Skip to main content
Premium Trial:

Request an Annual Quote

Better Biomarkers for the Diagnostics Labyrinth


Every year, more than 55 percent of men over 50 years old undergo screening for prostate cancer, a procedure that involves testing blood for the presence of elevated levels of prostate specific antigen, or PSA. As far as diagnostic tests go, it’s relatively successful: for the patient, who learns whether or not to worry about the potential for a tumor, and for the manufacturers and distributors of the assay, who sell the assay to clinical labs that charge up to $65 a pop, according to the National Prostate Cancer Coalition.

The promise of earning similar riches — not to mention the boon to early diagnosis of disease — has enticed many genomics and proteomics researchers and entrepreneurs into the search for biomarkers that could be turned into clinical diagnostics. Effective biomarkers, the argument goes, should help physicians more accurately treat patients, as well as provide cash-hungry biotechs with a faster route to translating genomic discoveries into stable revenue streams.

But what’s the best approach to developing a new diagnostic test? With all the newfangled methods for extracting information from the human genome — through gene expression, protein expression, SNP genotyping, and DNA methylation studies, to name just a few — are there certain approaches to deriving biomarkers that make more sense than others? One could argue that a single analyte, such as PSA, and a reliable assay for detection would seem the most tried and true approach, but what if such a single analyte can’t be found, or doesn’t exist?

Furthermore, given the ongoing struggle to develop new drugs, several researchers believe that the mechanisms of many diseases are too complicated for a single analyte to serve as a trustworthy indicator of disease status. One theory says that relying on patterns of gene or protein expression is necessary to take into account such inherent complexity. “There’s the realization that disease is a very heterogeneously complex issue,” says Emanuel Petricoin, co-director of the NCI-FDA clinical proteomics program. “A single marker is never going to do it, because humans — and disease — are too heterogeneous.”

Given the medical and business promise associated with an expanding diagnostics field, comparing the various approaches to biomarkers may give a sense of what’s to come — and which type of diagnostic is the best aim for your research. There are distinct differences that involve how easy it is to associate a biomarker with a disease condition, construct a robust assay, and in how regulatory agencies view the technology behind the assay. Scientists looking to find the right type of biomarker for a particular system, read on.


Obviously, choosing between the various approaches to developing biomarkers may depend in some cases more on necessity than convenience — if a protein expression pattern doesn’t correlate well with disease, but a gene expression pattern does, it would only make sense to go with the gene expression pattern. However, there are a few generalizations researchers make when justifying one approach over another. In some circumstances it may be easier to associate one type of biomarker with disease than another, or one type of assay technology may be generally more amenable to clinical reference labs. Alternatively, the more mature a particular technology for identifying and correlating a biomarker with disease, the more readily it may be accepted by the diagnostics community.

Overall, diagnostics as a distinct focus of medical innovation is relatively new, compared to the millennia-old hunt for cures to disease. Only after World War II did companies such as Roche, Bayer, and Abbott begin developing the technology for tests that could help pinpoint patients’ ailments. Many of the first assays were fairly simple, such as a blood alcohol level test, developed by Roche in the early ’50s, and a dry reagent blood sugar test developed in the early ’60s by Bayer for diabetic patients. In the ’70s, as the field of molecular biology blossomed and researchers developed new instrumentation for automating sample analysis, diagnostics companies took advantage of new enzyme immunoassay technology and began to significantly expand their repertoire of new tests.

By the early ’90s, the introduction of PCR allowed diagnostics companies, Roche in particular, to develop assays that could test directly for viral diseases, including the first direct tests for hepatitis C and HIV, which Roche launched in ’92.

But it was the Human Genome Project and the flurry of biotech activity that coincided with its completion that sparked the most recent uptick in diagnostics interest. Droves of industry and academic scientists flocked to genomics as a means of discovering new therapeutics, albeit with limited success so far. Over the past few years the idea of developing biomarkers has caught on as a quicker and easier route to cash-producing products, and as an aid for more accurately diagnosing and prescribing therapeutic regimens. The question then becomes: which type of biomarker works best?

Ease of disease association

Broadly speaking, a biomarker is anything that contributes to understanding a patient’s disease susceptibility, diagnosis, or prognosis. Thus, a biomarker could include patterns of SNP, mRNA, DNA methylation, protein, or metabolite expression — as long as the pattern can be shown to correlate with the phenotypic expression of the disease. Of course, that’s the hard part.

SNPs, which are likely to be conserved across generations and subpopulations, tend to be most useful in divining disease susceptibility, particularly when the disease in question has a strong genetic component. A classic example of such a disease is cystic fibrosis, according to Greg Hines, CEO of Tm Bioscience, a Toronto-based purveyor of genotyping assays. In a comparison study undertaken with the help of the Mayo Clinic, Tm Bioscience was able to show that its SNP test for cystic fibrosis performed better than four of its competitors’ SNP tests on the basis of cost, accuracy, and re-run rate, Hines says. The upshot, in other words, is that for certain diseases, a genotyping assay can be an effective way of singling out those more likely to become sick.

But cystic fibrosis is an extreme example. Most diseases, if they have a genetic component, can also be triggered by environmental or other factors. In such cases, researchers working with gene and protein expression say that genotyping may not always catch an individual likely to suffer from the disease. Nick Dracopoli, vice president for clinical discovery technologies at Bristol-Myers Squibb, says that there are many ways a gene potentially involved in disease can be inactivated, and that gene expression profiling could be used to detect that inactivation, regardless of how it comes about.

“If you’re looking at loss of function, there are many ways a gene can be damaged or broken,” he says. “So there isn’t a single assay we can use; the presence or absence of an individual SNP is not ever going to be useful.” This is particularly true in cancer, he adds, where there are genetically unstable cells that change somatically. In this respect, “SNP assays are not on the whole that important,” he says.

Dracopoli’s argument in favor of gene expression as the biomarker of choice also applies to DNA methylation, he says. Just as evidence of a patient’s SNPs are likely to show up as changes in that person’s gene expression profile, evidence of DNA methylation can also be discerned by looking at variations in mRNA expression, Dracopoli says. “With DNA methylation in tumors, we could do experiments to measure methylation at individual points in the genome, and then correlate that with response to drugs or disease outcome,” he says. “But presumably any changes in methylation that are occurring in tumors will be reflected in changes in gene expression downstream of those genes that are being methylated.”

Methylation researchers, however, make the point that their chosen technique may offer unique insights into particular diseases, especially cancer. Nathan Lakey, president and CEO of Orion Genomics, which is trying to develop diagnostic tests based on changes in DNA methylation, says that these changes are ideally suited to distinguishing between tumor and normal cells.

“Every disease has its own nuances,” Lakey says. “Recently it’s been shown that [DNA methylation changes are] if not the primary reason for cancer, they’re definitely a major reason. … DNA methylation changes are very pronounced in tumor cells, but they’re not present in regular cells to the extent they are in tumor cells.” And DNA methylation offers a much cleaner signal than mRNA expression, Lakey says, because gene expression patterns are much noisier than the DNA events that trigger them.

Meanwhile, in terms of the central dogma, many researchers involved in protein expression are fond of making an analogy to theater, in which genes contain the script, and proteins, as the actors, perform the play. Since gene expression is often poorly correlated with protein expression — in the 30 percent to 50 percent range, says Eric Fung, director of clinical affairs for Ciphergen Biosystems — “what actually happens with the proteins more accurately reflects the pathophysiology of disease,” he adds.

Petricoin at the FDA, who along with his colleague Lance Liotta at the NCI has developed expertise in correlating patterns of protein expression with disease, argues that while protein expression may be complementary to gene expression or SNP analysis, the fact that proteins are centrally involved in the mechanisms of practically every disease make them ideal biomarker candidates. “I don’t think that people are saying that protein profiling has any special attributes that others don’t that would allow it to dominate,” Petricoin says. “If anything it’s the fact that proteins are so central to every disease process that [makes taking] a proteomic-based approach really promising work.”

But Petricoin, Fung, and many other researchers not directly involved in protein expression profiling agree that the days of finding single analytes that accurately communicate the presence or absence of disease are numbered. The fact of the matter, they say, is that the vast majority of diseases are too complicated for a single analyte to describe their status. “We’re all trying to find the fewest number of proteins that we need to come up with the diagnostic answer,” says Fung. “If we could do it with one protein that would be ideal, because it’s always easiest to just measure one protein. But the reality of human biology and disease is that it’s much more complicated.”

And ultimately, proving a biomarker’s medical relevance depends heavily on a well-executed clinical trial and the proper statistical significance of the results — regardless of whether the trial is studying a biomarker derived from gene or protein expression, or any other platform. The choice of biomarker type can affect how easy it will be to prove its medical relevance, but the bottom line depends on convincing data.

Least complicated assay technology

Comparing the various approaches to biomarkers on the basis of researchers’ ability to associate their presence (or absence) with disease is only part of the story. Developing a diagnostic test for the clinical reference lab market also requires that the technology for performing the assay be affordable and readily automatable. In this respect, the currently ubiquitous ELISA, or enzyme-linked immunosorbent assay, has it made. Over the past 30 years, advances in robotics and enzyme kinetics have allowed monoclonal antibodies to take a lead role in diagnostics. To unseat the ELISA, genomics-inspired techniques for developing biomarkers will also have to be tailored to suit the demands of the clinical reference lab market.

And those demands span more than just providing a new, medically relevant test that performs well, says Dennis Smith, the executive vice president of genomic strategies and chief medical officer at AmeriPath, a provider of anatomic pathology and diagnostics services headquartered in Riviera Beach, Fla. In addition, clinical reference labs are looking for new assays that are easily automatable, thereby reducing staff costs and turn-around time, he says. “Economics and performance are the most important factors,” Smith says.

The maturity of the technology behind a new assay is one way to gauge its potential to break into clinical reference labs. In fact, one could argue that the easiest inroad would be to take advantage of instrumentation and assay technology already installed in the majority of clinical reference labs today. In this respect, tests that rely on PCR, such as DNA methylation profiling or some types of SNP and gene expression analysis, have a potential advantage. Establishing the validity and market viability of a new diagnostic test that relies on already-installed technology is “certainly easier,” says Dracopoli at BMS.

Researchers advocating the protein profiling approach to biomarkers face the challenge of proving the necessity of adopting a technology that’s still being put through its paces. Identifying protein expression patterns that correlate with disease, as it’s currently practiced by Petricoin, Fung, and many others both in academia and industry, requires SELDI — or surface-enhanced laser desorption ionization — mass spectrometry and pattern-recognition algorithms, neither of which is as well understood as ELISAs, or even gene-expression profiling technology.

According to Petricoin, protein profiling avoids the time-consuming hassle of developing the high-quality antibodies necessary for an ELISA, but SELDI mass spec is newly charted territory, and researchers have yet to prove it can perform robustly and reproducibly. “It might work — it’s certainly showing a lot of research promise,” he says. And Fung points out that mass spectrometry is not completely foreign to clinical reference laboratories, given that some labs perform metabolic and toxicological studies using the technology.

Simplest sample source

Aside from the maturity of the assay technology, the source of the sample is another factor that contributes to the ease with which a particular assay can be performed. In general, assays that rely on tissue biopsies tend to be more expensive because the patient must undergo an invasive procedure; assays that can be performed on serum are preferable, and assays that need only excreted fluids such as urine or feces are even simpler still.

Given this hierarchy of sample sources, it follows that biomarkers easily obtained from serum or excreted fluids would have the upper hand. DNA, as it turns out, tends to be much more stable structurally than mRNA and many proteins, and is more likely to be detected in serum or excreted fluids than more fragile or degradable mRNA or protein biomarkers. Extracting tumor biopsies is often necessary to acquire high-quality mRNA, researchers say, and while many proteins can be detected in plasma, lots of them may have degraded or denatured.

For this reason, DNA methylation may offer some advantages over more mainstream gene or protein expression-based assays, says research group leader Florian Eckhardt at Epigenomics, a Berlin-based company seeking to develop biomarkers based on DNA methylation. Furthermore, Eckhardt says, an assay designed to detect a specific strand of DNA has the advantage of being able to rely on PCR to amplify the desired signal if it is found in the sample.

“If the analyte is stable, then you’re going to have a better chance at seeing it, and you’re going to have an easier time moving samples from the patient to the testing site,” says Lakey at Orion Genomics. “DNA by far is one of the most stable analytes,” he says. RNA is unstable by design, he adds, because the cell doesn’t want it hanging around producing more protein than necessary.

Despite the stability of DNA as a sample source, however, it’s clear that mRNA and protein expression profiling have a future in diagnostics. Indeed, for diseases such as skin cancer, tissue biopsies may be readily available for gene expression profiling. Researchers are also trying to work around mRNA stability problems by developing new approaches to identify surrogate markers. One such effort involves studying mononuclear cells in the body that may carry changes in their gene expression associated with the presence of disease, says Dracopoli at BMS. Similarly, Petricoin at the FDA thinks new research suggesting that fragments of biomarker proteins attach themselves to carrier proteins in serum could address issues of protein stability.

Impact of the regulatory environment

For better or for worse, the field of diagnostics is not free from government regulation. A fully-approved clinical reference lab can initially offer some types of new diagnostic assays as “home brew” tests, a classification that doesn’t require FDA approval — as long as the providers of the test make no claims as to its specific clinical efficacy — but eventually all in vitro diagnostics require the FDA’s stamp. Often, the developer of a new diagnotic assay will seek to offer the test as a home brew to acquire data that can be used to apply for FDA approval as an IVD, as well as a means of acquiring revenue and market traction in the near term.

The complicating factor is that the FDA is in the process of rewriting the regulations that govern an analyte-specific reagent, a classification of diagnostic test that many new gene and protein expression assays will most likely fall into. The outcome of these rule revisions will affect the hoops diagnostics companies or their partners will have to jump through prior to FDA approval as an IVD. Last year, Roche ran into trouble when it attempted to sell its gene expression-based test for certain P450 gene mutations as an ASR, because the company made claims about the test’s specific utility in the clinic. In the future, however, as the FDA revises its guidelines, this issue may or not continue to be a problem, or there may be new regulations that differ from current practice. Currently, the guidelines are in flux.

Steve Gutman, the director of the FDA’s Office of In Vitro Diagnostic Device Evaluation and Safety, says that as far as his office is concerned, “We’re agnostic when it comes to technology.” Rather than try to stifle innovation by professing support for one platform over another, Gutman says his primary concern lies in making sure that a test measures what it purports to measure, that it’s accurate and reproducible, and that the result of the test has clinical benefit. “What FDA doesn’t look at is cost — it’s not relevant to us, although it may be to others,” he says, “and we also don’t look at the actual medical impact, that is, the morbidity [or] mortality over time.”

“I’d like to think that the questions that FDA is going to ask any company are the same questions everybody would want to know before they would actually be daring enough to order the test,” Gutman adds. “I would be disturbed to think we’re asking a bunch of stupid questions.”

That said, Gutman acknowledges that a new test based on a more established technology may have an easier route through the regulatory process. Furthermore, a test that involves detecting many analytes is by nature more complicated than one that seeks to detect just a few. While this is not necessarily injurious to a test’s chances for FDA approval, it might make grooming the assay for FDA approval more laborious.

The rise of multiplexing

Meanwhile, the steady march of new biomarkers continues on. In May, LabCorp began offering the first genetic test for alpha-antitrypsin deficiency, the most common factor leading to chronic obstructive pulmonary disease, and more companies are jumping into the biomarker business daily. One of the most recent converts is MDS Proteomics, the Toronto-based proteomics arm of healthcare giant MDS, which said in May it plans to refocus its discovery efforts away from therapeutics toward biomarkers and lead-optimization services.

In the long run, researchers see single-analyte tests fading into the background as they are replaced by multiple-analyte tests involving patterns of gene and protein expression. In fact, Petricoin at the FDA foresees a time when a diagnostic test could encompass more than one approach — a hybrid diagnostic, so to speak. “Multiplexing is where it’s going to go,” Petricoin says. “There are even hybrid technologies that could arise out of this field that haven’t been thought of yet.”


Roche’s AmpliChip: A Cautionary Tale

If the hordes clamoring to get into biomarkers had any illusions that regulatory agencies would turn a blind eye to new diagnostic tests, Roche’s experience with its AmpliChip CYP450 drug metabolism test may serve as a cautionary tale.

In June of last year, Roche announced that it would begin selling the multiplexed microarray test, which is based on Affymetrix technology, as an analyte-specific reagent or ASR, one of two ways for diagnostics companies to bring new products to market with minimal or no FDA oversight. The device, which tests for cytochrome P450 liver enzymes that play a role in drug metabolism, is the result of an R&D collaboration between Roche and Affymetrix that began in February 2003.

In July of last year, however, the FDA sent Roche a letter requesting that the company discuss the presumed status of the test with FDA officials. Several months later, after digesting the results of the discussion, the FDA sent another letter to Roche asking that the company submit the device for premarket review, an approval process designed to determine a test’s safety and efficacy. At this point, Roche could no longer sell the AmpliChip CYP450 product as an ASR (it could still sell the P450 chip to drug makers for pre-clinical research).

To get into the diagnostic market, Roche is waiting until the FDA determines whether the product could be approved as an in vitro diagnostic, a review process that could take three years and cost Roche at least $750,000, according to Gus Rosania, an assistant professor of pharmaceutical sciences at the University of California, Berkeley, who was interviewed by GT’s sister publication Pharmacogenomics Reporter. Roche had always planned to submit the P450 chip for regulatory review, says Roche spokesperson Melinda Baker, but the company had hoped to earn revenue on sales of the chip as an ASR too. Baker adds that Roche plans to submit applications for in vitro diagnostic use of the AmpliChip CYP450 test to regulatory agencies in the US and EU later this year.

The ramifications of Roche’s dance with the FDA for other companies trying to break into diagnostics are not exactly straightforward. Roche ran into trouble with the FDA initially because it claimed in a press release that the CYP450 test had clinical significance (FDA regulations prohibit a company from using descriptions of potential benefits to a patient as a means of marketing an analyte-specific reagent).

Greg Hines, CEO of Tm Bioscience, says his company’s SNP-based test for certain mutations in the genes that encode for P450 enzymes did not face FDA scrutiny, even though his company launched the test within days of Roche launching its AmpliChip CYP450 test. Tm Bioscience apparently avoided FDA scrutiny because it did not make specific claims about the test’s clinical utility.

The FDA has had little to say about how it would treat diagnostic products based on multiplexed protein expression, much to the chagrin of some big pharmas, like GlaxoSmithKline, who have invested in pharmacoproteomics. Currently, the FDA is in the process of rewriting the regulations covering analyte-specific reagents.  



Biomarkers for Diagnostics: How they Compare

  SNP patterns DNA methylation Gene expression Protein expression

Which approach is easiest to associate with disease?

Tend to be conserved across generations and subpopulations; tend to be most useful in divining disease susceptibility, particularly when the disease in question has a strong genetic component.

DNA methylation changes are ideally suited to distinguishing between tumor and normal cells, says Nathan Lakey, president and CEO of Orion Genomics

There are many ways a gene potentially involved in disease can be inactivated, and gene expression profiling can be used to detect that inactivation, regardless of how it came about, says Nick Dracopoli, vice president for clinical discovery technologies at Bristol-Myers Squibb

“What actually happens with the proteins more accurately reflects the pathophysiology of disease,” says Eric Fung, director of clinical affairs for Ciphergen Biosystems

How robust is the assay technology?

Tests that rely on technology already installed in clinical reference labs, such as PCR for DNA methylation profiling or some types of SNP and gene expression analysis, potentially have an advantage

(See listing under “SNP patterns”)

(See listing under “SNP patterns”)

Protein profiling avoids the time-consuming hassle of developing the high-quality antibodies necessary for an ELISA, but SELDI mass spec is newly charted territory, and researchers have yet to prove it can perform robustly and reproducibly, says Emanuel Petricoin, co-director of the NCI-FDA clinical proteomics program

How does the FDA view the technologies?

A new test based on a more established technology may have an easier route through the regulatory process, says Steve Gutman, director of the FDA’s Office of In Vitro Diagnostic Device Evaluation and Safety

(See listing under “SNP patterns”)

(See listing under “SNP patterns”)

The FDA is just beginning to evaluate assays based on protein profiling technology


The Scan

Y Chromosome Study Reveals Details on Timing of Human Settlement in Americas

A Y chromosome-based analysis suggests South America may have first been settled more than 18,000 years ago, according to a new PLOS One study.

New Insights Into TP53-Driven Cancer

Researchers examine in Nature how TP53 mutations arise and spark tumor development.

Mapping Single-Cell Genomic, Transcriptomic Landscapes of Colorectal Cancer

In Genome Medicine, researchers present a map of single-cell genomic and transcriptomic landscapes of primary and metastatic colorectal cancer.

Expanded Genetic Testing Uncovers Hereditary Cancer Risk in Significant Subset of Cancer Patients

In Genome Medicine, researchers found pathogenic or likely pathogenic hereditary cancer risk variants in close to 17 percent of the 17,523 patients profiled with expanded germline genetic testing.