Alzheimer’s disease risk protein SorLA regulates ER homeostasis and lipid metabolism in human microglia, with conserved effects in neurons

Abstract

Microglial dysfunction is a hallmark of Alzheimer’s disease (AD), yet the molecular mechanisms driving these impairments remain poorly defined. Genetic studies implicate several AD-associated genes in regulating microglial activity, including SORL1, which encodes the sorting receptor SorLA. Although SorLA is highly expressed in microglia, its functional role in cellular homeostasis has remained unclear. Here, we investigated SorLA function using human brain tissue, primary microglia from rapid autopsies, and CRISPR-engineered human iPSC-derived microglia and neurons. Integrated multi-omics analyses, including single-cell RNA sequencing, lipidomics, and proteomics, together with biochemical and functional assays, revealed that SorLA deficiency induces endoplasmic reticulum (ER) stress and interferon signaling, promotes lipid droplet accumulation, and impairs phagocytic and immune functions. Protein co-complex mapping and structural modeling identified ER-associated proteins co-enriched with SorLA, including SUN2, calnexin (CANX), and multiple COPI complex components (COPA, COPB1, COPG1, ARCN1), implicating SorLA in ER proteostasis and intracellular trafficking. Notably, SORL1 deletion in iPSC-derived neurons recapitulated key phenotypes observed in microglia, including lipid droplet accumulation and SorLA–SUN2 co-immunoprecipitation, indicating that this ER-associated pathway operates across distinct brain cell types. Together, these findings identify an ER-related role for SorLA that extends beyond its established function in endocytic trafficking. Loss of SorLA triggers maladaptive stress responses, perturbs lipid handling, and compromises cellular resilience, thereby contributing to AD-relevant cellular dysfunction.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00401-026-03002-9.

Introduction

Microglial dysfunction is being increasingly recognized as a crucial pathological component of neurodegenerative disorders, particularly Alzheimer’s disease (AD) [1, 2]. Genome-wide association studies have identified genetic risk variants for AD, many of which influence genes and pathways that are highly active in microglia [3]. Indeed, several of the best-established AD susceptibility genes are predominantly expressed in microglia within the central nervous system [4, 5]. Despite this strong genetic and cell-type-enriched association, the molecular mechanisms by which these genes contribute to AD risk and progression remain poorly understood.

One such gene is SORL1, which encodes a large multi-domain sorting receptor called sortilin-related receptor 1 (SorLA) and has a robust genetic link to AD. Candidate gene approaches and large meta-analysis [6, 7] have identified protective variants of SORL1 with significance at a genome-wide level. Concurrently, multiple sequencing studies have demonstrated the association of SORL1 with AD through rare [8–10], ultra-rare [11], and missense [12] variants. Notably, reduced SORL1 expression has been observed in the brains of AD patients [13, 14] even prior to the initial genetic association with AD [15]. Evidence from in vitro cell cultures [15], mouse models [16, 17], and association studies [8, 10], support the idea that SORL1 haploinsufficiency contributes to AD risk. Moreover, mutations resulting in truncated SorLA protein, frequently observed in AD patients, are highly penetrant [8, 12]. indicating that loss of SorLA function can substantially increase susceptibility to AD.

Although microglia express high levels of SORL1 transcripts [18–20], the primary focus of SorLA research in AD to date has been neuron-centric. SorLA’s domain architecture is complex, incorporating functional motifs including a VPS10 domain, low-density lipoprotein receptor (LDLR) class A repeats, EGF repeats, and a cassette of three fibronectin type III (FN3) domains [21], consistent with multifunctional roles in intracellular trafficking. In neurons, SorLA facilitates transport of amyloid precursor protein (APP) to the trans-Golgi network thereby limiting amyloidogenic processing and Aβ generation. Consistent with this role, multiple studies [22–24] have shown that SORL1 deficiency disrupts endosomal trafficking and lysosomal pathways in neurons, leading to defects in APP processing, autophagy, and other features commonly observed in the AD brain [25–27]. A multi-omic study [28] using embryonic stem cell-derived microglia further reported functional interactions between SORL1 and TREM2 that influence apolipoprotein E (APOE) expression. Consistent with this a recent study [29] showed that SorLA deficiency disrupts endolysosomal trafficking and lysosomal enzyme delivery in iPSC-derived microglia. In addition, SorLA has been implicated in lipid and glucose metabolism in adipocytes [30]. Collectively, these findings suggest that SorLA may play a broader role in regulating cellular protein and lipid homeostasis including in microglia, and that its disruption could contribute to microglial dysfunction in AD.

To investigate the role of SorLA in microglial biology and AD, we employed an integrated approach combining postmortem brain tissue from AD and non-demented donors, freshly purified human primary microglia, and in vitro systems, primarily human iPSC-derived microglia (iMG), with complementary analyses in THP-1 cells and iPSC-derived neurons. Notably, SorLA protein levels were significantly reduced in AD brain tissue compared to non-demented controls. A multi-omic approach, including single-cell RNA sequencing, lipidomics, proteomics, and CRISPR-based gene perturbation, revealed that loss of SorLA disrupts cellular homeostasis by inducing endoplasmic reticulum (ER) stress, activating interferon responses, and perturbing lipid metabolism. Protein co-complex analyses further identified SorLA associations with ER-resident and perinuclear proteins, including SUN2, implicating SorLA in ER proteostasis and intracellular trafficking. Together, these findings position SorLA as an important regulator of cellular homeostasis, with particularly prominent effects in microglia, and suggest that its dysfunction may contribute to AD-relevant cellular pathology.

Methods

Cell culture

A commercially available human episomal iPSC line carrying the APOE3/3 genotype, derived from a healthy female neonate donor (Gibco, cat. no. A18945) was used for differentiation into microglia (iMG) and neurons (iN). This line is footprint-free as it was generated from cord blood CD34⁺ cells using a non-integrating EBNA-based episomal reprogramming system, resulting in no genomic integration of reprogramming factors [31, 32]. iPSCs were maintained according to the manufacturer’s instructions in mTeSR^TM1 medium (StemCell Technologies, cat. no. 85850) and passaged every 3–4 days on Geltrex-coated 6-well plates. Cells were cultured at 37 °C in a humidified incubator with 5% CO₂ and routine mycoplasma testing and karyotype analysis were performed to ensure culture quality and genomic stability.

Normal human astrocytes for iPSC-differentiation into microglia (iMG) were sourced from LONZA (cat. no. CC-2565) and were maintained in ABM basal medium (Lonza, cat. no. CC-3187). Upon reaching 80% confluence, the cells were transferred to Poly-L-lysine (PLL) (Sigma, cat. no. P4707) coated plates. Conditioned media were prepared by switching ABM to microglia progenitor cell medium [33] after reaching 60% confluence. Conditioned media were collected every two days.

The human monocytic cell line THP-1 was obtained from ATCC (cat. no. TIB-202) and was cultured in RPMI medium (Gibco, cat. no. 61870127), supplemented with 10% fetal bovine serum (FBS) and 1% penicillin–streptomycin.

SH-SY5Y human neuroblastoma cells (ATCC cat. no. CRL-2266™) were maintained under standard culture conditions in Dulbecco’s modified Eagle medium/F12 (DMEM/F12) supplemented with 10% fetal bovine serum and 1% penicillin–streptomycin. Cells were cultured at 37 °C in a humidified atmosphere containing 5% CO₂ and routinely passaged at ~70–80% confluence using trypsin–EDTA. Culture medium was replaced every 2–3 days.

Two-step differentiation of iPSCs into microglia

Human iPSCs were differentiated into microglia (iMG) following our previously established protocol [33]. Briefly, iPSCs were first differentiated into hematopoietic progenitors using the STEMdiff Hematopoietic Kit (STEMCELL Technologies, cat. no. 05310). These progenitors were then cultured in microglia-inductive media (IMDM + 10% FBS + IL-3, GM-CSF, and M-CSF) supplemented with astrocyte-conditioned medium (ACM) in a 1:1 ratio. Floating cells were collected after 8 days and characterized using flow cytometry, immunocytochemistry (ICC), RNA/protein and ATAC-Seq profiling, all of which confirmed their iMG identity. This iMG phenotyping dataset, along with the results of morphological analysis and assessment of immune responses, is available at: https://sherlab.shinyapps.io/IPSC-derived-Microglia/. For functional assays iMG were collected every 2 to 3 days based on the confluency and plated on PLL-coated plates in near-homeostatic, serum-free medium adapted from Muffat et al. 2016 [34].

Differentiation of iPSCs into neurons

WT and SORL1 KO human iPSCs (Gibco cat. no. A18945), the same parental line used for iMG differentiation, were differentiated into excitatory neurons using a doxycycline-inducible NGN2 overexpression system. Lentiviral particles were generated from pCDH-EF1-rtTA-PGK-BSD (Addgene #192885) and pLV-TetO-NGN2-Puro (Addgene #79049). To establish inducible rtTA+ iPSC master lines, iPSCs were plated on Geltrex at ~50–60% confluence in mTeSR1, supplemented with Y-27632 when passaged as single cells, and transduced with rtTA lentivirus in the presence of polybrene (8 µg/mL). Cells were selected with blasticidin for three days, expanded, and cryopreserved as rtTA+ master stocks. For neuronal differentiation, rtTA+ iPSCs were replated on Geltrex at 50–70% confluence and transduced with TetO-NGN2 lentivirus (MOI 2–10) in mTeSR1 containing polybrene, followed by overnight incubation. NGN2 expression was induced the next day by switching to neuronal induction medium (DMEM/F12 supplemented with N2 and doxycycline at 1 µg/mL). From days 2–4, puromycin (1–2 µg/mL) was applied to select transduced cells. On day 4, cells were dissociated with Accutase and replated onto poly-D-lysine- and laminin-511-coated plates at ~30–50 × 10³ cells/cm² in induction medium supplemented with doxycycline and ROCK inhibitor. Neuronal induction medium was refreshed every other day until day 7, when cultures were transitioned to neuronal maturation medium consisting of Neurobasal supplemented with B27, BDNF (10 ng/mL), GDNF (10 ng/mL), cAMP (500 µM), and ascorbic acid (200 µM). Doxycycline was maintained for at least 10 days post-induction, with media changes every 2–3 days. By days 14–21, cultures consisted of mature neurons expressing MAP2, and NeuN (Supplementary extended data Fig. 5).

Postmortem human brain specimens and SorLA protein expression analysis

Dorsolateral prefrontal cortex (DLPFC) specimens from autopsy brains were obtained from the Religious Orders Study and Rush Memory and Aging Project (ROSMAP) at the Rush Alzheimer’s Disease Center (RADC) in Chicago [35–37], and Columbia University Medical Center/New York Brain Bank in New York, NY [38]. ROS and MAP were approved by the Institutional Review Board of Rush University Medical Center and all participants signed Informed and Repository consents and an Anatomic Gift Act. All brain specimens were obtained through informed consent and/or a brain donation program at the respective organizations. All procedures and research protocols were approved by the corresponding ethical committees of our collaborators’ institutions as well as the Institutional Review Board (IRB) of Columbia University Medical Center (protocol AAAR4962). The details of human brain specimens used in this study are provided in the Supplementary Table S1. Details of the ROS/MAP clinical and pathologic evaluation have been reported [39–41].

DLPFC tissue sections with 16 μm thickness were deparaffinized using CitriSolv (Decon Labs Inc., cat. no. 5989-27-5), rehydrated with decreasing grades of ethanol and washed with PBS. Antigen retrieval was performed using a commercial reagent from Sigma (cat. no. E1161) at pH 7.4, followed by blocking nonspecific binding with BSA. Tissue slides were incubated overnight with primary antibodies, washed, and then incubated with secondary antibodies. Lipofuscin signal was blocked using Biotium TrueBlack Lipofuscin Quencher (Fisher Scientific, cat. no. NC1125051) and the tissue sections were mounted with DAPI and dried overnight. Imaging was performed using an Olympus BX3 microscope and images were analyzed using CellProfiler software.

SorLA protein expression was analyzed in neurons (n = 2811), oligodendrocytes (n = 548), microglia (n = 248), and astrocytes (n = 24) from DLPFC sections of three aged, non-pathological human subjects using immunohistochemistry (IHC). Slides were imaged using an Olympus BX3 fluorescence microscope under identical acquisition settings across samples.

Automated cell segmentation and fluorescence quantification were performed using CellProfiler v4.2.1. Briefly, DAPI-positive nuclei were identified and masked using the IdentifyPrimaryObjects module. Cell-type-specific markers were thresholded to define cellular regions, which were converted into object masks using the ConvertImageToObjects module. Objects were filtered by size to exclude debris and artifacts using MeasureObjectSizeShape and FilterObjects, and further filtered to retain only objects containing a nucleus using RelateObjects and FilterObjects. For cells with complex morphology (e.g., microglia), the SplitOrMergeObjects module was applied to ensure that cellular processes were correctly associated with the corresponding cell bodies. Once individual cells were defined, fluorescence intensity and morphological features were extracted using MeasureObjectIntensity and MeasureObjectSizeShape. Cell-type-specific SorLA expression was quantified as the mean fluorescence intensity per cell (MeanIntensity) in the SorLA channel. Mean intensity values were then aggregated by cell type for downstream statistical analysis.

For statistical analysis, normality was assessed using the Shapiro–Wilk test, and homogeneity of variances was evaluated using Bartlett’s test. As data across cell types were non-normally distributed with unequal variances, a Kruskal–Wallis test was used followed by Dunn’s post hoc test with Bonferroni correction. Comparisons of SorLA expression between AD and non-AD subjects within microglia and neurons were normally distributed and analyzed using unpaired two-tailed t-tests. Statistical significance was defined as p < 0.05, and all analyses were performed using GraphPad Prism.

Isolation of human primary microglia from fresh brain autopsies

Primary human microglia were isolated from DLPFC using a non-enzymatic approach as previously used [42, 43]. All procedures were performed on ice. Briefly, cortical gray matter was dissected from DLPFC autopsy samples, the tissue underwent mechanical dissociation, the resulting cell suspension was filtered, and centrifuged. Pellets were resuspended, incubated with anti-myelin magnetic beads, and subjected to myelin depletion using magnetic separation columns. Cell suspensions were stained with antibodies, washed, filtered, and live microglia (CD11b⁺/CD45⁺/7AAD−) were sorted using a BD Influx cell sorter (Fig. S5c). Sorting used a 100-um nozzle and occurred at 8000–10,000 events per second. After isolation, 20,000 human primary microglia (pMG) were plated on a poly-L-lysine-coated chamber slide, before performing ICC.

Immunocytochemistry

ICC was conducted for various analyses on iMG and primary microglia using standard ICC protocol. Briefly, cells were plated on PLL laminin-coated chamber slides and fixed with 4% PFA. Following PBS washes, cells were blocked with 3% BSA and incubated with primary antibodies followed by secondary antibodies in 1% BSA/PBS. Following further washes, cells were mounted with DAPI (Invitrogen, cat. no. P36931).

Supplementary Table S2 provides information on the antibodies used for brain tissues and in vitro cellular immunofluorescence staining.

ER morphological analysis

Fluorescent images of WT and SORL1 KO iMG immunostained for calnexin (ER marker, red channel) and DAPI (nuclei, blue channel) were analyzed using ImageJ/Fiji. Nuclei were segmented via Otsu thresholding on DAPI; ER regions were thresholded on calnexin and analyzed for fragmentation (connected-component labeling), perinuclear overlap (dilated nuclear ROI), area fraction, texture entropy (Haralick), and centroid distance. Metrics were computed per cell. Data were simulated from observed means/SDs for export; statistical significance was assessed via unpaired t-test (p < 0.05).

Quantification of EEA1

EEA1 fluorescence was quantified using Fiji (ImageJ). Nuclei were segmented by thresholding the DAPI staining followed by particle analysis, then the ROIs were uniformly enlarged to include the whole cell. These ROIs were applied to the EEA1 channel, and per-cell area, mean intensity, and raw integrated density (RawIntDen) were measured. Background (bg) intensity was obtained from a nearby cell-free region, and background-corrected integrated density was calculated as:

Corrected IntDen = RawIntDen − (Area × Mean_bg). Corrected integrated density values were used as the per-cell measure of EEA1 burden. All images were acquired with identical microscope settings.

Single-cell RNA sequencing procedures and data analysis

Single-cell RNA sequencing (scRNA-Seq) was performed using the standard 10× Genomics platform. Briefly, WT and SORL1 KO iMG were collected, washed, and incubated with TruStain FcX blocking reagent (BioLegend, cat. no. 422301) followed by incubation with cell hashing antibodies (with different barcodes for WT and KO cells) as reported previously [44]. Following cell hashing, cell viability was assessed and found to be greater than 90%. ScRNA-Seq library preparation was conducted as per instructions provided by the 10× Genomics Chromium Next GEM single-cell 3′ Reagent kits v3.1 with Feature Barcode technology. Next generation sequencing (NGS) was performed on a NovaSeq 6000 platform, and the FASTQ files were processed using Cell Ranger with the human transcriptome (GRCh38-2020-A). UMIs of the cell hashing (HTO) library were counted using barcode sequences.

scRNA-Seq data analysis

scRNA-Seq data were analyzed using SCANPY. Gene expression matrices were loaded from 10× Cell Ranger (sc.read_10x_mtx()), and metadata were integrated from a manually curated cluster assignment file. Hashtag unlabeled cells were excluded, and quality control filters were applied to retain high-quality cells. Preprocessing and quality control steps were as follows: cells with fewer than 500 or more than 50,000 counts were removed. Cells with fewer than 200 or more than 8000 detected genes were excluded. Cells with mitochondrial gene content exceeding 25% were discarded. The retained cells were log-normalized (sc.pp.normalize_total(), sc.pp.log1p()), and highly variable genes (HVGs) were selected using sc.pp.highly_variable_genes(). Dimensionality reduction & clustering: Principal Component Analysis (PCA) was performed (sc.pp.pca()) for dimensionality reduction. A neighborhood graph was constructed (sc.pp.neighbors()), followed by clustering with the Leiden algorithm (sc.tl.leiden()). UMAP projections (sc.tl.umap()) were used for visualization.

Differential expression analysis: to ensure balanced group comparisons, the number of cells in WT and KO groups was matched using numpy.random.choice(). Differential gene expression was analyzed using the Wilcoxon rank-sum test (sc.tl.rank_genes_groups()), with Bonferroni correction for multiple testing. All analyses were conducted in Python using SCANPY, NumPy, Pandas, Seaborn, and Matplotlib. Functional annotation was performed using ShinyGO 0.82, a gene-set enrichment tool [45].

Analysis of endogenous SORL1 expression in human microglia subsets

Single-nucleus RNA-seq data from human microglia were obtained from Sun et al. Cell (2023) [46]. After QC and normalization (Scanpy), microglial nuclei were subsetted (n = 152,459). SORL1 expression was extracted from raw counts and log-normalized values. Cells were ranked by SORL1 expression and categorized into SORL1_high and SORL1_low groups using the upper and lower quartiles, respectively. Module scores for XBP1, CHOP, ATF4, ATF6, the composite ER stress module, ISG, lipid-stress, and KO_ERstress signatures were computed using Scanpy’s tl.score_genes() with equal gene weights. Microglial states (MG0–MG12) were taken directly from the Sun et al. annotation. Per-state module differences (SORL1_low − SORL1_high) were computed by calculating mean module scores within each state and subtracting group means. Boxplots and violin plots were generated using matplotlib/seaborn; heatmaps were created using seaborn’s heatmap().

Immunoprecipitation followed by mass spectrometry

Co-Immunoprecipitation (IP) was performed using 30 × 10⁶ THP-1 cells per sample. Cells were harvested, washed, and lysed in IP lysis buffer (30 mM Tris–HCl pH 7.4, 120 mM NaCl, 10% glycerol, 2 mM EDTA, 2 mM KCl, 1% NP-40) supplemented with protease inhibitor cocktail and PMSF. Lysates were subjected to sonication to facilitate solubilization of nuclear envelope-associated proteins, followed by centrifugation to remove insoluble material. The resulting supernatants were used for IP.

To identify SorLA-binding partners, liquid chromatography–tandem mass spectrometry (LC–MS/MS) was used to analyze 200 ng of the sample on a timeTOFPro. Peptides were separated on a reversed-phase C18 column, and mobile phases were adjusted. LC–MS/MS data analysis was done using MaxQuant and Andromeda for database searches. Searches were performed with the reference human proteome database from UniProt, with specified false discovery rates (FDR) and modifications. The output files from MaxQuant were submitted to the Perseus framework [47] for downstream analysis.

MS data processing and statistical analysis

The output files from MaxQuant were processed in Perseus for normalization, filtering, and quantitative analysis. Protein groups were filtered to remove contaminants, reverse hits, and proteins identified only by site. Log-transformed intensity values were normalized, and statistical comparisons between SorLA IP and IgG controls were performed.

SorLA-associated (co-complexed) proteins were classified into two confidence categories based on established significance thresholds. High-confidence SorLA-associated proteins (Class A) were defined by a false discovery rate (FDR) of ≤0.01%, while low-confidence SorLA-associated proteins (Class B) met an FDR threshold of ≤0.05%. These cutoffs are consistent with previously published affinity purification–mass spectrometry (AP–MS) studies that applied similar FDR thresholds to distinguish between high- and low-confidence protein associations detected by AP–MS, and were chosen to balance sensitivity and specificity. All proteins meeting either threshold were considered SorLA-enriched proteins and were included in downstream analysis.

To explore the biological significance of SorLA co-complexed proteins, functional enrichment analysis was performed using ShinyGO 0.82. Gene Ontology (GO) analysis identified enriched biological processes, molecular functions, and cellular components among SorLA-associated proteins. KEGG pathway analysis was used to map proteins onto known signaling and metabolic pathways. Protein association networks were constructed and visualized to highlight key functional clusters among SorLA-associated proteins.

Structural and statistical analysis of SorLA–SUN2 complex

The structural model of the SorLA–SUN2 complex was predicted using AlphaFold [48–50], an advanced deep learning algorithm for protein structure prediction. The generated complex was then processed and refined for visualization using PyMOL, a molecular graphics system. The structural components were recolored, with SorLA and SUN2 distinctly colored to differentiate between the two proteins in the predicted complex. To identify potential contact interfaces, molecular docking-based contact-site analysis was performed using the PRODIGY package [51]. To systematically characterize potential contact sites at the amino acid and domain levels, identified contact residues were mapped onto the protein sequence using a custom Python package designed for high-resolution feature annotation and contact-site profiling. Finally, for visualization, the predicted contact sites were mapped along the amino acid sequence, integrating intensity-weighted annotations to highlight high-confidence interface regions.

Cytokine expression analysis and bulk RNA sequencing

Cytokine expression in WT and SorLA KO iMG was assessed at both the protein (supernatant release) and mRNA levels. For cytokine release, iMG were cultured on poly-L-lysine (PLL)-coated chamber slides in near-homeostatic media and stimulated with 1 nM lipopolysaccharide (LPS) or vehicle-only for 12 h. Following stimulation, supernatants were collected for cytokine quantification. Human cytokine/chemokine levels were measured using a 71-Plex Multiplexing LASER Bead Technology assay (Eve Technologies). Fluorescence intensities were compared before and after LPS addition, and cytokine fold changes were calculated. A two-sided t-test was performed to determine statistical significance for individual cytokines, and results were visualized using a volcano plot.

In parallel, iMG were stimulated with IFN-γ (10 ng/ml) for 48 h, and bulk RNA sequencing (RNA-Seq) was performed to assess mRNA expression changes of key proinflammatory cytokines in SORL1 KO vs. WT iMG. Total RNA was extracted using RNeasy Plus Mini Kit (Qiagen), and library preparation was performed using standard poly(A) selection and cDNA synthesis protocols. Sequencing was conducted on an Illumina platform, generating paired-end reads. Bulk RNA-Seq data processing and differential expression analysis were as follows: Raw sequencing reads were subjected to quality control (QC) using FastQC, followed by adapter trimming. Reads were aligned to the human reference genome (GRCh38) using the STAR aligner, and gene expression was quantified using featureCounts. Downstream differential expression analysis (DEA) was conducted using DESeq2 [52] on iDEP platform [53]. Read counts were normalized using DESeq2’s median ratio normalization (MRN) method to account for differences in sequencing depth. Benjamini–Hochberg (BH) adjustment was applied to control the false discovery rate (FDR), and genes were considered significantly differentially expressed at FDR <0.05 (Supplementary extended data Fig. 4).

Immunoturbidimetric assay

ApoE in cell culture supernatants was measured using the K-ASSAY® ApoE kit (Kamiya Biomedical, Cat. KAI-007), an immunoturbidimetric assay designed for quantitative determination of human ApoE. Samples were analyzed on a Roche Integra 400 plus automated chemistry analyzer, which supports turbidimetric assays and allows reliable quantitation from low-volume biological fluids. Supernatants were centrifuged to remove debris, then loaded according to the manufacturer’s instructions; ApoE concentrations were calculated by comparison to kit calibrators run in parallel, following default Integra software protocols.

Proximity ligation assay on human brain tissue

Proximity ligation assay (PLA) was carried out on archival formalin paraffin-embedded DLPFC human brain tissues to detect proximity between SorLA and SUN2, using the Duolink PLA kit from MilliporeSigma (DUO92101). Briefly, slides were incubated with primary antibodies (SorLA and SUN2) at 4 °C overnight. Positive and negative (IgG) controls were employed to identify the background signals. Subsequently, slides were incubated with Duolink PLA probes (ACD, HybEZ II Oven), followed by ligation and amplification as per manufacturer’s protocol. Post-amplification, slides were blocked with 3% BSA, incubated with anti-IBA1 antibody, washed, and incubated with secondary antibody. TrueBlack Lipofuscin Quencher (Fisher Scientific, cat. no. NC1125051) was applied, followed by washes and mounting with Duolink In Situ Mounting Medium with DAPI. Imaging was done using an Olympus BX3 fluorescence microscope.

CRISPR-mediated gene knockout and gene overexpression

Gene editing in human iPSCs was achieved using a plasmid-free approach, where ribonucleoprotein (RNP) complexes were formed with synthetic single-guide RNA (sgRNA) and recombinant Cas9 protein, and were then delivered via electroporation. No plasmids were used in this method. Post-electroporation, gene editing was confirmed by Sanger sequencing. Edited cells were replated for single-cell cloning, followed by sequencing, and immunoblotting to validate gene knockout (KO), including SORL1.

For THP-1 KO lines, lentivirus was employed. sgRNAs targeting the coding sequence of the gene of interest were cloned into the lentiGuide-Puro plasmid [54] and transduced into THP-1 cells having constant expression of Cas9 in the LentiCas9-Blast vector [54], commercially purchased from Addgene (cat. no. 52962). Editing efficiencies were assessed by Sanger sequencing. Wherever applicable, cells were replated for single-cell clone formation. The absence of SorLA protein in SorLA KO THP-1 clones was verified by immunoblotting.

In all relevant instances, overexpression was achieved using the dCas9-VP64 CRISPR activation system [55]. sgRNAs aimed at the promoter of the target gene were cloned into lentiGuide-Puro plasmid [54] and delivered to target cells that constitutively expressed dCas9-VP64 expression. Gene overexpression was verified post-transduction using immunoblotting or qPCR.

Similarly, where applicable, transcriptional repression (CRISPR interference; CRISPRi) was achieved using tetracycline-inducible dCas9-KRAB [55–57] or constitutively expressed dCas9-KRAB systems, with sgRNAs targeting promoter regions of the corresponding genes [55–57]. Efficient gene repression was confirmed by immunoblotting. All CRISPRa/i relevant DNA sequences are provided in Supplementary Table S2.

Lipid droplet analysis using fluorescent dyes and PLIN2 antibody

Lipid droplet (LD) analysis was performed using BODIPY and Nile Red fluorescent dyes (ThermoFisher, cat. no. D3922 and N1142) and anti-PLIN2 antibody ICC. PLIN2 staining was conducted on primary human microglia (pMG) and differentiated WT and SORL1 KO iMG clones. For iMG (WT and SORL1 KO), cells were plated on 8-well chamber slides and stained with BODIPY (0.5 µM, 37 °C, 5–10 min), washed with PBS, fixed with 4% PFA, blocked with 3% BSA, and incubated with TMEM119 antibody in 1% BSA/PBS. For TMEM119 and PLIN2 ICC, cells were incubated with TMEM119 and PLIN2 antibodies, followed by secondary antibody incubation, PBS washes, and DAPI mounting. For primary human microglia (pMG) (CD11b⁺/CD45⁺/7AAD− sorted from fresh autopsies), cells were plated on 8-well chamber slides, fixed, blocked, and incubated with SorLA and PLIN2 antibodies in 1% BSA/PBS, followed by secondary antibody incubation, PBS washes, and DAPI mounting. All images were captured using an Olympus BX3 fluorescent microscope and analyzed with CellProfiler. Statistical analysis was performed in Prism software.

For lipid staining of THP-1 cells using BODIPY or Nile Red, 200,000 cells per replicate (with a minimum of three replicates) were incubated in PBS containing BODIPY (0.5 μM) or Nile Red (0.3 μM) at 37 °C for 5 min. Following incubation, cells were washed three times with PBS and analyzed by flow cytometry and FlowJo software.

Details of all used antibodies are provided in Supplementary Table S2.

For lipid droplet (LD) quantification and morphometry, LDs were visualized by BODIPY staining as described above and imaged on a fluorescence microscope using identical exposure settings for all conditions. For each experiment, ≥20 randomly selected cells per genotype were acquired by an investigator blinded to the experimental condition. LDs were segmented in Fiji/ImageJ using an intensity threshold followed by the “Analyze Particles” function. For each cell, the number of LDs was recorded and used to calculate LDs per cell. For Supplementary Fig. S5a,b, individual LDs were analyzed for projected area (pixels) and circularity (4πA/P²). LD area values from all cells were pooled to generate size-frequency histograms (S5a), and circularity values were compared between genotypes (S5b).

Lipidomics

Lipidomics analysis was conducted on three replicates of WT and SORL1 KO THP-1 cells using Ultra Performance Liquid Chromatography–Tandem Mass Spectrometry (UPLC-MSMS) as described previously [58]. Briefly, lipid extracts prepared from cell lysates, spiked with internal standards, were analyzed on an Agilent 1260 Infinity HPLC integrated with an Agilent 6490A QQQ mass spectrometer. Glycerophospholipids and sphingolipids were separated with normal-phase HPLC, and sterols and glycerolipids were separated with reverse-phase HPLC. Quantification of lipid species was done using multiple reaction monitoring (MRM) transitions under both positive and negative ionization modes, referencing appropriate internal standards. Lipid levels for each sample were calculated by summing the total number of moles of all lipid species measured, normalizing to mol %, and presented as mean mol %. For graphical illustration, replicates of WT and SORL1 KO were compared using a two-tailed t-test for each lipid species and subspecies concentration to calculate p values. −log10(p value) was plotted against log2 of fold change using matplotlib.

Filipin staining and flow cytometry

THP-1 cells were harvested, washed with PBS, and fixed in 4% paraformaldehyde for 15 min at room temperature. After two PBS washes, cells were incubated with filipin (free cholesterol probe MilliporeSigma cat. no. SAE0087-1ML) diluted in PBS containing 10% FBS for 30 min in the dark. Cells were then washed, resuspended in PBS, and analyzed immediately on a flow cytometer. Singlet, viable cells were gated, and the percentage of filipin-positive cells was determined using a fluorescence gate defined from non-targeting control samples. Data from three independent experiments were used for quantification in Fig. 5k,l.

Fig. 5.

SorLA deficiency leads to accumulation of lipid droplets in microglia. a Microscopic images showing stronger BODIPY staining in SORL1 KO iMG compared to WT, scale bars: 10 μm, TMEM119: microglial marker. b Single-cell quantification of BODIPY fluorescence intensity from three independent experiments (number of cells: WT = 383, 394, 379; SORL1 KO = 312, 312, 314 for R1–R3). Boxplots show pooled distributions with replicate identity preserved. Mann–Whitney tests revealed highly significant differences in all replicates (R1: U = 372.5; R2: U = 165.5; R3: U = 2157; all p < 1 × 10⁻¹⁵). b′ Mean BODIPY intensity per differentiation replicate. Unpaired t-test: t(4) = 3.73, p = 0.020; Δ = 0.00230 ± 0.00062 SEM; η² = 0.78. c Quantification of lipid droplets (LDs) per cell derived from BODIPY images. For each differentiation, ≥20 randomly selected cells per genotype were quantified by an investigator blinded to condition. Across experiments, SORL1 KO cells contained significantly more LDs per cell than WT (Welch’s t-test, p = 0.009). d Representative ICC images showing stronger PLIN2 immunoreactivity, in SORL1 KO iMG compared to WT (scale bars: 10 μm). e Single-cell PLIN2 fluorescence intensity quantified across three independent differentiations. Mann–Whitney tests showed strong significance for all replicates (R1–R2: p < 1 × 10⁻¹⁵; R3: p = 1.78 × 110⁻¹⁰). f Mean PLIN2 intensity per replicate (n = 3 per genotype), confirming increased PLIN2 abundance in SORL1 KO iMG (unpaired t-test: t(4) = 3.19, p = 0.033; Δ = 0.00267 ± 0.00084 SEM; η² = 0.72). g Immunoblot analysis of SorLA and PLIN2 in WT, SORL1 KO (S-KO), and SORL1-overexpressing (S-OE) iMG (CRISPRa) (n = 3). PLIN2 levels were elevated in SORL1 KO cells and reduced upon SORL1 overexpression, confirming a regulatory role for SorLA in LD accumulation. h Dual ICC in primary human microglia (pMG) freshly purified from autopsy DLPFC shows higher PLIN2 levels in cells with lower SorLA protein expression (scale bars: 10 μm). i shows quantification of (h). SorLA and PLIN2 protein levels are negatively correlated (Pearson r = −0.48, p = 6.3 × 10⁻⁴⁵; Spearman r = −0.41, p = 5.5 × 10⁻³²; n = 751 pMG from three DLPFC donors). j Volcano plot of lipidomic profiling from control and SORL1 KO THP-1 cells (n = 3), highlighting increased cholesteryl esters (CEs) in SORL1 KO cells. Y-axis: −log₁₀(p); X-axis: log₂ fold-change (KO/WT). k Flow cytometry contour plots showing filipin-positive THP-1 cells after CRISPR targeting with non-targeting or SORL1-targeting sgRNAs. Filipin stains free (unesterified) cholesterol. l Quantification from three experiments showing reduced filipin-positive cells in SORL1-targeted THP-1 cells (Welch’s t-test: sgRNA1 p = 0.0001, Δ = −20.53 ± 1.33 SEM; sgRNA2 p = 0.0006, Δ = −24.77 ± 1.74 SEM). m scRNA-seq data showing increased expression of CYP27A1, SOAT1, and CH25H, and decreased SQLE, in SORL1 KO iMG. Adjusted p values are shown above each panel. n Increased BODIPY staining upon SORL1 repression using doxycycline-inducible CRISPRi (unpaired t-test, p = 0.001; Δ = 9.67 ± 1.26 SEM). After doxycycline withdrawal for three days, BODIPY levels remained elevated (+D/−D), suggesting longer recovery may be required to restore lipid homeostasis

Extraction of synaptosome from human dorsolateral prefrontal cortex

A synaptosome fraction from human dorsolateral prefrontal cortex (DLPFC) was purified using a previously described protocol [59]. Briefly, the DLPFC tissue was minced, transferred to Syn-Per buffer (ThermoFisher Scientific, cat. no. 87793) with a protease inhibitor, homogenized, and centrifuged. The synaptosome-enriched pellet was resuspended, quantified using the BCA assay, and conjugated with pHrodo™ red dye (ThermoFisher Scientific, cat. no. P36014). The synaptosome enrichment was confirmed using immunoblotting with an anti-SNAP-25 antibody.

Phagocytosis assays

Phagocytosis assays in iPSC-derived microglia (iMG) and THP-1 cells were performed using pHrodo™ Red-labeled synaptosomes and fluorescently labeled amyloid-β42 (Aβ42; AnaSpec, cat. no. AS-64161). For Aβ assays, oligomeric Aβ42 was prepared according to the manufacturer’s instructions and used at a final concentration of 1 μM. Cells were incubated for 6 h at 37 °C unless otherwise indicated. Cytochalasin D was used as a control to inhibit actin-dependent phagocytosis and confirm uptake specificity. Phagocytosis of synaptosomes and Aβ42 by WT and SORL1 KO iMG was assessed by ICC and fluorescence microscopy. Following incubation, cells were washed, fixed, and immunostained with IBA1 to label microglia. Fluorescent signals were acquired using a widefield fluorescence microscope, and quantitative image analysis was performed using CellProfiler, with statistical analyses conducted in GraphPad Prism.

For flow cytometry-based phagocytosis assays, THP-1 cells were incubated with fluorescently labeled oligomeric Aβ42 (1 μM, 6 h), washed extensively to remove non-internalized peptide, and analyzed on a flow cytometer using the appropriate fluorescence channel for Aβ detection. For phagocytosis and recovery assays, cells were treated with fluorescently labeled oligomeric Aβ42, washed, and divided into two fractions. One fraction was analyzed immediately after washing, while the second fraction was cultured in Aβ-free medium for a 96-h recovery period prior to analysis to assess persistence or clearance of internalized material.

Western blotting

Western blotting (immunoblotting) was performed using a standard protocol. Briefly, to perform immunoblotting on postmortem human brain tissues, frozen DLPFC tissue was minced and transferred to ice-cold RIPA buffer (Cell Signaling Technology, cat. no. 9806S) with protease inhibitor cocktail (MilliporeSigma, cat. no. P8340) and phenylmethanesulfonylfluoride (PMSF) from MilliporeSigma (cat. no. 93482). Post homogenization, centrifugation, and sonication, the supernatants were collected, and total protein concentrations were determined using the BCA protein quantification method. Similarly, in vitro cultured cells were lysed in ice-cold RIPA buffer. Lysates were heated at 95 °C in Laemmli sample buffer supplemented with Cleland’s reagent for five minutes, and proteins were separated using standard electrophoretic procedures. Proteins were transferred to a PVDF membrane using a semi-wet transfer system. Proteins on the membrane were blocked with Intercept (TBS) Blocking buffer (LICOR, cat. no. 927-60001), incubated with primary and secondary antibodies, and imaged using the Odyssey LICOR machine and analyzed quantitatively using ImageJ packages. The details of the antibodies used are provided in Supplementary Table S2.

For uncropped Western blot images corresponding to the main and supplementary figures, see Supplementary extended data Fig. 10.

Flow cytometry

Standard flow cytometry procedures were employed for the analysis of gene expression and phagocytosis. Simply, cells were collected, washed with PBS, fixed with glutaraldehyde, and permeabilized with Triton X-100. Cells were then incubated with conjugated antibodies in BSA/PBS, washed, and analyzed using a BD analyzer at the Columbia University Core facility. The data were analyzed with the FlowJo package. Antibody details are provided in Supplementary Table S2.

Cell treatments

THP-1 cells were cultured under standard conditions and treated with tauroursodeoxycholic acid (TUDCA), 100 µM, 72 h (Selleck Chemicals, cat. no. S3654), Brefeldin A, 1 µg/mL, 72 h (eBioscience, cat. no. 00-4506-51), or Ruxolitinib, 1 µM, 72 h (InvivoGen, cat. no. tlrl-rux) or Bafilomycin A1, 1 nM, 24 h (Cell Signaling Technology cat. 54645S) or Phorbol 12-Myristate 13-Acetate (PMA), 80 ng/ml, 24 h. Following treatment, cells were collected for biochemical assays and LD accumulation analysis.

Quantitative real-time PCR

The quantitative real-time PCR (qPCR) was used to quantify mRNA levels in iMG and THP-1 cells. mRNA was extracted from cells using QIAGEN RNeasy Plus Mini kit (cat. no. 74104) according to the manufacturer’s instructions. Genomic DNA was removed using a gDNA Clean and Concentrate kit column. RNA was then purified using an RNase-free purification column. The purified RNA was used for cDNA synthesis using iScript™ cDNA synthesis kit from Bio-Rad (cat. no. 1708891). The cDNA was quantified with a Qubit fluorometer (ThermoFisher) and then subjected to qPCR analysis using Quant Studio 4.0 system from Applied Biosystem.

Correlation analysis

To assess relationships between protein expression levels in the human brain, correlation analysis between SorLA and SUN2 was performed using published large-scale proteomic datasets from the human DLPFC and orbitofrontal cortex. For the DLPFC analysis, we reanalyzed data generated by Johnson et al. [60] using the authors’ processed and publicly available file “Minimally_regressed_Batch_and_Site-corrected_LFQ_intensity.csv”, which contains label-free quantification (LFQ) intensities that were already normalized and corrected for batch and site effects. Protein intensities for SorLA and SUN2 were extracted directly from this dataset, and samples were grouped according to the diagnostic categories and Braak stages assigned in the original study. No additional normalization or batch correction was applied.

In parallel, correlations between SorLA and PLIN2 were assessed using protein expression measurements obtained from freshly purified primary human microglia isolated from rapid autopsies, analyzed across experimental conditions as described above. For all correlation analyses, both Pearson’s correlation coefficient (r), which assesses linear relationships under assumptions of normality, and Spearman’s rank correlation coefficient, a non-parametric measure robust to non-linearity and outliers, were computed. Corresponding p values were reported to assess statistical significance. In addition, ordinary least squares (OLS) linear regression was performed, and 95% confidence intervals for the regression lines were calculated. Data were visualized using scatter plots with fitted regression lines and confidence bands.

Statistical analysis

Statistical analysis was conducted using GraphPad Prism, Perseus, and SCANPY, applying appropriate tests based on data distribution and variance. Normality was assessed using the Shapiro–Wilk test, and homogeneity of variances was evaluated using Bartlett’s test. Depending on these assessments, parametric tests (e.g., one-sample t-test, two-sample t-test, ANOVA with post hoc corrections for multiple comparisons) were used for normally distributed data, while non-parametric alternatives (e.g., Mann–Whitney U test, Kruskal–Wallis test with Dunn’s post hoc correction) were applied for non-Gaussian data. For scRNA-Seq analysis, data were processed using SCANPY, and statistical testing for differential expression was performed using Wilcoxon rank-sum tests (sc.tl.rank_genes_groups()), with Benjamini–Hochberg correction for multiple comparisons. Clustering was performed using Leiden community detection, and dimensionality reduction was achieved through principal component analysis (PCA) and Uniform Manifold Approximation and Projection (UMAP). Permutation-based FDR methods were used for mass spectrometry-derived protein co-complex analysis in Perseus. The specific statistical tests applied for each dataset, along with effect sizes and multiple testing corrections (where applicable), are provided in the relevant figure legends and Results sections. Statistical significance was set at p < 0.05, unless otherwise specified.

Results

Microglia and neurons express SorLA protein in human brain

Although SORL1 transcripts are highly enriched in microglial transcriptomic datasets [20], their protein expression has primarily been characterized in neurons [61–63]. Given our focus on microglia, we first examined SorLA protein expression in microglia and then compared it across other major brain cell types. Immunohistochemistry (IHC) performed on paraffin-embedded sections from the dorsolateral prefrontal cortex (DLPFC) using a knockout-verified monoclonal anti-SorLA antibody (validated for ICC and immunoblotting [64]) revealed strong SorLA immunoreactivity in IBA1⁺ microglia (Fig. 1a,a’). To further validate this finding and ensure that the observed signal in IHC is not a tissue artifact, we examined purified microglia isolated from human brain autopsies, which again displayed robust SorLA staining (Fig. 1b). Neurons in the DLPFC, identified by NeuN staining, also showed prominent SorLA labeling (Fig. 1c), whereas GFAP⁺ astrocytes and CNPase⁺ oligodendrocytes (OLs) exhibited weak to undetectable signal (Fig. S1a,b, see quantification in Fig. 1d).

Fig. 1.

SorLA protein is expressed by microglia in the aged human brain and reduced in Alzheimer’s disease. a–c Immunofluorescence (IF) detection of SorLA protein in human dorsolateral prefrontal cortex (DLPFC). a Low-magnification view showing SorLA colocalization with IBA1⁺ microglia. a′ High-magnification image illustrating punctate SorLA immunoreactivity within IBA1⁺ microglia. b Immunocytochemistry (ICC) detection of SorLA in acutely isolated microglia from human DLPFC autopsy material. c SorLA localization in neurons (NeuN+). Scale bars in a: 50 μm, a’, b and c: 10 μm. d. Violin plots showing SorLA fluorescence intensity (arbitrary units, a.u.) across major brain cell types (Microglia (M): IBA1⁺, Neurons (N): NeuN⁺, Astrocytes (A): GFAP⁺, Oligodendrocytes (OL): CNPase⁺). ⁺) as assessed by IF and automated image analysis. Representative photomicrographs for A and OL can be found in Fig. S1a,b. Kruskal–Wallis test revealed significant differences among groups (H = 10,029.68, p < 1 × 10⁻¹⁵). Dunn’s post hoc test (Bonferroni corrected): N vs M (Z = 14.99, p < 1 × 10⁻¹⁵), A vs N (Z = 8.98, p < 1 × 10⁻¹⁵), OL vs N (Z = 36.43, p < 1 × 10⁻¹⁵), M vs A (Z = 3.9, p = 5.7 × 10⁻⁴), OL vs M (Z = 8.96, p < 1 × 10⁻¹⁵); OL vs A not significant (p > 0.99). e Representative IF images of SorLA expression in IBA1⁺ microglia in DLPFC from non-AD and AD donors (scale bars: 25 μm). e’ Dot plot quantification showing significantly reduced SorLA fluorescence intensity in AD microglia (t = 2.26, df = 13, p = 0.04, η² = 0.28). f Dot plot quantification showing reduced SorLA intensity in NeuN⁺ neurons in AD (t = 2.32, df = 8, p = 0.04, η² = 0.40). Each dot represents mean SorLA fluorescence for an individual, calculated from ≥150 microglia and ≥ 1500 neurons per subject. g–g’ Immunoblot analysis confirms lower SorLA levels in AD vs. non-AD DLPFC tissue. g Western blot of DLPFC lysates from AD and non-AD donors probed with antibodies against SorLA and GAPDH, a loading control. g’ Quantification of g (non-AD = 0.1498 ± 0.014, AD = 0.0356 ± 0.014; two-tailed t-test, p = 0.0002, t = 7.93, df = 6). Abbreviations: AD, Alzheimer’s disease; A, astrocytes; C, control; M, microglia; N, neurons; OL, oligodendrocytes

To quantitatively compare SorLA expression levels across these cell types, we measured mean fluorescence intensity per cell. Tests for data normality (Shapiro–Wilk) and variance homogeneity (Bartlett’s test) indicated non-normal distributions and unequal variances, prompting the use of a nonparametric approach. The Kruskal–Wallis test revealed a highly significant difference among groups (H = 10,029.68, p < 1 × 10⁻¹⁵). Dunn’s post hoc analysis (Bonferroni corrected) showed that neurons expressed the highest SorLA levels, followed by microglia, both of which were significantly higher than astrocytes and OLs (p < 1 × 10⁻¹⁵ for all comparisons). Importantly, microglia exhibited significantly greater SorLA expression than astrocytes (p = 5.7 × 10⁻⁴) and OLs (p < 1 × 10⁻¹⁵), while astrocyte and OL levels did not differ (p > 0.99; Fig. 1d).

SORL1 protein expression declines in Alzheimer’s disease

Given the strong genetic association between SORL1 and AD, we next examined whether SorLA protein expression is altered in AD brain tissue. Using IHC on DLPFC sections from late-onset AD (LOAD) patients and non-AD controls, (microglia: non-AD n = 6, AD n = 9; neurons: non-AD n = 5, AD n = 5; see Supplementary Table S1), we assessed SorLA expression in microglia and neurons. In microglia, SorLA mean fluorescence intensity was significantly lower in AD (0.00189 ± 0.00039) compared with controls (0.00248 ± 0.00063 a.u.; Fig. 1e,e’). Both groups passed normality (Shapiro–Wilk p = 0.45) and variance (F-test p = 0.21) checks; an unpaired two-tailed t-test confirmed a significant reduction (t = 2.26, df = 13, p = 0.04, η² = 0.28). A similar decrease was observed in neurons, where SorLA levels were reduced in AD (0.06682 ± 0.00291 vs. 0.07024 ± 0.00156; Fig. 1f). Data satisfied normality (p = 0.051) and variance (p = 0.25) assumptions, and an unpaired t-test indicated a significant reduction (t = 2.32, df = 8, p = 0.04, η² = 0.40). These results were corroborated by immunoblot analysis of DLPFC lysates, which also demonstrated lower SorLA protein levels in AD relative to controls (Fig. 1g,g’). This decrease in SorLA protein is consistent with prior transcriptomic and histological studies that reported reduced SORL1 mRNA expression in late-onset AD, particularly in neuronal populations.

Together, these findings reveal that SorLA is prominently expressed in microglia and neurons, and it is significantly reduced in AD in both cell populations, reinforcing its potential involvement in disease-associated disruptions of cellular homeostasis and highlighting the need for further mechanistic investigation.

CRISPR-Cas9-mediated generation of SorLA-deficient human microglia model system

To generate a model for studying the consequences of SorLA deficiency in human microglia, we used a non-plasmid CRISPR-Cas9 approach to create a SORL1 knockout (KO) single-cell clone from an APOE3/3 iPSC line. This line was previously validated [33] for differentiation into SORL1-expressing microglia-like cells (iMG) (Fig. S1c). Of note, a comprehensive characterization of iMG differentiated from this and other iPSC lines using our protocol [33] is accessible via an online platform associated with our previous publication [33]. This includes detailed comparisons to primary human microglia in terms of morphology, chromatin accessibility (ATAC-Seq), gene expression, proteomics, and immune response.

To generate SORL1 knockout (KO) iPSC clones, preassembled Cas9-sgRNA ribonucleoprotein (RNP) complexes were delivered into iPSCs via electroporation. ICC revealed complete loss of SorLA immunoreactivity in SORL1 KO iMG compared with control cells (Fig. S1c,d). This was further validated by Sanger sequencing, which confirmed CRISPR-Cas9-induced frameshift mutations in the SORL1 gene (Fig. S1e). Immunoblotting further verified the absence of SorLA protein in the KO clone, while control iMG expressed normal levels (Fig. S1f). Importantly, SORL1 KO iPSCs retained their ability to differentiate into iMG (Fig. S1g) and maintained a stable karyotype (Fig. S1h).

For experiments requiring large cell numbers, we generated independent SORL1 knockout (KO) clones from the THP-1 cell line using the same CRISPR-Cas9 strategy. The THP-1 cell line, derived from human peripheral blood and classified as a monocytic leukemia line, serves as a valuable research tool. Although not a perfect surrogate for microglia, THP-1 cells partially recapitulate key microglial characteristics in vitro, making them suitable for studying macrophage and microglial functions [65], [66], [67] under high-throughput experimental conditions. To assess their transcriptional relevance, we compared the global gene expression profiles of THP-1 cells with primary human microglia (pMG) [43] (Fig. S1i) and iMG [33] (Fig. S1i’), which revealed a strong positive correlation (pMG vs. THP-1: Spearman’s R = 0.81, p = 0.00; Kendall’s τ = 0.63, p = 0.00, iMG vs. THP-1: Spearman’s R = 0.75, p = 0.00; Kendall’s τ = 0.57, p = 0.00), supporting their suitability for downstream assays. Sanger sequencing confirmed frameshift indels in the SORL1 coding region (Fig. S1j), and immunoblotting verified the absence of SorLA protein in these clones (Fig. S1k).

Collectively, these results establish robust SORL1 KO microglial model systems, providing a foundation for downstream functional and mechanistic analyses.

SorLA deficiency impairs microglial phagocytosis and cargo clearance

Given SorLA’s established role in endolysosomal trafficking [68], we next assessed whether SORL1 loss compromises microglial phagocytic function. We compared the uptake and clearance of substrates in SORL1 KO and control iMG. Using fluorescently labeled amyloid beta 42 (Aβ42) as a substrate, microscopic analysis revealed markedly reduced phagocytic activity in SORL1 KO iMG compared to WT cells (Figs. 2a and S2a). Quantitative analysis of Aβ42 signal intensity confirmed a significant reduction in uptake in SORL1 KO iMG (Mann–Whitney U test, U = 8774, p = 0.02, n = 158 (WT), n = 132 (KO)) (Fig. 2b). Similarly, when challenged with pHrodo™ Red-labeled human synaptosomes purified from aged DLPFC tissue, SORL1 KO iMG exhibited a pronounced decrease in phagocytic activity relative to controls (Mann–Whitney U test, U = 2426, p = 8.2 × 11⁻¹³, n = 81 (WT), n = 136 (KO)) (Figs. 2c,d and S2b). The uptake of both of these substrates could be blocked by Cytochalasin D, confirming the uptake mechanism to be phagocytosis (Fig. S2a,b). To validate these findings using a higher-throughput approach, we assessed Aβ42 phagocytosis in SORL1 KO THP-1 clones by flow cytometry. All SORL1-deficient clones displayed significantly reduced phagocytic activity (Clone 1: Welch’s t-test p = 0.0002; Clone 2: p = 0.0003; Clone 3: Mann–Whitney test p = 0.0022; each analyzed ≥6 times) (Fig. 2e). Moreover, when Aβ42-treated SORL1 KO and control THP-1 cells were washed and cultured for 96 h in the absence of substrate, SORL1 KO cells retained significantly higher residual fluorescence (t-test p = 0.0002, t = 9.438, df = 5) (Fig. 2f), indicating impaired degradation and clearance of phagocytosed cargo. Collectively, these findings confirm that SorLA deficiency compromises phagocytosis and lysosomal cargo processing, validating the functional phenotype of SORL1 KO iMG and motivating subsequent in-depth molecular and cellular investigations.

Fig. 2.

Functional assessment of SorLA-deficient microglia shows impairment in both phagocytosis and cargo clearance. a Representative fluorescent images showing uptake of fluorescently labeled Aβ42 by WT and SORL1 KO iMG. Individual channels display TMEM119 (microglial marker), Aβ42 (Fluor 647), along with the merged overlay with DAPI. b Violin plot quantifying Aβ42 uptake in WT and SORL1 KO iMG (F-test p < 0.05, Mann–Whitney U = 8774, p = 0.02, n = 158 (WT), n = 132 (KO), pooled from three independent differentiations). c Representative images showing phagocytosis of human synaptosomes purified from aged DLPFC tissue and labeled with pHrodo™ Red. Individual panels display TMEM119, pHrodo Red, and their merged overlays with DAPI. d Violin plot quantifying synaptosome uptake (Mann–Whitney U = 2426, p = 8.2 × 10⁻¹³, n = 81 (WT), n = 136 (KO)). Cytochalasin D treatment was used to confirm specificity of phagocytosis for both Aβ42 and synaptosomes (see Fig. S2a,b). e Dot plot of flow cytometry data comparing Aβ42 phagocytosis in WT and SORL1 KO THP-1 cells, showing a significantly lower percentage of Aβ42-positive cells in SORL1 KO clones (mean ± SEM: Control = 44.4 ± 0.97, Clone1 = 19.1 ± 3.0, Clone2 = 20.1 ± 3.0, Clone3 = 26.3 ± 4.4). Each clone was assessed in at least six independent experiments performed on different days. Clone 1: Welch’s t-test p = 0.0002; Clone 2: p = 0.0003; Clone 3: Mann–Whitney test p = 0.002. f Fold-change analysis (flow cytometry) of Aβ42 fluorescence signal retention in matched control (WT is designated on x-axis as SORL1+) and SORL1 KO THP-1 clones 96 h after treatment. Each point represents an independent experiment (n = 3 per group), with horizontal bars showing mean ± SEM. KO clones displayed significantly higher cargo retention compared to WT (shown p values are from unpaired t-test). Cells were washed after Aβ42 exposure and cultured in fresh media, and values were normalized to WT control. Data in a–d are from iMG differentiated from iPSCs (Gibco A18945), and in e–f from THP-1 cells

Single-cell transcriptomics revealed divergent transcriptional programs in SORL1-deficient iMG

Following the generation and characterization of SORL1 KO iMG derived from the Gibco episomal iPSC line (A18945), we performed single-cell RNA-Seq analysis using the 10× Genomics platform. This iPSC line was selected because iMG differentiated from it have been extensively characterized and show robust, reproducible microglial differentiation [33]. After quality control (Supplementary extended data Fig. 3.1) we retained 4000 iMG consisting of 2000 control (WT) and 2000 SORL1 KO cells for further analysis. Canonical correlation analysis and dimensionality reduction were applied to perform unsupervised clustering, identifying 0–7 clusters corresponding to distinct microglial phenotypes based on marker gene expression. UMAP representation was then used to visualize these clusters (Fig. 3a,b). Cluster 0, the largest, was characterized by immune-related genes, such as FCGR2A, TGFBI, and PLAUR, indicating a role in immune signaling [69–72]. Cluster 1 displayed markers linked to lipid metabolism (DBI, PLIN2, APOC1, LIPA, CYP27A1) [73–76] and ER stress responses (TXN, FABP5, LGALS3, UCHL1, NUPR1, SGK1) [77–80] reflecting lipid-stressed microglia. Cluster 2 featured motility-associated genes (DAB2, FSCN1, IGFBP4) [81–84], indicative of migratory microglia. Cluster 3 was marked by proliferative markers (MKI67, HMGB1) [85, 86], while Cluster 4 expressed genes involved in DNA repair and cell cycle regulation (PCNA, RAD51AP1, ATAD2) [87, 88]. These profiles suggest that Clusters 3 and 4 represent proliferative subsets at different cell cycle phases. Cluster 5 was identified as metabolically hyperactive microglia due to its elevated expression of the late endosome/lysosome marker CD63 [43, 89], and an immune-related gene IGSF6 [90], reflecting heightened lysosomal function and immune activity. Cluster 6 was defined by antigen presentation markers (CD74, HLA-DRB1, HLA-DRA, CTSK, HLA-DPA1), indicating a role in antigen presentation. Lastly, Cluster 7 (smallest cluster) comprised transitioning microglia characterized by upregulation of stress and activation-associated genes (RPL26, S100B, S100A9, DBI) (Fig. 3b; Supplementary Table S3).

Fig. 3.

Single-cell RNA-Seq reveals transcriptional alterations in SORL1 KO iMG. a Principal Component Analysis (PCA) of WT and SORL1 KO iMG, showing transcriptional separation along PC1 and PC2, indicating distinct molecular states. b Uniform Manifold Approximation and Projection (UMAP) representation of unsupervised clustering reveals transcriptional subpopulations in WT and SORL1 KO iMG. Clusters are annotated based on dominant functional signatures, as indicated on the UMAP. c,d SORL1 depletion alters microglia subset distribution. c Stacked bar plot analysis shows expansion of the lipid-stressed cluster (Cluster 1; 12.6% to 27.7%) and contraction of the metabolically hyperactive cluster (Cluster 5; 15.3% to 0.7%) in SORL1 KO iMG. d UMAP visualization further confirms these shifts, showing a redistribution of cellular states upon SORL1 loss. e Single-cell differential expression analysis between WT and SORL1 KO iMG identified significantly altered genes across all cells (adjusted p ≤ 0.05; dashed lines indicate 1.5-fold change, significance threshold). f Violin plots display expression distribution of top upregulated genes in SORL1 KO iMG, with corresponding p values shown above. g Gene enrichment analysis of DEGs reveals that upregulated genes predominantly cluster within type I interferon response signaling pathways. h Focused volcano plot highlights selected upregulated genes from Cluster 1 (lipid-stressed microglia), involved in lipid metabolism (e.g., APOC1, LIPA, SGK1, TSPO, DBI) and ER stress (e.g., TXN, FABP5, LGALS3, UCHL1, NUPR1, SGK1, DBI). These genes are also shown in panel (e) but displayed here for clarity. i Volcano plot further highlights upregulation of classic ER stress markers, including DDIT3 (CHOP), HSPA5 (BiP), and OLR1, which are also among the DEGs in panel (e) but not clearly visible there due to scale. j Violin plots illustrate the expression distribution of top downregulated genes in SORL1 KO iMG, with p values indicated. k GO analysis of downregulated genes in SORL1 KO iMG indicates enrichment in acute inflammatory response, cell motility, and protein localization pathways, highlighting functional disruptions linked to SORL1 depletion. scRNA-Seq data shown are from iMG derived from Gibco cat. A18945 iPSC line

Next, we compared clusters between WT and SORL1 KO iMG (Figs. 3c,d, S3a,b) and observed significant differences. The relative abundance of Cluster 1, associated with lipid storage/metabolism and ER stress, expanded to 27.7% in SORL1 KO compared to 12.6% in WT (chi-squared statistic: 140.77, p value: 1.80e-32). Conversely, the relative abundance of cluster 5, reflecting metabolically hyperactive microglia, shrank to 0.7% in SORL1 KO compared to 15.3% in WT (chi-squared statistic: 287.63 p value: 1.62e-64) (Figs.c,d and Supplementary extended data Fig. 3.2, comparing WT (left) and KO (right)). Cluster 2, enriched in microglia motility-related genes, was also reduced from 20.5% to 15.5% in SORL1 KO iMG (chi-squared statistic: 16.60 p value: 4.61e-05). Interestingly, KEGG pathway analysis was performed specifically on Cluster 1, as it exhibited the most pronounced compositional shift between WT and SORL1 KO iMG. Analysis of the top 100 upregulated genes in this cluster revealed cholesterol metabolism and riboflavin metabolism as the most enriched pathways. Gene Ontology (GO) analysis further identified biological processes related to the regulation of lipid and protein transport and localization (Fig. S3c,d). This suggests a significant connection between SORL1 and metabolic processes, especially those of cholesterol and riboflavin.

To further investigate how SORL1 deficiency impacts microglial gene expression, we performed differential gene expression analysis at single-cell resolution, comparing all SORL1 KO and control iMG using Scanpy’s rank_genes_groups function. This analysis identified 218 significantly upregulated and 115 downregulated genes in SORL1 KO iMG (fold change ≥1.5; adjusted p value ≤0.05) (Fig. 3e and Supplementary Table S4). Gene enrichment analysis of the differentially expressed genes (DEGs) highlighted that the upregulated genes, including IFI6, ISG15, MX1, LY6E, IFITM3, OAS1 (Fig. 3f), were predominantly congregated in pathways related to type 1 interferon response signaling (Fig. 3g). Notably, in a previous scRNA-Seq study [43], these genes were identified as markers of a type I interferon response cluster in aged human DLPFC microglia (Fig. S3e), suggesting that the transcriptional changes observed in SORL1 KO microglia may reflect a type I interferon response signature similar to that seen in aged human microglia. In the present dataset, this signature may not have emerged as a distinct cluster because it likely represents a global shift across all SORL1 KO microglia, rather than being confined to a specific subpopulation. Additionally, a majority of the feature genes from Cluster 1, which are involved in lipid processing (e.g., APOC1, LIPA, SGK1, TSPO, DBI) and ER stress (e.g., TXN, FABP5, LGALS3, UCHL1, NUPR1, SGK1, DBI) were also significantly upregulated in the SORL1 KO iMG (Fig. 3h). Classic markers of ER stress, such as DDIT3 (CHOP), HSPA5 (BiP), and OLR1, were also found to be significantly upregulated (Fig. 3i). Conversely, differential gene expression analysis showed that a substantial proportion of downregulated genes in SORL1 KO iMG were associated with acute inflammatory responses, cell motility, and protein localization (Fig. 3j,k).

Together these results indicate that SORL1 expression depletion globally leads to a robust transcriptional activation of genes associated with interferon response, ER stress response and lipid processing. At the same time there is notable downregulation across all cells of genes linked to inflammatory response, cell motility and protein localization.

SorLA deficiency triggers ER stress and proinflammatory response in microglia

scRNA-seq data revealed increased expression of genes associated with ER stress, lipid processing, and proinflammatory interferon responses. Because SorLA regulates secretory and endosomal–Golgi trafficking, we reasoned that its loss disrupts proteostasis, thereby initiating ER stress and downstream inflammatory signaling [91]. To quantify this more rigorously, we computed untranslated protein response (UPR) transcription factor module scores for ATF4, ATF6, and CHOP in our scRNA-seq dataset. ATF6 and CHOP module scores were significantly elevated in SORL1 KO microglia compared with WT, while ATF4 showed a similar upward trend (p = 0.06) (Fig. 4a). Importantly, qPCR independently confirmed the upregulation of ER stress markers EIF2AK3 (PERK), HSPA5 (BiP) and ERN1 (IRE1a) [92, 93] in SORL1 KO iMG (Fig. S4a). To assess ER stress at the protein level, we performed immunoblotting on WT and SORL1 KO THP-1 cell lysates. Immunoblotting revealed a significant upregulation of PERK (37%), BiP (58%), and IRE1a (123%) in SORL1 KO THP-1 clones compared to WT cells (Fig. 4b,c), implicating SorLA in the regulation of ER stress. Because CRISPR-mediated knockouts can produce truncated or aberrant SorLA fragments that might independently trigger ER stress, we also repressed SORL1 expression using a CRISPR interference (CRISPRi) system, which reduces transcription without generating truncated protein products. Immunoblotting of CRISPRi-targeted THP-1 cells showed increased IRE1a protein levels relative to non-targeting (NT) controls (Fig. S4b,b’), supporting that ER stress induction reflects loss of SorLA function rather than artifacts from truncated SORL1 protein.

Fig. 4.

SorLA loss drives ER stress and inflammatory activation in iMG. a Boxplots showing UPR (Untranslated Protein Response) transcription factor module scores (ATF4, ATF6, CHOP) derived from scRNA-Seq data in WT and SORL1 KO iMG. ATF6 and CHOP modules were significantly elevated in SORL1 KO cells, whereas ATF4 showed a nonsignificant upward trend (Mann–Whitney test p values are shown above each plot. b Immunoblot showing increased levels of ER stress markers PERK, BiP, and IRE1a in SORL1 KO THP-1 clones (1–3) relative to WT controls. GAPDH served as the loading control. c Quantification of immunoblots in (b). Densitometric values normalized to GAPDH show significant upregulation of all three ER stress markers in KO clones (n = 3 independent experiments, each including WT and all three KO clones; mean ± SEM; displayed p values are from an unpaired t-test). d,e ER morphological defects in SORL1 KO iMG. Boxplots show ER fragment number per cell (d) and perinuclear ER overlap (e), measured across three independent iMG differentiations. Each point represents an individual cell (colored by replicate). WT and SORL1 KO iMG were stained with the ER marker calnexin (CANX) (see Fig. S4c). KO cells show increased ER fragmentation and reduced perinuclear ER enrichment, consistent with structural ER stress. Two-sided t-test p values are displayed on the plots; ≥20 cells were quantified per replicate. f RT-qPCR analysis showing increased expression of interferon-response genes (IFI6, MX1, ISG15, IFITM3) in SORL1 KO iMG compared to WT (mean ± SEM, n = 3; unpaired t-test; p < 0.05, p < 0.01, p < 0.0001). PLCG2 served as the control gene. g Cytokine profile (71-plex) of WT and SORL1 KO iMG culture supernatants. Volcano plot shows cytokines elevated in SORL1 KO relative to WT iMG (n = 3, independent differentiations). h Volcano plot of bulk RNA-seq data (n = 3) showing differential expression of proinflammatory cytokines in SORL1 KO vs. WT iMG after IFN-γ stimulation. Red dots: adjusted p < 0.05. i Global differential gene expression after IFN-γ stimulation in SORL1 KO vs. WT iMG. Blue: significantly upregulated; orange: significantly downregulated genes. X-axis: log₂ fold change; Y-axis: −log10 FDR (DESeq2 Wald test with Benjamini–Hochberg correction). j,k Human primary microglia with low endogenous SORL1 expression exhibit ER stress, UPR activation, interferon signaling, and lipid-stress signatures. Module score comparisons between SORL1_high and SORL1_low human microglia from Sun et al. Cell (2023) [46]. Boxplots show median, interquartile range, and 5–95% whiskers. SORL1_low nuclei display significantly elevated ER stress transcription factor modules (ATF4, ATF6, CHOP, XBP1; j) and increased composite ER stress scores (k). l Boxplots showing additional stress-responsive transcriptomic modules increased in SORL1_low microglia, including interferon-stimulated gene (ISG) signatures, lipid-stress signatures, and the SORL1-KO_ERstress signature derived from SORL1 KO iMG (Wilcoxon rank-sum test, p < 10⁻³⁰⁰ for all modules). m Heatmap showing per-state differences in stress-related transcriptional module scores between SORL1_low and SORL1_high human microglial states described in Sun et al. Cell (2023). Values represent the mean difference in the displayed module scores within each microglial state (MG0-MG12). Positive values indicate enrichment in SORL1_low microglia, whereas negative values indicate relative depletion. The color scale reflects the magnitude and direction of the difference

To assess whether SorLA deficiency affects ER architecture, we quantified ER morphology in WT and SORL1 KO iMG using immunostaining for the ER membrane marker CANX. SorLA-deficient iMG displayed pronounced structural ER stress (Fig. S4c), including a 166.4% increase in ER fragment number per cell (WT: 2.98 ± 0.14; KO: 7.9 ± 0.97; P = 0.01), a 44.2% reduction in perinuclear ER enrichment (WT: 70.58 ± 1.44%; KO: 39.37 ± 3.66%; P = 0.008) (Fig. 4d,e), and a 59.9% increase in ER texture entropy (P = 0.01) (Fig. S4d). These morphological disruptions are hallmarks of unresolved ER stress and align with the transcriptional and biochemical activation of all three UPR branches described above.

Next to confirm the proinflammatory phenotype of SORL1 KO iMG, we measured the expression of interferon response genes via qPCR. The results showed a more than two-fold increase in the expression of the tested genes (IFI6, ISG15, MX1, IFITM3), while the expression of PLCG2, used as a control, remained comparable in both cell types (Fig. 4f). Further, the cytokine analysis of culture supernatants using a 71-plex panel revealed significantly elevated levels of immune and inflammatory mediators in SORL1 KO iMG compared to controls. Notably, type I interferons (IFNA2), interleukins (TSLP, IL1A, IL8, IL9, IL12B, IL15, IL16, IL18, and IL25) and proinflammatory chemotactic cytokines with C–C (CCL1, CCL24) or C–X–C (CXCL5, CXCL9, CXCL13, CXCL15) motifs were significantly increased in the supernatants from the SORL1 KO iMG (Fig. 4g and Supplementary Table S5). The majority of these upregulated cytokines and chemokines are implicated in neuroinflammation [94–100] and neurodegeneration [101–105]. Finally, bulk RNA-Seq after IFN-γ stimulation of WT and SORL1 KO iMG confirmed the heightened proinflammatory response of SORL1 KO iMG. Transcription of key cytokines (CCL1, TNF, CXCL9, IL18, IL1B, CCL5) and chemokines (CXCL9–11, CXCL13) was significantly elevated in SORL1 KO cells, further validating their proinflammatory phenotype (Fig. 4h,i, QC for bulk RNA-Seq in Supplementary extended data Fig. 4).

Endogenous variation in SORL1 expression in human microglia mirrors the ER stress and inflammatory signatures observed in SorLA-deficient iMG

To determine whether reduced SORL1 expression in human microglia reflects the transcriptional phenotypes observed in SORL1-deficient iMG, we re-analyzed microglial nuclei (n = 152,459) from a published large snRNA-Seq dataset based on ROSMAP cohorts [46]. As expected from a cohort consisting of aged individuals, the distribution of SORL1 expression was strongly skewed (Fig. S4e). Next, we stratified microglia nuclei into SORL1_high and SORL1_low groups (Methods). SORL1_low microglia showed a consistent increase in ER stress-associated transcriptional programs. Module scores for ER stress transcription factors XBP1, CHOP, ATF4, and ATF6 were elevated in SORL1_low cells at the whole-population level (Fig. 4j), a pattern further supported by violin plots with individual nuclei displayed in the Supplementary Figure (Fig. S4f′,f‴). A composite ER stress score showed the same shift (Figs. 4k and S4g). Stress-responsive modules were similarly increased in SORL1_low cells, including interferon-stimulated gene (ISG) signatures, lipid-stress signatures, and the KO_ERstress signature derived from SORL1 KO iMG (Figs. 4l and S4h).

To assess whether SORL1-dependent transcriptional vulnerability varies across microglial phenotypes, we computed per-state module differences across all 12 microglial states (Fig. 4m) described in a published large snRNA-seq dataset [46]. Strikingly, the MG0 (Homeostatic) state showed the most consistent and pronounced increases across all three modules, ISG, SORL1 KO-ER stress (iMG), and lipid-stress, identifying MG0 as the microglial phenotype most broadly sensitive to SORL1 deficiency. MG2 (Inflammatory I), MG8 (Inflammatory II) and MG12 (Cycling) also showed strong increases in individual modules. These state-specific patterns were also corroborated by ER stress score comparisons across microglial subtypes (Fig. S4i). Importantly, the elevated lipid-stress module in SORL1_low microglia parallels the expansion of the lipid-stressed population observed in our SORL1 KO scRNA-seq data, suggesting that endogenous SORL1 variation in vivo may shift microglial metabolism toward a similar lipid-dysregulated state.

Together, these data show that endogenous reduction of SORL1 expression in human microglia recapitulates the ER stress, interferon-stimulated, and lipid-stress transcriptional phenotypes observed in SORL1 KO iMG, supporting a conserved, dose-dependent role for SorLA in maintaining microglial proteostasis and immune quiescence.

SorLA-deficient microglia exhibit disrupted lipid homeostasis and cholesteryl ester accumulation

The single‑cell transcriptomic analysis showed an increased proportion of Cluster 1 in SORL1‑KO iMG, a population defined by elevated expression of lipid‑metabolism-related genes. Considering the established role of ER stress in regulating lipid homeostasis [75, 106], we speculated that SorLA deficiency might disrupt lipid metabolism in microglia. To test this, we stained WT and SORL1 KO iMG with BODIPY [107], a fluorescent dye that labels neutral lipids and lipid droplets (LDs), to assess intracellular lipid accumulation (Fig. 5a). Across three independent differentiations, KO cells consistently exhibited higher BODIPY intensity than their matched WT controls (Fig. 5b). Replicate-wise analyses confirmed highly significant differences in R1 (WT = 0.001419 AU, n = 383; KO = 0.003906 AU, n = 312; U = 372.5), R2 (WT = 0.001183 AU, n = 394; KO = 0.003586 AU, n = 312; U = 165.5), and R3 (WT = 0.001114 AU, n = 379; KO = 0.002434 AU, n = 314; U = 2157), all p < 1 × 10⁻¹⁵). When replicate means were plotted together (Fig. 5b′), KO cultures again showed significantly elevated BODIPY intensity (p = 0.02), demonstrating a robust increase in lipid accumulation in SORL1-deficient iMG. To determine whether this reflected altered LD abundance, we quantified LD number per cell and found that KO cells contained significantly more LDs per cell than WT cells (Fig. 5c). In contrast, LD size distribution and circularity were largely unchanged (Fig. S5a,b), indicating that SorLA loss increases LD abundance without major changes in droplet size distribution or morphology in these cells.

We further validated these findings using ICC for PLIN2, a canonical LD-associated protein [108] (Fig. 5d). Similar to BODIPY, PLIN2 staining was markedly higher in SORL1 KO iMG, with replicate-wise Mann–Whitney tests again yielding low values (R1–R2 p < 1 × 10⁻¹⁵, R3 p = 1.79 × 10⁻¹⁰) (Fig. 5e,f). Immunoblot analysis confirmed increased PLIN2 protein levels in SORL1 KO cells, and importantly, SORL1 overexpression in WT iMG reduced PLIN2 abundance (Fig. 5g), supporting a direct role for SorLA in limiting LD accumulation.

To assess whether the lipid storage phenotype observed in vitro is also present in aged human microglia, we analyzed pMG purified from fresh autopsy DLPFC tissue [18, 43]. pMG were identified by positive expression of CD11B and CD45, with dead cells excluded by 7-AAD (Fig. S5c). Dual ICC for SorLA and the LD-associated protein PLIN2 revealed that pMG with lower SORL1 expression exhibited higher PLIN2 levels (Fig. 5h). Correlation analysis supported this association, showing a modest but significant negative correlation between SorLA and PLIN2 (Pearson r = −0.48, p = 6.3 × 10⁻⁴⁵, R² = 0.23; Spearman r = −0.41, p = 5.5 × 10⁻³²; n = 751 cells from three donors) (Fig. 5i). Although there were no human pMG with a complete lack of SorLA expression as in our engineered KO lines, these findings indicate that endogenous variation in SorLA expression in aged human microglia is linked to differences in lipid storage.

To further corroborate these findings in high-throughput assays that need cell numbers not feasible with pMG, we employed the human cell line THP-1. Flow cytometry revealed significantly higher BODIPY staining in a CRISPR-mediated SORL1 knockdown mixed-indel cell population (Fig. S5d) and in confirmed SORL1 KO clones (Fig. S5e) compared to matched controls. This result was replicated using Nile Red, another fluorescent dye commonly used to assess LD accumulation (Fig. S5f,g). To determine whether the LD phenotype persists under macrophage-like differentiation conditions, we differentiated THP-1 cells using phorbol 12-myristate 13-acetate (PMA), a standard approach for inducing macrophage-like maturation [109]. PMA-treated SORL1 KO cells exhibited a pronounced increase in BODIPY-positive LDs compared with WT controls, with an approximately sixfold elevation in BODIPY signal (Fig. S5h,i). These findings demonstrate that SorLA-dependent lipid accumulation is maintained following macrophage-like differentiation and is not restricted to undifferentiated THP-1 cells.

To identify the specific lipid subclasses affected by SorLA loss, we performed LC–MS-based lipidomics on extracts from control and SORL1 KO THP-1 cells. SORL1 KO cells exhibited a 64% increase in cholesteryl esters (CEs), with no significant change in diacylglycerols or triacylglycerols (DG/TG) (Fig. 5j, Supplementary Table S6). Additional lipid classes, including NAPE, GB3, AcylPG, sulfatide, and acyl-carnitine, were elevated, while PI, GM3, and dhSM were reduced, indicating a broader reorganization of lipid metabolism. Because CEs arise from lipoprotein processing and accumulate in the aged human brain, their selective increase suggests that SorLA plays a key role in regulating cholesterol esterification in myeloid cells.

To test whether elevated CE levels in SorLA-depleted THP-1 cells altered free cholesterol content, we performed filipin staining, a marker of unesterified cholesterol (Fig. 5k). Flow cytometry across three independent experiments showed a significant reduction in free cholesterol in cells targeted with two independent SORL1 sgRNAs, compared with non-targeting controls (Fig. 5l), suggesting enhanced channeling of cholesterol toward esterification in the absence of SorLA.

We next examined whether changes in cholesterol metabolism were reflected at the transcriptional level in our scRNA-Seq dataset from WT and SORL1 KO iMG. Expression of CYP27A1, SOAT1, and CH25H, genes involved in cholesterol hydroxylation and esterification, was significantly increased in SORL1 KO iMG, whereas SQLE, a key cholesterol-biosynthesis enzyme, was decreased (Fig. 5m). These transcriptomic signatures are consistent with enhanced cholesterol processing and esterification in SORL1-deficient microglia.

Finally, we investigated whether restoration of SORL1 expression could reverse LD accumulation. Using a doxycycline-inducible CRISPR interference (CRISPRi) system, we observed a marked increase in BODIPY staining upon SORL1 repression compared with cells expressing a non-targeting sgRNA (Fig. 5n). Notably, removing doxycycline for three days did not reverse the LD accumulation (Fig. 5n), suggesting that recovery from SorLA depletion may require a longer time or that lipid homeostasis resets slowly once perturbed.

Since immunostaining of human DLPFC tissue showed robust SORL1 expression in neurons as well as microglia, we tested whether SorLA-dependent lipid dysregulation is restricted to microglia or also occurs in neurons. MAP2/NeuN-positive iNeurons generated by NGN2 induction from the same iPSC line used for iMG (Supplementary extended data Fig. 5a,b) exhibited a marked increase in BODIPY staining in SORL1 KO neurons compared with WT (difference ± SEM: 50.64 ± 5.06; p = 0.0005; Supplementary extended data Fig. 5c). To exclude an iPSC line-specific effect, we generated SORL1 KO SH-SY5Y neuroblastoma cells, which similarly showed elevated BODIPY intensity relative to WT (difference ± SEM: 61.75 ± 16.07; p = 0.01; Supplementary extended data Fig. 5d). Consistent with this phenotype, SORL1 KO SH-SY5Y cells also displayed increased PERK protein levels, indicative of ER stress activation (Supplementary extended data Fig. 5e). These findings indicate that SorLA loss promotes LD accumulation across both microglia and neurons—two cell types that endogenously express high levels of SORL1 in human brain.

Together, these data demonstrate that loss of SorLA may trigger LD accumulation, increased cholesterol esterification, and lipid-stress signatures, changes relevant to AD-associated cellular pathology.

SorLA deficiency-mediated ER stress drives lipid droplet accumulation in microglia

To investigate the mechanisms linking ER stress, interferon response, and lipid accumulation in SorLA-deficient microglia, we performed a series of experiments. Given that SorLA binds mouse ApoE [110], a cholesterol transport protein, one plausible hypothesis could be that SorLA deficiency disrupts ApoE trafficking, leading to its intracellular accumulation and subsequent lipid dysregulation. However, Western blot analysis revealed significantly lower intracellular ApoE levels in SORL1 KO iMG compared to WT controls (t-test, t = 6.919, df = 3, p = 0.006, Fig. 6a,b). Conversely, the immunoturbidimetric assay of cell culture (iMG) supernatants revealed significantly elevated ApoE levels in SORL1 KO iMG (unpaired t-test, p = 0.044, t = -2.89, df = 4, η² = 0.67, Fig. 6c). This increase in extracellular ApoE was further confirmed by ELISA (Welch’s t-test, p = 4.6 × 10⁻⁵, t = 142.7, df = 2.011, Fig. 6d). However, elevated ApoE in KO supernatants could reflect reduced ApoE uptake rather than increased secretion. To test this, we quantified ApoE released from THP-1 cells cultured in serum-free media, eliminating external ApoE sources. ApoE levels remained significantly higher in SORL1 KO supernatants under serum-free conditions (Fig. S6a), suggesting that elevated ApoE release is likely cell-intrinsic. These findings are consistent with the established role of SorLA in endosomal sorting and cargo retention [13, 111] supporting a model in which SorLA deficiency redirects ApoE trafficking from intracellular compartments toward secretion. This alteration in ApoE handling aligns with reports that ApoE deficiency or dysregulation can influence microglial lipid metabolism. Complementary biochemical analysis showed elevated EEA1 protein expression in SORL1 KO THP-1 cells, consistent with early endosomal dysregulation accompanying the secretion-biased shift in ApoE trafficking (Fig. S6b,b’). ICC further revealed that SORL1 KO cells exhibit a visibly higher number of EEA1⁺ early endosomes with increased perinuclear clustering, indicating an expanded early endosomal burden (Fig. S6c,c’). Moreover, consistent with broader endolysosomal disruption, our single-cell data also showed a marked reduction of a CD63^hi microglial cluster in SORL1 KO iMG (Fig. 3c,d). Although previous work [23] has reported endosomal enlargement predominantly in neurons, these data suggest that SORL1 loss alters endosomal markers in microglia as well.

Fig. 6.

SorLA deficiency-induced ER stress promotes LD accumulation in microglia. a Representative Western blot (WB) showing reduced intracellular ApoE levels in SORL1 KO iMG vs. WT controls. Vinculin served as a loading control. b Densitometric quantification of the WB shown in (a) and two others confirms significantly decreased intracellular ApoE in SORL1 KO iMG (t-test, t = 6.919, df = 3, p = 0.006). c Immunoturbidimetric assay of supernatants reveals significantly higher ApoE levels in SORL1 KO iMG vs. controls (unpaired t-test, t = −2.89, df = 4, p = 0.044, η² = 0.67). d ELISA confirms significantly increased ApoE in SORL1 KO iMG supernatants (displayed p value: unpaired t-test with Welch’s correction, t = 142.7, df = 2.011). e RT-qPCR analysis shows significant upregulation of IFI6, ISG15, and MX1 in SORL1-intact iMG with CRISPR activation (displayed p values: two-tailed unpaired t-test). f Dot plot showing the quantification of flow cytometry data expressed as the fold change in the relative abundance of BODIPY+ iMG after IFI6, ISG15, and MX1 overexpression in SORL1-intact cells; only ISG15 overexpression shows a significant increase (displayed p value: two-tailed unpaired t-test). g BODIPY+ percentage in SORL1 KO THP-1 clones after Ruxolitinib, a JAK inhibitor, treatment reveals a significant reduction only in clone 2 (paired t-test, p = 0.008). h Representative flow cytometry histograms showing increased BODIPY staining in THP-1 cells following brefeldin A (BFA) treatment (1 µg/mL, 72 h). i Quantification of the proportion of BODIPY⁺ THP-1 cells in WT controls, untreated SORL1 KO clones (C1–C3), and the same genotypes following BFA treatment. BFA significantly increased BODIPY positivity in WT cells and further increased LD burden in each SORL1 KO clone relative to untreated conditions. Individual data points represent technical replicates; bars indicate mean ± SD. Welch’s t-tests were used to compare WT versus each SORL1 KO clone, and paired t-tests were used to compare untreated versus BFA-treated conditions within each genotype. Related p values are displayed on the graph. j Representative flow cytometry histograms showing BODIPY staining in WT and SORL1 KO THP-1 clones treated with the ER stress-relieving agent tauroursodeoxycholic acid (TUDCA; 100 µM, 72 h). k Quantification of the proportion of BODIPY⁺ cells following TUDCA treatment. TUDCA did not significantly alter LD burden in WT cells but significantly reduced BODIPY positivity in each SORL1 KO clone. Individual data points represent technical replicates; bars indicate mean ± SD. Statistical comparisons were performed as in panel (i) and p values are displayed on the graph. l WB analysis of the ER stress marker BiP in TUDCA-treated SORL1 KO THP-1 clone 3. l′ Densitometric quantification of WB showing reduced BiP expression following TUDCA treatment in SORL1 KO THP-1 clones (blue: clone 2; red: clone 3). Displayed p value was determined using a Mann–Whitney test on four experiments performed with two SORL1 KO THP-1 clones, indicating effective attenuation of ER stress. Data in panels (a–f) are from iMG, and panels (g–l′) are from THP-1 cells

Nonetheless, SORL1 KO cells exhibit reduced intracellular ApoE yet still accumulate LDs; ApoE mistrafficking alone is insufficient to account for the lipid phenotype. Therefore, we tested whether interferon activation contributes to lipid accumulation. We used a CRISPR activation system to induce IFI6, ISG15, and MX1 expression in WT iMG (Fig. 6e). Although BODIPY staining revealed a trend toward increased LDs, the effect was not statistically significant, with ISG15 showing the most pronounced increase (Fig. 6f), consistent with reports linking ISG15 to lipid metabolic regulation [112–114]. Conversely, inhibiting the JAK1/2-STAT1 pathway in SORL1 KO cells using the Ruxolitinib inhibitor [115, 116] only partially reduced BODIPY staining (Fig. 6g), indicating that interferon signaling may contribute but is not the principal driver of SORL1-linked lipid dysregulation.

We next examined whether ER stress alone is sufficient to induce LD accumulation and whether additional ER stress can further exacerbate the LD phenotype caused by SORL1 deficiency. Treatment of WT THP-1 cells with brefeldin A (BFA), a classical inducer of ER stress and ER–Golgi trafficking collapse [117, 118], significantly increased the proportion of BODIPY⁺ cells, indicating elevated LD accumulation (Fig. 6h,i). Importantly, although SORL1 KO cells already exhibited markedly higher LD levels under basal conditions compared to untreated WT cells, BFA treatment further increased LD accumulation in SORL1 KO clones (Fig. 6i), suggesting that SorLA deficiency does not saturate the LD phenotype and that additional, SorLA-independent ER stress can further aggravate lipid dysregulation. Notably, BFA also reduced SorLA protein levels in WT cells (Fig. S6d,d′), indicating that acute trafficking disruption suppresses SORL1 expression.

Consistent with increased LD burden, BFA robustly induced ER stress markers in WT cells and further exacerbated ER stress in SorLA-deficient THP-1 clones, as evidenced by increased BiP protein levels (Fig. S6d,d′). At the transcriptional level, BFA strongly induced BiP/GRP78, spliced XBP1, CHOP, and EDEM expression in WT cells, with a further enhanced induction observed in SORL1 KO clones (Supplementary extended data Fig. 6a), indicating heightened sensitivity of SorLA-deficient cells to trafficking-induced ER stress.

To determine whether alleviating ER stress is sufficient to reverse lipid accumulation, we treated THP-1 cells with tauroursodeoxycholic acid (TUDCA), a chemical chaperone and ER stress-relieving agent [119]. Flow cytometry-based quantification revealed that TUDCA treatment did not significantly alter the proportion of BODIPY⁺ cells in WT THP-1 cells (paired t-test, p = 0.69), indicating minimal effects on basal lipid content (Fig. 6j,k). In contrast, TUDCA significantly reduced LD accumulation in each SORL1 KO clone (paired t-tests: C1 p = 0.01; C2 p = 0.003; C3 p = 0.007) (Fig. 6j,k). This reduction in LD burden was accompanied by decreased BiP protein expression (Fig. 6l,l′) and a significant reduction in BiP mRNA levels, while other ER stress markers (sXBP1 and CHOP) showed trends toward reduction that did not reach statistical significance (Supplementary extended data Fig. 6b). Together, these data indicate that ER stress is sufficient to drive LD accumulation and that chemical alleviation of ER stress selectively alleviates lipid accumulation in SorLA-deficient cells without perturbing lipid homeostasis in WT cells.

Despite effective modulation of ER stress markers, neither BFA nor TUDCA normalized interferon-stimulated gene expression in SorLA-deficient THP-1 cells. Instead, BFA treatment increased IFI6, and ISG15 expression in SORL1 KO THP-1 clones, whereas WT cells exhibited reduced (IFI6, MX1) or increased (ISG15, IFITM3) expression under the same conditions (Supplementary extended data Fig. 6c). Interestingly, TUDCA treatment failed to suppress, and in some cases (IFI6, ISG15) was associated with further enhancement of interferon-stimulated gene expression in SORL1 KO clones. While in WT THP-1 cells, TUDCA treatment increased ISG15, MX1, and IFITM3 expression, IFI6 expression remained unchanged (Supplementary extended data Fig. 6d).

Together, these findings demonstrate that SorLA deficiency disrupts intracellular ApoE routing and sensitizes cells to ER stress, contributing to LD accumulation. However, modulation of ER stress alone is insufficient to normalize interferon-stimulated gene expression in SorLA-deficient cells, suggesting that interferon activation and ER stress represent partially uncoupled responses in this context.

SorLA-associated protein networks reveal ER enrichment

Given that we found that SorLA deficiency induces ER stress and disrupts lipid homeostasis, we next sought to identify the molecular pathways through which SorLA may associate with ER-related protein networks. To do this, we employed co-immunoprecipitation (IP) coupled with mass spectrometry (IP-MS) to map the SorLA protein association network in myeloid cells. Using THP-1 cell lysates, we performed IP with an anti-SorLA antibody, alongside an IgG isotype control to account for non-specific binding. After confirming antibody specificity, SorLA-associated protein complexes were purified (Fig. 7a) and subjected to MS analysis to identify SorLA co-complexed proteins and enriched pathways.

Fig. 7.

Identification of the SorLA-associated protein networks using immunoprecipitation followed by mass spectrometry (IP-MS). a Western blot confirms the specificity of SorLA co-immunoprecipitation (IP) using anti-SorLA antibody (IP-SorLA) compared to IgG control. Input shows baseline SorLA expression in THP-1 cells. Empty lanes were intentionally included between samples for clarity. b Scatterplot shows reproducibility of SorLA IP (as measured by SorLA intensity) between two IP-mass spectrometry (IP-MS) experiments (Pearson r = 0.97). c Scatterplot comparing anti-SorLA IP-MS and IgG IP-MS intensities. SorLA shows higher enrichment in the anti-SorLA condition (Pearson r = 0.31). d PCA plot highlighting separation between IgG IP-MS controls (IgG_MS, black) and SorLA IP-MS samples (SorLA_MS, blue). SorLA IP-MS replicates cluster tightly, whereas IgG IP-MS samples are dispersed and clearly separated from SorLA IP-MS samples, indicating high reproducibility of SorLA-specific IPs. Principal component 1 explains 77.8% of the total variance. e Volcano plot of SorLA-associated proteins identified by IP-MS. Significant enrichments were determined using a permutation-based FDR approach in Perseus with an S₀-modified t-test (S₀ = 0.1). High-confidence (Class A, FDR ≤ 0.01%) and low-confidence (Class B, FDR ≤ 0.05%) associations are separated by solid and dashed significance thresholds, respectively. The x-axis shows the log₂ fold change in protein abundance between anti-SorLA and IgG control samples. Proteins enriched in the SorLA IP, such as SUN2, SYNE3, and SorLA (bait), are highlighted in orange and red colors. f Western blot analysis of cell lysates from WT and SORL1 KO cells with antibodies against SorLA and a newly identified SorLA-associated calnexin (an ER chaperone protein) reveals reduced calnexin levels in SORL1 KO samples. g Bar graph shows the quantification of the calnexin/GAPDH ratio, showing a significant decrease in calnexin protein expression in the SORL1 KO condition (WB n = 3, two-tailed unpaired t-test p = 0.003). h,h” Volcano plots of SorLA co-complexed proteins by subcellular compartments. ER-associated proteins (pink, h”) dominate, followed by endosomal (orange, h) and lysosomal (cyan, h’) proteins. i,i’ “Representative images from ICC analysis of microglia, freshly isolated from postmortem human DLPFC, stained for SorLA (green), calnexin (red), and DAPI (blue). SorLA puncta exhibit perinuclear and cytoplasmic distribution with regions of partial overlap with calnexin-positive ER structures. Magnified inset (i’) highlights examples of SorLA-calnexin partial co-localization. Scale bar: 10 μm. j Bubble plot from GO cellular component (GO_CC) analysis of SorLA-binding proteins shows enrichment of ER-related terms, including “ER membrane,” “ER-Golgi intermediate compartment,” and “TAP complex”. k Biological process analysis of SorLA-binding partners highlights significant enrichment of “antigen processing via MHC class I TAP-dependent,” “cytosol to ER transport,” and “intracellular protein transport.” Bubble size represents gene count; color indicates enrichment magnitude. AP-P: antigen processing and presentation. Abbreviations: AP-P antigen processing and presentation, IP co-immunoprecipitation, IgG immunoglobulin, Expt. experiment, ER endoplasmic reticulum, nGenes number of genes. IP-MS data are from THP-1 cells.

The MS data demonstrated high similarity between technical replicates of IP-SorLA, Pearson r = 0.97, Figs. 7b and Supplementary extended data Fig. 7a,b), indicating robust reproducibility within the SorLA group. In contrast, replicates of IgG and IP-SorLA differed significantly (r = 0.3), highlighting clear distinctions in protein abundance between the two groups (Figs. 7c and S7a,b). Likewise, the principal component analysis (PCA) revealed that Component 1 explained 77.8% of the variance and Component 2 explained 14.4%, collectively capturing over 92% of the variance, indicating that the data are well represented in reduced dimensionality (Fig. 7d). These results suggested the successful IP capture of SorLA and its associated protein complexes.

After quality control, the Perseus MS data analysis pipeline [120] identified 81 significant SorLA-associated proteins by comparing protein abundance in anti-SorLA pulldowns versus IgG controls (Supplementary Table S7). A permutation-based FDR method with an S₀-modified t-test (S₀ = 0.1) was used to classify proteins into high-confidence (Class A, FDR = 0.01%) and low-confidence (Class B, FDR = 0.05%) categories, based on thresholds consistent with previous IP-MS studies [121, 122] (Fig. 7e). In line with our SORL1 KO iMG functional data, these SorLA-associated proteins included subunits of the coat protein complex I (COPI), such as COPA, COPB1, COPG1, and ARCN1, which are critical for sorting lipids and proteins between the Golgi and ER [123, 124]. Other identified proteins included those involved in protein folding (e.g., DNAJA2, DNAJC7, TSC1, CANX, TCP1, SPTLC1, STT3A, and GANAB) and lipid binding (e.g., ESYT1, ESYT2, GOLPH3, KIF16B, PSD4, SNX8), as well as cholesterol-associated proteins like HDLBP.

Since SorLA plays a key role in protein recycling [125–127], its absence may influence the abundance of associated proteins. To investigate this, we examined the protein levels of CANX (calnexin), identified as a SorLA-associated protein. CANX is an ER chaperone crucial for protein folding and quality control [128]. We observed reduced CANX protein levels in SORL1 KO THP-1 cells compared to controls (Fig. 7f,g), despite a slight increase in its transcription (Fig. S7c), indicating regulation at the protein level. Loss of CANX, along with other ER-associated SorLA co-enriched proteins (e.g., GANAB [129], STT3A [130], SPTLC1 [131], COPA [132]) could disrupt protein folding, glycosylation, membrane composition, and ER-Golgi trafficking, respectively. This can lead to ER stress observed in SORL1 KO cells.

Notably, highly enriched SorLA co-complexed proteins also included SUN2 and SYNE3, members of the LINC complex, which play roles in nuclear stability and in ER stress associated with the unfolded protein response (UPR) in premature aging [133]. Interestingly, cellular component analysis in Perseus revealed that while endolysosomal proteins were detected, the majority of SorLA-associated proteins were ER-associated, suggesting SorLA’s potential role in ER function in these cells (Fig. 7h,h’’).

To independently examine the ER association of SorLA suggested by the IP-MS data, we performed ICC on human microglia isolated from DLPFC autopsy tissue using antibodies against SorLA and the ER marker calnexin. SorLA displayed clear perinuclear and cytoplasmic puncta that partially overlapped or were closely apposed to calnexin-positive ER structures, consistent with proximity to ER-associated compartments in these cells (Fig. 7i,i’). In contrast, double labeling of SorLA with the early endosome marker EEA1 revealed only sparse co-localization, with most SorLA puncta positioned adjacent to rather than overlapping with EEA1-positive compartments (Supplementary Fig. S7d,d’). These observations indicate that, in microglia, SorLA is not strongly enriched in early endosomes, as previously reported in neurons, and instead frequently appears in perinuclear regions consistent with ER-proximal trafficking compartments, in line with the ER-associated protein network identified in the IP-MS analysis.

Using ShinyGO, a gene-set enrichment tool [45], we analyzed the 81 significant SorLA-binding partners for gene ontology (GO) enrichment. GO cellular component (GO_CC) analysis identified statistically enriched terms including the transporter associated with antigen processing (TAP) complex, ER protein-containing complex, nuclear outer membrane-ER membrane network, ER-Golgi intermediate compartment, COPI vesicle coat, and MHC class I peptide loading complex (Figs. 7j and S7e). Notably, the enrichment of TAP complex components, which mediate peptide translocation into the ER for MHC class I presentation, may point to an immune-related functional aspect of SorLA, particularly relevant in myeloid cells where antigen presentation machinery is more prominently engaged. GO biological process (GO_BP) analysis identified enrichment in terms related to antigen processing, ER, protein transport and localization (Fig. 7k).

Together, these findings suggest that SorLA contributes to ER-associated processes including antigen presentation and protein trafficking in microglia.

SorLA co-immunoprecipitates with SUN2 and regulates its abundance

Our MS analysis identified SYNE3 and SUN2, components of the LINC complex, as proteins co-enriched with SorLA, exhibiting large fold enrichment (Fig. 7e). To corroborate this finding, we performed IP followed by western blot (WB) analysis in THP-1 cells. SUN2 was consistently detected in anti-SorLA IP complexes but not in IgG controls (Fig. 8a). To assess the relevance of this association in the human brain, we next performed SorLA IP followed by SUN2 immunoblotting using human DLPFC tissue lysates. Consistent with our in vitro findings, SUN2 co-immunoprecipitated with SorLA in DLPFC lysates, supporting a physiological association between these proteins (Fig. 8b).

Fig. 8.

SorLA co-immunoprecipitates with SUN2 and regulates its abundance. a Co-immunoprecipitation (IP) of SorLA from THP-1 cells, followed by Western blot (WB) analysis shows SorLA and SUN2 co-precipitate specifically in IP-SorLA (n = 3). b IP of SorLA from human DLPFC lysates followed by WB confirms SorLA–SUN2 association in the brain (n = 3). c Immunohistochemistry (IHC) of SorLA and SUN2 in human dorsolateral prefrontal cortex (DLPFC) sections. Nuclei are labeled with DAPI (blue), microglia with IBA1 (green), SorLA in red, and SUN2 in white. Merged images show partial co-localization of SorLA and SUN2 (yellow arrow). Sections from four ROSMAP donors were analyzed. Scale bars 10 μm. d ICC detection of SorLA (red) and SUN2 (green) in acutely isolated CD11b⁺CD45⁺ human microglia. Merged images and magnified inset highlight the region of partial SorLA–SUN2 co-localization (yellow arrows). Microglia from three ROSMAP donors were analyzed. Scale bars, 10 μm. e Composite AlphaFold model scores for the predicted SorLA–SUN2 association. Bar plots show model ranking by predicted binding energy (-ΔG) weighted by VPS10 domain contact enrichment (left) or by enrichment of contacts within SUN2 low-complexity and unannotated regions (right). Model 3 consistently ranks among the highest-scoring models across both scoring schemes. f Relationship between predicted binding strength (-ΔG) and the fraction of contacts within VPS10 domains across AlphaFold models. Each point represents an individual model, illustrating the trade-off between overall binding energy and domain-specific contact enrichment. g Structural visualization of AlphaFold model 3 depicting the predicted SorLA–SUN2 interface. SUN2 is positioned in proximity to the VPS10-containing region (orange) of SorLA, consistent with biochemical and fluorescence imaging analyses. g′ Magnified view of the predicted contact interface shown in (g). In gg′, non-contact regions of SorLA are shown in slate, the VPS10 domain in orange, and predicted contact residues within VPS10 in red. Predicted contact residues of SUN2 are highlighted in yellow, while non-contact regions of SUN2 are shown in marine. h Western blot shows reduced SUN2 expression in SorLA KO (-) THP-1 cells compared to SorLA-intact (+) control, with GAPDH as the loading control. i Quantification of WB shows SUN2 protein level is significantly reduced in SorLA KO (−) THP-1 cells (n = 6, p: 0.001). SUN2 band intensities were normalized to GAPDH and subsequently normalized to the WT value within each blot to correct for inter-blot variability. j,k SUN2 mRNA expression in WT and SORL1 KO THP-1 cells (j) and iMG (k), measured by RT-qPCR and normalized to GAPDH. Scatter plots show increased SUN2 transcript levels in SORL1 KO cells compared to WT in both systems. Data are shown as mean ± SEM; shown p values were calculated using unpaired two-tailed t-tests (j) and Welch’s t-test (k). l Scatter plots show the relationship between SorLA and SUN2 protein levels (LFQ intensities) across Braak stages in DLPFC samples from a published large-scale proteomic dataset (Johnson et al., Nat Med, 2020). Correlation analysis was performed separately for Braak (B) I–II (early), Braak III–IV (intermediate), and Braak V–VI (late) stages. No significant correlation was observed at Braak I–II or V–VI, whereas a significant positive correlation between SorLA and SUN2 protein abundance was detected at Braak stages III–IV. Pearson and Spearman correlation coefficients and corresponding p values are indicated on each plot

Immunohistochemical analysis of aged human DLPFC sections revealed partial co-localization of SorLA and SUN2 at the nuclear periphery of IBA1-positive microglia (Fig. 8c). Similarly, ICC of freshly purified primary human microglia demonstrated close spatial proximity and partial overlap of SorLA and SUN2 signals in the perinuclear region (Fig. 8d). This spatial association was further supported at single-cell resolution by proximity ligation assay (PLA) in human DLPFC tissue sections (Fig. S8a), indicating close molecular proximity in situ.

Given that cell-specific ICC analysis in human DLPFC revealed SorLA expression in both microglia and neurons (Fig. 1), we next examined whether the SorLA–SUN2 association is conserved in neuronal systems. To this end, we performed SorLA IP in human iPSC-derived neurons (iNeurons) generated from the same iPSC line used for iMG differentiation, as well as in the human immortalized neuronal cell line SH-SY5Y [134]. In both iNeurons (Fig. S8b) and SH-SY5Y cells (Fig. S8c), SUN2 was co-immunoprecipitated with SorLA, demonstrating that the SorLA–SUN2 association is not restricted to myeloid cells but is also present in neuronal contexts.

To further assess the structural plausibility of the SorLA–SUN2 association, we performed AlphaFold multimer modeling [48, 135]. Across five independent models, AlphaFold consistently predicted SorLA–SUN2 interfaces with favorable binding energies and extensive interfacial contacts (Fig. S8d). Domain-level contact-site analysis revealed that predicted contacts were not uniformly distributed but were strongly enriched within VPS10-containing domains of SorLA, while SUN2 contact sites localized primarily to low-complexity regions (Fig. S8e). Residue-level analyses further demonstrated clustering of high-intensity contact residues within these domains across multiple models (Supplementary extended data Fig. 8a,b). Composite scoring that integrated predicted binding energy with VPS10 domain contact enrichment prioritized a subset of convergent models, supporting the robustness of the predicted interface (Fig. 8e,f). Structural visualization of the representative model (Model 3) placed SUN2 in close proximity to VPS10 domains of SorLA, consistent with the biochemical and ICC data (Fig. 8g,g’). Together, these computational predictions support a structurally plausible and domain-specific SorLA–SUN2 association interface.

To determine whether SorLA regulates SUN2 abundance, we examined SUN2 protein levels in WT and SORL1 KO THP-1 cells. Immunoblot analysis revealed a significant reduction in full-length SUN2 protein in SORL1 KO cells compared to WT controls (Fig. 8h,i). In contrast, SUN2 transcript levels were increased in SORL1 KO cells, an effect observed in both THP-1 cells and iMG (Fig. 8j,k), indicating a dissociation between SUN2 mRNA and protein abundance following SorLA loss. Consistent with post-transcriptional regulation, SORL1 overexpression using a CRISPR activation (CRISPRa) approach increased SUN2 protein levels (Fig. S8f,f’). In contrast, bafilomycin-mediated lysosomal inhibition did not restore SUN2 protein levels in SORL1 KO THP-1 cells and was accompanied by a reduction of both SorLA and SUN2 compared to untreated WT cells (Fig. S8g,g′), arguing against lysosomal degradation as the primary mechanism underlying SUN2 loss.

To assess the relevance of the SUN2–SorLA relationship in the human brain, we analyzed large-scale proteomic datasets from the DLPFC of AD and non-AD donors [60]. This analysis revealed a Braak stage-dependent positive correlation between SorLA and SUN2 protein abundance, with a significant correlation at intermediate Braak stages (III–IV), but not at early (I–II) or late (V–VI) stages (Fig. 8l). A positive correlation between SorLA and SUN2 protein abundance was also observed in an independent proteomic dataset derived from postmortem orbitofrontal cortex samples from an alcohol use disorder cohort (Fig. S8h). While these bulk proteomic datasets do not resolve cell-type specificity, publicly available protein expression resources [136] indicate that SUN2 is broadly expressed across major brain cell types, including neurons and glia, consistent with the observed SorLA–SUN2 relationship in the human brain.

Together, these findings indicate that coordinated variation in SorLA and SUN2 protein abundance is evident in the human brain and is dynamically regulated across disease progression, consistent with post-transcriptional coupling between these proteins during intermediate stages of AD.

CRISPR-mediated SUN2 depletion recapitulates select SorLA-associated phenotype

To determine whether SUN2 depletion phenocopies SorLA deficiency, we generated SUN2-deficient THP-1 cells using two independent CRISPR sgRNAs. Loss of SUN2 resulted in a significant accumulation of LDs compared to non-targeting controls, closely resembling the phenotype observed in SORL1 KO cells (Fig. 9a). Quantitative analysis confirmed robust LD accumulation in both SORL1 KO and SUN2 KO cells across independent experiments (Fig. 9b). SUN2 deficiency also elicited a partial ER stress response as indicated by increased BiP expression and a modest elevation of PERK protein levels (Fig. 9c,d). In contrast, activation of the IRE1a pathway was not significantly induced (Fig. S9a,a′). Overall, ER stress signaling was less pronounced in SUN2-deficient cells than in SORL1 KO cells (Fig. 9c,d).

Fig. 9.

SUN2 depletion recapitulates SorLA deficiency. a Flow cytometry histogram of BODIPY staining in THP-1 cells targeted with non-targeting (NT), SORL1, or SUN2 sgRNAs. b Quantification of BODIPY-positive cells from (a) and three additional experiments. Lipid accumulation is significantly increased in SorLA and SUN2-deficient cells (NT vs. SorLA: unpaired t-test, p = 1.4 × 10⁻⁵; NT vs. SUN2 sgRNA1: Mann–Whitney, p = 0.02; NT vs. SUN2 sgRNA2: unpaired t-test, p = 0.0003). c Western blot of PERK and BiP expression in SUN2 KO, WT, and SORL1 KO THP-1 cells, with GAPDH as a loading control. d Densitometric quantification of WB from (c) and two additional experiments, showing significantly increased BiP and PERK levels in SUN2 KO, though lower than SORL1 KO. BiP: One-way ANOVA, p = 0.0001; Tukey’s test: SUN2 KO vs. WT, p = 0.01; SUN2 KO vs. SORL1 KO, p = 0.001; WT vs. SORL1 KO, p = 9.0 × 10⁻⁵. PERK: One-way ANOVA, p = 0.0001; Tukey’s test: SUN2 KO vs. WT, p = 0.02; SUN2 KO vs. SORL1 KO, p = 0.01; WT vs. SORL1 KO, p = 0.00072. e Western blot of SorLA, BiP, and SUN2 in WT, SORL1 KO, and SORL1 KO-SUN2 overexpression (OE) (CRISPRa-sgRNA1) cells. SORL1 KO reduces SUN2 levels, which are restored by SUN2 OE. f,g Quantification of SUN2 (f) and BiP (g) levels (mean ± SD, n = 3). Data are shown as target protein/GAPDH ratios and normalized to WT within each independent experiment. In (f), p values were calculated using an unpaired two-sided t-test comparing SUN2 levels in SORL1 KO versus SORL1 KO-SUN2 OE. In (g), BiP levels in three SORL1 KO samples normalized to three WT samples (set to 1) were analyzed using a one-sample two-sided t-test, while comparisons between SORL1 KO and SORL1 KO-SUN2 OE were performed using an unpaired two-sided t-test. h Flow cytometry contour plots of BODIPY staining in SORL1 KO, SUN2 KO, and SORL1 KO-SUN2 OE THP-1 cells, with percentages of BODIPY-positive cells shown. i Quantification of (h) and two other experiments (mean ± SD, n = 3). Lipid accumulation is partially rescued by SUN2 OE in SORL1 KO cells. Displayed p value in (i) is from unpaired t-test. j Box plot of nuclear compactness in control, SORL1 KO, and SUN2 KO THP-1 cells, showing significantly reduced nuclear compactness in both knockouts (Kruskal–Wallis, p < 1.0 × 10⁻¹⁵; Dunn’s test: WT vs. SORL1 KO, p = 5.6 × 10⁻⁴; WT vs. SUN2 KO, p < 1.0 × 10⁻¹⁵; SORL1 KO vs. SUN2 KO, p < 1.0 × 10⁻¹⁵). Sample sizes (number of cells): Number of cells analyzed; WT = 619, SORL1 KO = 523, SUN2 KO = 363. Data represent three independent experiments. Data shown are from THP-1 cells

Given that SORL1 KO cells exhibit reduced SUN2 abundance, we next tested whether restoring SUN2 expression could mitigate SorLA-dependent phenotypes. CRISPR activation (CRISPRa)-mediated overexpression of SUN2 in SorLA-deficient THP-1 cells (Fig. 9e,f) partially alleviated ER stress, as reflected by reduced BiP protein levels (Fig. 9e,g). In contrast, PERK and IRE1a protein levels remained unchanged (Fig. S9b,b″ and c,c′). Consistent with this molecular effect, SUN2 overexpression in SORL1 KO THP-1 cells modestly reduced LD accumulation (Fig. 9h,i), indicating that restoration of SUN2 selectively impacts lipid storage with limited effects on global ER stress signaling.

As SUN2 is a core component of the LINC complex, we next assessed nuclear architecture. Both SORL1 KO and SUN2 KO cells exhibited significantly reduced nuclear compactness compared to WT cells, with SUN2 deletion producing a more severe defect (Fig. 9j). Consistent with these structural alterations, immunocytochemical analysis revealed fragmented and discontinuous perinuclear SUN2 staining in SORL1 KO iMG, accompanied by increased nuclear size and nuclear envelope irregularity relative to WT cells (Fig. S9d,e).

Together, these findings suggest SUN2 as a downstream effector of SorLA that selectively contributes to ER stress-associated LD accumulation and nuclear structural integrity.

Discussion

SorLA has been extensively studied as an intracellular sorting receptor implicated in Alzheimer’s disease (AD), primarily through its roles in neuronal endosomal trafficking and the processing of amyloid precursor protein. Here, we expand this framework by demonstrating that SorLA plays a broader role in maintaining microglial proteostasis, lipid homeostasis, and nuclear organization. Using complementary analysis across human brain tissue, iPSC-derived microglia (iMG), primary human microglia (pMG), and myeloid cell models, our findings position SorLA as an important regulator of ER-associated pathways in microglia, with downstream consequences for lipid metabolism and stress-responsive transcriptional programs, both of which are increasingly recognized as critical modifiers of neurodegenerative disease progression [137, 138].

A major finding of this study is that SorLA deficiency induces a robust ER stress response in microglia. This response emerged consistently across experimental systems and was reproduced using both CRISPR-mediated gene knockout and CRISPR interference-based SORL1 expression repression. The concordance of these approaches strongly argues against the possibility that ER stress is due to aberrant or truncated SorLA protein species. Instead, the data support a model in which reduced SorLA dosage directly compromises ER homeostasis. Consistent with this interpretation, meta-analysis of large-scale single-nucleus transcriptomic datasets revealed that endogenous variation in SORL1 expression within human microglia aligns with transcriptional signatures associated with ER stress and dysregulated lipid metabolism analogous to that observed in microglia derived from SorLA-deficient iPSCs. These convergent observations indicate a conserved, dose-sensitive requirement for SorLA in maintaining microglial proteostasis and lipostasis in the human brain and suggest that downregulation of SORL1 in AD may make microglia susceptible to chronic and maladaptive stress responses that could exacerbate neurodegenerative pathology.

Loss of SorLA consistently induced pronounced lipid droplet (LD) accumulation across all systems examined, including iMG, pMG, THP-1 cells, and SORL1-deficient iPSC-derived neurons, indicating that lipid dysregulation is a core, cell-autonomous consequence of SorLA depletion. While lipidomics was performed in THP-1 cells due to cell number requirements, convergent transcriptomic and protein-level analyses in iMG and human pMG support a shared ER stress–LD axis downstream of SorLA loss. Lipidomic profiling of SorLA-deficient THP-1 cells revealed selective enrichment of cholesteryl esters with a concomitant reduction in free cholesterol, consistent with impaired intracellular cholesterol flux rather than increased de novo synthesis. Concordantly, our scRNA-Seq analyses showed coordinated upregulation of genes involved in cholesterol esterification and hydroxylation alongside repression of cholesterol biosynthetic pathways, supporting disrupted cholesterol handling and storage. Pharmacological perturbation further linked these lipid changes to ER stress: SORL1-deficient THP-1 cells exhibited heightened sensitivity to ER-Golgi trafficking stress induced by brefeldin A, while attenuation of ER stress with the chemical chaperone tauroursodeoxycholic acid (TUDCA) significantly reduced LD accumulation. Together, these data support a model in which SorLA loss promotes ER stress-associated lipid dysregulation.

Notably, despite this connection, modulation of ER stress alone failed to normalize interferon-stimulated gene expression in SorLA-deficient cells, revealing a partial dissociation between proteostatic and inflammatory stress responses. scRNA-Seq revealed a robust and persistent interferon response program in SorLA-deficient iMG that remained largely intact after ER stress attenuation and was only incompletely suppressed by inhibition of JAK-STAT signaling. These findings indicate that interferon activation is not a secondary byproduct of ER stress but rather reflects the participation of parallel stress-responsive pathways. This uncoupling is consistent with emerging evidence that chronic proteostatic imbalance, altered lipid composition [139–141], and perturbations of nuclear architecture can independently modulate innate immune signaling in myeloid cells [142, 143]. Moreover, a recent study suggested that SorLA modulates microglial inflammatory responses via CD14 trafficking [144]. Within this framework, SorLA appears to function as a broader gatekeeper of microglial stress responses, restricting inflammatory activation through mechanisms that extend beyond the maintenance of ER homeostasis.

Our data suggest that SorLA regulates ApoE handling primarily through endosomal sorting rather than uptake, such that SorLA deficiency biases ApoE trafficking toward secretion. In the context of reduced free cholesterol levels and increased cholesterol esterification observed upon SorLA loss, this altered ApoE routing may contribute indirectly to lipid dysregulation in microglia, consistent with known links between ApoE availability, endosomal function, and lipid metabolism.

Unbiased proteomic analyses identified SUN2, together with other components of the LINC complex, as high-confidence SorLA-associated proteins. We validated this association in THP-1 cells, neurons derived from iPSCs, SH-SY5Y cells, and human brain tissue, thereby pointing toward potential ER and nuclear envelope-associated functions of SorLA. More broadly, SorLA immunoprecipitation–mass spectrometry datasets were strongly enriched for ER-resident proteins, suggesting that SorLA participates in a wider ER-associated protein network in which SUN2 represents one component. In addition, fluorescence imaging studies revealed a prominent perinuclear localization of SorLA in human microglia with partial overlap or close proximity to an ER marker, while co-localization with an early endosomal marker was limited. These observations suggest that SorLA frequently engages ER or perinuclear-associated compartments in microglia, consistent with previous annotations, placing SorLA in the ER [126, 145], Golgi [146] and endosomes [125].

AlphaFold-predicted structural models of the SorLA–SUN2 complex must be interpreted with caution, particularly given the opposing membrane topologies of SorLA (Type I) and SUN2 (Type II). Accordingly, these structural predictions should be considered hypothesis-generating rather than definitive. Despite these constraints, multiple lines of evidence support the plausibility of a spatial association within the ER or perinuclear compartments. Specifically, SUN2 contains a substantial luminal low-complexity region that could become transiently accessible to the luminal VPS10 domain of SorLA during biosynthesis or early trafficking within the ER-perinuclear space continuum. AlphaFold structural predictions consistently identified convergent putative contact interfaces between these regions, which are compatible with our biochemical and proximity-based imaging data. Functionally, SorLA deficiency selectively reduced a higher molecular weight SUN2 species, while a lower molecular weight form was preserved, and lysosomal inhibition did not restore the higher band. Together, these findings argue against excessive lysosomal degradation as the sole mechanism of SUN2 loss and instead suggest that SorLA may influence SUN2 maturation, stabilization, or early intracellular handling. While the precise molecular architecture and trafficking context of the SorLA–SUN2 association require further investigation, the cellular phenotypes identified here, including ER stress activation and LD accumulation, map directly onto pathways strongly implicated in AD pathogenesis.

Targeted deletion of SUN2 phenocopied several key features of SorLA deficiency, including LD accumulation, induction of ER stress, and abnormalities in nuclear architecture. Notably, SUN2 loss produced more pronounced LD accumulation than SorLA deficiency despite eliciting a comparatively attenuated ER stress response. This divergence indicates that LD accumulation is not strictly proportional to canonical unfolded protein response activation. Given SUN2’s central role within the LINC complex, its loss may directly disrupt nuclear-cytoskeletal coupling, nuclear mechanics, or lipid regulatory pathways that impinge on LD formation independently of ER stress. Indeed, depletion of SUN1/2 has been shown to promote nuclear LD formation and intranuclear membrane expansion in other systems, suggesting SUN-mediated nuclear organization intersects with lipid regulatory mechanisms [147]. Moreover, SUN2 contains a membrane-sensing amphipathic region implicated in inner nuclear membrane retention that is sensitive to lipid composition, consistent with the idea that local lipid environments can modulate SUN2 stability and function [148]. By contrast, SorLA deficiency broadly perturbs intracellular trafficking and ER function, triggering a more robust proteostatic response while simultaneously engaging adaptive mechanisms that may partially restrain lipid accumulation. Together, these observations suggest that SUN2-dependent regulation of nuclear membrane organization may represent a parallel pathway to ER stress-driven lipid dysregulation.

Forced overexpression of SUN2 in SorLA-deficient cells partially alleviated ER stress and reduced LD burden without fully rescuing the SorLA-null phenotype, indicating that SUN2 mediates only a subset of SorLA-dependent functions. This partial rescue aligns with large-scale human proteomic data from AD cohorts, demonstrating a Braak stage-dependent correlation between SorLA and SUN2 protein abundance, with the strongest association observed at intermediate Braak stages (III–IV). A similar positive relationship was also evident in an independent orbitofrontal cohort from individuals with alcohol use disorder. Notably, prior studies [149] have reported that full-length SORL1 expression declines with increasing AD neuropathology and is already reduced at moderate Braak stages, even in individuals without overt dementia, suggesting that SorLA dysregulation may emerge during intermediate disease stages. The emergence of SorLA–SUN2 co-variation at this stage may reflect a period in which neuronal and microglial SorLA-dependent regulatory mechanisms become increasingly stressed, yet neuronal and microglial populations remain sufficiently intact to permit detection of coordinated changes in protein abundance. At earlier stages, the system is largely buffered; at later stages, widespread cell loss obscures the underlying relationship. Collectively, these findings support a model in which SorLA contributes to ER and nuclear envelope proteostasis in part through regulation of SUN2. Disruption of this axis during the mid-stages of AD progression may therefore represent a discrete point of cellular vulnerability with potential relevance to disease mechanism and therapeutic timing.

Several limitations of our study merit consideration. We mainly used SORL1 loss-of-function models, while AD-associated genetic risk more often arises from haploinsufficiency or missense variants rather than complete gene ablation. We used knockout approaches as mechanistic tools to delineate pathways sensitive to SorLA dosage and disruption, rather than to replicate the precise genetic architecture of AD. Importantly, endogenous variation in SORL1 expression within human microglia recapitulated key transcriptional signatures identified in SorLA-deficient models, supporting the relevance of these pathways in a continuum of SorLA reduction consistent with partial loss or variant-mediated dysfunction. Although our data implicate SorLA in the regulation of SUN2 abundance and ER-nuclear envelope proteostasis, the specific trafficking and maturation mechanisms involved remain unresolved, and detailed biochemical characterization of the SorLA–SUN2 association, including reciprocal immunoprecipitation and domain-level mapping, will require further investigation. While we directly assessed phagocytic uptake and recovery using multiple substrates and quantitative approaches, we did not evaluate autophagic flux or lysosomal degradation pathways, which will be important to examine in future studies to fully define downstream consequences of impaired phagocytosis. Elucidating these pathways will be essential to understand how SorLA integrates ER, lipid, and nuclear homeostasis in microglia and how disruption of this network contributes to neurodegenerative disease.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

F.S. conceptualized the study and supervised the research. F.S. served as the project administrator and secured funding. I.H., J.C.N., N.R., M.A.C., M.O., R.S., and F.S. developed the methodology. I.H., J.C.N., N.R., M.A.C., and F.S. performed experiments and data collection. A.F.T. and D.A.B. provided human tissue specimens. Y.H., X.W., M.O., and F.S. conducted project validation. I.H., J.C.N., N.R., M.A.C., Y.H., X.W., M.O., and F.S. carried out formal analysis. I.H., J.C.N., N.R., and F.S. curated the data. Y.H., M.O., A.F.T., R.P.M., P.L.D., D.A.B., and F.S. contributed to discussion and interpretation. I.H., N.R., J.C.N., M.A.C., and F.S. contributed to visualization. F.S. wrote the original draft, and all authors reviewed and edited the manuscript.

Funding

F. Sher and this study are supported by the National Institute on Aging (NIA), part of the National Health Institute (NIH) grant number R01AG070118 and the Thompson Family Foundation Program for Accelerated Medicines Exploration in Alzheimer’s Disease and Related Disorders of The Nervous System (TAME-AD).

Data availability

The processed datasets for scRNA-Seq, mass spectrometry (MS), and cytokine secretion profiles are available in the Supplementary Material. Raw scRNA-Seq and bulk RNA-Seq data have been deposited in the Gene Expression Omnibus (GEO) under accession numbers GSE244653 and GSE289806. Reviewers can access the data using the following token: yxavkgcolvqdzaz. Mass spectrometry (MS) data have been submitted to ProteomeXchange via the PRIDE database under project accession number PXD061057. The data are currently private and will be made publicly available upon publication. Reviewers can access the dataset using the following token: fDVSoaqNwTxk. All data sets related to iMG characterization such as morphology, DNA accessibility profiling (ATAC-Seq), gene expression and proteomics profiles, and immune response are accessible at: https://sherlab.shinyapps.io/IPSC-derived-Microglia/. To request ROSMAP data, please visit http://www.radc.rush.edu.

Declarations

Competing interests

The authors declare no competing interests.

Institutional Review Board statement

The research was conducted in accordance with the guidelines of the Institutional Review Board of Columbia University, New York. The human iPSC line was procured from commercial suppliers (Gibco cat. A18945). Human tissue samples of dorsolateral prefrontal cortex were sourced from Rush Alzheimer’s Disease Center’s ROSMAP cohorts’ repositories, as well as The New York Brain Bank (NYBB) at Columbia University and were utilized in accordance with standard procedures.