Tau polarizes an aging transcriptional signature to excitatory neurons and glia

  1. Timothy Wu
  2. Jennifer M Deger
  3. Hui Ye
  4. Caiwei Guo
  5. Justin Dhindsa
  6. Brandon T Pekarek
  7. Rami Al-Ouran
  8. Zhandong Liu
  9. Ismael Al-Ramahi
  10. Juan Botas
  11. Joshua M Shulman  Is a corresponding author
  1. Jan and Dan Duncan Neurological Research Institute, Texas Children’s Hospital, United States
  2. Medical Scientist Training Program, Baylor College of Medicine, United States
  3. Department of Molecular and Human Genetics, Baylor College of Medicine, United States
  4. Department of Neuroscience, Baylor College of Medicine, United States
  5. Department of Neurology, Baylor College of Medicine, United States
  6. Department of Pediatrics, Baylor College of Medicine, United States
  7. Center for Alzheimer’s and Neurodegenerative Diseases, Baylor College of Medicine, United States

Abstract

Aging is a major risk factor for Alzheimer’s disease (AD), and cell-type vulnerability underlies its characteristic clinical manifestations. We have performed longitudinal, single-cell RNA-sequencing in Drosophila with pan-neuronal expression of human tau, which forms AD neurofibrillary tangle pathology. Whereas tau- and aging-induced gene expression strongly overlap (93%), they differ in the affected cell types. In contrast to the broad impact of aging, tau-triggered changes are strongly polarized to excitatory neurons and glia. Further, tau can either activate or suppress innate immune gene expression signatures in a cell-type-specific manner. Integration of cellular abundance and gene expression pinpoints nuclear factor kappa B signaling in neurons as a marker for cellular vulnerability. We also highlight the conservation of cell-type-specific transcriptional patterns between Drosophila and human postmortem brain tissue. Overall, our results create a resource for dissection of dynamic, age-dependent gene expression changes at cellular resolution in a genetically tractable model of tauopathy.

Editor's evaluation

Wu et al. have provided a revised manuscript that presents important new findings that start to explain cell type vulnerability and the types of transcriptional changes that occur in the context of neurodegenerative diseases. They cleverly use Drosophila for this as they have access to numerous brain cells and exquisite genetic control. They present compelling evidence of transcriptional deregulation and affected pathways in relation to Tau toxicity in a well-controlled study. They also tested if affected pathways modify toxicity but were not successful, however, as pointed out, this can have different reasons. This paper is of broad interest to those in the field of neurodegeneration and neuronal disease and from a methodological point of view to single-cell biologists.

https://doi.org/10.7554/eLife.85251.sa0

Introduction

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder characterized by extracellular amyloid-beta neuritic plaques and intracellular tau neurofibrillary tangles (DeTure and Dickson, 2019; Scheltens et al., 2021). Tau neuropathological burden is strongly correlated with cognitive decline, synaptic loss, and neuronal death (Arriagada et al., 1992; Braak and Braak, 1991; Gómez-Isla et al., 1997). Cell-type-specific vulnerability is also an important driver of AD clinical manifestations, including its characteristic amnestic syndrome. Neurofibrillary tangles first appear in the transentorhinal cortex, entorhinal cortex, and CA1 region of the hippocampus, affecting resident pyramidal cells and excitatory glutamatergic neurons; cholinergic neurons of the basal forebrain are also particularly vulnerable (Mrdjen et al., 2019; Fu et al., 2018). Single-cell RNA-sequencing (scRNAseq) or single-nucleus RNA-sequencing (snRNAseq) are promising approaches to pinpoint cell-type-specific mechanisms in AD, including those that may underlie neuronal vulnerability (Mathys et al., 2019; Grubman et al., 2019; Lau et al., 2020; Zhou et al., 2020). Emerging data highlight altered transcriptional states and/or cell proportions for vulnerable versus resilient neurons, including excitatory or inhibitory neurons, respectively (Leng et al., 2021). snRNAseq profiles also implicate important roles for non-neuronal cells, including oligodendrocytes, astrocytes, and microglia (Grubman et al., 2019; Lau et al., 2020; Zhou et al., 2020). Microglial expression signatures, including genes with roles in innate immunity, are sharply increased in brains with AD pathology, and an important causal role in AD risk and pathogenesis is reinforced by findings from human genetics (Bohlen et al., 2019; Deczkowska et al., 2018; Bellenguez et al., 2022).

One important limitation to gene expression studies from human postmortem tissue is that only cross-sectional analysis is possible, making it difficult to reconstruct dynamic changes over the full time course of disease. In fact, age is the most important risk factor for AD, which develops over decades (Masters et al., 2015; Villemagne et al., 2013). Another potential challenge is identifying molecularly specific changes since tau tangle pathology usually co-occurs with amyloid-beta plaques, along with other brain pathologies that can also cause dementia (e.g., Lewy bodies or infarcts) (Kapasi et al., 2017). By contrast, animal models permit experimentally controlled manipulations isolating specific triggers and their impact over time. For example, in mouse models of amyloid-beta pathology, scRNAseq and snRNAseq have implicated subpopulations of disease-associated microglia and astrocytes, and similar changes may also characterize brain aging (Keren-Shaul et al., 2017; Habib et al., 2020). Further, in tau transgenic models, activation of immune signaling by the nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB) transcription factor within microglia was found to be an important driver of pathological progression (Wang et al., 2022). We recently characterized tau- and aging-induced gene expression changes in a Drosophila melanogaster tauopathy model, revealing perturbations in many conserved pathways such as innate immune signaling (Mangleburg et al., 2020). Over 70% of tau-induced gene expression changes in flies were also observed in normal aging. In this study, we deploy scRNAseq in Drosophila to map the cell-specific contributions of age- and tau-driven brain gene expression and identify NFκB signaling as a promising marker of neuronal vulnerability.

Results

Single-cell transcriptome profiles of the tau transgenic Drosophila brain

Pan-neuronal expression of either wildtype or mutant forms of the human microtubule-associated protein tau (MAPT) gene in Drosophila recapitulates key features of AD and other tauopathies, including misfolded and hyperphosphorylated tau, age-dependent synaptic and neuron loss, and reduced survival (Wittmann et al., 2001). We performed scRNAseq of adult fly brains in tauR406W transgenic Drosophila (elav>tauR406W) and controls (elav-GAL4), including animals aged 1, 10, or 20 days (Figure 1—figure supplement 1A and B). The GAL4-UAS expression system is used to express human tau in neurons throughout the central nervous system (CNS) (Brand and Perrimon, 1993). The R406W variant in MAPT causes frontotemporal dementia with parkinsonism-17, an autosomal-dominant, neurodegenerative disorder with tau pathology (i.e., tauopathy). In flies, wild type and mutant forms of tau share conserved neurotoxic mechanisms and cause similar neurodegenerative phenotypes, but tauR406W induces a more robust transcriptional response and accelerated course (Wittmann et al., 2001; Bardai et al., 2018; Mangleburg et al., 2020). Following stringent quality control, transcriptome data from 48,111 single cells were available for our initial analyses, including from 6 total conditions (2 genotypes × 3 ages) (Figure 1—figure supplement 1C and E). In the integrated dataset, we identified 96 distinct cell clusters grouped by transcriptional signatures, and annotated cell-type identities to 59 clusters using available Drosophila brain scRNAseq reference data and established cell markers (Figure 1A, Figure 1—figure supplement 2, Figure 1—source data 1). As expected, most cells in the fly brain were neurons (CadN expression, n = 42,587), whereas glia were comparatively sparse (repo expression, n = 5524). Our dataset comprises a diverse range of cell types. Among all cell clusters, 49% were cholinergic neurons (VAChT), 20% were glutamatergic neurons (VGlut), 11% were GABAergic neurons (Gad1), and 7% were glia (repo, Gs2) (Figure 1B, Figure 1—figure supplement 3). We also identified several major glial subtypes in the fly brain (Kremer et al., 2017), including astrocyte-like, cortex, chiasm, subperineurial, perineurial, and ensheathing glia, along with a group of circulating macrophages (hemocytes). Overall, our findings are consistent with results from prior scRNAseq studies of whole adult Drosophila brains (Davie et al., 2018).

Figure 1 with 3 supplements see all
Single-cell RNA-sequencing of the adult Drosophila brain.

(A) Uniform manifold approximation and projection (UMAP) plot displays unsupervised clustering of 48,111 cells, including from control (elav-GAL4/+) and elav>tauR406W transgenic animals (elav-GAL4/+; UAS-tauR406W/+) at 1, 10, and 20 days. Expression of neuron- and glia-specific marker genes, CadN and repo, respectively, is also shown. Cell cluster annotations identify heterogeneous optic lobe neuron types, including from the lamina (L1-5, T1, C2/3, Lawf, Lai), medulla (Tm/TmY, Mi, Dm, Pm, T2/3), and lobula (T4/T5, LC). Other identified neuron types include photoreceptors (ninaC, eya), dopaminergic neurons (DAT, Vmat, ple), and central brain mushroom body Kenyon cells (ey, Imp, sNFP, trio). (B) Violin plot showing cell-type marker expression across annotated cell clusters. Selected markers include Elav (neurons), repo/Gs2 (glia), Gad1 (GABA), VGlut (glutamate), VAChT (acetylcholine), and DAT/Vmat/ple (dopamine). See also Figure 1—figure supplements 13 and Figure 1—source data 1–4.

Figure 1—source data 1

Drosophila scRNAseq cell cluster annotations.

Cluster refers to the numeric ID assigned by Seurat when FindClusters resolution is set to 2, and the Annotation column notes the cell identity assignment. This table can be used to obtain cell identities of Seurat cluster IDs in the result tables below.

https://cdn.elifesciences.org/articles/85251/elife-85251-fig1-data1-v2.xlsx
Figure 1—source data 2

Cell cluster markers.

Cluster markers are obtained by MAST differential expression analysis where each cell cluster is compared against all remaining cells. Only genes with a positive log2 fold change are displayed. Log2 fold change = expression fold change between a given cluster and all remaining cells. Pct.1 = percent of cells in the given cluster with non-zero expression of the gene. Pct.2 = percent of the remaining cells (not in the cluster) that have non-zero expression of the gene. Cluster ID = Seurat assigned cluster ID. BH-adjusted p-value = Benjamini–Hochberg-corrected p-value from the MAST differential expression analysis for each cluster.

https://cdn.elifesciences.org/articles/85251/elife-85251-fig1-data2-v2.xlsx
Figure 1—source data 3

Single-cell RNA-sequencing quality control parameters.

Cell library metrics from the 10x Genomics Cell Ranger output. Sample = cell library labels. Libraries from the replication experiment are labeled with ‘rep’ behind the final underscore. See also Figure 1—figure supplement 1. Additional details on the data provided in each column can be found in 10x Genomics support materials: https://support.10xgenomics.com/single-cell-gene-expression/software/pipelines/latest/output/gex-metrics.

https://cdn.elifesciences.org/articles/85251/elife-85251-fig1-data3-v2.xlsx
Figure 1—source data 4

Drosophila cell-type expression markers.

Table of established fly gene expression markers used as references for cell annotation. Genes = marker genes that can be used to identify a cell type. Cell type = Drosophila brain cell subpopulation. Reference = source publication used to obtain the genes.

https://cdn.elifesciences.org/articles/85251/elife-85251-fig1-data4-v2.xlsx

Tau drives changes in cell proportions in the brain

Leveraging our scRNAseq data and pooling longitudinal samples to permit robust comparisons, we first assessed how tau affects the relative abundance of cell-type subpopulations in the adult brain. We found 16 neuronal and 6 glial clusters with statistically significant changes in cell abundance when comparing tau and controls (Figure 2A and B, Figure 2—source data 1). Cholinergic mushroom body Kenyon cell neurons in the central complex, which are important in learning and memory, were sharply reduced, likely consistent with developmental toxicity of tau, as noted in prior studies of Drosophila tauopathy models (Mershin et al., 2004; Kosmidis et al., 2010). In fact, seven excitatory neuronal clusters, including several cholinergic and glutamatergic cell types, demonstrated significant declines, whereas inhibitory neuronal subpopulations (e.g., Pm and Mi4 GABAergic cells in the visual system) appeared resilient. Conversely, cluster 12 cells appeared more abundant in tau flies; this non-annotated cell type was enriched for neuroendocrine expression markers, Ms and Hug, as well as a regulator of synaptic plasticity, Arc1 (Figure 1—source data 2). Interestingly, several glial cell types also appeared increased in the brains of tau animals. Ensheathing glia, which showed the largest potential increase, are localized to neuropil in the fly brain and mediate phagocytosis following neuronal injury (Doherty et al., 2009; Freeman, 2015). In order to confirm these observations, which were based on pooled data across timepoints, we generated additional scRNAseq profiles from 10-day-old elav>tauR406W and control flies in triplicate samples (69,128 cells; Figure 2—figure supplement 1). Overall, 13 out of the 22 significant cell abundance changes were also observed in this replication dataset, including the sharp reduction of excitatory neurons (e.g., Kenyon cells), and the increase in multiple glial clusters (e.g., ensheathing glia) (Figure 2—figure supplement 1B, Figure 2—source data 1). Non-replicated changes in cell-type abundance may be driven by data from earlier (1 day) or later (20 day) timepoints (Figure 2B). Although our experimental design limits cross-sectional analyses at 1 and 20 days, the observed changes in cell abundance were suggestive of a combination of both developmental tau toxicity and progressive, age-dependent neurodegeneration (e.g., neuronal clusters 1, 9, and 12, and astrocyte-like glia). Selected cell-type proportion changes were also recapitulated based on computational deconvolution of available bulk-tissue RNAseq from tauR406W and control flies at 1, 10, and 20 days by using an independent, published scRNAseq reference dataset (Figure 2—figure supplement 2).

Figure 2 with 3 supplements see all
Tau-triggered cell proportion changes in the adult brain.

(A) Log2-fold change (log2FC) of normalized cell counts between elav>tauR406W (elav-GAL4/+; UAS-tauR406W/+) and control (elav-GAL4/+) animals. Timepoints are pooled for each cluster. Cell clusters with statistically significant changes (false discovery rate [FDR] < 0.05) are highlighted in black. Many of these cell abundance changes were replicated in an independent dataset generated from 10-day-old animals (Figure 2—figure supplement 1). Since cell-type abundance estimates are relative between clusters, we also performed an adjusted analysis in which glia were assumed to be unchanged (Figure 2—figure supplement 3A). (B) Plots highlight cluster cell counts with significant differences based on pooled timepoint comparisons between elav>tauR406W (red) and control (black) animals, including results for samples collected at 1 day (triangle), 10 days (cross-hatch square), or 20 days (filled square). See Figure 2—figure supplement 2 for complementary analysis based on deconvolution of bulk brain RNA-sequencing. (C) Whole-mount immunofluorescence of adult brains from 10-day-old flies. Glia are stained using the Anti-Repo antibody (red) in control (elav-GAL4/+) and elav>tauR406W transgenic flies. Full Z-stack projection is shown. Scale bar = 100 microns. See also Figure 2—figure supplement 3B for additional immunostains for nuclei and actin. (D) Quantification of glia (Repo-positive puncta), brain volume, and glial density is shown. Statistical analysis employed Welch’s T-test with n=9 animals per group and significance threshold p < 0.05. Error bars denote the 95% confidence interval. See also Figure 2—figure supplements 13 and Figure 2—source data 1.

Figure 2—source data 1

Tau-triggered cell proportion changes.

Analysis of cell abundance changes between elav>tauR406W and control animals as quantified by DESeq2. In the discovery dataset, the 1, 10, and 20-day timepoints are pooled, such that n = 3 values for each comparison. The replication dataset is comprised of n = 3 elav>tauR406W and control (elav-GAL4) animals all prepared at day 10. baseMean = mean of normalized cell counts for the given cell cluster across all samples. Log2FoldChange = log2 fold change of elav>tauR406W vs. control cell counts. lfcSE = standard error of log2 fold change value. Pvalue = p-value from Wald test of the genotype log2 fold change value. Differences in cell count are quantified by negative binomial GLM, such that count ~genotype + age. Padj = adjusted p values using the Benjamini–Hochberg procedure. Experiment = denotes if data is from the discovery or replication analysis.

https://cdn.elifesciences.org/articles/85251/elife-85251-fig2-data1-v2.xlsx

Similar to our Drosophila tauopathy model, snRNAseq from postmortem human brain tissue has consistently suggested AD-associated increases in glial cell abundance, including astrocytes, oligodendrocytes, microglia, and endothelial cells (Lau et al., 2020; Zhou et al., 2020). However, one major limitation of both scRNAseq and snRNAseq analysis is that cell-type abundance estimates are relative across the dataset. Therefore, a decline in neuronal subpopulations could lead to inflated abundance estimates of other, stable cell types. Indeed, whereas widespread neuronal loss is highly characteristic of AD (Davies and Maloney, 1976; Braak and Braak, 1991; Leng et al., 2021), systematic histopathological studies in postmortem brain tissue do not support an absolute increase in microglia or astrocyte numbers, but rather a proportional increase in reactive glia in diseased tissues (Serrano-Pozo et al., 2013; Davies et al., 2017; Paasila et al., 2019). We therefore computed confidence intervals for cell abundance changes under an alternative model in which glia were assumed to be unchanging (Figure 2—figure supplement 3A). In this more conservative, adjusted analysis, only the neuroendocrine group (cluster 12) was increased and 15 excitatory neuronal subtypes were decreased.

In order to resolve the remaining ambiguity in potential glial cell changes, we performed immunofluorescence on whole-mount Drosophila brains (Figure 2C). Although the overall intensity of glial nuclear staining (anti-Repo) was increased in elav>tauR406W flies, quantification revealed no significant increase in absolute glial numbers. Instead, we found nominally increased glial density in tau animals after considering their reduced total brain volumes (Figure 2D). The increased intensity of antibody staining in tau brains may arise from enhanced antibody penetration since similar changes are also seen for other markers (Figure 2—figure supplement 3B and C). Moreover, increased repo gene expression was not observed in either scRNAseq or in our previously published bulk-tissue RNAseq (Mangleburg et al., 2020). Overall, our results suggest that the apparent increase in glial cell abundance from scRNAseq is likely a consequence of proportional changes in single-cell suspensions due to neuronal loss: in our replication dataset from 10-day-old flies, the proportion of neurons were reduced from 90% to 83% in control versus elav>tauR406W flies. While it is difficult to exclude more modest or selective regional changes, we conclude that similar to human postmortem tissue findings (Serrano-Pozo et al., 2013), absolute glial numbers are largely stable following tau expression in the Drosophila brain.

Tau and aging exert cell-specific effects on brain gene expression

To our knowledge, the specific contributions of tau and aging on gene expression across heterogeneous cell types in the adult brain have not been systematically examined. In order to define the impact of aging on brain gene expression, we first quantified cell-specific transcriptional signatures in control flies (elav-GAL4) by performing differential expression analyses between the three timepoints from matched cell clusters (Figure 3A, Figure 3—source data 1). Overall, we define 5998 unique, aging-induced differentially expressed genes. Based on Gene Ontology term enrichment, ribosome/protein translation and energy metabolism pathways were broadly dysregulated during aging, involving the majority of cell types (Figure 3—source data 2). We next used linear regression to examine tau-induced differential gene expression within each cell type, including adjustment for age as a covariate. Overall, a total of 5280 unique genes were differentially expressed in at least one or more cell types (Figure 3B, Figure 3—figure supplement 1A), and these results overlap significantly with our prior bulk RNA-seq in elav>tauR406W flies (Figure 3—figure supplement 2). Importantly, 93% of tau-induced differentially expressed genes (n = 4917 out of 5280) were also triggered by aging in control flies (among n = 5998 genes). However, tau and aging appeared to have markedly distinct impacts when considering the distribution of gene perturbations across heterogeneous cell types (Figure 3C). Whereas aging broadly perturbed gene expression, tau-triggered changes were sharply polarized to excitatory neurons and glia. Further, the overlap between tau and aging varied across clusters (range = 0–75%) and tau-specific signatures predominated in selected cell types. For example, cholinergic Kenyon cells from the α'/β' mushroom body lobes were among the most vulnerable cell types (Figure 2A) and also had the greatest number of tau-induced gene perturbations (Figure 3B), which were approximately equally divided between up- and downregulated changes (Figure 3—figure supplement 1A, Figure 3—source data 1). In fact, among 2289 tau-induced differentially expressed genes within α'/β' Kenyon cells, 2139 (93%) were unique to tau and not similarly triggered in the corresponding cell type in aging control animals. We confirmed that the number of differentially expressed genes and affected cell types does not correspond to the spatial pattern of MAPT transgene pan-neuronal expression in the brain (Figure 3—figure supplement 4); however, it is difficult to exclude the possibility that some vulnerable cell types with high MAPT expression might be inadvertently censored from our analyses.

Figure 3 with 5 supplements see all
Aging- versus tau-triggered brain gene expression changes.

(A) Aging has widespread transcriptional effects on most brain cell types. Number of aging-induced differentially expressed genes (false discovery rate [FDR] < 0.05) within each cell cluster is shown, based on comparisons of day 1 vs. day 10 and day 10 vs. day 20 in control animals only (elav-GAL4/+). For each cell cluster, the number of gene expression changes unique to aging (white) or overlapping with tau-induced changes (gray) is highlighted. Labels for cell clusters with significant tau-induced cell abundance changes are shown in bold. (B) In contrast with aging, tau induces a more focal transcriptional response, with greater selectivity for excitatory neurons and glia. Number of tau-induced, differentially-expressed genes (FDR < 0.05) within each cell cluster is shown, based on regression models including age as a covariate and considering both control and elav>tauR406W animals (elav-GAL4/+; UAS-tauR406W/+) at 1, 10, and 20 days. For each cell cluster, the number of gene expression changes unique to tau (black) or overlapping with aging-induced changes (gray) is highlighted. Labels for cell clusters with significant tau-induced cell abundance changes are shown in bold. Tau-induced gene expression changes from single-cell profiles significantly overlap with prior analyses conducted using bulk brain RNA-sequencing (Figure 3—figure supplement 2). (C) Uniform manifold approximation and projection (UMAP) plots show the number of aging- (red) versus tau- (green) triggered differentially expressed genes within each cell cluster. Color intensity represents the number of differentially expressed genes. See also Figure 3—figure supplements 15 and Figure 3—source data 1–5.

Figure 3—source data 1

Tau- and aging-triggered gene expression changes.

Tau-induced differentially expressed genes were adjusted for aging by including a covariate in the regression model, based on comparisons of scRNAseq data elav>tauR406W vs. control (elav-GAL4) at 1, 10, and 20 days. Aging-induced differentially expressed genes are based on comparisons in control (elav-GAL4) flies, including between day 1 (d1) and day 10 (d10), and day 10 vs. day 20 (d20); comparisons are noted in age_comparisons column. The cell cluster being compared is denoted in the cluster column. Avg_logFC is the log2 fold change of gene expression between day 10 vs. day 1 or day 20 vs. day 10; in each entry, the former is the numerator, and the latter is the denominator. For tau vs. control comparisons, the numerator is tau, and the denominator is control. Pct.1 and Pct.2 refer to the percent of cells that have non-zero expression for the given gene in the numerator and denominator, respectively. P_val = uncorrected p-values from the MAST linear regression. Padj = Benjamini–Hochberg-adjusted p-values. Analysis = specifies either ‘control aging’ or ‘tau age-adjusted’ for the respective analyses.

https://cdn.elifesciences.org/articles/85251/elife-85251-fig3-data1-v2.xlsx
Figure 3—source data 2

Functional pathways from differential expression analysis.

Significantly enriched functional terms based on overrepresentation analysis (ORA) of cell-specific differentially expressed gene sets, including from either (i) aging (controls), (ii) tau age-adjusted (elav>tauR406W vs. control (elav-GAL4)), or the (iii) ‘tau-specific’ gene set, which is the unique subset of genes from ii not seen in i. Genes used for functional enrichment analysis have a false discovery rate (FDR) < 0.05 in all differential expression analyses, and all functional enrichment terms listed have a hypergeometric FDR < 0.05. Analysis = source of gene set used for functional enrichment (i–iii, above). Age = relevant age groups of source comparison. Cluster = cell cluster source of gene set. Term_id = identifier of enrichment term. term = description of enriched term. FDR = FDR-corrected p-values. Database = database origin of term.

https://cdn.elifesciences.org/articles/85251/elife-85251-fig3-data2-v2.xlsx
Figure 3—source data 3

Tau-induced gene expression changes in the replication dataset.

Cross-sectional replication analysis comparing differentially expressed genes in an independent dataset from day 10 (elav >tauR406W vs. control (elav-GAL4)). The cell cluster being compared is denoted in the cluster column. Avg_log2FC is the log2 fold change of gene expression between tau vs. control comparisons, the numerator is tau, and the denominator is control. Pct.1 and Pct.2 refer to the percent of cells that have non-zero expression for the named gene in the numerator and denominator, respectively. P_val = unadjusted, raw p-values from the MAST linear regression. Padj = Benjamini–Hochberg-adjusted p-values.

https://cdn.elifesciences.org/articles/85251/elife-85251-fig3-data3-v2.xlsx
Figure 3—source data 4

Cell-type-specific overlaps between tau-induced differentially expressed genes.

Cell-cluster overlaps are quantified between the age-adjusted discovery data (Figure 3—source data 1) and the day 10 cross-sectional replication data (Figure 3—source data 3). Cluster = annotated cell identities or Seurat ID of unannotated clusters. ageAdj_discovery_DEG_n = number of tau-induced differentially expressed genes (FDR < 0.05) in the discovery dataset. d10_CS_replicate_DEG_n = number of differentially expressed genes in the replication dataset. Intersect = number of overlapping differentially expressed genes between results of the two comparisons. percent_of_original = percent of differentially expressed genes in the discovery dataset that is also observed in the replication dataset. Phyper = p-value of hypergeometric tests evaluating whether the number of overlapping genes observed is greater than by chance. tot_genes = total number of unique genes detected in each cell type and shared between datasets used for hypergeometric test.

https://cdn.elifesciences.org/articles/85251/elife-85251-fig3-data4-v2.xlsx
Figure 3—source data 5

Cross-sectional tau-induced differential expression.

Cross-sectional analysis of tau-induced changes (elav>tauR406W vs. control (elav-GAL4)) from the discovery dataset at 1, 10, and 20 days; age = the age being compared. The specific cell cluster being compared is denoted in the cluster column. Avg_log2FC is the log2 fold change of gene expression between tau vs. control comparisons, the numerator is tau, and the denominator is control. Pct.1 and Pct.2 refer to the percent of cells that have non-zero expression for the named gene in the numerator and denominator, respectively. P_val = unadjusted, raw p-values from the MAST linear regression. Padj = Benjamini–Hochberg-adjusted p-values.

https://cdn.elifesciences.org/articles/85251/elife-85251-fig3-data5-v2.xlsx

Using functional enrichment analysis, we identify tau transcriptional signatures implicating altered inflammation, oxidative phosphorylation, and ribosomal gene expression (Figure 3—figure supplement 1B, Figure 3—source data 2). These pathways were prominently disrupted in excitatory neurons of the fly visual system, along with other central brain cholinergic and glutamatergic cell clusters. The pattern of transcriptional perturbation is also consistent with the established susceptibility of the mushroom body and optic lobes to tau-mediated neurodegeneration (Wittmann et al., 2001; Kosmidis et al., 2010). In other cases, we noted functional enrichments with greater specificity for selected cell clusters, such as altered signatures for mTOR signaling in glutamatergic cluster 21 and Foxo signaling in a subset of neuron types, including lamina intrinsic amacrine (Lai) cells and a cluster receptive to columnar motion (T4/T5). In addition, genes involved in mRNA splicing regulation were perturbed in another group of visual processing cells (T2a) as well as cholinergic cluster 7. Among non-neuronal cells, ensheathing glia, cortex glia, astrocyte-like glia, and hemocytes had the greatest number of tau-driven differential expression changes (Figure 3—figure supplement 1C), highlighting signatures related to fatty acid metabolism and synaptic regulation (Figure 3—source data 2).

To examine the robustness of our findings, we compared our results on tau-induced, cell-type-specific gene expression changes with the independent dataset from 10-day-old flies. Based on hypergeometric overlap tests of differentially expressed gene sets, expression profiles in two-thirds (61 out of 90) of cell-type clusters from our longitudinal analysis were replicated at 10 days, including several vulnerable excitatory neuron and glial cell clusters (Figure 3—source data 4). In secondary analyses, we also analyzed differential expression cross-sectionally, permitting examination of age-dependent changes in specific genes or pathways (Figure 3—figure supplement 5, Figure 3—source data 5). Overall, when aggregated across all clusters, there was a 90% overlap between the total unique, tau-triggered differentially expressed genes at 10 days between the discovery and replication dataset.

Tau triggers changes in neuronal innate immune signaling

Whereas most tau-induced genes strongly overlapped with aging, a minority overall were tau-specific (363 out of 5280 gene perturbations). Interestingly, this gene set was significantly enriched for mediators of the innate immune response, particularly NFκB signaling pathway components (Figure 3—source data 2). From Drosophila bulk brain RNA-seq data, we previously identified seven gene coexpression modules perturbed by tauR406W expression using weighted correlation network analysis (WGCNA) (Mangleburg et al., 2020). Among these, a 236-gene module was strongly enriched for innate immune response genes downstream of NFκB. In our bulk brain RNA-seq data, this module was also activated by wildtype tau, but the mutant form, tauR406W, caused a more robust, accelerated response (Figure 4—figure supplement 1). In order to better understand the cell type-specific expression patterns, we next examined the innate immune coexpression module in our scRNAseq data. This immune signature was broadly detected in the adult fly brain, including both glia and many neuron types (Figure 4A, Figure 4—figure supplement 2A). Moreover, expression of the immune module was strongly dysregulated by tau, with 50 out of 90 clusters showing significant changes (Figure 4B, Figure 4—source data 1). Tau activated the immune signature in the majority of affected cell types (86%, 43 out of 50 clusters). In particular, tau-triggered increases were noted in multiple excitatory neuron clusters (e.g., Dm3 glutamatergic cells in the visual system) as well as non-neuronal cells, including glia (e.g., ensheathing and cortex glia) and hemocytes. Conversely, in a selected subset of seven clusters, tau attenuated expression of the innate immune module (Figure 4B), including excitatory neurons in the lamina and several Kenyon cell types that were among the most vulnerable to tau-triggered neuronal loss, based on cell abundance estimates (Figure 2A). Other tau-perturbed coexpression modules revealed distinct cell-type-specific patterns (Figure 4—figure supplement 3). For example, a module enriched for synaptic regulators was markedly reduced in glia in response to tau, whereas expression was increased in multiple glutamatergic neuron subtypes.

Figure 4 with 8 supplements see all
Tau-induced changes in innate immune response genes and neuronal vulnerability.

(A) Innate immune genes are expressed broadly in the adult fly brain, including both neurons and glia. Plot shows mean overall normalized expression by cell cluster among n = 236 genes belonging to a tau-induced coexpression module that is significantly enriched for innate immune response pathways (Mangleburg et al., 2020). In this plot, gene expression was averaged across both elav>tauR406W and control cells; similar results are seen when stratifying by either age or genotype (Figure 4—figure supplement 2A). See also Figure 4—figure supplement 2D for experimental confirmation of NFκB/Rel protein expression in neurons and glia. (B) Tau activates or suppresses innate immune response genes in a cell-type-specific manner. Plot shows log2 fold-change mean expression per cell cluster for the same 236-gene immune response coexpression module, based on comparisons between elav>tauR406W (elav-GAL4/+; UAS-tauR406W/+) and control (elav-GAL4/+) flies. See also Figure 4—figure supplement 2B and C for plots of curated NFκB signaling pathway genes and Figure 4—figure supplement 3 for similar analyses of other coexpression modules. (C) Log2 fold-change in Relish (Rel) regulon gene expression per cluster is shown, based on comparisons between tauR406W and control flies. All results were significant (false discovery rate [FDR] < 0.05) based on regression models including age as a covariate. (D) Plot shows overall mean expression of the Rel-regulon (x-axis) versus tau-induced cell abundance change (y-axis). Among clusters with significant, tau-induced cell loss (denoted in blue, FDR < 0.05; see also Figure 2A), cell abundance change was inversely correlated with Rel regulon expression (Pearson correlation: R = –0.9, p=0.0021). Many other cell types without significant cell abundance changes are also shown in gray. Both control and tau cells are pooled for this analysis. See also Figure 4—figure supplements 18 and Figure 4—source data 1–4.

Figure 4—source data 1

Tau-induced expression changes in innate immune response genes.

Differential expression of the immune response coexpression module (magenta, Mangleburg et al., 2020), based on comparisons of elav>tauR406W and control (elav-GAL4) animals, adjusting for age. Cluster = cell cluster identity. lrt.pvalues = uncorrected p-values from likelihood ratio test. lrt.padj = Benjamini–Hochberg-adjusted p-value. log2FC = log2 fold change of mean immune module expression between tau and controls for each cell cluster.

https://cdn.elifesciences.org/articles/85251/elife-85251-fig4-data1-v2.xlsx
Figure 4—source data 2

Regulon coexpression networks.

183 regulons and member genes are denoted with the (+) notation, indicating that genes within these modules are positively co-expressed.

https://cdn.elifesciences.org/articles/85251/elife-85251-fig4-data2-v2.xlsx
Figure 4—source data 3

Differential regulon expression analysis.

Regulon expression per cell was defined as the mean expression of regulon member genes. For each cell type, regulon expression across all cells was regressed on genotype [elav>tauR406W vs. control (elav-GAL4)] and age, and a likelihood ratio test was performed against a reduced model with age only. Cluster = cell identity. Regulons = regulon used in statistical testing. lrt.pvalues = unadjusted p-values from likelihood ratio test. lrt.padj = Benjamini–Hochberg-adjusted p-values. log2FC = log2 fold change of mean regulon expression between tau and control.

https://cdn.elifesciences.org/articles/85251/elife-85251-fig4-data3-v2.xlsx
Figure 4—source data 4

Predictors of tau-triggered cell proportion changes.

Comprehensive list of retained, non-zero coefficients from the elastic net regression models considering expanded list of 2993 predictor variables and cell clusters showing significant, tau-induced reductions in cell abundance. Term = a variable in the elastic net multiple regression. Elastic net coefficient = the coefficient for the specified variable in the final regression model.

https://cdn.elifesciences.org/articles/85251/elife-85251-fig4-data4-v2.xlsx

To confirm and extend our analysis of tau- and cell-type-specific gene expression perturbations, we derived a complementary set of 183 transcription factor coexpression networks (regulons) based on our scRNAseq data. Specifically, regulons define coexpressed gene sets in which members are also predicted targets of a specific transcription factor (Van de Sande et al., 2020). Overall, clustering cells based on regulon enrichment recapitulates similar, expected relationships between annotated cell types (Figure 4—figure supplement 4, Figure 4—source data 2), and differential regulon analysis also revealed consistent tau-induced, cell-type-specific transcriptional perturbations (Figure 4—source data 3). In particular, we examined the 442-gene regulon comprised of targets of the NFκB transcription factor ortholog in Drosophila, Relish (Rel), which is activated downstream of the Drosophila Imd (Immune deficiency) pathway, similar to the tumor necrosis factor receptor pathway in mammals (Myllymäki et al., 2014). The expression pattern of the Rel regulon and its differential expression in tau versus control flies were consistent with our findings for the immune coexpression module derived from bulk RNAseq, which includes both Imd, Rel, and multiple antimicrobial peptides that are activated by Rel (Figure 4C). We also obtained consistent results based on a manually-curated, 62-gene set including well-established NFκB signaling pathway members (Figure 4—figure supplement 2B and C). Based on our cross-sectional analyses, the pattern of tau-triggered activation of the Rel regulon in selected clusters (e.g., L1-5 lamina neurons and astrocyte-like glia) was age-dependent (Figure 4—figure supplement 5). We also experimentally confirmed Rel expression in both neurons and glia in the adult fly brain using an available strain in which the endogenous protein harbors an amino-terminal GFP tag (Figure 4—figure supplement 2D).

Expression signatures for neuronal vulnerability in tauopathy

In order to more directly model the relationship of transcriptional regulation and cellular vulnerability in tauopathy, we integrated regulon expression levels with cell abundance estimates from scRNAseq (Figure 4—figure supplement 7A). We hypothesized that innate immune signatures may be predictors of neuronal subtype vulnerability in tauopathy. We implemented regularized multiple regression in which cell-type-specific regulon mean expression served as the predictor variable and tau-triggered cell abundance changes from scRNAseq provided the response variable. The analysis was restricted to cell clusters that show significant declines in elav>tauR406W flies. Out of 183 total regulons, Rel/NFκB activity was prioritized among the top predictors of vulnerability to tau-induced cell loss (Figure 4D, Figure 4—figure supplement 7B). The Rel regulon remained a robust predictor in an expanded analysis including multiple technical variables as well as expression levels for an additional 2793 curated functional pathways (Figure 4—source data 4). Importantly, for this analysis, regulon expression was averaged across both elav>tauR406W and control cells, rather than considering differential expression, and the vulnerable clusters include cell types in which Rel and its targets (Rel regulon) are either activated (e.g., Dm3) or suppressed (e.g., Gamma lobe of the Kenyon cells) in response to tau (Figure 4C). Interestingly, the inverse relationship with cell abundance is recapitulated when restricting consideration of Rel regulon activity in control animals, suggesting that basal NFκB signaling—in the absence of tau—may be a predictive marker for neurodegeneration (Figure 4—figure supplement 7C). Specifically, among those cells vulnerable to tau-triggered cell abundance changes, Rel regulon expression is associated with the severity of decline. Besides Rel, the top 3 predictors of vulnerability for tau-induced cell loss include the CrebB and CHES-1 regulons (see Figure 4—figure supplement 7B for full list). Interestingly, CrebB—the cAMP response element-binding protein—and its target genes were previously shown to be dysregulated in the Drosophila tauopathy model (Mahoney et al., 2020), consistent with our finding of CrebB regulon downregulation across many cell types (Figure 4—source data 3). In mammals, the conserved CrebB ortholog, CREB, is linked to synaptic plasticity and long-term memory storage, and has also been proposed to interact with the NFκB pathway (Kaltschmidt et al., 2006).

In order to directly test whether Rel/NFκB may modify tau-mediated neurodegeneration in a cell-autonomous manner, we used RNA-interference (RNAi) for neuron-specific knockdown of Rel and performed histology to detect structural brain degeneration. In these experiments, elav-GAL4 is used to drive pan-neuronal expression of both UAS-tauR406W and the UAS-Relish.RNAi transgenes. However, we did not detect any significant difference in the vacuolar degeneration caused by tau following Rel knockdown (Figure 4—figure supplement 8). Additional experiments will likely be required definitively resolve the cell-type-specific causal mechanisms (see ‘Discussion’); however, our results identify NFκB targets and innate immune signaling as potential markers and/or mediators of vulnerability to tau-mediated neurodegeneration.

Cross-species overlap of cell-type-specific transcriptional signatures

To establish translational relevance, we next examined the conservation of cell-type-specific transcriptional signatures between Drosophila and human brain (Figure 5A, Figure 5—figure supplement 1A). Using Pearson correlation and considering 5630 conserved genes (1:1 fly/human mapping), we assessed pairwise correspondences between gene expression profiles for all clusters from either our Drosophila scRNAseq data (tauR406W + control) and published snRNAseq from human dorsolateral prefrontal cortex (AD cases and control) (Mathys et al., 2019). Overall, inferred neuronal and glial cellular identities correlated well across species. Cross-species correlations in cell-type-specific signatures were further replicated in an independent AD case/control snRNAseq dataset from the human entorhinal cortex (Grubman et al., 2019; Figure 5—figure supplement 1B). Similar results were also obtained in a complementary analysis leveraging a published Drosophila scRNAseq dataset (wildtype flies only) and excluding human brains with AD pathology (controls only) (Figure 5—figure supplement 2). The resulting correlation map can enable integrative, cross-species analyses. For example, a human microglial subcluster (Mic1) notable for association with high tau neuropathological burden was correlated with the ensheathing glia cluster from Drosophila, indicating shared characteristic transcriptional signatures (Figure 5A). Moreover, these two cell types showed significantly overlapping gene expression changes in association with AD pathology (human brain) or following pan-neuronal expression of tauR406W (Drosophila) (hypergeometric test, p=4.83 × 10–5) (Figure 5—source data 1). This result suggests that tau pathology may indeed be an important driver of Mic1 transcriptional changes in disease.

Figure 5 with 3 supplements see all
Conservation of cell-type-specific gene expression signatures.

(A) Heatmap shows Pearson correlation of gene expression (5630 conserved, orthologous genes) between annotated cell clusters from Drosophila (rows) and human postmortem brain (column). Human brain single-nucleus RNA-sequencing (snRNAseq) was obtained from Mathys et al., 2019, including published cell-type associations with amyloid plaque burden and neurofibrillary tangle Braak staging (braaksc) (top). Annotated human cell types include endothelial cells (End), microglia (Mic), oligodendrocytes (Oli), pericytes (Per), astrocytes (Ast), oligodendrocyte precursor cells (Opc), excitatory neurons (Ex), and inhibitory neurons (In). (B) Innate immune mediators are expressed broadly in the human brain, including in neurons and glia. Plot shows mean expression by cell cluster for 85 human orthologs of NFκB signaling pathway members, based on reprocessing and analysis of the Mathys et al. snRNAseq data. (C) Alzheimer’s disease (AD) is associated with cell-type-specific perturbation in NFκB signaling genes. Plot shows log2 fold-change mean expression per cell cluster for the same 85 NFκB signaling genes, based on comparisons of brains with AD pathology versus controls. See also Figure 5—figure supplements 13 and Figure 5—source data 1.

Figure 5—source data 1

Cell-type-specific, Alzheimer’s disease (AD)-associated gene expression changes from human brain.

snRNAseq data from human postmortem brain tissues (Mathys et al., 2019) was analyzed for differentially expressed genes in AD neuropathological cases versus controls without AD pathology. Cell subcluster (subcluster) labels are as defined by the original publication. Pct.1 and Pct.2 refer to the percent of cells that have non-zero expression for the given gene in the numerator and denominator, respectively. Avg_log2FC is the log2 fold change of gene expression between AD vs. control comparisons; the numerator is AD, and the denominator is control. P_val = unadjusted p-values from the MAST linear regression analysis. Padj = Benjamini–Hochberg-adjusted p-values.

https://cdn.elifesciences.org/articles/85251/elife-85251-fig5-data1-v2.xlsx

As introduced above, mediators of innate immunity are also highly conserved across species. Similar to elav>tauR406W flies, we confirmed consistent NFκB pathway expression in excitatory neurons and microglia in transgenic mice harboring a MAPTP301S transgene (Lee et al., 2021; Figure 5—figure supplement 3). Next, leveraging the Mathys et al. human snRNAseq data, we confirmed that NFκB signaling pathway genes are expressed across most cell types in human postmortem brain tissue, including both neurons and glia (Figure 5B). In the context of AD pathology, NFκB pathway gene expression appeared strongly downregulated in most neurons from the dorsolateral prefrontal cortex, which are highly susceptible to degeneration, whereas expression was increased among oligodendrocytes, microglia, and astrocytes (Figure 5C). Interestingly, a subset of excitatory and inhibitory neuronal subclusters (Ex8 and In4, respectively) showed an AD-associated increase in expression. Thus, human brains with AD pathology are also characterized by widespread changes in NFκB innate immune signaling, including either activation or attenuation in many distinct neuronal and non-neuronal subtypes.

Discussion

Aging is the most important risk factor for AD, influencing both disease onset and progression. Based on longitudinal, single-cell analysis in Drosophila, we discover that tau and aging activate strongly overlapping transcriptional responses: 93% of tau-induced differentially expressed genes are also perturbed by aging in control animals. Instead, tau and aging are distinguished by their spatial and cell-type-specific impacts. Aging has a global influence on brain gene expression, affecting most brain cell types. By contrast, tau has a focal impact, polarizing the transcriptional response to a handful of cell types, including excitatory neurons and glia. The strong overlap between tau- and aging-induced gene expression signatures agrees with our prior analyses of bulk brain tissue (Mangleburg et al., 2020). We and others have also documented similar findings in AD mouse models, including both MAPT and amyloid precursor protein transgenics (Wan et al., 2020; Cummings et al., 2015; Gjoneska et al., 2015; Matarin et al., 2015; Hargis and Blalock, 2017). By contrast with animal models, cross-sectional studies of human postmortem tissue make it difficult to disambiguate the impact of aging from disease pathology on the brain transcriptome. However, our cross-species analyses highlight that most human brain cell types share transcriptional signatures with counterparts in the Drosophila brain. These correspondences comprise a cross-species atlas enabling studies of controlled experimental manipulations (e.g., tau vs. aging) on homologous cell clusters between humans and flies.

Mechanistic dissection of cell-type-specific vulnerability promises to reveal drivers for the earliest clinical manifestations of AD, such as the characteristic memory impairment accompanying the loss of excitatory neurons in hippocampus and associated limbic regions (Mrdjen et al., 2019; Fu et al., 2018). Given the transcriptional overlaps, one attractive model is that aging establishes a spatial pattern of vulnerable cell states that templates the subsequent tau-triggered neurodegeneration. However, as noted above, aging has wide-ranging impact across the brain and many cell types with robust aging-induced transcriptional responses in Drosophila are, in fact, resilient to tau-mediated neurodegeneration based on cell proportion changes (e.g., clusters 2, 3, and T4/T5; Figures 2A and 3A). Moreover, the overlap between tau and aging does not reliably predict those cell types that are most vulnerable to neuronal loss (Figure 3A and B). Differentially expressed genes triggered by tau and aging are nevertheless similarly enriched for many common biological pathways that may provide clues to cell-type-specific mechanisms of vulnerability in neurodegeneration. Specifically, we document shared expression signatures for altered synaptic regulation, protein translation, lipid metabolism, and oxidative phosphorylation across heterogeneous cell populations, including excitatory neuron types that are particularly vulnerable to tau. Similar pathways have been implicated based on snRNAseq analyses from human postmortem brain (Mathys et al., 2019; Grubman et al., 2019; Lau et al., 2020) and several mouse AD models, including MAPT transgenics (Wang et al., 2022; Lee et al., 2021; Habib et al., 2020; Zhou et al., 2020).

Among the many dysregulated molecular processes, aging is characterized by a systemic pro-inflammatory state that has been called ‘immunosenescence’ or ‘inflamma-aging’ (Shaw et al., 2013; Hou et al., 2019). Genes encoding regulators of immunity, including TREM2, CR1, and many others, have been strongly implicated in AD susceptibility by human genetics (Bellenguez et al., 2022), and abundant evidence now supports a key role for many such genes among glial cells (Wang et al., 2015; Zhou et al., 2020; Keren-Shaul et al., 2017). We previously identified an age-associated Drosophila innate immune response signature that is amplified by tau (Mangleburg et al., 2020). Here, we significantly extend these observations, leveraging the cellular resolution afforded by single-cell profiles. First, we discover that this immune coexpression module, including many NFκB/Rel signaling factors and targets, is broadly expressed in the adult fly brain, including both neurons and glia, and we confirm similar findings in snRNAseq data from human postmortem brain. Second, we show that tau can either activate or attenuate NFκB immune pathways in a cell-type-specific manner, with tau-triggered decreases in expression apparent in neurons with the greatest proportional cell loss. Lastly, models integrating cell-type-specific gene transcriptional expression and cell abundance changes suggest that basal Imd signaling strength (i.e., Rel regulon activity) predicts the severity of tau-triggered neuronal decline among susceptible cell types. Overall, our results suggest that besides the well-established requirements in glia (see below), innate immune response pathways may also have important, cell-autonomous roles in modulating neuronal vulnerability to tau pathology in AD. Indeed, both insect and mammalian neurons express evolutionary-conserved innate immune signaling pathways, including Toll-like receptors and NFκB signal transduction components, and these pathways can be triggered by infection or other cellular insults (Lehnardt et al., 2003; Tang et al., 2007; Cao et al., 2013; Cho et al., 2013; Petersen et al., 2013; Welch et al., 2022). In addition, NFκB immune signaling pathways have been coopted for diverse, non-canonical functions, such as in neurodevelopment and synaptic plasticity (Okun et al., 2011; Gutierrez and Davies, 2011; Nguyen et al., 2020). Knockdown of Rel in Drosophila neurons has previously been shown to promote survival in non-transgenic, wildtype animals (Kounatidis et al., 2017), whereas activation of the Rel signaling pathway leads to neurodegeneration (Cao et al., 2013). In addition, a recent reanalysis of snRNAseq data from Mathys et al., 2019 revealed AD-associated perturbation of NFκB immune pathways in excitatory neurons, possibly triggered by DNA double-strand breaks (Welch et al., 2022). Although experimental manipulation of Rel in the elav>tauR406W model did not alter tau-mediated neurodegeneration, additional studies may be required to definitively resolve the potential cell-type-specific causal contribution(s) of NFκB/Relish. Our negative result could reflect poor sensitivity and variability of the histologic assay or it may be necessary to use alternate neuronal drivers restricted to the adult brain.

Our scRNAseq analyses also highlight a robust, tau-induced transcriptional response among Drosophila glia. This result is consistent with several brain gene expression studies from both humans and mouse models that strongly implicate altered transcriptional states and/or increased numbers of AD-associated glial subtypes, including oligodendrocytes, astrocytes, and microglia (Mathys et al., 2019; Grubman et al., 2019; Lau et al., 2020; Zhou et al., 2020; Habib et al., 2020). Although our analyses initially suggested a possible tau-triggered increase in glial abundance in the brain, on direct examination, we documented stable absolute numbers but increased density of glia due to brain atrophy. Systematic histopathological studies in human brain tissue have similarly revealed predominantly reactive changes with overall stable numbers of both astrocytes and microglia (Serrano-Pozo et al., 2013). We conclude that potential increases in disease-associated glia inferred exclusively from single-cell profiles should be interpreted cautiously, and additional experimental investigations may ultimately be required to resolve whether they result from (i) absolute changes in cell number, (ii) activation and/or transformation of cell states, or (iii) proportional changes due to primary perturbations in other brain cell types. Nevertheless, glial-specific experimental manipulations of immune regulators in both Drosophila and mammalian models, including NFκB signaling (flies and mice) and the AD susceptibility gene TREM2 (mice), can potently modify neurodegeneration, consistent with cell non-autonomous requirements (Walter, 2016; Kounatidis et al., 2017; Petersen et al., 2012; Hakim-Mishnaevski et al., 2019; Fuhrmann et al., 2010; Town et al., 2008; Leyns and Holtzman, 2017; Wang et al., 2022). By contrast with mammals, glia represent only 5–10% of all cells in the Drosophila brain (Ito et al., 1995; Schmidt et al., 1997; Awasaki et al., 2008). Nevertheless, Drosophila glial subtypes recapitulate the diversity of functions and morphologies of mammalian glia (Doherty et al., 2009; Freeman, 2015; Kremer et al., 2017; Stork et al., 2012). Although the myeloid hematopoietic lineage is not present in flies, which therefore lack microglia, ensheathing glia can similarly respond to cellular injury and scavenge debris (Doherty et al., 2009). Indeed, our cross-species analysis demonstrates shared transcriptional signatures between corresponding glial subtypes, consistent with our findings of conserved responses to tau-mediated neuronal injury. In future work, it will be interesting to further dissect both the cell-autonomous and non-cell-autonomous drivers underlying both the neuronal and glial responses to tauopathy.

The elav>tauR406W flies selected for this study share conserved downstream mechanisms of neurotoxicity with wildtype tau (Bardai et al., 2018) and have been widely used as an experimental model for investigations of both AD and other tauopathies, including frontotemporal dementia. Nevertheless, one potential caveat is the absence of amyloid-beta peptide, which is also an important driver of gene expression changes in AD, including innate immune transcriptional signatures (Keren-Shaul et al., 2017; Wan et al., 2020). Another potential limitation is that the elav-GAL4 driver activates tau expression during developmental stages, and the observed changes in cell-abundance or gene expression may therefore reflect this time course. For example, tau developmental toxicity has been shown to cause malformation of mushroom body structures (Kosmidis et al., 2010), and this phenotype likely explains the reductions in several cell clusters in our dataset. While our study was under review, a complementary, single-cell transcriptome analysis using the nsyb>tauP301L model was published, in which transgene expression is expected to be more restricted within the adult brain (Praschberger et al., 2023). While there were overlaps in the vulnerable cell types for both the elav>tau and nsyb>tau models (e.g., excitatory cholinergic neuron subtypes, like γ-KC, α'/β'-KC, and T4/5), there were also some notable distinctions—the inhibitory C2 cell cluster, which is GABAergic, was highlighted only in the nsyb model. Further comparisons are somewhat limited by other experimental and analytic design differences between the studies. Nevertheless, the elav>tauR406W model is well established to recapitulate aging-dependent, neuronal loss and progressive CNS dysfunction (Wittmann et al., 2001). Indeed, our longitudinal design reveals suggestive age-dependent cell abundance changes among several cell types (e.g., clusters 1, 9, and 12, along with astrocyte-like glia), and cross-sectional analyses also reveal evidence for progressive transcriptional changes. It will be important to perform additional studies, perhaps using inducible driver systems, to more systematically dissect the dynamic time course of tau neurotoxic mechanisms, including differentiating developmental versus degenerative changes that accompany brain aging.

Materials and methods

Key resources table
Reagent type (species) or resourceDesignationSource or referenceIdentifiersAdditional information
AntibodyRabbit polyclonal anti-GFPGeneTexCat#GTX113617; RRID:AB_1950371IF(1:500)
AntibodyAlexa 647 goat polyclonal anti-rabbit IgG (H+L)Jackson ImmunoResearchCat#111-605-003IF(1:500)
AntibodyCyTM3 AffiniPure goat polyclonal anti-mouse (H+L)Jackson ImmunoResearchCat#115-165-003IF(1:500)
AntibodyAlexa Fluor 488 donkey polyclonal anti-mouse IgG (H+L)Jackson ImmunoResearchCat#715-545-150IF(1:500)
AntibodyCy3TM3 AffiniPure goat polyclonal anti-rat IgG (H+L)Jackson ImmunoResearchCat#112-165-003IF(I:500)
AntibodyMouse monoclonal anti-repoDSHBCat#8D12IF(1:500) – glial counting
IF(1:50) – Rel costain
AntibodyRat monoclonal anti-ElavDSHBCat#7E8A10; RRID:AB_528218IF(1:100)
AntibodyMouse monoclonal anti-RelDSHBCat#21F3;
RRID:AB_1553772
IF(1:500)
Chemical compound, reagentConjugated A488-PhalloidinThermo FisherCat#A12379IF(1:500)
Chemical compound, drugDispaseSigma-AldrichCat#D4818;
Chemical compound, drugCollagenase IInvitrogenCat#17100-100
Chemical compound, drugNucBlue and Propidium iodideInvitrogenCat#R37610
Chemical compound, drugVectashield antifade mounting mediumVector LaboratoriesCat#H-1000-10
Commercial assay or kitChromium Single Cell Gene Expression 3’ v3.110x GenomicsCat#PN-1000268
Genetic reagent (Drosophila melanogaster)elavC155-GAL4Bloomington Drosophila
Stock Center
BDSC:458
Genetic reagent (D. melanogaster)w1118; UAS-TauR406WLab: Dr. Mel B. Feany,
PMID:11408621
N/AWittmann et al., 2001
Genetic reagent (D. melanogaster)Rel-GFPBloomington Drosophila
Stock Center
BDSC:81268y1 w*; PBac{GFP.FPTB-Rel}VK00037
Genetic reagent (D. melanogaster)UAS-Rel.RNAi-2Bloomington Drosophila
Stock Center
BDSC:33661y1; P{TRiP.HMS00070}attP2
Genetic reagent (D. melanogaster)UAS-Rel.RNAi-1Vienna Drosophila
Resource Center
VDRC:49414P{GD1199}v49414
Software, algorithmImaris Microscopy Image Analysis Software 9.9.1https://imaris.oxinst.com/Oxford Instruments
Software, algorithmPrism 9.4.1https://www.graphpad.com/scientific-software/prism/GraphPad
Software, algorithmImageJhttps://imagej.nih.gov/ij/NIH
Software, algorithmCell Ranger 4.0.0https://support.10xgenomics.com/single-cell-gene-expression/software/pipelines/latest/what-is-cell-ranger10x Genomics
Software, algorithmSeurat v3https://doi.org/10.1016/j.cell.2019.05.031Stuart et al., 2019
Software, algorithmDoubletFinder 2.0.3https://github.com/chris-mcginnis-ucsf/DoubletFinderMcGinnis et al., 2019
Software, algorithmScmap 1.9.3https://bioconductor.org/packages/release/bioc/html/scmap.htmlKiselev et al., 2018
Software, algorithmOptic lobe neural network classifierhttps://static-content.springer.com/esm/art%3A10.1038%2Fs41586-020-2879-3/MediaObjects/41586_2020_2879_MOESM7_ESM.zipÖzel et al., 2021,
Supplementary
Data Appendix 1,
Python/R code
Software, algorithmpySCENIC 0.12.0https://github.com/aertslab/pySCENICVan de Sande et al., 2020
Software, algorithmDESeq2 1.34.0https://bioconductor.org/packages/release/bioc/html/DESeq2.htmlLove et al., 2014
Software, algorithmMuSiC 0.1.1https://github.com/xuranw/MuSiCWang et al., 2019
Software, algorithmMAST 1.20.0https://bioconductor.org/packages/release/bioc/html/MAST.htmlFinak et al., 2015
Software, algorithmWEBGESTALTR
0.4.4
https://github.com/bzhanglab/WebGestaltRWang et al., 2013
Software, algorithmGlmnet 4.1-4https://cran.r-project.org/web/packages/glmnet/index.htmlFriedman et al., 2010
Software, algorithmCaret 6.0-92https://cran.r-project.org/web/packages/caret/index.htmlKuhn, 2008
Software, algorithmDRSC Integrated Ortholog Prediction Tool (DIOPT)https://www.flyrnai.org/dioptHu et al., 2011
Software, algorithmgProfiler2 0.2.1https://cran.r-project.org/web/packages/gprofiler2/index.htmlRaudvere et al., 2019
Software, algorithmSCTransform 0.3.3https://github.com/satijalab/sctransformStuart et al., 2019

Human subjects

Request a detailed protocol

No new data from human subjects were generated for this study. Previously published, available snRNAseq data from human postmortem brain were obtained from Mathys et al., 2019 and Grubman et al., 2019 in order to evaluate cross-species correspondences in cell-type-specific expression signatures. The Mathys data is comprised of snRNAseq from the dorsolateral prefrontal cortex (DLPFC) from 48 brain autopsies with varying AD neuropathology (amyloid plaque and tau neurofibrillary tangle burden), including 24 with no significant pathology (controls) and 24 cases with mild to severe AD pathology. Subjects were balanced for sex (12 males and 12 females), and age (median age at death = 87 for both groups). The Grubman data is comprised of snRNAseq from the entorhinal cortex of 12 brain autopsies, including 6 AD pathological cases and 6 controls without significant AD pathology. Subjects in the Grubman data were also age-matched, with a median age of 83 and 80 for the AD case and control groups, respectively.

Drosophila stocks and husbandry

Request a detailed protocol

For scRNAseq libraries generated in this study, w1118; UAS-tauR406W flies (0N4R isoform, 383 amino acids), described in Wittmann et al., 2001; Mangleburg et al., 2020 were crossed with the pan-neuronal driver elavC155-Gal4, producing the experimental genotypes: elav-Gal4/+;UAS-tauR406W/+ or elav-Gal4/Y; UAS-tauR406W/+. In order to minimize genetic background as a potential confounder, UAS-tauR406W strains used in this study were backcrossed with w1118 for five generations as previously described (Guo et al., 2018). Controls were generated by outcrossing elav-Gal4 with w1118 animals, producing elav-Gal4/+ or elav-Gal4/Y. Adult progeny from experimental crosses were subsequently aged to 1, 10, or 20 d for dissection and library generation. Flies were raised on standard molasses-based media at 25°C in ambient lighting. We also utilized a Rel-GFP strain (y, w; PBac{GFP.FPTB-Rel}VK00037), which is an endogenous protein trap allele, encoding a fusion protein with GFP at the Rel amino-terminus. For the histology experiments, elav-Gal4/Y;UAS-tauR406W/+ animals were crossed with UAS-Rel.RNAi-1 (VDRC: v49414), UAS-Rel.RNAi-2 (TRiP: HMS00070), or w1118. Resulting female progeny with both the UAS-tauR406W transgene and RNAi (or controls) were aged to 10 d and prepared for histology.

Drosophila brain histology

Request a detailed protocol

Drosophila heads were fixed in 8% glutaraldehyde (Electron Microscopy Sciences) at 4°C for 10 d, followed by paraffin embedding and microtome sectioning as previously described in Chouhan et al., 2016. Serial 5-µm-thick coronal sections were prepared for the whole head, mounted onto microscopy slides, and stained with hematoxylin and eosin. Bright-field microscopy images were acquired using the Leica DM 6000B system. For quantification, the number of vacuoles greater than 5 um in diameter in an ~50 um stack comprising of the ellipsoid body, fan-shaped body, and posterior commissure. The mean number of vacuoles per section was computed per animal. Statistical testing between conditions was performed using Welch’s t-test.

Drosophila brain dissociation

Request a detailed protocol

For scRNAseq profiling of elav >tauR406W and control flies, 16–18 dissected and intact Drosophila brains were combined and dissociated for each experimental condition (six total samples: 2 genotypes × 3 timepoints). An equal number of male and female animals were combined for each condition. For the replication dataset, triplicate samples (biological replicates) for the identical elav>tauR406W and control genotypes were prepared at day 10 (six total samples). Adult fly brains were dissected out of the cuticle using sharp forceps in 1X PBS and dissociated following published protocols (Davie et al., 2018). Dissected brains in solution were first centrifuged at 800 × g for 3 min, resuspended, and dissociated by incubating with 50 uL of dispase (3 mg/mL, Sigma) and 75 uL of collagenase I (100 mg/mL, Invitrogen) for 2 hr at 25°C while shaking at 500 RPM. Cell suspensions were mixed by gentle pipetting 3–4 times every 5 min in the first hour, and every 10 min in the second hour. Resulting cell suspensions were pelleted by centrifugation at 400 × g for 5 min at 4°C, washed in 1000 uL ice-cold PBS, pelleted, and resuspended in 400 uL ice-cold PBS with 0.04% bovine serum albumin. Cell suspensions were passed through a 10 um pluriStrainer cell strainer (pluriSelect) to ensure that undissociated tissue were removed and a single-cell suspension was obtained. Cell concentration and viability were assessed using a hemocytometer under a fluorescent microscope after staining with NucBlue and Propidium iodide (Invitrogen). Fresh, intact single-cell suspensions were immediately used for single-cell library preparation.

Single-cell library preparation and sequencing

Request a detailed protocol

Single-cell libraries were prepared per the manufacturer’s protocol for the Chromium Single Cell Gene Expression 3’ v3.1 kit (10x Genomics) by the BCM Single Cell Genomics Core. 16,000 cells were added to each channel with a target recovery rate of 10,000 cells per library. Cells, reverse transcription (RT) reagents, gel beads containing barcoded oligonucleotides, and oil were loaded on a Chromium controller (10x Genomics) to generate single-cell Gel Bead-In-Emulsions (GEMs) where full-length cDNA was synthesized and barcoded for each individual cell. GEMs were subsequently broken and cDNAs from each single cell were pooled. Following clean up using Dynabeads MyOne Silane Beads (Invitrogen), cDNA was amplified by PCR. The amplified product was fragmented to optimal size before end-repair, A-tailing, and adaptor ligation. Final library was generated by amplification. Completed libraries were sequenced using the Baylor Genomic and RNA Profiling Core on the Illumina NovaSeq 6000 platform with a minimum depth of 300,000,000 reads per sample (on average 463 M reads per sample). A total of 12 high-quality libraries were generated (six libraries for the discovery and replication datasets, respectively). Illumina BCL files were demultiplexed into FASTQ files by calling the Cell Ranger 4.0.0 mkfastq function. FASTQ files were aligned to the Drosophila reference genome (BDGP6.22.98) and quantified using the Cell Ranger 4.0.0 count pipeline. The human microtubule-associated protein tau (MAPT) mRNA coding sequence (CDS) (isoform 3, NCBI Reference Sequence NM_016834.5:151–1302) along with a short SV40 3’UTR sequence was appended to the Drosophila reference genome for assessing MAPT transgene expression levels. Given the 10× recovery rate estimations, the cell calling algorithm in Cell Ranger was applied by setting the --expect-cells parameter in count to 10,000 for each library, thus filtering out partitions that likely did not contain single cells. Cell ranger alignment metrics for each library are available in Figure 1—source data 3. Filtered count matrices were loaded into Seurat v3 in R for additional quality control and downstream analyses. Cells were removed from the data object if the number of unique genes per cell were less than 200 or greater than 3000, or if the proportion of mitochondrial reads per cell was greater than 20%. Filtered count matrices from Cell Ranger are available to download with the Drosophila scRNAseq data on the Synapse AMP-AD Knowledge Portal.

Normalization, integration, and clustering

Request a detailed protocol

Gene expression was first normalized independently per library using a regularized negative binomial regression approach as implemented by SCTransform (Stuart et al., 2019). 5000 highly variable features (HVG) were used for normalization while accounting for percent mitochondrial reads. Variable features were defined and ranked by computing the variance of standardized gene counts after loess-based adjustment of mean–variance relationships (Stuart et al., 2019). Residuals of the fitted regression models were used as normalized gene expression values for HVGs. All libraries normalized via SCTransform were integrated using the canonical correlation analysis (CCA) pipeline in Seurat v3 to correct for batch effects and facilitate identification of similar cell identities across conditions. Highly ranked HVGs shared across all libraries were used as integration features. Integration anchors across libraries (correspondences of the selected features between cells) were computed over the first 30 CCA dimensions in the combined dataset and then used to inform the subsequent integration and grouping of cells. After integration, Seurat v3 was used for principal component analysis (PCA) and cell clustering. 100 principal components (PCs) of the integrated dataset were used for graph-based clustering and Louvain algorithm optimization as implemented in FindNeighbors and FindClusters. The final resolution in FindClusters was set to resolution = 2, yielding 96 cell clusters in our dataset. We selected this resolution to replicate the clustering pattern of a similarly processed Drosophila whole-brain scRNA-seq dataset (Davie et al., 2018). 100 PCs were used to embed cells in 2D space via uniform manifold approximation and projection (UMAP). Normalization of gene counts used in differential expression analysis, cell cluster marker gene computation, cell identity annotation, and other applications directly comparing gene expression levels between cell clusters were computed separately on the non-integrated gene expression data using the NormalizeData function in Seurat v3. In brief, for each gene in each cell, unique molecular identifiers (UMI) were divided by the sum UMIs in that cell, multiplied by a scalar (10,000), and log transformed. However, cell cluster membership (clusters 0–95) was defined using the integrated dataset as described above. The six additional libraries that comprise the day 10 replication dataset were clustered, integrated, and analyzed separately using the identical pipeline.

Doublet detection

Request a detailed protocol

DoubletFinder was applied per library to predict and remove heterotypic doublets, leaving a total of 48,111 high-quality single cells in the discovery dataset. For each library, artificial doublets were generated from the existing data. PCA was performed after merging the real and artificial data and a distance matrix was generated with the first 40 PCs to compute the proportion of artificial K-nearest-neighbors (pANN) for each cell. PC neighborhood size (pK) for computing pANN was estimated for each library as previously described (McGinnis et al., 2019). The number of suspected doublets per library was estimated and cells were ranked by pANN for removal. Total doublet proportion for each library was computed based on a custom linear equation of the input-to-multiplet estimation provided by the 10x Chromium documentation: Y = 5.272x10-4 + 7.589x10-6 (x), x being the number of recovered intact cells after the initial filtering criteria described above. The linear equation was generated based on recovery estimations in the manufacturer’s protocol. Adjustment of the estimated doublet proportion for undetectable homotypic doublets was applied in DoubletFinder by using the Seurat clustering classifications at resolution = 2 as described above.

SCENIC regulons

Request a detailed protocol

Gene regulatory networks (regulons) were computed using the Python implementation of SCENIC (pySCENIC). Raw gene abundances (UMIs) for 48,111 high-quality cells were exported as a loom object via loompy, and pySCENIC was implemented as described in Van de Sande et al., 2020. Putative gene targets for the published list of 815 Drosophila transcription factors (TFs) (see Key Resources Table) were inferred by tree-based regression (GRNBoost2) where expression of each gene was regressed on TFs, producing a list of adjacencies connecting TFs to their target genes (non-mutually exclusive). In the cisTarget step, modules were retained for further analysis if the regulatory motif of its parent TF was enriched among most gene members. Within retained modules, genes lacking enrichment of the appropriate motif were pruned. TF-motif annotations and pre-computed motif-gene rankings were obtained from https://resources.aertslab.org/cistarget/, Drosophila v8; motif search space encompassed up to 5 kb upstream of transcription start sites and intronic regions. This pipeline identified 183 regulons, encompassing 7134 out of 14,907 genes in the transcriptome dataset (Figure 4—source data 2), and cell-level activity for each regulon was computed by a ranking and recovery approach using pySCENIC AUCell. Within each cell, genes were ranked by expression level in a descending order, then the cumulative number of genes recovered belonging to a regulon at each rank was recorded. An area under the curve (AUC) was calculated after applying a default cutoff at the 95th percentile of gene ranks and is used to infer regulon activity. High AUC scores indicate greater representation of a given regulon among the top 5% of highly expressed genes in a cell. AUC scores for the 183 regulons across 48,111 cells were used for unsupervised clustering by UMAP for visualization of cell relationships based on gene regulatory networks (Figure 4—figure supplement 4A).

Annotation of cell identity/abundance

Request a detailed protocol

We searched for cell identities of the 96 defined clusters by consolidating a series of four analytic approaches (Figure 1—figure supplement 2A). Two published datasets were used as references for our annotation procedure, including 56,902 cells from adult wildtype Drosophila (w1118 and DGRP-551) brains profiled at days 0, 1, 3, 6, 9, 15, 30, and 50 (Davie et al., 2018), as well as 109,743 cells from adult Canton-S Drosophila optic lobes at day 3 (Özel et al., 2021). Cell clusters in these references were previously annotated using available literature-based cell markers or statistical inference with published bulk RNA-sequencing of reporter-targeted cell types. The Davie et al. dataset contained 87 cell clusters (Seurat FindClusters res = 2.0) with 41 assigned cell identities. The Özel et al., 2021 dataset was clustered at a higher resolution (Seurat FindClusters res = 10), containing 200 cell clusters and 87 assigned cell identities. First, Scmap-cluster was used to compute gene expression correlation between each cell in our dataset to all defined clusters in the Davie and Özel datasets. 500 genes with higher-than-expected dropouts were selected as correlation features as described in Kiselev et al., 2018. The cosine similarity, Spearman and Pearson correlations of these features were subsequently computed between each cell in our dataset and all reference cluster centroids. For a cell to be mapped to a reference cluster, two out of three similarity scores must be concordant, and at least one must be greater than 0.7. Second, we intersected the top 20 cluster markers for each cell cluster (ranked by log2 fold change) in our dataset with the top 20 markers in each reference cluster. Cluster markers (cluster-enriched genes) for our 96 cell clusters were computed by differential expression analysis of the non-integrated, normalized gene abundances, comparing each cell cluster against all remaining cells. Markers were defined as positively differentially expressed genes (log2 fold-change greater than 0.1, Benjamini–Hochberg [BH]-corrected p-value 0.05) when comparing cells in given cluster versus all remaining cells in the dataset. Cell clusters were ‘mapped’ to a reference cluster in the Davie dataset (whole brain reference) if at least 13/20 top markers were shared. Likewise, a cluster was mapped in the Özel dataset (optic lobe reference) if at least 7/20 top markers were shared. These cutoffs were empirically determined by maximizing the number of best matches. Cell cluster markers for our dataset are listed in Figure 1—source data 2. Third, a trained neural network classifier for adult neurons as described in Özel et al., 2021 was implemented in Python to label optic lobe neurons in our dataset. Log-normalized expression of 533 genes (out of the 587 genes in the Özel adult training set) across all cells were used as the input for the classifier. Finally, we checked for positive expression of well-established cell markers in each cluster (Figure 1—source data 4, Figure 1—figure supplement 2B, and Figure 1—figure supplement 3), using published cell marker datasets (Davie et al., 2018; Konstantinides et al., 2018; Kremer et al., 2017). Most cell cluster annotations were evaluated and consolidated based on best agreement across two or more approaches within or across the Davie and Özel references. Less certain annotations were visually inspected in UMAP space to check for proximity with adjacent clusters and manually evaluated for cell marker expression. Cell-level confidence for scmap assignment (similarity score) or neural network classifications (confidence score) were also manually evaluated. Results from the Özel reference was prioritized for optic lobe neurons, especially for cell clusters that may be heterogeneous in the Davie reference (Dm8/Tm5c, TmY14, Tm9, Tm5ab, Mt1). Pm neurons, chiasm glia, and subperineurial glia did not reach consensus across two or more approaches and were thus deemed less confident annotations. Several other optic lobe cell types were well mapped in a single approach to the Özel et al., 2021 dataset (TmY8, TmY3, Tm5c, Tm5ab, Tm20, Dm2, Dm8, Mi9, LC12, and LC17), where robust metrics were observed from the optic lobe neural network predictor or with scmap. Confirming our cell identity correspondences with the published scRNAseq datasets, we found high correlation among normalized gene expression when comparing individual cells in our dataset with the cluster-level means of the transcriptome in reference clusters as computed by cosine similarity (Figure 1—figure supplements 1 and 2C and D). The cosine similarity score between each annotated cell in our dataset and its cluster-level counterpart in the Davie or Özel references were computed based on shared non-dropout (count >0) genes, that is, the transcriptome of each cell in our data was correlated to the cluster-level mean of corresponding genes in a reference cluster. Lower similarity scores may reflect gene expression changes induced by tau pathology, less confident annotation (in this study or in the references used), clustering resolution differences, or high variance in the reference cluster.

To annotate the replication scRNAseq data (69,128 cells), labels from the completed dataset above (48,111 cells) were transferred using the Seurat v3 FindTransferAnchors and TransferData functions. In brief, pairs of similar cells between the reference and query dataset were identified using a mutual nearest-neighbor approach after projecting the replication dataset onto the reference dataset in PCA-reduced space. The 5000 most variable genes in the new dataset were used for the dimensional reduction. Each cell was assigned a score and a predicted label from the reference dataset. Cell-level metadata for both the discovery and replication datasets are uploaded to the Synapse AMP-AD Knowledge Portal as noted in the Key Resources Table.

After annotation, cell counts for each assigned cluster (90 clusters) were first quantified per library (six libraries, ages: days 1, 10, and 20; genotypes: control, tau), and treated as count data. To adjust for extreme proportional differences in cell composition across libraries and differences in the total number of cells captured per library, cell counts were normalized using negative binomial generalized linear models (NB-GLM) as implemented in DESeq2 (Love et al., 2014). In brief, raw cell counts were modeled using NB-GLM with a fitted mean and a cluster-specific dispersion factor. Dispersion factors were computed based on mean count values using an empirical Bayes approach as described in Love et al., 2014. The fitted mean is composed of a library-specific size factor and a parameter proportional to the true counts in each cluster per library. To compute size factors per library, raw counts were organized in a matrix such that rows represent clusters and columns represent samples (libraries). Raw counts were first divided by the row-wise geometric means and then divided by the per-column median of resulting quotients (size factor) to obtain normalized cell count values per cluster. These normalized cell counts were used to generate the plots in Figure 2B. The three age groups for each genotype (days 1, 10, 20) were combined to produce an n = 3 comparison of cell counts between tau and control animals.

Deconvolution of fly RNA-sequencing data

Request a detailed protocol

Deconvolving bulk-tissue RNA-sequencing data into estimated proportions of cell populations was performed by implementing Multi-subject Single-cell Deconvolution (Wang et al., 2019) using default parameters. MuSiC leverages cell-specific expression data from annotated scRNA-seq datasets and weighted non-negative least-squares regression to characterize cell compositions of bulk tissue gene expression data. This approach accounted for gene expression variability across samples and cells, thus upweighting the most consistently expressed genes across samples or cells for deconvolution. Whole-head RNA-sequencing counts of experimental conditions identical to those in this study were taken from Mangleburg et al., 2020 and used as input for deconvolution. Specifically, cell counts for n = 2 control and n = 3 tauR406W samples at days 1, 10, and 20 were deconvolved (15 samples total, each sample is a homogenate of 100 heads). 56,902 cells from a published Drosophila whole-brain scRNAseq dataset was used as an orthogonal reference for deconvolution, providing cell-specific transcriptional profiles from wildtype control animals (w1118 and DGRP-551) (Davie et al., 2018). Individual scRNA libraries were treated as subjects in the MuSiC pipeline for evaluating gene expression variability in marker gene weighting. Both annotated and unannotated cell clusters in the reference scRNAseq dataset were included in the deconvolution pipeline. Select cell clusters with non-zero estimated proportions across two or more timepoints are plotted in Figure 2—figure supplement 2.

Immunofluorescence and confocal microscopy

Request a detailed protocol

10-day-old female controls (elav-GAL4/+) or elav>tauR406W (elav-GAL4/+; UAS-tauR406W/+) were used for glial quantification immunofluorescence experiments. The animals were anesthetized with CO2 and brains were dissected with forceps and fixed in 4% paraformaldehyde (PFA) overnight at 4°C. After fixation, PFA was aspirated and replaced with PBS with 2% Triton-X (PBST) and incubated at 4°C overnight for tissue penetration. Residual air trapped in brain tissues were removed by placing samples under a vacuum for 1 hr at room temperature. The brains were then incubated in blocking solution (5% normal goat serum in PBST) at room temperature, rocking for 1 hr. Primary antibodies were diluted in 0.3% PBST and samples were incubated in primary at 4°C, rocking for at least 24 hr. The primary antibody solution was aspirated, and the samples were washed with PBST (two quick washes followed by three 15 min washes). Samples were incubated in secondary antibodies at room temperature, rocking for 2 hr. The secondary antibody solution was then aspirated and the samples were washed with PBST (two quick washes followed by three 15 min washes). DAPI stain, when applicable, was added in the secondary antibody step. Whole brains were then mounted in Vectashield antifade mounting medium (Vector Laboratories, H-1000-10) and stored in the dark at 4°C until imaged. Samples were imaged on a Leica Microsystems SP8X confocal microscope. Z-stacks covered the entirety of whole-mount brains. We used the following antibodies and dilutions: mouse anti-Repo (8D12, 1:500 for glial quantification experiment, 1:50 for Rel experiment, DSHB), rat anti-Elav (7E8A10, 1:100, DSHB); rabbit anti-GFP (1:500; GeneTex), mouse anti-Rel (21F3, 1:500, DSHB), conjugated A488-Phalloidin (1:500; Thermo Fisher), Cy3 AffiniPure goat anti-mouse (H+L) (1:500; Jackson ImmunoResearch Laboratories), Alexa 647-conjugated goat anti-Rabbit IgG (1:500; Jackson ImmunoResearch), Alexa Fluor 488 donkey anti-mouse IgG (H+L) (1:500; Jackson ImmunoResearch), and Cy3 AffiniPure goat anti-rat IgG (H+L) (1:500; Jackson ImmunoResearch).

Quantification of glia from confocal immunofluorescence digital microscopy was performed using Imaris (v9.9.1) imaging software. We counted the total number of repo-positive cells using the ‘spots’ object and automatic detection parameters with local thresholding and background subtraction. Brain volume was determined by using the ‘surfaces’ object on the Phalloidin channel to encompass the entire three-dimensional volume of the brain. Graphs of raw repo-positive counts per brain as well as glial density (repo-positive counts divided by brain volume) were created in GraphPad Prism (v9.4.1) software. Glial quantifications were performed using full Z-stacks of whole-mount brains. Welch’s t-test was used for comparisons between control and tauR406W animals (n = 9 animals per group). The significance threshold was set to p<0.05. Error bars represent the 95% confidence interval.

Mean pixel intensity for DAPI, phalloidin, or repo signal was calculated for n = 9 brains per genotype using ImageJ/Fiji (units: corrected total cell fluorescence [CTCF]). In GraphPad Prism, the mean pixel intensity for each channel (DAPI, phalloidin, or repo) was compared between control vs. tau-expressing animals using parametric, unpaired, two-tailed t-tests. All experimental groups passed the Shapiro–Wilk test for normality except for phalloidin intensity in tau-expressing animals, so this comparison (mean phalloidin intensity in control vs. tau-expressing animals) was done with a nonparametric (Mann–Whitney) t-test. The BIOP JACoP plugin on ImageJ/Fiji was used to calculate colocalization between relish-GFP signal and elav or repo signal. Area of overlap between signals (in pixels) was calculated for each slice (n = 85 slices total) using a stack histogram as the threshold. The area of overlap was then divided by the total elav or repo area to find the percentage of elav or repo area that was also positive for relish-GFP signal.

Bulk-tissue RNA-sequencing data

Request a detailed protocol

Bulk-RNA sequencing data and WGCNA co-expression modules of the experimental conditions described in this study were obtained from Mangleburg et al., 2020. WGCNA module expression activity in scRNAseq was computed by taking the mean of module member genes within each cell. Cluster-level expression activity of WGCNA modules was then estimated by averaging the cell-level activity across all cells in a given cluster and was subsequently used to compute the log2 fold change of tau versus control expression activity.

Drosophila NFκB signaling mediators

Request a detailed protocol

A list of Drosophila NFκB signaling pathway members was generated based on manual curation from published studies (Valanne et al., 2011; Myllymäki et al., 2014; Kounatidis et al., 2017; Li et al., 2020) and included the following genes: PGRP-LE, imd, Tak1, key, Rel, eff, PGRP-LC, bsk, akirin, Jra, sick, Tab2, IKKbeta, Uev1A, ben, Dredd, Fadd, PGRP-LA, Diap2, Diap1, egr, Traf6, trbd, pirk, casp, PGRP-LB, PGRP-LF, dnr1, scny, RYBP, PGRP-SC1a, PGRP-SC1b, PGRP-SC2, CYLD, POSH, spirit, spheroide, spz, PGRP-SA, PGRP-SD, pll, Myd88, dl, Gprk2, Deaf1, Tl, psh, grass, modSP, Dif, mop, tub, cact, nec, Pli, 18w, Toll-4, Tehao, Toll-6, Toll-7, Tollo, and Toll-9.

Analysis of differential cellular abundance

Request a detailed protocol

Statistical testing of the log2 fold change (log2FC) of tau versus control normalized cell abundance was performed using negative binomial-generalized linear models (NB-GLM) as implemented in DESeq2. Age was treated as a covariate, and a Wald test was performed on the coefficient of the genotype variable using the following model: log2(cell count) ~ age + genotype. Using DESeq2, log2FC was computed for each cluster (elav>tauR406W vs. control) based on maximum-likelihood estimation after fitting the GLM. Raw log2FC values were transformed using an adaptive shrinkage estimator from the ‘ashr’ R package as implemented in DESeq2 to account for clusters with high dispersion or low counts. These transformed log2FC values were then used for cell abundance analysis and interpretation. A BH-adjusted p-value<0.05 was used to establish significance of Wald test statistic. In order to generate the plots for Figure 2B, normalized cell counts were obtained using the counts function in DESeq2. Results were visualized using box and whisker plots, including the following values: median, minimum/maximum, and lower/upper quartiles.

In order to better understand how relative changes might influence cell abundance estimates, we inferred confidence intervals for cell cluster log2FC values (Figure 2—figure supplement 3A). Based on experimental ground truth (Serrano-Pozo et al., 2013), the log2FC value for seven cell clusters (Ensheathing glia, Perineurial glia, Astrocyte-like glia, Cortex glia, Chiasm glia, Subperineurial glia, and Hemocytes) was centered to zero. Specifically, the value of each glial cluster was iteratively subtracted from the log2FC values for all other clusters, establishing a minimum and maximum log2FC value for all cell clusters. We predict that the true cell abundance falls within this computed range, after accounting for potential proportional influences. A range that includes zero thus suggests there may be no true change between tau and control.

Analysis of differential gene expression

Request a detailed protocol

Differential gene expression analyses were performed using Model-based Analysis of Single-cell Transcriptomics (MAST) for each cell cluster (Finak et al., 2015). In brief, generalized linear hurdle models were used to compute differential expression, where logistic regression was used to account for stochastic dropouts, and a Gaussian linear model was fitted to predict gene expression levels. Differential expression was determined by a likelihood ratio test. We required that differentially expressed genes meet a significance threshold of BH-adjusted p-value<0.05; absolute log2 fold-change > 0.1; and detectable (non-zero) expression in at least 10% of cells in the cluster. Cellular detection rate (CDR, fraction of genes reliably detected in each cell) was included as a covariate in all regression models, as in published protocols (Finak et al., 2015). CDR acts as a proxy for estimating the effect of dropout events, amplification efficiency, cell volume, and other extrinsic factors while performing expression-related regression analyses. Analyses of tau-induced differential expression (age-adjusted) also included age as a regression model covariate. Separately, aging-induced changes within each cell cluster were computed from control data (elav-GAL4/+), comparing differential gene expression between days 1 and 10, and days 10 and 20. In order to evaluate robustness and replicability, cross-sectional, tau-induced differentially expressed genes were also computed in day 10 animals (tau vs. control) in the replication data and results were compared between the discovery and replication datasets. Similarly, cross-sectional tau-induced differential expression was also computed for each timepoint in the discovery dataset. For differential expression of the human cell subclusters reported in Mathys et al., 2019, normalized counts between individuals with AD pathology (n = 24) and low/no pathology (n = 24) were compared using MAST for each cell subcluster as described above.

To assess cell-type-specific differences in regulon gene expression levels (Figure 4C), we performed differential regulon analysis using base statistical packages in R. In brief, mean expression of gene members in each regulon were computed per cell, and cell-type-matched comparisons were made between control and elav>tauR406W using linear regression, including age as a covariate. Specifically, for each cell type, a likelihood ratio test compared the fit of a full model (Regulon Expression ~ Genotype + age) and a reduced model (Regulon Expression ~ age), evaluating the contribution of genotype to model fit. Significance was set at a BH-adjusted p-value<0.05. Regulon log2-fold changes were computed for elav>tauR406W versus control mean expression in each cluster.

Overrepresentation analysis (ORA) of differentially expressed gene sets were completed using the R implementation of WEBGESTALT (Wang et al., 2013). The following databases were used: Gene Ontology (GO) biological processes, GO molecular functions, GO cellular component, KEGG, and Panther. Enrichment significance was defined by hypergeometric test, followed by p-value adjustment using the BH-procedure; significance was set at p<0.05. ORA of the tau unique gene set (n = 363 genes) was performed using gProfiler (Raudvere et al., 2019). The Gene Ontology (GO), Human phenotype ontology (HP), KEGG, miRTarBase (MIRNA), Transfac (TF), and WikiPathways (WP) databases were used for querying genes. The organism parameter was set to ‘dmelanogaster’ and the ‘fdr’ correction method was used to apply the BH multiple testing correction. A false discovery rate (FDR) < 0.05 was the threshold for significance.

Multiple regression with elastic net

Request a detailed protocol

To identify features driving cell vulnerability in our scRNAseq dataset, we pooled information across cell clusters by performing elastic net regression. For all clusters showing significant elav>tauR406W vs. control cell abundance changes, log2FC values were regressed on the cell-type-specific mean expression for 183 regulons. Given our goal to identify the factors that influence cell-type-specific vulnerability, we focused on eight cell clusters with significant cell loss (FDR < 0.05). In a secondary analysis, we repeated elastic net regression and considered a larger number of potential predictor variables including (i) the 183 regulons (as above); (ii) 2793 unique GO, KEGG, and Panther pathways found to be significantly enriched among elav>tauR406W differentially expressed genes; (iii) 7 WGCNA modules altered in elav>tauR406W (Mangleburg et al., 2020); and (iv) curated NFκB signaling pathways. In addition, we also considered a large number of (v) cell cluster technical parameters as potential predictors, including normalized cell counts, mean tau transgene expression, sum of UMIs, mean percent mitochondrial reads, and number of tau-induced differentially expressed genes (age-adjusted). For this analysis, all computed variables (e.g., GO pathways, WGCNA modules, regulons, NFκB genes) were first averaged within each cell, then averaged across all cells in order to determine a mean value for each cell cluster. Cluster-level means for all gene sets were computed using pooled cell data from both elav>tauR406W and controls and all ages. For gene sets derived from ORA analyses, we restricted consideration to those differentially expressed genes driving enrichment. We generated a matrix consisting of rows for each cell cluster and columns with values/means for each potential predictor variable.

We used the caret and glmnet packages in R to organize the data and perform elastic net regularized regression. Alpha (ridge vs. lasso characteristic) and lambda (shrinkage parameter) values were tuned in a 1000 × 1000 grid using repeated threefold cross-validation in caret, and the average root mean squared errors (RMSE) from testing the partitions were used to assess model performance. Threefold cross-validation was repeated 100 times for all alpha-lambda pairs using a different data fold split for each iteration in order to account for variability in model performance from random sample partitioning. The mean of all prediction errors was used to assess the final performance of each alpha-lambda pair, and we selected the model with the lowest RMSE. Lastly, we generated a ranklist of predictor variables for tau-induced cell abundance changes based on the magnitude of coefficients from the selected model (Figure 4—figure supplement 7B, Figure 4—source data 4).

Cross-species analysis with human and mouse RNAseq datasets

Request a detailed protocol

Previously published human snRNAseq data (Mathys et al., 2019; Grubman et al., 2019) were reprocessed and filtered using the identical pipeline as describe above for Drosophila data. Raw gene counts were normalized using the NormalizeData function in Seurat v3. The resulting 70,634 filtered cells from the Mathys data were re-clustered using our pipeline above for visual representation in Figure 5; however, the cell cluster annotations from the original publication were preserved. 13,214 filtered cells from the Grubman data were similarly promoted for analysis. Drosophila orthologs of genes detected in human datasets were determined using the DRSC Integrated Ortholog Prediction Tool (DIOPT) (Hu et al., 2011), requiring a minimum DIOPT score threshold of 5 or greater. If more than one fly ortholog was identified, we selected the ortholog with either (i) the highest DIOPT score, (ii) the highest weighted DIOPT score, or (iii) the highest ranked option (best score when mapped both forward and reverse). Thus, 5630 or 4145 human-fly gene ortholog pairs, respectively, were considered for cross-species analyses of the Mathys and Grubman datasets. We scaled normalized expression of each gene with mean = 0 and variance = 1. Cluster-level gene expression was computed by averaging scaled expression values from all cells. Subsequently, we performed Pearson correlation analysis for all cluster pairs to quantify transcriptional similarities between fly and human cell, examining pairwise correlation coefficients for all gene-orthologs across all clusters. For visualization, we generated heatmaps representing Pearson correlation coefficients by seriation with hierarchical clustering. Association statistics for human neuropathological traits (heatmap at top of Figure 5A) were repurposed directly from the published supplementary from Mathys et al., 2019. For quantification of overlap between human microglia and fly ensheathing glia, we examined conserved differentially expressed genes using the hypergeometric overlap test. To demonstrate control-only correlations, scRNAseq from the filtered Davie et al. dataset (subsetted for annotated cell types) were compared to snRNAseq profiles of 24 control individuals in the Mathys et al. data. To compute Rel regulon differential expression in a tauopathy mouse model, we used normalized scRNAseq pseudobulk counts from MAPT P301L mice (n = 3) and non-transgenic controls (n = 2) as published in Lee et al., 2021. Conserved mice genes (DIOPT > 4) in the fly Rel regulon were averaged per cell cluster (554 mice genes mapped to at least one fly ortholog) and a likelihood ratio test was used to evaluate the contribution of genotype to differential expression in each cluster.

Data availability

All original single cell sequencing data have been uploaded to the Accelerating Medicines Parternship (AMP)-AD Knowledge Portal on Synapse and can be accessed through the DOI: https://doi.org/10.7303/syn35798807.1.

The following data sets were generated
    1. Wu T
    2. Deger JM
    3. Ye H
    4. Guo C
    5. Dhindsa J
    6. Pekarek BT
    7. Al-Ouran R
    8. Liu Z
    9. Al-Ramahi I
    10. Botas J
    11. Shulman JM
    (2022) Synapse
    Tau polarizes an aging transcriptional signature to excitatory neurons and glia.
    https://doi.org/10.7303/syn35798807.1
The following previously published data sets were used
    1. Mathys H
    (2019) Synapse
    The Single-cell transcriptomic analysis of Alzheimer's disease (snRNAseqPFC_BA10) Study.
    https://doi.org/10.7303/syn18485175
    1. Grubman, et al
    (2019) NCBI Gene Expression Omnibus
    ID GSE138852. A single-cell atlas of the human cortex reveals drivers of transcriptional changes in Alzheimer's disease in specific cell subpopulations.
    1. Davie, et al
    (2018) NCBI Gene Expression Omnibus
    ID GSE107451. A single-cell transcriptome atlas of the ageing Drosophila brain.
    1. Ozel, et al
    (2020) NCBI Gene Expression Omnibus
    ID GSE142789. Neuronal diversity and convergence in a visual system developmental atlas.
    1. Mangleburg, et al
    (2020) Synapse
    Integrated analysis of the aging brain transcriptome and proteome in tauopathy.
    https://doi.org/10.7303/syn7274101
    1. Van de Sande B
    (2020) Aerts lab resource
    ID allTFs_dmel. pySCENIC: List of Drosophila transcription factors.
    1. Van de Sande B
    (2020) Aerts lab resource
    ID m0.001-o0.0.tbl. psySCEnIC: Transcription factor to motif annotations.

References

    1. Bellenguez C
    2. Küçükali F
    3. Jansen IE
    4. Kleineidam L
    5. Moreno-Grau S
    6. Amin N
    7. Naj AC
    8. Campos-Martin R
    9. Grenier-Boley B
    10. Andrade V
    11. Holmans PA
    12. Boland A
    13. Damotte V
    14. van der Lee SJ
    15. Costa MR
    16. Kuulasmaa T
    17. Yang Q
    18. de Rojas I
    19. Bis JC
    20. Yaqub A
    21. Prokic I
    22. Chapuis J
    23. Ahmad S
    24. Giedraitis V
    25. Aarsland D
    26. Garcia-Gonzalez P
    27. Abdelnour C
    28. Alarcón-Martín E
    29. Alcolea D
    30. Alegret M
    31. Alvarez I
    32. Álvarez V
    33. Armstrong NJ
    34. Tsolaki A
    35. Antúnez C
    36. Appollonio I
    37. Arcaro M
    38. Archetti S
    39. Pastor AA
    40. Arosio B
    41. Athanasiu L
    42. Bailly H
    43. Banaj N
    44. Baquero M
    45. Barral S
    46. Beiser A
    47. Pastor AB
    48. Below JE
    49. Benchek P
    50. Benussi L
    51. Berr C
    52. Besse C
    53. Bessi V
    54. Binetti G
    55. Bizarro A
    56. Blesa R
    57. Boada M
    58. Boerwinkle E
    59. Borroni B
    60. Boschi S
    61. Bossù P
    62. Bråthen G
    63. Bressler J
    64. Bresner C
    65. Brodaty H
    66. Brookes KJ
    67. Brusco LI
    68. Buiza-Rueda D
    69. Bûrger K
    70. Burholt V
    71. Bush WS
    72. Calero M
    73. Cantwell LB
    74. Chene G
    75. Chung J
    76. Cuccaro ML
    77. Carracedo Á
    78. Cecchetti R
    79. Cervera-Carles L
    80. Charbonnier C
    81. Chen H-H
    82. Chillotti C
    83. Ciccone S
    84. Claassen JAHR
    85. Clark C
    86. Conti E
    87. Corma-Gómez A
    88. Costantini E
    89. Custodero C
    90. Daian D
    91. Dalmasso MC
    92. Daniele A
    93. Dardiotis E
    94. Dartigues J-F
    95. de Deyn PP
    96. de Paiva Lopes K
    97. de Witte LD
    98. Debette S
    99. Deckert J
    100. del Ser T
    101. Denning N
    102. DeStefano A
    103. Dichgans M
    104. Diehl-Schmid J
    105. Diez-Fairen M
    106. Rossi PD
    107. Djurovic S
    108. Duron E
    109. Düzel E
    110. Dufouil C
    111. Eiriksdottir G
    112. Engelborghs S
    113. Escott-Price V
    114. Espinosa A
    115. Ewers M
    116. Faber KM
    117. Fabrizio T
    118. Nielsen SF
    119. Fardo DW
    120. Farotti L
    121. Fenoglio C
    122. Fernández-Fuertes M
    123. Ferrari R
    124. Ferreira CB
    125. Ferri E
    126. Fin B
    127. Fischer P
    128. Fladby T
    129. Fließbach K
    130. Fongang B
    131. Fornage M
    132. Fortea J
    133. Foroud TM
    134. Fostinelli S
    135. Fox NC
    136. Franco-Macías E
    137. Bullido MJ
    138. Frank-García A
    139. Froelich L
    140. Fulton-Howard B
    141. Galimberti D
    142. García-Alberca JM
    143. García-González P
    144. Garcia-Madrona S
    145. Garcia-Ribas G
    146. Ghidoni R
    147. Giegling I
    148. Giorgio G
    149. Goate AM
    150. Goldhardt O
    151. Gomez-Fonseca D
    152. González-Pérez A
    153. Graff C
    154. Grande G
    155. Green E
    156. Grimmer T
    157. Grünblatt E
    158. Grunin M
    159. Gudnason V
    160. Guetta-Baranes T
    161. Haapasalo A
    162. Hadjigeorgiou G
    163. Haines JL
    164. Hamilton-Nelson KL
    165. Hampel H
    166. Hanon O
    167. Hardy J
    168. Hartmann AM
    169. Hausner L
    170. Harwood J
    171. Heilmann-Heimbach S
    172. Helisalmi S
    173. Heneka MT
    174. Hernández I
    175. Herrmann MJ
    176. Hoffmann P
    177. Holmes C
    178. Holstege H
    179. Vilas RH
    180. Hulsman M
    181. Humphrey J
    182. Biessels GJ
    183. Jian X
    184. Johansson C
    185. Jun GR
    186. Kastumata Y
    187. Kauwe J
    188. Kehoe PG
    189. Kilander L
    190. Ståhlbom AK
    191. Kivipelto M
    192. Koivisto A
    193. Kornhuber J
    194. Kosmidis MH
    195. Kukull WA
    196. Kuksa PP
    197. Kunkle BW
    198. Kuzma AB
    199. Lage C
    200. Laukka EJ
    201. Launer L
    202. Lauria A
    203. Lee C-Y
    204. Lehtisalo J
    205. Lerch O
    206. Lleó A
    207. Longstreth W Jr
    208. Lopez O
    209. de Munain AL
    210. Love S
    211. Löwemark M
    212. Luckcuck L
    213. Lunetta KL
    214. Ma Y
    215. Macías J
    216. MacLeod CA
    217. Maier W
    218. Mangialasche F
    219. Spallazzi M
    220. Marquié M
    221. Marshall R
    222. Martin ER
    223. Montes AM
    224. Rodríguez CM
    225. Masullo C
    226. Mayeux R
    227. Mead S
    228. Mecocci P
    229. Medina M
    230. Meggy A
    231. Mehrabian S
    232. Mendoza S
    233. Menéndez-González M
    234. Mir P
    235. Moebus S
    236. Mol M
    237. Molina-Porcel L
    238. Montrreal L
    239. Morelli L
    240. Moreno F
    241. Morgan K
    242. Mosley T
    243. Nöthen MM
    244. Muchnik C
    245. Mukherjee S
    246. Nacmias B
    247. Ngandu T
    248. Nicolas G
    249. Nordestgaard BG
    250. Olaso R
    251. Orellana A
    252. Orsini M
    253. Ortega G
    254. Padovani A
    255. Paolo C
    256. Papenberg G
    257. Parnetti L
    258. Pasquier F
    259. Pastor P
    260. Peloso G
    261. Pérez-Cordón A
    262. Pérez-Tur J
    263. Pericard P
    264. Peters O
    265. Pijnenburg YAL
    266. Pineda JA
    267. Piñol-Ripoll G
    268. Pisanu C
    269. Polak T
    270. Popp J
    271. Posthuma D
    272. Priller J
    273. Puerta R
    274. Quenez O
    275. Quintela I
    276. Thomassen JQ
    277. Rábano A
    278. Rainero I
    279. Rajabli F
    280. Ramakers I
    281. Real LM
    282. Reinders MJT
    283. Reitz C
    284. Reyes-Dumeyer D
    285. Ridge P
    286. Riedel-Heller S
    287. Riederer P
    288. Roberto N
    289. Rodriguez-Rodriguez E
    290. Rongve A
    291. Allende IR
    292. Rosende-Roca M
    293. Royo JL
    294. Rubino E
    295. Rujescu D
    296. Sáez ME
    297. Sakka P
    298. Saltvedt I
    299. Sanabria Á
    300. Sánchez-Arjona MB
    301. Sanchez-Garcia F
    302. Juan PS
    303. Sánchez-Valle R
    304. Sando SB
    305. Sarnowski C
    306. Satizabal CL
    307. Scamosci M
    308. Scarmeas N
    309. Scarpini E
    310. Scheltens P
    311. Scherbaum N
    312. Scherer M
    313. Schmid M
    314. Schneider A
    315. Schott JM
    316. Selbæk G
    317. Seripa D
    318. Serrano M
    319. Sha J
    320. Shadrin AA
    321. Skrobot O
    322. Slifer S
    323. Snijders GJL
    324. Soininen H
    325. Solfrizzi V
    326. Solomon A
    327. Song Y
    328. Sorbi S
    329. Sotolongo-Grau O
    330. Spalletta G
    331. Spottke A
    332. Squassina A
    333. Stordal E
    334. Tartan JP
    335. Tárraga L
    336. Tesí N
    337. Thalamuthu A
    338. Thomas T
    339. Tosto G
    340. Traykov L
    341. Tremolizzo L
    342. Tybjærg-Hansen A
    343. Uitterlinden A
    344. Ullgren A
    345. Ulstein I
    346. Valero S
    347. Valladares O
    348. Broeckhoven CV
    349. Vance J
    350. Vardarajan BN
    351. van der Lugt A
    352. Dongen JV
    353. van Rooij J
    354. van Swieten J
    355. Vandenberghe R
    356. Verhey F
    357. Vidal J-S
    358. Vogelgsang J
    359. Vyhnalek M
    360. Wagner M
    361. Wallon D
    362. Wang L-S
    363. Wang R
    364. Weinhold L
    365. Wiltfang J
    366. Windle G
    367. Woods B
    368. Yannakoulia M
    369. Zare H
    370. Zhao Y
    371. Zhang X
    372. Zhu C
    373. Zulaica M
    374. EADB
    375. Laczo J
    376. Matoska V
    377. Serpente M
    378. Assogna F
    379. Piras F
    380. Piras F
    381. Ciullo V
    382. Shofany J
    383. Ferrarese C
    384. Andreoni S
    385. Sala G
    386. Zoia CP
    387. Zompo MD
    388. Benussi A
    389. Bastiani P
    390. Takalo M
    391. Natunen T
    392. Laatikainen T
    393. Tuomilehto J
    394. Antikainen R
    395. Strandberg T
    396. Lindström J
    397. Peltonen M
    398. Abraham R
    399. Al-Chalabi A
    400. Bass NJ
    401. Brayne C
    402. Brown KS
    403. Collinge J
    404. Craig D
    405. Deloukas P
    406. Fox N
    407. Gerrish A
    408. Gill M
    409. Gwilliam R
    410. Harold D
    411. Hollingworth P
    412. Johnston JA
    413. Jones L
    414. Lawlor B
    415. Livingston G
    416. Lovestone S
    417. Lupton M
    418. Lynch A
    419. Mann D
    420. McGuinness B
    421. McQuillin A
    422. O’Donovan MC
    423. Owen MJ
    424. Passmore P
    425. Powell JF
    426. Proitsi P
    427. Rossor M
    428. Shaw CE
    429. Smith AD
    430. Gurling H
    431. Todd S
    432. Mummery C
    433. Ryan N
    434. Lacidogna G
    435. Adarmes-Gómez A
    436. Mauleón A
    437. Pancho A
    438. Gailhajenet A
    439. Lafuente A
    440. Macias-García D
    441. Martín E
    442. Pelejà E
    443. Carrillo F
    444. Merlín IS
    445. Garrote-Espina L
    446. Vargas L
    447. Carrion-Claro M
    448. Marín M
    449. Labrador M
    450. Buendia M
    451. Alonso MD
    452. Guitart M
    453. Moreno M
    454. Ibarria M
    455. Periñán M
    456. Aguilera N
    457. Gómez-Garre P
    458. Cañabate P
    459. Escuela R
    460. Pineda-Sánchez R
    461. Vigo-Ortega R
    462. Jesús S
    463. Preckler S
    464. Rodrigo-Herrero S
    465. Diego S
    466. Vacca A
    467. Roveta F
    468. Salvadori N
    469. Chipi E
    470. Boecker H
    471. Laske C
    472. Perneczky R
    473. Anastasiou C
    474. Janowitz D
    475. Malik R
    476. Anastasiou A
    477. Parveen K
    478. Lage C
    479. López-García S
    480. Antonell A
    481. Mihova KY
    482. Belezhanska D
    483. Weber H
    484. Kochen S
    485. Solis P
    486. Medel N
    487. Lisso J
    488. Sevillano Z
    489. Politis DG
    490. Cores V
    491. Cuesta C
    492. Ortiz C
    493. Bacha JI
    494. Rios M
    495. Saenz A
    496. Abalos MS
    497. Kohler E
    498. Palacio DL
    499. Etchepareborda I
    500. Kohler M
    501. Novack G
    502. Prestia FA
    503. Galeano P
    504. Castaño EM
    505. Germani S
    506. Toso CR
    507. Rojo M
    508. Ingino C
    509. Mangone C
    510. Rubinsztein DC
    511. Teipel S
    512. Fievet N
    513. Deramerourt V
    514. Forsell C
    515. Thonberg H
    516. Bjerke M
    517. Roeck ED
    518. Martínez-Larrad MT
    519. Olivar N
    520. GR@ACE
    521. Aguilera N
    522. Cano A
    523. Cañabate P
    524. Macias J
    525. Maroñas O
    526. Nuñez-Llaves R
    527. Olivé C
    528. Pelejá E
    529. DEGESCO
    530. Adarmes-Gómez AD
    531. Alonso MD
    532. Amer-Ferrer G
    533. Antequera M
    534. Burguera JA
    535. Carrillo F
    536. Carrión-Claro M
    537. Casajeros MJ
    538. Martinez de Pancorbo M
    539. Escuela R
    540. Garrote-Espina L
    541. Gómez-Garre P
    542. Hevilla S
    543. Jesús S
    544. Espinosa MAL
    545. Legaz A
    546. López-García S
    547. Macias-García D
    548. Manzanares S
    549. Marín M
    550. Marín-Muñoz J
    551. Marín T
    552. Martínez B
    553. Martínez V
    554. Martínez-Lage Álvarez P
    555. Iriarte MM
    556. Periñán-Tocino MT
    557. Pineda-Sánchez R
    558. Real de Asúa D
    559. Rodrigo S
    560. Sastre I
    561. Vicente MP
    562. Vigo-Ortega R
    563. Vivancos L
    564. EADI
    565. Epelbaum J
    566. Hannequin D
    567. campion D
    568. Deramecourt V
    569. Tzourio C
    570. Brice A
    571. Dubois B
    572. GERAD
    573. Williams A
    574. Thomas C
    575. Davies C
    576. Nash W
    577. Dowzell K
    578. Morales AC
    579. Bernardo-Harrington M
    580. Turton J
    581. Lord J
    582. Brown K
    583. Vardy E
    584. Fisher E
    585. Warren JD
    586. Rossor M
    587. Ryan NS
    588. Guerreiro R
    589. Uphill J
    590. Bass N
    591. Heun R
    592. Kölsch H
    593. Schürmann B
    594. Lacour A
    595. Herold C
    596. Johnston JA
    597. Passmore P
    598. Powell J
    599. Patel Y
    600. Hodges A
    601. Becker T
    602. Warden D
    603. Wilcock G
    604. Clarke R
    605. Deloukas P
    606. Ben-Shlomo Y
    607. Hooper NM
    608. Pickering-Brown S
    609. Sussams R
    610. Warner N
    611. Bayer A
    612. Heuser I
    613. Drichel D
    614. Klopp N
    615. Mayhaus M
    616. Riemenschneider M
    617. Pinchler S
    618. Feulner T
    619. Gu W
    620. van den Bussche H
    621. Hüll M
    622. Frölich L
    623. Wichmann H-E
    624. Jöckel K-H
    625. O’Donovan M
    626. Owen M
    627. Demgene
    628. Bahrami S
    629. Bosnes I
    630. Selnes P
    631. Bergh S
    632. FinnGen
    633. Palotie A
    634. Daly M
    635. Jacob H
    636. Matakidou A
    637. Runz H
    638. John S
    639. Plenge R
    640. McCarthy M
    641. Hunkapiller J
    642. Ehm M
    643. Waterworth D
    644. Fox C
    645. Malarstig A
    646. Klinger K
    647. Call K
    648. Behrens T
    649. Loerch P
    650. Mäkelä T
    651. Kaprio J
    652. Virolainen P
    653. Pulkki K
    654. Kilpi T
    655. Perola M
    656. Partanen J
    657. Pitkäranta A
    658. Kaarteenaho R
    659. Vainio S
    660. Turpeinen M
    661. Serpi R
    662. Laitinen T
    663. Mäkelä J
    664. Kosma V-M
    665. Kujala U
    666. Tuovila O
    667. Hendolin M
    668. Pakkanen R
    669. Waring J
    670. Riley-Gillis B
    671. Liu J
    672. Biswas S
    673. Diogo D
    674. Marshall C
    675. Hu X
    676. Gossel M
    677. Graham R
    678. Cummings B
    679. Ripatti S
    680. Schleutker J
    681. Arvas M
    682. Carpén O
    683. Hinttala R
    684. Kettunen J
    685. Mannermaa A
    686. Laukkanen J
    687. Julkunen V
    688. Remes A
    689. Kälviäinen R
    690. Peltola J
    691. Tienari P
    692. Rinne J
    693. Ziemann A
    694. Waring J
    695. Esmaeeli S
    696. Smaoui N
    697. Lehtonen A
    698. Eaton S
    699. Lahdenperä S
    700. van Adelsberg J
    701. Michon J
    702. Kerchner G
    703. Bowers N
    704. Teng E
    705. Eicher J
    706. Mehta V
    707. Gormley P
    708. Linden K
    709. Whelan C
    710. Xu F
    711. Pulford D
    712. Färkkilä M
    713. Pikkarainen S
    714. Jussila A
    715. Blomster T
    716. Kiviniemi M
    717. Voutilainen M
    718. Georgantas B
    719. Heap G
    720. Rahimov F
    721. Usiskin K
    722. Lu T
    723. Oh D
    724. Kalpala K
    725. Miller M
    726. McCarthy L
    727. Eklund K
    728. Palomäki A
    729. Isomäki P
    730. Pirilä L
    731. Kaipiainen-Seppänen O
    732. Huhtakangas J
    733. Lertratanakul A
    734. Hochfeld M
    735. Bing N
    736. Gordillo JE
    737. Mars N
    738. Pelkonen M
    739. Kauppi P
    740. Kankaanranta H
    741. Harju T
    742. Close D
    743. Greenberg S
    744. Chen H
    745. Betts J
    746. Ghosh S
    747. Salomaa V
    748. Niiranen T
    749. Juonala M
    750. Metsärinne K
    751. Kähönen M
    752. Junttila J
    753. Laakso M
    754. Pihlajamäki J
    755. Sinisalo J
    756. Taskinen M-R
    757. Tuomi T
    758. Challis B
    759. Peterson A
    760. Chu A
    761. Parkkinen J
    762. Muslin A
    763. Joensuu H
    764. Meretoja T
    765. Aaltonen L
    766. Mattson J
    767. Auranen A
    768. Karihtala P
    769. Kauppila S
    770. Auvinen P
    771. Elenius K
    772. Popovic R
    773. Schutzman J
    774. Loboda A
    775. Chhibber A
    776. Lehtonen H
    777. McDonough S
    778. Crohns M
    779. Kulkarni D
    780. Kaarniranta K
    781. Turunen JA
    782. Ollila T
    783. Seitsonen S
    784. Uusitalo H
    785. Aaltonen V
    786. Uusitalo-Järvinen H
    787. Luodonpää M
    788. Hautala N
    789. Loomis S
    790. Strauss E
    791. Chen H
    792. Podgornaia A
    793. Hoffman J
    794. Tasanen K
    795. Huilaja L
    796. Hannula-Jouppi K
    797. Salmi T
    798. Peltonen S
    799. Koulu L
    800. Harvima I
    801. Wu Y
    802. Choy D
    803. Pussinen P
    804. Salminen A
    805. Salo T
    806. Rice D
    807. Nieminen P
    808. Palotie U
    809. Siponen M
    810. Suominen L
    811. Mäntylä P
    812. Gursoy U
    813. Anttonen V
    814. Sipilä K
    815. Davis JW
    816. Quarless D
    817. Petrovski S
    818. Wigmore E
    819. Chen C-Y
    820. Bronson P
    821. Tsai E
    822. Huang Y
    823. Maranville J
    824. Shaikho E
    825. Mohammed E
    826. Wadhawan S
    827. Kvikstad E
    828. Caliskan M
    829. Chang D
    830. Bhangale T
    831. Pendergrass S
    832. Holzinger E
    833. Chen X
    834. Hedman Å
    835. King KS
    836. Wang C
    837. Xu E
    838. Auge F
    839. Chatelain C
    840. Rajpal D
    841. Liu D
    842. Call K
    843. Xia T
    844. Brauer M
    845. Kurki M
    846. Karjalainen J
    847. Havulinna A
    848. Jalanko A
    849. Palta P
    850. della Briotta Parolo P
    851. Zhou W
    852. Lemmelä S
    853. Rivas M
    854. Harju J
    855. Lehisto A
    856. Ganna A
    857. Llorens V
    858. Laivuori H
    859. Rüeger S
    860. Niemi ME
    861. Tukiainen T
    862. Reeve MP
    863. Heyne H
    864. Palin K
    865. Garcia-Tabuenca J
    866. Siirtola H
    867. Kiiskinen T
    868. Lee J
    869. Tsuo K
    870. Elliott A
    871. Kristiansson K
    872. Hyvärinen K
    873. Ritari J
    874. Koskinen M
    875. Pylkäs K
    876. Kalaoja M
    877. Karjalainen M
    878. Mantere T
    879. Kangasniemi E
    880. Heikkinen S
    881. Laakkonen E
    882. Sipeky C
    883. Heron S
    884. Karlsson A
    885. Jambulingam D
    886. Rathinakannan VS
    887. Kajanne R
    888. Aavikko M
    889. Jiménez MG
    890. della Briotta Parola P
    891. Lehistö A
    892. Kanai M
    893. Kaunisto M
    894. Kilpeläinen E
    895. Sipilä TP
    896. Brein G
    897. Awaisa G
    898. Shcherban A
    899. Donner K
    900. Loukola A
    901. Laiho P
    902. Sistonen T
    903. Kaiharju E
    904. Laukkanen M
    905. Järvensivu E
    906. Lähteenmäki S
    907. Männikkö L
    908. Wong R
    909. Mattsson H
    910. Hiekkalinna T
    911. Paajanen T
    912. Pärn K
    913. Gracia-Tabuenca J
    914. ADGC
    915. Abner E
    916. Adams PM
    917. Aguirre A
    918. Albert MS
    919. Albin RL
    920. Allen M
    921. Alvarez L
    922. Apostolova LG
    923. Arnold SE
    924. Asthana S
    925. Atwood CS
    926. Ayres G
    927. Baldwin CT
    928. Barber RC
    929. Barnes LL
    930. Barral S
    931. Beach TG
    932. Becker JT
    933. Beecham GW
    934. Beekly D
    935. Below JE
    936. Benchek P
    937. Benitez BA
    938. Bennett D
    939. Bertelson J
    940. Margaret FE
    941. Bird TD
    942. Blacker D
    943. Boeve BF
    944. Bowen JD
    945. Boxer A
    946. Brewer J
    947. Burke JR
    948. Burns JM
    949. Bush WS
    950. Buxbaum JD
    951. Cairns NJ
    952. Cao C
    953. Carlson CS
    954. Carlsson CM
    955. Carney RM
    956. Carrasquillo MM
    957. Chasse S
    958. Chesselet M-F
    959. Chesi A
    960. Chin NA
    961. Chui HC
    962. Chung J
    963. Craft S
    964. Crane PK
    965. Cribbs DH
    966. Crocco EA
    967. Cruchaga C
    968. Cuccaro ML
    969. Cullum M
    970. Darby E
    971. Davis B
    972. De Jager PL
    973. DeCarli C
    974. DeToledo J
    975. Dick M
    976. Dickson DW
    977. Dombroski BA
    978. Doody RS
    979. Duara R
    980. Ertekin-Taner N
    981. Evans DA
    982. Fairchild TJ
    983. Fallon KB
    984. Farlow MR
    985. Farrell JJ
    986. Fernandez-Hernandez V
    987. Ferris S
    988. Frosch MP
    989. Fulton-Howard B
    990. Galasko DR
    991. Gamboa A
    992. Gearing M
    993. Geschwind DH
    994. Ghetti B
    995. Gilbert JR
    996. Grabowski TJ
    997. Graff-Radford NR
    998. Grant SFA
    999. Green RC
    1000. Growdon JH
    1001. Haines JL
    1002. Hakonarson H
    1003. Hall J
    1004. Hamilton RL
    1005. Harari O
    1006. Harrell LE
    1007. Haut J
    1008. Head E
    1009. Henderson VW
    1010. Hernandez M
    1011. Hohman T
    1012. Honig LS
    1013. Huebinger RM
    1014. Huentelman MJ
    1015. Hulette CM
    1016. Hyman BT
    1017. Hynan LS
    1018. Ibanez L
    1019. Jarvik GP
    1020. Jayadev S
    1021. Jin L-W
    1022. Johnson K
    1023. Johnson L
    1024. Kamboh MI
    1025. Karydas AM
    1026. Katz MJ
    1027. Kaye JA
    1028. Keene CD
    1029. Khaleeq A
    1030. Kim R
    1031. Knebl J
    1032. Kowall NW
    1033. Kramer JH
    1034. Kuksa PP
    1035. LaFerla FM
    1036. Lah JJ
    1037. Larson EB
    1038. Lee C-Y
    1039. Lee EB
    1040. Lerner A
    1041. Leung YY
    1042. Leverenz JB
    1043. Levey AI
    1044. Li M
    1045. Lieberman AP
    1046. Lipton RB
    1047. Logue M
    1048. Lyketsos CG
    1049. Malamon J
    1050. Mains D
    1051. Marson DC
    1052. Martiniuk F
    1053. Mash DC
    1054. Masliah E
    1055. Massman P
    1056. Masurkar A
    1057. McCormick WC
    1058. McCurry SM
    1059. McDavid AN
    1060. McDonough S
    1061. McKee AC
    1062. Mesulam M
    1063. Mez J
    1064. Miller BL
    1065. Miller CA
    1066. Miller JW
    1067. Montine TJ
    1068. Monuki ES
    1069. Morris JC
    1070. Myers AJ
    1071. Nguyen T
    1072. O’Bryant S
    1073. Olichney JM
    1074. Ory M
    1075. Palmer R
    1076. Parisi JE
    1077. Paulson HL
    1078. Pavlik V
    1079. Paydarfar D
    1080. Perez V
    1081. Peskind E
    1082. Petersen RC
    1083. Phillips-Cremins JE
    1084. Pierce A
    1085. Polk M
    1086. Poon WW
    1087. Potter H
    1088. Qu L
    1089. Quiceno M
    1090. Quinn JF
    1091. Raj A
    1092. Raskind M
    1093. Reiman EM
    1094. Reisberg B
    1095. Reisch JS
    1096. Ringman JM
    1097. Roberson ED
    1098. Rodriguear M
    1099. Rogaeva E
    1100. Rosen HJ
    1101. Rosenberg RN
    1102. Royall DR
    1103. Sager MA
    1104. Sano M
    1105. Saykin AJ
    1106. Schneider JA
    1107. Schneider LS
    1108. Seeley WW
    1109. Slifer SH
    1110. Small S
    1111. Smith AG
    1112. Smith JP
    1113. Song YE
    1114. Sonnen JA
    1115. Spina S
    1116. George-Hyslop PS
    1117. Stern RA
    1118. Stevens AB
    1119. Strittmatter SM
    1120. Sultzer D
    1121. Swerdlow RH
    1122. Tanzi RE
    1123. Tilson JL
    1124. Trojanowski JQ
    1125. Troncoso JC
    1126. Tsuang DW
    1127. Valladares O
    1128. Van Deerlin VM
    1129. van Eldik LJ
    1130. Vassar R
    1131. Vinters HV
    1132. Vonsattel J-P
    1133. Weintraub S
    1134. Welsh-Bohmer KA
    1135. Whitehead PL
    1136. Wijsman EM
    1137. Wilhelmsen KC
    1138. Williams B
    1139. Williamson J
    1140. Wilms H
    1141. Wingo TS
    1142. Wisniewski T
    1143. Woltjer RL
    1144. Woon M
    1145. Wright CB
    1146. Wu C-K
    1147. Younkin SG
    1148. Yu C-E
    1149. Yu L
    1150. Zhang Y
    1151. Zhao Y
    1152. Zhu X
    1153. CHARGE
    1154. Adams H
    1155. Akinyemi RO
    1156. Ali M
    1157. Armstrong N
    1158. Aparicio HJ
    1159. Bahadori M
    1160. Becker JT
    1161. Breteler M
    1162. Chasman D
    1163. Chauhan G
    1164. Comic H
    1165. Cox S
    1166. Cupples AL
    1167. Davies G
    1168. DeCarli CS
    1169. Duperron M-G
    1170. Dupuis J
    1171. Evans T
    1172. Fan F
    1173. Fitzpatrick A
    1174. Fohner AE
    1175. Ganguli M
    1176. Geerlings M
    1177. Glatt SJ
    1178. Gonzalez HM
    1179. Goss M
    1180. Grabe H
    1181. Habes M
    1182. Heckbert SR
    1183. Hofer E
    1184. Hong E
    1185. Hughes T
    1186. Kautz TF
    1187. Knol M
    1188. Kremen W
    1189. Lacaze P
    1190. Lahti J
    1191. Grand QL
    1192. Litkowski E
    1193. Li S
    1194. Liu D
    1195. Liu X
    1196. Loitfelder M
    1197. Manning A
    1198. Maillard P
    1199. Marioni R
    1200. Mazoyer B
    1201. van Lent DM
    1202. Mei H
    1203. Mishra A
    1204. Nyquist P
    1205. O’Connell J
    1206. Patel Y
    1207. Paus T
    1208. Pausova Z
    1209. Raikkonen-Talvitie K
    1210. Riaz M
    1211. Rich S
    1212. Rotter J
    1213. Romero J
    1214. Roshchupkin G
    1215. Saba Y
    1216. Sargurupremraj M
    1217. Schmidt H
    1218. Schmidt R
    1219. Shulman JM
    1220. Smith J
    1221. Sekhar H
    1222. Rajula R
    1223. Shin J
    1224. Simino J
    1225. Sliz E
    1226. Teumer A
    1227. Thomas A
    1228. Tin A
    1229. Tucker-Drob E
    1230. Vojinovic D
    1231. Wang Y
    1232. Weinstein G
    1233. Williams D
    1234. Wittfeld K
    1235. Yanek L
    1236. Yang Y
    1237. Farrer LA
    1238. Psaty BM
    1239. Ghanbari M
    1240. Raj T
    1241. Sachdev P
    1242. Mather K
    1243. Jessen F
    1244. Ikram MA
    1245. de Mendonça A
    1246. Hort J
    1247. Tsolaki M
    1248. Pericak-Vance MA
    1249. Amouyel P
    1250. Williams J
    1251. Frikke-Schmidt R
    1252. Clarimon J
    1253. Deleuze J-F
    1254. Rossi G
    1255. Seshadri S
    1256. Andreassen OA
    1257. Ingelsson M
    1258. Hiltunen M
    1259. Sleegers K
    1260. Schellenberg GD
    1261. van Duijn CM
    1262. Sims R
    1263. van der Flier WM
    1264. Ruiz A
    1265. Ramirez A
    1266. Lambert J-C
    (2022) New insights into the genetic etiology of Alzheimer’s disease and related dementias
    Nature Genetics 54:412–436.
    https://doi.org/10.1038/s41588-022-01024-z

Decision letter

  1. Patrik Verstreken
    Reviewing Editor; KU Leuven, Belgium
  2. Claude Desplan
    Senior Editor; New York University, United States
  3. Simon G Sprecher
    Reviewer; Universität Fribourg, Switzerland

Our editorial process produces two outputs: (i) public reviews designed to be posted alongside the preprint for the benefit of readers; (ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "Tau polarizes an aging transcriptional signature to excitatory neurons and glia" for consideration by eLife and please accept our apologies for the longer than usual reviewing time. Your article has been reviewed by 3 peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Claude Desplan as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Simon G Sprecher (Reviewer #3).

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

1) Given that all vulnerable cell types were already lost at day 1, the reviewers were unclear whether the model assesses age-dependent neurodegeneration. This may also be developmental toxicity. There should be a balanced discussion on this or alternatively, data could be included making use of models that show defects only at older age.

2) There were some concerns about genetic background (cf rev 1) and controls (cf rev 2): ie is there a possibility to include wt-tau or carefully discussing this; likewise, given the depth of analysis one achieves with single cell seq approaches, genetic background issues can be real confounding factors. Was this addressed in the experimental design.

3) The finding of involvement of the NFkB pathway is interesting, but causality has not been shown. All reviewers thought it would be rather simple to put the idea to test by genetically modulating this pathway and assessing if neuronal loss is rescued.

4) the last comment by reviewer 2 was also deemed important. The comparison between species and of an FTD-Tau mutation with AD needs to be toned down.

5) the other issues can likely be addressed by textual changes or added discussion.

Reviewer #1 (Recommendations for the authors):

I have these points that would improve the paper:

– Can the authors test whether the neuronal loss in their model is due neurodegeneration rather than developmental toxicity to tau.

– Given that they find the NFkB pathway to be involved in tauopathy in a model organism, it would be fascinating if they put this idea to test and show causality by genetically modulating this pathway and rescuing neuronal loss.

– Can the authors please mention the genetic background of all lines used in the Methods. Where UAS-tau and the wild-type fly that was crossed to elav-Gal4 to serve as a control in the same genetic background?

– The authors should include in the counting step whatever 3' UTR the tau transformation vector had that was used for generating the fly model, since the fast majority of reads should map there rather than in the tau sequence. For now, it seems that NM_016834.5:151-1302 (Methods) represents the CDS.

– Figure S6B and S10D are missing a quantification. In addition, for 10D it would be helpful to add a negative control to see how specific the signal is, such as a control fly without the endogenous GFP tag.

– It is interesting that glia seem to react strongly to tau. However, it is not clear whether this is cell autonomous – because they also express tau – or as a reaction of the neuronal tau expression. Is the promoter they use neuronal and would we expect any expression in glia? Can they maybe add a panel to Figure S8 with a boxplot showing tau expression levels in glia cell types and neurons.

– It is interesting how the authors find multiple regulons (some with >2x larger coefficients than the Rel regulon) to be associated with the degree of vulnerability. For the curious reader it would be helpful to at least point them out and briefly mention the underlying biology.

– Why were in Figure S7 KEGG pathways only annotated for few cell types. This should be explained in legend or annotated more widely, e.g. in all cell types that are lost.

– In Figure S9B not all cell-types shown. Does this mean that no pathways were found in those, maybe they could add this to the legend? And why are the KEGG terms in Figure S7B for a'/b'-KC different than in S9B?

Also a typo is in this legend: 'including pathways that are actively in cell-type specific vs. more global patterns'

– Figure S8 should add whether this is counts or log-scale.

– Figure 4D can the authors add explicitly whether this is control and tau cells pooled?

Reviewer #2 (Recommendations for the authors):

Wu et al. conducted longitudinal single-nucleus RNA sequencing in a Drosophila transgenic line expressing pathogenic tau (Arg406 ->Trp) and control to study presenile degenerative dementia with bitemporal atrophy. Their data is consistent with previous findings on Tau neurotoxicity, which significantly affects excitatory neurons in human brain samples and transgenic mice. Intriguingly, intracellular transgenic Tau induced strong transcriptional signatures, aging-like signatures, and an innate immune response, including the NFKB pathway, in the transgenic animals. This dataset provides a valuable resource for exploring dynamic, age-dependent gene expression changes at a cellular level. The authors propose that innate immune signatures may serve as predictors of neuronal subtype vulnerability in tauopathies. However, the observed skewing of cell proportions in day-1 animals necessitates stronger evidence to support this hypothesis, which is currently lacking in the manuscript. The paper is primarily descriptive and lacks mechanistic insights. Furthermore, the identified pathways/genes presented in the paper lack orthogonal validation.

1. About the controls: Authors compared Tau transgenic line (Arg406 ->Trp) with the control (GAL4 expressing animals) but not with wt-tau line. They may potentially lead to misinterpretation. Although Wittmann et al. 2001 noted toxicity when wt-tau is expressed, the toxicity is much less compared to Tau transgenic line. Or would another alternative be to use mutant tau animals lacking aggregation-prone regions?

2. It is striking to see the drastic cell proportion changes on day-1 (figure 2B), which may reflect the deficits in neuronal development. Did authors check the expression of transgene expression levels across neuronal subtypes to make sure the vulnerability is not due to a difference in the Tau transgene expression?

3. Figure 2B-D, although authors admit that the difference is likely due to the "increase in glial cell abundance from scRNAseq is likely a consequence of proportional changes in single cell suspensions due to neuronal loss," is there a way to quantitatively assess this? Especially authors know the amount of neuronal loss and increase in the glial cells through scRNAseq.

4. Line 204-205, authors claim, "93% of tau-induced differentially expressed genes were also triggered by aging in control flies". However, figure 3A does not reflect the 93% similarity. There are more DEGs in age-specific conditions compared to the Tau. The same holds true for the Figure 2B. Moreover

5. Figure 2B, the number of DEG in cluster Lai and Kenyon cells is highly skewed in the Tau transgenic lines. I find it is intriguing to see high number of DEGs in the cells that are degenerating. Since these plots don't tell much about whether they are up or down, it would be good to mention what proportion of the genes are up and down.

6. The authors claim that among non-neuronal cell types, ensheathing glia, cortex glia, astrocyte-like glia, and hemocytes have the highest number of tau-driven DEGs, but this is not clear from the UMAP in Figure 3C. Additionally, Figure 3C lacks a scale bar, making it difficult to interpret and compare the figure with Figure 3B.

7. In line 249, the authors claim 90% concordance with previously published datasets, but the data representing this is missing in the paper. Additionally, performing DEG with pseudo-bulk from different clusters and performing DEG to find the concordance may not be very informative. For example, did the authors find consistent gene signatures per cluster when compared with previous datasets? This data should be provided.

8. Authors have created an excellent data resource, and it would be interesting to explore the resilience of inhibitory neurons or the vulnerability of excitatory neurons to gain more insights into the cell-type-specific vulnerability or resilience mechanisms. The authors should present a couple of volcano plots showing the differentially expressed genes between important clusters, such as LAI, Kenyon cells, ensheathing glia, etc.

9. It would be beneficial for the authors to explore these pathways in greater depth and perform further experimental validation to strengthen the findings using orthogonal approaches. For instance, a rescue experiment where NFKB/Relish is knocked out to see if this modifies Tau toxicity.

10. In Figure 5, the authors compared cell-type-specific transcriptional signatures between human Alzheimer's disease (AD) and Drosophila. However, some readers may find this comparison difficult to comprehend as the two species differ vastly. Moreover, the Tau mutation that the authors investigated is not associated with AD. Also, in AD, amyloid pathology significantly drives gene expression in immune cells, which is absent in Drosophila. Authors should consider taking the relevant dataset derived from the Arg406 ->Trp patients or from the iPSC-derived cells to validate the observations.

Reviewer #3 (Recommendations for the authors):

As mentioned above, I feel this is an elegant and nicely described study. It provides a further example of the power of single-cell transcriptomic to assess disease genes, showing that in such a fashion one can assess the impact of cell-types, but also the actual genes that are altered. It also shows that the findings can be transferred to patient tissue, thus integrating previous published human sc-data.

There are a few points that I feel might be important to take into account.

– The authors show that ratios of cells are changed in response to tau-GOF, however no explanation is given. What is the basis of this alteration? Cell-death, changes of differentiation/proliferation (it seems the GOF is throughout development in "elav" cells).

– The integration and comparison of fly/human data is a bit short, I could not follow the process of how this was achieved, what framework the authors used etc.

– Conceptually, I think it is nice to see the in-silico analysis of the tau-GOF and aging, however I feel some in vivo validation might have been beneficial to support the claims of affected cell-types and differential expressed genes.

While I fully agree that an extensive genetic analysis may be beyond the scope of the current paper, a proof-of-concept analysis would have been supportive of the validity of the data.

https://doi.org/10.7554/eLife.85251.sa1

Author response

Essential revisions:

1) Given that all vulnerable cell types were already lost at day 1, the reviewers were unclear whether the model assesses age-dependent neurodegeneration. This may also be developmental toxicity. There should be a balanced discussion on this or alternatively, data could be included making use of models that show defects only at older age.

The reviewers raise an important point, and we agree that “developmental toxicity” may contribute to some of our observations. The elav-GAL4 pan-neuronal driver activates expression of tauR406W during developmental stages and observed changes in cell-abundance or gene expression may reflect this.

For example, tau developmental toxicity has been shown to cause malformation of mushroom body structures (Kosmidis et al., 2010), and this phenotype likely explains the reductions in several cell clusters in our dataset. However, the widely-used elav>tauR406W model also recapitulates aging-dependent, progressive neuronal loss and CNS dysfunction (Wittmann et al., 2001). Consequently, our longitudinal design can additionally highlight the accompanying cell-specific transcriptional changes. While we acknowledge that most of the changes in cell abundance highlighted in Figure 2B appear to be established at day 1, in many cases, age-dependent changes are also strongly suggested (e.g. Cluster 1, 9, and 12, along with astrocyte-like glia). One important caveat is that the dataset lacks replicate samples at either day 1 or day 20. Therefore, while changes in cell abundance at these timepoints are suggestive in many cases, our experimental design does not permit cross-sectional statistical analysis at either timepoint, nor can we quantitatively examine changes over time (e.g. 1 vs. 10 days or 1 vs. 20 days). For this reason, our analyses of cell abundance (Figure 2A) relied on pooled data from all 3 time points. We also performed a replication analysis at 10 day (Figure 2—figure supplement 1A), based an independent generated dataset with triplicate samples. In a complementary analysis, we also employed the deconvolution algorithm MuSiC to examine cell type proportions from elav>tauR406W bulk RNAseq data (Figure 2—figure supplement 2). Notably, our experimental design does permit robust analysis of age-dependent changes in cell-specific gene expression. Our primary analysis (Figure 3B) leverages the longitudinal data and includes adjustment for age as a covariate. However, in order to address the reviewer’s feedback, we have now included cross-sectional analyses of cell-specific differential expression changes (Results text and Figure 3–Source Data 5). These new data readily allow interrogation of specific genes and pathways for mapping of dynamic changes over time. For example, in our revisions, we note that tau-induced Relish regulon activity is amplified (or attenuated) in selected cell types with aging, highlighting potential dynamic changes in immune signaling pathways (Results text and Figure 4—figure supplement 5). For example, in the L1-5 lamina neuron cluster, NFkB responsive genes are only significantly differentially expressed at 20-days. Similarly, the innate-immune signature in astrocyte-like glia appears driven by changes in 10- and 20-day-old flies; no significant change is observed at 1-day. As requested, we have also added text to the results and discussio in order to provide a more balanced and nuanced discussion of these issues and emphasizing the challenge to disentangle developmental toxicity from age-dependent neurodegeneration. We also include new discussion of a recently published study that appeared while our manuscript was under review, which uses the nsyb>tauP301L model expected to have more restricted expression within the adult brain (Praschberger et al., Neuron 2023, PMID: 36948206). Finally, we have also added text to the Figure 2 legend clarifying that significant differences in tau-induced cell abundance were based on comparisons of pooled timepoints.

2) There were some concerns about genetic background (cf rev 1) and controls (cf rev 2): ie is there a possibility to include wt-tau or carefully discussing this; likewise, given the depth of analysis one achieves with single cell seq approaches, genetic background issues can be real confounding factors. Was this addressed in the experimental design.

We agree that genetic background is an important potential confounder to consider. The UAS-tauR406W strain used in this study was iteratively backcrossed to w1118 for 5 generations as described in our earlier published work (Guo et al., 2018, PMID: 29874575). For comparison, the control strain, elav-GAL4 / + was generated by outcrossing elav-GAL4 with the identical w1118 strain used for backcrossing (elav-GAL4 / w1118). In order to clarify these experimental design considerations, our revision includes new explanatory text in the Method. Please also see #4, below, for discussion of tauWT.

3) The finding of involvement of the NFkB pathway is interesting, but causality has not been shown. All reviewers thought it would be rather simple to put the idea to test by genetically modulating this pathway and assessing if neuronal loss is rescued.

We thank the reviewer for this suggestion. Given the intriguing association between the Relish (NFkB) regulon and tau-associated cell abundance changes, we tested whether neuron-specific knockdown of Rel alters structural brain degeneration in the elav>tauR406W model. These experiments leveraged 2 independent RNA-interference (RNAi) strains that were previously validated from other published work. Following adult brain histology, we did not detect any significant difference in the vacuolar degeneration caused by tau following pan-neuronal expression of Relish RNAi (expression of the UAS-tauR406W and UAS-RelRNAi were both driven by the elav-GAL4 driver). Our revision includes these new experimental data (Figure 4—figure supplement 8). We have also added new text to our Discussion; however, we interpret these negative data cautiously, since similar genetic manipulations of Relish have previously been demonstrated to modulate neurodegeneration in other Drosophila neurodegenerative models (Cao et al., 2013, PMID: 23613578; Petersen et al., 2013, PMID:23502677; Kounatidis et al., 2017, PMID:28445733). Moreover, recent studies in mouse models of neurodegeneration suggest that NFkB signaling in neurons may be required for the cell non-autonomous recruitment and activation of microglia (Welch et al., 2022, PMID: 36170369). Thus, while evidence from human genetics strongly supports a causal role for immune processes in AD risk and pathogenesis, further experimentation will be required to definitively resolve cell-type specific mechanisms and contribution of NFkB/Relish in tau-mediated neurodegeneration.

4) the last comment by reviewer 2 was also deemed important. The comparison between species and of an FTD-Tau mutation with AD needs to be toned down.

For the analyses presented in the final section of results and in Figure 5, we had 2 main goals. First, we wanted to confirm the degree of overlap for cell-type specific transcriptional signatures between the Drosophila and human brain, providing a useful map for reciprocal forward and reverse translation from single-cell data generated across species. Cross-species overlap between brain cell types is already well established—many studies have noted correspondences between gene expression markers of many excitatory and inhibitory neuronal subtypes, including cholinergic, dopaminergic, and glutamergic cells. Homologies between Drosophila and mammalian glial subtypes are also well-established (Freeman, 2015, PMID:25722465, Yildrim et al., 2019; PMID:30443934). Notably, the overlaps shown in Figure 5A (and independently replicated in Figure 5—figure supplement 1B) are not specific to the AD / tau model context; these are broadly generalizable for cross-species interpretation of many other single cell profiling studies of the brain, which we believe that Reviewer 3 appreciated as a strength of the study. In order to further highlight this and help clear up any ambiguity, we repeated our analysis, but excluding human brains from AD cases from Mathys et al. (controls only) and considering overlap with an independent Drosophila single-cell dataset comprised of wildtype controls (Davies et al. 2018). The resulting heatmap is consistent with our findings in Figure 5A, revealing cross-species correspondences between many neuron and glial subtypes. These new data are included in our revision (Figure 5—figure supplement 2).

Our second goal was to examine whether NFkB pathway genes are expressed in human neurons and potentially dysregulated in AD, similar to our findings in elav>tauR406W flies. Our results in Figure 5B-C thus provide support for translation of our findings. The reviewer raises concerns about the validity of this comparison due to differences between our fly tauopathy model and AD, including the (i) reliance on a mutant form of MAPT associated with a distinct disease (frontotemporal dementia) and (ii) also the lack of amyloidbeta pathology. In response, we begin by noting that for many decades, similar FTD mutant MAPT transgenic mice have been widely used in AD research (e.g., the rTg4510 and PS19 strains which harbor P301L/S variants). Further, while β-amyloid pathology appears to be an important trigger for innate immunity gene expression signatures, several recent studies also highlight an important role for tau (Lee et al., 2021, Wang et al., 2022, Chen et al., 2023).

Regardless, we have performed some new analyses to further address the reviewer’s critiques. First, as suggested, we leveraged a published analysis of scRNAseq “pseudobulk” data from a MAPTP301L mouse of tauopathy (Lee et al., 2021, PMID:34965428) to evaluate cell-type specific expression of the Rel/NFkB regulon. This analysis demonstrates broadly consistent results with that from human AD brain, including increased NFkB pathway expression in excitatory neurons and microglia (Figure 5—figure supplement 3). Second, in our prior published work on bulk RNAseq (Mangleburg et al., 2020, PMID:32993812), we found an approximately 70% overlap between differentially expressed genes in elav>tauWT vs. elav>tauR406W. In order to further examine the conservation between the transcriptional signature induced by wildtype and mutant tau, we plotted longitudinal normalized expression for the innate immune module in each case (vs. control). The results highlight overall similar Tau-triggered increases in innate immunity, with elav>tauR406W causing a more severe and accelerated gene expression signature (Figure 4—figure supplement 5). Our result is consistent with a notable in vivo study by Bardai et al. (J Neurosci 2018, PMID: 29138281) in which 5 different FTDP17 mutant forms of MAPT, including R406W, were compared with wildtype MAPT in Drosophila models. The results strongly suggest that R406W (and several other protein coding mutations examined) increase MAPT phosphorylation but share conserved downstream mechanisms leading to neurodegeneration, likely including NFkB signaling.

Besides incorporating these new data (above), we have also carefully reviewed this section of the manuscript for clarity, making a number of additional textual revisions to present a more balanced and cautious interpretation. In the discussion, we also add new tex on the (i) potential limitations for drawing conclusions about AD from models lacking β-amyloid pathology and (ii) the evidence supporting common mechanisms for FTDP17 mutant and wildtype MAPT. We also note the caveat of cross-species comparisons given the uncertain conservation of glia, especially microglia, between mammals and flies.

5) The other issues can likely be addressed by textual changes or added discussion.

Please see additional responses and edits below.

Reviewer #1 (Recommendations for the authors):

I have these points that would improve the paper:

– Can the authors test whether the neuronal loss in their model is due neurodegeneration rather than developmental toxicity to tau.

Please see our detailed response to Essential revision #1. Briefly, while the changes in cell abundance that we observe at either 1 or 20 days are highly suggestive, except for day 10 where we have replicate samples, our experimental design does not permit cross-sectional robust, statistical analysis, nor can we quantitatively examine changes over time (e.g. 1 vs. 10 days or 1 vs. 20 days). For this reason, our primary analyses highlighting significant cell abundance changes in tau vs. control (Figure 2A) relied on pooled data from all 3 time points, and we also present a cross-sectional replication analysis at day 10. However, our experimental design does permit robust analysis of age-dependent changes in gene expression that provide independent, albeit indirect, support for neurodegenerative pathophysiology. In order to address the reviewer request, our revision includes new cross-sectional analyses of cell-specific differential expression changes (Figure 3–Source Data 5). In particular, we highlight neuronal and glial cell types in which tau-induced Relish regulon activity is amplified (or attenuated) with aging (Figure 4—figure supplement 5). We have also added text to provide a more balanced and nuanced discussion of these issues and emphasizing the challenge to disentangle developmental toxicity from age-dependent neurodegeneration.

– Given that they find the NFkB pathway to be involved in tauopathy in a model organism, it would be fascinating if they put this idea to test and show causality by genetically modulating this pathway and rescuing neuronal loss.

As noted in response to Essential revision #3, we have performed new experiments to directly test the hypothesis that neuronal immune pathways are causally linked to tau-mediated neurodegeneration; however, the results were negative. These data are included in the revision and we also carefully discuss published work from other fly models of aging and neurodegeneration as well as mouse tauopathy that strongly suggest NFkB can directly modulate neurodegeneration.

– Can the authors please mention the genetic background of all lines used in the Methods. Where UAS-tau and the wild-type fly that was crossed to elav-Gal4 to serve as a control in the same genetic background?

Please see response to Essential revision #2. We have added the relevant explanatory text to the Methods, and indeed, the control (elav-GAL4) strain was outcrossed to same w1118 strain that we previously used for outcrossing of UAS-tauR406W.

– The authors should include in the counting step whatever 3' UTR the tau transformation vector had that was used for generating the fly model, since the fast majority of reads should map there rather than in the tau sequence. For now, it seems that NM_016834.5:151-1302 (Methods) represents the CDS.

As detailed above in response to the related comment in the public review, we have repeated the analysis of MAPT expression including the UTR sequence, as requested, and updated Figure 3—figure supplement 4. The overall results and interpretation are not substantially changed.

– Figure S6B and S10D are missing a quantification. In addition, for 10D it would be helpful to add a negative control to see how specific the signal is, such as a control fly without the endogenous GFP tag.

We have performed the requested quantifications. Our new results in Figure 2—figure supplement 3C confirm that the signal intensity for multiple markers (DAPI, phalloidin, and repo) are all significantly increased in the elav>tauR406W adult brain. We have also performed quantification for Figure S10D (now Figure 4—figure supplement 2D in the revision), demonstrating that 78% of neurons and 51% of glia costain for Relish. In order to address specificity, we co-stained flies harboring the GFP-tagged Relish protein with both anti-GFP and anti-Rel. The observed colocalization supports specificity of the Rel-GFP strain that we used to confirm Rel expression in neurons (Figure 4—figure supplement 6).

– It is interesting that glia seem to react strongly to tau. However, it is not clear whether this is cell autonomous – because they also express tau – or as a reaction of the neuronal tau expression. Is the promoter they use neuronal and would we expect any expression in glia? Can they maybe add a panel to Figure S8 with a boxplot showing tau expression levels in glia cell types and neurons.

The elav-GAL4 driver line used for MAPT expression in our model is a well-established and widely used pan-neuronal driver. As requested, we have generated a plot highlighting the specificity of MAPT neuronal expression, and we have included this in our revision (Figure 3—figure supplement 4D). Indeed, while we favor a model in which the glial reaction is cell non-autonomous, we cannot completely exclude either a

transient or low level of MAPT expression in glia that may contribute in part via an alternative (or additional) cell autonomous mechanism. Indeed, while not a common pathology of AD, other tauopathies like PSP have prominent glial tau aggregates (e.g., tufted astrocytes), and glial tau toxicity has also been modeled in Drosophila using a glial-specific driver (Colodner and Feany 2010, PMID: 11408621). Our revision includes new text in the discussion to draw attention to the need for additional studies to dissect cell autonomous vs. cell non-autonomous mechanisms contributing to both neuronal and glial responses in tauopath.

– It is interesting how the authors find multiple regulons (some with >2x larger coefficients than the Rel regulon) to be associated with the degree of vulnerability. For the curious reader it would be helpful to at least point them out and briefly mention the underlying biology.

We thank the reviewer for this suggestion. We have added new text to the results drawing attention to other noteworthy findings from the analysis and the implicated biology.

– Why were in Figure S7 KEGG pathways only annotated for few cell types. This should be explained in legend or annotated more widely, e.g. in all cell types that are lost.

We had originally only provided annotations on pathway enrichment for selected clusters that were discussed in the text, but we agree that this may be confusing. Based on the feedback from the reviewer and also the question raised below regarding the difference between Figure S7B and S9B (now Figure 3—figure supplements 1 and 2) we have removed these highly selective annotations. Instead, in the Figure legend, we refer to Figure 3–Source Data 2 which includes comprehensive pathway enrichment results.

– In Figure S9B not all cell-types shown. Does this mean that no pathways were found in those, maybe they could add this to the legend? And why are the KEGG terms in Figure S7B for a'/b'-KC different than in S9B?

The renamed Figure 3—figure supplement 1B includes all cell clusters with at least 1 enriched KEGG term; we have added clarification to the figure legend. We also note in the legend that readers may refer to Figure 3–Source Data 2 which includes comprehensive annotations, including other curated pathways in addition to KEGG (e.g., GO, Panther and others). For simplicity, in the figure annotation, we restricted our consideration to KEGG pathways. In Figure 3—figure supplement 2B (previously Figure S7B), the selective annotations with functional enrichment analysis considered a subset of differentially expressed genes that were seen consistently in both scRNAseq and bulk tissue RNAseq data; whereas Figure 3—figure supplement 1B considers all differentially expressed genes from the scRNAseq data. Based on the feedback, we agree that this is confusing; we have removed these highly selective annotations and instead refer to Figure 3—figure supplement 1B and Figure 3–Source Data 2 with comprehensive pathway enrichment results.

– Figure S8 should add whether this is counts or log-scale.

The gene expression data in Figure 3—figure supplement 4 are displayed as normalized gene counts. This has been clarified in the figure legend.

– Figure 4D can the authors add explicitly whether this is control and tau cells pooled?

Control and tau cells are pooled in Figure 4D. We have clarified this in the revised figure legend.

Reviewer #2 (Recommendations for the authors):

Wu et al. conducted longitudinal single-nucleus RNA sequencing in a Drosophila transgenic line expressing pathogenic tau (Arg406 ->Trp) and control to study presenile degenerative dementia with bitemporal atrophy. Their data is consistent with previous findings on Tau neurotoxicity, which significantly affects excitatory neurons in human brain samples and transgenic mice. Intriguingly, intracellular transgenic Tau induced strong transcriptional signatures, aging-like signatures, and an innate immune response, including the NFKB pathway, in the transgenic animals. This dataset provides a valuable resource for exploring dynamic, age-dependent gene expression changes at a cellular level.

We thank the reviewer for this positive feedback. We also want to clarify that beyond frontotemporal dementia / FTDP17, which results from the MAPTR406W mutation, we believe that our studies also provide important insights on the cell-specific transcriptional mechanisms of tau-mediated neurodegeneration in Alzheimer’s disease (AD). As noted in our response, FTD mutant MAPT transgenic mice have been widely used in AD research for decades (e.g. rTg4510 and PS19 which harbor P301L/S variants). In our prior study of Drosophila head bulk RNAseq (Mangleburg et al., 2020), we found an approximately 70% overlap between differentially expressed genes in elav>TauWT vs. elav>TauR406W. This result is consistent with the systematic in vivo study by Bardai et al. (J Neurosci 2018, PMID: 29138281) using Drosophila transgenic models in which 5 different FTDP17 mutant forms of MAPT, including R406W, were compared with wildtype. The results of this work strongly suggest that R406W (and several other protein coding mutations included) increase MAPT phosphorylation but share conserved downstream mechanisms with wildtype MAPT toxicity, leading to neurodegeneration.

The authors propose that innate immune signatures may serve as predictors of neuronal subtype vulnerability in tauopathies. However, the observed skewing of cell proportions in day-1 animals necessitates stronger evidence to support this hypothesis, which is currently lacking in the manuscript. The paper is primarily descriptive and lacks mechanistic insights. Furthermore, the identified pathways/genes presented in the paper lack orthogonal validation.

As noted in the response to Essential revisions #1 and #3, we have performed additional analyses and made textual revisions to address questions concerning tau developmental toxicity, and we have also attempted additional experiments to directly test whether NFkB pathways may be causal in neurons, which we include in the revision (Figure 4—figure supplement 8). While we feel that evidence from other publications suggest our findings have causal implications, we have taken an overall cautious approach in our interpretations. For example, our abstract notes only that NFkB signaling maybe a “marker for cellular vulnerability”. Our revision includes substantial new text clarifying these issues and the pertinent caveats.

1. About the controls: Authors compared Tau transgenic line (Arg406 ->Trp) with the control (GAL4 expressing animals) but not with wt-tau line. They may potentially lead to misinterpretation. Although Wittmann et al. 2001 noted toxicity when wt-tau is expressed, the toxicity is much less compared to Tau transgenic line. Or would another alternative be to use mutant tau animals lacking aggregation-prone regions?

Our goal was to understand mechanisms of tau-mediated neurodegeneration relevant broadly across tauopathies, including AD. As explained above, prior published work, including our own data, support conserved molecular mechanisms of neurodegeneration for wildtype and mutant forms of MAPT. Subsequent to the Wittman et al. study, the same group performed additional analyses in which wildtype and mutant MAPT transgene expression levels were more tightly controlled, and these lines were found to have strongly overlapping mechanisms (Bardai et al., J Neurosci 2018, PMID: 29138281). Indeed, FTDP17 mutant forms of MAPT have been integral to studies of tau-mediated neurodegeneration across flies, mice, and cellular models, and such findings have been broadly translatable to AD, albeit with important caveats, similar to any experimental model. Nevertheless, as discussed in the response to Essential revision #4, we have performed some new analyses to address some of the reviewer’s concerns (see response to Reviewer 2, point #10). In particular, we leveraged our previously published bulk RNAseq (Mangleburg et al., 2020) in order to examine the conservation of innate immune transcriptional signatures induced by wildtype and mutant tau (Figure 4—figure supplement 1). The results highlight overall similar Tau-triggered increases in innate immunity, with elav>tauR406W causing a more severe and accelerated gene expression change as expected.

2. It is striking to see the drastic cell proportion changes on day-1 (figure 2B), which may reflect the deficits in neuronal development. Did authors check the expression of transgene expression levels across neuronal subtypes to make sure the vulnerability is not due to a difference in the Tau transgene expression?

Our originally submitted manuscript included an analysis confirming that the affected cell types and differentially expressed genes do not simply correspond to the spatial pattern of MAPT transgene expression in the Drosophila brain (Figure 3—figure supplement 4). Based on the feedback from Reviewer 1, our revision includes an updated, improved version of this analysis, and we have additionally included a new plot highlighting the lack of correlation between MAPT expression and cell abundance (Figure 3—figure supplement 4B). In addition, in our expanded elastic-net regression model (Figure 4–Source Data 4) MAPT expression was excluded as a significant predictor of cell abundance change. Please also see response to Essential revision #1 for discussion of the concern related to tau developmental toxicity.

3. Figure 2B-D, although authors admit that the difference is likely due to the "increase in glial cell abundance from scRNAseq is likely a consequence of proportional changes in single cell suspensions due to neuronal loss," is there a way to quantitatively assess this? Especially authors know the amount of neuronal loss and increase in the glial cells through scRNAseq.

The analyses presented in Figure 2 are indeed an indirect quantification of the proportional changes in neuronal versus glial numbers. Based on our cell count data from the 10-day old flies with triplicate samples, the proportion of neurons in 10x Chromium libraries were reduced from 90% to 83% in controls vs. elav>tauR406W flies, respectively. We have added these estimates to the Results text. Our conclusion that the observed changes likely reflect neuronal loss is additionally informed by our quantification of our experimental data showing largely preserved glial numbers and reduced overall brain volumes (Figure 2D). Lastly, we include an analysis in which confidence intervals for cell abundance changes were computed using an alternative model in which glia were assumed to be unchanging (Figure 2—figure supplement 3A).

4. Line 204-205, authors claim, "93% of tau-induced differentially expressed genes were also triggered by aging in control flies". However, figure 3A does not reflect the 93% similarity. There are more DEGs in age-specific conditions compared to the Tau. The same holds true for the Figure 2B.

The 93% figure is based on consideration of the overall count of unique differentially expressed genes (DEGs), pooling across all cell types. Of the 5,280 tau-triggered DEGs, there is a 93% overlap with the 5,998 aging-induced gene set. However, the degree of overlap varies when considered separately in each cell type, with a range of 0-75%. Indeed, we want to draw attention to the striking difference between overall brain vs. cell type-specific transcriptome overlaps between tau and aging, as well as the distribution of these changes in the adult fly brain. Our revision includes edits to clarify the 93% figure, and we also present the range of cell-specific overlap in Results.

5. Figure 2B, the number of DEG in cluster Lai and Kenyon cells is highly skewed in the Tau transgenic lines. I find it is intriguing to see high number of DEGs in the cells that are degenerating. Since these plots don't tell much about whether they are up or down, it would be good to mention what proportion of the genes are up and down.

Figure 3—figure supplement 1A displays the number of up versus down tau DEGs for each cell cluster. The number of up- vs. down-regulated DEGs for Lai and Kenyon cells are roughly equal, and we have added a mention of this result in the text.

6. The authors claim that among non-neuronal cell types, ensheathing glia, cortex glia, astrocyte-like glia, and hemocytes have the highest number of tau-driven DEGs, but this is not clear from the UMAP in Figure 3C. Additionally, Figure 3C lacks a scale bar, making it difficult to interpret and compare the figure with Figure 3B.

The visualization provided in Figure 3C is intended to complement the plots in panels A and B, highlighting the striking difference between tau- and age- induced cell-type-specific DEGs. Because of the dynamic range and quantitatively stronger impact of aging, the applied scaling tends to minimize the tau-triggered DEGs. In order to address this, we have regenerated Figure 3C without the prior scaling, which better highlights transcriptional changes in the ensheathing glia, and we also have added a label to the scale bar. We have also generated a new dedicated plot of tau DEGs, based on the Figure 3B data with its own, independent scaling which we include as Figure 3—figure supplement 1C.

7. In line 249, the authors claim 90% concordance with previously published datasets, but the data representing this is missing in the paper. Additionally, performing DEG with pseudo-bulk from different clusters and performing DEG to find the concordance may not be very informative. For example, did the authors find consistent gene signatures per cluster when compared with previous datasets? This data should be provided.

We apologize for this misunderstanding. This replication analysis (6 total scRNAseq libraries), including triplicate samples collected each from 10-day-old controls (elav-GAL4) and elav>tauR406W, is in fact a completely new, unpublished scRNAseq dataset that was generated as part of this study. The replication dataset is described in the Methods and we have included associated analytics in the supplemental information (Figure 2—figure supplement 1 and Figure 3–Source Data 3,4); all data has also been deposited in Synapse so that it is available for the community. This dataset permitted our replication analysis for cell abundance changes discussed in the response to Essential revision #1 as well as the replication of cell-type specific differentially expressed genes. The 90% figure refers to the overlap between 3,937 and 4,957 total unique tau-triggered differentially expressed genes at 10-days between the discovery and replication dataset, respectively. In order to examine cell-specific overlaps, we newly considered the full discovery dataset (results from primary age-adjusted analysis, including days 1, 10 and 20. Two-thirds of clusters (61 out of 90)) show a significant overlap with our 10-day-old replication dataset, including excitatory neuron and glial subtypes that are the major focus of our manuscript. We have added these new replication analyses to the Results (Figure 3–Source Data 4).

8. Authors have created an excellent data resource, and it would be interesting to explore the resilience of inhibitory neurons or the vulnerability of excitatory neurons to gain more insights into the cell-type-specific vulnerability or resilience mechanisms. The authors should present a couple of volcano plots showing the differentially expressed genes between important clusters, such as LAI, Kenyon cells, ensheathing glia, etc.

We thank the reviewer for this suggestion. Our revision includes new volcano plots in Figure 3—figure supplements 3 and 5, highlighting selected cell clusters of interest that we discuss in the manuscript, including vulnerable excitatory neurons (e.g., Kenyon cells, Dm3a/b, and Lawf1) and glial subtypes (e.g., ensheathing and perineurial glia and astrocyte-like cells).

9. It would be beneficial for the authors to explore these pathways in greater depth and perform further experimental validation to strengthen the findings using orthogonal approaches. For instance, a rescue experiment where NFKB/Relish is knocked out to see if this modifies Tau toxicity.

Within the scope of this manuscript, we have focused our experimental validation on Relish / NFkB immune pathway. This includes confirming Relish expression in neurons, and we have now added new experimental data directly testing whether genetic manipulation of Relish modulates tau-mediated neurodegeneration. As detailed in the response to Essential revision #3, this result is negative but we have incorporated these data in our revised manuscript along with new discussion. As requested by Reviewer 1, we have also added new text to the discussion highlighting additional pathways of interest based on the results of our elastic net regression analysis.

10. In Figure 5, the authors compared cell-type-specific transcriptional signatures between human Alzheimer's disease (AD) and Drosophila. However, some readers may find this comparison difficult to comprehend as the two species differ vastly. Moreover, the Tau mutation that the authors investigated is not associated with AD. Also, in AD, amyloid pathology significantly drives gene expression in immune cells, which is absent in Drosophila. Authors should consider taking the relevant dataset derived from the Arg406 ->Trp patients or from the iPSC-derived cells to validate the observations.

As detailed in our response to Essential revision #4, we have revised the Results and Discussion text to clarify the goals and present a more balanced and cautious interpretation of these cross-species analyses. Also, as noted above, our revision includes new text to justify the use of elav>tauR406W model for insights relevant to AD, noting the potential caveats, including lack of amyloid-β pathology. We also performed new analyses, as suggested. First, to address the reviewer’s concerns, we repeated the analysis presented in Figure 5A, but excluding human brains from AD cases from Mathys et al. (controls only) and considering overlap with an independent Drosophila single-cell dataset comprised of wildtype controls (Davies et al. 2018). The resulting heatmap is consistent with our findings in Figure 5A, revealing cross-species correspondences between many neuron and glial subtypes independent of AD pathology and disease models (Figure 5—figure supplement 2). Second, we leveraged a published analysis of scRNAseq data from a MAPTP301L mouse of tauopathy (Lee et al., 2021, PMID:34965428) to evaluate cell-type specific expression of the Rel/NFkB regulon in the mammalian brain. This analysis demonstrates broadly consistent results with that from human AD brain, including increased NFkB pathway expression in excitatory neurons and microglia (Figure 5—figure supplement 3). Lastly, we also present new analyses from our prior published work on bulk RNAseq (Mangleburg et al., 2020), where we found an approximately 70% overlap between differentially expressed genes in elav>tauWT vs. elav>tauR406W. We therefore plotted longitudinal normalized expression for the innate immune module in each case (vs. control flies), highlighting the conservation of transcriptional signatures induced by wildtype and mutant tau (Figure 4—figure supplement 1). The results highlight overall similar tau-triggered increases in innate immunity, with elav>tauR406W causing a more severe and accelerated gene expression signature. This result is consistent with Bardai et al. (J Neurosci 2018, PMID: 29138281) in which R406W mutant and wildtype MAPT were shown to share conserved downstream mechanisms leading to neurodegeneration, likely including NFkB immune pathways.

Reviewer #3 (Recommendations for the authors):

As mentioned above, I feel this is an elegant and nicely described study. It provides a further example of the power of single-cell transcriptomic to assess disease genes, showing that in such a fashion one can assess the impact of cell-types, but also the actual genes that are altered. It also shows that the findings can be transferred to patient tissue, thus integrating previous published human sc-data.

There are a few points that I feel might be important to take into account.

– The authors show that ratios of cells are changed in response to tau-GOF, however no explanation is given. What is the basis of this alteration? Cell-death, changes of differentiation/proliferation (it seems the GOF is throughout development in "elav" cells).

As detailed in response to Essential revision #1, we believe that the changes in cell proportion are likely due to a reduction in neuronal numbers, reflecting a combination of developmental toxicity and aging-dependent neurodegeneration. Based on our experimental studies and other published evidence, we believe that glial proliferation is unlikely a major contributor (if at all present). Our revision includes new text in the discussion providing a more balanced and nuanced discussion of these issues, and emphasizing the challenge to disentangle developmental toxicity from age-dependent neurodegeneration.

– The integration and comparison of fly/human data is a bit short, I could not follow the process of how this was achieved, what framework the authors used etc.

We regret that this was unclear. As detailed in response to Essential revision #4, we have tried to clarify with new text in the Results, Discussion and Figure 5 legend, as well as the addition of new analytic results to address concerns raised primarily by Reviewer 2.

– Conceptually, I think it is nice to see the in-silico analysis of the tau-GOF and aging, however I feel some in vivo validation might have been beneficial to support the claims of affected cell-types and differential expressed genes.

While I fully agree that an extensive genetic analysis may be beyond the scope of the current paper, a proof-of-concept analysis would have been supportive of the validity of the data.

Our original manuscript included experimental validation demonstrating that (1) the apparent increases in glial cell abundance is likely due to changes in cell proportions, and we also (2) confirmed the expression of Relish in both neurons and glia of the adult fly brain. For our revision, we were guided by the requested Essential Revision #3 (see detailed response), and we have therefore performed additional experiments directly testing whether manipulation of Relish/NFkB in neurons alters tau-induced neurodegeneration. While the results of these experiments were negative, we have incorporated them into the results and discussion, along with discussion of other published studies and potential future work that might further support a causal role for NFkB immune pathways in tauopathy.

https://doi.org/10.7554/eLife.85251.sa2

Article and author information

Author details

  1. Timothy Wu

    1. Jan and Dan Duncan Neurological Research Institute, Texas Children’s Hospital, Houston, United States
    2. Medical Scientist Training Program, Baylor College of Medicine, Houston, United States
    Contribution
    Conceptualization, Formal analysis, Investigation, Writing - original draft, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-5296-2023
  2. Jennifer M Deger

    1. Jan and Dan Duncan Neurological Research Institute, Texas Children’s Hospital, Houston, United States
    2. Medical Scientist Training Program, Baylor College of Medicine, Houston, United States
    3. Department of Molecular and Human Genetics, Baylor College of Medicine, Houston, United States
    4. Department of Neuroscience, Baylor College of Medicine, Houston, United States
    Contribution
    Investigation, Writing – review and editing
    Competing interests
    No competing interests declared
  3. Hui Ye

    1. Jan and Dan Duncan Neurological Research Institute, Texas Children’s Hospital, Houston, United States
    2. Department of Neurology, Baylor College of Medicine, Houston, United States
    Contribution
    Investigation, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-3965-9702
  4. Caiwei Guo

    1. Jan and Dan Duncan Neurological Research Institute, Texas Children’s Hospital, Houston, United States
    2. Department of Molecular and Human Genetics, Baylor College of Medicine, Houston, United States
    3. Department of Neuroscience, Baylor College of Medicine, Houston, United States
    Present address
    Department of Genetics, Stanford University School of Medicine, Standford, United States
    Contribution
    Investigation, Writing – review and editing
    Competing interests
    No competing interests declared
  5. Justin Dhindsa

    1. Jan and Dan Duncan Neurological Research Institute, Texas Children’s Hospital, Houston, United States
    2. Medical Scientist Training Program, Baylor College of Medicine, Houston, United States
    Contribution
    Formal analysis, Investigation, Writing – review and editing
    Competing interests
    No competing interests declared
  6. Brandon T Pekarek

    Jan and Dan Duncan Neurological Research Institute, Texas Children’s Hospital, Houston, United States
    Contribution
    Formal analysis, Investigation, Writing – review and editing
    Competing interests
    No competing interests declared
  7. Rami Al-Ouran

    Jan and Dan Duncan Neurological Research Institute, Texas Children’s Hospital, Houston, United States
    Present address
    School of Computing and Informatics, Al Hussein Technical University, Amman, Jordan
    Contribution
    Formal analysis, Investigation, Writing – review and editing
    Competing interests
    No competing interests declared
  8. Zhandong Liu

    1. Jan and Dan Duncan Neurological Research Institute, Texas Children’s Hospital, Houston, United States
    2. Department of Pediatrics, Baylor College of Medicine, Houston, United States
    3. Center for Alzheimer’s and Neurodegenerative Diseases, Baylor College of Medicine, Houston, United States
    Contribution
    Conceptualization, Supervision, Funding acquisition, Writing – review and editing
    Competing interests
    No competing interests declared
  9. Ismael Al-Ramahi

    1. Jan and Dan Duncan Neurological Research Institute, Texas Children’s Hospital, Houston, United States
    2. Center for Alzheimer’s and Neurodegenerative Diseases, Baylor College of Medicine, Houston, United States
    Contribution
    Conceptualization, Supervision, Writing – review and editing
    Competing interests
    No competing interests declared
  10. Juan Botas

    1. Jan and Dan Duncan Neurological Research Institute, Texas Children’s Hospital, Houston, United States
    2. Center for Alzheimer’s and Neurodegenerative Diseases, Baylor College of Medicine, Houston, United States
    Contribution
    Conceptualization, Supervision, Funding acquisition, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-5476-5955
  11. Joshua M Shulman

    1. Jan and Dan Duncan Neurological Research Institute, Texas Children’s Hospital, Houston, United States
    2. Department of Molecular and Human Genetics, Baylor College of Medicine, Houston, United States
    3. Department of Neuroscience, Baylor College of Medicine, Houston, United States
    4. Department of Neurology, Baylor College of Medicine, Houston, United States
    5. Center for Alzheimer’s and Neurodegenerative Diseases, Baylor College of Medicine, Houston, United States
    Contribution
    Conceptualization, Supervision, Funding acquisition, Writing - original draft, Writing – review and editing
    For correspondence
    joshua.shulman@bcm.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-1835-1971

Funding

National Institute on Aging (R01AG057339)

  • Zhandong Liu
  • Juan Botas
  • Joshua M Shulman

National Institute on Aging (R01AG053960)

  • Joshua M Shulman

National Institute on Aging (U01AG061357)

  • Joshua M Shulman

National Institute on Aging (U01AG046161)

  • Joshua M Shulman

Eunice Kennedy Shriver National Institute of Child Health and Human Development (P50HD103555)

  • Joshua M Shulman

National Institutes of Health (S10OD023469)

  • Brandon T Pekarek

National Institutes of Health (S10OD025240)

  • Brandon T Pekarek

Cancer Prevention and Research Institute of Texas (RP200504)

  • Joshua M Shulman

Parkinson's Foundation (PF-PRF-830012)

  • Hui Ye

Huffington Foundation

  • Zhandong Liu
  • Juan Botas
  • Joshua M Shulman

McGee Family Foundation

  • Joshua M Shulman

Duncan Neurological Research Institute

  • Zhandong Liu
  • Ismael Al-Ramahi
  • Juan Botas
  • Joshua M Shulman

Effie Marie Caine Endowed Chair for Alzheimer's Research

  • Joshua M Shulman

Alzheimer’s Association (AARF-21-848017)

  • Hui Ye

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We are grateful to Pinghan Zhao, Tom Lee, Alma Perez, Katy Zhu, Akash Tarkunde and Bismark Amoh for assistance with Drosophila brain dissections. We also thank the Bloomington Drosophila Stock Center, the Vienna Drosophila RNAi Center, FlyBase (Gramates et al., 2017), the Developmental Studies Hybridoma Bank, and BioRender. This study was supported by grants from the National Institutes of Health (NIH) (R01AG057339, R01AG053960, U01AG061357, and U01AG046161). In addition, we utilized BCM core resources, including the Intellectual and Developmental Disabilities Research Center (P50HD103555), Genomic and RNA Profiling (S10OD023469), and Single Cell Genomics (S10OD025240 and Cancer Prevention Research Institute of Texas grant RP200504). HY is additionally supported by the Parkinson’s Foundation (PF-PRF-830012) and the Alzheimer’s Association (AARF-21-848017). JMS was additionally supported by the Huffington Foundation, McGee Family Foundation, the Jan and Dan Duncan Neurological Research Institute at Texas Children’s Hospital, and The Effie Marie Caine Endowed Chair for Alzheimer’s Research.

Senior Editor

  1. Claude Desplan, New York University, United States

Reviewing Editor

  1. Patrik Verstreken, KU Leuven, Belgium

Reviewer

  1. Simon G Sprecher, Universität Fribourg, Switzerland

Version history

  1. Preprint posted: November 16, 2022 (view preprint)
  2. Received: November 29, 2022
  3. Accepted: May 22, 2023
  4. Accepted Manuscript published: May 23, 2023 (version 1)
  5. Version of Record published: June 9, 2023 (version 2)

Copyright

© 2023, Wu et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,165
    Page views
  • 145
    Downloads
  • 1
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Timothy Wu
  2. Jennifer M Deger
  3. Hui Ye
  4. Caiwei Guo
  5. Justin Dhindsa
  6. Brandon T Pekarek
  7. Rami Al-Ouran
  8. Zhandong Liu
  9. Ismael Al-Ramahi
  10. Juan Botas
  11. Joshua M Shulman
(2023)
Tau polarizes an aging transcriptional signature to excitatory neurons and glia
eLife 12:e85251.
https://doi.org/10.7554/eLife.85251

Share this article

https://doi.org/10.7554/eLife.85251

Further reading

    1. Evolutionary Biology
    2. Genetics and Genomics
    Thomas A Sasani, Aaron R Quinlan, Kelley Harris
    Research Article

    Maintaining germline genome integrity is essential and enormously complex. Although many proteins are involved in DNA replication, proofreading, and repair, mutator alleles have largely eluded detection in mammals. DNA replication and repair proteins often recognize sequence motifs or excise lesions at specific nucleotides. Thus, we might expect that the spectrum of de novo mutations – the frequencies of C>T, A>G, etc. – will differ between genomes that harbor either a mutator or wild-type allele. Previously, we used quantitative trait locus mapping to discover candidate mutator alleles in the DNA repair gene Mutyh that increased the C>A germline mutation rate in a family of inbred mice known as the BXDs (Sasani et al., 2022, Ashbrook et al., 2021). In this study we developed a new method to detect alleles associated with mutation spectrum variation and applied it to mutation data from the BXDs. We discovered an additional C>A mutator locus on chromosome 6 that overlaps Ogg1, a DNA glycosylase involved in the same base-excision repair network as Mutyh (David et al., 2007). Its effect depends on the presence of a mutator allele near Mutyh, and BXDs with mutator alleles at both loci have greater numbers of C>A mutations than those with mutator alleles at either locus alone. Our new methods for analyzing mutation spectra reveal evidence of epistasis between germline mutator alleles and may be applicable to mutation data from humans and other model organisms.

    1. Evolutionary Biology
    2. Genetics and Genomics
    Ban Wang, Alexander L Starr, Hunter B Fraser
    Research Article

    Although gene expression divergence has long been postulated to be the primary driver of human evolution, identifying the genes and genetic variants underlying uniquely human traits has proven to be quite challenging. Theory suggests that cell-type-specific cis-regulatory variants may fuel evolutionary adaptation due to the specificity of their effects. These variants can precisely tune the expression of a single gene in a single cell-type, avoiding the potentially deleterious consequences of trans-acting changes and non-cell type-specific changes that can impact many genes and cell types, respectively. It has recently become possible to quantify human-specific cis-acting regulatory divergence by measuring allele-specific expression in human-chimpanzee hybrid cells—the product of fusing induced pluripotent stem (iPS) cells of each species in vitro. However, these cis-regulatory changes have only been explored in a limited number of cell types. Here, we quantify human-chimpanzee cis-regulatory divergence in gene expression and chromatin accessibility across six cell types, enabling the identification of highly cell-type-specific cis-regulatory changes. We find that cell-type-specific genes and regulatory elements evolve faster than those shared across cell types, suggesting an important role for genes with cell-type-specific expression in human evolution. Furthermore, we identify several instances of lineage-specific natural selection that may have played key roles in specific cell types, such as coordinated changes in the cis-regulation of dozens of genes involved in neuronal firing in motor neurons. Finally, using novel metrics and a machine learning model, we identify genetic variants that likely alter chromatin accessibility and transcription factor binding, leading to neuron-specific changes in the expression of the neurodevelopmentally important genes FABP7 and GAD1. Overall, our results demonstrate that integrative analysis of cis-regulatory divergence in chromatin accessibility and gene expression across cell types is a promising approach to identify the specific genes and genetic variants that make us human.