Preview

PDF, English - main document Download (2MB) | Terms of use

Abstract

The dissertation aimed at the establishment of blood-based microRNA and protein signatures for prediction and early detection of tumour recurrence in patients who had undergone resection of pancreatic ductal adenocarcinoma (PDAC). Utilising a microarray of 2,977 antibodies, variations were detected in the protein content of serum samples collected from 101 patients, who had experienced tumour recurrence or not, including consecutively collected samples from the same patients. Secretome analyses of non-tumours and cancer cells indicated tumour-related variations. Selected biomarkers were utilised to train support vector machine classifiers. They were validated on new, prospectively collected samples from 36 patients in order to document applicability. By combination of biomarkers selected by both a focussed tumour-centred approach and a broader systemic analysis, a classifier of 10 proteins was defined that discriminated patients with recurrence from those without at 91% accuracy. Validation on prospectively collected samples achieved an accuracy of 85%. Recurrence detection was on average 3.5 months earlier than that with current processes. Besides diagnosis, protein signatures were established that allow predicting the period, after which tumour recurrence is likely to occur. I studied the microRNA (miRNA) content of 149 serum samples by means of small RNA sequencing. For the discovery phase, 75 serum samples were analysed. Libraries were prepared and sequenced at the sequencing core facility of DKFZ. Data was obtained in Fastq files and processed using the Heidelberg Unix System Analysis Resource (HUSAR). In total, I analysed 135 miRNA variations between recurrence and nonrecurrence samples using logistic regression and selected informative miRNA biomarkers after removal of unnecessary covariates by Least Absolute Shrinkage Selection Operator (LASSO) regression. To find the best possible miRNA combination, I used Recursive Feature Elimination (RFE) with 5-fold cross validation. A miRNA classifier made of hsa-mir-100, hsa-mir-215, hsa-mir-3916, hsa-mir-484, hsa-mir-6752, hsa-mir-6773, hsa-mir-6883-5P was constructed and trained. The algorithm parameters were optimized to avoid over- or underfitting. The signature was validated in an independent cohort with all parameters being fixed. The miRNA classifier could discriminate between recurrence from nonrecurrence at an accuracy of 97% and 91% in the discovery and validation cohort, respectively. Furthermore, I established four miRNAs classifiers that accurately predicted the time when recurrence is likely to happen. 2 Combining both miRNAs and proteins was done using the samples that were tested simultaneously by antibody microarray and small RNA sequencing. The data was randomly divided into a training and validation cohorts and RFE with 5-fold cross validation was applied to the 17 miRNAs and protein markers. Using a marker signature of two miRNAs and two proteins, I was able to detect pancreatic cancer recurrence at 91% accuracy, which was slightly reduced to 83% upon validation. Analysis of the protein and miRNA contents of blood permits prediction and detection of tumour recurrence in PDAC patients after curative surgery with an accuracy that substantially surpasses the performance of currently used processes, in particular CA19-9 testing. The analysis also indicated the existence of changes that are either directly due to the tumour’s presence or based on the body’s systemic reaction to it. Combining both miRNA and proteins reduced the number of molecules required to achieve an accurate and robust diagnosis. The results could have a direct and immediate benefit for patients with pancreatic cancer and could be translated to clinical practice quickly. In addition, the process could proof the applicability of the signatures for early diagnosis of the primary tumour. Thus, the results could be applied to screening individuals who are at high risk of pancreatic cancer, potentially having a clinical impact beyond the detection of tumour relapse. In addition, the established assays could serve as a means for monitoring disease progression during chemotherapeutic treatment.