Post by Admin on Oct 13, 2023 5:23:29 GMT
Figure S3
Loss of NDNF-PROX1 inhibitory neurons is associated with an upregulation of excitatory neuron DE signature, related to Figure 3
(A) Distribution of Z scores per cell class for the analysis shown in Figure 3A. Each small line indicates one cell type and the tick lines represent the mean.
(B) Logistic mixed-effect model regression (STAR Methods) of NDNF-PROX1 proportion versus ExN cell-type transcriptional signature in Aꞵ+ subjects with added poisson noise. Poisson noise counts were added to the UMI counts of each gene in each cell prior to computing the regression. Boxplots show the distribution of −log10 transformed p values over 30 noise iterations. Center line, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range; points, outliers.
(C) Logistic mixed-effect model regression of NDNF-PROX1 proportion versus ExN cell-type transcriptional signature in Aꞵ+ subjects with downsampling of cells. Iterating 30 times, we randomly downsampled each ExN type to 7,000 cells, unless the cell-type size was less than this number. Boxplots show the Z score distributions over the 30 downsampling iterations. The dashed line indicates the FDR threshold of 0.05. Center line, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range; points, outliers.
(D) Association between NDNF-PROX1 loss and LINC00507-COL5A2 cell type assessed using an alternative strategy of constructing a meta gene of the ExN DE signature from the first principal component (STAR Methods). Each small line indicates one cell type and the tick lines represent the mean.
(E) Logistic mixed-effect model regression of NDNF-PROX1 proportion versus ExN cell-type transcriptional signature in Aꞵ+ subjects after randomizing assignment of cells to excitatory cell types. Dashed line represents FDR threshold of 0.05.
(F and G) Logistic mixed-effect model regression (STAR Methods) of NDNF-PROX1 proportion versus ExN cell-type transcriptional signature in control (F) and Aꞵ+Tau+ (G) subjects. The inset boxplot in (F) shows the overall distribution of Z scores among the 17 excitatory neuron types. Dashed line represents FDR threshold of 0.05.
(H) Boxplots representing the association (as measured by Z score from logistic mixed-effect model regression) between each inhibitory neuron cell type (x axis) with the ExN DE signature across 17 ExN types (boxplots). The red dot represents the Z score of the LINC00507-COL5A2 type. Analysis is based on Aꞵ+ individuals only. In boxplots, center line, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range.
(I) Barplots representing the relative frequencies of the five astrocyte cell types.
(J) Dot plots showing expression of astrocyte marker genes in each astrocytic cell type.
(K) LogFC expression of APP, CDK5, and SNAP25 genes in LINC00507-COL5A2 excitatory neurons across increasing Aꞵ burden scores. Each dot represents the logFC in a sample with Aꞵ pathology versus control samples. Regression line is illustrated in blue with associated standard error.
We next sought to better understand the association of the ExN DE signature with Aꞵ plaque pathology. Comparison of ExN DE genes between Aꞵ+ and Aꞵ+Tau+ biopsies revealed a bimodal pattern among upregulated DE genes (Figure 3C), in which one DE gene set was evident solely in the early stage of pathology, whereas another was present in both early- and late-stage samples. Given their differences in expression trajectory during AD pathological progression, we asked whether these two sets of DE genes were differentially correlated with NDNF-PROX1 inhibitory loss. Only the DE genes specifically found in response to early AD pathology, particularly within the LINC00507-COL5A2 population, correlated with NDNF-PROX1 proportional loss (Figure 3D).
The activity of L1 NDNF-expressing InNs plays a crucial role in the integration of long-range inputs into cortex, particularly through gain modulation of whole cortical columns.24,25 We wondered if their loss might alter the excitability of nearby L2/3 pyramidal cells. Indeed, we identified a significant association between NDNF-PROX1 loss and upregulation of neural activity response genes23 specifically within LINC00507-COL5A2 ExNs in Aꞵ+ individuals (FDR-adjusted p value < 0.01; Figure 3E). Furthermore, Aꞵ+ biopsy samples with a greater proportional loss of NDNF-PROX1 cells showed a higher percentage of LINC00507-COL5A2 cells expressing canonical activity-regulated genes26 (FOS, JUNB, ARC, NPAS4, ERG1, and ERG2; p value < 0.014; Figure 3F). Increased activity of ExNs would be expected to alter their metabolism. Consistent with this, gene set enrichment analysis (GSEA) demonstrated increased expression of metabolism- and mitochondria-related gene sets specifically in biopsies with the lowest level of Aꞵ plaque burden, further reinforcing the relevance of the hyperactivity phenotype to the early stages of AD pathology (Figure 3G). The enrichment of these terms was diminished in biopsies with higher AD pathological burden (Figure 3G), a pattern that was stronger in upper-layer ExNs. Congruently, comparing samples with the lowest Aꞵ burden with Aꞵ+Tau+ demonstrated a significantly higher divergence of the DE patterns of upper-layer neurons compared with the lower layer ExNs (p value < 0.003; Student’s t test; Figure 3I). Collectively, our results demonstrate that NDNF-PROX1 InN loss is correlated with hyperactivity and preferential loss of L2/3 ExNs in the prefrontal cortex with low Aꞵ plaque burden.
Hyperactivity of neurons can perturb pre- and post-synaptic mechanisms.27 In ExNs from subjects with low Aꞵ burden, we identified upregulation of SNAP25, SYT1, and CDK5, which are involved in presynaptic vesicle release28,29,30 (Figures 3G and 3I). Increased activity of the presynaptic vesicle cycle can elevate Aꞵ production.31 Congruently, we found upregulation of genes encoding for protein components involved in Aꞵ fibril formation, such as APP itself, only in the Aꞵ-low disease samples (Figures 3G and 3I). The oligomeric Aꞵ receptor genes PRNP, ATP1A3, and PGRMC1, whose protein products influence neuronal activity and synapse functioning through the modulation of N-methyl-D-aspartate (NMDA) receptors32 and neuronal calcium signaling,33 were similarly upregulated in ExNs at the early stages of AD pathology (Figure 3I). HM astrocytes also play critical roles in supporting synaptic function and coordinating antioxidant responses, especially in the context of neuronal hyperactivity.34,35 In our integrative analysis of astrocytes, we identified one WIF1+ type with low expression of GFAP and high expression of EAAT1, EAAT2, and GSTP1 genes, which encode for critical components of glutamate/glutathione cycling (Figures S3I and S3J). The WIF1-expressing astrocytes showed enrichment of DE genes related to glutathione metabolism, lysosomal machinery, and fatty acid degradation specifically in subjects with low Aꞵ burden (Figure 3J), consistent with gene sets previously reported to be upregulated in the astrocytic response to hyperactive neurons.35 Together, these results suggest that aberrant activity and metabolism of upper-layer pyramidal cells perturb synapse homeostasis and astrocyte functioning in the brain.
Through multiple lines of evidence, we identified hyperactivity of the upper-layer pyramidal neurons in biopsy samples with low Aꞵ burden that is associated with a selective loss of upper-layer inhibitory and excitatory neurons. To confirm these patterns and establish their functional relevance to AD pathological course, we conducted electrophysiological experiments on acute slices from frontal cortex biopsy samples of an independent cohort of 26 living individuals, including eight samples with low levels of Aꞵ, eight samples with a high burden of Aꞵ, and 10 Aβ-free control samples. To assess hyperactivity, baseline neuronal activity was induced in control and Aꞵ+ samples through treatment with NMDA (Figure S4). In addition to L2/3 pyramidal neurons, we also measured spike activity levels in L5 ExNs as a negative control. Consistent with our results from the analysis of single-nucleus transcriptome data, we observed higher bursting activity in upper-layer ExNs from samples with low Aꞵ burden (Figure 3K). In contrast, L5 ExNs exhibited similar activities between Aꞵ+ and control samples (Figure 3K). These findings provide direct functional evidence of hyperactivity of L2/3 ExNs in the human brain at the early stages of AD pathology.