Sequence-intrinsic barriers define a class of translation-restricted uORFs in the human genome

Abstract

Ribosome profiling has revealed widespread translation of upstream open reading frames (uORFs) within the human 5' untranslated region, greatly expanding the apparent coding potential of the genome. However, despite pervasive translational activity, uORF-encoded proteins (uORFps) are only rarely detected by proteomic approaches, creating a major unresolved discrepancy between translation and protein accumulation. Here, we systematically quantified the intracellular accumulation of 111 evolutionarily conserved human uORFps using a unified reporter platform. Although we found that most conserved uORFps accumulated at extremely low levels despite robust transcript expression, protein length emerged as the strongest baseline predictor of this low accumulation. Intriguingly, a subset of conserved uORFs remained poorly accumulated even after experimental extension with a C-terminal EGFP tag, defining a distinct class of translation-restricted uORFs. Sequence and structural analysis of these translation-restricted uORFs revealed a cooperative enrichment of basic amino acids and stable local RNA secondary structures. Synonymous substitutions designed to disrupt these RNA structures substantially restored protein accumulation, demonstrating a major contribution of local transcript stratification to this translation-restricted phenotype. Using these identified rules, we developed a multivariable predictive model that classified 1,432 translation-restricted uORFs across the human transcriptome. Genes harboring these predicted translation-restricted uORFs exhibited significantly reduced downstream translation efficiency and elevated sequence conservation, demonstrating that these sequence-encoded barriers are under purifying selection to act as cis-regulatory elements scattered across the human transcriptome. Together, our findings identify a conserved class of human uORFs that encode intrinsic barriers to productive translation and provide a rigorous framework for understanding how noncanonical coding sequences shape the human proteome.