Strengths and limitations of non-disclosive data analysis: a comparison of breast cancer survival classifiers using VisualSHIELD

Preserving data privacy is an important concern in the research use of patient data. The DataSHIELD suite enables privacy-aware advanced statistical analysis in a federated setting. Despite its many applications, it has a few open practical issues: the complexity of hosting a federated infrastructure, the performance penalty imposed by the privacy-preserving constraints, and the ease of use by non-technical users. In this work, we describe a case study in which we review different breast cancer classifiers and report our findings about the limits and advantages of such non-disclosive suite of tools in a realistic setting. Five independent gene expression datasets of breast cancer survival were downloaded from Gene Expression Omnibus (GEO) and pooled together through the federated infrastructure. Three previously published and two newly proposed 5-year cancer-free survival risk score classifiers were trained in a federated environment, and an additional reference classifier was trained with unconstrained data access. The performance of these six classifiers was systematically evaluated, and the results show that i) the published classifiers do not generalize well when applied to patient cohorts that differ from those used to develop them; ii) among the methods we tried, the classification using logistic regression worked better on average, closely followed by random forest; iii) the unconstrained version of the logistic regression classifier outperformed the federated version by 4% on average. Reproducibility of our experiments is ensured through the use of VisualSHIELD, an open-source tool that augments DataSHIELD with new functions, a standardized deployment procedure, and a simple graphical user interface.

Fear circuit-based neurobehavioral signatures mirror resilience to chronic social stress in mouse

Consistent evidence from human data points to successful threat-safety discrimination and responsiveness to extinction of fear memories as key characteristics of resilient individuals. To promote valid cross-species approaches for the identification of resilience mechanisms, we establish a translationally informed mouse model enabling the stratification of mice into three phenotypic subgroups following chronic social defeat stress, based on their individual ability for threat-safety discrimination and conditioned learning: the Discriminating-avoiders, characterized by successful social threat-safety discrimination and extinction of social aversive memories; the Indiscriminate-avoiders, showing aversive response generalization and resistance to extinction, in line with findings on susceptible individuals; and the Non-avoiders displaying impaired aversive conditioned learning. To explore the neurobiological mechanisms underlying the stratification, we perform transcriptome analysis within three key target regions of the fear circuitry. We identify subgroup-specific differentially expressed genes and gene networks underlying the behavioral phenotypes, i.e., the individual ability to show threat-safety discrimination and respond to extinction training. Our approach provides a translationally informed template with which to characterize the behavioral, molecular, and circuit bases of resilience in mice.

Discovering common pathogenetic processes between COVID-19 and diabetes mellitus by differential gene expression pattern analysis

Coronavirus disease 2019 (COVID-19) is an infectious disease caused by the newly discovered coronavirus, SARS-CoV-2. Increased severity of COVID-19 has been observed in patients with diabetes mellitus (DM). This study aimed to identify common transcriptional signatures, regulators and pathways between COVID-19 and DM. We have integrated human whole-genome transcriptomic datasets from COVID-19 and DM, followed by functional assessment with gene ontology (GO) and pathway analyses. In peripheral blood mononuclear cells (PBMCs), among the upregulated differentially expressed genes (DEGs), 32 were found to be commonly modulated in COVID-19 and type 2 diabetes (T2D), while 10 DEGs were commonly downregulated. As regards type 1 diabetes (T1D), 21 DEGs were commonly upregulated, and 29 DEGs were commonly downregulated in COVID-19 and T1D. Moreover, 35 DEGs were commonly upregulated in SARS-CoV-2 infected pancreas organoids and T2D islets, while 14 were commonly downregulated. Several GO terms were found in common between COVID-19 and DM. Prediction of the putative transcription factors involved in the upregulation of genes in COVID-19 and DM identified RELA to be implicated in both PBMCs and pancreas. Here, for the first time, we have characterized the biological processes and pathways commonly dysregulated in COVID-19 and DM, which could be in the next future used for the design of personalized treatment of COVID-19 patients suffering from DM as comorbidity.

New Molecular Insights into Immune Cell Development

During development innate lymphoid cells and specialized lymphocyte subsets colonize peripheral tissues, where they contribute to organogenesis and later constitute the first line of protection while maintaining tissue homeostasis. A few of these subsets are produced only during embryonic development and remain in the tissues throughout life. They are generated through a unique developmental program initiated in lympho-myeloid-primed progenitors, which lose myeloid and B cell potential. They either differentiate into innate lymphoid cells or migrate to the thymus to give rise to embryonic T cell receptor-invariant T cells. At later developmental stages, adaptive T lymphocytes are derived from lympho-myeloid progenitors that colonize the thymus, while lymphoid progenitors become specialized in the production of B cells. This sequence of events highlights the requirement for stratification in the establishment of immune functions that determine efficient seeding of peripheral tissues by a limited number of cells.

Hippo pathway effectors YAP1/TAZ induce an EWS-FLI1-opposing gene signature and associate with disease progression in Ewing sarcoma

YAP1 and TAZ (WWTR1) oncoproteins are the final transducers of the Hippo tumor suppressor pathway. Deregulation of the pathway leads to YAP1/TAZ activation fostering tumorigenesis in multiple malignant tumor types, including sarcoma. However, oncogenic mutations within the core components of the Hippo pathway are uncommon. Ewing sarcoma (EwS), a pediatric cancer with low mutation rate, is characterized by a canonical fusion involving the gene EWSR1 and FLI1 as the most common partner. The fusion protein is a potent driver of oncogenesis, but secondary alterations are scarce, and little is known about other biological factors that determine the risk of relapse or progression. We have observed YAP1/TAZ expression and transcriptional activity in EwS cell lines. Analyses of 55 primary human EwS samples revealed that high YAP1/TAZ expression was associated with progression of the disease and predicted poorer outcome. We did not observe recurrent SNV or copy number gains/losses in Hippo pathway-related loci. However, differential CpG methylation of the RASSF1 locus (a regulator of the Hippo pathway) was observed in EwS cell lines compared with mesenchymal stem cells, the putative cell of origin of EwS. Hypermethylation of RASSF1 correlated with the transcriptional silencing of the tumor suppressor isoform RASFF1A, and transcriptional activation of the pro-tumorigenic isoform RASSF1C, which promotes YAP1/TAZ activation. Knockdown of YAP1/TAZ decreased proliferation and invasion abilities of EwS cells and revealed that YAP1/TAZ transcription activity is inversely correlated with the EWS-FLI1 transcriptional signature. This transcriptional antagonism could be explained partly by EWS-FLI1-mediated transcriptional repression of TAZ. Thus, YAP1/TAZ may override the transcriptional program induced by the fusion protein, contributing to the phenotypic plasticity determined by dynamic fluctuation of the fusion protein, a recently proposed model for disease dissemination in EwS. © 2019 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of Pathological Society of Great Britain and Ireland.

Decomposing Oncogenic Transcriptional Signatures to Generate Maps of Divergent Cellular States

The systematic sequencing of the cancer genome has led to the identification of numerous genetic alterations in cancer. However, a deeper understanding of the functional consequences of these alterations is necessary to guide appropriate therapeutic strategies. Here, we describe Onco-GPS (OncoGenic Positioning System), a data-driven analysis framework to organize individual tumor samples with shared oncogenic alterations onto a reference map defined by their underlying cellular states. We applied the methodology to the RAS pathway and identified nine distinct components that reflect transcriptional activities downstream of RAS and defined several functional states associated with patterns of transcriptional component activation that associates with genomic hallmarks and response to genetic and pharmacological perturbations. These results show that the Onco-GPS is an effective approach to explore the complex landscape of oncogenic cellular states across cancers, and an analytic framework to summarize knowledge, establish relationships, and generate more effective disease models for research or as part of individualized precision medicine paradigms.

Reverse Engineering Cancer: Inferring Transcriptional Gene Signatures from Copy Number Aberrations with ICAro

The characterization of a gene product function is a process that involves multiple laboratory techniques in order to silence the gene itself and to understand the resulting cellular phenotype via several omics profiling. When it comes to tumor cells, usually the translation process from in vitro characterization results to human validation is a difficult journey. Here, we present a simple algorithm to extract mRNA signatures from cancer datasets, where a particular gene has been deleted at the genomic level, ICAro. The process is implemented as a two-step workflow. The first one employs several filters in order to select the two patient subsets: the inactivated one, where the target gene is deleted, and the control one, where large genomic rearrangements should be absent. The second step performs a signature extraction via a Differential Expression analysis and a complementary Random Forest approach to provide an additional gene ranking in terms of information loss. We benchmarked the system robustness on a panel of genes frequently deleted in cancers, where we validated the downregulation of target genes and found a correlation with signatures extracted with the L1000 tool, outperforming random sampling for two out of six L1000 classes. Furthermore, we present a use case correlation with a published transcriptomic experiment. In conclusion, deciphering the complex interactions of the tumor environment is a challenge that requires the integration of several experimental techniques in order to create reproducible results. We implemented a tool which could be of use when trying to find mRNA signatures related to a gene loss event to better understand its function or for a gene-loss associated biomarker research.

Clonal Analysis Delineates Transcriptional Programs of Osteogenic and Adipogenic Lineages of Adult Mouse Skeletal Progenitors

Bone, cartilage, and marrow adipocytes are generated by skeletal progenitors, but the relationships between lineages and mechanisms controlling their differentiation are poorly understood. We established mouse clonal skeletal progenitors with distinct differentiation properties and analyzed their transcriptome. Unipotent osteogenic and adipogenic cells expressed specific transcriptional programs, whereas bipotent clones combined expression of those genes and did not show a unique signature. We tested potential regulators of lineage commitment and found that in the presence of interferon-γ (IFNγ) adipogenic clones can be induced to osteogenesis and that their adipogenic capacity is inhibited. Analysis of IFNγ-regulated genes showed that lineage signatures and fate commitment of skeletal progenitors were controlled by EGR1 and EGR2. Knockdown experiments revealed that EGR1 is a positive regulator of the adipogenic transcriptional program and differentiation capacity, whereas EGR2 inhibits the osteogenic program and potency. Therefore, our work revealed transcriptional signatures of osteogenic and adipogenic lineages and mechanism triggering cell fate.

•

Bone marrow osteo- and adipogenic progenitors have specific transcriptional profiles

•

Bipotent progenitors combine expression of osteogenic and adipogenic programs

•

IFNγ inhibits adipogenesis and induces osteogenesis via downregulation of Egr1/Egr2

•

Egr1 maintains adipogenic and Egr2 suppresses osteogenic lineage commitment

Prospective patient stratification into robust cancer-cell intrinsic subtypes from colorectal cancer biopsies

Colorectal cancer (CRC) biopsies underpin accurate diagnosis, but are also relevant for patient stratification in molecularly-guided clinical trials. The consensus molecular subtypes (CMSs) and colorectal cancer intrinsic subtypes (CRISs) transcriptional signatures have potential clinical utility for improving prognostic/predictive patient assignment. However, their ability to provide robust classification, particularly in pretreatment biopsies from multiple regions or at different time points, remains untested. In this study, we undertook a comprehensive assessment of the robustness of CRC transcriptional signatures, including CRIS and CMS, using a range of tumour sampling methodologies currently employed in clinical and translational research. These include analyses using (i) laser-capture microdissected CRC tissue, (ii) eight publically available rectal cancer biopsy data sets (n = 543), (iii) serial biopsies (from AXEBeam trial, NCT00828672; n = 10), (iv) multi-regional biopsies from colon tumours (n = 29 biopsies, n = 7 tumours), and (v) pretreatment biopsies from the phase II rectal cancer trial COPERNCIUS (NCT01263171; n = 44). Compared to previous results obtained using CRC resection material, we demonstrate that CMS classification in biopsy tissue is significantly less capable of reliably classifying patient subtype (43% unknown in biopsy versus 13% unknown in resections, p = 0.0001). In contrast, there was no significant difference in classification rate between biopsies and resections when using the CRIS classifier. Additionally, we demonstrated that CRIS provides significantly better spatially- and temporally- robust classification of molecular subtypes in CRC primary tumour tissue compared to CMS (p = 0.003 and p = 0.02, respectively). These findings have potential to inform ongoing biopsy-based patient stratification in CRC, enabling robust and stable assignment of patients into clinically-informative arms of prospective multi-arm, multi-stage clinical trials. © 2018 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of Pathological Society of Great Britain and Ireland.

Unmasking Transcriptional Heterogeneity in Senescent Cells

Cellular senescence is a state of irreversibly arrested proliferation, often induced by genotoxic stress [1]. Senescent cells participate in a variety of physiological and pathological conditions, including tumor suppression [2], embryonic development [3, 4], tissue repair [5-8], and organismal aging [9]. The senescence program is variably characterized by several non-exclusive markers, including constitutive DNA damage response (DDR) signaling, senescence-associated β-galactosidase (SA-βgal) activity, increased expression of the cyclin-dependent kinase (CDK) inhibitors p16INK4A (CDKN2A) and p21CIP1 (CDKN1A), increased secretion of many bio-active factors (the senescence-associated secretory phenotype, or SASP), and reduced expression of the nuclear lamina protein LaminB1 (LMNB1) [1]. Many senescence-associated markers result from altered transcription, but the senescent phenotype is variable, and methods for clearly identifying senescent cells are lacking [10]. Here, we characterize the heterogeneity of the senescence program using numerous whole-transcriptome datasets generated by us or publicly available. We identify transcriptome signatures associated with specific senescence-inducing stresses or senescent cell types and identify and validate genes that are commonly differentially regulated. We also show that the senescent phenotype is dynamic, changing at varying intervals after senescence induction. Identifying novel transcriptome signatures to detect any type of senescent cell or to discriminate among diverse senescence programs is an attractive strategy for determining the diverse biological roles of senescent cells and developing specific drug targets.

Large-scale integration of small molecule-induced genome-wide transcriptional responses, Kinome-wide binding affinities and cell-growth inhibition profiles reveal global trends characterizing systems-level drug action

The Library of Integrated Network-based Cellular Signatures (LINCS) project is a large-scale coordinated effort to build a comprehensive systems biology reference resource. The goals of the program include the generation of a very large multidimensional data matrix and informatics and computational tools to integrate, analyze, and make the data readily accessible. LINCS data include genome-wide transcriptional signatures, biochemical protein binding profiles, cellular phenotypic response profiles and various other datasets for a wide range of cell model systems and molecular and genetic perturbations. Here we present a partial survey of this data facilitated by data standards and in particular a robust compound standardization workflow; we integrated several types of LINCS signatures and analyzed the results with a focus on mechanism of action (MoA) and chemical compounds. We illustrate how kinase targets can be related to disease models and relevant drugs. We identified some fundamental trends that appear to link Kinome binding profiles and transcriptional signatures to chemical information and biochemical binding profiles to transcriptional responses independent of chemical similarity. To fill gaps in the datasets we developed and applied predictive models. The results can be interpreted at the systems level as demonstrated based on a large number of signaling pathways. We can identify clear global relationships, suggesting robustness of cellular responses to chemical perturbation. Overall, the results suggest that chemical similarity is a useful measure at the systems level, which would support phenotypic drug optimization efforts. With this study we demonstrate the potential of such integrated analysis approaches and suggest prioritizing further experiments to fill the gaps in the current data.