Molecular mechanisms of functional phase separation in eukaryotic gene transcription

Transcription of protein-coding genes by RNA polymerase II (Pol II) is highly regulated during cell differentiation and organismal development. In human cells, Pol II transcription is regulated by ~1,600 transcription factors, which bind specific DNA sequences in the vicinity of target genes and contain low-complexity regions that often function in transcription activation. Evidence for the involvement of liquid-liquid phase transitions in transcription initiation, pausing, and elongation has continued to accumulate. We showed, for instance, that the highly conserved C-terminal domain (CTD) of Pol II can phase separate and mediate the clustering of Pol II enzymes in human cells. These results provide a great starting point for analyzing how transcription factors interact with Pol II, using an integrated experimental-theoretical approach. Our efforts will elucidate the phenomenon of phase separation and its functional importance for gene transcription, and will address the enigmatic question of how the different transcription factors can interact with the same general transcription machinery.

Current State of Research

The Cramer lab purifies the transcription factors in a recombinant form and investigates their phase separation propensity as well as in vitro specificity regarding known transcriptional condensates (Cramer, 2019). To this end, they recently co-discovered a distinct, previously unknown transcriptional condensate that is essential for stress-induced transcriptional downregulation and cellular stress survival (Rawat, Boehning et al., 2021). Extending the analysis of the governing intermolecular interactions and in vitro specificity towards transcription factors will help to shed light on the sequence determinants that regulate condensate residency.

The Soeding lab has systematically analyzed data from an unbiased, high-throughput screen for transcriptional activation using a deep neural network. The analysis highlights which sequence features determine the potency of transcriptional activation (Erijman et al, Mol. Cell 2020). They are currently extending this analysis by training a machine learning classifier to predict condensate membership for transcription factors and other phase-separating proteins with intrinsically disordered domains (IDRs). They use representation learning with a pretrained transformer neural network to learn which sequence features predict the tendency to phase separate and the target condensates.  

The Zweckstetter lab analyzes residual structure in phase-separated CTD and transcription factor samples by nuclear magnetic resonance, aiming at the molecular basis of the interactions.