Cellular underpinnings of the selective vulnerability to tauopathic insults in Alzheimer’s disease

Neurodegenerative diseases such as Alzheimer’s disease (AD) exhibit pathological changes in the brain that proceed in a stereotyped and regionally specific fashion, but the cellular and molecular underpinnings of regional vulnerability are currently poorly understood. Recent work has identified certain subpopulations of neurons in a few focal regions of interest, such as the entorhinal cortex, that are selectively vulnerable to tau pathology in AD. However, the cellular underpinnings of regional susceptibility to tau pathology are currently unknown, primarily because whole-brain maps of a comprehensive collection of cell types have been inaccessible. Here, we deployed a recent cell-type mapping pipeline, Matrix Inversion and Subset Selection (MISS), to determine the brain-wide distributions of pan-hippocampal and neocortical neuronal and non-neuronal cells in the mouse using recently available single-cell RNA sequencing (scRNAseq) data. We then performed a robust set of analyses to identify general principles of cell-type-based selective vulnerability using these cell-type distributions, utilizing 5 transgenic mouse studies that quantified regional tau in 12 distinct PS19 mouse models. Using our approach, which constitutes the broadest exploration of whole-brain selective vulnerability to date, we were able to discover cell types and cell-type classes that conferred vulnerability and resilience to tau pathology. Hippocampal glutamatergic neurons as a whole were strongly positively associated with regional tau deposition, suggesting vulnerability, while cortical glutamatergic and GABAergic neurons were negatively associated. Among glia, we identified oligodendrocytes as the single-most strongly negatively associated cell type, whereas microglia were consistently positively correlated. Strikingly, we found that there was no association between the gene expression relationships between cell types and their vulnerability or resilience to tau pathology. When we looked at the explanatory power of cell types versus GWAS-identified AD risk genes, cell type distributions were consistently more predictive of end-timepoint tau pathology than regional gene expression. To understand the functional enrichment patterns of the genes that were markers of the identified vulnerable or resilient cell types, we performed gene ontology analysis. We found that the genes that are directly correlated to tau pathology are functionally distinct from those that constitutively embody the vulnerable cells. In short, we have demonstrated that regional cell-type composition is a compelling explanation for the selective vulnerability observed in tauopathic diseases at a whole-brain level and is distinct from that conferred by risk genes. These findings may have implications in identifying cell-type-based therapeutic targets.


Supplemental Figures
Refer to Table S1 and the original manuscript for further details on these cell types.S3 for descriptions of these datasets.Transmembrane protein 41A Unclear primary biological function

Trem2
Triggering receptor expressed on myeloid cells 2 Disease-associated microglia activation Table S5: AD-associated genes.List of the names and brief descriptions of the genes examined using univariate (Figure 4) and multivariate (Figure S11 and Figure S13) selective vulnerability analyses, each of which has one or more variants associated with AD incidence.These genes represent an intersection between the list given by the Alzheimer's Disease Sequencing Project [8,9] and the coronal series of the Allen Gene Expression Atlas (AGEA) [10], which yielded 24 genes.Gene annotations were obtained from the UniProt database [11]

Figure S1 :Figure S2 :
Figure S1: Correlation structure of the Yao, et al. cell types.A. Heat map of the Pearson correlations of the gene expression profiles of the Yao, et al. [1] cell types.B. Heat map of the Pearson correlations of the regional distributions of the Yao, et al. cell types as inferred by MISS [2].

Figure S3 :Figure S4 :Figure S5 :
Figure S3: Distributions of hippocampal glutamatergic neurons.Sagittal views of the threedimensional reconstructions of brain-wide densities of the hippocampal glutamatergic neurons in the Yao, et al. dataset [1].Refer to TableS1and the original manuscript for further details on these cell types.

Figure S6 :
Figure S6: Statistical significance of the correlations in Figure 2. Heat map of the nominal -log 10 (p) values for the correlations presented in Figure 2A.The critical value corresponding to a Bonferroni-corrected significance level of 0.05 is -4.0 (red arrow).

Figure S8 :
Figure S8: Correlation structure of the mouse tauopathy datasets Heat maps of the Pearson correlations between time points of the nine mouse tauopathy experiments analyzed in this study [3, 4, 5, 6, 7].

Figure S9 :BFigure S10 :
Figure S9: BIC plots for the multivariate linear models in Figure 3. Scatter plots of the BIC criterion with respect to the number of cell types added to the model (n) to determine the optimal sets for each tauopathy dataset.

Figure S11 :
FigureS11: Multivariate analysis of end-timepoint pathology, AD genes (BIC).Scatter plots of the optimal cell-type-based models of tau pathology at the end time points for each of the nine mouse tauopathy studies, along with their associated R 2 values and the BIC-selected genes.

Figure S12 :
Figure S12: BIC plots for the multivariate linear models in Figure S11.Scatter plots of the BIC criterion with respect to the number of cell types added to the model (n) to determine the optimal sets for each tauopathy dataset.

Figure S13 :
FigureS13: Multivariate analysis of end-timepoint pathology, AD genes.Scatter plots of the optimal cell-type-based models of tau pathology at the end time points for each of the nine mouse tauopathy studies, along with their associated R 2 values and the 5 genes with the highest correlations to pathology.

Figure S14 :
Figure S14: BIC plots for the multivariate linear models in Figure 4. Scatter plots of the BIC criterion with respect to the number of cell types added to the model (n) to determine the optimal sets the Hurtado, et al. dataset [4] for each timepoint.

Table S1 :
[1]tamatergic cell types.List of the abbreviations and names of the glutamatergic cell types used in this study, each of which corresponds to a taxonomic subclass annotated by Yao et al.[1].We have delineated these subclasses as being either cortical or hippocampal based on the annotations of their lower-level clusters.

Table S2 :
[1]Aergic and non-neuronal cell types.List of the abbreviations and names of the GABAergic and non-neuronal cell types used in this study, each of which corresponds to a taxonomic subclass annotated by Yao et al.[1].At the level of subclasses, these types are not uniquely defined between cortical and hippocampal regions.

Table S3 :
Mouse tauopathy datasets.List of the tauopathy datasets explored here with descriptions of four key experimental features: mouse genetic background, injection site, type of τ injected, and the number of regions for which τ pathology was quantified.All studies quantified τ pathology within hemispheres ipsilateral and contralateral to the injection site separately with the exception of Hurtado, which was bilaterally averaged.RH -right hemisphere; LH -left hemisphere; PFF -preformed fibrils; DSAD -Down Syndrome Alzheimer's disease; CBD -corticobasal dengeneration.

Table S4 :
T-test results for cell-type classes.Summary of one-way and two-way t-test results for distributions of Pearson's R values of the four cell-type classes within the Yao, et al. dataset: cortical glutamatergic neurons, hippocampal glutamatergic neurons, GABAergic neurons, and non-neuronal cells (see Figure2C).T-tests were performed after first using the Fisher's R-to-Z transformation on the individual Pearson's R values displayed in Figure2A.The p-values reported have been multiple-hypothesis corrected using the Bonferroni criterion.See TablesS1 and S2for a complete list of the cell types within each class.

Table S6 :
unless otherwise noted.Top-five feature linear model statistics.Statistics corresponding to the linear models shown in Figure S10 and Figure S13.Bold font indicates the best model by the Bayesian Information Criterion (BIC).*: p < 0.01; **: p < 0.001; ***: p < 0.0001.