Expression of α4β2α5 receptor component genes is affected by single nucleotide polymorphisms
To identify effects in aging and dementia of gene variants previously shown in younger adults to influence CHRNA5 expression (33) and α5 coding (21, 38), we examined brain expression quantitative trait loci (eQTL). The variants were in weak-moderate linkage disequilibrium in our European ancestry sample (r2 = 0.34), in agreement with previous work (33). We also found that dosage of the A allele of the missense SNP rs16969968 (minor allele frequency (MAF) = 0.33) in the coding region of CHRNA5 (Fig. 1B) was associated with lower CHRNA5 expression (t = -13.93 p = 3.98*10− 40), consistent with existing data (39). A different SNP haplotype in the regulatory region upstream of CHRNA5 (Fig. 1B), denoted here by the A allele of the tag SNP rs1979905 (MAF = 0.43), was associated with higher CHRNA5 expression (t = 27.87, p = 5.94*10− 124) (Fig. 1C). Furthermore, analyses of the coding-SNP rs16969968 and the regulatory-SNP rs1979905 together (Fig. 1D) showed that CHRNA5 expression is predominantly regulated by the regulatory-SNP rs1979905 rather than the coding-SNP, rs16969968, as all rs1979905 A allele non-carriers showed similar levels of CHRNA5 mRNA regardless of the rs16969968 A allele (Nested one-way ANOVA: F(2) = 229.6; Šidák’s post-hoc test for multiple comparisons: rs1979905 A1 vs. A0 p = 8*10− 6, A2 vs. A1 p = 0.004, A2 vs A0 p = 4*10− 6). This was also demonstrated using a conditional eQTL model, where the effect of rs16969968 A allele on CHRNA5 expression was lost when co-varying for rs1979905 A allele (rs16969968A: t = -1.068, p = 0.2815; rs1979905A: t = 21.931, p = 6.140*10− 86). No trans-eQTL effects were detected (Fig. 1E) between either of these SNPs and the expression of required partner nicotinic subunit genes, CHRNA4 and CHRNB2.
To assess the α4 and β2 nicotinic subunits required for the formation of α5-containing α4β2α5 receptors, we extended our eQTL analyses to SNPs in CHRNA4 and CHRNB2, focusing on those associated with altered gene expression or clinical effects(35, 36, 40). Without exception, eQTL SNP effects for these genes were weaker than those of rs16969968 and rs1979905 for CHRNA5: the T allele of CHRNB2 intronic variant rs2072660 (MAF = 0.23) was associated with lower CHRNB2 expression (t = -5, p = 7.05*10− 7), and a similar association with CHRNB2 expression was seen with the T allele of the CHRNB2 non-coding variant rs4292956 (MAF = 0.07) (t = -5.244, p = 2.02*10− 7) (Fig. 1E). For CHRNA4, the G allele of missense variant rs1044396 (MAF = 0.45) in the coding region of CHRNA4 was associated with higher CHRNA4 expression (t = 3.67, p = 0.0003). We also used the Gene Query function of the xQTLServe online tool (DLPFC of 534 ROS/MAP participants) to identify the T allele of the intronic variant rs45497800 as associated with decreased CHRNA4 expression (t = -6.92, p = 1.32 * 10− 11). We then replicated this association in our larger cohort of 924 ROS/MAP individuals (MAF = 0.07) (t = -4.57, p = 5.54*10− 6) (Fig. 1E). No associations were found between these SNPs and the expression of other high-affinity nicotinic receptor subunit genes (Fig. 1E).
CHRNA5 polymorphisms are not associated with smoking status in this largely nonsmoking population
To investigate previously-reported (25, 41) associations between genotype for the CHRNA5 SNPs and smoking status at baseline (current/former/never smoked), we used a Chi-squared test and found no relationship (rs1979905_A : χ2(4) = 1.575, p = 0.813; rs16969968_A: χ2(4) = 1.317, p = 0.858) in this largely nonsmoking population. Smoking status was not used in further analysis, unless specifically indicated.
A CHRNA5 polymorphism is negatively associated with brain β-amyloid levels
To address interrelationships among nicotinic subunit expression, nicotinic SNPs, and neuropathological and cognitive phenotypes, we used clinical and post-mortem data in a multi-step model. As illustrated in Fig. 1F, both β-amyloid and tau pathology were negatively associated with global cognitive performance proximal to death (tau: t = -21 p = 2.46*10− 84; amyloid: t = -10.4, p = -3.14*10− 24) and positively associated with each other (t = 17.68, p = 6.96*10− 63). Of the nicotinic receptor subunits and SNPs examined, only the SNP increasing CHRNA5 expression had a significant association with AD neuropathology, with the A allele of the regulatory-SNP rs1979905 negatively associated with β-amyloid load (t = -2.551, p = 0.011). This association remained significant after FDR correction (pFDR = 0.032). By contrast, the expression of the major nicotinic subunit genes CHRNA4 and CHRNB2 showed significant positive associations with the last global cognition score, which remained significant after false discovery rate (FDR) correction for multiple comparisons (CHRNA4: t = 2.98, pFDR = 0.01; CHRNB2: t = 3.71, pFDR = 0.0009). Conversely, the rs2072660 T allele, associated with lower CHRNB2 expression, was negatively associated with the last global cognition score (t = -2.14, p = 0.032). A similar negative association with the last global cognition score (t = -2.46, p = 0.014) was found for the T allele of the CHRNB2 SNP rs4292956, and this association remained significant after FDR correction (pFDR = 0.039).
CHRNA5 expression does not tightly correlate with other components of the cholinergic system
To further investigate the interrelationships among CHRNA5 and the major nicotinic subunits as well as other components of the cholinergic system, we performed a series of expression correlation analyses. Most of the major components of the cholinergic system which were detected in bulk DLPFC data of the ROS/MAP individuals (CHRNA2, CHRNA4, CHRNA7, CHRNB2, CHRM3, CHRM1 and ACHE) showed significant positive correlation with each other (Fig. 1G). By contrast, CHRNA5 stood out as showing no positive correlation with any of the other major cholinergic genes and only weak negative correlations with the expression of CHRNB2 and CHRM1 (Table 1). Considering that the expression of CHRNA5, CHRNB2, and CHRNA4 are required for the assembly of the high-affinity α4β2α5 nicotinic receptor, it was surprising to see that CHRNA5 expression was not positively correlated with either CHRNA4 or CHRNB2 (Fig. 1G and Table 1). Therefore, we next investigated whether this lack of correlation may arise from differences in the cell-type specific expression of CHRNA5 compared to the major nicotinic receptor subunits, CHRNA4 and CHRNB2, which are more broadly expressed.
CHRNA5 shows stronger expression in chandelier interneurons than most other cell classes
To assess the cell-type specificity of CHRNA5 expression in the ROS/MAP cohort, we calculated the average CHRNA5 expression per cell type per individual using the genotype-matched single-nucleus RNAseq data available from the DLPFC in a subset of 22 ROS/MAP participants (31)(2 individuals lacked genotyping data). In this small dataset, CHRNA5 was expressed at a low level across a number of different excitatory, inhibitory, and nonneuronal cell types (Fig. 2A), with significantly higher expression in inhibitory PVALB + chandelier cells (as identified by Cain et al. 2020). Chandelier cells had significantly-higher expression of CHRNA5 compared to most other cell types (One-way ANOVA: F(21) = 3.439, p = 6.1*10− 7; Tukey’s post-hoc t-test Chandelier cells vs. 18 out of 19 other cell types: p < 0.05) (Fig. 2A). By contrast, chandelier cell expression of CHRNA4 and CHRNB2 were at a similar level in chandelier cells to their expression levels in many other cell types (Fig. 2B).
To confirm the novel finding that chandelier cells show stronger CHRNA5 expression compared to other classes of neurons, we probed publicly available cell-type specific gene expression databases of human brain tissue. Using the Allen Institute SMART-seq single-cell transcriptomics data from multiple cortical areas (https://celltypes.brain-map.org/rnaseq/human_ctx_smart-seq), we found CHRNA5 expression to be highest in a PVALB+/SCUBE3 + inhibitory cell type (0.06 trimmed mean CHRNA5 expression) likely representing chandelier cells (42), and in a co-clustering PVALB+/MFI + cell type (0.06 trimmed mean CHRNA5 expression). In the Human Protein Atlas database (brain single cell tissue) (https://www.proteinatlas.org/ENSG00000169684-CHRNA5/single+cell+type/brain), CHRNA5 showed highest expression in a PVALB + inhibitory cell type (c-41) which also showed highest expression of SCUBE3 (Inhibitory neurons c-41, 15.1 normalized CHRNA5 transcripts per million), likely also representing chandelier cells.
To investigate the cell type-specificity of the CHRNA5 eQTL effects of the regulatory-SNP rs1979905 in the single nucleus data from ROS/MAP, we stratified the averaged CHRNA5 expression by genotype for the rs1979905 A allele. We found that higher allelic dosage of the rs1979905 A allele was associated with greater CHRNA5 expression (Fig. 2C), and that this pattern was most pronounced in subtypes of layer 5 (L5 RORB IT: t = 2.460, p = 0.0249) and layer 6 (L6 IT THEMIS: t = 2.402, p = 0.028) excitatory neurons (Fig. 2D). In the stronger CHRNA5-expressing PVALB + chandelier cells, however, the additive pattern did not reach significance. Other neuronal cell types appeared to diverge completely from the typical eQTL pattern of rs1979905 (Fig. 2C,D). This analysis suggests that only a subset of cell-types contribute to the stepwise expression pattern observed in the prefrontal cortex by rs1979905 genotype.
Cell type specific data from ROS/MAP and the Allen database supports the hypothesis that CHRNA5 possesses a distinctive expression pattern with enrichment in chandelier cell interneurons, compared to its more abundant and widely-expressed subunit partners.
A genotype-specific reduction in proportion of chandelier cells with increasing brain β-amyloid levels
To determine whether impaired function/trafficking α5-containing nicotinic receptors might promote chandelier neuron vulnerability to neurodegeneration, we examined the interaction of rs16969968 genotype and AD neuropathology on estimated proportions of chandelier neurons in the bulk RNAseq dataset. This investigation was based on cell type proportion estimates for chandelier cells and several other interneuron subclasses, from sets of single-cell-informed marker genes, in a subset of 640 ROS/MAP participants Overall, β-amyloid levels were negatively associated with the proportion of chandelier cells (t = -4.024, p = 7*10− 5). However, the missense rs16969968 A allele homozygotes showed significantly lower chandelier cell proportions with increasing β-amyloid load (Fig. 4A), compared to rs16969968 A allele non-carriers (interaction term t = -2.799, p = 0.005)( Fig. 4B). In a secondary analysis, a near-significant positive association of rs1979905 A allele and β-amyloid levels with chandelier cells was observed (Fig. 4C), with rs1979905 A allele homozygotes showing a trend towards a smaller drop in chandelier cell proportions with increasing β-amyloid, compared to rs1979905 A allele non-carriers (interaction term t = 1.769, p = 0.078)(Fig. 4D). The observed relationships between chandelier cell proportions, CHRNA5 genotypes and amyloid were not altered by the inclusion of smoking status as a covariate in the analysis. Chandelier cell proportions did not correlate with tau pathology (t = -0.307, p = 0.759), nor was there an interaction of CHRNA5 genotype and tau pathology with chandelier cell proportions (data not shown).
To assess whether the interaction between β-amyloid levels and CHRNA5 SNPs was driven by the effects of these SNPs on CHRNA5 expression, we assessed the effect of the interaction of CHRNA5 levels and β-amyloid load on chandelier cell proportion but found no significant effect (t = 2.711, p = 0.544). This suggests that the genotype-specific association between amyloid load and chandelier cell proportion is more likely driven by changes in nicotinic α5 protein structure and/or trafficking (52) as a consequence of having two copies of the missense SNP in CHRNA5.
The schematic in Fig. 5 illustrates a working model of the impact of the rs16969968 A allele homozygosity for chandelier cell vulnerability, as well as example mechanisms enriched in chandelier cells and known to alter β-amyloid processing.