A tau homeostasis signature is linked with the cellular and regional vulnerability of excitatory neurons to tau pathology

Abstract

Excitatory neurons are preferentially impaired in early Alzheimer's disease but the pathways contributing to their relative vulnerability remain largely unknown. Here we report that pathological tau accumulation takes place predominantly in excitatory neurons compared to inhibitory neurons, not only in the entorhinal cortex, a brain region affected in early Alzheimer's disease, but also in areas affected later by the disease. By analyzing RNA transcripts from single-nucleus RNA datasets, we identified a specific tau homeostasis signature of genes differentially expressed in excitatory compared to inhibitory neurons. One of the genes, BCL2-associated athanogene 3 (BAG3), a facilitator of autophagy, was identified as a hub, or master regulator, gene. We verified that reducing BAG3 levels in primary neurons exacerbated pathological tau accumulation, whereas BAG3 overexpression attenuated it. These results define a tau homeostasis signature that underlies the cellular and regional vulnerability of excitatory neurons to tau pathology.

Figure 1.. Excitatory and inhibitory neurons are differentially vulnerable to tau pathology in primary and secondary affected regions of EC-tau mice.

(a, b) Representative images of MC1-positive (+) tau staining co-localized with TBR1+ and SATB2+ excitatory (EX) neurons, but not PVALB+, SST+ or CALB2+ inhibitory (IN) neurons, in the MEC of EC-tau mice at 22 months (a) and at 30+ months (b). Three independent experiments were repeated with similar results. Scale bar, 20 μm. c) Co-localization ratio of MC1+ tau with neuronal marker+ neurons, which was quantified in the MEC, PRH and NC (layers II-IV) of EC-tau mice at 22 and 30+ months. (d, e) Number of neuronal marker+ neurons (d) and MC1+ cells (e), which was counted in the above regions of EC-tau mice at 22 and 30+ months. (f) Number of TBR1+ and SATB2+ EX neurons, which was compared in the MEC of non-transgenic (WT) mice at 22 and 30+ months. Data are presented as mean ± SEM (n = 6 animals, 2 sections each animal. If the section does not have any MC1+ neurons, it will be removed out of the analysis, e.g. PRH-22 mo: n = 9; PRH-30+ mo: n = 11; NC-22 mo: n = 6; NC-30+ mo: n = 11 independent sections) (c, d, f) and (n = 7 independent experiments, each value is the average of 12 biological independent sections) (e). Statistical significance was assessed by Kruskal-Wallis test with the Dunn’s multiple comparison test (c) and one-way ANOVA with Tukey’s post-test (d). *** P < 0.0001 vs PVALB, SST and CALB2 (the same brain regions and ages of the mice) (The Kruskal-Wallis statistic is 53.16, 53.09, 41.17, 49.65, 29.37 and 48.02, respectively.) (c); *** P < 0.0001 vs 30+ months (the same brain regions and neuronal markers) (The R squared = 0.6996, F = 34.16) (d). In (e, f), statistical significance was assessed by two-tailed unpaired t test. NS, not significant; *** P < 0.0001 vs 22 months (The statistic is t=6.921, df=12; t=8.833, df=12; t=16.56, df=12 (e); t=0.2748, df=22; t=0.2040, df=22 (d), respectively.).

Figure 2.. EX and IN human neurons are differentially vulnerable to tau pathology in primary affected regions of AD brain.

(a, b) Representative images of MC1-positive (+) tau staining co-localized with TBR1+ and SATB2+ EX neurons, but not PVALB+, SST+ or CALB2+ IN neurons, in the EC of AD patient brain at Braak stage II (a) and Braak stage V-VI (b). Three independent experiments were repeated with similar results. Scale bar, 20 μm. (c) Co-localization ratio of MC1+ tau with neuronal marker+ neurons, which was quantified in the EC layer II-IV of AD brains at different Braak stages; data are presented as mean ± SEM (n = 3 cases, 2 sections each case), and the statistical significance was assessed by Kruskal-Wallis test with the Dunn’s multiple comparison test. *** P < 0.0001 vs PVALB, SST and CALB2 (the same Braak stage) (The Kruskal-Wallis statistic is 9.280, 25.82 and 24.90, respectively.). (d, e) Number of neuronal marker+ neurons (d) and MC1+ cells (e), which were assessed in EC layer II-IV of AD brains at different Braak stages; data are shown as the percentage of the average number of neuronal marker+ cells at Braak stage I-II and are presented as mean ± SEM (n = 3 cases, 2 sections each case), and the statistical significance was assessed by one-way ANOVA with Tukey’s multiple comparison post hoc tests (The R squared = 0.6026, F = 11.37; R squared = 0.5187, F = 8.082) (d) or two-tailed Unpaired t test with Welch’s correction (e). ** P < 0.01; *** P < 0.001 vs Braak stage I-II (The statistic is t=6.369, df=6; t=4.150 df=6).

Figure 3.. Single-nucleus RNA-seq analysis reveals a specific tau homeostasis signature in EX neurons in human brains.

(a, b) Comparison of the differential expression of relevant subproteomes for different cell types. For each subproteome (and the whole transcriptome as a control), the difference between the mean expression in EX and IN neurons (measured by the Δ score, see Methods) was calculated, and the values are presented as mean ± SEM. In (a, b) results are reported for the SNS and the DroNc-Seq datasets, respectively. (c, d) Comparison of Δ scores for five subproteomes (and the whole transcriptome as a control) within the EX neurons, between regions affected relatively early or late in AD for the SNS and DroNc-Seq datasets, respectively. The significance was evaluated by building a null model for each subproteome (see Methods, Supplementary Table 1 and Supplementary Figures 4–6) and corrected with a Benjamini-Hochberg multiple hypothesis testing correction *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. Subproteomes (where n_sns and n_drnc are the sample sizes corresponding to SNS and DroNc-seq datasets respectively): EX markers (excitatory markers, a set of genes specific for excitatory neurons): n_sns=n_drnc=2; promoters (a set of proteins promoting tau aggregation): n_sns=n_drnc=6; MS (metastable subproteome, a subset of highly expressed and aggregation-prone proteins, which are supersaturated – i.e. proteins whose concentration in the cellular environment is higher than a critical value keeping them soluble and functional – and downregulated in AD): n_sns=162, n_drnc=179; transcriptome (the whole transcriptome, here reported as a negative control); tangles (proteins co-aggregating with tau and found in neurofibrillary tangles): n_sns=57, n_drnc=68; protectors (a set of proteins protecting tau from aggregating): n_sns=n_drnc=6; IN markers (inhibitory markers, a set of genes specific for inhibitory neurons): n_sns=n_drnc=3.

Figure 4.. Single-nucleus RNA-seq analysis shows high levels of tau aggregation protectors in glia.

Differential expression of relevant subproteomes for different cell types. For each subproteome (and the transcriptome of reference as a control) the difference between the mean expression in glia and neurons (measured by the Δ score, see Methods), within cell-types from different regions was calculated. In (a, b, c) differential expression values between glia and EX neurons are reported. Specifically, results are reported for (a) microglia (MG), (b) astrocytes (ASC1, ASC2), and (c) oligodendrocytes (ODC1, ODC2), respectively. (d, e, f) Differential expression between glia and IN neurons are reported, with values corresponding to (d) microglia (MG), (e) astrocytes (ASC1, ASC2), and (f) oligodendrocytes (ODC1, ODC2), respectively. For each bar, the significance was evaluated by building a null model for each subproteome and corrected with a Benjamini-Hochberg multiple hypothesis testing correction *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001 (see Methods, Supplementary Table 2 and Supplementary Figures 8–10). Results are reported for the DroNc-Seq dataset. Subproteomes: tau (MAPT gene); the definition of promoters, transcriptome, tangles, and protectors, and the sample sizes are the same as Figure 3.

Figure 5.. Co-expression network analysis of the subproteomes relevant to tau homeostasis.

Sketch of the co-expression network to identify hub genes of the subproteomes related to tau homeostasis. The network is fully connected, and the edges linking the genes (nodes) are weighted with the Pearson’s correlation coefficient. The hubs, which are defined as the genes more tightly co-expressed with every other gene in the network, and here defined as master regulators, are highlighted with the labels (top 10% of the most co-expressed genes). The size of each node is proportional to the sum of the weights of the edges connected to it. BAG3 is a hub in the protectors region of the network (lower left). The color code identifies the different subproteomes: MS (red), tangles (green), protectors (blue), promoters (yellow), tau (black), and with the genes shared between MS and tangles colored in brown.

Figure 6.. Validation by single-molecule FISH of the localization and mRNA expression levels of representative tau homeostasis signature genes in human EC and prefrontal cortex.

(a, b) Representative sm-FISH images of the co-staining of EX neuronal marker (SLC17A7, red), IN neuronal marker (GAD1, purple), and target probe (MAPT, MAPK1, FKBP5 and ENC1, green) in the EC (a) and the BA9 (b) of human brain without pathological hallmarks of neurodegenerative diseases (54–66 years old); dotted ovals represent individual EX or IN neurons. Three independent experiments were repeated with similar results. Scale bar, 10 μm. (c, d) Comparison of the number of single RNAs of the target probe in individual EX and IN neuron in the EC (c) and the BA9 (d) regions (n = 4 human brains, 10 neurons from each case). Data are presented as mean ± SEM. The statistical significance was assessed by two-tailed unpaired t test with Welch’s correction. *** P < 0.0001 vs IN neurons (The statistic is t=8.061, df=47; t=6.181 df=42; t=10.77, df=39 (c); and t=7.981, df=48; t=4.675, df=56; t=12.60, df=41 (d), respectively.).

Figure 7.. Modulating the expression of one of the tau aggregation protectors, BAG3, affects tau accumulation in primary cortical neurons.

(a) Representative western blot images of primary cortical neurons transduced with lentivirus expressing scrambled BAG3 or shBAG3, or overexpressing BAG3 (OE) as described in Online Methods. GAPDH is a housekeeping protein used as the loading control. Three independent experiments were repeated with similar results. Full length of the blot can be found in the Supplementary Fig. 11. (b) The percentage of EX and IN neurons (n = 55 from 11 coverslips each group) with 12E8 (pS262 and/or pS356 tau)-positive (+) puncta (≥ 5) in the neurites was quantified as described in Online Methods. (c) Representative immunocytochemical images of 12E8+ (red) puncta (white arrow heads) in the neurites of TBR1+ (green) EX neurons. White arrow indicates a neuron with high expression of TBR1; yellow arrow indicates a neuron with low expression of TBR1. GAD1+ (purple) IN neurons were also transduced with shBAG3 lentivirus and tau was shown to accumulate in neurites (white arrowheads). Three independent experiments were repeated with similar results. (d) Representative immunocytochemical images of tau inclusions (green) in TBR1+ (red) EX neurons (white dotted circle) (white arrow, high expression of TBR1; yellow arrow, low expression of TBR1) and tau inclusions (green) in GAD1+ (purple) IN neurons (yellow dotted circle) transduced with different lentiviruses as described in Online Methods. The nuclei were counterstained with Hoechst33342 (blue). Three independent experiments were repeated with similar results. Scale bars, 50 μm (c); 20 μm (d). (e, f) The quantitation of the number of TBR1+ EX and GAD1+ IN neurons with tau inclusions (n = 80 region of interests (ROI) from 4 coverslips per group). (b, e and f) Data are presented as mean ± SEM. Statistical significance was assessed by nonparametric Kruskal-Wallis test with the post hoc test of Dunn’s multiple comparisons. *** P < 0.0001 vs neurons transduced with scramble BAG3 (The Kruskal-Wallis statistic is 34.54, 164.6 and 20.09, respectively.).