Venture Capital Goes Genomic

During the 1960s, 70s and into the 90s, a field of investigation arose that examined buyer’s practices when it came to the consumption of goods and services. Algorithms were developed to interrogate consumer choice. One such treatise was reported in 1994 (Carson, RT et al, Experimental Analysis of Choice, Marketing Letters 1994). What these researchers explored were the motivations and forces that drove consumption. When choices are offered, decisions are driven by such factors as complexity and utility. Complexity demands personal expertise or failing that, input from experts, while utility places a value on the good or service.

A recent report from a small biotechnology company called Foundation Medicine has brought this field of endeavor to mind. It seems that this group will be offering DNA sequencing to select chemotherapy drugs. This service, currently priced at $5,800, will focus upon a small cassette of genes that they described as “key” in tumor growth. Based on their technology they have already raised $33.5 million from the likes of Third Rock, Google and Kleiner Perkins Caulfield & Byers, venture capital sources. The CEO of Foundation substantiates the approach by pointing out that fully 150 people have already used their services. One hundred and fifty!

It seems from this report that our colleagues in the field of molecular profiling have studied the dictates of “Experimental Analysis of Choice” to a “T.” What we have is the perfect storm of medical marketing.

First, the technology is so complex as to be beyond the ken of both patients and physicians alike. Thus, expertise is required and that expertise is provided by those engaged in the field. Second, the utility of drug selection is beyond reproach. Who in their right mind wouldn’t want to receive a drug with a higher likelihood of a response when we consider the toxicities and costs, as well as the consequences of the wrong treatment? Dazzled by the prospect of curative outcomes, patients will, no doubt, be lining up around the block.

But, let’s deconstruct what this report is actually telling us. First, a scientifically interesting technology has been brought to the market. Second, it exists to meet an unmet need. So far, so good. What is lacking, however, is evidence. Not necessarily evidence in the rarefied Cochrane sense of idealized survival curves, nor even Level II evidence, but any evidence at all. Like whirling dervishes, patients and their physicians are drawn into a trancelike state, when terms like NextGen sequencing, SNP analysis and splice variants are bandied about.

Despite the enthusiastic reception by investors, I fear a lack of competent due diligence. To wit, a recent article in Biotechniques, “Will the Real Cancer Cell Please Stand Up,” comes to mind. It seems that cancer cells are not individual entities but networks. A harmonic oscillation develops between tumor, stroma, vasculature and cytokines. In this mix, the cancer cell is but one piece of the puzzle.

Indeed, according to recent work from Baylor, some of the tumor promotion signals in the form of small interfering RNAs, may arise not from the cancer cells, but instead from the surrounding stroma. How then, will even the most punctiliously perfect genomic analyses of a cancer cells play out in the real world of human tumor biology and clinical response prediction? Not very well I fear. But then again such a discussion would require data on the predictive validity of the method, something that appears to be sorely lacking.

Will today’s gene profile companies prove to be the biotech Facebook IPOs of tomorrow?

What is Personalized Cancer Therapy?

Personalized therapy is the right treatment, at the right dose for the right patient. Like the weather, however, it seems that everyone’s talking about it, but no one is doing anything about it.

In its simplest form personalized care is treatment that is designed to meet an individual’s unique biological features. Like a key in a lock, the right drug or combination opens the door to a good outcome.

When over the years I lectured on the development of the cisplatin/gemcitabine doublet, my two boys were quite young. I would show a slide depicting a doorknob with a key in the keyhole. I likened our lab’s capacity to identify sensitivity to the cisplatin/gemcitabine combination as “unlocking” an individual’s response.

At the time my wife and I would leave the key in the inside of the front door enabling us to unlock it when going out. We reasoned at the time that our 2-year-old would not be strong enough, nor tall enough to turn the key and let himself outside.  We reasoned wrong, for one day our son Alex reached up, turned the key and opened the door right in front of us. Lesson learned: Given the right key, anyone can open a door.

I continued my analogy by saying that even Arnold Schwarzenegger would be unable to open a door given the wrong key, but might, if he continued trying, snap it off in the lock.

The right key is the right treatment, effortlessly unlocking a good response, while the wrong key is the wrong treatment more often than not too much, too late, akin to a solid tumor bone marrow transplant.

In recent years, personalized care has come to be considered synonymous with genomic profiling. While we applaud breakthroughs in human genomics today, there is no molecular platform that can match patients to treatments.  The objective response rate of just 10 percent, almost all in breast and ovarian cancer patients in one study (Von Hoff J Clin Oncol 2010 Nov 20:28(33): 4877-83), suggests that cancer biology is demonstrably more complex than an enumeration of its constituent DNA base pairs. The unilateral focus on this area of investigation over others might be described as “the triumph of hope over experience” (James Boswell, Life of Samuel Johnson, 1791).

But hope springs eternal and with it the very real possibility of improving our patients outcomes. By accepting, even embracing, the complexity of human tumor biology we are at the crossroads of a new future in cancer medicine.

William Withering (1741-1799) the English physician and botanist credited with discovering digitalis as the therapy for dropsy, e.g. congestive heart failure (An Account of the Foxglove and some of its Medical Uses, Withering W. 1785), had absolutely no idea what a membrane ATPase was, when he made his remarkable discovery. It didn’t matter. Cardiac glycosides provided lifesaving relief to those who suffered from this malady for fully two centuries before Danish scientist, Jens Christian Skou, identified these membrane bound enzymes, for which he was awarded a Nobel Prize in 1997.

Similarly, penicillin, aspirin, and morphine were in all use for decades, centuries, even millenia before their actual modes of action were unraveled. Medical doctors must use any and all resources at their disposal to meet the needs of their patients. They do not need to know “how” something works so much as they (and their patients) need to know “that” it works.

The guiding principle of personalized medicine is to match patients to therapies. Nowhere in this directive is there a prescription of the specific platform to be used. Where genomic signatures provide useful insights for drug selection, as they do in APL (ATRA, Arsenic trioxide); NSCLC (EGFr, ROS1, ALK); CML (Imatinib, Dasatanib) then they should be used.

However, in those disease where we haven’t the luxury of known targets or established pathways, i.e. most human malignancies, then more global assessments of human tumor biology should, indeed must, be used if we are to meet the needs of our patients.  Primary culture analyses like the EVA/PCD® provide a window onto human tumor biology. They are vehicles for therapy improvement and conduits for drug discovery.  Scientists and clinicians alike need to apply any and all available methodologies to advance their art. The dawn of personalized medicine will indeed be bright if we use all the arrows in our quiver to advance clinical therapeutics and basic research.

The Unfulfilled Promise of Genomic Analysis

In the March 8 issue of the New England Journal of Medicine, investigators from London, England, reported disturbing news regarding the predictive validity and clinical applicability of human tumor genomic analysis for the selection of chemotherapeutic agents.

As part of an ongoing clinical trial in patients with metastatic renal cell carcinoma (the E-PREDICT) these investigators had the opportunity to conduct biopsies upon metastatic lesions and then compare their genomic profiles with those of the primary tumors. Their findings are highly instructive, though not terribly unexpected. Using exon-capture they identified numerous mutations, insertions and deletions. Sanger sequencing was used to validate mutations. When they compared biopsy specimens taken from the kidney they found significant heterogeneity from one region to the next.

Similar degrees of heterogeneity were observed when they compared these primary lesions with the metastatic sites of spread. The investigators inferred a branched evolution where tumors evolved into clones, some spreading to distant sites, while others manifested different features within the primary tumor themselves. Interestingly, when primary sites were matched with metastases that arose from that site, there was greater consanguinity between the primary and met than between one primary site and another primary site in the same kidney. Another way of looking at this is that your grandchildren look more like you, than your neighbor.

Tracking additional mutations, these investigators found unexpected changes that involved histone methyltransferase, histone d-methyltransferase and the phosphatase and tensin homolog (PTEN). These findings were perhaps among the most interesting of the entire paper for they support the principal of phenotypic convergence, whereby similar genomic changes arise by Darwinian selection. This, despite the observed phenotypes arising from precursors with different genomic heritages. This fundamental observation suggests that cancers do not arise from genetic mutation, but instead select advantageous mutations for their survival and success.

The accompanying editorial by Dr. Dan Longo makes several points worth noting.  First he states that “DNA is not the whole story.” This should be familiar to those who follow my blogs, as I have said the same on many occasions.  In his discussion, Dr Longo then references Albert Einstein, who said “Things should be made as simple as possible, but not simpler.” Touché.

I appreciate and applaud Dr. Longo’s comments for they echo our sentiments completely. This article is only the most recent example of a growing litany of observations that call into question molecular biologist’s preternatural fixation on genomic analyses. Human biology is not simple and malignantly transformed cells more complex still. Investigators who insist upon using genomic platforms to force disorderly cells into artificially ordered sub-categories, have once again been forced to admit that these oversimplifications fail to provide the needed insights for the advancement of cancer therapeutics. Those laboratories and corporations that offer “high price” genomic analyses for the selection of chemotherapy drugs should read this and related articles carefully as these reports portend a troubling future for their current business model.

If It is Too Good to Be True . . .

The February 12, 2012, CBS 60 Minutes covered a story that has sparked a great deal of interest among cancer patients and medical professionals. The topic was an investigator named Anil Poti who, while working at Duke University developed a laboratory platform for the study of human lung cancer.

Using molecular profiling, Dr. Poti and his collaborators, reported their capacity to distinguish responding and non-responding cancer patients, providing survival curves that were nothing short of astonishing. I recall attending the original lectures given by these investigators at the American Association of Cancer Research meeting several years ago.

As an investigator in the field of drug response prediction, working in lung cancer I had a particular interest in their platform and I was extremely impressed by the outcomes they reported. At the time, I wondered how the static measurement of gene profiles could possibly characterize the nuances of human biology, to encompass the epigenetic, siRNA, pseudogene, non-coding DNA and protein kinetics that ultimately characterize the human phenotype. Nonetheless, with such compelling data I was prepared to be convinced.

That is until a relatively unheralded report in the Cancer Letter raised concerns by several biostatisticians regarding the reproducibility of Dr. Poti’s findings. And then more comments were followed by a full NIH investigation. A panel of biostatisticians was convened and a formal report provided the explanation for Dr. Poti’s excellent results.

They had been invented. The clinical outcomes were not real results. The findings had been retrofitted to match the patient responses and this was the subject of the 60 Minutes report.

What the 60 Minutes report did not address however, was the real problem. That being the inability of contemporary genetic profiling to truly define human biology. For all the reasons enumerated above, siRNA, non-coding DNA, etc., the simple measurement of gene sequences cannot accurately predict biological behavior. This is what the 60 Minutes reporters and the physicians they interviewed, never discussed. The problem at hand is not an errant investigator but an errant scientific community. Our love affair with the gene that began in 1953 (Watson and Crick) has now been confronted by a most heartbreaking example of infidelity (pun intended).

Genes do not make us what we are; they only (sometimes) permit us to become what we are, with the vagaries of transcription and translation lying between.

This leads us to the reasons I find this so critically important:

1. I cannot stress strongly enough that this is NOT what I do. Genomic analysis (their work) and functional analysis (our work) are distinctly different platforms.
2. I strenuously resist any attempt on the part of anyone to tar me or my work with this brush.
3. It is precisely because genomic analysis cannot accurately predict cancer patient outcomes, that these investigators found it necessary to invent their data.
4. Despite this, functional analyses can and do provide these types of predictive results in lung cancers and other diseases as we have reported in numerous publications.
5. Finally, while imitation is the sincerest form of flattery, this is one instance in which I would prefer to decline the compliment.

The False Economy of Genomic Analyses

We are witness to a revolution in cancer therapeutics. Targeted therapies, named for their capacity to target specific tumor related features, are being developed and marketed at a rapid pace. Yet with an objective response rate of 10 percent (Von Hoff et al JCO, Nov 2011) reported for a gene array/IHC platform that attempted to select drugs for individual patients we have a long way to go before these tests will have meaningful clinical applications.

So, let’s examine the more established, accurate and validated methodologies currently in use for patients with advanced non-small cell lung cancer. I speak of patients with EGFR mutations for which erlotinib (Tarceva®) is an approved therapy and those with ALK gene rearrangements for which the drug crizotinib (Xalkori®) has recently been approved.

The incidence of ALK gene rearrangement within patients with non-small cell lung cancer is in the range of 2–4 percent, while EGFR mutations are found in approximately 15 percent. These are largely mutually exclusive events. So, let’s do a “back of the napkin” analysis and cost out these tests in a real life scenario.

One hundred patients are diagnosed with non-small cell lung cancer.
•    Their physicians order ALK gene rearrangement     $1,500
•    And EGFR mutation analysis     $1,900
•    The costs associated $1,500 + $1,900 x 100 people =    $340,000
Remember, that only 4 percent will be positive for ALK and 15 percent positive for EGFR. And that about 80 percent of the ALK positive patients respond to crizotinib and about 70 percent of the EGFR positive patients respond to erlotinib.

So, let’s do the math.

We get three crizotinib responses and 11 erlotinib responses: 3 + 11 = 14 responders.
Resulting in a cost per correctly identified patient =     $24,285

Now, let’s compare this with an ex-vivo analysis of programmed cell death.

Remember, the Rational Therapeutics panel of 16+ drugs and combinations tests both cytotoxic drugs and targeted therapies. In our soon to be published lung cancer study, the overall response rate was 65 percent. So what does the EVA/PCD approach cost?

Again one hundred patients are diagnosed with non-small cell lung cancer.
•    Their physicians order an EVA-PCD analysis    $4,000
•    The costs associated: $4,000 x 100 people =    $400,000
•    With 65 percent of patients responding, this
constitutes a cost per correctly identified patient =     $6,154

Thus, we are one quarter the cost and capable of testing eight times as many options. More to the point, this analysis, however crude, reflects only the costs of selecting drugs and not the costs of administering drugs. While, each of those patients selected for therapy using the molecular profiles will receive an extraordinarily expensive drug, many of the patients who enjoy prolonged benefit using EVA/PCD receive comparatively inexpensive chemotherapeutics.

Furthermore, those patients who test negative for ALK and EGFR are left to the same guesswork that, to date has provided responses in the range of 30 percent and survivals in the range of 12 months.

While the logic of this argument seems to have escaped many, it is interesting to note how quickly organizations like ASCO have embraced the expensive and comparatively inefficient tests. Yet ASCO has continued to argue against our more cost-effective and broad-based techniques.

No wonder we call our group Rational Therapeutics.

Lots of Heat No Light – ASCO Technology Assessment Update 2011

“Once more unto the breach, dear friends.”

This famous line from Shakespeare’s Henry V, describes the Battle of Agincourt and England’s unexpected victory over the French. Not unlike Henry V a small coterie of relatively underfunded and embattled investigators around the world continue to fight an entrenched medical community who refuse to relinquish their grip on the clinical trial process.

Their re-review updated from 2004, sheds no new light on the field, as the authors conclude that their 2004 recommendations stand without modification.

The authors, to their credit, have updated their database to include cell death endpoints. They cite the ovarian cancer study by Dr. Ian Cree, that assigned 180 patients, (of which 147 were evaluable), with recurrent disease, and reported a response rate of 40.5 percent for assay directed versus 31.3 percent for physician choice, yet failed to achieve significance. The reasons for this trial’s failure however were obvious, as it was underpowered and more importantly allowed the physician’s choice arm to include Dr. Cree’s own drug combinations as the trial accrued. This left Dr. Cree in the uncomfortable position of having to compete with himself.

More disturbing is their dismissal of a paper by Selma Ugurel, MD, from Clinical Cancer Research 2006 in which, patients with metastatic melanoma received assay-directed treatment for this otherwise chemo resistant and lethal disease. Patients found drug sensitive in the laboratory had a response rate of 36.4 percent, while those found drug resistant had a response rate of only 16.1 percent (a two-fold improvement). The overall survivals were similarly improved with assay-directed patients 14.6 months vs. drug resistant patients of 7.4 months. Again a doubling. Furthermore these results achieved statistical significance.

The ASCO group concludes with the comment, “However, the investigator did not compare the two interventions.” As I know this paper well, and was extremely impressed that some of the responders went out to 30 months, I find the ASCO group’s insouciance surprising.

This reminds me of an old joke by the comedian Jerry Seinfeld. It seems that he had watched a television program where a man caught bullets shot from a gun with his bare teeth. Seinfeld went on to say, that despite being immensely impressed by this man’s prowess, he just couldn’t seem remember his name. “What do you got to do to impress people”?

As I am familiar with the Ugurel paper, I have been very impressed with these investigators completing a study by dint of their dedication to the field. Stranded without funding or cooperative group support, laboratory-based therapeutics remains unconfirmed, not by the unwillingness of the investigators but by the unwillingness of the cooperative and funding agencies to test the hypotheses.

While we squander billions of dollars on genomic analyses that are increasingly leading us nowhere, these ASCO study groups and their colleagues continue to refuse to formally evaluate human tissue studies. In light of the lack of improvement in survival for most cancers over the past 50 years, despite the expenditure of hundreds of billions of dollars on research, perhaps assay-directed therapy is just the solution that medical oncology needs.

Using Function to Inform Genomics

Recently, I was asked to evaluate a charming young woman with an unusual gynecologic primary. She had received numerous forms of therapy and surgery for her low-grade carcinoma. Her most recent surgery provided tissue to our laboratory for analysis. The results were consistent with the clinical presentation, revealing relative resistance to virtually all conventional chemotherapeutic drugs, but a very striking pattern of sensitivity to three compounds associated with the insulin like growth factor signaling pathway.

As all three compounds pointed to activity in this pathway, I reasoned that the patient had a mutation upstream, as I wrote in my report to her physician. I suggested that they should investigate this pathway.

I was subsequently apprised that, upon my recommendation and at the request of patient’s husband, an analysis had been submitted to a laboratory that identified a mutation in PI3K, the very pathway that I had identified in our functional analysis. Thus, this patient’s resistance to chemotherapeutics and sensitivity to PI3K inhibitors reflected the profound survival signal provided by this mutation. Of interest, when the family originally requested a molecular profile be conducted in parallel with our functional profile, the commercial lab in Arizona did not include PI3K mutational studies. It wasn’t until the functional results pointed to the PI3K pathway that the specific mutational analysis was undertaken and found positive.

This experience is very similar to our original work with the EGFR inhibitors gefitinib and erlotinib. Several years before the EGFR mutation was identified and long before the mutational analysis was commercially available we identified activity in patients using the functional platform. Patients were then treated based upon the EVA-PCD results under protocol with every one of these patients responding, as we reported (Nagourney Proc ASCO, 2007). This speaks to the robustness of the functional platform and to its capacity to guide drug development.

Has the Era of Genomics as we Know it Come and Gone?

I have often described my personal misgivings surrounding the application of gene profiles for the prediction of response to therapeutics. My initial concerns regarded the oversimplification of biological processes and the attempt of analyte-driven investigators to ascribe linear pathways to non-linear events.

The complexities of human tumor biology, however vast, took a turn toward the incomprehensible with the publication of a lead article in Nature by the group from Harvard under Dr. Pier Paulo Pandolfi. (Poliseno, L., et al. 2010. A coding-independent function of gene and pseudogene mRNAs regulates tumor biology. Nature. 2010 Jun 24; 465(7301):1016-7.) I sat in as Dr. Pandolfi reviewed his work during the Pezcoler Award lecture, held Monday, April 4, 2011, in Orlando at the AACR meeting.

What Dr. Pandolfi’s group found was that gene regulation is under the control of messenger RNA (mRNA) that are made both by coding regions and non-coding regions of the DNA. By competing for small interfering RNAs (siRNA) the gene and pseudogene mRNAs regulate one another. That is to say that RNA speaks to RNA and determines what genes will be expressed.

To put this in context, Dr. Pandolfi’s findings suggest that the 2 percent of the human genome that codes for known proteins (that is, the part that everyone currently studies) represents only 1/20 of the whole story. Indeed, one of the most important cancer related genes (known as PTEN, is under the regulation of 250 separate, unrelated genes. Thus, PTEN, KRAS and, for all we know, all genes, are under the direct regulation and control of genetic elements that no one has ever studied!

This observation represents one more nail in the coffin of those unidimensional thinkers who have attempted to draw straight lines from genes to functions. This further suggests that attempts on the part of gene profilers to characterize patients likelihoods of response based on gene mutations are not only misguided but, may actually be dishonest.

The need for phenotype analyses like the EVA-PCD performed at Rational Therapeutics has never been greater. As the systems biologists point out complexity is the hallmark of biological existence. Attempts to oversimplify phenomena that cannot be simplified, have, and will continue to, lead us in the wrong direction.

What’s the Best Treatment for Metastatic Colorectal Cancer?

The answer is: nobody knows.

We have previously described a patient with a small bowel cancer for whom a treatment regimen contrary to the most widely used triplet was recommended. While it is arguable that small bowel adenocarcinoma is rare enough that no one really has a favorite regimen, colorectal management has become somewhat rigidly focused on FOLFOX. Yet, this popular combination may not be right for every patient with colon cancer.

We know, for example, that FOLFOX combined with Avastin provided no advantage in the adjuvant setting. We also know that the random addition of Erbitux to FOLFOX similarly failed to provide an advantage. As the modes of action differ between drugs, it is not surprising that subsets of colon cancer patients may do better with Irinotecan based therapies. Indeed, clinical trials combining the new monoclonal antibodies with Irinotecan have proven quite favorable, including the 2007 BOND-2 trial reported by investigators at Memorial Sloan Kettering in New York.

With this in mind, patients who present with both untreated colon cancer and a favorable profile for Irinotecan based combinations always interest us. One such patient presented to our attention in the last few weeks. This patient, in his mid 30s, was found to have inoperable, widely metastatic disease with extensive liver involvement. Confirmatory biopsies provided tissue for analysis and revealed no evidence of mismatch repair.

The results of the EVA-PCD platform were interesting on many levels. First, the EGFr active drugs provided a uniquely favorable profile, as did the down-stream inhibition of the MEK-ERK inhibitor we studied. These findings strongly suggested that the patient was RAS wild type (i.e. non-mutated). It is known that RAS mutation confers resistance to the EGFr active drugs. By inference, his sensitivity to the EGFr active drugs was prima facie evidence of RAS wild type, a finding that was confirmed later by molecular analysis. There was also a favorable profile for VEGF active drugs. Most favorable of all was the combination of Irinotecan with inhibitors of both VEGF and EGFr. This was the regimen that we selected.

We wait with interest the results of the therapy, as re-staging for response will be conducted in the coming months.

When Fluff Isn’t Enough

Recent press coverage from the San Antonio Breast Cancer Symposium (SABCS) touched upon the development of multi-gene predictors for clinical response in breast cancer. One report from that meeting described correlations between a laboratory assay model in use at the University of Pittsburgh and microarray analyses. However, the suggestion that this laboratory technique — described by its proponents as a chemosensitivity assay — could accurately identify gene profiles that would predict response seems at odds with the current literature.

Although the press coverage concluded that this technique showed “promising performance” it was largely exploratory and defined by the authors as a “validation study.” What is interesting is that a team of highly reputable investigators from M.D. Anderson recently reported a very negative study using a similar approach of identifying target genes in cell lines and then correlating them with patient outcomes. In the paper, published in the June 2010 issue of Breast Cancer Research and Treatment (Liedtke, C. et al. Breast Cancer Res Treat. 2010 Jun; 121(2):301-9) the authors reported “cell line derived predictors of response to four commonly used chemotherapy drugs did not predict response accurately in patients.”

Indeed, differential gene expression seemed only to correlate with paclitaxel. The authors found that false discovery rates were high for all other drugs tested. Thus, the report from the SABCS will need to be carefully examined to determine whether truly relevant clinically predictive information can be provided by this particular laboratory platform.