Gene expression profiles as biomarkers for the prediction of chemotherapy drug response in human tumour cells

High-accuracy prediction of colorectal cancer chemotherapy efficacy using machine learning applied to gene expression data

Introduction: FOLFOX and FOLFIRI chemotherapy are considered standard first-line treatment options for colorectal cancer (CRC). However, the criteria for selecting the appropriate treatments have not been thoroughly analyzed. Methods: A newly developed machine learning model was applied on several gene expression data from the public repository GEO database to identify molecular signatures predictive of efficacy of 5-FU based combination chemotherapy (FOLFOX and FOLFIRI) in patients with CRC. The model was trained using 5-fold cross validation and multiple feature selection methods including LASSO and VarSelRF methods. Random Forest and support vector machine classifiers were applied to evaluate the performance of the models. Results and Discussion: For the CRC GEO dataset samples from patients who received either FOLFOX or FOLFIRI, validation and test sets were >90% correctly classified (accuracy), with specificity and sensitivity ranging between 85%-95%. In the datasets used from the GEO database, 28.6% of patients who failed the treatment therapy they received are predicted to benefit from the alternative treatment. Analysis of the gene signature suggests the mechanistic difference between colorectal cancers that respond and those that do not respond to FOLFOX and FOLFIRI. Application of this machine learning approach could lead to improvements in treatment outcomes for patients with CRC and other cancers after additional appropriate clinical validation.

Analysis of Saliva Gene Expression during Head and Neck Cancer Radiotherapy: A Pilot Study

Saliva, a biological fluid, is a promising candidate for novel approaches to prognosis, clinical diagnosis, monitoring and management of patients with both oral and systemic diseases. However, to date, saliva has not been widely investigated as a biomarker for radiation exposure. Since white blood cells are also present in saliva, it should theoretically be possible to investigate the transcriptional biomarkers of radiation exposure classically studied in whole blood. Therefore, we collected whole blood and saliva samples from eight head and neck cancer patients before the start of radiation treatment, at mid-treatment and after treatment. We then used a panel of five genes: BAX, BBC3, CDKN1A, DDB2 and MDM2, designated for assessing radiation dose in whole blood to evaluate gene expression changes that can occur during radiotherapy. The results revealed that the expression of the five genes did not change in whole blood. However, in saliva, CDKN1A and DDB2 were significantly overexpressed at the end, compared to the start, of radiotherapy, and MDM2 was significantly underexpressed between mid-treatment and at the end of treatment. Interestingly, CDKN1A and DDB2 expressions also showed an increasing monotonic relationship with total radiation dose received during radiotherapy. To our knowledge, these results show for the first time the ability to detect gene expression changes in saliva after head and neck cancer radiotherapy, and pave the way for further promising studies validating saliva as a minimally invasive means of biofluid collection to directly measure radiation dose escalation during treatment.

Gene and MicroRNA Expression Are Predictive of Tumor Response in Rectal Adenocarcinoma Patients Treated With Preoperative Chemoradiotherapy

Preoperative chemoradiotherapy (pCRT) followed by surgery is the standard treatment for locally advanced rectal cancer (LARC). However, tumor response to pCRT is not uniform, and there are no effective predictive methods. This study investigated whether specific gene and miRNA expression are associated with tumor response to pCRT. Tissue biopsies were obtained from patients before pCRT and resection. Gene and miRNA expression were analyzed using a one-color microarray technique that compares signatures between responders (R) and non-responders (NR), as measured based on tumor regression grade. Two groups composed of 38 "exploration cohort" and 21 "validation cohort" LARC patients were considered for a total of 32 NR and 27 R patients. In the first cohort, using SAM Two Class analysis, 256 genes and 29 miRNAs that were differentially expressed between the NR and R patients were identified. The anti-correlation analysis showed that the same 8 miRNA interacted with different networks of transcripts. The miR-630 appeared only with the NR patients and was anti-correlated with a single transcript: RAB5B. After PAM, the following eight transcripts were strong predictors of tumor response: TMEM188, ITGA2, NRG, TRAM1, BCL2L13, MYO1B, KLF7, and GTSE1. Using this gene set, an unsupervised cluster analysis was applied to the validation cohort and correctly assigned the patients to the NR or R group with 85.7% accuracy, 90% sensitivity, and 82% specificity. All three parameters reached 100% when both cohorts were considered together. In conclusion, gene and miRNA expression profiles may be helpful for predicting response to pCRT in LARC patients. J. Cell. Physiol. 232: 426-435, 2017. © 2016 Wiley Periodicals, Inc.

Genomic approach towards personalized anticancer drug therapy

Stratification of patients for multidrug response is a promising strategy for cancer treatment. Genome-based prediction models have great potential for this purpose because the extent of drug sensitivity may be attributed to the heterogeneity of the underlying genetic characteristics of cancer. However, microarray data is difficult to analyze and is not reproducible. Several machine-learning algorithms have therefore been developed in a repeatable manner. Random forests algorithm, which uses an ensemble approach based on classification and regression trees, appears to be superior for predicting multidrug sensitivity. This is because ensemble methods are more effective when there are much more predictors than samples. Here, we review recent advances in the development of classification algorithms using microarray technology for prediction of anticancer sensitivity, discuss the availability of ensemble methods for prediction models, and present data regarding the identification of potential responders to FOLFOX therapy using random forests algorithm.

Potential responders to FOLFOX therapy for colorectal cancer by Random Forests analysis

Background:

Molecular characterisation using gene-expression profiling will undoubtedly improve the prediction of treatment responses, and ultimately, the clinical outcome of cancer patients.

Methods:

To establish the procedures to identify responders to FOLFOX therapy, 83 colorectal cancer (CRC) patients including 42 responders and 41 non-responders were divided into training (54 patients) and test (29 patients) sets. Using Random Forests (RF) algorithm in the training set, predictor genes for FOLFOX therapy were identified, which were applied to test samples and sensitivity, specificity, and out-of-bag classification accuracy were calculated.

Results:

In the training set, 22 of 27 responders (81.4% sensitivity) and 23 of 27 non-responders (85.1% specificity) were correctly classified. To improve the prediction model, we removed the outliers determined by RF, and the model could correctly classify 21 of 23 responders (91.3%) and 22 of 23 non-responders (95.6%) in the training set, and 80.0% sensitivity and 92.8% specificity, with an accuracy of 69.2% in 29 independent test samples.

Conclusion:

Random Forests on gene-expression data for CRC patients was effectively able to stratify responders to FOLFOX therapy with high accuracy, and use of pharmacogenomics in anticancer therapy is the first step in planning personalised therapy.

Promise and progress in environmental genomics: a status report on the applications of gene expression-based microarray studies in ecologically relevant fish species

The advent of any new technology is typically met with great excitement. So it was a few years ago, when the combination of advances in sequencing technology and the development of microarray technology made measurements of global gene expression in ecologically relevant species possible. Many of the review papers published around that time promised that these new technologies would revolutionize environmental biology as they had revolutionized medicine and related fields. A few years have passed since these technological advancements have been made, and the use of microarray studies in non-model fish species has been adopted in many laboratories internationally. Has the relatively widespread adoption of this technology really revolutionized the fields of environmental biology, including ecotoxicology, aquaculture and ecology, as promised? Or have these studies merely become a novelty and a potential distraction for scientists addressing environmentally relevant questions? In this review, the promises made in early review papers, in particular about the advances that the use of microarrays would enable, are summarized; these claims are compared to the results of recent studies to determine whether the forecasted changes have materialized. Some applications, as discussed in the paper, have been realized and have led to advances in their field, others are still under development.

Identification of biomarkers in human head and neck tumor cell lines that predict for in vitro sensitivity to gefitinib

Potential biomarkers were identified for in vitro sensitivity to the epidermal growth factor receptor (EGFR) tyrosine kinase inhibitor gefitinib in head and neck cancer. Gefitinib sensitivity was determined in cell lines, followed by transcript profiling coupled with a novel pathway analysis approach. Eleven cell lines were highly sensitive to gefitinib (inhibitor concentration required to give 50% growth inhibition [GI(50)] < 1 microM), three had intermediate sensitivity (GI(50) 1-7 microM), and six were resistant (GI(50) > 7 microM); an exploratory principal component analysis revealed a separation between the genomic profiles of sensitive and resistant cell lines. Subsequently, a hypothesis-driven analysis of Affymetrix data (Affymetrix, Inc., Santa Clara, CA, USA) revealed higher mRNA levels for E-cadherin (CDH1); transforming growth factor, alpha (TGF-alpha); amphiregulin (AREG); FLJ22662; EGFR; p21-activated kinase 6 (PAK6); glutathione S-transferase Pi (GSTP1); and ATP-binding cassette, subfamily C, member 5 (ABCC5) in sensitive versus resistant cell lines. A hypothesis-free analysis identified 46 gene transcripts that were strongly differentiated, seven of which had a known association with EGFR and head and neck cancer (human EGF receptor 3 [HER3], TGF-alpha, CDH1, EGFR, keratin 16 [KRT16], fibroblast growth factor 2 [FGF2], and cortactin [CTTN]). Polymerase chain reaction (PCR) and enzyme-linked immunoabsorbant assay analysis confirmed Affymetrix data, and EGFR gene mutation, amplification, and genomic gain correlated strongly with gefitinib sensitivity. We identified biomarkers that predict for in vitro responsiveness to gefitinib, seven of which have known association with EGFR and head and neck cancer. These in vitro predictive biomarkers may have potential utility in the clinic and warrant further investigation.

Induction of 1C aldoketoreductases and other drug dose-dependent genes upon acquisition of anthracycline resistance

OBJECTIVES

Recent studies suggest that tumor cells overexpressing aldoketoreductases (AKRs) exhibit increased resistance to DNA damaging agents such as anthracyclines. AKRs may induce resistance to the anthracycline doxorubicin by catalyzing its conversion to the less toxic 13-hydroxy metabolite doxorubicinol. However, it has not been established whether during selection for anthracycline resistance, AKR overexpression in tumor cells can be correlated with the onset or magnitude of drug resistance and with appreciable conversion of anthracyclines to 13-hydroxy metabolites.

METHODS AND FINDINGS

Through microarray and quantitative polymerase chain reaction studies involving rigid selection criteria and both correlative discriminate statistics and time-course models, we have identified several genes whose expression can be correlated with the onset and/or magnitude of anthracycline resistance, including AKR1C2 and AKR1C3. Also associated with the onset or magnitude of anthracycline resistance were genes involved in drug transport (ABCB1, ABCC1), cell signaling and transcription (RDC1, CXCR4), cell proliferation or apoptosis (BMP7, CAV1), protection from reactive oxygen species (AKR1C2, AKR1C3, FTL, FTH, TXNRD1, MT2A), and structural or immune system proteins (IFI30, STMN1). As expected, doxorubicin-resistant and epirubicin-resistant cells exhibited higher levels of doxorubicinol than wild-type cells, although at insufficient levels to account for significant drug resistance. Nevertheless, an inhibitor of Akr1c2 (5beta-cholanic acid) almost completely restored sensitivity to doxorubicin in ABCB1-deficient doxorubicin-resistant cells, while having no effect on ABCB1-expressing epirubicin-resistant cells.

CONCLUSION

Taken together, we show for the first time that a variety of genes (particularly redox genes such as AKR1C2 and AKR1C3) can be temporally and causally correlated with the acquisition of anthracycline resistance in breast tumor cells.

Antitumor effects and biomarkers of activity of AZD0530, a Src inhibitor, in pancreatic cancer

PURPOSE

To determine the efficacy of AZD0530, an orally active small molecule Src inhibitor, in human pancreatic cancer xenografts and to seek biomarkers predictive of activity.

EXPERIMENTAL DESIGN

Sixteen patient-derived pancreatic cancer xenografts from the PancXenoBank collection at Johns Hopkins were treated with AZD0530 (50 mg/kg/day, p.o.) for 28 days. Baseline gene expression profiles of differently expressed genes in 16 tumors by Affymetrix U133 Plus 2.0 gene array were used to predict AZD0530 sensitivity in an independent group of eight tumors using the K-Top Scoring Pairs (K-TSP) method.

RESULTS

Three patient tumors of 16 were found to be sensitive to AZD0530, defined as tumor growth <50% compared with control tumors (100%). Western blot and/or immunohistochemistry results showed that AZD0530 administration resulted in the down-regulation of Src, FAK, p-FAK, p-paxillin, p-STAT-3, and XIAP in sensitive tumor xenografts compared with control tumors. The K-TSP classifier identified one gene pair (LRRC19 and IGFBP2) from the 16 training cases based on a decision rule. The classifier achieved 100% and 83.3% of sensitivity and specificity in an independent test set that consists of eight xenograft cases.

CONCLUSIONS

AZD0530 treatment significantly inhibits the tumor growth in a subset of human pancreatic tumor xenografts. One gene pair (LRRC19 and IGFBP2) identified by the K-TSP classifier has high predictive power for AZD0530 sensitivity, suggesting the potential for this gene pair as biomarker for pancreatic tumor sensitivity to AZD0530.

Potent antitumor activity of the anti-CD19 auristatin antibody drug conjugate hBU12-vcMMAE against rituximab-sensitive and -resistant lymphomas

Despite major advances in the treatment of non-Hodgkin lymphoma (NHL), including the use of chemotherapeutic agents and the anti-CD20 antibody rituximab, the majority of patients eventually relapse, and salvage treatments with non-cross-resistant compounds are needed to further improve patient survival. Here, we evaluated the antitumor effects of the microtubule destabilizing agent monomethyl auristatin E (MMAE) conjugated to the humanized anti-CD19 antibody hBU12 via a protease-sensitive valine-citrulline (vc) dipeptide linker. hBU12-vcMMAE induced potent tumor cell killing against rituximab-sensitive and -resistant NHL cell lines. CD19 can form heterodimers with CD21, and high levels of CD21 were reported to interfere negatively with the activity of CD19-targeted therapeutics. However, we observed comparable internalization, intracellular trafficking, and drug release in CD21(low) and CD21(high), rituximab-sensitive and -refractory lymphomas treated with hBU12-vcMMAE. Furthermore, high rates of durable regressions in mice implanted with these tumors were observed, suggesting that both rituximab resistance and CD21 expression levels do not impact on the activity of hBU12-vcMMAE. Combined, our data suggest that hBU12-vcMMAE may represent a promising addition to the treatment options for rituximab refractory NHL and other hematologic malignancies, including acute lymphoblastic leukemia.

Genome-wide studies in thyroid neoplasia

There is much interest in the application of genome biology to the field of thyroid neoplasia, despite the relatively low mortality rate associated with thyroid cancer in general. The principal reason for this interest is that the field of thyroid neoplasia stands to benefit from the application of genomic information to address a variety of pathologic and clinical issues. In addition to practical patient care issues, there is an excellent opportunity of expand the basic understanding of thyroid carcinogenesis. In this article, the most relevant genomic work on thyroid tumors performed to date is reviewed along with some general comments about the potential impact of genomic biology on thyroid pathology and the management of patients with thyroid nodules and cancer.

Loss of putative tumor suppressor EI24/PIG8 confers resistance to etoposide

Expression of p53-target gene EI24/PIG8 is lost in invasive breast cancers, suggesting that EI24/PIG8 is a tumor suppressor that prevents tumor spreading, and partially mediates p53-attributed tumor suppressor activity. EI24/PIG8 also has pro-apoptotic activity indicating that loss of EI24/PIG8 may modulate sensitivity to chemotherapy. Here it is demonstrated that suppression of EI24/PIG8 in fibroblasts and breast cancer cells significantly inhibits the apoptotic response to etoposide treatment. These findings suggest that loss of EI24/PIG8 contributes significantly to resistance of cells to chemotherapeutic agents that function through p53, and identify the EI24/PIG8 status as a potentially new prognostic marker of chemotherapy responsiveness.

Prediction potential of candidate biomarker sets identified and validated on gene expression data from multiple datasets

BACKGROUND

Independently derived expression profiles of the same biological condition often have few genes in common. In this study, we created populations of expression profiles from publicly available microarray datasets of cancer (breast, lymphoma and renal) samples linked to clinical information with an iterative machine learning algorithm. ROC curves were used to assess the prediction error of each profile for classification. We compared the prediction error of profiles correlated with molecular phenotype against profiles correlated with relapse-free status. Prediction error of profiles identified with supervised univariate feature selection algorithms were compared to profiles selected randomly from a) all genes on the microarray platform and b) a list of known disease-related genes (a priori selection). We also determined the relevance of expression profiles on test arrays from independent datasets, measured on either the same or different microarray platforms.

RESULTS

Highly discriminative expression profiles were produced on both simulated gene expression data and expression data from breast cancer and lymphoma datasets on the basis of ER and BCL-6 expression, respectively. Use of relapse-free status to identify profiles for prognosis prediction resulted in poorly discriminative decision rules. Supervised feature selection resulted in more accurate classifications than random or a priori selection, however, the difference in prediction error decreased as the number of features increased. These results held when decision rules were applied across-datasets to samples profiled on the same microarray platform.

CONCLUSION

Our results show that many gene sets predict molecular phenotypes accurately. Given this, expression profiles identified using different training datasets should be expected to show little agreement. In addition, we demonstrate the difficulty in predicting relapse directly from microarray data using supervised machine learning approaches. These findings are relevant to the use of molecular profiling for the identification of candidate biomarker panels.