Model selection for prognostic time-to-event gene signature discovery with applications in early breast cancer data.

Model selection between competing models is a key consideration in the discovery of prognostic multigene signatures. The use of appropriate statistical performance measures as well as verification of biological significance of the signatures is imperative to maximise the chance of external validation of the generated signatures. Current approaches in time-to-event studies often use only a single measure of performance in model selection, such as logrank test p-values, or dichotomise the follow-up times at some phase of the study to facilitate signature discovery. In this study we improve the prognostic signature discovery process through the application of the multivariate partial Cox model combined with the concordance index, hazard ratio of predictions, independence from available clinical covariates and biological enrichment as measures of signature performance. The proposed framework was applied to discover prognostic multigene signatures from early breast cancer data. The partial Cox model combined with the multiple performance measures were used in both guiding the selection of the optimal panel of prognostic genes and prediction of risk within cross validation without dichotomising the follow-up times at any stage. The signatures were successfully externally cross validated in independent breast cancer datasets, yielding a hazard ratio of 2.55 [1.44, 4.51] for the top ranking signature.

A gene expression map of Arabidopsis thaliana development.

Regulatory regions of plant genes tend to be more compact than those of animal genes, but the complement of transcription factors encoded in plant genomes is as large or larger than that found in those of animals. Plants therefore provide an opportunity to study how transcriptional programs control multicellular development. We analyzed global gene expression during development of the reference plant Arabidopsis thaliana in samples covering many stages, from embryogenesis to senescence, and diverse organs. Here, we provide a first analysis of this data set, which is part of the AtGenExpress expression atlas. We observed that the expression levels of transcription factor genes and signal transduction components are similar to those of metabolic genes. Examining the expression patterns of large gene families, we found that they are often more similar than would be expected by chance, indicating that many gene families have been co-opted for specific developmental processes.

Accurate molecular characterization of formalin-fixed, paraffin-embedded tissues by microRNA expression profiling.

Formalin-fixed, paraffin-embedded tissues are an invaluable tool for biomarker discovery and validation. As these archived specimens are not always compatible with modern genomic techniques such as gene expression arrays, we assessed the use of microRNA (miRNA) as an alternative means for the reliable molecular characterization of formalin-fixed, paraffin-embedded tissues. Expression profiling using two different microarray platforms and multiple mouse and human formalin-fixed, paraffin-embedded tissue types resulted in the correlation ratios of miRNA expression levels between frozen and fixed tissue pairs ranging from 0.82 to 0.99, depending on the cellular heterogeneity of the tissue type. The same miRNAs were identified as differentially expressed between tissues using both fixed and frozen specimens. While formalin fixation time had only marginal effects on microarray performance, extended storage times for tissue blocks (up to 11 years) resulted in a gradual loss of detection of miRNAs expressed at low levels. Method reproducibility and accuracy were also evaluated in two different tissues stored for different lengths of time. The technical variation between full process replicates, including independent RNA isolation methods, was approximately 5%, and the correlation of expression levels between microarray and real-time quantitative reverse transcriptase polymerase chain reaction was 0.98. Together, these data demonstrate that miRNA expression profiling is an accurate and robust method for the molecular analysis of archived clinical specimens, potentially extending the use of miRNAs as new diagnostic, prognostic, and treatment response biomarkers.

Analyzing micro-RNA expression using microarrays.

The discovery of micro-RNAs (miRNAs) and the growing appreciation of the importance of micro-RNAs in the regulation of gene expression are driving increasing interest in miRNA expression profiling. Early studies have suggested prominent roles for these genetically encoded regulatory molecules in a variety of normal biological processes and diseases, particularly cancer. However, the field of miRNA expression profiling is in its infancy. Several factors, including the small size, the unknown but limited number of miRNAs, and the tissue-to-tissue and tissue-to-disease state variability in miRNA expression, make the adaptation of microarray technology to the evaluation of miRNA expression nontrivial. This chapter describes the unique features of miRNA microarray experiments and analysis and provides a case study demonstrating our approach to miRNA expression analysis.

Association Between Results of a Gene Expression Signature Assay and Recurrence-Free Interval in Patients With Stage II Colon Cancer in Cancer and Leukemia Group B 9581 (Alliance).

PURPOSE: Conventional staging methods are inadequate to identify patients with stage II colon cancer (CC) who are at high risk of recurrence after surgery with curative intent. ColDx is a gene expression, microarray-based assay shown to be independently prognostic for recurrence-free interval (RFI) and overall survival in CC. The objective of this study was to further validate ColDx using formalin-fixed, paraffin-embedded specimens collected as part of the Alliance phase III trial, C9581. PATIENTS AND METHODS: C9581 evaluated edrecolomab versus observation in patients with stage II CC and reported no survival benefit. Under an initial case-cohort sampling design, a randomly selected subcohort (RS) comprised 514 patients from 901 eligible patients with available tissue. Forty-nine additional patients with recurrence events were included in the analysis. Final analysis comprised 393 patients: 360 RS (58 events) and 33 non-RS events. Risk status was determined for each patient by ColDx. The Self-Prentice method was used to test the association between the resulting ColDx risk score and RFI adjusting for standard prognostic variables. RESULTS: Fifty-five percent of patients (216 of 393) were classified as high risk. After adjustment for prognostic variables that included mismatch repair (MMR) deficiency, ColDx high-risk patients exhibited significantly worse RFI (multivariable hazard ratio, 2.13; 95% CI, 1.3 to 3.5; P < .01). Age and MMR status were marginally significant. RFI at 5 years for patients classified as high risk was 82% (95% CI, 79% to 85%), compared with 91% (95% CI, 89% to 93%) for patients classified as low risk. CONCLUSION: ColDx is associated with RFI in the C9581 subsample in the presence of other prognostic factors, including MMR deficiency. ColDx could be incorporated with the traditional clinical markers of risk to refine patient prognosis.

Identification and validation of an anthracycline/cyclophosphamide-based chemotherapy response assay in breast cancer.

BACKGROUND: There is no method routinely used to predict response to anthracycline and cyclophosphamide-based chemotherapy in the clinic; therefore patients often receive treatment for breast cancer with no benefit. Loss of the Fanconi anemia/BRCA (FA/BRCA) DNA damage response (DDR) pathway occurs in approximately 25% of breast cancer patients through several mechanisms and results in sensitization to DNA-damaging agents. The aim of this study was to develop an assay to detect DDR-deficient tumors associated with loss of the FA/BRCA pathway, for the purpose of treatment selection. METHODS: DNA microarray data from 21 FA patients and 11 control subjects were analyzed to identify genetic processes associated with a deficiency in DDR. Unsupervised hierarchical clustering was then performed using 60 BRCA1/2 mutant and 47 sporadic tumor samples, and a molecular subgroup was identified that was defined by the molecular processes represented within FA patients. A 44-gene microarray-based assay (the DDR deficiency assay) was developed to prospectively identify this subgroup from formalin-fixed, paraffin-embedded samples. All statistical tests were two-sided. RESULTS: In a publicly available independent cohort of 203 patients, the assay predicted complete pathologic response vs residual disease after neoadjuvant DNA-damaging chemotherapy (5-fluorouracil, anthracycline, and cyclophosphamide) with an odds ratio of 3.96 (95% confidence interval [Cl] =1.67 to 9.41; P = .002). In a new independent cohort of 191 breast cancer patients treated with adjuvant 5-fluorouracil, epirubicin, and cyclophosphamide, a positive assay result predicted 5-year relapse-free survival with a hazard ratio of 0.37 (95% Cl = 0.15 to 0.88; P = .03) compared with the assay negative population. CONCLUSIONS: A formalin-fixed, paraffin-embedded tissue-based assay has been developed and independently validated as a predictor of response and prognosis after anthracycline/cyclophosphamide-based chemotherapy in the neoadjuvant and adjuvant settings. These findings warrant further validation in a prospective clinical study.

Implications for powering biomarker discovery studies.

This study examined variations in gene expression between FFPE blocks within tumors of individual patients. Microarray data were used to measure tumor heterogeneity within and between patients and disease states. Data were used to determine the number of samples needed to power biomarker discovery studies. Bias and variation in gene expression were assessed at the intrapatient and interpatient levels and between adenocarcinoma and squamous samples. A mixed-model analysis of variance was fitted to gene expression data and model signatures to assess the statistical significance of observed variations within and between samples and disease states. Sample size analysis, adjusted for sample heterogeneity, was used to determine the number of samples required to support biomarker discovery studies. Variation in gene expression was observed between blocks taken from a single patient. However, this variation was considerably less than differences between histological characteristics. This degree of block-to-block variation still permits biomarker discovery using either macrodissected tumors or whole FFPE sections, provided that intratumor heterogeneity is taken into account. Failure to consider intratumor heterogeneity may result in underpowered biomarker studies that may result in either the generation of longer gene signatures or the inability to identify a viable biomarker. Moreover, the results of this study indicate that a single biopsy sample is suitable for applying a biomarker in non-small-cell lung cancer.

The MicroArray Quality Control (MAQC)-II study of common practices for the development and validation of microarray-based predictive models.

Gene expression data from microarrays are being applied to predict preclinical and clinical endpoints, but the reliability of these predictions has not been established. In the MAQC-II project, 36 independent teams analyzed six microarray data sets to generate predictive models for classifying a sample with respect to one of 13 endpoints indicative of lung or liver toxicity in rodents, or of breast cancer, multiple myeloma or neuroblastoma in humans. In total, >30,000 models were built using many combinations of analytical methods. The teams generated predictive models without knowing the biological meaning of some of the endpoints and, to mimic clinical reality, tested the models on data that had not been used for training. We found that model performance depended largely on the endpoint and team proficiency and that different approaches generated models of similar performance. The conclusions and recommendations from MAQC-II should be useful for regulatory agencies, study committees and independent investigators that evaluate methods for global gene expression analysis.