Gene structure heterogeneity drives transcription noise within human chromosomes

Abstract

Classical observations have long suggested there is a link between 3D gene structure and transcription1–4. However, due to the many factors regulating gene expression, and to the challenge of visualizing DNA and chromatin dynamics at the same time in living cells, this hypothesis has been difficult to quantitatively test experimentally. Here we take an orthogonal approach and use computer simulations, based on the known biophysical principles of genome organisation5–7, to simultaneously predict 3D structure and transcriptional output of human chromatin genome wide. We validate our model by quantitative comparison with Hi-C contact maps, FISH, GRO-seq and single-cell RNA-seq data, and provide the 3DGene resource to visualise the panoply of structures adopted by any active human gene in a population of cells. We find transcription strongly correlates with the formation of protein-mediated microphase separated clusters of promoters and enhancers, associated with clouds of chromatin loops, and show that gene noise is a consequence of structural heterogeneity. Our results also indicate that loop extrusion by cohesin does not affect average transcriptional patterns, but instead impacts transcriptional noise. These findings provide a functional role for intranuclear microphase separation, and an evolutionary mechanism for loop extrusion halted at CTCF sites, to modulate transcriptional noise.