CoChat
Sign Up
Cited by 297Alzheimer's disease research and treatmentsDementia and Cognitive Impairment ResearchNeuroinflammation and Neurodegeneration Mechanisms

Meta-Analysis of the Alzheimer’s Disease Human Brain Transcriptome and Functional Dissection in Mouse Models

Open access
Wan, Ying-Wooi, Al-Ouran, Rami, Mangleburg, Carl G., Perumal, Thanneer M., Lee, Tom V., Allison, Katherine, Swarup, Vivek, Funk, Cory C. +56 more·2020·Cell Reports
DOI 10.1016/j.celrep.2020.107908PMC PMC7428328
Open this paper in CoChat — free accountOpen PDF

Abstract

We present a consensus atlas of the human brain transcriptome in Alzheimer's disease (AD), based on meta-analysis of differential gene expression in 2,114 postmortem samples. We discover 30 brain coexpression modules from seven regions as the major source of AD transcriptional perturbations. We next examine overlap with 251 brain differentially expressed gene sets from mouse models of AD and other neurodegenerative disorders. Human-mouse overlaps highlight responses to amyloid versus tau pathology and reveal age- and sex-dependent expression signatures for disease progression. Human coexpression modules enriched for neuronal and/or microglial genes broadly overlap with mouse models of AD, Huntington's disease, amyotrophic lateral sclerosis, and aging. Other human coexpression modules, including those implicated in proteostasis, are not activated in AD models but rather following other, unexpected genetic manipulations. Our results comprise a cross-species resource, highlighting transcriptional networks altered by human brain pathophysiology and identifying correspondences with mouse models for AD preclinical studies.

AI snapshot

Written from the paper’s full text.

TL;DR

Meta-analysis of 2,114 human postmortem brain RNA-seq samples identifies 30 AD-associated coexpression modules across seven brain regions, then compares them with 251 mouse brain expression signatures from 96 studies. It finds broad but largely non-specific overlap, especially in microglial and neuronal modules, with sex and aging effects and poor recapitulation of proteostasis modules.

Key findings

  • From 2,114 samples in 1,234 subjects, the human meta-analysis defined as many as 2,355 upregulated and 2,130 downregulated transcripts in AD brains and 30 AD-associated consensus coexpression modules across seven regions.
  • Across 96 mouse studies and 376 curated comparisons, 251 mouse DEG sets were tested against the 30 human modules; 1,569 significant overlaps were detected (padj < 0.01), 77% were concordant in direction, and 28 of 30 human modules overlapped at least one mouse signature.
  • Microglial-enriched cluster B and neuronal-enriched cluster C modules broadly overlapped APP/MAPT and other neurodegenerative/aging models, while cluster E proteostasis modules showed little overlap with AD models but overlapped unexpected manipulations such as DNMT1 knockout and Gnasxl deficiency.
  • Sex-stratified human analysis found ~2.5 times more DEGs in women than men; female rTg4510 mice showed earlier cluster C/B overlap than males, and clusters C/D were selectively enriched for female-specific DEGs while cluster E upregulated transcripts overlapped male-specific DEGs.

Methods

  • Human data: ROSMAP, MSSM, and Mayo RNA-seq cohorts totaling 2,114 samples from 1,234 subjects, including 478 AD cases and 300 controls, from seven brain regions; AD required both clinical diagnosis and neuropathologic confirmation.
  • Mouse data: 96 studies and 2,279 tissue samples were uniformly reprocessed; 376 experimental comparisons yielded 251 DEG sets after QC (gene membership >10, fold change >1.2, FDR <1%).
  • Analysis: fixed- and random-effects meta-analysis of differential expression; five coexpression algorithms plus graph clustering to define 30 consensus modules; hypergeometric tests with Benjamini-Hochberg adjustment for mouse-human module overlaps.

Limitations

  • Stated by the paper: transcriptome-proteome correlation is modest; meta-analysis may miss modest or unevenly represented gene sets; large consensus modules may limit sensitivity; human postmortem tissue is susceptible to artifact; bulk tissue mixes cell types; whole-brain profiles may obscure regional overlaps; many mouse studies were omitted.
  • Also stated: human RNA-seq can be evaluated only at death, limiting resolution of age-dependent changes or definitive links to clinical-pathologic progression.
  • Evident from the text: cross-species ortholog mapping is ambiguous when human genes map to multiple mouse orthologs; the primary analysis considered all possible pairs, with a 1:1 sensitivity test recovering 83% of significant overlaps.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1016/j.celrep.2020.107908

Cites

129 of 129 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

45 of 62 cited sources checked · 25 supported · 0 flagged · accuracy not yet measured

  • Findingown data

    Gene sets from prior AD-associated oligodendrocyte-enriched coexpression modules significantly overlapped with consensus cluster D.

    “independent analyses of RNA-seq from the Mayo and MSSM cohorts previously identified AD-associated coexpression modules enriched for oligodendrocyte expression signatures (Allen et al., 2018; McKenzie et al., 2017), and these gene sets significantly overlapped with consensus cluster D (Figure 1C).”
    Quote in the paperNot a substantial claim2 of 2 sources checked
    • cites Allen et al., 2018 source
    • cites McKenzie et al., 2017 source
  • Findingown data

    Module m109 showed significant overlap with four of five consensus clusters.

    “Interestingly, m109 showed significant overlap with four of the five consensus clusters (Figure 1C), suggesting that it may represent a coordinated brain transcriptional response involving multiple cell types.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Mostafavi et al., 2018 source

      The Results sections on constructing the module-trait network and identifying disease-associated modules describe 47 modules and identify m109, but do not report m109 overlapping four of five consensus clusters. Searches for “five consensus clusters” and “m109 overlap four clusters” found no passage addressing that specific overlap. (LLM, conf=0.75)

      Experimental — accuracy not yet measured
  • Premisecitation

    RNA sequencing of human postmortem brain tissue reinforces the importance of microglial and inflammatory mechanisms in AD pathogenesis, among other pathways.

    “Results from RNA sequencing (RNA-seq) of human postmortem brain tissue further reinforce the importance of microglial and inflammatory mechanisms in AD pathogenesis, among other pathways (Allen et al., 2018; Conway et al., 2018; McKenzie et al., 2017; Patrick et al., 2017; Wang et al., 2016, 2019; Zhang et al., 2013).”
    Quote in the paperSubstantial claim3 of 3 sources checked
    • cites Allen et al., 2018 source

      Methods §2.1 confirms that the study used RNA-seq on human postmortem brain tissue, and the Discussion reports modules enriched in microglial transcripts. But the paper’s principal RNA-seq finding concerns myelination networks; it does not establish that those results reinforce inflammatory mechanisms in AD pathogenesis. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
    • cites Conway et al., 2018 source
    • cites McKenzie et al., 2017 source

      The Discussion (page unavailable) reports that a co-expression module derived from human postmortem AD brains was enriched for immune and microglial genes, supporting the claim’s biological point alongside other pathways. But the Methods describe RNA-seq for mouse knockout experiments, not for the human postmortem analysis; they do not support calling the human result an RNA-seq finding. (LLM, conf=0.88)

      Experimental — accuracy not yet measured
  • Premisecitation

    Human genome-wide association studies support a causal role for immune mechanisms in AD.

    “human genome-wide association studies (GWASs) support a causal role for immune mechanisms (Jansen et al., 2019; Kunkle et al., 2019; Raj et al., 2014).”
    Quote in the paperSubstantial claim3 of 3 sources checked
    • cites Jansen et al., 2019 source
    • cites Kunkle et al., 2019 source
    • cites Raj et al., 2014 source
  • Premisecitation

    Prior observations in AD mouse models and human epidemiology have implicated sex as a modifier of AD manifestation and progression.

    “Prior observations in AD mouse models (Jiao et al., 2016) and human epidemiology (Altmann et al., 2014; Li and Singh, 2014; Mayeux and Stern, 2012) have implicated sex as a modifier of AD manifestation and progression.”
    Quote in the paperSubstantial claim2 of 3 sources checked
    • cites Jiao et al., 2016 source
    • cites Altmann et al., 2014 source
    • cites Li and Singh, 2014 source
  • Classified as probably not a substantial claim. 31
  • May combine two claims. 38

  • Classified as probably not a substantial claim.

    Premisecitation

    Other studies have questioned the overall degree of conservation between AD mouse models and human brains.

    “Although some studies have identified selected overlaps in expression changes between AD mouse models and human brains (Bennett et al., 2018; Castillo et al., 2017; Mostafavi et al., 2018; Neuner et al., 2019; Rojo et al., 2017), other studies have questioned the overall degree of conservation (Burns et al., 2015; Galatro et al., 2017; Hargis and Blalock, 2017).”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites Burns et al., 2015 source
    • cites Galatro et al., 2017 source

      The Discussion (p. 10) reports very little overlap between human and mouse microglial aging signatures, supporting concern about cross-species conservation. But the paper compares microglia during aging, not AD mouse models with human AD brains; it does not establish the claim at that disease-specific level. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites Hargis and Blalock, 2017 source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    Some studies have identified selected overlaps in expression changes between AD mouse models and human brains.

    “Although some studies have identified selected overlaps in expression changes between AD mouse models and human brains (Bennett et al., 2018; Castillo et al., 2017; Mostafavi et al., 2018; Neuner et al., 2019; Rojo et al., 2017), other studies have questioned the overall degree of conservation (Burns et al., 2015; Galatro et al., 2017; Hargis and Blalock, 2017).”
    Quote in the paperSubstantial claim3 of 3 sources checked
    • cites Bennett et al., 2018 source

      The Results section on the novel drug and biomarker discovery pipeline discusses human-brain omics and an ex vivo model, but does not compare expression changes in AD mouse models with human brains. Searches for mouse models, cross-species comparisons, and overlapping expression changes found no passage in our copy reporting the claimed overlap; pages were not provided. (LLM, conf=0.72)

      Experimental — accuracy not yet measured
    • cites Castillo et al., 2017 source
    • cites Mostafavi et al., 2018 source

  • May combine two claims.

    Findingcitation

    Hargis and Blalock (2017) reported that DEGs were concordant between aging in humans and rodent models, a conclusion supported by our analysis.

    “Hargis and Blalock (2017) reported that DEGs were concordant between aging in humans and rodent models, a conclusion supported by our analysis.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Hargis and Blalock, 2017 source
    • Rests on an assumption: Aging is the strongest known AD risk factor.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    Our findings are consistent with a complementary study of mouse data from gene expression arrays.

    “Our findings are consistent with a complementary study of mouse data from gene expression arrays (Hargis and Blalock, 2017).”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Hargis and Blalock, 2017 source

  • May combine two claims.

    Premisecitation

    CRND8 APP mice reveal reduced synaptic markers and hippocampal neuronal loss at 6 months, when overlaps are first detected with cluster B modules.

    “CRND8 APP mice reveal reduced synaptic markers and hippocampal neuronal loss at 6 months, when overlaps are first detected with cluster B modules (Adalbert et al., 2009; Brautigam et al., 2012).”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites Adalbert et al., 2009 source
    • cites Brautigam et al., 2012 source

  • May combine two claims.

    Findingcitation

    Expression of a mutant form of neuroserpin in mouse neural progenitor cells activates a transcriptional signature (M205) that significantly overlapped with cluster A module IF-Gyellow (padj = 8.1 × 10−17) as well as FPblue (padj = 4.7 × 10−6).

    “In another example, expression of a mutant form of neuroserpin in mouse neural progenitor cells (Guadagno et al., 2017) activates a transcriptional signature (M205) that significantly overlapped with cluster A module IF-Gyellow (padj = 8.1 × 10−17) as well as FPblue (Figure 5B; padj = 4.7 × 10−6), a cluster D module showing some selectivity for AD transgenic models (above).”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Guadagno et al., 2017 source

  • May combine two claims.

    Findingown data

    Virtually all cluster D modules failed to overlap with modules showing significant associations with neuropsychiatric disorders and were therefore "AD specific."

    “Interestingly, virtually all cluster D modules failed to overlap with modules showing significant associations with neuropsychiatric disorders (Figure S6A) and were therefore “AD specific.””
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Gandal et al., 2018 source

      The indexed copy (section labeled Abstract; page unavailable) identifies co-expression modules significantly associated with psychiatric disorders, including CD4 and CD11. It does not examine the citing paper’s cluster D modules, test their overlap with those modules, or establish that they are AD-specific. (LLM, conf=0.84)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    Loss of the Gnasxl isoform causes a hypermetabolic mouse phenotype, resulting in growth retardation, hypoglycemia, and reduced adiposity.

    “Loss of the Gnasxl isoform causes a hypermetabolic mouse phenotype, resulting in growth retardation, hypoglycemia, and reduced adiposity (Nunn et al., 2013).”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Nunn et al., 2013 source

      The New Findings section (page unavailable) says Gnasxl knockout mice are lean and hypermetabolic, supporting the hypermetabolic phenotype and broadly supporting reduced adiposity. The passages in our copy do not establish growth retardation or hypoglycemia; searches for those outcomes found no supporting passage. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Interpretationcitation

    Although cluster B changes may occur subsequent to disease onset, they are likely causal, perhaps affecting progression.

    “which, along with other studies (Sala Frigerio et al., 2019), strongly suggests that although these changes may occur subsequent to disease onset, they are likely causal, perhaps affecting progression.”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites Kunkle et al., 2019 source

      The Results (Candidate gene prioritization at genome-wide loci; page unavailable) report Alzheimer’s risk genes expressed in microglia and genes correlated with pathological stage. Searches for “cluster B,” changes subsequent to disease onset, and effects on progression found no passage addressing whether those changes are causal or when they occur. (LLM, conf=0.75)

      Experimental — accuracy not yet measured
    • cites Sala Frigerio et al., 2019 source

      The Introduction proposes that Aβ accumulation triggers pathogenic processes in which microglia play a central role, and the Discussion links AD risk genes to the microglial response to amyloid. This supports the general idea that changes following amyloid accumulation may contribute to disease. The paper does not identify the citing document’s ‘cluster B’ changes or establish that those changes affect progression after disease onset; its Discussion instead discusses a possible effect on disease onset. (LLM, conf=0.82)

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim.

    Premisecitation

    LOAD cases were defined as those with both clinical and neuropathological evidence for definitive late onset Alzheimer's disease, i.e., high burden of neurofibrillary tangles, neuritic amyloid plaques, and cognitive impairment with little evidence of other pathology.

    “The motivation was to define LOAD cases as those with both clinical and neuropathological evidence for definitive late onset Alzheimer’s disease - i.e., a high burden of neurofibrillary tangles, neuritic amyloid plaques, and cognitive impairment with little evidence of other pathology (Jack et al., 2018).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Jack et al., 2018 source

      Section 6 (Definition of AD; page unavailable) supports the amyloid-and-tau component: it says both are required for the paper’s neuropathologic criteria for AD. But it applies the AD definition independently of clinical symptoms, rather than requiring cognitive impairment. Section 9.7 discusses multiple contributing pathologies; the copy does not establish ‘little evidence of other pathology’ as a requirement for defining late-onset cases. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim.

    Premisecitation

    Ensemble inference methodologies were identified as more robust for identification of gene-gene interactions from coexpression data.

    “where ensemble inference methodologies were identified as more robust for identification of gene-gene interactions from coexpression data.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Marbach et al., 2012 source

  • May combine two claims.

    Findingown data

    The 30 modules fell into five distinct consensus clusters (A-E) that were highly preserved across study and tissue type and demonstrated minimal overlap.

    “Evaluation across tissues demonstrated that these 30 modules fell into five distinct “consensus clusters” that were highly preserved across study and tissue type and demonstrated minimal overlap (henceforth denoted clusters A–E) (Figure 1A; Figure S1).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The majority (68%) of experimental comparisons had gene expression changes corresponding to multiple human modules (mean 6 overlapping modules per mouse model DEG set).

    “Overall, we detected 1,569 significant overlaps (padj < 0.01), with the majority (68%) of experimental comparisons having gene expression changes that corresponded to multiple human modules (mean 6 overlapping modules per mouse model DEG set) (Figure 3; Table S5).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Consensus clusters were enriched for astrocytes and pericytes (A and B), endothelial cells (B), microglia (B), neurons (C and E), oligodendroglia (D), and oligodendroglial precursors (A and D).

    “Consensus clusters were enriched for astrocytes and pericytes (clusters A and B), endothelial cells (B), microglia (B), neurons (C and E), oligodendroglia (D), and oligodendroglial precursors (A and D).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingcitation

    Coincident changes in both cluster B and cluster C module genes were characteristic of selected MAPT strains and were also seen in the CDK5-P25 mouse.

    “Interestingly, coincident changes in both cluster B and cluster C module genes (Figure 4A, circles) were characteristic of selected MAPT strains and were also seen in the CDK5-P25 mouse, which similarly develops extensive tau pathology and brain atrophy (Cruz et al., 2003).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Cruz et al., 2003 source

  • May combine two claims.

    Findingown data

    Cluster B genes were consistently upregulated in human AD brain transcriptomes and mouse AD models, whereas cluster C genes were predominantly downregulated.

    “For example, cluster B genes were consistently upregulated in human AD brain transcriptomes and mouse AD models, whereas cluster C genes were predominantly downregulated.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    In TgCRND8 APP transgenic mouse, cluster B module activation was seen by 6 months and sustained at 12 months.

    “For example, in the TgCRND8 APP transgenic mouse, cluster B module activation was seen by 6 months, and this signature was sustained at 12 months.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Upregulated transcripts from male subjects selectively overlapped cluster E modules, whereas downregulated cluster E transcripts showed stronger enrichment for female DEGs.

    “Specifically, upregulated transcripts from male subjects selectively overlapped cluster E modules, whereas downregulated cluster E transcripts showed stronger enrichment for female DEGs.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Mouse models show overall poor correspondence to human disease based on brain transcriptomes, except for neuronal and microglial-enriched coexpression modules.

    “mouse models show overall poor correspondence to human disease, based on brain transcriptomes, with the exception of neuronal and microglial-enriched coexpression modules.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The C57BL6/J (B6J)-nmf205 mouse (M182) overlaps significantly with modules from consensus clusters B and C (FPturquoise, padj = 6.4 × 10−6, and PHGbrown, padj = 9.6 × 10−39, respectively), behaving similarly to MAPT transgenics and CDK-P25 mice.

    “For example, the C57BL6/J (B6J)-nmf205 mouse (M182) overlaps significantly with modules from consensus clusters B and C (FPturquoise, padj = 6.4 × 10−6, and PHGbrown, padj = 9.6 × 10−39, respectively), behaving similarly to MAPT transgenics (M230, M243) and CDK-P25 mice (M64, M65).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Consensus cluster E includes signatures for certain inhibitory and excitatory cortical neuronal subtypes and was enriched for genes that regulate proteostasis.

    “Among these, consensus cluster E includes signatures for certain inhibitory and excitatory cortical neuronal subtypes and was enriched for genes that regulate proteostasis.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Clusters A and B overlapped with DEGs from either sex, with approximately 3-fold increased enrichment for female DEGs compared with male DEGs.

    “Although clusters A and B overlapped with DEGs from either sex, approximately 3-fold increased enrichment was seen for female DEGs compared with male DEGs.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    We highlight several overlaps with transcriptomic endophenotypes from completely unexpected mouse experimental manipulations that manifest brain expression changes that mimic human AD and in some cases even overlap coexpression modules better than currently available AD mouse models.

    “we highlight several overlaps with transcriptomic endophenotypes from completely unexpected mouse experimental manipulations that manifest brain expression changes that mimic human AD and in some cases even overlap coexpression modules better than currently available AD mouse models.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductown data

    AD cases were defined by requiring both clinical diagnosis and neuropathologic confirmation (n = 478).

    “We applied a stringent definition of AD (n = 478 cases), requiring both a clinical diagnosis and neuropathologic confirmation (Table 1).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    These models recapitulate features of AD neuropathology, including plaques or tangles, along with variable neuronal dysfunction or loss and progressive behavioral impairment.

    “These models recapitulate features of AD neuropathology, including plaques or tangles, along with variable degree of neuronal dysfunction or loss and progressive behavioral impairment (Ballatore et al., 2007; Esquerda-Canals et al., 2017).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Ballatore et al., 2007 source
    • cites Esquerda-Canals et al., 2017 source

  • May combine two claims.

    Study design and conductnone given

    All control brains (n = 300) were devoid of significant AD pathology, and in two of three cohorts, control subjects were clinically confirmed as non-demented.

    “All control brains (n = 300) were devoid of significant AD pathology, and in two of the three cohorts, control subjects were clinically confirmed as non-demented.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    The most widely used transgenic models express mutant forms of APP, with or without PSEN1/2, associated with autosomal-dominant early-onset AD, or MAPT mutations, which cause familial frontotemporal dementia.

    “The most widely used transgenic models express mutant forms of the amyloid precursor protein (APP) gene, with or without presenilin-1/2 (PSEN1/2), which are associated with autosomal-dominant, early-onset AD, or, alternatively, MAPT mutations, which cause familial frontotemporal dementia (FTD).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Ballatore et al., 2007 source
    • cites Esquerda-Canals et al., 2017 source

  • May combine two claims.

    Interpretationown data

    The poor correspondence of mouse models to human disease is an important caveat for interpreting studies using these animal models and may explain in part their poor predictive power as preclinical models for AD.

    “This is an important caveat for the interpretation of studies using these animal models and may explain in part their poor predictive power as preclinical models for AD.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    These patterns may not be specific for neurodegenerative disease, but rather may accompany brain aging more generally.

    “Overall, our results suggest that these patterns may not be specific for neurodegenerative disease, but rather may accompany brain aging more generally.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    These signatures are consistently observed across multiple cohorts and several brain regions and were identified by multiple independent coexpression algorithms.

    “These signatures are consistently observed across multiple cohorts and several brain regions and were identified by multiple independent coexpression algorithms.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    It is possible that we could miss relevant gene sets with more modest coexpression or uneven representation across studies because of sample ascertainment differences.

    “it is possible that we could miss relevant gene sets with more modest coexpression or uneven representation across studies because of sample ascertainment differences.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductnone given

    Human brain RNA-seq can be evaluated only at the time of death, creating significant challenges to resolve age-dependent changes or definitively establish links with clinical-pathologic progression.

    “As human brain RNA-seq can be evaluated only at the time of death, there are significant challenges to resolve age-dependent changes or to definitively establish links with clinical-pathologic progression.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Autosomal-dominant mutations in neuroserpin cause familial encephalopathy with Neuroserpin inclusion bodies, a rare, early-onset neurodegenerative dementia caused by protein aggregation within the ER and associated ER stress.

    “Autosomal-dominant mutations in neuroserpin cause familial encephalopathy with Neuroserpin inclusion bodies, a rare, early-onset neurodegenerative dementia caused by protein aggregation within the endoplasmic reticulum (ER) and associated ER stress (Roussel et al., 2013).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Roussel et al., 2013 source

  • May combine two claims.

    Interpretationown data

    Consensus clusters B and C account for only 14 of 30 coexpression modules (47%) and do not appear to be specific for AD pathophysiology.

    “However, consensus clusters B and C account for only 14 of 30 coexpression modules (47%) and do not appear to be specific for AD pathophysiology.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    Our meta-analytic approach was designed to identify signatures of disease that are robust to technical and study-specific heterogeneity.

    “although our meta-analytic approach was designed to identify signatures of disease that are robust to technical and study-specific heterogeneity”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductreasoning

    To avoid technical variabilities arising from RNA-seq alignment and quantification, we reprocessed and realigned all RNA-Seq reads from the source studies.

    “To avoid some of the technical variabilities arising due to RNA-seq alignment and quantification, and also to account for some of the technical variabilities we reprocessed and realigned all the RNA-Seq reads from the source studies”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    We applied 5 distinct network module identification methodologies to each of the 7 tissue-specific expression datasets.

    “We apply 5 distinct network module identification methodologies to each of the 7 tissue specific expression datasets.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    We selected nearly 100 independent mouse RNA-seq studies for inclusion in our analyses, prioritizing those most relevant to AD and neurodegeneration.

    “although we selected nearly 100 independent mouse RNA-seq studies for inclusion in our analyses, prioritizing those most relevant to AD and neurodegeneration”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductown data

    We required a minimum of n = 2 samples for each group (experimental and control); the average for all samples included in each comparison was n = 8.4 (range = 4-28 total samples).

    “We required a minimum of n = 2 samples for each group (experimental and control); the average for all samples included in each comparison was n = 8.4 (range = 4-28 total samples).”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductown data

    P values were adjusted for multiple hypothesis testing using false discovery rate (FDR) estimation, and differentially expressed genes were determined as those with an estimated FDR below, or at, 5% with a corresponding absolute expression and fold-change cutoffs.

    “P values were then adjusted for multiple hypothesis testing using false discovery rate (FDR) estimation, and the differentially expressed genes were determined as those with an estimated FDR below, or at, 5% with a corresponding absolute expression and fold-change cutoffs.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductreasoning

    Two of the three studies (MSBB and Mayo RNaseq) obtained more than one tissue from the same donors; therefore, except ROSMAP, donor-specific effects were explicitly modeled as random effects.

    “Two of the three studies - MSBB and Mayo RNaseq - obtained more than one tissue from the same donors. Therefore, except the ROSMAP study, donor-specific effects were explicitly modeled as random effects.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    The cell type-specific enrichment of human coexpression modules may reflect global changes in cell proportions, such as neuronal loss or microgliosis.

    “which may therefore reflect global changes in cell proportions, such as neuronal loss or microgliosis.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductreasoning

    All p values reported in the text were adjusted for multiple comparisons using the Benjamini-Hochberg method.

    “The Benjamini-Hochberg method was applied to adjust for multiple comparisons, using the p.adjust function.”
    • Also stated in: § QUANTIFICATION AND STATISTICAL ANALYSIS · plain
      “All p values reported in the text were adjusted in this manner.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductreasoning

    We harmonized the LOAD definition across three studies to compare analysis results and understand LOAD biology across different tissues.

    “To compare analysis results across studies and to get an understanding of LOAD biology across different tissues, we harmonized the LOAD definition across three studies.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductreasoning

    Fisher's exact test was used to test enrichment of each gene set with the annotated set, and resulting p values were corrected independently for each set using Benjamini-Hochberg method.

    “Fisher’s exact test was used to test enrichment of each gene set with the annotated set. Resulting p values were corrected independently for each set using Benjamini-Hochberg method for significance testing”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    The combined human cohort consisted of 2,114 total samples from 1,234 subjects, including 478 AD cases, 300 controls without AD pathology, and 456 subjects with progressive supranuclear palsy or other neurodegenerative pathologies.

    “the combined cohort consisted of 2,114 total samples from 1,234 subjects (Table 1; Table S1), including 478 AD cases, 300 controls (without AD pathology), and 456 subjects with progressive supranuclear palsy or other neurodegenerative pathologies.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    All mouse RNA-seq data were reprocessed using a single, complementary pipeline to facilitate cross-species and cross-model comparisons.

    “All mouse RNA-seq data were reprocessed using a single, complementary pipeline to facilitate cross-species and cross-model comparisons.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    A secondary sensitivity analysis was performed by restricting to human-mouse gene ortholog pairs with 1:1 relationships.

    “In a secondary analysis (Table S6), we performed a sensitivity test by restricting to human-mouse gene ortholog pairs with 1:1 relationships.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    We omitted many other mouse RNA-seq studies with the potential to provide additional insights.

    “we omitted many others with the potential to provide additional insights.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductown data

    We excluded all DEG sets consisting of fewer than 10 conserved mouse genes, resulting in 251 sets of DEGs for consideration in subsequent analyses.

    “We excluded all DEG sets from consideration consisting of fewer than 10 conserved mouse genes, resulting in 251 sets of DEGs for consideration in our subsequent analyses.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisereasoning

    Controls were defined as patients with a low burden of plaques and tangles, as well as no evidence of cognitive impairment if available.

    “Controls were concordantly defined as patients with a low burden of plaques and tangles, as well as no evidence of cognitive impairment if available.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    RNA-seq profiles from whole brain might obscure more localized transcriptome overlaps because the extent and tempo of neurodegeneration can vary widely across different brain regions.

    “as the extent and tempo of neurodegeneration, including in both human and mouse models, can vary widely across different brain regions, RNA-seq profiles from whole brain might obscure more localized transcriptome overlaps.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The extent and tempo of neurodegeneration can vary widely across different brain regions.

  • May combine two claims.

    Study design and conductnone given

    Human RNA-seq data, along with the majority of included mouse studies, derive from bulk brain tissue, which includes mixed cell types.

    “the human RNA-seq data, along with the majority of included mouse studies, derive from bulk brain tissue, which includes mixed cell types.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Study design and conductnone given

    Extraction and processing of human brain tissue is likely more susceptible to postmortem artifact compared with laboratory preparation of mouse mRNA.

    “Compared with the laboratory preparation of mouse mRNA, the extraction and processing of human brain tissue is likely more susceptible to postmortem artifact”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    The reprocessing was done using a consensus set of tools with only library type-specific parameters varying between pipelines.

    “The reprocessing was done using a consensus set of tools with only library type-specific parameters varying between pipelines.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    A unified RNA-seq analysis pipeline was used for reprocessing of all mouse datasets, with the exception of 2 HD studies where count files were downloaded directly from GEO.

    “A unified RNA-seq analysis pipeline was used for reprocessing of all datasets, with the exception of 2 HD studies where count files were downloaded directly from GEO.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductreasoning

    Sub-samples were selected to harmonize the late onset AD (LOAD) case-control definition across the three studies for all differential expression analyses.

    “Sub-samples were selected to harmonize the late onset AD (LOAD) case - control definition across the three studies for all differential expression analyses.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    The strengths of our study include coexpression modules based on more than 2,000 human brain samples and consideration of a large and diverse number of mouse studies.

    “The strengths of our study include coexpression modules based on more than 2,000 human brain samples and consideration of a large and diverse number of mouse studies.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductreasoning

    To account for differences between samples, studies, experimental batch effects and unwanted RNA-Seq specific technical variations, we performed library normalization and covariate adjustments for each study separately using fixed/mixed effects modeling.

    “To account for differences between samples, studies, experimental batch effects and unwanted RNA-Seq specific technical variations we performed library normalization and covariate adjustments for each study separately using fixed/mixed effects modeling.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductreasoning

    The random effect model was selected as a conservative approach to correct for variation across studies.

    “The random effect model was selected as a conservative approach to correct for variation across studies.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Our analyses adjusted for sample variability in postmortem interval.

    “although our analyses adjusted for sample variability in postmortem interval.”
    Quote in the paperNot a substantial claim
61 of 61

  • The paper relies on this without support.

    The mechanisms by which the AD pathologic cascade is triggered and propagated remain incompletely defined.

    “The mechanisms by which this AD pathologic cascade is triggered and propagated remain incompletely defined and are likely influenced by heterogeneous and dynamic perturbations involving multiple biological pathways.”

  • The paper relies on this without support.

    Transcript abundance is highly influenced by multiple sources of technical and biological variation.

    “Because transcript abundance is highly influenced by multiple sources of technical and biological variation, we first evaluated the consistency and robustness of findings across three independent human brain transcriptome studies”

  • The paper relies on this without support.

    Mouse models allow controlled experimental manipulations that can be used to establish causation, isolate the effects of specific molecular lesions, and investigate dynamic, age-dependent changes.

    “Compared with transcriptome profiles in human brains, mouse models allow controlled experimental manipulations that can be used to establish causation, isolate the effects of specific molecular lesions, and investigate for dynamic, age-dependent changes.”

  • The paper relies on this without support.

    The AMP-AD Consortium has generated RNA-seq profiles from more than 1,200 human brains.

    “The Accelerating Medicines Partnership-Alzheimer’s Disease (AMP-AD) Consortium has generated RNA-seq profiles from more than 1,200 human brains and is applying systems biology approaches toward the goal of elucidating AD mechanisms and potential therapeutic targets.”

  • The paper relies on this without support.

    Aging is the strongest known AD risk factor.

    “Aging is the strongest known AD risk factor.”

  • The paper relies on this without support.

    The extent and tempo of neurodegeneration can vary widely across different brain regions.

    “as the extent and tempo of neurodegeneration, including in both human and mouse models, can vary widely across different brain regions”
6 of 6

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • These modules were grouped into five consensus clusters and annotated using Reactome, representing conserved AD-related molecular signatures across cohorts and brain regions.

    “These modules were grouped into five consensus clusters and annotated using Reactome, representing conserved AD‐related molecular signatures across cohorts and brain regions. 37 , 41 , 51”

    From Unique transcriptomic alterations in 5XFAD;PS19 mouse model identify glial lipid dysregulation and coordinated microglial–oligodendrocyte responses (2026) · cites this paper as [37]

    The Results section, “A Consensus Atlas of AD-Associated Transcriptional Modules from Human Brain,” says that 30 AD-associated modules formed five consensus clusters preserved across studies and brain regions. It also reports enrichment for biological processes, but the Methods section lists several annotation resources without saying that the clusters were annotated using Reactome; the only explicit Reactome mention is in the title of a cited PANTHER paper. (LLM, conf=0.88)

    Experimental — accuracy not yet measured

  • The AMP-AD study from Wan et al. performed network analysis separately from each brain region.

    “The AMP-AD study from Wan et al.63 performed network analysis separately from each brain region, while, in our previous study,56 we performed consensus network analysis across the different brain regions.”

    From hdWGCNA identifies co-expression networks in high-dimensional transcriptomics data (2023) · cites this paper as [63]

  • PHGgreen was enriched with oligodendrocyte- and myelin-associated genes (Mog, Mag, Cldn11, and Ugt8a), consistent with findings from Wan et al.

    “Notably, PHGgreen was enriched with oligodendrocyte‐ and myelin‐associated genes (Mog, Mag, Cldn11, and Ugt8a; Figure 6F), consistent with findings from Wan et al. 37”

    From Unique transcriptomic alterations in 5XFAD;PS19 mouse model identify glial lipid dysregulation and coordinated microglial–oligodendrocyte responses (2026) · cites this paper as [37]

  • A previously published analysis of AMP-AD data identified 30 co-expression modules significantly enriched for DEGs between AD patients and controls.

    “We leveraged a previously published analysis of the AMP‐AD data, which identified 30 co‐expression modules significantly enriched for DEGs between AD patients and controls. 37”

    From Unique transcriptomic alterations in 5XFAD;PS19 mouse model identify glial lipid dysregulation and coordinated microglial–oligodendrocyte responses (2026) · cites this paper as [37]

  • Among 26 mouse transcriptomic cohorts where Abi3 and Gngt2 were both differentially regulated, their expression changes were concordant in 24 studies.

    “In addition, we investigated 96 additional mouse transcriptomic datasets [33, 34] and in 26 of these cohorts, we found that Abi3 and Gngt2 were both differentially regulated. Among these 26 cohorts, Abi3 and Gngt2 expression changes were concordant in 24 studies (Additional File 6: Table S4), showing that these genes are consistently co-regulated across different mouse models and experimental”

    From Deletion of Abi3/Gngt2 influences age-progressive amyloid β and tau pathologies in distinctive ways (2022) · cites this paper as [33]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Mouse DEG-set size threshold conflicts with the reported range

    Minor

    The results state that gene sets were retained only if they had more than 10 genes, but give a minimum size of 10. The methods instead say that sets with fewer than 10 genes were excluded, which permits sets of exactly 10 genes.

1 of 1

Cites

Flagged first, then by citations

7 of the 136 references couldn’t be matched to a known paper.

Cited by

Sorted by citations

Related

Authors
Wan, Y.-W., Al-Ouran, R., Mangleburg, C. G., Perumal, T. M., Lee, T. V., Allison, K., Swarup, V., Funk, C. C., Gaiteri, C., Allen, M., Wang, M., Neuner, S. M., Kaczorowski, C. C., Philip, V. M., Howell, G. R., Martini-Stoica, H., Zheng, H., Mei, H., Zhong, X., Kim, J. W., Dawson, V. L., Dawson, T. M., Pao, P.-C., Tsai, L.-H., Haure-Mirande, J.-V., Ehrlich, M. E., Chakrabarty, P., Levites, Y., Wang, X., Dammer, E. B., Srivastava, G., Mukherjee, S., Sieberts, S. K., Omberg, L., Dang, K. D., Eddy, J. A., Snyder, P., Chae, Y., Amberkar, S., Wei, W., Hide, W., Preuss, C., Ergun, A., Ebert, P. J., Airey, D. C., Mostafavi, S., Yu, L., Klein, H.-U., Carter, G. W., Collier, D. A., Golde, T. E., Levey, A. I., Bennett, D. A., Estrada, K., Townsend, T. M., Zhang, B., Schadt, E., De Jager, P. L., Price, N. D., Ertekin-Taner, N., Liu, Z., Shulman, J. M., Mangravite, L. M., & Logsdon, B. A.
Affiliations
  • Baylor College of Medicine (US)
  • Sage Bionetworks (US)
  • University of California, Irvine (US)
  • Institute for Systems Biology (US)
  • Rush University Medical Center (US)
  • Mayo Clinic in Florida (US)
  • Icahn School of Medicine at Mount Sinai (US)
  • Jackson Laboratory (US)
  • GlaxoSmithKline (China) (CN)
  • Johns Hopkins University (US)
  • Broad Institute (US)
  • University of Florida (US)
  • Mayo Clinic (US)
  • Emory University (US)
  • AbbVie (United States) (US)
  • University of Manchester (GB)
  • Durham University (GB)
  • Beth Israel Deaconess Medical Center (US)
  • Eli Lilly (United States) (US)
  • University of British Columbia (CA)
  • Eli Lilly (United Kingdom) (GB)
  • Biogen (United States) (US)
Corresponding author
  • Liu, Zhandong
  • Shulman, Joshua M.
  • Mangravite, Lara M.
  • Logsdon, Benjamin A.
ORCID
Venue
Cell Reports
Publisher
Elsevier BV
Published
2020-07
Volume
32
Issue
2
Pages
107908
ISSN
2211-1247 (ISSN-L)
DOI
10.1016/j.celrep.2020.107908
PMC
PMC7428328
OpenAlex
W3043048574
Semantic Scholar
f816457288d6c6c6dd6024a992a8383a1e4cc92a
Cited by
295
References
136
Citations by year
2026: 35 · 2025: 32 · 2024: 42 · 2023: 59 · 2022: 59 · 2021: 55 · 2020: 13
FWCI
8.08
Type
journal article
Language
en
Open access
Licence
Locations
  • Cell Reports (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Durham Research Online (Durham University) (acceptedVersion, other-oa)
  • Deep Blue (University of Michigan) (submittedVersion, cc-by)
  • PubMed Central (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • DSpace@MIT (Massachusetts Institute of Technology) (submittedVersion, cc-by-nc)
  • Durham Research Online (Durham University) (submittedVersion, cc-by)
  • The Mouseion at the JAXlibrary (Jackson Laboratory) (submittedVersion, cc-by)
Funders
  • Division of Mathematical Sciences (DMS-1263932)
  • Burroughs Wellcome Fund
  • Michael J. Fox Foundation for Parkinson&apos;s Research
  • Mayo Clinic
  • Cullen Foundation
  • National Science Foundation
  • Translational Genomics Research Institute (P50 AG016574, R01 AG003949, R01 AG017216, R01 AG018023, R01 AG025711, R01 AG032990, U01 AG006576, U01 AG006786, U01 AG046139)
  • National Institutes of Health (F31AG050357, P50 NS38377, R01AG050631, R01AG053960, R01AG057339, R01AG057473, R01AG057914, RF1 AG057440, RF1AG054014, U01AG046161)
  • Alzheimer&apos;s Disease Research Center, University of Pittsburgh
  • National Institute on Aging (P30AG10161, R01AG15819, R01AG17917, R01AG30146, R01AG36836, U01AG32984, U01AG46152)
  • Veterans Administration Medical Center
  • National Institute of Neurological Disorders and Stroke (R01 NS080820)
  • Johns Hopkins Hospital
  • Huffington Foundation
  • JPB Foundation
  • Boler Family Foundation
  • Arizona Biomedical Research Commission (0011, 05-901, 1001, 4001)
  • School of Medicine, Johns Hopkins University
  • Arizona Department of Health Services (211002)
  • Illinois Department of Public Health
  • Adrienne Helis Malvin Medical Research Foundation
  • Mayo Foundation for Medical Education and Research (U24 NS072026)
  • Texas Children&apos;s Hospital
  • Icahn School of Medicine at Mount Sinai (P50 AG025688, U01 AG046161, U01 AG061357, U01AG046170)
  • Cancer Prevention and Research Institute of Texas (RP170387)
Topics
  • Health Sciences › Medicine › Physiology › Alzheimer's disease research and treatments
  • Health Sciences › Medicine › Psychiatry and Mental health › Dementia and Cognitive Impairment Research
  • Life Sciences › Neuroscience › Neurology › Neuroinflammation and Neurodegeneration Mechanisms
Keywords
Alzheimer's disease, human brain transcriptome, gene co-expression modules, differential gene expression, mouse models of Alzheimer's disease, amyloid pathology, tau pathology, microglia, neuronal genes, proteostasis, Huntington's disease, amyotrophic lateral sclerosis, neurodegeneration, aging
MeSH terms
Alzheimer Disease, Animals, Brain, Disease Models, Animal, Female, Gene Expression Regulation, Humans, Male, Sex Characteristics, Species Specificity, Transcription, Genetic, Case-Control Studies, Gene Expression Profiling, Mice, Gene Regulatory Networks, Transcriptome

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-14.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-14.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review