The Challenges of Optimizing Biomarkers to Guide Clinical Decision Making

Defining an optimal therapeutic approach for prostate cancer remains a dilemma for patients and physicians. Although substantial progress has been made in screening, grading, and staging of prostate cancer, predictive biomarkers to define those tumors that are indolent vs those that are aggressive and likely to metastasize remain lacking, notwithstanding the challenges of the diversity and sequela of available therapeutic interventions for organ-confined tumors. Gleason Grading is a powerful tool to predict outcome (1); however, it is the underlying pathobiology of the individual tumors that drives outcome. The complex pathobiology of the prostate underlies these issues, best demonstrated by the complexity of identifying and sampling the clinically significant prostate cancer by imaging at the time of diagnosis (2).

In this issue of the Journal, Jamaspishvili et al. (3) advance our knowledge and raise new questions in defining the “bad actor” prostate tumors. Although loss of Phosphatase and tensin homolog (PTEN) has been known to be associated with poor outcomes in prostate cancer (4–7), making PTEN loss an actionable biomarker has been elusive.

In this study conducted at Queen’s University, Cancer Research Institute, Jamaspishvili and colleagues (3) refined a robust PTEN assay based on immunohistochemistry and either manual or automated interpretation to measure PTEN loss. The robustness of the assay and the extensive validation that the authors performed are demonstrated by the fact the authors were able to use different antibodies to PTEN and manual evaluation based on a defined rubric (5,7) and measure an impact on outcome on a diversity of tissue microarrays from different patient cohorts. The data were benchmarked against ERG Fusions (8) and CAPRA.S (9,10), both of which demonstrate substantial prognostic utility, which in multivariable analysis showed loss of PTEN was a better predictor of biochemical recurrence-free survival.

What makes this study noteworthy beyond the fact that Jamaspishvili et al. (3) have credentialed PTEN loss as a biomarker of progression in prostate cancer? The authors have moved the goalpost of what is adequate sampling within the context of fit-for-purpose application of biomarker detection, elevating the issue of tumor heterogeneity in tissue-based prognostic biomarkers.

The importance of adequate sampling in the context of pathology has been well known over a century. The quality of the diagnosis and capacity to prognosticate outcome are only as good as the tissue reviewed. In practice, the success of the current paradigm of histomorphologic evaluation is rather impressive. Although the diagnostic paradigm of histomorphologic evaluation from a needle biopsy rather than a surgically extirpated organ is well documented, in the instance of prostate cancer, the limited sample size of a needle biopsy may limit the prognostic capacity of biomolecular markers.

The application of tissue biomarkers is generally focused on the measurement of a presence or amplification of expression of a protein or nucleic acid. From a practical perspective, the presence or gain of a marker is easily identified. The application of these biomarkers tends toward minimal levels of expression, such that the risk of false negatives due to high thresholds is reduced. A limited number of cells for the denominator may be adequate if the markers are found to be present. The extreme of this led to the recommendation than any estrogen receptor-positive tumor cell was sufficient to classify a breast cancer as ER positive (11). In context, these recommendations were appropriate at the time, the assays were suboptimal, and the risk assessment of misclassification favored the use of estrogen blockade.

However, the measurement of the loss of expression of a protein or nucleic acid is far more complex. The number of cells that must be evaluated to determine if a biomarker is absent is critical. The denominator of events must be sufficient, and this denominator is variable depending on the cut point of loss for the biomarker as well as the performance of the test, including the complexity of preanalytic, analytic, and interpretative variables.

In the instance of PTEN loss in prostate cancer, the question is much more complex. As demonstrated by Jamaspishvili et al. (3) the evaluation of additional tissue microarray cores, resulting in a greater number of independent sampling events, led to better prediction of clinical behavior. Although a tissue microarray is not a tumor core biopsy, it is a surrogate. Tissue microarrays in this study were constructed from prostatectomy specimens that were evaluated after submission of the entire prostate sample to enable a survey of the prostate and Gleason grading of individual clones. The loss of PTEN is the negative predictive event, and the loss is sufficiently independent—within the context of both the intra- and intertumoral heterogeneity—to anticipate that application of this prognostic biomarker on core needle biopsies may demonstrate weaker prognostic performance solely because of tissue sampling. It is clear that increased biopsy number improves detection, even in the setting of imaging (12); however, it is unlikely that this sampling is adequate for robust evaluation of loss of PTEN compared with prostatectomy.

There has always been tension between the risk and benefits of biopsy vs surgical excision in oncology with reference to the adequacy of diagnosis, prognostic accuracy, and suitability for evaluation of predictive biomarkers. The goal of more precise information to forecast outcome and determine optimal therapeutic approach is offset by the goal of less invasive, rapid evaluation in the early stages of diagnosis and treatment planning. The advances in imaging technologies allow detection and sampling of smaller lesions, further adding to the diagnostic and treatment planning calculus. As Jamaspishvili et al. (3) demonstrate, the adequacy of tumor sampling remains a critical issue in treatment planning. This issue is not limited to the prostate, or to oncology, but is a challenge across the broad field of biomarkers for both prognostic and predictive applications across many fields of medicine. There is no simple solution to this dilemma; rather, it is critical that we are aware of these issues and continue to work to refine the performance of biomarkers to guide therapy.

Notes

There are no directly relevant disclosures related to the content of this article.

The author thanks Drs Peter Choyke and Robert Star for their critical input and consultation.

References