Ernst, Corinna 
ORCID: 0000-0001-7756-8815
(2021).
Computational methods for improved cancer risk prediction based on multi-gene panel analysis in a routine diagnostic setting.
PhD thesis, Universität zu Köln.
|
PDF
Thesis_final.pdf - Published Version Download (785kB) | Preview |
Abstract
Multi-gene panel approaches allow for screening for putative disease-causing genetic factors in several genes simultaneously, while keeping costs, storage requirements, and computational times comparatively low compared to whole genome or exome sequencing. Therefore, multi-gene panel sequencing has become a standard approach for the investigation and diagnosis of various diseases with a hereditary component in clinical labs wordwide, and several tools have become established for multi-gene panel data processing, providing automized, easy-to-use solutions for the detection and annotation of single nucleotide variants (SNVs) and short insertions and deletions (indels) in exons of well-established disease-associated genes. This thesis demonstrates, in the context of personalized risk prediction for familial breast and ovarian cancer (BC/OC), how bioinformatic analyses that go beyond standard variant calling with automized workflows, can contribute to an improvement of genetic testing based on multi-gene panels in clinical diagnostics. These improvements include multi-gene panel design, variant detection, and variant interpretation. Since only less than one third of BC/OC cases with a familial burden can be explained by a germline mutation in confirmed high- to moderate-risk genes, the search for further genetic risk factors is ongoing, but may be hindered by low mutation prevalences in the corresponding genes that demand for huge sample sizes in order to achieve sufficient statistical power. Here, the potential association of pathogenic variants in suspected risk genes FANCM, BARD1, and BRIP1 with hereditary BC and/or OC was assessed in case-control studies including well-characterized index patients and geographically matched female controls. FANCM and BARD1 were confirmed as risk genes for hereditary BC, and BRIP1 was confirmed as a highly penetrant OC risk gene without pronounced effects on BC risk. Consequently, coding regions of FANCM and BARD1 should be included in sequencing targets of multi-gene panels for diagnostic germline testing of individuals at risk for familial BC, and coding regions of BRIP1 for individuals at risk for familial OC, respectively. Recent studies revealed that BC/OC risks are modified by additional genetic factors,i.e., common SNVs and indels which are usually not even located in coding genomic regions. Investigation of these BC/OC-associated polymorphisms represents a paradigm shift in contrast to the analysis of rare (pathogenic) variants. Their effects are not sufficiently large to contribute individually to BC/OC risks, but they can be combined into polygenic risk scores (PRS), which could achieve clinically useful degrees of risk discrimination. In collaboration with Julika Borde, I assessed the utility of PRSs for BC risk prediction in a clinical cohort of females carrying a heterozygous protein-truncating variant (PTV) in CHEK2 and were independent of former genome-wide association studies. We found that, based on PRSs, BC risk can be stratified such that CHEK2 PTV carriers may have both a BC risk equivalent to that of the general population but, on the other hand, may also fall into risk groups for which access to intensified prophylactic measures is recommended. The SNP sets employed in our study each comprise less than 100 loci, and hence, have the potential to be straightforwardly implemented into multi-gene panel analyses. In addition, SNPs can be used for ethnicity checks and quality assurance purposes. Detection of large genomic insertions or deletions, so-called copy number variants (CNVs), from sequencing data requires read depth-based approaches that go beyond standard variant calling. In a joint work with Louisa Lepkes, the utility of these in silico CNV detection approaches for multi-gene panel data in clinical diagnostics was evaluated, and the prevalence of CNVs in cancer predisposition genes in individuals at risk for familial BC/OC was assessed. We showed that CNVs constitute a non-negligible fraction in the spectrum of putative BC/OC-causing variants, namely 1.81% in our study sample of 4208 female index patients. However, due to high proportions of false positive predictions, which primarily accumulated at the extremes of the length or GC content distribution of sequencing targets, wet lab verification of in silico predicted CNVs is required in the framework of clinical diagnostics. The third part of the thesis deals with considerations regarding the interpretation of genetic testing outcomes in the context of genetic counseling. Interpretation of missense mutations is a particularly challenging task, as their impact to protein function are difficult to predict and they can therefore often only be classified as variants of uncertain significance (VUS). Thus, the use of in silico approaches for automated variant classification has become established in many laboratories. I evaluated the performance of four in silico prediction tools embedded in the widely-used, commercial AlamutTM Visual software (Interactive Biosoftware, Rouen, France) and found that all tools under investigation suffered from poor specificities, resulting in an unacceptable proportion of variants falsely classified as pathogenic, and that this shortcoming could not be bypassed by considering the predictions in combination. Thus, clinical consequences should never be based solely on in silico forecasts, but my findings indicate that in silico prediction tools provide clues to the benignity of variants. In collaboration with Dr. Jan Hauke, the determination of variant pathogenicity based on the comparison of observed variant allele fractions (VFs) in paired blood- and tumor-derived samples was assessed, considering 208 rare BRCA1/2 germline variants in 181 OC patients. Our results demonstrate that a significantly increased VF in tumor in comparison to the corresponding blood-derived sample are insufficient to infer pathogenicity, but decreased VFs may provide a suitable criterion for the assessment of BRCA1/2 variants as benign. In collaboration with Konstantin Weber-Lassalle, I investigated pairwise blood- and tumor-derived DNA samples of OC patients with the aim to prove the existence of pathogenic variants in TP53 and PPM1D in blood cells arising from clonal hematopoiesis (CH) rather than from germline inheritance, and to evaluate the frequency of CH occurences in dependence to the exposure to chemotherapy. We found that CH represents a frequent event following chemotherapy, affecting 26 out of the 523 OC index patients enrolled in our study sample. Therefore, the possibility of CH always has to be considered prior to a potential misdiagnosis of Li-Fraumeni syndrome 1, a cancer predisposition syndrome linked to pathogenic variants in TP53.
| Item Type: | Thesis (PhD thesis) | ||||||||
| Creators: |
|
||||||||
| URN: | urn:nbn:de:hbz:38-524483 | ||||||||
| Date: | 2021 | ||||||||
| Language: | English | ||||||||
| Faculty: | Faculty of Mathematics and Natural Sciences | ||||||||
| Divisions: | Faculty of Medicine > Sonstiges > Centrum für integrierte Onkologie (CIO) | ||||||||
| Subjects: | Data processing Computer science Life sciences |
||||||||
| Uncontrolled Keywords: |
|
||||||||
| Date of oral exam: | 22 June 2021 | ||||||||
| Referee: |
|
||||||||
| Refereed: | Yes | ||||||||
| URI: | http://kups.ub.uni-koeln.de/id/eprint/52448 |
Downloads
Downloads per month over past year
Export
Actions (login required)
 |
View Item |