Origin Unknown? Not For Long...

Although most cancer patients present with a primary tumor, up to 15 percent first come to the oncologist’s attention with metastatic disease without a clear origin (1). Pathological study of cancers of unknown primary site (CUP) is challenging, and in roughly two to five percent of cases, no primary site is found (2) – resulting in a lengthy diagnostic process, and potentially delaying treatment. But what if a computer program could help identify the source?

Uniting genetics and computer science, an international team of researchers have created a potential solution – a program, known as TumorTracer, which analyzes DNA mutations and mutation patterns in tissue samples to identify the location of the primary tumor. “We had been doing research comparing somatic mutations across different types of cancer, to determine which ones might respond to specific chemotherapies. And various groups had published pan-cancer analyses of the somatic mutations found in various cancers. One day, it occurred to us that we could turn the problem sideways – using somatic mutations to identify the cancer type instead,” says Aron Eklund, co-author of the associated paper (3).

The team analyzed three aspects of somatic mutations – point mutations in cancer driver genes, copy number variations, and base substitution frequencies – and discovered that all three contributed independently to cancer identification, explains Eklund. They used this information to build and validate their algorithm. The initial results show promise: the algorithm classified some initial tumors with known primary sites with 85 percent accuracy. Analysis currently takes around 48 hours, but as sequencing becomes faster, this could be reduced.

The next steps will be to extend the range of cancers the algorithm can identify, and further optimize it. “One obvious application is to help diagnose metastatic tumors whose primary site hasn’t been identified. We don’t yet know whether or not our method will be more accurate than existing methods based on histopathology, various scans and examinations, gene expression signatures, and so on. But we imagine that it won’t be long before every tumor biopsy gets sequenced to identify targetable mutations — and then applying TumorTracer would require only negligible incremental time and cost. So even if TumorTracer is only marginally useful in aiding diagnosis, we think it could find widespread use,” says Eklund.