ABSTRACT
For a biomarker to work for early detection, it must show a measurable signal in apparently healthy persons without symptoms. Th is requirement poses unique challenges to biomarker discovery and validation. For biomarker discovery, it is common to use clinical samples, i.e. samples collected aft er clinical diagnosis of cancer, usually due to symptoms. Such an approach may yield biomarkers that are abundant in clinical samples and absent in control samples. However, this approach may not be optimal or effi cient for identifying early detection biomarkers, because biomarkers present in high quantity in clinical samples may not be present in measurable amounts in samples from early disease; conversely, biomarkers of early occult disease may not be present at high levels in clinical samples collected at diagnosis when the disease became clinically apparent. Even when a study is conducted with the utmost care and best study design, a biomarker discovered in clinical samples must be validated in pre-diagnostic samples, i.e. samples collected from healthy subjects who later developed the cancer. Our experience suggests that biomarkers identifi ed in clinical samples tend not to hold up in pre-diagnostic samples. Th ere are several possible reasons for this. First, when using samples at the extreme end of a spectrum, more abundant molecules can easily overwhelm much less abundant ones. Th e abundant molecules may be the result of exponential tumor growth, but are not present at high enough level during early stages of carcinogenesis. Meanwhile, the less abundant, “true signals” of early events may be missed. Second, the temporal expression pattern of most cancer biomarkers is unknown. Although most tumor-associated markers are up-regulated in tumors, some genes or proteins have been reported to be down-regulated in tumors as compared to normal tissues (Ali et al. 2006; Zhao et al. 2010). It is possible that some biomarkers, such as those associated with tumor tissue microenvironment, show diff erential signals in the early events of carcinogenesis, but the signals attenuate as the tumor progresses. Early stage tumors are likely to be very small, in the millimeter range. Using mathematical modeling, it was estimated that occult early stage serous ovarian cancers have a median diameter of less than 3 mm (Brown and Palmer 2009). Th e ability to detect a biomarker secreted by a tumor this small depends not only on assay sensitivity (limit of detection), but also on physiological factors such as, how much of the biomarker is secreted by the tumor, how stable the
protein is in the circulation, and many other factors. According to another study that relates tumor size and biomarker level in the blood, based also on mathematical modeling, detecting tumors in the millimeter range (diameter) is theoretically possible with current technologies, but only under unlikely physiological scenario of zero background noise, a high rate of biomarker secretion by the tumor, and a high proportion of the biomarker reaching the blood (Lutz et al. 2008) – an unlikely situation for most biomarkers in the blood. In addition, tumor heterogeneity poses another challenge. For example, molecular genetic analyses showed that diff erent ovarian cancer histological subtypes contain diff erent mutations spectra (Lalwani et al. 2011). Furthermore, a recent Cancer Genome Atlas project on high grade serous ovarian cancer showed signifi cant molecular heterogeneity within the subtype (Cancer Genome Atlas Research Network 2010). For these reasons, it is commonly considered that no single marker can detect a cancer with high enough sensitivity, and that a panel of biomarkers, combined in a model, is necessary in order to achieve requisite performance characteristics.
