Molecular Subtypes of Pancreatic Cancer: A Proteomics Approach

In this issue of Clinical Cancer Research, Son and colleagues proposed a molecular signature comprising 24 proteins, quantified using multiple reaction monitoring-based mass spectrometry (MS), to classify pancreatic ductal adenocarcinomas (PDAC) into four risk subgroups (1). Unsupervised molecular subtyping has been highly effective in classifying various cancers with shared molecular and clinical characteristics. Precise molecular subtyping can help identify actionable targets for drug design, predict responses, and guide the clinicians for treatment planning. Patients with PDAC will benefit immensely from clinically relevant and easily identifiable subtypes. At present, the 5-year survival for patients with PDAC is less than 10%, a grim reminder of the urgent need to improve treatment approaches. The heterogeneous nature of the disease is a critical obstacle in disease characterization and treatment. Therefore, recent studies have switched focus to molecular characterization. Advancements in omics technologies and the evolution of various clustering methods continuously improve our understanding of disease biology and associated heterogeneity.

Finding a concordant signature from a small sample cohort that could represent a tumor type is a daunting task. In the past decade, various efforts have been made to classify PDACs at the molecular level, utilizing genetic and transcriptomic information to predict patient survival and therapy responses. Multiple studies have attempted to subgroup PDACs based on mRNA expression data (2). In this article, the authors used surgically resected tumor tissues from a local cohort of 225 patients with PDAC and postulated stable, exocrine-like, activated, and extracellular matrix (ECM)-remodeling subtypes based on proteomic signatures without considering tumor purity. It is to be noted that the defined subtypes showed a good correlation with the existing subtypes identified previously using transcriptomics data (2). For example, the “ECM remodeling” and “activated” subtypes from this study showed associations with basal-like and activated stroma subtypes, respectively, as defined by Puleo and colleagues (3). These subtypes from both the studies were associated with the worst survival. Similarly, subtypes with better prognosis (stable) overlapped with the classical subtype [defined by Moffitt and colleagues (4) and Collisson and colleagues (2)] and immunogenic [by Bailey and colleagues (5)], immune classical, and desmoplastic subtypes [both by Puleo and colleagues (3)]. Significant parallels between proteomic and transcriptomic profiling demonstrate that the data gathered in different labs using distinct omics platforms show concordance and could be applied globally to predict prognosis (Fig. 1). However, the actual utility of the proteomic method by Son and colleagues (1) lies in that it can be used on a relatively small amount of sample, without considering tumor purity, and can produce results much faster.

Tumors are a heterogeneous mix of both biotic (cancer cells and stromal cells) and abiotic (ECM, oxygen levels, pH, and nutrient availability) components, which are generally heterogeneous. The abiotic factors such as hypoxia may govern stemness and inadequate response to therapy; however, the relative enrichment of the biotic and abiotic factors in different tumor subtypes is not clear. The present study by Son and colleagues (1) indicates the two most aggressive tumor types, that is, “ECM remodeling” and “activated” subtypes, are relatively enriched in the hypoxic microenvironments as indicated by the expression of VEGFA, SLC2A1, and ENO2.

Another important aspect of molecular subtyping is to identify drug targets, design therapies, and predict responses. Previous subtypes achieved some success in defining the therapies for respective subtypes. For example, Collisson and colleagues identified that the classical subtype is sensitive to gemcitabine, while the quasi-mesenchymal tumors had higher sensitivity toward erlotinib (2). Subsequent studies also found correlations between the subtypes and putative therapeutic targets and therapies (2). The initial success in the clinic was observed during the COMPASS trial, where transcriptomics data was successfully used to identify previously defined classical and basal-like subtypes among the patient cohort (2). The trial further established that the classical tumor cohort responded better to the first-line chemotherapy than the basal-like tumor cohort (2). In the current study, based on the protein signatures, the authors were able to identify signaling pathways enriched among defined subtypes. For example, the higher expression of GATA-binding protein 6 (GATA6), galectin 4, and aldehyde dehydrogenase 1 A1 expression, observed in this study and previous studies, was associated with the subtypes with the best prognosis. The activated subtype associated with poor prognosis was enriched in the PI3K-Akt and MAPK signaling pathway proteins. Targeting PI3K-Akt and MAPK pathways has shown promising results in preclinical models of PDAC. Thus, activated subtype patients might benefit from agents targeting PI3K-Akt and MAPK signaling. Similarly, the ECM remodeling subtype patients could benefit from FAK and WNT-β-Catenin inhibitors. Various FAK inhibitors have shown promising results in preclinical models, and few of them are currently being tested in clinical settings. Another important signature associated with the ECM remodeling subtype and previous subtypes associated with worst survival outcomes is the expression of glucose transporters (6). The glucose transporters indicate a distinct metabolic phenotype among tumors of these cohorts. Metabolic reprogramming associated with a shift toward the Warburg effect is a crucial PDAC hallmark. Increased expression of glucose transporters and glycolytic enzymes is the prime feature associated with it. Patients with this tumor cohort can benefit from inhibitors targeting metabolic shifts toward the Warburg effect (7). Currently, radiotherapy and surgery are the only treatment options for patients with early and locally advanced disease, and chemotherapy is the sole option for metastatic disease. Chemotherapeutic regimens including FOLFIRINOX and gemcitabine-abraxane have shown some degree of efficacy in improving patients' survival with metastatic disease. However, the prognosis of patients with PDAC remains poor. Improved understanding of the disease will lead to personalized therapies to improve clinical outcomes.

Despite promises from the current and past studies, there are various challenges in utilizing these subtypes in the clinic. For example, assigning a single patient data confidently to any subtype remains a challenge (8). This could be due to the built of current models that rely on a large training cohort; thus, resulting in efficient population-based predictors but poor single sample predictors. Another reason could be tumor plasticity, and there is a larger possibility that tumors transition between various subtypes during disease progression, thus convoluting predictions. The current study and some previous studies utilized samples from surgical resections, which may not efficiently define subtypes for higher stage tumors. These predictions could be improved further by the inclusion of more samples and tissues from diverse PDAC stages, and the focus should be on identifying molecular signatures to predict a single patient outcome. In addition, efforts should be made to devise algorithms that could assign a single sample to a subgroup. Such tools should be made available to the clinicians in the most user-friendly interfaces. Another aspect is to educate clinicians and patients to facilitate their participation and an effortless bench to bedside transition of these efforts.

The use of proteomics for subtyping has emerged as a promising tool in the past couple of years. With the improvements made in the speed, sensitivity, mass accuracy, and resolution of MS, nearly the whole proteome can be quickly profiled. It can be further utilized to identify posttranslational modifications and expression profiling of proteins parallelly and to help improve our understanding of the molecular signatures. Future classifications will potentially use integrated proteomics, genomics, transcriptomics, epigenomics, and metabolomics approaches to comprehend the disease better.

No disclosures were reported.

This work was supported in part by funding from the NIH grant (R01 CA163649, R01 CA210439, and R01 CA216853, NCI) to P.K. Singh; P01 (P01 CA217798, NCI) to P.K. Singh.