Application of transfer learning to predict drug-induced human in vivo gene expression changes using rat in vitro and in vivo data

The liver is the primary site for the metabolism and detoxification of many compounds, including pharmaceuticals. Consequently, it is also the primary location for many adverse reactions. As the liver is not readily accessible for sampling in humans; rodent or cell line models are often used to evaluate potential toxic effects of a novel compound or candidate drug. However, relating the results of animal and in vitro studies to relevant clinical outcomes for the human in vivo situation still proves challenging. In this study, we incorporate principles of transfer learning within a deep artificial neural network allowing us to leverage the relative abundance of rat in vitro and in vivo exposure data from the Open TG-GATEs data set to train a model to predict the expected pattern of human in vivo gene expression following an exposure given measured human in vitro gene expression. We show that domain adaptation has been successfully achieved, with the rat and human in vitro data no longer being separable in the common latent space generated by the network. The network produces physiologically plausible predictions of human in vivo gene expression pattern following an exposure to a previously unseen compound. Moreover, we show the integration of the human in vitro data in the training of the domain adaptation network significantly improves the temporal accuracy of the predicted rat in vivo gene expression pattern following an exposure to a previously unseen compound. In this way, we demonstrate the improvements in prediction accuracy that can be achieved by combining data from distinct domains.

Use of deep learning methods to translate drug-induced gene expression changes from rat to human primary hepatocytes

In clinical trials, animal and cell line models are often used to evaluate the potential toxic effects of a novel compound or candidate drug before progressing to human trials. However, relating the results of animal and in vitro model exposures to relevant clinical outcomes in the human in vivo system still proves challenging, relying on often putative orthologs. In recent years, multiple studies have demonstrated that the repeated dose rodent bioassay, the current gold standard in the field, lacks sufficient sensitivity and specificity in predicting toxic effects of pharmaceuticals in humans. In this study, we evaluate the potential of deep learning techniques to translate the pattern of gene expression measured following an exposure in rodents to humans, circumventing the current reliance on orthologs, and also from in vitro to in vivo experimental designs. Of the applied deep learning architectures applied in this study the convolutional neural network (CNN) and a deep artificial neural network with bottleneck architecture significantly outperform classical machine learning techniques in predicting the time series of gene expression in primary human hepatocytes given a measured time series of gene expression from primary rat hepatocytes following exposure in vitro to a previously unseen compound across multiple toxicologically relevant gene sets. With a reduction in average mean absolute error across 76 genes that have been shown to be predictive for identifying carcinogenicity from 0.0172 for a random regression forest to 0.0166 for the CNN model (p < 0.05). These deep learning architecture also perform well when applied to predict time series of in vivo gene expression given measured time series of in vitro gene expression for rats.

FTICR mass spectrometry-based multivariate analysis to explore distinctive metabolites and metabolic pathways: A comprehensive bioanalytical strategy toward time-course metabolic profiling of Thymus vulgaris plants responding to drought stress

In this research, metabolic profiling/pathways of Thymus vulgaris (thyme) plant were assessed during a water deficit stress using an FTICR mass spectrometry-based metabolomics strategy incorporating multivariate data analysis and bioinformatics techniques. Herein, differences of MS signals in specific time courses after water deficit stress and control cases without any timing period were distinguished significantly by common pattern recognition techniques, i.e., PCA, HCA-Heatmap, and PLS-DA. Subsequently, the results were compared with supervised Kohonen neural network (SKN) ones as a non-linear data visualization and capable mapping tool. The classification models showed excellent performance to predict the level of drought stress. By assessing variances contribution on the PCA-loadings of the MS data, the discriminant variables related to the most critical metabolites were identified and then confirmed by ANOVA. Indeed, FTICR MS-based multivariate analysis strategy could explore distinctive metabolites and metabolic pathways/profiles, grouped into three metabolism categories including amino acids, carbohydrates (i.e., galactose, glucose, fructose, sucrose, and mannose), and other metabolites (rosmarinic acid and citrate), to indicate biological mechanisms in response to drought stress for thyme. It was achieved and approved through the MS signals, genomics databases, and transcriptomics factors to interpret and predict the plant metabolic behavior. Eventually, a comprehensive pathway analysis was used to provide a pathway enrichment analysis and explore topological pathway characteristics dealing with the remarkable metabolites to demonstrate that galactose metabolism is the most significant pathway in the biological system of thyme.

Wnt/Tcf1 pathway restricts embryonic stem cell cycle through activation of the Ink4/Arf locus

Understanding the mechanisms regulating cell cycle, proliferation and potency of pluripotent stem cells guarantees their safe use in the clinic. Embryonic stem cells (ESCs) present a fast cell cycle with a short G1 phase. This is due to the lack of expression of cell cycle inhibitors, which ultimately determines naïve pluripotency by holding back differentiation. The canonical Wnt/β-catenin pathway controls mESC pluripotency via the Wnt-effector Tcf3. However, if the activity of the Wnt/β-catenin controls the cell cycle of mESCs remains unknown. Here we show that the Wnt-effector Tcf1 is recruited to and triggers transcription of the Ink4/Arf tumor suppressor locus. Thereby, the activation of the Wnt pathway, a known mitogenic pathway in somatic tissues, restores G1 phase and drastically reduces proliferation of mESCs without perturbing pluripotency. Tcf1, but not Tcf3, is recruited to a palindromic motif enriched in the promoter of cell cycle repressor genes, such as p15^Ink4b, p16^Ink4a and p19^Arf, which mediate the Wnt-dependent anti-proliferative effect in mESCs. Consistently, ablation of β-catenin or Tcf1 expression impairs Wnt-dependent cell cycle regulation. All together, here we showed that Wnt signaling controls mESC pluripotency and proliferation through non-overlapping functions of distinct Tcf factors.

Studying how to safely expand stem cells in culture is essential for regenerative medicine applications. Hence there is a clear need to decode how the cell cycle of mouse embryonic stem cells (mESCs) is regulated. Tcf3 and Tcf1 belong to the Tcf family of proteins. Tcf/Lef are effectors of the Wnt/β-catenin pathway and Tcf3 controls mESC pluripotency. Here we identified a recruitment site for Tcf1 embedded into a number of cell cycle repressor genes such as p15^Ink4b, p16^Ink4a and p19^Arf. Tcf1-mediated activation of these genes drastically slows down proliferation of mESCs. In conclusion, here we showed that the Wnt pathway, besides controlling mESC pluripotency via Tcf3, also regulates mESC cell cycle through the recruitment of Tcf1 to the regulatory sites of key cell cycle genes.

Abstract 1918: Synergy between epithelial and mesenchymal cells in breast cancer: from mammospheres to predicting patient outcomes

Targeting breast cancer stem cells promises to eliminate the root of metastasis and recurrence and thus to eradicate the ultimate cause of cancer-related death of patients. However, the identity of epithelial cancer stem-like is still elusive; cancer stem-like and metastasis initiating cells have been described to be either epithelial (E) or mesenchymal (M), and thus their identity appears to critically depend on the context of the analyzed cell line.

Here, we have studied the heterogeneous HMLER cell line to identify the relationships between epithelial and mesenchymal cells while they are differentiated and selected under suspension culture conditions. We find that CD24+/CD44- epithelial cells and CD44+/CD24- mesenchymal cells synergize in causing the proliferation of both sub-populations, increasing mammosphere production, and causing the emergence of aggressive motile cells when replated. We observed that when grown in suspension, isolated epithelial as well as isolated mesenchymal cell populations converged towards a hybrid E/M state characterized by coexpression of genes of both lineages, from population down to the single cell level. For epithelial populations convergence at the hybrid state during growth in suspension was thus achieved by proliferation of the most mesenchymal cells resembling an epithelial to mesenchymal transition at the population level. For mesenchymal populations, the hybrid state was reached by the reverse, MET, involving proliferation of the most epithelial cells in a context dependent manner.We use these findings to resolve which genes are predictive of poor outcome in primary human breast tumors. Here we show for the first time that high expression of mesenchymal genes in luminal B/epithelial tumors predicts poor outcome of patients, whereas expression of epithelial genes in basal/mesenchymal tumors predicts poor outcome and thus increased metastasis. Therefore, functional synergy between E and M cells can explain how the tumor context determines whether a minority of mesenchymal or epithelial cells can serve as metastatic stem cells.

Citation Format: Anne Grosse-Wilde, Aymeric d'Herouel, Rolf Kuestner, Gokhan Ertaylan, Alexander Skupin, Sui Huang, Adrian Ozinsky. Synergy between epithelial and mesenchymal cells in breast cancer: from mammospheres to predicting patient outcomes. [abstract]. In: Proceedings of the 105th Annual Meeting of the American Association for Cancer Research; 2014 Apr 5-9; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2014;74(19 Suppl):Abstract nr 1918. doi:10.1158/1538-7445.AM2014-1918

Abstract LB-252: Synergy of epithelial and mesenchymal cells under anoikis-inducing conditions

Tumor metastasis is the ultimate cause for death of cancer patients. In epithelial tumors such as breast cancer the early steps of the metastatic cascade require tumor cells to migrate, invade and survive detachment from the extracellular matrix in suspension which in epithelial (E) cells results in cell death (anoikis). Therefore, all of these earlier steps appear to be accomplished by mesenchymal (M) cells. In a process called epithelial to mesenchymal transition (EMT) E tumor cells can become M. However, a central paradox of this model is that in most patients the metastatic tumor has an E morphology just as the primary tumor. Analogous to primary tumors and their corresponding metastases, in transformed and immortalized mammary epithelial HMLER cells we frequently found E cells in cultures of replated derivatives of HMLER cells grown in suspension, as well as in suspended mammospheres. In order to disentangle the role of E and M cells under suspension culture conditions in the heterogeneously E and M HMLER cells, we generated mammospheres of HMLER cells, and of isolated E or M HMLER clones. As expected, only the heterogeneous HMLER cells and the M clones were able to form mammospheres, and in gene expression arrays we detected a significant upregulation of genes encoding the pluripotency factor OCT4 and its targets. Importantly, the M cell clone significantly upregulated E genes during suspension conditions. Intriguingly, amongst those were many secreted growth factors and inflammatory cytokines such as SLPI, IL1B, MMP7, S100 proteins and CXCL1, several of which have previously been described to enhance aggressive cancer growth and suppress apoptosis at the metastatic site. In order to directly test the effect of E cells on M cells in suspended mammospheres we generated suspension cultures of isolated E and M cells or of E and M co-cultures. We found that i) E cells can survive in suspension only in conjunction with M cells, and ii) co-culture of E and M cells in suspension resulted in significantly larger and higher number of mammospheres than culture of isolated M cells. Moreover, treatment of M cells with supernatant of E cells could mimic the coculture effect of increased and enlarged mammosphere formation, suggesting that soluble E factors are responsible for enhanced proliferation of M HMLER cells in suspension. In summary, in stem cell-enriched mammospheres of M cells we found high induction of secreted growth and survival factors which are constitutively high expressed in adherent as well as in suspended E cells.

Indeed, co-culture of suspended E and M cells resulted in a synergistic effect of enhanced proliferation of M cells in mammospheres and survival of E cells. Synergy of E and M cells in suspension can explain how epithelial cells could be retained under anoikis-inducing conditions in the vasculature and why metastatic tumors are E like their primary counterparts.

Citation Format: Anne Grosse-Wilde, Gokhan Ertaylan, Robert West, Kathie Walters, Adrian Ozinsky. Synergy of epithelial and mesenchymal cells under anoikis-inducing conditions. [abstract]. In: Proceedings of the 104th Annual Meeting of the American Association for Cancer Research; 2013 Apr 6-10; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2013;73(8 Suppl):Abstract nr LB-252. doi:10.1158/1538-7445.AM2013-LB-252

Inference of surface membrane factors of HIV-1 infection through functional interaction networks

BACKGROUND

HIV infection affects the populations of T helper cells, dendritic cells and macrophages. Moreover, it has a serious impact on the central nervous system. It is yet not clear whether this list is complete and why specifically those cell types are affected. To address this question, we have developed a method to identify cellular surface proteins that permit, mediate or enhance HIV infection in different cell/tissue types in HIV-infected individuals. Receptors associated with HIV infection share common functions and domains and are involved in similar cellular processes. These properties are exploited by bioinformatics techniques to predict novel cell surface proteins that potentially interact with HIV.

METHODOLOGY/PRINCIPAL FINDINGS

We compiled a set of surface membrane proteins (SMP) that are known to interact with HIV. This set is extended by proteins that have direct interaction and share functional similarity. This resulted in a comprehensive network around the initial SMP set. Using network centrality analysis we predict novel surface membrane factors from the annotated network. We identify 21 surface membrane factors, among which three have confirmed functions in HIV infection, seven have been identified by at least two other studies, and eleven are novel predictions and thus excellent targets for experimental investigation.

CONCLUSIONS

Determining to what extent HIV can interact with human SMPs is an important step towards understanding patient specific disease progression. Using various bioinformatics techniques, we generate a set of surface membrane factors that constitutes a well-founded starting point for experimental testing of cell/tissue susceptibility of different HIV strains as well as for cohort studies evaluating patient specific disease progression.

Identifying potential survival strategies of HIV-1 through virus-host protein interaction networks

Background

The National Institute of Allergy and Infectious Diseases has launched the HIV-1 Human Protein Interaction Database in an effort to catalogue all published interactions between HIV-1 and human proteins. In order to systematically investigate these interactions functionally and dynamically, we have constructed an HIV-1 human protein interaction network. This network was analyzed for important proteins and processes that are specific for the HIV life-cycle. In order to expose viral strategies, network motif analysis was carried out showing reoccurring patterns in virus-host dynamics.

Results

Our analyses show that human proteins interacting with HIV form a densely connected and central sub-network within the total human protein interaction network. The evaluation of this sub-network for connectivity and centrality resulted in a set of proteins essential for the HIV life-cycle. Remarkably, we were able to associate proteins involved in RNA polymerase II transcription with hubs and proteasome formation with bottlenecks. Inferred network motifs show significant over-representation of positive and negative feedback patterns between virus and host. Strikingly, such patterns have never been reported in combined virus-host systems.

Conclusions

HIV infection results in a reprioritization of cellular processes reflected by an increase in the relative importance of transcriptional machinery and proteasome formation. We conclude that during the evolution of HIV, some patterns of interaction have been selected for resulting in a system where virus proteins preferably interact with central human proteins for direct control and with proteasomal proteins for indirect control over the cellular processes. Finally, the patterns described by network motifs illustrate how virus and host interact with one another.

HIV decision support: from molecule to man

Human immunodeficiency virus (HIV) is recognized to be one of the most destructive pandemics in recorded history. Effective highly active antiretroviral therapy and the availability of genetic screening of patient virus data have led to sustained viral suppression and higher life expectancy in patients who have been infected with HIV. The sheer complexity of the disease stems from the multiscale and highly dynamic nature of the system under study. The complete cascade from genome, proteome, metabolome and physiome to health forms a multidimensional system that crosses many orders of magnitude in temporal and spatial scales. Understanding, quantifying and handling this complexity is one of the biggest challenges of our time, which requires a highly multidisciplinary approach. In order to supply researchers with an interactive framework and to provide the medical professional with appropriate tools and information for making a balanced and reliable clinical decision, we have developed 'ViroLab', a collaborative decision-support system (http://www.virolab.org/). ViroLab contains computational models that cover various spatial and temporal scales from atomic-level interactions in nanoseconds up to sociological interactions on the epidemiological level, spanning years of disease progression. ViroLab allows for personalized drug ranking. It is on trial in six hospitals and various virology and epidemiology laboratories across Europe.