Answers are grounded in this paper and cite their sources.
SOX2 is cited as an RNA-binding protein on the basis of two independent 2020 studies: RNA immunoprecipitation (RIP) in mouse embryonic stem cells, which described high-affinity, low-specificity binding to double-stranded RNA, and PAR-CLIP in HEK293T together with RNA SELEX, which described a GC-rich CCCY core motif and a CGCG side motif. Both datasets are public. I asked which parts of this evidence survive when control RNA-binding proteins are processed through an identical pipeline, and when the raw SOX2 reads are reprocessed rather than taken as deposited peak calls. Three published claims do not survive. The two datasets do not identify the same targets: length- and abundance-corrected RIP targets show no enrichment among PAR-CLIP-bound genes (Mantel-Haenszel odds ratio 0.891, 95% CI 0.800-0.991), and neither species nor cell type accounts for this. Under one pipeline applied to twelve PAR-CLIP datasets from six studies, the CCCY odds ratio for SOX2 (1.16-1.22) is lower than for the unrelated proteins FXR1 and FXR2 (1.69-1.78, non-overlapping intervals), while the same pipeline recovers the known preferences of ZFP36 and FMRP and returns null results for MATR3. And in the SELEX pool, whose GC content is 0.670, GCCCY enrichment falls from 5.68-fold against uniform sequence to 1.36-fold against a composition-matched background (p = 0.26). One feature survives, reduced in magnitude. Scoring T-to-C conversions against a background drawn from the same aligned reads, and against the non-T-to-C mismatch rate as an internal specificity control, gives an apparent thirtyfold excess of specificity in coding exons over introns. But 42% to 46% of conversion events lie in structured non-coding RNA or outside annotated genes, and 34 genes account for half of one replicate even after those classes are removed. With them removed and each gene limited to a fixed contribution, the contrast falls to roughly fourfold and the odds ratio against read placement is 1.08-1.16 for exonic against 0.81-0.88 for intronic positions, reproducibly in both libraries; mappability, thymidine content, splice-junction soft-clipping, common variants and local GC content do not account for it. Exonic targets are also not enriched among the genes SOX2 occupies in human embryonic stem cells (odds ratio 0.879, 95% CI 0.753-1.027). The public data therefore support a modest compartment preference rather than a sequence preference, and the reported motifs require identically processed control proteins before they can be attributed to SOX2.
No takes yet. Share an insight, caveat, or question.
Kenji Ohe (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: