Molecular and cellular pathology of monogenic Alzheimer’s disease at single cell resolution

Cell and molecular biology analyses of sporadic Alzheimer’s disease brain are confounded by clinical variability, ageing and genetic heterogeneity. Therefore, we used single-nucleus RNA sequencing to characterize cell composition and gene expression in the cerebral cortex in early-onset, monogenic Alzheimer’s disease. Constructing a cellular atlas of frontal cortex from 8 monogenic AD individuals and 8 matched controls, provided insights into which neurons degenerate in AD and responses of different cell types to AD at the cellular and systems level. Such responses are a combination of positively adaptive and deleterious changes, including large-scale changes in synaptic transmission and marked metabolic reprogramming in neurons. The nature and scale of the transcriptional changes in AD emphasizes the global impact of the disease across all brain cell types. One Sentence Summary Alzheimer’s disease brain atlas provides insights into disease mechanisms

3 genes encode for a protease and one of its substrates, and act in a common pathogenic pathway (9). Given the early onset of monogenic AD (4,5), comparisons with age-matched controls reduces the impact of age as a confounding factor for interpreting cellular changes (10)(11)(12).
We analyzed gene expression in single nuclei from post-mortem frontal cortex (Brodmann area 5 9) of 8 individuals with monogenic AD carrying PSEN1 Intron4, M146I or APP V717 mutations, and 8 age-and gender-matched controls (Fig. 1A, Data S1). Neuronal and non-neuronal/glial nuclei were separated by FACS, enabling equal representation of neurons and glial cells in the dataset (NeuN + and NeuNrespectively; Fig. S1, A to D). Droplet-based single-nucleus RNA sequencing (snRNA-seq; see Methods for details) was carried out separately for neuronal and 10 glial nuclei. A two-step process using cell types of human middle temporal gyrus from snRNAseq data generated by the Allen Institute for Brain Science as a reference (13) resulted in a final dataset of 89,325 high confidence nuclei (64,408 from controls and 24,917 from monogenic AD), which was used for all subsequent analyses. Consistent with previous studies of sporadic (8) and monogenic AD (4,5), nuclear sorting identified a marked reduction in the number of 15 neurons (NeuN + ) in monogenic AD patients with PSEN1 or APP mutations ( Fig. 1B; Fig. S1, B and C) (14,15). The neuronal nuclei from non-demented control brains had a higher mRNA content compared to glial nuclei (13,16) (Fig. 1C and Fig. S1, E and F). In contrast we found that there was a marked reduction in the mRNA content of neuronal and glial nuclei from monogenic AD cortex, compared with their counterparts in non-demented controls (Fig. 1C and 20 Fig. S1, E and F).
Previous studies of sporadic AD brain identified loss of excitatory neurons from the cerebral cortex, most notably from entorhinal cortex early in the disease (17)(18)(19)(20). To determine the degree 4 of selective neuronal vulnerability in monogenic AD, we compared the proportions of the 10 types of excitatory and 10 types of inhibitory neurons, and 6 glial cell types in the monogenic AD frontal cortex with that of non-demented controls ( Fig. 1D-G). For broad categories of neuronal types, we found that both excitatory and inhibitory neurons were significantly reduced in AD (Fig. 1E). In contrast, the proportions of glial cells were not significantly different, 5 although there was a relative increase in both astrocytes and oligodendrocytes, most probably reflecting the loss of neurons (Fig. 1E). While excitatory neurons were broadly lost from the AD cortex, certain classes of excitatory neurons were disproportionately reduced, most notably classes of layer 3/4 (ExcB1) and layer 4-6 neurons (ExcB4; Fig. 1F). In contrast, almost all subtypes of interneurons were significantly reduced in the frontal cortex of monogenic AD 10 patients (Fig. 1G). These findings were confirmed by cell counting in tissue sections from the same monogenic AD and control individuals (Fig. 1, H and I), with, for example, one of the most abundant interneuron subtypes (Parvalbumin + , PVALB) reduced by 62% in AD cortex (Fig. 1I).
The large-scale loss of inhibition due to interneuron degeneration would be expected to impair excitation:inhibition balance, leading to epilepsy. This is consistent with the high incidence of 15 seizures observed in monogenic AD (4,5,(21)(22)(23)(24), with five of the AD patients studied here manifesting seizures and/or myoclonus, and with the hypothesis that seizures may occur before widespread neurodegeneration and even before clinical symptoms of dementia (25).
For the analysis of gene expression changes and their relevance to AD pathogenesis, we focused 20 on those alterations shared between PSEN1 and APP AD. To interrogate the biological relevance of changes in gene expression in each cell type, we analyzed not only individual genes exhibiting the strongest differences in gene expression in each cell type between monogenic AD and non-5 demented controls, but also functional groups ( Fig. 2A and 3, A to C; Fig. S5 to S8; Data S2 and S3). Down-regulation of gene expression dominates in many cell types, including excitatory neurons, inhibitory neurons, oligodendrocytes and oligodendrocyte precursor cells (Fig. S5, A, C and E). Notably amongst functional categories, many genes encoding pre-and post-synaptic proteins involved in synaptic transmission were down-regulated in both excitatory and inhibitory 5 neurons ( Fig. 2A; Fig. S6B). Furthermore, both inhibitory and excitatory neurotransmitter receptors were downregulated in expression, as were genes required for GABA production in interneurons. Therefore, in addition to neuron loss, reduced expression of synapse and neurotransmission genes may contribute to declining neurological function and to the development of epilepsy in monogenic AD. 10 In addition to defects in neurotransmission, differential gene expression pointed to a marked switch in neuronal metabolism. A large number of genes encoding multiple elements of the mitochondrial electron transport chain, as well as a number of enzymes required for the Krebs cycle, were downregulated in monogenic AD neurons ( Fig. 2B and Fig. S6B). This was seen 15 across multiple neuronal types and indicates widespread mitochondrial dysfunction and defects in oxidative phosphorylation in neurons. Accompanying this was an upregulation of genes involved in glycolysis, many of which are known HIF-1 targets (Fig 2B, Fig S6B and Data S3).
Reduced expression of electron transport chain complexes and changes in activity of Krebs cycle enzymes have both previously been noted in AD neurons (26)(27)(28), and the upregulation of 20 glycolysis genes is consistent with functional imaging evidence for a relative increase in aerobic glycolysis in sporadic AD (29). The combination of both reduced oxidative phosphorylation and increased glycolysis is reminiscent of metabolic reprogramming observed in cancer (30) and 6 during immune cell activation (31), which is typically associated with increased metabolic demands. Since mitochondrial dysfunction and oxidative damage are thought to be early events in AD (32), it is possible that the shift to glycolysis is required to meet the neurons' metabolic needs and as a protective mechanism to generate antioxidants. It will be important to determine whether mitochondrial dysfunction or oxidative stress are the primary drivers of metabolic 5 reprogramming in monogenic AD, and more importantly whether this is a protective or pathological process.
To determine possible links between monogenic and sporadic AD, we examined the cellular expression of sporadic AD GWAS-associated genes (2, 3) in monogenic AD patients. Mapping 10 the set of 37 currently known sporadic AD GWAS-associated genes (2, 3), as well as APP and PSEN1/2, to our dataset, we found that almost all were expressed in at least one cell type in monogenic AD or non-demented controls (Fig. 2C), with many expressed in neurons. Of note, nine AD GWAS genes show significant changes in at least two neuronal subtypes in monogenic AD. This includes genes that are up-regulated (Fig. 2D), despite the overall reduction in gene 15 expression in the monogenic AD cortex (Fig. S5A). For instance, clusterin (CLU), the tau kinase PTK2B (33), the APP-processing regulator ABCA7 (34) and the regulator of intracellular trafficking BIN1 were all upregulated in neurons (Fig. 2D). Conversely, SORL1, an intracellular sorting receptor for APP (35), the endocytosis regulator PICALM, the transmembrane protein CNTNAP2 and the transcription factor MEF2C were down-regulated in neurons (Fig. 2D). The 20 net consequence of these changes is a mixture of protective and pathogenic effects. For example, increasing clusterin levels could potentially be a response to endolysosomal dysfunction (9, 36), and would be predicted to support increased flux through that system. In contrast, reducing 7 SORL1 levels would accelerate pathogenesis, as loss of function SORL1 mutations are themselves causal for monogenic AD (37).
As in neuronal cells, all glial cell types had altered gene expression in monogenic AD (Fig. 3 and   S8). Microglia and astrocytes exhibited signs of activation due to inflammatory or damaging 5 stimuli (Fig. 3, A to C; Fig. S8, A and B). Specifically, APOE, SPP1 and complement CQ1 were upregulated in microglia (38) and GFAP, CHI3L1 and GJA1 were upregulated in astrocytes (39)(40)(41). In addition, microglia exhibited hallmarks of innate immune cell activation, with upregulation of genes essential for antigen presentation, C1q components, and lysosome components (Fig. 3, A and C; Fig. S8A). Astrocytes also demonstrated several signatures of 10 cellular stress, such as increased expression of molecular chaperones and metallothioneins, and upregulation of lysosomal genes (Fig. 3, B and C; Fig. S8B).
The inflammatory response in monogenic AD was distinct from that in an individual with intracerebral hemorrhage (ICH). For astrocytes and microglia, cells from the ICH cortex formed 15 specific clusters distinct from both monogenic AD and non-demented controls (Fig. 3, D and G).
ICH microglia expressed genes associated with acute activation, including SPP1, FTH1 and S100A11 (42) (Fig. 3E). The activation of monogenic AD microglia was distinct from that of ICH ( Fig. 3E; Data S4). In particular, the monogenic AD microglial phenotype was more similar to the recently described human AD microglia phenotype (HAM; Fig. 3F) than to the murine 20 damage-associated microglia phenotype (DAM; Fig. 3F) (43). In contrast, ICH microglia were more similar to murine DAM than human AD microglia (Fig. 3F). A similar trend is observed in AD astrocytes, which have increased expression of a number of genes that have been identified 8 in reactive astrocytes in mouse models of AD (Fig. 3, G to I; Data S4) (39). In contrast, reactive astrocyte genes were expressed at a higher level in ICH astrocytes (Fig. 3I). Overall, we conclude that microglia and astrocytes in the monogenic AD brain have an activation phenotype that is disease-specific and distinct from acute activation due to brain hemorrhage. 5 To complement the analysis above that focused on intracellular signaling, we also studied how changes in gene expression affect intercellular signaling in monogenic AD. To do so, we analyzed co-expression of ligands and their cognate receptors across different cell types, comparing AD and non-demented controls. This analysis revealed an overall decrease in potential cell-cell signaling among neurons in monogenic AD ( Fig. 4A; Fig. S9, A to D), but an 10 increase in neuron-microglia and neuron-astrocyte signaling ( Fig. 4A; Fig. S9, A to D). Some changes in neuron-glia signaling are likely to be positively adaptive to the ongoing disease process and others deleterious (Fig. 4, B to D, Fig. S9, E to H). These include a positive adaptive change in neuronal scavenging of granulins, with both upregulation of GRN expression by microglia and increased neuronal expression of the SORT1 receptor (Fig. 4B). These changes 15 would increase neuronal accumulation of granulins, improving lysosomal function in PSEN1 and APP mutant neurons, which is compromised by these mutations (9). Conversely, microglial homeostasis via the chemokine CX3CL1 appears compromised by reduced expression in neurons of both the ADAM chemokine processing enzymes (44) and the CX3CR1 receptor (Fig.   4C). Similarly, neuron-oligodendrocyte/astrocyte signaling via neuregulins is also compromised 20 in AD, with down-regulation of neuronal neuregulin expression and reduced expression of the relevant receptor ERBB4 (45) in OPCs and oligodendrocytes and EGFR in astrocytes (46) (Fig.   4D). 9 Single-cell analysis of the frontal cortex in monogenic AD led to a number of insights into the molecular and cellular pathology of monogenic AD, conserved between patients carrying the PSEN1 and APP mutations. First, we observed widespread degeneration of almost all classes of inhibitory interneurons in monogenic AD, consistent with high incidence of epilepsy in the 5 patients. Second, we found that neuronal cells undergo metabolic reprogramming, similar to cancer cells. Third, we observed both adaptive and deleterious changes in neuronal and glial cells in monogenic AD patients. The presence of positive adaptive changes may explain the relatively slow development of clinical symptoms over decades (9,(47)(48)(49)(50)(51), and may also point towards resilience mechanisms that support survival of these neurons late in the disease process. In 10 addition to specific gene expression changes, the pervasive and global nature of the diseaseassociated changes across multiple cell types, and almost all cells within each class, underlines the global nature of the disease in its latter stages. As such, it is consistent with a disease process that begins in early adulthood (52), and supports the hypothesis that successful treatments for monogenic AD will likely need to be administered decades before the typical age at onset of 15 clinical symptoms (4,5).  10 signature is more similar to that of human Alzheimer's microglia (HAM) than murine damageassociated microglia (DAM). (G-I) Astrocytes similarly demonstrate an AD-specific activated phenotype, distinct from acute activation. Elements of previously described reactive astrocyte phenotypes (Pan-reactive and A1-/A2-specific) are found in AD astrocytes, but to a lesser degree than in acute activation due to brain hemorrhage (ICH). Heatmaps show log-transformed fold change between APP and PSEN1 and matched controls.
Differential expression in specific cell types of particular biological relevance is highlighted by 10 black boxes and the likely biological net effect summarized diagrammatically.