Biomarker Discovery from We to Me: Is Learning from Each Patient a New Approach?

In this issue of Clinical Cancer Research, Bellone and colleagues reported an exceptional response to pembrolizumab in ovarian cancer with clear cell features after progression to different lines of chemotherapy, bevacizumab, and radiation (1). Clinical trials with immune checkpoint inhibitors as a single agent in recurrent ovarian cancer showed objective response of approximately 10%, with encouraging activity in clear cell ovarian cancer subtype. The investigation of this patient's tumor showed a low number of mutations and an aberrant PD-L1 expression secondary to 3′ untranslated region (UTR) disruption of the PD-L1 gene. The PD-L1 overexpression was associated with the presence of T-cell lymphocytes in the initial tumor tissue, suggesting the importance of the PD-1/PD-L1–mediated immune escape in this cancer development. By analyzing this super responder, the researchers have discovered the genetic changes that underlie the response to treatment. From this translational study driven by a single patient, it is important to define what we need to learn to generate meaningful information.

In this case report, the data provide insights on the biology of how cancer cells can exploit gene regulation to evade immunosurveillance. The team described that the aberrant PD-L1 surface expression was secondary to a peculiar structural variant disrupting the 3′ region of the PD-L1 gene. This gain of function of the PD-L1 gene highlights the critical role of the 3′UTR sequence in the regulation of PD-L1 expression. This is consistent with a recent article showing that marked elevation of the PD-L1 transcripts was secondary to the truncation of the 3′UTR of the PD-L1 gene (2). The 3′UTR significantly determines the stability, localization, translation, and degradation of mRNA. The most prevalent class of regulatory elements in the 3′UTR are miRNA-binding sites, miRNAs being small noncoding RNAs regulating the majority of protein-coding genes. Further investigations are needed to determine the underlying molecular mechanisms of the 3′UTR-mediated regulation of PD-L1 expression and how this affects the immunosuppressive network.

In this report, immunohistochemistry (IHC) serves as a surrogate of the PD-L1 3′UTR disruption that actively evades antitumor immunity through PD-L1 overexpression. However, the use of PD-L1 expression on the tumor cells for patient selection remains controversial (3). Several factors contribute to the debate and are mainly related to the role of PD-L1 in this specific tumor/microenvironment and techniques used (cutoff definition and type of assays). In addition, the expression of PD-L1 is a dynamic process that contributes to the complexity of the PD-L1 status analysis. Recurrent serous ovarian cancer can harbor significantly increased PD-L1 expression compared with primary tumors (4). The function of this change is not well described, yet targeting this axis alone at the time of progression showed modest activity. In this case report, the aberrant overexpression of PD-L1 was a baseline inherent abnormality that probably drove the disease progression and thus was targetable for treatment. Previous chemotherapy and/or radiation can lead to increased expression of PD-L1 and be used for therapeutic intent. However, inducing a response to immune therapy is a multistep process, including multiple agonist and antagonist signals. Beyond the PD-1/PD-L1 axis, there are other immune escape mechanisms and various pathways that have to be considered simultaneously. Given the different mechanisms of actions and the regulation through a balance of positive and negative signals involved in the response to immune therapy (Fig. 1), identifying predictive biomarkers to immune therapy has been challenging.

Ovarian cancer is a heterogeneous disease evolving within an individual patient with genetically distinct subtypes of cancer cells, particularly in high-grade serous ovarian cancer characterized by TP53 mutation and genomic instability. Therefore, immune therapy is felt to have the potential to eliminate durably a broader range of tumor cell subtypes. However, targeting immune response is complex given the continuously evolving characteristics of the cancer cells, tumor microenvironment, and host immune system, which are constantly in interaction with one another (Fig. 1; ref. 3). At the cancer cell level, the mutational burden correlates with response to immune therapy, particularly in tumors with high clonal neoantigens. The changes of the tumor mutation/expression landscape as induced by the immune therapy may also be associated with response (4). The DNA repair pathway can impact the response to treatment, as tumors with mismatch repair deficiency may have a greater tumor mutation load. At the microenvironment level, tumor-infiltrating immune cells, expression of molecules with immunosuppressive activity, and the presence of inflammatory cells and cytokines are all factors that can influence the immune response. Hypoxia also regulates the immune response through secretion of diverse cytokines. At the host immune system level, peripheral blood, which includes tumor antigen–specific antibodies and antigen-reactive T cells, also mediates the immune system. Recently, response to PD-1 immunotherapy was also shown to be modulated by the gut microbiome, with differences observed in the diversity and composition of responders versus nonresponders (5).

Automated algorithms integrating all of the interactions may be needed to predict the immune response. This approach, coupled with careful evaluation of who responds, and who does not, is driven by the need to define biomarkers that can be applied for patient selection. Advances in next-generation sequencing (NGS) have generated abundant genomic data that may identify predictive biomarkers. Such approaches, driven by big data analyses, have significantly improved our understanding of disease biology and impacted our treatment management, as BRCA1/2 mutations, for example, in ovarian cancer. Bioinformatics analyses compiling multiple patients' data generate powerful information to identify prognostic and predictive biomarkers. Alongside this approach, NGS is also facilitating deeper understanding of resistance and response, in particular, the analysis of exceptional responders that allows discovery of markers and potentially revitalizes or repositions the use of agents in enriched populations (1). This individualized approach to biomarker analysis can be invaluable when it comes to a patient's treatment decision. At recurrence, the identification of the mechanisms involved in treatment resistance may be specific to each patient and each time. Innovative approaches will be needed to detect the resistance process involved and how to target it for treatment strategy. Big data are crucial to populate sufficient information and provide evidence of clinical utility; however, the investigation of the outlier patients can help in discovery and lead to personalized medicine, moving data generated by “we” to “me.” Both approaches are required depending on the question asked. The “we” (big data) provides strong data for prognostic biomarkers and population selection. The “me” (individual patient) allows pilot and hypothesis-driven data underlying the mechanisms involved in response and resistance to treatment for a specific patient.

Studying a single patient based on unusual response or resistance to treatment may lead to discovery of unknown mechanisms of action and enrich both science and patient care. We have come a long way from treating cancer as a single entity to specific histology subtype and now as the individual cancer patient with unique biology. Therefore, the vision of moving precision therapy to individualized medicine requires translating discovery into innovative clinical trial designs. Comprehensive tools are required to combine all data, leverage big data, and interpret patient data. This requires a change from conventional statistical approaches to innovative bioinformatics interpretation at an individual level. Laboratory information management systems have been developed to support this translational medicine. Artificial intelligence is also a new area in health care by integrating these complex networks to provide individualized targets for a patient at a specific time.

No potential conflicts of interest were disclosed.