Noise genetics: inferring protein function by correlating phenotype with protein levels and localization in individual human cells

Mapping Phenotypic Plasticity upon the Cancer Cell State Landscape Using Manifold Learning

ABSTRACT

Phenotypic plasticity describes the ability of cancer cells to undergo dynamic, nongenetic cell state changes that amplify cancer heterogeneity to promote metastasis and therapy evasion. Thus, cancer cells occupy a continuous spectrum of phenotypic states connected by trajectories defining dynamic transitions upon a cancer cell state landscape. With technologies proliferating to systematically record molecular mechanisms at single-cell resolution, we illuminate manifold learning techniques as emerging computational tools to effectively model cell state dynamics in a way that mimics our understanding of the cell state landscape. We anticipate that "state-gating" therapies targeting phenotypic plasticity will limit cancer heterogeneity, metastasis, and therapy resistance.

SIGNIFICANCE

Nongenetic mechanisms underlying phenotypic plasticity have emerged as significant drivers of tumor heterogeneity, metastasis, and therapy resistance. Herein, we discuss new experimental and computational techniques to define phenotypic plasticity as a scaffold to guide accelerated progress in uncovering new vulnerabilities for therapeutic exploitation.

Inferring the structures of signaling motifs from paired dynamic traces of single cells

Individual cells show variability in their signaling dynamics that often correlates with phenotypic responses, indicating that cell-to-cell variability is not merely noise but can have functional consequences. Based on this observation, we reasoned that cell-to-cell variability under the same treatment condition could be explained in part by a single signaling motif that maps different upstream signals into a corresponding set of downstream responses. If this assumption holds, then repeated measurements of upstream and downstream signaling dynamics in a population of cells could provide information about the underlying signaling motif for a given pathway, even when no prior knowledge of that motif exists. To test these two hypotheses, we developed a computer algorithm called MISC (Motif Inference from Single Cells) that infers the underlying signaling motif from paired time-series measurements from individual cells. When applied to measurements of transcription factor and reporter gene expression in the yeast stress response, MISC predicted signaling motifs that were consistent with previous mechanistic models of transcription. The ability to detect the underlying mechanism became less certain when a cell's upstream signal was randomly paired with another cell's downstream response, demonstrating how averaging time-series measurements across a population obscures information about the underlying signaling mechanism. In some cases, motif predictions improved as more cells were added to the analysis. These results provide evidence that mechanistic information about cellular signaling networks can be systematically extracted from the dynamical patterns of single cells.

Discovery and characterization of variance QTLs in human induced pluripotent stem cells

Quantification of gene expression levels at the single cell level has revealed that gene expression can vary substantially even across a population of homogeneous cells. However, it is currently unclear what genomic features control variation in gene expression levels, and whether common genetic variants may impact gene expression variation. Here, we take a genome-wide approach to identify expression variance quantitative trait loci (vQTLs). To this end, we generated single cell RNA-seq (scRNA-seq) data from induced pluripotent stem cells (iPSCs) derived from 53 Yoruba individuals. We collected data for a median of 95 cells per individual and a total of 5,447 single cells, and identified 235 mean expression QTLs (eQTLs) at 10% FDR, of which 79% replicate in bulk RNA-seq data from the same individuals. We further identified 5 vQTLs at 10% FDR, but demonstrate that these can also be explained as effects on mean expression. Our study suggests that dispersion QTLs (dQTLs) which could alter the variance of expression independently of the mean can have larger fold changes, but explain less phenotypic variance than eQTLs. We estimate 4,015 individuals as a lower bound to achieve 80% power to detect the strongest dQTLs in iPSCs. These results will guide the design of future studies on understanding the genetic control of gene expression variance.

Common genetic variation can alter the level of average gene expression in human tissues, and through changes in gene expression have downstream consequences on cell function, human development, and human disease. However, human tissues are composed of many cells, each with its own level of gene expression. With advances in single cell sequencing technologies, we can now go beyond simply measuring the average level of gene expression in a tissue sample and directly measure cell-to-cell variance in gene expression. We hypothesized that genetic variation could also alter gene expression variance, potentially revealing new insights into human development and disease. To test this hypothesis, we used single cell RNA sequencing to directly measure gene expression variance in multiple individuals, and then associated the gene expression variance with genetic variation in those same individuals. Our results suggest that effects on gene expression variance are smaller than effects on mean expression, relative to how much the phenotypes vary between individuals, and will require much larger studies than previously thought to detect.

The fitness cost and benefit of phase-separated protein deposits

Phase separation of soluble proteins into insoluble deposits is associated with numerous diseases. However, protein deposits can also function as membrane-less compartments for many cellular processes. What are the fitness costs and benefits of forming such deposits in different conditions? Using a model protein that phase-separates into deposits, we distinguish and quantify the fitness contribution due to the loss or gain of protein function and deposit formation in yeast. The environmental condition and the cellular demand for the protein function emerge as key determinants of fitness. Protein deposit formation can influence cell-to-cell variation in free protein abundance between individuals of a cell population (i.e., gene expression noise). This results in variable manifestation of protein function and a continuous range of phenotypes in a cell population, favoring survival of some individuals in certain environments. Thus, protein deposit formation by phase separation might be a mechanism to sense protein concentration in cells and to generate phenotypic variability. The selectable phenotypic variability, previously described for prions, could be a general property of proteins that can form phase-separated assemblies and may influence cell fitness.

Image-based spatiotemporal causality inference for protein signaling networks

Motivation

Efforts to model how signaling and regulatory networks work in cells have largely either not considered spatial organization or have used compartmental models with minimal spatial resolution. Fluorescence microscopy provides the ability to monitor the spatiotemporal distribution of many molecules during signaling events, but as of yet no methods have been described for large scale image analysis to learn a complex protein regulatory network. Here we present and evaluate methods for identifying how changes in concentration in one cell region influence concentration of other proteins in other regions.

Results

Using 3D confocal microscope movies of GFP-tagged T cells undergoing costimulation, we learned models containing putative causal relationships among 12 proteins involved in T cell signaling. The models included both relationships consistent with current knowledge and novel predictions deserving further exploration. Further, when these models were applied to the initial frames of movies of T cells that had been only partially stimulated, they predicted the localization of proteins at later times with statistically significant accuracy. The methods, consisting of spatiotemporal alignment, automated region identification, and causal inference, are anticipated to be applicable to a number of biological systems.

Availability and implementation

The source code and data are available as a Reproducible Research Archive at http://murphylab.cbd.cmu.edu/software/2017_TcellCausalModels/

Accurate Classification of Protein Subcellular Localization from High-Throughput Microscopy Images Using Deep Learning

High-throughput microscopy of many single cells generates high-dimensional data that are far from straightforward to analyze. One important problem is automatically detecting the cellular compartment where a fluorescently-tagged protein resides, a task relatively simple for an experienced human, but difficult to automate on a computer. Here, we train an 11-layer neural network on data from mapping thousands of yeast proteins, achieving per cell localization classification accuracy of 91%, and per protein accuracy of 99% on held-out images. We confirm that low-level network features correspond to basic image characteristics, while deeper layers separate localization classes. Using this network as a feature calculator, we train standard classifiers that assign proteins to previously unseen compartments after observing only a small number of training examples. Our results are the most accurate subcellular localization classifications to date, and demonstrate the usefulness of deep learning for high-throughput microscopy.

A metric and workflow for quality control in the analysis of heterogeneity in phenotypic profiles and screens

Heterogeneity is well recognized as a common property of cellular systems that impacts biomedical research and the development of therapeutics and diagnostics. Several studies have shown that analysis of heterogeneity: gives insight into mechanisms of action of perturbagens; can be used to predict optimal combination therapies; and can be applied to tumors where heterogeneity is believed to be associated with adaptation and resistance. Cytometry methods including high content screening (HCS), high throughput microscopy, flow cytometry, mass spec imaging and digital pathology capture cell level data for populations of cells. However it is often assumed that the population response is normally distributed and therefore that the average adequately describes the results. A deeper understanding of the results of the measurements and more effective comparison of perturbagen effects requires analysis that takes into account the distribution of the measurements, i.e. the heterogeneity. However, the reproducibility of heterogeneous data collected on different days, and in different plates/slides has not previously been evaluated. Here we show that conventional assay quality metrics alone are not adequate for quality control of the heterogeneity in the data. To address this need, we demonstrate the use of the Kolmogorov-Smirnov statistic as a metric for monitoring the reproducibility of heterogeneity in an SAR screen, describe a workflow for quality control in heterogeneity analysis. One major challenge in high throughput biology is the evaluation and interpretation of heterogeneity in thousands of samples, such as compounds in a cell-based screen. In this study we also demonstrate that three heterogeneity indices previously reported, capture the shapes of the distributions and provide a means to filter and browse big data sets of cellular distributions in order to compare and identify distributions of interest. These metrics and methods are presented as a workflow for analysis of heterogeneity in large scale biology projects.

The Arp2/3 inhibitory protein arpin induces cell turning by pausing cell migration

Branched actin networks generated by the Arp2/3 complex provide the driving force for leading edge protrusion in migrating cells. We recently identified Arpin, a protein that inhibits the Arp2/3 complex in lamellipodia. Arpin is activated by the small GTPase Rac, which triggers lamellipodium formation, and thus Arpin renders protrusions unstable. A conserved role of Arpin is to induce migrating cells to turn in different migration models. Here we investigated the mechanism by which Arpin controls directional persistence. For this analysis, we segmented migration trajectories into alternating phases of active migration and pauses, based on a speed threshold. Regardless of the threshold value, Arpin induced more frequent pausing, during which the cell was more likely to change the direction of its migration. Arpin simultaneously acts on cell speed and directional persistence, which are strongly coupled parameters. Induction of frequent pausing by Arpin is consistent with Arpin circuitry: by inhibiting the Arp2/3 complex as a response to Rac activation, Arpin antagonizes a positive feedback loop that sustains protrusions at the leading edge and maintains active migration. We propose the 'duration of active migration' as a useful proxy to measure feedbacks associated with cell migration.

Surviving apoptosis: life-death signaling in single cells

Tissue development and homeostasis are regulated by opposing pro-survival and pro-death signals. An interesting feature of the Tumor Necrosis Factor (TNF) family of ligands is that they simultaneously activate opposing signals within a single cell via the same ligand-receptor complex. The magnitude of pro-death events such as caspase activation and pro-survival events such as Nuclear Factor (NF)-κB activation vary not only from one cell type to the next but also among individual cells of the same type due to intrinsic and extrinsic noise. The molecules involved in these pro-survival and/or pro-death pathways, and the different phenotypes that result from their activities, have been recently reviewed. Here we focus on the impact of cell-to-cell variability in the strength of these opposing signals on shaping cell fate decisions.

Non-monotonic cellular responses to heterogeneity in talin protein expression-level

Talin is a key cell-matrix adhesion component with a central role in regulating adhesion complex maturation, and thereby various cellular properties including adhesion and migration. However, knockdown studies have produced inconsistent findings regarding the functional influence of talin in these processes. Such discrepancies may reflect non-monotonic responses to talin expression-level variation that are not detectable via canonical "binary" comparisons of aggregated control versus knockdown cell populations. Here, we deployed an "analogue" approach to map talin influence across a continuous expression-level spectrum, which we extended with sub-maximal RNAi-mediated talin depletion. Applying correlative imaging to link live cell and fixed immunofluorescence data on a single cell basis, we related per cell talin levels to per cell measures quantitatively defining an array of cellular properties. This revealed both linear and non-linear correspondences between talin expression and cellular properties, including non-monotonic influences over cell shape, adhesion complex-F-actin association and adhesion localization. Furthermore, we demonstrate talin level-dependent changes in networks of correlations among adhesion/migration properties, particularly in relation to cell migration speed. Importantly, these correlation networks were strongly affected by talin expression heterogeneity within the natural range, implying that this endogenous variation has a broad, quantitatively detectable influence. Overall, we present an accessible analogue method that reveals complex dependencies on talin expression-level, thereby establishing a framework for considering non-linear and non-monotonic effects of protein expression-level heterogeneity in cellular systems.

On comparing heterogeneity across biomarkers

Microscopy reveals complex patterns of cellular heterogeneity that can be biologically informative. However, a limitation of microscopy is that only a small number of biomarkers can typically be monitored simultaneously. Thus, a natural question is whether additional biomarkers provide a deeper characterization of the distribution of cellular states in a population. How much information about a cell's phenotypic state in one biomarker is gained by knowing its state in another biomarker? Here, we describe a framework for comparing phenotypic states across biomarkers. Our approach overcomes the current limitation of microscopy by not requiring costaining biomarkers on the same cells; instead, we require staining of biomarkers (possibly separately) on a common collection of phenotypically diverse cell lines. We evaluate our approach on two image datasets: 33 oncogenically diverse lung cancer cell lines stained with 7 biomarkers, and 49 less diverse subclones of one lung cancer cell line stained with 12 biomarkers. We first validate our method by comparing it to the "gold standard" of costaining. We then apply our approach to all pairs of biomarkers and use it to identify biomarkers that yield similar patterns of heterogeneity. The results presented in this work suggest that many biomarkers provide redundant information about heterogeneity. Thus, our approach provides a practical guide for selecting independently informative biomarkers and, more generally, will yield insights into both the connectivity of biological networks and the complexity of the state space of biological systems.

Computational analysis of signaling patterns in single cells

Signaling proteins are flexible in both form and function. They can bind to multiple molecular partners and integrate diverse types of cellular information. When imaged by time-lapse microscopy, many signaling proteins show complex patterns of activity or localization that vary from cell to cell. This heterogeneity is so prevalent that it has spurred the development of new computational strategies to analyze single-cell signaling patterns. A collective observation from these analyses is that cells appear less heterogeneous when their responses are normalized to, or synchronized with, other single-cell measurements. In many cases, these transformed signaling patterns show distinct dynamical trends that correspond with predictable phenotypic outcomes. When signaling mechanisms are unclear, computational models can suggest putative molecular interactions that are experimentally testable. Thus, computational analysis of single-cell signaling has not only provided new ways to quantify the responses of individual cells, but has helped resolve longstanding questions surrounding many well-studied human signaling proteins including NF-κB, p53, ERK1/2, and CDK2. A number of specific challenges lie ahead for single-cell analysis such as quantifying the contribution of non-cell autonomous signaling as well as the characterization of protein signaling dynamics in vivo.