Today we cannot predict the behavior of individal patient tumors and need to overtreat some patients and undertreat others.  However, the science behind lung cancer has now moved a step forward by identifying a ”molecular signature” of key genes that may predict patient survival.  This week’s New England Journal of Medicine (NEJM) included a very provocative manuscript by a group in Taiwan that studied tumor tissue from a series of patients who had surgery for NSCLC and identified five key genes that could help them separate those with a good prognosis from those with a poor prognosis.  Similar to a prior NEJM article from a research group at Duke that identified a large collection of genes that distinguished better prognosis and worse prognosis patients with early stage NSCLC, the article this week by Chen and colleagues out of Taiwan was designed to do a better job than just relying on our current staging system to predict clinical outcomes for patients with lung cancer. 

   Their technique was complex, and these results need to be reproduced widely before this approach becomes established and incorporated into clinical decision-making.  But there were certainly robust differences between the two groups they identified by molecular signatures.  The median survival was twice as long in the low-risk group as in the high risk group (40 months vs. 20 months), and there was a more than doubling in the median relapse-free survival (29 vs. 13 months for high- vs. low-risk, respectively). 

   I’ll review some of how they came to this point, but even trying to simplify this, it’s pretty scientifically complex.  You may want to take some Tylenol in advance.

   The Taiwan group started with 185 frozen tissue specimens that had several hundred genes tested all at once with a process called gene microarray analysis.  This is a complicated scientific technique that is expensive, involves analyzing huge amounts of data from the gene profiles, is not at all standardized, and generally not validated (reproduced and confirmed) beyond the center that initiated the research.  Because of this, the gene profiling studies that have been done thus far have been promising but not broadly useful yet.  A different technique that can focus on a smaller number of genes would be of much greater utility. 

  After starting from 672 genes in the microarray analysis, they correlated survival from 125 patients who had undergone surgery for NSCLC with gene  expression levels for these, then used a complex computer analysis to identify 16 genes that were particularly discriminating for better or worse survival.  For these 16 genes, the investigators developed “risk scores”, which predict the likelihood of an individual patient’s survival on the basis of the level of expression of these “good risk” or “bad risk” genes.  After totaling the risk scores for these genes, they separated the patient groups into a favorable “gene signature” (combination of these genes overall) or unfavorable, with the cut-off at the halfway point for total combined risk of the gene collection.

  But wait, there’s more.  They confirmed the level of expression of these 16 genes by a process called reverse transcriptase polymerase chain reaction, or RT-PCR, which is more widely done in various labs than gene microarray work.  From those 16 genes, the investigators whittled the list down to just 5 genes (known as DUSP6, MMD, STAT1, ERBB3, and LCK, but what they are is not very important to this story) that were most important in predicting survival.   And this 5-gene signature, which can be determined by RT-PCR, turned out to be a robust predictor of relapse-free and overall survival.    An example of the difference is in this figure, stolen appropriated from the NEJM website (the public domain portion, actually), which shows the overall survival difference over time for stage I patients with a high risk vs. low risk 5-gene signature.

  (sorry, this one is just a small version, but you can see the difference)

   There are very similar splayed curves for other stages, both in overall survival and recurrence-free survival.  Importantly, the 5 key genes did well not only in this training set, but from another set of tumor tissue from 60 more patients at the same hospital, as well as a set of published microarray data from 86 other patients with NSCLC from the US.  So this wasn’t specific just to the population in Taiwan.  This is an important consideration, as we have come to recognize that there may be differences in the biology of different lung cancers and how they respond to treatment depending on the racial group and underlying genetic differences between them.

  In the accompanying editorial, authors Roy Herbst and Scott Lippman from MD Anderson suggest that the article by Chen represents the maturation of a first phase of lung cancer genomics, the study of the human genome, or the full collection of genes in the DNA.  This phase includes looking back at tissue already collected and how patients did, and trying to connect those factors with the benefit of hindsight.  So what is the present and future of this work?  The next phase involves changing treatment plans for patients based on their identified level of risk based on the molecular signature.  The fields of lymphoma and breast cancer are a bit ahead of lung cancer in this respect, and some trials with such a design have started for these cancers.  We’re designing trials in lung cancer to follow this approach over the next few years.

   In addition, we hope to use molecular signatures soon to help us identify which patients are more or less likely to respond to different treatments, whether chemo or targeted therapies.  We’ve already made some small steps in this regard.  For instance, expression of the DNA repair gene excision repair cross-complementation group 1 gene (ERCC1) is associated with resistance to cisplatin-based adjuvant therapy, as reported in an important retrospective study (also published in NEJM), while EGFR tyrosine kinase mutations have been associated with a high likelihood of response to EGFR inhibitors that is often of prolonged duration (detailed in another post).  

   These approaches are not yet generally available, and they’re not yet standard practice.  But I think we can all see the increasing momentum as more and more of these studies that help us discriminate which patients are most likely to recur and which are more or less likely to respond to a particular treatment strategy.  Since these all have potentially very significant side effects, sparing some patients from unnecessary or unhelpful treatments while directing them to the most beneficial approaches gets us closer to a holy grail of individualized plans rather than a “one size fits all”, sledgehammer approach that overtreats some and undertreats others.