Microglia maintain structural integrity during fetal brain morphogenesis.
The main results reproduced, with only marginal, non-material deviations.
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 reproduce the pipeline DATA point, with an important provenance caveat. The harvester's RU links are partly off-target: the linked code (github.com/satijalab/sctransform) is NOT the authors' own repo but IS the paper's genuine normalization tool (P16-valid); the linked data GSE121654 is NOT the authors' own data but Hammond et al. 2019 (PMID 30471926), which the paper REUSES (authors' own scRNA-seq is in ArrayExpress E-MTAB-13581, not linked). On «our HPC»/«infra» I (1) downloaded GSE121654_RAW.tar (sha256 b7f4747..., 47 Drop-seq DGE matrices) and counted 114,050 deposited barcodes vs the paper's reported 76,149 post-QC reanalyzed cells -- consistent (76,149 is a QC subset of 114,050), not a mismatch; (2) ran the linked tool SCTransform(glmGamPoi) end-to-end on the merged data (110,513 post-QC cells -> 3,000 variable genes -> 25 clusters), demonstrating the linked pipeline reproduces on the linked data. NOT a 1:1 of the exact 76,149 integer: that requires the authors'/Hammond's private tSNE coordinates + cluster labels ('provided by the authors'), not in the public deposit -- a data-provenance limitation, NOT a fabrication signal (76,149 is independently anchored to the GSE121654 title). NOT attempted (the hard 20%): the 2,517-ATM sub-count, the authors' own E-MTAB-13581 data (115,256 cells), the La Manno reuse (229,948 / 1,711 cells), and all wet-lab/imaging/Metascape results (non-pipeline). Verdict: partial reproduction with no fabrication concern detected; a human reviewer should confirm via AUDIT.md.
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 68assessed: 2026-06-15 ⛓ 9fe71fa8a79d
✎ 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 2026-06-15no human curator yet
- Last updated
- 2026-09-19
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: sonnetThe paper tests whether embryonic microglia, which display heterogeneous transcriptomic states and accumulate at specific fetal brain locations, have functional roles—specifically whether ATM-like microglia accumulating at cortical boundaries maintain structural integrity of the developing brain.
- ★ Embryonic ATM-like microglia accumulate at two fetal cortical boundaries, the cortico-striato-amygdalar boundary (CSA) and cortico-septal boundary (CSB), resembling post-natal axon-tract-associated microglia (ATM) finding
- ★ Embryonic ATM-like MG share a core transcriptomic signature (Spp1, Csf1, Igf1, Lgals3, Gpnmb) with post-natal ATM/PAM populations finding
- ★ MG accumulation with ATM-core marker expression at the CSA is conserved in human fetal brain from GW9 to GW14 finding
- ★ Microglia are required to prevent formation of large cavitary lesions at the CSA and CSB during embryonic development finding
- ★ Cavitary lesion formation upon MG loss is specific to microglial depletion and not observed in Cx3cr1, Dap12/TyroBP, or CR3 mutants, or after maternal immune activation finding
- ★ Spp1 (osteopontin), an ATM-core factor, contributes mechanistically to microglia's role in preventing cavitary lesions mechanism
- ★ Microglia and Spp1 contribute to rapid repair of cavitary lesions once formed finding
- CSA ATM-like microglia display amoeboid morphology and elevated phagocytic activity/CD68 expression compared to neighboring microglia finding
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| single-cell RNA-seq (reanalysis of La Manno dataset) | mouse embryonic brain, E9–E18 | none | microglial cluster identity (cycling, non-cycling, ATM-like) and marker gene expression | — |
| immunolabeling/immunohistochemistry | Cx3cr1(gfp/+) and CD11c-EYFP mouse embryonic brain sections (E14.5–E18.5) | none | co-expression of microglial (P2Y12, IBA1) and ATM-core markers (CD11c, Mac2/Lgals3, Clec7a, Spp1, GPNMB) at CSA/CSB | — |
| immunolabeling | human fetal brain sections, GW9–GW14 | none | co-expression of ATM-core markers with IBA1 at CSA | — |
| pHrodo phagocytosis assay | ex vivo mouse embryonic brain slices, E14.5 | none | phagocytic activity of CSA microglia | — |
| CD68 immunolabeling/quantification | mouse embryonic brain, E14.5 | none | lysosomal marker coverage in CSA ATM vs. neocortical microglia | — |
| electron microscopy | mouse embryonic brain (PLX3397-treated) | PLX3397-induced microglial depletion | ultrastructure of cavitary lesions (absence of cells/basal lamina, membrane debris) | electron microscope |
| histological lesion analysis after pharmacologic/genetic microglial depletion | mouse embryos (CSF1R-blocking antibody AFS98, PLX3397 diet E6.5–E15 or E12.5–E15, Pu.1 mutants, Csf1r(ΔFIRE/ΔFIRE)) | microglia/macrophage depletion or loss (KO/pharmacological) | formation of cavitary lesions at CSA and CSB, E14.5–E18.5 | — |
| histological lesion analysis in functional mutants | mouse embryos (Cx3cr1, Dap12/TyroBP, CR3 mutants; maternal immune activation model) | gene knockout or prenatal inflammatory challenge | presence/absence of cavitary lesions at CSA | — |
- – scRNA-seq identified 3 distinct embryonic MG clusters: cycling, non-cycling, and ATM-like, with ATM-like cells prominent from E14 onward
- ▲ Embryonic ATM-like MG DEGs overlap significantly with post-natal ATM (Hammond) and PAM (Li) signatures FC>1.5, Bonferroni-adjusted p<1e-10
- ▲ Approximately 60% of CSA microglia co-expressed ATM-core markers (CD11c, Mac2, Clec7a, Spp1, GPNMB) at E15 ~60%
- ▲ Approximately 70% of CSB microglia co-expressed ATM markers at E18.5 ~70%
- – Human fetal brains (GW9–GW14) showed a conserved pattern of ATM-marker-expressing microglia at the CSA
- ▲ All microglia depletion/loss models (AFS98 antibody, PLX3397 long and short treatment, Pu.1 mutants, Csf1r-ΔFIRE) produced large cavitary lesions at the CSA starting E14.5 and persisting to E18.5, and similar lesions at the CSB
- – No cavitary lesions observed in Cx3cr1, Dap12/TyroBP, or CR3 mutants, or after maternal immune activation
- ▲ CSA ATM microglia showed higher CD68 coverage than neocortical ramified microglia
- count 1,711 microglial cells (embryonic MG extracted from La Manno scRNA-seq dataset for clustering analysis)
- fold_change FC > 1.5, Bonferroni-adjusted p < 1e-10 (overlap between embryonic and post-natal ATM DEGs (Venn diagram, Figure 1D))
- other ~60% (proportion of CSA microglia co-expressing ATM-core markers at E15)
- other ~70% (proportion of CSB microglia co-expressing ATM markers at E18.5)
- count n Ncx = 28, n CSA = 29 cells (4 mice, 2 litters) (CD68 coverage comparison between neocortical and CSA microglia)
- count E14.5: n controls = 21, n AFS = 6; E18.5: n controls = 15, n AFS = 17 (sample sizes for AFS98 CSF1R-blocking antibody depletion experiment)
- count E14.5: n controls = 5, n PLX3397 = 5; E18.5: n controls = 12, n PLX3397 = 11 (sample sizes for PLX3397 feeding E6.5–E15 depletion experiment)
- count E15.5: n controls = 5, n PLX3397-E12 = 9; E18.5: n controls = 12, n PLX3397-E12 = 10 (sample sizes for short PLX3397 feeding E12.5–E15 depletion experiment)
Statistical methods review
Model: opusA 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 combines single-cell transcriptomic re-analysis of published datasets with imaging-based quantification in genetically modified and pharmacologically treated mouse models. Differential gene expression between microglial clusters was assessed with fold-change thresholds and Bonferroni-adjusted p values, while quantitative comparisons of marker co-expression and cellular measures between anatomical regions or experimental groups were made with non-parametric tests. Results were reported as means ± SEM with significance indicated by p-value thresholds, and sample sizes were given as numbers of cells and/or embryos per condition.
| Test | Applied to | n | Assumptions |
|---|---|---|---|
| Mann-Whitney U test | comparisons of ATM marker co-expression and CD68 coverage between neocortex and CSA microglia (Figure S1, e.g., S1I, S1K) | e.g., n_Ncx = 28, n_CSA = 29 cells in 4 mice from 2 litters (CD68 coverage); other comparisons from at least 3 mice from 2 litters | not stated |
| Differential expression with fold-change threshold and Bonferroni-adjusted p value | identification/overlap of embryonic vs post-natal ATM DEGs (Figure 1D; Venn diagrams) | FC > 1.5, Bonferroni-adjusted p < 1e-10; based on cells in scRNA-seq datasets (e.g., n = 1,711 microglial cells extracted) | not stated |
-
Group differences in marker co-expression and CD68 coverage were assessed with the Mann-Whitney U test.↳ Could also: A parametric two-sample t-test (with a normality check), or a mixed-effects model that nests cells within mice/litters, could also have been used. — A mixed-effects/hierarchical model would explicitly account for cells being clustered within animals and litters, and would treat the animal as the unit of replication; this can be informative when many cells come from a few mice.
-
Dispersion was summarized as means ± SEM.↳ Could also: Standard deviation or a 95% confidence interval could also be reported, optionally alongside individual data points. — SD conveys the spread of the data and a CI conveys precision of the estimate; both are often favored for small samples and make the variability directly interpretable to readers.
-
Significance was reported as p-value thresholds (*, **, ***).↳ Could also: Exact p values together with effect-size estimates (e.g., differences in means or rank-biserial correlation) and their confidence intervals could also be presented. — Exact p values and effect sizes give a fuller, quantitative picture of the magnitude and certainty of an effect beyond a binary threshold.
-
Differential expression used a fold-change cutoff with Bonferroni-adjusted p values.↳ Could also: A Benjamini-Hochberg false discovery rate approach, or model-based single-cell DE frameworks (e.g., MAST, DESeq2, or pseudobulk per animal), could also be applied. — FDR control is often preferred in high-dimensional genomics for its sensitivity, and pseudobulk/model-based methods can better account for biological replicate structure in single-cell data.
-
Multiple pairwise comparisons across regions and several depletion models were each evaluated individually.↳ Could also: A single ANOVA (or its non-parametric Kruskal-Wallis analog) with a post-hoc multiple-comparison correction could also be used when more than two groups are compared. — An omnibus test with post-hoc correction controls the family-wise error rate across the set of related comparisons within an experiment.
-
Sample sizes were reported as observed counts of embryos and cells per condition.↳ Could also: An a priori power/sample-size justification could also accompany these counts. — A stated power analysis clarifies the effect size the study was designed to detect and supports interpretation of comparisons with smaller group sizes.
Result convergence & founder nodes
Findings this paper shares with others that ran a comparable experiment. A node’s strength is how many independent papers report it (replication breadth) — not how often it is cited, so a heavily-replicated but under-cited founder still stands out.
-
ATM-core-marker-expressing microglia are present at the CSA in human fetal brain from GW9 to GW14, demonstrating cross-species conservation of this developmental microglial subtype.imaging human fetal brain 2024×1papers★ This paper is the founder (earliest)
-
Approximately 60% of microglia at the CSA co-express ATM core proteins (CD11c, Mac2, Clec7A, SPP1, GPNMB), indicating selective enrichment of ATM-like microglia at this embryonic brain boundary.imaging mouse embryonic brain up 2024×1papers★ This paper is the founder (earliest)
-
No cavitary lesions arise in Cx3cr1, Dap12/TyroBP, or CR3 mutants or in maternal immune activation models, indicating the structural phenotype requires total microglial loss rather than general immune or signalling perturbation.imaging mouse embryonic brain none 2024×1papers★ This paper is the founder (earliest)
-
Microglial depletion by antibody, pharmacological, or genetic means induces cavitary lesions at the corpus callosum anlage from E14.5, demonstrating that ATM-like microglia are required for structural integrity at this boundary.imaging mouse embryonic brain up 2024×1papers★ This paper is the founder (earliest)
-
ATM-like microglia at the CSA show higher CD68 coverage than neighboring SPP1-negative ramified microglia, indicating elevated phagocytic activity at the embryonic brain boundary.imaging mouse embryonic brain up 2024×1papers★ This paper is the founder (earliest)
-
Re-analysis of published scRNA-seq data identifies a distinct embryonic ATM-like microglial cluster in mouse brain present from E14 onward, alongside cycling and non-cycling populations.scRNA-seq mouse embryonic brain 2024×1papers★ This paper is the founder (earliest)
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-38309258
Paper: Lawrence AR, Canzi A, Bridlance C, … Garel S. "Microglia maintain structural integrity during fetal brain morphogenesis." Cell 187(4):962-980.e19, 2024-02-15. DOI 10.1016/j.cell.2024.01.012 · PMCID PMC10869139.
RU-linked artifacts (as harvested) — verified
| Artifact | RU link | Reality (verified) |
|---|---|---|
| Code | github.com/satijalab/sctransform | Legit & used by the paper. Methods cite "SCTransform v0.3.5" + glmGamPoi v1.10.2 (RRID SCR_014601) for normalization/scaling. NOT the authors' own analysis repo (they ship no own repo), but P16-valid: the linked third-party tool IS part of the paper's pipeline. |
| Data | geo:GSE121654 | Reused external dataset, not the paper's own. GSE121654 = Hammond et al. 2019 (PMID 30471926, Stevens lab), "76,149 microglia across the mouse lifespan." The Lawrence paper reanalyzes it (postnatal ATM signature; Table S1). The paper's OWN scRNA-seq is deposited separately under ArrayExpress E-MTAB-13581. |
Reproducible pipeline-derived results (in scope)
The paper's computational outputs come from a standard Seurat (v4.1.1) + SCTransform (v0.3.5, glmGamPoi) workflow applied to three scRNA-seq datasets:
| # | Reported value | Dataset | Pipeline | Paper location |
|---|---|---|---|---|
| C1 | 76,149 cells (incl. 2,517 ATM) reanalyzed | GSE121654 (Hammond, RU-linked) | filter + reuse authors' tSNE/cluster labels | Results / Fig. legend "tSNE of the 76,149 cells" |
| C2 | 115,256 cells sequenced (5,830 M reads) | E-MTAB-13581 (own) | CD45+ purification → SCTransform → PCA(5 PCs) → UMAP | Methods / STAR |
| C3 | 229,948 cells reused | La Manno loom (mousebrain.org) | reuse | Methods |
| C4 | 1,711 cells (510 atm, 415 cycling, 786 non-cycling) | La Manno subset | filter ≥10 cells/gene | Results |
What we attempt (80/20)
- PRIMARY (C1): Download the RU-linked GSE121654 deposited DGE matrices on «infra», sum per-sample cell-barcode counts, compare to the reported 76,149. Uses BOTH RU-linked artifacts. Cheap, crisp, falsifiable.
- SECONDARY (tool run): Execute the RU-linked tool SCTransform v0.3.5 (Seurat + glmGamPoi) on the GSE121654 matrices per the paper's stated parameters; report derived stats (variable genes, PCs, clusters) — demonstrates the linked pipeline runs on the linked data and yields sensible output.
Out of scope / not attempted (the hard 20%)
- The 2,517-ATM sub-count and reproduced tSNE coordinates depend on cluster labels the authors obtained privately from the Hammond authors ("coordinates and reported clusters were provided by the authors") — not derivable from the public deposit. Skipped.
- C2 (E-MTAB-13581) own-data full reprocessing: feasible but heavier; attempt only if cheap. The 115,256 figure is a sequencing/loading count, weakly pipeline-derived.
- All wet-lab, imaging (Imaris), Metascape enrichment, electrophysiology — out of scope (non-pipeline).
Honesty notes (possible-fabrication watch)
- 76,149 is independently checkable: it is the title number of GSE121654, so it is a strong external anchor. Deposited raw barcode columns may EXCEED 76,149 (raw DGE before the authors' final QC) — any gap is expected and reported, not a mismatch in the paper.
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.
This is a partial, pipeline-level reproduction with no fabrication concern. The RU-linked data (GSE121654) is reused Hammond et al. 2019 data, and the linked code (satijalab/sctransform) is the paper's genuine normalization tool, which ran end-to-end (110,513 cells → 25 clusters) with only minor version drift. The single comparable number, 76,149 post-QC cells, is consistent with the 114,050 deposited barcodes (a QC subset) and is independently anchored to the GSE121654 title, but it is not recoverable 1:1 because Hammond's/the authors' private tSNE coordinates and cluster labels were not deposited. The deviations are input-side (cohort/QC definition + data provenance), on our-method/data-availability side rather than the authors' — moderate in magnitude and fully explainable, so overall yellow.
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 [email protected].
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.