Independent validation of a model using cell line chemosensitivity to predict response to therapy

A machine learning-based gene signature of response to the novel alkylating agent LP-184 distinguishes its potential tumor indications

BACKGROUND

Non-targeted cytotoxics with anticancer activity are often developed through preclinical stages using response criteria observed in cell lines and xenografts. A panel of the NCI-60 cell lines is frequently the first line to define tumor types that are optimally responsive. Open data on the gene expression of the NCI-60 cell lines, provides a unique opportunity to add another dimension to the preclinical development of such drugs by interrogating correlations with gene expression patterns. Machine learning can be used to reduce the complexity of whole genome gene expression patterns to derive manageable signatures of response. Application of machine learning in early phases of preclinical development is likely to allow a better positioning and ultimate clinical success of molecules. LP-184 is a highly potent novel alkylating agent where the preclinical development is being guided by a dedicated machine learning-derived response signature. We show the feasibility and the accuracy of such a signature of response by accurately predicting the response to LP-184 validated using wet lab derived IC50s on a panel of cell lines.

RESULTS

We applied our proprietary RADR® platform to an NCI-60 discovery dataset encompassing LP-184 IC50s and publicly available gene expression data. We used multiple feature selection layers followed by the XGBoost regression model and reduced the complexity of 20,000 gene expression values to generate a 16-gene signature leading to the identification of a set of predictive candidate biomarkers which form an LP-184 response gene signature. We further validated this signature and predicted response to an additional panel of cell lines. Considering fold change differences and correlation between actual and predicted LP-184 IC50 values as validation performance measures, we obtained 86% accuracy at four-fold cut-off, and a strong (r = 0.70) and significant (p value 1.36e-06) correlation between actual and predicted LP-184 sensitivity. In agreement with the perceived mechanism of action of LP-184, PTGR1 emerged as the top weighted gene.

CONCLUSION

Integration of a machine learning-derived signature of response with in vitro assessment of LP-184 efficacy facilitated the derivation of manageable yet robust biomarkers which can be used to predict drug sensitivity with high accuracy and clinical value.

Prediction of fulvestrant efficacy in patients with advanced breast cancer: retrospective-prospective evaluation of the predictive potential of a multigene expression assay

BACKGROUND

Fulvestrant is a selective oestrogen receptor (ER) degrader used as monotherapy and combination therapy for ER positive HER2 negative advanced breast cancer (ABC) in postmenopausal women. The drug response predictor (DRP), is a mathematical algorithm based on the expression of multiple genes in the tumour. The fulvestrant DRP algorithm has previously shown effect in BC. In this study, we investigated the DRP's potential in predicting fulvestrant benefit.

METHOD

Among 695 patients with ABC prospectively included in a Danish Breast Cancer Cooperative Group (DBCG) cohort we retrospectively included 226 patients who received fulvestrant as monotherapy. The DRP result was based on mRNA extracted from tumour biopsies and analysed using Affymetrix array. Primary endpoint was time to progression (TTP).

RESULTS

For patients who received fulvestrant in line one to four and were previously unexposed to adjuvant endocrine therapy, we identified a hazard ratio (HR) of 0.44 (90% confidence interval (90% CI) upper limit of 1.08, one sided p = 0.066) for a predicted positive vs negative outcome. A weaker association was seen when including patients exposed to adjuvant endocrine treatment or received fulvestrant in fifth or later lines. Exploratory analyses showed that the DRP was efficient when using recent biopsies for DRP estimate and among recently treated patients.

CONCLUSION

The DRP showed a potential in predicting fulvestrant treatment but was not significant in the overall analysis. Use of older biopsies, long-term endocrine treatment and multiple therapies between biopsy used for analysis and fulvestrant treatment, probably affect the predictive accuracy.

Integration of Tumor Genomic Data with Cell Lines Using Multi-dimensional Network Modules Improves Cancer Pharmacogenomics

Leveraging insights from genomic studies of patient tumors is limited by the discordance between these tumors and the cell line models used for functional studies. We integrate omics datasets using functional networks to identify gene modules reflecting variation between tumors and show that the structure of these modules can be evaluated in cell lines to discover clinically relevant biomarkers of therapeutic responses. Applied to breast cancer, we identify 219 gene modules that capture recurrent alterations and subtype patients and quantitate various cell types within the tumor microenvironment. Comparison of modules between tumors and cell lines reveals that many modules composed primarily of gene expression and methylation are poorly preserved. In contrast, preserved modules are highly predictive of drug responses in a manner that is robust and clinically relevant. This work addresses a fundamental challenge in pharmacogenomics that can only be overcome by the joint analysis of patient and cell line data.

Predicting efficacy of epirubicin by a multigene assay in advanced breast cancer within a Danish Breast Cancer Cooperative Group (DBCG) cohort: a retrospective-prospective blinded study

Purpose

Anthracyclines remain a cornerstone in the treatment of primary and advanced breast cancer (BC). This study has evaluated the predictive value of a multigene mRNA-based drug response predictor (DRP) in the treatment of advanced BC with epirubicin. The DRP is a mathematical method combining in vitro sensitivity and gene expression with clinical genetic information from > 3000 clinical tumor samples.

Methods

From a DBCG cohort, 140 consecutive patients were treated with epirubicin between May 1997 and November 2016. After patient informed consent, mRNA was isolated from archival formalin-fixed paraffin-embedded primary breast tumor tissue and analyzed using Affymetrix arrays. Using time to progression (TTP) as primary endpoint, the efficacy of epirubicin was analyzed according to DRP combined with clinicopathological data collected retrospectively from patients’ medical records. Statistical analysis was done using Cox proportional hazards model stratified by treatment line.

Results

Median TTP was 9.3 months. The DRP was significantly associated to TTP (P = 0.03). The hazard ratio for DRP scores differing by 50 percentage points was 0.55 (95% CI –0.93, one-sided). A 75% DRP was associated with a median TTP of 13 months compared to 7 months following a 25% DRP. Multivariate analysis showed that DRP was independent of age and number of metastases.

Conclusion

The current study prospectively validates the predictive capability of DRP regarding epirubicin previously shown retrospectively allowing the patients predicted to be poor responders to choose more effective alternatives. Randomized prospective studies are needed to demonstrate if such an approach will lead to increased overall survival.

Molecular prediction of adjuvant cisplatin efficacy in Non-Small Cell Lung Cancer (NSCLC)-validation in two independent cohorts

INTRODUCTION

Effective predictive biomarkers for selection of patients benefiting from adjuvant platinum-based chemotherapy in non-small cell lung cancer (NSCLC) are needed. Based on a previously validated methodology, molecular profiles of predicted sensitivity in two patient cohorts are presented.

METHODS

The profiles are correlations between in vitro sensitivity to cisplatin and vinorelbine and baseline mRNA expression of the 60 cell lines in the National Cancer Institute panel. An applied clinical samples filter focused the profiles to clinically relevant genes. The profiles were tested on 1) snap-frozen tumors from 133 patients with completely resected stage 1B-2 NSCLC randomized to adjuvant cisplatin and vinorelbine (ACV, n = 71) or no adjuvant treatment (OBS, n = 62) and 2) formalin-fixed paraffin-embedded (FFPE) tumors from 95 patients with completely resected stage 1A-3B NSCLC receiving adjuvant cisplatin and vinorelbine.

RESULTS

The combined cisplatin and vinorelbine profiles showed: 1) univariate Hazard Ratio (HR) for sensitive versus resistant of 0.265 (95% CI:0.079-0.889, p = 0.032) in the ACV cohort and a HR of 0.28 in a multivariate model (95% CI:0.08-1.04, p = 0.0573); 2) significant prediction at 3 year survival from surgery in univariate (HR = 0.138 (95% CI:0.035-0.537), p = 0.004) and multivariate analysis (HR = 0.14 (95% CI:0.030-0.6), p = 0.0081). No discrimination was found in the OBS cohort (HR = 1.328, p = 0.60). The cisplatin predictor alone had similar figures with 1) univariate HR of 0.37 (95% CI:0.12-1.15, p = 0.09) in the ACV cohort and 2) univariate HR of 0.14 (95% CI:0.03-0.59, p = 0.0076) to three years. Functional analysis on the cisplatin profile revealed a group of upregulated genes related to RNA splicing as a part of DNA damage repair and apoptosis.

CONCLUSIONS

Profiles derived from snap-frozen and FFPE NSCLC tissue were prognostic and predictive in the patients that received cisplatin and vinorelbine but not in the cohort that did not receive adjuvant treatment.

Novel biomarkers in cancer: The whole is greater than the sum of its parts

The major issues hampering progress in the treatment of cancer patients are distant metastases and drug resistance to chemotherapy. Metastasis formation is a very complex process, and looking at gene signatures alone is not enough to get deep insight into it. This paper reviews traditional and novel approaches to identify gene signature biomarkers and intratumoural fluid pressure both as a novel way of creating predictive markers and as an obstacle to cancer therapy. Finally recently developed in vitro systems to predict the response of individual patient derived cancer explants to chemotherapy are discussed.

Cell Line Derived 5-FU and Irinotecan Drug-Sensitivity Profiles Evaluated in Adjuvant Colon Cancer Trial Data

PURPOSE

This study evaluates whether gene signatures for chemosensitivity for irinotecan and 5-fluorouracil (5-FU) derived from in vitro grown cancer cell lines can predict clinical sensitivity to these drugs.

METHODS

To test if an irinotecan signature and a SN-38 signature could identify patients who benefitted from the addition of irinotecan to 5-FU, we used gene expression profiles based on cell lines and clinical tumor material. These profiles were applied to expression data obtained from pretreatment formalin fixed paraffin embedded (FFPE) tumor tissue from 636 stage III colon cancer patients enrolled in the PETACC-3 prospective randomized clinical trial. A 5-FU profile developed similarly was assessed by comparing the PETACC-3 cohort with a cohort of 359 stage II colon cancer patients who underwent surgery but received no adjuvant therapy.

RESULTS

There was no statistically significant association between the irinotecan or SN-38 profiles and benefit from irinotecan. The 5-FU sensitivity profile showed a statistically significant association with relapse free survival (RFS) (hazard ratio (HR) = 0.54 (0.41-0.71), p<1e-05) and overall survival (HR = 0.47 (0.34-0.63), p<1e-06) in the PETACC-3 subpopulation. The effect of the 5-FU profile remained significant in a multivariable Cox Proportional Hazards model, adjusting for several relevant clinicopathological parameters. No statistically significant effect of the 5-FU profile was observed in the untreated cohort of 359 patients (relapse free survival, p = 0.671).

CONCLUSION

The irinotecan predictor had no predictive value. The 5-FU predictor was prognostic in stage III patients in PETACC-3 but not in stage II patients with no adjuvant therapy. This suggests a potential predictive ability of the 5-FU sensitivity profile to identify colon cancer patients who may benefit from 5-FU, however, any biomarker predicting benefit for adjuvant 5-FU must be rigorously evaluated in independent cohorts. Given differences between the two study cohorts, the present results should be further validated.

Predicting response to multidrug regimens in cancer patients using cell line experiments and regularised regression models

Background

Patients suffering from cancer are often treated with a range of chemotherapeutic agents, but the treatment efficacy varies greatly between patients. Based on recent popularisation of regularised regression models the goal of this study was to establish workflows for pharmacogenomic predictors of response to standard multidrug regimens using baseline gene expression data and origin specific cell lines. The proposed workflows are tested on diffuse large B-cell lymphoma treated with R-CHOP first-line therapy.

Methods

First, B-cell cancer cell lines were tested successively for resistance towards the chemotherapeutic components of R-CHOP: cyclophosphamide (C), doxorubicin (H), and vincristine (O). Second, baseline gene expression data were obtained for each cell line before treatment. Third, regularised multivariate regression models with cross-validated tuning parameters were used to generate classifier and predictor based resistance gene signatures (REGS) for the combination and individual chemotherapeutic drugs C, H, and O. Fourth, each developed REGS was used to assign resistance levels to individual patients in three clinical cohorts.

Results

Both classifier and predictor based REGS, for the combination CHO, were of prognostic value. For patients classified as resistant towards CHO the risk of progression was 2.33 (95% CI: 1.6, 3.3) times greater than for those classified as sensitive. Similarly, an increase in the predicted CHO resistance index of 10 was related to a 22% (9%, 36%) increased risk of progression. Furthermore, the REGS classifier performed significantly better than the REGS predictor.

Conclusions

The regularised multivariate regression models provide a flexible workflow for drug resistance studies with promising potential. However, the gene expressions defining the REGSs should be functionally validated and correlated to known biomarkers to improve understanding of molecular mechanisms of drug resistance.

Electronic supplementary material

The online version of this article (doi:10.1186/s12885-015-1237-6) contains supplementary material, which is available to authorized users.

Development and blind clinical validation of a microRNA based predictor of response to treatment with R-CHO(E)P in DLBCL

MicroRNAs (miRNA) are a group of short noncoding RNAs that regulate gene expression at the posttranscriptional level. It has been shown that microRNAs are independent predictors of outcome in patients with diffuse large B-cell lymphoma (DLBCL) treated with the drug combination R-CHOP. Based on the measured growth inhibition of 60 human cancer cell lines (NCI60) in the presence of doxorubicine, cyclophosphamide, vincristine and etoposide as well as the baseline microRNA expression of the 60 cell lines, a microRNA based response predictor to CHOP was developed. The response predictor consisting of 20 microRNAs was blindly validated in a cohort of 116 de novo DLBCL patients treated with R-CHOP or R-CHOEP as first line treatment. The predicted sensitivity based on diagnostic FFPE samples matched the clinical response, with decreasing sensitivity in poor responders (P = 0.03). When the International Prognostic Index (IPI) was included in the prediction analysis, the separation between responders and non-responders improved (P = 0.001). Thirteen patients developed relapse, and five patients predicted sensitive to their second and third line treatment survived a median 1194 days, while eight patients predicted not sensitive to their second and third line treatment survived a median 187 days (90% CI: 485 days versus 227 days). Among the latter group it was predicted that four would have been sensitive to another second line treatment than the one they received. The predictions were almost the same when diagnostic biopsies were used as when relapse biopsies were used. These preliminary findings warrant testing in a larger cohort of relapse patients to confirm whether the miRNA based predictor can select the optimal second line treatment and increase survival.

Development and validation of a gene expression score that predicts response to fulvestrant in breast cancer patients

Fulvestrant is a selective estrogen receptor antagonist. Based on the measured growth inhibition of 60 human cancer cell lines (NCI60) in the presence of fulvestrant, as well as the baseline gene expression of the 60 cell lines, a gene expression score that predicts response to fulvestrant was developed. The score is based on 414 genes, 103 of which show increased expression in sensitive cell lines, while 311 show increased expression in the non-responding cell lines. The sensitivity genes primarily sense signaling through estrogen receptor alpha, whereas the resistance genes modulate the PI3K signaling pathway. The latter genes suggest that resistance to fulvestrant can be overcome by drugs targeting the PI3K pathway. The level of this gene expression score and its correlation with fulvestrant response was measured in a panel of 20 breast cancer cell lines. The predicted sensitivity matched the measured sensitivity well (CC = -0.63, P = 0.003). The predictor was applied to tumor biopsies obtained from a Phase II clinical trial. The sensitivity of each patient to treatment with fulvestrant was predicted based on the RNA profile of the biopsy taken before neoadjuvant treatment and without knowledge of the subsequent response. The prediction was then compared to clinical response to show that the responders had a significantly higher sensitivity prediction than the non-responders (P = 0.01). When clinical covariates, tumor grade and estrogen receptor H-score, were included in the prediction, the difference in predicted senstivity between responders and non-responders improved (P = 0.003). Using a pre-defined cutoff to separate patients into predicted sensitive and predicted resistant yielded a positive predictive value of 88% and a negative predictive value of 100% when compared to clinical data. We conclude that pre-screening patients with the new gene expression predictor has the potential to identify those postmenopausal women with locally advanced, estrogen-receptor-positive breast cancer most likely to respond to fulvestrant.