Cell subtype-specific effects of genetic variation in the Alzheimer's disease brain

Abstract

The relationship between genetic variation and gene expression in brain cell types and subtypes remains understudied. Here, we generated single-nucleus RNA sequencing data from the neocortex of 424 individuals of advanced age; we assessed the effect of genetic variants on RNA expression in cis (cis-expression quantitative trait loci) for seven cell types and 64 cell subtypes using 1.5 million transcriptomes. This effort identified 10,004 eGenes at the cell type level and 8,099 eGenes at the cell subtype level. Many eGenes are only detected within cell subtypes. A new variant influences APOE expression only in microglia and is associated with greater cerebral amyloid angiopathy but not Alzheimer's disease pathology, after adjusting for APOEε4, providing mechanistic insights into both pathologies. Furthermore, only a TMEM106B variant affects the proportion of cell subtypes. Integration of these results with genome-wide association studies highlighted the targeted cell type and probable causal gene within Alzheimer's disease, schizophrenia, educational attainment and Parkinson's disease loci.

Figure 1.. Study design and summary of cell type and subtype specific cis-eQTL.

a, Schema of our study. b, UMAP visualization of 1,509,626 nuclei from 424 donors. Each of the seven major cell types is labeled with a different color. c, Number of eGenes (genes targeted by a cis-eQTL effect) detected within each of the 7 cell types. d, Number of eGenes detected in each of the 64 cell subtypes that were retained for analysis. e,f, Relationship between cell (sub)type proportions and number of eGenes detected. Dashed line shows a least-squares fit with zero intercept, and β is its slope. The shaded area represents the 95 percent confidence interval. e, cell type proportions. f, cell subtype proportions; the slope is much steeper than for cell types, as illustrated by the inset which enlarges the plotting of data near the origin. g, Number of eGenes that are unique to the analysis of cell subtypes: for each cell type, we present a Venn diagram summarizing the extent to which cell type-level eGenes are found once the cells assigned to a given cell type are partitioned into the subtypes of that cell type; the six most common cell types are shown. For each cell type, the set of eGenes identified in all subtypes of a given cell type are shown in gray, and, in each cell type, a subset of these cell subtype eGenes are not recovered in the cell-type level analysis, suggesting that they may be specific to a cell subtype context.

Figure 2.. Similarities and differences of cell type-specific eQTL.

a, Number of cell type-specific and non-specific eGenes. b, Number of eGenes that were unique to or shared between cell types. Only the top 20 intersections are shown. In the vertical bar chart, proportion of eGenes specifically expressed in one cell type are colored gray, and that expressed in two or more cell types are colored blue: most genes are expressed in more than one cell type. c, π₁ statistic to quantitate the extent of eQTL sharing between each pair of cell type. d, Proportion of shared eQTL that had consistent direction of effect in each pair of cell types. e,f, Examples of cell type-specific eQTL. p-values were computed by a simple linear regression between allele dosage and cell type-level gene expression. The numbers of participants are shown in parentheses. e, oligodendrocyte-specific eQTL between rs128648 and APP gene expression. f, microglia-specific eQTL between rs2288911 and APOE gene expression. Elements of boxes show the following statistics: center line, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range; points, outliers. Measurements were taken from distinct samples in e and f.

Figure 3.. Replication of eQTL by bulk cortical RNA and RNA from induced cell lines.

a, Number of eGenes detected in our single-nucleus eQTL study vs. a bulk cortical RNA eQTL study using the same brain region (DLPFC). For the single-nucleus eQTL results, unique eGenes of the seven cell types were combined. For bulk cortical eQTL, we mapped cis-eQTL using 1,092 individuals from the ROSMAP studies. b, π₁ statistic of single-nucleus eQTL and bulk eQTL. The top and bottom rows used bulk eQTL as query and reference, respectively. c, Comparison of our single nucleus-derived eGenes and those from an earlier study that had a smaller sample size and combined a mixture of brain tissues to produce results only at the cell type level. d, Number of cell type-level eGenes that were replicated in 44 neuronal (iN) and 38 astrocyte (iA) cell lines derived from iPSC of the ROSMAP participants. The lead eSNP of each eGene was tested whether its allele dosage was associated with the eGene expression level in the corresponding iPSC-derived data with an FDR < 0.05. We replicate more of the eGenes than expected by chance. ***, P < 1 × 10⁻⁶ by one-sided permutation test. e,f, Effect size β of eQTL shared between single-nucleus and induced cells. r, Pearson’s correlation coefficient. p-values were computed by two-sided t-tests. e, Excitatory neurons and iNs. 234 eGenes; 95% confidence interval (CI) of r, 0.60 – 0.74. f, Astrocytes and iAstro. 121 eGenes; 95% CI of r, 0.59 – 0.78. g, The opposite direction of effect for the MAPT eQTL at rs111600065. A linear regression was applied to MAPT expression and allele dosage. Effect size β and its S.E. are shown in the figure. The major “A” allele tags the H1 haplotype of MAPT, and the minor “C” allele tags the H2 haplotype, associated with Parkinson’s disease, Parkinsonism, and perhaps AD^,. Elements of boxes show the following statistics: center line, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range; points, outliers. Measurements were taken from distinct samples. The numbers of participants are shown in parentheses.

Figure 4.. Chromatin states of cell type-level eQTL.

a, Distance between eSNPs and transcription start sites (TSSs) of eGenes. b, Enrichment of eSNP in chromatin states. Chromatin states of a human DLPFC tissue were obtained from sample E073 of Roadmap Epigenomics. TssA, active TSS; TssAFlnk, flanking active TSS; TxFlnk, transcription at gene 5’ and 3’; Tx, strong transcription; TxWk, weak transcription; EnhG, genic enhancers; Enh, enhancers; ZNF/Rpts, ZNF genes & repeats; Het, heterochromatin; TssBiv, bivalent/poised TSS; BivFlnk, flanking bivalent TSS/enhancer; EnhBiv, bivalent enhancer; ReprPC, repressed PolyComb; ReprPCWk, weak repressed PolyComb; Quies, quiescent/low. c, Enrichment of eQTL in cell type-specific enhancers and promoters. Enhancers and promoters of four brain cell types were obtained from a published report. d, Microglia-specific eQTL for APOE (rs2288911) in the context of microglial-specific chromatin conformation data. These data were repurposed from Nott et al.. The x-axis denotes the physical position along a segment of chromosome 19 containing the APOE gene and several related genes; their exon structure is presented in the top horizontal track. The next four tracks report chromatin immunoprecipitation followed by sequencing (ChIP-seq) data against the H3K27Ac epitope, a mark found in active TSS and enhancers; each track presents data from a different cell type, isolated as purified nuclei. The peaks denote segments that are in a transcriptionally active conformation. The next four tracks present data from the same samples using the Assay for Transposase Accessible Chromatin (ATAC) which denotes chromosomal segments that in an open conformation and accessible for transcription. The four H3K4Me3 tracks present ChIP-seq data for that epitope, which is also correlated with active promoter regions of a chromosome. Proximity ligation-assisted ChIP-seq (PLAC-seq) data are presented in the last 4 tracks and denote pairs of chromosomal segments that are in physical proximity to one another as the chromatin loops.

Figure 5.. Fraction QTL (fQTL) between SNPs and cell subtype proportions.

a, Number of independent significant fQTL per cell subtypes. b, Locuszoom plot around the TMEM106B locus. A lead risk SNP of AD in the locus (rs5011436) was used as the reference SNP given that it tags a risk haplotype that contains many other SNPs with similar statistical properties given their strong LD. The y-axis shows −log₁₀ of p-values for the Excitatory neuron subtype 3 (Exc.3) cell subtype fQTL. c, Manhattan plot for genome-wide fQTL results for Exc.3. The dashed line shows the significance threshold p < 5.6 × 10⁻¹⁰ (accounting for all tested fQTL GWAS). d, Genotypes of rs5011436 and proportion of Exc.3 among excitatory neurons. The numbers of participants are shown in parentheses. e, Genotypes of rs5011436 and gene expression levels of TMEM106B in bulk DLPFC tissues, illustrating the local cis-eQTL. The numbers of participants are shown in parentheses. Elements of boxes show the following statistics: center line, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range; points, outliers. Measurements were taken from distinct samples in c, d, and e. f, Schematic summary of the major allele of rs5011436 and its relationship to three phenotypes. FTD, frontotemporal dementia. Red and blue lines show positive and negative correlations, respectively. p-values were computed by two-sided t-tests in b, c, d, and e.

Figure 6.. Overlap of results for our eQTL and GWAS of selected neurodegenerative and neuropsychiatric disease.

a–c, Colocalization of cell type-specific eQTL with risk SNPs of (a) AD GWAS, (b) Parkinson’s disease GWAS, and (c) schizophrenia GWAS. Each of the heatmaps report the posterior probabilities of the H4 hypothesis (PP.H4) of the coloc method, which assumes GWAS and eQTL share a single causal SNP. Rows report overlap for individual genes and SNP pair; columns report PP.H4 score in each of our cell types. The color of each cell is based on the code found to the right of each panel; the darker color denotes higher confidence that the same variant influences susceptibility and gene expression in that cell type. Grey cells indicate that the gene was not an eQTL target in that cell type. Top bar chart shows the number of colocalized eGenes with high confidence (PP.H4 > 0.8) in each cell type. d, Cell type-level transcriptome-wide association studies (TWAS). Using the FUSION method, we deployed instruments inferring the expression of 28,305 genes across all cell types in the summary statistics for AD, PD, ALS, and schizophrenia. The count of genes meeting a transcriptome-wide threshold of significance in each cell type is presented for each disease, with the expected excess of microglial genes in AD and intriguing number of oligodendroglial genes in PD. Cell types are in order of descending expression heritability. e, Illustration of the TWAS result from one cell type in one disease: the statistical significance and effect direction of all inferred microglial genes are presented with the physical position along the chromosome being presented on the x-axis and the significance threshold, on the y-axis. Each dot represents a gene. Positive and negative y coordinates show that transcript abundance was associated with increased and decreased risk of AD, respectively. The y-axis between −10 and 10 are enlarged to enhance visibility. Novel and known candidates for AD risk genes in microglia are colored black and gray, respectively. Red dashed lines highlight the threshold of FDR = 0.05. p-values were computed by two-sided z-tests.