I recently had an interesting conversation with a physician regarding her patient with an aggressive breast cancer.
A portion of tumor had been submitted to our laboratory for analysis and we identified activity for the alkylating agents and the Taxanes, but not for Doxorubicin. After our report was submitted to the treating physician she contacted me to discuss our findings, as well as the results from a genomic/proteomic laboratory that conducted a parallel analysis upon a portion of the patient’s tumor. The physician was kind enough to forward me their report. Their results recommended doxorubicin while ours did not. The treating physician asked for my input. Here, I thought, was a “teachable moment.”
Our discussion turned to the profound difference between analyte-based laboratory tests e.g. genomic and proteomic, and functional platforms like our own (EVA-PCD). Genomic, transcriptomic and proteomic platforms measure the presence or absence of genes, RNA or protein. Gene amplification, deletions or mutations and protein and phosphoprotein expressions are examined. These platforms dichotomize patients into those who do and those who do not express the given analyte, with cutoffs for gene copy number or intensity of staining.
These platforms have worked very well in diseases where there is a linear connection between the gene (or protein) and the disease state, e.g. BCR-ABL in CML for which imatinib has proven so effective. These tests have worked reasonably well in EGFR mutated and ALK gene rearranged in lung cancer, but even here response rates and response durations have been less dramatic. However, they have not worked very well at all for the vast majority of cancers that do not carry specific and well-characterized targets. These cancers reflect polygenic phenomena and are not defined by a single gene or protein expression.
Functional platforms look at cellular response to injury at the systems level and measure the end result of drug exposure to gauge the likelihood of a clinical response. Our focus on cell biology allows us to determine whether a drug or combination induces programmed cell death. After all, regardless of what gene elements are operational, it is the ultimate eradication of the cancer clone (its loss of viability) that results in clinical response.
As we reported in a recent paper in non-small cell lung cancer, patients who revealed the most sensitive ex-vivo profile to erlotinib (Tarceva) lived far longer than the general clinical experience for those patients who were selected for erlotinib by EGFr mutation analysis alone. Some of these patients are alive at 5, even 9 years since diagnosis.
We live in a technocracy where process has taken precedence over results. We are enamored with complex scientific technologies sometimes at the expense of simple answers. A metallurgist, familiar with every last detail of the alloys used in a Boeing 747 wouldn’t necessarily be your first choice for pilot. A skilled pathologist, intimately familiar with the most detailed intricacies of human diagnostics would not likely be your preferred surgeon for cardiac bypass.
Cancer diagnosis and cancer treatment are two distinctly different disciplines. While we use the ER (estrogen receptor) status in breast cancer to select treatment, few oncologists would select Tamoxifen for their NSCLC patients even though many NSCLC patients express ER in their tissue. ER + NSCLC does not respond to tamoxifen and V600E BRAF mutated (+) colon cancer patients do not respond to vemurafenib, the very drug that works so well in BRAF V600E (+) melanoma.
Cancer is contextual and responses are not solely predicated upon the presence or absence of a gene element alone. We must use a broader brush when we paint the likeness of our patients in the laboratory, one that encompasses the vicissitudes of human biology in all of its complexities.
Where I took issue with the report, however, was its “evidence-based” moniker. The evidentiary manuscripts cited to support the drug recommendations, with titles like “Overexpression of COX-2 in celecoxib-resistant breast cancer cell lines” provided little evidence that a (+) COX-2 finding by IHC on this patient’s biopsy specimen would offer any real hope of response. It seemed that with all of the really interesting science going on here, no one had taken the time to do the hard work to figure out whether any of these observations had a basis in reality. The failure of ERCC1 expression in lung cancer to correlate with response and survival or the Duke University debacle with gene profiling in NSCLC are just the most recent examples of how “lovely theories can be spoiled by a little fact.”
As we and our colleagues in cell profiling have actually taken the time to correlate predictions with clinical outcomes we have shown a 2.04 fold higher objective response rate (p 0.001) and significantly improved 1-year survival (p=0.02). (Apfel, C. et al Proc ASCO, 2013). To the contrary, it is of interest to examine the comparatively scant published literature on genomic and IHC profiling for drug selection under similar circumstances. While one group reported an underwhelming objective response rate of 10 percent in their study, (Von Hoff, J Clin Oncol 2010) a more recent study is even more illuminating. A Spanish group used genomic profiling in 254 colon cancer patients to select candidates for gene-targeted agents (KRAS/BRAF/PI3K/PTEN/MET) and provided therapy for 82. They reported a significantly shorter time to progression for targeted treatments compared with conventional therapies 7.9 vs 16.3 week (P<0.001) and an overall objective response rate of 1.2 percent, yes that’s 1.2% (1/82).
Human tumor biology is many things, but simple is not one of them. Reductionist thinking is not providing the insights that our patients desperately need. While we await the arrival of a perfect test for the prediction of response to cancer therapy, perhaps we as physicians and our patients should use a good one, one that works.