Complementing genomics and transcriptomics: Phosphoproteomics illuminates systems biology in prostate cancer

ABSTRACT

Integration of phosphoproteomics with traditional genomics and transcriptomics provides a more comprehensive overview of the signaling networks in advanced prostate cancer for immediate preclinical and future clinical use. Our recent publication introduces computational approaches for integrating the phosphoproteome, specifically with the intent of identifying important kinase signaling networks in advanced-stage prostate cancer.

Prostate cancer is the most common male cancer and second leading cause of cancer-related death of men in the United States. Although treatable at early stages, prostate cancer can ultimately progress into a metastatic and lethal state. Successful first-line treatment options range from radiation and surgical removal of the prostate (prostatectomy) to physical (orchiectomy) or chemical (e.g., leuprolide) castration that blocks the androgen/androgen receptor axis that fuels the tumor. Unfortunately, a majority of patients eventually develop resistance to these treatments and the disease progresses into a castration-resistant form, usually concomitant with metastases, termed metastatic castration-resistant prostate cancer (mCRPC). This has led to the development of second-generation anti-androgen drugs such as enzalutamide and abiraterone acetate with greater efficacy. However, this vicious cycle of resistance and treatment continues as the disease develops new ways of evading therapy.¹

Previous studies have largely focused on the mutational and transcriptional landscape of mCRPC to identify new markers of resistance, with mixed results.^2,3 These and other studies have provided evidence that the mutation rates in prostate cancer, and subsequent mCRPC, are relatively low compared to those in other cancers.^2,3 It is therefore imperative to identify novel approaches that will improve our understanding of treatment resistance in advanced prostate cancer in order to better tailor treatment options for patients. One concept is the implementation of phosphoproteomics, which enables identification of pathway activity in lieu of mutational burden.

We previously used immunohistochemical staining to show increased tyrosine phosphorylation in mCRPC patient tumor samples compared to treatment naïve or locally confined prostate cancer, suggestive of heightened kinase activity in these more aggressive tumor types.⁴ We followed up our findings by performing unbiased shotgun phosphopeptide enrichment coupled to quantitative mass spectrometry (known as phosphoproteomics) to aid in the identification of these kinases in mCRPC samples. Based on our previous work and the paucity of mutations in mCRPC, we predicted that integration of the phosphoproteome would complement analyses from genomic and transcriptomic studies to aid in the identification of potential pathway drivers in this disease (Fig. 1). This was indeed suggested by the findings described in our recent publication in Cell.⁵

Figure 1.

Value of multi-omic integration. Focusing on singular approaches such as genomics (A), transcriptomics (B), proteomics (C), or phosphoproteomics (D) reveals context-dependent pathways within a given tumor type. Adopting an integrated, multi-omic analysis (E) can unveil the overall systems biology of the entire tumor and provide new mechanisms for therapeutic intervention.

We procured treatment-naïve prostate cancers and mCRPC tumor samples via an IRB-approved rapid autopsy program from the University of Michigan and utilized phosphoproteomics to generate a compendium of phosphotyrosine (pY) and phosphoserine/phosphothreonine (pS/pT) peptides. To identify differential kinase activity between the sample types we applied the master regulator inference algorithm (MARINa), which infers differences in kinase activity based on differential phosphorylation of its targets.⁶ We also applied MARINa to the patients' transcriptomic data and identified aberrantly activated transcription factor master regulators in mCRPC.

To integrate our phosphoproteomic data with other omic datasets, we used the Tied Diffusion through Interacting Events (TieDIE) algorithm.⁷ This pathway-based method expands upon heat diffusion strategies, such as the HotNet algorithm,⁸ to integrate information from several different sources. We employed TieDIE to integrate our phosphoproteomic data with genomic data, transcriptomic data, and a priori knowledge from pathway databases⁵ and generated a cohort-level scaffold network for mCRPC. Importantly, integration of the phosphoproteomic data enhanced, and in some cases validated, the pathway networks provided by genomic or transcriptional analyses. We observed that the AKT/mTOR/MAPK signaling pathway was significantly enriched in the integrated analysis but only marginally enriched when phosphoproteomic data were excluded. Proteins involved in other cancer hallmarks, including cell cycle pathway, DNA repair pathway, and nuclear receptor pathway, were also enriched when we included the phosphoproteomic input.⁵ These results provide compelling evidence for inclusion of the phosphoproteome in the identification of potential therapeutic targets in metastatic CRPC.

In an attempt at personalized medicine via dataset integration, we used a sample-specific version of MARINa called the Virtual Inference of aberrant Protein activity by Enriched Regulon analysis (VIPER).⁹ We used the VIPER software to analyze individual patient transcriptome and phosphoproteome data and summarized them into transcriptional and kinase master regulators, respectively, which were then applied to a pre-established cohort-level scaffold network to generate personalized integrated network models for each patient. To visualize a patient's network model in a simpler manner, we created the Phosphorylation-based Cancer Hallmarks using Integrated Personalized Signatures (pCHIPS) diagram, which can be accessed online (https://sysbiowiki.soe.ucsc.edu/pchips).⁵ This tool illustrates interpatient variability via differential pathway activation and can also reveal differences in pathway networks pre- and post-treatment. In our study, we used the individual pCHIPS and created a kinase hierarchy to prioritize targetable kinases in the patients that we were able to analyze.

Although our network-based approach may demonstrate future clinical utility, we see our strategy as having a more immediate contribution toward advancing preclinical research in mouse models and cell lines. While the current use of genomics and transcriptomics has identified key signaling networks in prostate cancer,^2,3 we believe that the addition of phosphoproteomics will complement these technologies to increase discovery and action of novel targets in this disease. Integration of phosphoproteomics will allow us to better interpret the systems biology of current preclinical models, thus identifying additional kinases or their substrates to test as targets for therapy. Current mouse models typically originate from genomic studies that identify key driver genes or tumor suppressors of interest.¹⁰ The phosphoproteome would provide additional possibilities that include both the kinases and their downstream substrates as specific players in this disease. Overall, this would allow us to create more specific and representative mouse models of prostate cancer for use in preclinical research.

In summary, we have shown in a panel of mCRPC patients that integration of phosphoproteomic data complements traditional genomics studies and yields a more complete map of the signaling network in this disease. The Cancer Genome Atlas (TCGA) has revealed an enormous amount of genomic information for cancers, including prostate cancer, but there are instances where the genomic data are insufficient to determine the next plan of action; for example, cancers with low mutation frequencies often provide few actionable targets. Phosphoproteomics can help to bridge that gap and aid in the identification of tangible targets for therapy. Ultimately, it is the functional aspects of the signaling networks that are both drivers and targets in cancer, hence the necessity to understand more about these pathways at the protein level. Programs such as the Clinical Proteomic Tumor Analysis Consortium (CPTAC) in collaboration with the Cancer Moonshot have recognized the importance of linking genomic data with functional proteomic data. It will therefore be necessary to refocus our strategy on a more global effort that encompasses analyses from all areas of research. We hope that this new computational approach will provide a foundation to encourage system biologists to develop novel algorithms that can utilize the increase in phosphoproteomics knowledge.

Disclosure of potential conflicts of interest

No potential conflicts of interest were disclosed.

Acknowledgments

We thank all the co-authors from the paper by Drake, Paull et al. (Cell 2016), especially members of the Dr. Owen N. Witte and Dr. Joshua M. Stuart laboratories and the University of Michigan rapid autopsy program. The computational algorithm described in this manuscript was developed by Dr. Evan O. Paull in the laboratory of Dr. Stuart (University of California, Santa Cruz). Due to citation limitations, we were not able to cite all of the important work that has contributed to the identification of new targets in mCRPC.

Funding

LCC is supported by the National Institute of General Medical Sciences of the National Institutes of Health under award number T32 GM008339. JMD is supported by the Department of Defense Prostate Cancer Research Program W81XWH-15-1-0236, Prostate Cancer Foundation Young Investigator Award, and by a grant from the New Jersey Health Foundation.

Author contributions

VMT, LCC, and JMD wrote the manuscript.