRB ArrayTools was used to develop signatures capable of predicting; (1) normal lung/lung cancer for identification of somatic DNA methylation changes, and (2) ARLC/NARLC. The predictive capability of each signature was assessed by predicting sample class in independent samples (test sets) using publicly available datasets where possible. In addition to this, a class comparison approach was utilised to identify potential candidates that could be targeted for therapy.
Finally, gene expression differences in subjects with and without prior asbestos exposure were assessed using Illumina’s 48K Beadchip Human_HT12 V3.0 for SCC and Operon’s 22K oligonucleotide chip for AC. Class comparison analyses identified ADAM28 as a potential oncogene involved in asbestos-related lung adenocarcinomas, with expression verified in three independent test sets of phenotypic relevance: 1) TPCH Test Set (n=58 lung cancers), 2) Wikman Lung tumour set (n=20 lung cancers) and 3) Nymark cell line dataset. In contrast, gene expression profiling of asbestos related lung squamous cell carcinomas identified MS4A1 as a potential candidate. However, immunohistochemical staining showed that expression of MS4A1 was primarily localised to stromal lymphocytes rather than tumour cells. Although this signal was predominantly from infiltrating lymphocytes it was a reproducible finding that alludes to a specific tumour-stroma interaction generated by asbestos.
In conclusion, this thesis has used a whole-genome based approach to identify genetic and epigenetic changes underlying the biology of ARLC. Copy number profiling failed to identify asbestos-specific regions of copy number gain and loss and reflected general regions of alteration in lung cancer. Gene expression profiling identified two genes, ADAM28 and MS4A1, with a plausible role in AC and SCC asbestos carcinogenicity respectively that were biologically validated in independent test sets of phenotypic relevance. Finally, DNA methylation profiling identified prediction panels capable of distinguishing normal lung/lung cancer and AC/SCC histology, however, DNA methylation panels were not able to predict ARLC or NARLC class with any confidence suggesting that methylation may not be a major regulatory process in ARLC.