Multi-omic identification of perineurial hyperplasia and lipid-associated nerve macrophages in human polyneuropathies.
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.
- ✓Same input data as the authors
- ✓Reported values were directly comparable
- ✓No relevant deviation in data/preprocessing
- ✓No authors-side cause for any deviation
- ✓Reported values are derivable from the shared data
- ✓Any deviation was negligible
- ✓The central claim held under reproduction
- ✓Overall, the reproduction was clean
- Every checked point held up.
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 -> clean 1:1 reproduction. The paper (Heming et al, Nat Commun 2025, human PNS atlas) ships its own reproducibility path: a self-contained Quarto doc (docs/index.qmd) that auto-downloads the exact figure-source objects from Zenodo 15108216 and regenerates every figure, plus an renv.lock and a Docker image. We ran the authors' figure path on «our HPC» (SLURM 2176066, conda R env, ~2.5 min): downloaded the shipped .qs objects to «infra» and read off the summary statistics the paper reports as text. ALL FOUR headline text numbers matched EXACTLY: 365,708 total nuclei (C1), 37 donors = 33 PNP + 4 CTRL across 3 centers (C2), 24 main clusters (C3), and 35 immune subclusters / 18,436 immune nuclei / 18 macrophage clusters Macro1-18 / 3 perineurial clusters periC1-3 (C4). The shipped figure-source dataframes (Fig 3C pseudobulk DEG counts, Fig 3A propeller log2-ratios, Fig 2C Macro18 GO enrichment) reproduce their panels by construction and are direction/magnitude-consistent with the narrative (myelinating SC reduced and repair/damage SC expanded in PNP; Macro18 enriched for fatty-acid/cholesterol GO terms = 'lipid-associated'). NO FABRICATION SIGNAL: every checked headline number is exactly derivable from the publicly deposited data. NOT ATTEMPTED (out of 80/20 scope): re-running the upstream raw-FASTQ pipeline (Cell Ranger 7.0.1 -> CellBender 0.3.0 -> scVI 1.0.4 integration -> clustering), which is stochastic and not bit-reproducible; Xenium spatial pipeline; g-ratio/axon/EMA wet-lab quantifications; miloDE/LIANA/CNA objects (plottable but not reduced to a single text number). Note: the BRIEF's data accession (GSE182098) is the Yim et al. RODENT comparison set, not this study's own data (own data = GSE285983 snRNA-seq + GSE285984 Xenium; figure source = Zenodo 15108216).
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 90assessed: 2026-06-14 ⛓ be12e75766e8
✎ 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: opusHuman polyneuropathies (PNPs) of diverse etiologies lack mechanistic and cellular characterization; the authors test whether multi-omic single-nucleus and spatial transcriptomics of human sural nerves can define nerve cell types and identify shared and disease-specific cellular/molecular mechanisms across PNPs.
- ★ A single-nucleus transcriptomic atlas of 37 human sural nerves (365,708 nuclei) defines 24 cell clusters and reveals unexpected heterogeneity of perineurial cells (periC1-3). resource
- ★ PNPs share a loss of myelinating Schwann cells with an increase in repair/damage Schwann cells and endoneurial lipid-phagocytizing macrophages. finding
- ★ CXCL14 is a novel perineurial cell marker whose increased expression accompanies focal perineurial hyperplasia, particularly in immune-mediated PNPs. finding
- ★ PNPs affect multiple cell types beyond the endoneurium, supporting a concept of PNPs as 'pan-nerve diseases'. finding
- ★ The Macro18 cluster represents lipid-associated macrophages (LAMs) with a myelin-phagocytizing, lipid-sensing phenotype shared with stroke-associated myeloid cells. mechanism
- ★ Novel cell-type markers were identified and multi-modally confirmed: MLIP (mySC), GRIK3 and PRIMA1 (nmSC), and CXCL14 (periC). finding
- A venous/capillary endothelial cluster (ven_capEC2) expressing ABCB1, SLC1A1, MFSD2A and GJA1 represents blood-nerve barrier cells. finding
- Perineurial cells (periC2) are predicted to interact with B cells via CXCL14-CXCR4 signaling. mechanism
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| single-nucleus RNA-sequencing (snRNA-seq) | human sural nerve biopsies (33 PNP patients, 4 controls; 37 nerves) | none (disease vs control comparison) | transcriptome / nuclei cluster composition (365,708 nuclei) | — |
| subcellular spatial transcriptomics (spatial-seq, in situ amplification-based) | human sural nerve FFPE cross-sections (8 patients) | none | spatial expression of 99-gene custom marker panel; transcript localization | custom 99-marker panel |
| immunofluorescence (IF) | human sural nerve sections (4 patients, min 2 sections each) | none | CXCL14 protein localization in perineurium | — |
| RNA in situ hybridization combined with immunofluorescence (RNA-ISH/IF) | rodent (murine) peripheral nerve and positive control tissues | none | Grik3, Prima1, Cxcl14 expression and colocalization with cell markers | — |
| histology / immunohistochemistry (g-ratio, axon quantification, perineurium thickness) | human sural nerve sections (37 tissues; perineurium in 12 CIDP and 18 CTRL) | none | myelin thickness (g-ratio), number of intactly myelinated axons, axon diameters, perineurium thickness (EMA staining) | IHC anti-epithelial membrane antigen |
| histological lipid/myelin co-staining (BODIPY + myelin + marker IF) | human sural nerve sections from PNP patients (3 patients, min 2 sections each) | none | colocalization of Macro18 markers (SPP1, PLIN2, CD36, FABP5) with myelin and neutral lipids | BODIPY neutral lipid label, DAPI |
- ▼ Loss of myelinating Schwann cells with emergence of damageSC and repairSC in PNP
- ▲ Increase of Macro18/endoneurial lipid-associated macrophage subcluster in PNP patients
- ▲ Perineurial cluster periC3 increased and positively correlated with PNP disease status
- – mySC nuclei negatively correlate with PNP status, INCAT score, and g-ratio; leukocyte and repairSC clusters positively correlate
- – Stroke-associated myeloid cells (SAMC) projected predominantly onto the Macro18 cluster
- – CXCL14 colocalized with known perineurial markers CLDN1, SLC2A1, KRT19 in the perineurium
- ▲ BNB markers ABCB1 and SLC1A1 enriched in endoneurial vessels confirming ven_capEC2 as BNB cells
- – Macro18 markers (SPP1, PLIN2, CD36, FABP5) colocalized with myelin and BODIPY-labeled neutral lipids
- count 365,708 high-quality nuclei (total nuclei from 37 human sural nerves after QC)
- count 24 clusters (main cell clusters identified in snRNA-seq)
- count 18,436 nuclei (immune cell nuclei subclustered into 35 IC subclusters)
- count 79% of all IC nuclei (proportion of immune cell clusters of myeloid lineage)
- count 99 transcripts (custom spatial-seq marker panel size)
- fold_change log2 fold change > 2, adjusted p-value < 0.001 (threshold for Macro18/LAM marker genes in GO enrichment (one-sided Fisher's exact, Benjamini-Hochberg))
- count 33 PNP patients and 4 controls (cohort composition)
- count 12 CIDP and 18 CTRL (patients with perineurium thickness measured by EMA IHC)
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 performed single-nucleus RNA-sequencing of 37 human sural nerve biopsies (33 PNP patients, 4 controls; 365,708 nuclei after QC and batch correction) integrated with subcellular spatial transcriptomics on 8 samples. Cell clustering, differential cluster abundance (reported as log2 fold change), and pseudobulk differential gene expression were used to characterize disease-associated compositional and transcriptional changes. Clinical associations between nucleus cluster membership and PNP status, INCAT disability score, and g-ratio were assessed with confounder correction. Gene ontology enrichment was evaluated via one-sided Fisher's exact test with Benjamini-Hochberg FDR correction using enrichR.
| Test | Applied to | n | Assumptions |
|---|---|---|---|
| one-sided Fisher's exact test with Benjamini-Hochberg FDR correction (via enrichR) | Gene Ontology term enrichment analysis of Macro18/LAM cluster marker genes (Fig. 2D) | Input gene set: markers with log2 fold change > 2 and adjusted p-value < 0.001 in one-vs.-all immune cluster comparison | not stated |
| pseudobulk differential expression (specific test not stated in available text) | Number of differentially expressed genes per cell cluster, PNP vs. CTRL (Fig. 3C) | null | not stated |
| correlation analysis with confounder correction (specific method not stated in available text) | Association of individual nucleus cluster membership with PNP status, INCAT disability score, and g-ratio (Fig. 3B) | null | not stated |
| marker gene identification test (specific test not stated; log2FC > 2, adjusted p-value < 0.001 used as thresholds) | One-vs.-all immune cell cluster comparison to define Macro18/LAM markers input to GO enrichment | 18,436 immune cell nuclei | not stated |
-
GO term enrichment was computed using a one-sided Fisher's exact test on a binary gene set defined by fold-change and p-value thresholds↳ Could also: A rank-based gene set enrichment method such as fgsea or GSEA applied to all genes ranked by fold change or test statistic could also be used — Rank-based methods use the full distribution of gene-level statistics rather than a binary inclusion threshold, which can increase sensitivity for pathways with moderate but consistent signals and avoids the need to choose arbitrary cutoffs for the input gene list
-
Cell-type cluster abundance differences between PNP and controls were reported as log2 fold change with a display cutoff of > 0.5, without formal statistical testing stated in the available text↳ Could also: Dedicated differential abundance testing tools such as Milo (negative binomial regression on k-nearest-neighbor neighborhoods) or propeller could also be applied — These methods provide formal p-values, FDR estimates, and effect-size confidence intervals for abundance changes, enabling inference about which shifts exceed sampling variability and accounting for donor-level variability in cell-type proportions
-
The specific pseudobulk differential expression method used for per-cluster DEG analysis is not stated in the available text↳ Could also: Pseudobulk approaches implemented in DESeq2 (Wald test), edgeR (quasi-likelihood F-test), or limma-voom are each widely used and make distinct distributional assumptions — Explicitly reporting the pseudobulk aggregation strategy (sum vs. mean), the count model, and any offset or normalization aids reproducibility and allows readers to assess which assumptions were applied to donor-level aggregate counts
-
Clinical associations (PNP status, INCAT score, g-ratio) were assessed by correlating individual nucleus cluster membership with confounder correction↳ Could also: A donor-level linear or mixed model using per-donor cell-type proportions as the outcome (e.g., linear model with center as covariate, or lme4-based mixed model with donor as random effect) could also be used — Nucleus-level analyses can inflate effective sample size through pseudo-replication across thousands of nuclei from the same donor; donor-level models make the patient the unit of inference, naturally handle within-donor correlation, and more directly map to the clinical question
-
The control group comprised four samples repurposed from surgical sural nerve autografts in traumatic nerve injury patients↳ Could also: Sensitivity analyses comparing PNP subgroups internally against each other, or supplementary analyses using a larger or age/sex-matched control cohort, could also serve as complementary reference comparisons — With n = 4 controls from a specific clinical subpopulation, the reference distribution for 'healthy nerve' rests on a small sample; internal cross-etiology comparisons or additional controls can provide complementary anchor points and support generalizability of findings
-
Batch correction was applied at the embedding/clustering stage after pooling nuclei from three collection centers↳ Could also: A hierarchical or mixed-effects pseudobulk model (e.g., dreamlet) that retains donor and center as explicit random effects at the differential expression stage could also be used — Retaining batch structure as explicit random effects in the inferential model allows uncertainty from center and donor variability to propagate into test statistics and confidence intervals, rather than treating the corrected representation as if batch variation had been fully removed
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-40849297
Paper: Heming et al., "Multi-omic identification of perineurial hyperplasia and lipid-associated nerve macrophages in human polyneuropathies." Nat Commun 2025. DOI 10.1038/s41467-025-62964-8 · PMID 40849297 · PMCID PMC12375038.
Code: https://github.com/mihem/pns_atlas (authors' own code; snRNA-seq + Xenium).
Archived: Zenodo 10.5281/zenodo.15750105 (code v1.0.0). renv.lock pins R 4.3.1,
Seurat 5.0.1.9004, qs 0.25.7. Docker image mihem/pns_atlas:v3.0.
Data accessions (corrected vs BRIEF):
- Own snRNA-seq raw/processed: GEO GSE285983
- Own Xenium spatial: GEO GSE285984
- Figure-source processed objects: Zenodo 15108216 (
.qsfiles; auto-downloaded by the authors'docs/index.qmdQuarto reproducibility document). - GSE182098 (named in the BRIEF) is not this study's data — it is the Yim et al. rodent comparison dataset. GSE142541 / GSE189432 are other external comparison sets.
In scope (pipeline-derived, attempted)
The paper's main results derive from a snRNA-seq pipeline: Cell Ranger 7.0.1 → CellBender 0.3.0 → Seurat 5 (+BPCells) → scDblFinder → scVI 1.0.4 integration → Leiden clustering → annotation; then propeller (abundance), Libra/edgeR (pseudobulk DE), miloDE (neighborhood DE), enrichR (GO), LIANA (cell-cell), rcna (CNA).
The authors ship the exact figure-source objects on Zenodo and the figure code in
docs/index.qmd. The clearly-specified, low-compute, checkable summary statistics
that the paper reports as text/figure numbers are reproduced by loading those shipped
objects and reading off their metadata (no re-clustering required):
| id | reported claim | object (Zenodo 15108216) | how |
|---|---|---|---|
| C1 | 365,708 total high-quality nuclei | umap_figure.qs | ncol() |
| C2 | 37 samples (33 PNP + 4 CTRL), 3 centers | umap_figure.qs meta | unique(sample/center) |
| C3 | 24 main clusters | umap_figure.qs | nlevels(Idents) / @misc$cluster_col |
| C4 | 35 immune subclusters; 18,436 IC nuclei; Macro1-18; periC1-3 | ic_figure.qs / umap_figure.qs | counts |
| C5 | Fig 3C: # pseudobulk DEGs per cluster (PNP vs CTRL) | pnp_ctrl_pseudobulk_de.qs | dataframe = figure source |
| C6 | Fig 3A: propeller abundance log2-ratios PNP vs CTRL | propeller_PNP_CTRL(_ic).qs | dataframe = figure source |
| C7 | Fig 2C: Macro18/LAM GO enrichment (lipid-associated) | enrichr_macro18.qs | top GO terms |
Out of scope (NOT attempted, by 80/20)
- Upstream pipeline from raw FASTQ (Cell Ranger → CellBender → scVI integration → clustering). This is the hard, expensive, non-deterministic ~80% (scVI is stochastic; exact cluster boundaries are not bit-reproducible). We instead verify that the shipped processed objects — which ARE the published figure source — are internally consistent with the reported text numbers. This is the fabrication-relevant check: are the headline numbers actually derivable from the deposited data?
- Xenium spatial pipeline (GSE285984), g-ratio/axon morphometry, EMA histology (wet-lab/manual), miloDE neighborhood DE object, LIANA, CNA feature plots — plottable from shipped objects but not reduced to a single checkable text number here.
- Re-deriving DE/propeller statistics from raw counts (the shipped dataframes are themselves the figure source, so plotting them is exact by construction; we record them as figure-level reproduction, distinct from the text-number checks C1-C4).
Reproduction venue
All compute on «infra» «our HPC»-2 SLURM via «host». conda prefix env on «infra»
(r-base 4.3.3, r-qs 0.27.3, r-seurat 5.3.0). «job». Objects downloaded from
Zenodo inside the compute job to «infra». ~2.5 min wall.
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.
Clean 1:1 reproduction. All four headline text claims (365,708 nuclei; 37 donors = 33 PNP + 4 CTRL across 3 centers; 24 main clusters; 35 immune subclusters / 18,436 immune nuclei / Macro1-18 / periC1-3) matched exactly when read off the authors' publicly deposited figure-source objects (Zenodo 15108216), and the figure-source dataframes (Fig 3C DEGs, Fig 3A propeller, Fig 2C Macro18 GO) reproduce their panels by construction and are direction/magnitude-consistent with the lipid-associated-macrophage and SC-shift narrative. No fabrication signal — every checked value is exactly derivable from the public data. The only limitation is on data-availability/scope, not on the authors: the upstream stochastic raw-FASTQ→object pipeline was deliberately not re-run, so this confirms deposited-object↔text consistency rather than full end-to-end reproducibility.
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.