Rbfox2 controls autoregulation in RNA-binding protein networks.
Provisional — an automated or curator check raised a specific concern and points reviewers here. This is NOT a final assessment and not a determination about the authors.
The main result did not reproduce in this reproduction attempt. Where our recomputation produced values that differ from the published ones, those discrepancies are listed below. This is a single automated attempt — not peer review and not a finding of error or misconduct — and differences can also arise from data access, undocumented parameters or the computing environment. The verdict can be contested via “report an error”.
Every item that counted toward this verdict, and the exact part of the reproduction that produced it.
- ✓Same input data as the authors
- 🟡Reported values were only indirectly comparable
- 🟡A deviation arose in the data or preprocessing
- 🟡A deviation was attributed to the published material
- 🟡Reported values were not (fully) derivable from the shared data
- 🟡The deviation was non-trivial in magnitude
- 🟡The central claim did not (fully) hold under reproduction
- 🟡Overall, the reproduction showed a material discrepancy
A 0–100 reproducibility-quality score from the per-question grades, shown as a z-score: standard deviations above (+) or below (−) the mean of comparable assessments.
▸Reproduction agent’s raw note
Partial, honest reproduction with a mix of clean confirmations and clear mismatches -- not a 1:1 numeric reproduction, but the core biology is qualitatively supported on several fronts while specific quantitative claims and one central claim (Rbfox2 self-splicing autoregulation) do not reproduce. CONFIRMED: (1) Rbfox2 knockdown is strongly and significantly reflected in the RNA-seq data (edgeR: logFC=-1.62, FDR=1.4e-05); (2) Rbfox2-specific iCLIP peaks (FHFOX2) are clearly and substantially more enriched for the UGCAUG binding motif (24.0% of peaks, ~11.8x over flanking background) than the non-specific control sample (UntagFOX2, 3.8%, ~2.8x) -- a clean, independent confirmation of Rbfox2's known RNA-binding specificity; (3) widespread splicing disruption across all 5 rMATS event types upon knockdown, qualitatively matching the paper's general finding, despite the MISO-to-rMATS tool substitution. MISMATCH/PARTIAL: (1) raw iCLIP cluster counts are 3-6x higher than reported for both samples (UntagFOX2: 18,339 vs ~2,961; FHFOX2: 130,573 vs ~40,243), most plausibly due to a simplified PCR-duplicate-removal method (position+strand collapse instead of the paper's original randomer/UMI-based approach) and undocumented additional post-filtering in the original pipeline; (2) only 3 of 6 RT-PCR-validated splicing target genes (Tia1, Srsf7, Tra2a) show statistically significant, modest-effect-size splicing changes in our rMATS analysis, the other 3 (Snrnp70, Ptbp2, Sf1) do not, likely reflecting differences between MISO's and rMATS's exact exon-event definitions rather than a true biological discrepancy; (3) the paper's central autoregulation claim -- that Rbfox2 splices its own transcript to autoregulate its expression -- was NOT reproduced: no significant self-splicing event was found for Rbfox2 in any of the 5 rMATS event types. NOT ATTEMPTED: genomic-feature distribution of iCLIP peaks, any analysis of the broader multi-RBP autoregulatory network the paper's title alludes to (only Rbfox2's own iCLIP+RNA-seq data was reproduced), and independent re-derivation of the paper's specific iCLIP barcode/randomer structure or 'high-confidence cluster' filtering criteria. Disclosed tool substitutions: STAR for TopHat2/Cufflinks2.1.1 (RNA-seq alignment), edgeR for DESeq2 (DESeq2 was not actually installed in the environment despite initial assumption), rMATS for MISO (splicing analysis). One infrastructure interruption occurred mid-run: a portal restart briefly severed the VPN tunnel to the HPC cluster, but all in-flight SLURM jobs survived unaffected and were picked back up once connectivity was restored -- no data or compute was lost. Large reusable intermediates (raw/trimmed FASTQ, BAM files, STAR genome index) remain on «infra» scratch and were NOT deleted, per this session's standing rule that deletions always require explicit confirmation, which could not be obtained in this autonomous run; the operator may reclaim ~10s of GB by removing «path»,ref/star_index,iclip/_star,rnaseq/_star} once the results are reviewed.
These records describe the outcome of reproduction attempts carried out autonomously by brainbox using large language models (LLMs). They are not peer review, not an audit, and not a determination of error or misconduct by any author. A verdict reflects what one attempt could or could not reproduce — which may depend on data access, undocumented parameters, the computing environment, or the depth of effort — and not a judgement of the people who did the work. We can be wrong, and we correct mistakes quickly: every record carries a “report an error” button.
✎ I am an author of this paper
Updated or fixed a deposit, or is there an erratum? Ask us to re-run the metrics. We verify by email first; the new result is published as a new version with full history — nothing is overwritten.
Provenance — full disclosure
When this reproduction was carried out, which methodology version was used, and by whom — so the record can be audited and checked independently.
- Reproduced
- 2026-07-31
- Rubric version
- v1.0
- Assessed by
-
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-07-31no human curator yet
- Last updated
- 2026-07-31
Provisional, curator- or AI-assessed, and independently checkable. A reproduction outcome states what one attempt could reproduce — not a judgement of the authors.
Deep full-text extraction
Model: opusHow is negative autoregulation of splicing factors modulated so that different steady-state RNA-binding protein levels can arise in different physiological contexts? The paper tests whether Rbfox2 cross-regulates alternative splicing-coupled nonsense-mediated decay (AS-NMD) events in other RNA-binding protein genes, thereby tuning their autoregulatory networks.
- ★ Rbfox2 cross-regulates AS-NMD events within RNA-binding protein genes to alter their expression, tuning autoregulatory splicing networks and placing Rbfox2 at a critical node of a multilayer regulatory network. mechanism
- ★ Combining iCLIP and mRNA sequencing identified >200 AS-NMD splicing events bound by Rbfox2 in mouse embryonic stem cells. finding
- ★ These 'silent' AS-NMD events show minimal apparent splicing changes but appreciable gene expression changes upon Rbfox2 knockdown, because the NMD-inducing isoform is degraded; CLIP-seq data are therefore uniquely able to uncover them. finding
- ★ Nearly 70 of the Rbfox2-bound AS-NMD events fall within genes encoding RNA-binding proteins, many of which are autoregulated. finding
- ★ Most in vivo Rbfox2 binding is driven by the canonical (U)GCAUG motif, revising the earlier estimate that only one-third of binding sites contain the motif; silent and coding splicing events are evolutionarily conserved and frequently contain UGCAUG. finding
- ★ Rbfox2 regulation is position-dependent: cross-linking within ~200 nt downstream from a cassette exon (and near the downstream constitutive 3' splice site) correlates with exon activation, whereas cross-linking ~300-400 nt upstream correlates weakly with repression. mechanism
- Rbfox2-bound constitutive introns have significantly weaker splice sites than unbound constitutive introns, suggesting Rbfox2 also increases the efficiency of a subset of weaker constitutive splicing events. finding
- ★ A modified iCLIP protocol using doxycycline-inducible Flag-HA-tagged RBFOX2 with sequential Flag/HA immunoprecipitation and an untagged-RBFOX2 background control yields a high-specificity, single-nucleotide-resolution Rbfox2 RNA-binding map in mESCs. method
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| iCLIP (individual nucleotide-resolution cross-linking immunoprecipitation coupled to high-throughput sequencing) | V6.5 mouse embryonic stem cells stably transduced with lentiviral doxycycline-inducible Flag-HA-tagged human RBFOX2 (FHFOX2) | overexpression of tagged RBFOX2 transgene titrated with doxycycline to near-endogenous Rbfox2 levels | single-nucleotide cross-link sites and significant binding clusters across the transcriptome | sequential Flag and HA immunoprecipitation; high-throughput sequencing |
| iCLIP background control | V6.5 mESCs expressing untagged human RBFOX2 (UntagFOX2) | untagged RBFOX2 expression (negative control for tag-based IP) | background cross-link site distribution used for enrichment/binomial testing | sequential Flag and HA immunoprecipitation; high-throughput sequencing |
| mRNA sequencing (poly(A)-selected RNA-seq), biological duplicates | V6.5 mouse embryonic stem cells | lentiviral shRNA knockdown of Rbfox2 3'UTR (shFox2-1, shFox2-2) versus control hairpins (shLuc, shGFP) | alternative splicing changes (percent spliced in, psi/Δψ) and gene expression changes | Illumina |
| RT-PCR splicing validation | V6.5 mouse embryonic stem cells | shRNA-mediated Rbfox2 knockdown | inclusion levels of candidate cassette exons predicted as Rbfox2-regulated | — |
| Computational cluster calling and enrichment analysis | FHFOX2 vs UntagFOX2 iCLIP cross-link data (mESC transcriptome) | none | significant binding clusters, genomic/intronic region fold enrichment | CLIPper algorithm (https://github.com/YeoLab/clipper); χ2 and binomial tests |
| Computational motif analysis | Top 5000 FHFOX2 iCLIP clusters (mESC) | none | enriched hexamers, position-specific probability matrices, motif co-occurrence and cross-link coverage around motifs | MEME (Bailey et al. 2009) |
| Computational splice site strength and conservation analysis | Introns/cassette exons of the mESC transcriptome, bound vs expression-matched unbound loci | none | splice site strength scores and mean per-nucleotide conservation across placental mammals | MaxEnt algorithm; phastCons; Wilcoxon test |
| Computational alternative splicing quantification and gene ontology analysis | mESC RNA-seq data (control vs Rbfox2 knockdown); bound/regulated vs bound/unregulated cassette exon sets | Rbfox2 shRNA knockdown (in silico comparison) | Δψ with Bayes' factors; enriched GO terms over background of all mESC-expressed cassette exons | MISO (Katz et al. 2010); DAVID (Huang et al. 2009) |
- ▲ iCLIP yielded 5.64 million tagged (FHFOX2) and 1.21 million untagged cross-link sites; 40,243 FHFOX2 clusters versus only 2961 UntagFOX2 clusters, giving 35,639 high-confidence clusters 5.64 million vs 1.21 million cross-links; 40,243 vs 2961 clusters
- ▲ FHFOX2 cross-links were strand-specific and enriched in introns over UntagFOX2, and dramatically enriched in introns flanking cassette and mutually exclusive exons while slightly but significantly depleted in constitutively spliced introns ~80% sense strand, 3% antisense
- ▲ Most clusters contain the canonical motif: 51% of intronic clusters contained a GCAUG, 6.1% a conserved GCAUG, and 71% contained a match to the UGCAUG position-specific probability matrix; UGCAUG was the top-scoring hexamer 51%, 6.1%, 71%
- ▲ Cross-links map at nucleotide resolution onto conserved UGCAUG motifs, peaking at motif positions 2 and 6; 16% of conserved intronic UGCAUG matches were within 100 nt of a cluster 28% of UGCAUG within 5 nt of a cross-link; 53% within 300 nt; ~5-fold drop in signal beyond positions 1 and 7
- – Secondary motifs resembling hnRNPL ([CAG]C[AU]CAC) and CELF/Cugbp1 (UGUGUG) consensus occur near Rbfox2 sites but mostly co-occur with UGCAUG; clusters lacking the UGCAUG-related motif yielded no significant high-information motif CCUCAC in 6.8% of clusters, UGUGUG in 18%; 39% and 72% respectively co-occurred with UGCAUG within 10 nt of cluster summits
- – Rbfox2 knockdown produced hundreds of splicing changes, with concordant high-confidence changes in all annotated splicing categories; 18 of 21 tested cassette exons validated by RT-PCR 4%-13% of expressed events concordant; 18/21 validated
- ▲ Cassette exons and their flanking splice site regions bound by FHFOX2 clusters were significantly more conserved across placental mammals than expression-matched unbound loci, and bound intronic UGCAUG motifs were more conserved than unbound motifs
- ▼ Introns with an iCLIP cluster within 200 nt of a splice site had weaker average splice site strength than unbound introns, an effect significant for constitutive introns; weak 5' splice sites were more strongly bound just downstream
- count 5.64 million tagged and 1.21 million untagged cross-link sites (FHFOX2 vs UntagFOX2 iCLIP libraries in mESCs)
- count 40,243 FHFOX2 clusters and 2961 UntagFOX2 clusters (gene-level FDR ≤ 0.05; P ≤ 0.05); 35,639 high-confidence clusters after binomial enrichment (Q-value threshold 0.05) (CLIPper cluster calling and enrichment filtering)
- count >200 Rbfox2-bound AS-NMD splicing events; nearly 70 within RNA-binding protein genes (iCLIP plus mRNA-seq in mESCs (abstract))
- other 51% of intronic clusters contain GCAUG; 6.1% contain a conserved GCAUG (phastCons placental mammals >0.4); 71% contain a UGCAUG PSPM match (FDR < 0.1; P < 0.001) (motif content of iCLIP clusters; motif analysis on top 5000 clusters by cross-link frequency)
- other 28% of conserved UGCAUG occurrences within 5 nt of a cross-link site; 53% within 300 nt; 16% of conserved exact UGCAUG matches in expressed introns within 100 nt of a cluster (distance of cross-link sites to conserved intronic UGCAUG motifs)
- other CCUCAC motif in 6.8% of clusters (FDR < 0.3; P < 0.0002); UGUGUG in 18% of clusters (FDR < 0.1; P < 0.0003); 39% of CCUCAC and 72% of UGUGUG matches co-occur with UGCAUG within 10 nt of cluster summits (secondary motifs in FHFOX2 iCLIP clusters)
- count 18 of 21 cassette exons validated by RT-PCR (validation of MISO-predicted Rbfox2-regulated cassette exons after knockdown)
- other 4% to 13% of expressed events showed concordant splicing changes (Δψ ≥ |5%|, Bayes' factor > 5) (Rbfox2 knockdown RNA-seq, two shRNA comparisons in biological duplicate)
Statistical methods review
Model: sonnetA neutral, descriptive read of the statistical approach — what was done, and (for shared learning, not as criticism) what could also have been done.
The paper combines iCLIP (cross-linking immunoprecipitation with high-throughput sequencing) mapping of Rbfox2 binding sites with RNA-seq of Rbfox2 knockdown mouse embryonic stem cells to identify binding-associated splicing and expression changes. Enrichment and distributional comparisons (e.g., binding site location, splice-site strength, sequence conservation) are assessed with chi-squared and Wilcoxon rank-sum tests, cluster-calling significance is assessed with FDR/Q-value and binomial thresholds via the CLIPper algorithm, and splicing changes upon knockdown are called using the MISO Bayesian framework (percent-spliced-in, Δψ, with a Bayes factor cutoff). Results are reported primarily as fold enrichments, box plots, average signal/conservation profiles, and P-value or FDR thresholds rather than as point estimates with confidence intervals.
| Test | Applied to | n | Assumptions |
|---|---|---|---|
| Chi-squared (χ2) test | Fold enrichment of FHFOX2 over UntagFOX2 cross-links in genomic regions and intronic categories (Fig. 1A,B); intersection of Rbfox2-regulated exons with iCLIP-bound exons (Fig. 3D) | — | not stated |
| Wilcoxon (rank-sum) test | Splice-site strength (MaxEnt score) comparisons between bound and unbound introns (Fig. 1C); conservation score comparisons around bound vs. unbound cassette exons and UGCAUG motifs (Fig. 1D, Fig. 2C) | — | not stated |
| Binomial test | Confirming FHFOX2 cluster enrichment over UntagFOX2 background within called cluster bounds | 35,639 high-confidence clusters derived from 40,243 FHFOX2 and 2,961 UntagFOX2 CLIPper clusters | not stated |
| MISO Bayesian splicing analysis (Bayes factor) | Calling Rbfox2-dependent alternative splicing changes upon knockdown (Δψ ≥ |5%|, Bayes factor ≥ 5, required concordant in two comparisons) | biological duplicate RNA-seq: shLuc/shGFP controls vs. shFox2-1/shFox2-2 knockdowns | not stated |
| FDR/Q-value thresholding | Gene-level CLIPper cluster calling (FDR ≤ 0.05, P ≤ 0.05) and hexamer motif enrichment (e.g., UGCAUG: FDR < 0.1, P < 0.001; CCUCAC: FDR < 0.3, P < 0.0002; UGUGUG: FDR < 0.1, P < 0.0003) | top 5000 iCLIP clusters ranked by cross-link frequency (motif analysis) | not stated |
-
Chi-squared tests were used across several enrichment comparisons (e.g., Fig. 1A,B; Fig. 3D) without a stated joint multiplicity correction across all such tests.↳ Could also: Reporting FDR- or Bonferroni-adjusted q-values across the full set of chi-squared comparisons — This would formally control the false discovery or family-wise error rate when several enrichment tests are examined together, complementing the per-test P-values already shown.
-
Wilcoxon rank-sum tests were used to compare splice-site strength and conservation distributions between bound and unbound groups.↳ Could also: Reporting a non-parametric effect size (e.g., rank-biserial correlation or Cliff's delta) alongside the P-value — An effect size would convey the magnitude of the shift between distributions in addition to statistical significance, which is useful when sample sizes are large enough that small differences can reach significance.
-
Rbfox2-dependent splicing changes were called using a Δψ cutoff combined with a MISO Bayes factor threshold.↳ Could also: A frequentist FDR-controlled test on isoform ratios (e.g., a beta-binomial or DEXSeq-style model) — This would express splicing-change confidence in a P-value/FDR framework, which can facilitate direct comparison with other differential-splicing tools that report FDR rather than Bayes factors.
-
RNA-seq comparisons used two biological replicates per condition (control and knockdown hairpins).↳ Could also: Increasing replicate number (e.g., three or more per condition) with replicate-aware dispersion modeling (e.g., DESeq2/edgeR-style shrinkage estimators) — Additional replicates and dispersion modeling can improve precision in estimating variability, particularly for transcripts with lower expression or fold-change estimates near the detection threshold.
-
Concordance between the two independent shRNA comparisons (shLuc vs. shFox2-1; shGFP vs. shFox2-2) was assessed qualitatively by requiring the same direction of change.↳ Could also: A joint statistical model treating hairpin as a covariate (e.g., a generalized linear model combining both knockdown/control pairs) — This would yield a single formal test statistic and p-value for concordant regulation, in addition to the directional concordance criterion already used.
-
Distributional comparisons (e.g., splice-site strength, conservation) are summarized with box plots.↳ Could also: Reporting summary statistics such as median with a 95% confidence interval, or violin plots showing the full distribution — This can convey distribution shape and estimation uncertainty beyond the quartile/whisker summary a box plot provides.
What was reproduced
The exact results taken into scope, with each reported value next to the value our attempt produced.
Assessments & scoring basis
Each contributor’s verdict, the per-question basis, and the auditable, itemised worksheet behind it.
An automated assessment. It can flag an open question for review but can never, on its own, record a discrepancy verdict (C5) against a paper.
Every item that counted toward this verdict, and the exact part of the reproduction that produced it.
Input data is a clean 1:1 match (all 10 GSE54794 runs, complete, exact n-concordance) and two claims reproduce convincingly on our side: Rbfox2 knockdown (edgeR logFC=-1.616, FDR=1.39e-05) and UGCAUG motif specificity (24.0% of FHFOX2 peaks vs 3.8% of UntagFOX2 peaks, ~11.8x vs ~2.8x over flanking background). The largest deviations are on our side: iCLIP cluster counts are 3-6x the reported numbers (18,339 vs ~2,961; 130,573 vs ~40,243), traceable to position+strand dedup substituting for the paper's randomer/UMI dedup, and the splicing analysis used rMATS instead of MISO so the reported RT-PCR-validated exons (only 3/6 significant: Tia1, Srsf7, Tra2a) were never tested in kind. The paper's central autoregulation claim did not reproduce (no significant Rbfox2 self-splicing event, all FDR>0.4), but because the event-definition tool differs and the authors' filtering criteria are underspecified, this is a limited rather than a refuted conclusion. No fabrication signal, no 'too perfect' pattern -- a solid reproduction with fully disclosed, explainable methodological deviations.
Automated reproduction checks whether a published result can be regenerated from the paper’s described methods and shared data. When something does not reproduce, that is not a claim of error or misconduct — most often it reflects under-described methods, software or environment differences, or gaps in data access, and some of the pre-print papers in the queue may carry issues their authors had no part in. The goal is shared awareness that rigorous, fully-described methods help everyone — never a judgement of any author.
Are you an author? We would genuinely like to hear from you — to clarify the record, add data or code, re-run the pipeline after an accession update, and publish your response right next to the assessment. Everything here is open and auditable.
🚩 Report an error in this record
Spotted something wrong — a verdict you’d contest, a data or value error, or a private detail that slipped through? Tell us, with a short justification. Authors and readers are equally welcome to write in; we review every report.
Prefer email, or the form below not working? Contact us at support@doesitreproduce.com.