Hypothesis

Recurrent SF3B1, SRSF2, U2AF1, and ZRSR2 mutations increase the formation of splicing factor condensates that trap nascent transcripts, raising local intron‑retention noise until a concentration‑dependent entropy threshold is crossed; cells that exceed this threshold become selectively vulnerable to spliceosome inhibition because their splicing machinery is already operating near collapse.

Mechanistic basis

Mutant spliceosomes alter Pol II kinetics and promote R‑loop formation, which stabilizes intron‑containing RNAs [https://pmc.ncbi.nlm.nih.gov/articles/PMC12765304/]. These RNAs phase‑separate with splicing factors, enlarging nuclear condensates and concentrating mutant spliceosome complexes. As condensate volume grows, the effective concentration of aberrant splice sites rises, increasing the probability that any given transcript will be mis‑spliced (intron retention or cryptic splice‑site usage). When the fraction of retained introns surpasses a critical entropy value—estimated from single‑cell transcriptomic noise metrics—the cell experiences a loss of buffering capacity for essential genes (e.g., MAP3K7, RAD51) and shifts from adaptive splicing plasticity to deterministic splicing failure.

This model extends the observation that cancers with spliceosome mutations show dependency on compromised splicing machinery [https://pmc.ncbi.nlm.nih.gov/articles/PMC12709091/] by proposing that the dependency is not static but emerges only after condensate‑driven entropy exceeds a threshold. Below the threshold, cells tolerate isoform noise; above it, they rely on residual spliceosome activity to keep essential transcripts above a viability floor, creating a therapeutic window for inhibitors like H3B-8800.

Testable predictions

1. In isogenic cell lines expressing SF3B1 K700E, SRSF2 P95H, U2AF1 S34F, or ZRSR2 loss, intron‑retention levels measured by long‑read RNA‑seq will correlate positively with the size and number of nuclear speckle‑like condensates visualized by live‑cell imaging of SRSF2‑GFP.
2. Using a titration of low‑dose spliceosome inhibitor, cells with condensate volume above the 75th percentile will show a steeper decline in viability (IC50 shift >2‑fold) compared with cells below this percentile, despite similar mutant allele frequencies.
3. Artificially increasing intron retention via CRISPR‑dCas9‑KRAB targeting of intronic splicing enhancers will lower the inhibitor concentration needed to induce apoptosis, while over‑expressing a constitutively active spliceosome co‑factor (e.g., SRPK1) will raise the threshold and confer resistance.
4. Single‑cell entropy scores derived from splice‑junction variance will predict response to H3B-8800 in patient‑derived xenografts; tumors with entropy scores exceeding a defined cutoff will exhibit >50% tumor growth inhibition, whereas those below will show <20% inhibition.

Experimental approach

• Generate CRISPR knock‑in models of each mutation in a myeloid progenitor line; tag endogenous SRSF2 with HaloTag for condensate quantification.
• Perform 4SU‑labelled nascent RNA sequencing and nanopore direct RNA sequencing to quantify intron retention and condensate-associated RNA.
• Apply gradient concentrations of H3B-8800 and measure viability (CellTiter‑Glo), γH2AX foci, and Annexin V staining.
• Use intravital speckle microscopy to correlate condensate dynamics with drug response in vivo.
• Analyze publicly available splice‑junction datasets from TCGA and BeatAML to compute entropy thresholds and test predictive power against clinical spliceosome‑inhibitor trial outcomes.

If these experiments confirm that condensate‑driven intron‑retention entropy predicts inhibitor sensitivity, the hypothesis provides a mechanistic, quantifiable framework for patient stratification and for designing combination strategies that push malignant cells over the entropy threshold while sparing normal tissue.