Abstract:
Background: Messenger RNA (mRNA) therapeutics offer immense clinical promise, yet achieving cell- and tissue-selective expression remains a critical challenge. Tumor cells frequently hyperactivate the eIF4A translation initiation axis, granting them a selective translational advantage for transcripts containing highly structured 5' untranslated regions (5'UTRs). This tumor-specific translational phenotype represents a powerful lever for designing oncogene-targeted mRNA payloads. However, extracting precise sequence-to-translation rules directly from patient tissue is severely confounded by cellular composition: bulk ribosome profiling (Ribo-seq) and RNA sequencing (RNA-seq) average signals across malignant, stromal, and immune populations, masking true tumor-intrinsic translational efficiency (TE).
Methods: To isolate unconfounded translational signals, we developed a context-aware transcriptomic analysis pipeline that integrates tissue deconvolution with differential translation modeling. Utilizing single-cell RNA-seq references, the framework estimates sample-specific cell-type proportions and incorporates them as covariates within an empirical-Bayes moderated differential TE analysis. Using the resulting composition-corrected translatome maps, we applied minimum free energy folding models to canonical 5'UTRs. Rather than relying on standard segmentation heuristics, we implemented a rigorous segmented regression framework and evaluated the slope transitions using the Davies test to identify statistically robust thermodynamic thresholds governing structure-dependent translation.
Results: Application of this framework to existing datasets demonstrates that standard bulk analyses miscall or entirely miss key translationally regulated genes due to immune and stromal cell infiltration. By correcting for these shifts, our pipeline successfully recovered hidden, tumor-intrinsic translational signals. Furthermore, structural rule mining revealed a distinct, highly non-linear relationship between UTR folding energy and TE in the tumor state. The Davies test successfully identified the exact thermodynamic onset where secondary structure begins to significantly alter translation, establishing a clear biological threshold for engineering tumor-selective expression.
Conclusion: This work provides a reproducible, statistically rigorous approach to mining patient translatomes. By defining the exact structural boundaries of translation initiation in patient tumors, this framework establishes the fundamental, mathematically validated design rules required to build next-generation, context-selective mRNA therapeutics.