Loss of ATG7 in microglia impairs UPR, triggers ferroptosis, and weakens amyloid pathology control

Microglia are professional phagocytes of the central nervous system that contribute to development, homeostasis, and responses to injuries. Microglia clear apoptotic cells, myelin debris, and protein aggregates (Borst et al., 2021; Paolicelli et al., 2022); sculpt neuronal synapses (Stevens et al., 2007; Schafer et al., 2012); and secrete cytokines, chemokines, and neurotrophic factors (Cserép et al., 2021; Ueno et al., 2013). Like many cell types, microglia rely on autophagy to maintain homeostasis. Autophagy is required for removing misfolded or aggregated proteins, eliminating intracellular pathogens, and clearing damaged organelles, such as mitochondria, ER, and peroxisomes (Pohl and Dikic, 2019; Rocchi et al., 2017). Additionally, autophagy provides a critical source of energy for microglia in response to nutrient starvation and/or defective mammalian target of rapamycin signaling (Ulland et al., 2017). Microglia and other phagocytes also utilize autophagy-related (ATG) proteins to conjugate the microtubule-associated protein 1A/1B light chain 3 (LC3) with the membranes of phagosomes and endosomes. This process, known as “noncanonical” autophagy, facilitates fusion of phagosomes and endosomes with lysosomes, promoting the degradation of engulfed material (Heckmann et al., 2019). A microglial deficit of noncanonical autophagy due to deletion of Atg5 or Rubicon attenuated the clearance of brain aggregates of amyloid-β (Aβ) and augmented behavior defects in a mouse model of Alzheimer’s disease (AD) (Heckmann et al., 2019). Myeloid cell deletion of Atg7, a crucial mediator of canonical and noncanonical autophagy, impaired microglia-mediated pruning of immature synapses, resulting in increased dendritic spine densities, altered neuronal connectivity, and autism-like behaviors (Kim et al., 2017). Microglia-specific deletion of Atg7 also led to increased accumulation of phagocytosed myelin and a more severe clinical outcome in the experimental autoimmune encephalomyelitis model of multiple sclerosis (Berglund et al., 2020). A recent study showed that a deletion of Atg7 induced microglia senescence in a model of amyloid pathology (Choi et al., 2023).

In this study, we discovered that the absence of Atg7-dependent autophagy in microglia within a mouse model of Aβ plaque accumulation reduced microglia density both near and distant from the plaques. This impairment hindered the microglia’s ability to contain Aβ-induced neurotoxicity. Single-cell RNA sequencing (scRNA-seq) of microglia demonstrated that deletion of Atg7 was associated with reduced expression of unfolded protein response (UPR) pathway genes, which was paralleled by increased expression of oxidative stress response genes, with the appearance of a population of microglia-expressing ferritin light and heavy chain genes. In vitro experiments corroborated reduced UPR in Atg7-deficient microglia exposed to the ER stressor tunicamycin or Aβ peptides. Conversely, Atg7-deficient microglia exhibited increased oxidative stress and lipid peroxidation, leading to ferroptosis. Although Atg7 deficiency also reduced UPR in aged non-5xFAD mice, microglia numbers were maintained even at a late age. We conclude that an Atg7 defect in microglia disrupts adaptive chaperone-encoding UPR pathways and increases oxidative stress. Combined with long-term Aβ proteotoxicity, this results in ferroptosis, although microglia still retain enough capacity to manage the physiological buildup of misfolded proteins that occurs with aging.

We crossed Atg7^fl/fl mice with Cx3cr1^CreERT2 mice, enabling inducible deletion of Atg7 in CX3CR1⁺ cells by administration of tamoxifen (TAM). Although CX3CR1 is broadly expressed in myeloid cells, TAM-induced Cre recombinase permanently deletes Atg7 in yolk sac–derived self-renewing microglia, while Atg7-deleted peripheral myeloid cells are replaced over time by Atg7-sufficient myeloid cells continuously generated in the bone marrow (Yona et al., 2013). Thus, we refer to these mice as Atg7^ΔMG. To see whether Atg7 deficiency affected microglia responses in AD, we crossed Atg7^ΔMG mice to 5xFAD transgenic mice, a model that accumulates Aβ plaques characteristic of AD pathology (Oakley et al., 2006). 1-mo-old Atg7^ΔMG-5xFAD mice and Atg7^fl/fl-5xFAD littermates were fed with TAM-containing chow until 2 mo of age, returned to regular chow, and aged to 5 or 12 mo to allow accumulation of Aβ plaques. A substantial deletion of Atg7 was validated by immunoblotting of microglia isolated from Atg7^ΔMG-5xFAD and Atg7^fl/fl-5xFAD mice (Fig. 1 A). Furthermore, microglia from Atg7^ΔMG-5xFAD displayed a marked reduction in LC3 expression, reflecting attenuated generation of autophagic vesicles (Fig. 1 A). Transmission electron microscopy (TEM) imaging showed that Atg7^ΔMG-5xFAD microglia had fewer multivesicular structures representative of autophagosomes (Fig. 1, B and C), corroborating that conditional deletion of Atg7 leads to an autophagic defect in microglia.

To compare microglial responses to amyloid plaques in Atg7^ΔMG-5xFAD and Atg7^fl/fl-5xFAD mice, we stained brain sections with antibodies for the microglia markers IBA1 and PU.1, along with Methoxy-X04, which detects fibrillar Aβ (Fig. 1 D). Co-staining of PU.1 and IBA1 revealed significantly fewer cortical microglia nuclei (quantified by PU.1⁺IBA1⁺ spots) in Atg7^ΔMG-5xFAD mice than in Atg7^fl/fl-5xFAD mice at 5 and 12 mo of age (Fig. 1, D and E). Analysis of anti-PU.1 and Methoxy-X04 showed that Atg7^ΔMG-5xFAD mice had fewer plaque-associated microglia than Atg7^fl/fl-5xFAD controls within the 15-μm radius from the border of Aβ plaques (Fig. 1, F and G). Microglia density outside of a 15-μm radius from Aβ plaques was similarly reduced in Atg7^ΔMG-5xFAD mice at both 5 and 12 mo of age (Fig. 1, H and I). Co-staining for IBA1 and the activation marker CD11c revealed fewer cortical IBA1⁺CD11c⁺ microglia in Atg7^ΔMG-5xFAD than in Atg7^fl/fl-5xFAD mice (Fig. 1, J and K). We conclude that deletion of Atg7 in microglia of 5xFAD mice results in a reduction of microglia density, which prevents proper accumulation and activation of microglia around amyloid plaques.

Microglia have been shown to control brain Aβ load as well as the spreading of Aβ plaques and their neurotoxicity (Hansen et al., 2018; Condello et al., 2015). Thus, we asked whether the decrease of microglia density around Aβ plaques in Atg7^ΔMG-5xFAD mice increased Aβ pathology. Assessment of Aβ load by staining brain sections with Methoxy-X04 showed similar levels of Aβ accumulation in Atg7^ΔMG-5xFAD and Atg7^fl/fl-5xFAD mice at 5 and 12 mo of age (Fig. S1, A and B). This result was confirmed by ELISA of insoluble Aβ42 on brain lysates (Fig. S1 C). However, analysis of Aβ plaque morphology revealed differences between Atg7^ΔMG-5xFAD and Atg7^fl/fl-5xFAD mice. We stained brain sections with Methoxy-X04 (which detects fibrillar Aβ) and anti-6E10 antibody (which detects both fibrillar and oligomeric Aβ) and divided amyloid plaques into three categories: inert plaques (Methoxy-X04⁺, 6E10⁻), which are more compact and less neurotoxic; filamentous plaques (Methoxy-X04⁻, 6E10⁺), which are more diffuse and neurotoxic; and mixed plaques, which show a dense core surrounded by a filamentous ring (Methoxy-X04⁺, 6E10⁺) (Fig. 2 A) (Condello et al., 2015). At 5 mo of age, Atg7^ΔMG-5xFAD mice had significantly fewer inert plaques than Atg7^fl/fl-5xFAD mice (Fig. 2 B); at 12 mo of age, Atg7^ΔMG-5xFAD mice had significantly fewer inert plaques and more mixed plaques (Fig. 2 B). Thus, Atg7-deficient microglia were unable to efficiently constrain plaque spreading. We further stained brain sections for LAMP1, which labels vesicles accumulating in damaged axons and dendrites, collectively called dystrophic neurites. While LAMP1 volumes surrounding the plaques were similar in Atg7^ΔMG-5xFAD and Atg7^fl/fl-5xFAD mice at 5 mo of age, larger LAMP1 volumes were evident in Atg7^ΔMG-5xFAD mice at 12 mo of age (Fig. 2, C and D), indicating that the increase of diffuse plaques in Atg7^ΔMG-5xFAD mice exacerbates neurite dystrophy in the long term. Co-staining of brain sections with LAMP1 and IBA1 corroborated that LAMP1 was primarily located outside of microglia (Fig. S1 D). We conclude that diminution of microglia density around plaques in Atg7^ΔMG-5xFAD mice is associated with increased Aβ plaque toxicity and neurite damage.

We next investigated whether Atg7 deficiency impacted microglia ability to phagocytose fibrillar Aβ. We injected Methoxy-X04 dye into the peritoneum of Atg7^ΔMG-5xFAD and Atg7^fl/fl-5xFAD mice and measured its uptake by microglia 3 h after injection via flow cytometry (Fig. 2 E). A smaller proportion of Methoxy-X04⁺ microglia was present in Atg7^ΔMG-5xFAD mice than in Atg7^fl/fl-5xFAD mice (Fig. 2, F and G), suggesting that Atg7 deficiency is associated with curtailed phagocytosis of fibrillar Aβ. Immunofluorescence staining for the phagosome marker CD68 showed that Atg7^ΔMG-5xFAD mice had fewer phagocytic microglia compared with Atg7^fl/fl-5xFAD mice (Fig. 2 H). This finding was supported by the quantification of the percentage of CD68⁺ area (Fig. 2 I), which aligned with the observed overall reduction in microglia density. In contrast, Atg7 deficiency did not affect the density of CD68⁺ puncta per IBA1⁺ microglia (Fig. 2 J). These data indicated that Atg7 deficiency affects clearance of Aβ plaques and increases their toxicity mostly because of reduced numbers of microglia around the plaques.

To elucidate the mechanisms underpinning altered responses of Atg7-deficient microglia to amyloid plaques, we performed scRNA-seq. CD45⁺ live single cells were sorted from the cortices of 5-mo-old Atg7^fl/fl, Atg7^ΔMG, Atg7^fl/fl-5xFAD, and Atg7^ΔMG-5xFAD mice, and single-cell transcriptomes were generated using the 10x Genomics platform. We removed cells with low unique molecular identifier (UMI) counts, low gene counts, and high mitochondria ratio. After quality controls and doublet removal, we performed SC Transform to normalize the dataset and integration to eliminate batch effects. A total of 99,307 single cells were plotted on uniform manifold approximation projection (UMAP) dimensions for visualization. Unsupervised clustering revealed a total of 45 distinct clusters across all mice (Fig. S2 A). The preponderance of cells captured by our scRNA-seq dataset were microglia, with fewer border-associated macrophages, monocytes, T cells, B cells, and neutrophils (Fig. S2 B). Despite a slight increase of T cells in Atg7^ΔMG mice, none of the genotypes had a significant impact on the clusters representing non-microglial cell types (Fig. S2 C). Re-clustering of microglia resulted in eight distinct clusters comprising of 45,087 microglia (Fig. 3 A). Based on the expression of published microglia subtype-specific marker genes (Wang et al., 2020; Zhou et al., 2020), we identified homeostatic microglia (HM) (clusters 0, 1, and 2), transitional microglia (TM) (cluster 5), disease-associated microglia (DAM) (cluster 3), IFN-responsive microglia (cluster 4), and proliferating microglia (cluster 7) (Fig. 3 A and B; and Table S1). We also detected a cluster of microglia (cluster 6) that was, in addition to DAM genes, enriched for ferritin heavy polypeptide 1 (Fth1), ferritin light polypeptide 1 (Ftl1), peroxiredoxin-1 (Prdx1), and several ribosomal proteins (Rplp1, Rpl41, and Rps29) (Fig. 3, B and C). We refer to this cluster as ferritin microglia (FTM). This microglial population resembled the “microglia inflamed in MS–iron (MIMS-iron)” previously identified in scRNA-seq of multiple sclerosis brain samples, where it was associated with iron overload (Absinta et al., 2021). Indeed, FTM and MIMS-iron clusters shared 57 common signature genes, including Fth1, Ftl1, and 46 ribosomal genes (Fig. 3 D). Next, we analyzed the abundance of each cluster in relation to the different genotypes. DAM, TM, and IFN-responsive microglia were highly enriched in 5xFAD versus non-5xFAD mice, consistent with their involvement in the response to Aβ, while HM were less abundant (Fig. 3 E and Fig. S2 D). FTM were more enriched in Atg7-deficient microglia, in both 5xFAD and non-5xFAD background, as were module scores based on MIMS-iron signature genes (Fig. 3 E and Fig. S2 E). Fth1 and Ftl1 form the primary intracellular iron storage complex. This complex decreases the pool of bioavailable reactive iron, limiting its participation in oxidative reactions that generate reactive oxygen species (ROS) (Levi et al., 2024). Thus, the enrichment of FTM and upregulation of ferritin genes in Atg7^ΔMG-5xFAD versus Atg7^fl/fl-5xFAD controls suggested that microglia from Atg7^ΔMG-5xFAD mice are more exposed to oxidative stress.

To further investigate the transcriptional impact of Atg7 deficiency on microglia, we performed differential gene expression analysis on HM in all genotypes and on DAM together with TM in 5xFAD genotypes only. Analysis of differentially expressed genes (DEGs) in Atg7^fl/fland Atg7^ΔMG HM delineated decreased expression of calreticulin (Calr), BiP (Hspa5), protein disulfide isomerase family A member 3 (Pdia3, also known as ERp57), protein disulfide isomerase family A member 6 (Pdia6), stromal cell–derived factor 2 like 1 (Sdf2l1), and X-box–binding protein 1 (Xbp1) in Atg7^ΔMG HM (Fig. 3 F and Table S2). Calr, ERp57, PDIA6, and BiP are chaperone proteins residing in the ER that facilitate protein folding and protein quality control. BiP plays a crucial role in initiating UPR pathways. Unfolded proteins accumulated in the ER bind to BiP, causing BiP to detach from the UPR sensors—inositol-requiring enzyme-1α (IRE1α), protein kinase RNA-like ER kinase (PERK), and activating transcription factor 6 (ATF6), which trigger the downstream signaling (Hetz et al., 2020). Xbp1 is the transcription factor spliced by IRE1α into the active isoform Xbp1s. Sdf2l1 has been shown to interact with the BiP chaperone cycle and to manage ER-associated protein degradation; its expression is regulated by Xbp1s (Tiwari et al., 2013; Sasako et al., 2019; Fujimori et al., 2017). The downregulation of chaperone- and UPR-related genes suggested a weakened UPR in Atg7-deficient microglia. We also observed that the reduced expression of ER stress-related genes was accompanied by an increase in the expression of genes for MHC-I and MHC-II molecules (H2-D1, H2-K1, H2-Ab1, and Cd74), which are assembled in the ER and therefore may be also affected by alterations in the ER compartment.

Comparison of HM from Atg7^fl/fl-5xFAD and Atg7^ΔMG-5xFAD mice also exposed a reduced expression of UPR genes, including Calr, Hspa5, Pdia3, Pdia6, Sdf2l1, mesencephalic astrocyte-derived neurotrophic factor (Manf), and FK-506–binding protein 2 (Fkbp2) (Fig. 3 G and Table S3). Similar to aforementioned genes, mesencephalic astrocyte-derived neurotrophic factor is an ER chaperone protein with various functions (Yu et al., 2021). FK-506–binding protein 2 is a prolyl isomerase involved in the ER stress response (Jeong et al., 2017). Comparison of DEGs in DAM and TM from Atg7^fl/fl-5xFAD and Atg7^ΔMG-5xFAD mice showed that downregulated UPR-related genes were highly concordant with those observed in HM (Fig. 3 H and Table S4). Gene Ontology (GO) enrichment analysis of downregulated genes using clusterProfiler showed enrichment for terms related to unfolded protein, ER-associated protein degradation, and glycoprotein synthesis (Fig. S2, F–H). Comparison of the module score of “response to unfolded protein” GO pathway among the four genotypes examined corroborated that deficiency of Atg7 is associated with a defect of UPR regardless of the disease context or cell state (HM or DAM) (Fig. 3 I). The H2-DMa gene that promotes MHC-II peptide loading was downregulated in Atg7^ΔMG-5xFAD microglia (Fig. 3 H), and the module score of “antigen processing and presentation” was also affected by Atg7 deficiency (Fig. S3 A). Staining brain sections for I-A/I-E showed reduced MHC-II expression (Fig. S3 B), corroborating a defect in the posttranslational assembly of MHC-II. Along with the increased expression of genes for MHC-I and MHC-II molecules, these findings suggest an imbalance between MHC synthesis and assembly in the altered ER. Overall, suppression of ER stress in Atg7-deficient microglia is accompanied by suppression of MHC-II expression.

A notable set of genes upregulated in the HM and DAM clusters of Atg7^ΔMG-5xFAD mice compared with those in Atg7^fl/fl-5xFAD mice included oxidative stress response genes, such as Fth1, Ftl1, biliverdin reductase B (Blvrb), Prdx1, and transaldolase 1 (Taldo1) (Fig. 3, F–H;,Table S2, Table S3, and Table S4). Biliverdin reductase B is an oxidoreductase catalyzing multiple NADPH-dependent reduction processes, and one of its products, bilirubin, has a protective antioxidative role (Barañano et al., 2002); Prdx1 resolves oxidative stress by reducing peroxide and alkyl hydroperoxide (Neumann et al., 2009); transaldolase 1 helps maintaining the reduced state of glutathione and protects cellular integrity from oxygen radicals (Samland and Sprenger, 2009). The increased response to oxidative stress in Atg7-deficient microglia was confirmed by comparing the module scores for the “positive regulation of ROS metabolic process” GO pathway across the four genotypes (Fig. 3 J). Genes upregulated in the microglia of DAM and TM clusters of Atg7^ΔMG-5xFAD mice compared with those in Atg7^fl/fl-5xFAD mice also included Cdkn1a, which encodes for p21, which halts cell cycle progression, allowing for cellular responses to damage (Di Micco et al., 2021). We also noted an upregulation of Ninjurin-1 (Ninj1), which promotes ferroptosis-associated plasma membrane rupture (Ramos et al., 2024), suggesting that oxidative stress is associated to ferroptosis.

To confirm that Atg7 deficiency results in reduced UPR, we conducted an analysis of the three main UPR pathways: the IRE1α pathway, the PERK pathway, and the ATF6 pathway. Each of these pathways activates specific bZIP transcription factors to induce UPR gene expression (Hetz et al., 2020; Walter and Ron, 2011). We generated Atg7^fl/flCx3cr1^Cre mice where Atg7 is permanently deleted in microglia without the need for TAM. We cultured primary microglia from the brains of Atg7^fl/flCx3cr1^Cre mice and Atg7^fl/fl control mice, exposed them to tunicamycin, a well-known ER stress inducer, and assessed UPR protein levels via immunoblotting. The Atg7-deficient primary microglia displayed lower levels of Xbp1s (indicating the IRE1–Xbp1 pathway), ATF4 and CHOP (representing the PERK pathway), ATF6 (indicating the ATF6 pathway), and BiP (a chaperone protein that regulates various UPR pathways) (Fig. 4 A). Similarly, primary microglia of Atg7^fl/flCx3cr1^Cre mice exposed to Aβ₄₂ peptides showed reduced mRNA expression of Xbp1, Ddit3 (CHOP), and Hspa5 (BiP) compared with Atg7^fl/fl microglia (Fig. 4 B). These findings confirmed that all three UPR pathways were downregulated in Atg7-deficient microglia. This reduction in UPR was paralleled by an increase in protein synthesis rate as determined by quantifying the amount of nascent peptides by O-propargyl-puromycin labeling (Fig. 4 C). Finally, staining of brain sections for FTL showed that Atg7^ΔMG-5xFAD mice had significantly more microglia expressing this oxidative response molecule than 5xFAD at various pathological stages (Fig. 4, D and E), validating that Atg7 deficiency increases microglia exposure to ROS.

To investigate the mechanism behind microglia reduction in Atg7^ΔMG mice, we focused on potential cellular pathways. scRNA-seq data revealed an increase in p21 (Cdkn1a) expression, a key regulator of cell cycling. To determine if Atg7 deficiency affected proliferation, we stained brain sections for Ki67. The frequency of Ki67⁺ microglia in 5-, 8-, and 12-mo-old Atg7^ΔMG-5xFAD mice was similar to that in littermate controls (Fig. S4), suggesting that Atg7 deficiency did not significantly impact microglia proliferation. We then explored whether Atg7-deficient microglia were more susceptible to canonical apoptosis. Because apoptotic cells are rapidly cleared in vivo and difficult to detect with standard methods, we assessed the viability of primary microglia exposed to Aβ₄₂ peptide using Annexin V/7AAD staining. The results showed similar percentages of Annexin V⁺ early apoptotic cells in both Atg7-deficient and Atg7^fl/fl microglia, but there was an increase in Annexin V⁺ 7AAD⁺ late apoptotic/necrotic cells in the Atg7-deficient group (Fig. 4 F). Considering the enrichment of oxidative stress response signature genes and Ninjurin1 as well as FTL protein in Atg7-deficient microglia, we hypothesized that microglial loss might result from ferroptosis—a form of cell death triggered by excessive ROS production and subsequent peroxidation of membrane lipids—rather than from canonical autophagy. Analyzing ferroptosis levels in primary microglia cultured with or without Aβ₄₂ peptide showed higher lipid peroxidation levels in Atg7-deficient primary microglia under both conditions (Fig. 4 G). Atg7-deficient microglia had a higher number of mitochondria compared with Atg7^fl/fl microglia, regardless of Aβ exposure (Fig. 4 H). This could partly account for the increased oxidative stress observed. Overall, these findings suggest that Atg7 deficiency induces oxidative stress and lipid peroxidation, leading to increased ferroptosis of microglia.

Aging is the greatest risk factor for neurodegeneration, and cells in the aged brain also bear more proteotoxicity (Hetz, 2021). Moreover, aging is associated with the accumulation of toxic protein aggregates caused by increased ROS-mediated macromolecular damage. Thus, we asked whether microglial numbers decline in aged mice with microglial-specific Atg7 deficiency, as observed in Atg7^ΔMG-5xFAD mice. 1-mo-old Atg7^ΔMG and Atg7^fl/fl mice were fed with TAM-containing chow until 2 mo of age, returned to regular chow, and aged up to 20 mo prior to analysis. Microglia isolated ex vivo from Atg7^ΔMG mice contained significantly less ATG7 protein than did microglia from Atg7^fl/fl mice (Fig. 5 A). Moreover, Atg7^ΔMG microglia accumulated much more p62 (Sequestosome—SQSTM1) protein than did Atg7^fl/fl microglia (Fig. 5 A), consistent with a blockade of autophagic flux. However, IBA1 staining of brain sections established that the overall density of microglia in brain cortex was unaffected by Atg7 deficiency (Fig. 5 B). Bulk RNA-seq of microglia purified from the brains of 20-mo-old mice revealed lower expression of UPR genes, such as Calr, Pdia3, Pdia6, and Hspa5, in Atg7^ΔMG mice than in Atg7^fl/fl mice (Fig. 5, C and D). Atg7^ΔMG microglia also expressed more oxidative stress–related genes, some of which were not detected in scRNA-seq, such as xanthine dehydrogenase (Xdh) (Harrison, 2002) and oxidative stress–induced growth inhibitor 1 (Osgin1) (Li et al., 2007) (Fig. 5, C and D). Comparison of DEGs from the scRNA-seq dataset (Atg7^fl/fl versus Atg7^ΔMG HM, Atg7^fl/fl-5xFAD versus Atg7^ΔMG-5xFAD HM, and Atg7^fl/fl-5xFAD versus Atg7^ΔMG-5xFAD TM and DAM) with those from the bulk RNA dataset (Atg7^fl/fl versus Atg7^ΔMG-aged microglia) revealed six shared genes (Calr, Pdia3, Pdia6, Hspa5, Sdf2l1, and Cx3cr1), all of which were ER chaperone genes, with the exception of Cx3cr1, which reflects the halved gene dosage in mice-expressing Cx3cr1^CreERT2 (Fig. 5 E). We conclude that Atg7 deficiency diminishes UPR programs and increases oxidative stress regardless of the physiological/pathological states of microglia; however, a second hit, such as prolonged proteotoxic stress induced by accumulation of Aβ plaques, is necessary to disrupt microglia viability and numbers (Fig. S5).

This study demonstrates that microglia deletion of Atg7 reduces UPR and increases oxidative stress, making microglia susceptible to ferroptosis when challenged with proteotoxic stress induced by amyloid plaques. Previous studies have detected UPR in AD brains (Hoozemans et al., 2005, 2009; Stutzbach et al., 2013), mouse models of Aβ plaques (Cui et al., 2017; Soejima et al., 2013), and a model of tauopathy (Ho et al., 2012). A scRNA-seq study also revealed upregulation of genes involved in protein folding and molecular chaperones in AD patients with late pathology versus early pathology (Mathys et al., 2019), suggesting that long-term disease disrupts proteostasis. During AD, UPR has primarily been observed in neurons, showing both adaptive and harmful effects (Ajoolabady et al., 2022). In contrast, most research on UPR in microglia has focused on its role in cytokine production (Fernández et al., 2021). Xbp1s was found to bind the IFNβ promoter (Zeng et al., 2010) and promote the expression of type I IFN in a microglia cell line (Studencka-Turski et al., 2019). In a model of HIV-associated neurocognitive disorder, blockade of the IRE1α–Xbp1 (Pinto et al., 2022) or PERK pathway (Silveira et al., 2022) attenuated the conversion of microglia to an M1 inflammatory phenotype by the HIV trans-activator Tat. A microglia inflammatory response was also shown to be induced by ATF6 in the experimental autoimmune encephalomyelitis model of multiple sclerosis (Ta et al., 2016). Our results indicate that ER stress response in microglia requires ATG7, as does the expression of the MHC-II complex, which is assembled in the ER.

While ATG7’s role in autophagy and noncanonical autophagy, such as LC3-associated phagocytosis and endocytosis, is well established (Heckmann and Green, 2019; Peña-Martinez et al., 2022), its role in maintaining UPR in microglia was unexpected. Interestingly, an ATG7 defect appears to influence the ER stress response in other cell types as well. Atg7-deficient pancreatic β cells downregulated UPR-related genes; as a result, autophagy-deficient β cells were unable to meet the increased demand for UPR in obesity, facilitating occurrence of diabetes (Quan et al., 2012). Hepatocellular-specific deletion of Atg7 in hepatocytes also impaired UPR, although in a pathway-selective pattern (Kwanten et al., 2016): while the activity of ATF6 pathway was abolished, the PERK pathway escalated, resulting in apoptosis. Such dysregulation of autophagy and UPR induced severe hepatocellular injury and damage of liver architecture. Together with our findings, these studies establish a new paradigm in which ATG7 supports ER stress response pathways. ATG7 likely impacts the ER due to its role in regulating the formation and trafficking of membrane-bound organelles, including autophagosomes, endosomes, and lysosomes. However, the detailed molecular mechanisms underlying the ATG7–UPR pathway remain to be defined.

Our study demonstrates that ATG7 curtails oxidative stress in microglia, which may, in part, occur through mitophagy, reducing the persistence of dysfunctional mitochondria. Loss of Atg7 in microglia led to higher mitochondrial content, increased ROS, lipid peroxidation, and ferroptosis, ultimately reducing the microglia numbers and their ability to contain plaques neurotoxicity. Enhanced oxidative stress was marked by expansion of FTM. Ferritin-positive iron-laden microglia with dystrophic features have been seen in both AD patients and mouse models (Lopes et al., 2008; Kenkhuis et al., 2021; McIntosh et al., 2019). Furthermore, scRNA-seq of Parkinson’s disease brain tissues has revealed a population of FTH1⁺ microglia (Ryan et al., 2023). Similarly, single-nucleus RNA sequencing (snRNA-seq) of white matter lesions from multiple sclerosis patients identified an iron⁺ microglia cluster, featured by the expression of FTH1, FTL1, ribosomal genes, and MHC class II genes (Absinta et al., 2021), which closely resembles the FTM cluster observed in our study. Although microglial response to oxidative stress appears to be a common event in neurodegeneration, no studies have yet explored its connection with the expression and function of autophagy genes, which warrants further investigation. A recent study linked Atg7 deficiency to microglial senescence in an AD model (Choi et al., 2023). While we observed an increase in the expression of the senescence marker Cdkn1a, this gene may not only induce senescence but also facilitate the microglia response to cellular stress by inhibiting cell cycle progression.

Although there is no direct link between ATG7 variants in humans and AD, a previous study reported that patients with loss-of-function ATG7 variants exhibited abnormalities in the cerebellum and corpus callosum, as well as neurological symptoms like ataxia, seizures, learning difficulties, schizophrenia, and even late-onset dementia (Collier et al., 2021). Notably, the neurological symptoms in these patients are due to a global ATG7 deficiency, not just a deficiency specific to microglia. Additionally, mutations in other autophagy genes, such as ATG5, WIPI2, and SQSTM1, have also been linked to neurological diseases (Yamamoto et al., 2023). These studies indicate a wider link between autophagy deficits and neurological disorders, highlighting the need for further research. In conclusion, our study highlights the role of ATG7 in supporting the UPR and controlling oxidative stress in microglia in response to amyloid plaques. Therefore, autophagy, ER stress, and modulation of ROS should be considered important therapeutic targets for promoting beneficial microglial responses in AD.

Mice used in this study were age-matched female mice. Atg7^fl/fl mice were generated by Dr. Tomoki Chiba (Komatsu et al., 2005) and provided by Dr. Herbert Skip Virgin (Washington University in St. Louis, St. Louis, MO, USA). Atg7^fl/fl mice were crossed with Cx3cr1^CreERT2 mice (https://www.jax.org/strain/020940) to generate Atg7^ΔMG. They were then crossed with 5xFAD mice to generate Atg7^fl/fl-5xFAD and Atg7^ΔMG-5xFAD mice. To induce Atg7 deletion, 1-mo-old Atg7^fl/fl, Atg7^ΔMG, Atg7^fl/fl-5xFAD, and Atg7^ΔMG-5xFAD mice were fed with TAM diet (Envigo td.130857) for 4 wk and then switched to regular chow diet. For primary microglia culture, Atg7^fl/fl mice were crossed with Cx3cr1^Cre mice (https://www.jax.org/strain/025524) to generate Atg7^fl/flCx3cr1^Cre mice. Mice were housed in the animal facilities of Washington University in St. Louis. All animal experiments were conducted in compliance with institutional regulations, under authorized protocols #22–0274 approved by the Institutional Animal Care and Use Committee.

Brain tissues were isolated from postnatal day 2–day 4 Atg7^fl/flCx3cr1^Cre pups and their Atg7^fl/fl littermates, with olfactory bulbs and cerebella removed. Tissues were digested with trypsin (T1426; Sigma-Aldrich) and DNase I (Thermo Fisher Scientific) and homogenized into single-cell suspensions, which were then cultured with complete DMEM medium in T25 flasks precoated with poly-L-lysine (Thermo Fisher Scientific). Medium were changed on day 1, day 3, and day 5. Starting from ∼day 7 (when cells reach ∼90% confluency), microglia were detached from the mixed glia culture every other day by shaking the flasks on a 110-rpm orbital shaker for 30 min. After shaking, the supernatants were collected from the flask and transferred into 6-well plates. Here, the primary microglia in the supernatant were allowed to settle and attach to the bottom of the wells. DMEM was supplemented with 10% heat-inactivated FBS, 1% GlutaMAX (35050061; Gibco), 100 U/ml penicillin-streptomycin (15140122; Gibco), and 1 mM sodium pyruvate (11360070; Gibco). The medium was also supplemented with 10% L929 cell–conditioned medium to support microglia differentiation.

Primary microglia were collected and replated into 6-well plates at the density of 300,000 cells per well. The next day, cells were treated with 0.5 µg/ml tunicamycin (T7765; Sigma-Aldrich) for 18 h and harvested for immunoblotting (Fig. 4 A).

Αβ_1–42 peptides were obtained from AnaSpec and reconstituted as suggested in their manufacturing instructions. Primary microglia were collected and replated into non-tissue culture 96-well plates at the density of 20,000 cells per well. Cells were treated with 40 µM of Αβ_1–42 peptides for 48 h and harvested for RNA extraction (Fig. 4 B) or Annexin-V/7AAD staining (Fig. 4 F). For Annexin-V/7AAD staining, we followed the protocol from PE Annexin V Apoptosis Detection Kit I (BD Biosciences).

Primary microglia were harvested and replated in non-tissue culture 96-well plates at the density of 50,000/well. The next day, cells were harvested, and protein synthesis rates were measured using the Protein Synthesis Assay Kit (ab239725; Abcam).

Primary microglia were harvested and replated in non-tissue culture 96-well plates at the density of 20,000/well. Cells were cultured using complete DMEM and treated with or without 40 µM of Αβ_1–42 peptides for 48 h. Lipid peroxidation was measured using Bodipy 581/591 C11 (D3861; Thermo Fisher Scientific). Mitochondria mass was measured using MitoTracker Green FM Dye (M46750; Thermo Fisher Scientific).

Microglia were stained with an antibody cocktail and Fc block for 30 min on ice. Events were acquired by Canto II (BD Bioscience). The following antibodies were used in the study: CD45-PE (30-F11, 103106; BioLegend), CD45-APC-Cy7 (30-F11, 103116; BioLegend), and CD11b-APC (M1/70, 101212; BioLegend).

Mice were perfused with ice-cold PBS containing 1 U/ml of heparin. Brain tissues were collected into PBS supplemented with 1 mM EDTA (Corning) and mechanically dissociated using the Dounce homogenizer. Myelin lipids were removed by density gradient centrifugation using 30% isotonic Percoll (GE Healthcare). The resulting cell suspensions were stained with CD45-PE (30-F11, 103106; BioLegend), CD11b-APC (M1/70, 101212; BioLegend), and DAPI. Microglia, identified as DAPI⁻ CD45^lo CD11b⁺ populations, were sorted by BD FACSAria II. Sorted cells were collected into PBS containing 2% FBS for TEM or immunoblotting, or into PBS with 0.04% BSA for scRNA-seq.

Sorted microglia or primary microglia were lysed with RIPA buffer (Cell Signaling Technology). Protein concentrations were measured by the Detergent Compatible Protein Assay Kit (# 5000111; Bio-Rad). Equal amounts of protein from each sample were mixed with 4× Laemmli Sample Buffer (#1610747; Bio-Rad) and separated on 4–20% Mini-PROTEAN TGX Precase Protein Gels (#4561094; Bio-Rad). Following electrophoresis, proteins were transferred onto polyvinylidene difluoride membranes. Membranes were washed in Tris-buffered saline with 0.1% Tween 20 and blocked with 5% BSA in Tris-buffered saline with 0.1% Tween 20 for 1 h at room temperature. Primary antibody incubation was performed overnight at 4°C, after which membranes were washed and incubated in HRP-conjugated secondary antibody for 1 h at room temperature. Membranes were washed again and developed using SuperSignal West Pico PLUS Chemiluminescent Substrate (Thermo Fisher Scientific). Immunoblot images were taken using the Bio-Rad ChemiDoc MP imaging system.

The primary antibodies used in this study included: anti-ATG7 (D12B11, Rabbit mAb, #8558; Cell Signaling Technology), anti-SQSTM1/p62 (D1Q5S, Rabbit mAb, #39749; Cell Signaling Technology), anti-LC3B (Rabbit, #2775; Cell Signaling Technology), anti-CHOP (L63F7, Mouse mAb, #2895; Cell Signaling Technology), anti-β-actin (13E5, Rabbit mAb, HRP-conjugated, #5125; Cell Signaling Technology), anti-BiP (C50B12, Rabbit mAb, #3177; Cell Signaling Technology), anti-Xbp1s (D2C1F, Rabbit mAb, #12782; Cell Signaling Technology), anti-ATF-4 (D4B8, Rabbit mAb, #11815; Cell Signaling Technology), and anti-ATF-6 (D4Z8V, Rabbit mAb, #65880; Cell Signaling Technology).

RNA was extracted from primary microglia using the RNeasy Plus Micro Kit (QIAGEN) according to the manufacturer’s instructions. RT was carried out using the iScript cDNA synthesis kit (#1708890; Bio-Rad). The resulting cDNA was diluted 30-fold prior to quantitative RT-PCR. Quantitative PCR was performed using iTaq Universal SYBR Green Supermix (#1725121; Bio-Rad) on a CFX96 Touch Real-Time PCR Detection System (Bio-Rad). Gene expression was normalized to the reference gene Actb (β-actin), and relative gene expression levels were determined by the ΔΔCt method. The following primers were used: Ddit3 forward: 5′-GTC​CCT​AGC​TTG​GCT​GAC​AGA-3′; Ddit3 reverse: 5′-TGG​AGA​GCG​AGG​GCT​TTC-3′; Actb forward: 5′-GGA​GGG​GGT​TGA​GGT​GTT-3′; Actb reverse: 5′-TGT​GCA​CTT​TTA​TTG​GTC​TCA​AG-3′; Hspa5 forward: 5′-TCA​TCG​GAC​GCA​CTT​GGA-3′; Hspa5 reverse: 5′-CAA​CCA​CCT​TGA​ATG​GCA​AGA-3′; and Xbp1 forward: 5′-AAG​AAC​ACG​CTT​GGG​AAT​GG-3′; Xbp1 reverse: 5′-ACT​CCC​CTT​GGC​CTC​CAC-3′.

Cortices were isolated for Aβ ELISA. PBS lysates were obtained by homogenizing tissues with PBS containing cOmplete Protease Inhibitor Cocktail (#1873580; Roche). Homogenates were spined at 4°C for 14,000 rpm/15 min, and the supernatants were collected as PBS lysates (soluble fraction). The pellets were then homogenized with guanidine solution (5.5 M guanidine with 50 mM Tris + protease inhibitor, pH = 8.0), gently mixed on a rotor for 3 h, and spun down. The supernatants were collected as the guanidine fraction (insoluble fraction). The volumes of PBS and guanidine solution used were proportional to the tissue weight. A protein assay was conducted before the ELISA to determine the protein concentration of each sample.

Plates were coated with 10 µg/ml Aβ42 antibody (HJ7.4) at 4°C overnight; standards were made by serial dilution of Aβ (1–42) protein (AS-24224; AnaSpec). For ELISA of insoluble fraction, we loaded the plates with 10,000× diluted guanidine fractions and incubated overnight at 4°C. After incubation, the wells were washed with PBS-T and incubated with mHJ5.1-Biotin for 90 min at 37°C. Plates were then washed, incubated with Strep-poly-HRP40 for 90 min at room temperature, and read at 650-nm absorbance.

Mice were peritoneally injected with 10 mg/kg of Methoxy-X04 dye. After 3 h, mice were sacrificed, and microglia were isolated by mechanical homogenization with Dounce homogenizer and Percoll gradient separation. Cells were stained with CD45, CD11b, and aqua live/dead dye. Flow cytometry was performed on Canto II (BD Bioscience).

Microglia were prepared as previously described (Wang et al., 2022). In brief, sorted microglia were fixed, embedded, and stained with Eponate 12 resin (Ted Pella Inc.). Ultrathin sections (95 nm) were cut with Leica Ultracut UCT ultramicrotome (Leica Microsystems Inc.) and counterstained with uranyl acetate and lead citrate. Samples were imaged on the JEOL 1200 EX II transmission electron microscope (JEOL USA Inc.). Quantitative evaluation was performed by taking random images of 30 cells of comparable size. Data were expressed as the total number of multivesicular/multilamellar structures per cell.

Mice were perfused with ice-cold PBS containing 1 U/ml of heparin. Brains were isolated, fixed by 4% paraformaldehyde in PBS overnight at 4°C, and dehydrated by 30% sucrose in PBS overnight at 4°C. Brain tissues were then embedded in a 2:1 mixture of 30% sucrose in PBS and Tissue-Plus O.C.T. Compound (Thermo Fisher Scientific). 40-μm thick coronal sections were obtained on Cryostat (CM1950; Leica). Sections were blocked and permeabilized by 5% BSA and 0.5% Triton-X100 in PBS solution for 1–2 h at room temperature. Primary antibodies were added at indicated concentrations and incubated overnight at 4°C. After three washes, secondary antibody staining was performed at room temperature for 1 h. Sections were embedded with Fluoromount-G (SouthernBiotech) for imaging. For most of the data (Fig. 1, D–K; Fig. 2, A–D and H–J; Fig. 4, D and E; Fig. 5 B; Fig. S1 D; Fig. S3 B; and Fig. S4), confocal images were taken by Nikon A1Rsi+. For the quantification of plaques (Fig. S1, A and B), images were taken by Zeiss Axio Scan 7 Fluorescence Slide Scanner.

Primary antibodies and the concentrations used in this study: 1:1,000 for anti-Iba1/AIF1 (Rabbit mAb, #17198; Cell Signaling Technology), 1:500 for anti-Iba1 (Goat polyclonal, ab5076; Abcam), 1:1,000 for anti-CD68 (#137002; BioLegend), 1:500 for anti-PU.1 (#2258; Cell Signaling Technology), 1:1,000 for anti-LAMP1 (#121602; BioLegend), 1:1,000 for anti-CD11c (#97585; Cell Signaling Technology), 1:1,000 for anti-β-amyloid, 1–16 (6E10, mouse IgG1, #803013; BioLegend), 1:500 for anti-FTL (ab69090; Abcam), and 1:200 for anti-I-A/I-E (#107618; BioLegend), 1:200 for anti-Ki67 (ab16667; Abcam).

Secondary antibodies or dyes and their concentrations: anti-goat IgG Alexa-Fluor 488, 1:1,000 (donkey polyclonal; Abcam), anti-rabbit IgG Alexa Fluor 647, 1:1,000 (goat recombinant polyclonal; Invitrogen), anti-rabbit IgG Alexa Fluor 555, 1:1,000 (donkey polyclonal; Abcam), and Methoxy-X04 1:3,000 (3 mg/ml; Tocris).

Images were acquired using the Zeiss Axio Scan 7 Fluorescence Slide Scanner and analyzed with ImageJ. The cortex region in each image was marked by the “selection” tool. To account for variabilities in background staining, the intensity threshold was dynamically set as “0.95 * mean intensity” rather than a fixed value. The area above/below the threshold is quantified by functions “analyze particles” and “summarize,” which were automatically calculated by batch processing. Two to three brain sections per mouse were used, and the data in Fig. S1 are the average value of each mouse.

We classified plaque morphology as previously described (Wang et al., 2020). Briefly, “inert plaques” are those only made of fibrillar Aβ, which is stained by Methoxy-X04; “mixed plaques” exhibit a core of Methoxy-X04⁺ fibrillar Aβ, surrounded by a halo of Methoxy-X04⁻ 6E10⁺ non-fibrillar Aβ; and “filamentous plaques” have little content of β-sheet structure or branched amyloid fibrils, and are only positively stained by 6E10 antibody. The proportion of each plaque type was calculated by normalizing to the total number of plaques in each image.

The plaque-associated microglia were identified using the “Spots” and “Surface” functions of Imaris (Bitplane), followed by running a MATLAB script as described before (Ulland et al., 2017). Briefly, we applied the spots function to PU.1⁺ staining to locate each microglia and applied the surface function to Methoxy-X04⁺ staining to capture the locations and volumes of each dense-core plaque. With the three inputs (microglia locations, plaque locations, and plaque volumes), the MATLAB script was able to calculate the overall microglia density within a 15-μm distance from plaque surfaces in each image; it also calculated the average number of microglia within the same 15-μm shell regions surrounding the plaques. The data in Fig. 1, F and G are the average number of microglia within the shell regions, which reflects how dense microglia cluster around the plaques.

The volumes of dystrophic neurites around the plaques were calculated by a similar approach. The surface function was applied to both LAMP1⁺ staining and Methoxy-X04⁺ staining. Thus, there were four inputs for the MATLAB scripts: locations of dystrophic neurites, volumes of each dystrophic neurites, plaque locations, and plaque volumes. The output of the MATLAB scripts is the average volume of LAMP1⁺ dystrophic neurites within the 15-μm shell region in each image.

The volume of IBA1⁺ staining was quantified by the surface function in Imaris. To calculate IBA1⁺CD11c⁺ colocalization volumes, individual surfaces were created for the IBA1 and CD11c channels. The built-in “Batch Colocalization” function in Imaris was then applied to these two channels to determine the overlapping volumes. The same approach was applied for the quantification of IBA1⁺CD68⁺ colocalization, IBA1⁺FTL⁺ colocalization, and IBA1⁺MHC-II⁺ colocalization.

Cells were isolated from the cortex of 20-mo-old mice (three Atg7^fl/fl and three Atg7^ΔMG) and stained for CD45, CD11b, and DAPI. Live CD45^intCD11b⁺ microglia were FACS sorted into RLT Lysis Buffer (provided in QIAGEN RNeasy Plus Micro Kit). More than 20,000 microglia were collected from each mouse. RNA was extracted by QIAGEN RNeasy Plus Micro Kit. RNA integrity numbers for all six samples were >9. The following steps were described in the paper from Peng et al. (2022). Data were analyzed using DESeq2 package. GO analysis was performed using clusterProfiler (Wu et al., 2021). Venn diagram in Fig. 5 E was made by R package ggvenn.

We performed scRNA-seq on 5-mo-old female mice, including three Atg7^fl/fl, three Atg7^ΔMG, two Atg7^fl/fl-5xFAD, and three Atg7^ΔMG-5xFAD mice. Mice were perfused with cold PBS, and cortices were dissociated. CD45⁺ cells were sorted into 0.04% BSA in PBS, followed by library construction using the 10x Genomics Chromium Single Cell 3′ v3 Gene Expression Kit. Libraries were sequenced by Illumina sequencer at the McDonell Genome Institute. Cell Ranger Software Suite (v6.0.1) from 10x Genomics was used for sample demultiplexing, barcode processing, and single-cell counting. The reference genome GRCm39 was used for alignment. Seurat (v4.0) package in R (v4.1.0) was used for downstream analysis (Hao et al., 2021).

During quality control steps, we filtered out cells with >10% mitochondrial content, as well as cells with low UMI counts (<1,000) and low gene number per cell (<500). Cutoffs for UMI and gene number were empirically determined by histograms showing cell density as a function of UMI per gene counts. Genes expressed in <10 cells were also removed from the dataset. A total of 99,307 cells were remained after quality control.

Microglia clusters were digitally isolated and further filtered for clusters containing doublets, stress genes, and apoptotic cells. After filtering, the dataset from Atg7^fl/fl, Atg7^ΔMG, Atg7^fl/fl-5xFAD, and Atg7^ΔMG-5xFAD mice (11 mice in total) contained a total of 45,087 microglia cells with a median 2,825 UMI and median 1,426 genes. We normalized the data using SC Transform (Hafemeister and Satija, 2019) (regressed on mitochondria ratio) and integration using FindIntegrationAnchors function (Stuart et al., 2019). Principle component analysis was performed, and the top 30 principal components were selected for dimensionality reduction using the uniform approximation and projection algorithm. Clustering was performed using the FindNeighbors and FindClusters functions using a resolution of 0.4. Marker genes were identified by comparing each cluster against all other clusters using the FindAllMarkers function. Cell clusters from each tissue were annotated based on marker gene expression. GO analysis was performed using clusterProfiler (Wu et al., 2021). P values in violin plots were calculated using stat_compare_means() function of ggpubr package (0.4.0).

All graphs display the group mean ± SEM or ± SD, as specified in the figure legends. Statistical analysis was performed using GraphPad Prism software. P < 0.05 was considered as significant difference between different conditions, determined by two-tailed unpaired t test, one-way ANOVA with Tukey’s multiple comparisons test, or two-way ANOVA with Sidak’s multiple comparisons test, as indicated in the figure legends.

We are including five supplemental figures and four supplemental tables that support and extend the findings in the manuscript. Fig. S1 shows the Aβ plaque burden in Atg7^fl/fl-5xFAD and Atg7^ΔMG-5xFAD mice. Fig. S2 provides an expansion of the scRNA-seq analysis of microglia. Fig. S3 shows the reduction of MHC-II surface expression in microglia from Atg7^ΔMG-5xFAD mice. Fig. S4 illustrates the abundance of proliferative microglia in Atg7^fl/fl-5xFAD and Atg7^ΔMG-5xFAD mice. Fig. S5 is a schematic representation illustrating the outcomes of microglia-specific ATG7 deficiency. Table S1 listed the feature genes of each microglia sub-cluster of scRNA-seq study. Table S2 listed the DEGs in HM between Atg7^fl/fl and Atg7^ΔMG mice. Table S3 listed the DEGs in HM between Atg7^fl/fl-5xFAD and Atg7^ΔMG-5xFAD mice. Table S4 listed the DEGs in “TM + DAM” between Atg7^fl/fl-5xFAD and Atg7^ΔMG-5xFAD mice.

scRNA-seq and bulk RNA-seq data generated in this study were deposited in the Gene Expression Omnibus (GSE286094 and GSE286095 respectively). scRNA-seq data was also deposited on the Broad Institute Single Cell Portal (SCP2885).

We would like to thank Dr. Christina Stallings for the helpful discussions and Dr. Susan Gilfillan for the critical reading. We thank the Genome Technology Access Center at the McDonnell Genome Institute for scRNA-seq and bulk RNA-seq. The center is supported by the Washington University Institute of Clinical and Translational Sciences grant UL1TR002345 from the National Center for Advancing Translational Sciences of the National Institutes of Health (NIH). We thank the Pathology and Immunology Flow Cytometry Core for cell sorting.

Z. Cai was supported by American Heart Association (24PRE1194958). This work was supported by the Needleman Center for Autophagy Therapeutics and Research, the Fred and Ginger Haberle Charitable Fund at East Texas Communities Foundation, and the NIH (R01 AG051485, P01 AG078106, R01 AG081631, R01 AI158579, and R21 AG085133).

Author contributions: Z. Cai: conceptualization, data curation, formal analysis, funding acquisition, investigation, visualization, and writing—original draft, review, and editing. S. Wang: conceptualization, formal analysis, investigation, methodology, project administration, validation, visualization, and writing—original draft, review, and editing. S. Cao: conceptualization, investigation, and methodology. Y. Chen: resources. S. Penati: formal analysis and investigation. V. Peng: data curation, formal analysis, methodology, and visualization. C.M. Yuede: methodology and resources. W.L. Beatty: data curation, investigation, and resources. K. Lin: data curation and methodology. Y. Zhu: investigation. Y. Zhou: conceptualization, data curation, and software. M. Colonna: conceptualization, funding acquisition, project administration, resources, supervision, and writing—review and editing.