In search of tumor primary site in cervical CUP syndrome using DNA methylation patterns

Introduction Cervical CUP syndrome (cancer of unknown primary) is defined as a lymph node metastasis and the lack of a corresponding primary tumor. This leads to invasive diagnostic algorithms and unspecific therapy of the patient. Publicly available DNA methylation patterns, which show high tissue specificity, of different cancer entities were analyzed to evaluate the prediction of the tumor primary site.

Methods Methylome data of 1874 cases of four types of cancer (HNSC, BRCA, ESCA and LUSC) from The Cancer Genome Atlas (TCGA) was analyzed in the programming language R. Clustering was performed by t-distributed stochastic neighbor embedding (t-SNE), a classification algorithm commonly used in bioinformatics. Furthermore, a random forest classifier, another machine learning algorithm, was implemented to predict the cancer entities based on DNA methylation patterns.

Results The clustering analysis showed distinguishable groups which could be attributed directly to the four different entities of cancer. The classifier was able to correctly predict tumor type in a randomly selected validation set of the original 1874 cases correctly with an accuracy of 99 % . The cluster of head and neck cancer patients showed further subgroups, which were associated with clinical data. One group showed statistically significant increased overall survival (p<0.01)

Conclusions A variety of different tumor entities can be differentiated by analyzing DNA methylation patterns of the primary tumor site. DNA methylation pattern analysis therefore has potential as a novel technique to detect the tumor primary site of cervical CUP syndrome. It is to be evaluated whether DNA methylation patterns of metastatic lymph node tissue are also suited to predict the primary tumor.

Poster-PDF A-1262.pdf

Publication History

Article published online:
13 May 2021