Paul Geeleher

Clinical drug response can be predicted using baseline gene expression levels and in vitro drug sensitivity in cell lines

We demonstrate a method for the prediction of chemotherapeutic response in patients using only before-treatment baseline tumor gene expression data. First, we fitted models for whole-genome gene expression against drug sensitivity in a large panel of cell lines, using a method that allows every gene to influence the prediction. Following data homogenization and filtering, these models were applied to baseline expression levels from primary tumor biopsies, yielding an in vivo drug sensitivity prediction. We validated this approach in three independent clinical trial datasets, and obtained predictions equally good, or better than, gene signatures derived directly from clinical data.

Gene-set analysis is severely biased when applied to genome-wide methylation data

MOTIVATION DNA methylation is an epigenetic mark that can stably repress gene expression. Because of its biological and clinical significance, several methods have been developed to compare genome-wide patterns of methylation between groups of samples. The application of gene set analysis to identify relevant groups of genes that are enriched for differentially methylated genes is often a major component of the analysis of these data. This can be used, for example, to identify processes or pathways that are perturbed in disease development. We show that gene-set analysis, as it is typically applied to genome-wide methylation assays, is severely biased as a result of differences in the numbers of CpG sites associated with different classes of genes and gene promoters. RESULTS We demonstrate this bias using published data from a study of differential CpG island methylation in lung cancer and a dataset we generated to study methylation changes in patients with long-standing ulcerative colitis. We show that several of the gene sets that seem enriched would also be identified with randomized data. We suggest two existing approaches that can be adapted to correct the bias. Accounting for the bias in the lung cancer and ulcerative colitis datasets provides novel biological insights into the role of methylation in cancer development and chronic inflammation, respectively. Our results have significant implications for many previous genome-wide methylation studies that have drawn conclusions on the basis of such strongly biased analysis. CONTACT cathal.seoighe@nuigalway.ie SUPPLEMENTARY INFORMATION Supplementary data are available at Bioinformatics online.

Integrative analyses of genetic variation, epigenetic regulation, and the transcriptome to elucidate the biology of platinum sensitivity

BackgroundUsing genome-wide genetic, gene expression, and microRNA expression (miRNA) data, we developed an integrative approach to investigate the genetic and epigenetic basis of chemotherapeutic sensitivity.ResultsThrough a sequential multi-stage framework, we identified genes and miRNAs whose expression correlated with platinum sensitivity, mapped these to genomic loci as quantitative trait loci (QTLs), and evaluated the associations between these QTLs and platinum sensitivity. A permutation analysis showed that top findings from our approach have a much lower false discovery rate compared to those from a traditional GWAS of drug sensitivity. Our approach identified five SNPs associated with 10 miRNAs and the expression level of 15 genes, all of which were associated with carboplatin sensitivity. Of particular interest was one SNP (rs11138019), which was associated with the expression of both miR-30d and the gene ABCD2, which were themselves correlated with both carboplatin and cisplatin drug-specific phenotype in the HapMap samples. Functional study found that knocking down ABCD2 in vitro led to increased apoptosis in ovarian cancer cell line SKOV3 after cisplatin treatment. Over-expression of miR-30d in vitro caused a decrease in ABCD2 expression, suggesting a functional relationship between the two.ConclusionsWe developed an integrative approach to the investigation of the genetic and epigenetic basis of human complex traits. Our approach outperformed standard GWAS and provided hints at potential biological function. The relationships between ABCD2 and miR-30d, and ABCD2 and platin sensitivity were experimentally validated, suggesting a functional role of ABCD2 and miR-30d in sensitivity to platinating agents.

The regulatory effect of miRNAs is a heritable genetic trait in humans

BackgroundmicroRNAs (miRNAs) have been shown to regulate the expression of a large number of genes and play key roles in many biological processes. Several previous studies have quantified the inhibitory effect of a miRNA indirectly by considering the expression levels of genes that are predicted to be targeted by the miRNA and this approach has been shown to be robust to the choice of prediction algorithm. Given a gene expression dataset, Cheng et al. defined the regulatory effect score (RE-score) of a miRNA as the difference in the gene expression rank of targets of the miRNA compared to non-targeted genes.ResultsUsing microarray data from parent-offspring trios from the International HapMap project, we show that the RE-score of most miRNAs is correlated between parents and offspring and, thus, inter-individual variation in RE-score has a genetic component in humans. Indeed, the mean RE-score across miRNAs is correlated between parents and offspring, suggesting genetic differences in the overall efficiency of the miRNA biogenesis pathway between individuals. To explore the genetics of this quantitative trait further, we carried out a genome-wide association study of the mean RE-score separately in two HapMap populations (CEU and YRI). No genome-wide significant associations were discovered; however, a SNP rs17409624, in an intron of DROSHA, was significantly associated with mean RE-score in the CEU population following permutation-based control for multiple testing based on all SNPs mapped to the canonical miRNA biogenesis pathway; of 244 individual miRNA RE-scores assessed in the CEU, 214 were associated (p < 0.05) with rs17409624. The SNP was also nominally significantly associated (p = 0.04) with mean RE-score in the YRI population. Interestingly, the same SNP was associated with 17 (8.5% of all expressed) miRNA expression levels in the CEU. We also show here that the expression of the targets of most miRNAs is more highly correlated with global changes in miRNA regulatory effect than with the expression of the miRNA itself.ConclusionsWe present evidence that miRNA regulatory effect is a heritable trait in humans and that a polymorphism of the DROSHA gene contributes to the observed inter-individual differences.

Consistency in large pharmacogenomic studies

Haibe-Kains et al.1 reported inconsistency between two large-scale pharmacogenomic studies—the Cancer Cell Line Encyclopedia (CCLE)2 and the Cancer Genome Project (CGP)3. Upon careful analysis of the same data, we have come to quite different and much more positive conclusions. Here we highlight the most important reasons for this. There is a Reply to this Comment by Safikhani, Z. et al. Nature 540, http://dx.doi.org/10.1038/nature19839 (2016). To assess the concordance of two large studies of the efficacy of cancer drugs, Haibe-Kains et al.1 compared the correlation in drug sensitivity measures with correlation in gene expression measured on the same human cancer cell lines. The authors reported correlation ‘between’ cell lines for gene expression but, inconsistently, ‘across’ cell lines for drug sensitivity (see Methods). On re-analysis, we found much higher correlations between cell lines than across cell lines for both gene expression and drug sensitivity measures (median Spearman’s rank correlation coefficient (rs) = 0.88 between cell lines, rs = 0.56 across cell lines for expression; median rs = 0.62 between cell lines and rs = 0.35 across cell lines for area under the curve (AUC), a drug sensitivity measure). Thus, by correcting this inconsistency, the correlations for expression and drug sensitivity data were far more similar than was originally reported, which markedly undermines the authors’ interpretation of the relative quality of expression and drug sensitivity datasets. In addition, the fundamental issue is that the authors’ reported Spearman’s correlation coefficients do not fairly reflect the concordance of drug sensitivity between the studies, because of the lack of variability in drug response, which arises owing to the highly targeted nature of many of the drugs assessed. To see why correlation is not an appropriate measure of biological concordance for these data, consider the hypothetical example of a drug that is not effective against any cell lines, which is a possibility for an experimental drug. In such a case, the randomly fluctuating measurement error, which is inherent in biological assays, will dominate over the non-existent biological variability, meaning that there could be no expectation of correlation between repeated measures of drug sensitivity—assuming other experimental variables are held constant. In this study, many of the drugs were highly targeted agents, which by design require specific, and often rare, molecular targets for response (see Supplementary Table 1). Consider nilotinib, which targets the BCR-ABL1 fusion gene and was suggested in ref. 1 to exhibit ‘poor consistency’ between CGP and CCLE (rs = 0.1 for AUC). In CGP, BCR-ABL1 status was reported to be strongly associated with drug sensitivity (P = 2.54 × 10−65), accurately reflecting the known biology. BCR-ABL1 status was not reported by CCLE; however, upon re-analysis we identified three BCR-ABL1-positive cell lines among the 189 nilotinib-treated cell lines that overlapped CGP, and these were also the three most sensitive samples (P = 9 × 10−7). Hence, despite the fact that these drug sensitivity data were accurately recapitulating biological expectations in both studies, the authors’ criteria classified nilotinib sensitivity incorrectly. Of the 577 cell lines screened in CGP, 573 do not contain the nilotinib target, that is, the BCR-ABL1 fusion gene. Thus, given (as expected) no drug response in almost all cell lines screened (median AUC across all cell lines = 0.99; AUC of 1 represents no drug response; Fig. 1a, Supplementary Table 1), there was little biological variability across most of the cell lines, resulting in low correlation between the repeated measurements made by CCLE and CGP, despite clearly concordant results. Similarly, most other drugs that the authors compared were also targeted agents, meaning that this lack of drug response was common; for 10 of the 15 drugs, the median

Discovering novel pharmacogenomic biomarkers by imputing drug response in cancer patients from large genomics studies

Obtaining accurate drug response data in large cohorts of cancer patients is very challenging; thus, most cancer pharmacogenomics discovery is conducted in preclinical studies, typically using cell lines and mouse models. However, these platforms suffer from serious limitations, including small sample sizes. Here, we have developed a novel computational method that allows us to impute drug response in very large clinical cancer genomics data sets, such as The Cancer Genome Atlas (TCGA). The approach works by creating statistical models relating gene expression to drug response in large panels of cancer cell lines and applying these models to tumor gene expression data in the clinical data sets (e.g., TCGA). This yields an imputed drug response for every drug in each patient. These imputed drug response data are then associated with somatic genetic variants measured in the clinical cohort, such as copy number changes or mutations in protein coding genes. These analyses recapitulated drug associations for known clinically actionable somatic genetic alterations and identified new predictive biomarkers for existing drugs.

Cancer biomarker discovery is improved by accounting for variability in general levels of drug sensitivity in pre-clinical models

We show that variability in general levels of drug sensitivity in pre-clinical cancer models confounds biomarker discovery. However, using a very large panel of cell lines, each treated with many drugs, we could estimate a general level of sensitivity to all drugs in each cell line. By conditioning on this variable, biomarkers were identified that were more likely to be effective in clinical trials than those identified using a conventional uncorrected approach. We find that differences in general levels of drug sensitivity are driven by biologically relevant processes. We developed a gene expression based method that can be used to correct for this confounder in future studies.

pRRophetic: An R Package for Prediction of Clinical Chemotherapeutic Response from Tumor Gene Expression Levels

We recently described a methodology that reliably predicted chemotherapeutic response in multiple independent clinical trials. The method worked by building statistical models from gene expression and drug sensitivity data in a very large panel of cancer cell lines, then applying these models to gene expression data from primary tumor biopsies. Here, to facilitate the development and adoption of this methodology we have created an R package called pRRophetic. This also extends the previously described pipeline, allowing prediction of clinical drug response for many cancer drugs in a user-friendly R environment. We have developed several other important use cases; as an example, we have shown that prediction of bortezomib sensitivity in multiple myeloma may be improved by training models on a large set of neoplastic hematological cell lines. We have also shown that the package facilitates model development and prediction using several different classes of data.

Predicting Response to Histone Deacetylase Inhibitors Using High-Throughput Genomics

BACKGROUND Many disparate biomarkers have been proposed as predictors of response to histone deacetylase inhibitors (HDI); however, all have failed when applied clinically. Rather than this being entirely an issue of reproducibility, response to the HDI vorinostat may be determined by the additive effect of multiple molecular factors, many of which have previously been demonstrated. METHODS We conducted a large-scale gene expression analysis using the Cancer Genome Project for discovery and generated another large independent cancer cell line dataset across different cancers for validation. We compared different approaches in terms of how accurately vorinostat response can be predicted on an independent out-of-batch set of samples and applied the polygenic marker prediction principles in a clinical trial. RESULTS Using machine learning, the small effects that aggregate, resulting in sensitivity or resistance, can be recovered from gene expression data in a large panel of cancer cell lines.This approach can predict vorinostat response accurately, whereas single gene or pathway markers cannot. Our analyses recapitulated and contextualized many previous findings and suggest an important role for processes such as chromatin remodeling, autophagy, and apoptosis. As a proof of concept, we also discovered a novel causative role for CHD4, a helicase involved in the histone deacetylase complex that is associated with poor clinical outcome. As a clinical validation, we demonstrated that a common dose-limiting toxicity of vorinostat, thrombocytopenia, can be predicted (r = 0.55, P = .004) several days before it is detected clinically. CONCLUSION Our work suggests a paradigm shift from single-gene/pathway evaluation to simultaneously evaluating multiple independent high-throughput gene expression datasets, which can be easily extended to other investigational compounds where similar issues are hampering clinical adoption.

BioconductorBuntu - A Linux Distribution that Implements a Web-Based DNA Microarray Analysis Server

SUMMARY BioconductorBuntu is a custom distribution of Ubuntu Linux that automatically installs a server-side microarray processing environment, providing a user-friendly web-based GUI to many of the tools developed by the Bioconductor Project, accessible locally or across a network. System installation is via booting off a CD image or by using a Debian package provided to upgrade an existing Ubuntu installation. In its current version, several microarray analysis pipelines are supported including oligonucleotide, dual-or single-dye experiments, including post-processing with Gene Set Enrichment Analysis. BioconductorBuntu is designed to be extensible, by server-side integration of further relevant Bioconductor modules as required, facilitated by its straightforward underlying Python-based infrastructure. BioconductorBuntu offers an ideal environment for the development of processing procedures to facilitate the analysis of next-generation sequencing datasets. AVAILABILITY BioconductorBuntu is available for download under a creative commons license along with additional documentation and a tutorial from (http://bioinf.nuigalway.ie).