Sequence-encoded organisation of RNA-protein condensates in embryonic development

Cells compartmentalise much of their biochemistry without membranes, in condensates built from RNA and protein. Which  molecules join which compartment has been hard to answer, because the determinants are distributed along whole sequences rather than concentrated in short motifs. We develop assays and interpretable deep learning models that read out this code from both sides of the RNA-protein interface.

Combining two orthogonal assays, semi-extractability and organic-phase separation, we defined a group of condensation-prone transcripts that resist standard extraction and carry high RNA-binding protein occupancy. These smOOPs (semi-extractable and orthogonal organic phase separation-enriched RNAs) localise to larger intracellular foci, are more densely interconnected with one another than their abundance predicts, and can be separated from control transcripts by sequence alone using an explainable deep learning model. They encode proteins unusually rich in intrinsically disordered regions, linking the condensation propensity of a transcript to that of the protein it specifies.
 

To resolve composition inside a single condensate, we developed HCR-Proxy, which couples hybridisation chain reaction signal amplification to in situ proximity biotinylation. Tuning molecular crowding confines the labelling radius enough to separate nested subcompartments. Targeting three positions along the nascent pre-rRNA recovered the distinct proteomes of the fibrillar centre, dense fibrillar component and granular component, and identified a previously uncharacterised nucleolar protein. Modelling these proteomes showed that interpretable features, molecular weight, net charge and RNA-binding domain content, predict where a protein partitions.

Both results point the same way: membership of a condensate is predictable from sequence. That turns the next question into a design problem. We are extending these models from classifying known transcripts to predicting behaviour across full-length sequences, and using them to specify synthetic RNA elements whose localisation and partitioning we can state in advance and then test by building them. Writing a sequence that obeys a proposed rule is the strongest validation available here, and it moves a descriptive account of condensate organisation towards a programmable one. The same framework applies where this organisation fails, including repeat-expansion disorders in which transcripts sequester proteins into aberrant assemblies.

Publications
Klobučar T, Novljan J, Iosub IA, et al., Ule J, Modic M. Integrative profiling of condensation-prone RNAs during early development. Cell Genomics 6, 101065 (2026). doi:10.1016/j.xgen.2025.101065

Trupej A, Bergant V, Novljan J, et al., Mardakheh FK, Modic M. HCR-Proxy resolves site-specific proximal RNA microenvironments at subcompartmental resolution. Nucleic Acids Research gkag086 (2026). doi:10.1093/nar/gkag086

KIT press release: New RNA class discovered that helps keep cells organized (088/2025)