Alternative tandem transcription initiation links noncoding variants to human disease through translational control

Alternative tandem transcription initiation is a pervasive mechanism of gene regulation, yet its genetic impact on human disease remains largely unknown. Here, we systematically quantify the genetic regulation of alternative tandem transcription initiation across 25,859 samples from 49 normal human tissues and 33 tumor tissues. We identify approximately 0.4 million genetic variants associated with alternative transcription initiation in 5295 genes, with 32% operating independently of gene expression. Moreover, we discover 2238 multi-tissue alternative tandem transcription initiation outliers enriched for rare deleterious promoter and 5' UTR variants, demonstrating that both common and rare variants modulate transcription initiation. Strikingly, 74% of disease variants that colocalize with genetic variants regulating alternative transcription initiation cannot be identified through expression quantitative trait loci. Transcriptome-wide association studies identify 614 disease susceptibility genes associated with alternative transcription initiation, including known cancer drivers such as MAFF and MLLT10. Functional validation uncovers OSGEP as a breast cancer risk gene, where the alternative allele lengthens the 5' UTR and reduces protein abundance through upstream open reading frame-mediated translation repression, and suppresses breast cancer cell proliferation. Our findings establish alternative transcription initiation as a major, underappreciated mechanism associating noncoding variation with disease, providing a critical resource for interpreting disease risk loci.