In March 2010, Austin, Texas-based diagnostics firm Vermillion launched sales of its OVA1 ovarian cancer test. A panel of five proteins intended to guide treatment decisions in women with ovarian masses, the test was the culmination of — to summarize briefly — 10-plus years of research, two corporate name changes, and a Chapter 11 bankruptcy proceeding. As such, its release was greeted by Vermillion investors and observers as something of a watershed event.
The company's stock, which only months before had traded for as little as $0.04 per share, spiked to nearly $34 per share in the week before the test's launch. In a US Securities and Exchange Commission filing, Vermillion management predicted that OVA1 would do $9.7 million in sales in 2010 and $34 million in sales in 2011. On an August 2010 earnings call during which the company reported its first revenue from sales of the test, CEO Gail Page put the long-term market opportunity for OVA1 at more than 1 million units per year.
Three years later, Page is gone, let go last November after a string of disappointing quarters. Sales of OVA1 have plateaued at around 17,000 tests per year — far below the number needed for the company to break even, let alone meet its former lofty projections. At its annual meeting in March, the company's investors signaled their dissatisfaction with management by electing a dissident shareholder to the board of directors. Since the beginning of April, Vermillion's stock has been trading in a range of $1.03 to a high of $2.44 per share today, following the announcement of an equity investment earlier this week.
A biotech outfit generating excitement with a promising, but untested, new product and then struggling to deliver on this early potential certainly isn't unheard of. What makes Vermillion interesting, though, is this: Even allowing for OVA1's lackluster showing in the marketplace, the company may be the biggest success story in the cancer protein biomarker space.
Such an assessment speaks less to the strength of Vermillion's accomplishments than to the grim performance of the field as a whole. Asked for examples of cancer protein markers in clinical practice, Swiss Federal Institute of Technology Zurich researcher Ruedi Aebersold offered this appraisal: "There are not a lot. And most [that are in use] are not really that great."
There was hope that proteomics, by allowing scientists to more easily analyze massive numbers of potential markers, would improve this situation. Thus far, that hasn't proven to be the case. The last decade has seen the publication of thousands of studies reporting to have discovered new protein biomarkers. During this same period, however, new proteins have made it into the clinic at a rate of around 1.5 per year. Against this landscape, Vermillion's ability to push a five-protein panel through the US Food and Drug Administration and to market stands out.
The Vermillion path
The company originally incorporated in 1993 as Abiotic Systems, a California-based mass spectrometry firm. In 1995, it changed its name to Ciphergen Biosystems and, two years later, launched its ProteinChip SELDI-TOF mass spec system. Though SELDI-MS is rarely used today due to issues such as poor reproducibility, upon its launch the ProteinChip system became a popular platform for protein biomarker discovery, and Ciphergen, which in 2007 would change its name to Vermillion, developed its own internal biomarker programs, including the ovarian cancer effort that ultimately led to OVA1.
This effort was led by Johns Hopkins University researcher Daniel W. Chan, director of the school's Center for Biomarker Discovery and an early user of the ProteinChip device. Looking back on Ciphergen's nascent efforts to market the system for protein biomarker discovery, Chan recalled a certain naivety to the company's expectations for the platform.
"This was back in the late 1990s when the SELDI technology was beginning to be developed and Ciphergen had started marketing it," he said. "And initially in my discussions with the company, they figured that with this technology, [researchers] would be able to find biomarkers in a short period of time."
Chan and his colleagues soon came to realize that this wasn't the case. "There was still work to do to improve the [ProteinChip] technology," he said. More important than that, however, he noted, were the improvements needed in researchers' applications of the technology.
In the early days of proteomics, there was an assumption among many in the field that "you could just get a bunch of samples and analyze them using mass spectrometry and identify markers that could be used clinically," Chan said. In reality, effective biomarker discovery and implementation has turned out to be considerably more demanding.
Sudhir Srivastava is chief of the Cancer Biomarkers Research Group at the Early Detection Research Network, an organization established by the National Cancer Institute to help speed the translation of cancer biomarkers. The network was launched in 2000 following an investigation by NCI into why so few of the biomarker discoveries published each year were making their way into the clinic.
"We took about eight or nine months of discussion," Srivastava said, "and then we found the following: Number one, people are not as interested in validating their findings as in discovering them. Number two, there was a disconnect between the discovery process and the intended clinical use of the biomarkers. And, number three, the [researchers] who were discovering [the markers] were often validating them using bad study designs."
The EDRN aims to remedy these problems by offering resources and infrastructure for moving potential markers from discovery through analytical and clinical validation and, ultimately, into medical practice. In the 13 years since its founding, Srivastava noted, the organization has helped five cancer biomarker tests gain FDA approval, including OVA1.
"There are a few key things we learned" during OVA1's development, Chan said — for instance, he noted, the importance of using large sample sets for initial discovery experiments.
Due to the difficulty of obtaining large numbers of clinical samples, discovery is often done in relatively small sets — 20 cancer cases and 20 controls, for example. "But I can tell you that this approach doesn't always work," Chan said, observing that by using such small cohorts, researchers risk mistaking natural variability between individuals for disease markers.
Another lesson, he said, is that researchers should take care to perform their discovery and validation work in sample sets that accurately reflect their markers' intended clinical use. Ongoing research into biomarkers for aggressive prostate cancer — another of Chan's areas of focus — offers a case in point.
"You want to find markers that allow you to detect prostate cancers that you need to treat, not prostate cancer that isn't aggressive and probably won't cause harm to patients later on," Chan said. "And I've reviewed a lot of grant submissions where people say, 'OK, I'll take a set of prostate cancer [samples] versus non-cancer, and once I find a marker then I'll test it in the aggressive population versus the non-aggressive population.
"But what you need to do is use those [aggressive and non-aggressive] populations for [the initial discovery]," he said. "The selection of the proper patient population … is very critical to the success of biomarker discovery and validation."
The biobank bottleneck
Indeed, as researchers have grown more aware of the need to carefully select patient cohorts, sample collection and access have emerged as fundamental issues in the cancer protein marker field.
"I've been in numerous discussion groups and strategy meetings at NIH or the EU or other agencies where the bottlenecks of biomarker research are discussed," ETH Zurich's Aebersold said. "And every time, when there is a summary report issued, the matter of biobanking is way up at the top. Access to samples that are well annotated and well preserved is absolutely critical."
Biobanking "is slow. It's tedious. It costs a lot. It isn't glamorous," Aebersold said. "So it's a huge bottleneck. Generally, for a laboratory like ours, it's difficult to get access to suitable sample cohorts."
Such access can be difficult — meaning expensive — to come by in industry, too, noted Donald Munroe, Vermillion's chief scientific officer and senior vice president of business development.
"The first thing you have to accept is that you're prepared to pay for what is needed," he said. "It's cheap and easy to just get some cancer samples and some healthy controls, but they probably won't be informative."
"Clinical specimens gate all [biomarker] product development," added Albert Luderer, CEO of Seattle-based proteomics firm Integrated Diagnostics, which plans to launch its first test, a proteomic diagnostic for lung cancer, later this year.
While sample access is, of course, an issue for all omics efforts, and, indeed, for clinical research in general, the challenges are particularly acute for protein work. Unlike DNA, proteins — and especially certain post-translationally modified variants like phosphoproteins — are relatively labile, and so subject to alterations during sample collection and handling.
This is compounded by the fact that, given the newness of proteomics as a field and the relative dominance of genomics in biomarker research, few existing sample sets were collected specifically for protein work.
"There's an attraction to using archived samples" for protein biomarker research, said Emmanuel Petricoin, director of George Mason University's Center for Applied Proteomics and Molecular Medicine. "They are there. They are banked. You have the outcome data. And so some people don't want to believe that there might be issues with them from a pre-analytical variability standpoint."
Or, he noted, in the case of samples collected for genomics work, "the attitude sometime is, 'Well, let's just collect any proteomic data we can on them because it's all gravy. Some of it might be flawed, but it's better than nothing.'"
This has led to what Petricoin called a "dichotimized state" with regards to protein marker data. One the one side are the aforementioned researchers, optimists with regard to use of potentially problematic samples. And on the other, he said, "are the people who are extremely anxious about [these sample collection issues], who view a lot of the repositories with a jaundiced eye and feel like perhaps none of the proteomic [data] from these sample sets can be believed."
The reality, Petricoin said, "probably lies somewhere in between."
More specifically, cancer proteomics has made notable progress in sample collection procedures where blood-based markers for purposes like early detection or recurrence monitoring are concerned, he added.
"That area, I think, has improved significantly because of all the different studies where it has been documented that whatever [markers] were found were due to the way the sample was handled," Petricoin said.
Collection of tissue samples such as biopsies for personalized medicine applications, on the other hand, still needs improvement, he said. "There are a lot of people who just want to grind up the tissue and analyze it. They don't want to think about how long that tissue sat on the bench. They don't want to think about the fact that formalin penetrates very slowly."
In addition, "on the tissue side there are so many hands in the pot," Petricoin said. "Radiology, pathology, the community hospital oncologist. And then there are all the kinds of tests people want to do — PCR [and fluorescent in situ hybridization]. And the personalized medicine space has been so dominated by genomics for the last 10 years that most of the [standard operating procedures] have been developed for genomics analysis. And since DNA is stable, the attention to detail for that has been pretty shoddy. The methodologies for preserving proteins and phosphoproteins have been basically an afterthought."
"Tissue samples are much more complicated [than blood samples] to extract and store properly and even categorize" for protein biomarker research, added Mary Lopez, director of Thermo Fisher Scientific's Biomarker Research Initiatives in Mass Spectrometry Center.
For instance, she noted, the basic process of tissue normalization — establishing baseline measurements across which protein content in different samples can be compared — remains a vexing issue for the field. "Do we use wet weight, dry weight, protein assays?" she said. "Intellectually, it's not that hard a problem, but practically it's very hard to standardize methods and protocols, and for everyone to agree on what the right way is to do this."
And while in Petricoin's telling, tissue collection challenges primarily impact personalized medicine applications, Lopez said that they're relevant to blood-based protein biomarker research as well, given that tissue is often used in the discovery stages of such work.
"It's easier, perhaps, to identify biomarkers in tissue samples because they are higher abundance and there's less of a dynamic range issue," she said. "But it's not so great with respect to understanding how to prepare the sample, how the sample was stored, what the morphology of the sample is. So, it's a catch-22, if you will."
Efforts are underway to improve sample collection and access for protein biomarker research, Lopez noted, citing in particular the work of Lund University researcher Gyorgy Marko-Varga, who is heading up the biobanking and biomarker efforts for Sweden's BIG3 clinical study — a prospective longitudinal study of lung cancer, cardiovascular disease, and chronic obstructive pulmonary disease that aims to collect roughly five million samples from 100,000 subjects in the country's Skane county.
Formerly a principal scientist at AstraZeneca, where he led a 4,000-patient protein biomarker study investigating the efficacy and safety of treating various cohorts with that company's lung cancer drug Iressa, Marko-Varga said that, given the multitude of uses the BIG3 samples will ultimately be put to, the researchers have adopted a strategy of simply getting their samples into storage as quickly as possible.
"For DNA this isn't a big issue," he said. "If you [wait] 24 hours or 36 hours it's OK, because it's a stable molecule. However, if you want to look at proteins ... some are [stable] at eight hours, some are at 10, some are at 36. If you ask 10 clinicians, they will all tell you different requirements, so the strategy we chose was to just do [the collection and freezing] as fast as possible."
Currently, all samples for the project are frozen within two hours, Marko-Varga said. They are stored in small aliquots in a 384 well-plate format, meaning "that we can send the samples all over the world and they will only be taken out once from the biobank and then discarded," he says. "We don't freeze, store, freeze, store, because as soon as you do that, the samples are no longer the same."
Since the BIG3 study's launch in 2010, Marko-Varga and his colleagues have established fully-automated sample processing systems in two hospitals, he notes, and work is currently underway on installing them in an additional two.
"People ask why there hasn't been more success [in protein biomarker research.] And quality of samples is one main reason," Marko-Varga said. Good sample collection and effective marker development, he noted, "go hand in hand."
Signs of progress
Cancer protein researchers are pursuing advances along other lines as well. Assay throughput, for instance — a critical requirement for clinical validation and implementation — has seen significant improvement in recent years, particularly with regard to mass spec-based work.
"I'm a little bit embarrassed to say that in the 5,000 or 10,000 papers on biomarker proteomics, I don't know of a single one in which anybody has actually run 1,000 samples," proteomics researcher Leigh Anderson observed during a 2011 presentation at the Association for Mass Spectrometry's annual Applications to the Clinical Lab meeting.
"It's well known in the diagnostics community that if you can't run a few thousand samples, you can't know if a biomarker is clinically relevant," he said. "So this is a huge limitation that we need to overcome."
Two years later, due in no small part to research efforts led by Anderson in collaboration with Agilent Technologies, mass spec instruments and workflows capable of running thousands of clinical samples do, in fact, exist, with protein biomarker firms including Integrated Diagnostics and San Diego, Calif.-based Applied Proteomics prepping cancer panels designed to run on such platforms.
Researchers have also developed savvier attitudes toward discovery, moving away from broad screening strategies to more deliberate and targeted methods.
"Initially, questions were defined as just, 'Let's find a cancer biomarker,'" Aebersold said, recalling early protein marker discovery efforts. "Now it's, 'Let's find a biomarker for a particular type or stage of cancer, or a biomarker of whether a person will respond to a certain type of drug.' So the questions are much more granular and refined."
This shift in approach, Petricoin noted, has led to what might be seen as a reigning in of the field's ambitions. While much of the initial excitement surrounding proteomics focused on its potential as a tool for the early detection of cancer, such a goal was probably always somewhat unrealistic.
"A whole population screening early detection test is really, really tough," Petricoin said. Because of cancer's relatively low prevalence in the general population "your markers have to be unbelievably fantastic — like 100 percent sensitive and specific" to reliably pick up cancer cases without also generating large numbers of false positives.
Given this, much cancer protein marker research today tends toward narrower applications like monitoring recurrence of disease or helping to evaluate results obtained via more established screening procedures like imaging.
"Early detection is a great idea," but focusing for now on more modest aims offers researchers a good way "to walk their way through," cancer marker development, Petricoin said. "If you can't find evidence that you can detect recurrence early, then why do you think you can detect the presence of disease early in a general population?"
Another potentially interesting development, Aebersold noted, is the recent trend toward connecting cancer proteomics with genomics, as in the ongoing collaboration between the National Cancer Institute's Cancer Genome Atlas project and its Clinical Proteomic Tumor Analysis Consortium. In that project, CPTAC researchers are investigating potential protein cancer markers in tumor samples that have undergone extensive genomic characterization by TCGA.
"There's not a lot of data [on such approaches] out yet, but I think this is where currently the most excitement in the field is," he said. "Cancer genomics can sequence hundreds of tumors [along with] adjacent health tissue in the same individual," allowing profiling of genomic differences from which researchers can identify potential protein markers.
This means "we can basically do a lot of discovery by computer [by] taking advantage of the enormous richness of existing genomics data and trying to organize the effect of genomics on the level of the proteome," Aebersold said, adding that in a paper published last year in Science Translational Medicine he and his co-authors used such an approach to identify protein markers for ovarian cancer. From that, they ultimately put together a panel that, in initial findings, distinguished ovarian cancer and benign tumors with performance roughly equivalent to that of Vermillion's OVA1.
Broadly speaking, Chan said, protein marker development has grown more rigorous and sophisticated since he and Vermillion began work on OVA1 more than a decade ago.
"I think the field has recognized some of the lessons" that proved key to the test's development, Chan said. "Some of these things are not things we knew 10 years ago. But it's a learning process. These concepts are still evolving, and people are beginning to appreciate them more and more."
"I think that today — and maybe it's from watching companies like Vermillion move through the [marker development] process — there's a much better appreciation for what it takes," said Vermillion CEO Thomas McLain.
But while Chan and Vermillion have demonstrated that a sufficiently exacting approach can move cancer protein markers out of discovery and into the marketplace, they haven't yet shown that firms or their investors can make money doing it.
"I've always viewed the [cancer biomarkers] market as having three inherent risk factors: scientific risk, regulatory risk, and market risk," Peter Levine, former CEO of now-defunct protein diagnostics firm Correlogic, said. Having led that company from its founding in 2000 through to its bankruptcy and ultimate dissolution 11 years later, Levine has a more intimate experience than most with these risk factors.
Correlogic was a competitor of Vermillion's in the ovarian cancer space, with its own protein biomarker test, OvaCheck, aimed at distinguishing benign and malignant ovarian masses. Unlike OVA1, OvaCheck never made it to market as Correlogic ran out of money while still pushing the test through the FDA approval process.
Levine placed much of the blame for Correlogic and OvaCheck's failure on the exigencies and vagaries of the FDA process — regulatory risk, in other words. Having observed Vermillion's struggles since OVA1's launch in 2010, however, he didn't discount the market risk for such tests either.
"Vermillion's [roughly $600 per OVA1 test] reimbursement is a very nice fee per test, but will that recoup [investors' money] if you're only doing around 16,000 tests per year?" he asked. "So, it really is a questionable business model."
To break even given its current spend rate, Vermillion needs to sell in the range of 60,000 to 70,000 OVA1 tests per year — more than triple the number sold in 2012. The company has embarked on a number of initiatives — from additional clinical studies to programs aimed at improving reimbursement and physician reorders — with an eye towards driving sales.
The question, though, is whether such efforts will prove sufficient — and, perhaps more to the point, whether they will work in time. In its most recent quarterly report, filed last November, Vermillion noted that to continue operations through 2013 and beyond it would require additional capital, adding that, given this need, there was "substantial doubt about the company's ability to continue as a going concern."
Vermillion did obtain some breathing room this week, announcing an equity financing agreement under which investors including Oracle Investment Management, Jack W. Schuler, and Matthew W. Strobeck are to purchase 8 million shares of the company's common stock for $13.2 million. The deal also calls for Vermillion to issue these investors 12.5 million warrants with a strike price of $1.46, bringing the total potential investment to $31.5 million.
"Over the last five years, you've had very few [cancer biomarker] companies that have made it through to profitability," said Paul Beresford, vice president of business development and strategic marketing at Boulder, Colo.-based proteomics firm Biodesix. Biodesix is also in the early stages of driving adoption and reimbursement of a serum protein test — Veristrat, in its case, a diagnostic for use in guiding drug therapy in second-line advanced lung cancer.
A major challenge the proteomic cancer testing business has run into, Beresford said, is investors, and in some cases the firms themselves, underestimating the time and money required to validate and then drive adoption of a test after the initial discovery phase.
"Discovery may be in the thousands of dollars in terms of expense," he said. "But the markers then have to be fully validated, so you're looking at companies spending millions of dollars to take them through retrospective and prospective trials."
And then there is the adoption component, Beresford noted. "Securing a sales force, driving through to reimbursement, getting paid for the test, and ultimately getting into the guidelines and standard of care. Once again that's millions of dollars to truly get that test to where it is used day in and day out by physicians and in the clinic."
Unlike Vermillion, Biodesix remains privately held, so information on its revenues from Veristrat is not available. However, according to a company press release, it has performed the test on around 5,000 patients since launching it in 2009.
Many investors, Beresford said, "are just learning this business and hadn't realized it was such a long road to profitability. The [investor] community is on that learning curve now, and I think they're moving forward with the understanding that it is going to take a bit longer for these [tests] to get into standard of care and that the business model for diagnostics generally is still being sorted out."
According to Adair Newhall, an associate at venture capital firm Domain Associates, although the number of VCs interested in investing in early stage life science outfits has dwindled, those that remain have become increasingly willing to extend their timelines for exiting these investments — particularly, he noted, as the "appetite for acquisition" in this space has waned.
Domain is an investor in Applied Proteomics, which is currently preparing validation studies for its lead product — a protein panel for identifying patients with colon polyps or adenomas who should get colonoscopies — with the goal of launching it commercially within the next few years.
"Ideally we like a five-year timeline from investment to liquidity event," Newhall said. But, he added, in the case of Applied, Domain expects to go longer.
"I think we always had a longer-term view with this investment because it started at such an early stage," he said, noting that the firm is willing, in this case, to extend its time horizon given the market potential of the Applied test. According to Applied, the diagnostic has a potential US market of around 14 million patients a year.
In retrospect, Levine said, he and Correlogic likely underestimated the importance of maximizing their market potential. Comparing OvaCheck to tests like Applied's, he said that it "was a mistake going after ovarian cancer" as the company's first target.
"While a terrible disease, it pales in market size compared to something like colorectal cancer," he said. "There, you have tens and tens of millions of people potentially at risk who are not being tested at all. So I certainly think that for companies that are working on simplified blood tests for diseases like colorectal cancer, if those tests can become a substitute or adjunct to standard tests like colonoscopies, they will become very, very successful."
Indeed, "the experiences of predecessor [cancer protein marker] companies" demonstrate the importance of pursuing indications "that have enough available market to justify commercial development," Integrated Diagnostics' Luderer said. His firm estimates that its lung cancer diagnostic, which is intended to help doctors determine whether lesions picked up by imaging techniques like CT scans are likely benign or malignant, has a US market potential of around 3.3 million patients per year.
Another issue hindering commercialization efforts, Luderer said, is that many companies in the space have lacked "the skills or resources required to perform 'missionary selling,' where physician education, coupled with the ability to close sales, are critical skills."
With release of its lung diagnostic slated for this year, Integrated Diagnostics will soon get the chance to put these insights to the test.
Thus far, it hasn't particularly paid to be an optimist where protein cancer markers are concerned. Surveying the field's prospects, though, Luderer sounded a note of confidence — as is perhaps only befitting a CEO on the cusp of a key launch.
"New diagnostic content that can drive breakthroughs in diagnosis is at a premium," he said. "And the largest unmet needs in diagnostic medicine await those with the vision and fortitude to be change agents."