Co-expression Extrapolation Research Articles

Cell-free DNA Methylation as a Predictive Biomarker of Response to Neoadjuvant Chemotherapy for Patients with Muscle-invasive Bladder Cancer in SWOG S1314

BackgroundNeoadjuvant chemotherapy (NAC) is the standard of care in muscle-invasive bladder cancer (MIBC). However, treatment is intense, and the overall benefit is small, necessitating effective biomarkers to identify patients who will benefit most. ObjectiveTo characterize cell-free DNA (cfDNA) methylation in patients receiving NAC in SWOG S1314, a prospective cooperative group trial, and to correlate the methylation signatures with pathologic response at radical cystectomy. Design, setting, and participantsSWOG S1314 is a prospective cooperative group trial for patients with MIBC (cT2-T4aN0M0, ≥5 mm of viable tumor), with a primary objective of evaluating the coexpression extrapolation (COXEN) gene expression signature as a predictor of NAC response, defined as achieving pT0N0 or ≤pT1N0 at radical cystectomy. For the current exploratory analysis, blood samples were collected prospectively from 72 patients in S1314 before and during NAC, and plasma cfDNA methylation was measured using the Infinium MethylationEPIC BeadChip array. InterventionNo additional interventions besides plasma collection. Outcome measurements and statistical analysisDifferential methylation between pathologic responders (≤pT1N0) and nonresponders was analyzed, and a classifier predictive of treatment response was generated using the Random Forest machine learning algorithm. Results and limitationsUsing prechemotherapy plasma cfDNA, we developed a methylation-based response score (mR-score) predictive of pathologic response. Plasma samples collected after the first cycle of NAC yielded mR-scores with similar predictive ability. Furthermore, we used cfDNA methylation data to calculate the circulating bladder DNA fraction, which had a modest but independent predictive ability for treatment response. In a model combining mR-score and circulating bladder DNA fraction, we correctly predicted pathologic response in 79% of patients based on their plasma collected at baseline and after one cycle of chemotherapy. Limitations of this study included a limited sample size and relatively low circulating bladder DNA levels. ConclusionsOur study provides the proof of concept that cfDNA methylation can be used to generate classifiers of NAC response in bladder cancer patients. Patient summaryIn this exploratory analysis of S1314, we demonstrated that cell-free DNA methylation can be profiled to generate biomarker signatures associated with neoadjuvant chemotherapy response. With validation in additional cohorts, this minimally invasive approach may be used to predict chemotherapy response in locally advanced bladder cancer and perhaps also in metastatic disease.

SWOG S1314: A randomized phase II study of co-expression extrapolation (COXEN) with neoadjuvant chemotherapy for localized, muscle-invasive bladder cancer with overall survival follow up.

536 Background: This trial evaluated COXEN, a gene expression model, as a predictive biomarker in muscle-invasive bladder cancer (BC) patients randomized to Gemcitabine-Cisplatin (GC) or dose-dense Methotrexate-Vinblastine-Adriamycin/doxorubicin-Cisplatin (ddMVAC). Primary results correlating COXEN with pathologic response at surgery have been reported. This secondary analysis includes progression-free (PFS) and overall survival (OS). Methods: Eligibility included Stage cT2-T4a N0 M0, urothelial BC (mixed histology allowed), ≥ 5 mm of viable tumor, cisplatin eligible, with plan for cystectomy. 237 patients were randomized between ddMVAC, given every 14 days for 4 cycles, and GC, given every 21 days for 4 cycles. Cox regression was used to evaluate COXEN score or treatment arm association with PFS and OS, adjusting for stratification factors (stage and PS). Results: 167 patients were included in the primary COXEN analysis all having either at least 3 cycles of chemo and surgery within 100 days of last chemo or having progressed while receiving chemo. The COXEN scores were not significantly prognostic for OS or PFS in their respective arms; the COXEN GC score was a significant predictor for OS in pooled arms. OS and PFS data are shown for both scores in the table. In the intent to treat analysis (n=227), there was no significant difference in OS or PFS for ddMVAC versus GC (for OS, HR =0.87, 95% CI 0.54-1.40), p = 0.57); for PFS (HR= 0.76 95% CI 0.58-1.01, p = 0.055). Association of path response with OS will be presented. Conclusions: The COXEN GC score may be prognostic of survival in those receiving platinum-based neoadjuvant treatment. The randomized, prospective design provides estimates of OS and PFS for GC and ddMVAC that appear comparable, but this phase II trial is underpowered for definitive comparisons. The prospective data and correlative samples from S1314 will allow for further assessment of COXEN and other RNA and DNA based predictive and prognostic biomarkers. Clinical trial information: NCT02177695. [Table: see text]

A pilot clinical trial of genomic-based therapy assignment with co-expression extrapolation (COXEN) in advanced/metastatic urothelial carcinoma.

452 Background: We investigated the Co-eXpression ExtrapolatioN (COXEN) algorithm to select “next best therapy” for patients with advanced/metastatic urothelial cancer based on gene expression profiles. Methods: This is a single-arm, open label pilot study, that investigates the COXEN algorithm’s ability to use patients’ tumor gene expression models to predict sensitivities to 75 FDA-approved anticancer agents. The COXEN results were reviewed and discussed in a multi-disciplinary molecular tumor board. From that a treatment plan was chosen based on monotherapy clinical efficacy data in urothelial carcinoma, combination safety data (combinations of agents were allowed provided phase I data were available), side effect profile/tolerability, response or resistance to prior therapies, therapy contraindications, and feasibility of therapy. The objective of this study was to determine the feasibility of using COXEN as a clinical decision algorithm to make a real-time treatment decision within 21 days of specimen collection in patients with advanced/metastatic urothelial carcinoma. Results: A total of 8 patients enrolled and underwent tumor biopsy for COXEN analysis. Five out of 8 patients received COXEN therapy within 21 days of biopsy with various regimens including: vorinostat and etoposide; doxorubicin; sunitinib; doxorubicin and paclitaxel; erlotinib and nab-paclitaxel. There were no objective responses. The 5 patients treated with COXEN therapy had a median potential follow up of 36 months, median overall survival of 8.4 months (95% CI: 5.2 months – not estimable), and median progression free survival of 2.2 months (95% CI: 1.4 – 3.9 months). Grade 3 or 4 adverse events included: febrile neutropenia (n=2), hypophosphatemia (n=1) and hyponatremia (n=1). This trial was terminated early due to a lack of treatment responses. Conclusions: The COXEN algorithm is a unique personalized method for selecting therapy for patients with advanced urothelial carcinoma. Although the COXEN method was feasible, it did not demonstrate clinical efficacy in this small cohort of patients. Clinical trial information: NCT02788201.

Enhanced Co-Expression Extrapolation (COXEN) Gene Selection Method for Building Anti-Cancer Drug Response Prediction Models.

The co-expression extrapolation (COXEN) method has been successfully used in multiple studies to select genes for predicting the response of tumor cells to a specific drug treatment. Here, we enhance the COXEN method to select genes that are predictive of the efficacies of multiple drugs for building general drug response prediction models that are not specific to a particular drug. The enhanced COXEN method first ranks the genes according to their prediction power for each individual drug and then takes a union of top predictive genes of all the drugs, among which the algorithm further selects genes whose co-expression patterns are well preserved between cancer cases for building prediction models. We apply the proposed method on benchmark in vitro drug screening datasets and compare the performance of prediction models built based on the genes selected by the enhanced COXEN method to that of models built on genes selected by the original COXEN method and randomly picked genes. Models built with the enhanced COXEN method always present a statistically significantly improved prediction performance (adjusted p-value ≤ 0.05). Our results demonstrate the enhanced COXEN method can dramatically increase the power of gene expression data for predicting drug response.

Review of SWOG S1314: Lessons from a Randomized Phase II Study of Co-Expression Extrapolation (COXEN) with Neoadjuvant Chemotherapy for Localized, Muscle-Invasive Bladder Cancer.

SWOG1314 is a randomized phase II study of co-expression extrapolation (COXEN) with neoadjuvant chemotherapy for localized, muscle-invasive bladder cancer. COXEN is a biomarker approach in which predictive biomarkers are developed using in vitro data, which may then be applied directly into a clinical testing application. Two separate Gene Expression Models (GEMs), one for the Methotrexate, Vinblastine, Doxorubicin, Cisplatin (MVAC) regimen and another for gemcitabine plus cisplatin (GC) were tested in S1314. The lessons learned from the development and operationalization of the S1314 clinical trial are described in detail, which may help to inform the future trials of predictive biomarkers for urothelial carcinoma in the neoadjuvant setting. Specific areas addressed include: The need for broad support from the bladder cancer community, planning for non-evaluable subjects, defining adequate neoadjuvant treatment, defining adequate tissue collection, setting expectations in phase II clinical studies of predictive biomarkers, and maximizing the impact of the samples collected in these studies for broader biomarker development. With a large number of newly available treatments in advanced urothelial carcinoma in the last few years, more investigations of these agents in the neoadjuvant setting is anticipated. There will be a great need for the development of predictive biomarkers in conjunction with the use of these agents in the preoperative setting. Insights from S1314 may provide useful information and lessons learned in this development.

Association of molecular subtypes with pathologic response in a phase II study of co-expression extrapolation (COXEN) with neoadjuvant chemotherapy (NAC) for localized, muscle-invasive bladder cancer (SWOG S1314; NCT02177695).

5028 Background: Cisplatin-based NAC is recommended for patients with MIBC prior to radical cystectomy (RC) but the majority will not have a pathologic response. To identify responders the COXEN gene expression model with chemotherapy-specific scores (for DD-MVAC and GC) was developed and in a prospective rPII clinical trial (SWOG S1314) the GC score was associated with path downstaging in the pooled arms. We investigated RNA based molecular subtypes as additional predictive biomarkers for response to NAC in patients treated in S1314. Methods: Eligibility required cT2-T4a N0 M0, predominant urothelial, > 5 mm tumor, cisplatin eligible, and plan for RC and PLND. 237 patients were randomized between 4 cycles of ddMVAC and GC. Based on Affymetrix transcriptomic data used to assign COXEN scores, we determined subtypes using 3 classifiers: TCGA (k=5), Consensus (k=6), and MD Anderson (MDA; k=3). Primary objective was to assess subtype association with pathologic response to NAC in the pooled arms and to determine any association with COXEN. TCGA and Consensus classifiers were collapsed into 3 groups for ROC analyses. We tested whether each classifier contributed additional predictive power when added to a model based on pre-defined stratification factors (PS 0 vs. 1; T2 vs. T3, T4a). Results: 161 patients had adequate tissue and gene expression results, received at least 3 of 4 cycles of NAC and had pT-N response based on RC. Covariates were 78% PS=0, 89% T2, 84% male, median age 65, 51% randomized to ddMVAC, 49% GC with 33% pT0 and 52% downstaging. Although the TCGA 3 group classifier (Basal-Squamous (BS)/Neuronal, Luminal, Luminal infiltrated) and GC Coxen score yielded the largest AUCs (0.607, 0.610) for pT0 response, neither reached statistical significance (p=0.20, p=0.22). For downstaging (<pT2), the 3 category Consensus classifier (BS/NE-like, Luminal, Stroma-rich) significantly increased the AUC from 0.568 (strat factors alone) to 0.620 (p=0.044). The MDA classifier AUC was 0.640 and the GC Coxen score AUC was 0.626, but neither were significant (p=0.076, p=0.14. The MVAC Coxen score did not improve the AUC beyond the stratification factors. Conclusions: The Consensus classifier, which is based in part on the TCGA and MDA classifiers, modestly improved prediction for pathologic downstaging when added to clinical stage and PS. With additional followup, we will assess the association of COXEN scores and subtypes with overall survival. Clinical trial information: NCT02177695 .

Enhanced co-expression extrapolation (COXEN) gene selection method for building anticancer drug response prediction models.

e14073 Background: Accurate prediction of tumor response to a drug treatment is of paramount importance for precision oncology. The co-expression extrapolation (COXEN) gene selection approach has been successfully used in multiple studies to select genes for predicting the response of tumor cells to a specific drug. Here, we enhance the original COXEN approach to select genes that are predictive of the efficacies of multiple drugs simultaneously for building general drug response prediction model. Methods: We implemented two methods to select predictive genes. The first method ranks the genes according to their prediction power for each individual drug and then takes a union of top predictive genes of all the drugs. The second method uses a linear regression model to evaluate the prediction power of a gene for all drugs while the drugs are one-hot encoded in the regression model. Among the predictive genes, we further select genes by evaluating the preservation of co-expression patterns between cancer cases with drug response data available and cancer cases for which drug response needs to be predicted, because the preservation of co-expression patterns indicates the similarity of genomic regulations between cancer cases. Results: To test the enhanced COXEN method, we used a lightGBM regression model to predict drug response based on the selected genes on two benchmark in vitro drug screening datasets. The table below compares the performance of prediction models built based on 200 genes selected by the enhanced COXEN method to that of models built on 200 genes randomly picked from the LINCS gene set, which includes 976 “landmark” genes well-representing cellular transcriptomic changes identified in the Library of Integrated Network-Based Cellular Signatures (LINCS) project. The enhanced COXEN approach selects genes better than random LINCS genes as demonstrated by the increased average coefficient of determination (R2) for predicting the area under the dose response curve through cross-validation. Pair-wise t-test indicates the improvement is statistically significant (p-value ≤ 0.05) on both datasets. Conclusions: Our result demonstrates the benefit of using an enhanced COXEN approach to select genes for building general drug response prediction model. [Table: see text]

SWOG S1314: A randomized phase II study of co-expression extrapolation (COXEN) with neoadjuvant chemotherapy for localized, muscle-invasive bladder cancer.

4506 Background: Both dose-dense Methotrexate-Vinblastine-Adriamycin/doxorubicin-Cisplatin (ddMVAC) and Gemcitabine-Cisplatin (GC) are accepted neoadjuvant regimens for muscle-invasive bladder cancer (BC). We investigated COXEN, a gene expression model, as a predictive biomarker. Methods: Eligibility included Stage cT2-T4a N0 M0, urothelial BC (mixed histology allowed), ≥ 5 mm of viable tumor, cisplatin eligible, with plan for cystectomy. 237 patients were randomized between ddMVAC, given every 14 days for 4 cycles, and GC, given every 21 days for 4 cycles. The primary objective was to assess whether the pre-specified dichotomous treatment-specific COXEN gene expression profile is prognostic of pT0 rate or ≤ pT1 at surgery, and to assess whether COXEN score is a predictive factor between regimens and response. Logistic regression was used to model response, adjusting for stratification factors. Results: 167 patients were included; the ddMVAC/GC arms had a median age of 65/64, PS = 0 in 80%/75%, Male proportion of 88%/79% and T2 stage of 87%/92%. All had at least 3 cycles of chemo and surgery/progression within 100 days of last chemo. There were favorable COXEN ddMVAC scores in 32% and GC score in 26%. The pT0 rates for ddMVAC and GC were 32% and 35%; the rates of ≤ pT1 were 55% and 49%, respectively. Conclusion: The COXEN scores were not significantly prognostic for response in their individual arms; The COXEN GC score was significant predictor for downstaging in pooled arms. There was no evidence of an interaction between COXEN score and regimen in predicting response. The prospective data and samples from this study will allow for further development of COXEN and other predictive biomarkers. Clinical trial information: NCT02177695. [Table: see text]

Abstract SY36-01: Optimizing neoadjuvant therapy prior to surgery: Biomarker and target discovery

Abstract SY36-01: Optimizing neoadjuvant therapy prior to surgery: Biomarker and target discovery

COXEN prediction of antineoplastic drug sensitivity in bladder cancer patients.

365 Background: COXEN (Co-eXpression ExtrapolatioN) uses molecular profiles as a “rosetta stone” for translating drug sensitivities of one set of cancers into predictions for another completely independent set of cell lines or human tumors. The ability of COXEN to predict drug effectiveness in pts using tumor samples from in vitro assays is unique. Methods: We tested the predictive value of COXEN for standard chemotherapies in a cohort of bladder cancer pts. Total RNA was extracted from formalin fixed paraffin embedded (FFPE) tissue and converted to cDNA, amplified with Ovation FFPE WTA, and hybridized to a GeneChip Human Genome U133 Plus 2.0 array. Using gene expression data from 278 independent bladder tumors, COXEN scores were generated using bioinformatics models originally built using the NCI-60 cell line panel and a model building algorithm (MiPP). Gene expression data was processed to score 76 FDA approved antineoplastic drugs. Results: A total of 24 samples were tested (15 tumors with 1 sample and 9 tumors with 2 biological replicas (2 samples from the same tumor)) from 15 pts who received chemotherapy (median age 64 (41-74); 73% male; with muscle invasive bladder cancer (MIBC) (12/15, 80%) or metastatic bladder cancer (mBC) (3/15, 20%)). Response to therapy was confirmed by pathologic response in MIBC pts and radiologic response in mBC pts. Chemotherapies evaluated included: methotrexate/vinblastine/doxorubicin/cisplatin; gemcitabine/cisplatin; gemcitabine/carboplatin; and cisplatin/etoposide. COXEN accurately predicted antineoplastic drug sensitivity in 11/15 (73%) pts (75% MIBC and 67% mBC), of which 7/11 pts had 2 biological samples. However, only 3/7 (43%) biological replicas confirmed COXEN prediction. COXEN accurately predicted drug sensitivity in 9/10 (90%) pts with response and 2/5 (40%) pts with resistance to therapy. Conclusions: COXEN did well in predicting antineoplastic drug response for the majority of bladder cancer pts in this cohort. However, predictions from 2 samples within the same tumor were not always consistent, likely due to the expected tumor heterogeneity found in bladder cancer tumors. A prospective clinical trial in patients with mBC using COXEN to select next best therapy is in development.

Drug Selection in the Genomic Age: Application of the Coexpression Extrapolation Principle for Drug Repositioning in Cancer Therapy.

The use of patient-specific data in drug and dose selection is becoming an increasingly important component in cancer therapy. Basing drug choice on molecular aspects of the tumor is consistent with the identification of cancer as a molecular disease or diseases, even within the same histological type, and its treatment specific to the background from which it arose and exists. Multiple examples exist of single-gene mutations, and over- or underexpression that convey either sensitivity or resistance to a given agent. What about the picture that global gene expression in a cancer presents regarding drug sensitivity or resistance? Coexpression extrapolation (COXEN) is a methodology that acts as a Rosetta stone between patterns of gene expression that correlate to drug responses in vitro and those in tumors of untreated patients to predict chemosensitivity in such tumors even to drugs that are not specifically indicated for that histotype. Further applications of COXEN in drug discovery allow for in silico screens correlating drug response and gene expression in a genetically diverse cell panel to gene expression patterns in a target tumor with the potential for identifying and repurposing compounds. Here we discuss how COXEN is being developed and tested for application in drug selection, repositioning, and repurposing in oncology.

Abstract 2179: Gene expression models for predicting doxorubicin response in canine osteosarcoma

Abstract Purpose: The use of gene expression models for predicting individual response to cancer therapy is a promising area of research for improving outcome. Osteosarcoma is the most common primary bone cancer in dogs, with an average 1 year survival after treatment. The co-expression extrapolation (COXEN) method has been established as a gene expression method for accurately predicting drug sensitivity in human cancer. Our purpose is to explore the ability of COXEN for the intra- or interspecies extrapolation of cancer cell line data for prediction of chemosensitivity in canine osteosarcoma patients. Methods: Gene expression and drug sensitivity data to doxorubicin (DOX) for the human NCI60 cell line panel were publicly available. Corresponding data was obtained and generated for the canine ACC29 cell line panel and canine osteosarcoma tumors via cell viability assays, hospital record search, and microarray gene expression analysis. Differentially expressed genes (DEGs) were identified in reference sets by Significance Analysis of Microarrays or t-test analysis between sensitive and resistant samples to DOX. Probe matching between species was done based on highest target sequence homology or by collapsing probes to gene level by selecting probes with maximum variance between samples. Hierarchical clustering was done using the CIMminer program. DEGs are filtered for genes sharing strong co-expression between reference and co-expression datasets, followed by prediction model building using the Misclassification Penalized Posterior (MiPP) algorithm. Results: The NCI60 and ACC29 panels are similarly sensitive to DOX with log GI50 molar ranges of -8.3 to -5.7 and -8.2 to -5.9, respectively. Samples from ACC29 and COS16 clustered according to DOX sensitivity 67 and 100% correctly when DEGs from NCI60 were used. NCI60-trained models using DEGs co-expressed with the ACC29 were 83% accurate in predicting sensitivity to DOX in ACC29 (p = 0.019, binomial). NCI60-trained models using DEGs co-expressed with COS49 were 67% accurate in predicting response to DOX in COS49 (p = 0.049, binomial, p = 0.045, Log Rank test). Models with NCI60 DEGs but co-expressed and built on COS16 were 73% accurate in COS33 (p = 0.026, binomial, p = 0.001, Log Rank test). ACC29-trained models using DEGs co-expressed with COS49 were 73% accurate in COS49 (p = 0.008, binomial). Models with ACC29 or ACCosteo DEGs but co-expressed and built on COS16 were each 70% accurate in COS33 (p = 0.047, binomial, p = 0.071 and 0.071, Log Rank test). Conclusions: Both human and canine COXEN models were able to accurately predict response of canine osteosarcoma patients to DOX. The best performing model involved identifying DEGs from human cell lines, co-expressing and building the model with dog tumors, suggesting that genomic strategies in canine oncology could benefit from the wealth of available human genomic data, and could provide validation for personalized medicine in human cancer. Citation Format: Jared S. Fowles, Kristen C. Brown, Ann M. Hess, Dawn L. Duval, Daniel L. Gustafson. Gene expression models for predicting doxorubicin response in canine osteosarcoma. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr 2179. doi:10.1158/1538-7445.AM2015-2179

Drug Selection in the Genomic Age: Application of the Coexpression Extrapolation Principle for Drug Repositioning in Cancer Therapy

The use of patient-specific data in drug and dose selection is becoming an increasingly important component in cancer therapy. Basing drug choice on molecular aspects of the tumor is consistent with the identification of cancer as a molecular disease or diseases, even within the same histological type, and its treatment specific to the background from which it arose and exists. Multiple examples exist of single-gene mutations, and over- or underexpression that convey either sensitivity or resistance to a given agent. What about the picture that global gene expression in a cancer presents regarding drug sensitivity or resistance? Coexpression extrapolation (COXEN) is a methodology that acts as a Rosetta stone between patterns of gene expression that correlate to drug responses in vitro and those in tumors of untreated patients to predict chemosensitivity in such tumors even to drugs that are not specifically indicated for that histotype. Further applications of COXEN in drug discovery allow for in silico screens correlating drug response and gene expression in a genetically diverse cell panel to gene expression patterns in a target tumor with the potential for identifying and repurposing compounds. Here we discuss how COXEN is being developed and tested for application in drug selection, repositioning, and repurposing in oncology.

Abstract A30: Interspecies applications of gene expression-based prediction of chemosensitivity

Abstract Introduction: Treatment of cancer based on molecular characteristics of individual tumors is a promising area of research that can lead to improved outcomes. Validation of these new genomic methods is crucial for their success in future clinical trials. The co-expression extrapolation (COXEN) method has been shown to extrapolate data from a reference dataset to accurately predict drug sensitivity in vitro and in vivo in human cancer. Canine cancer models have great translational potential because of their biologic and genetic similarities with humans. Implementing genomic strategies across species allows dogs to take advantage of the wealth of available human data, and humans to take advantage of data from veterinary clinical trials. Our purpose is to explore the ability of COXEN for interspecies extrapolation of human cancer cell line data to generate gene expression models for the prediction of chemosensitivity in canine cancer cell lines and tumors. Methods: Publicly available gene expression and drug sensitivity data for the human NCI60 panel was obtained for doxorubicin (DOX), vinblastine (VBL), carboplatin (CAR), paclitaxel (PTX), lomustine (LOM), and cisplatin (CIS). Gene expression data was publicly obtained or generated via microarray gene expression analysis after RNA extraction of archived tumor samples for canine osteosarcoma (COS49) or from 16 canine cancer cell lines (ACC16). Alamar blue assays were performed to calculate drug sensitivity in the ACC16. Gene signatures predictive of drug sensitivity were identified by comparing the 12 least and most sensitive cell lines from the NCI60 via Significance Analysis of Microarrays (SAM). Human probesets were matched with canine orthologs according to highest sequence homology. A subset of genes from the signatures that shared strong co-expression with a second dataset (ACC16 or COS49) was identified through correlation matrices followed by percentile cutoffs from a permutation-based null distribution. The co-expressed genes were used in the Missclassification Penalized Posterior (MiPP) algorithm to generate models using the NCI60 panel with or without the second dataset for external cross-validation. Drug sensitivity in cell lines or disease free interval (DFI) in tumor patients was predicted from final models. Results: NCI60 models with co-expressed genes with the ACC16 were on average 66% accurate in predicting sensitivity to 6 drugs when built on the NCI60 panel alone (significant by binomial test for LOM (P = 0.0327)) and 82% accurate when using the ACC16 for external cross-validation (significant by binomial test for DOX, CAR, and LOM (P = 0.0461, 0.0327, 0.0005), and by spearman rank correlation for VBL, PTX, and LOM (P = 0.0202, 0.0105, 0.0306)). NCI60 models with co-expressed genes with the COS49 were 70% and 50% accurate for DOX and CAR when built on NCI60 panel and 75% and 66% accurate when using the COS49 for external cross-validation (significant by binomial test for DOX (P = 0.0414)). Conclusions: The results show the potential of the COXEN method in translating known human in vitro chemosensitivity data to both canine cancer cells and tumors. NCI60 models that used canine datasets for external cross-validation were more accurate in most cases, suggesting the need for a canine component during the model building process. Future expansion of the canine cancer panels may lead to improvement with accuracy and robustness in predicting chemosensitivity in dogs. Likewise, incorporating canine models with these genomic strategies in the veterinary clinical trial setting could provide needed validation for personalized medicine in human cancer. Citation Format: Jared S. Fowles, Ann M. Hess, Dawn L. Duval, Douglas H. Thamm, Daniel L. Gustafson. Interspecies applications of gene expression-based prediction of chemosensitivity. [abstract]. In: Proceedings of the AACR Special Conference: The Translational Impact of Model Organisms in Cancer; Nov 5-8, 2013; San Diego, CA. Philadelphia (PA): AACR; Mol Cancer Res 2014;12(11 Suppl):Abstract nr A30.

Retrospective Analysis of Survival Improvement by Molecular Biomarker-Based Personalized Chemotherapy for Recurrent Ovarian Cancer

Aggressive tumors such as epithelial ovarian cancer (EOC) are highly heterogeneous in their therapeutic response, making it difficult to improve overall response by using drugs in unselected patients. The goal of this study was to retrospectively, but independently, examine whether biomarker-based personalized chemotherapy selection could improve survival of EOC patients. Using in vitro drug sensitivity and patient clinical outcome data, we have developed co-expression extrapolation (COXEN) biomarker models for predicting patient response to three standard chemotherapy drugs used to treat advanced EOC: paclitaxel, cyclophosphamide, and topotecan, for which sufficient patient data were available for our modeling and independent validation. Four different cohorts of 783 EOC patients were used in our study, including two cohorts of 499 patients for independent validation. The COXEN predictors for the three drugs independently showed high prediction both for patient short-term therapeutic response and long-term survival for recurrent EOC. We then examined the potential clinical benefit of the simultaneous use of the three drug predictors for a large diverse EOC cohort in a prospective manner, finding that the median overall survival was 21 months longer for recurrent EOC patients who were treated with the predicted most effective chemotherapies. Survival improvement was greater for platinum-sensitive patients if they were treated with the predicted most beneficial drugs. Following the FDA guidelines for diagnostic prediction analysis, our study has retrospectively, yet independently, showed a potential for biomarker-based personalized chemotherapy selection to significantly improve survival of patients in the heterogeneous EOC population when using standard chemotherapies.

Ovarian cancer stem cell-specific gene expression profiling and targeted drug prescreening

Cancer stem cells, with unlimited self-renewal potential and other stem cell characteristics, occur in several types of cancer, including ovarian cancer (OvC). Although CSCs can cause tumor initiation, malignant proliferation, relapse and multi-drug resistance, ways to eliminate them remain unknown. In the present study, we compared ovarian cancer stem cell (OVCSC) expression profiles in normal ovarian surface epithelium and ovarian cells from patients with advanced disease to identify key pathways and specific molecular signatures involved in OVC progression and to prescreen candidate small-molecule compounds with anti-OVCSC activity. Comparison of genome-wide expression profiles of OvC stemness groups with non-stemness controls revealed 6495, 1347 and 509 differentially expressed genes in SDC, SP1 and SP2 groups, respectively, with a cut-off of fold-change set at >1.5 and P<0.05. NAB1 and NPIPL1 were commonly upregulated whereas PROS1, GREB1, KLF9 and MTUS1 were commonly downregulated in all 3 groups. Most differentially expressed genes consistently clustered with molecular functions such as protein receptor binding, kinase activity and chemo-repellent activity. These genes regulate cellular components such as centrosome, plasma membrane receptors, and basal lamina, and may participate in biological processes such as cell cycle regulation, chemoresistance and stemness induction. Key Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways such as ECM receptor, ErbB signaling, endocytosis and adherens junction pathways were enriched. Gene co-expression extrapolation screening by the Connectivity Map revealed several small-molecule compounds (such as SC-560, disulfiram, thapsigargin, esculetin and cinchonine) with potential anti-OVCSC properties targeting OVCSC signature genes. We identified several key CSC features and specific regulation networks in OVCSCs and predicted several small molecules with potential anti-OVCSC pharmacological properties, which may aid the development of OVCSC-specific drugs.

Abstract 2890: NCI60-based gene expression models for predicting chemosensitivity in canine cancer.

Abstract Background: Personalized medicine employing treatment strategies based on molecular characteristics of patient tumors is a promising area of research. Strong experimental validation of such methods is essential before initiating clinical trials involving patients receiving personalized treatment versus standard of care. The recently developed co-expression extrapolation (COXEN) method has proven successful in predicting drug sensitivity in vitro and in vivo in human cancer. Biologic and genetic similarities make canine cancer models highly attractive for human translational research. Our purpose in this study is to explore the COXEN method's utility in a cross-species extrapolation of gene expression models (GEMs) to predict chemosensitivity in canine cancer cells from a human reference set. Materials and Methods: Microarray gene expression and drug sensitivity data for doxorubicin (DOX), vinblastine (VBL), carboplatin (CAR), paclitaxel (PTX), lomustine (CCNU), and cisplatin (CIS) was publicly available for the human NCI60 panel. Gene expression data from 15 canine osteosarcoma (COS) tumors from dogs treated with DOX, CARBO, or CIS was obtained. Microarray gene expression analysis was performed for 16 canine cancer cell lines (ACC16). Sensitivity of the ACC16 to these chemotherapeutics was calculated via Alamar Blue assays. Significance Analysis of Microarrays (SAM) with a false discovery rate of 0.1 was performed on the 12 most sensitive and resistant cell lines from the NCI60 to identify gene expression signatures predicting sensitivity for each drug. Corresponding expression data from the ACC16 was compared to the NCI60 signatures via correlation matrices to identify co-expressed genes for prediction model development using the Misclassification Penalized Posterior (MiPP) algorithm. Results: GI50 ranges for all 6 drugs in the NCI60 panel were comparable to the ranges calculated for the ACC16. COXEN analysis generated GEMs for each drug that on average were 79% accurate in predicting sensitivity in the ACC16. GEMs for VBL, PTX, and CCNU resulted in COXEN prediction scores that significantly correlated with actual GI50 data (R=0.70, 0.78, 0.70 and p=0.02, 0.01, 0.03, Spearman). COXEN GEMs based on co-expressed genes between the NCI60 and COS tumors were 73% accurate in predicting disease free interval (DFI), as opposed to GEMs based on the NCI60 and ACC16, which were 47% accurate in DFI prediction in COS patients. Conclusions: Initial results are encouraging despite the limited sample set for the use of COXEN in cross-species prediction models and for future use in veterinary clinical trials. Future and ongoing studies will expand the ACC16 panel and generate microarray data for 50 additional COS tumor samples. Incorporating dog models in these genomic strategies may provide powerful validation that is needed for the continuing advancement of personalized medicine in cancer. Citation Format: Jared S. Fowles, Ann M. Hess, Dawn L. Duval, Daniel L. Gustafson. NCI60-based gene expression models for predicting chemosensitivity in canine cancer. [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 2890. doi:10.1158/1538-7445.AM2013-2890

Abstract 50: An informatic approach to rational selection of cell lines modeling complex clinicopathologic phenotypes based on gene expression: The COXEN algorithm

Abstract A perennial issue in biomedical research regards validation of the relevance of experimentally tractable model systems to disease characteristics of interest. In the study of many types of cancer, this often operationally means selection of a cell line or a number of cell line models for perturbation within in vitro or in vivo models of disease progression or treatment. However, many classic cell lines are of uncertain provenance, may have been isolated from tumors staged and graded decades ago in a fashion of limited interpretability with regards to current care, or have exhibited drifting phenotypes in intervening years. One example is that of the classic bladder cancer cell line, T24, isolated from a sample of “verified grade III papillomatosis of the bladder” in 1973, which has given rise to isogenically verified metastatic variants like T24T (1). We recently have reported an algorithm, called COXEN (COeXpression ExtrapolatioN), which uses concordant gene expression signatures of complex clinicopathologic phenotypes to infer interpretable and predictable relationships between disparate cell line and tumor populations, despite significantly different global transcriptional differences. While prior reports have used COXEN to predict chemotherapeutic response in clinical trials a priori based on signatures from cell lines (2,3), herein we report an application of the reverse analysis, using signatures of clinicopathologic characteristics in tumors including invasion, nodal metastasis, and poor survival to select from a panel of forty bladder cancer cell lines models reflecting with highest fidelity the transcriptional patterns underlying the parameter of interest. We find that this approach validates some cell line model systems of bladder cancer and proposes new uses for others. This approach, widely adaptable to other tumor histologies, may provide important guidance for selection of appropriate models to test therapies targeted at clinically important phenotypes. (1) Harding MA et al. Cancer Res 2002. (2) Lee JK et al. PNAS 2007. (3) Williams et al. Cancer Res 2009. 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 50. doi:10.1158/1538-7445.AM2011-50

Multigene expression-based predictors for sensitivity to Vorinostat and Velcade in non-small cell lung cancer.

The ability to predict the efficacy of molecularly targeted therapies for non-small cell lung cancer (NSCLC) for an individual patient remains problematic. The purpose of this study was to identify, using a refined "coexpression extrapolation (COXEN)" algorithm with a continuous spectrum of drug activity, tumor biomarkers that predict drug sensitivity and therapeutic efficacy in NSCLC to Vorinostat, a histone deacetylase inhibitor, and Velcade, a proteasome inhibitor. Using our refined COXEN algorithm, biomarker prediction models were discovered and trained for Vorinostat and Velcade based on the in vitro drug activity profiles of nine NSCLC cell lines (NCI-9). Independently, a panel of 40 NSCLC cell lines (UVA-40) were treated with Vorinostat or Velcade to obtain 50% growth inhibition values. Genome-wide expression profiles for both the NCI-9 and UVA-40 cell lines were determined using the Affymetrix HG-U133A platform. Modeling generated multigene expression signatures for Vorinostat (45-gene; P = 0.002) and Velcade (15-gene; P = 0.0002), with one overlapping gene (CFLAR). Examination of Vorinostat gene ontogeny revealed a predilection for cellular replication and death, whereas that of Velcade suggested involvement in cellular development and carcinogenesis. Multivariate regression modeling of the refined COXEN scores significantly predicted the activity of combination therapy in NSCLC cells (P = 0.007). Through the refinement of the COXEN algorithm, we provide an in silico method to generate biomarkers that predict tumor sensitivity to molecularly targeted therapies. Use of this refined COXEN method has significant implications for the a priori examination of targeted therapies to more effectively streamline subsequent clinical trial design and cost.

Highlights of This Issue

The ability to predict the efficacy of molecularly targeted therapies for non-small cell lung cancer (NSCLC) for an individual patient remains problematic. Nagji and colleagues used the refined coexpression extrapolation (COXEN) algorithm to identify tumor biomarkers that predict drug sensitivity and therapeutic efficacy in NSCLC to Vorinostat (histone deacetylase inhibitor) and Velcade (proteasome inhibitor). Importantly, they provide an in silico method to generate biomarkers that predict tumor sensitivity to novel molecularly targeted therapies. Use of this refined COXEN methodology has significant implications for the a priori examination of targeted therapies to more effectively streamline subsequent clinical trial design and cost.Since αvβ3 integrin plays a key role in angiogenesis and metastasis of human tumors, ligands against this integrin are potentially useful in targeted therapy and cancer imaging. Xiao and colleagues report that a novel RGD peptide ligand, LXW7 with a built-in handle, was identified using OBOC combinatorial chemistry. LXW7 demonstrated high targeting efficacy and specificity for the αvβ3 integrin expressed in tumor cells. Moreover, LXW7 permits easier functionalization for payload conjugation without attenuating the binding affinity of the peptide compared with RGD cyclopentapeptide ligands. LXW7, therefore, has great potential as a highly efficient peptide ligand for targeted imaging and drug delivery.Non-small cell lung cancer (NSCLC) with an activating mutation of the epidermal growth factor receptor (EGFR) ultimately develops resistance to EGFR tyrosine kinase inhibitors (TKI). Okamoto and colleagues show that EGFR mutation–positive NSCLC (HCC827) cells engineered to stably overexpress HGF (HCC827-HGF) exhibit gefitinib resistance. TAK-701, a humanized monoclonal antibody to HGF, in combination with gefitinib suppressed their growth in HCC827-HGF cells both in vitro and in vivo. Our findings suggest that the addition of TAK-701 to gefitinib is a promising strategy to overcome EGFR-TKI resistance induced by HGF in NSCLC with an activating EGFR mutation.Aurora A kinase (AAK), a protein kinase critical for tumor proliferation and survival, is a novel target for cancer treatment. MLN8054, an oral small-molecule AAK inhibitor, produces broad preclinical antitumor activity. In this Phase I study, Macarulla and colleagues evaluated safety, pharmacokinetics, and pharmacodynamics in serial tumor and skin biopsies. Total daily doses of 70 or 80 mg led to marked changes in mitotic cell chromosome alignment and spindle bipolarity in some patients. These results provide evidence for the dominant mechanism to inhibit AAK and support ongoing efforts to target this enzyme in the clinical treatment setting.