gms | German Medical Science

Introduction: Melanoma is the most serious type of skin cancer that frequently spreads to other organs of the human body. Especially melanoma metastases to the brain (intracranial metastases) are hard to treat and a major cause of death of melanoma patients. Little is known about molecular alterations and altered mechanisms that distinguish brain from extracranial melanoma metastases. Almost all existing studies compared brain metastases from one set of patients to extracranial metastases of other melanoma patients. This neglects the important facts that each melanoma is highly individual and that brain and extracranial melanoma metastases from the same patient are more similar to each other than to melanoma metastases from other patients in the same organ. To overcome this, a personalized analysis of transcriptomes of patient-matched metastasis pairs is presented focusing on results of our recently developed Hidden Markov Model (HMM) approach [1]. This study was done in the frame of the BMBF-funded e:Med junior research alliance MelBrainSys (01ZX1913A/B).

Methods: Bulk RNA-sequencing of the transcriptomes of patient-matched melanoma metastasis pairs of 16 patients was performed. Chromosomal log-ratio gene expression profiles were computed for each metastasis pair and analyzed by a three-state Hidden Markov Model (HMM) to identify differentially expressed genes for each individual metastasis pair. A global HMM was trained for all patients using a Bayesian Baum Welch algorithm that integrates prior knowledge about expected gene expression changes. The most likely expression state of each gene was predicted by state-posterior decoding. Basics of the utilized HMM-approach were transferred from our prior methodological developments for the HMM-based analysis of tumor gene expression data measured on microarrays [2], [3] and our recently published HMM-based analysis of genomewide DNA-methylation profiles of the patient-matched melanoma metastasis pairs [4].

Results: The utilized HMM-approach for the analysis of patient-matched metastasis pairs will be introduced. The importance of the Bayesian Baum Welch training to obtain biologically relevant models will be presented. Prior findings of comparisons to related tools for DNA copy number analyses (e.g. GLAD, CBS, BioHMM) and paired t-tests will be used to demonstrate the value of the HMM-approach. Moreover, the predictions of the HMM support the existence of a brain-like phenotype and the downregulation of immune pathways in brain metastases. Further, many genes that were recurrently predicted to be altered were significantly overlapping with two related bulk transcriptome studies and a single-cell transcriptome study. The expression levels of several of these genes were also significantly associated with survival of melanoma patients from TCGA indicating a potential clinical relevance of our findings.

Conclusions: Our findings clearly demonstrate the great value of the HMM-approach for the comparative personalized analysis of metastasis transcriptomes and contribute to a better characterization of molecular differences between brain and extracranial melanoma metastases.

The authors declare that they have no competing interests.

The authors declare that a positive ethics committee vote has been obtained.