NEW YORK (GenomeWeb News) – In a paper scheduled to appear online this week in the Proceedings of the National Academy of Sciences, a team of Ontario researchers used a computer algorithm to come up with a new six-gene expression signature for predicting survival in non-small cell lung cancer that they subsequently validated in four datasets.
While the researchers expected to find just one or two gene signatures, their algorithm suggested that there are thousands — or even hundreds of thousands — of distinct gene expression signatures that could provide prognostic information regarding NSCLC.
While the potential clinical application of the new six-gene signature is the most important immediate outcome of the work, lead author Paul Boutros told GenomeWeb Daily News, the multitude of other gene signatures detected raises many new questions about cancer biology and biomarkers in general. Boutros, a graduate student in senior author Igor Jurisica's lab at the University of Toronto at the time of the study, is now an institute fellow at the Ontario Institute for Cancer Research.
Non-small-cell lung cancer accounts for the vast majority of lung cancer cases. Although surgical resection of NSCLC tumors can cure some individuals, as many as a third to half of NSCLC patients relapse and die within five years of surgery.
Genetic signatures based on gene expression profiles are one potential means for identifying individuals who might need more aggressive cancer therapy. Still, despite the surge in research aimed at identifying gene expression signatures associated with lung cancer outcomes, relatively few genes in the prognostic signatures identified so far overlap with one another.
Suspecting that different statistical methodologies could be to blame for the discrepancy, Jurisica and his colleagues developed a modified Steepest Descent or mSD algorithm to assess published quantitative RT-PCR profiles for NSCLC.
Using this non-linear, semi-supervised approach, they came up with a six-gene prognostic signature — including STX1A, HIF1A, CCT3, HLA-DPB1, MAFK, and RNF5 — that could accurately predict survival outcomes in four public microarray studies of lung cancer.
When the signature was applied to data from eight different studies representing 589 NSCLC patients, the researchers found that they could accurately stratify individuals with Stage I cancer into two groups with different survival outcomes.
Even so, the signature was not the only one they detected. They found another 1,789 distinct signatures. And based on their calculations, the researchers estimated that there are probably at least half a million different six-gene prognostic signatures for this type of lung cancer.
Although the researchers say the main six-gene signature described in the paper holds the greatest potential for clinical significance in the near future, the numerous additional gene signatures they detected may provide new insights into the biology of everything from lung cancer to biomarkers in general.
By comparing the genes present in the various gene expression signatures, it may be possible to learn more about the biology of NSCLC, for instance. In this study, the team found three genes in the mSD signature that were highly enriched in the other signatures. And by examining the top ten genes that were enriched in the various gene signatures, the researchers found that signatures containing the calcitonin-related polypeptide alpha neuropeptide were statistically significant more than 40 percent of the time.
The researchers predict that multiple gene signatures may be present in other types of cancer as well — providing a possible explanation for the frequent lack of overlap between gene expression signatures identified by different groups. But, they noted, more research is needed to determine whether that's the case.
And while new findings should not diminish the applicability of previously identified and validated gene signatures, Jurisica told GenomeWeb Daily News, they do raise the possibility that there are additional, undiscovered gene signatures that actually perform better than those described so far.
The researchers are continuing to explore such questions. For his part, Boutros is currently investigating samples from various types of cancer as well as non-cancer samples at the OICR to see if the preponderance of potential prognostic signatures is unique to lung cancer or whether this pattern turns up in other cancers or conditions.