SPRR1B+ keratinocytes prime oral mucosa for rapid wound healing via STAT3 activation.
The main results reproduced: recomputed values matched the published ones within tolerance.
Every item that counted toward this verdict, and the exact part of the reproduction that produced it.
- Nothing in this column.
- 🟡Could not use the authors’ exact input data
- 🟡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
Described well enough to PARTIALLY reproduce the bulk-RNA-seq branch of Fig.1 using the listed data (GSE97615) and listed tool (clusterProfiler). 1:1 on the GO themes (C3) and ~1:1 on the DEG count (C1: 479 vs 485, -1.2%). The P2=146 sub-module (C2) did NOT reproduce: the paper's 'unsupervised clustering' of the 485 DEGs into P1/P2 is under-specified (no algorithm/distance/k), and a reasonable ward.D2 k=2 split gives 307/172. Important honesty flag: the paper attributes the 485 DEGs to an injured-vs-uninjured contrast, but GSE97615 contains ONLY post-wound samples (no uninjured baseline); the only computable contrast (oral vs skin) reproduces 479 and matches the paper's own description of P2 as constitutively expressed in oral but absent in skin — likely imprecise methods wording rather than a fabricated number, flagged for human review. NOT attempted (out of scope): all single-cell analyses (GSE164241 + PRJCA006797, Seurat/Harmony, 9 cell types, KC1-KC5), SCENIC TF regulons (STAT3/KLF5/PRDM1/GRHL1), STRING/Cytoscape/CytoHubba PPI hubs, Monocle2 pseudotime, and ChIP-seq (SRP070705, bowtie2/MACS2) — these use different data and different tools than the listed clusterProfiler, and are heavy.
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.
Assessment versions
Every reproduction run is kept as an immutable version — anchored to the data as it stood, with a tamper-evident chain hash. A rerun (e.g. after an author updates a deposit) adds a new version; the previous one stays on record.
-
v1 current initial assessment Score 78assessed: 2026-06-14 ⛓ 0eb58393a3cb
✎ 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
- 2026-06-14
- Rubric version
- v1.0
- Assessed by
-
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-06-15no human curator yet
- Last updated
- 2026-08-05
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: sonnetOral mucosa heals faster and with less scarring than skin, and the paper tests whether a specific keratinocyte subpopulation, constitutively present in unwounded oral mucosa and induced via STAT3 activation, primes oral tissue for rapid wound healing.
- ★ A shared wound healing gene set (P2-WHGs, 146 genes) is constitutively expressed in uninjured oral mucosa but not in uninjured skin finding
- ★ Keratinocytes, and oral keratinocytes in particular, are the major cell type expressing P2-WHGs among all sequenced skin/mucosa cell types finding
- ★ A keratinocyte subcluster, SPRR1B+ KC4, shows the highest P2-WHGs score and is more enriched in normal oral mucosa than in skin finding
- ★ STAT3 transcriptionally regulates SPRR1B by binding its promoter region, and JAK/STAT3 inhibition reduces SPRR1B protein levels mechanism
- ★ SPRR1B knockdown significantly inhibits mucosal keratinocyte migration finding
- ★ SPRR1B+ keratinocytes are induced during wound healing in both skin and oral mucosa in a murine model finding
- Oral KC4 shows increased expression of inflammation (S100A8, S100A9, IL36G) and EMT (VIM, LUM, COL1A1) genes compared to skin KC4 finding
- Oral mucosa contains multiple layers of KRT14+ basal keratinocytes compared to a single basal layer in skin finding
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| bulk mRNA-seq / DEG analysis | human injured/uninjured oral mucosa and skin (GSE97615) | wound (injury) vs uninjured | differentially expressed genes, gene panels P1/P2 | — |
| single-cell RNA-seq | human normal skin (PRJCA006797, 5 samples) and oral mucosa (GSE164241, 21 samples: gingiva/buccal) | none | cell type clustering, P2-WHGs enrichment score per cell/cluster | — |
| gene ontology / GSEA enrichment analysis | P2-WHGs gene list and KC4 marker genes (bioinformatic) | none | enriched biological processes/pathways | — |
| SCENIC transcription factor regulon analysis | oral KC4 keratinocytes (scRNA-seq derived) | none | TF regulon activity scores | SCENIC |
| protein-protein interaction network analysis | TFs identified in KC4 (bioinformatic) | none | hub TF identification (STAT3) | STRING; CytoHubba MCC algorithm |
| ChIP-seq re-analysis / ChIP-PCR | epithelial cells / HOK (human oral keratinocytes) | none | STAT3 binding to SPRR1B promoter (0-500 bp upstream of TSS) | — |
| Western blot | HOK cells | JAK1/2 inhibitor ruxolitinib | STAT3, p-STAT3, JAK1, p-JAK1, SPRR1B protein levels | — |
| immunofluorescence staining | human normal skin and oral mucosa biopsies | none | KRT14, KRT10, SPRR1B protein localization/expression | — |
- ▲ 485 DEGs identified as shared wound healing-associated gene sets between injured/uninjured skin and oral mucosa adjusted P<0.05, log2FC>1
- ▲ 146 P2-WHGs genes constitutively expressed in uninjured oral mucosa but not skin
- ▲ Keratinocytes show significantly higher P2 gene set scores than other cell types; oral keratinocytes score higher than skin keratinocytes despite lower abundance p<2.2e-16
- ▲ SPRR1B+ KC4 shows the highest P2-WHGs score among 5 keratinocyte subclusters, higher in oral than skin KC4 p<2.2e-16
- – STAT3 identified as hub transcription factor in KC4 PPI network, connecting 6 nodes 6 nodes
- – ChIP-PCR confirms STAT3 binds directly to SPRR1B promoter region 0-500 bp upstream of TSS
- ▼ Ruxolitinib (JAK1/2 inhibitor) decreases p-STAT3 levels and reduces SPRR1B protein in HOK cells
- ▼ SPRR1B knockdown inhibits mucosal keratinocyte migration
- count 485 (differentially expressed genes defining shared wound healing-associated gene sets)
- count 146 (P2 wound healing genes (P2-WHGs) constitutively expressed in uninjured oral mucosa)
- fold_change log2 fold change > 1 (DEG threshold for injured vs uninjured comparison)
- pvalue adjusted P < 0.05 (DEG and GO enrichment significance threshold)
- count 103,758 (cells retained after QC for scRNA-seq clustering of skin and oral mucosa)
- pvalue p < 2.2e-16 (Wilcoxon rank-sum test, P2-WHGs score comparison between oral and skin keratinocytes)
- count 9 (distinct cell types identified by scRNA-seq clustering)
- count 5 (keratinocyte subtypes (KC1-KC5) identified by subclustering)
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 study integrated publicly available bulk RNA-seq (GSE97615) and single-cell RNA-seq data (103,758 cells from 5 skin and 21 oral mucosal samples) to characterize wound-healing gene programs across tissue types and keratinocyte subtypes. Bulk DEGs were called with fold-change and adjusted-P thresholds; single-cell group differences were assessed with two-sided Wilcoxon rank-sum tests; gene module enrichment was visualized on UMAP and tested with GSEA. Transcription-factor regulon activity was estimated computationally (SCENIC), and key findings were validated by ChIP-PCR and western blot in human oral keratinocyte (HOK) cells.
| Test | Applied to | n | Assumptions |
|---|---|---|---|
| Differential expression analysis, bulk RNA-seq (specific test not named; 'adjusted P' implies a count-model-based test) | Injured vs uninjured oral mucosa and skin (GSE97615) to define 485 shared wound-healing DEGs (P1 and P2 gene panels) | Not stated; derived from public dataset GSE97615 | not stated |
| Two-sided Wilcoxon rank-sum test | P2-WHG enrichment score comparison across all cell types (oral mucosa vs skin) — Fig. 2e | 103,758 cells from 5 skin and 21 oral mucosal samples | not stated |
| Two-sided Wilcoxon rank-sum test | P2-WHG enrichment scores across KC1–KC5 subclusters (oral vs skin) — Fig. 3d | Keratinocyte subset of 103,758 cells; per-subcluster n not stated | not stated |
| Gene Set Enrichment Analysis (GSEA) | P2-WHG signaling in oral KC4 vs skin KC4 — Fig. 3e | KC4 cells from 5 skin and 21 oral mucosal samples; exact cell count not stated | not stated |
| Two-sided Wilcoxon rank-sum test | Individual gene expression (S100A8, S100A9, IL36G, VIM, LUM, COL1A1) in oral vs skin KC4 — Fig. 4h | KC4 subset; exact n not stated | not stated |
| Two-sided Wilcoxon rank-sum test on gene-signature module scores | EMT, wound healing, and inflammatory response scores in oral vs skin KC4 — Fig. 4i | KC4 subset; exact n not stated | not stated |
| Gene Ontology (GO) overrepresentation enrichment analysis | P2-WHG gene list (Fig. 1c); cell-type marker genes (Fig. 2c); KC4 marker genes (Fig. 4g); all with adjusted P < 0.05 | Gene lists derived from scRNA-seq and bulk RNA-seq; correction algorithm not named | not stated |
-
Cell-level gene module scores were compared with Wilcoxon rank-sum tests, treating each cell as an independent observation across tissue types↳ Could also: A pseudo-bulk approach — aggregating expression or scores per donor, then applying a donor-level t-test or linear model — would also account for within-donor correlation among cells — Cells from the same donor share biological and technical co-variation; pseudo-bulk methods propagate donor-level variance into the test statistic, reducing inflation of effective sample size that can arise from treating thousands of cells as independent units
-
Multiple pairwise Wilcoxon tests were applied across KC1–KC5 subclusters and across tissue types without a stated multiplicity correction↳ Could also: A Kruskal-Wallis omnibus test followed by Dunn post-hoc correction, or Benjamini-Hochberg FDR applied across the family of pairwise Wilcoxon comparisons, would also control the family-wise error rate across the full set of subcluster comparisons — Applying a global test or explicit FDR correction across the full comparison family conveys control over false discovery rate when multiple subclusters are being contrasted simultaneously
-
Wilcoxon test results were reported only as p < 2.2e-16 without exact p values or effect size measures↳ Could also: Reporting the rank-biserial correlation alongside the p value would also quantify the magnitude of the group difference for each Wilcoxon test — With tens of thousands of cells, Wilcoxon tests approach their numerical precision floor (p ≈ 2.2e-16) regardless of practical effect size; an explicit effect size metric conveys the biological magnitude of the difference independently of sample size
-
Bulk RNA-seq differential expression was called with an adjusted-P threshold without naming the underlying statistical model or correction algorithm↳ Could also: DESeq2 (negative-binomial Wald test with Benjamini-Hochberg FDR) or edgeR (quasi-likelihood F-test) are standard, named methods for count-based bulk RNA-seq DEG calling — Naming the specific model and correction method improves reproducibility and allows readers to assess whether distributional assumptions (e.g., negative-binomial dispersion estimation) are appropriate for the data
-
GO enrichment was performed as overrepresentation analysis (ORA) on a threshold-defined gene list (|log2FC| > 1, adjusted P < 0.05)↳ Could also: Ranked GSEA on the full fold-change-ranked gene list would also test pathway enrichment without requiring a binary inclusion threshold — Ranked methods use the continuous fold-change signal across all genes and are less sensitive to the choice of fold-change/FDR cutoff; the paper already applies GSEA in one context (Fig. 3e), so extension to GO pathways would be consistent with the existing analytical repertoire
-
No dispersion measure (SD, SEM, or CI) is reported for any quantitative comparison, including those involving a small number of biological donors (5 skin samples)↳ Could also: Reporting inter-sample SD or 95% CI at the donor level (especially for the n = 5 skin group) would also convey between-donor variability alongside the cell-level statistics — Donor-level dispersion metrics allow readers to assess consistency across individuals and distinguish biological heterogeneity from cell-level technical noise, which is particularly informative when the number of donors is small relative to the number of cells
Citation network
Where this publication sits in the reproducibility-weighted citation graph — what it is built on, and what is built on it. Citation data from OpenAlex.
No assessed neighbours yet — the network grows as more papers are assessed.
Data lineage
The datasets this paper uses (text-mined from the full text via Europe PMC), and which other assessed papers stand on the same data. A shared dataset is a factual link — not a judgement.
What was reproduced
The exact results taken into scope, with each reported value next to the value our attempt produced.
Scope — pmid-39300285
Paper: Xuanyuan et al. 2024, Commun Biol — "SPRR1B+ keratinocytes prime oral mucosa for rapid wound healing via STAT3 activation." PMID 39300285 / PMC11413210 / DOI 10.1038/s42003-024-06864-5.
Listed code: https://github.com/YuLab-SMU/clusterProfiler (a third-party GO/KEGG enrichment tool — P16: applying an existing tool to the paper's data is equally valid). Listed data: GEO GSE97615.
What GSE97615 actually is
Bulk RNA-seq (Illumina HiSeq 2000), Homo sapiens, 24 samples = 12 skin (S1–S12)
- 12 oral mucosa (O1–O12), each tissue with biopsies at day 1 (n=4), day 3 (n=4),
day 6 (n=4) of wound healing. This is the Iglesias-Bartolome et al. wound-healing
dataset. Processed values shipped as
GSE97615_..._RPKM.xlsx. Note: GSE97615 contains only post-wound timepoints — there is NO "uninjured" baseline arm in this accession, so the paper's "uninjured vs injured" wording cannot be the contrast within GSE97615. The only well-defined within-dataset contrast that yields genes "constitutively expressed in oral mucosa" (their P2 panel) is oral mucosa (12) vs skin (12). We reproduce that contrast.
IN SCOPE (pipeline-derived, low-hanging, uses listed code+data)
| # | Reported result | Paper loc | Pipeline |
|---|---|---|---|
| C1 | 485 DEGs between tissues, limma, |log2FC|>1 & adj.p<0.05 | Results/Methods (bulk RNA-seq) | limma on GSE97615 RPKM |
| C2 | 146 P2-WHGs (oral-high module, "constitutively expressed in oral mucosa") | Results, Fig.1 | direction-split of C1 DEGs |
| C3 | GO enrichment of P2-WHGs, clusterProfiler, P<0.05 & FDR<0.05, top 15 GO terms | Methods + Fig | clusterProfiler::enrichGO (org.Hs.eg.db, BP) |
OUT OF SCOPE (not the listed code; heavy; not attempted)
- All single-cell analysis: GSE164241 + PRJCA006797, 103,758 cells, Seurat + Harmony, 9 cell types, KC1–KC5, AddModuleScore (different data, heavy).
- SCENIC TF regulons (STAT3/KLF5/PRDM1/GRHL1), STRING/Cytoscape PPI + CytoHubba hub genes, Monocle2 pseudotrajectory — separate tools, not clusterProfiler.
- ChIP-seq (SRP070705, bowtie2/MACS2/annotatr) — separate raw-data pipeline.
- All wet-lab (IHC, organoids, scratch assays) — out of scope by definition.
Reproduction strategy
One «our HPC» SLURM R job (conda prefix env: r-base, bioconductor-limma, bioconductor-clusterprofiler, bioconductor-org.hs.eg.db, r-readxl, r-openxlsx). Download RPKM xlsx inside the job (compute node has internet), run limma oral-vs- skin on log2(RPKM+1), count DEGs, split by direction, run clusterProfiler GO on the oral-high set, dump small result JSON/TSV. All data stays on «infra»; only small result values come back to «host».
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.
The bulk-RNA-seq branch of Fig.1 reproduces well: 485 DEGs ≈ 479 (-1.2%) and the GO themes (keratinization q=3e-12, defense response to bacterium q=0.031) match 1:1 from the public GSE97615 data. The deviations are on the authors'/methods side but non-fabrication: the paper mis-describes the contrast as injured-vs-uninjured (no such samples in GSE97615) and under-specifies the clustering, so the reported 146 P2 genes is not derivable (ward.D2 k=2 → 307/172). Severity is moderate — direction and themes hold, exact sub-module does not — and most of the paper's mechanistic claims were out of scope, so the core conclusion is only limitedly confirmed.
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.
Reproduction footprint
claude-opus-4-8Measured resources invested to assess this paper — sanitised (machine class only, no job ids/paths). Compute = HPC accounting (SLURM); tokens = the AI agent's session.