Time-series deconvolution methods and predictive models to link chromatin dynamics and transcriptional regulation

Abstract

Though the sequence of the genome within each eukaryotic cell is essentially fixed, it exists in a complex and changing chromatin state. This state is determined, in part, by the dynamic binding of proteins and factors to the DNA. These factors—including nucleosome histones, transcription factors (TFs), and polymerases—interact with one another, the genome, and other molecules to allow the chromatin to adopt one of innumerable possible configurations. Understanding how changes in the chromatin state associate with changes in transcription remains a fundamental research question.We sought to tackle this question by studying cell populations in two biological contexts: response to acute cadmium stress and progression through the cell division cycle. In each case, we analyzed time series data to identify temporal linkages between changes in chromatin and changes in transcription, and found these linkages were often part of cascading regulatory interactions. To better understand these temporal linkages, we developed specific models for each context.After placing cells under cadmium stress, we observed highly correlated patterns between expression and the chromatin state for many stress response genes. Using these patterns, we developed models to predict expression from chromatin changes for ~4,400 genes.While the context of cadmium stress yielded strong changes in both chromatin and transcription, identifying these kinds of regulation patterns within the cell cycle was far more challenging. In the context of the cell cycle, these patterns are subtler and confounded by factors such as loss of cell cycle synchrony, asymmetric cell division, and variable copy number during S-phase. To address these challenges, we developed a model to temporally deconvolve the chromatin landscape genome-wide. Our model learns jointly from replicate experiments to increase signal, reduce noise, and create high-resolution profiles for millions of chromatin measurements and ~5,700 gene transcripts throughout the cell cycle. Using these deconvolution profiles, we found that chromatin changes generally cycle independently from gene expression. However, for almost 100 genes, chromatin-transcription dynamics cycled together, allowing us to precisely characterize their temporal coordination.Together, our models for cadmium stress and cell cycle progression provide a framework for linking chromatin dynamics to transcriptional regulation, not only at well-characterized genes but also in overlooked regions, including those with non-genic transcription. With these models, we establish a systematic methodology for studying regulatory chromatin dynamics in diverse biological contexts.