Background: Head and neck squamous cell carcinomas (HNSCCs) are complex and highly aggressive tumours that develop in the mouth, throat, salivary glands and nose. HNSCCs account for more than half a million cases annually and are the sixth most common cancer worldwide. Alcohol, tobacco and human papilloma virus (HPV) infection are the well-known causes for HNSCC. The current options for treatment are surgery, radiotherapy, chemotherapy or a combination of therapies. Locally advanced HNSCC patients show heterogenous response to standard treatments and the survival after 5 years is about 50%. Therefore, there is a need to identify biomarkers to predict outcome and improve personalised therapies. The recent advancement in next generation sequencing technologies has allowed for understanding the molecular characteristics of the tumour and identify patients at high risk that are unresponsive to the standard treatment. HPV-associated oropharyngeal carcinoma have shown a very high rate of loco-regional control (LRC) and overall survival (OS) after postoperative radio- chemotherapy (PORT-C) and are being assessed for treatment de-escalation strategies to reduce toxicity in clinical trials. The treatment response of HPV-negative HNSCC, however, is still heterogeneous and novel biomarkers are required to identify subgroups of patients for treatment adaptation. Objectives: The overall aim of the thesis is to develop biomarkers to identify patients at high risk for future treatment adaptations and improve personalised radiotherapy based on the biological differences in HNSCC patients. For this purpose, novel gene signatures were developed and validated using machine learning approaches and biological information in order to predict LRC in patients with locally advanced HNSCC. The novel gene signatures will help to identify patients at high risk that do not respond to standard treatments and to further understand the molecular mechanisms involved in heterogenous treatment response. Materials and methods: The data from a total of 504 locally advanced HNSCC patients of the German Cancer Consortium - Radiation Oncology Group (DKTK-ROG) treated with postoperative radiotherapy (PORT) or postoperative radiochemotherapy (PORT-C) were evaluated. Data from 60 mice bearing xenografts of ten established human HNSCC cell lines were also evaluated. Gene expression analyses was performed using the GeneChip Human Transcriptome Array 2.0 and nanoString analyses. Differential gene expression analysis, Cox regression analysis and machine learning algorithms were used to develop gene signatures. Models were built on the training cohort and then applied on an independent validation cohort. Results: The patients with HPV-negative HNSCC that were treated with PORT-C were classified into the four molecular subtypes basal, mesenchymal, atypical and classical that were previously reported for HNSCC patients treated with primary radio(chemo)therapy or surgery and were related to LRC. The mesenchymal subtype had the worst prognosis as compared to the other subtypes. These tumours were associated with overexpression of epithelial-mesenchymal transition genes and DNA repair genes. A novel 6-gene signature was developed and validated based on full-transcriptome data using machine-learning approaches that was prognostic for LRC in patients with HPV-negative HNSCC treated with PORT-C. The 6-gene signature consisted of four individual genes CAV1, GPX8, IGLV3-25, TGFBI and one metagene combining the highly correlated genes, INHBA and SERPINE1. The identified gene signature was combined with the clinical parameters, T stage and tumour localisation as well as the stem-cell marker CD44 and the 15-gene hypoxia- associated classifier and this improved the performance of the model. Previously identified prognostic gene signatures and molecular-subtype classification were back-translated from HNSCC patients to pre-clinical tumour models. The tumour models were classified into the four subtypes basal, mesenchymal, atypical and classical, similar to the patients. The mesenchymal tumours were significantly associated with a higher TCD50 as compared to other subtypes. A novel 2-gene signature consisting of FN1 and SERPINE1 was developed based on tumour models and patient data using differential gene expression analysis. The 2-gene signature was prognostic for the TCD50 in tumour models and was successfully validated on an independent PORT-C patient cohort for LRC. A matched-pair analysis was performed between patients that were treated with postoperative radiochemotherapy and patients that were treated with postoperative radiotherapy. A 2- metagene signature, consisting of KRT6A, KRT6B, KRT6C forming one metagene and SPRR1A, SPRR1B, SPRR2A, SPRR2C forming the second metagene, was identified. The novel predictive signature stratified patients into high and low risk groups. The high-risk group patients that received PORT-C showed higher LRC as compared to the high-risk patients that received PORT. Thus, the predictive gene signature identified patients that were considered to be at intermediate risk according to clinical factors but were at biologically high risk for the development of loco-regional recurrences after PORT. These patients might benefit from PORT-C treatment. Conclusions: In this thesis, novel gene signatures were identified by combining machine learning and biological information to stratify locally advanced HNSCC patients into high and low risk groups for loco-regional control. This information could be used in the future, e.g. to adjust radiotherapy doses based on the risk group. The developed gene signatures could be combined with other gene signatures or the molecular subtype stratification to develop potential combined treatment approaches. Within the DKTK-ROG framework, the gene signatures will be incorporated with biomarkers developed on the same cohort at the other DKTK-ROG partner sites using the data from different omics platforms in the future. This would help to better understand the molecular basis of heterogenous treatment response in HNSCC patients and uncover novel targets for therapies. The thesis also provides a valuable insight into the applicability of preclinical tumour models to study the efficacy of personalised radiotherapy treatments. Overall, the gene signatures identified in this thesis were from retrospective studies and have to be validated in prospective studies before their application in interventional clinical trials to improve personalised radiotherapy treatments. Additionally, the methods used in the thesis to identify the gene signatures could be used and applied across different cancer datasets for identification of biomarkers. Therefore, this thesis has provided a basis for future studies on personalized treatment of HNSCC based on their genetic profile.:Content
Abbreviations VII
Tables XII
1 Introduction 1
2 Biological and statistical background 6
2.1 Head and neck squamous cell carcinoma 6
2.1.1 Tumourigenesis 6
2.1.2 Biomarkers: clinical and genomics 9
2.2 Statistics 12
General statistical analyses 17
2.3 Gene expression analyses 18
3 Molecular subtypes and mechanisms of radioresistance 20
3.1 Introduction and motivation 20
3.2 Patient cohort and experimental design 21
3.2.1 Patient cohort 21
3.2.2 Clinical endpoints and statistical analysis 23
3.2.3 Experimental design 23
3.3 Results 26
3.3.1 Prognostic factors for LRC and OS 26
3.3.2 Death as competing risk 26
3.3.3 Multivariable Cox regression for improved prognosis 29
3.3.4 Molecular subtypes in HPV-negative HNSCC patients 31
3.3.5 Molecular subtypes are prognostic for LRC after PORT-C 33
3.4 Discussion 36
4 A novel 6-gene signature for LRC prognosis 39
4.1 Introduction and motivation 39
4.2 Patient cohort and experimental design 40
4.2.1 Patient cohorts 40
4.2.2 Clinical endpoints and statistical analysis 41
4.2.3 Experimental design 41
4.3 Results 44
4.3.1 Characteristics of the patient cohorts 44
4.3.2 Development of the 6-gene signature prognostic for LRC 45
4.3.3 Combination of the 6-gene signature and clinical parameters 47
4.3.4 Extension with CD44 and the 15-gene hypoxia signature 48
4.3.5 Prognostic for secondary endpoints 49
4.3.6 Technical validation using nanoString technology 52
4.3.7 Death as competing risk 56
4.4 Discussion 58
5 Biomarker development in preclinical tumour models and HNSCC patients 62
5.1 Introduction and motivation 62
5.2 Patient cohort and experimental design 64
5.2.1 Patient derived xenograft tumour models 64
5.2.2 Patient cohorts 64
5.2.3 Clinical endpoints and statistical analysis 65
5.2.4 Experimental design 65
5.3 Results 68
5.3.1 Molecular subtypes 68
5.3.2 Development of the 2-gene signature 70
5.3.3 Technical validation using the nanoString technology 71
5.3.4 Back-translation of gene signatures in xenograft models 75
5.4 Discussion 79
6 PORT-C improves LRC in intermediate-risk patients 82
6.1 Introduction and motivation 82
6.2 Patient cohort and experimental design 83
6.2.1 Patient cohorts 84
6.2.2 Clinical endpoints and statistical analysis 84
6.2.3 Experimental design 84
6.3 Results 87
6.3.1 Characteristics of the patient cohorts 87
6.3.2 Propensity score matching analysis 88
6.3.3 Development of the predictive 2-metagene signature 90
6.4 Discussion 93
7 Conclusion and future perspectives 96
8 Summary 99
9 Zusammenfassung 102
Appendix 105
A. Supplementary Figures 105
B. Supplementary Tables 110
Bibliography 116
Erklärungen 149
| Identifer | oai:union.ndltd.org:DRESDEN/oai:qucosa:de:qucosa:82766 |
| Date | 22 December 2022 |
| Creators | Patil, Shivaprasad |
| Contributors | Löck, Steffen, Börries, Melanie, Technische Universität Dresden |
| Source Sets | Hochschulschriftenserver (HSSS) der SLUB Dresden |
| Language | English |
| Detected Language | English |
| Type | info:eu-repo/semantics/publishedVersion, doc-type:doctoralThesis, info:eu-repo/semantics/doctoralThesis, doc-type:Text |
| Rights | info:eu-repo/semantics/openAccess |
| Relation | 10.1016/j.radonc.2021.12.049, 10.1016/j.radonc.2022.04.006, 10.3390/cancers14123031 |
Page generated in 0.0035 seconds