Ovarian cancer: Peak experience guides biomarker hunt
Finding the right biomarkers for
cancer is a little like looking for love. You could waste a lot of time on worthless contenders. The quest can seem endless, and you may have to reconsider candidates you rejected the first time around. People you respect may strongly disagree with your choice once you find one you like. And even after making a commitment, there’s still no guarantee that what you’ve found is the real thing. Worst-case-scenario: You’re looking for Mr. Goodbar.
Few are willing to call off the search in either love or medicine, though—the
potential reward is just too magnificent. That’s one of the reasons for the
excitement over the discovery of three potential biomarkers for detecting early-stage
ovarian cancer (Cancer Res. 2004;64: 5882-5890).
The biomarkers, identified from serum proteomic analysis, are apolipoprotein A1 (down-regulated in cancer), a truncated form of transthyretin (also down-regulated), and a cleavage fragment of inter-XXX-trypsin inhibitor heavy chain H4 (up-regulated). Their discovery "is an excellent step toward identifying biomarkers for early detection of ovarian cancer," says Steven Skates, PhD, assistant professor of medicine, Harvard Medical School, and a biostatistician at Massachusetts General Hospital, whose own ovarian cancer research has led to the development of an algorithm for using longitudinal measurements of the CA 125 biomarker.
But just as exciting—perhaps even more so—is the means by which the biomarkers were discovered. The study was designed to take into account confounding variables that many previous studies may have overlooked, say the study’s authors, making it a more sophisticated approach to conducting proteomics-based biomarker research.
The markers emerged from a five-center case-control study, which was designed to contain internal as well as external validation based on the origin of samples. This is important for several reasons, say the researchers.
Previous proteomic studies have pointed to "an unfortunate fact of life," says Eric T. Fung, MD, PhD, vice president of clinical affairs for the diagnostics division of Ciphergen Biosystems Inc., Fremont, Calif., and one of the study’s coauthors. Studies in which samples are obtained from a single site make it possible for researchers to obtain results that suggest very high levels of accuracy. But like a great first date, they can be hard to replicate later on. "When these results are attempted to be validated on a sample obtained from a different hospital, they are generally unsuccessful," says Dr. Fung.
First, there may be demographic and epidemiological differences between hospitals. Each institution may also have its own protocols for sample processing, and those differences, no matter how subtle, can be reflected in proteomic profiles. Finally, says Dr. Fung, the power of bioinformatics and biostatistics as applied to the complex data generated by proteomics platforms makes it possible "to create what we call ’overfit solutions’—that is, mathematical solutions that are tailored to the data that are given."
In the current study, researchers collected 503 specimens from M.D. Anderson Cancer Center; Duke University Medical Center; Groningen University Hospital, the Netherlands; and the Royal Hospital for Women, Australia. This included 153 patients with invasive epithelial ovarian cancer (65 with stages I/II, 88 with stages III/IV), 42 with other ovarian cancers, 166 with benign pelvic masses, and 142 healthy women. In addition to these specimens, they tested 142 independent archived serum specimens initially collected at Johns Hopkins Medical Institutions for routine clinical laboratory testing. "It’s the largest cohort of ovarian cancer patients in a proteomics study that I know of," says Dr. Fung.
Their approach was twofold. For the biomarker discovery, researchers used stage I/II samples and healthy controls from Duke and Groningen, analyzed separately according to each site, to come up with a short list of candidate markers; only those that shared the same up or down dysregulation patterns in analyses of data from both sites and that were deemed statistically informative made the cut. "The reality is very, very few were on the list," says lead author Zhen Zhang, PhD, associate professor of pathology and associate director of the Center for Biomarker Discovery, both at Johns Hopkins. The markers that survived from the first two sites were then further validated, using all the samples at Royal Hospital and M. D. Anderson as well as the remaining samples from the first two sites. Samples from Johns Hopkins were used to independently confirm the findings using traditional immunoassays. Protein expression profiling was done using Ciphergen’s ProteinChip Biomarker System, a platform for employing surface-enhanced laser desorption/ionization time-of-flight mass spectrometry. Each sample was fractionated multiple times, says Dr. Zhang, with each fraction run in triplicate on multiple array surfaces.
The researchers moved beyond simply determining which mass spectrometry peaks were important for classification to obtaining their underlying proteomic identification, which serves several purposes. While the proteins identified in the discovery phase may not be ideal—they may not survive at room temperature, for example, or may otherwise not be viable—the researchers might nonetheless use them to search further along the protein path for better markers if the need arises. Identifying them also may help illuminate the role they play in the disease pathway. Finally and most important, says Dr. Zhang, it provides biological soundness to what otherwise would be mere theoretical potential. Good results from a computer don’t always translate into real-world success, but understanding the actual biological plausibility of a marker’s value should help researchers make intelligent decisions about whether to proceed with larger scale validation studies.
This practical thrust of the study is no surprise, given the involvement of the Center for Biomarker Discovery at Johns Hopkins. The purpose of the center, which was established a little more than four years ago, is to use proteomic technology to find new markers that can be translated into diagnostic clinical laboratory use, says the center’s founder, Daniel W. Chan, PhD, another of the study’s coauthors. "So our focus is perhaps different from most basic science researchers. We really want to convert our discoveries into patient use," says Dr. Chan, a director of the center and professor of pathology, oncology, radiology, and urology, Johns Hopkins. "That’s why we think it’s important to look beyond just proteomic profiling. We want to identify specific biomarkers so we can study the proteins, learn about them, and make sure they are consistent and stable and so on."
The feet-firmly-on-the-ground approach led the researchers to pay close attention to the study’s blueprint.
"More and more, people are realizing that the design of a proteomic study is an issue," says Dr. Zhang. "They may not use our specific design, but at least people are starting to realize the issue of confounding variables." In the early days of proteomic research, he says, linking mass spectrometry peak patterns to proteins and peptides was the end of the road for many researchers. "They got excited by the technology, they put on some samples, they saw the results, and they had markers."
Dr. Zhang and his colleagues took that a step further, says Dr. Skates, when they used immunologically based assays to measure the peptides in serum, independently of the SELDI analysis. "That last step is something that I don’t think anyone else has done yet," says Dr. Skates.
"It’s a step that is very important," he continues, "because there is a belief that the SELDI measurement—peak height—is quite variable." The strengths of ELISA measurements, on the other hand, are known to all: They are very accurate, with a much smaller coefficient of variation (repeated measurements have a very small deviation from each other), and with results that can be replicated by other laboratories.
Whether SELDI measurements are indeed highly variable is not yet clear, Dr. Skates suggests. Some might argue that mass spectrometry peaks are sufficient to differentiate cancers from noncancers; others argue that the dynamic nature of the human proteome can lead to wildly divergent measurements. But what is clear, he says, is that the stability of mass spectrometry peak heights raises plenty of questions. So using ELISA-based assays "is a lot more convincing to a lot more people."
The next step, he says, would be to use a collection of serial serum samples gathered as part of a screening study, measuring the markers in the samples over time to see if they identified ovarian cancers before clinical diagnoses, without too many false-positive identifications. "Now that’s the acid test. That would really set these markers apart from many other potential ovarian cancer markers." He adds that the researchers "have done the work to justify essentially using up an aliquot of precious serum from such studies."
Those words aren’t spoken lightly. Obtaining serum samples before ovarian cancer is detected is challenging, to say the least, given the low incidence of the disease. Obtaining serial samples from 10 such cases of ovarian cancer would require screening approximately 25,000 postmenopausal women in one year, or the equivalent, such as 5,000 women for five years. "It’s a huge amount of work," says Dr. Skates. "So you don’t want to use those serum samples on a marker that hasn’t passed a significant hurdle" using more readily obtainable samples drawn at the time of clinical diagnosis.
Dr. Skates says if the three markers pass the test using serial samples from previous screening studies, investigators may consider using them as a panel in prospective ovarian cancer screening studies.
Biorepositories of serial serum samples are being established to test emerging markers as part of ongoing ovarian cancer screening studies, including studies based on the longitudinal CA 125 approach. One such effort is the UK Collaborative Trial of Ovarian Cancer Screening, led by professor Ian Jacobs of University College London, which involves 200,000 normal-risk, postmenopausal women randomized to three groups, including 50,000 to an arm using the longitudinal CA 125 risk of ovarian cancer algorithm approach. (Another 50,000 are randomized to an arm using annual ultrasound as a first-line test, and the remaining 100,000 are part of a control group.) The study began in mid-2001 and is scheduled to end in 2011, with an additional year’s followup.
In another effort, Dr. Skates and colleagues in the U.S. are looking at high-risk women in a pilot study involving pre- and postmenopausal women, measuring CA 125 every three months. High-risk subjects include women from a family with a mutation in the BRCA1 or BRCA2 gene, or with multiple ovarian or breast cancers, or both, in the same lineage.
The need for new biomarkers is a given. But researchers are not necessarily looking for the equivalent of the perfect mate. "We’re really not looking for a silver bullet, at least not right now," says Dr. Zhang. Cancer’s heterogeneity makes it unlikely that a single marker will be enough.
Adds Dr. Chan: "I think we’re asking for too much for one protein to be able to diagnose a cancer. We can’t expect one protein to be able to do all the work for us." They’re biomarkers, not Cinderella, in other words.
The epidemiology of ovarian cancer is such that any test should have 99.7 percent specificity, says Dr. Zhang. With their multiple marker approach, the researchers were able to obtain 98 percent specificity, with a sensitivity of approximately 70 percent. By adding ultrasound, the researchers boosted specificity to the desired 99.7 percent, he reports. And while the sensitivity is "marginally good," he says, "it’s not acceptable. We’re looking to get in the 80s."
As it turns out, the researchers are not convinced other proteomic studies—including their own earlier work—have truly reached optimal sensitivity and specificity, either. While some studies have reported sensitivities and specificities of close to 100 percent, says Dr. Chan, "often when it’s too good to be true, it’s too good to be true." Studies whose numbers were derived solely by profiling may not be as accurate as they appear; the same may be true for studies using specimens from only a single institution. "We believe our findings are more realistic, more practical, and perhaps can stand in validation, and other people can reproduce the work later on," says Dr. Chan.
The three markers that emerged from this study offer some tantalizing glimpses for anyone interested in ovarian cancer proteomics.
The markers most likely represent host-response activity, says Dr. Fung.
"We actually believe this is logical in terms of the biology of the disease," he says. Conventional wisdom with respect to tumor markers is that these proteins are synthesized by the cancerous cells; that’s at least partially true, he says, because many tumor antigens such as PSA and CA 125 are indeed synthesized in such manner. "That being said, most conventional tumor markers are not useful for diagnosis of early-stage cancer but rather have a very good ability to pick up only late-stage disease." There’s a logic to that, say the researchers: Early-stage tumors may not be large enough to shed detectable levels of tumor marker into the general circulation.
Host-tumor interaction occurs even at the earliest stages, on the other hand. "So our hypothesis—and we’re doing additional research to confirm it—is that there’s a host-response protein amplification cascade," Dr. Fung says. This would be primarily an inflammatory cascade that incorporates proteases, protease inhibitors, and other types of processing enzymes. "So what we could detect is the host response by measuring how proteins are modified during the course of the host-response activity."
It’s possible, for example, that the truncated transthyretin, which lacks the NH2-terminal 10 amino acids, is a more specific marker than native transthyretin. "As long as we can find biomarkers or proteins that indicate the presence of the cancer, it is useful as a diagnostic marker," Dr. Chan suggests. "It does not necessarily have to be produced by the cancer itself."
The researchers are also doing additional discovery using what Dr. Fung calls the Deep Proteome suite of discovery tools. "They allow us to mine the layers of the proteome to a greater depth," he says, which may enable them to further characterize the thousands of proteins present in the serum proteome at low concentrations. "We believe we’ll find additional candidate markers using this approach." They’re also interested in using the SELDI platform to look at other candidate biomarkers identified in the literature. "Some of these might be useful in combination with the markers we just described," Dr. Fung says. "We know that some of these proteins are in fact post-translationally modified and may therefore be preferentially measured using SELDI."
Ciphergen is embarking on a retrospective and prospective clinical trial to validate and extend the findings of the study, Dr. Fung reports. "It’s our view that during the course of these efforts, we’ll be able to show how SELDI-based assays are going to change the way clinical laboratory testing can be done for certain diseases, because of the power of SELDI to quantitate and measure these different post-translational modifications." The paper reported that the truncated transthyretin had undergone post-translational modification; further analysis revealed that the same was true for apolipoprotein A1. "We don’t have the complete biochemical information to tell you what that modification of apolipoprotein A1 is, nor do we yet have the clinical data to tell you whether it’s clinically significant," says Dr. Fung. "But we are characterizing it."
In early October Dr. Fung reported on further work on two of the markers described
in the Cancer Research paper at the biennial meeting of the International
Gynecologic Cancer Society, in Edinburgh, Scotland. He says researchers were
able to build a reproducible SELDI-based assay for transthyretin and apolipoprotein
A1. And, he says, the markers were validated in an additional 700-sample set.
The markers also appear to be responsive to treatment, further evidence that
the markers may be related to the biology of the disease. Finally, very preliminary
data suggest that the markers occur at relatively normal levels when a woman
is in a disease-free stage, then drop off when the cancer recurs. "We’ve seen
that in several patients, and that’s work we’ll be continuing to develop."
Above all, however, the researchers emphasize their sights are set on finding markers for early detection. While markers for monitoring therapy and predicting appropriate therapy will also be helpful, "early detection is still the way to go," says Dr. Chan. It may be a little like love at first sight—no matter how hard it may be to find, the search continues.
Karen Titus is CAP TODAY contributing editor and co-managing editor.