Abstract
In the context of cancer, clonal hematopoiesis of indeterminate potential (CHIP) is associated with the development of therapy-related myeloid neoplasms and shorter overall survival. Cell-free DNA (cfDNA) sequencing is becoming widely adopted for genomic screening of cancer patients but has not been used extensively to determine CHIP status due to a requirement for matched blood and tumor sequencing. Here we present an accurate machine learning approach to determine clonal hematopoiesis (CH) status from cfDNA sequencing alone and apply our model to 4,096 oncology clinical cfDNA samples. Using this method, we determine that 26% of patients in this cohort have evidence of CH and CH is most common in lung cancer patients. Matched RNAseq data shows signals of increased inflammation, especially neutrophil activation, within the tumor microenvironment of CH-positive patients. Additionally, CH patients showed evidence of neutrophil activation systemically, pointing to a potential mechanism of action for the worse outcomes associated with CH status. Neutrophil activation may be one of many mechanisms however, as estrogen positive breast cancer patients harboring TET2 frameshift mutations had worse outcomes but similar neutrophil levels to CH-negative patients.
One Sentence Summary We train an accurate machine learning model to detect clonal hematopoiesis in cancer and characterize associated changes in the tumor microenvironment.
Competing Interest Statement
All authors were employed by Novartis during their work on this manuscript.
Footnotes
↵* These authors jointly supervised this work