Genome-wide epigenomic analyses of cell-free DNA from anaplastic lymphoma kinase-rearranged non-small cell lung cancer patients

Targeted therapies improve the prognosis of advanced anaplastic lymphoma kinase-driven non-small cell lung cancer (ALK+ NSCLC) patients. However, clinical courses vary considerably due to acquired drug resistance. Thus, timely detection of treatment failure is crucial to guide subsequent therapies and optimize patient outcome. The analysis of tumor alterations in cell-free DNA (cfDNA) represents a novel approach to monitor cancer dynamics during therapy in longitudinal plasma samples. Besides mutations and copy number alterations, cancer-specific epigenomic changes have emerged as promising biomarkers for cfDNA-based tumor assessment. This thesis aimed to identify tumor-derived methylation (5mC) and hydroxymethylation (5hmC) alterations in cfDNA of metastatic ALK+ NSCLC patients, associate these epigenetic biomarkers to gene expression in lung cancer and assess their suitability for monitoring of tyrosine kinase inhibitor therapy in serial plasma samples. To this end, 79 longitudinal plasma samples from 31 patients were collected alongside plasma of 14 healthy individuals. Genome-wide 5mC and 5hmC profiles were generated by cell-free methylation immunoprecipitation and 5hmC selective chemical labeling, followed by sequencing. Additionally, 5(h)mC profiles of primary monocytes, neutrophils and erythroid progenitor cells were prepared using the same methods. These hematopoietic cells constitute the major non-tumor contributors (72.2%) to cfDNA of cancer patients. A technical novelty of this study was the enrichment for tumor-derived 5(h)mC signals in cfDNA by excluding genomic regions highly (hydroxy-)methylated in the reference blood cell types. Of 9,603,454 300-bp genomic loci, 577,701 (5mC; 6.0%) and 499,681 (5hmC; 5.2%) exhibited low (or no) signal in the profiled blood cells. The blood cell signal-reduced 5mC regions demonstrated an increased correlation to lung cancer tissue methylation (Spearman, r = 0.26), compared to the entire dataset (r = 0.11), and revealed cancer- as well as tissue-specific 5mC signals. Cancer versus control analysis at the remaining genomic regions identified 5,499 differentially methylated (DMRs) and 495 differentially hydroxymethylated regions (DhMRs). Hierarchical clustering analysis based on the D(h)MRs cleanly separated patient from control samples and clustered patient cfDNA according to the inferred tumor burden within the samples. This suggests that sample separation is primarily driven by tumor-derived signals and confirms that the identified D(h)MRs are enriched for cancer 5(h)mC alterations. DMRs proximal to transcription start sites were enriched at genes downregulated in lung cancer tissue, demonstrating that cancer-specific gene regulatory 5mC marks can be retrieved from cfDNA. Many of these genes (e.g. GATA4 and HOXA9) were previously described to confer tumor suppressive functions in NSCLC. 5(h)mC levels in cfDNA correlated with tumor-derived genomic alterations (e.g. EML4-ALK fusion and global chromosomal instability [t-MAD score]) determined in matched plasma samples. The highest correlation was observed between PTGER4 methylation and t-MAD scores (Pearson, r = 0.86). Four 5mC (SOX9-AS1, HOXA10-AS, PRAC1, and PTGER4) and three 5hmC biomarkers (IL1RAP, KPNA7 and KIF25) were employed for therapy monitoring in ten patients with available longitudinal samples (³2). In particular 5mC biomarkers mirrored cancer dynamics found by radiologic imaging and genomic tumor alterations in cfDNA. At four instances, cfDNA 5mC levels anticipated therapy relapse in advance of imaging with a maximum lead time of 481 days. In one patient, both 5mC and 5hmC biomarkers detected disease progression ahead of imaging and genomic alterations in cfDNA, highlighting the sensitivity of 5(h)mC-based tumor assessment. In conclusion, 5mC and 5hmC profiling from cfDNA provides an opportunity for sensitive cancer detection and therapy monitoring. The tissue-specificity and the regulatory functions of these DNA modifications provide data about the tumors that currently cannot be obtained by copy number or single nucleotide variation profiling.