Proteomics as a guiding tool for more effective personalized therapy

Abstract

Ovarian cancer remains the deadliest gynecological malignancy in the Western world and is most often diagnosed at a rarely curable late stage. Examination of protein end points has been employed as an investigative mechanism to guide targeted therapy and to stratify ovarian cancer. Proteomics allows characterization of the proteins and the associated protein and peptide modifications. This has given us insight into the perturbations of signaling pathways within tumor cells and has improved the discovery of new drug targets and possible prognostic indicators of outcome and disease response to therapy. Development of validated assays that survey the genetic and/or proteomic make-up of an individual tumor will add greatly to the histological classification of the tumor and may lead to different treatment approaches tailored to the unique expression pattern of each individual patient. It is anticipated that application of proteomics may bring to reality the clinical adoption of molecular stratification, e.g. not, ‘is the gene overexpressed?’, but ‘is the pathway upregulated?’ This will be successful if validated peptide biomarkers are applied for patient selection prospectively and with inclusion of preplanned biological correlates. These events will guide future directions of proteomics as a selector and as a validator and will guide how we integrate proteomics information daily into patient care and into selecting therapy of advanced and recurrent ovarian cancer and other cancers.

introduction

Ovarian cancer remains the leading cause of death from gynecological malignancy among US women, with a lifetime probability of developing the disease of 1 in 59. Most women will be diagnosed with advanced stage disease. With current treatment modalities, the 5-year survival rate is 30%–40% with advanced stage compared with 90%–95% with organ-confined disease [1, 2]_. A key need is to understand the molecular and cellular microenvironmental events driving ovarian cancer behavior in a fashion that will allow individualization of therapeutic interventions. Progress is being made in this direction. Genomic studies now indicate that there may be two subtypes of epithelial ovarian cancer (EOC), one subtype readily identified by genomic means as the BRCA-like high-grade serous cancer [3]. This leaves the majority of ovarian cancers consisting of grade 1 and 2 serous, non-serous and mixed histology. Protein end points are being incorporated into these models to build our knowledge and to allow development of mechanisms to guide targeted therapy and substratify this group.

Proteomics allows characterization of the proteins and the associated protein and peptide modifications that make up the complex signaling networks that mediate cellular activity. Proteomics has given us insight into the perturbations of signaling pathways within tumor cells and has facilitated the discovery and characterization of new drug targets. Proteomic analysis of tissue samples has also allowed us to discover possible early markers for the presence of new disease and for the progression of established cancer through the course of treatment [4]. Mass spectrometry (MS) has been employed to evaluate differential expression patterns of proteins and peptides in patient serum or other biospecimens [5, 6]. Matrix-assisted laser desorption and ionization with time-of-flight (MALDI-TOF) and surface-enhanced laser desorption and ionization with time-of-flight (SELDI-TOF) have been the most common methods used [7]. More recently, reverse phase protein assay (RPPA) has been used to examine tumor protein lysates in an array format analyzed with standardized antibodies specific for total and activated protein targets [8, 9]. This allows examination of the degree of activation of the pathway as well as the amount of the target present. Serial investigative samples can be compared side by side for response to treatment or target validation and inhibition in clinical trials. Ultimately, one of our biggest challenges may be managing the flood of information. Critical clinical and scientific questions are addressing how we integrate proteomics information into patient care and into developments in therapy of advanced and recurrent ovarian and other cancers. While progress has been made with high-throughput and single/sets of proteins for ovarian cancer diagnostics, we have yet to make progress in the identification of proteomic or other biomarkers that successfully guide therapy.

has proteomic profiling led to identification of predictive biomarkers for ovarian cancer?

A predictive biomarker is a biomarker that forecasts the differential benefit of a particular intervention based on marker status. For example, in advanced colorectal cancer, the benefit of cetuximab appears to be limited to patients with tumors that have the wild-type K-ras genotype [10]. Using K-ras genotype status as a predictive biomarker could, therefore, guide the choice of therapy, stratifying application of cetuximab, targeting the upstream epidermal growth factor receptor (EGFR) tyrosine kinase. A prognostic biomarker allows identification of patients with differing risks of a specific outcome, such as progression or death. Thus, a prognostic marker can distinguish populations into subgroups where different treatment options are appropriate (possibly including no treatment if that is standard), but it cannot guide the choice of a particular therapy [11, 12].

Screening of validated predictive biomarkers, e.g. hormone receptor status and HER2 amplification in breast cancer [13], K-ras genotype in colon cancer and EGFR mutation status in lung cancer [14], have guided clinical decisions regarding both targeted and traditional cytotoxic chemotherapy. These biomarkers were identified retrospectively during examination of tumors in order to understand why and how the cancers did or did not respond to therapy. None were identified prospectively in a discovery setting to aid in guiding therapeutics. Resources, both genomic and proteomic, are now accumulating in ovarian cancer to allow us to apply that type of analysis. A probable differential genomic signature of type I and II ovarian cancers comes from the combination of genomic analyses and the recognized differential patterns of histology and outcome of epithelial ovarian cancer subsets. Further work is needed to validate these and to identify protein biomarkers with which to stratify type I and II patients.

Application of proteomic technologies to the management of ovarian cancer is both necessary for and dependent upon the discovery and validation of reliable biomarkers. As new therapeutic options develop, it is desirable to use our increasing knowledge of tumor molecular biology to discover and validate predictive biomarkers. But, simply assaying available specimens is an inadequate study design. Although such a strategy may result in some biological insight into the behavior of the cancer, it is not sufficient for development of a clinical utility. It is essential to identify and validate biomarkers, currently most likely successful through retrospective analyses such as were successful in the other cancers, then to design, conduct and analyze biomarker-stratified prospective clinical trials.

Proteomic profiling has been valuable for discovery and may be applied for identification of predictive biomarkers. Most reported studies are of limited power and great data volume, and most lack confirmation. Focused attention to evaluation and validation will more rapidly advance our ability to use end point results with confidence. Data mining of the newly developing proteomic databases is just beginning and where generated from pre- and post-treatment tissues, will be the most rewarding.

proteomic profiling as a guiding tool in making clinical decisions

Two types of proteomic profile have been described and there are two predominant clinical applications to consider. The first, like genomics signatures, consists of defined proteins and protein states, such as activation of AKT, mTOR and inhibition of bad, constituting an AKT pathway activation signature. The second consists of a multi-analyte discriminator derived from complex measurements and algorithms, a ‘black box’ approach, which is more difficult to confirm and frequently does not provide sequence-specific output. These proteomic profiles offer opportunities to identify new disease biomarkers that can be used in the clinical setting to distinguish populations into subgroups where different treatment options are appropriate and for stratification of patients for specific treatment.

Both prognostic and predictive applications have been introduced for multiple cancers. A study in patients with non-small-cell lung cancer (NSCLC) demonstrated an important stepping-stone in proteomics research, the ability to risk-stratify patients based on a proteomic pattern. Yanagisawa et al. [15] showed not only that their training set-derived multiplex pattern could accurately diagnose and classify their blinded independent validation set, but that it also demonstrated a proteomic pattern that could distinguish between patients with resected NSCLC who had poor prognosis and those who had good prognosis. Clinical trials are underway at multiple institutions analyzing whether using these proteomics data to make clinical decisions is beneficial in NSCLC patients [16]. Routine screening of breast cancer tumor samples for hormone receptor and HER2 status has proved that the ‘right’ proteomic factor can aid in decisions regarding both targeted and traditional cytotoxic chemotherapy. Recent interim analysis of the Breast Cancer International Research Group (BCIRG) 006 study demonstrated that addition of trastuzumab to adjuvant chemotherapy was associated with significant disease-free survival (DFS) and overall survival (OS) benefit in women with HER2-overexpressing breast cancer [17].

It is anticipated that application of similar technologies may guide clinical decisions for ovarian cancer. Köbel et al. [18] have evaluated expressed proteins of 21 candidate genes by tissue microarray immunohistochemical (IHC) analysis. Included are CA125, CRABP-II, EpCAM, ER, F-Spondin, HE4, IGF2, K-cadherin, Ki-67, KISS1, matriptase, mesothelin, MIF, MMP7, p21, p53, PAX8, PR, SLPI, TROP2 and WT1, all reported in at least one study to have potential prognostic utility in ovarian cancer. Survival analyses show that 9 of the 21 biomarkers are prognostic indicators in the entire cohort but when analyzed by subtype only, three are prognostic indicators in the high-grade serous cancers and none in the clear-cell subtype. Ki-67 staining varied markedly between different subtypes and is an unfavorable prognostic marker in the entire cohort, but is not of prognostic significance within any subtype. WT1 is more frequently expressed in high-grade serous carcinomas, an aggressive subtype. It is an unfavorable prognostic marker within the entire cohort of ovarian carcinomas, but is a favorable prognostic marker within the high-grade serous subtype. Differentially expressed proteomic profiles across histological subtypes indicate that distinct biochemical events are more associated with subtype than with stage. This may explain why these patients have different responses to targeted and chemotherapies and different survival outcomes, and should guide our approaches in the future. Unfortunately, few properly powered and validated biomarkers or biomarker signatures in EOC have been identified to date. Further studies are needed to validate these findings and to determine whether these genetic and proteomic differences require different treatment approaches.

Before proteomic profiling can be introduced into the clinical setting at a broader level, standardization of the pre-analytical phase is needed, including patient preparation, sample collection, sample preparation, sample storage, measurement and data analysis. These advances relate to pre-analytical factors, analytical standardization and quality-control measures required for effective implementation into routine laboratory testing in order to generate clinically reliable and useful information. It will be crucial to design and perform prospective clinical studies that can identify novel therapeutic strategies based on these techniques, and to validate their prognostic and predictive impact on clinical decision-making.

proteomic profiling as a selector and as a validator

The relationship between prognostic or predictive marker expression and outcome should be explored by prospective study for validation before its application to clinical use, such as population stratification or targeted therapy selection. Retrospective studies of the marker(s) in available tissues of patients with known outcome who have been treated similarly have been conducted in many tumors. A marker that is independently associated with outcome after adjusting for such factors is considered promising for clinical application. This approach requires proper design, power and validation. Retrospective analyses may be fraught with suboptimal power to discern an effect because it is unlikely that tumor specimens will be available for all enrolled patients thus introducing greater possibility for bias in the available sample subset (large tumor, more tissue available; more aggressive center, more tissue available, etc.). Quality control of paraffin-embedded tissue may not always be consistent. However, if the marker-based analyses were planned prospectively pre- and post-treatment, and tissue was available from all or most patients, adequate statistical power may remain with which to compare the clinical outcome against the biomarker. The benefit of this approach is the ability to obtain new knowledge with increasing likelihood of success. Information regarding the prognostic properties of the putative marker relative to existing (standard) treatment, as well as the predictive effect of the marker relative to the new (targeted) treatment, is necessary to propose specific, testable hypotheses to allow optimal design of a validation trial. Proteomic analysis in studies may also help to drive development of new targeted therapies. The proteomics data in aggregate will allow dissection of numerous pathways and guide key investigators as to where to look for pathway activation by genetic, genomic and epigenetic events.

proteomic profiling for targeted therapy of ovarian cancer

Targeted therapy implies a therapy against a specific molecular event(s). This pleads the question that with targeted therapy; do we mean that the therapy has only one target? Some of our most active molecular therapeutics (e.g. imatinib) have more than one confirmed molecular target. And, in some cases, an initially ‘off-target’ effect is later reclassified as biologically pertinent, such as inhibition of vascular endothelial growth factor receptor 2 (VEGFR2) with sorafenib. Caution is needed in claiming that targeted therapy requires an exacting degree of specificity; this can mislead when selective stratification for a single biomarker is used without strong prior validation. The simple equation, ‘drug + molecular target = targeted therapy’ underestimates the complexity of molecular therapeutics. At best, it can be considered necessary but not sufficient [19].

Before we discover and characterize molecular pathways for a targeted therapy, critical criteria should be considered: (i) should we use target-associated biomarkers to select patients for therapy? or, (ii) should we select and validate biomarkers from targeted agent trials using translational output and patient outcome? The first question requires validation of biomarkers to the target(s) and only the outcome of that target modulation matters, thus allowing for patient stratification before therapy. Molecular events leading to a dominant biological effect is an ideal molecular therapeutic target that can be applied as a population selection biomarker after validation. Again, this is well illustrated with HER2-overexpressing breast cancer, selectively successful with trastuzumab treatment. Many effective targeted agents have validated and efficient response prediction biomarkers. The second question requires identification of potential biomarkers to be examined and altered in the context of molecular therapeutic intervention. This requires scientific examination in the context of the trial to define and validate biomarkers for later clinical application. An example of this is advanced colorectal cancer where the stratification of benefit of cetuximab as a function of EGFR downstream pathway activation was studied retrospectively. Here, mutational activation of K-ras was shown to be a negative predictive biomarker for outcome [20, 21].

We can evaluate end point analyses from translational clinical trial measures to evaluate the potential for those end points to inform subsequent studies. For example, proteomic findings of two negative clinical trials both indicate regulation of activation of downstream end points and were hypothesis generating as to why the studies were not successful clinically. Posadas et al. [22] examined EGFR and downstream pathway end points in a phase II clinical trial of gefitinib. They examined by RPPA protein expression of total and phosphorylated (p) EGFR, AKT and extracellular regulated kinase (ERK) in epithelial ovarian cancer tumor samples obtained before and at 1 month of daily therapy. A decrease in the quantity of both EGFR and p-EGFR in tumor tissue was observed with gefitinib therapy in >50% of patients, although this was not associated with clinical benefit. This lack of association between the biochemical and clinical effects of gefitinib indicates that the EGFR downstream pathway may be important, but inhibiting the upstream event of phosphorylation of EGFR in EOC is not sufficient for clinical activity. This provides proof of target in a clinical setting, albeit without activity, and indicates that combination therapy with molecular therapeutics against complementary targets may prove successful. c-Kit and platelet-derived growth factor receptor (PDGFR) have been studied as potential molecular targets in EOC by several groups [23]. A proteomics end point analysis in a phase II trial of imatinib identified a statistically significant trend between pretreatment p-Kit levels in both microdissected tumor and stroma with gastrointestinal toxicity, between tumor EGFR and PDGFR with grade of fatigue, and EGFR and p-AKT levels with ascites and edema (P ≤ 0.01). However, the on-target activity was not associated with clinical activity as a single agent. Both negative studies indicate insufficiency of inhibition of these single receptor tyrosine kinases to impact downstream signaling cascades in EOC, with the possibility that there is promiscuity of the agent in vivo in an unrecognized fashion that alters the signaling balance, and that the target(s) of imatinib or gefitinib are neither necessary nor sufficient to alter the natural history of ovarian cancer. The consistent finding of modulation of ERK with the different targeted agents implies that activation and modulation of ERK may be a measurable biomarker common to many kinase inhibitors. One might stratify exposure to kinase inhibitors by high p-ERK and measure changes to validate that finding then prospectively stratify by ERK activation. These studies hint at a promise that proteomic markers may be identified and used to guide treatment decisions in ovarian cancer.

There are numerous reports in the EOC literature documenting expression of a marker or mutant protein, albeit in a limited proportion of clinical cases. To date, few if any of these have been defined as high-frequency, cancer-driving molecular events for focused molecular therapeutics. Genetic abnormalities in BRCA1/2 are the best-validated genetic molecular target in ovarian cancer. Lesnock et al. [24] recently reported the prognostic association of BRCA1 expression by IHC analysis in sporadic EOC. Reduced BRCA1 expression in optimally resected stage III EOC patients treated with intraperitoneal cisplatin and paclitacxel was associated with improved survival compared with high BRCA1 expression (84 versus 58 months, hazard ratio 0.69, 95% confidence interval 0.474–1.00; P = 0.05). These findings require validation but are compelling because of the availability of poly (ADP-ribose) polymerase (PARP) inhibitors. The advent of the PARP inhibitor (PARPi) class of agents and the demonstration of their possible selective benefit in BRCA1/2 mutation carriers [25] is an example of a completed molecular target–molecular therapeutic circle (Fig. 1). Clinical activity of PARPi in cancers associated with BRCA1/2 mutation demonstrates great promise in the selective use of this class of inhibitors in patients with defective DNA repair pathways. New approaches including measurement of PARP1 protein and proteomic analysis of BRCA1/2 expression in tissue are being evaluated as potential treatment selection tools. Assessing quantity of proteins, such as the presence or attenuation of BRCA1/2, or functionality of proteins, such as reduction in activation of ERK, would move beyond assessment of simple expression and take into consideration the genetic, epigenetic and proteomic derangements that drive cancer progression. The development and validation of biomarkers of activity, toxicity, drug exposure is also necessary. Proteomic analysis of clinical samples obtained during well-designed clinical trials using novel agents, such as the PARPi, will provide resources with which to address the complexity of the signaling events associated with advanced ovarian cancer and hone down on selected signals for examination.

Fig. 1.

Molecular target–molecular therapeutic circle.

Recent technologies employing protein microarrays such as RPPA has allowed quantification of multiple end points in a high-throughput fashion. These include expression levels of key proteins and their activated forms that compose critical signaling nodes involved in proliferation, survival and angiogenesis. For example, the phosphatidylinositol 3'-kinase (PI3K) pathway has been shown to be a driving pathway in serous ovarian cancer in vitro, due to a common gain-of-function mutation on chromosome 3p causing mutational activation of the p110 kinase subunit of PI3K. This propagates activation of the Akt pathway, yielding a strong survival signal [26]. Agents against PI3K itself and its downstream effectors Akt and mTOR are now in early clinical investigation in EOC. Downstream of these important targets are mitogen-activated protein kinase, MEK, and its effector, ERK. They can be modulated directly by small molecule MEK inhibitors such as AZD6244. In addition to being indirect effectors of the PI3K pathway, they are directly downstream of the Src/Ras/Raf pathway. The moderate frequency of K-ras mutations in mucinous EOC and low-grade serous cancers (13%–33%) led to the ongoing phase II GOG study of AZD6244 that recently entered its second stage of accrual.

Targeted therapy may be effective for only the subset of patients who harbor tumors with susceptible and specific protein network defects. Although targeted cancer therapy has been directed at a single molecular target in many cases, in the future we can imagine targeting an entire set of interconnected (kinase-driven) events along a deranged signaling pathway with a ‘smart’ drug selection approach. Proteomics will help identify the optimal targeted agent and effective dose for each patient's disease, which will allow the monitoring of response and relapse, and engineering of new drugs and strategies to circumvent resistance mechanisms.

future directions

Comprehensive systems biology proteomic approaches applied to ovarian cancer and other cancers will both help the patient and tumor community and create a comprehensive database of information pairing the genomic change with its proteomic expression and agents that may successfully target that proteomic drive. Organized studies are needed in order to initiate this direction. Such is the biomarker-strategy based prospective trial design conducted by Dr G. Mills, applied to numerous cancers, including ovarian cancer. ‘T9’ (Ten Thousand Tumors Ten Thousand Tests Ten Thousand Therapies) project will genetically test the tumors of 10 000 patients with relapsed refractory cancer for whom there remains no therapeutic standard of care and who are at high risk of recurrence within 18 months. The results will be applied towards identification of the best treatment considerations. This project was designed prospectively to develop an atlas of mutation and co-mutation in patients entering clinical trials and to develop cohorts of patients with rare aberrations for clinical trials. The results will help identify association of mutation and co-mutation with outcomes and molecular evolution in metastasis and with treatment.

This example project sits on the backbone of genomic change, the findings to be applied clinically are those that alter the protein target and the protein signaling pathway, and involve a final step of proteomic validation. It is also important to complete cancer genome projects to identify directions from which to optimize therapies for each tumor type. This has many difficulties such as the high cost of recruitment of a large number of patients and capture and analysis of a larger number of tissue samples, need for compliance and a skilled laboratory.

Comprehensive proteomic profiling and trial-focused end point profiling will be critical for selection of patient-tailored molecular therapy. Proteomics will help dissect these protein signaling pathways to define the preferred targets of molecular therapy. Emerging proteomic technology can be applied to help select patients who may be more likely to benefit from select targeted therapies and will bring to reality the clinical adoption of molecular stratification. Incorporating translational end points and preplanned biological correlates in a prospective validation study, remains an important method to guide future direction of proteomics as a selector and as a validator. This will guide how we integrate proteomics information daily into patient care and into selecting therapy of advanced and recurrent ovarian cancer and other cancers. Proteomic profiling for ovarian cancer shows tremendous promise in addressing the goals of predictive biomarkers for a more effective therapy. The discovery of novel, validated biomarker signatures will broaden our understanding of the disease and will further define the driving forces behind this heterogeneous condition.

disclosures

Dr Kohn has reported no conflicts of interest.