Tracing cancer back to its source using computer dissection

Cancer can spread through the body without the patient knowing that they had a primary tumour in the colon, lung, breast, prostate or other organ. When patients with metastatic cancer visit their doctor with symptoms it is then a difficult task to identify the tissue of origin of the cancer and so treat it accordingly. A European team has now tested software that can classify superficially identical tumours based on data obtained from a gene expression microarray analysis of biopsied tumour tissue. Given that metastatic cancer of unknown primary site (CUP) is one of the 10 most frequent cancer diagnoses worldwide the software could prove indispensible in treating cancer that has reached this stage.

Previous studies have shown that it is possible to distinguish between metastatic cancer tissues even though they look almost identical under the microscope by analysing them chemically using DNA micro-arrays. Unfortunately, such arrays, while widely used in research laboratories, are complex and expensive and not viable in the clinical setting of a hospital or oncology unit. Commonly, samples would have to be sent to specialist laboratories, which delays results and is expensive. A much simpler approach is to use automated software that can analyse tissues "in silico."

Torik Ayoubi of KU Leuven in Belgium and colleagues at the Maastricht University Medical Centre in The Netherlands have now tested the TCLASS software from http://www.mastergenix.com/ on a random set of 100 metastatic cancers that make up part of the large Expression Project for Oncology and found that that they could classify them according to their original cancer type just as accurately as any of the available laboratory tools. The advantage is that the TCLASS approach is extremely simple to implement and could potentially be further developed as a major diagnostic tool.

The software was able to correctly identify the bladder as the primary cancer site in patient data associated with numerous cancers: the set consisted of 3 bladder, 9 breast, 4 cervix, 22 colon, 10 endometrium, 2 kidney, 6 lung, 3 lymphoma, 9 melanoma, 2 myometrium, 17 ovary, 1 pancreas, 1 pharynx, 1 prostate, 1 salivary gland, 1 skin, 3 soft tissue and 3 stomach cancers. It identified pelvic lymph nodes, metastatic cells in the lung, abdominal skin metastases as originally deriving from breast tumours, liver cancer cells as having their source in colon cancer and many other combinations with an overall accuracy of 88%. The team points out that an analysis of the 12 metastatic tumours where the software failed to identify the tumour origin revealed that in 6 of these cases, the tumour had unambiguous squamous characteristics but the web version of the software did not distinguish these whereas the commercial version of the software does.

"Perhaps no other oncologic disease presents as great a challenge to clinicians as metastatic cancer of unknown primary site. Few diagnoses bring about as much uncertainty, pessimism, and therapeutic nihilism," the team explains. The software-based analysis of metastatic tissues could remedy this situation and extend the common one-year prognosis for this type of cancer and give patients the opportunity to receive far better therapy than is usually available for CUPs.