Impact of 5xFAD and Trem2 R47H mutations across major brain cell types

In this study we investigate regional, plaque proximal, and genotype specific gene expression changes induced by the Trem2R47H mutation. As this mutation is not sufficient to induce amyloid plaque pathology in mice, we utilize a hemizygous 5xFAD/homozygous Trem2R47H mouse model which induces Aβ pathology in concert with the Trem2R47H mutation, compared with matched 5xFAD (Aβ pathology only), Trem2R47H, and WT controls (Fig. 1A). By comparing these four genotypes, we uncover the transcriptomic alterations induced specifically by 5xFAD transgenes (independent of Trem2R47H mutation), specifically by Trem2R47H (independent of 5xFAD), and those induced by a combination of 5xFAD and Trem2R47H mutations.

Fig. 1: MERFISH spatial transcriptomics enables spatial variation analysis of the transcriptome at the cell type level.

A Dataset overview consisting of 15 samples, from WT, Trem2R47H, 5xFAD and Trem2R47H; 5xFAD mice. Cell counts in batch 1 are aggregated across two technical replicates, resulting in approximately double the cell counts of the other batches. B Integration of cell by gene matrix with RNA spatial location enables spatial analysis of transcriptomic variation on a regional and genotype basis. C 300 gene overlay on a single coronal section, at increasing resolutions. D UMAP displaying 37 annotated cell types after integration across all samples. E UMAP of cell genotypes. Note the distinct subpopulations specific to 5xFAD and Trem2R47H; 5xFAD genotypes, particularly in microglia and astrocyte cell populations. F Hierarchical organization of cell clusters, combined with raw cell type proportions per genotype.

We performed spatial transcriptomic analysis using MERFISH on 19 coronal half sections from 15 total animals with WT, 5xFAD, Trem2R47H and Trem2R47H; 5xFAD genotypes at 12 months of age, which age was chosen to match the later timepoint in our previous studies, which we have extensively characterized [22, 27]. Mice at this age exhibit an extensive plaque burden throughout the brain, combined with neuritic damage and glial responses.

After quality control, this dataset resulted in 432,794 cells. Using a 300 gene panel, we identified 37 major cell types, and transcriptomically and spatially mapped 5xFAD and Trem2R47H transcriptomic alterations at the single-cell level. We also identified Aβ plaque locations in the same samples and assessed their relationship to spatial gene expression (Fig. 1B).

After spatial transcript decoding (Fig. 1C), cells were processed using our single-cell pipeline (Supplementary Fig. 1A, B), and clusters were identified based on reference to known cell type markers (Supplementary Fig. 1C), in conjunction with spatial location (Fig. 1D). Color coding genotype information on the UMAP shows strong 5xFAD induced cell type composition changes, reaching significance only in microglia (p = 0.0012, Wilcoxon rank sum test) (Fig. 1E). Hierarchical clustering identified initial splits between non-neuronal and neuronal cells, followed by excitatory vs. inhibitory, and spatial (subcortical, hippocampal, cortical) based splits in excitatory cell types (Fig. 1F).

Visualization of neuronal cell types (Fig. 2A, Supplementary Fig. 2) show strong spatial localization, commensurate with previous region-based studies and atlases. Hippocampal excitatory cells define the primary structures of the hippocampal formation (DG, CA1, CA3), while cortical excitatory neurons divide into distinct layers across most of the cortex. We identified and visualized cell type-specific markers for these distinct neuron types (Fig. 2B) to verify spatial fidelity with raw decoded transcripts.

Fig. 2: Spatial and transcriptomic analysis of coronal brain slices enables analysis of the spatial distribution of individual genes.

A Spatial position of neuron subpopulations from single coronal sample. B Raw transcript overlay on PolyT cell body staining, of cell type markers for a subset of the neuron subpopulations in A. C Annotated spatial regions for a single coronal sample. Annotation performed based on transcriptomic cellular locations, combined with the Allen mouse brain reference atlas. D Raw transcript overlays of Tmem119 (homeostatic microglia), Itgax (disease associated microglia), and Trem2 on top of DAPI nuclei staining (blue). Point brightness indicates pixels where multiple individual transcripts were aggregated. E Violin plots of normalized expression of the genes indicated in D, divided by genotype and aggregated across samples. Asterisks indicate statistical significance (p < 0.05, linear mixed effects model). F Regional mean normalized expression in microglia aggregated within genotypes.

Next, we segmented major brain regions, subdividing the cortex into three subregions: the neocortex (somatosensory, visual, parietal, retrosplenial, and auditory cortices), the limbic cortex (perirhinal, ectorhinal, entorhinal, and piriform cortices), and the cortical amygdala, and identify major structures in hippocampal and subcortical regions. This resulted in 10 identified major brain regions (Fig. 2C).

We visualized raw transcript counts of Tmem119 and Itgax to confirm microglia activation in the 5xFAD and Trem2R47H; 5xFAD mice. Tmem119 is a homeostatic microglia marker, while Itgax is a marker for disease associated microglia (DAM), a distinctive microglia subset whose activation is associated with neuroinflammatory responses, including response to Aβ plaque pathology. As expected, 5xFAD and Trem2R47H; 5xFAD mice show Itgax expression upregulation, indicating increased microglial activation (Fig. 2D–E, Itgax: p < 0.02, Tmem119: p < 10−10, linear mixed effects model). Microglia transition to a fully activated state via a two-stage Trem2 dependent pathway, highlighting the importance of this gene in AD progression [48]. We note that Trem2 expression is significantly increased in the microglia of both 5xFAD and Trem2R47H; 5xFAD mice (Fig. 2E, 5xFAD: adjusted p = 2.6 × 10−3, fold change = 1.88, Trem2R47H; 5xFAD: adjusted p = 7.1 × 10−6, fold change = 1.89. Linear mixed effects model). We also note that Itgax and Trem2 expression is consistently upregulated in 5xFAD and Trem2R47H; 5xFAD mice across most regions (Fig. 2F).

Finally, we analyzed neuronal density for each cell type by computing the total number of detected neurons divided by the estimated volume of the associated regions. No statistically significant trend was identified, however this may be due to our relatively small number of sample sections per genotype and the section variance across samples.

Overall, MERFISH spatial transcriptomics enables detection of high-level cell type clusters, visually identifiable and quantifiable transcriptomic differences in microglia and regional annotation and assignment of individual cells to specific coarse-grained spatial regions.

Glial and neuronal transcriptomes are affected by nearby plaques

Spatial transcriptomics can reveal local effects of pathology, such as Aβ plaques, on the regulation of gene expression in nearby cells. By co-staining coronal brain slices with both DAPI and thioflavin S (a canonical stain for Aβ plaques), we observed that DAPI brightly labels Aβ plaques in addition to nuclei [49] (Fig. 3A). We therefore applied DAPI staining to MERFISH prepared coronal slices and a machine learning approach to automatically detect and segment plaques in each of the MERFISH samples.

Fig. 3: Machine learning enables accurate identification of plaque locations across brain slices.

A DAPI (nuclei staining), ThioS (plaque staining) and overlay indicate that DAPI stains both nuclei and plaques. B Manual annotation of Aβ plaques (differentiated from cells by size, brightness, and morphology) is used as the basis for a machine learning model to detect plaque locations. C Detected plaques (yellow) in a single 5xFAD sample. Zooming in (right panel) we see that the machine learning model identifies the plaque, but not the cells surrounding it (manually circled, green). D Plaques exhibit significantly lower transcript density than cells (p < 0.0001, t-test), and significantly higher volume (p < 0.0001, t-test).

DAPI stained plaques are visually distinguishable from nuclei by their large size, greater brightness, and fibrous morphology and lack of circular cell soma shape (Fig. 3B, C). These features enable manual annotation of plaques in individual fields of view. We trained a modified cellpose model [34] to detect large plaques (mean diameter 22.4 µm), but not cells (mean diameter 8.5 µm) (Fig. 3B, Supplementary Fig. 3A, B).

We analyzed each 5xFAD and Trem2R47H; 5xFAD section using this model (Supplementary Fig. 3C, D) and verified that (1) the model does not detect cells (Fig. 3B), (2) the predicted plaques are morphologically distinct from cells (Fig. 3C), and (3) the predicted plaques have significantly lower transcript density when compared to cells (Fig. 3D, t-test, p < 0.0001), as well as greater volume (t-test, p < .0001). Across all 5xFAD and Trem2R47H; 5xFAD samples, we identified a total of 5616 plaques (per sample: 5xFAD- 659.2 ± 160.1, Trem2R47H; 5xFAD- 464 ± 146.9, mean ± s.e.).

Across brain regions, we found the closest cell to each identified plaque was most frequently microglial (62.6% of plaques in 5xFAD and 60.0% in Trem2R47H; 5xFAD) (Fig. 4A). Additionally, microglia density in the region within 100 µm of a plaque (proximal) was significantly higher than in the 100–500 µm region(distal) (5xFAD, proximal density = 17.6 ± .935 × 10−5, distal density = 7.81 ± 0.665 × 10−5, p = 7.55 × 10−4; Trem2R47H; 5xFAD, proximal density = 17.6 ± 0.790 × 10−5, distal density = 6.87 ± 2.15 × 10−5, 1.31 × 10−3, mean ± s.e, plaques/µm2), while no genotype difference was detected for density either proximal or distal to plaques (p > 0.19). Astrocytes were the second most common cell type identified near plaques (7.9% of plaques in 5xFAD and 10.1% in Trem2R47H; 5xFAD) (Fig. 4A), however, overall astrocyte density showed no differences in density between proximal or distal areas in either genotype (p > 0.16). Thus, the typical microenvironment around plaques includes microglia, with astrocytes and other cell types at greater distances from the plaque (Fig. 4B) [24, 50]. We then analyzed cell type proportions in annuli around individual plaques measured at 25 µm intervals. The smallest 25 µm circle around each plaque center was populated almost exclusively by microglia, with other cell types becoming more prevalent with increasing distance to plaque (Fig. 4C).

Fig. 4: Aβ plaque proximity causes transcriptomic dysregulation in both glia and neuronal cell types.

A Example plaque (arrow) with associated annotated cell types. Note the congregation of microglia around the plaque. B Proportion of cell types identified as closest to plaques. For each plaque, the closest cell was identified, and the proportion of resulting cell types was computed. C Cell type proportions within annuli at specific distances from plaque centers. The top row indicates raw cell proportions, while the bottom row shows cell type proportions after normalization by the total number of cells in that group. D Aβ plaque density by region. Statistical comparison (p < 0.01, linear mixed effects model) identifies three regions with significantly lower plaque density in Trem2R47H; 5xFAD animals. E Differential expression results for microglia and astrocytes using distance to plaque as the continuous dependent variable. Cells were selected such that all tested cells were within 100 µm of the center of a plaque. Cells aggregated across 5xFAD and Trem2R47H; 5xFAD samples. Genes filtered by expression in the associated cell type as identified in previous studies. No other cell types exhibited more than one differentially expressed gene in this test. Results indicate expected expression changes per µm distance increase from closest plaque center. F Differential expression results testing cells within 100 µm of the center of a plaque against those 100–500 µm from the center of a plaque. The cell types with the largest number of DE genes among glia and neurons are visualized here. Genes filtered by expression in the associated cell type as identified in previous studies. Red points indicate genes exceeding both adjusted p-value and log fold change thresholds, green points only exceed log fold change thresholds, blue points only exceed adjusted p threshold, and gray points exceed no threshold.

We assessed whether plaques appear proximal to neurons. The distance from a plaque to the closest neuron was significantly larger than the distance from a neuron to its closest neuronal neighbor (5xFAD: minimal plaque to neuron distance 56.4 ± 10.7 µm, minimal neuron to neuron distance 21.6 ± 1.24 µm, p = 0.012; Trem2R47H; 5xFAD: minimal plaque to neuron distance 48.7 ± 2.53 µm, minimal neuron to neuron distance 21.0 ± 0.733 µm, p = 1.47 × 10−4, mean ± s.e., plaques in corpus callosum excluded from analysis due to lack of nearby neurons, t-test). We examined the typical distance of each neuronal cell type to the nearest plaque. This analysis showed that subiculum, layer 5, and layer 6 excitatory neurons have the lowest median distance to plaques among identified cell types (Supplementary Fig. 3E). However, none of the top 5 neuron types (ranked by median distance to plaque, excluding subiculum excitatory and SST-Chodl cells due to low cell numbers), exhibited significant density variation between plaque proximal (<100 µm) and distal (100–500 µm) regions. This implies that neuronal plaque proximity is driven primarily by plaque density in the associated regions. Additionally, plaques on average form in regions nearly twice as far from the nearest neuron as the typical distance between neurons, but the average neuronal density does not appear to be decreased in plaque proximal vs. plaque distal regions, implying a variation in plaque to neuron distance at the microscale (<100 µm), but not at larger scales (<500 µm).

The highest plaque density occurred in the corpus callosum (CC) (5xFAD average 5.61 × 10−5, Trem2R47H; 5xFAD average 6.23 × 10−5 plaques/µm2) and hippocampal areas (5xFAD average 2.89 × 10−5, Trem2R47H; 5xFAD average 2.04 × 10−5 plaques/µm2, averaged across CA1, CA3, and DG), followed by cortex (5xFAD average 3.92 × 10−5, Trem2R47H; 5xFAD average 2.08 × 10−5 plaques/µm2, averaged across neocortex, limbic cortex, and cortical amygdala), with the lowest densities in the subcortical regions (5xFAD average 2.71 × 10−5, Trem2R47H; 5xFAD average 0.323 × 10−5 plaques/µm2, averaged across midbrain, thalamus and hypothalamus) (Fig. 4D). Mice with the 5xFAD genotype had higher plaque density compared with Trem2R47H; 5xFAD mice in the midbrain, thalamus, and neocortex (p < 0.05, linear mixed effects model), but not the CC. This distribution is consistent with the pattern of median minimum distance to plaques (Supplementary Fig. 3E), with roughly all cell types showing larger distance to plaques in Trem2R47H; 5xFAD samples. High plaque density regions such as the subiculum and lower cortical layers contained neurons with the lowest median distance to the nearest plaque. Trem2R47H; 5xFAD animals exhibited larger plaque sizes than 5xFAD animals (1108 µm3 vs. 984 µm3, p = 0.025, Wilcoxon rank sum test), though this appears to be gender and pathology dependent, as male animals showed the reverse effect (741.96 µm3 vs. 799.60 µm3, p = 0.0046) as well as lower pathology levels (Supplementary Fig. 3C, D).

We next tested whether cells within 100 µm of the nearest Aβ plaque have altered patterns of gene expression (Fig. 4E). Due to the relatively low cell abundance proximal to plaques, we aggregated 5xFAD and Trem2R47H; 5xFAD samples, and separated individual cells by cluster. We analyzed plaque proximity based differential expression with two techniques. First, we identified cells within 100 µm of a plaque center. Using DESeq2 and treating cells as independent samples, we identified genes whose expression correlated with proximity to the nearest plaque (Fig. 4E, Supplementary Table 3). Continuous effects were identified primarily in microglia and astrocytes, with microglia showing an upregulation of typical DAM associated genes (e.g. Csf1, Apoe, Cst7), and a downregulation of P2ry12, a homeostatic microglia associated gene [51]. Similarly, C4b, Clu, and Gfap, markers of a previously known disease associated astrocyte (DAA) phenotype were also upregulated near plaques [52].

To validate these findings and to account for variability across biological replicates, we additionally performed a pseudobulk analysis of differential expression between plaque-proximal (within 100 µm of the closest plaque) and plaque-distal (100–500 µm to closest plaque). We applied a linear mixed effects model to pseudobulk expression for each cell type in each sample, accounting for batch as a random effect (Fig. 4F, Supplementary Table 4). Additionally, we filtered genes based on their known expression in each cell type from previous single-cell atlases [37, 38], to avoid spurious identification of differentially expressed genes due to technical (errors in segmentation) or biological (phagocytosis, overlapping cellular processes) effects [53].

Pseudobulk analysis was generally consistent with the DESeq2 results and identified both glial and neuronal changes (Fig. 4F). Microglia and astrocytes exhibited typical disease associated profiles in cells located proximal (<100 µm) to plaque centers. However, Nnat expression in astrocytes and oligodendrocytes, and Mmp14 expression in astrocytes decreased near plaques. This result contrasts with previous findings in humans and other mouse models showing upregulation of Mmp14 in reactive astrocytes in AD [54].

The pseudobulk analysis also revealed notable changes in gene expression affecting neurons proximal to plaques (Fig. 4F, bottom row). L6b neurons showed lower Ngf expression near plaques, a gene therapy target in AD [55]. Nr2f2, upregulated near plaques, is known to be dysregulated by AD associated single nucleotide polymorphisms in the APOE enhancer [56]. Cnr1 and Htr1a, also upregulated near plaques, are linked to regulation of the serotonergic system, which is known to affect memory in the context of AD [57]. L2 intratelencephalic (IT) neurons near plaques showed downregulation of Dkk3 (a WNT signaling modulator whose presence reduces Aβ pathology in mouse models [58]) and of the potassium ion channel subunit Kcnd2 [59] near plaques. L5 NP cells show Grm1 upregulation and Chrna7 downregulation near plaques. Parvalbumin-expressing inhibitory cells shows Grin2a, Zbtb20, and Plagl1 downregulation near plaques. Excitatory neurons in the cortical amygdala exhibited downregulation of Ntf3 (associated with nervous system maintenance [60]), Nptx1 (associated with synapse remodeling, but typically upregulated in previous studies of cortical neurons near plaques [61]), and Camk2g (implicated in synaptic plasticity [62]). Because there were few plaques in subcortical regions, we did not test plaque-associated differential expression for neuronal cell types in this region.

Microglia and astrocytes exhibit distinct cell type-specific spatial patterns of activation associated with 5xFAD mutation

We next directly analyzed spatial and transcriptomic variation of glia between genotypes. We made four pairwise comparisons (5xFAD vs. WT, Trem2R47H; 5xFAD vs. Trem2R47H, Trem2R47H vs. WT, Trem2R47H; 5xFAD vs. 5xFAD), to identify 5xFAD and Trem2R47H dependent variations, which we then compared with differential expression between Trem2R47H; 5xFAD and WT (Supplementary Table 5).

We identified 19 differentially expressed genes in microglia and 8 in astrocytes across all four pairwise comparisons. By contrast, we found 1–2 differentially expressed genes in oligodendrocyte (OGC) and oligodendrocyte precursor cells (OPC) cell populations (Fig. 5A), and none in the other non-neuronal cell types. Microglia and astrocytes primarily exhibited 5xFAD dependent changes (similar differential expression results for both 5xFAD vs. WT, and Trem2R47H; 5xFAD vs. Trem2R47H), replicating the DAM/DAA gene upregulation and homeostatic gene downregulation identified in the plaque proximity analysis. These were widely replicated in differential comparison of Trem2R47H; 5xFAD and WT. Interestingly, neither Itgax nor Cd74 were identified as differentially expressed in plaque proximity analysis of microglia, whereas they were upregulated 9.58 and 15.7-fold in 5xFAD compared with WT animals, and 19.7 and 26.0-fold in Trem2R47H; 5xFAD compared with Trem2R47H animals. The Trem2 gene itself showed a small reduction in expression dependent on the Trem2R47H mutation, which we hypothesize may be due to reduced binding efficiency of gene probes overlapping the mutated region, as the effect was not seen in our previous study [22]. Differential expression also shows a small but significant Trem2R47H specific upregulation in homeostatic microglia genes, including Tmem119 (fold change = 1.13, adjusted p = 0.019, Trem2R47H; 5xFAD vs. 5xFAD), and P2ry12 (fold change = 1.27, adjusted p = 4.60 × 10−4, Trem2R47H; 5xFAD vs. 5xFAD). The consistent variation in both Trem2R47H vs. WT and Trem2R47H; 5xFAD vs. 5xFAD comparisons indicates this may be a plaque independent effect and corroborates the overall lower plaque burden in Trem2R47H; 5xFAD samples.

Fig. 5: Microglia and astrocytes exhibit 5xFAD induced transcriptome alterations.

A Pairwise differential expression between genotypes among glia populations. Five pairwise comparisons are indicated (5xFAD vs. WT, Trem2R47H; 5xFAD vs. Trem2R47H, Trem2R47H vs. WT, Trem2R47H; 5xFAD vs. 5xFAD, Trem2R47H; 5xFAD vs. WT). Heatmaps display log fold change for each comparison, with genes not exceeding significance set to 0. Heatmaps are thresholded to the range (−1, 1). Genes not exhibiting significant expression in the associated cell type according to the Allen or mousebrain references were removed. Cell types exhibiting no differentially expressed genes not shown. B (1) Subclustering results for microglia. Clusters with transcriptomes influenced by spatial colocalization with other cell types removed. (2) Diffusion pseudotime results, indicating a non-bifurcating differentiation trajectory. (3) Genotype proportions for each subcluster. Clusters C1-4 are found primarily in 5xFAD and Trem2R47H; 5xFAD mice, with C6-7 localized to WT and Trem2R47H animals. Asterisked clusters pass threshold for overabundance. Results indicate the pseudotime trajectory (1) describes a genotype specific transition. (4) Proportions of DAM and homeostatic microglia within annotated regions. Significant variations in distribution include decreased DAM proportions in cortex, DG, CA3 and hypothalamus. (5) Upregulated genes in each subcluster divide into homeostatic and DAM associated genes. C Astrocyte subclustering analysis. (1) Subclustering results, unbiased (by genotype proportion) clusters combined and relabeled as C1. (2) Pseudotime trajectories indicate no clear differentiation pattern. (3) C2-3 exhibit 5xFAD and Trem2R47H; 5xFAD upregulation, with C4-5 upregulated in WT and Trem2R47H mice. (4) Significant spatial variation indicates C5 and C4 are differentiated by spatial location (subcortical vs. cortex/hippocampus), while C2 also appears upregulated in Cortex and Hippocampus, and C3 is distributed evenly across regions. (5) C2-3 exhibit upregulation of C4b and Gfap, part of the disease associated astrocyte (DAA) phenotype, while the spatially variable C4 and C5 differentiate by Cspg5 and Camk2g expression. D (1) spatial distribution of DAM and homeostatic microglia overlaid on a 5xFAD sample. (2) DAM proportion of total microglia in each region. Error bars indicate standard errors. Asterisks indicate regional differences between 5xFAD and Trem2R47H; 5xFAD mice (p < 0.05, linear mixed effects model). Highest concentrations of DAM in CC, midbrain, and thalamus. (3) DAM proportions of microglia divided by cortical layer. No statistically significant change detected between 5xFAD and Trem2R47H; 5xFAD mice, but statistically significant increases in DAM proportion in lower cortical layers. E (1) spatial distribution of DAA and homeostatic astrocytes overlaid on a 5xFAD sample. (2) DAA proportion of total microglia in each region. Error bars indicate standard errors. No statistically significant variations identified between genotypes. Highest concentrations of DAA in CC and surrounding regions. (3) DAA proportions of microglia divided by cortical layer. No statistically significant change detected between 5xFAD and Trem2R47H; 5xFAD mice, but statistically significant increases in DAA proportion in lower cortical layers.

To explore the effects of AD risk genes in specific glial subtypes, we subclustered the microglia and astrocyte subpopulations. We identified several small clusters of microglia that appear to express neuronal or other glial markers, and we confirmed that these cells are located near cells expressing these markers. We removed these cells from this portion of the analysis. After removal, subclustering identifies 7 microglia clusters (Fig. 5B1). Pseudotime analysis identified a single linear trajectory across all microglial cell types (Fig. 5B2). We next examined the genotype proportions of these clusters. After normalizing by the number of cells per sample, we averaged across samples of the same genotype, and computed cluster proportions. This identifies a clear 5xFAD dependent bias, with two clusters (labeled homeostatic) exhibiting > 80% proportion coming from non 5xFAD (i.e. WT and Trem2R47H) samples. The remaining five clusters corresponded to disease associated microglia (DAM) enriched in 5xFAD and Trem2R47H; 5xFAD mice (Fig. 5B3).

We aggregated homeostatic and DAM subgroupings and identified regional spatial biases (Fig. 5B4). DAMs were enriched in hippocampal area CA1. They were also enriched in thalamus, and midbrain, despite the relative lack of plaque density in these regions compared to the CA1 and CC (Fig. 4D). Finally, we identified markers for the individual microglia subpopulations, and plot normalized expression (Fig. 5B5).

We focused on the analysis of the genes differentially associated with late-stage DAMs as several genes exclusive to late-stage DAM (DAM II) were included (Itgax, Cst7, Csf1, Ccl6), as well as genes present across both stages (Apoe). Except for Ccl6, all of these genes are differentially expressed in 5xFAD and Trem2R47H; 5xFAD, with primary expression of DAM2 genes in C3-5 (later pseudotime). Apoe is evenly distributed across C2-5, reflecting its overexpression across the DAM developmental timeline (Fig. 5B5) [63]. While we do not have explicit genes that encode DAM I compared with DAM II in the gene panel, clusters C1-2 likely contain DAM I microglia, based on the pseudodevelopmental timeline.

Subclustering the astrocyte subpopulations, we aggregated clusters not exhibiting genotype specific bias (see methods for thresholds) into a single cluster (C1), retaining the genotype biased clusters (Fig. 5C1). Pseudotime trajectory analysis (Fig. 5C2) did not yield a distinctive pattern, however after analysis of genotype bias (identifying C1 as unbiased, C2/C3 as DAA, and C4/C5 as upregulated in WT/Trem2R47H samples, Fig. 5C3), we note that C5 and C4 exhibited distinct spatial distributions, with C4 appearing exclusively in cortex and hippocampus, and C5 appearing in subcortical regions (Fig. 5C4). The DAA exhibited a similar regional specificity, with C2 primarily restricted to cortex and hippocampus. Cluster markers are identified and plotted (Fig. 5C5).

We next examined the spatial distribution of DAM and DAA cells by region. Disease associated microglia were enriched in the CC, subiculum and subcortical regions (Fig. 5D1). Computing the proportion of microglia identified as DAM by region (Fig. 5D2) showed similar proportions of DAMs between 5xFAD and Trem2R47H; 5xFAD samples by region, except in the DG, thalamus, midbrain, and hypothalamus. This corresponds with the plaque density bias in 5xFAD samples (Fig. 4D). In the cortex, we saw a significant increase in DAMs in the lower cortical layers (L5/L6) compared with the upper cortical layers (L2/L3) (p < 0.0001, linear mixed effects model, Fig. 5D3).

Disease associated astrocytes were concentrated in the CC and surrounding areas (Fig. 5E1-2). Virtually no disease associated astrocytes were present in upper cortical layers, but this population was significantly upregulated in deeper cortical layers (Fig. 5E3). We did not find significant genotype specific effects in other glial cells.

To further analyze the spatial variation of gene expression, we performed direct pseudobulk differential expression analysis of microglia, astrocytes, oligodendrocytes and oligo-precursors across the 10 identified major brain regions (Supplementary Fig. 4, Supplementary Table 6). We compared each region with the average across the remaining 9 regions. We also computed regional cell density and cell proportion for each of these cell types.

Analysis of microglia (Supplementary Fig. 4A) showed that ~60% of spatially variable genes were also differentially expressed across genotypes. For example, the canonical late-stage DAM markers Cst7 and Itgax were significantly upregulated in the corpus callosum. A small number of other genes (Ctss, C1qa, Zbtb20, Ly9, Tmem119) had spatially variable patterns of expression that were consistent across WT and Trem2R47H mice and dysregulated in 5xFAD and Trem2R47H; 5xFAD mice. Microglia cell populations also showed drastic increases in both cell proportion and density across all brain regions.

Astrocytes (Supplementary Fig. 4B) exhibited large numbers of spatially variable genes with consistent patterns of expression across all genotypes (e.g. Erbb4, Nnat, Grin3a, Mmp14, Id4, Pax6, etc). We also found spatial variation in several disease associated genes (Aqp4, Gfap). These spatial variations were primarily observed between cortical and subcortical (thalamus, midbrain, hypothalamus) regions. However, astrocytes exhibited little genotype specific cell proportion or density variations between regions.

Oligodendrocytes exhibited two separate gene groupings (Supplementary Fig. 4C). One set (Snca, Dlg4, Nnat, Robo1, S100b, Ptgds) showed spatially variable expression across multiple regions, particularly between cortical/hippocampal and subcortical regions. The other set of genes (Adam10, Psen1, Olig1, etc) is primarily upregulated in CC and downregulated in amygdala, with very little variation in other regions. This pattern is not 5xFAD or Trem2R47H dependent and was observed even in WT oligodendrocytes cells. This second pattern is not replicated in oligodendrocyte precursor cells, though a significant cortex vs. subcortical divide is present in OPCs (Supplementary Fig. 4D). Neither cell type exhibits significant genotype dependent cell proportion changes within regions.

Overall, our data show that spatial variation in microglia and astrocyte gene expression is more affected by 5xFAD than by Trem2R47H. Both disease-associated microglia and astrocytes exhibit specific spatial distributions. DAMs were distributed across the coronal section, but concentrated in the CC and subcortical regions, and DAA were biased almost exclusively to the CC and surrounding regions. Regional transcriptional variations were primarily impacted in 5xFAD for microglia and astrocytes, and both 5xFAD and Trem2R47H mutations were independent of regional variations in oligodendrocytes and oligodendrocyte precursors.

Neurons exhibit complex transcriptomic impacts of 5xFAD and Trem2R47H mutations

We performed differential expression analysis for each of the four comparisons (5xFAD vs. WT, Trem2R47H; 5xFAD vs. Trem2R47H, Trem2R47H vs. WT, Trem2R47H; 5xFAD vs. 5xFAD), and compared with Trem2R47H; 5xFAD vs. WT differentially expressed genes, followed by subclustering analysis for each of the neuronal cell types (Supplementary Table 5). Analysis of cortical neurons identifies differentially expressed genes for all these comparisons in each cell type (Fig. 6A–C), as well as genotype biased subclusters for most neuron cell types (Fig. 6D, E, Supplementary Fig. 5). We first considered genes consistently identified as differentially expressed across multiple cortical neuronal cell types.

Fig. 6: Cortical neurons exhibit consistent Trem2 associated transcriptomic variations and spatially localized genotype biased subclusters.

A Pseudobulk, linear mixed effects model differential expression results for cortical IT neurons. Heatmaps indicate log fold changes. Fold changes for genes not exhibiting significance set to 0 (white). B Differential expression results for other cortical excitatory cell types. C Log fold expression changes for each comparison for genes identified as consistently differentially expressed across multiple cell types. D Subclustering of L2 IT neurons (UMAP, top left) identifies a single subcluster overrepresented in WT and Trem2R47H samples (top right, asterisked). This cluster is homogeneously distributed along layer 2 with bias for the neocortex (bottom left), and exhibits overrepresentation of Grp, Nos1, Nptx1, and Ptk2b. E Subclustering of L5 NP neurons (UMAP, top