The identification of a Distinct Astrocyte Subtype that Diminishes in Alzheimer's Disease.
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.
- ✓No authors-side cause for any deviation
- 🟡Could not use the authors’ exact input data
- 🟡Reported values were only indirectly comparable
- 🟡A deviation arose in the data or preprocessing
- 🟡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
Re-analysis paper (Wei 2024) over two snRNA-seq datasets. The flagship CellChat result (119 control / 123 AD ligand-receptor interactions, Fig 3) and the human astrocyte subtype (Fig 2/4) are computed on syn18485175 = Mathys 2019 ROSMAP PFC = AMP-AD CONTROLLED access (DUC required) -> not reproducible on public data (data_restricted). The brief pairs Code=CellChat with Data=GSE143758, but in the paper these are decoupled: CellChat runs on the restricted human data while public GSE143758 (mouse Habib DAA) feeds a Seurat astrocyte subclustering. We reproduced that public pipeline on «our HPC»: Seurat 4.3.0.1 on the pre-extracted hippocampus astrocyte UMI matrix (20970 genes x 25076 nuclei). The reported 12 subclusters (res=0.5) came out as 10 at res=0.5 and exactly 12 at res=0.6 (partial; expected Seurat version/seed sensitivity, no version pinned in paper). The canonical DAA population reproduced cleanly as a distinct AD-enriched subcluster (within-tol). NOT attempted: the restricted-human CellChat/Fig2/Fig4/GSEA results, and wet-lab IHC (non-pipeline). Honest outcome: described well enough; partial 1:1 on the public-data portion, flagship number behind controlled access.
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 57assessed: 2026-06-15 ⛓ 67a26a06a9bf
✎ 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
- 2026-08-05
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: opusBecause astrocytes are heterogeneous and play complex protective and harmful roles in Alzheimer's disease (AD), the authors hypothesize that distinct astrocyte subpopulations differ in their response to amyloid beta and tau, and aim to identify AD-associated astrocyte subtypes via single-nuclei transcriptomics.
- ★ A distinct astrocyte subpopulation marked by low GFAP, plus AQP4 and CD63 expression, exists in normal brain but is diminished in AD samples in both human and mouse. finding
- ★ This GFAP-low/AQP4+/CD63+ astrocyte subpopulation is functionally enriched in amyloid beta (Aβ) clearance and tau protein binding. finding
- ★ The diminished astrocyte subpopulation was verified by immunohistochemistry of marker genes in human control/AD prefrontal cortex and in mouse AD models. finding
- ★ Ligand-receptor interactions between astrocytes and other cell types are significantly altered in AD. finding
- ★ Integrative analysis of publicly available human and mouse snRNA-seq AD datasets to characterize astrocyte heterogeneity and subpopulation-specific transcriptomic changes. method
- Astrocytes can clear amyloid beta and internalize/bind tau, including via cell-surface tau-binding proteins such as HSPGs. mechanism
- Targeting specific astrocyte subpopulations may offer new therapeutic avenues for AD. finding
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| single-nuclei RNA sequencing (snRNA-seq) data mining and integration (Seurat, PCA, UMAP, FindClusters) | human prefrontal cortex (syn18485175: 24 no-AD-pathology and 24 AD-pathology individuals) and mouse hippocampus/cortex (GSE143758: 21 control, 16 5xFAD, age 1.5-19.6 months); additional human datasets GSE140399, GSE157827, GSE174367, GSE138852 | none (disease vs control comparison; 5xFAD genotype in mouse) | astrocyte lineage cell clustering, subpopulation proportions, marker/DEG expression | 10X Genomics; CellRanger 3.0.2; mm10/GRCm38; Seurat |
| gene set enrichment analysis (GSEA) / DEG identification | human snRNA-seq astrocyte lineage clusters (syn18485175) | none (cluster1 vs other clusters) | enriched gene sets (Aβ clearance, tau binding, lysosome, etc.) and fold changes | fgsea v1.16.0, MSigDB C5 gene sets |
| cell-cell communication / ligand-receptor interaction analysis | human snRNA-seq control vs AD datasets (syn18485175) | none (AD vs control) | number and strength of intercellular interactions, signaling pathways | CellChat with CellChatDB |
| multiplex immunohistochemistry / immunofluorescence (confocal) | human frontal lobe cortex paraffin sections from 3 AD patients and 3 age-matched controls (autopsy) | none (AD vs control) | counts of DAPI+GFAP-low/AQP4+/CD63+ cells vs total GFAP+ cells | Nikon AR Confocal Microscope 20x; NIS-Elements HC v5.41.02; antibodies anti-GFAP, anti-CD63, anti-AQP4 |
| immunohistochemistry / immunofluorescence (confocal) | mouse brain sections, APP NL-F/NL-F knock-in (APP NL-F) mice vs aged C57BL/6 WT controls | genetic knock-in (APP NL-F) | GFAP/AQP4/CD63 (and CD31, Aβ) immunostaining, Aβ engulfment | Nikon AXR confocal microscope 40x; antibodies anti-GFAP, anti-AQP4, anti-CD63-APC, anti-CD31, anti-Aβ |
| immunocytochemistry (ICC) Aβ internalization assay | mouse primary astrocyte culture from neonatal (P-7) mouse brain | FAM-Aβ42 treatment (2 hours) | Aβ42 internalization in GFAP/CD63 stained astrocytes | Nikon AXR confocal microscope 40x; FAM-Aβ42 (AnaSpec) |
- – In the human dataset, five astrocyte subpopulations (clusters 0-4) were identified; proportion of GFAP-positive cells in clusters 0 and 2 increased while cluster1 decreased in AD.
- ▼ In the mouse AD dataset, 12 astrocyte subpopulations were identified; cluster 7 (low Gfap expression) was almost completely lost in AD samples.
- ▼ GFAP-low astrocytes detected in both control mice and humans, with a significant decrease in a specific GFAP-low subpopulation in AD.
- ▲ Human cluster1 (GFAP-low, diminished in AD) showed GSEA enrichment for Aβ clearance and tau protein binding, plus secondary lysosome/secretory vesicle, antioxidant activity, mitochondrial protein complex, synapse, and endocytic/phagocytic vesicle gene sets.
- – Total number and strength of intercellular interactions differed between control and AD groups.
- – Astrocytes were heterogeneous with distinct GFAP-low and GFAP-high expression patterns in both mouse and human datasets.
- count 24 individuals with no AD pathology and 24 with AD pathology (human prefrontal cortex snRNA-seq dataset (syn18485175))
- count 21 control and 16 5xFAD (mouse hippocampus/cortex snRNA-seq dataset (GSE143758), age 1.5-19.6 months)
- count 3 AD patients and 3 age-matched controls (human frontal cortex autopsy samples for IHC)
- pvalue P-value < 0.05 (significance threshold for GSEA gene sets and CellChat signaling pathways)
- pvalue *** p < 0.001 (significance level indicated for IHC cell-count t-tests)
- count 1000 permutations (fgsea GSEA parameter)
- count top 30 PCs (PCA components selected for downstream snRNA-seq analysis)
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 study primarily employs computational reanalysis of publicly available snRNA-seq datasets (human and mouse) using Seurat-based clustering, differential gene expression (DEG) analysis, and gene set enrichment analysis (GSEA via fgsea) to characterize astrocyte subpopulations in Alzheimer's disease. Cell-cell communication was assessed with the CellChat framework. IHC on a small validation cohort (3 AD, 3 control) and primary astrocyte culture experiments were analyzed with independent t-tests, with results reported as mean ± SEM. Statistical significance was defined at p<0.05 throughout.
| Test | Applied to | n | Assumptions |
|---|---|---|---|
| Independent two-sample t-test | IHC cell counting comparisons between AD and control brain sections (human frontal cortex and mouse APP NL-F model) and primary astrocyte culture Aβ internalization assay | n = number of sections; 1 image from each of 3 slides per patient, 3 patients per group (exact section-level n per comparison not explicitly stated) | stated (normal distribution assumed) |
| Seurat FindAllMarkers / FindMarkers (Wilcoxon rank-sum test by default) | Differential gene expression between astrocyte subpopulations and between AD and control cells within clusters | Cell-level counts within clusters: 24 AD and 24 control individuals (human dataset); 21 control and 16 5xFAD mice (mouse dataset) | not stated |
| GSEA permutation test via fgsea (1000 permutations) | Gene set enrichment analysis against MSigDB C5 gene sets for each astrocyte subpopulation | — | not stated |
| CellChat mass-action model with permutation-based p-value | Ligand-receptor interaction strengths and signaling pathway activity between cell types in control vs. AD | Minimum 10 cells per group required; cell counts from integrated multi-donor datasets | not stated |
-
Differential gene expression across astrocyte subpopulations was identified using Seurat's FindAllMarkers (Wilcoxon rank-sum at the single-cell level) across a 24-donor human dataset↳ Could also: Pseudobulk methods such as DESeq2 or edgeR applied to per-donor aggregated counts could also be used — Pseudobulk approaches treat biological replicates (donors) as the unit of analysis rather than individual cells, which better accounts for within-donor correlation and is widely recommended for multi-donor snRNA-seq designs to control type I error
-
IHC group comparisons used independent t-tests with n=3 biological samples per group↳ Could also: A non-parametric Mann-Whitney U test could also be used — With only 3 observations per group, the normality assumption underlying the t-test cannot be empirically verified; non-parametric alternatives make no distributional assumptions and are a common choice in small-n histological studies
-
IHC results are reported as mean ± SEM with n=3 per group↳ Could also: Mean ± SD or individual data points overlaid on summary plots could also be used — With n=3, SD directly conveys the spread of raw observations rather than precision of the mean; plotting individual data points alongside summary statistics is increasingly recommended for small samples to make the underlying data visible to readers
-
GSEA significance was defined as p<0.05; the paper does not explicitly state whether raw or fgsea-adjusted (BH-FDR) p-values were used↳ Could also: Explicitly thresholding and reporting on fgsea's padj (BH-FDR adjusted across all tested gene sets) could also be used — Testing hundreds of gene sets simultaneously increases the expected number of false positives; fgsea computes BH-adjusted p-values by default and reporting these is a widely adopted convention for pathway enrichment analyses
-
Multiple t-tests were performed across IHC outcomes without a stated correction for multiple comparisons↳ Could also: A Bonferroni or Benjamini-Hochberg FDR correction across the family of IHC comparisons could also be applied — When several comparisons are evaluated across the same tissue sections or experimental conditions, stating and applying a correction method makes the false-discovery rate explicit, even when the number of comparisons is small
-
Cell-cell communication was inferred using CellChat's mass-action ligand-receptor model↳ Could also: Alternative tools such as NicheNet, LIANA, or CellPhoneDB could also be used — Different frameworks rely on distinct ligand-receptor databases and statistical models; cross-tool comparison is an approach some studies use to identify interactions that are robust across methodological assumptions
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.md — pmid-38502590
Paper: Wei H et al. (2024) The identification of a Distinct Astrocyte Subtype that Diminishes in Alzheimer's Disease. Aging Dis. PMID 38502590 / PMC11567244 / DOI 10.14336/ad.2024.0205-1.
Nature of the study: A re-analysis paper. It mines two pre-existing public/ controlled snRNA-seq datasets to identify a "GFAP^low AQP4+ CD63+" astrocyte subtype that is reduced in AD, then validates it with its own wet-lab IHC.
Datasets named in the paper
| label | accession | host | access | used for |
|---|---|---|---|---|
| Mouse 5xFAD | GSE143758 | GEO | public | astrocyte subclustering (Supp Fig 2) — Habib DAA data |
| Human PFC | syn18485175 ("snRNAseqPFC_BA10", Mathys 2019) | Synapse | CONTROLLED (ROSMAP/AMP-AD, DUC required) | Fig 2 (5 astro subclusters), Fig 3 (CellChat), Fig 4, GSEA |
| validation | GSE140399, GSE157827, GSE174367, GSE138852 | GEO | public | secondary validation (not the headline) |
Pipeline-derived results — classification
IN SCOPE (public data + standard pipeline) → attempted
- R1. Mouse astrocyte subclustering (GSE143758). Seurat re-clustering of the pre-extracted hippocampus astrocyte UMI matrix; PCA top 30 PCs, FindClusters res=0.5. Reported: 12 subclusters (clusters 0–11) (Supp Fig 2A-B). Pipeline: Seurat. → REPRODUCE on «our HPC».
- R2. Marker identification of the GFAP^low/AQP4+/CD63+ population within the mouse astrocytes (qualitative: a subcluster low in Gfap, high Aqp4/Cd63). Pipeline: Seurat FindAllMarkers / per-cluster mean expression. → check.
OUT OF SCOPE — data_restricted (cannot obtain) → NOT attempted
- R3. CellChat ligand-receptor counts (Fig 3C): 119 interactions in control, 123 in AD. Per Methods, CellChat was run on the human syn18485175 data ("we compared ast_cluster1 with other cell types in the human dataset"). syn18485175 = Mathys 2019 ROSMAP PFC = AMP-AD controlled access (requires a signed Data Use Certificate via Synapse/RADC governance). The brief's named code artifact (CellChat) thus has its input behind controlled access → the flagship CellChat numbers are not reproducible on public data.
- R4. Human 5 astrocyte subclusters (Fig 2A-B), GSEA enrichment (Fig 2E-H), GFAP^low proportion control vs AD (Fig 4F). All on syn18485175 → restricted.
OUT OF SCOPE — non-pipeline (wet-lab) → NOT attempted
- Human IHC / confocal quantification (3 AD + 3 control), NIS-Elements segmentation. Manual/experimental, not a bioinformatic pipeline.
Note on the code/data pairing in the brief
The brief lists Code=CellChat + Data=GSE143758. In the paper these are not
paired: CellChat ran on the human restricted data, while GSE143758 (mouse)
feeds a Seurat subclustering. We therefore reproduce the public-data pipeline
(Seurat subclustering on GSE143758) and record the CellChat/human results as
data_restricted. Running CellChat on the mouse data would have no reported value
to compare against (the paper reports no mouse CellChat counts) → no_expected_result.
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 re-analysis paper whose public-data portion reproduces well: on GSE143758 the canonical DAA population came out as a distinct AD-enriched cluster7 (~2.3x AD:WT) and the reported 12 mouse subclusters were recovered at res=0.6 (10 at the stated res=0.5) — an expected Seurat version/seed sensitivity, on our methodological side, not the authors'. The flagship results (CellChat 119/123 interactions, Fig 3C; human GFAPlowAQP4+CD63+ subtype diminishing in AD, Fig 4F p<0.001) run on controlled-access human data (syn18485175, AMP-AD) and were not attempted — a legitimate data-availability limit, not a derivability or fabrication defect. Net: solid where checkable, central conclusion untested but unimpeached; 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 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.