Final Report Summary - PC INTERACTOME (Assembly of a prostate cancer genome-wide molecular interactome for the identification of the key regulatory genes of malignant transformation and new targets for therapeutic intervention)
Description of the work carried out to achieve the project's objectives, main results and conclusions
The proposed project number 253290 with acronym “PC interactome” aimed at characterizing fundamental mechanisms and signaling pathways deregulated in prostate cancer progression through the assembly of a prostate cancer interactome (PCi), which is an accurate genome-wide network of transcriptional and post-translational molecular interactions. Moreover, interrogation of this interactome should allow the identification of novel druggable targets for prostate cancer therapy. The general aim of this project was thus to exploit the prostate cancer interactome to elucidate the key regulatory genes driving prostate carcinogenesis, in particular late-stage aggresive tumors.
The mouse prostate cancer interactome was generated during the outgoing phase of the Project at the laboratory of Dr. Cory Abate-Shen at Columbia University, New York. We used genetically engineered models of prostate cancer from our laboratory as well as those from the laboratories of our colleagues within and outside the Mouse Models of Human Cancer Consortium (NCI). As discussed in the grant, the production of accurate interactomes is predicated on the availability of large Gene Expression Profiling (GEP) datasets (~300), representative of natural phenotypic variability, which can be achieved by genetic diversity and/or pharmacological perturbations. Thus, the generation of the mouse interactome required an ambitious undertaking. In particular, to generate both genetic and pharmacologic diversity, we obtained 14 different GEM models that cumulatively represent the spectrum of prostate cancer phenotypes and encompass a range of differing cancer initiating/progression events. As for the mice in Dr. Abate-Shen lab, as planned, we have included models that are dependent on the specific loss of function of Pten (Ptenflox/flox) or p53 (p53flox/flox) and the gain of function of K-ras (K-rasLSL-G12D/+) or B-raf (B-rafLSL-V600E/+) in the prostate epithelium. The rest of the mice have been generously provided by colleagues in the field (e.g. Drs. Terry Van Dyke, Barbara Foster, Charles Sawyers, Bart Williams). For pharmacological diversity, each models was treated with various pharmacological agents, including those affecting steroid hormone signaling (e.g. testosterone), PI3 kinase signaling (e.g. rapamycin), tyrosine kinase inhibitors (e.g. Imantinib) and other relevant pathways. Using this approach, we obtained mouse prostate tissues having 196 independent perturbations (i.e. 14 mouse models each treated with different 14 agents), which including experimental duplicates makes up to 392 gene expression profiles. The interactome was then constructed using the ARACNe algorithm, developed in Dr. Califano's lab. Once assembled, the mouse prostate cancer interactome network includes 19.321 genes expressed in the mouse prostate of which 1.707 are transcription factors. In order to validate the accuracy of the predicted interactions we examined published ChIP-seq and ChIP datasets for a variety of transcription factors and computed the significance on the overlap between the targets of those TF and their targets as predicted in our interactome. The results showed highly significant overlap for all TFs assessed, including BCL6, MYC, AR and ER.
The next step was to generate clinically meaningful gene expression signatures to interrogate the mouse interactome to identify drivers of the phenotype transition. To this aim, I relied on a mouse model of metastatic prostate cancer that has also been characterized by the fellow. Specifically, the Nkx3.1CreERT2/+; Ptenflox/flox; KrasLSL-G12D/+; R26-YFP mouse model, that develops adenocarcinoma of the prostate as early as 2 weeks after tumor induction upon tamoxifen administration to activate the CreERT2 fusion protein. These mice show a fully penetrant metastatic phenotype with disseminated cells been visible by means the YFP lineage tracing allele as soon as 15 days after induction in the lymph nodes. Hematogenous dissemination of isolated single cells in the lung or liver occurs at 1 month after induction while by 3 months those tumors are able to form overt metastasis. The results of the mouse model characterization were recently published in PNAS (Aytes, A., et al. PNAS, 2013). By comparing these metastatic primary tumors to the non-metastatic tumors of the Nkx3.1CreERT2/+; Ptenflox/flox; R26-YFP, and by using the MARINa algorithm (published by the Califano laboratory), we identified a list of candidate master regulators of the non-metastatic to metastatic transition. Cox regression models were used to stratify patients based on their risk to biochemical recurrent using published datasets. We found that FOXM1 and CENPF synergistically predict disease outcome in this datasets. The expression levels for FOXM1 and CENPF were further investigated in silico, using publicly available datasets on Oncomine, and experimentally, using patients prostate cancer specimens on tissue microarrays (TMA). We discovered that both FOXM1 and CENPF are upregulated in metastasis compared to primary tumors and that at the protein level, they are overexpressed in tumors with high Gleason scores (7 to 9) compared to low Gleason scores (4-6). Additionally, we performed functional validation studies to demonstrate the addiction that prostate cancer cells have to the synergistic activation of these two transcription factors. To this end, human prostate cancer cell lines were engineered so that we can induce silencing of FOXM1 and CENPF. The experiment was designed so that silencing of each gene resulted in expression of either an RFP or GFP reporter allele to allow investigating the phenotype of the surviving tumor cells. We were able to demonstrate that although silencing of either gene resulted in tumor growth reduction, silencing of both genes nearly completely abrogated tumor formation. Moreover, in an additional competition assay in vivo we showed that the surviving cells after gene silencing were only those that had undergone silencing of either one gene, but not the ones in which both genes were knocked down.
Next, we investigated whether expression of FOXM1 and/or CENPF proteins are associated with prostate cancer progression and/or are prognostic for disease outcome using independent cohorts on tissue microarrays. Our studies revealed patients whose prostate tumors have elevated expression of both FOXM1 and CENPF are associated with the worst outcome for all three endpoints, corresponding to time to biochemical-free recurrence, death due to prostate cancer, and time to metastasis. In contrast, elevated expression of either FOXM1 or CENPF individually was either not significant or only marginally significant for biochemical recurrence and prostate-specific survival, and was 10 to 13 orders of magnitude less significant than co-expression for time to metastasis.
Finally, we sought to elucidate the molecular mechanisms that lead to the activation of FOXM1 or CENPF transcriptional programs. We silenced FOXM1 and CENPF either alone or in combination in human prostate cancer cell lines an investigated the signaling pathways that were altered. Interestingly, silencing of either gene alone led to the identified of biological pathways that are consistent with their known function as regulators of cellular proliferation, reduction of stress response and/or regulation of mitosis. However, co-silencing of both genes However, co-silencing of FOXM1 and CENPF revealed a new repertoire of significantly differentially expressed genes and enriched biological pathways that had not been identified following their individual silencing. In particular, co-silencing of FOXM1 and CENPF revealed the enrichment of biological pathways associated with signaling pathways relevant to tumorigenesis, including: “PI3-Akt signaling”, “MAP kinase pathway”, “signaling by NGF”, “alanine, aspartate and glutamate metabolism”, “regulation of insulin-like growth factor”. Of particular interest was the enrichment of PI3-kinase and MAP kinase signaling and related pathways following co-silencing, but not individual silencing, of FOXM1 and CENPF, as these constitute established hallmarks of aggressive prostate cancer. Indeed, among genes that most contributed to pathway enrichment were key regulators of MAP kinase signaling (LAMTOR1, HRAS, RAC1, MAPKAPK5, and MAPK4) and PI3 kinase signaling (RHOA, RPTOR, PIK33R2, TSC2, AKT, PI3KCA, PI3KCB and RICTOR).
Socio-economic impact of the project
Patients with advanced disease often undergo androgen deprivation therapy, which initially leads to tumor regression but typically results in recurrence of castration-resistant prostate cancer that is highly metastatic. Appearance of metastatic disease is almost always a harbinger of eventual cancer mortality. Therefore, given the complexity of the metastatic process together with the low mutational landscape that characterize prostate cancer, we hypothesized that the dissection of the molecular events responsible for disease progression may inform subsequent preclinical and clinical studies against new therapeutic targets.
Our results open new perspectives in the management of patients with a diagnosis of localized prostate cancer. Further investigation to confirm the association of FOXM1 and CENPF with patients’ outcome
We anticipate that investigating these processes is of paramount importance for several reasons: (i) first, because there are virtually no molecular biomarkers for patients with an early diagnosis of localized high-grade prostate cancer that can predict the occurrence of distant metastasis; (ii) second, because even for patients with metastatic disease, there very few biomarkers of progression after treatment beyond prostate specific antigen (PSA) serum levels, and (iii) third, because despite recent advances in chemotherapy and androgen signaling blocking have shown to be effective at improving survival of metastatic prostate cancer patients, there is yet no available treatment that is curative.
The expected final results or the proposed project can be divided in three as corresponding to the different aims. Firstly, the mouse prostate cancer interactome is a very valuable resource for the scientific community. We expect our results to be a proof of concept that this regulatory network can be interrogated with signatures of different biological contexts to predict drivers of the observed phenotypes. In our case we have been able to successfully identify and validate master regulators of disease progression but one can foresee applications in contexts like castration resistance, tumor initiation or drug resistance. Secondly, by identifying FoxM1 and Cenpf as master regulators of disease progression, we have provided new biomarkers that can potentially be translated into the human clinical practice as tools to help predicting risk in patients with low-grade tumors. Thirdly, we expect that the completion of the objectives for Aim 3 of the current proposal, which will be pursued during the last year, will lead to the identification of new druggable targets for therapeutic intervention. We speculate those targets to be upstream modulators of the transcriptional activity of FoxM1 and Cenpf so that we will be able to investigate the mechanisms of action of candidate drugs and the potential mechanisms of resistance.
The proposed project number 253290 with acronym “PC interactome” aimed at characterizing fundamental mechanisms and signaling pathways deregulated in prostate cancer progression through the assembly of a prostate cancer interactome (PCi), which is an accurate genome-wide network of transcriptional and post-translational molecular interactions. Moreover, interrogation of this interactome should allow the identification of novel druggable targets for prostate cancer therapy. The general aim of this project was thus to exploit the prostate cancer interactome to elucidate the key regulatory genes driving prostate carcinogenesis, in particular late-stage aggresive tumors.
The mouse prostate cancer interactome was generated during the outgoing phase of the Project at the laboratory of Dr. Cory Abate-Shen at Columbia University, New York. We used genetically engineered models of prostate cancer from our laboratory as well as those from the laboratories of our colleagues within and outside the Mouse Models of Human Cancer Consortium (NCI). As discussed in the grant, the production of accurate interactomes is predicated on the availability of large Gene Expression Profiling (GEP) datasets (~300), representative of natural phenotypic variability, which can be achieved by genetic diversity and/or pharmacological perturbations. Thus, the generation of the mouse interactome required an ambitious undertaking. In particular, to generate both genetic and pharmacologic diversity, we obtained 14 different GEM models that cumulatively represent the spectrum of prostate cancer phenotypes and encompass a range of differing cancer initiating/progression events. As for the mice in Dr. Abate-Shen lab, as planned, we have included models that are dependent on the specific loss of function of Pten (Ptenflox/flox) or p53 (p53flox/flox) and the gain of function of K-ras (K-rasLSL-G12D/+) or B-raf (B-rafLSL-V600E/+) in the prostate epithelium. The rest of the mice have been generously provided by colleagues in the field (e.g. Drs. Terry Van Dyke, Barbara Foster, Charles Sawyers, Bart Williams). For pharmacological diversity, each models was treated with various pharmacological agents, including those affecting steroid hormone signaling (e.g. testosterone), PI3 kinase signaling (e.g. rapamycin), tyrosine kinase inhibitors (e.g. Imantinib) and other relevant pathways. Using this approach, we obtained mouse prostate tissues having 196 independent perturbations (i.e. 14 mouse models each treated with different 14 agents), which including experimental duplicates makes up to 392 gene expression profiles. The interactome was then constructed using the ARACNe algorithm, developed in Dr. Califano's lab. Once assembled, the mouse prostate cancer interactome network includes 19.321 genes expressed in the mouse prostate of which 1.707 are transcription factors. In order to validate the accuracy of the predicted interactions we examined published ChIP-seq and ChIP datasets for a variety of transcription factors and computed the significance on the overlap between the targets of those TF and their targets as predicted in our interactome. The results showed highly significant overlap for all TFs assessed, including BCL6, MYC, AR and ER.
The next step was to generate clinically meaningful gene expression signatures to interrogate the mouse interactome to identify drivers of the phenotype transition. To this aim, I relied on a mouse model of metastatic prostate cancer that has also been characterized by the fellow. Specifically, the Nkx3.1CreERT2/+; Ptenflox/flox; KrasLSL-G12D/+; R26-YFP mouse model, that develops adenocarcinoma of the prostate as early as 2 weeks after tumor induction upon tamoxifen administration to activate the CreERT2 fusion protein. These mice show a fully penetrant metastatic phenotype with disseminated cells been visible by means the YFP lineage tracing allele as soon as 15 days after induction in the lymph nodes. Hematogenous dissemination of isolated single cells in the lung or liver occurs at 1 month after induction while by 3 months those tumors are able to form overt metastasis. The results of the mouse model characterization were recently published in PNAS (Aytes, A., et al. PNAS, 2013). By comparing these metastatic primary tumors to the non-metastatic tumors of the Nkx3.1CreERT2/+; Ptenflox/flox; R26-YFP, and by using the MARINa algorithm (published by the Califano laboratory), we identified a list of candidate master regulators of the non-metastatic to metastatic transition. Cox regression models were used to stratify patients based on their risk to biochemical recurrent using published datasets. We found that FOXM1 and CENPF synergistically predict disease outcome in this datasets. The expression levels for FOXM1 and CENPF were further investigated in silico, using publicly available datasets on Oncomine, and experimentally, using patients prostate cancer specimens on tissue microarrays (TMA). We discovered that both FOXM1 and CENPF are upregulated in metastasis compared to primary tumors and that at the protein level, they are overexpressed in tumors with high Gleason scores (7 to 9) compared to low Gleason scores (4-6). Additionally, we performed functional validation studies to demonstrate the addiction that prostate cancer cells have to the synergistic activation of these two transcription factors. To this end, human prostate cancer cell lines were engineered so that we can induce silencing of FOXM1 and CENPF. The experiment was designed so that silencing of each gene resulted in expression of either an RFP or GFP reporter allele to allow investigating the phenotype of the surviving tumor cells. We were able to demonstrate that although silencing of either gene resulted in tumor growth reduction, silencing of both genes nearly completely abrogated tumor formation. Moreover, in an additional competition assay in vivo we showed that the surviving cells after gene silencing were only those that had undergone silencing of either one gene, but not the ones in which both genes were knocked down.
Next, we investigated whether expression of FOXM1 and/or CENPF proteins are associated with prostate cancer progression and/or are prognostic for disease outcome using independent cohorts on tissue microarrays. Our studies revealed patients whose prostate tumors have elevated expression of both FOXM1 and CENPF are associated with the worst outcome for all three endpoints, corresponding to time to biochemical-free recurrence, death due to prostate cancer, and time to metastasis. In contrast, elevated expression of either FOXM1 or CENPF individually was either not significant or only marginally significant for biochemical recurrence and prostate-specific survival, and was 10 to 13 orders of magnitude less significant than co-expression for time to metastasis.
Finally, we sought to elucidate the molecular mechanisms that lead to the activation of FOXM1 or CENPF transcriptional programs. We silenced FOXM1 and CENPF either alone or in combination in human prostate cancer cell lines an investigated the signaling pathways that were altered. Interestingly, silencing of either gene alone led to the identified of biological pathways that are consistent with their known function as regulators of cellular proliferation, reduction of stress response and/or regulation of mitosis. However, co-silencing of both genes However, co-silencing of FOXM1 and CENPF revealed a new repertoire of significantly differentially expressed genes and enriched biological pathways that had not been identified following their individual silencing. In particular, co-silencing of FOXM1 and CENPF revealed the enrichment of biological pathways associated with signaling pathways relevant to tumorigenesis, including: “PI3-Akt signaling”, “MAP kinase pathway”, “signaling by NGF”, “alanine, aspartate and glutamate metabolism”, “regulation of insulin-like growth factor”. Of particular interest was the enrichment of PI3-kinase and MAP kinase signaling and related pathways following co-silencing, but not individual silencing, of FOXM1 and CENPF, as these constitute established hallmarks of aggressive prostate cancer. Indeed, among genes that most contributed to pathway enrichment were key regulators of MAP kinase signaling (LAMTOR1, HRAS, RAC1, MAPKAPK5, and MAPK4) and PI3 kinase signaling (RHOA, RPTOR, PIK33R2, TSC2, AKT, PI3KCA, PI3KCB and RICTOR).
Socio-economic impact of the project
Patients with advanced disease often undergo androgen deprivation therapy, which initially leads to tumor regression but typically results in recurrence of castration-resistant prostate cancer that is highly metastatic. Appearance of metastatic disease is almost always a harbinger of eventual cancer mortality. Therefore, given the complexity of the metastatic process together with the low mutational landscape that characterize prostate cancer, we hypothesized that the dissection of the molecular events responsible for disease progression may inform subsequent preclinical and clinical studies against new therapeutic targets.
Our results open new perspectives in the management of patients with a diagnosis of localized prostate cancer. Further investigation to confirm the association of FOXM1 and CENPF with patients’ outcome
We anticipate that investigating these processes is of paramount importance for several reasons: (i) first, because there are virtually no molecular biomarkers for patients with an early diagnosis of localized high-grade prostate cancer that can predict the occurrence of distant metastasis; (ii) second, because even for patients with metastatic disease, there very few biomarkers of progression after treatment beyond prostate specific antigen (PSA) serum levels, and (iii) third, because despite recent advances in chemotherapy and androgen signaling blocking have shown to be effective at improving survival of metastatic prostate cancer patients, there is yet no available treatment that is curative.
The expected final results or the proposed project can be divided in three as corresponding to the different aims. Firstly, the mouse prostate cancer interactome is a very valuable resource for the scientific community. We expect our results to be a proof of concept that this regulatory network can be interrogated with signatures of different biological contexts to predict drivers of the observed phenotypes. In our case we have been able to successfully identify and validate master regulators of disease progression but one can foresee applications in contexts like castration resistance, tumor initiation or drug resistance. Secondly, by identifying FoxM1 and Cenpf as master regulators of disease progression, we have provided new biomarkers that can potentially be translated into the human clinical practice as tools to help predicting risk in patients with low-grade tumors. Thirdly, we expect that the completion of the objectives for Aim 3 of the current proposal, which will be pursued during the last year, will lead to the identification of new druggable targets for therapeutic intervention. We speculate those targets to be upstream modulators of the transcriptional activity of FoxM1 and Cenpf so that we will be able to investigate the mechanisms of action of candidate drugs and the potential mechanisms of resistance.