De novo gene signature identification from single-cell RNA-seq with hierarchical Poisson factorization.

Common approaches to gene signature discovery in single-cell RNA-sequencing (scRNA-seq) depend upon predefined structures like clusters or pseudo-temporal order, require prior normalization, or do not account for the sparsity of single-cell data. We present single-cell hierarchical Poisson factorization (scHPF), a Bayesian factorization method that adapts hierarchical Poisson factorization (Gopalan et al, 2015, Proceedings of the 31st Conference on Uncertainty in Artificial Intelligence, 326) for de novo discovery of both continuous and discrete expression patterns from scRNA-seq. scHPF does not require prior normalization and captures statistical properties of single-cell data better than other methods in benchmark datasets. Applied to scRNA-seq of the core and margin of a high-grade glioma, scHPF uncovers marked differences in the abundance of glioma subpopulations across tumor regions and regionally associated expression biases within glioma subpopulations. scHFP revealed an expression signature that was spatially biased toward the glioma-infiltrated margins and associated with inferior survival in glioblastoma.

Consensus scHPF Identifies Cell Type-Specific Drug Responses in Glioma by Integrating Large-Scale scRNA-seq.

Single-cell transcriptomic analyses now frequently involve elaborate study designs including samples from multiple individuals, experimental conditions, perturbations, and batches from complex tissues. Dimensionality reduction is required to facilitate integration, interpretation, and statistical analysis. However, these datasets often include subtly different cellular subpopulations or state transitions, which are poorly described by clustering. We previously reported a Bayesian matrix factorization algorithm called single-cell hierarchical Poisson factorization (scHPF) that identifies gene co-expression patterns directly from single-cell RNA-seq (scRNA-seq) count matrices while accounting for transcript drop-out and noise. Here, we describe consensus scHPF, which analyzes scHPF models from multiple random initializations to identify the most robust gene signatures and automatically determine the number of factors for a given dataset. Consensus scHPF facilitates integration of complex datasets with highly multi-modal posterior distributions, resulting in factors that can be uniformly analyzed across individuals and conditions. To demonstrate the utility of consensus scHPF, we performed a meta-analysis of a large-scale scRNA-seq dataset from drug-treated, human glioma slice cultures generated from surgical specimens across three major cell types, 19 patients, 10 drug treatment conditions, and 52 samples. In addition to recapitulating previously reported cell type-specific drug responses from smaller studies, consensus scHPF identified disparate effects of the topoisomerase poisons etoposide and topotecan that are highly consistent with the distinct roles and expression patterns of their respective protein targets.

Deconvolution of cell type-specific drug responses in human tumor tissue with single-cell RNA-seq.

BACKGROUND: Preclinical studies require models that recapitulate the cellular diversity of human tumors and provide insight into the drug sensitivities of specific cellular populations. The ideal platform would enable rapid screening of cell type-specific drug sensitivities directly in patient tumor tissue and reveal strategies to overcome intratumoral heterogeneity. METHODS: We combine multiplexed drug perturbation in acute slice culture from freshly resected tumors with single-cell RNA sequencing (scRNA-seq) to profile transcriptome-wide drug responses in individual patients. We applied this approach to drug perturbations on slices derived from six glioblastoma (GBM) resections to identify conserved drug responses and to one additional GBM resection to identify patient-specific responses. RESULTS: We used scRNA-seq to demonstrate that acute slice cultures recapitulate the cellular and molecular features of the originating tumor tissue and the feasibility of drug screening from an individual tumor. Detailed investigation of etoposide, a topoisomerase poison, and the histone deacetylase (HDAC) inhibitor panobinostat in acute slice cultures revealed cell type-specific responses across multiple patients. Etoposide has a conserved impact on proliferating tumor cells, while panobinostat treatment affects both tumor and non-tumor populations, including unexpected effects on the immune microenvironment. CONCLUSIONS: Acute slice cultures recapitulate the major cellular and molecular features of GBM at the single-cell level. In combination with scRNA-seq, this approach enables cell type-specific analysis of sensitivity to multiple drugs in individual tumors. We anticipate that this approach will facilitate pre-clinical studies that identify effective therapies for solid tumors.

Single-cell transcriptomics of human T cells reveals tissue and activation signatures in health and disease.

Human T cells coordinate adaptive immunity in diverse anatomic compartments through production of cytokines and effector molecules, but it is unclear how tissue site influences T cell persistence and function. Here, we use single cell RNA-sequencing (scRNA-seq) to define the heterogeneity of human T cells isolated from lungs, lymph nodes, bone marrow and blood, and their functional responses following stimulation. Through analysis of >50,000 resting and activated T cells, we reveal tissue T cell signatures in mucosal and lymphoid sites, and lineage-specific activation states across all sites including distinct effector states for CD8+ T cells and an interferon-response state for CD4+ T cells. Comparing scRNA-seq profiles of tumor-associated T cells to our dataset reveals predominant activated CD8+ compared to CD4+ T cell states within multiple tumor types. Our results therefore establish a high dimensional reference map of human T cell activation in health for analyzing T cells in disease.

Single-Cell Analysis of Regional Differences in Adult V-SVZ Neural Stem Cell Lineages.

The ventricular-subventricular zone (V-SVZ) harbors adult neural stem cells. V-SVZ neural stem cells exhibit features of astrocytes, have a regional identity, and depending on their location in the lateral or septal wall of the lateral ventricle, generate different types of neuronal and glial progeny. We performed large-scale single-cell RNA sequencing to provide a molecular atlas of cells from the lateral and septal adult V-SVZ of male and female mice. This revealed regional and sex differences among adult V-SVZ cells. We uncovered lineage potency bias at the single-cell level among lateral and septal wall astrocytes toward neurogenesis and oligodendrogenesis, respectively. Finally, we identified transcription factor co-expression modules marking key temporal steps in neurogenic and oligodendrocyte lineage progression. Our data suggest functionally important spatial diversity in neurogenesis and oligodendrogenesis in the adult brain and reveal molecular correlates of adult NSC dormancy and lineage specialization.

Single-Cell Transcriptomic Analysis of Tumor Heterogeneity.

Intratumoral heterogeneity is among the greatest challenges in precision cancer therapy. However, developments in high-throughput single-cell RNA sequencing (scRNA-seq) may now provide the statistical power to dissect the diverse cellular populations of tumors. In the future these technologies might inform the selection of targeted combination therapies and enrollment criteria for clinical trials.

Single-cell transcriptome analysis of lineage diversity in high-grade glioma.

BACKGROUND: Despite extensive molecular characterization, we lack a comprehensive understanding of lineage identity, differentiation, and proliferation in high-grade gliomas (HGGs). METHODS: We sampled the cellular milieu of HGGs by profiling dissociated human surgical specimens with a high-density microwell system for massively parallel single-cell RNA-Seq. We analyzed the resulting profiles to identify subpopulations of both HGG and microenvironmental cells and applied graph-based methods to infer structural features of the malignantly transformed populations. RESULTS: While HGG cells can resemble glia or even immature neurons and form branched lineage structures, mesenchymal transformation results in unstructured populations. Glioma cells in a subset of mesenchymal tumors lose their neural lineage identity, express inflammatory genes, and co-exist with marked myeloid infiltration, reminiscent of molecular interactions between glioma and immune cells established in animal models. Additionally, we discovered a tight coupling between lineage resemblance and proliferation among malignantly transformed cells. Glioma cells that resemble oligodendrocyte progenitors, which proliferate in the brain, are often found in the cell cycle. Conversely, glioma cells that resemble astrocytes, neuroblasts, and oligodendrocytes, which are non-proliferative in the brain, are generally non-cycling in tumors. CONCLUSIONS: These studies reveal a relationship between cellular identity and proliferation in HGG and distinct population structures that reflects the extent of neural and non-neural lineage resemblance among malignantly transformed cells.