Astrid Murumägi

Quantitative scoring of differential drug sensitivity for individually optimized anticancer therapies

We developed a systematic algorithmic solution for quantitative drug sensitivity scoring (DSS), based on continuous modeling and integration of multiple dose-response relationships in high-throughput compound testing studies. Mathematical model estimation and continuous interpolation makes the scoring approach robust against sources of technical variability and widely applicable to various experimental settings, both in cancer cell line models and primary patient-derived cells. Here, we demonstrate its improved performance over other response parameters especially in a leukemia patient case study, where differential DSS between patient and control cells enabled identification of both cancer-selective drugs and drug-sensitive patient sub-groups, as well as dynamic monitoring of the response patterns and oncogenic driver signals during cancer progression and relapse in individual patient cells ex vivo. An open-source and easily extendable implementation of the DSS calculation is made freely available to support its tailored application to translating drug sensitivity testing results into clinically actionable treatment options.

FusionCatcher - a tool for finding somatic fusion genes in paired-end RNA-sequencing data

FusionCatcher is a software tool for finding somatic fusion genes in paired-end RNA-sequencing data from human or other vertebrates. FusionCatcher achieves competitive detection rates and real-time PCR validation rates in RNA-sequencing data from tumor cells. Fusion-Catcher is available at http://code.google.com/p/fusioncatcher/.

Cytokinesis failure due to derailed integrin traffic induces aneuploidy and oncogenic transformation in vitro and in vivo

Aneuploidy is frequently detected in solid tumors but the mechanisms regulating the generation of aneuploidy and their relevance in cancer initiation remain under debate and are incompletely characterized. Spatial and temporal regulation of integrin traffic is critical for cell migration and cytokinesis. Impaired integrin endocytosis, because of the loss of Rab21 small GTPase or mutations in the integrin β-subunit cytoplasmic tail, induces failure of cytokinesis in vitro. Here, we describe that repeatedly failed cytokinesis, because of impaired traffic, is sufficient to trigger the generation of aneuploid cells, which display characteristics of oncogenic transformation in vitro and are tumorigenic in vivo. Furthermore, in an in vivo mouse xenograft model, non-transformed cells with impaired integrin traffic formed tumors with a long latency. More detailed investigation of these tumors revealed that the tumor cells were aneuploid. Therefore, abnormal integrin traffic was linked with generation of aneuploidy and cell transformation also in vivo. In human prostate and ovarian cancer samples, downregulation of Rab21 correlates with increased malignancy. Loss-of-function experiments demonstrate that long-term depletion of Rab21 is sufficient to induce chromosome number aberrations in normal human epithelial cells. These data are the first to demonstrate that impaired integrin traffic is sufficient to induce conversion of non-transformed cells to tumorigenic cells in vitro and in vivo.

Reanalysis of RNA-Sequencing Data Reveals Several Additional Fusion Genes with Multiple Isoforms

RNA-sequencing and tailored bioinformatic methodologies have paved the way for identification of expressed fusion genes from the chaotic genomes of solid tumors. We have recently successfully exploited RNA-sequencing for the discovery of 24 novel fusion genes in breast cancer. Here, we demonstrate the importance of continuous optimization of the bioinformatic methodology for this purpose, and report the discovery and experimental validation of 13 additional fusion genes from the same samples. Integration of copy number profiling with the RNA-sequencing results revealed that the majority of the gene fusions were promoter-donating events that occurred at copy number transition points or involved high-level DNA-amplifications. Sequencing of genomic fusion break points confirmed that DNA-level rearrangements underlie selected fusion transcripts. Furthermore, a significant portion (>60%) of the fusion genes were alternatively spliced. This illustrates the importance of reanalyzing sequencing data as gene definitions change and bioinformatic methods improve, and highlights the previously unforeseen isoform diversity among fusion transcripts.

DNA methylation signatures of the AIRE promoter in thymic epithelial cells, thymomas and normal tissues.

Mutations in the AIRE gene cause autoimmune polyendocrinopathy candidiasis ectodermal dystrophy (APECED), which is associated with autoimmunity towards several peripheral organs. The AIRE protein is almost exclusively expressed in medullary thymic epithelial cells (mTEC) and CpG methylation in the promoter of the AIRE gene has been suggested to control its tissue-specific expression pattern. We found that in human AIRE-positive medullary and AIRE-negative cortical epithelium, the AIRE promoter is hypomethylated, whereas in thymocytes, the promoter had high level of CpG methylation. Likewise, in mouse mTECs the AIRE promoter was uniformly hypomethylated. In the same vein, the AIRE promoter was hypomethylated in AIRE-negative thymic epithelial tumors (thymomas) and in several peripheral tissues. Our data are compatible with the notion that promoter hypomethylation is necessary but not sufficient for tissue-specific regulation of the AIRE gene. In contrast, a positive correlation between AIRE expression and histone H3 lysine 4 trimethylation, an active chromatin mark, was found in the AIRE promoter in human and mouse TECs.

Drug response prediction by inferring pathway-response associations with kernelized Bayesian matrix factorization

MOTIVATION A key goal of computational personalized medicine is to systematically utilize genomic and other molecular features of samples to predict drug responses for a previously unseen sample. Such predictions are valuable for developing hypotheses for selecting therapies tailored for individual patients. This is especially valuable in oncology, where molecular and genetic heterogeneity of the cells has a major impact on the response. However, the prediction task is extremely challenging, raising the need for methods that can effectively model and predict drug responses. RESULTS In this study, we propose a novel formulation of multi-task matrix factorization that allows selective data integration for predicting drug responses. To solve the modeling task, we extend the state-of-the-art kernelized Bayesian matrix factorization (KBMF) method with component-wise multiple kernel learning. In addition, our approach exploits the known pathway information in a novel and biologically meaningful fashion to learn the drug response associations. Our method quantitatively outperforms the state of the art on predicting drug responses in two publicly available cancer datasets as well as on a synthetic dataset. In addition, we validated our model predictions with lab experiments using an in-house cancer cell line panel. We finally show the practical applicability of the proposed method by utilizing prior knowledge to infer pathway-drug response associations, opening up the opportunity for elucidating drug action mechanisms. We demonstrate that pathway-response associations can be learned by the proposed model for the well-known EGFR and MEK inhibitors. AVAILABILITY AND IMPLEMENTATION The source code implementing the method is available at http://research.cs.aalto.fi/pml/software/cwkbmf/ CONTACTS muhammad.ammad-ud-din@aalto.fi or samuel.kaski@aalto.fi SUPPLEMENTARY INFORMATION Supplementary data are available at Bioinformatics online.

Consistency in drug response profiling

The comparative analysis by Haibe-Kains et al.1 concluded that data from two large-scale studies of cancer cell lines2,3 showed highly discordant results for drug sensitivity measurements, whereas gene expression data were reasonably concordant. Here, we crosscompared the two original datasets2,3 against our own data of drug response profiles in overlapping cancer cell line panels. Our results indicate that it is possible to achieve concordance between different laboratories for drug response measurements by paying attention to the harmonization of assays and experimental procedures. There is a Reply to this Comment by Safikhani, Z. et al. Nature 540, http://dx.doi.org/10.1038/ nature20172 (2016). Haibe-Kains et al.1 reported on a comparative evaluation of two drug sensitivity and molecular profiling datasets, one from the Cancer Genome Project (CGP)2 and the other from the Cancer Cell Line Encyclopedia (CCLE)3. In their analyses, gene expression profiles between hundreds of common cancer cell lines across all genes showed high consistency between the two studies (median rank correlation (MRC) = 0.85), whereas the drug response data for 15 common compounds were highly discordant (MRC = 0.28 for halfmaximum inhibitory concentration (IC50) values). This report1 and the accompanying commentary4 suggested that differences in laboratory protocols, compounds and their tested concentration ranges, and computational methods may account for the differences, but these reports did not elaborate which of these factors are important and whether they can be controlled for. Here, we reanalysed the dose–response data from both CGP and CCLE using a standardized area under the curve (AUC) response metric, which we call the drug sensitivity score (DSS)5. We then compared the CGP and CCLE data with a new dataset of drug responses profiled using the Institute for Molecular Medicine Finland (FIMM) compound testing assay6, covering 308 drugs across 106 cancer cell lines. The FIMM data included 45 compounds in common with CGP and 14 with the CCLE in 50 cell lines (Supplementary Data 1). In the AUC calculation, we unified the drug concentration ranges across the CGP, CCLE and FIMM assays. We observed a significantly higher level of consistency (P = 4.2 × 10−5), especially between the CCLE and FIMM drug response data (MRC = 0.74), as compared to the consistency between FIMM and CGP data (MRC = 0.54) (Fig. 1a). Similar experimental protocols were applied at FIMM and CCLE, including the same readout (CellTiter-Glo, Promega), similar controls (vehicle as negative control and positive controls of toxic compounds 100 μ M benzethonium chloride or 1 μ M MG132). However, there were also differences, such as the plate format used (1,536 versus 384 wells). Importantly, there was no effort made to standardize cell numbers used or any other parameters between the three laboratories, such as the source, passage number and media used for cells, nor the origin and handling of drugs. Therefore, this observed level of drug response agreement could be substantially improved by further standardization of the laboratory protocols. The CGP experimental protocol differed from the two others in terms of the readout (fluorescent nucleic acid stain Syto 60, Life Technologies), in the use of controls (drug-free cells as negative and no cells as positive controls), and the plate format used (96or 384-well plates). We compared the drug response profiles between the same cell lines from different laboratories, in line with the approach of Haibe-Kains et al.1, in which they showed consistency in gene expression profiles from CGP and CCLE (MRC = 0.85)1. The Haibe-Kains et al.1 approach, in which the correlation is calculated for each drug separately across the cell lines, showed more variability (Fig. 1b), owing to the fact that some drugs show minimal efficacy in all the tested cell lines. Analogously, gene expression correlations vary more widely when analysed at the level of genes across cell lines (MRC = 0.58 between CGP and CCLE), as certain genes are not expressed above technical noise. Although both ways to compare the data are relevant to the overall goal of personalized

Enhanced sensitivity to glucocorticoids in cytarabine-resistant AML.

We sought to identify drugs that could counteract cytarabine resistance in acute myeloid leukemia (AML) by generating eight resistant variants from MOLM-13 and SHI-1 AML cell lines by long-term drug treatment. These cells were compared with 66 ex vivo chemorefractory samples from cytarabine-treated AML patients. The models and patient cells were subjected to genomic and transcriptomic profiling and high-throughput testing with 250 emerging and clinical oncology compounds. Genomic profiling uncovered deletion of the deoxycytidine kinase (DCK) gene in both MOLM-13- and SHI-1-derived cytarabine-resistant variants and in an AML patient sample. Cytarabine-resistant SHI-1 variants and a subset of chemorefractory AML patient samples showed increased sensitivity to glucocorticoids that are often used in treatment of lymphoid leukemia but not AML. Paired samples taken from AML patients before treatment and at relapse also showed acquisition of glucocorticoid sensitivity. Enhanced glucocorticoid sensitivity was only seen in AML patient samples that were negative for the FLT3 mutation (P=0.0006). Our study shows that development of cytarabine resistance is associated with increased sensitivity to glucocorticoids in a subset of AML, suggesting a new therapeutic strategy that should be explored in a clinical trial of chemorefractory AML patients carrying wild-type FLT3.

Intrinsic resistance to PIM kinase inhibition in AML through p38α-mediated feedback activation of mTOR signaling

Although conventional therapies for acute myeloid leukemia (AML) and diffuse large B-cell lymphoma (DLBCL) are effective in inducing remission, many patients relapse upon treatment. Hence, there is an urgent need for novel therapies. PIM kinases are often overexpressed in AML and DLBCL and are therefore an attractive therapeutic target. However, in vitro experiments have demonstrated that intrinsic resistance to PIM inhibition is common. It is therefore likely that only a minority of patients will benefit from single agent PIM inhibitor treatment. In this study, we performed an shRNA-based genetic screen to identify kinases whose suppression is synergistic with PIM inhibition. Here, we report that suppression of p38α (MAPK14) is synthetic lethal with the PIM kinase inhibitor AZD1208. PIM inhibition elevates reactive oxygen species (ROS) levels, which subsequently activates p38α and downstream AKT/mTOR signaling. We found that p38α inhibitors sensitize hematological tumor cell lines to AZD1208 treatment in vitro and in vivo. These results were validated in ex vivo patient-derived AML cells. Our findings provide mechanistic and translational evidence supporting the rationale to test a combination of p38α and PIM inhibitors in clinical trials for AML and DLBCL.

Colorectal Cancer Consensus Molecular Subtypes Translated to Preclinical Models Uncover Potentially Targetable Cancer Cell Dependencies

Purpose: Response to standard oncologic treatment is limited in colorectal cancer. The gene expression–based consensus molecular subtypes (CMS) provide a new paradigm for stratified treatment and drug repurposing; however, drug discovery is currently limited by the lack of translation of CMS to preclinical models. Experimental Design: We analyzed CMS in primary colorectal cancers, cell lines, and patient-derived xenografts (PDX). For classification of preclinical models, we developed an optimized classifier enriched for cancer cell–intrinsic gene expression signals, and performed high-throughput in vitro drug screening (n = 459 drugs) to analyze subtype-specific drug sensitivities. Results: The distinct molecular and clinicopathologic characteristics of each CMS group were validated in a single-hospital series of 409 primary colorectal cancers. The new, cancer cell–adapted classifier was found to perform well in primary tumors, and applied to a panel of 148 cell lines and 32 PDXs, these colorectal cancer models were shown to recapitulate the biology of the CMS groups. Drug screening of 33 cell lines demonstrated subtype-dependent response profiles, confirming strong response to EGFR and HER2 inhibitors in the CMS2 epithelial/canonical group, and revealing strong sensitivity to HSP90 inhibitors in cells with the CMS1 microsatellite instability/immune and CMS4 mesenchymal phenotypes. This association was validated in vitro in additional CMS-predicted cell lines. Combination treatment with 5-fluorouracil and luminespib showed potential to alleviate chemoresistance in a CMS4 PDX model, an effect not seen in a chemosensitive CMS2 PDX model. Conclusions: We provide translation of CMS classification to preclinical models and uncover a potential for targeted treatment repurposing in the chemoresistant CMS4 group. Clin Cancer Res; 24(4); 794–806. ©2017 AACR.