Spatial-reprogramming derived GPNMB+ macrophages interact with COL6A3+ fibroblasts to enhance vascular fibrosis in
The main results reproduced, with only marginal, non-material deviations.
- 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 reproduce the tractable half 1:1. The paper ships NO original code (Data/Code Availability: 'did not generate ... original code'); the only code link, SCP, is a generic third-party single-cell visualization toolkit, so reproduction used brief rule P16: a standard Seurat v4 pipeline (the paper's described preprocessing: HVG2000, 50 PCs, Harmony, FindClusters, FindAllMarkers min.pct=0.2 logfc=0.2) run on the paper's own public dataset GSE131928 (10X GBM) on «our HPC» (SLURM 2180168). RESULTS: dataset shape EXACT (16,201 cells); the macrophage half of the central axis REPRODUCES cleanly -- GPNMB is 8-27x enriched in the macrophage compartment (237 vs <=28 TPM), confirming the 'GPNMB+ MDM' identity; canonical cell-type structure (malignant/macrophage/T/oligodendrocyte) recovered with correct markers. The fibroblast half does NOT reproduce from this single dataset: GSE131928 10X is malignant+immune-dominated and contains no stromal/fibroblast/mural cluster, so COL6A3+ TAF, the SMC markers, and the collagen signature cannot be shown (COL6A3 is near-zero everywhere -- consistent with, but not a positive confirmation of, 'restricted to stroma'). NOT ATTEMPTED (hard ~20%): the novel spatial-reprogramming / vascular-fibrosis mechanism (COL6A3+ TAF -> GPNMB+ MDM, GPNMB-ITGB5), which depends on the neoadjuvant cohort OMIX003593 / HRA007138 -- HRA007138 is NGDC GSA-Human CONTROLLED-ACCESS (data_restricted, DAC application required); the 101-algorithm ML prognostic signature; CIBERSORTx deconvolution (external tool); and all wet-lab validation. No fabrication flagged: every checkable claim is qualitatively consistent with the public data. Verdict: PARTIAL (clean 1:1 on the macrophage/atlas half; stromal half blocked by dataset coverage + access-restricted data).
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 64assessed: 2026-06-15 ⛓ 96e451e6863f
✎ 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
- 2026-06-15
- Rubric version
- v1.0
- Assessed by
-
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-06-16no 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: opusThe study investigates the mechanisms underlying resistance/failure of combined antiangiogenic plus PD-1 inhibitor neoadjuvant therapy in glioblastoma, testing whether COL6A3+ tumor-associated fibroblasts spatially reprogram monocyte-derived macrophages into immunosuppressive GPNMB+ macrophages that promote vascular fibrosis and treatment resistance.
- ★ A distinct subset of COL6A3+ tumor-associated fibroblasts (TAFs) with matrix-fibroblast characteristics exists in GBM and is significantly enriched in non-responders to neoadjuvant combination therapy. finding
- ★ COL6A3+ TAFs drive the spatial-reprogramming of anti-tumorigenic MDMs into a pro-tumorigenic, immunosuppressive GPNMB+ phenotype. mechanism
- ★ Reprogrammed GPNMB+ MDMs promote COL6A3+ TAF-mediated vascular fibrosis through the GPNMB-ITGB5 interaction. mechanism
- ★ COL6A3+ TAFs and GPNMB+ MDMs have prognostic and immunological value across multiple GBM and immunotherapy cohorts, evaluated with 101 machine learning algorithms. finding
- ★ Targeting COL6A3+ TAFs with cilengitide is a potential therapeutic strategy. finding
- ★ A stromal cell atlas in GBM was constructed by integrating single-cell and spatial transcriptomic data. resource
- COL6A3+ TAF-mediated vascular fibrosis leads to T cell exclusion, limiting efficacy of neoadjuvant combination therapy. mechanism
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| single-cell RNA-seq (integrative analysis) | newly diagnosed, recurrent, responder and non-responder GBM and LGG patient samples | neoadjuvant anlotinib + pembrolizumab (responder vs non-responder) | cellular composition, cell subtype proportions, DEGs | — |
| spatial transcriptomics | newly diagnosed GBM and non-responder GBM patient tissue | none/neoadjuvant therapy response | spatial organization of TME cell types | — |
| bulk RNA-seq / microarray with machine learning and deconvolution | glioma and immunotherapy patient cohorts (TCGA, CGGA, Ivy-GAP, GLASS, GSE108474, immunotherapy cohorts) | PD-1/PD-L1 blockade in immunotherapy cohorts | survival/prognosis and immunotherapy efficacy prediction | — |
| multiplex immunohistochemistry (mIHC) | FFPE human GBM tissue | none | GPNMB and ITGB5 protein co-localization | Mantra system (CaseViewer) |
| immunofluorescence / confocal imaging | cultured cells (TAFs, MDMs) | in vitro treatments | ICAM1, GPNMB, COL6A3, PDGFRA protein expression | Nikon-CSU-W1 spinning disk / Nikon Eclipse Ti2 |
| atomic force microscopy | cultured cells on slides | various treatments | Young's modulus / cell elastic behavior (vascular fibrosis) | MFP-3D-BIO AFM (Asylum Research), MLCT-D probe (Bruker) |
| flow cytometry | primary MDMs / THP-1-derived macrophages | polarization / sorting | CD45+CD11b+ICAM1+GPNMB- (ICAM1+ MDM) vs CD45+CD11b+ICAM1-GPNMB+ (GPNMB+ MDM) | — |
| ELISA / chemotaxis (Transwell) / RT-qPCR | COL6A3+ TAFs, primary tumor cells, MDMs/THP-1 | COL6A3+ TAF supernatant, sotuletinib, TGFβ3 antibody, sGPNMB, cilengitide, hypoxia (1% O2) | CSF1/TGFβ3 secretion, MDM migration, mRNA expression | FineTest/Jonlnbio ELISA kits; Bio-Rad real-time PCR |
- ▲ COL6A3+ TAFs are significantly enriched in non-responders to neoadjuvant combination therapy
- – COL6A3+ TAF supernatant induces differentiation of macrophages into GPNMB+ MDM phenotype in vitro
- – GPNMB+ MDMs promote vascular fibrosis via GPNMB-ITGB5 interaction; cilengitide proposed to block
- other TAMs make up 30–50% of all cells in the TME of GBM (abundance of tumor-associated macrophages in GBM)
- other ~85% MDMs and 15% microglia among TAMs (composition of tumor-associated macrophages)
- other median survival < 15 months (GBM standard-of-care outcomes)
- count 101 machine learning algorithms (evaluation of prognostic/immunological value of COL6A3+ TAFs and GPNMB+ MDMs)
- count 33 newly diagnosed GBM; 7 responders; 5 non-responders to neoadjuvant therapy (scRNA-seq patient cohort)
- count 691 (TCGA), 693 and 325 (CGGA), 270 (Ivy-GAP), 371 (GLASS), 487 (GSE108474) glioma patients (bulk RNA-seq survival/ML cohorts)
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.
This paper integrates scRNA-seq, spatial transcriptomics, and bulk RNA-seq data from multiple patient cohorts to characterize the GBM tumor microenvironment, focusing on COL6A3+ tumor-associated fibroblasts and GPNMB+ macrophages. Bioinformatic analyses encompass dimensionality reduction, graph-based clustering, differential expression testing, pathway enrichment (GSEA, GSVA, clusterProfiler), and 101 machine learning algorithms applied to bulk cohorts for prognostic modeling. Experimental validation was conducted via multiplex immunohistochemistry, flow cytometry, RT-qPCR, ELISA, chemotaxis assays, and atomic force microscopy, with in vitro experiments performed in three independent replicates. Cross-cohort validation was performed across five bulk RNA-seq survival cohorts and four immunotherapy response cohorts.
| Test | Applied to | n | Assumptions |
|---|---|---|---|
| GSEA two-sided permutation test with Benjamini-Hochberg adjustment | Comparison of differentially expressed genes between two groups across gene sets; NES > 1 and adjusted p < 0.05 used as significance threshold | — | not stated |
| Seurat FindAllMarkers (Wilcoxon rank-sum test, default) | Cluster-level differentially expressed gene identification (min.pct = 0.2, logfc.threshold = 0.2) | Cell counts per cluster; not stated explicitly | not stated |
| clusterProfiler over-representation / enrichment analysis (hypergeometric test) | Pathway enrichment of DEGs against human C2 and C5 MSigDB gene sets; adjusted p < 0.05 threshold | — | not stated |
| GSVA (gene set variation analysis) | Gene set scoring on cell count matrix | — | not stated |
| 101 machine learning algorithms (ensemble evaluation) | Prognostic and immunological model building across bulk RNA-seq cohorts (TCGA n=691, CGGA n=693 and n=325, Ivy-GAP n=270, GLASS n=371, GSE108474 n=487) | Varies by cohort; largest n=693 | not stated |
| inferCNV hierarchical clustering | Distinguishing malignant from non-malignant cells based on CNV profiles; myeloid and NK/T cells used as baseline | — | not stated |
| 2^(-ΔΔCt) relative quantification (RT-qPCR) | Relative mRNA expression levels; GAPDH as internal control; three independent experiments | 3 independent experiments | not stated |
-
Differential expression between clusters was identified using Seurat's FindAllMarkers, which applies a Wilcoxon rank-sum test by default↳ Could also: Pseudobulk differential expression methods such as DESeq2 or edgeR, which aggregate single-cell counts to the sample level before testing — Pseudobulk approaches model biological variability across donor samples rather than treating individual cells as independent observations, which can reduce inflation of false positives when multiple samples per group are available — an active area of methodological discussion in the scRNA-seq literature
-
Batch effects across patient samples were corrected using Harmony applied to principal components↳ Could also: scVI (variational autoencoder-based integration) or BBKNN (batch-balanced k-nearest neighbor graph) as alternative integration strategies — Different integration methods vary in how they handle non-linear batch effects and may preserve or remove biological signal to different degrees; benchmarking studies suggest the optimal choice can be dataset-dependent, and reporting multiple methods' outputs is one way to assess robustness
-
Pathway activity in individual cells was scored using Seurat's AddModuleScore↳ Could also: AUCell, which ranks gene expression within each cell and computes an area-under-the-curve score for each gene set — AUCell is less sensitive to cell size differences and uses a rank-based approach that can be more robust when comparing cells with highly variable total transcript counts, and explicitly models the recovery curve per cell
-
Cluster resolution was explored across a range (0.1–1) without a stated criterion for final selection↳ Could also: Stability-based resolution selection using tools such as clustree or scclust, which visualize cluster membership across resolutions to identify stable partitions — A quantitative stability criterion provides a reproducible, data-driven rationale for the chosen resolution and allows readers to assess sensitivity of downstream findings to this hyperparameter
-
Prognostic modeling used 101 machine learning algorithms evaluated across bulk RNA-seq cohorts↳ Could also: A pre-specified, smaller set of models evaluated with nested cross-validation and calibration assessment (e.g., Harrell's C-index with confidence intervals) — Prospectively constraining the model set and reporting calibration alongside discrimination reduces the risk of selecting an algorithm that performs well by chance across cohorts, and confidence intervals on performance metrics convey uncertainty in a way that point estimates from many algorithms do not
-
In vitro experiments (RT-qPCR, chemotaxis, flow cytometry) were conducted with three independent replicates↳ Could also: Reporting results with 95% confidence intervals or individual data points alongside a measure of central tendency (e.g., mean ± SD), together with a pre-stated statistical test for each comparison — With small n (n=3), individual data points and CIs communicate the full distribution and estimation uncertainty more transparently than summary statistics alone, and pre-stating the test avoids post-hoc analytical flexibility
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-41174767
Paper: Du Y, Long X, Li X, et al. Spatial-reprogramming derived GPNMB+ macrophages interact with COL6A3+ fibroblasts to enhance vascular fibrosis in glioblastoma. Genome Medicine 2025. DOI 10.1186/s13073-025-01553-2. PMID 41174767 · PMCID PMC12577258.
Nature of the artifact (important)
- The paper's Data/Code Availability states verbatim: "This study did not generate new sequencing data or original code for analysis. All data and code used for analysis in this study are publicly available."
- The only "code" link (https://github.com/zhanghao-njmu/SCP) is the SCP R package — a generic third-party single-cell toolkit used for visualization, not the authors' analysis code. There is no authors' pipeline repo.
- Per brief rule P16, reproduction by applying a standard third-party pipeline to the paper's own public data is fully valid. We therefore reproduce pipeline-derivable marker / cell-identity claims by running a standard Seurat scRNA-seq pipeline (matching the paper's described method) on one of the paper's own public datasets.
Datasets used by the paper (from Methods → "Datasets analyzed in this study")
| accession | source | content | obtainability |
|---|---|---|---|
| GSE131928 | GEO | GBM scRNA-seq (Neftel 2019; 10X 16,201 + SS2 7,930 cells) | public, processed TPM matrices directly downloadable ← chosen |
| GSE138794 | GEO | GBM scRNA-seq | public |
| GSE182109 | GEO | glioma immune-cell atlas (Abdelfattah 2022; ~3 GB RAW) | public |
| OMIX003593 | NGDC | neoadjuvant scRNA + spatial (the key novel cohort) | open archive (OMIX) |
| HRA007138 | NGDC GSA-Human | GBM scRNA (controlled-access class) | controlled / on-request (DAC) |
| TCGA, CGGA, Ivy-GAP, GLASS | bulk | survival / deconvolution cohorts | public |
In scope (pipeline-derived, low-hanging, public-data-derivable) — ATTEMPTED
Run standard Seurat (v4) on GSE131928 10X processed TPM (the room's named accession) and verify cell-identity claims that the paper states are derivable from public single-cell data:
- C1 — GPNMB marks a macrophage/myeloid population ("GPNMB+ MDMs"). Check: GPNMB mean expression is highest in the macrophage/myeloid compartment.
- C2 — COL6A3 expression is cell-type-restricted to the stromal compartment (paper: "high expression of COL6A3 occurred only in stromal cells"). Check: COL6A3 mean expression per major cell type.
- C3 — Marker-gene identity of cell types: macrophage (CD68/AIF1/C1QB/PTPRC), T cell (CD3D/CD3E), oligodendrocyte (MBP/PLP1), malignant (EGFR/SOX2/OLIG1), mural/SMC (ACTA2/MYH11/TAGLN), fibroblast/TAF collagens (COL1A1/COL3A1/COL6A3).
- C4 — Dataset shape: recover the 10X cell count (~16,201) and a sensible cluster/cell-type structure with the paper's preprocessing parameters (HVG 2000, 50 PCs, Harmony batch correction, FindAllMarkers min.pct=0.2, logfc.threshold=0.2).
Out of scope — NOT attempted (and why)
- The novel spatial-reprogramming / vascular-fibrosis mechanism (COL6A3+ TAF
→ GPNMB+ MDM reprogramming, GPNMB–ITGB5 interaction, spatial co-localization):
depends primarily on the neoadjuvant cohort OMIX003593 / HRA007138.
HRA007138 is GSA-Human controlled-access (DAC application required) →
data_restrictedfor the mechanism-defining data. This is the hard ~20%; not attempted. (80/20 rule.) - 101-algorithm ML prognostic signature (TCGA/CGGA/Ivy-GAP/GLASS): large multi-cohort modelling; out of the low-hanging core; not attempted.
- CIBERSORTx deconvolution: requires the CIBERSORTx web tool / token; not attempted.
- All wet-lab validation (mIHC, AFM, flow, ELISA, organoids, qRT-PCR, cilengitide): non-computational → out of scope by definition.
Pipeline named per in-scope result
Seurat v4 (LogNormalize-equivalent on the deposited TPM, FindVariableFeatures vst nfeatures=2000, ScaleData, RunPCA npcs=50, Harmony by patient, FindNeighbors dims=1:50, Fin
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.
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.