GERMLINE VARIANT CALLING IN FORMALIN-FIXEDPARAFFIN-EMBEDDED TUMOURSbyShyong Quin YapB.Sc. (Hons), Trent University, 2011A THESIS SUBMITTED IN PARTIAL FULFILLMENTOF THE REQUIREMENTS FOR THE DEGREE OFMASTER OF SCIENCEinTHE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES(Experimental Medicine)The University of British Columbia(Vancouver)December 2017c© Shyong Quin Yap, 2017AbstractGermline alterations can have clinical implications for both cancer patients and their families. Be-cause the tumour genome may contain both germline and somatic variants, the increasingly com-mon practice of clinical tumour sequencing presents an opportunity to also pre-screen for germlinevariants. This framework is time- and cost-effective because only patients with potential germlinevariants are referred to downstream confirmatory testing. However, a key challenge is that tumourspecimens are commonly formalin-fixed and paraffin-embedded (FFPE), which induces DNA dam-age that may interfere with molecular testing. Another challenge is distinguishing between germlineand somatic variants in the tumour in order to accurately select candidates for follow-up screening.In order to leverage tumour sequencing for identifying germline variants, these challenges must beaddressed.To this end, we retrospectively analyzed clinical amplicon sequencing data from 213 patientswith a range of tumours, for whom matched-normal samples were available. We assessed formalin-induced DNA damage by comparing amplicon enrichment and sequencing results of FFPE DNAto the matched-normal DNA isolated from peripheral blood mononuclear cells, a gold standardfor germline testing. Although formalin-induced DNA fragmentation and cytosine deaminationwere detectable, we determined that the discrepancies were minor and could be mitigated by usingshorter amplicons and enriching for longer DNA templates. We also found that 98.0% of germlinealterations identified in the blood were retained in the tumours, suggesting that FFPE tumour DNAcan be a reliable source for germline variant calling. Finally, we applied variant allele frequency(VAF) thresholds to delineate germline and somatic variants in tumour-only analyses. We reportedthat a VAF cut-off of 15% would correctly identify 99% of germline alterations in FFPE tumours,but erroneously submit 14% of somatic mutations (false positives) for follow-up germline testing.This underscores the high sensitivity and positive predictive value of using VAF to discriminatebetween germline and somatic variants. Collectively, our results demonstrate that clinical tumouramplicon sequencing could also be used to provide cost-efficient first-line germline testing.iiLay SummaryHereditary genetic changes have clinical impacts on cancer patients and their families. Tumourscontain tumour-specific and inherited genetic variations. Using tumour DNA for pre-screening ofhereditary variants is time- and cost-saving because only patients with potential hereditary variantsrequire follow-up. Follow-up testing involves analyzing blood or saliva to confirm the presence ofthe potential hereditary variants before making clinical decisions. A key challenge in implement-ing this approach is differentiating between tumour-specific and inherited variants in the tumourDNA. Furthermore, the commonly-used tumour fixative, formalin, induces DNA damage, whichinterferes with using tumour DNA for genetic testing. We showed that the effects of formalin ontumour DNA were minor, and we established a highly sensitive and precise method for separatinghereditary variants from tumour-specific mutations. Our findings imply that extracting hereditaryinformation from tumour DNA analysis could serve as a practical, cost-effective approach to pro-viding hereditary genetic testing in the clinic.iiiPrefaceThis dissertation is based on next-generation sequencing data from The OncoPanel Pilot (TOP)study. The TOP study was designed to optimize and validate a clinical targeted NGS panel for itsutility as standard of care. Approval of this pilot study was covered by the Human Research EthicsProtocol H14-01212. Sample preparation and sequencing, as well as processing of raw data, readalignment, and variant calling were collaboratively performed by members of Canada’s MichaelSmith Genome Sciences Centre and the Centre for Clinical Genomics. Data analyses in Chapter 3and 4 are my original work.Analysis of pharmacogenomic genes in Chapter 3 and 4 was presented as a poster titled “Com-parative Analysis of Pharmacogenomic Variants in Amplicon-sequenced DNA from PeripheralBlood and Formalin-fixed Paraffin-embedded Tumours” at the 2016 Intelligent Systems for Molecu-lar Biology conference and the Translational Medicine (TransMed) Special Interest Group meeting.ivTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiLay Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xList of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiiAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 The emergence of precision oncology . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Overview of next-generation sequencing technologies . . . . . . . . . . . . . . . . 31.2.1 Illumina sequencing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51.2.2 Clinical applications of NGS . . . . . . . . . . . . . . . . . . . . . . . . . 71.3 Variant analysis pipeline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101.3.1 Quality control and pre-processing of raw sequencing reads . . . . . . . . 111.3.2 Read alignment and post-alignment processing . . . . . . . . . . . . . . . 121.3.3 Variant calling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121.3.4 Variant annotation and interpretation . . . . . . . . . . . . . . . . . . . . . 141.4 ACCE model process for evaluating genetic tests . . . . . . . . . . . . . . . . . . 151.5 Clinical implications of germline alterations in cancer . . . . . . . . . . . . . . . . 161.5.1 Cancer predisposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171.5.2 Pharmacogenomics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17v1.6 Technical challenges in implementing germline testing in clinical oncology . . . . 201.6.1 Tumour-only sequencing . . . . . . . . . . . . . . . . . . . . . . . . . . . 201.6.2 Formalin-fixed paraffin-embedded tumours . . . . . . . . . . . . . . . . . 221.7 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242.1 Overview of study design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242.2 Patient samples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252.3 Sample preparation, library construction, and Illumina sequencing . . . . . . . . . 262.4 OncoPanel (Amplicon-based targeted sequencing panel for solid tumours) . . . . . 272.5 Variant calling pipeline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282.5.1 Read alignment and variant calling . . . . . . . . . . . . . . . . . . . . . . 282.5.2 Variant filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292.5.3 Variant annotation and interpretation . . . . . . . . . . . . . . . . . . . . . 302.6 Sequence analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332.7 Application of VAF thresholds to separate germline alterations from somatic mutations 333 Results: Assessment of Formalin-Induced DNA Damage in FFPE Specimens . . . . 353.1 Comparison of efficiency in amplicon enrichment and sequencing results betweenblood and FFPE specimens . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353.2 Reduced coverage depth in FFPE specimens is more pronounced for longer amplicons 423.3 Deamination effects lead to increased C>T/G>A transitions in FFPE specimens . 483.4 Increased age of paraffin block results in reduced amplicon yield and elevated levelof C>T/G>A sequence artifacts . . . . . . . . . . . . . . . . . . . . . . . . . . . 564 Results: Identification of Germline Alterations in FFPE Tumours . . . . . . . . . . 604.1 Frequency and interpretation of germline alterations in patients from the TOP cohort 604.2 Germline alterations are highly concordant between blood and FFPE specimens . . 894.3 Application of VAF thresholds to separate germline alterations from somatic muta-tions in tumour-only analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 955 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 995.1 Formalin-induced DNA damage has minor effects on sequencing metrics . . . . . . 1005.2 Sequence artifacts induced by cytosine deamination tend to occur at low allelic fre-quency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1005.3 Sequence artifacts other than those caused by cytosine deamination are detected . . 1015.4 Storage time of FFPE blocks correlates with the extent of formalin-induced DNAdamage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102vi5.5 Germline variants are highly retained in the tumour genome . . . . . . . . . . . . 1025.6 The use of VAF thresholds is feasible for distinguishing between germline and so-matic alterations in tumour-only analyses . . . . . . . . . . . . . . . . . . . . . . 1035.7 Limitations and future directions . . . . . . . . . . . . . . . . . . . . . . . . . . . 104Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107A Supporting Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126viiList of TablesTable 1.1 Different NGS platforms by read length, chemistry, and detection method. . . . 4Table 2.1 Distribution of cancer types in the TOP cohort. . . . . . . . . . . . . . . . . . . 26Table 2.2 Gene reference models for HGVS nomenclature of OncoPanel genes. . . . . . . 27Table 2.3 Potential risk alleles in the hg19 human reference genome within the target re-gions of the OncoPanel. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28Table 2.4 Thresholds for parameters of VarScan2 fpfilter used for filtering raw variantoutput. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30Table 2.5 Spurious variants removed by the variant filtering pipeline. . . . . . . . . . . . 31Table 3.1 Comparison of coverage uniformity between blood and FFPE specimens usingthe Wilcoxon signed-rank test. . . . . . . . . . . . . . . . . . . . . . . . . . . 41Table 3.2 Multiple linear regression to predict log2 fold change between amplicon cover-age depth in blood and FFPE specimens based on amplicon length and GC content. 47Table 3.3 Summary statistics of fraction of base changes in blood and FFPE specimens. . 51Table 3.4 Multiple pairwise comparison of log2 fold change in fraction of base changesbetween blood and FFPE specimens using Dunn’s test with Benjamini-Hochbergmultiple hypothesis testing correction. . . . . . . . . . . . . . . . . . . . . . . 52Table 3.5 Summary statistics of fraction of base changes in blood and FFPE specimenswithin 1-10% allele frequency. . . . . . . . . . . . . . . . . . . . . . . . . . . 55Table 3.6 Spearman’s rank correlation between pre-sequencing variables (e.g. enrichmentefficiency and age of paraffin block) and sequencing metrics (e.g. fraction ofC>T/G>A, average per base normalized coverage, and on-target aligned reads). 59Table 4.1 Frequency of germline variants in cancer-related genes in blood specimens fromTOP patients. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63Table 4.2 Interpretation of germline alterations in cancer-related genes detected in bloodspecimens of TOP patients. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67viiiTable 4.3 Frequency of germline variants in pharmacogenomic genes detected in bloodspecimens of TOP patients. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77Table 4.4 Interpretation of germline alterations in pharmacogenomic genes detected inblood specimens of TOP patients. . . . . . . . . . . . . . . . . . . . . . . . . . 79Table 4.5 Distribution of discordant germline alterations in patients from the TOP cohort. 91Table 4.6 Sensitivity of identifying germline variants in tumour-only analyses at variousvariant allele frequency thresholds. . . . . . . . . . . . . . . . . . . . . . . . . 97Table 4.7 Positive predictive values for referral of potential germline variants to down-stream confirmatory testing at various variant allele frequency thresholds. . . . . 98Table A.1 Target regions and amplicons of the OncoPanel. . . . . . . . . . . . . . . . . . 127ixList of FiguresFigure 1.1 Sequencing cost per human-sized genome between 2001 and 2015. . . . . . . 2Figure 1.2 Workflow for Illumina sequencing. . . . . . . . . . . . . . . . . . . . . . . . . 6Figure 1.3 Comparison of testing content across targeted gene panels, whole exome se-quencing, and whole genome sequencing. . . . . . . . . . . . . . . . . . . . . 7Figure 1.4 Different approaches in target enrichment. . . . . . . . . . . . . . . . . . . . . 9Figure 1.5 File formats of raw output from NGS instruments. . . . . . . . . . . . . . . . 12Figure 1.6 VCF format for storing sequence variation data. . . . . . . . . . . . . . . . . . 14Figure 1.7 Four main criteria of the ACCE model process for evaluating a genetic test:Analytical validity, Clinical validity, Clinical utility, and Ethical, legal and so-cial implications. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16Figure 1.8 Involvement of TS, DPD, TP, and MTHFR in 5-FU mechanism of action. . . . 19Figure 1.9 Deviations of VAF as a result of tumour content and heterogeneity. . . . . . . . 21Figure 2.1 Schematic description of study design and data analyses. . . . . . . . . . . . . 25Figure 2.2 Pipelines for (A) variant calling and (B) filtering. . . . . . . . . . . . . . . . . 32Figure 2.3 2x2 contingency table for determination of true positive, false positive, truenegative, and false negative variant calls in tumour-only analyses. . . . . . . . 34Figure 3.1 Comparison of efficiency in amplicon enrichment between blood and FFPEspecimens. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38Figure 3.2 Assessment of read alignments between blood and FFPE specimens (Wilcoxonsigned-rank test). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39Figure 3.3 Evaluation of coverage uniformity in blood and FFPE specimens (Wilcoxonsigned-rank test, ****p < 0.0001, ns = not significant). . . . . . . . . . . . . . 40Figure 3.4 Amplicon-specific differences in coverage depth between blood and FFPE spec-imens. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44Figure 3.5 The relationship between amplicon GC content and amplicon length (Pearson’scorrelation). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45xFigure 3.6 Scatter plots showing log2 fold change between amplicon coverage depth inblood and FFPE specimens in relation to (A) amplicon length, (B) GC content,(C) top 100 longest amplicons, and (D) top 100 amplicons with the highest GCcontent (Pearson’s correlation). . . . . . . . . . . . . . . . . . . . . . . . . . . 46Figure 3.7 Assessment of formalin-induced sequence artifacts in FFPE specimens. . . . . 50Figure 3.8 Comparison of relative difference in fraction of base changes in FFPE speci-mens compared to blood (Kruskal-Wallis test). . . . . . . . . . . . . . . . . . 51Figure 3.9 Assessment of formalin-induced sequence artifacts in FFPE specimens at dif-ferent ranges of allele frequency. . . . . . . . . . . . . . . . . . . . . . . . . . 54Figure 3.10 Scatter plots showing (A) amplicon yield and (B) efficiency in amplicon enrich-ment, which is represented by the log2 fold change between the amount of DNAinput for producing amplicons and amplicon yield, in relation to age of paraffinblocks (Spearman’s rank correlation). . . . . . . . . . . . . . . . . . . . . . . 58Figure 3.11 The relationship between fraction of base changes and age of paraffin block fordifferent types of base changes (Spearman’s rank correlation). . . . . . . . . . 58Figure 4.1 Distribution of germline alterations in cancer-related genes in patients from theTOP study. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87Figure 4.2 Distribution of germline alterations in PGx genes in patients from the TOP study. 88Figure 4.3 Venn diagram demonstrating concordance of variants identified in 217 tumour-blood paired samples. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90Figure 4.4 Assessment of using a VAF cut-off approach to identify germline alterations intumour-only analyses. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97Figure 4.5 Assessment of using a VAF cut-off approach to refer potential germline alter-ations in tumour-only analyses to follow-up testing. . . . . . . . . . . . . . . . 98xiList of AbbreviationsACCE Analytical validity, Clinical validity, Clinical utility, and Ethical, legal and socialimplicationsACMG American College of Medical Genetics and GenomicsALK Anaplastic lymphoma kinase geneAPC Adenomatous polyposis coli geneASCII American Standard Code for Information InterchangeBAM Binary Alignment/MapBAQ Base quality scoreBCR-ABL1 Breakpoint cluster region and Abelson murine leukemia viral oncogene homolog 1fusion geneBRAF B-Raf proto-oncogeneBRCA1 Breast cancer type 1 susceptibility geneBRCA2 Breast cancer type 2 susceptibility geneBWA Burrows-Wheeler alignerBWT Burrows-Wheeler transformCDC Centers for Disease Control and PreventionCI Confidence intervalCNV Copy number variationCOSMIC Catalogue of Somatic Mutations in Cancer databaseCPG Cancer predisposing genexiiCRC Colorectal cancerdbSNP Single Nucleotide Polymorphism DatabasedNTP DeoxyribonucleotidesdTMP Deoxythymidine monophosphateDNA Deoxyribonucleic acidDPD Dihydropyrimidine dehydrogenase enzymeDPYD Dihydropyrimidine dehydrogenase geneEGFR Epidermal growth factor receptor geneExAC Exome Aggregation ConsortiumFAP Familial adenomatous polyposisFET Fisher’s exact testFFPE Formalin-fixed Paraffin-embeddedGIST Gastrointestinal stromal tumourGSTP Glutathione S-transferasesGSTP1 Glutathione S-transferases geneHER2 Human epidermal growth factor receptor 2 geneHGP Human Genome ProjectICGC International Cancer Genome ConsortiumIGV Intergrative Genomics ViewerLOH Loss of heterozygosityMAPQ Mapping quality scoreMIT Massachusetts Institute of TechnologyMEN2 Multiple endocrine neoplasia type 2MR Methylenetetrahydrofolate reductase enzymeMTHFR Methylenetetrahydrofolate reductase genexiiiNHGRI National Human Genome Research InstituteNGS Next-generation sequencingPGx PharmacogenomicsPCR Polymerase chain reactionPPV Positive predictive valueRB1 Retinoblastoma 1 geneRET Ret proto-oncogeneSAM Sequence Alignment/MapSNV Single nucleotide variantSRA Sequence Read ArchiveTCGA The Cancer Genome AtlasTOP The OncoPanel PilotTP53 Tumour protein p53 geneTP Thymidine phosphorylase enzymeTS Thymidylate synthase enzymeTYMP Thymidine phosphorylase geneTYMS Thymidylate synthase geneUDG Rracil-DNA glycosylaseUGT1A Uridine diphosphate glycosyltransferase 1 family enzymesUGT1A1 Uridine diphosphate glycosyltransferase 1A1 geneVAF Variant allele frequencyVCF Variant Call FormatWES Whole exome sequencingWGS Whole genome sequencingxivAcknowledgmentsI would like to thank my supervisor, Dr. Aly Karsan for giving me the opportunity to undertake abioinformatics project for my Master’s thesis and funding me throughout the course of my research.My gratitude is extended to my committee members, Dr. Martin Hirst and Dr. Ryan Morin forproviding guidance and ideas in data analyses for my project. I am also grateful for the productivediscussions I had with my colleagues from the Centre for Clinical Genomics and Karsan lab.There were many individuals who endured my frustration yet provided me with endless supportand encouragement. For that, I would like to express my heartfelt gratitude to my family, myacademic mentor, Dr. David Walker, and my friends. Thank you James Topham, Jenny Zhao,Jessica Pilsworth, Joey Ip, Kate Slowski, Ka Ming Nip, Laura Graziano, Patrick Coulombe, SamiahAlam, Samantha Jones, Santina Lin, Tehmina Masud, and Yuko Goto for keeping me motivated andsane throughout this journey. Last but not least, I would like to thank my partner, Florian Krauthanfor comforting me in my moments of failures, celebrating my successes, and understanding whydirty dishes were left in the sink.xvChapter 1Introduction1.1 The emergence of precision oncologyCancers are fundamentally a group of genetic disorders. The role of genetic alterations in drivingmalignant transformation has been implicated in studies dating back to the late nineteenth and earlytwentieth centuries by David von Hansemann and Theodor Boveri. Von Hansemann suggested thataberrant cell division accounted for unequal chromosome distributions in tumour cells [1, 213].Motivated by von Hansemann’s findings, Boveri explored the outcomes of sea urchin embryos thatwere induced to divide abnormally. An intriguing observation that drew Boveri’s attention wasnot all chromosomal imbalanced cells proliferated uncontrollably and formed tumours, there weresome that resulted in cell death, thus indicating that genetic materials were functionally distinct.This led to Boveri’s hypothesis that tumour development is promoted by retention of chromatinparts that stimulate growth or elimination of those that inhibit growth, concepts that manifested inthe present-day knowledge of oncogenes and tumour-suppressor genes, respectively [1, 32].Major strides have been made in understanding the molecular basis of cancer, including the dis-covery of recurrent gene mutations and elucidation of oncogenic pathways. Some of these findingshave been successfully translated into clinical applications wherein patients who harbour actionablesomatic mutations benefitted from treatment with targeted anti-cancer drugs. Notable examples in-clude treatment of HER2-overexpressed breast cancer with trastuzumab [10, 68, 186, 191, 193, 212],BCR-ABL1-translocated chronic myeloid leukemia with imatinib [69, 175], and BRAF-mutatedmelanoma with vemurafenib [42, 168]. The ability to improve clinical outcome by exploiting tu-mour genetic vulnerabilities has contributed to the advent of precision oncology, a framework thattailors patient care based on tumour genetic makeup.As more actionable somatic mutations are revealed, precision oncology regimens begin to facelimitations caused by single-gene assays, which pose challenges in scaling to meet diagnosticneeds. Fortunately, these barriers have been surmounted by advances in next-generation sequencing1(NGS) technologies. By harnessing the high-throughput nature of NGS, traditional gene-by-geneapproaches are being rapidly supplanted by targeted gene panels and genome-scale profiling, whichsurveys the whole exome or genome. The dramatic decline in sequencing cost [217] and low re-quirement for DNA input [48, 179, 194] have also accelerated the adoption of NGS-based genomictesting in clinical practice (Figure 1.1). Furthermore, concurrent progress in algorithmic devel-opment has enabled efficient storage, processing, and interpretation of massive genomic data setsproduced by NGS platforms [156, 206]. Automated variant analysis pipelines can be established byintegrating these bioinformatics tools to allow accurate reporting of clinically significant genomicalterations [30, 98, 113, 187]. Hence, while the precision oncology framework has been instigatedby the discovery of actionable somatic mutations, its translation into clinical use is largely catalyzedby advancements in DNA sequencing technologies and analysis algorithms. Although research ef-forts are still underway in refining these technological components, it is undeniable that the pre-cision oncology paradigm holds great potential in enhancing disease management and therapeuticintervention for cancer patients.Figure 1.1: Sequencing cost per human-sized genome between 2001 and 2015. Data by cour-tesy of the National Human Genome Research Institute (NHGRI) [217].21.2 Overview of next-generation sequencing technologiesThe Human Genome Project (HGP) was completed in 2003, approximately 13 years after its launchdate, producing the first human reference genome at an estimated expense of US$2.7 billion [2].While the HGP provided a wealth of information, which led to major breakthroughs in the fieldof genomics, the completion time and cost of the project were apparent rate limiting steps. Theneed for more time- and cost-efficient DNA sequencing methods stimulated the development ofNGS technologies. NGS is a general term that describes various high-throughout DNA sequencingtechnologies that can vary based on read length, chemistry, and detection methods (Table 1.1).These differences give rise to the strengths and weaknesses of each NGS platform. Recognition ofthese system specifications are essential for users to capitalize on the strengths and compensate forthe limitations of the different NGS technologies.In general, the sequencing process of most NGS platforms can be summarized into three steps.The first step involves nucleotide addition, which can be accomplished by DNA polymerase reac-tion or ligation (sequencing by synthesis vs. sequencing by ligation). This is followed by a detectionstep to identify the nucleotide species that was incorporated on single molecule or clonally ampli-fied DNA templates. Nucleotide detection can be performed using optical or non-optical sensing.Illumina and Pacific Biosciences (PacBio) platforms use optical sensing to detect fluorescence forbase calling [18, 36, 70, 72, 89], whereas the Ion Torrent platform uses non-optical sensing to detectchange in pH to determine nucleotide identity [173]. Lastly, a wash step re-initiates the cycle for thenext base on the DNA templates by removing anchor-probe complexes or fluorophores and block-ing groups. A key feature of NGS is its ability to simultaneously carry out this stepwise processfor many millions of DNA templates; hence, NGS is also known as massively parallel sequenc-ing. There are also NGS technologies that deviate from this sequencing cycle, such as the OxfordNanopore Technologies (ONT) platform, which directly measures DNA sequence using currentshifts produced as DNA translocates through nanopore sensors [216].3Table 1.1: Different NGS platforms by read length, chemistry, and detection method. Adapted from a table created by [118] under theCreative Commons Attribution 4.0 International License.Platform Read length† Amplification Chemistry Detection WebsiteComplete Genomics Short Clonal Sequencing by ligation Optical http://www.completegenomics.com/Illumina Short Clonal Sequencing by synthesis Optical http://www.illumina.comIon Torrent Short Clonal Sequencing by synthesis Solid state http://www.thermofisher.com/ca/en/home/brands/ion-torrent.htmlOxford Nanopore Long SinglemoleculeNanopore Nanopore https://nanoporetech.com/Pacific Biosciences Long SinglemoleculeSequencing by synthesis Optical http://www.pacb.com/Roche 454 Short Clonal Sequencing by synthesis Optical http://www.454.comSoLiD, ThermoFisherApplied BiosystemsShort Clonal Sequencing by ligation Optical http://www.thermofisher.com/ca/en/home/brands/applied-biosystems.html†Short-read platforms range from 35 bp to 700 bp, whereas long-read platforms range from 8 Kbp to 200 Kbp [84].41.2.1 Illumina sequencingAt present, the most widely used NGS technology is the Illumina short-read platform as evidentby its prevalence in the literature and the Sequence Read Archive (SRA). In 2011, 84% of se-quence reads in the SRA were generated by Illumina sequencing [109]. The Illumina platformuses a sequencing-by-synthesis approach with reversible dye terminators, which enable base callingthrough detection of fluorescent signals while blocking the ribose 3’-OH group to prevent additionof the next nucleotide by DNA polymerase [18, 89]. Briefly, an Illumina NGS workflow begins byligating adapters to the ends of fragmented DNA, followed by hybridizing these templates to com-plementary adapter sequences on flow cell surfaces. Bridge amplification is then performed to gen-erate clusters of clonally amplified DNA templates. DNA sequencing starts by annealing primerscomplementary to adapter sequences, which enable DNA polymerase to carry out the elongationprocess. All four reversible dye terminator-bound deoxyribonucleotides (dNTPs) are simultane-ously added during each cycle and are distinguishable by unique fluorophore-labelling. The dNTPsare also terminally blocked, allowing the incorporation of only one dNTP molecule per cycle. Sub-sequent to dNTPs addition, unbound dNTPs are washed away. The flow cells are then imaged usinglaser channels and fluorescence corresponding to the incorporated dNTP is emitted at each clus-ter. Finally, a new cycle is initiated by cleaving the fluorophores and unblocking the 3’-OH groups(Figure 1.2) [84, 118, 135, 136].5Figure 1.2: Workflow for Illumina sequencing. (1) Hybridization of adapter-ligated templates to flow cell surfaces. (2) Bridge amplifi-cation to generate clonally amplified clusters. (3) Sequencing by synthesis.61.2.2 Clinical applications of NGSThe emergence of NGS has revolutionized biological inquiry, particularly in cancer genomics re-search. Collaborative efforts such as The Cancer Genome Atlas (TCGA) and the InternationalCancer Genome Consortium (ICGC) have leveraged NGS technologies to characterize genomiclandscapes of different subtypes of tumours [6, 150, 166, 172, 203]. This has resulted in the iden-tification of novel driver mutations, thereby enhancing knowledge of tumour biology and treatmentstrategies. The ability to sequence multiple genes and samples in parallel with less DNA and in acost- and time-effective manner, also makes NGS an attractive clinical tool to complement preci-sion medicine initiatives. These advantages demonstrate that the viability and efficiency of NGS aresuperior to traditional Sanger sequencing, which is typically limited to sequencing a specific generegion of a given sample per run. Currently, tumour sequencing assays ranging from targeted genepanels up to genome-scale profiling have been employed in clinical oncology to guide diagnosis,prognosis, and therapeutic decision-making (Figure 1.3).Figure 1.3: Comparison of testing content across targeted gene panels, whole exome sequenc-ing, and whole genome sequencing.Targeted gene panelsSeveral considerations must be made when developing clinical-grade tumour genomic tests,including turnaround time, testing cost per patient, and depth of sequencing coverage. For thesereasons, many clinical laboratories have resorted to targeted gene panels, which focus on muta-tional hotspots, actionable genes, or genomic regions of known clinical relevance. Furthermore,despite the growing catalog of tumour genetic variants contributed by large consortium projects,the impact of the majority of variants on cancer development remains unknown [171, 197, 199].Genomic information of unknown significance not only pose limitations in clinical translation, butalso challenges in communicating such results to patients [53, 171, 183, 197]. Hence, targeted genepanels are more practical for prospective clinical use at the present time than whole exome andgenome approaches.Strategies to interrogate genomic regions of interest in targeted NGS assays include amplicon-based and capture-based methods (Figure 1.4). Amplicon-based method enriches target regions7using PCR amplification prior to NGS. PCR can be performed in uniplex, in which a single primerpair is used to generate amplicons within one reaction, or in multiplex, in which multiple primerpairs are used to generate amplicons in a single reaction. Conventional multiplex PCR faces chal-lenges such as interactions between primers and competition for reagents, which could lead to am-plification failures. These limitations of conventional multiplex PCR can be circumvented by themicrodroplet-based PCR platform developed by RainDance Technologies. In microdroplet-basedPCR enrichment, droplets containing a single primer pair are merged with droplets containing frag-mented genomic DNA and PCR reagent mix, which includes dNTPs and DNA polymerase. PCRenrichment is then performed for a library of droplets, which simultaneously amplifies several targetregions in a single reaction. Within each PCR droplet, amplicon generation is confined to a singleprimer pair and other reagents. This eliminates interaction between primer pairs and competitionfor reagents, thereby enhancing amplification uniformity [130].The amplicon-based approach typically requires lower amount of DNA input and offers quickturnaround relative to hybridization capture methods [41, 79]. However, PCR amplification can dis-tort detection of copy number variation (CNV), although bioinformatics tools, such as ONCOCNV[23], have been developed to perform copy number analysis using amplicon sequencing data. More-over, amplicon sequencing is prone to enrichment bias, especially in samples with low amounts ofDNA templates. For example, Wong et al. [221] reported higher prevalence of formalin-inducedsequence artifacts in samples with lower amounts of amplifiable templates. Clinical specimens withlow template copies tend to have reduced amplicon enrichment and higher probability of amplify-ing DNA templates with sequence artifacts [221]. Another limitation of amplicon sequencing is itsinability to detect novel gene fusions because PCR primer pairs would fail to amplify the translo-cated DNA [79, 190]. PCR primers can also mask genetic variants that are located within the primerregion, but this disadvantage could be mitigated by designing overlapping amplicons.Hybridization capture methods involve the use of complementary oligonucleotide probes tobind targeted regions. There are two methods for target capture, namely array-based and in-solutioncapture (Figure 1.4B). In array-based capture, complementary oligonucleotide probes of targetedregions are fixed on microarrays. Fragmented genomic DNA library is hybridized to the probeson the microarray and subjected to NGS after unbound library DNA is washed away. In the in-solution capture approach, the target-specific probes are biotinylated. The pool of probes is mixedwith fragmented genomic DNA library and hybridization occurs “in solution.” Hybridized DNAis pulled down using streptavidin-labelled magnetic beads and then subjected to NGS [79, 190].The in-solution capture method is typically preferred over array-based capture because it omitsthe necessity for expensive instrument to process microarrays, and it requires lower quantity ofDNA as starting material [22, 52]. In contrast to the amplicon-based method, hybridization captureapproaches can detect gene fusions, as well as yield more reliable inference of CNVs [79, 190].8Figure 1.4: Different approaches in target enrichment. (A) Amplicon-based enrichment. (B)Capture-based enrichment, which can be categorized as array-based and in-solution cap-turing.9Whole exome sequencingWhole exome sequencing (WES) interrogates all protein-coding regions, which constitute ap-proximately 1% of the genome [164, 190]. Target capture methods are commonly used to en-rich coding sequences before massively parallel sequencing. To date, it is estimated that 85% ofpathogenic variants are present within exons [164]. Therefore, WES assays have the potential tofacilitate prospective medical decision-making, as well as contribute to retrospective studies to un-cover the functional and clinical impacts of newly discovered genetic alterations in tumours.There are several disadvantages associated with WES assays. This includes the inability to de-tect mutations in non-coding regions and structural variants, which can promote cancer formation.WES assays also tend to achieve lower depth of coverage compared to targeted gene panels, in-creasing the rates of false positives. Because of the increased testing content, analytical validationof WES assays is more challenging and time-consuming. Nevertheless, whole exome testing hasalready been offered by a few academic centres such as Broad Institute of MIT and Harvard, BaylorCollege of Medicine, and Washington University in St. Louis, to assist in precision cancer medicine[164, 190].Whole genome sequencingWhole genome sequencing (WGS) scans the entire genome, including coding and non-codinggenomic regions. Similar to WES, WGS faces drawbacks in terms of depth of coverage and diffi-culty in ensuring analytic validity [190]. Although the Illumina HiSeq X Ten System has made itpossible to sequence an entire genome at 30x coverage under US$1000 [66], application of WGSin routine clinical testing is still challenging due to limitations in variant interpretation. As a re-sult of limited clinically annotated genetic variants, WGS is expected to yield a high burden ofvariants of unknown significance, which are problematic in clinical practice [170, 190]. Despitethese constraints, Laskin et al. [113] reported that the Personalized OncoGenomics (POG) study,which integrates whole genome analysis in making therapeutic decision, can benefit patients withadvanced cancers by matching them with targeted agents that are approved or currently in clinicaltrials. Thus, while there are challenges that need to be overcome to implement WGS as standard ofcare, WGS has proven its utility in conducting additional search for actionable mutations that areundetected by less comprehensive sequencing strategies. In particular, follow-up testing with WGScan broaden the treatment options for patients with incurable cancers.1.3 Variant analysis pipelineThe increase in affordability of NGS technologies has led to a marked surge in data production. Forinstance, the Illumina HiSeq 2500 platform is capable of sequencing 150–180 whole exomes fromhuman samples at 50x coverage, generating approximately 1TB of raw data in a single run [66].Processing these enormous data sets and extracting useful results for research and clinical purposes10rely heavily on bioinformatics algorithms and tools. A general workflow for variant analysis inmedical genomics consists of four stages: (1) quality control and pre-processing of raw sequencingreads, (2) read alignment to the reference genome and post-alignment processing, (3) variant calling,and (4) variant annotation and interpretation.1.3.1 Quality control and pre-processing of raw sequencing readsBase-calling algorithms convert signals, such as fluorescence, light intensity, or electrical current,captured by NGS instruments to DNA sequences. For each base identified, a measure of uncertainty,known as base quality (BAQ) score, is derived by taking into account background noise. BAQ scoresare commonly reported in Phred scale, given by the equation below.BAQPhred =−10log10P(error)Hence, 1/1000 probability of base error would correspond to a Phred-scaled BAQ score of 30[114, 149].The raw output of NGS instruments, which contains information from base calling, are stored inFASTQ and FASTA text-based formats. FASTQ files contain DNA sequences with sequence namesand Phred-scaled BAQ scores, whereas FASTA files only contain DNA sequences with sequencenames (Figure 1.5) [15, 143]. Quality of raw NGS data can be evaluated using bioinformaticstools such as FastQC, which generates a diagnostic report consisting of various quality controlparameters. These include sequence length distribution, GC content distribution, degree of sequenceduplication, presence of overrepresented and adapter sequences, and average Phred-scaled BAQscores at each base across reads [9].Quality control results are used to assist in pre-processing of raw NGS data before further anal-ysis takes place. For example, NGS libraries with poor quality bases near the 3’ ends of reads mayrequire read trimming before alignment. Removal of adapters is also typically performed in thepre-processing step [15]. Computational tools that are commonly used to accomplish these tasksinclude Trimmomatic [26] and Cutadapt [132]. Moreover, assessment of quality control metricsalso enables the recognition of poor quality NGS libraries and those that are potentially contami-nated. Flagging of these libraries would allow downstream analyses and result interpretation to beperformed with caution.11Figure 1.5: File formats of raw output from NGS instruments. (A) FASTQ format. Sequencename starting with @ is presented in line 1, followed by read sequence in line 2. Line 3begins with + and may be followed by the sequence name again, and line 4 consists ofquality scores corresponding to bases in line 2. Quality scores are represented as ASCIIcharacters. (B) FASTA format. Line 1 contains sequence name, which starts with >,whereas line 2 contains the read sequence.1.3.2 Read alignment and post-alignment processingNext, pre-processed reads are aligned to the reference genome. Alignment algorithms essentiallymatch read sequences to sequences in the reference genome, while accounting for sequencing errorsand true genomic alterations [15, 123, 143, 149]. Because of the large data output produced by NGS,alignment algorithms must also be time- and memory-efficient.Widely-used alignment softwares implement either the Burrows-Wheeler transform (BWT)compression algorithm or hash table indexing [15, 123, 149, 156]. Popular BWT-based aligners in-clude BWA [120–122] and Bowtie2 [112], which are well-known for reduced runtimes and memoryrequirements. Conversely, algorithms based on hash tables such as Novoalign (http://novocraft.com),SHRiMP2 [55], and Stampy [129] have longer computational times but tend to yield more accu-rate alignments [15, 123, 149]. The standard formats for storing aligned read data are the Se-quence Alignment/Map (SAM) text-based format and its compressed binary version, the BinaryAlignment/Map (BAM) format [124].Several post-alignment steps are performed such as removal of read duplicates, which could beintroduced by PCR bias. Local realignment is also often performed at genomic regions surroundinginsertions-deletions (indels) to reduce errors caused by inaccurate alignments. As raw BAQ scoresmight be erroneously assigned, recalibration of BAQ scores is sometimes recommended prior tovariant calling. Quality scores are adjusted by accounting for differences in quality between machinecycles and neighbouring dinucleotides [15, 143, 149, 155].1.3.3 Variant callingVariant calling identifies genomic differences between aligned read sequences and the referencegenome. When matched normal samples are sequenced, some variant callers are able to detect12germline, somatic, and loss of heterozygosity (LOH) events through paired analysis of tumour andmatched normal samples [108]. Variant callers can be categorized as heuristic or probabilistic [15,143, 156]. The standard output of variant callers for storing sequence variant data is the text-basedVariant Call Format (VCF) (Figure 1.6) [54].VarScan2 is a variant caller that uses heuristic factors such as cut-offs for coverage depth, BAQscore, and variant allele frequency (VAF), as well as the one-sided Fisher’s exact test (FET) onmapped read counts to identify variants [15, 107, 108, 143, 156]. In single sample analysis (sam-ple vs. reference genome), a variant is detected through applying the one-sided FET to comparereference- and variant-supporting read counts to an expected read distribution that reflects sequenc-ing error rate at a reference site [107, 108]. For example, the expected read distribution for a ref-erence site with 2000x coverage depth and a 0.1% sequencing error rate would comprise 2 variant-supporting reads, which are reads with errors, and 1998 reference-supporting reads.Variant callers that are based on probabilistic methods employ Bayes’ theorem to measure pos-terior probabilities of all possible genotypes at a loci. These posterior probabilities are then usedto infer genotype. Probabilistic methods also take into account prior information such as popula-tion allele frequencies, which can be obtained from the Single Nucleotide Polymorphism Database(dbSNP) or through multi-sample variant calling, and patterns of linkage disequilibrium, as wellas genotype likelihood, which can be computed using BAQ scores [143, 149]. Examples of vari-ant callers that implement probabilistic approaches include GATK [138], MuTect [49], and Strelka[182].13Figure 1.6: VCF format for storing sequence variation data. The VCF header, which beginswith ##, contains information describing the data in the file. The line starting with #displays the column names and indicates the beginning of the body of the VCF file. Datacolumns must include eight mandatory fields: chromosome (CHROM), 1-based genomicposition of the variant (POS), variant identifier (ID), reference base (REF), alternate base(ALT), quality score (QUAL), indicator whether the variant passed the specified filteringcriteria (FILTER), variant annotations separated by semicolon (INFO). The FORMATfield contains colon-separated descriptors of the values in the genotype column, which isnamed after the sample(s) reported in the VCF file [54].1.3.4 Variant annotation and interpretationSubsequent to variant calling, variant annotation is performed by adding structural and functionalinformation. Structural annotation provides information on the genomic location of the variant (e.g.intronic, intergenic, 5’UTR, 3’UTR, etc.), the impact of the variant on the gene transcript (e.g.missense, non-synonymous, synonymous, frameshift, etc.), and changes in codon and amino acid[15, 145, 155]. These information are typically reported in the Human Genome Variation Society(HGVS) nomenclature to ensure a consistent format in describing sequence variants [59].Functional annotation, on the other hand, provides information on the effect of genomic vari-ants on protein function. Pathogenicity of a variant cannot be exclusively presumed based on thepredicted variant effect. For example, not all truncating, missense, and frameshift variants leadto deleterious effects. Conversely, not all synonymous variants are benign. Thus, functional pre-diction algorithms compute functional inference by incorporating additional data such as sequencehomology, 3D protein structure, genomic context, protein interaction network, and evolutionaryconservation [155, 163]. Examples of functional prediction tools include PolyPhen-2 [3], SIFT14[147], MutPred [119], Condel [83], and PhD-SNP [35].Population allele frequencies are also taken into account when interpreting variants [15, 145,155]. For instance, a common variant (minor allele frequency >1%) is unlikely to be involved incancer predisposition [14, 16]. Population allele frequencies can be obtained by annotating vari-ants with the dbSNP and Exome Aggregation Consortium (ExAC) database. Furthermore, variantdatabases such as ClinVar and Catalogue of Somatic Mutations in Cancer (COSMIC) are also usefulresources that provide information on clinical significance of variants and presence of variants incancers [145, 155].The original output of variants is then filtered and prioritized based on the different types ofannotations, as well as clinical and biological relevance, generating a list of candidate variants[145, 155]. These variants are typically visualized using a genomic browser such as the IntergrativeGenomics Viewer (IGV) for quality control purposes [155, 156]. Finally, a report is producedand reviewed by a board certified medical pathologist or clinical molecular geneticist for approval[145, 197].1.4 ACCE model process for evaluating genetic testsDevelopment of a genetic test, including NGS-based genomic testing, for clinical use must be ac-companied by an evaluation process to establish robustness and clinical benefits of the test. Oneapproach to assess genetic tests is the ACCE model process, which consists of four criteria thatmake up its acronym: Analytical validity, Clinical validity, Clinical utility, and Ethical, legal andsocial implications [181, 232].Analytic validation ensures that a clinical assay detects the genetic changes it is designed toidentify with sufficient sensitivity and specificity [181, 232]. For example, analytic validity of atargeted NGS panel can be determined by running the panel on samples with known mutations thatwere previously identified using Sanger sequencing. Sensitivity and specificity of the targeted NGSpanel can be measured using the results from Sanger sequencing as reference standards. Analyticvalidity also refers to quality assurance of a clinical assay. For instance, a targeted NGS panel mustbe capable of producing similar sequencing metrics (e.g. read depth and coverage uniformity) forsamples with comparable DNA quantity and integrity.Clinical validation determines whether results of the genetic test correspond to the clinical con-dition it is meant to detect [181, 232]. Example of a genetic test with high clinical validity is RETmutational testing, which can identify individuals with multiple endocrine neoplasia type 2 (MEN2)at a sensitivity of 95–98%. MEN2 is a heritable disorder transmitted in an autosomal dominantpattern, resulting in increased susceptibility to tumours in endocrine tissues, especially medullarythyroid carcinoma [33]. On the other hand, clinical utility of a genetic test is defined by its abilityto enhance clinical outcome, such as survival and progression-free survival, after weighing in therisks, benefits, and economic impact of the test [181, 232]. The clinical utility of RET mutational15testing is demonstrated by its efficacy in identifying MEN2 susceptible children who would benefitfrom prophylactic surgery to remove all or parts of the thyroid gland. This results in reduced risk ofdeveloping thyroid cancers, thereby improving survival of these individuals [33].Lastly, the ACCE model includes evaluation of ethical, legal and social implications of a genetictest. This component considers how the genetic test can lead to ramifications such as violationof privacy and confidentiality, stigmatization and discrimination based on genetic makeup (e.g.accessibility to insurance), and complications pertaining to consent for disclosure and ownership ofthe data. As well, this component of the framework ensures that safeguards, such as relevant policiesand genetic counseling protocols, are implemented to prevent societal repercussions [181, 232].Figure 1.7: Four main criteria of the ACCE model process for evaluating a genetic test:Analytical validity, Clinical validity, Clinical utility, and Ethical, legal and social im-plications. Image by courtesy of Centers for Disease Control and Prevention (CDC).1.5 Clinical implications of germline alterations in cancerScreening for somatic and germline alterations are essential in delivering precision medicine to can-cer patients. Somatic mutations can influence disease management and treatment of cancer patientswith targeted agents, whereas clinical implications of germline alterations extend beyond the pa-tients, affecting their families as well. Germline variants in cancer predisposing genes (CPGs) can16predict the risk of disease onset, allowing for preventive measures to be administered [165]. Further-more, germline variants in pharmacogenomic (PGx) genes can predict response to chemotherapeuticdrugs, including drug sensitivity and adverse drug reactions [144, 158]. Therefore, germline test-ing should be offered to ensure more precise cancer care if resources are available to analyze andinterpret germline findings, and appropriate protocols are established to communicate results withpatients and affected family members.1.5.1 Cancer predispositionGermline variants in CPGs can indicate increased cancer risks. Between 1982 and 2014, 114 CPGshave been discovered using approaches such as candidate gene, genome-wide mutation, and link-age analyses. The majority of CPGs act as tumour suppressors; hence, loss-of-function mutationsin these genes, which inactivate gene function, predispose carriers to cancer. Genes in this cate-gory, including TP53, BRCA1, BRCA2, APC, and RB1, are usually involved in DNA repair andcell-cycle regulation. Conversely, there are fewer CPGs that promote cancer formation throughgain-of-function mutations. These CPGs, typically protein kinases such as ALK, EGFR, and RET ,predispose carriers to cancer through activation of gene function [165].Clinical testing of CPGs can improve various aspects of patient care such as disease managementand treatment. For instance, patients with BRCA-deficient breast and ovarian tumours can be treatedwith PARP inhibitors, which target tumour cells and impair growth through synthetic lethality [57,91]. Notably, a key benefit of testing for germline variants in CPGs is the window of opportunityto implement cancer preventive measures for patients and affected relatives. Cancer prevention caninvolve approaches such as early and regular cancer screening, as well as prophylactic surgery andchemotherapy. For example, patients with familial adenomatous polyposis (FAP), which is causedby germline alterations in the APC gene and associated with high risk of developing colorectalcancer (CRC), are recommended to begin early colonoscopy-based screening [21].1.5.2 PharmacogenomicsDespite the expanding spectrum of targeted anti-cancer drugs, cytotoxic chemotherapy remains theprimary treatment for several types of cancers. However, germline variants in PGx genes that affectthe function and/or expression of drug targets and drug disposition proteins (proteins involved indrug metabolism and transport) can give rise to chemotherapy-related toxicity [144, 158]. Exam-ples of chemotherapy-related toxicity include hand-foot syndrome, hearing loss, cardiomyopathy,and high-grade neutropenia, diarrhea, nausea and vomiting [58, 95, 110, 115, 178]. Chemotherapy-related toxicity can be debilitating and fatal, as well as culminate significant expenditures towardscancer supportive care [17, 157, 167]. To alleviate the occurrence of chemotherapy-related tox-icity, germline PGx testing should be implemented in clinical practice to guide the selection ofchemotherapeutic drugs and optimization of drug dosage for cancer patients.175-fluorouracil5-fluorouracil (5-FU) is a fluoropyrimidine drug that is commonly administered in chemother-apy regimens for patients with gastrointestinal cancers, including CRC. Inter-patient variability inresponse to 5-FU treatment can be caused by germline variants in the TYMS gene, which encodesfor the drug target, the thymidylate synthase enzyme (TS). One of the 5-FU mechanisms of ac-tion involves the conversion of 5-FU to fluorodeoxyuridine monophosphate (5-FdUMP). 5-FdUMPthen sequesters TS by forming a ternary complex with TS and the 5,10-methylenetetrahydrofolate(CH2THF) cofactor, thereby impeding DNA synthesis (Figure 1.8). Germline alterations that resultin a higher expression of TS such as the triple repeats of a 28 bp sequence upstream of the TYMStranslational start site (rs45445694) are indicators of reduced likelihood of experiencing 5-FU toxi-city. Unfortunately, this also means that treatment with 5-FU might not be effective due to high TSlevels in the tumours [144, 158].Germline variants in DPYD and TYMP genes, which encode for the 5-FU metabolizing proteins,dihydropyrimidine dehydrogenase (DPD) and thymidine phosphorylase (TP), respectively, can alsoserve as predictors for 5-FU-induced toxicity. DPD catabolizes 5-FU into dihydrofluorouracil (5-FUH2), which mainly occurs in the liver, and the inactive products are subsequently excreted in theurine (Figure 1.8). Hence, germline variants resulting in DPD deficiency or total loss contribute toa longer half-life of 5-FU, which can cause severe or fatal toxicity in cancer patients. Several stud-ies implied that TP may play a causative role in tumour growth and metastasis. In fact, higher TPexpression was observed in tumours than normal tissues in CRC patients. Consequently, the 5-FUprodrug, capecitabine, is administered to target TP-overexpressed tumours because TP can metab-olize capecitabine to the thymidylate synthase inhibitor, 5-FdUMP (Figure 1.8). Hence, this affectstumour growth with minimal toxic effects in normal cells. However, the presence of germline vari-ants that increase expression of TP in normal cells could potentially lead to adverse drug reactionsin patients receiving 5-FU-based chemotherapy [144, 158].Efficacy of 5-FU depends on the intracellular reduced folate, CH2THF, which together withthe 5-FU active metabolite, 5-FdUMP, inhibit TS. This blocks the synthesis of deoxythymidinemonophosphate (dTMP), causing imbalanced nucleotide levels in the cell and DNA damage. Oneof the enzymes that regulates intracellular CH2THF levels is methylenetetrahydrofolate reductase(MR), which irreversibly converts CH2THF to CH3THF (Figure 1.8). Germline variants in theMTHFR gene that reduce enzymatic activity such as c.677C>T and c.1298A>C polymorphismscan increase chemosensitivty of tumours to 5-FU through cellular accumulation of CH2THF. Nev-ertheless, several studies suggested that the combined presence of MTHFR c.1298A>C and TYMS3’UTR indels could serve as predictors for 5-FU toxicity in CRC patients [158].18Figure 1.8: Involvement of TS, DPD, TP, and MTHFR in 5-FU mechanism of action.Dashed lines indicate more than one process is involved in producing the out-put, whereas solid lines indicate direct reactions. 5-FU, fluorouracil; 5-FdUMP,fluorodeoxyuridine monophosphate; 5-FUH2, dihydrofluorouracil; CH2THF, 5,10-methylenetetrahydrofolate; CH3THF, 5-methyltetrahydrofolate.OxaliplatinOxaliplatin is a platinum derivative commonly used in combination with 5-FU for treating gas-tric and colorectal cancers. Deactivation of oxaliplatin can be induced by conjugation of the plat-inum derivative with glutathione (GSH), which is catalyzed by glutathione S-transferases (GSTP).While there are studies suggesting that germline variants in the GSTP1 gene are associated with in-creased neurotoxicity in patients treated with oxaliplatin combination therapy, there are also groupsreporting conflicting results. Therefore, additional studies are required to confirm the impact ofGSTP1 polymorphisms on oxaliplatin treatment [144, 158].19IrinotecanIrinotecan is a camptothecin analog widely used in chemotherapy regimens for treating lungcancers and CRC. The active metabolite of irinotecan, SN-38, blocks type I DNA topoisomerase,impairing DNA replication. SN-38 is inactivated in the liver by uridine diphosphate glycosyltrans-ferase 1 family enzymes (UGT1A) through glucuronidation and then excreted. Thus, patients withdeficiency in UGT1A, which can be caused by germline variants in the UGT1A1 gene, are at higherrisk of experiencing toxicity due to a longer half-life of SN-38. An example of a germline variant inUGT1A1 that results in reduced activity is the UGT1A1*28 allele, which corresponds to an extra TArepeat within position -53 and -42 of the translational start codon. Dose reduction is recommendedfor carriers to prevent toxic effects induced by irinotecan [144, 158].1.6 Technical challenges in implementing germline testing in clinicaloncology1.6.1 Tumour-only sequencingOne of the challenges in integrating germline testing in clinical oncology is tumours are often se-quenced without matched normal samples. Although sequencing of matched normal samples wouldallow accurate identification of somatic mutations and simultaneous detection of clinically impor-tant germline variants, it is common for clinical laboratories to only sequence tumour samples tominimize cost and turnaround time. However, genomic analyses of tumours can reveal clinicallyrelevant germline variants [27, 102, 140, 141, 184]. For examples, Schrader et al. [184] reportedthat pathogenic germline variants in CPGs were retained in the tumour genomes of 91.9% of pa-tients in the study cohort. Hence, clinical laboratories could leverage tumour genomic testing foridentification of germline variants and subsequently refer potential germline variants to downstreamconfirmatory testing.A clinical pipeline that leverages tumour genomic testing to perform initial screening for germlinealterations could provide germline testing in a cost-effective manner because only selected patientswould require follow-up testing. However, as the tumour genome contains both germline and so-matic variants, the difficulty remains in devising an approach to accurately separate germline vari-ants from somatic mutations. Jones et al. [102] used public databases such as dbSNP and COSMIC,as well as effect prediction tool to distinguish between germline and somatic variants, but reportedhigh false positive rates. The use of these public databases cannot reliably differentiate betweenvariant statuses because it is possible for a germline variant to occur somatically, and vice versa.For instance, an evaluation of 468 genes with known somatic driver mutations recorded in the COS-MIC database showed that 49 of these genes were also known to harbour germline alterations thatare associated with inherited predisposition to cancer [165].20One possible approach is the use of VAF to discriminate between germline and somatic variants.Because tumour biopsies are typically admixtures of tumour and normal cells, there is a high like-lihood that somatic mutations might deviate from diploid zygosity (i.e. heterozygous variants areexpected to have VAF close to 50%, whereas homozygous variants are expected to have VAF closeto 100%; Figure 1.9A). Moreover, tumour heterogeneity might also give rise to VAF deviations(Figure 1.9B). Therefore, the use of VAF threshold could be a potential solution in distinguishingbetween germline and somatic alterations in genomic analyses of tumours without matched normalDNA.Figure 1.9: Deviations of VAF as a result of tumour content and heterogeneity. Blue and redcells represent tumour cells, whereas grey cells represent normal cells.211.6.2 Formalin-fixed paraffin-embedded tumoursAnother disadvantage of performing germline variant analysis using tumour DNA is tumour sam-ples in the clinic are often formalin-fixed paraffin-embedded (FFPE). Formalin fixation preservestissue morphology for histological assessment, whereas paraffin embedding enables stable storageof specimens at room temperature, which is cost- and space-saving compared to maintaining freshfrozen specimens in freezers [63, 66]. DNA isolated from FFPE tumours pose technical challengesin molecular testing because formalin fixation induces several types of DNA damage [63]. There-fore, assessment of these different forms of DNA damage is essential to establish quality control fora clinical genomic assay.The main component of formalin, formaldehyde, can react with DNA bases and proteins, pro-ducing DNA-DNA, DNA-protein, and protein-protein crosslinks. Additionally, formaldehyde-DNAadducts can also be generated in formalin-fixed tissues. Crosslinking induced by formaldehydedestabilizes the DNA structure, resulting in degradation and low DNA yields extracted from FFPEtissues [63]. Another predominant form of formalin-induced DNA damage is DNA fragmentation.Hence, FFPE tissues not only produce low quantities of DNA, but also DNA with short fragmentsizes. Particularly, this interferes with amplicon-based methods by reducing the amount of ampli-fiable DNA templates [61, 188, 220]. Severity of DNA fragmentation also increases with age ofparaffin blocks and acidity of formalin solution used in tissue fixation [38, 128].FFPE DNA also constitutes increased frequency of sequence artifacts. This is problematic inclinical practice because there is a high risk of misinterpreting artifactual base changes as true mu-tations that may influence patient care. Oxidization of formaldehyde, which generates formic acid,creates an acidic environment that catalyzes hydrolytic cleavage of N-glycosidic bonds betweenpurines and the sugar backbone [63]. This produces abasic sites at which sequence artifacts canoccur as most DNA polymerases tend to selectively incorporate adenines across abasic sites duringthe extension stage. In fewer cases, guanines and short deletions ranging from 1–3 bases could alsobe introduced by DNA polymerase when synthesizing through abasic sites [92].A well-documented source of sequence artifacts in FFPE DNA is cytosine deamination. Thisgenerates uracil lesions, which leads to artifactual C>T/G>A transitions because adenines areadded opposite of uracils during synthesis of complementary DNA strands [63]. Wong et al. [221]showed increased levels of C>T/G>A artifacts in amplicon sequencing data generated from highlyfragmented DNA samples. This observation was attributed to a higher probability of amplifyingDNA templates containing sequence artifacts in samples with reduced amount of amplifiable DNAtemplates as a result of fragmentation damage [221]. While cytosines can be restored by treatingFFPE DNA with uracil-DNA glycosylase (UDG) to eliminate uracil lesions, there is currently nomethod to repair deamination of 5-methylcytosine (5-mC). 5-mC are common at CpG dinucleotidesand are more susceptible to deamination in formalin-fixed tissues. Deamination of 5-mC gives riseto thymine instead of uracil, thus cannot be reinstated through treatment with the UDG enzyme [63].221.7 ObjectivesGermline alterations have clinical implications for cancer patients and their family members. Be-cause the tumour genome contains both somatic and germline variants, clinical tumour sequencingpresents an opportunity for pre-screening of germline variants. This framework is a time- and cost-effective approach for providing germline testing because only patients with potential germlinevariants would require downstream confirmatory testing. A primary challenge in implementingthis framework for identifying clinically significant germline variants is differentiating betweengermline and somatic alterations in the tumour genome. Tumour specimens also tend to be FFPE,which causes DNA damage that could affect the use of FFPE DNA for clinical genomic testing.To date, no study has evaluated the detection of germline alterations in FFPE tumours usingNGS-based tests. As formalin fixation is known to cause DNA damage, the usability of FFPEDNA for NGS-based germline testing must be compared to a gold standard such as DNA extractedfrom the peripheral blood mononuclear cell fraction. To determine reliability of using tumour DNAfor germline variant calling, the retention rate of germline variants in the tumour genome mustbe measured. This is because tumour-specific mutations might result in loss of germline variants.Currently, there is also no standard method in distinguishing between germline and somatic variantsin tumour-only analyses. Hence, an approach to distinguish between germline and somatic variantsin tumour genomes must be established and assessed for its sensitivity and precision.In this study, we aimed to determine whether potential germline alterations can be accuratelyidentified through genomic analyses of FFPE tumours. We performed analytic validation of a clini-cal amplicon-based targeted sequencing panel for FFPE solid tumours by comparison with sequenc-ing of DNA isolated from the peripheral blood mononuclear cell fraction, which is the gold standardfor germline testing. We identified three objectives in this study:1. Assess the degree of formalin-induced DNA damage in FFPE DNA2. Determine the retention rate of germline alterations in FFPE tumours, and3. Evaluate the use of VAF thresholds to distinguish germline alterations from somatic mutationsin tumour-only analyses.Through these analyses, we hoped to characterize formalin-induced DNA damage to facilitate qual-ity control and improve robustness of our assay. Finally, we also aimed to measure the sensitivitiesfor identifying potential germline alterations and positive predictive values (PPVs) for referringgermline alterations to downstream germline testing at various VAF cut-offs. Establishing theseperformance parameters would serve as important guidelines in clinical practice.23Chapter 2Materials and Methods2.1 Overview of study designThis study examined whether potential germline alterations can be accurately identified in FFPEtumours. Targeted sequencing data from 213 cancer patients with FFPE tumour and matched bloodsamples were retrospectively analyzed (Figure 2.1). DNA was extracted from the peripheral bloodmononuclear cell fraction, hereafter referred to as blood, and FFPE specimens. The DNA sampleswere sheared and enriched for amplicons in the OncoPanel, a clinical targeted NGS panel for solidtumours. Amplicons were barcoded and subjected to NGS. Sequencing data were processed andanalyzed with a custom variant analysis pipeline. To assess the degree of formalin-induced DNAdamage, the efficiency in amplicon enrichment and sequencing results of FFPE samples were com-pared to blood. Furthermore, variant concordance between blood and FFPE tumours was measuredto determine whether tumour DNA is a reliable resource for detecting germline alterations. Lastly,the use of VAF thresholds in distinguishing between germline and somatic alterations in tumour-only analyses was evaluated.24Figure 2.1: Schematic description of study design and data analyses.2.2 Patient samplesBlood and FFPE tumour samples were acquired from 213 patients who provided informed consentfor The OncoPanel Pilot (TOP) study (Human Research Ethics Protocol H14-01212), a pilot studyto optimize the OncoPanel, which is an amplicon-based targeted NGS panel for solid tumours.The TOP study also assessed the OncoPanel’s application for guiding disease management andtherapeutic intervention. One blood sample and four FFPE tumours were sequenced in duplicates,which resulted in 217 tumour-normal paired samples (434 sequencing libraries were included inour analyses). Patients in the TOP study were those with advanced cancers including CRC, lungcancer, melanoma, gastrointestinal stromal tumour (GIST), and other cancers (Table 2.1). The ageof paraffin block for tumour samples ranged from 18 to 5356 days with a median of 274 days.25Table 2.1: Distribution of cancer types in the TOP cohort.Cancer Type Number of Cases Percentage (%)Colorectal 97 46Lung 60 28Melanoma 18 8Other† 16 8GIST 7 3Sarcoma 4 2Neuroendocrine 4 2Cervical 2 0.9Ovarian 2 0.9Breast 2 0.9Unknown 1 0.5†This category includes thyroid, peritoneum, Fallopian tube, gastric, endometrial, squamous cell carcinoma,anal, salivary gland, peritoneal epithelial mesothelioma, adenoid cystic carcinoma, pancreas, breast, gallbladder, parotid epithelial myoepithelial carcinoma, carcinoid, and small bowel cancers.2.3 Sample preparation, library construction, and IlluminasequencingGenomic DNA was extracted from the peripheral blood mononuclear cell fraction and FFPE tumoursamples using the Gentra Autopure LS DNA preparation platform and QIAamp DNA FFPE tissuekit (Qiagen, Hilden, Germany), respectively. The extracted DNA was sheared according to a previ-ously described protocol [30] to obtain approximate sizes of 3 Kb followed by PCR primer merging,amplification of target regions, and adapter ligation using the Thunderstorm NGS Targeted Enrich-ment System (RainDance Technologies, Lexington, MA) as per manufacturer’s protocol. Barcodedamplicons were sequenced with the Illumina MiSeq system for paired end sequencing with a v2250-bp kit (Illumina, San Diego, CA).262.4 OncoPanel (Amplicon-based targeted sequencing panel for solidtumours)The OncoPanel assesses coding exons and clinically relevant hotspots of 15 cancer-related genes andsix PGx genes. Germline alterations in the six PGx genes could serve as predictors of susceptibilityto chemotherapy-induced toxicity. Primers were designed by RainDance Technologies (Lexington,MA) using the GRCh37/hg19 human reference genome to generate 416 amplicons between 56 bpand 288 bp in size, which interrogate ∼ 20 Kb of target bases. Complete list of genes and genereference models for the OncoPanel is presented in Table 2.2, whereas OncoPanel target regionsand amplicons are presented in Table A.1.Table 2.2: Gene reference models for HGVS nomenclature of OncoPanel genes.Gene Protein Reference ModelCancer-relatedAKT1 Protein kinase B NM 001014431.1ALK Anaplastic lymphoma receptor tyrosine kinase NM 004304.3BRAF Serine/threonine-protein kinase B-Raf NM 004333.4EGFR Epidermal growth factor receptor NM 005228.3HRAS GTPase HRas NM 005343.2MAPK1 Mitogen-activated protein kinase 1 NM 002745.4MAP2K1 Mitogen-activated protein kinase kinase 1 NM 002755.3MTOR Serine/threonine-protein kinase mTOR NM 004958.3NRAS Neuroblastoma RAS viral oncogene homolog NM 002524.3PDGFRA Platelet-derived growth factor receptor alpha NM 006206.4PIK3CA Phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit alpha NM 006218.2PTEN Phosphatase and tensin homolog NM 000314.4STAT1 Signal transducer and activator of transcription 1 NM 007315.3STAT3 Signal transducer and activator of transcription 3 NM 139276.2TP53 Tumor protein P53 NM 000546.5Pharmacogenomic-relatedDPYD Dihydropyrimidine dehydrogenase NM 000110.3GSTP1 Glutathione S-rransferase pi 1 NM 000852.3MTHFR Methylenetetrahydrofolate reductase NM 005957.4TYMP Thymidine phosphorylase NM 001113755.2TYMS Thymidylate synthetase NM 001071.2UGT1A1 Uridine diphosphate (UDP)-glucuronosyl transferase 1A1 NM 000463.2272.5 Variant calling pipeline2.5.1 Read alignment and variant callingReads that passed the Illumina Chastity filter were aligned to the hg19 human reference genomeusing the BWA mem algorithm (version 0.5.9) with default parameters, and the alignments wereprocessed and converted to the BAM format using SAMtools (version 0.1.18). The SAMtoolsmpileup function (samtools mpileup -BA -d 500000 -L 500000 -q 1)was usedto generate pileup files for all target bases followed by variant calling with the VarScan2 mpileup2cns(version 2.3.6) function with parameter thresholds of VAF ≥ 0.1 and Phred-scaled BAQ score ≥ 20(--min-var-freq 0.1 --min-avg-qual 20 --strand-filter 0 --p-value 0.01--output-vcf --variants).Four genomic positions at which the hg19 human reference genome contained potential riskalleles were identified (Table 2.3). Hence, patients homozygous for these four risk alleles wouldnot be identified by our standard variant calling procedure. For these four genomic sites, ourmethod for variant calling was modified to provide calls for every patient in the cohort. TheVarScan2 mpileup2cns function was used with parameter thresholds of VAF ≥ 0.25, VAF tocall homozygote ≥ 0.9, BAQ score ≥ 20, and fraction of variant reads from each strand ≥ 0.1(--min-var-freq 0.25 --min-freq-for-hom 0.9 --min-avg-qual 20--strand-filter 1 --p-value 0.01 --output-vcf). Next, allelic statuses werere-assigned, in which wild type calls were re-assigned as homozygous variants, while homozy-gous variants were re-assigned as wild type calls. Corrections to the VAFs of these four genomicsites were also made to ensure that the VAFs reflected percentage of reads with the risk alleles.Table 2.3: Potential risk alleles in the hg19 human reference genome within the target regionsof the OncoPanel.Gene Chr Pos Risk Allele dbSNP ID HGVS*DPYD chr1 98348885 C rs1801265 p.Cys29Argc.85T>CMTOR chr1 11205058 G rs386514433;rs1057079p.Ala1577Alac.4731A>Gchr1 11288758 C rs1064261 p.Asn999Asnc.2997T>CTP53 chr17 7579472 C rs1042522 p.Arg72Proc.215G>C*Description of sequence variants according to the HGVS recommendations.282.5.2 Variant filteringVariant calls were filtered using the VarScan2 fpfilter function with fraction of variant readsfrom each strand ≥ 0.1 and default thresholds for other parameters (Table 2.4). The VarScan2fpfilter removed 247 low quality variants. Seventy germline variants in the blood were alsoexcluded from our analysis because these variants in the tumours were filtered by the VarScan2fpfilter. There were also 16 risk allele calls in tumour samples that did not pass the strand filter,causing the removal of 10 risk allele calls in the blood samples from our evaluation. Overall, a totalof 343 calls were excluded by the VarScan2 fpfilter and strand filter. Eleven low coverage calls(≤ 100x) were also excluded from our analysis. Manual inspection was performed for a subset ofvariants, including variants detected within primer regions and in PGx genes, using the IntegrativeGenomics Viewer (IGV, version 2.3). This resulted in the removal of 500 spurious calls, whichstemmed from misalignment near indels, sequencing artifacts, primer masking, and primer artifacts(Table 2.5). Implementation of this filtering pipeline reduced the raw variant output of 5288 callsfrom 217 paired tumour-blood samples (434 sequencing libraries) to 4434 calls (Figure 2.2B).29Table 2.4: Thresholds for parameters of VarScan2 fpfilter used for filtering raw variantoutput.Parameter Description Threshold--min-var-count Min number of var-supporting reads 4--min-var-count-lc Min number of var-supporting reads when depth belowsomaticPdepth2--min-var-freq Min variant allele frequency 0.1--max-somatic-p Max somatic p-value 0.05--max-somatic-p-depth Depth required to test max somatic p-value 10--min-ref-readpos Min average read position of ref-supporting reads 0.1--min-var-readpos Min average read position of var-supporting reads 0.1--min-ref-dist3 Min average distance to effective 3’ end of ref reads 0.1--min-var-dist3 Min average distance to effective 3’ end of variant reads 0.1--min-strandedness Min fraction of variant reads from each strand 0.1--min-strand-reads Min allele depth required to perform the strand tests 5--min-ref-basequal Min average base quality for ref allele 15--min-var-basequal Min average base quality for var allele 15--min-ref-avgrl Min average trimmed read length for ref allele 90--min-var-avgrl Min average trimmed read length for var allele 90--max-rl-diff Max average relative read length difference (ref - var) 0.25--max-ref-mmqs Max mismatch quality sum of ref-supporting reads 100--max-var-mmqs Max mismatch quality sum of var-supporting reads 100--max-mmqs-diff Max average mismatch quality sum (var - ref) 50--min-ref-mapqual Min average mapping quality for ref allele 15--min-var-mapqual Min average mapping quality for var allele 15--max-mapqual-diff Max average mapping quality (ref - var) 502.5.3 Variant annotation and interpretationSnpEff (version 4.2) was used for effect prediction, and the SnpSift package in SnpEff was usedto annotate variants with databases such as dbSNP (b138), COSMIC (version 70), 1000 GenomesProject, and ExAC (release 0.3) for interpretation. Clinical significance reported by the ClinVardatabase and literature review were also used for variant interpretation.30Table 2.5: Spurious variants removed by the variant filtering pipeline.Gene Chr Pos Ref Alt ReasonEGFR chr7 55249112 G A Sequencing artifact, alignment of different sized ampliconsEGFR chr7 55249115 C T Sequencing artifact, alignment of different sized ampliconsKIT chr4 55595593 G A Sequencing artifactKIT chr4 55599268 C T Variant masked by primer in FFPE specimenMAPK1 chr22 22162126 A G Variant masked by primer in FFPE specimenMTHFR chr1 11856378 G A Sequence artifactMTOR chr1 11186783 G A Sequencing artifact within primer regionMTOR chr1 11190646 G A Variant masked by primer in FFPE specimenMTOR chr1 11199428 G T Sequencing artifactMTOR chr1 11269434 C T Sequencing artifactMTOR chr1 11298014 C T Sequencing artifactSTAT3 chr17 40476769 C T Sequencing artifact, alignment of different sized ampliconsTYMP chr22 50964446 A T Poor target region, alignment of different sized ampliconsTYMP chr22 50964862 A T Poor target region, alignment of different sized ampliconsTYMS chr18 673449 G C Alignment error near the indel, chr18:673443c.*447 *452delTTAAAGUGT1A1 chr2 234668879 CAT C Sequencing artifact at poly-AT sequence in promoterUGT1A1 chr2 234668881 T TAC Alignment error/sequencing artifact at poly-AT sequence inpromoter31Figure 2.2: Pipelines for (A) variant calling and (B) filtering.322.6 Sequence analysisA custom Python script was used to process BAM files to quantify the number of on-target aligned(reads that map to target regions), off-target aligned (reads that map to hg19 but not target re-gions), and unaligned reads with a Phred-scaled mapping quality (MAPQ) score ≥ 10. Unalignedreads were also screened against microbial sequences, including viruses, archaea, bacteria, andfungi, to ensure that samples did not contain significant amount of microbial contaminants. Cov-erage depth for target bases with MAPQ ≥ 1 and BAQ ≥ 20 were obtained using bam-readcount(https://github.com/genome/bam-readcount). To measure coverage depth of amplicons, the SAM-tools view function was used to filter for reads with MAPQ ≥ 1 (samtools view -b -q1) followed by the bedtools intersect function (version 2.25.0) to quantify the number ofreads that overlap with amplicon positions (intersect -a $AMPLICON POSITIONS -b$BAM FILE -f 0.85 -r -c).Per-base metrics generated using bam-readcount were also used for assessment of sequenceartifacts. A custom R script was used to count and categorize the different groups of base changes(i.e. C>T/G>A, A>G/T>C, C>A/G>T, A>C/T>G, C>G/G>C, and A>T/T>A). Unless statedotherwise, analysis of sequence artifacts excluded true variants identified by our VarScan2 variantcalling pipeline and base changes with VAF < 1%, which were considered sequencing errors. Allstatistical analyses and data visualization were performed using the R statistical software package(version 3.3.2) and associated open source packages.2.7 Application of VAF thresholds to separate germline alterationsfrom somatic mutationsVariants in the tumours that passed our filtering criteria were subjected to VAF thresholds between10–45%. At each VAF cut-off, variants that were not filtered out were considered predicted germlinevariants. Given that all tumour samples have matched blood samples, true positives were identifiedas predicted germline variants that overlap with variants in the blood (Figure 2.3). Conversely, falsenegatives were identified as variants that were filtered out by the VAF cut-off (predicted as somatic),but were present in the blood samples. Sensitivity at each VAF threshold was calculated by dividingthe number of true positives with the sum of true positives and false negatives. Because predictedgermline variants would be referred to follow-up germline testing, PPVs were calculated at eachVAF cut-off to evaluate precision of our approach. False positives were identified as predictedgermline variants that were absent in the blood, and PPV was calculated by dividing the number oftrue positives with the sum of true positives and false positives.33Figure 2.3: 2x2 contingency table for determination of true positive, false positive, true nega-tive, and false negative variant calls in tumour-only analyses.34Chapter 3Results: Assessment ofFormalin-Induced DNA Damage inFFPE SpecimensTumour biopsies and resections are often FFPE to preserve cellular morphology for pathologicalreview, which is a requirement for standard of care. The FFPE method also enables storage oftissues at room temperature, minimizing cost and mitigating logistical difficulties in storing largearchives of clinical specimens [127]. However, formaldehyde, the main component of formalin,is known to induce DNA damage such as fragmentation and cytosine deamination, which couldaffect the use of FFPE DNA in clinical genomic testing [63, 106, 153, 154, 189, 220, 221]. Wecharacterized formalin-induced DNA damage in our data to assess its impact on the utility of FFPEDNA for germline variant calling. As DNA derived from blood is one of the gold standards forgermline testing, we compared efficiency in amplicon enrichment and sequencing results of FFPEspecimens to blood.3.1 Comparison of efficiency in amplicon enrichment and sequencingresults between blood and FFPE specimensFormalin fixation causes DNA fragmentation that would reduce template DNA for PCR amplifica-tion, leading to decreased efficiency in amplicon enrichment methods for FFPE DNA [61, 63, 220,221]. To investigate this effect, we first compared the amplicon yield between blood and FFPEspecimens, and a Wilcoxon signed-rank test indicated that amplicon yield in FFPE specimens wassignificantly lower than blood specimens (W = 23613, Z = 12.7, p = 8.3×10−62, r = 0.61; Fig-ure 3.1A). However, the amount of DNA input for amplicon enrichment varied across specimensin our study design, and we demonstrated that amplicon yield was weakly correlated with DNA35input for both blood and FFPE specimens (Spearman’s rank correlation: blood, rs = 0.29, 95% CI= 0.16–0.41, p = 2.1×10−5; FFPE, rs = 0.25, 95% CI = 0.12–0.37, p = 2.5×10−4; Figure 3.1B).To account for the difference in DNA input across specimens, we derived the log2 fold changebetween DNA input and amplicon yield (log2 (Amplicon Yield/DNA Input)) to measure the effi-ciency in amplicon enrichment. We compared the log2 fold change in FFPE specimens to blood,and we found a significant decrease in enrichment efficiency in FFPE specimens compared to blood(Wilcoxon signed-rank test, W = 24754, Z = 12.7, p = 4.6×10−57, r = 0.61; Figure 3.1C). This re-sult implies that production of amplicons was less efficient in FFPE specimens compared to blood,demonstrating the drawback of using FFPE DNA in amplicon-based NGS.To examine whether blood and FFPE specimens produce comparable sequencing results, wecompared read alignments between blood and FFPE specimens. Inspection of on-target alignedreads, which are reads that align to target regions used for variant calling, revealed no significant dif-ference in the percentage of on-target aligned reads between blood and FFPE specimens (Wilcoxonsigned-rank test, W = 10178.5, Z = -1.69, p = 0.091, r = -0.081; Figure 3.2). However, there weremore outliers with slightly lower percentage of on-target aligned reads (< 75%) in FFPE specimenscompared to blood, and the distribution of percentage of on-target aligned reads was also wider inFFPE specimens (range: FFPE = 32.5–97.4%, blood = 74.0–95.9%), suggesting more variabilityin the rate of on-target alignment in FFPE specimens than blood. Similarly, no significant differ-ence in the percentage of off-target aligned reads, which are reads that map to the human referencegenome but not to target regions, was observed between specimen types (Wilcoxon signed-ranktest, W = 11494.5, Z = -0.359, p = 0.72, r = -0.017; Figure 3.2). Although a Wilcoxon signed-ranktest indicated that the percentage of unaligned reads was significantly different between blood andFFPE specimens (W = 19069, Z = 7.82, p = 2.4×10−16, r = 0.38; Figure 3.2), there was only asmall decrease in the median percentage of unaligned reads in FFPE specimens compared to blood(median: FFPE = 0.8%, blood = 1.3%). Moreover, our data showed no significant difference inpercentage of contaminant reads between specimen types (W = 14877, Z = 3.29, p = 9.2×10−4,r = 0.16; Figure 3.2), although there was one extreme outlier in FFPE specimens (range: FFPE =0.028–64%, blood = 0.082–8.1%). While there were minor differences in percentage of unalignedreads between sequencing libraries generated from blood and FFPE DNA, blood and FFPE librariesresulted in comparable percentage of on-target aligned reads, thereby providing equivalent amountof aligned reads for variant calling.Although blood and FFPE specimens demonstrated no significant difference in the percentageof on-target aligned reads, this result did not reflect the coverage depth of target regions in bloodand FFPE specimens. To examine whether discrepancy in coverage depth exists between specimentypes, we obtained coverage depth of target bases for all sequencing libraries and normalized perbase coverage depth to account for difference in library size. We derived the average per basecoverage depth for each library and compared this sequencing metric between blood and FFPE36specimens. The average per base coverage depth was significantly different between FFPE andblood specimens (Wilcoxon signed-rank test, W = 20864, Z = 9.76, p = 2.5×10−26, r = 0.66), butthere was only a slight decrease in the average per base coverage depth in FFPE specimens comparedto blood (median: FFPE = 1194, blood = 1271). We also calculated the percentages of target basesthat met coverage thresholds ranging from 0x to 1000x to evaluate coverage uniformity of targetbases between blood and FFPE specimens. While coverage uniformity was significantly differentbetween blood and FFPE specimens at coverage levels except at the 0x and 100x coverage depthcut-off (Wilcoxon signed-rank test, p < 0.0001; Figure 3.3), we considered these discrepancies tobe minor because the absolute difference in median percentage of target bases only exceeded 5%at 500x, 900x, and 1000x coverage thresholds (Table 3.1). Nevertheless, there were more outlierswith lower percentage of target bases than median values in FFPE specimens at coverage thresholdsbetween 100x to 1000x, implying that poor coverage uniformity was more profound for a subset ofFFPE specimens. Together, our findings reveal that FFPE specimens demonstrated lower efficiencyin amplicon enrichment and minor discrepancies in coverage depth and uniformity compared toblood specimens, whereas comparable proportion of on-target read alignments could be attainedbetween specimen types.37Figure 3.1: Comparison of efficiency in amplicon enrichment between blood and FFPE spec-imens. (A) Distributions of amplicon yield in blood and FFPE specimens (Wilcoxonsigned-rank test). Dashed lines indicate median amplicon yield in blood and FFPE spec-imens, which are 299.3 ng and 103.6 ng, respectively. (B) Correlations between ampliconyield and the amount of DNA input for amplicon enrichment in blood and FFPE spec-imens (Spearman’s rank correlation). (C) Distributions of fold change between DNAinput and amplicon yield (log2), which is used to measure efficiency in amplicon enrich-ment in blood and FFPE specimens (Wilcoxon signed-rank test). Dashed lines indicatemedian log2 fold change in blood and FFPE specimens, which are 1.04 and -0.332, re-spectively.38Figure 3.2: Assessment of read alignments between blood and FFPE specimens (Wilcoxonsigned-rank test). Box plots show the median (horizontal bar within) and interquartilerange (IQR) of percentage of reads, with whiskers representing the range of data ≤ 1.5xthe IQR and circles indicating outliers.39Figure 3.3: Evaluation of coverage uniformity in blood and FFPE specimens (Wilcoxonsigned-rank test, ****p < 0.0001, ns = not significant). Per base coverage was normal-ized to account for difference in library size. Percentage of target bases that met variouscoverage thresholds was calculated. Box plots show the median (horizontal bar within)and IQR of percentage of target bases that met the respective coverage thresholds, withwhiskers representing the range of data ≤ 1.5x the IQR and circles indicating outliers.40Table 3.1: Comparison of coverage uniformity between blood and FFPE specimens using theWilcoxon signed-rank test.Blood FFPE TumourThreshold Median (%) Range (%) Median (%) Range (%) D† (%) p (< 0.0001*)≥ 0x 100 100–100 100 97.0–100 0.0 1.0≥ 100x 100 100–100 100 37.0–100 0.0 2.4×10−4≥ 200x 100 100–100 100 29.0–100 0.0 2.9×10−11*≥ 300x 100 98.0–100 99.0 24.0–100 1.0 4.1×10−18*≥ 400x 99.0 94.0–100 97.0 17.0–100 2.0 5.0×10−28*≥ 500x 97.0 84.0–99.0 89.5 13.0–99.0 7.5 2.1×10−38*≥ 600x 92.0 77.0–97.0 87.0 9.0–96.0 5.0 1.5×10−32*≥ 700x 84.0 70.0–91.0 80.0 6.0–91.0 4.0 5.7×10−25*≥ 800x 77.0 63.0-84.0 73.0 5.0–83.0 4.0 4.7×10−27*≥ 900x 73.0 54.0–78.0 66.0 4.0–77.0 7.0 4.6×10−40*≥ 1000x 68.5 41.0–73.0 59.0 3.0-74.0 9.5 3.6×10−42*†Absolute difference between median of blood and FFPE specimens.413.2 Reduced coverage depth in FFPE specimens is more pronouncedfor longer ampliconsThe OncoPanel consists of 416 amplicons that interrogate coding exons and mutational hotpotsof 21 genes, and these amplicons vary in length and GC content. Since we observed discrep-ancy in sequencing coverage between blood and FFPE specimens, we sought to determine whetherthis discrepancy was influenced by amplicon length and GC content. We obtained the coveragedepth for each amplicon and normalized the coverage depth to account for difference in librarysize. We found significant differences in coverage depth between blood and FFPE specimens for336 out of 416 amplicons (Wilcoxon signed-rank test with Benjamini-Hochberg correction, ad-justed p < 0.0001; Figure 3.4). To quantify the amplicon-specific differences in coverage depth,we derived the log2 fold change in the median coverage depth between blood and FFPE speci-mens (log2 (Median CoverageFFPE/Median CoverageBlood)) for each amplicon. Hence, a negativefold change indicates lower coverage depth of the amplicon in FFPE specimens relative to bloodspecimens, whereas a positive fold change indicates higher coverage depth of the amplicon in FFPEspecimens relative to blood specimens. The volcano plot shows that 223 out of the 336 ampliconshad negative log2 fold changes, whereas 113 out of the 336 amplicons had positive log2 fold changes(Figure 3.4). These results indicate that there were differences in coverage depth between FFPE andblood specimens for a large proportion of amplicons in the panel, with more amplicons exhibitinglower coverage depth in FFPE specimens than blood specimens.We subsequently examined the impact of amplicon length and GC content on the amplicon-specific differences in coverage depth between specimen types, which we measured as the log2 foldchange in median coverage depth between blood and FFPE specimens. We first confirmed that nosignificant correlation existed between amplicon GC content and length (Pearson’s correlation, r =0.045, 95% CI = -0.051–0.14, p = 0.36; Figure 3.5). We then evaluated the correlation between log2fold change in amplicon coverage depth and amplicon length, and Pearson’s correlation demon-strated a strong, negative correlation between the two variables (r = -0.77, 95% CI = -0.81– -0.73,p = 1.4×10−82; Figure 3.6A). This result indicates that coverage depth in FFPE specimens tendedto be lower relative to blood specimens as amplicon length increased. On the other hand, coveragedepth tended to be enriched in FFPE specimens relative to blood for shorter amplicons. We alsoassessed the correlation between log2 fold change in amplicon coverage depth and amplicon GCcontent, and Pearson’s correlation demonstrated a weak, negative correlation between the two vari-ables (r = -0.32, 95% CI = -0.41– -0.23, p = 1.8×10−11; Figure 3.6B). Although the correlation isweak, this finding still implies that coverage depth in FFPE specimens tended to be lower relativeto blood specimens as amplicon GC content increased, whereas enriched coverage depth in FFPEspecimens with respect to blood was observed for amplicons with lower GC content.Because amplicon length and GC content demonstrated significant correlations with amplicon-specific differences in coverage depth, we determined which contributing factor had a greater effect.42We used a multiple linear regression to predict log2 fold change in amplicon coverage depth basedon amplicon length and GC content (Table 3.2). A significant equation was found (F(2, 413) =427.6, p = 2.41×10−101), with an adjusted R2 of 0.673. Predicted log2 fold change in ampliconcoverage depth between blood and FFPE specimens is equal to1.63−6.97×10−3(Length)−1.03×10−2(GC Content),in which amplicon length is expressed in base pairs (bp) and GC content is expressed as percent-age (%). Both amplicon length and GC content were significant predictors of log2 fold change inamplicon coverage depth. Based on the standardized coefficients, we compared the strength of pre-dictors within the model to identify the predictor with a greater effect on the response variable. Ourassessment showed that one standard deviation increase in amplicon length would lead to a 0.756standard deviation decrease in log2 fold change in amplicon coverage depth, whereas one standarddeviation increase in amplicon GC content would lead to a 0.288 standard deviation decrease in log2fold change in amplicon coverage depth. This result indicates that amplicon length had a strongerassociation with amplicon-specific differences in coverage depth between specimen types, whichwe measured as the log2 fold change in amplicon coverage depth between blood and FFPE speci-mens, than GC content. Collectively, these findings reveal the challenge imposed by fragmentationdamage in FFPE DNA, which resulted in shorter template DNA that would not be amenable to PCRamplification of longer amplicons.43Figure 3.4: Amplicon-specific differences in coverage depth between blood and FFPE spec-imens. Difference in amplicon coverage depth between specimen types was deter-mined using the Wilcoxon signed-rank test with Benjamini-Hochberg correction (ad-justed p < 0.0001). Volcano plot illustrates the -log10 adjusted p-value in relationto log2 fold change between median coverage depth in blood and FFPE specimens(log2 (Median CoverageFFPE/Median CoverageBlood)) for amplicons in the panel. Nega-tive log2 fold change indicates lower coverage depth of the amplicon in FFPE specimensrelative to blood (↓ CoverageFFPE), whereas positive log2 fold change indicates highercoverage depth of the amplicon in FFPE specimens relative to blood (↑ CoverageFFPE).Color of points represents length of amplicons in base pairs (bp). N = number of ampli-cons44Figure 3.5: The relationship between amplicon GC content and amplicon length (Pearson’scorrelation). Solid line represents the fitted linear relationship between the two variables,and the shaded band indicates pointwise 95% confidence interval of the fitted linear re-gression line.45Figure 3.6: Scatter plots showing log2 fold change between amplicon coverage depth in bloodand FFPE specimens (log2 (Median CoverageFFPE/Median CoverageBlood)) in relation to(A) amplicon length, (B) GC content, C) top 100 longest amplicons, and (D) top 100amplicons with the highest GC content (Pearson’s correlation). Solid line representsthe fitted linear relationship between the two variables, and the shaded band indicatespointwise 95% confidence interval of the fitted linear regression line.46Table 3.2: Multiple linear regression to predict log2 fold change between amplicon coveragedepth in blood and FFPE specimens (log2 (Median CoverageFFPE/Median CoverageBlood))based on amplicon length and GC content.Variable Unstandardized Coefficient Standard Error Standardized Coefficient p-valueLength (bp) −6.97×10−3 2.59×10−4 −7.56×10−1 7.45×10−93GC Content (%) −1.03×10−2 1.01×10−3 −2.88×10−1 4.71×10−22Intercept = 1.63, Adjusted R2 = 0.673F(2, 413) = 427.6, p-value = 2.41×10−101473.3 Deamination effects lead to increased C>T/G>A transitions inFFPE specimensFormalin fixation not only induces DNA fragmentation, but also base modifications that give riseto sequence artifacts [62–64, 94, 106, 153, 154, 221]. A prominent type of formalin-induced se-quence artifact is C>T/G>A transitions as a result of deamination of cytosine bases [63, 106, 125,154, 221]. To measure the level of formalin-induced artifacts in FFPE specimens, we quantifiedthe fraction of base changes that were not identified as true single nucleotide variants (SNVs)by our variant calling pipeline. We only considered high quality bases (BAQ ≥ 20) and basechanges that were ≥ 1% allele frequency to exclude sequencing errors from our analysis. Basechanges were categorized into C>T/G>A and A>G/T>C, which are nucleotide transitions, as wellas C>A/G>T, A>C/T>G, C>G/G>C, and A>T/T>A, which are nucleotide transversions. Wecompared the fraction of base changes between specimen types and found significant differences infraction of C>T/G>A and A>G/T>C between blood and FFPE specimens (Wilcoxon signed ranktest, p < 0.0001; Figure 3.7A). As blood DNA is not affected by formalin fixation, we evaluatedthe prevalence of artifactual base changes in FFPE specimens with respect to blood by calculat-ing the fold change between the median fraction of base changes in blood and FFPE specimens(Table 3.3). We noted a substantially larger fold change for C>T/G>A compared to A>G/T>C:fraction of C>T/G>A was 23 times higher in FFPE specimens relative to blood, whereas fractionof A>G/T>C was 3.1 times higher in FFPE specimens relative to blood. Increased C>T/G>Aartifacts was consistent with cytosine deamination effects that are reportedly predominant in FFPEDNA. On the other hand, A>G/T>C artifacts could be caused by deamination of adenine to gen-erate hypoxanthine, which forms base pairs with cytosine instead of thymine, changing A-T basepairs to G-C base pairs. Deamination of adenine to hypoxanthine can be catalyzed by an acidicenvironment [214], which can arise in FFPE specimens because formaldehyde can be oxidized togenerate formic acid [63].To assess the relative difference in fraction of base changes in FFPE specimens compared toblood specimens, we calculated the log2 fold change in fraction of base changes between pairedblood and FFPE specimens (log2 (Fraction of Base ChangesFFPE/Fraction of Base ChangesBlood)).We compared the relative difference in fraction of base changes across different types of basechanges, and a Kruskal-Wallis test indicated that type of base changes had a significant effecton the relative difference in fraction of base changes (H = 428.5, p = 2.1×10−90; Figure 3.8).Multiple pairwise comparison of the relative difference in fraction of base changes was performedusing a post-hoc Dunn’s test with Benjamini-Hochberg correction. Relative difference in fractionof C>T/G>A was significantly different compared to the five other types of base changes, andthis was similar for A>G/T>C (adjusted p < 0.0001; Table 3.4). Although both C>T/G>A andA>G/T>C were elevated in FFPE specimens compared to the other base transversions, the magni-tude of difference was larger for C>T/G>A than A>G/T>C (median log2 fold change: C>T/G>A48= 4.2, A>G/T>C = 1.6), which further confirmed that deamination of cytosine bases is the mostfrequent form of sequence artifact in FFPE DNA.Formalin-induced sequence artifacts often occur at low allele frequency; hence, we examinedthe prevalence of sequence artifacts at different ranges of allele frequency, including 1–10%, 10-20%, and 20-30%. Because variants were not called within the 1–10% allele frequency range,we did not remove true SNVs detected by our variant calling pipeline to ensure consistency whencomparing fraction of base changes across different ranges of allele frequency. Nevertheless, weadhered to the previous criterion of only including base changes with BAQ ≥ 20 in this analysis.For all types of base changes, we noted that the range of allele frequency had a significant effect onfraction of base changes in blood and FFPE specimens (Kruskal-Wallis test, p < 0.0001; Figure 3.9),with increased levels of base changes at the 1-10% allele frequency range compared to 10-20%and 20-30%. Because blood DNA represents good quality DNA that is unaffected by formalinfixation, we also compared the fraction of base changes at the 1-10% allele frequency range inFFPE specimens to blood. Similar to previous analyses (Figure 3.7; Table 3.3), there was a markedincrease in C>T/G>A and a modest increase in A>G/T>C in FFPE specimens relative to bloodwithin the 1-10% allele frequency (fold change: C>T/G>A = 33, A>G/T>C = 3.1; Table 3.5).Collectively, our assessment demonstrates that high frequency of C>T/G>A transition was presentand detectable in FFPE specimens, which indicated that deamination of cytosine is the primary formof formalin-induced sequence artifact, and these artifactual transitions were more prevalent at low,but clinically relevant allele frequency.49Figure 3.7: Assessment of formalin-induced sequence artifacts in FFPE specimens. (A) Com-parison of fraction of base changes in blood and FFPE specimens (Wilcoxon signed-rank test). Box plots show the median (horizontal bar within) and IQR of fraction ofbase changes for different types of base changes, with whiskers representing the rangeof data ≤ 1.5x the IQR and circles indicating outliers. (B) Box plots showing squareroot-transformed fraction of base changes on the Y-axis.50Table 3.3: Summary statistics of fraction of base changes in blood and FFPE specimens.Blood FFPE TumourBase Changes Median Range Median Range FC†C>T/G>A 8.9×10−5 0–2.6×10−3 2.0×10−3 0–8.6×10−2 23A>G/T>C 1.2×10−4 0–1.3×10−3 3.7×10−4 0–5.0×10−3 3.1C>A/G>T 6.0×10−5 0–1.8×10−4 6.0×10−5 0–2.6×10−3 1.0A>C/T>G 2.4×10−4 5.9×10−5–5.3×10−4 1.8×10−4 5.8×10−5–1.4×10−3 0.77C>G/G>C 1.2×10−4 0–2.4×10−4 1.2×10−4 0–4.8×10−4 1.0A>T/T>A 6.0×10−5 0–8.4×10−4 5.9×10−5 0–8.6×10−4 0.99†Fold change (FC) between the median of blood and FFPE specimens.Figure 3.8: Comparison of relative difference in fraction of base changes in FFPE spec-imens compared to blood (Kruskal-Wallis test). Relative difference was measuredas log2 fold change between fraction of base changes in blood and FFPE specimens(log2 (Fraction of Base ChangesFFPE/Fraction of Base ChangesBlood)). Box plots showthe median (horizontal bar within) and IQR of log2 fold change for different types ofbase changes, with whiskers representing the range of data ≤ 1.5x the IQR and circlesindicating outliers.51Table 3.4: Multiple pairwise comparison of log2 fold change in fraction of base changes be-tween blood and FFPE specimens using Dunn’s test with Benjamini-Hochberg multiplehypothesis testing correction. Top values represent Dunn’s pairwise z statistics, whereasbottom values represent adjusted p-value. Asterisk(*) indicates significance level of ad-justed p-value < 0.0001.Base Changes A>C/T>G A>G/T>C A>T/T>A C>A/G>T C>G/G>CA>G/T>C -11.74.15×10−31*A>T/T>A -0.399 9.573.45×10−1 1.31×10−21*C>A/G>T -3.46 6.39 -2.734.00×10−4 1.52×10−10* 3.99×10−3C>G/G>C -3.02 8.63 -2.17 0.9181.73×10−3 6.76×10−18* 1.71×10−2 1.92×10−1C>T/G>A -17.1 -5.60 -14.3 -11.1 -14.17.78×10−65* 1.76×10−8* 5.10×10−46* 1.32×10−28* 6.46×10−45*5253Figure 3.9: Assessment of formalin-induced sequence artifacts in FFPE specimens at differentranges of allele frequency. (A) Comparison of fraction of base changes across differentranges of allele frequency (Kruskal-Wallis test). Box plots show the median (horizontalbar within) and IQR of fraction of base changes for different types of base changes, withwhiskers representing the range of data ≤ 1.5x the IQR and circles indicating outliers.(B) Box plots demonstrating square root-transformed fraction of base changes acrossdifferent ranges of allele frequency. Dashed lines equal to 0.05 to indicate that the Y-axisscales are different for blood and FFPE tumour plots.54Table 3.5: Summary statistics of fraction of base changes in blood and FFPE specimens within1-10% allele frequency.Blood FFPE TumourBase Changes Median Range Median Range †FCC>T/G>A 6.2×10−5 0–2.6×10−3 2.0×10−3 0–8.6×10−2 33A>G/T>C 1.2×10−4 0–1.3×10−3 3.7×10−4 0–4.9×10−3 3.1C>A/G>T 6.0×10−5 0–2.4×10−4 6.0×10−5 0–1.9×10−3 1.0A>C/T>G 2.4×10−4 5.9×10−5–4.8×10−4 1.8×10−4 0–1.1×10−3 0.77C>G/G>C 1.2×10−4 0–2.4×10−4 1.2×10−4 0–4.8×10−4 1.0A>T/T>A 6.0×10−5 0–8.4×10−4 5.9×10−5 0–8.6×10−4 0.99†Fold change (FC) between the median of blood and FFPE specimens.553.4 Increased age of paraffin block results in reduced amplicon yieldand elevated level of C>T/G>A sequence artifactsThe amount of amplifiable DNA derived from FFPE specimens is dependent on the extent of frag-mentation damages. Given two FFPE DNA samples of similar quantity, the sample with moreextensive DNA fragmentation would yield reduced amount of PCR amplicons compared to the lessfragmented sample [61, 221]. As the age of paraffin blocks in our study ranged from 18 to 5356days, we hypothesized that older paraffin blocks would result in more extensively fragmented DNA,leading to reduced efficiency in amplicon enrichment. Inspection of the relationship between am-plicon yield and age of paraffin block demonstrated a moderate, negative correlation (Spearman’srank correlation, rs = -0.42, 95% CI = -0.52– -0.30, p = 1.2×10−10; Figure 3.10A), suggesting thatDNA extraction from older paraffin blocks tended to yield lower amount of amplicons. Because theamount of DNA input for production of amplicons varied across specimens in our study design, arepresentation of efficiency in amplicon enrichment would be the log2 fold change between DNA in-put and amplicon yield. Thus, we assessed the correlation between log2 fold change and the storagetime of FFPE blocks. There was a moderate, negative correlation between log2 fold change and ageof paraffin block (Spearman’s rank correlation, rs = -0.42, 95% CI = -0.53– -0.30, p = 1.2×10−10;Figure 3.10B), implying that production of amplicons was less efficient in FFPE DNA extractedfrom older paraffin blocks, which was likely caused by more substantial DNA fragmentation.As DNA fragmentation results in reduced template DNA for PCR amplification, this leads to ahigher probability for enrichment of sequence artifacts [38, 221]. Our previous evaluation indicatedthat older paraffin blocks were associated with lower efficiency in amplicon enrichment, whichwas possibly due to increased fragmentation damages in the extracted DNA (Figure 3.10). Thisled to our hypothesis that older paraffin blocks would yield elevated levels of sequence artifacts,particularly C>T/G>A transitions, which are the most prominent type of formalin-induced basemodifications. To address our hypothesis, we assessed the relationship between fraction of basechanges and age of paraffin blocks for different types of base changes (Figure 3.11). There was amoderate, positive correlation between fraction of C>T/G>A transitions and age of paraffin block(Spearman’s rank correlation, rs = 0.54, 95% CI = 0.43–0.63, p = 1.0×10−17). We also noteda positive correlation between fraction of A>G/T>C and age of paraffin block (Spearman’s rankcorrelation, rs = 0.29, 95% CI = 0.16–0.40, p = 2.1×10−5), albeit a weak one. As for transversionbase changes (i.e. C>A/G>T, A>C/T>G, C>G/G>C, and A>T/T>A), no significant correlationswith age of paraffin block were observed (Spearman’s rank correlation, p < 0.05). These findingsreveal that increased detection of sequence artifacts, especially the common C>T/G>A changes inFFPE specimens, was associated with long term storage of FFPE blocks.We subsequently examined how pre-sequencing variables such as age of paraffin block andefficiency in amplicon enrichment correlate with sequencing metrics like average per base cover-age (normalized to account for library size), percentage of on-target alignments, and fraction of56C>T/G>A changes (Table 3.6). This assessment would provide insight on how pre-sequencingvariables can affect sequencing results, thereby facilitating sample selection if multiple specimenswere available before sequencing. We noted a moderate, negative correlation between average perbase coverage and age of paraffin block (Spearman’s rank correlation, rs = -0.47, 95% CI = -0.57–-0.36, p = 2.0×10−13), and a weak, negative correlation between percentage of on-target alignedreads and age of paraffin block (Spearman’s rank correlation, rs = -0.35, 95% CI = -0.46– -0.23, p= 1.4×10−7). Conversely, we observed a moderate, positive correlation between average per basecoverage and efficiency in amplicon enrichment (Spearman’s rank correlation, rs = 0.52, 95% CI= 0.42–0.61, p = 3.8×10−17), and a weak, positive correlation between percentage of on-targetaligned reads and efficiency in amplicon enrichment (Spearman’s rank correlation, rs = 0.34, 95%CI = 0.22–0.45, p = 1.2×10−7). Since efficiency in amplicon enrichment was inversely correlatedwith storage time of FFPE blocks, opposing correlations with sequencing metrics were expectedfor both pre-sequencing variables. Furthermore, there was also a moderate, negative correlation be-tween fraction of C>T/G>A and efficiency in amplicon enrichment (Spearman’s rank correlation,rs = -0.55, 95% CI = -0.64– -0.45, p = 1.4×10−18). As reduced efficiency in amplicon enrichment isan indicator for low amount of template DNA, the consequent increase in C>T/G>A changes couldbe the outcome of stochastic enrichment of sequence artifacts. Together, these results reveal thatpre-sequencing variables such as age of paraffin block and efficiency in amplicon enrichment couldbe predictors of sequencing metrics, in which older FFPE blocks were more likely to yield lowerefficiency in amplicon enrichment, leading to poorer sequencing results and increased prevalenceof artifactual C>T/G>A transitions.57Figure 3.10: Scatter plots showing (A) amplicon yield and (B) efficiency in amplicon enrich-ment, which is represented by the log2 fold change between the amount of DNA inputfor producing amplicons and amplicon yield, in relation to age of paraffin blocks (Spear-man’s rank correlation). Solid lines represent locally weighted smoothing (LOESS)curves, with shaded bands indicating 95% confidence interval of the LOESS curves.Figure 3.11: The relationship between fraction of base changes and age of paraffin block fordifferent types of base changes (Spearman’s rank correlation).58Table 3.6: Spearman’s rank correlation between pre-sequencing variables (e.g. enrichment efficiency and age of paraffin block) andsequencing metrics (e.g. fraction of C>T/G>A, average per base normalized coverage, and on-target aligned reads). Top valuesrepresent Spearman’s rho and 95% confidence interval in brackets, whereas bottom values represent p-value. Asterisk(*) indicatessignificance level of p-value < 0.05.Variable Enrichment Age of Paraffin Fraction of Average Per BaseEfficiency† Block (Day) C>T/G>A Normalized CoverageAge of Paraffin -0.42 (-0.53– -0.30)Block (Day) 1.2×10−10*Fraction of -0.55 (-0.64– -0.45) 0.54 (0.43–0.63)C>T/G>A 1.4×10−18* 1.0×10−17*Average Per Base 0.52 (0.42–0.61) -0.47 (-0.57– -0.36) -0.80 (-0.84– -0.75)Normalized Coverage 3.8×10−17* 2.0×10−13* 1.3×10−48*On-target 0.34 (0.22–0.45) -0.35 (-0.46– -0.23) -0.57 (-0.65– -0.47) 0.73 (0.66–0.79)Aligned Reads (%) 1.2×10−7* 1.4×10−7* 2.2×10−20* 0*†log2 fold change between DNA input for amplicon enrichment and amplicon yield.59Chapter 4Results: Identification of GermlineAlterations in FFPE TumoursClinical tumour sequencing can guide disease management and therapeutic intervention for can-cer patients. Although simultaneous identification of clinically important germline variants couldbe performed if tumour-normal pairs are sequenced, matched normal samples are not routinelyprocessed in the clinical setting due to logistical constraints and limited funding and time [77,78, 125, 219]. Genomic analyses of tumours can reveal both germline and somatic alterations[27, 102, 140, 141, 184]. However, this requires a method to separate germline variants from so-matic mutations. The TOP study is comprised of 213 patients with tumour and matched bloodspecimens. This enabled us to measure the retention rate of germline alterations in tumour DNA toconfirm that tumour DNA is a reliable source for identification of germline alterations. We also eval-uated the use of VAF thresholds to distinguish between germline and somatic statuses of variantsidentified in the tumour DNA. Our goal was to establish the sensitivities for identifying potentialgermline alterations and PPVs for referring germline alterations to downstream germline testing atvarious VAF cut-offs. These parameters could serve as guidelines in the clinical practice.4.1 Frequency and interpretation of germline alterations in patientsfrom the TOP cohortWe examined 15 cancer-related genes and six PGx genes in DNA isolated from blood samples fromthe 213 cancer patients in the TOP cohort. We identified a total of 1990 germline alterations thatpassed our filtering criteria (Figure 2.2B). In 212 out of 213 patients, we detected a total of 1205variants in the 15 cancer-related genes screened by the OncoPanel, with an average of 5.7 variantsper patient (standard error = 0.15, range = 1–11 variants; Table 4.1). These germline alterationswere found at 50 genomic positions and interpreted using various bioinformatics approaches andliterature review (Table 4.2). Through effect prediction using the SnpEff software, we demonstrated60that 39/50 of these variants were synonymous, 8/50 were missense variants, 2/50 occurred withinsplice regions, and 1/50 were frameshift variants. Based on the ExAC database, 28/50 germlinevariants had population allele frequency < 1%, whereas 18/50 germline variants had populationallele frequency ≥ 1%. Four out of 50 germline variants were not reported in the ExAC database.To assess clinical significance of the 50 germline alterations in cancer-related genes, we usedinformation in the ClinVar database. Our assessment revealed 8/50 benign variants, 8/50 likelybenign variants, 6/50 annotated as benign/likely benign, 2/50 with conflicting interpretations ofpathogenicity, and 1/50 with uncertain significance. We were unable to determine the clinicalsignificance of 24/50 germline variants because these variants were not reported in the ClinVardatabase. While we found no variants that were pathogenic or likely pathogenic, we identified oneTP53 variant, p.Arg72Pro/c.215G>C (rs1042522), that is associated with drug response. Based onliterature review, clinical studies revealed that the Pro/Pro genotype results in severe neutropeniain ovarian cancer patients receiving cisplatin-based chemotherapy, and poor survival and treatmentresponse in gastric cancer patients receiving paclitaxel and capecitabine combination chemotherapyand 5-fluorouracil-based adjuvant chemotherapy [25, 28, 31, 46, 97, 103, 105, 224, 226, 227, 230,231]. The combination of evidence from our literature review and the ClinVar database suggeststhat the TP53 p.Arg72Pro/c.215G>C (rs1042522) could be potentially useful in guiding clinicalmanagement for cancer patients.Furthermore, we identified a total of 785 variants in the six PGx genes screened by the Onco-Panel in 212 out of 213 patients, with an average of 3.7 germline alterations per patient (standarderror = 0.10, range = 1–8 variants; Table 4.3). These PGx variants occurred at 23 genomic positionsand were interpreted using similar methods to the germline alterations identified in cancer-relatedgenes (Table 4.4). Effect prediction using the SnpEff software demonstrated that 13/23 germlinevariants were missense variants, 4/23 were synonymous, 2/23 occurred within splice regions, 2/23occurred upstream of a gene, 1/23 were located at splice donor sites, and 1/23 were present at the3’ untranslated region. Based on the ExAC database, 8/23 germline variants had population allelefrequency < 1%, whereas 10/23 germline variants had population allele frequency ≥ 1%. Five outof 23 germline variants were not reported in the ExAC database.We also assessed clinical significance of the germline alterations in the PGx genes using theClinVar database. This assessment demonstrated that 5/23 variants were categorized as either be-nign or likely benign, 4/23 with conflicting interpretations of pathogenicity, 2/23 submitted withoutassessment of clinical significance, and 1/23 with uncertain significance. There was also 4/23 ofvariants that were not reported in the ClinVar database. Although our analysis showed no variantsthat were pathogenic or likely pathogenic in the PGx genes, we identified 7/23 germline alterationsthat were associated with drug response. These alterations were DPYD p.Asp949Val/c.2846A>T(rs67376798), c.1906G>A (rs3918290), p.Met166Val/c.496A>G (rs2297595), GSTP1 p.Ile105Val/c.313A>G (rs1695), MTHFR p.Glu429Ala/c.1286A>C (rs1801131), p.Ala222Val/c.665C>T61(rs1801133), and TYMS c.*447 *452delTTAAAG (rs151264360), which could serve as predictorsfor response to chemotherapy. While the germline variants in DPYD, MTHFR, and TYMS are as-sociated with fluoropyrimidine-related toxicities, the germline variant in GSTP1 is associated withadverse drug reactions in response to oxaliplatin treatment [144, 158].Overall, we found an average of 5.7 variants per patient in cancer-related genes and an av-erage of 3.7 variants per patient in PGx genes in the TOP cohort. Our assessment also revealedgermline alterations at 50 and 23 genomic positions in cancer-related and PGx genes, respectively.While annotation with the ClinVar database did not identify any pathogenic or likely pathogenicgermline alterations, this analysis revealed a total of eight variants (one in a cancer-related geneand seven in PGx genes) that could serve as predictors for drug response. We showed that theTP53 p.Arg72Pro/c.215G>C (rs1042522) was present in 97 out of 213 patients (46%), and 208out of 213 (98%) TOP patients had at least one germline PGx variant that was associated with drugresponse (Figure 4.1; Figure 4.2).62Table 4.1: Frequency of germline variants in cancer-related genes in blood specimens fromTOP patients.Gene Chr Pos ID? HGVS* Zygosity Total Pct‡ (%)wt-var†, var-var††ALK 2 29443662 NA p.Val1185Valc.3555G>A1, 0 1 0.5EGFR 7 55242453 NA p.Pro741Proc.2223C>T1, 0 1 0.57 55242500 COSM133588 p.Lys757Argc.2270A>G2, 0 2 0.97 55249063 rs1050171;COSM1451600p.Gln787Glnc.2361G>A96, 60 156 737 5524915 rs56183713;COSM13400p.Val819Valc.2457G>A2, 0 2 0.97 55259450 rs2229066;COSM85893;rs17290559p.Arg836Argc.2508C>T9, 0 9 4KIT 4 55592059 rs151016327;COSM3760661p.Thr461Thrc.1383A>G2, 0 2 0.94 55599268 rs55789615;COSM1307p.Ile798Ilec.2394C>T14, 0 14 74 55602765 rs3733542;COSM1325p.Leu862Leuc.2586G>C37, 3 40 18MAPK1 22 22162126 rs386488966;rs3729910p.Tyr43Tyrc.129T>C13, 1 14 722 22221623 rs201495639 p.Tyr36Tyrc.108C>T3, 0 3 1MTOR 1 11169420 rs41274506 p.Asp2485Aspc.7455C>T1, 0 1 0.51 11172909 NA p.Glu2456Lysc.7366G>A1, 0 1 0.51 11174452 NA p.Arg2408Glnc.7223G>A1, 0 1 0.51 11181327 rs11121691 p.Leu2303Leuc.6909G>A70, 6 76 361 11184593 rs56051835 p.Leu2208Leuc.6624T>C2, 0 2 0.963Gene Chr Pos ID? HGVS* Zygosity Total Pct‡ (%)wt-var†, var-var††1 11188172 rs370318222 p.Tyr1974Tyrc.5922C>T1, 0 1 0.51 11190646 rs2275527 p.Ser1851Serc.5553C>T65, 0 65 311 11190730 rs17848553 p.Ala1823Alac.5469C>T8, 0 8 0.51 11194521 COSM180791 c.5133C>T 1, 0 1 0.51 11205058 rs386514433;rs1057079p.Ala1577Alac.4731A>G81, 12 93 441 11269506 NA p.Leu1222Phec.3664C>T1, 0 1 0.51 11272468 rs17036536 p.Arg1154Argc.3462G>C8, 0 8 41 11288758 rs1064261 p.Asn999Asnc.2997T>C85, 0 85 401 11298038 rs55752564 p.Ala690Alac.2070G>A1, 0 1 0.51 11298640 rs55881943 p.Ala607Alac.1821G>A1, 0 1 0.51 11301714 rs1135172 p.Asp479Aspc.1437T>C80, 114 194 921 11308007 rs35903812 p.Ala329Thrc.985G>A3, 0 3 11 11316244 rs12120294 p.Leu170Leuc.510G>C1, 0 1 0.5PDGRRA 4 55141055 rs1873778;COSM1430082p.Pro567Proc.1701A>G0, 183 183 864 55152040 rs2228230;COSM22413p.Val824Valc.2472C>T57, 5 62 29STAT1 2 191851646 rs41270237 p.Thr385Thrc.1155G>A2, 0 2 0.92 191856001 rs41509946 p.Gln330Glnc.990G>A3, 0 3 164Gene Chr Pos ID? HGVS* Zygosity Total Pct‡ (%)wt-var†, var-var††2 191859906 rs61756197 p.Gln275Glnc.825G>A1, 0 1 0.92 191859935 rs41473544 p.Val266Ilec.796G>A2, 0 2 0.92 191872307 rs45463799 p.Asn118Asnc.354C>T3, 0 3 12 191874667 rs386556119;rs2066802p.Leu21Leuc.63T>C42, 3 45 21STAT3 17 40469241 COSM979464 c.2100C>T 1, 0 1 0.517 40475056 rs117691970 p.Gly618Glyc.1854C>T4, 0 4 217 40486040 rs200098006 p.Leu275Leuc.825T>G2, 0 2 0.917 40486043 NA p.Gln274Glnc.822A>G1, 0 1 0.517 40498635 rs146184566;COSM979479p.Ser75Serc.225G>A1, 0 1 0.517 40498713 NA p.Lys49Lysc.147A>G1, 0 1 0.517 40498722 NA p.Ala46Alac.138G>T1, 0 1 0.5TP53 17 7577069 rs55819519;COSM44017p.Arg290Hisc.869G>A1, 0 1 0.517 7577553 COSM44368 p.Met243fsc.727delA1, 0 1 0.517 7578210 rs1800372;COSM249885p.Arg213Argc.639A>G1, 0 1 0.517 7578420 COSM1386804 p.Thr170Thrc.510G>A1, 0 1 0.517 7579472 rs1042522;COSM250061p.Arg72Proc.215G>C73,24 97 4617 7579579 rs1800370 p.Pro36Proc.108G>A5, 0 5 265Gene Chr Pos ID? HGVS* Zygosity Total Pct‡ (%)wt-var†, var-var††Total variants in cancer-related genes = 1205Average number of variants per patient = 5.7Standard error = 0.15?dbSNP and/or COSMIC IDs.*Description of sequence variants according to the HGVS recommendations.†wt-var represents heterozygous variant.††var-var represents homozygous variant.‡Percentage of patients with the variant.66Table 4.2: Interpretation of germline alterations in cancer-related genes detected in blood specimens of TOP patients.Gene Chr:Pos ID? HGVS* AF** VariantEffect†ClinicalSignificance††Functional/Clinical Impacts Ref.ALK 2:29443662 NA p.Val1185Valc.3555G>A0.00082 Syn. NA NA NAEGFR 7:55242453 NA p.Pro741Proc.2223C>T0.0074 Syn. NA NA NA7:55242500 COSM133588 p.Lys757Argc.2270A>G0.00082 Missense UncertainsignificanceHomozygous mutation was iden-tified in a patient with intra-hepatic cholangiocarcinoma, lead-ing to activation of downstreamEGFR pathways as demonstratedby MAPK and Akt phosphoryla-tions.[117]7:55249063 rs1050171;COSM1451600‡p.Gln787Glnc.2361G>A52 Syn. Benign/LikelybenignConflicting evidence on predictiveand prognostic values in lung can-cer patients. Poorer response toanti-EGFR therapy in colorectalcancer patients compared to pa-tients with the GG genotype.[29,116,215,228]67Gene Chr:Pos ID? HGVS* AF** VariantEffect†ClinicalSignificance††Functional/Clinical Impacts Ref.7:5524915 rs56183713;COSM13400p.Val819Valc.2457G>A0.035 Syn. Likely benign One study reported that this vari-ant in combination with rs1050171was correlated with TNM stage ofsquamous cell lung carcinoma.[215]7:55259450 rs2229066;COSM85893;rs17290559p.Arg836Argc.2508C>T1.7 Syn. Benign/LikelybenignNA NAKIT 4:55592059 rs151016327;COSM3760661p.Thr461Thrc.1383A>G0.28 Syn. Benign NA NA4:55599268 rs55789615;COSM1307p.Ile798Ilec.2394C>T2.1 Syn. Benign/LikelybenignNA NA4:55602765 rs3733542;COSM1325p.Leu862Leuc.2586G>C12 Syn. Benign/LikelybenignNA NAMAPK1 22:22162126 rs386488966;rs3729910p.Tyr43Tyrc.129T>C4.5 Syn. NA NA NA22:22221623 rs201495639 p.Tyr36Tyrc.108C>T0.052 Syn. NA NA NA68Gene Chr:Pos ID? HGVS* AF** VariantEffect†ClinicalSignificance††Functional/Clinical Impacts Ref.MTOR 1:11169420 rs41274506 p.Asp2485Aspc.7455C>T0.33 Syn. NA NA NA1:11172909 NA p.Glu2456Lysc.7366G>A0.00082 Missense NA NA NA1:11174452 NA p.Arg2408Glnc.7223G>ANA Missense NA NA NA1:11181327 rs11121691 p.Leu2303Leuc.6909G>A22 Syn. NA Likely has an effect on exonicsplicing enhancer or exonic splic-ing silencer binding site activity.[233]1:11184593 rs56051835 p.Leu2208Leuc.6624T>C0.49 Syn. Benign NA NA1:11188172 rs370318222 p.Tyr1974Tyrc.5922C>T0.00082 Syn. NA NA NA1:11190646 rs2275527 p.Ser1851Serc.5553C>T22 Syn. Benign NA NA1:11190730 rs17848553 p.Ala1823Alac.5469C>T2.4 Syn. Benign NA NA69Gene Chr:Pos ID? HGVS* AF** VariantEffect†ClinicalSignificance††Functional/Clinical Impacts Ref.1:11194521 COSM180791 c.5133C>T 0.029 SpliceregionNA NA NA1:11205058 rs386514433;rs1057079‡p.Ala1577Alac.4731A>G32 Syn. NA One study reported improved clin-ical response and progression-freesurvival in advanced esophagealsquamous cell carcinoma patientswith the AG genotype comparedto the AA genotype who weretreated with paclitaxel plus cis-platin chemotherapy.[126]1:11269506 NA p.Leu1222Phec.3664C>T0.00082 Missense NA NA NA1:11272468 rs17036536 p.Arg1154Argc.3462G>C1.8 Syn. Benign NA NA70Gene Chr:Pos ID? HGVS* AF** VariantEffect†ClinicalSignificance††Functional/Clinical Impacts Ref.1:11288758 rs1064261‡ p.Asn999Asnc.2997T>C26 Syn. NA C allele likely influences exonicsplicing enhancer or exonic splic-ing silencer binding site activity ordisrupts a protein domain. Meta-analysis found no association withcancer risk.[233]1:11298038 rs55752564 p.Ala690Alac.2070G>A0.077 Syn. NA NA NA1:11298640 rs55881943 p.Ala607Alac.1821G>A0.017 Syn. ConflictinginterpretationsofpathogenicityNA NA1:11301714 rs1135172‡ p.Asp479Aspc.1437T>C72 Syn. NA NA NA1:11308007 rs35903812 p.Ala329Thrc.985G>A0.27 Missense Likely benign NA NA1:11316244 rs12120294 p.Leu170Leuc.510G>C0.36 Syn. NA NA NA71Gene Chr:Pos ID? HGVS* AF** VariantEffect†ClinicalSignificance††Functional/Clinical Impacts Ref.PDGFRA 4:55141055 rs1873778;COSM1430082‡p.Pro567Proc.1701A>G99 Syn. Benign No association with PDGFRαexpression in colorectal cancer.[73]4:55152040 rs2228230;COSM22413p.Val824Valc.2472C>T18 Syn. Benign NA NASTAT1 2:191851646 rs41270237 p.Thr385Thrc.1155G>A0.42 Syn. Likely benign NA NA2:191856001 rs41509946 p.Gln330Glnc.990G>A0.36 Syn. Likely benign NA NA2:191859906 rs61756197 p.Gln275Glnc.825G>A0.025 Syn. NA NA NA2:191859935 rs41473544 p.Val266Ilec.796G>A0.20 Missense Likely benign Functional testing indicated thatthe variant was not a gain-of-function mutation in STAT1[60]2:191872307 rs45463799 p.Asn118Asnc.354C>T0.32 Syn. Likely benign NA NA72Gene Chr:Pos ID? HGVS* AF** VariantEffect†ClinicalSignificance††Functional/Clinical Impacts Ref.2:191874667 rs386556119;rs2066802p.Leu21Leuc.63T>C8.5 Syn. Benign High frequency among patientswith multiple sclerosis and chronichepatitis C.[76]STAT3 17:40469241 COSM979464 c.2100C>T NA SpliceregionNA NA NA17:40475056 rs117691970 p.Gly618Glyc.1854C>T0.37 Syn. Likely benign NA NA17:40486040 rs200098006 p.Leu275Leuc.825T>G0.066 Syn. NA NA NA17:40486043 NA p.Gln274Glnc.822A>G0.00082 Syn. NA NA NA17:40498635 rs146184566;COSM979479p.Ser75Serc.225G>A0.029 Syn. Likely benign NA NA17:40498713 NA p.Lys49Lysc.147A>G0.012 Syn. NA NA NA17:40498722 NA p.Ala46Alac.138G>TNA Syn. NA NA NA73Gene Chr:Pos ID? HGVS* AF** VariantEffect†ClinicalSignificance††Functional/Clinical Impacts Ref.TP53 17:7577069 rs55819519;COSM44017p.Arg290Hisc.869G>A0.016 Missense ConflictinginterpretationsofpathogenicityA conservative amino acid substi-tution that was predicted to be pos-sibly damaging by in silico anal-ysis. Reported in patients withLi-Fraumeni syndrome and can-cer patients without family histo-ries of Li-Fraumeni syndrome orLi-Fraumeni-like syndrome.[11,12, 47,160,162,210]17:7577553 COSM44368 p.Met243fsc.727delANA Frameshift NA Reported in esophageal squamouscell carcinoma of patients fromnorthern Iran.[20]17:7578210 rs1800372;COSM249885p.Arg213Argc.639A>G1.2 Syn. Benign/LikelybenignOne study demonstrated thatthis variant was not a predictivebiomarker for initiation and pro-gression of gastroesophageal refluxdisease, Barrett’s Esophagus, andesophageal cancer in the Brazilianpopulation.[161]74Gene Chr:Pos ID? HGVS* AF** VariantEffect†ClinicalSignificance††Functional/Clinical Impacts Ref.17:7578420 COSM1386804 p.Thr170Thrc.510G>A0.012 Syn. NA One study reported that TP53 mu-tations in exon 5, which includethis variant, were associated withthe worst prognosis for patientswith non-small-cell lung cancer.[209]17:7579472 rs1042522;COSM250061‡p.Arg72Proc.215G>C34 Missense Drug response p53 protein with Arg72 wasassociated with increased apop-tosis, while p53 protein withPro72 demonstrated increased G1cell-cycle arrest and activationof p53-dependent DNA repair.Pro/Pro genotype resulted insevere neutropenia in ovariancancer patients receiving cisplatin-based chemotherapy, and poorsurvival and treatment response ingastric cancer patients receivingpaclitaxel and capecitabine com-bination chemotherapy, as wellas 5-fluorouracil-based adjuvantchemotherapy. Conflicting evi-dence on risk of predispositon tovarious cancer types.[25,28, 31,46, 97,103,105,224,226,227,230,231]75Gene Chr:Pos ID? HGVS* AF** VariantEffect†ClinicalSignificance††Functional/Clinical Impacts Ref.17:7579579 rs1800370 p.Pro36Proc.108G>A1.3 Syn. Benign/LikelybenignNA NA?dbSNP and/or COSMIC IDs.*Description of sequence variants according to the Human Genome Variation Society (HGVS) recommendations.**AF = Allele frequency reported by the Exome Aggregation Consortium (ExAC) and presented in percentage.†Effect of genetic variants as predicted by the SnpEff software.††Clinical significance on ClinVar database.‡Human reference genome hg19 contains the minor allele. If the minor allele is associated with functional and/or clinical impacts reportedin the literature, this will be indicated in the functional/clinical impacts column.76Table 4.3: Frequency of germline variants in pharmacogenomic genes detected in blood spec-imens of TOP patients.Gene Chr Pos dbSNP ID HGVS* Zygosity Total Pct‡ (%)wt-var†, var-var††DPYD 1 97547947 rs67376798 p.Asp949Valc.2846A>T2, 0 2 0.91 97770920 rs1801160 p.Val732Ilec.2194G>A24, 0 24 111 97915614 rs3918290 c.1906G>A 1, 0 1 0.51 97915615 rs3918289 c.1905C>T 1, 0 1 0.51 97981421 rs1801158 p.Ser534Asnc.1601G>A3, 0 3 21 98039419 rs56038477 p.Glu412Gluc.1236G>A7, 0 7 31 98165091 rs2297595 p.Met166Valc.496A>G34, 0 34 161 98348885 rs1801265 p.Cys29Argc.85T>C69, 11 80 37GSTP1 11 67352689 rs1695 p.Ile105Valc.313A>G89, 20 109 51MTHFR 1 11854476 rs1801131 p.Glu429Alac.1286A>C86, 16 102 471 11856378 rs1801133 p.Ala222Valc.665C>T90, 20 110 51TYMP 22 50964236 rs11479 p.Ser471Leuc.1412C>T51, 6 57 2722 50964255 rs112723255 p.Ala465Thrc.1393G>A16, 1 17 822 50964493 NA p.Glu413Lysc.1237G>A1, 0 1 0.522 50964907 rs201685922 c.929 932delCCGC 1, 0 1 0.522 50965102 rs8141558 p.Leu277Leuc.831G>A1, 0 1 0.522 50965597 rs373478014 p.Thr254Thrc.762G>A1, 0 1 0.522 50965624 rs139223629 p.Gln245Glnc.735G>A1, 0 1 0.577Gene Chr Pos dbSNP ID HGVS* Zygosity Total Pct‡ (%)wt-var†, var-var††22 50965683 rs200497106 p.Gly226Argc.676G>A1, 0 1 0.522 50966082 NA p.Ala194Valc.581C>T1, 0 1 0.5TYMS 22 673443 rs151264360 c.*447 *452delTTAAAG 89, 43 132 62UGT1A1 2 234668870 rs873478 c.-64G>C 1, 0 1 0.52 234668879 rs34983651 c.-55 -54insAT 81, 17 98 46Total variants in PGx genes = 785Average number of variants per patient = 3.7Standard error = 0.10*Description of sequence variants according to the HGVS recommendations.†wt-var represents heterozygous variant.††var-var represents homozygous variant.‡Percentage of patients with the variant.78Table 4.4: Interpretation of germline alterations in pharmacogenomic genes detected in blood specimens of TOP patients.Gene Chr:Pos dbSNP ID HGVS* AF** VariantEffect†ClinicalSignificance††Functional/Clinical Impacts Ref.DPYD 1:97547947 rs67376798 p.Asp949Valc.2846A>T0.26 Missense Drug response Close to iron sulfur motif, whichcould interfere with electron trans-port or cofactor binding. Re-duced DPD activity with strongclinical evidence indicating associ-ation with severe fluoropyrimidine-related toxicity.[7, 24,40,58, 65,115,137,142,146,152,185,200,205,207,208]1:97770920 rs1801160 p.Val732Ilec.2194G>A4.6 Missense Benign/Likelybenign,not providedReduced DPD activity and associ-ated with severe fluoropyrimidine-related toxicity.[24,58, 81,185,207,208]79Gene Chr:Pos dbSNP ID HGVS* AF** VariantEffect†ClinicalSignificance††Functional/Clinical Impacts Ref.1:97915614 rs3918290 c.1906G>A 0.52 SplicedonorDrug response Exon 14 is skipped, producing aninactive enzyme with no uracil-binding site. Reduced DPD ac-tivity with strong clinical evidenceindicating association with severefluoropyrimidine-related toxicity.[7, 40,58, 81,115,142,146,185,200,205,207,208]1:97915615 rs3918289 c.1905C>T 0.030 SpliceregionNot provided Benign variant as predicted byPolyPhen-2, a functional predic-tion software. No association withfluoropyrimidine-related toxicity.[24,152]1:97981421 rs1801158 p.Ser534Asnc.1601G>A1.4 Missense Conflictinginterpretationsofpathogenicity,not providedConflicting evidence on changes toDPD activity. Conflicting clini-cal evidence on association withfluoropyrimidine-related toxicity.[142,152,185,205,208]80Gene Chr:Pos dbSNP ID HGVS* AF** VariantEffect†ClinicalSignificance††Functional/Clinical Impacts Ref.1:98039419 rs56038477 p.Glu412Gluc.1236G>A1.5 Syn. Benign Synonymous variant in highlinkage disequilibrium withc.1129-5923C>G (rs75017182)in haplotype B3 (HapB3).rs75017182 causes nonsensemutation in exon 11, resulting inreduced DPD activity. Associatedwith fluoropyrimidine-relatedtoxicity.[7, 58,142,148]1:98165091 rs2297595 p.Met166Valc.496A>G8.6 Missense Drug response Conflicting evidence on changesto DPD activity. Associated withfluoropyrimidine-related toxicity.[58,81,152,200,207,208]1:98348885 rs1801265‡ p.Cys29Argc.85T>C23 Missense Not provided C allele causes reduced DPDactivity. Conflicting clinicalevidence on association withfluoropyrimidine-related toxicity.[40,81,146,201,208]81Gene Chr:Pos dbSNP ID HGVS* AF** VariantEffect†ClinicalSignificance††Functional/Clinical Impacts Ref.GSTP1 11:67352689 rs1695 p.Ile105Valc.313A>G33 Missense Drug response Disrupts the enzyme’s electrophile-binding active site, thereby lower-ing catalytic efficiency. Increasedrisk of oxaliplatin-related toxicityand efficacy of oxaliplatin treat-ment.[5, 45,95,139,176,196]MTHFR 1:11854476 rs1801131 p.Glu429Alac.1286A>C30 Missense Drug response Reduced MTHFR activity withconflicting evidence on efficacy oftreatment with fluoropyrimidines.[74,75,100,134,176]1:11856378 rs1801133 p.Ala222Valc.665C>T30 Missense Drug response Reduced MTHFR activity, result-ing in stronger inhibition of DNAsynthesis. Increased effective-ness of fluoropyrimidine treatment,although conflicting clinical evi-dence exists. Conflicting evidenceon fluoropyrimidine-related toxic-ity.[50,74,75, 90,100,134,176,185,198]82Gene Chr:Pos dbSNP ID HGVS* AF** VariantEffect†ClinicalSignificance††Functional/Clinical Impacts Ref.TYMP 22:50964236 rs11479 p.Ser471Leuc.1412C>T12 Missense Benign/LikelybenignHigh expression in tumour cells,correlated with poor overall sur-vival in the presence of highplatelet counts. Limited clini-cal evidence suggesting associationwith adverse reactions from fluo-ropyrimidine treatment.[37,96,101]22:50964255 rs112723255 p.Ala465Thrc.1393G>A4.4 Missense Benign/LikelybenignNo association withfluoropyrimidine-related toxicity.Increased risk of transplant-relatedtoxicity from HLA-matchedsibling allogeneic stem cell trans-plantation. Increased risk ofchronic graft-versus-host diseasewhen donor is a carrier of theminor allele and recipient ishomozygous for the major allele.[88,101,192]22:50964493 NA p.Glu413Lysc.1237G>ANA Missense NA NA NA83Gene Chr:Pos dbSNP ID HGVS* AF** VariantEffect†ClinicalSignificance††Functional/Clinical Impacts Ref.22:50964907 rs201685922 c.929 932delCCGC 0.49 SpliceregionConflictinginterpretationsofpathogenicityObserved in a German Americanpatient with mitochondrial neuro-gastrointestinal encephalomyopa-thy (MNGIE), but relation with TPenzymatic defect was not estab-lished.[151]22:50965102 rs8141558 p.Leu277Leuc.831G>A0.58 Syn. Benign/LikelybenignNA NA22:50965597 rs373478014 p.Thr254Thrc.762G>A0.0016 Syn. NA NA NA22:50965624 rs139223629 p.Gln245Glnc.735G>A0.26 Syn. ConflictinginterpretationsofpathogenicityNA NA22:50965683 rs200497106 p.Gly226Argc.676G>A0.0091 Missense UncertainsignificanceNA NA22:50966082 NA p.Ala194Valc.581C>TNA Missense NA NA NA84Gene Chr:Pos dbSNP ID HGVS* AF** VariantEffect†ClinicalSignificance††Functional/Clinical Impacts Ref.TYMS 22:673443 rs151264360 c.*447 *452delTTAAAG 48‡‡ 3’ UTR Drug response Decreased stability of secondarymRNA structure and lower TSexpression. Conflicting evidenceon survival, response to fluoropy-rimidine treatment, and risk offluoropyrimidine-related toxicity.[4, 67,85, 90,131,196]UGT1A1 2:234668870 rs873478 c.-64G>C 1.1‡‡ UpstreamgeneNA Unknown [43,225,229]2:234668879 rs34983651 c.-55 -54insAT 33‡‡ UpstreamgeneConflictinginterpretationsofpathogenicity,affects,associationLower UGT1A1 expression andassociated with irinotecan-relatedtoxicity.[8, 56,82, 99,111,133,139,174,177,204]85*Description of sequence variants according to the Human Genome Variation Society (HGVS) recommendations.**AF = Allele frequency reported by the Exome Aggregation Consortium (ExAC) and presented in percentage.†Effect of genetic variants as predicted by the SnpEff software.††Clinical significance on ClinVar database.‡Human reference genome hg19 contains the minor allele. If the minor allele is associated with functional and/or clinical impacts reportedin the literature, this will be indicated in the functional/clinical impacts column.‡‡Allele frequency from the 1000 Genomes Project is reported when the allele frequency is unavailable in the ExAC database.86Figure 4.1: Distribution of germline alterations in cancer-related genes in patients from the TOP study. Percentage of patients is calcu-lated for each variant and annotated above individual bars. Color of bars represent options for clinical significance in the ClinVardatabase. The TP53 variant, p.Arg72Pro/c.215G>C, that is associated with drug response is present in 97 out of 213 (45.5 %)patients in the TOP cohort. log(1 + x) transformation is applied to change the scale of set values on the Y-axis.87Figure 4.2: Distribution of germline alterations in PGx genes in patients from the TOP study. Percentage of patients is calculated foreach variant and annotated above individual bars. Color of bars represent options for clinical significance in the ClinVar database.208 out of 213 patients in the TOP cohort have at least one germline PGx variant that is associated with drug response. log(1 + x)transformation is applied to change the scale of set values on the Y-axis.884.2 Germline alterations are highly concordant between blood andFFPE specimensThe tumour genome consists of germline and somatic alterations. In fact, several studies demon-strated that a germline cancer-predisposing variant is present in 3-10% of patients undergoingtumour-normal sequencing [102, 141, 169, 184]. While we were unable to detect any pathogenic orlikely pathogenic germline variants due to the rarity of these variants and the small cohort size of theTOP study, we were still able to identify eight germline alterations that could serve as predictors fordrug response, in addition to other germline alterations. Because paired tumour-blood samples werecollected for patients in the TOP cohort, we sought to determine variant concordance of germlinealterations between tumour and blood specimens. This analysis would reveal the extent to whichgermline alterations can be detected in DNA isolated from tumours.Because there are four tumour specimens in the TOP cohort with duplicates, we examined atotal of 217 tumour-normal paired samples. A total of 4434 variants were identified, in which 4003variants were germline and 431 variants were somatic. Out of the 4003 germline variants, 2022germline variants were detected in the blood. We found that 1981/2022 germline variants in theblood were retained in the tumours, giving rise to a concordance rate of 98.0% (Figure 4.3). Eightyfive out of 1981 germline variants did not retain the same allele status between blood and tumour:83/85 were heterozygous in the blood specimens but homozygous in the tumours, whereas 2/85were homozygous in the blood specimens but heterozygous in the tumours. We also identified41 germline variants in the blood that were discordant with the tumours. These germline variantswere present in the blood but not detected in the tumours. Thirty four out of 41 discordant variantswere heterozygous in the blood specimens but wild type in the tumours. The sum of these 34germline variants with the 83 germline variants that were heterozygous in the blood specimens buthomozygous in the tumours gave rise to a total of 117 LOH events out of 2022 germline variants inthe blood, which resulted in a LOH rate of 5.8%. The remaining 7/41 discordant germline variantswere caused by low sequencing depth (< 100x) in the tumours. Table 4.5 shows the 85 germlinevariants that were discordant between blood and tumour due to allelic status and the 41 discordantgermline variants that were present in the blood but not detected in the tumours.Multiple factors could contribute to the discordant calls aside from LOH events, including posi-tion of the variant within regions of somatic copy number mutations, genomic rearrangements dueto the presence of intragenic fragile sites, and DNA damage caused by formalin fixation [13, 87].Nevertheless, despite the presence of discordant germline alterations, our analysis revealed that themajority of germline alterations identified in the blood could be detected in tumour specimens.89Figure 4.3: Venn diagram demonstrating concordance of variants identified in 217 tumour-blood paired samples.90Table 4.5: Distribution of discordant germline alterations in patients from the TOP cohort.Gene Chr:Pos ID? HGVS* Clinical Significance† Reason for discordance(Blood/Tumour)CountDPYD 1:97547947 rs67376798 p.Asp949Valc.2846A>TDrug response Het/WT 11:97770920 rs1801160 p.Val732Ilec.2194G>ABenign/Likely benign,Not providedHet/Hom 11:98165091 rs2297595 p.Met166Valc.496A>GDrug response Het/Hom 11:98348885 rs1801265 p.Cys29Argc.85T>CNot provided Low coverage in tumour 21:98348885 rs1801265 p.Cys29Argc.85T>CNot provided Het/WT 21:98348885 rs1801265 p.Cys29Argc.85T>CNot provided Het/Hom 3EGFR 7:55249063 rs1050171;COSM1451600p.Gln787Glnc.2361G>ABenign/Likely benign Het/Hom 1GSTP1 11:67352689 rs1695 p.Ile105Valc.313A>GDrug response Het/WT 311:67352689 rs1695 p.Ile105Valc.313A>GDrug response Het/Hom 7KIT 4:55602765 rs3733542;COSM1325p.Leu862Leuc.2586G>CBenign/Likely benign Het/Hom 4MTHFR 1:11854476 rs1801131 p.Glu429Alac.1286A>CDrug response Het/Hom 691Gene Chr:Pos ID? HGVS* Clinical Significance† Reason for discordance(Blood/Tumour)Count1:11856378 rs1801133 p.Ala222Valc.665C>TDrug response Het/Hom 61:11856378 rs1801133 p.Ala222Valc.665C>TDrug response Het/WT 3MTOR 1:11169420 rs41274506 p.Asp2485Aspc.7455C>TNA Het/WT 11:11181327 rs11121691 p.Leu2303Leuc.6909G>ANA Het/Hom 11:11181327 rs11121691 p.Leu2303Leuc.6909G>ANA Low coverage in tumour 11:11181327 rs11121691 p.Leu2303Leuc.6909G>ANA Het/WT 21:11190646 rs2275527 p.Ser1851Serc.5553C>TBenign Het/WT 11:11190730 rs17848553 p.Ala1823Alac.5469C>TBenign Het/Hom 21:11205058 rs1057079;rs386514433p.Ala1577Alac.4731A>GNA Het/Hom 41:11205058 rs1057079;rs386514433p.Ala1577Alac.4731A>GNA Het/WT 41:1272468 rs17036536 p.Arg1154Argc.3462G>CBenign Het/Hom 21:11288758 rs1064261 p.Asn999Asnc.2997T>CNA Het/Hom 292Gene Chr:Pos ID? HGVS* Clinical Significance† Reason for discordance(Blood/Tumour)Count1:11288758 rs1064261 p.Asn999Asnc.2997T>CNA Het/WT 31:11301714 rs1135172 p.Asp479Aspc.1437T>CNA Low coverage in tumour 11:11301714 rs1135172 p.Asp479Aspc.1437T>CNA Het/Hom 4PDGFRA 4:55141055 rs1873778;COSM1430082p.Pro567Proc.1701A>GBenign Low coverage in tumour 34:55152040 rs2228230;COSM22413p.Val824Valc.2472C>TBenign Het/WT 24:55152040 rs2228230;COSM22413p.Val824Valc.2472C>TBenign Het/Hom 2STAT1 2:191872307 rs45463799 p.Asn118Asnc.354C>TLikely benign Het/WT 12:191874667 rs386556119;rs2066802p.Leu21Leuc.63T>CBenign Het/WT 1STAT3 17:40498713 NA p.Lys49Lysc.147A>GNA Het/WT 1TP53 17:7577553 COSM44368 p.Met243fsc.727delANA Het/WT 117:7579472 COSM250061;rs1042522p.Arg72Proc.215G>CDrug response Het/Hom 1317:7579472 COSM250061;rs1042522p.Arg72Proc.215G>CDrug response Het/WT 493Gene Chr:Pos ID? HGVS* Clinical Significance† Reason for discordance(Blood/Tumour)Count17:7579579 rs1800370 p.Pro36Proc.108G>ABenign/Likely benign Het/Hom 1TYMP 22:50964236 rs11479 p.Ser471Leuc.1412C>TBenign/Likely benign Het/Hom 7TYMS 18:673443 rs151264360 c.*447 *452delTTAAAG Drug response Het/Hom 1618:673443 rs151264360 c.*447 *452delTTAAAG Drug response Het/WT 1UGT1A1 2:234668870 rs873478 c.-64G>C NA Het/WT 12:234668879 rs34983651 c.-55 -54insAT Conflicting interpreta-tions of pathogenicity,AssociationHom/Het 22:234668879 rs34983651 c.-55 -54insAT Conflicting interpreta-tions of pathogenicity,AssociationHom/WT 2Total discordant variants = 126?dbSNP and/or COSMIC IDs.*Description of sequence variants according to the HGVS recommendations.†Clinical significance on ClinVar database.Het/Hom = Loss of heterozygosity in the tumourHet/WT = Heterozygous in the blood, but wild type in the tumourHom/Het = Homozygous in the blood, but heterozygous in the tumour944.3 Application of VAF thresholds to separate germline alterationsfrom somatic mutations in tumour-only analysesThrough variant analysis of DNA from blood specimens, we identified germline alterations thatwere associated with drug response, which could predict risk of developing chemotherapy-inducedtoxicity. Furthermore, we assessed the concordance of germline variants between blood and tumoursamples, which demonstrated a high concordance rate of 98.0%. Together, these analyses confirmedthat clinically relevant germline alterations were present in our data set and a large proportion ofgermline alterations could be identified in the tumour DNA. Next, we sought to evaluate the use ofVAF thresholds to separate germline alterations from somatic mutations in tumour-only analyses.This assessment would determine whether application of VAF thresholds is an accurate method toidentify potential germline alterations in clinical tumour sequencing for follow-up germline testing.While our data set did not contain pathogenic germline variants, we anticipated that this approachcould be used to detect germline genetic events associated with cancer predisposition and drugresponse for future patients.We compared the VAF distributions of germline variants detected in blood and tumour speci-mens, and we found a significant difference (Kolmogorov-Smirnov test, D = 0.17, p = 0; Figure 4.4).As expected, we showed that heterozygous alterations in blood tended to have VAFs close to 50%,whereas homozygous alterations in the blood tended to have VAFs close to 100%. However, theVAF distribution of germline variants in the tumours tended to deviate from 50% and 100% forheterozygous and homozygous statuses, respectively. This variation in VAF distributions betweenblood and tumour samples, which could be caused by tumour content, tumour heterogeneity, LOH,or DNA damage as a result of formalin fixation, indicated that the sensitivity of using a VAF cut-offto distinguish between germline and somatic alterations in tumour-only analyses could be com-promised. Thus, we explored the sensitivity of identifying germline alterations at various VAFthresholds. At each VAF cut-off, we determined the number of true positives by identifying variantsin the tumours that overlapped with germline variants in matched blood samples. True positive rate(sensitivity) was then calculated as the fraction of variants that were correctly identified as germlineusing the VAF threshold over the total number of germline variants in the tumours. We noted thatthe sensitivity of germline variant detection in FFPE tumours had lowered with the increase ofVAF thresholds. At VAF thresholds of 15%, 20%, 25%, and 30%, 0.99 (95% CI = 0.99–1.0), 0.98(95% CI = 0.97–0.98), 0.96 (95% CI = 0.95–0.97), and 0.94 (95% CI = 0.93–0.95) sensitivities ofdetecting germline variants in the tumours were achieved, respectively (Table 4.6).Because clinical genomics requires accurate identification of genetic alterations that are clin-ically important, potential germline alterations identified through tumour-only analyses must bereferred to follow-up testing [27, 86, 169]. Hence, not only must our approach for discriminatingbetween germline and somatic alterations be highly sensitive, but also highly precise to minimizesubmission of somatic mutations (false positives) for downstream germline testing, which could in-95cur additional cost and time. For similar reasons that caused VAFs of germline alterations in tumoursamples to differ from germline alterations in the blood, we presumed VAFs of somatic mutationsto be lower. We assessed this variation in VAF distributions between germline and somatic alter-ations in the tumours and found a significant difference (Kolmogorov-Smirnov test, D = 0.52, p =0; Figure 4.5). Indeed, VAFs of somatic mutations tended to be concentrated at lower percentagescompared to VAFs of germline variants. We measured PPVs at various VAF thresholds to examinethe precision of referring germline variants to follow-up testing. At each VAF cut-off, we identifiedtrue germline alterations by overlapping the variants in the tumours with germline variants called inmatched blood samples. PPV was then calculated as the fraction of true positives over total num-ber of variants identified in the tumours, including somatic mutations (false positives). We notedthat as VAF thresholds increased, PPV of referring germline variants to follow-up testing increased.At VAF thresholds of 15%, 20%, 25%, and 30%, 0.86 (95% CI = 0.85–0.87), 0.88 (95% CI =0.86–0.89), 0.89 (95% CI = 0.87–0.90), and 0.90 (95% CI = 0.89–0.91) PPVs of correctly referringpredicted germline variants to downstream germline testing were achieved, respectively (Table 4.7).Despite the difference in VAF distributions between germline alterations in blood and tumoursamples, VAF thresholds at 15%, 20%, 25%, and 30% managed to achieve high sensitivity (≥ 0.94)for detecting germline variants in the tumours. We were also able to leverage the difference inVAFs of germline and somatic variants to distinguish germline variants from somatic mutations intumour-only analyses. Evidently, VAF cut-offs of 15%, 20%, 25%, and 30% could enable precisereferral of potential germline variants to follow-up testing (PPV ≥ 0.86). Missed cases in detect-ing clinically important germline variants could lead to severe clinical ramifications. For example,failure to follow-up on a patient with a potential germline variant that is associated with cancer pre-disposition would mean losing out on an opportunity to inform his or her family members. Becauseconfirmatory testing is mandatory for potential germline variants detected in the tumour, we sug-gest selecting a VAF cut-off with a high sensitivity to minimize the number of missed cases. Basedon our evaluation, 15% VAF threshold would maximize sensitivity by identifying 99% of germlinevariants in the tumours, although 14% of the predicted germline variants would be somatic mu-tations (false positives). Overall, we demonstrated that the use of VAF thresholds is a promisingapproach to accurately identify potential germline alterations in clinical tumour sequencing.96Figure 4.4: Assessment of using a VAF cut-off approach to identify germline alterations intumour-only analyses. (A) Comparison of VAF distributions of germline alterations be-tween blood and tumour. (B) Empirical cumulative distribution of VAFs of germlinealterations in blood and tumour samples (Kolmogorov-Smirnov test).Table 4.6: Sensitivity of identifying germline variants in tumour-only analyses at various vari-ant allele frequency thresholds. 95% confidence interval is the binomial confidence inter-val calculated using the Clopper-Pearson method.VAF (%) False Negative True Positive Sensitivity 95% CI Miss rate 95% CI10 0 1981 1.0 1.0–1.0 0 0–0.001915 13 1968 0.99 0.99-1.0 0.0066 0.0035–0.01120 46 1935 0.98 0.97–0.98 0.023 0.017–0.03125 77 1904 0.96 0.95–0.97 0.039 0.031–0.04830 117 1864 0.94 0.93–0.95 0.059 0.049–0.07035 192 1789 0.90 0.89–0.92 0.097 0.084–0.1140 313 1668 0.84 0.83–0.86 0.16 0.14–0.1745 458 1523 0.77 0.75–0.79 0.23 0.21–0.2597Figure 4.5: Assessment of using a VAF cut-off approach to refer potential germline alter-ations in tumour-only analyses to follow-up testing. (A) Comparison of VAF distri-butions between germline and somatic alterations in tumour specimens. (B) Empiricalcumulative distribution of VAFs of germline and somatic alterations in tumour samples(Kolmogorov-Smirnov test).Table 4.7: Positive predictive values for referral of potential germline variants to downstreamconfirmatory testing at various variant allele frequency thresholds. 95% confidence inter-val is the binomial confidence interval calculated using the Clopper-Pearson method.VAF (%) False Positive True Positive Total Calls Positive Predictive Value 95% CI10 431 1981 2412 0.82 0.81–0.8415 319 1968 2287 0.86 0.85–0.8720 273 1935 2208 0.88 0.86–0.8925 245 1904 2149 0.89 0.87–0.9030 203 1864 2067 0.90 0.89–0.9135 178 1789 1967 0.91 0.90–0.9240 146 1668 1814 0.92 0.91–0.9345 118 1523 1641 0.93 0.91–0.9498Chapter 5DiscussionClinical tumour sequencing can inform medical decision-making for cancer patients. While se-quencing of tumour-normal pairs would enable accurate identification of somatic mutations andsimultaneous detection of clinically important germline variants, matched normal samples are of-ten not obtained in clinical practice. Genomic analyses of tumours can reveal actionable somaticmutations and clinically relevant germline alterations [102, 141, 184]. However, tumour tissues arecommonly FFPE, which represents a challenge in clinical genomics. Formalin fixation damages nu-cleic acid through fragmentation and cytosine deamination, affecting molecular testing with FFPEDNA [63, 106, 153, 154, 189, 220, 221]. Hence, usability of FFPE DNA for germline testing andapproaches to discriminate between germline and somatic variants in tumour-only analyses mustbe evaluated. These assessments would facilitate optimization of workflows to identify potentialgermline alterations using clinical tumour sequencing.In this study, we retrospectively analyzed targeted sequencing data from tumour and matchedblood specimens of 213 cancer patients. Our findings were consistent with DNA fragmentation andcytosine deamination being common forms of DNA damage in FFPE specimens. While the impactof formalin fixation on amplicon enrichment and sequencing results was detectable, we determinedthat these discrepancies were minor and could be minimized using methods such as using shorteramplicons and enriching for longer DNA templates. We also found that 98.0% of germline alter-ations identified in the blood using our panel test were present in the FFPE tumours. This impliesthat a high proportion of germline genetic changes were retained in the tumour genome, demon-strating the reliability of using tumour DNA for germline variant calling. Finally, we assessed theapplication of VAF thresholds to delineate germline and somatic variants in tumour-only analyses.We reported that a VAF cut-off of 15995.1 Formalin-induced DNA damage has minor effects on sequencingmetricsSeveral studies have reported findings that are consistent with our assessment of formalin-inducedDNA damage in FFPE specimens. To assess the usability of FFPE DNA for germline testing,we compared efficiency in amplicon enrichment and sequencing results of FFPE DNA to blood,which is a gold standard for germline testing. We noted lower efficiency in amplicon enrichment inFFPE DNA, with a more pronounced decrease in coverage depth for longer amplicons in the panel.Similarly, Shi et al. [188], Didelot et al. [61], and Wong et al. [220] demonstrated that shorteramplicons gave rise to better PCR amplification success in FFPE DNA, indicating the presence offragmentation damage, which yielded template DNA of shorter fragment lengths. While we ob-served comparable proportion of on-target aligned reads between FFPE and blood DNA, there wereminor discrepancies in coverage depth and uniformity of target bases in FFPE DNA. Various groupshave also reported disparities in coverage depth and uniformity in FFPE DNA when compared toDNA extracted from either fresh frozen or unfixed specimens [19, 195, 220]. Additionally, Wonget al. [221] and Didelot et al [61] showed inverse correlations between coverage depth and the de-gree of DNA fragmentation in FFPE DNA, suggesting that formalin-induced fragmentation damagecould be accountable for such discrepancies in sequencing results. Although we detected differ-ences in sequencing results between FFPE and blood DNA, we concluded that these effects wereeither minor or statistically insignificant. As for the discrepancy in amplicon enrichment, shorteramplicons can be designed to circumvent the drawback of fragmentation damage in FFPE samples.5.2 Sequence artifacts induced by cytosine deamination tend to occurat low allelic frequencyCytosine deamination is a major cause of sequence artifacts in formalin-fixed specimens [44, 62,64, 106, 154, 195, 221]. Herein, we observed increased C>T/G>A artifacts in FFPE DNA com-pared to blood. Artifactual C>T/G>A changes are formed by incorporation of adenines in thecomplementary DNA strand at uracil lesions generated by deamination of cytosines [63]. Whenmeasuring frequency of sequence artifacts at different allele frequency ranges, Wong et al. [221]reported higher C>T/G>A transitions at a lower allele frequency range (1–10% vs. 10–25%). Thisfinding led us to compare the fraction of base changes at different allele frequency ranges, including1–10%, 10–20%, and 20–30%. Indeed, we observed a substantial increase in C>T/G>A withinthe 1–10% allele frequency range. However, we were unable to separate FFPE artifacts from low-allelic-fraction somatic mutations within the different allele frequency ranges due to the lack ofmatched fresh frozen or unfixed tumour tissues. Somatic mutations can occur at VAFs that devi-ate significantly from a diploid zygosity (i.e. heterozygous variant should have VAF close to 50%,whereas homozygous variant should have VAF close to 100%) because of low tumour content or100tumour heterogeneity [34, 39, 104, 202, 222]. Therefore, further workflow optimization shouldbe performed for the purpose of identifying clinically relevant somatic mutations in the tumourgenome. A method to reduce sequence artifacts caused by cytosine deamination is treatment withuracil-DNA glycosylase (UDG) before sequencing. UDG is an enzyme capable of depleting uracillesions in DNA, giving rise to abasic sites. During PCR amplification, cytosine bases are restored atabasic sites by using the complementary DNA strand as template, which consists of guanine basesopposite of the uracil lesions [63]. Several studies showed that pre-treatment of FFPE DNA withUDG can markedly reduce C>T/G>A sequence artifacts [62, 64, 106]. However, this approachcannot correct sequence artifacts at CpG dinucleotides because these cytosines are typically methy-lated, and deamination of 5-methyl cytosines generates thymines instead of uracil bases, which areresistant to UDG repair [64].5.3 Sequence artifacts other than those caused by cytosinedeamination are detectedWe also observed elevated levels of A>G/T>C artifacts in FFPE DNA, albeit to a lesser extentcompared to C>T/G>A artifacts. Likewise, Wong et al. [218] reported that 35% of sequence arti-facts in Sanger sequencing of the BRCA1 gene were A>G/T>C nucleotide changes. We speculatethat increase in A>G/T>C artifacts is caused by deamination of adenine to generate hypoxanthine,which forms base pairs with cytosine instead of thymine. This results in transformation of A-T basepairs to G-C base pairs. Deamination of adenine to hypoxanthine can be catalyzed by an acidicenvironment [214], which can arise in FFPE specimens because formaldehyde can be oxidized togenerate formic acid [63].Acidic conditions also promotes depurination, creating abasic sites. Many DNA polymerasesselectively incorporate adenines across abasic sites, while guanines and small deletions are inte-grated in fewer cases [92]. Despite being statistically insignificant, we observed a subset of FFPEspecimens with higher fractions of C>A/G>T artifacts. These artifactual changes could have re-sulted from depurination of guanines, followed by incorporation of adenines by DNA polymerase inthe complementary strand, which alters G-C base pairs to A-T base pairs. Heyn et al. [92] reportedthat DNA polymerases demonstrated varying bypass rates at abasic sites. For instance, AmpliTaqGold, Pfu, and Platinum Taq HiFi extended across lower frequency of abasic sites compared toPlatinum Taq, Bst and Sso-Dpo4 (<34% vs. >77%) [92]. Thus, selection of a high fidelity DNApolymerase could lessen these forms of sequence artifacts.Costello et al. [51] discovered that C>A/G>T artifacts can also occur due to oxidation of DNAduring the shearing process, converting guanines to 8-oxoguanine lesions. This conversion is highlydependent on the surrounding 5’ and 3’ bases of the guanine, in which guanines within GGC arethe most susceptible to oxidation. 8-oxoguanine can form base pairs with cytosine and adenine, andmispairing with adenine would give rise to artifactual C>A/G>T transversions. However, this was101not the cause of C>A/G>T artifacts in our data because both blood and FFPE DNA were sheared,and we did not observe simultaneous C>A/G>T increments in both specimen types compared toother types of base changes.5.4 Storage time of FFPE blocks correlates with the extent offormalin-induced DNA damageLudyga et al. [128] demonstrated that long-term storage of FFPE blocks led to increased DNA frag-mentation, producing shorter template DNA for PCR amplification. Furthermore, Carrick et al. [38]showed that increased storage time of FFPE blocks affects sequencing coverage and depth in NGSdata. These findings are in agreement with our results, in which we found negative associations be-tween age of paraffin blocks and efficiency in amplicon enrichment, coverage depth of target bases,and percentage of on-target aligned reads. As well, we observed a positive correlation between ageof paraffin blocks and fraction of C>T/G>A artifacts, an outcome of stochastic enrichment. Due toexposure to environmental conditions, older FFPE blocks tend to produce increasingly fragmentedDNA, which results in lower amounts of amplifiable DNA. Consequently, there is a higher chance ofamplifying template DNA with sequence artifacts caused by formalin, yielding increased frequencyof artifactual nucleotide changes in older FFPE specimens [221]. These results demonstrating thecorrelations between storage time of paraffin blocks and sequencing variables suggest that if mul-tiple FFPE blocks are available, the specimen with the shorter storage time should be selected formolecular testing. However, clinical specimens are often limited, making sample selection a rareoption in the diagnostic setting. As such, other approaches to eliminate sequence artifacts should beconsidered such as application of molecular barcodes and hybridization-capture enrichment, whichallow tracking of DNA templates [71, 159, 180, 220]. This would enable detection of variants thatare only supported by the same template DNA, indicating a higher chance that these variants aresequence artifacts and should be interpreted with caution.5.5 Germline variants are highly retained in the tumour genomeVarious groups have identified clinically significant germline alterations through analyzing tumourgenomes [102, 140, 141, 184]. Schrader et al. [184] reported that potential pathogenic germlinevariants in cancer-predisposing genes were conserved in the tumours of 91.9% of patients in theirstudy cohort (182 of 198 patients), whereas 21.4% of these patients (39 of 182 patients) demon-strated LOH or other forms of mutations in the remaining wild type allele. We found that 98.0%(1981/2022) of germline alterations identified in the blood were retained in the tumour, a findingthat is in line with previous work. This suggests that tumour DNA could be a reliable source fordetecting germline alterations, implying that a tumour-only sequencing protocol could be leveragedfor pre-screening of germline variants before submission to downstream confirmatory testing. A102framework as such could provide germline testing in a time- and cost-effective manner becauseonly selected patients (i.e. those with potential germline alterations that are clinically important)would require follow-up.Out of the 98.0% of germline alterations that were retained in the tumours, 4.3% (85/1981) ofvariants demonstrated discordant allelic status between blood and tumour. We also identified dis-cordant germline variants, which were present in the blood but not detected in the tumour, that werecaused by LOH (heterozygous variant in the blood but wild type in the tumour; 34/41 variants)and low sequencing coverage (< 100x; 7/41 variants). All tumour specimens in our study wereformalin-fixed, therefore it is possible that DNA damage induced by formaldehyde exposure playeda role in creating discordant germline variants. Variant discordance can also be caused by muta-genesis in the tumour, such as somatic CNVs in the region of the germline variant. For instance,Gross et al. [87] showed a high prevalence of DPYD CNVs in high-grade triple negative breastcancer, particularly in cases with copy number loss of the BRCA1 DNA-repair gene. The commonfragile site FRA1E is located within the DPYD gene and its stability is highly dependent on intactBRCA1 [13]. Hence, deficiency in BRCA1 protein would result in increased fragility of FRA1E,leading to genomic rearrangements in DPYD. As germline variants in the DPYD gene can predictsusceptibility to 5-FU-related toxicity, somatic CNVs in DPYD could affect the detection of thesegermline variants in tumour genomic sequencing.5.6 The use of VAF thresholds is feasible for distinguishing betweengermline and somatic alterations in tumour-only analysesAlthough sequencing of tumour-normal pairs would enable accurate identification of germline andsomatic variants, this approach is not routinely practiced in clinical genomics due to inadequatefunding and facilities to store additional specimens. Methods to distinguish between germline andsomatic alterations in tumour-only analyses have been described by different groups [80, 93, 102].Hiltemann et al. [93] used a virtual normal that was assembled by aggregating whole-genome-sequenced normal samples from 931 healthy and unrelated individuals, whereas Jones et al. [102]resorted to using an unmatched normal sample and public databases such as dbSNP, 1000 GenomesProject, and COSMIC, as well as effect prediction tools. We leveraged the fact that the VAFsof somatic mutations typically deviate from 50% and 100% for heterozygous and homozygousvariants, respectively, and employed VAF thresholds to differentiate between variant statuses. Ourapproach managed to achieve high sensitivity and precision, therefore verifying the feasibility ofusing VAF threshold to differentiate between germline and somatic alterations in the absence ofmatched normal samples.The VAF threshold method takes advantage of genetic impurity and heterogeneity of tumours,which render the deviation of somatic VAFs from diploid zygosity. Jones et al. [102] discoveredthat performance of the VAF threshold approach was highly dependent on tumour purity. While103the use of VAF can correctly identify all germline and somatic alterations in tumours with < 50%purity, this accuracy was not observed for specimens with higher tumour content. In fact, only12.5% of cancer-predisposing germline variants and an average of 48% of somatic mutations wereaccurately predicted using the VAF method in tumours with more than 50% tumour content [102].Unfortunately, pathologic estimation of tumour content was not available for our analyses. How-ever, we speculate that the tumour specimens in our data set are highly impure or heterogeneous,thereby contributing to the high sensitivity and precision attained by the VAF threshold approach.While there are bioinformatic algorithms available to infer clonality and impurity estimates of tu-mours, many of these methods require matched normal controls or are not compatible with targetedsequencing data [223]. Nevertheless, this information should be integrated into clinical pipelines toenhance the performance of using a VAF threshold approach to distinguish between germline andsomatic alterations in tumour-only genomic analyses.5.7 Limitations and future directionsThere are several limitations in our study. First, we did not manually review every single vari-ant called by our pipeline. Only variants located within primer regions were manually inspected,while our variant filter also included common artifacts that were curated during clinical assess-ment. Hence, it is highly possible that sequence artifacts are present in our data set, particularlylow-allelic-fraction variants (i.e. < 10%) detected in the blood. Variant inspection using a genomebrowser is routinely conducted by genomic analysts in clinical practice to decrease the risk of re-porting false positive results [80, 197]. However, manual review of variants was not implementedin our study because our analyses were focused on evaluating analytical validity instead of produc-ing genomic information for clinical decision-making. Moreover, the large number of variants inour study would be time-consuming and unfeasible for manual inspection. Our evaluation of theVAF threshold approach in differentiating between germline and somatic variants is favourable ofthe framework to implement initial screening for germline variants in clinical tumour sequencingbefore follow-up germline testing.The small gene panel and patient cohort size are caveats in our study. Although we were ableto identify germline variants that can influence drug response, we did not report any pathogenicgermline variants that were associated with cancer predisposition in our data set. Hence, we can onlyspeculate that our approach could be applicable to variants in cancer-predisposing genes. Studiesthat were able to identify pathogenic germline variants were performed with cohort sizes and genepanels that were substantially larger than this study. For instance, the study by Schrader et al. [184],which revealed pathogenic germline variants in 16% of patients, was performed in a cohort of 1566patients and 341 genes were screened. To determine whether the VAF threshold method can beapplied to detect genetic alterations linked to cancer susceptibility, further assessment involving alarger patient cohort and surveying known cancer-predisposing genes should be carried out.104The present study addresses two problems faced by using tumour genomic sequencing to iden-tify germline alterations: the widespread use of FFPE tumours and the challenge in differentiatingbetween germline and somatic variants in tumour genomes. Archival FFPE tissues remain a sizableresource for cancer genomic studies and clinical genomic sequencing. Thus, there is a need to un-derstand the extent of the different forms of DNA damage induced by formalin. Our analyses notonly provided insights on the impact of formalin-induced DNA damage on amplicon-based NGSdata, but also helped us devised guidelines to minimize these effects. Formalin fixation followedby paraffin embedding is an attractive method to preserve tissue morphology for histologic assess-ment because it allows storage at ambient temperature, which reduces cost incurred by maintainingfreezers required for fresh-frozen samples. Yet, many studies, including ours, have indicated theside effects of formaldehyde exposure on nucleic acid [63, 106, 153, 154, 189, 220, 221]. Instead ofinvesting efforts into mitigating these side effects, a potential solution is to transition from the useof formalin to the UMFIX (Sakura Finetek USA, Inc.) fixative, which is capable of preserving bothcellular morphology for pathologic review and macromolecules, including DNA [211].Most clinical laboratories conduct tumour-only sequencing and apply approaches to distinguishbetween germline and somatic alterations. Without matched normal samples, interpretation of vari-ants becomes complicated. Jones et al. [102] and Garofalo et al. [80] concluded that sequencing oftumour-normal pairs is the best practice to accurately identify variant statuses. For a center to pro-vide tumour-normal paired sequencing, it must be equipped to collect, analyze, and report germlinefindings. This includes establishing appropriate pre-test and post-test counseling, protocols to se-cure patient consent and manage variant of uncertain significance, and frameworks to communicateresults that may implicate the patients’ relatives. While various groups recommend the sequencingof tumour-normal pairs, some centers simply do not have the funding or infrastructure to implementthis as a standard practice. Furthermore, the American College of Medical Genetics and Genomics(ACMG) recommended that clinical laboratories report incidental variants in 56 genes that are asso-ciated with disease risk in DNA derived from germline samples, including matched normal samplesthat only serve the purpose of subtracting germline variants to identify somatic mutations in tumours[86]. Interrogation of these genes suggested by the ACMG guidelines could result in detection ofmore variants with uncertain significance, which might pose more harm than good to patients. Addi-tionally, cases in which only FFPE tumour blocks exist for a deceased patient would greatly benefitfrom approaches in differentiating between germline and somatic variants. For example, if thedeceased individual is suspected to be a carrier of an inheritable disease, the ability to accuratelyidentify the germline risk allele could prompt germline testing for the individual’s relatives and fa-cilitate preventive care. Thus, establishing approaches to tell apart germline and somatic variants intumour genomic analyses still has its advantages from a clinical and financial perspective.To summarize, we showed that the common forms of formalin-induced DNA damage in tu-mour samples were DNA fragmentation and cytosine deamination. Because these effects were105either minor or statistically insignificant compared to DNA extracted from blood, this justified thefeasibility of using FFPE DNA for germline testing. Characterization of formalin-induced DNAdamage could also assist in establishing recommendations to enhance amplicon enrichment andsequencing results. We also reported a high retention rate of germline alterations in the tumourgenomes, suggesting the reliability of using tumour DNA for germline variant calling. Finally,we showed that application of VAF thresholds can achieve high sensitivity and precision in dis-tinguishing germline alterations from somatic mutations in tumour-only analyses. This supportsthe framework of leveraging clinical tumour sequencing for identification of germline alterations.Subsequently, only patients with potential germline variants are referred to follow-up testing. Aframework in which selected patients are referred to downstream confirmatory testing represents atime- and cost-effective approach to deliver germline testing.106Bibliography[1] Nature Milestones in Cancer. Nature Reviews Cancer, 6:S8, 2006. → pages 1[2] The Human Genome Project Completion: Frequently Asked Questions, 2010. URLhttps://www.genome.gov/11006943/human-genome-project-completion-frequently-asked-questions/. Accessed September 12,2017. → pages 3[3] I. A. Adzhubei, S. Schmidt, L. Peshkin, et al. A method and server for predicting damagingmissense mutations. Nature Methods, 7(4):248–249, 2010. → pages 14[4] S. Afzal, M. Gusella, B. Vainer, et al. Combinations of polymorphisms in genes involved inthe 5-fluorouracil metabolism pathway are associated with gastrointestinal toxicity inchemotherapy-treated colorectal cancer patients. Clinical Cancer Research, 17(11):3822–3829, 2011. → pages 85[5] F. Ali-osman, O. Akande, G. Antoun, et al. Molecular Cloning , Characterization , andExpression in Escherichia coli of Full-length cDNAs of Three Human Glutathione S-Transferase Pi Gene Variants. The Journal of Biological Chemistry, 272(15):10004–10012,1997. → pages 82[6] A. Ally, M. Balasundaram, R. Carlsen, et al. Comprehensive and Integrative GenomicCharacterization of Hepatocellular Carcinoma. Cell, 169(7):1327–1341.e23, 2017. → pages7[7] U. Amstutz, S. Farese, S. Aebi, et al. Dihydropyrimidine dehydrogenase gene variation andsevere 5-fluorouracil toxicity: a haplotype assessment. Pharmacogenomics, 10(6):931–944,2009. → pages 79, 80, 81[8] Y. Ando, H. Saka, M. Ando, et al. Polymorphisms of UDP-glucuronosyltransferase geneand irinotecan toxicity: A pharmacogenetic analysis. Cancer Research, 60(24):6921–6926,2000. → pages 85[9] S. Andrews. FastQC. URL https://www.bioinformatics.babraham.ac.uk/projects/fastqc/.Accessed October 1, 2017. → pages 11[10] I. L. Andrulis, S. B. Bull, M. E. Blackstein, et al. Neu/erbB-2 amplification identifies apoor-prognosis group of women with node-negative breast cancer. Toronto Breast CancerStudy Group. Journal of Clinical Oncology, 16(4):1340–9, 1998. → pages 1107[11] N. A˚nensen, J. Skavland, C. Stapnes, et al. Acute myelogenous leukemia in a patient withLiFraumeni syndrome treated with valproic acid, theophyllamine and all-trans retinoic acid:a case report. Leukemia, 20(4):734–736, 2006. → pages 74[12] S. L. Arcand, C. M. Maugard, P. Ghadirian, et al. Germline TP53 mutations in BRCA1 andBRCA2 mutation-negative French Canadian breast cancer families. Breast Cancer Researchand Treatment, 108(3):399–408, 2008. → pages 74[13] M. F. Arlt, B. Xu, S. G. Durkin, et al. BRCA1 Is Required for Common-Fragile-SiteStability via Its G 2 / M Checkpoint Function BRCA1 Is Required for Common-Fragile-SiteStability via Its G 2 / M Checkpoint Function. Molecular and Cellular Biology, 24(15):6701–6709, 2004. → pages 89, 103[14] S. G. Baker and J. Kaprio. Common susceptibility genes for cancer: search for the end ofthe rainbow. BMJ (Clinical research ed.), 332(7550):1150–2, 2006. → pages 15[15] R. Bao, L. Huang, J. Andrade, et al. Review of Current Methods, Applications, and DataManagement for the Bioinformatics Analysis of Whole Exome Sequencing. LibertasAcademica, 13:67–82, 2014. → pages 11, 12, 13, 14, 15[16] C. Baynes, C. S. Healey, K. a. Pooley, et al. Common variants in the ATM, BRCA1,BRCA2, CHEK2 and TP53 cancer susceptibility genes are unlikely to increase breastcancer risk. Breast cancer research: BCR, 9(2):R27, 2007. → pages 15[17] C. L. Bennett and E. A. Calhoun. Evaluating the total costs of chemotherapy-induced febrileneutropenia: results from a pilot study with community oncology cancer patients. TheOncologist, 12(4):478–83, 2007. → pages 17[18] D. R. Bentley, S. Balasubramanian, H. P. Swerdlow, et al. Accurate whole human genomesequencing using reversible terminator chemistry. Nature, 456(7218):53–59, 2008. →pages 3, 5[19] J. Betge, G. Kerr, T. Miersch, et al. Amplicon Sequencing of Colorectal Cancer: VariantCalling in Frozen and Formalin-Fixed Samples. Plos One, 10(5):e0127146, 2015. → pages100[20] F. Biramijamal, A. Allameh, P. Mirbod, et al. Unusual profile and high prevalence of p53mutations in esophageal squamous cell carcinomas from northern Iran. Cancer Research,61(7):3119–23, 2001. → pages 74[21] G. D. V. Blanco, O. A. Paoluzi, P. Sileri, et al. Familial colorectal cancer screening: Whenand what to do? World Journal of Gastroenterology, 21(26):7944–7953, 2015. → pages 17[22] K. Bodi, A. G. Perera, P. S. Adams, et al. Comparison of commercially available targetenrichment methods for next-generation sequencing. Journal of Biomolecular Techniques,24(2):73–86, 2013. → pages 8[23] V. Boeva, T. Popova, M. Lienard, et al. Multi-factor data normalization enables thedetection of copy number aberrations in amplicon sequencing data. Bioinformatics, 30(24):3443–3450, 2014. → pages 8108[24] V. Boige, M. Vincent, P. Alexandre, et al. DPYD Genotyping to PredictAdverse EventsFollowing Treatment With Fluorouracil-BasedAdjuvant Chemotherapy in Patients WithStage III Colon Cancer. JAMA Oncology, 2(5):655–662, 2016. → pages 79, 80[25] S. E. Bojesen and B. G. Nordestgaard. The common germline Arg72Pro polymorphism ofp53 and increased longevity in humans. Cell Cycle, 7(2):158–163, 2008. → pages 61, 75[26] A. M. Bolger, M. Lohse, and B. Usadel. Trimmomatic: A flexible trimmer for Illuminasequence data. Bioinformatics, 30(15):2114–2120, 2014. → pages 11[27] Y. Bombard, M. Robson, and K. Offit. Revealing the Incidentalome When Targeting theTumor Genome. The Journal of the American Medical Association, 310(8):795–6, 2014. →pages 20, 60, 95[28] M. Bonafe´, S. Salvioli, C. Barbi, et al. The different apoptotic potential of the p53 codon 72alleles increases with age and modulates in vivo ischaemia-induced cell death. Cell Deathand Differentiation, 11(9):962–973, 2004. → pages 61, 75[29] S. Bonin, M. Donada, G. Bussolati, et al. A synonymous EGFR polymorphism predictingresponsiveness to anti-EGFR therapy in metastatic colorectal cancer patients. TumorBiology, 37(6):7295–7303, 2016. → pages 67[30] I. E. Bosdet, T. R. Docking, Y. S. Butterfield, et al. A clinically validated diagnosticsecond-generation sequencing assay for detection of hereditary BRCA1 and BRCA2mutations. Journal of Molecular Diagnostics, 15(6):796–809, 2013. → pages 2, 26[31] G. Bougeard, S. Baert-Desurmont, I. Tournier, et al. Impact of the MDM2 SNP309 and p53Arg72Pro polymorphism on age of tumour onset in Li-Fraumeni syndrome. Journal ofMedical Genetics, 43(6):531–3, 2006. → pages 61, 75[32] T. Boveri. Zur Frage der Entstehung Maligner Tumoren. Gustav Fischer, 1914. → pages 1[33] W. Burke. Genetic tests: clinical validity and clinical utility. Current Protocols in HumanGenetics, 81(9):1–8, 2014. → pages 15, 16[34] L. Cai, W. Yuan, Z. Zhang, et al. In-depth comparison of somatic point mutation callersbased on different tumor next-generation sequencing depth data. Scientific Reports, 6(November):36540, 2016. → pages 101[35] E. Capriotti, R. Calabrese, and R. Casadio. Predicting the insurgence of human geneticdiseases associated to single point protein mutations with support vector machines andevolutionary information. Bioinformatics, 22(22):2729–2734, 2006. → pages 15[36] M. O. Carneiro, C. Russ, M. G. Ross, et al. Pacific biosciences sequencing technology forgenotyping and variation discovery in human data. BMC Genomics, 13(1):375, 2012. →pages 3[37] D. Caronia, M. Martin, J. Sastre, et al. A polymorphism in the cytidine deaminase promoterpredicts severe capecitabine-induced hand-foot syndrome. Clinical Cancer Research, 17(7):2006–2013, 2011. → pages 83109[38] D. M. Carrick, M. G. Mehaffey, M. C. Sachs, et al. Robustness of next generationsequencing on older formalin-fixed paraffin-embedded tissue. PLoS ONE, 10(7):3–10,2015. → pages 22, 56, 102[39] J. Carrot-Zhang and J. Majewski. LoLoPicker: Detecting Low Allelic-Fraction Variants inLow-Quality Cancer Samples from Whole-exome Sequencing Data. bioRxiv, 8(23):043612,2016. → pages 101[40] K. E. Caudle, C. F. Thorn, T. E. Klein, et al. Clinical Pharmacogenetics ImplementationConsortium Guidelines for Dihydropyrimidine Dehydrogenase Genotype andFluoropyrimidine Dosing. Clinical Pharmacology & Therapeutics, 94(6):640–645, 2013.→ pages 79, 80, 81[41] F. Chang and M. M. Li. Clinical application of amplicon-based next-generation sequencingin cancer. Cancer Genetics, 206(12):413–419, 2013. → pages 8[42] P. B. Chapman, A. Hauschild, C. Robert, et al. Improved Survival with Vemurafenib inMelanoma with BRAF V600E Mutation. New England Journal of Medicine, 364(26):2507–2516, 2011. → pages 1[43] S. Cheli, F. Pietrantonio, E. Clementi, et al. LightSNiP assay is a good strategy forpharmacogenetics test. Frontiers in Pharmacology, 6(JUN):1–5, 2015. → pages 85[44] G. Chen, S. Mosier, C. D. Gocke, et al. Cytosine Deamination is a Major Cause of BaselineNoise in Next Generation Sequencing. Molecular Diagnosis & Therapy, 18(5):587–593,2014. → pages 100[45] Y. C. Chen, C. H. Tzeng, P. M. Chen, et al. Influence of GSTP1 I105V polymorphism oncumulative neuropathy and outcome of FOLFOX-4 treatment in Asian patients withcolorectal carcinoma. Cancer Science, 101(2):530–535, 2010. → pages 82[46] H. Cheng, B. Ma, R. Jiang, et al. Individual and combined effects of MDM2 SNP309 andTP53 Arg72Pro on breast cancer risk: An updated meta-analysis. Molecular BiologyReports, 39(9):9265–9274, 2012. → pages 61, 75[47] K. N. Chitrala and S. Yeguvapalli. Computational screening and molecular dynamicsimulation of breast cancer associated deleterious non-synonymous single nucleotidepolymorphisms in TP53 gene. PLoS ONE, 9(8), 2014. → pages 74[48] J. Chung, D.-S. Son, H.-J. Jeon, et al. The minimal amount of starting DNA for Agilent’shybrid capture-based targeted massively parallel sequencing. Scientific Reports, 6(1):26732,2016. → pages 2[49] K. Cibulskis, M. S. Lawrence, S. L. Carter, et al. Sensitive detection of somatic pointmutations in impure and heterogeneous cancer samples. Nature Biotechnology, 31(3):213–219, 2013. → pages 13[50] V. Cohen, V. Panet-raymond, N. Sabbaghian, et al. Methylenetetrahydrofolate ReductasePolymorphism in Advanced Colorectal Cancer : A Novel Genomic Predictor of Clinical110Response to Fluoropyrimidine-based Chemotherapy Advances in BriefMethylenetetrahydrofolate Reductase Polymorphism in Advanced Colorecta. ClinicalCancer Research, 9(May):1611–1615, 2003. → pages 82[51] M. Costello, T. J. Pugh, T. J. Fennell, et al. Discovery and characterization of artifactualmutations in deep coverage targeted capture sequencing data due to oxidative DNA damageduring sample preparation. Nucleic Acids Research, 41(6):1–12, 2013. → pages 101[52] H. Cui. Methods of Gene Enrichment and Massively Parallel Sequencing Technologies. InL.-J. C. Wong, editor, Next Generation Sequencing: Translation to Clinical Diagnostics,chapter 3, pages 39–58. Springer Science+Business Media, New York, 2013. ISBN9781461470014. → pages 8[53] J. Culver, C. Brinkerhoff, J. Clague, et al. Variants of uncertain significance in BRCAtesting: Evaluation of surgical decisions, risk perception, and cancer distress. ClinicalGenetics, 84(5):464–472, 2013. → pages 7[54] P. Danecek, A. Auton, G. Abecasis, et al. The variant call format and VCFtools.Bioinformatics, 27(15):2156–2158, 2011. → pages 13, 14[55] M. David, M. Dzamba, D. Lister, et al. SHRiMP2: Sensitive yet practical short readmapping. Bioinformatics, 27(7):1011–1012, 2011. → pages 12[56] F. A. de Jong. Prophylaxis of Irinotecan-Induced Diarrhea with Neomycin and PotentialRole for UGT1A1*28 Genotype Screening: A Double-Blind, Randomized,Placebo-Controlled Study. The Oncologist, 11(8):944–954, 2006. → pages 85[57] K. J. Dedes, P. M. Wilkerson, D. Wetterskog, et al. Synthetic lethality of PARP inhibition incancers lacking BRCA1 and BRCA2 mutations. Cell Cycle, 10(8):1192–1199, 2011. →pages 17[58] M. J. Deenen, J. Tol, A. M. Burylo, et al. Relationship between single nucleotidepolymorphisms and haplotypes in DPYD and toxicity and efficacy of capecitabine inadvanced colorectal cancer. Clinical Cancer Research, 17(10):3455–3468, 2011. → pages17, 79, 80, 81[59] J. T. den Dunnen, R. Dalgleish, D. R. Maglott, et al. HGVS Recommendations for theDescription of Sequence Variants: 2016 Update. Human Mutation, 37(6):564–569, 2016.→ pages 14[60] M. Depner, S. Fuchs, J. Raabe, et al. The Extended Clinical Phenotype of 26 Patients withChronic Mucocutaneous Candidiasis due to Gain-of-Function Mutations in STAT1. Journalof Clinical Immunology, 36(1):73–84, 2016. → pages 72[61] A. Didelot, S. K. Kotsopoulos, A. Lupo, et al. Multiplex picoliter-droplet digital PCR forquantitative assessment of DNA integrity in clinical samples. Clinical Chemistry, 59(5):815–823, 2013. → pages 22, 35, 56, 100111[62] H. Do and A. Dobrovic. Dramatic reduction of sequence artefacts from DNA isolated fromformalin-fixed cancer biopsies by treatment with uracil- DNA glycosylase. Oncotarget, 3(5):546–58, 2012. → pages 48, 100, 101[63] H. Do and A. Dobrovic. Sequence artifacts in DNA from formalin-fixed tissues: Causes andstrategies for minimization. Clinical Chemistry, 61(1):64–71, 2015. → pages 22, 35, 48, 99,100, 101, 105[64] H. Do, S. Q. Wong, J. Li, et al. Reducing sequence artifacts in amplicon-based massivelyparallel sequencing of formalin-fixed paraffin-embedded DNA by enzymatic depletion ofuracil-containing templates. Clinical Chemistry, 59(9):1376–1383, 2013. → pages 48, 100,101[65] D. Dobritzsch, G. Schneider, K. D. Schnackerz, et al. Crystal structure ofdihydropyrimidine dehydrogenase, a major determinant of the pharmacokinetics of theanti-cancer drug 5-fluorouracil. EMBO Journal, 20(4):650–660, 2001. → pages 79[66] L. Dong, W. Wang, A. Li, et al. Clinical Next Generation Sequencing for PrecisionMedicine in Cancer. Current Genomics, 16(4):253–63, 2015. → pages 10, 22[67] E. Dotor, M. Cuatrecases, M. Martı´nez-Iniesta, et al. Tumor thymidylate synthase 1494del6genotype as a prognostic factor in colorectal cancer patients receiving fluorouracil-basedadjuvant treatment. Journal of Clinical Oncology, 24(10):1603–1611, 2006. → pages 85[68] J. A. Drebin, V. C. Link, D. F. Stern, et al. Down-modulation of an oncogene proteinproduct and reversion of the transformed phenotype by monoclonal antibodies. Cell, 41(3):695–706, 1985. → pages 1[69] B. J. Druker, F. Guilhot, S. G. O’Brien, et al. Five-Year Follow-up of Patients ReceivingImatinib for Chronic Myeloid Leukemia. New England Journal of Medicine, 355:2408–2417, 2006. → pages 1[70] J. Eid, A. Fehr, J. Gray, et al. Real-Time DNA Sequencing from. Science, New Series, 323(5910):133–138, 2009. → pages 3[71] A. Eijkelenboom, E. J. Kamping, A. W. Kastner-van Raaij, et al. Reliable Next-GenerationSequencing of Formalin-Fixed, Paraffin-Embedded Tissue Using Single Molecule Tags. TheJournal of Molecular Diagnostics, 18(6):851–863, 2016. → pages 102[72] A. C. English, S. Richards, Y. Han, et al. Mind the Gap: Upgrading Genomes with PacificBiosciences RS Long-Read Sequencing Technology. PLoS ONE, 7(11):1–12, 2012. →pages 3[73] P. Estevez-Garcia, A. Castan˜o, A. C. Martin, et al. PDGFRα/β and VEGFR2polymorphisms in colorectal cancer: incidence and implications in clinical outcome. BMCCancer, 12:514, 2012. → pages 72[74] M. C. Etienne, J. L. Formento, M. Chazal, et al. Methylenetetrahydrofolate reductase genepolymorphisms and response to fluorouracil-based treatment in advanced colorectal cancerpatients. Pharmacogenetics, 14(12):785–792, 2004. → pages 82112[75] M. C. Etienne-Grimaldi, G. Milano, F. Maindrault-Gœbel, et al. Methylenetetrahydrofolatereductase (MTHFR) gene polymorphisms and FOLFOX response in colorectal cancerpatients. British Journal of Clinical Pharmacology, 69(1):58–66, 2010. → pages 82[76] G. Fortunato, G. Calcagno, V. Bresciamorra, et al. Multiple sclerosis and hepatit.PDF.Journal of Interferon & Cytokine Research, 28:141–152, 2008. → pages 73[77] G. M. G. Frampton, A. Fichtenholtz, G. a. G. Otto, et al. Development and validation of aclinical cancer genomic profiling test based on massively parallel DNA sequencing. Nature,31(October):1–11, 2013. → pages 60[78] D. Fumagalli, P. G. Gavin, Y. Taniyama, et al. A rapid, sensitive, reproducible andcost-effective method for mutation profiling of colon cancer and metastatic lymph nodes.BMC Cancer, 10:101, 2010. → pages 60[79] J. Gagan and E. M. Van Allen. Next-generation sequencing to guide cancer therapy.Genome Medicine, 7(1):80, 2015. → pages 8[80] A. Garofalo, L. Sholl, B. Reardon, et al. The impact of tumor profiling approaches andgenomic data strategies for cancer precision medicine. Genome Medicine, 8(1):79, 2016. →pages 103, 104, 105[81] G. Gentile, A. Botticelli, L. Lionetto, et al. Genotypephenotype correlations in5-fluorouracil metabolism: a candidate DPYD haplotype to improve toxicity prediction. ThePharmacogenomics Journal, 16(4):320–325, 2016. → pages 79, 80, 81[82] B. Glimelius, H. Garmo, A. Berglund, et al. Prediction of irinotecan and 5-fluorouraciltoxicity and response in patients with advanced colorectal cancer. The PharmacogenomicsJournal, 11(1):61–71, 2011. → pages 85[83] A. Gonza´lez-Pe´rez and N. Lo´pez-Bigas. Improving the assessment of the outcome ofnonsynonymous SNVs with a consensus deleteriousness score, Condel. American Journalof Human Genetics, 88(4):440–449, 2011. → pages 15[84] S. Goodwin, J. D. McPherson, and W. R. McCombie. Coming of age: ten years ofnext-generation sequencing technologies. Nature Reviews Genetics, 17(6):333–351, 2016.→ pages 4, 5[85] F. Graziano, A. Ruzzo, F. Loupakis, et al. Liver-only metastatic colorectal cancer patientsand thymidylate synthase polymorphisms for predicting response to 5-fluorouracil-basedchemotherapy. British Journal of Cancer, 99(5):716–21, 2008. → pages 85[86] R. C. Green, J. S. Berg, W. W. Grody, et al. ACMG recommendations for reporting ofincidental findings in clinical exome and genome sequencing. Genetics in Medicine, 15(7):565–74, 2013. → pages 95, 105[87] E. Gross, C. Meul, S. Raab, et al. Somatic copy number changes in DPYD are associatedwith lower risk of recurrence in triple-negative breast cancers. British Journal of Cancer,109(9):2347–2355, 2013. → pages 89, 103113[88] V. Guillem, J. C. Hernandez-Boluda, D. Gallardo, et al. A polymorphism in the TYMP geneis associated with the outcome of HLA-identical sibling allogeneic stem cell transplantation.American Journal of Hematology, 88(10):883–889, 2013. → pages 83[89] J. Guo, N. Xu, Z. Li, et al. Four-color DNA sequencing with 3’-O-modified nucleotidereversible terminators and chemically cleavable fluorescent dideoxynucleotides.Proceedings of the National Academy of Sciences, 105(27):9145–9150, 2008. → pages 3, 5[90] M. Gusella, G. Crepaldi, C. Barile, et al. Pharmacokinetic and demographic markers of5-fluorouracil toxicity in 181 patients on adjuvant therapy for colorectal cancer. Annals ofOncology, 17(11):1656–1660, 2006. → pages 82, 85[91] T. Helleday. The underlying mechanism for the PARP and BRCA synthetic lethality:Clearing up the misunderstandings. Molecular Oncology, 5(4):387–393, 2011. → pages 17[92] P. Heyn, U. Stenzel, A. W. Briggs, et al. Road blocks on paleogenomes–polymeraseextension profiling reveals the frequency of blocking lesions in ancient DNA. Nucleic AcidsResearch, 38(16):e161, 2010. → pages 22, 101[93] S. Hiltemann, G. Jenster, J. Trapman, et al. Discriminating somatic and germline mutationsin tumor DNA samples without matching normals. Genome Research, 25(9):1382–1390,2015. → pages 103[94] M. Hofreiter, V. Jaenicke, D. Serre, et al. DNA sequences from multiple amplificationsreveal artifacts induced by cytosine deamination in ancient DNA. Nucleic Acids Research,29(23):4793–4799, 2001. → pages 48[95] J. Hong, S. W. Han, H. S. Ham, et al. Phase II study of biweekly S-1 and oxaliplatincombination chemotherapy in metastatic colorectal cancer and pharmacogenetic analysis.Cancer Chemotherapy and Pharmacology, 67(6):1323–1331, 2011. → pages 17, 82[96] L. Huang, F. Chen, Y. Chen, et al. Thymidine phosphorylase gene variant, platelet countsand survival in gastrointestinal cancer patients treated by fluoropyrimidines. ScientificReports, 4(1):5697, 2014. → pages 83[97] Z.-H. Huang, D. Hua, L.-H. Li, et al. Prognostic role of p53 codon 72 polymorphism ingastric cancer patients treated with fluorouracil-based adjuvant chemotherapy. Journal ofCancer Research and Clinical Oncology, 134(10):1129–1134, 2008. → pages 61, 75[98] D. M. Hyman, D. B. Solit, M. E. Arcila, et al. Precision medicine at Memorial SloanKettering Cancer Center: Clinical next-generation sequencing enabling next-generationtargeted therapy trials. Drug Discovery Today, 20(12):1422–1428, 2015. → pages 2[99] F. Innocenti, S. D. Undevia, L. Iyer, et al. Genetic variants in theUDP-glucuronosyltransferase 1A1 gene predict the risk of severe neutropenia of irinotecan.Journal of Clinical Oncology, 22(8):1382–1388, 2004. → pages 85[100] A. Jakobsen, J. N. Nielsen, N. Gyldenkerne, et al. Thymidylate synthase andmethylenetetrahydrofolate reductase gene polymorphism in normal tissue as predictors offluorouracil sensitivity. Journal of Clinical Oncology, 23(7):1365–1369, 2005. → pages 82114[101] B. A. Jennings, Y. K. Loke, J. Skinner, et al. Evaluating Predictive PharmacogeneticSignatures of Adverse Events in Colorectal Cancer Patients Treated with Fluoropyrimidines.PLoS ONE, 8(10):1–9, 2013. → pages 83[102] S. Jones, V. Anagnostou, K. Lytle, et al. Personalized genomic analyses for cancer mutationdiscovery and interpretation. Science Translational Medicine, 7(283ra53), 2015. → pages20, 60, 89, 99, 102, 103, 104, 105[103] A. V. Khrunin, A. Moisseev, V. Gorbunova, et al. Genetic polymorphisms and the efficacyand toxicity of cisplatin-based chemotherapy in ovarian cancer patients. ThePharmacogenomics Journal, 10(1):54–61, 2010. → pages 61, 75[104] J. Kim, D. Kim, J. S. Lim, et al. Accurate detection of low-level somatic mutations withtechnical replication for next-generation sequencing. bioRxiv, 2017. → pages 101[105] J. G. Kim, S. K. Sohn, Y. S. Chae, et al. TP53 codon 72 polymorphism associated withprognosis in patients with advanced gastric cancer treated with paclitaxel and cisplatin.Cancer Chemotherapy and Pharmacology, 64(2):355–360, 2009. → pages 61, 75[106] S. Kim, C. Park, Y. Ji, et al. Deamination Effects in Formalin-Fixed, Paraffin-EmbeddedTissue Samples in the Era of Precision Medicine. Journal of Molecular Diagnostics, 19(1):137–146, 2017. → pages 35, 48, 99, 100, 101, 105[107] D. C. Koboldt, Q. Zhang, D. E. Larson, et al. VarScan 2 : Somatic mutation and copynumber alteration discovery in cancer by exome sequencing VarScan 2 : Somatic mutationand copy number alteration discovery in cancer by exome sequencing. Genome Research,22(3):568–576, 2012. → pages 13[108] D. C. Koboldt, D. E. Larson, and R. K. Wilson. Using VarScan 2 for Germline VariantCalling and Somatic Mutation Detection. Current Protocols in Bioinformatics, 44:15.4.1–15.4.17, 2013. → pages 13[109] Y. Kodama, M. Shumway, and R. Leinonen. The sequence read archive: Explosive growthof sequencing data. Nucleic Acids Research, 40(D1):2011–2013, 2012. → pages 5[110] S. Kumar, R. Marfatia, S. Tannenbaum, et al. Doxorubicin-induced cardiomyopathy 17years after chemotherapy. Texas Heart Institute Journal, 39(3):424–427, 2012. → pages 17[111] D. M. Kweekel, H. Gelderblom, T. Van der Straaten, et al. UGT1A1*28 genotype andirinotecan dosage in patients with metastatic colorectal cancer: a Dutch Colorectal CancerGroup study. British Journal of Cancer, 99(2):275–282, 2008. → pages 85[112] B. Langmead and S. L. Salzberg. Fast gapped-read alignment with Bowtie 2. NatureMethods, 9(4):357–359, 2013. → pages 12[113] J. Laskin, S. Jones, S. Aparicio, et al. Lessons learned from the application ofwhole-genome analysis to the treatment of patients with advanced cancers. Molecular CaseStudies, 1(1):a000570, 2015. → pages 2, 10115[114] C. Ledergerber and C. Dessimoz. Base-calling for next-generation sequencing platforms.Briefings in Bioinformatics, 12(5):489–497, 2011. → pages 11[115] A. M. Lee, Q. Shi, E. Pavey, et al. DPYD variants as predictors of 5-fluorouracil toxicity inadjuvant colon cancer treatment (NCCTG N0147). Journal of the National Cancer Institute,106(12):1–12, 2014. → pages 17, 79, 80[116] J. Leichsenring, A.-L. Volckmar, N. Magios, et al. Synonymous EGFR Variant p.Q787Q isNeither Prognostic Nor Predictive in Patients with Lung Adenocarcinoma Jonas. Genes,Chromosomes & Cancer, 56(3):214–220, 2017. → pages 67[117] F. Leone, G. Cavalloni, Y. Pignochino, et al. Somatic mutations of epidermal growth factorreceptor in bile duct and gallbladder carcinoma. Clinical Cancer Research, 12(6):1680–1685, 2006. → pages 67[118] S. E. Levy and R. M. Myers. Advancements in Next-Generation Sequencing. AnnualReview of Genomics and Human Genetics, 17(1):95–115, 2016. → pages 4, 5[119] B. Li, V. G. Krishnan, M. E. Mort, et al. Automated inference of molecular mechanisms ofdisease from amino acid substitutions. Bioinformatics, 25(21):2744–2750, 2009. → pages15[120] H. Li. Exploring single-sample snp and indel calling with whole-genome de novo assembly.Bioinformatics, 28(14):1838–1844, 2012. → pages 12[121] H. Li and R. Durbin. Fast and accurate short read alignment with Burrows-Wheelertransform. Bioinformatics, 25(14):1754–1760, 2009. → pages[122] H. Li and R. Durbin. Fast and accurate long-read alignment with Burrows-Wheelertransform. Bioinformatics, 26(5):589–595, 2010. → pages 12[123] H. Li and N. Homer. A survey of sequence alignment algorithms for next-generationsequencing. Briefings in Bioinformatics, 11(5):473–483, 2010. → pages 12[124] H. Li, B. Handsaker, A. Wysoker, et al. The Sequence Alignment/Map format andSAMtools. Bioinformatics, 25(16):2078–2079, 2009. → pages 12[125] M.-T. Lin, S. L. Mosier, M. Thiess, et al. Clinical validation of KRAS, BRAF, and EGFRmutation detection using next-generation sequencing. American Journal of ClinicalPathology, 141(6):856–66, 2014. → pages 48, 60[126] Y. Liu, S. N. Xu, Y. S. Chen, et al. Study of single nucleotide polymorphisms of FBW7 andits substrate genes revealed a predictive factor for paclitaxel plus cisplatin chemotherapy inChinese patients with advanced esophageal squamous cell carcinoma. Oncotarget, 7(28):44330–44339, 2016. → pages 70[127] J. J. Lou, L. Mirsadraei, D. E. Sanchez, et al. A review of room temperature storage ofbiospecimen tissue and nucleic acids for anatomic pathology laboratories andbiorepositories. Clinical Biochemistry, 47:267–273, 2014. → pages 35116[128] N. Ludyga, B. Gru¨nwald, O. Azimzadeh, et al. Nucleic acids from long-term preservedFFPE tissues are suitable for downstream analyses. Virchows Archiv, 460(2):131–140, 2012.→ pages 22, 102[129] G. Lunter and M. Goodson. Stampy: A statistical algorithm for sensitive and fast mappingof Illumina sequence reads. Genome Research, 21(6):936–939, 2011. → pages 12[130] L. Mamanova, A. J. Coffey, C. E. Scott, et al. Target-enrichment strategies fornext-generation sequencing. Nature Methods, 7(2):111–118, 2010. → pages 8[131] M. V. Mandola, J. Stoehlmacher, W. Zhang, et al. A 6 bp polymorphism in the thymidylatesynthase gene causes message instability and is associated with decreased intratumoral TSmRNA levels. Pharmacogenetics, 14(5):319–327, 2004. → pages 85[132] Marcel and Martin. Cutadapt removes adapter sequences from high-throughput sequencingreads. EMBnet.journal, 17(1), 2011. → pages 11[133] E. Marcuello, A. Alte´s, A. Menoyo, et al. UGT1A1 gene variations and irinotecan treatmentin patients with metastatic colorectal cancer. British Journal of Cancer, 91(4):678–682,2004. → pages 85[134] E. Marcuello, A. Alte´s, A. Menoyo, et al. Methylenetetrahydrofolate reductase genepolymorphisms: Genomic predictors of clinical response to fluoropyrimidine-basedchemotherapy? Cancer Chemotherapy and Pharmacology, 57(6):835–840, 2006. → pages82[135] E. R. Mardis. Next-Generation Sequencing Platforms. Annual Review of AnalyticalChemistry, 6(1):287–303, 2013. → pages 5[136] E. R. Mardis. DNA sequencing technologies: 20062016. Nature Protocols, 12(2):213–218,2017. → pages 5[137] L. K. Mattison, M. R. Johnson, and R. B. Diasio. A comparative analysis of translateddihydropyrimidine dehydrogenase cDNA; conservation of functional domains and relevanceto genetic polymorphisms. Pharmacogenetics, 12(2):133–44, 2002. → pages 79[138] A. McKenna, M. Hanna, E. Banks, et al. The Genome Analysis Toolkit: A MapReduceframework for analyzing next-generation DNA sequencing data. Genome Research, 20:1297–1303, 2010. → pages 13[139] H. L. McLeod, D. J. Sargent, S. Marsh, et al. Pharmacogenetic predictors of adverse eventsand response to chemotherapy in metastatic colorectal cancer: Results from North AmericanGastrointestinal Intergroup Trial N9741. Journal of Clinical Oncology, 28(20):3227–3233,2010. → pages 82, 85[140] S. R. McWhinney and H. L. McLeod. Using germline genotype in cancer pharmacogeneticstudies. Pharmacogenomics, 10(3):489–493, 2009. → pages 20, 60, 102[141] F. Meric-Bernstam, L. Brusco, M. Daniels, et al. Incidental germline variants in 1000advanced cancers on a prospective somatic genomic profiling protocol. Annals of Oncology,27(5):795–800, 2016. → pages 20, 60, 89, 99, 102117[142] D. Meulendijks, L. M. Henricks, G. S. Sonke, et al. Clinical relevance of DPYD variantsc.1679T>G, c.1236G>A/HapB3, and c.1601G>A as predictors of severefluoropyrimidine-associated toxicity: A systematic review and meta-analysis of individualpatient data. The Lancet Oncology, 16(16):1639–1650, 2015. → pages 79, 80, 81[143] M. Mielczarek and J. Szyda. Review of alignment and SNP calling algorithms fornext-generation sequencing data. Journal of Applied Genetics, 57(1):71–79, 2016. → pages11, 12, 13[144] B. Mohelnikova-Duchonova, B. Melichar, and P. Soucek. FOLFOX/FOLFIRIpharmacogenetics: The call for a personalized approach in colorectal cancer therapy. WorldJournal of Gastroenterology, 20(30):10316–10330, 2014. → pages 17, 18, 19, 20, 62[145] S. Moorthie, A. Hall, and C. F. Wright. Informatics and clinical genome sequencing:opening the black box. Genetics in Medicine, 15(3):165–171, 2013. → pages 14, 15[146] A. Morel, M. Boisdron-Celle, L. Fey, et al. Clinical relevance of differentdihydropyrimidine dehydrogenase gene single nucleotide polymorphisms on 5-fluorouraciltolerance. Molecular Cancer Therapeutics, 5(11):2895–2904, 2006. → pages 79, 80, 81[147] P. C. Ng and S. Henikoff. SIFT: Predicting amino acid changes that affect protein function.Nucleic Acids Research, 31(13):3812–3814, 2003. → pages 15[148] Q. Nie, S. Shrestha, E. E. Tapper, et al. Quantitative contribution of rs75017182 todihydropyrimidine dehydrogenase mRNA splicing and enzyme activity. ClinicalPharmacology & Therapeutics, 00(00):1–9, 2017. → pages 81[149] R. Nielsen, J. S. Paul, A. Albrechtsen, et al. Genotype and SNP calling from next-generationsequencing data. Nature Reviews Genetics, 12(6):443–51, jun 2011. → pages 11, 12, 13[150] S. Nik-Zainal, H. Davies, J. Staaf, et al. Landscape of somatic mutations in 560 breastcancer whole-genome sequences. Nature, 534(7605):47–54, 2016. → pages 7[151] I. Nishino, A. Spinazzola, A. Papadimitriou, et al. Mitochondrial neurogastrointestinalencephalomyopathy: An autosomal recessive disorder due to thymidine phosphorylasemutations. Annals of Neurology, 47(6):792–800, 2000. → pages 84[152] S. M. Offer, C. C. Fossum, N. J. Wegner, et al. Comparative functional analysis of dpydvariants of potential clinical relevance to dihydropyrimidine dehydrogenase activity. CancerResearch, 74(9):2545–2554, 2014. → pages 79, 80, 81[153] R. Ofner, C. Ritter, S. Ugurel, et al. Non-reproducible sequence artifacts in FFPE tissue : anexperience report. Journal of Cancer Research and Clinical Oncology, 143(7):1199–1207,2017. → pages 35, 48, 99, 105[154] E. Oh, Y.-L. Choi, M. J. Kwon, et al. Comparison of Accuracy of Whole-ExomeSequencing with Formalin-Fixed Paraffin-Embedded and Fresh Frozen Tissue Samples.PloS One, 10(12):e0144162, 2015. → pages 35, 48, 99, 100, 105118[155] G. R. Oliver, S. N. Hart, and E. W. Klee. Bioinformatics for clinical next generationsequencing. Clinical Chemistry, 61(1):124–135, 2015. → pages 12, 14, 15[156] S. Pabinger, A. Dander, M. Fischer, et al. A survey of tools for variant analysis ofnext-generation genome sequencing data. Briefings in Bioinformatics, 15(2):256–278, 2014.→ pages 2, 12, 13, 15[157] B. J. Paessens, C. von Schilling, K. Berger, et al. Health resource consumption and costsattributable to chemotherapy-induced toxicity in german routine hospital care inlymphoproliferative disorder and NSCLC patients. Annals of Oncology, 22(10):2310–2319,2011. → pages 17[158] M. Panczyk. Pharmacogenetics research on chemotherapy resistance in colorectal cancerover the last 20 years. World Journal of Gastroenterology, 20(29):9775–9827, 2014. →pages 17, 18, 19, 20, 62[159] Q. Peng, R. Vijaya Satya, M. Lewis, et al. Reducing amplification artifacts in high multiplexamplicon sequencing by using molecular barcodes. BMC Genomics, 16(1):589, 2015. →pages 102[160] K. P. Pennington, T. Walsh, M. Lee, et al. BRCA1, TP53, and CHEK2 germline mutationsin uterine serous carcinoma. Cancer, 119(2):332–338, 2013. → pages 74[161] D. A. Pilger, P. L. Da Costa Lopez, F. Segal, et al. Analysis of R213R and 13494 g > apolymorphisms of the p53 gene in individuals with esophagitis, intestinal metaplasia of thecardia and Barrett’s Esophagus compared with a control group. Genomic Medicine, 1(1-2):57–63, 2007. → pages 74[162] S. Quesnel, S. Verselis, C. Portwine, et al. p53 compound heterozygosity in a severelyaffected child with Li-Fraumeni syndrome. Oncogene, 18(27):3970–3978, 1999. → pages74[163] B. Quinta´ns, A. Ordo´n˜ez-Ugalde, P. Cacheiro, et al. Medical genomics: The intricate pathfrom genetic variant identification to clinical interpretation. Applied and TranslationalGenomics, 3(3):60–67, 2014. → pages 14[164] B. Rabbani, M. Tekin, and N. Mahdieh. The promise of whole-exome sequencing inmedical genetics. Journal of Human Genetics, 59(1):5–15, 2014. → pages 10[165] N. Rahman. Realizing the promise of cancer predisposition genes. Nature, 505(7483):302–308, 2014. → pages 17, 20[166] B. J. Raphael, R. H. Hruban, A. J. Aguirre, et al. Integrated Genomic Characterization ofPancreatic Ductal Adenocarcinoma. Cancer Cell, 32(2):185–203.e13, 2017. → pages 7[167] N. Rashid, H. A. Koh, H. C. Baca, et al. Economic burden related to chemotherapy-relatedadverse events in patients with metastatic breast cancer in an integrated health care system.Breast Cancer: Targets and Therapy, 8:173–181, 2016. → pages 17119[168] M. C. Ravnan and M. S. Matalka. Vemurafenib in Patients With BRAF V600EMutation-Positive Advanced Melanoma. Clinical Therapeutics, 34(7):1474–1486, 2012. →pages 1[169] V. M. Raymond, S. W. Gray, S. Roychowdhury, et al. Germline findings in tumor-onlysequencing: Points to consider for clinicians and laboratories. Journal of the NationalCancer Institute, 108(4):1–5, 2016. → pages 89, 95[170] H. L. Rehm. Disease-targeted sequencing: a cornerstone in the clinic. Nature ReviewsGenetics, 14(4):295–300, 2013. → pages 10[171] S. Richards, N. Aziz, S. Bale, et al. Standards and guidelines for the interpretation ofsequence variants: a joint consensus recommendation of the American College of MedicalGenetics and Genomics and the Association for Molecular Pathology. Genetics inMedicine, 17(5):405–423, 2015. → pages 7[172] A. G. Robertson, J. Shih, C. Yau, et al. Integrative Analysis Identifies Four Molecular andClinical Subsets in Uveal Melanoma. Cancer Cell, 32(2):204–220.e15, 2017. → pages 7[173] J. M. Rothberg, W. Hinz, T. M. Rearick, et al. An integrated semiconductor device enablingnon-optical genome sequencing. Nature, 475(7356):348–352, 2011. → pages 3[174] E. Rouits, V. Charasson, A. Pe´tain, et al. Pharmacokinetic and pharmacogeneticdeterminants of the activity and toxicity of irinotecan in metastatic colorectal cancerpatients. British Journal of Cancer, 99:1239–45, 2008. → pages 85[175] J. D. Rowley. A New Consistent Chromosomal Abnormality in Chronic MyelogenousLeukaemia identified by Quinacrine Fluorescence and Giemsa Staining. Nature, 243(5405):290–293, 1973. → pages 1[176] A. Ruzzo, F. Graziano, F. Loupakis, et al. Pharmacogenetic profiling in patients withadvanced colorectal cancer treated with first-line FOLFOX-4 chemotherapy. Journal ofClinical Oncology, 25(10):1247–1254, 2007. → pages 82[177] A. Ruzzo, F. Graziano, F. Loupakis, et al. Pharmacogenetic profiling in patients withadvanced colorectal cancer treated with first-line FOLFIRI chemotherapy. ThePharmacogenomics Journal, 8(4):278–288, 2008. → pages 85[178] L. P. Rybak, D. Mukherjea, S. Jajoo, et al. Cisplatin ototoxicity and protection: clinical andexperimental studies. The Tohoku Journal of Experimental Medicine, 219(3):177–86, 2009.→ pages 17[179] V. N. Rykalina, A. A. Shadrin, V. S. Amstislavskiy, et al. Exome sequencing from nanogramamounts of starting DNA: Comparing three approaches. PLoS ONE, 9(7), 2014. → pages 2[180] E. Samorodnitsky, B. M. Jewell, R. Hagopian, et al. Evaluation of Hybridization CaptureVersus Amplicon-Based Methods for Whole-Exome Sequencing. Human Mutation, 36(9):903–914, 2015. → pages 102120[181] S. Sanderson, R. Zimmern, M. Kroese, et al. How can the evaluation of genetic tests beenhanced? Lessons learned from the ACCE framework and evaluating genetic tests in theUnited Kingdom. Genetics in Medicine, 7(7):495–500, 2005. → pages 15, 16[182] C. T. Saunders, W. S. W. Wong, S. Swamy, et al. Strelka: Accurate somatic small-variantcalling from sequenced tumor-normal sample pairs. Bioinformatics, 28(14):1811–1817,2012. → pages 13[183] C. L. Scherr, N. M. Lindor, T. L. Malo, et al. A preliminary investigation of geneticcounselors’ information needs when receiving a variant of uncertain significance result: amixed methods study. Genetics in Medicine, 17(9):739–746, 2015. → pages 7[184] K. A. Schrader, D. T. Cheng, V. Joseph, et al. Germline Variants in Targeted TumorSequencing Using Matched Normal DNA. JAMA Oncology, 2(1):1–8, 2015. → pages 20,60, 89, 99, 102, 104[185] M. Schwab, U. M. Zanger, C. Marx, et al. Role of genetic and nongenetic factors forfluorouracil treatment-related severe toxicity: A prospective clinical trial by the German5-FU toxicity study group. Journal of Clinical Oncology, 26(13):2131–2138, 2008. →pages 79, 80, 82[186] A. D. Seidman, D. Berry, C. Cirrincione, et al. Randomized phase III trial of weeklycompared with every-3-weeks paclitaxel for metastatic breast cancer, with trastuzumab forall HER-2 overexpressors and random assignment to trastuzumab or not in HER-2nonoverexpressors: final results of Cancer and Leu. Journal of Clinical Oncology, 26(10):1642–9, 2008. → pages 1[187] B. S. Sheffield, B. Tessier-Cloutier, H. Li-Chang, et al. Personalized oncogenomics in themanagement of gastrointestinal carcinomas-early experiences from a pilot study. CurrentOncology, 23(6):e571–e575, 2016. → pages 2[188] S.-R. Shi, R. J. Cote, L. Wu, et al. DNA extraction from archival formalin-fixed,paraffin-embedded tissue sections based on the antigen retrieval principle: heating under theinfluence of pH. Journal of Histochemistry and Cytochemistry, 50(8):1005–1011, 2002. →pages 22, 100[189] J. A. Sikorsky, D. A. Primerano, T. W. Fenger, et al. DNA damage reduces Taq DNApolymerase fidelity and PCR amplification efficiency. Biochemical and BiophysicalResearch Communications, 355(2):431–437, 2007. → pages 35, 99, 105[190] R. Simon and S. Roychowdhury. Implementing personalized cancer genomics in clinicaltrials. Nature Reviews Drug Discovery, 12(5):358–369, 2013. → pages 8, 10[191] S. Sjogren, M. Inganas, A. Lindgren, et al. Prognostic and predictive value of c-erbB-2overexpression in primary breast cancer, alone and in combination with other prognosticmarkers. Journal of Clinical Oncology, 16:462–469, 1998. → pages 1[192] E. H. Slager, M. W. Honders, E. D. V. D. Meijden, et al. Identification of the angiogenicendothelial-cell growth factor-1 / thymidine phosphorylase as a potential target for121immunotherapy of cancer Identification of the angiogenic endothelial-cell growth factor-1 /thymidine phosphorylase as a potential target. Blood, 107(12):4954–4960, 2006. → pages83[193] D. J. Slamon, G. M. Clark, S. G. Wong, et al. Human Breast Cancer: Correlation of Relapseand Survival with Amplification of the HER- 2/neu Oncogene. Science, 235(4785):177–182, 1987. → pages 1[194] A. So, A. Vilborg, Y. Bouhlal, et al. A Robust Targeted Sequencing Approach for Low Inputand Variable Quality DNA from Clinical Samples. bioRxiv, 2017. → pages 2[195] D. H. Spencer, J. K. Sehn, H. J. Abel, et al. Comparison of clinical targeted next-generationsequence data from formalin-fixed and fresh-frozen tissue specimens. Journal of MolecularDiagnostics, 15(5):623–633, 2013. → pages 100[196] J. Stoehlmacher, D. J. Park, W. Zhang, et al. A multivariate analysis of genomicpolymorphisms: prediction of clinical outcome to 5-FU/oxaliplatin combinationchemotherapy in refractory colorectal cancer. British Journal of Cancer, (May):344–354,2004. → pages 82, 85[197] S. P. Strom. Current practices and guidelines for clinical next-generation sequencingoncology testing. Cancer Biology & Medicine, 13(1):3–11, 2016. → pages 7, 15, 104[198] K. W. Suh, J. H. Kim, D. Y. Kim, et al. Which Gene is a Dominant Predictor of ResponseDuring FOLFOX Chemotherapy for the Treatment of Metastatic Colorectal Cancer, theMTHFR or XRCC1 Gene? Annals of Surgical Oncology, 13(11):1379–1385, 2006. →pages 82[199] M. A. Sukhai, K. J. Craddock, M. Thomas, et al. A classification system for clinicalrelevance of somatic variants identified in molecular profiling of cancer. Genetics inMedicine, 18(2):128–136, 2016. → pages 7[200] J. J. Swen, M. Nijenhuis, A. de Boer, et al. Pharmacogenetics: From Bench to Byte AnUpdate of Guidelines. Clinical Pharmacology & Therapeutics, 89(5):662–673, 2011. →pages 79, 80, 81[201] D. Tanaka, A. Hishida, K. Matsuo, et al. Polymorphism of dihydropyrimidinedehydrogenase (DPYD) Cys29Arg and risk of six malignancies in Japanese. NagoyaJournal of Medical Science, 67(Fig 1):117–124, 2005. → pages 81[202] R. Tian, M. K. Basu, and E. Capriotti. Computational methods and resources for theinterpretation of genomic variants in cancer. BMC genomics, 16 Suppl 8(Suppl 8):S7, 2015.→ pages 101[203] F. Tirode, D. Surdez, X. Ma, et al. Genomic landscape of ewing sarcoma defines anaggressive subtype with co-association of STAG2 and TP53 mutations. Cancer Discovery, 4(11):1342–1353, 2014. → pages 7122[204] G. Toffoli, E. Cecchin, G. Corona, et al. The role of UGT1A1*28 polymorphism in thepharmacodynamics and pharmacokinetics of irinotecan in patients with metastatic colorectalcancer. Journal of Clinical Oncology, 24(19):3061–3068, 2006. → pages 85[205] G. Toffoli, L. Giodini, A. Buonadonna, et al. Clinical validity of a DPYD-basedpharmacogenetic test to predict severe toxicity to fluoropyrimidines. International Journalof Cancer, 137(12):2971–2980, 2015. → pages 79, 80[206] F. Torri, I. D. Dinov, A. Zamanyan, et al. Next generation sequence analysis andcomputational genomics using graphical pipeline workflows. Genes, 3(3):545–575, 2012.→ pages 2[207] A. B. P. van Kuilenburg, J. Haasjes, D. J. Richel, et al. Clinical implications ofdihydropyrimidine dehydrogenase (DPD) deficiency in patients with severe5-fluorouracil-associated toxicity: Identification of new mutations in the DPD gene.Clinical Cancer Research, 6(12):4705–4712, 2000. → pages 79, 80, 81[208] A. B. P. van Kuilenburg, J. Meijer, M. W. T. Tanck, et al. Phenotypic and clinicalimplications of variants in the dihydropyrimidine dehydrogenase gene. Biochimica etBiophysica Acta - Molecular Basis of Disease, 1862(4):754–762, 2016. → pages 79, 80, 81[209] F. J. Vega, P. Iniesta, T. Caldes, et al. P53 Exon 5 Mutations as a Prognostic Indicator ofShortened Survival in Non-Small-Cell Lung Cancer. British Journal of Cancer, 76(1):44–51, 1997. → pages 75[210] A. Villani, U. Tabori, J. Schiffman, et al. Biochemical and imaging surveillance in germlineTP53 mutation carriers with Li-Fraumeni syndrome: A prospective observational study. TheLancet Oncology, 12(6):559–567, 2011. → pages 74[211] V. Vincek, M. Nassiri, M. Nadji, et al. A Tissue Fixative that Protects Macromolecules(DNA, RNA, and Protein) and Histomorphology in Clinical Samples. LaboratoryInvestigation, 83(10):1427–1435, 2003. → pages 105[212] C. L. Vogel, M. A. Cobleigh, D. Tripathy, et al. Efficacy and safety of trastuzumab as asingle agent in first-line treatment of HER2-overexpressing metastatic breast cancer.Journal of Clinical Oncology, 20(3):719–26, 2002. → pages 1[213] D. von Hansemann. Ueber asymmetrische Zelltheilung in epithel Krebsen und derenbiologische Bedeutung. Virchows Arch. Path. Anat., 119(2):299–326, 1890. → pages 1[214] H. Wang and F. Meng. Theoretical study of proton-catalyzed hydrolytic deaminationmechanism of adenine. Theoretical Chemistry Accounts, 127(5):561–571, 2010. → pages48, 101[215] Y. Wang, W. Bao, H. Shi, et al. Epidermal growth factor receptor exon 20 mutationincreased in post-chemotherapy patients with non-small cell lung cancer detected withpatients’ blood samples. Translational Oncology, 6(4):504–10, 2013. → pages 67, 68[216] Y. Wang, Q. Yang, and Z. Wang. The evolution of nanopore sequencing. Frontiers inGenetics, 5(DEC):1–20, 2014. → pages 3123[217] K. Wetterstrand. DNA sequencing costs: data from the NHGRI genome sequencingprogram (GSP), 2016. URLhttps://www.genome.gov/27541954/dna-sequencing-costs-data/. Accessed September 12,2017. → pages 2[218] C. Wong, R. A. DiCioccio, H. J. Allen, et al. Mutations in BRCA1 from fixed,paraffin-embedded tissue can be artifacts of preservation. Cancer Genetics andCytogenetics, 107(1):21–27, 1998. → pages 101[219] N. A. Wong, D. Gonzalez, M. Salto-Tellez, et al. RAS testing of colorectal carcinoma-aguidance document from the Association of Clinical Pathologists Molecular Pathology andDiagnostics Group. Journal of Clinical Pathology, 67(9):751–757, 2014. → pages 60[220] S. Q. Wong, J. Li, R. Salemi, et al. Targeted-capture massively-parallel sequencing enablesrobust detection of clinically informative mutations from formalin-fixed tumours. ScientificReports, 3(3):3494, 2013. → pages 22, 35, 99, 100, 102, 105[221] S. Q. Wong, J. Li, A. Y-C Tan, et al. Sequence artefacts in a prospective series offormalin-fixed tumours tested for mutations in hotspot regions by massively parallelsequencing. BMC Medical Genomics, 7(1):1–10, 2014. → pages 8, 22, 35, 48, 56, 99, 100,102, 105[222] C. Xu, M. Nezami Ranjbar, Z. Wu, et al. Detecting very low allele fraction variants usingtargeted DNA sequencing and a novel molecular barcode-aware variant caller. BMCGenomics, 18(1):1–11, 2017. → pages 101[223] V. K. Yadav and S. De. An assessment of computational methods for estimating purity andclonality using genomic data derived from heterogeneous tumor tissue samples. Briefings inBioinformatics, 16(2):232–241, 2015. → pages 104[224] M. Yang, Y. Guo, X. Zhang, et al. Interaction of P53 Arg72Pro and MDM2 T309Gpolymorphisms and their associations with risk of gastric cardia cancer. Carcinogenesis, 28(9):1996–2001, 2007. → pages 61, 75[225] S. S. Yea, S. S. Lee, W.-Y. Kim, et al. Genetic variations and haplotypes ofUDP-glucuronosyltransferase 1A locus in a Korean population. Therapeutic DrugMonitoring, 30(1):23–34, 2008. → pages 85[226] T. Yoneda, A. Kuboyama, K. Kato, et al. Association of MDM2 SNP309 and TP53Arg72Pro polymorphisms with risk of endometrial cancer. Oncology Reports, 30(1):25–34,2013. → pages 61, 75[227] Y. Zha, P. Gan, Q. Liu, et al. TP53 Codon 72 Polymorphism Predicts Efficacy of PaclitaxelPlus Capecitabine Chemotherapy in Advanced Gastric Cancer Patients. Archives of MedicalResearch, 47(1):13–18, 2016. → pages 61, 75[228] W. Zhang, L. P. Stabile, P. Keohavong, et al. Mutation and polymorphism in the EGFR-TKdomain associated with lung cancer. Journal of Thoracic Oncology, 1(7):635–47, 2006. →pages 67124[229] X. Zhang, G. Ao, Y. Wang, et al. Genetic variants and haplotypes of the UGT1A9, 1A7 and1A1 genes in Chinese Han. Genetics and Molecular Biology, 35(2):428–434, 2012. →pages 85[230] Y. Zhang, L. Liu, Y. Tang, et al. Polymorphisms in TP53 and MDM2 contribute to higherrisk of colorectal cancer in Chinese population: A hospital-based, case-control study.Molecular Biology Reports, 39(10):9661–9668, 2012. → pages 61, 75[231] Z. Z. Zhu, A. Z. Wang, H. R. Jia, et al. Association of the TP53 codon 72 polymorphismwith colorectal cancer in a Chinese population. Japanese Journal of Clinical Oncology, 37(5):385–390, 2007. → pages 61, 75[232] R. L. Zimmern and M. Kroese. The evaluation of genetic tests. Journal of Public Health, 29(3):246–250, 2007. → pages 15, 16[233] J. Zining, X. Lu, H. Caiyun, et al. Genetic polymorphisms of mTOR and cancer risk: asystematic review and updated meta-analysis. Oncotarget, 7(35), 2016. → pages 69, 71125Appendix ASupporting Materials126Table A.1: Target regions and amplicons of the OncoPanel. Genomic regions are presented in the 0-based coordinate system.Gene Target Target region Amplicon region Amplicon length(bp)Amplicon GC content(%)AKT1 AKT1 a chr14:105246545-105246554 chr14:105246426-105246549 123 56chr14:105246508-105246651 143 62AKT1 AKT1 b chr14:105246449-105246458 chr14:105246426-105246549 123 56ALK ALK a chr2:29445212-29445274 chr2:29445125-29445298 173 55ALK ALK b chr2:29443629-29443703 chr2:29443598-29443783 185 52BRAF BRAF a chr7:140481395-140481418 chr7:140481369-140481472 103 44BRAF BRAF b chr7:140453129-140453152 chr7:140453054-140453256 202 37chr7:140453084-140453218 134 38DPYD DPYD a chr1:97915607-97915621 chr1:97915561-97915715 154 39chr1:97915565-97915675 110 39DPYD DPYD b chr1:98348878-98348892 chr1:98348806-98349041 235 33chr1:98348854-98349048 194 34DPYD DPYD c chr1:97981336-97981350 chr1:97981294-97981367 73 47chr1:97981302-97981501 199 46chr1:97981339-97981478 139 44DPYD DPYD d chr1:97547940-97547954 chr1:97547867-97547976 109 41chr1:97547890-97547980 90 40DPYD DPYD e chr1:97981417-97981425 chr1:97981302-97981501 199 46chr1:97981339-97981478 139 44DPYD DPYD f chr1:98039415-98039423 chr1:98039299-98039458 159 45chr1:98039303-98039444 141 44DPYD DPYD g chr1:97770916-97770924 chr1:97770882-97771019 137 43127Gene Target Target region Amplicon region Amplicon length(bp)Amplicon GC content(%)DPYD DPYD h chr1:98165087-98165095 chr1:98164981-98165128 147 44chr1:98164995-98165134 139 43EGFR EGFR 18 chr7:55241611-55241738 chr7:55241576-55241789 213 56chr7:55241580-55241751 171 52chr7:55241626-55241822 196 55EGFR EGFR 19 chr7:55242412-55242515 chr7:55242355-55242548 193 47chr7:55242357-55242539 182 46EGFR EGFR 20 chr7:55248983-55249173 chr7:55248931-55249219 288 60chr7:55248958-55249121 163 63chr7:55249036-55249220 184 57EGFR EGFR 21 chr7:55259409-55259569 chr7:55259367-55259564 197 54chr7:55259367-55259618 251 54chr7:55259486-55259683 197 51ERBB2 ERBB2 20 chr17:37880976-37881166 chr17:37880956-37881156 200 60chr17:37881009-37881184 175 58GSTP1 GSTP1 a chr11:67352682-67352696 chr11:67352653-67352763 110 55HRAS HRAS a chr11:534281-534293 chr11:534207-534319 112 56chr11:534221-534328 107 58HRAS HRAS b chr11:533870-533884 chr11:533768-533926 158 57chr11:533839-533944 105 60HRAS HRAS c chr11:533549-533557 chr11:533469-533606 137 60chr11:533492-533604 112 60HRAS HRAS d chr11:533462-533470 chr11:533362-533490 128 69chr11:533375-533532 157 68128Gene Target Target region Amplicon region Amplicon length(bp)Amplicon GC content(%)chr11:533469-533606 137 60IDH1 IDH1 a chr2:209113110-209113113 chr2:209113081-209113224 143 45chr2:209113081-209113233 152 44IDH2 IDH2 a chr15:90631932-90631935 chr15:90631870-90631993 123 52chr15:90631870-90632013 143 51IDH2 IDH2 b chr15:90631836-90631839 chr15:90631716-90631861 145 62chr15:90631730-90631890 160 62KIT KIT 11 chr4:55593579-55593710 chr4:55593484-55593753 269 39chr4:55593515-55593749 234 39KIT KIT 13 chr4:55594174-55594289 chr4:55594134-55594327 193 44KIT KIT 14 chr4:55595498-55595653 chr4:55595415-55595657 242 36chr4:55595461-55595701 240 37chr4:55595591-55595691 100 39KIT KIT 17 chr4:55599233-55599360 chr4:55599192-55599390 198 38chr4:55599249-55599471 222 37KIT KIT 18 chr4:55602661-55602777 chr4:55602617-55602825 208 40KIT KIT 9 chr4:55592020-55592218 chr4:55591945-55592179 234 42chr4:55592075-55592262 187 38KRAS KRAS a chr12:25398274-25398291 chr12:25398185-25398329 144 39chr12:25398185-25398332 147 39KRAS KRAS b chr12:25380269-25380283 chr12:25380213-25380316 103 49KRAS KRAS c chr12:25378554-25378568 chr12:25378521-25378619 98 34chr12:25378521-25378625 104 35KRAS KRAS d chr12:25378644-25378653 chr12:25378617-25378774 157 34129Gene Target Target region Amplicon region Amplicon length(bp)Amplicon GC content(%)chr12:25378618-25378777 159 33MAP2K1 MAP2K1 a chr15:66727449-66727486 chr15:66727398-66727542 144 54chr15:66727420-66727512 92 53MAPK1 MAPK1 1 chr22:22221609-22221732 chr22:22221487-22221662 175 76chr22:22221490-22221659 169 75chr22:22221595-22221688 93 68chr22:22221601-22221702 101 68chr22:22221634-22221798 164 78chr22:22221637-22221803 166 77MAPK1 MAPK1 2 chr22:22161950-22162137 chr22:22161907-22162096 189 43chr22:22161907-22162136 229 42chr22:22161958-22162176 218 41chr22:22162016-22162178 162 43MAPK1 MAPK1 3 chr22:22160136-22160330 chr22:22160021-22160175 154 33chr22:22160122-22160286 164 45chr22:22160220-22160371 151 41MAPK1 MAPK1 4 chr22:22153298-22153419 chr22:22153248-22153419 171 44chr22:22153264-22153479 215 40chr22:22153343-22153540 197 38MAPK1 MAPK1 5 chr22:22142980-22143099 chr22:22142925-22143098 173 46chr22:22143051-22143177 126 45MAPK1 MAPK1 6 chr22:22142543-22142679 chr22:22142465-22142665 200 36chr22:22142499-22142749 250 36chr22:22142546-22142745 199 36130Gene Target Target region Amplicon region Amplicon length(bp)Amplicon GC content(%)MAPK1 MAPK1 7 chr22:22127159-22127273 chr22:22127112-22127303 191 47chr22:22127112-22127307 195 47MAPK1 MAPK1 8 chr22:22123490-22123611 chr22:22123448-22123624 176 40chr22:22123459-22123654 195 38chr22:22123499-22123697 198 38MTHFR MTHFR a chr1:11856371-11856385 chr1:11856347-11856476 129 53chr1:11856349-11856435 86 51MTHFR MTHFR b chr1:11854469-11854483 chr1:11854433-11854519 86 52chr1:11854433-11854568 135 52MTOR MTOR 0 chr1:11319302-11319468 chr1:11319264-11319432 168 57chr1:11319360-11319552 192 48chr1:11319403-11319553 150 48MTOR MTOR 1 chr1:11318539-11318652 chr1:11318494-11318653 159 45chr1:11318549-11318730 181 40MTOR MTOR 10 chr1:11298456-11298676 chr1:11298401-11298570 169 52chr1:11298473-11298672 199 56chr1:11298608-11298733 125 48MTOR MTOR 11 chr1:11297897-11298107 chr1:11297829-11298062 233 56chr1:11297830-11298025 195 56chr1:11297932-11298128 196 57chr1:11297938-11298137 199 56chr1:11298015-11298209 194 55MTOR MTOR 12 chr1:11294197-11294324 chr1:11294140-11294320 180 54chr1:11294179-11294376 197 51131Gene Target Target region Amplicon region Amplicon length(bp)Amplicon GC content(%)MTOR MTOR 13 chr1:11293452-11293546 chr1:11293416-11293575 159 42chr1:11293416-11293587 171 41chr1:11293502-11293640 138 34MTOR MTOR 14 chr1:11292490-11292587 chr1:11292453-11292614 161 44chr1:11292478-11292624 146 42MTOR MTOR 15 chr1:11291354-11291493 chr1:11291283-11291482 199 51chr1:11291301-11291523 222 51chr1:11291395-11291540 145 51MTOR MTOR 16 chr1:11290979-11291113 chr1:11290930-11291107 177 50chr1:11290995-11291191 196 48MTOR MTOR 17 chr1:11288722-11288977 chr1:11288694-11288837 143 52chr1:11288783-11288938 155 52chr1:11288866-11289048 182 49MTOR MTOR 18 chr1:11276202-11276293 chr1:11276136-11276316 180 44chr1:11276254-11276394 140 45MTOR MTOR 19 chr1:11273453-11273625 chr1:11273386-11273568 182 50chr1:11273419-11273657 238 47chr1:11273523-11273673 150 46MTOR MTOR 2 chr1:11316987-11317224 chr1:11316917-11317167 250 57chr1:11316937-11317132 195 56chr1:11317039-11317219 180 53chr1:11317059-11317290 231 53chr1:11317122-11317289 167 52MTOR MTOR 20 chr1:11272850-11272967 chr1:11272815-11272975 160 49132Gene Target Target region Amplicon region Amplicon length(bp)Amplicon GC content(%)chr1:11272871-11273029 158 48MTOR MTOR 21 chr1:11272366-11272533 chr1:11272332-11272525 193 54chr1:11272412-11272590 178 51MTOR MTOR 22 chr1:11270868-11270965 chr1:11270785-11270983 198 39chr1:11270830-11271009 179 40chr1:11270901-11271087 186 38MTOR MTOR 23 chr1:11269366-11269517 chr1:11269329-11269510 181 50chr1:11269329-11269544 215 48chr1:11269415-11269611 196 43MTOR MTOR 24 chr1:11264615-11264762 chr1:11264525-11264725 200 53chr1:11264583-11264812 229 59chr1:11264668-11264812 144 56MTOR MTOR 25 chr1:11259595-11259762 chr1:11259547-11259742 195 49chr1:11259621-11259794 173 48MTOR MTOR 26 chr1:11259312-11259462 chr1:11259269-11259496 227 49chr1:11259283-11259481 198 51chr1:11259299-11259490 191 53MTOR MTOR 27 chr1:11227496-11227576 chr1:11227454-11227608 154 44MTOR MTOR 28 chr1:11217206-11217350 chr1:11217177-11217376 199 54chr1:11217177-11217384 207 54chr1:11217262-11217457 195 46MTOR MTOR 29 chr1:11210180-11210285 chr1:11210155-11210336 181 48chr1:11210155-11210351 196 48MTOR MTOR 3 chr1:11316046-11316251 chr1:11315980-11316216 236 54133Gene Target Target region Amplicon region Amplicon length(bp)Amplicon GC content(%)chr1:11315982-11316181 199 55chr1:11316099-11316281 182 53chr1:11316161-11316373 212 48MTOR MTOR 30 chr1:11206730-11206850 chr1:11206694-11206859 165 49chr1:11206742-11206922 180 50MTOR MTOR 31 chr1:11205022-11205104 chr1:11204986-11205137 151 49MTOR MTOR 32 chr1:11204702-11204814 chr1:11204662-11204861 199 52chr1:11204675-11204861 186 52MTOR MTOR 33 chr1:11199587-11199717 chr1:11199411-11199595 184 51chr1:11199557-11199732 175 56chr1:11199615-11199777 162 46MTOR MTOR 34 chr1:11199358-11199494 chr1:11199316-11199494 178 49chr1:11199321-11199531 210 47chr1:11199411-11199595 184 51MTOR MTOR 35 chr1:11194405-11194525 chr1:11194361-11194533 172 56chr1:11194435-11194632 197 53MTOR MTOR 36 chr1:11193134-11193256 chr1:11193062-11193253 191 57chr1:11193124-11193296 172 53MTOR MTOR 37 chr1:11190583-11190836 chr1:11190480-11190665 185 52chr1:11190523-11190768 245 60chr1:11190618-11190812 194 61chr1:11190643-11190869 226 60chr1:11190664-11190864 200 60MTOR MTOR 38 chr1:11189792-11189897 chr1:11189760-11189906 146 53134Gene Target Target region Amplicon region Amplicon length(bp)Amplicon GC content(%)chr1:11189828-11189939 111 54MTOR MTOR 39 chr1:11188909-11189010 chr1:11188860-11189060 200 44chr1:11188881-11189061 180 44MTOR MTOR 4 chr1:11313893-11314032 chr1:11313839-11314007 168 55chr1:11313920-11314062 142 47MTOR MTOR 40 chr1:11188508-11188611 chr1:11188475-11188658 183 46chr1:11188475-11188689 214 44MTOR MTOR 41 chr1:11188058-11188185 chr1:11188017-11188196 179 55chr1:11188090-11188285 195 45MTOR MTOR 42 chr1:11187678-11187865 chr1:11187630-11187824 194 48chr1:11187644-11187880 236 51chr1:11187745-11187945 200 50chr1:11187755-11187964 209 50MTOR MTOR 43 chr1:11187064-11187203 chr1:11187021-11187193 172 49chr1:11187083-11187254 171 47MTOR MTOR 44 chr1:11186676-11186855 chr1:11186642-11186789 147 50chr1:11186733-11186894 161 45MTOR MTOR 45 chr1:11184552-11184692 chr1:11184522-11184710 188 48chr1:11184527-11184726 199 49MTOR MTOR 46 chr1:11182033-11182185 chr1:11181999-11182166 167 56chr1:11182026-11182217 191 54MTOR MTOR 47 chr1:11181300-11181427 chr1:11181248-11181448 200 57chr1:11181341-11181492 151 49MTOR MTOR 48 chr1:11177058-11177145 chr1:11177000-11177154 154 38135Gene Target Target region Amplicon region Amplicon length(bp)Amplicon GC content(%)chr1:11177010-11177180 170 40chr1:11177111-11177167 56 46chr1:11177123-11177247 124 38MTOR MTOR 49 chr1:11175450-11175527 chr1:11175417-11175568 151 49MTOR MTOR 5 chr1:11307873-11308153 chr1:11307713-11307911 198 48chr1:11307824-11307989 165 54chr1:11307947-11308136 189 53chr1:11307996-11308224 228 46chr1:11308035-11308224 189 42MTOR MTOR 50 chr1:11174867-11174946 chr1:11174829-11174981 152 41MTOR MTOR 51 chr1:11174372-11174512 chr1:11174347-11174531 184 53chr1:11174366-11174566 200 52MTOR MTOR 52 chr1:11172906-11172976 chr1:11172880-11173039 159 45chr1:11172882-11173039 157 45MTOR MTOR 53 chr1:11169703-11169788 chr1:11169648-11169834 186 42chr1:11169670-11169834 164 43MTOR MTOR 54 chr1:11169344-11169429 chr1:11169300-11169473 173 39chr1:11169300-11169490 190 40MTOR MTOR 55 chr1:11168235-11168345 chr1:11168184-11168362 178 47chr1:11168230-11168424 194 47MTOR MTOR 56 chr1:11167539-11167559 chr1:11167453-11167651 198 38chr1:11167457-11167616 159 38MTOR MTOR 6 chr1:11307679-11307792 chr1:11307627-11307819 192 50chr1:11307713-11307911 198 48136Gene Target Target region Amplicon region Amplicon length(bp)Amplicon GC content(%)MTOR MTOR 7 chr1:11303168-11303359 chr1:11303118-11303318 200 53chr1:11303125-11303369 244 51chr1:11303216-11303408 192 50MTOR MTOR 8 chr1:11301607-11301740 chr1:11301563-11301748 185 56chr1:11301635-11301811 176 50MTOR MTOR 9 chr1:11300357-11300606 chr1:11300300-11300533 233 55chr1:11300308-11300507 199 56chr1:11300388-11300637 249 58chr1:11300393-11300590 197 58chr1:11300494-11300669 175 51NRAS NRAS a chr1:115258737-115258754 chr1:115258618-115258778 160 51chr1:115258707-115258813 106 49NRAS NRAS b chr1:115256522-115256536 chr1:115256435-115256573 138 44chr1:115256499-115256573 74 44NRAS NRAS c chr1:115252286-115252294 chr1:115252219-115252339 120 45NRAS NRAS d chr1:115252199-115252207 chr1:115252152-115252262 110 46PDGFRA PDGFRA 12 chr4:55141005-55141142 chr4:55140958-55141182 224 45PDGFRA PDGFRA 14 chr4:55144060-55144175 chr4:55144023-55144217 194 47PDGFRA PDGFRA 18 chr4:55152005-55152132 chr4:55151966-55152172 206 50PIK3CA PIK3CA a chr3:178936076-178936102 chr3:178936001-178936244 243 34chr3:178936046-178936198 152 36PIK3CA PIK3CA b chr3:178952078-178952092 chr3:178952032-178952128 96 39chr3:178952039-178952141 102 39PTEN PTEN a chr10:89692985-89692999 chr10:89692941-89693051 110 39137Gene Target Target region Amplicon region Amplicon length(bp)Amplicon GC content(%)PTEN PTEN b chr10:89717666-89717680 chr10:89717587-89717717 130 45chr10:89717589-89717728 139 46PTEN PTEN c chr10:89717765-89717785 chr10:89717709-89717831 122 36chr10:89717744-89717891 147 29STAT1 STAT1 0 chr2:191874599-191874731 chr2:191874528-191874718 190 42chr2:191874613-191874793 180 41STAT1 STAT1 1 chr2:191873686-191873835 chr2:191873624-191873871 247 38chr2:191873635-191873803 168 38chr2:191873697-191873892 195 38STAT1 STAT1 10 chr2:191851762-191851796 chr2:191851613-191851799 186 35chr2:191851649-191851803 154 31chr2:191851690-191851887 197 26chr2:191851777-191851888 111 24STAT1 STAT1 11 chr2:191851577-191851675 chr2:191851498-191851620 122 48chr2:191851576-191851671 95 49chr2:191851613-191851799 186 35chr2:191851649-191851803 154 31STAT1 STAT1 12 chr2:191850342-191850388 chr2:191850312-191850430 118 38STAT1 STAT1 13 chr2:191849033-191849121 chr2:191848981-191849156 175 37chr2:191849006-191849181 175 39chr2:191849069-191849244 175 41STAT1 STAT1 14 chr2:191848365-191848468 chr2:191848325-191848479 154 55chr2:191848356-191848530 174 51STAT1 STAT1 15 chr2:191847106-191847246 chr2:191847028-191847218 190 38138Gene Target Target region Amplicon region Amplicon length(bp)Amplicon GC content(%)chr2:191847030-191847274 244 39chr2:191847178-191847370 192 43STAT1 STAT1 16 chr2:191845343-191845397 chr2:191845312-191845453 141 42STAT1 STAT1 17 chr2:191844495-191844594 chr2:191844415-191844614 199 39chr2:191844445-191844651 206 35chr2:191844553-191844675 122 34STAT1 STAT1 18 chr2:191843579-191843729 chr2:191843525-191843725 200 59chr2:191843601-191843759 158 58STAT1 STAT1 19 chr2:191841563-191841753 chr2:191841511-191841754 243 45chr2:191841512-191841692 180 46chr2:191841629-191841810 181 41chr2:191841677-191841810 133 41STAT1 STAT1 2 chr2:191872286-191872389 chr2:191872197-191872377 180 33chr2:191872207-191872447 240 31chr2:191872329-191872389 60 40chr2:191872342-191872470 128 28STAT1 STAT1 20 chr2:191840535-191840615 chr2:191840499-191840649 150 40STAT1 STAT1 21 chr2:191839553-191839660 chr2:191839503-191839699 196 48chr2:191839516-191839699 183 48STAT1 STAT1 22 chr2:191835426-191835445 chr2:191835395-191835502 107 34STAT1 STAT1 3 chr2:191865797-191865891 chr2:191865727-191865932 205 40chr2:191865750-191865938 188 41STAT1 STAT1 4 chr2:191864349-191864432 chr2:191864320-191864483 163 38chr2:191864320-191864484 164 38139Gene Target Target region Amplicon region Amplicon length(bp)Amplicon GC content(%)STAT1 STAT1 5 chr2:191862940-191863036 chr2:191862850-191863096 246 33chr2:191862910-191863091 181 35STAT1 STAT1 6 chr2:191862579-191862735 chr2:191862544-191862710 166 51chr2:191862648-191862845 197 30chr2:191862668-191862840 172 29STAT1 STAT1 7 chr2:191859784-191859947 chr2:191859744-191859926 182 42chr2:191859744-191859962 218 44chr2:191859799-191859994 195 45chr2:191859800-191859996 196 44STAT1 STAT1 8 chr2:191855951-191856048 chr2:191855914-191856111 197 48chr2:191855915-191856093 178 50STAT1 STAT1 9 chr2:191854338-191854402 chr2:191854239-191854418 179 28chr2:191854270-191854480 210 25chr2:191854374-191854505 131 31STAT3 STAT3 0 chr17:40500404-40500536 chr17:40500316-40500516 200 47chr17:40500411-40500610 199 51STAT3 STAT3 1 chr17:40498584-40498733 chr17:40498525-40498708 183 46chr17:40498525-40498749 224 46chr17:40498650-40498844 194 42chr17:40498692-40498844 152 41STAT3 STAT3 10 chr17:40481762-40481796 chr17:40481632-40481773 141 45chr17:40481704-40481863 159 45chr17:40481731-40481843 112 45STAT3 STAT3 11 chr17:40481569-40481667 chr17:40481494-40481693 199 50140Gene Target Target region Amplicon region Amplicon length(bp)Amplicon GC content(%)chr17:40481533-40481703 170 47chr17:40481632-40481773 141 45STAT3 STAT3 12 chr17:40481425-40481477 chr17:40481366-40481506 140 49chr17:40481373-40481505 132 50STAT3 STAT3 13 chr17:40478131-40478219 chr17:40478074-40478256 182 51STAT3 STAT3 14 chr17:40476978-40477081 chr17:40476949-40477109 160 56chr17:40476950-40477142 192 56STAT3 STAT3 15 chr17:40476726-40476866 chr17:40476640-40476780 140 58chr17:40476674-40476910 236 53chr17:40476716-40476913 197 49STAT3 STAT3 16 chr17:40475588-40475645 chr17:40475546-40475687 141 41STAT3 STAT3 17 chr17:40475275-40475374 chr17:40475235-40475407 172 48STAT3 STAT3 18 chr17:40475019-40475163 chr17:40474995-40475190 195 54chr17:40475030-40475210 180 51STAT3 STAT3 19 chr17:40474297-40474514 chr17:40474229-40474410 181 46chr17:40474229-40474479 250 45chr17:40474329-40474561 232 47chr17:40474343-40474515 172 46chr17:40474468-40474602 134 48STAT3 STAT3 2 chr17:40497574-40497677 chr17:40497541-40497718 177 51chr17:40497542-40497739 197 51STAT3 STAT3 20 chr17:40469197-40469244 chr17:40469143-40469279 136 45chr17:40469162-40469305 143 47STAT3 STAT3 21 chr17:40468804-40468921 chr17:40468739-40468917 178 49141Gene Target Target region Amplicon region Amplicon length(bp)Amplicon GC content(%)chr17:40468862-40469026 164 48STAT3 STAT3 22 chr17:40467760-40467820 chr17:40467728-40467860 132 54STAT3 STAT3 3 chr17:40491329-40491429 chr17:40491290-40491450 160 57chr17:40491313-40491458 145 57STAT3 STAT3 4 chr17:40490746-40490832 chr17:40490645-40490892 247 34chr17:40490692-40490890 198 33STAT3 STAT3 5 chr17:40489778-40489877 chr17:40489711-40489910 199 53chr17:40489732-40489909 177 53STAT3 STAT3 6 chr17:40489450-40489606 chr17:40489368-40489564 196 48chr17:40489411-40489659 248 50chr17:40489491-40489686 195 50STAT3 STAT3 7 chr17:40485906-40486069 chr17:40485844-40486043 199 43chr17:40485844-40486071 227 43chr17:40485925-40486165 240 43chr17:40485962-40486143 181 40STAT3 STAT3 8 chr17:40485688-40485785 chr17:40485635-40485806 171 54chr17:40485696-40485860 164 60STAT3 STAT3 9 chr17:40483487-40483551 chr17:40483423-40483600 177 32TP53 TP53 0 chr17:7579836-7579914 chr17:7579781-7579946 165 60chr17:7579810-7579966 156 57TP53 TP53 1 chr17:7579697-7579723 chr17:7579553-7579751 198 57chr17:7579599-7579757 158 58TP53 TP53 2 chr17:7579309-7579592 chr17:7579268-7579457 189 60chr17:7579378-7579534 156 62142Gene Target Target region Amplicon region Amplicon length(bp)Amplicon GC content(%)chr17:7579442-7579617 175 56chr17:7579553-7579751 198 57TP53 TP53 3 chr17:7578368-7578556 chr17:7578326-7578501 175 65chr17:7578418-7578597 179 58TP53 TP53 4 chr17:7578174-7578291 chr17:7578076-7578331 255 54chr17:7578084-7578279 195 54chr17:7578238-7578327 89 55TP53 TP53 5 chr17:7577496-7577610 chr17:7577463-7577641 178 56TP53 TP53 6 chr17:7577016-7577157 chr17:7576986-7577057 71 61chr17:7576988-7577210 222 55chr17:7576995-7577060 65 63chr17:7577041-7577214 173 53TP53 TP53 7 chr17:7576850-7576928 chr17:7576793-7576973 180 45TP53 TP53 8 chr17:7573924-7574035 chr17:7573874-7574063 189 61chr17:7573894-7574073 179 59TP53 TP53 9 chr17:7572924-7573010 chr17:7572851-7573045 194 52chr17:7572895-7573046 151 50TYMP TYMP 0 chr22:50967922-50968140 chr22:50967858-50968047 189 67chr22:50967971-50968139 168 69chr22:50968080-50968243 163 71TYMP TYMP 1 chr22:50967562-50967769 chr22:50967507-50967698 191 62chr22:50967604-50967783 179 64chr22:50967718-50967883 165 68TYMP TYMP 2 chr22:50966938-50967041 chr22:50966860-50967060 200 55143Gene Target Target region Amplicon region Amplicon length(bp)Amplicon GC content(%)chr22:50966976-50967117 141 60TYMP TYMP 3 chr22:50966014-50966148 chr22:50965980-50966177 197 57TYMP TYMP 4 chr22:50965591-50965714 chr22:50965553-50965744 191 61TYMP TYMP 5 chr22:50965002-50965169 chr22:50964791-50965035 244 77chr22:50964973-50965125 152 71chr22:50965012-50965194 182 70chr22:50965108-50965319 211 66TYMP TYMP 6 chr22:50964672-50964907 chr22:50964619-50964809 190 73chr22:50964644-50964892 248 75chr22:50964739-50964935 196 77chr22:50964791-50965035 244 77TYMP TYMP 7 chr22:50964427-50964587 chr22:50964354-50964541 187 79chr22:50964354-50964585 231 79chr22:50964448-50964636 188 77TYMP TYMP 8 chr22:50964196-50964349 chr22:50964160-50964348 188 72chr22:50964184-50964377 193 73TYMS TYMS a chr18:673437-673451 chr18:673367-673498 131 35chr18:673367-673516 149 36UGT1A1 UGT1A1 a chr2:234668849-234668909 chr2:234668782-234668979 197 49chr2:234668782-234669030 248 51chr2:234668806-234668958 152 46144