Tumor Heterogeneity and Personalized Medicine

Eric Schuur

My recent blog post, Tumor heterogeneity, revealed…, discussed the New England Journal of Medicine article by Gerlinger and colleagues describing the genetic heterogeneity found both within a patient’s individual tumor nodules and between spatially separate nodules. There has been a substantial amount of discussion of this work and angst about how it might signal the end of personalized medicine even before it really got started. I don’t believe that will be the case at all. To the contrary, this paper made interesting contributions in three conceptual areas that may help pull the field forward. These areas are the 1) relevance of prognostic gene expression profiles, 2) the nature of “driver” genetic mutations, and 3) the pathogenesis of cancer itself. All of these areas are, in my opinion, very important to make headway in before personalized cancer medicine can become a truly effective tool in medicine.

Heterogeneity in gene expression profiles across the tumor specimen

The result that most seized on to proclaim the demise of personalized medicine was the finding that gene expression signature from spatially separated parts of a tumor nodule yielded different assessments of prognosis. The implication is that a single biopsy specimen is inadequate to generate an accurate prediction of clinical course or response to treatment. Most likely that is at least partially true. However, the issue is with sampling, rather than the molecular biology. We have known for decades that tumors have variable histology within their mass, with some regions indicating poorer prognosis than others via their histologic grade. Rather than reflecting a conceptual disconnect that dooms a new paradigm, it looks more like a technical problem to solve, which should be no surprise along this new path.

Convergent evolution

Both the Gerlinger paper, as well as others (e.g. Walter et al, NEJM), using NGS have now demonstrated that within a single patient the same gene can be found to be mutated multiple independent times, suggesting that this mutation creates a change in gene function that participates in the development of the cancer. This had not been shown in humans before. This finding will be useful for clinical diagnostics and it may be game changing in basic research. In clinical diagnostics identification of a multiply-mutated gene would give additional confidence that the damage it represents is causal and may help select targeted therapy. In basic research, identification of such genes would represent novel evidence of the causality of specific genetic changes in the disease process. This type of evidence is a smoking gun, a sign post saying “Needs to be mutated to reach this disease state”. This type of evidence, which only deep sequencing can yield, is a new and useful application of NGS that was not previously available.


The picture that the Gerlinger paper, Walters paper, and others paints is one of clonal evolution of cancer. This type of work paints this picture with clarity that has not been achievable before. What is striking to me is that these results make it harder to ignore the concept that these genetic alterations, as important as they clearly are in the progression of cancer, may not be the cause of cancer. They beg the question, “what initiated this evolutionary process?”. Certainly, oncogenes, tumor suppressors, and the like are a part of cancer pathogenesis, carrying the developing disease along. But it seems to me that there is still a “first cause” of some sort that we have not put our collective fingers on. Genomic instability is certainly key, but then what is the genesis of the genomic instability? What are the inputs that kick this process off? Efforts to answer these questions will move us closer to effective treatments for cancer and other diseases that may share these pathogenic processes.