Transcriptome and proteome profiling reveals TREM2-dependent and -independent glial response and metabolic perturbation in an Alzheimer’s mouse model

Abstract

Elucidating the intricate molecular mechanisms of Alzheimer’s disease (AD) requires a multidimensional analysis incorporating various omics data. In this study, we employed transcriptome and proteome profiling of App^NL-G-F, a human APP knock-in model of amyloidosis, at the early and mid-stages of amyloid-beta (Aβ) pathology to delineate the impacts of Aβ deposition on brain cells. By contrasting App^NL-G-F mice with TREM2 (Triggering receptor expressed on myeloid cells 2) knockout models, our study further investigates the role of TREM2, a well-known AD risk gene, in influencing microglial responses to Aβ pathology. Our results highlight altered microglial states as a central feature of Aβ pathology, characterized by the significant upregulation of microglia-specific genes related to immune responses such as complement system and antigen presentation, and catabolic pathways such as phagosome formation and lysosome biogenesis. The absence of TREM2 markedly diminishes the induction of these genes, impairs Aβ clearance, and exacerbates dystrophic neurite formation. Importantly, TREM2 is required for the microglial engagement with Aβ plaques and the formation of compact Aβ plaque cores. Furthermore, this study reveals substantial disruptions in energy metabolism and protein synthesis, signaling a shift from anabolism to catabolism in response to Aβ deposition. This metabolic alteration, coupled with a decrease in synaptic protein abundance, occurs independently of TREM2, suggesting the direct effects of Aβ deposition on synaptic integrity and plasticity. In summary, our findings demonstrate altered microglial states and metabolic disruption following Aβ deposition, offering mechanistic insights into Aβ pathology and highlighting the potential of targeting these pathways in AD therapy.

AD is the most prevalent form of dementia globally and is characterized by its progressive neurodegenerative nature. The hallmark pathologies of AD are the presence of extracellular Aβ plaques and intracellular neurofibrillary tau tangles, which have been the focus of research for decades (1). In recent years, the role of glial cells in AD pathogenesis has gained increasing attention. Specifically, activated microglia and reactive astrocytes have been identified as key players in responding to the proteinopathies (Aβ and tau) of AD. Genome-wide association studies (GWAS) also highlighted many AD-risk genes are predominantly expressed in microglia. Among the AD risk genes enriched in microglia, TREM2 has drawn significant attention from researchers. In AD mouse models, TREM2 acts as a master regulator controlling microglial responses to Aβ plaques and the acquisition of unique disease-associated phenotypes (2). TREM2 R47H variant impairs the affinity of TREM2 for its ligands, resulting in a 3- to 4-fold increased risk of developing AD (3, 4). While GWAS data suggest the protective role of TREM2-mediated signaling in AD progression, the TREM2-dependent microglial response has also been shown to aggravate neuroinflammation and neurodegeneration (5). The contribution of TREM2-dependent microglial response to AD pathology remains an area of active investigation.

Beyond glial responses, the disruption of various biological processes, such as energy metabolism and oxidative stress, have been well-documented in AD pathology (6). All these processes are believed to contribute to the initiation and progression of AD. To fully understand these complex disease mechanisms, a holistic approach that incorporates systemic biological methods is required. The integration of multi-omics approaches, in particular, offers a comprehensive characterization of the AD pathological phenotype. By capturing molecular signatures and interactions across biological levels, these powerful phenotyping technologies have the potential to significantly accelerate our understanding of the pathophysiological mechanisms underlying AD.

In this study, we performed transcriptome and proteome profiling of single human APP knock-in App^NL-G-F mice at the onset and middle stages of Aβ pathology to comprehensively investigate the pathway alterations and molecular mechanisms associated with Aβ deposition. App^NL-G-F knock-in mice were selected as they eliminate artifacts related to APP overexpression, such as the mis-localization of APP and the accumulation of APP fragments including CTF-β (C-terminal fragment of APP) and AICD (APP intracellular domain) (7, 8). Consequently, this model is particularly suited for exploring downstream events, such as glial activation, in response to Aβ deposition. Furthermore, we also profiled App^NL-G-F, TREM2 knockout mice to examine how the absence of TREM2 affects the responses of microglia and other glial cells to Aβ deposition. Our results revealed extensive glial responses and metabolic disturbances following Aβ deposition and demonstrated the crucial role of TREM2 in modulating microglial responses and its potential interactions with other glial cells.

Result

Transcriptomic and proteomic profiling of App^NL-G-F mice brains

To gain systemic insights into the biological impacts of Aβ deposition in the brain, we conducted both transcriptomic and proteomic analyses on the brains of App^NL-G-F and App^NL-G-F; Trem2^KO mice at 3- and 9-month-old corresponding to the onset and middle stages of Aβ pathology, respectively (Fig. 1A)—β plaque deposition starts to appear at 3-month-old and becomes abundant at 9-month-old (Fig. S1). We performed transcriptomic and proteomic analyses using the hippocampus and cortex, respectively. Both brain regions showed prominent Aβ pathology associated with cognitive decline. For the transcriptome profiling, RNA libraries were prepared from the dissected hippocampus using the SMART-Seq mRNA LP kit. Differentially expressed genes (DEGs) were identified between different genotypes and age groups. For the quantitative proteome profiling, the cortices were lysed and digested with Trypsin, and the peptides from each sample were labeled with assigned Tandem Mass Tags (TMT) channels for LC-MS/MS Analysis. TNT quantification was performed with Proteome Discoverer 2.5 followed by the identification of differentially abundant proteins (DAPs). We first compared transcriptome and proteome between 3- and 9-month-old App^NL-G-F mice and identified 1321 DEGs and 748 DAPs. Among them, only 156 genes/proteins were co-differentially expressed (Fig. 1B and File S1), suggesting the influence of other biological factors on mRNA/protein stability and translation. Spearman correlation analysis across 156 shared molecules showed that the correlation coefficient between the DEPs and DAPs was 0.66 (Fig. 1C), revealing a moderate correlation between mRNA and protein levels similar to previous publications (9).

Figure 1.

Comparative transcriptomic and proteomic analysis of App^NL-G-Fand App^NL-G-F; Trem2^KOmouse brains.A, experimental design schematic for the preparation of brain samples and subsequent ‘omics’ analyses. B, Venn diagram displaying the overlap between differentially expressed genes (DEGs, FDR < 0.05) and differentially abundant proteins (DAPs, FDR < 0.05) in App^NL-G-F mice at 3 and 9 months of age (n = 3 per age group). C, Spearman correlation scatter plot for the 156 differentially-expressed genes shared by DEGs and DAPs, showing a moderate correlation (ρ = 0.66).

The altered microglial state is a dominant feature associated with Aβ deposition

To better understand the function of 156 shared DEGs/DAPs, we conducted a functional analysis using Over-representation Analysis (ORA) and Ingenuity Pathway Analysis (IPA). Our ORA, based on Gene Ontology (GO) annotations, revealed significant enrichment in GO terms, particularly within the categories of Biological Process (BP) and Cellular Component (CC), related to immune response (such as ‘immune effector process’, and ‘adaptive immune response’) and catabolic processes (including ‘lytic vacuole’, ‘lysosome’, and ‘phagocytic vesicle’, Fig. 2, A and B). A gene-concept network of the top enriched GO terms further revealed genes involved in the complement pathway (C1qa, C1qb, C1qc, C4b, and Cfh), regulators of immune response (Inpp5d, Ptprc, and Ptpn6), the cathepsin family of cysteine proteases (Ctsh, Ctsd, Ctss, and Ctsz), lysosomal proteins (Grn, Hexa, Hexb, LAMP1, and LAMP2), and lipid transfer and transport (Clu, ApoE, Prdx6, NPC2, and ABCA1), as shown in Figure 2C. Notably, many of these genes, including ApoE, C1q, Inpp5d, Grn, and ABCA1, are known to be associated with AD risks, underscoring the significance of these pathways in modulating AD pathology (10).

Figure 2.

Characterization of microglial response to Aβ deposition.A, overrepresentation analysis (ORA) of 156 genes identified as both differentially expressed (DEGs) and differentially abundant proteins (DAPs) according to Gene Ontology (GO) terms within the ‘Biological Process’ (BP) category. B, ORA for the same gene set within the ‘Cellular Component’ (CC) category. In both (A and B), the dot color represents the adjusted p-value significance, while the dot size indicates the number of genes encompassed. C, a gene-concept network illustrates the relationship between selected GO terms (adaptive immune response, immune effector process, lysosome, phagocytic vesicle) and corresponding genes. The color intensity denotes the fold-change of the genes, and the size of the grey nodes reflects the gene count for each GO term. D, Ingenuity Pathway Analysis (IPA) of canonical pathways for the shared DEGs and DAPs, with bar colors indicating predicted pathway activation (positive z-score) or inhibition (negative z-score), and grey bars signifying pathways with no activity prediction by IPA. Significance is determined using a right-tailed Fisher’s exact test, with a threshold p-value ≤0.05, denoted by a dashed line at −log10(p-value) = 1.3. E, Volcano plot presenting significant DEGs and F, DAPs in App^NL-G-F mouse brains at 3 and 9 months of age (n = 3 per age group), highlighting the microglia-enriched gene expression and protein abundance shifts associated with Aβ pathology. Upregulated genes are represented as brown dots, and downregulated genes as navy dots.

The IPA results were largely in agreement with the ORA findings. The top canonical pathways among the shared DEGs include immune responses (e.g., ‘neutrophil degranulation’, ‘complement system’, ‘interferon signaling’, ‘MHC II antigen presentation’), as well as lysosome and phagosome-related pathways (e.g., ‘phagosome maturation’ and ‘CLEAR signaling pathway’), as shown in Figure 2D. Given that microglia are the primary immune cells in the brain, responsible for surveillance and phagocytic clearance of debris, these findings suggest that altered microglial state is a dominant feature in Aβ pathology.

Indeed, the volcano plot showed the top-ranked DEGs in response to Aβ deposition are predominantly found in microglia (Fig. 2E and File S2). Many of these upregulated genes are the well-known markers of disease-associated microglia (DAM), such as components of the complement system (e.g., C1qa and C1qc), antigen presentation genes (e.g., H2-Aa and CD74), and lysosomal genes (e.g., Ctsd and Lyz2). Similarly, the top DAPs in response to Aβ deposition also include genes related to the complement system (e.g., C1qa and C1qc) and lysosomal function (e.g., Hexb, Ctsz, and Ctsd) (Fig. 2F). The observed induction of the immune response in 9-month-old App^NL-G-F mice appears to be mainly driven by microglia, as our immunostaining results showed minimal infiltration of T cells (CD3+), B cells (B220+), and granulocytes (Ly6G+) in 9-month-old App^NL-G-F mice (Fig. S2). Collectively, our results suggest an altered microglial state as a dominant feature associated with Aβ deposition, characterized by an upregulation of genes involved in the immune response and enhanced lysosomal capacity.

Microglial response to Aβ deposition depends on TREM2

TREM2 has been reported as a master regulator of microglial response (11, 12, 13). To evaluate the role of TREM2 in AD pathology, we compared the transcriptome of microglia isolated from 9-month-old App^NL-G-F and App^NL-G-F; Trem2^KO. Loss of TREM2 led to predominantly downregulated DEGs (245 Downs vs. 115 Ups) (Fig. 3A and File S3). Most top-downregulated DEGs are the microglial genes induced in Figure 2E, such as Cst7 (Cystatin F), Itgax (Integrin Alpha X), Clec7a (C-type lectin domain family 7-member A), Ccl6 (C-C motif chemokine ligand 6), and CD74 (Major histocompatibility complex, class II invariant chain). These results support the crucial role of TREM2 in mediating microglial response to Aβ deposition.

Figure 3.

Trem2-dependent regulation of genes expression and protein abundance in Aβ pathology.A, Volcano plot illustrating DEGs when comparing the brain of App^NL-G-F mice to that of App^NL-G-F; Trem2^KO mice at 9 months of age (n = 3 per age group). Upregulated genes (FDR < 0.05) are represented as brown dots, and downregulated genes as navy dots). B, IPA of canonical pathways for DEGs comparing App^NL-G-F mouse brains with App^NL-G-F; Trem2^KO counterparts at 9 months old. C, gene Set Enrichment Analysis (GSEA) of DEGs between 9-month-old App^NL-G-Fand App^NL-G-F; Trem2^KO mouse brains. Individual dots represent distinct GO terms, with the dot size reflecting the number of genes within each term. The GeneRatio, indicating the proportion of genes enriched in the identified pathway relative to the total number of genes in the data set, is depicted. GO terms with positive normalized enrichment scores (NES) suggest activation (left side), whereas terms with negative NES suggest suppression (right side). D, Volcano plot highlighting significant differentially abundant proteins (DAPs) comparing App^NL-G-F mouse brain to App^NL-G-F; Trem2^KO mouse brains at 9 months of age (n = 3 per age group).

Results of IPA further confirm that loss of TREM2 resulted in a significant reduction in immune responses, as evidenced by the decreased activity in the ‘Neutrophil degranulation’, ‘Neuroinflammation Signaling Pathway,’ ‘Complement System’, and ‘Toll-like Receptor signaling’, as well as reduced phagocytosis, indicated by ‘Phagosome formation’ (Fig. 3B). Gene set enrichment analysis (GSEA) using the Gene Ontology annotations echoed IPA results, pointing to a broad suppression of pathways involved in phagocytic vesicle formation, lysosomal function, and immune response (Fig. 3C). Interestingly, TREM2 deficiency led to increased expression of gene sets related to ‘extracellular matrix organization’ and ‘cartilage development.’ This observation aligns with recent studies showing that microglia are involved in the remodeling of the extracellular matrix and suggests loss of TREM2 may compromise this process in AD pathology (14).

At the protein level, the absence of TREM2 resulted in significant changes in the abundance of 37 proteins (Fig. 3D and File S5). The down-regulated DAPs are predominantly associated with immune responses, including components of the complement system such as C1qa, C1qb, and C1qc, along with Ptprc (Protein tyrosine phosphatase, receptor type C), all critical for immune activation and regulation. Lysosomal proteins, such as Hexa (Beta-hexosaminidase subunit alpha), Hexb (Beta-hexosaminidase subunit beta), Ctsz (Cathepsin Z), Ctsh (Cathepsin H), and Ctss (Cathepsin S) were also reduced. Interestingly, the TREM2 knockout led to increased levels of Tmem106b, a lysosomal protein predominantly expressed in neurons. Overall, the proteomic results align with the transcriptomic data, collectively supporting that TREM2 is central to the orchestration of microglial response in AD.

To validate our transcriptomic and proteomic findings, we conducted immunostaining to visualize the microglial response to Aβ deposition. Along with the increase of Aβ deposition from 3 to 9 months in the App^NL-G-F mice, there is a significant expansion and clustering of microglia around the Aβ plaques, evidenced by over a 100% increase in the area occupied by IBA1-positive (IBA1+) microglia and over 50% of the plaque area was found to be covered by Iba1+ microglia at 9 months old (Fig. 4, A and B). Remarkably, loss of TREM2 caused a significant decrease in both the Iba1+ area and the percentage of plaque area covered by Iba1+ microglia (Fig. 4, A and B), leading to a substantial reduction of plaque-associated microglia. This result aligns with recent studies showing that TREM2 deficiency impairs microglial viability and proliferation (15, 16, 17). Consistent with the key role of microglia in clearing Aβ plaques, we observed a significant increase in total Aβ plaque areas (82E1) in App^NL-G-F; Trem2^KO mice, a phenotype shown as early as 3 months old (Fig. 4, A and B).

Figure 4.

Knockout of TREM2 diminishes plaque-associated microglia.A, representative immunofluorescence images of Iba1 (microglia, green) and Aβ (82E1 antibody, red) in brain sections from App^NL-G-F and App^NL-G-F; Trem2^KO mice at both 3 and 9 months of age, with eight mice analyzed per age group. B, quantitative analysis of areas positive for Iba1 and Aβ, along with the percentage of Aβ area co-stained with Iba1. Statistical significance is denoted as ∗∗p < 0.01 and ∗∗∗∗p < 0.0001. The scale bar represents 200 μm.

Microglia are known to undergo significant morphological changes in response to injury and environmental cues (18). To investigate the role of TREM2 in modulating microglial morphology, we analyzed the morphological changes in microglia located proximal and distal to Aβ plaques using immunostaining for two microglial markers, IBA1 and TMEM119 (Fig. 5A). Our analysis showed microglia close to Aβ plaques (interacting microglia) exhibited a distinct morphology compared to those further away (distal microglia) (Fig. 5B). Quantitative Sholl analysis revealed that plaque-associated microglia had reduced morphological complexity and smaller convex hull volumes compared to microglia distant from the plaques (Fig. 5, C–F). However, these morphological changes were similar between App^NL-G-F and App^NL-G-F; Trem2^KO mice, indicating that TREM2 does not significantly influence the morphological dynamics of microglia in the context of Aβ plaques.

Figure 5.

Spatial and morphological analysis of microglia.A, representative single-slice immunofluorescence images of Iba1 (microglia, green), TMEM119 (microglia, magenta), and corresponding mask segmentation labels showing branches (yellow) and soma (magenta) in brain sections from 9-month-old App^NL-G-F and App^NL-G-F; Trem2^KO mice. B, depth-coded images representing 3D volumes of selected isolated microglia categorized according to proximity to Aβ plaques. Color represents depth. Scale bar, 20 μm. Average Sholl profiles of microglia located proximal or distal to Aβ plaques in 9-month-old App^NL-G-F (C) and App^NL-G-F; Trem2^KO (D) mice. Comparison of morphological complexity (E) and convex hull volume (F) of microglia across spatial locations and genotypes. Statistical significance is denoted as ∗∗p < 0.01.

Our transcriptome and proteome results suggested that TREM2 mediates the induction of immune response and lysosomal expansion in microglia. Consistently, we observed a significant increase in the expression of CD74, a well-established marker of DAM, among plaque-associated microglia in App^NL-G-F mice. However, this induction was absent in App^NL-G-F; Trem2^KO mice (Fig. 6, A and B). Similarly, lysosomal genes Ctsd (Cathepsin D) and CD68 are significantly induced in the App^NL-G-F mice. In contrast, in the App^NL-G-F; Trem2^KO mice, the induction of both Ctsd and CD68 was significantly reduced (Fig. 6, C and D). Notably, this reduced staining for Ctsd and CD68 may largely be attributed to the decreased number of plaque-associated microglia in the absence of TREM2. Overall, these immunostaining results corroborate our earlier transcriptomic and proteomic data, underscoring the pivotal role of TREM2 in mediating microglial responses to Aβ deposition.

Figure 6.

Loss of TREM2 reduces expression of genes related to MHC Class II antigen presentation and lysosomal function.A, representative immunofluorescence images of Iba1 (microglia, green) and CD74 (MHC class II invariant chain, red) in brain sections from App^NL-G-F and App^NL-G-F; Trem2^KO mice at 9 months of age (n = 8 per group). B, quantitative analysis of the stained areas positive for Iba1 and CD74 from the brain sections described in (A). C, immunofluorescence images showcasing Iba1 (microglia, green), Ctsd (cathepsin D, red), and CD68 (lysosomal marker, magenta) in brain sections from the same mice groups. D, quantitative analysis of areas positive for Ctsd and CD68 in the brain sections. Statistical significance indicated by ∗∗p < 0.01 and ∗∗∗∗p < 0.0001. Scale bar: 200 μm.

Microglia facilitate the formation of compact plaques and restrict neuronal damage

The observation of the diminished plaque-associated microglia in App^NL-G-F; TREM2^KO mice prompted us to further examine its effects on Aβ pathology. Microglia are known to influence the morphology and density of Aβ plaques. To assess plaque compactness, we employed Thioflavin S (Thio-S) staining to visualize the compact core of Aβ plaques, complemented by total Aβ plaque visualization using the 82E1 antibody. At the age of 3 months, Aβ plaques in both App^NL-G-F and App^NL-G-F; Trem2^KO mice were found to be Thio-S negative, indicating the initial formation of diffuse plaques. By 9 months, a significant portion of plaques in App^NL-G-F mice developed Thio-S positive cores (Fig. 7, A and B), a change that was associated with an increase in microglia clustering around the plaques (Fig. 4). Conversely, the loss of TREM2 resulted in a near absence of Thio-S positive plaques and a lower ratio of Thio-S positive to 82E1 positive areas (Fig. 7, A and B), accompanied by diminished microglial engagement around Aβ plaques (Fig. 4). These results suggest plaque-associated microglia are required in the development of compact Aβ plaque cores.

Figure 7.

Plaque-associated microglia facilitate the formation of compact Aβ plaques.A, immunofluorescence staining of Thioflavin S (Thio-S, indicating compact Aβ plaque cores, green) and total Aβ (identified by 82E1 antibody, red) in brain sections from App^NL-G-F and App^NL-G-F; Trem2^KO mice at both 3 and 9 months of age (n = 8 per group). B, quantification of the areas positive for Thio-S and Aβ, as well as the ratio of Thio-S-positive areas to Aβ-positive areas, was performed across both age groups and genotypes. Statistical significance is indicated by ∗∗p < 0.01 and ∗∗∗∗p < 0.0001. Scale bar: 200 μm.

Aβ plaques are known to induce microtubule disruption and lysosome accumulation in presynaptic dystrophic neurites. The immunostaining of Lysosome-associated membrane protein 1 (LAMP-1), a widely used marker for dystrophic neurites, revealed an age-associated increase in App^NL-G-F mice (Fig. 8A). The deletion of TREM2 further exacerbates Aβ-associated neuropathology, as evidenced by an increased LAMP+ area in both 3- and 9-month-old App^NL-G-F; Trem2^KO mice (Fig. 8A). Our proteomic analysis revealed that loss of TREM2 increased the abundance of another lysosomal protein TMEM106b enriched in neurons (Fig. 3D). Aligned with proteome data, immunostaining of TMEM106b is increased in 9-month-old App^NL-G-F mice and further exacerbated in App^NL-G-F; Trem2^KO mice (Fig. 8, C and D). Interestingly, the staining of TMEM106b localized at the vicinity of Aβ plaques (Fig. S3), mirroring the pattern of LAMP1 staining. To further validate that plaque-associated increase of TMEM106b staining, like LAMP1, indicates the accumulation of lysosomes and autophagic vacuoles in dystrophic neurites, we performed co-staining for TMEM106b, LAMP1, and the axonal marker Neurofilament. Confocal imaging revealed that TMEM106b largely colocalizes with LAMP1, with both proteins accumulating in the swollen, bulbous-shaped dystrophic neurites marked by Neurofilament staining (Fig. 8E, blue arrow). Quantification using Pearson’s correlation coefficient yielded an R-value of .84, indicating a strong positive correlation in their spatial distribution. Manders’ overlap coefficients showed that 81% of the TMEM106b signal overlaps with LAMP1 (M1 = 0.81), and 76% of LAMP1 overlaps with TMEM106b (M2 = 0.76). These results suggest that the increased TMEM106b staining observed around Aβ plaques may also serve as a marker for plaque-associated dystrophic neurites.

Figure 8.

Loss of TREM2 exacerbates Aβ-associated neuropathology.A, immunofluorescence staining for LAMP1 (lysosomal-associated membrane protein 1, a marker for dystrophic neurites) in brain sections from App^NL-G-F and App^NL-G-F; Trem2^KO mice at both 3 and 9 months of age (n = 8 per group). B, quantitative analysis of LAMP1-positive areas, comparing both genotypes at 3 and 9 months of age (n = 8 per group). C, immunofluorescence staining for TMEM106b (a lysosomal protein enriched in neurons) in brain sections from the same groups of mice. D, quantification of TMEM106b-positive areas across both genotypes and age groups (n = 8 per group). Statistical significance is denoted by ∗p < 0.05 and ∗∗p < 0.01. Scale bar represents 200 μm. E, confocal image of co-immunofluorescence staining for LAMP1, TMEM106b, and neurofilament in brain sections from 9-month-old App^NL-G-F mice. Co-localization analysis yielded a Pearson’s R-value of 0.84 between TMEM106b and LAMP1 channels. The scale bar represents 50 μm.

Aβ pathology reduces mitochondrial function and protein translation

In addition to altered microglial state, our pathway enrichment analysis of the transcriptome highlighted profound disruptions in energy metabolism and protein translation caused by Aβ plaques. Using IPA to compare DEGs between 3- and 9-month-old App^NL-G-F mice, the top enriched Canonical Pathways suggested a marked reduction in “ATP synthesis” and “Oxidative Phosphorylation,” as evidenced by a negative Z-score, and an increase in “Mitochondrial Dysfunction,” indicated by a positive Z-score (Fig. 9A, supplemental file 3). GSEA results further validated the reduced expression of genes associated with the “Respiratory chain complex,” “ATP synthase complex,” as well as ‘mitochondrial matrix’ (Fig. 9B). An examination of genes implicated in ‘Mitochondrial Dysfunction’ revealed not only a decrease in the expression of genes related to the mitochondrial respiratory chain, such as Ndufa8, Sdhb, Cox4i, and Atp5d, but also a reduction in genes responsible for antioxidant defense, including GPX4, Sod1, and GST. Conversely, there was an upregulation in genes associated with calcium mobilization (Capn3, Itpr2) and the production of reactive oxygen species (ROS) (Nos1, Pik3r5) (Fig. 9C). Collectively, these findings suggest that Aβ plaque accumulation suppresses ATP production in mitochondria, and weakens antioxidant defenses while concurrently increasing ROS production and promoting calcium mobilization.

Figure 9.

Aβ pathology suppresses mitochondrial function and ribosomal protein expression.A, IPA of canonical pathways based on DEGs revealed changes in mitochondrial function when comparing 3- and 9-month-old App^NL-G-F mouse brains. B, GSEA of DEGs in App^NL-G-F mouse brains between 3 and 9 months. Each dot corresponds to a specific GO term, with the GeneRatio reflecting the proportion of genes enriched within the pathway relative to the total genes in the dataset. The color gradation from yellow to blue represents the Benjamini-Hochberg (BH) adjusted p-value, and the dot size indicates the count of enriched genes. C, Heatmap illustrating the expression patterns of DEGs related to mitochondrial function in App^NL-G-F mouse brains at 3 and 9 months, grouped by enriched pathways. Expression values are normalized counts extracted from DESeq2, and z-scored with red indicating upregulation and blue indicating downregulation. D, IPA reveals similar suppression of mitochondrial function and ribosomal protein levels in App^NL-G-F; TREM2^KO mouse brains at 3 and 9 months, compared to App^NL-G-F mouse brains.

Concomitant with the decline in mitochondrial function, IPA analysis also highlighted a suppression of protein translation, including ‘Eukaryotic Translation initiation,’ ‘Eukaryotic Translation Elongation,’ ‘Eukaryotic Translation termination,’ and ‘EIF2 signaling,’ crucial signaling to regulate global protein synthesis (Fig. 9A). This finding was corroborated by GSEA, which confirmed the reduced expression of “ribosome” and “ribosomal subunit,” and “mitochondrial ribosome” (Figs. 9B and S4). As protein synthesis (PS) machinery consumes nearly half of the total energy produced in mammalian cells, the rate of PS is tightly coupled with cellular ATP levels (19, 20). Therefore, the downregulation of both mitochondrial and ribosomal genes observed in 9-month-old App^NL-G-F mice is likely causally linked and signifies a reduction in anabolic activities. Together with increased ‘lysosome’ and ‘phagocytic vesicles’ (Fig. 9B), our results suggest that brain cells switch from anabolism to catabolism in response to the accumulation of Aβ plaques.

Interestingly, the suppression of mitochondrial and ribosomal genes does not appear to be dependent on TREM2, as IPA canonical pathway analysis revealed similar trends of decreased ‘Oxidative Phosphorylation’ and increased ‘Mitochondrial Dysfunction’ when comparing DEGs between 3- and 9-month-old App^NL-G-F; Trem2^KO mice (Fig. 9D and File S4). This suggests that the observed mitochondrial and ribosomal alterations are intrinsic responses to Aβ plaque accumulation, rather than being affected by Trem2-dependent microglial response.

Aβ pathology reduces the abundance of excitatory synaptic proteins

IPA of DAPs between 3- and 9-month-old App^NL-G-F mice revealed a marked reduction in the ‘synaptogenesis signaling pathway’, as indicated by a negative Z-score (Fig. 10A and File S5). This alteration, unique to the proteome, contrasts with the transcriptome profiling outcomes, underscoring the significance of post-transcriptional and post-translational modifications in AD pathology. GSEA on these DAPs further validated a decrease of synaptic proteins in both “postsynaptic” and “presynaptic” compartments (Fig. 10B). Notably, the decline in protein levels was particularly pronounced in the “glutamatergic synapse” and “excitatory synapse.” This supports recent findings that excitatory synapses are disproportionately susceptible to degeneration in the context of AD (21, 22).

Figure 10.

Aβ pathology reduces the abundance of excitatory synaptic proteins. IPA of canonical pathways based on DAPs revealed changes in the abundance of synaptic proteins when comparing 3- and 9-month-old App^NL-G-F mouse brains (A), and App^NL-G-F; TREM2^KO mouse brains (B). Bar colors denote predicted pathway activation (positive z-score) or inhibition (negative z-score). p-values were derived from a right-tailed Fisher’s exact test, with a threshold for significant enrichment set at p-value ≤0.05. GSEA of DAPs between 3- and 9-month-old in App^NL-G-F mouse brains (C), and App^NL-G-F; TREM2^KO mouse brains (D). Each dot represents a unique GO term. The GeneRatio indicates the proportion of genes enriched within each pathway relative to the total gene count in the dataset, with the color gradient from yellow to blue displaying the Benjamini-Hochberg adjusted p-value, and dot size showing the count of genes enriched per pathway.

Concordant with transcriptomic observations, GSEA on DAPs also underscored an increased abundance in proteins linked to catabolic organelles, such as “lysosomes,” “lytic vacuoles,” and “phagocytic vesicles,” contrasted with a decline in anabolic machinery like ‘ribosomes’ (Fig. 10B). This transition from anabolic to catabolic processes likely reflects a metabolic reprogramming in neurons triggered by Aβ accumulation. Intriguingly, TREM2 does not affect the observed reduction in synaptic proteins and ribosomes. Comparative analysis of DAPs between 3- and 9-month-old App^NL-G-F; Trem2^KO mice similarly indicated a decrease in the ‘synaptogenesis signaling pathway’ and ribosomal abundance, suggesting a TREM2-independent mechanism (Fig. 10, C and D). Conversely, the abundance of the ‘complement system’ exhibits TREM2 dependency. The significance (p value) of its induction, noted in App^NL-G-F mice, is markedly reduced in the absence of TREM2, indicating a microglia-mediated response (Fig. 10, C and D).

In addition, GSEA results suggested that Aβ deposition correlates with an elevated presence of “myelin sheath” proteins in both genotypes. Western blot analysis of myelin proteolipid protein (PLP1), the major myelin protein in the brain, confirmed its increased abundance in both genotypes associated with Aβ deposition (Fig. S5). This finding aligns with earlier reports that Aβ deposition led to thicker myelin and enhanced oligodendrogenesis (23).

To validate that Aβ deposition leads to the loss of glutamatergic synapses, we performed co-immunostaining of PSD95 (a postsynaptic density marker), VGLUT1 (a presynaptic marker of glutamatergic synapses), and Aβ in brain sections from App^NL-G-F and App^NL-G-F; Trem2^KO mice at 9 months of age (Fig. 11A). Quantitative analysis revealed a marked reduction in the density of PSD95 puncta and glutamatergic synapses around Aβ plaque in 9-month-old App^NL-G-F. A similar decrease in the density of PSD95 and glutamatergic synapses was observed in age-matched App^NL-G-F; Trem2^KO mice (Fig. 11B). Traditionally, the reduction in synaptic proteins in AD has been attributed to aberrant microglial-mediated synaptic pruning (24). However, our findings suggest that this reduction is microglia-independent. Instead, we propose that the decrease in synaptic proteins may result from reduced protein synthesis and ATP production, reflecting a shift in neuronal metabolism from anabolism to catabolism due to Aβ deposition. Further investigation is needed to establish a direct causal relationship between these factors.

Figure 11.

Aβ deposition caused a similar loss of glutamatergic synapses in App^NL-G-Fand App^NL-G-F; Trem2^KOmice.A, confocal images of the co-staining of PSD95 (postsynaptic density marker), VGLUT1 (presynaptic markers of glutamatergic synapses), and Aβ deposition in brain sections from App^NL-G-F and App^NL-G-F; Trem2^KO mice at 9 months of age (n = 8–9 per group). B, quantitative analysis of PSD95 (left) and synapses (right) density inside and outside of Aβ plaques in both genotypes. Statistical significance is denoted by ∗∗∗p < 0.001. The scale bar represents 20 μm.

Discussion

In this study, we conducted comprehensive transcriptomic and proteomic analyses of App^NL-G-F and App^NL-G-F; Trem2^KO mice to elucidate the pathological alterations induced by Aβ deposition and to assess the contribution of TREM2 in modulating the microglial response to Aβ pathology. Our results reveal that altered microglial states represent a predominant feature of Aβ pathology. Top-ranked DEGs associated with Aβ deposition are predominantly associated with immune functions such as the complement system and antigen presentation, as well as catabolic activities such as phagosome formation and lysosome biogenesis. This underscores the pivotal role of microglia in immune surveillance and the phagocytic elimination of Aβ. Our findings also underscore the essential role of TREM2 in facilitating microglial response. The absence of TREM2 substantially reduces the expression of genes related to - immune response, phagosome formation, and lysosome biogenesis. In the pathologic analysis, loss of TREM2 led to a notable decrease in plaque-associated microglia, which compromises Aβ clearance and the formation of compact Aβ plaque cores, thus aggravating the progression of dystrophic neurites. Moreover, we observed that staining for TMEM106b, akin to LAMP1 staining, marks the accumulation of lysosomes and autophagic vacuoles within dystrophic neurites associated with Aβ plaques. Furthermore, our study revealed marked disruptions in energy metabolism and protein synthesis caused by Aβ deposition. Both the transcriptomic and proteomic analyses indicate a significant shift in brain cells from anabolic to catabolic processes in response to Aβ pathology. This metabolic shift is closely associated with a reduction in the abundance of synaptic proteins, suggesting a direct link between metabolic reprogramming and the decreased abundance of synaptic proteins. This relationship highlights the profound impact of Aβ deposition on synaptic integrity and plasticity, offering insights into the mechanisms that underlie cognitive impairment in AD.

Most of the current research in the AD field has focused on transcriptome profiling. While transcriptome analysis is a valuable tool, it may not capture the full complexity of AD pathology. Our study provides compelling evidence that Aβ pathology exerts a significant impact on brain cells at both transcriptional and translational levels. This conclusion is supported by several observations: (1) the existence of only a partial overlap and moderate correlation between transcriptomic and proteomic data; (2) a noticeable decline in the expression of ribosomal and mitochondrial genes, which provide machinery and energy for protein synthesis; and (3) our data showing that reductions in ribosomal components correlate with a decrease of synaptic proteins. Previous studies have established that a global suppression in protein synthesis marks an early milestone in AD pathology, likely contributing to synaptic dysfunction (25, 26, 27). Mechanistically, phosphorylation of eukaryotic initiation factor 2a (EIF2α) is known to mediate the integrated stress response (ISR) and suppress protein synthesis in AD pathology (28, 29). Targeting the ISR or EIF2α-dependent defects in brain protein synthesis has been shown to enhance synaptic plasticity and ameliorate cognitive deficits in AD models (30, 31). Building on previous research, our results suggest that Aβ deposition triggers a metabolic shift from anabolism to catabolism in the brain. This shift, characterized by reduced energy production and diminished protein synthesis, plays a key role in contributing to synaptic dysfunction. Our study underscores the importance of integrating both transcriptomic and proteomic analyses for a holistic understanding of AD pathology.

Our study elucidates the indispensable role of TREM2 in orchestrating the microglial response to Aβ plaques. We demonstrate that TREM2 is required for the clustering of microglia around Aβ plaques, while its absence leads to a marked reduction in plaque-associated microglia. Our results also suggested that TREM2 may affect microglial survival near Aβ plaques. Consistently, previous research showed plaque-associated microglia adopt a unique transcriptomic profile, termed the disease-associated microglia (DAM) signature, and the microglial transition from a homeostatic to a DAM state requires TREM2 (2, 32). The loss of TREM2 impairs various microglial functions, including metabolism, survival, clustering, immune response, and phagocytosis (12, 33, 34, 35). Furthermore, our study highlights the essential role of microglia in the formation of compact cores within Aβ plaques, aligning with recent studies that have shown microglial phagocytosis to be instrumental in the formation of dense-core plaques (36). The lack of such dense-core plaques in App^NL-G-F; Trem2^KO mice is associated with increased dystrophic neurites, suggesting the protective role of TREM2-mediated microglial response in countering Aβ-induced neuropathology. Collectively, our findings resonate with GWAS data, supporting that the loss of TREM2 function exacerbates AD pathology. These results call for further research into TREM2-dependent signaling, which may offer a promising avenue for developing therapeutic strategies targeting AD.

Our study, while providing valuable insights into the mechanisms of Aβ pathology, is subject to several limitations. First, the transcriptomic and proteomic analyses were conducted on different brain regions: the hippocampus and the cortex, respectively. Despite observing similar patterns of Aβ deposition and glial responses in both regions, our comparative approach between transcriptome and proteome did not account for potential regional variations. Secondly, our analysis focused on AD mouse models at 3 and 9 months to represent the onset and mid-stages of Aβ pathology. This temporal framing introduces the possibility that some of the observed changes could stem from natural aging processes rather than the AD pathology itself. However, considering that mature adult mice are typically defined as being 2 to 8 months old (37), we anticipate that aging per se may not significantly contribute to the differences observed in our study. Thirdly, our pathway enrichment analysis, which infers the direction of regulatory changes based on gene expression levels or protein abundance, overlooks the impact of post-translational modifications. Such modifications can critically influence protein function and interaction networks. To fully capture the complexity of the disease mechanism, future research employing multi-layer proteomic profiling of AD models is warranted and may provide novel insights into the intricate mechanisms underlying Aβ pathology.

Experimental procedures

Animals

App^NL-GF mice were a gift from Dr Saido’s lab in RIKEN (7). TREM2 knockout mice were purchased from The Jackson Laboratory (Strain #: 027197). App^NL-GF mice were crossed to TREM2 knockout mice to generate App^NL-GF; Trem2^KO homozygous for both genetic manipulations. Mice were housed less than five per cage with ad libitum access to food and water in a pathogen-free barrier facility. Mice of both sexes were used for all experiments. All the animal work was performed following NIH guidelines and protocols approved by The Ohio State University Institutional Animal Care and Use Committee (IACUC).

Bulk RNA seq and data analysis

Hippocampus was dissected from brain tissues, submerged in RNAlater solution (Thermo Fisher), and stored in −80 °C freezer till further processing. Total RNA was extracted using Quick-RNA miniprep (R1055, Zymo Research). RNA quality was evaluated by TapeStation using high Sensitivity RNA ScreenTape (5067-5579, Agilent). RNA samples with RNA integrity numbers greater than eight were used for cDNA library construction. RNA seq libraries were prepared using SMART Seq mRNA LP Kit (Takara Bio) following the manufacturer’s instructions. The qualities of the cDNA library were assessed using TapeStation using High Sensitivity D5000 ScreenTape (5067-5592, Agilent). cDNA library samples were then pooled and sequenced with the HiSeq 4000 System (Illumina) by AZENTA life sciences.

Demultiplexed FASTQ files of bulk RNA sequencing data were aligned to the mouse genome (Mus_musculus.GRCm39) using STAR (version 2.7.10a) (38). Adapters were trimmed using Flexbar (version 3.5.0.). Reads mapped to genomic features were counted using featureCounts (version 2.0.3) (39). The count matrix was imported in R (version 4.2.0) for analysis. Gene expression signatures of different experimental conditions were compared using DeSeq2 (version 1.36.0), using the Wald test for hypothesis testing (40). Genes with adjusted p values (using a Bonferroni correction) <0.05 and log2 fold changes >0.25 were considered significantly differentially expressed.

Correlation analysis

Spearman correlation coefficients between DEGs (mRNA) and DAPs (protein) were calculated using only molecules passing thresholds (FDR < 0.05). Values for which no protein measurement value exists were not considered in this as well as the following analysis, i.e., no imputed values were taken into account.

Gene network and functional enrichment analysis

Canonical Pathway analysis was performed by QIAGEN’s Ingenuity (IPA, QIAGEN Redwood City, www.qiagen.com/ingenuity). Complete lists of DEGs and DAPs, along with their log2 fold change expression values and FDR were inputted into IPA for identifying canonical pathways, biological functions, and upstream regulators using a cutoff of FDR <0.05. The p-value of overlap, calculated using the right-tailed Fischer’s Exact Test with a statistical threshold of 0.05, is used to indicate the probability of association of molecules from test dataset with the canonical pathway by random chance alone. A positive or negative regulation z-score value indicates that a function is predicted to be activated or inhibited. No activity prediction by IPA results in ineligible z-score which is represented by grey bars.

Over-representation Analysis (ORA) with Gene Ontology (GO) annotation, Gene Set Enrichment Analysis (GSEA), and gene-concept network were performed by using the R package clusterProfiler (version 4.4.4) (41, 42, 43). For ORA, the functional categories of Biological Process (BP) and Cellular Component (CC) were included to identify significant GO terms associated with the differentially expressed genes/proteins shared between the transcriptomic and proteomic data.

As GSEA by clusterProfiler requires the input of a list of gene sets/pathways to check and a uniquely-named vector of ranking metric values, values of log2 Fold Change and −log10(p-value) × sign(log2 Fold Change) were used to form the ranking metrics for transcriptomic and proteomic datasets respectively, to score and rank the genes in descending order. As GSEA is more sensitive and has the advantage of detecting phenotypic differences manifested by small but consistent changes across a set of genes, we applied GSEA to the complete list of DEGs and DAPs, aiming to capture as many changes as possible, including subtle yet coordinated ones that might be missed by ORA.

Quantitative proteomic analysis

Brain samples were lysed using 5% SDS and 50 mM TEAB buffer, sonicated, cleaned, and quantified via BCA assay. Samples were then prepared using the STrap Midi MS sample prep device (Protifi), with each sample containing 2000 μg of protein. This preparation involved reduction with dithiothreitol (DTT), alkylation with iodoacetamide (IAM), quenching of the IAM reaction with DTT, and overnight digestion with Trypsin/Lys-C within the STrap device. Peptides were eluted from the STrap, with 60 μg per sample dried via SpeedVac and reconstituted in 50% acetonitrile for tandem mass tag (TMT) labeling in 50 mM TEAB (pH 8.5). After TMT labeling, samples were pooled, acidified with 1% formic acid, and analyzed for label check on a nano LC-MS/MS system. Following successful label checks, samples were dried, reconstituted in 2% formic acid, desalted using EVOLUTE EXPRESS ABN (Biotage), and fractionated via offline HPLC into 12 fractions. These fractions were analyzed by LC-MS/MS after being reconstituted with mobile phase A; approximately 5% of each was injected using the TMT method. TMT quantification and identity discovery were performed using Proteome Discoverer 2.5 (Thermo Fisher). False discovery rate (FDR) was calculated based on The Benjamini-Hochberg Procedure. FDR ≤0.05 was considered to be significant.

Protein analysis

Proteins were sequentially extracted from brain tissues with RIPA in the presence of protease inhibitors. For Western blots, equal amounts of proteins (20–40 mg) were separated on NuPAGE bis-tris gels (Invitrogen), and membranes were probed with primary and secondary antibodies. Signals were visualized by chemiluminescence ECL Plus (GE Healthcare). Blots were quantified by densitometry with ImageJ software.

Histological analysis

Brains were sectioned on a cryostat at 40-mm thickness. For immunofluorescence staining, free-floating sections were blocked with PBS containing 10% normal goat serum (NGS) at room temperature for 30 min, incubated with primary antibody in blocking solution (PBS with 1% NGS) at 4 °C for 24 to 48 h, and then incubated with secondary antibody at room temperature for 2 h. Sections were mounted on slides with ProLong Diamond (Life Technologies). Images were captured on a ZEISS Axio Observer and/or the Nikon AXR point scanning confocal microscope. Image quantification was performed using ImageJ software. Auto Threshold methods “Otsu” or “Triangle” were used to define the region of interest (ROI). Statistical analyses were conducted using a two-tailed unpaired t test or one-way ANOVA.

Primary antibodies used in this study are: Anti-Human Amyloid β (N) (82E1) from IBL Co., LTD, Anti-Iba1 from Wako Co, Anti CD74, Anti-Neurofilament Marker (pan-neuronal, cocktail) Antibody (SMI-311R) from Biolegend, Anti-Ctsd and Anti-TMEM106b from Cell Signaling Technology, Anti-CD68 (ab53444) and anti-LAMP1 (ab25245) from Abcam. Anti-TMEM119, Anti-PSD95 sdAb, and Anti-VGLUT1 sdAb from Synaptic Systems. All secondary antibodies were purchased from Thermo Fisher or Jackson Immunoresearch. For thioflavin S staining, sections were stained with 0.05% thioflavin-S in 50% ethanol for 8 min and differentiated in two changes of 80% ethanol for 10 s each. Sections were then washed in large volumes of distilled water three times and incubated in a high concentration of 3× PBS buffer for 30 min. Sections were then briefly rinsed with distilled water and coverslipped for imaging. The Coloc2 plugin in ImageJ is used to quantify the degree of co-localization between two fluorescently labeled channels in confocal images. Statistical measures, including Pearson’s correlation coefficient and Manders’ overlap coefficients, were used to assess the spatial correlation and overlap between the intensity distributions of the two channels.

Microglia morphology analysis

To quantify microglia morphology and their spatial relationships to plaques, image volumes of ∼800 × 440 × 30 μm were acquired using resonant scanning with 16× averaging on a Nikon AXR confocal equipped with a 20 × 0.75 NA Plan Apo objective. Images had a lateral pixel size of 0.215 μm and a z-step size of 0.8 μm and were subsequently denoised using NIS Elements (Nikon) denoise.ai. For automated plaque and microglia segmentation, U-Net models using an adaptive and self-configuring nnU-Net architecture were trained using 20 image volumes and corresponding annotations (44). For plaques, the plaque regions were annotated by hand using signals from Iba1 and Aβ channels to define plaque boundaries. For microglia, a 3D Ilastik pixel classifier was trained to classify microglia soma and branch voxels using both Iba1 and TMEM119 for enhanced detection of complete microglia morphology (45). The subsequent predictions were thresholded into binary masks for soma and branches and manually corrected to improve accuracy.

Training of the nnU-Net models was performed on an RTX 6000 Ada 48 GB NVIDIA GPU, with the 3D full-resolution model being trained for 1000 epochs with an initial learning rate of 0.01 and parameters previously described (44). Following the training of the model, the output accuracy compared to ground truth labels in validation volumes was evaluated. The model achieved a pseudo-dice score of 0.997 for plaques, 0.950 for microglia branches, and 0.989 for microglia soma, with one indicating perfect overlap. These models were used to segment plaques microglia branches, and soma in the denoised confocal volumes.

Following plaque segmentation, a 3D distance map was created to measure distances between microglia and plaques. For each image volume, we isolated microglia that were most distant from plaques, as well as those that were interacting with plaques. These cells were also inspected to ensure they were intact and completely within the image volume. Isolated microglia volumes were processed to remove branches from neighboring cells and traced using SNT (46). Traces were checked for accuracy before 3D Sholl analysis (47), and convex hull volume measurements being made. Morphological complexity was calculated from the area under the Sholl profile (48).

Synapse analysis

To quantify PSD-95 puncta and putative synapses with VGLUT1 image volumes of ∼98 × 98 × 15 μm were acquired using resonant scanning with 16× averaging on a Nikon AXR confocal equipped with a 60 × 1.4 NA Plan Apo objective. Images had a lateral pixel size of 0.048 μm and a z-step size of 0.3 μm and were subsequently denoised using NIS Elements (Nikon) denoise.ai.

For automated segmentation of plaques, a nnU-Net model was trained as previously described for microglia analysis using 34 image volumes paired with annotations. This model achieved a pseudo-dice score of 0.98 for plaques, with one indicating perfect overlap. A 3D Ilastik pixel classifier was trained to classify VGLUT1 positive voxels and PSD-95 puncta were automatically segmented by first preprocessing image volumes using a rolling ball background subtraction with a radius of five pixels followed by an unsharp mask filter with a radius of one pixel and a mask weight of 0.8 in Fiji followed by seeded watershed detection in Python (49). PSD-95 alone or putative synapses, identified as PSD-95 objects overlapping with VGLUT1 positive classified voxels, were quantified based on their presence inside and distance from plaques using scikit-image (50) and SciPy (51).

Data availability

The differentially expressed genes (DEGs) and differentially abundant proteins (DAPs) derived from transcriptome and proteome analyses are provided in Files S1–S5. The raw data supporting the findings of this study are available upon request. Please contact the principal investigator Jie Gao (Jie.Gao@osumc.edu) for data sharing arrangements.

Supporting information

This article contains supporting information.

Ethics approval and consent to participate

All experiments were performed in compliance with institutional guidelines (The Ohio State University).

Conflict of interest

The authors declare that they have no conflicts of interest with the contents of this article.

Reviewed by members of the JBC Editorial Board. Edited by Elizabeth J. Coulson