Discovery of Colorectal Cancer PIK3CA Mutation as Potential Predictive Biomarker: Power and Promise of Molecular Pathological Epidemiology

Abstract

Regular use of aspirin reduces incidence and mortality of various cancers, including colorectal cancer. Anti-cancer effect of aspirin represents one of the “Provocative Questions” in cancer research. Experimental and clinical studies support a carcinogenic role for PTGS2 (cyclooxygenase-2), which is an important enzymatic mediator of inflammation, and a target of aspirin. Recent “Molecular Pathological Epidemiology” (MPE) research has shown that aspirin use is associated with better prognosis and clinical outcome in PIK3CA-mutated colorectal carcinoma, suggesting somatic PIK3CA mutation as a molecular biomarker that predicts response to aspirin therapy. The PI3K enzyme plays a pivotal role in the PI3K-AKT signaling pathway. Activating PIK3CA oncogene mutations are observed in various malignancies including breast cancer, ovarian cancer, brain tumor, hepatocellular carcinoma, lung cancer and colon cancer. The prevalence of PIK3CA mutations increases continuously from rectal to cecal cancers, supporting the “colorectal continuum” paradigm, and an important interplay of gut microbiota and host immune/inflammatory reaction. MPE represents an interdisciplinary integrative science, conceptually defined as “epidemiology of molecular heterogeneity of disease”. Because exposome and interactome vary from person to person and influence disease process, each disease process is unique (the unique disease principle). Hence, MPE concept and paradigm can extend to non-neoplastic diseases including diabetes mellitus, cardiovascular diseases, metabolic diseases, etc. MPE research opportunities are currently limited by paucity of tumor molecular data in existing large-scale population-based studies. However, genomic, epigenomic, and molecular pathology testing (e.g., analyses for microsatellite instability, MLH1 promoter CpG island methylation, and KRAS and BRAF mutations in colorectal tumors) is becoming routine clinical practice. In order for integrative molecular and population science to be routine practice, we must first reform education curricula by integrating both population and molecular biologic sciences. As consequences, next-generation hybrid molecular biological and population scientists can advance science, moving closer to personalized precision medicine and health care.

Introduction: Aspirin as Anti-Cancer Drug

Epidemiologic studies as well as evidence from randomized controlled trials indicates that regular use of aspirin reduces incidence of cancers including colorectal cancer,^1-10 and that regular use of aspirin after colorectal cancer diagnosis improves patient outcomes.^9-16 Experimental data also highlight an oncogenic role for PTGS2 (cyclooxygenase-2), enzyme central to the inflammatory response and a primary target of aspirin.^11,17-22 However, beyond the anti-inflammatory effect of aspirin and its influence on PTGS2, the mechanisms underlying improved outcome associated with regular aspirin use remain unclear. Mechanisms of anti-cancer effect of aspirin represent one of the “Provocative Questions” in cancer research.^23,24

Colorectal Cancer: Heterogenous Diseases

Colorectal cancers consist of a group of heterogenous disorders with diverse sets of genetic and epigenetic changes which accumulate during the carcinogenesis process.^25-29 Genomic and epigenomic analyses of colorectal cancer have revealed enormous heterogeneity of the disease.^30-38 In addition, molecular features and behavior of tumor cells are influenced by host immunity and inflammation^39-46 as well as by exposome (a totality of exposures from environment) and interactome (a totality of interactions of various molecules).⁴⁷ Evidence also suggests variability and continuum of functions of oncogenes, tumor suppressors and passenger genes/mutations.^47-52 This vast array of influences on both the initiation as well as progression of colorectal cancers poses a significant challenge to accurately predict the clinical behavior of any given tumor. Essentially, each tumor goes through its own unique pathway to cancer and ultimate killing of host.⁴⁷ If pathway A (to tumor A) is similar to pathway B (to tumor B) but not to pathway C (to tumor C), we can classify tumors A and B into one subtype, and tumor C into another subtype, using tumor biomarker(s). Tumor biomarkers can help classify molecularly-similar cancers into one subtype or another, to better predict their behaviors and response to therapy.⁴⁷ Thus, molecular testing of tumors has grown increasingly routine in clinical settings.^53-58 In addition to inter-tumor heterogeneity, intratumor heterogeneity adds another layer of complexity,⁵⁹ which can not only lead to heterogeneous biological behavior, but also a practical issue in tumor molecular testing.

Activating mutations in PIK3CA (official HGNC ID: HGNC:8975, phosphatidylinositol-4,5-bisphosphate 3-kinase, catalytic subunit alpha) are present in many cancer types including colorectal cancer. The prevalence of PIK3CA exon 9 and/or exon 20 hotspot mutations in colorectal cancers is approximately 15-20% in large population-based studies,^60-62 and more variable in clinical trials^63-67 and other studies.^68-82 Studies which used Pyrosequencing assay^60,83,84 (which is more sensitive than Sanger sequencing⁸⁵) generally show higher frequencies of PIK3CA mutations than Sanger sequencing studies.^{61,62,69,80,86} Considering mutations in other less-commonly mutated exons as well as false negativity in molecular assays (particularly, Sanger sequencing), it is estimated that approximately 20% of colorectal cancers in the general population harbor PIK3CA mutations. Interestingly, the prevalence of PIK3CA mutations in colorectal cancer increases continuously from rectum (approximately 10%) to cecum (approximately 25%),^62,87 supporting the “colorectal continuum” paradigm,⁸⁸ and an important interplay of gut microbiota and host response.^39,89-93PIK3CA mutation in colorectal cancer is associated with phosphorylated AKT expression,⁹⁴, phosphorylated RPS6 expression,⁸³ inactive CTNNB1 status,⁸³ VDR (vitamin D receptor) expression,⁹⁵ and KRAS mutations,^60,62,83,96 including codon 61 and 146 mutations.⁵⁰PIK3CA G>A substitutions are associated with loss of MGMT mismatch repair enzyme.^62,83 Associations of PIK3CA mutations with other molecular features such as microsatellite instability (MSI), CpG island methylator phenotype (CIMP), and BRAF mutations are less consistent.^{60-62,65,71,83}PIK3CA amplification has also been reported.⁹⁷PIK3CA mutation may predict resistance to anti-EGFR therapy in stage IV colorectal cancer.^63,86,98-100 A prognostic role of overall PIK3CA mutation status in colorectal cancer remains uncertain;^60,65 however, the presence of coexisting exon 9 and exon 20 mutations may be associated with shorter patient survival,⁶⁰ which is supported by experimental data.¹⁰¹

Colorectal Cancer PIK3CA Mutation Predicts Response to Aspirin

Recently, Liao et al.¹⁰² tested the hypothesis that aspirin might be effective specifically on PIK3CA-mutated colorectal cancer, based on experimental evidence for an interplay of the PI3K (phosphatidylinositol-4,5-bisphosphonate 3-kinase) and PTGS2 (cyclooxygenase-2) pathways.^19,103,104 The study demonstrated that the effect of aspirin use on patient survival appeared significantly stronger in PIK3CA-mutated cancer than in PIK3CA-wild-type cancer.¹⁰² These findings not only highlight the potential cross-talk between the PI3K and PTGS2 pathways, but also suggest PIK3CA mutation in colorectal cancer as a potential biomarker to predict response to aspirin therapy.^102,105,106

Possible Mechanisms of Interaction between Aspirin and Tumor PIK3CA Mutation

The study by Liao et al.¹⁰² demonstrated strong interactive effects of aspirin and tumor PIK3CA mutation in a late phase of tumor progression, while the frequencies of tumor PIK3CA mutations were similar in both tumors that arose in aspirin users compared with non-users prior to diagnosis. These data imply that the interactive effect of aspirin and tumor PIK3CA mutation evolves as a tumor develops and progresses. The data also suggest important roles of exposures (including drug) and the tumor microenvironment in modifying a tumor phenotype (Figure 1). Recent evidence attests to the plasticity of BRAF-mutated melanoma cells which is conferred by secretion of hepatocyte growth factor (HGF) from stromal cells, leading to tumor resistance to targeted RAF inhibitor therapy.¹⁰⁷ In colorectal carcinogenesis, changes in the local tumor microenvironment may also be related to the biogeography of the colon, which could in turn reflect variation in the gut microbiome, an increasingly important area of investigation.

Figure 1.

Outcome of patients with PIK3CA-mutated colorectal cancer appears to differ according to the status of regular aspirin use after cancer diagnosis. The likelihood of long term survival among aspirin users is higher than that among non-aspirin users. The findings suggest that aspirin use may interact with molecular status of tumor to modify a tumor phenotype and clinical behavior.

Accumulating evidence implies a role for gut microbiota and contents in the immune response and carcinogenesis.^89-91 In parallel, a recent provocative study showed that the prevalence of major molecular events in colorectal cancer (including PIK3CA and BRAF mutations, microsatellite instability, and CpG island methylator phenotype) appears to increase gradually along detailed subsites from rectum to ascending colon.⁸⁷ These data challenge the existing dichotomy model of proximal vs. distal colorectum in terms of the molecular features.^87,88 Taken together, it is possible that the effect of aspirin on the local microenvironment and its interaction with tumor PIK3CA mutation may change according to gut microbiota, contents, and biogeography.

An alternative mechanism for the interaction between aspirin and tumor PIK3CA mutation may relate to platelet function and tumor thrombosis. The well-characterized anti-platelet effects of aspirin are evident even with low dose of the drug. Distant tumor metastasis consists of a complex process of tumor cell invasion into stroma and vascular wall, survival in the bloodstream, formation of tumor thrombus, attachment of vascular wall at a metastatic site, and tumor cell invasion of the vascular wall into stroma at the metastatic site. Thus, given the central role of thrombosis in tumor metastasis, it is quite plausible that an influence of aspirin on tumor metastatic potential and patient outcomes may be mediated by its anti-platelet effects, which could in turn differ according to PIK3CA mutation status.

What's Next?

Given the intriguing observation by Liao et al.¹⁰² what should we do next? First, these findings require validation in independent datasets. Although additional observational and interventional studies would be valuable,¹⁰⁸ only a few cohorts have collected detailed long-term data on aspirin use and characterized a substantial number of tumors for PIK3CA mutation. Beyond observational cohort studies, several clinical trials of aspirin as well as the PTGS2 (COX-2) selective inhibitor celecoxib for colorectal cancer patients are underway. Within these studies, the interactive effect of aspirin (or celecoxib) and tumor PIK3CA mutation can be examined. A distinct effect for celecoxib and tumor PIK3CA mutation compared with that for aspirin would support an anti-cancer effect of aspirin mediated through platelets since celecoxib does not appear to significantly inhibit platelets.

Beyond human studies, additional experimental models may be useful to test an interactive effect of aspirin and tumor PIK3CA mutation with the caveat that the local microenvironment in human tumors likely differs from in vitro and animal models.

Thus, for future human correlative studies or experimental systems, it will be increasingly necessary to assess the exposome (a totality of exposures) and interactome (a totality of interactions between various molecules in the local microenvironment). Both molecular pathological epidemiology (MPE) and systems biology¹⁰⁹ approaches consider network perturbations beyond a single exposure – molecular pathway interaction in isolation.¹¹⁰

Molecular Pathological Epidemiology (MPE): Integrative Science Enables the Discovery

The study by Liao et al.¹⁰² represents a prototypical example of “Molecular Pathological Epidemiology (MPE)” research, which has emerged as the transdisciplinary integration of molecular pathology and epidemiology.^43,111,112 The term and concept of MPE have been gaining popularity worldwide.^{13,92,113-134} Conceptually, MPE is defined as “epidemiology of molecular heterogeneity of disease (both intra-individual and inter-individual)”. Hence, MPE differs from conventional epidemiology where a given disease (eg, colon cancer) is regarded as a single entity without explicit consideration of inherent disease heterogeneity.¹³⁵ A similar concept of “etiologic heterogeneity” has also been used.^136,137 While much of MPE data have been derived from neoplastic diseases, the MPE design and paradigm can be applied to research of non-neoplastic conditions.¹¹⁰ Using MPE design, we can examine interactions between influences of exogenous factors (eg, aspirin use and smoking status) and molecular alterations in cancer (eg, PIK3CA mutation and BRAF mutation), which affect cancer cell behavior.^{11,102,105,138-140} Thus, much beyond conventional epidemiology research, MPE not only enables further insights into disease development and progression, but also provides potential disease biomarkers which can predict response or resistance to lifestyle or pharmacological intervention.^43,112 With conventional epidemiology design, the potential predictive value of PIK3CA mutation in colorectal cancer for response to aspirin could not have been uncovered.

Currently, it requires enormous resources, time and effort to build a database of tumor molecular pathology to make MPE research a reality. Comprehensive database is necessary to assess not only tumor markers, but also various other exposures to control for confounding. In analysis of aspirin use, energy balance status such as body mass index and physical activity may influence systemic inflammation status.^141,142 Investigators need to design a study, select populations, follow participants, collect information and biospecimens, and establish an infrastructure for specimen and data management.¹⁴³ For example, the study by Liao et al.¹⁰² utilized databases of longitudinal prospective cohort studies, the Nurses’ Health Study¹⁴⁴ and the Health Professionals Follow-up Study, which started in 1976 and 1986, respectively. Both studies have not only accumulated enormous amounts of information on diet, lifestyle factors, environmental exposures, and personal and family history of various diseases, but also established repositories of tumor tissues and other biospecimens including blood, buccal cells, toenail and urine. Thus, Liao et al.¹⁰² could readily test the intriguing hypothesis of an interactive effect of aspirin and tumor PIK3CA mutation, utilizing this valuable resource. Substantial resources and effort have been devoted to these and other similar large-scale studies, which can help us better understand various diseases and develop effective clinical and public health strategies to decrease disease burden in our societies. Thus, utilization of existing resource can be a very cost effective approach for public health research.¹⁴⁵

The MPE concept takes into account the exposome as well as genome, epigenomes, and interactomes in disease pathogenesis. Integration of MPE and systems biology (which typically utilizes experimental model systems) will facilitate the development of new research areas and clinical practice. In line with this integration of MPE and systems biology, nanotechnologies and biosensors can be utilized for improvement of personalized risk stratification, screening, and early detection of diseases.^35,146,147

Can MPE Approach Be Used Routinely in Epidemiology Research?

Because most epidemiologic studies and public disease databases lack tumor tissue repositories and molecular data, MPE study designs may be regarded as the exception rather than the rule. MPE has caveats; there is inherent multiple hypothesis testing.¹¹² Because multiple subtypes need to be assessed, sample size in MPE should ideally be large; however, a given MPE dataset is usually confined to existing epidemiologic or clinical cohorts and pathology specimen availability.¹¹² It is well known that many tumor molecular features are associated with each other.^25,148-156 Thus, often multiple biomarkers need to be simultaneously assessed. Moreover, to move steps closer to personalized precision medicine, we should be able to address research and clinical questions on less deleterious mutations and mutations in genes which are rarely mutated in a given cancer. All of these factors necessitate large sample sizes in MPE research.

Given the power and promise of MPE research exemplified by Liao et al.,¹⁰² how can we facilitate integrative MPE-type research to advance public health science? One near-term solution is the use of non-molecular tumor subtyping, which can reflect tumor molecular biology. For example, colorectal cancer can be classified as fatal cancer vs. non-fatal cancer, which can reflect biological aggressiveness. However, such tumor subtyping is not an optimal classification method since it only captures the behavior of tumor subtypes only to a limited extent.

Molecular testing is increasingly routine in colorectal cancer. For example, KRAS mutation testing is common for prediction of resistance to anti-EGFR therapy.^55,157,158 MSI and BRAF mutation testing is common for screening of hereditary colorectal cancer.¹⁵⁹ More recently, multi-gene testing utilizing next generation sequencing platform is increasingly common. Therefore, it may be possible that population cancer registries can integrate this routine clinical molecular testing data with standard demographic, clinical, and pathological information. Tumor molecular data accumulated in population disease registries will enable investigators to conduct MPE research without necessarily assuming the substantial expense of collecting tissue materials for analysis of tumor molecular changes. Such a population MPE database will be useful to test basic science findings in a population-based human sample,¹⁶⁰ towards timely research translation into clinical use. Figure 2 illustrates an example of integrative population-based MPE database, which enables us to readily test a multitude of research hypotheses towards potential translation into clinical and public health practice.

Figure 2.

An example of population-based molecular pathological epidemiology (MPE) database. Such a database enables us to test a multitude of research hypotheses, with a potential for translation into public health and clinical practice.

It may be a challenging task to transform population disease registries into “population molecular pathology registries”. On a basic level, we should introduce MPE concepts within public health schools,^125,135,161 so that students can learn not only traditional epidemiologic concepts, but also the molecular pathologic basis of human diseases. This may lead to a new generation of public health practitioners and researchers with a solid understanding of MPE that can lead to effective establishment, development and utilization of population-based molecular pathology registry databases.

As we develop such population molecular pathology registries, additional challenges will emerge. For example, next generation sequencing assays will enable us to sequence an individual's whole exome or genome at a reasonable cost, perhaps as a routine clinical test. Such technology may lead to ready determination of not only the identities of individuals but also the presence of potential known disease-causing mutations that can have profound clinical implications for an individual as well as his/her relatives. How molecular information can or should be accessed in population registries for research remains a significant issue that has yet to be resolved.

Conclusions and Future Perspectives

We are moving more towards increasingly interdisciplinary, integrative science. Epidemiology also needs to transform for the new era of medicine and public health.¹⁶² The importance of team science and an interdisciplinary education system has repeatedly been discussed.^{125,133,135,161,163,164} Currently, the MPE study such as the one by Liao et al.¹⁰² represents a rare unique example of interdisciplinary research for which a replication study cannot easily be performed. However, in the future it may be possible to transform population disease registries into molecular pathology databases, since molecular testing is increasingly routine in clinical practice. Such population molecular pathology registries will enable investigators to perform MPE research as routine practice. In order to achieve this goal, we should consider introducing MPE concepts within the standard population science training. Providing such an education will hopefully lead to the next generation of population scientists who can firmly understand the principles of MPE and can advance integrative science to move us closer to our goal of personalized medicine and health care.

Search Strategy and Selection Criteria.

A comprehensive search for relevant English language publications was performed in Pubmed utilizing Medical Subject Heading (MeSH) terms, and other terms (“epidemiology”, “cancer”, “tumor”, “molecular pathology”, “molecular pathological epidemiology”, “exposure”, “biomarker”, “aspirin”, “NSAIDs”, “anti-inflammatory drug”, “survival”, “response”), in various combinations. The list of references in retrieved articles was assessed for additional relevant articles. A final decision to include or exclude a given article was based on the quality, relevance, and uniqueness of the article.