The road to personalized medicine depends on development of biomarker assays that can reliably—with ultrahigh sensitivity and specificity—identify what is going on in the patient’s system. At CHI’s “Biomarker Assay Development” conference held recently in San Diego, several presentations focused on the challenges that continue to confront researchers as they progress toward this goal.
Among the speakers was Shannon Payne, senior scientist at Epigenomics. She described her group’s identification of a region of the Septin 9 gene that is methylated in more than 90% of colorectral cancer (CRC) tissues, with little or no methylation in normal colorectal tissue and other controls.
Because CRC typically develops with few or no symptoms, early detection is the key. The American Cancer Society’s goal is to screen 75% of average risk people 50 years of age and older by 2015, Dr. Payne noted. The percent of people currently being screened is around 50%. The current standard noninvasive test, the fecal blood test, requires too much patient involvement in an unpleasant process and is used at a <14% compliance rate.
Including all existing screening test modalities, 61% of patients are first diagnosed with late-stage disease. Dr. Payne and others are confident that a blood-based alternative test method would increase patient acceptance of asymptomatic testing.
DNA methylation provides unique advantages, Dr. Payne noted. It is biologically relevant as fundamental to gene regulation and genome stability; it is a rich source of biomarkers for diagnosis, prognosis, and drug response prediction; it provides technically robust detection and quantification, which makes it ideal for the clinical routine; it can use both routine body fluid samples and paraffin-fixed samples; and—of no small importance—comprehensive patent protection of biomarkers and technologies makes it commercially attractive.
Using an Affymetrix microarray platform that was custom made for Epigenomics, Dr. Payne’s group identified thousands of candidates identified as methylated in target tissue. From these the goal for optimal oncology biomarkers was to identify those with high methylation levels in target cancer tissue (high sensitivity) and low methylation levels in healthy tissue and other diseases (high specificity).
In one experiment, methylation levels were compared for the target CRC tissue samples, other cancers, healthy tissue, and benign disease. The sensitivity and specificity of the SEPT9 methylation was further characterized in a larger sample set and confirmed in real-time PCR studies. SEPT9 methylation was present in >90% of all CRC tissues.
As a member of the Septin family, SEPT9 has “its fingers in a lot of pies,” Dr. Payne explained, including breast and ovarian cancer, myeloid-lymphoid leukemia (MLL), and many other malignancies. But its methylation is highly specific for colorectal cancer. Detection of minute amounts of methylated SEPT9 DNA shed by the tumor into the blood requires a highly sensitive real-time qPCR assay. The average amount of free DNA in plasma is 3–10 ng/mL, and the target DNA can be 1/1000 or less of that concentration.
Hence a biomarker assay must detect very low copy numbers. “We are looking for signal from a small tumor in a sea of blood,” she stated. The test must detect all stages of disease; detect cases, not controls; have a low false positive rate; and be practical in the clinic. The Epigenomics process of DNA extraction, bisulfite treatment, and qPCR resulted in a 90% limit of detection equal to 6.9 picograms. When scaled-up it met all four goals: lower input volume of plasma, higher throughput, potential for automation, and lower cost.
Epigenomics has now verified the performance of the SEPT9 biomarker in blood samples from three cohorts of colonoscopy-verified individuals that identified CRC of all stages in 72%, 74%, and 69% of the samples, respectively. Controls were positive in 10%, 8%, and 11%, respectively. Prospective population-based studies are under way and commercial partners have been selected.
The success of many investigational drugs is dependent on matching treatments with the appropriate target populations, noted Daniel Chelsky, CSO at Caprion Proteomics, in his presentation. “Variability in response to therapy, both with regard to efficacy and to adverse events, is leading the pharmaceutical industry down the path of personalized medicine. Further pushing the process along are government and private insurance payers who are faced with expensive treatments that can help some but provide little benefit and possible harm to others.”
One promising solution to the problem, he observed, is to identify predictive biomarkers of drug efficacy; circulating proteins that stratify patients into populations of likely responders and nonresponders to a proposed therapy. Finding such biomarkers has been challenging due to the complexity of human plasma—the sample of choice—and to the available technologies for detection and quantification of thousands of proteins.
“Based on the experience of over two dozen preclinical and clinical proteomic studies with pharmaceutical partners, an industrialized and productive approach to biomarker discovery and validation has been developed by Caprion.” Dr. Chelsky stated.
On the path to personalized medicine, discovering biomarker candidates is not enough. The Caprion approach identifies proteins that predict drug efficacy or that stratify patients by stage and severity of disease through a well-controlled and industrialized mass spectrometry analysis of plasma samples, Dr. Chelsky added.
The process begins with uniform blood sample collection into tubes containing protease inhibitors. Plasma is depleted by antibody affinity of the high- and medium-abundance proteins that typically obscure biomarkers of interest. Isolated proteins from each sample are digested to peptides that are more accurately identified and quantified by a quadrupole time of flight mass spectrometer. Peptides are matched across all samples and compared for peak intensity in each cohort of patients.
Those peptides, which are differentially expressed, are targeted for fragmentation to generate the amino acid sequence that is matched to the parent protein. Peptides identifying the same protein are clustered and subjected to a consistency filter that requires all peptides from the same protein to show similar behavior in each patient.
In a study performed by Caprion for a pharmaceutical company, plasma samples were compared between patients with ovarian cancer and breast cancer, as well as with healthy control subjects. Of approximately 50,000 peptide ions tracked across the samples, 4,000 were found to be differentially expressed.
These peptides were used to determine the proteomic relationship between the patients by multidimensional scaling (MDS). Each cohort was found to be distinct from the perspective of plasma protein profiles. “Thus, not only can patients with disease be distinguished, but those with two related diseases can be separated as well,” Dr. Chelsky observed.
A similar study involved the identification of circulating prostate cancer biomarkers. To determine whether markers could be found in spite of significant differences in sample acquisition, three groups of samples were compared—healthy control samples from a commercial supplier and two sample sets from different suppliers of patients with prostate cancer.
The three groups were analyzed and compared to find 3,569 peptides out of approximately 43,000 detected across the sample population that distinguished the two cancer groups from the healthy controls. These peptides were used to compare the patients by MDS analysis. The peptides that distinguished the two prostate cancer sample sets from the controls were found to significantly overlap and the combined cancer group separated well from the healthy controls in the MDS plot.
Peptides that separated each cancer group from the controls were sequenced to identify approximately 200 proteins in each comparison. “The interesting finding,” Dr. Chelsky noted, “is that 141 of the proteins in each group were shared, demonstrating that the impact of collecting samples from different sources was relatively minor compared with the effect of the disease on the plasma proteome.”
Each of the prostate cancer patient sample groups contained individuals with stage T2 (n=14) and stage T3 (n=10) prostate cancer. A bioinformatics software similar to MDS was used to compare the patients. Most patients at the two stages could be separated based on their plasma proteomic profiles.
“This indicates that protein expression in plasma is not limited to distinguishing disease, but can be applied to measuring the stage or severity of the disease. The implications of this finding are important to diagnosis and treatment, as well as for monitoring response to therapy,” Dr. Chelsky concluded.