HomeResearchReconstruction of Dynamically Rewiring Circuits: From Population to Single Cells

Our Scientists

Reconstruction of Dynamically Rewiring Circuits: From Population to Single Cells

Research Summary

Aviv Regev studies how gene regulation rewires when cells adapt to their environments, when they differentiate, and when species evolve.

Molecular circuits are the information-processing devices of cells and organisms, transforming extra- and intracellular signals into coherent cellular responses. Past studies to chart key circuits in mechanistic and functional detail typically have required decades of serial work to identify and connect a few components or interactions at a time. In recent years, there has been hope that genomic approaches could make it possible to reconstruct circuitry on a systems level. However, genomic studies have largely been observational and rarely involve large-scale testing of the models and subsequent refinement.

Our goal is to bring the rigor of classical experimental circuit research into genome-scale circuit reconstruction. We use a systematic computational and experimental approach based on iterating three steps. First, we measure the circuit's output (e.g., mRNA levels) or internal state (e.g., protein–DNA or protein modification states) along a relevant time course using genomic tools. Next, using machine learning algorithms we develop, we create a computational model that explains the observed data. We perturb every key component proposed by our model (e.g., using CRISPR). Then, we repeat the process by measuring the circuit's output or internal state after the perturbation, refining the model, and testing it again until data and model converge.

All circuits are dynamic and rewire in response to perturbation at time scales from minutes to eons, as cells respond to new environmental conditions, differentiate, or evolve. We focus on a carefully selected model system at each time scale. For short-term responses, we study the regulatory circuit of primary mouse dendritic cells (DCs), responding to pathogen components through Toll-like receptors. For long-term responses, we study the differentiation of hematopoietic cells either through discrete differentiation states in the hematopoietic lineage or continuously in T cell differentiation, as well as in the context of tissues like the complex ecosystem of the gut or tumors. For evolutionary changes, we study the rewiring of nutrient responses in a yeast phylogeny spanning 300 million years and 15 species.

We develop and apply an extensive experimental and computational toolbox for each step. Experimentally, this process includes novel methods to profile RNA, proteins and their interactions, nanotechnology-based delivery to primary cells, and mesoscale signatures to monitor the effect of hundreds of perturbations on hundreds of transcripts, proteins, or interactions. Computationally, we have pioneered sophisticated algorithms to reconstruct dynamic circuit models from time course and perturbation data; to design time course, perturbation, and signature experiments; and to facilitate analysis of large-scale genomic data sets, especially RNA-Seq.

Short-term Responses: The Transcriptional Circuitry of Dendritic Cells
We used mouse DCs to establish the measure-model-perturb approach, while building quantitative, high-resolution models. Our models associate regulatory and signaling proteins with their downstream transcriptional targets, quantify the contribution of different steps in the RNA life cycle to the circuit's output, position genetic variants (from mice and humans) within the circuits, analyze the role of posttranslational modifications of circuit components, and more.

For example, using temporal transcriptional profiles and a novel algorithm, we constructed a circuit of transcription factors and chromatin regulators controlling mRNA outputs of 1,800 genes in 80 modules. To distinguish direct from indirect interactions, we measured transcription factor binding profiles during DC stimulation and reconstructed a temporal hierarchical model of factor binding. Combining binding, expression, and perturbation we can distinguish which binding events are functional and identify how changes in binding set the timing and amplitude of gene induction. We then examined the signaling pathways that lie upstream of the transcriptional machinery. Using RNA profiles, we nominated, perturbed, and experimentally validated signaling regulators. We also leverage natural genetic variation as perturbation.  Using primary DCs from 96 recombinant inbred mouse strains and a novel algorithm, we identified loci that contribute to heritable variation in the response and positioned their product in our circuit model.

The emerging circuit substantially extends and revises the previously described system and proposes many novel regulators and interactions, with implications for human disease. For example, we have uncovered a module of five cell cycle transcription factors, which were coopted to control antiviral gene expression in postmitotic DCs, and a feed-forward loop involving the chromatin regulator CBX4, which is important in the specific induction of type I interferons. We discovered new signaling factors involved in the antiviral response and suggested a new potential use for a PLK inhibitor drug in autoimmune disease. Human orthologs of several of the novel regulators were associated with autoimmune diseases in genome-wide association studies.

A Deeper Understanding: Toward Single-Cell Genomic Analysis of Dynamic Circuits
Studies in recent years have shown that the level and activity of transcripts and proteins can vary substantially between isogenic cells in seemingly identical states. However, such studies typically relied on a handful of transcripts or proteins measured per cell. Recent technological advances, including single-cell DNA-Seq, RNA-Seq, and mass-flow cytometry (CyTOF), can help transform our understanding of the extent, causes, and function of cell-to-cell variation.

In particular, we are beginning to harness single-cell genomics for circuit reconstruction by addressing several key challenges in experimental measurements and parallelization, devising methods to build temporal dynamic models from single cell and population level data, and combining perturbations with single-cell analysis. In early work, we used single-cell transcriptome profiling to show that stimulated DCs show bimodal heterogeneity in transcript levels and splice forms. Some highly expressed transcripts, such as housekeeping genes, show little variation among cells, while others, including key immune response genes, are bimodal, either highly expressed in a cell or not present at all. Analyzing transcripts' covariation across cells, we inferred regulatory modules and the regulators that may underlie this variation.

Rewiring in Immune Cell Differentiation
Tremendous progress has been made in identifying individual factors that regulate cell differentiation, but many aspects of the global architecture of differentiation circuits remain unknown, including the number of key factors and how cell states are maintained or change. We are developing learning algorithms that can handle multiple, partly overlapping waves of rewiring over longer time scales and that can determine how circuits change along a differentiation lineage tree.

In particular, we study the regulatory circuit of Th17 cell differentiation. Th17 cells are proinflammatory immune cells involved in many autoimmune diseases. We measured transcriptional profiles along a time course of Th17 differentiation and developed a model with 80 key regulators embedded within a time-varying protein–protein, protein–DNA network. We used nanowires—a technology developed by the Park lab at Harvard—to deliver RNAi to test and refine our model, showing the network is organized into two coupled, self-reinforcing yet mutually antagonistic modules, explaining how balance is maintained in T cell differentiation. The nodes defined by our model have identified pathways essential in Th17 differentiation and autoimmunity. For example, our models proposed that the inducible kinase SGK1 (not previously implicated in Th17 differentiation) has the strongest effect on a Th17 cell maintenance circuit. Using T cells from conditional Sgk1 knockout mice, we showed in collaboration with Vijay Kuchroo's lab that SGK1 is essential to maintain the Th17 transcriptional program in vitro and to sustain pathogenic Th17 cells and autoimmunity in mice. Since SGK1 is induced by a high salt concentration, we hypothesized a novel role for a high salt diet in promoting Th17 pathogenesis and autoimmunity; we have demonstrated this process experimentally in mice.

From Ontogeny to Phylogeny: Circuit Rewiring in Evolution
We hypothesize that similar principles may apply when networks rewire epigenetically at short time scales and genetically during evolution. For the latter, we are tracing the evolution of regulatory circuits in 15 Ascomycota yeast species over 300 million years of evolution. We measured transcriptional and chromatin profiles in each species in nutritional and environmental stresses and used a new algorithm, Arboretum, to infer regulatory circuits in each extant and ancestral species. We identified changes in chromatin regulators and transcription factors that govern rewiring in the phylogeny. We perturbed their orthologs and paralogs across species and measured their effect on the transcriptional response and on chromatin organization.

We identified key principles of regulatory evolution and specific mechanisms by which gene regulation rewires in evolution. In general, regulatory modules diverge proportionally to phylogenetic distance, with prominent changes accompanying changes in lifestyle and ploidy. Gene paralogs significantly contribute to regulatory divergence within a very short window from their duplication, which is extended for paralogs from a whole-genome duplication.

Focusing on the control of a conserved module of growth genes (such as ribosomal protein and ribosome biogenesis genes), we showed: how the regulators controlling this module have switched in the phylogeny, possibly driven by selection to maintain coordination with telomere regulation; how changes in antinucleosomal sequences and general regulatory factors underlie changes in their metabolism-dependent regulation; how duplication and loss of one regulator led to the loss of stress-dependent repression in a human pathogen; and how duplication and divergence of another regulator led to sub- and neofunctionalization of its transcriptional role and its effect on cell growth and size in several lineages.

These projects are partially supported by grants from the National Institutes of Health, the Manton Foundation, and the Klarman Family Foundation.

As of April 12, 2016

Scientist Profile

Massachusetts Institute of Technology
Computational Biology, Systems Biology