Guest Commentary: Back to cancerís drawing board
Back to cancer’s drawing board
By Murali Prahalad of Epic Sciences
A curious problem arises for each new precision oncology therapeutic with a companion diagnostic: Many predicted responders exhibit primary resistance.
Frequently, such events beget studies that bring to bear more sensitive genomics techniques and deeper sequencing to uncover the confounding variants. After each new discovery, however, unexpected primary resistance remains curiously pervasive.
While biology is rarely clear-cut, a recurrent problem may cause one to reexamine a field’s foundational assumptions. Decades after the precision medicine revolution began, it is worth assessing whether the evidence still supports genomics as the primary pillar upon which to understand cancer and make treatment decisions. Perhaps, as Claude Bernard has been attributed as saying, “It is what we think we know that keeps us from learning.”
Let’s critically examine three tenets that have propelled the rise of precision oncologics and cancer diagnostics.
1: Can genomic testing identify driver mutations?
Two decades after the launch of the first commercial BRCA test for breast cancer, more than 95 percent of diagnosed women under 40 are now tested for BRCA mutations. Across many cancer types, tumor boards now often review a mutation list from tumor sequencing results, alongside other clinical data, to determine a treatment plan. Oncologists consider these mutation lists with care; studies in the past decade have detailed now commonly discussed limitations of this practice—as well as less frequently acknowledged issues that may better explain confounding results.
Perhaps the most commonly acknowledged limitation is spatial resolution. For stage III and IV patients, who comprise approximately half of newly diagnosed cancer patients, many recent metastases are yet too small to detect by radiology (perhaps having only 100,000 cells). Lesions may also be too numerous to sample practically or humanely. Spatial resolution is further limited because fine needle aspirates, which remove just thousands of cells, may miss adjacent areas of a billion-cell tumor where more aggressive and clinically relevant subclones reside.
The other commonly acknowledged limitation is temporal resolution; tumors are known to evolve rapidly in response to therapeutic pressure. When clinicians return to data from primary tumor biopsies after a patient has undergone multiple rounds of therapies, those test results likely no longer reflect all current disease drivers. Subsequent resampling of tumors may be impractical, painful and introduce delays.
While oncologists and researchers acknowledge that genetic tests produce a spatially and temporally incomplete view of disease, two other issues are less overtly acknowledged, but can introduce more severe complications.
The most fundamental limitation of genetic testing is its protocol. By pooling genetic material from thousands of lysed cancer cells, these tests erase how mutations cluster in individual cells. By contrast, cancer plays out at the level of a tumor cell, with each cell’s metabolic characteristics, metastatic potential and drug susceptibility influenced by its own stack of genetic and epigenetic abnormalities. A driver gene does not kill a patient. Rather, aggressive tumor cells do. Averaging cells’ genetic signals together thus expunges the context that may explain perplexing responses to perceived vulnerabilities, such as when a proposed driver mutation is actually present only in non-metastatic cells. Charles Swanton’s team at Cancer Research UK has described in numerous studies the prevalence of subclonal populations with unique combinations of genetic mutations across a wide number of cancers, noting in one study last year that cellular heterogeneity “complicates” genetic alterations’ typical application “as biomarkers for treatment personalization.”
Furthermore, the material these tests evaluate—tissue biopsies—reflect static tumors, not the cells that have entered the lymph or blood and may soon deposit in distal sites to seed new tumors. The key disease drivers are the circulating tumor cells (CTCs) present outside of known tumors, particularly for the 90 percent of patients who die in the metastatic stage of disease and the large numbers of patients who are diagnosed at stage III or IV. Consider a case in which tissue biopsy results indicate a potential response to a single targeted agent, but the majority of CTCs detected in the patient’s blood sample represent numerous genetically heterogeneous subclonal populations. Would a broad-spectrum chemotherapy or combination therapy better control disease progression?
Howard Scher at Memorial Sloan Kettering Cancer Center confirmed such a hypothesis in a study of 221 metastatic castrate-resistant prostate cancer (mCRPC) patients, which he presented at the ASCO Genitourinary Cancers Symposium this past January. By individually cataloging each of the 9,225 observed CTCs based on a combination of proteomic, genomic and morphologic markers, he constructed a heterogeneity index that represented the extent of clonal diversity among the subpopulations of CTCs in every patient’s blood.
Some patients’ liquid biopsies harbored as many as 15 distinct types of CTCs. Scher found that high heterogeneity predicted shorter radiographic progression-free survival and overall survival to targeted therapy (enzalutamide and abiraterone). However, high heterogeneity was not associated with resistance to taxane-based chemotherapy, suggesting an abstract measure of heterogeneity can identify patients likely to fail targeted therapies but benefit from chemotherapy. The signature was strong enough to demonstrate a therapy interaction and potential utilization as a predictive biomarker in the management of mCRPC patients.
Scher also found that many CTC subtypes, even when comprising just a fraction of the total observed CTC population, predict shorter overall survival and drug resistance. Indeed, patients whose blood contained even a few cells of one particular, newly identified CTC subtype would universally fail standard-of-care therapies and experience much shorter overall survival. That cell type’s unique genomic signature, however, would be likely undetectable due to the limited sensitivity of even the most sensitive tissue biopsy sequencing techniques.
These data suggest that—to borrow an ice hockey analogy—to control the game, we must skate to where the puck will be (CTCs) and not where the puck is at the moment (a tumor). Cancer is more complex, however, as we often must watch multiple pucks’ trajectories simultaneously.
2: Do genetic abnormalities reflect cellular function?
A one-to-one correlation between genetic mutations and protein expression is acknowledged to be an oversimplification due to the influence of epigenetics and other factors.
It’s interesting, however, to consider the frequency at which mutational status and cell function may diverge in tumor cells that survived successive rounds of therapy. Organizations conducting pan-omics studies, such as Patrick Soon-Shiong’s NantHealth, have reported that many observed somatic mutations in actionable genes have displayed little to no corresponding expression at the transcriptomic level. Similarly, through CTC-based liquid biopsies, we have found that late-stage prostate cancer patients who are resistant to anti-androgen therapeutics harbor CTCs with androgen receptor amplification, but lack elevated androgen receptor protein expression, indicating the cells have evolved from their earlier resistance mechanisms to promote androgen-independent proliferation.
3: Does overall survival measure the efficacy of precision oncologics?
When viewing cancer as a cellular disease driven by heterogeneous and independently evolving subclonal populations, does a primary endpoint of overall survival accurately reflect a drug’s utility for each patient in a large Phase 3 study?
A series of recent single-cell studies support a conclusion that, across many cancer types, tumor heterogeneity extends beyond current assumptions and independent clonal evolution is a critical driver of relapse and drug resistance. By example, in a Seattle Fred Hutchinson Cancer Research Center study, Jerald P. Radich found that mutations in two clinically challenging and recurrence-associated genes, FLT3 and NPM1, occurred in myriad states across a minimum of nine distinct clonal populations in each patient—and with both convergent and differential evolutionary trajectories.
These studies indicate that the same factors that confound tissue biopsy-based genomic tests also undermine overall survival as an imprecise measure of a targeted therapy’s utility in a heterogeneous disease. For instance, a patient’s overall survival may not be extended significantly by a drug that indeed eliminates a highly aggressive subpopulation of cells as intended, but cannot combat the patient’s other aggressive populations of CTCs that elude detection. Useful therapies may thus be inadvertently lost in the current clinical paradigm.
While overall survival will not be an endpoint that can be eliminated, a more modern and nuanced view of cancer therapeutics as combination agents against selected components of disease necessitates an evolution in trial design.
Drawing a new model for cancer
Upon reexamination of the key tenets behind current cancer diagnostics and clinical development, we must humbly go back to cancer’s drawing board.
In some situations, answers to simple questions, such as the presence of a specific mutation in an early-stage patient, may well be addressed sufficiently by sensitive tissue biopsy or circulating tumor DNA-based approaches. But in more complex disease, we must begin to desegregate the full heterogeneity of a patient’s disease-driving CTCs and their genomic and proteomic status. A cellular view is necessary for the resolution to rationalize emerging combination therapy strategies and to understand the confounding outcomes that affect many patients receiving targeted therapies today.
Finally, CTC-based approaches must also evolve to not mirror tissue biopsies’ sampling limitations. That is, we need methods that do not restrict analyses to CTCs of a certain size or that express certain proteins. For a heterogeneous disease we are just beginning to untangle, we cannot assume to know which subtypes of cells are irrelevant to sample.
Murali Prahalad has been the president and CEO of San Diego-based Epic Sciences since 2013. He has more than 14 years of experience in technology and life-sciences companies. Prior to joining Epic, he was vice president of corporate strategy at Life Technologies, now part of Thermo Fisher Scientific, where he helped shape the organic and inorganic investment priorities for research tools, clinical diagnostics and applied market portfolios. After joining Life Technologies in 2005, his experience ranged from mergers and acquisitions to in-licensing and general management, including large businesses that spanned everything from genomics to cell biology.