Aaron Boudreau

Splicing program of human MENA produces a previously undescribed isoform associated with invasive, mesenchymal-like breast tumors

Human mena (hMENA), a member of the actin cytoskeleton regulators Ena/VASP, is overexpressed in high-risk preneoplastic lesions and in primary breast tumors and has been identified as playing a role in invasiveness and poor prognosis in breast cancers that express HER2. Here we identify a unique isoform, hMENAΔv6, derived from the hMENA alternative splicing program. In an isogenic model of human breast cancer progression, we show that hMENA11a is expressed in premalignant cells, whereas hMENAΔv6 expression is restricted to invasive cancer cells. “Reversion” of the malignant phenotype leads to concurrent down-regulation of all hMENA isoforms. In breast cancer cell lines, isoform-specific hMENA overexpression or knockdown revealed that in the absence of hMENA11a, overexpression of hMENAΔv6 increased cell invasion, whereas overexpression of hMENA11a reduced the migratory and invasive ability of these cells. hMENA11a splicing was shown to be dependent on the epithelial regulator of splicing 1 (ESRP1), and forced expression of ESRP1 in invasive mesenchymal breast cancer cells caused a phenotypic switch reminiscent of a mesenchymal-to-epithelial transition (MET) characterized by changes in the cytoskeletal architecture, reexpression of hMENA11a, and a reduction in cell invasion. hMENA-positive primary breast tumors, which are hMENA11a-negative, are more frequently E-cadherin low in comparison with tumors expressing hMENA11a. These data suggest that polarized and growth-arrested cellular architecture correlates with absence of alternative hMENA isoform expression, and that the hMENA splicing program is relevant to malignant progression in invasive disease.

Gene Co-Expression Modules as Clinically Relevant Hallmarks of Breast Cancer Diversity

Co-expression modules are groups of genes with highly correlated expression patterns. In cancer, differences in module activity potentially represent the heterogeneity of phenotypes important in carcinogenesis, progression, or treatment response. To find gene expression modules active in breast cancer subpopulations, we assembled 72 breast cancer-related gene expression datasets containing ∼5,700 samples altogether. Per dataset, we identified genes with bimodal expression and used mixture-model clustering to ultimately define 11 modules of genes that are consistently co-regulated across multiple datasets. Functionally, these modules reflected estrogen signaling, development/differentiation, immune signaling, histone modification, ERBB2 signaling, the extracellular matrix (ECM) and stroma, and cell proliferation. The Tcell/Bcell immune modules appeared tumor-extrinsic, with coherent expression in tumors but not cell lines; whereas most other modules, interferon and ECM included, appeared intrinsic. Only four of the eleven modules were represented in the PAM50 intrinsic subtype classifier and other well-established prognostic signatures; although the immune modules were highly correlated to previously published immune signatures. As expected, the proliferation module was highly associated with decreased recurrence-free survival (RFS). Interestingly, the immune modules appeared associated with RFS even after adjustment for receptor subtype and proliferation; and in a multivariate analysis, the combination of Tcell/Bcell immune module down-regulation and proliferation module upregulation strongly associated with decreased RFS. Immune modules are unusual in that their upregulation is associated with a good prognosis without chemotherapy and a good response to chemotherapy, suggesting the paradox of high immune patients who respond to chemotherapy but would do well without it. Other findings concern the ECM/stromal modules, which despite common themes were associated with different sites of metastasis, possibly relating to the “seed and soil” hypothesis of cancer dissemination. Overall, co-expression modules provide a high-level functional view of breast cancer that complements the “cancer hallmarks” and may form the basis for improved predictors and treatments.

An "elite hacker": breast tumors exploit the normal microenvironment program to instruct their progression and biological diversity.

The year 2011 marked the 40 year anniversary of Richard Nixon signing the National Cancer Act, thus declaring the beginning of the “War on Cancer” in the United States. Whereas we have made tremendous progress toward understanding the genetics of tumors in the past four decades, and in developing enabling technology to dissect the molecular underpinnings of cancer at unprecedented resolution, it is only recently that the important role of the stromal microenvironment has been studied in detail. Cancer is a tissue-specific disease, and it is becoming clear that much of what we know about breast cancer progression parallels the biology of the normal breast differentiation, of which there is still much to learn. In particular, the normal breast and breast tumors share molecular, cellular, systemic and microenvironmental influences necessary for their progression. It is therefore enticing to consider a tumor to be a “rogue hacker”—one who exploits the weaknesses of a normal program for personal benefit. Understanding normal mammary gland biology and its “security vulnerabilities” may thus leave us better equipped to target breast cancer. In this review, we will provide a brief overview of the heterotypic cellular and molecular interactions within the microenvironment of the developing mammary gland that are necessary for functional differentiation, provide evidence suggesting that similar biology—albeit imbalanced and exaggerated—is observed in breast cancer progression particularly during the transition from carcinoma in situ to invasive disease. Lastly we will present evidence suggesting that the multigene signatures currently used to model cancer heterogeneity and clinical outcome largely reflect signaling from a heterogeneous microenvironment—a recurring theme that could potentially be exploited therapeutically.

Subcellular Localization and Ser-137 Phosphorylation Regulate Tumor-suppressive Activity of Profilin-1

Background: The actin-binding protein profilin-1 is a eukaryotic protein essential for growth, with poorly understood antitumor function. Results: Profilin-1 antitumor activity requires nuclear localization and is inhibited by Ser-137 phosphorylation. Conclusion: Profilin-1 has spatially defined functions and is post-translationally regulated. Significance: Our data support a model to reconcile the seemingly oppositional functions of profilin-1 and may have implications for novel anticancer therapies. The actin-binding protein profilin-1 (Pfn1) inhibits tumor growth and yet is also required for cell proliferation and survival, an apparent paradox. We previously identified Ser-137 of Pfn1 as a phosphorylation site within the poly-l-proline (PLP) binding pocket. Here we confirm that Ser-137 phosphorylation disrupts Pfn1 binding to its PLP-containing ligands with little effect on actin binding. We find in mouse xenografts of breast cancer cells that mimicking Ser-137 phosphorylation abolishes cell cycle arrest and apoptotic sensitization by Pfn1 and confers a growth advantage to tumors. This indicates a previously unrecognized role of PLP binding in Pfn1 antitumor effects. Spatial restriction of Pfn1 to the nucleus or cytoplasm indicates that inhibition of tumor cell growth by Pfn1 requires its nuclear localization, and this activity is abolished by a phosphomimetic mutation on Ser-137. In contrast, cytoplasmic Pfn1 lacks inhibitory effects on tumor cell growth but rescues morphological and proliferative defects of PFN1 null mouse chondrocytes. These results help reconcile seemingly opposed cellular effects of Pfn1, provide new insights into the antitumor mechanism of Pfn1, and implicate Ser-137 phosphorylation as a potential therapeutic target for breast cancer.

Abstract P4-05-02: Phospho-reactome measurements reveal heterogenic kinase signatures

Background: Treating cancer increasingly relies on targeting kinases, because their oncogenic activity drive tumorigenesis. Discovering which active mechanisms of disease progression can be efficiently targeted, and knowing whether kinase networks circumvent therapeutic interventions, are challenges researchers and clinicians face. Surprisingly however, measuring the phosphorylating activity of kinases, and potentially monitoring the functionality of the entire human phospho-reactome at once, remains largely unexplored. We developed a semi-high throughput assay to monitor the phospho-catalytic activity of kinase enzymes, using their biological targets as phospho-sensors. We successfully used this assay to identify oncogenic phospho-signatures prevalent in breast cancer, and can be used to establish drug-sensitivity profiles in models of kinase-targeted therapies. Methods: We first defined how to computationally build a library of peptide sensors established from confirmed kinase substrates’ phosphorylation sites. Precisely, we used computational methods to create a unique phospho-repertoire cataloguing 3,408 peptide sequences established from validated human proteins’ phosphorylation sites, curated from 38 public databases. Second, we experimentally used these biologically relevant probes in multiplex assays to quantify the catalytic state of kinases. Specifically, a kinome-representative 242-peptide set was developed into an ATP-consumption screen to identify the activity signatures of EGFR, MAPK, AKT, ABL and SRC family kinases, and explore 642 kinase/substrate nodes. Next, we described analysis methods to derive phospho-signatures from semi-high throughput ATP-consumption measurements. We validated the assay using isogenic culture model of basal-like breast cancer (HMT-3522 S1 and T4-2), and cell lines harboring EGFR/HER2-oncogenic alterations such as MDA-MB-231, MCF7 or T47D. Results: The differential phosphorylation activity of 25 recombinant, active kinase enzymes was successfully captured. In cancer cell extracts, hyper-activated EGFR, ERK, MEK, AKT, and SRC kinases originally identified by immuno-detection were reliably and specifically detectable using the peptide-based kinase-activity assay. The phospho-sensing assay revealed the heterogeneity of active kinase signaling circuits among different breast cancer cells. Conclusion: This unique strategy and resources allow to comprehensively measure the catalytic activity of multitude kinases at once, representing a new molecular dimension to characterize biological samples. We will use such new phospho-reactome profiling system to determine the efficacy of new combinatorial therapies, and define how chemotherapeutic interventions lead to the reprogramming of phospho-circuits. Citation Information: Cancer Res 2013;73(24 Suppl): Abstract nr P4-05-02.

Abstract P1-08-01: MammaPrint ultra-high risk score is associated with response to neoadjuvant chemotherapy in the I-SPY 1 TRIAL (CALGB 150007/150012; ACRIN 6657)

Background: Increasingly, gene expression tests are being used with standard clinicopathological markers to assess risk of recurrence in breast cancer and to guide treatment. The 70-gene signature (MammaPrint™) was recently validated in a prospective study as an effective tool for identifying low risk patients who may avoid chemotherapy without compromising outcome. MammaPrint scores have been shown to differentiate likelihood of neoadjuvant chemotherapy response. Here we hypothesize that a further stratification of the 70-gene signature into ‘high-risk’ and ‘ultra-high risk’ groups might yield improved predictors of chemo-sensitivity and patient outcome. Methods: Agilent 44K array data from pre-treatment biopsies were obtained from 149 I-SPY 1 TRIAL patients treated with neoadjuvant anthracycline-based chemotherapy. These data were used to compute each patient9s MammaPrint score and risk category (good vs. poor threshold 0.4). Of these patients, 138 were either in the ‘poor’ outcome group (136) or in the ‘good’ outcome group but Her2+ (2). The median score cut-point based on these 138 patients equals -0.154, and was used to further stratify patients into MammaPrint High1 (MP1) or MammaPrint (ultra) High2 (MP2) groups, with MP1 defined as ≤ -0.154 and MP2 defined as >-0.154 (69 MP1, 69 MP2). Outcome parameters included pathologic complete response (pCR) after therapy and recurrence free survival (RFS). Fisher9s exact test was used to assess association with pCR overall and within hormone receptor (HR) and Her2 subtypes, and Cox proportional hazards modeling to assess association with RFS. 29 patients who received Herceptin and 4 without pCR data were excluded from analysis, yielding 105 evaluable patients. Results: Though all receptor subtypes were represented in MP1 and MP2 subgroups, the majority of MP1 patients were HR+/Her2- (54%), whereas MP2 patients were more evenly distributed among triple negative (TN; 38%), HR+/Her2- (22%), HR-/Her2+ (22%) and HR+/Her2+ (18%) subtypes. 25/105 patients achieved pCR: 8 (15%) in the MP1 group and 17 (33%) in the MP2 group. Applying Fisher9s exact test, we found that a significantly higher percentage of pCR was observed for MP2 patients in the overall population (p = 0.038), but not within receptor subtypes; though there was a trend towards a higher pCR rate in MP2 patients within the HR+/Her2- subset (p = 0.071). The greater sensitivity of MP2 to chemotherapy may in part be driven by differences in receptor subtype distribution, as well as a relationship between MP2 status and proliferation; we found that MP1/2 classification was highly associated with Ki67 (low/intermediate: 25% positive) (p = 3.3E-5), with MP2 patients having more Ki67 positive cells. No significant differences in RFS were observed between MP1 and MP2 subsets, likely because MP2 patients, who respond best, are also at the highest risk of relapse. Conclusion: These analyses suggest that additional MammaPrint score stratification within the ‘poor’ biology group might be a useful for developing companion diagnostics to neoadjuvant therapies, a hypothesis currently being tested in the adaptive randomization engine of the I-SPY 2 clinical trial. Citation Information: Cancer Res 2013;73(24 Suppl): Abstract nr P1-08-01.

Abstract 5221: Co-expression modules as hallmarks of breast cancer survival and response.

Background: Co-expression modules in cancer are cohesive groups of genes with highly correlated expression patterns, presumably regulating phenotypes important in carcinogenesis, metastasis, or response to treatment. With the growing database of gene expression data, it is now possible to deduce co-expression modules active in breast cancer subpopulations. Methods: We assembled 74 breast-cancer related gene expression datasets containing ∼5,500 samples altogether. Per dataset, we identified genes with bimodal expression analyzed using mixture-model clustering to ultimately find gene groups that are consistently co-regulated across multiple datasets. Functional and pathway enrichment of modules was assessed using DAVID and g:profiler software tools. Associations between module expression and patient outcome, chemotherapy response, clinical variables, epithelial vs. stromal expression, intrinsic subtype and other published signatures were assessed using standard statistical methods. Results: Our meta-analysis identified 11 modules ranging in size from ∼5-200 genes. As expected, there were modules representing estrogen signaling, cell proliferation, and ERBB2 signaling that correlate with intrinsic subtype and receptor status. In addition, we found modules associated with immune signaling, development, histone modification, and the ECM. The immune modules were highly correlated to previously published Tcell/Bcell, STAT1, and IFN immune signatures. In multivariate analysis, the combination of a downregulated Tcell/Bcell immune module and an upregulated proliferation module associated with recurrence, suggesting that cancers with a high proliferation rate in the absence of an activated immune system are prone to recur. This association was especially strong in TN and basal cancers. Comparing modules associated with response to chemotherapy to those associated with the prognosis of untreated patients, the most common pattern was that of a module associating with good prognosis or a good response to chemotherapy (but not both). For instance, high expression of the estrogen module associates with a good prognosis but a poor response to chemotherapy, whereas upregulation of the proliferation module associates with a poor prognosis but a good response to chemotherapy. Another pattern we observed, of a biomarker that associates with good prognosis without chemotherapy and a good response to chemotherapy, was found in the T/B cell immune modules. Conclusion: Co-expression modules provide a high-level functional view of breast cancer that complements the ‘cancer hallmarks’. These results suggest that in some high-immune patients, the same host processes contributing to an excellent response to chemotherapy might preclude its necessity, and support a treatment strategy boosting anti-tumoral immunity in low-immunity patients with highly proliferating tumors. Citation Format: Denise M. Wolf, Christina Yau, Aaron Boudreau, Laura Esserman, Laura Van ‘t Veer, Marc Lenburg. Co-expression modules as hallmarks of breast cancer survival and response. [abstract]. In: Proceedings of the 104th Annual Meeting of the American Association for Cancer Research; 2013 Apr 6-10; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2013;73(8 Suppl):Abstract nr 5221. doi:10.1158/1538-7445.AM2013-5221

Abstract P4-09-02: A robust signature of long-term clinical outcome in breast cancer

Background: Multigene prognostic signatures derived from high-dimensional mRNA expression data have been proposed to forecast patient outcome and predict chemotherapy benefit more accurately than standard clinical parameters. However, many prognostic signatures fail to predict late recurrences occurring 10 or more years following initial diagnosis. Furthermore, breast cancer subtypes canonically associated with favorable biology, particularly estrogen-receptor positive disease, are characterized by a higher frequency of late recurrences. Patients identified as being at a higher risk of late recurrence might benefit from more prolonged systemic (hormonal) therapy; as such, the goal of this research has been to develop a prognostic signature that can faithfully stratify patient risk up to and beyond 10 years of follow-up. Methods: A novel multiplexed Cox modeling approach was applied to microarray data obtained from an untreated, node-negative patient cohort with long-term follow-up (n = 141) to train the signature. The long-term prognostic signature was subsequently validated in an additional group of patients (n = 154) that were mostly node-positive (94%) and administered adjuvant chemotherapy (71%). The performance of the signature was compared to existing clinicopathological parameters and genomic signatures using Cox proportional hazards analysis. Logistic regression modeling was employed to evaluate the added benefit to discriminatory accuracy obtained by incorporating the long-term prognostic signature alongside current biomarkers to predict 10-year overall survival. Results: The long-term signature was able to stratify patient risk with unprecedented accuracy compared to standard clinicopathological and genomic features in both the training and the validation cohorts; none of the patients predicted to have good biology (n = 47 and 38) died within 10 years in either cohort, whereas only 34% and 44% of patients predicted to have poor biology (n = 47 and 46) survived 10 years in the training and validation cohorts, respectively. Within the validation series, the signature was able to identify patients at risk of metastasis with the highest hazard ratio in comparison to other prognostic signatures (univariate hazard ratio of 14.0 [95% CI 3.2–58; p = 0.00083], adjusted hazard ratio of 5.2 [95% CI 1.2–22; p = 0.0257] when corrected for standard clinicopathological markers). Adding the long-term prognostic signature to existing prognostic biomarkers led to significantly improved classification of patients into appropriate 10-year overall survival risk categories (Net Reclassification Improvement of 25.1% in validation series at >5% risk threshold, p = 0.0123). Conclusions: We were able to identify a 200-gene long-term signature able to stratify patient risk with superior accuracy over a relatively long follow-up period. We are currently using this algorithm to develop prognostic signatures for other cancer types, and are using similar multiplexed algorithms to develop gene signatures able to predict response to neoadjuvant therapy. Citation Information: Cancer Res 2012;72(24 Suppl):Abstract nr P4-09-02.

Abstract LB-311: Combinatorial logic over breast cancer control modules predicts survival and chemotherapy response of ISPY1 breast cancer patients (CALGB 150007/150012; ACRIN 6657)

Proceedings: AACR 102nd Annual Meeting 2011‐‐ Apr 2‐6, 2011; Orlando, FL Background : Breast cancer is a heterogeneous disease. Clinically useful predictors for breast cancer management may therefore need to consider the variety of molecular processes that can impact chemotherapy response and survival. With the growing database of gene expression data, it is now possible to deduce co-expression modules active in breast cancer subpopulations. Viewed from a combinatorial perspective supporting ‘OR’ and ‘AND’ reasoning, the activities of these modules might provide insights into breast cancer biology and form the basis for improved predictors. Methods : We assembled 74 breast-cancer related gene expression datasets containing ∼5,500 samples altogether. Per dataset, we identified genes with bimodal expression analyzed using mixture-model clustering to ultimately find gene groups that are consistently co-regulated across multiple datasets. We scored the AFFY U133 ISPY1 dataset (117 pts) for module expression, and explored the relationships between module state, chemotherapy response, patient outcome, receptor status, intrinsic subtype, and other signatures. To derive logical functions consisting of nested ‘AND’ and ‘OR’ gates relating module expression to clinical variables, we discretized the module scores and applied a methodology in the spirit of Karnaugh maps for digital circuit design. Results : Our meta-analysis identified 11 modules ranging in size from ∼5–200 genes. The expression level of some modules were associated with known molecular markers such as intrinsic subtype or wound signature scores, whereas others – three immune related modules, two ECM related modules, a mixed immune/ECM module, and a small module encoding histones – do not. Disease free survival of patients with triple negative disease is predicted (p=1.93e-08) by the logical function fDFS_TN =[ER_module>LOW] OR [TN_Immune_ECM=HIGH] OR [(AURK/PROLIF_module=HIGH) AND ((Histone_module>LOW) OR (TN_Immune_ECM=HIGH)], where TN_Immune_ECM is a composite function over the three immune modules and the immune/ECM hybrid module. The best predictor of chemo-sensitivity for triple negatives is a hybrid over subtype and module state, fRCB01_TN =Subtype_basal AND TN\_immune\_ECM, correctly classifying 95% of the RCB01 sensitives and 90% of the RCBIII resistants. Conclusion : In triple negatives, high proliferation can be rescued by immune/ECM upregulation, though neither immune rescue nor low RCB is necessary for survival if the estrogen module is even slightly upregulated or if the histone module is high. Similar analyses of ER/Pr+ and Her2+ patients demonstrate receptor-subtype specificity in the logic predicting response. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 102nd Annual Meeting of the American Association for Cancer Research; 2011 Apr 2-6; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2011;71(8 Suppl):Abstract nr LB-311. doi:10.1158/1538-7445.AM2011-LB-311

P1-06-09: Patient-Specific Integrative Pathway Analysis Using PARADIGM Identifies Key Activities in I-SPY 1 Breast Cancer Patients (CALGB 150007/150012; ACRIN 6657).

Background: A major challenge in interpreting high-throughput multianalyte genomic data sets such as those produced by the ISPY clinical trials is data integration and interpretation within the context of biologically relevant pathways. To address this need, the data analysis tool PARADIGM (PAthway Recognition Algorithm using Data Integration on Genomic Models) was developed to infer the activities of genetic pathways by integrating any number of functional genomic data sets for a given patient sample into a pathway activity profile. Methods: We used PARADIGM to integrate gene expression (Agilent 44K) and DNA copy number data (AFFY 22K and 330K MIP) from 133 ISPY-1 patients into pathway component activity levels for approximately 1400 curated signal transduction, transcriptional and metabolic pathways superimposed onto a single non-redundant ‘SuperPathway9. These pathway activities then become the substrate for statistical analyses to identify pathways characterizing different breast cancer subtypes, as well as those associated with recurrence and response to neoadjuvant chemotherapy within breast cancer subgroups. To identify subtype-specific pathway activities, we used ANOVA for initial feature filtering followed by Tukey analysis with Benjamini Hochberg multiple testing correction. For other binary outcome comparisons we used Mann-Whitney (2-sample Wilcoxon) analysis. PARADIGM results were corroborated with pathway enrichment analysis and filtered for significance. Results: In agreement with breast cancer cell line and other prior studies, basal-like and triple negative cancers are dominated by upregulation of the FOXM1 and MYC/Max subnetworks and downregulation of the FOXA1/ER signal transduction pathway, the converse of the activity pattern seen in luminal breast cancers. These and other subtype associations pass stringent multiple testing corrected significance tests. Though an association study of recurrence over the entire patient cohort mostly yields pathways characteristic of basal-like tumors, alternative pathway associations emerge when subtypes are analyzed individually for outcome and significance tests are relaxed to include features that pass un-corrected Wilcoxon significance tests and also generate highly significant pathway enrichment scores. Subtype-specific drivers of recurrence and chemo-resistance supported by this level of evidence include ALK1/2 (TGFB-BMP) and p53 effector signaling for basals and Syndecan-1 and c-MYC for luminals. Chemo-sensitivity pathways, assessed by association with pCR and RCB1, appear to be subtype-specific as well, with HDAC class 1 signaling, LRP6-Wnt, and IRE1alpha chaperones dominating basal-like cancers and c-MYB activity dominating Her2+ cancers, whereas chemo-sensitivity of HR+Her2- cancers though rare appears to be driven by the DNA damage axis (BRCA/BARD1). Conclusion: These and other similar analyses suggest that patients with TN or basal-like disease might benefit from the addition of ALK1 pathway inhibitors to treatment, whereas high risk HR+ patients might benefit from Syndecan-1 inhibitors. C-MYC/MAX inhibitors might benefit all high risk patients. Citation Information: Cancer Res 2011;71(24 Suppl):Abstract nr P1-06-09.