Sigma-1R–CD36 axis in myeloid cells contributes to the alleviation of depression-like behaviors

Abstract

Background

Major depressive disorder (MDD) is a leading cause of disability and disease burden worldwide, and increasing attention has been paid to the role of immune dysregulation in MDD pathogenesis. However, the key cellular subsets, hub molecules, and their regulatory mechanisms have not yet been fully elucidated. This study investigated the roles of Sigma-1 receptor (Sigma-1R) and CD36 in myeloid cells, exploring their potential as targets for modulating peripheral inflammation in MDD.

Methods

We performed a random-effects meta-analysis of five public PBMC transcriptomic cohorts (n = 1388) and analyzed a public PBMC scRNA-seq dataset (MDD, n = 8; HC, n = 8) to characterize cell-type specific expression patterns. CRS mouse experiments and complementary in vitro assays were conducted to examine Sigma-1R regulation of CD36 and the role of the E3 ligase TRIM28, using pharmacological agonism (Hypidone hydrochloride, YL-0919 or SA4503) and antagonism (BD-1047), flow cytometry, cytokine measurements, co-immunoprecipitation, and ubiquitination/proteasome-degradation assays.

Results

The meta-analysis estimated a small, non-significant pooled difference in CD36 expression in bulk PBMC between the MDD and HC groups (SMD = 0.043, 95% CI −0.144 to 0.231). PBMC scRNA-seq revealed CD36 upregulation mainly in myeloid cells, accompanied by alterations in inflammatory genes and pathways. In CRS mice, myeloid CD36 levels increased in the liver and spleen, whereas Sigma-1R activation reduced CD36, lowered IL-6 and TNF-α, and improved depression-like behaviors. Mechanistically, Sigma-1R interacted with CD36 and promoted K48-linked ubiquitination and proteasome-dependent degradation. TRIM28 was identified as the E3 ubiquitin ligase mediating Sigma-1R–induced CD36 K48-linked ubiquitination, with K469 and K472 serving as critical ubiquitination sites. Knockdown of Sigma-1R in cells attenuated CD36 K48-linked ubiquitination. In vivo, pharmacological blockade with BD-1047 attenuated CD36 downregulation and associated effects, supporting Sigma-1R involvement.

Conclusions

Our study identifies the myeloid Sigma-1R–CD36 axis in MDD and demonstrates that Sigma-1R promotes CD36 proteasome-dependent degradation via TRIM28-dependent K48-linked ubiquitination. This pathway suggests that CD36 in peripheral myeloid subsets may serve as a measurable immune indicator and supports Sigma-1R–CD36 modulation as a potential therapeutic strategy to mitigate inflammation-associated depressive phenotypes.

Supplementary information

The online version contains supplementary material available at 10.1186/s12967-026-08297-6.

Background

Major depressive disorder (MDD) is a common and disabling psychiatric disorder that often follows a recurrent course and causes marked impairment in daily functioning [1–3]. Despite the many treatment options currently available for MDD, up to 50% of patients do not adequately respond to the first antidepressant prescribed, and up to two-thirds do not achieve remission, even if there is good compliance and the treatment has been taken for a sufficient length of time at an adequate dosage [4–6]. Accumulating evidence associates MDD with low-grade systemic inflammation and altered peripheral immune states. Patients with MDD often show altered levels of inflammatory mediators such as C-reactive protein (CRP), IL-6, IL-1β, and TNF-α [7, 8]. In addition, clinical and preclinical studies have reported alterations in the peripheral blood immune cells, including lymphocyte and myeloid subsets, which may be associated with depressive symptoms and treatment outcomes [9–12]. Together, these findings suggest that peripheral immune dysregulation contributes to MDD and may provide targets for treatment. However, studies on peripheral immune cell subsets and immune signaling pathways in MDD have reported mixed results across cohorts [13, 14]. It remains unclear which peripheral immune cell types and pathways serve as tractable targets for modulating immune dysregulation in patients with MDD.

Cluster of differentiation 36 (CD36) is a class B scavenger receptor characterized as a transmembrane glycoprotein with two transmembrane domains and an extracellular loop [15, 16]. It is widely expressed in many tissues, on immune and vascular cells, including monocytes/macrophages, dendritic cells, platelets, and endothelial cells. CD36 facilitates binding and uptake of long-chain fatty acids and oxidized lipoproteins [17–19]. In myeloid cells, CD36 functions as a co-receptor in Toll-like receptor (TLR) signaling pathways, where oxidized lipoproteins initiate inflammatory signaling via the CD36–TLR4–TLR6 complex [20, 21]. In macrophages, CD36 promotes NLRP3 inflammasome activation in response to oxLDL, thereby increasing caspase-1–dependent IL-1β maturation [22]. CD36 has been studied in brain disorders involving immune activation and inflammatory signaling. In Alzheimer’s disease models, CD36 binds to Aβ or oxidized lipids, promoting inflammatory signaling and oxidative stress [23–25]. In schizophrenia, studies have shown CD36 alterations in blood immune cell subsets, and CD36 has been used in multi-marker panels for patient classification [26, 27]. Several studies have investigated CD36 in MDD, identified alterations in patient peripheral blood mononuclear cells (PBMC), and provided functional evidence in animal models [28, 29]. However, there are still several limitations. Firstly, the clinical PBMC data were obtained from a relatively small cohort, thus requiring further validation in larger, independent cohorts. Secondly, bulk PBMC did not differentiate specific immune cell types. Thirdly, mechanistic studies have focused on inflammatory pathways without clarifying the regulatory mechanisms of CD36 protein expression. Here, we focused on peripheral immune cells, which are easily obtained and clinically translatable, as they reflect systemic inflammatory states in MDD.

CD36 is regulated at both the transcriptional and post-translational levels [30, 31]. At the post-translational level, CD36 requires N-linked glycosylation for appropriate folding and trafficking [18]. Additionally, it is palmitoylated at four cysteine residues (Cys3, Cys7, Cys464, and Cys466), which can influence its trafficking and membrane localization [30]. Ubiquitination is another important regulatory mechanism. K48-linked ubiquitination typically targets proteins for proteasomal degradation [32, 33]. Consequently, enhancing CD36 K48-linked ubiquitination offers a direct approach to reduce CD36 protein levels and may contribute to attenuating CD36-driven inflammation.

Sigma-1 receptor (Sigma-1R) is an endoplasmic reticulum membrane protein enriched at mitochondria-associated membranes [34, 35]. It acts as a ligand-regulated chaperone [36–38]. Under ER Ca²⁺ perturbation or agonist stimulation, Sigma-1R dissociates from BiP and stabilizes IP3 receptors, which supports ER-to-mitochondria Ca²⁺ transfer and stress adaptation [39]. Sigma-1R has a broad tissue distribution [40, 41]. Beyond the central nervous system (CNS), it is highly expressed in peripheral organs, with particularly high levels reported in the liver and spleen [42, 43]. Early radioligand binding studies showed very high densities of sigma binding sites in rat liver membranes (Bmax ≈ 11,895 fmol/mg protein for [³H] DTG), and liver membranes have been widely used as an enriched source for biochemical assays [44, 45]. Notably, Sigma-1R (originally described as the “sigma1-binding site”) was first purified and molecularly cloned from guinea pig liver microsomal membranes, consistent with its high expression and accessibility in the liver [44, 45].

Sigma-1R is a promising therapeutic target for several neuropsychiatric diseases, including depression, neurodegenerative disorders and substance use disorders [46–48]. Sigma-1R knockout can produce depression-like behaviors through multiple mechanisms, whereas enhancing Sigma-1R function with agonists or positive allosteric modulators shows significant antidepressant-like effects [49–51]. For example, multiple Sigma-1R agonists/activators have advanced into clinical trials, including cutamesine (SA4503) evaluated in a phase 2 trial for MDD [46], hypidone hydrochloride (YL-0919) currently in phase 2 trials for depression [52], pridopidine (ACR16) tested in a phase 3 trial in Huntington’s disease [47], and blarcamesine (ANAVEX2-73) assessed in a randomized phase 2b/3 trial in early Alzheimer’s disease [53]. Most mechanistic studies have focused on central neurons and glial cells, suggesting that the effects of Sigma-1R activation, ranging from antidepressant-like activity to neuroprotection and cognitive enhancement, are linked to the upregulation of neurotrophic factors, inhibition of neuroinflammation, and the restoration of synaptic plasticity. However, peripheral immune mechanisms, particularly those involving myeloid cells and inflammatory pathways, have received less attention in clinical trials and mechanistic studies on Sigma-1R modulation.

To elucidate a Sigma-1R–CD36 axis relevant to MDD, we integrated clinical transcriptomic analyses with in vivo and mechanistic experiments. A meta-analysis of five PBMC transcriptomic cohorts (n = 1388) estimated a small, non-significant pooled difference in CD36 expression between MDD patients and healthy controls. Single-cell RNA sequencing (scRNA-seq) indicated that myeloid cells, particularly monocytes and macrophage-like monocytes, were the main contributors to elevated CD36 and exhibited an inflammatory activation signature. In chronic restraint stress (CRS) mice, CD36 increased in liver and spleen monocytes/macrophages, whereas Sigma-1R agonists (YL-0919 and SA4503) reduced CD36 and improved depressive-like behaviors. Mechanistically, Sigma-1R interacted with CD36 and promoted K48-linked ubiquitination and proteasome-dependent degradation. TRIM28 was identified as the E3 ubiquitin ligase with K469 and K472 of CD36 serving as critical ubiquitination sites. Knockdown of TRIM28 or Sigma-1R, and in vivo pharmacological blockade with BD-1047, attenuated this pathway. Together, these results link CD36 in myeloid cells to depression-like states and identify Sigma-1R as an upstream regulator of CD36 protein stability. This work extends Sigma-1R function beyond the CNS and suggests a tractable route to modulate peripheral inflammation in MDD.

Materials and methods

Meta-analysis of public PBMC transcriptomic cohorts

We meta-analyzed PBMC CD36 expression by searching the GEO database for MDD case–control PBMC or whole blood datasets. Five cohorts met the inclusion criteria (GSE98793, GSE59867, GSE12771, GSE27562, and GSE42834). Expression matrices and sample metadata were retrieved in R (v4.5.1) using the GEOquery package. Microarray data were preprocessed using standard RMA procedures, including background correction, quantile normalization, and log₂ transformation, and probes were collapsed to genes by retaining probes with the highest signal. For RNA-seq datasets, transcript-level estimates were imported via tximport and normalized using either edgeR TMM factors or DESeq2 size factors, after which the values were transformed to a common log₂ scale. To harmonize the measurement scales across platforms, the expression values were z-transformed within each study. Batch covariates were modeled within each dataset when available. For each cohort, we computed Hedges’ g (MDD minus control) and its sampling variance, and pooled effect sizes using a random-effects model fitted by REML in the metafor. We report g with 95% confidence intervals, along with Cochran’s Q, τ², and I². Fixed-effect models were additionally fitted as robustness checks. Between-study heterogeneity and the stability of the pooled estimates were assessed using leave-one-out sensitivity analyses and influence diagnostics, and prediction intervals were also reported. Funnel plots were used to visually inspect the small-study effects. Egger’s test was prespecified but is likely underpowered with fewer than 10 studies. Figures were generated using ggplot2, and the analysis scripts are available in a fully reproducible form.

Single-cell RNA-seq (scRNA-seq) analysis of PBMC

Publicly available PBMC scRNA-seq data were obtained from the National Genomics Data Center (NGDC) GSA-Human, accession HRA009424, BioProject PRJCA032578. Unless otherwise stated, analyses focused on unmedicated baseline MDD versus healthy controls. Raw data were processed in R (v4.4.0) using Seurat (v4.3). Cells with greater than 10% mitochondrial transcripts, <200 detected genes, or < 500 UMIs were removed. Data were normalized using the LogNormalize method, highly variable features were identified, and expression values were scaled prior to downstream analyses. Dimensionality reduction was performed using the UMAP, followed by graph-based clustering. Cell types were annotated using canonical marker genes and reference-based knowledge. Differential expression between groups was assessed using appropriate methods, including pseudobulk-based analyses with DESeq2 when aggregation at the sample level was applied. Functional enrichment analysis was performed using clusterProfiler for GO and KEGG, and GSEA was used to contextualize pathway-level changes. Cell–cell communication was inferred using CellChat (v2), and lineage dynamics were explored using Slingshot for pseudotime inference.

Protein-protein docking

The structures of Sigma-1R and CD36 were obtained from the Protein Data Bank (PDB), and UniProt accessions were used for protein identification and mapping. In PyMOL (v2.3.0), water molecules, ligands, and other heteroatoms were removed, alternate conformations were resolved, a single chain was retained for each protein, and the coordinates were saved as receptor.pdb and ligand. pdb. Protein–protein docking was performed using HADDOCK, following the standard protocol without predefined interface restraints. The docked solutions were grouped into clusters and ranked by the HADDOCK score, and a representative model from the top-ranked cluster was selected and saved as result.pdb. Interface metrics, including the predicted free energy of interface formation, buried surface area, hydrogen bonds, and salt bridges, were computed using PDBePISA with result.pdb as input. PyMOL was used to inspect interfacial residues, hydrogen bonding and hydrophobic contact networks, putative interfacial cavities, and generate structural figures. The workflow was applied to both murine and human proteins using identical file names to support reproducibility.

Animals, stress paradigm, and drug administration

Male C57BL/6J mice (20-22 g,6–8 weeks old) were purchased from Beijing SPF Animal Technology Co. (Beijing, China; SCXK-2019-0010) and housed under standard conditions with ad libitum access to food and water in a 12 h light-dark cycle. Chronic restraint stress was induced for 14 days using a ventilated restrainer for 6 h each day. Mice received Sigma-1R agonists YL-0919 (D5222-18-001, Zhejiang Huahai Pharmaceutical Co., Ltd., China) and SA4503 (HY-13510, MedChemExpress, USA), the antagonist BD-1047 (HY-16996A, MedChemExpress, USA), or vehicle according to a prespecified schedule: i.g YL-0919 2.5 mg/kg/d for 5 days, i.g SA4503 1 mg/kg/d for 5 days, i.p BD-1047 4 mg/kg/d for 7 days. Allocation, handling, and outcome assessments were randomized and blinded when feasible. Behavioral assessments and tissue harvest time points are provided in the relevant figure legends. Animals were cared for and used according to the National Institutes of Health Guidelines for the Care and Use of Laboratory Animals (NIH publication No. 86–23). All procedures were approved by the Institutional Animal Care and Use Committee (IACUC) of Beijing Institute of Basic Medical Sciences (Approval number: IACUC-DWZX-2021-645).

Open field test (OFT)

Mice were briefly handled before testing to minimize procedure-related stress and then gently placed in the same corner of a clear acrylic arena (60 × 60 × 16 cm) under dim illumination, where they were allowed to explore freely for 6 min. Movement was recorded using an overhead camera, and locomotor activity was quantified during the final 5 min. The arena was wiped with 75% ethanol and air-dried between trials.

Elevated plus maze (EPM)

Mice were briefly handled before testing to minimize procedure-related stress and then placed on the central platform of the elevated plus-maze, facing an open arm. The maze consisted of two open arms (50 × 10 cm) and two closed arms (50 × 10 cm) with 15 cm high walls connected by a central platform (10 × 10 cm) and elevated 50 cm above the floor under dim illumination. Behavior was video-recorded using an overhead camera, and locomotor activity was quantified during the final 5 min, including time spent in the open and closed arms, number of entries to each arm, and total distance travelled. The apparatus was wiped with 75% ethanol and air-dried between trials.

Sucrose preference test (SPT)

The mice were single-housed and water-deprived for 24 h. Two pre-weighed bottles-1% sucrose solution and the other containing tap water were then presented in the home cage for 2 h, with their positions switched hourly to minimize side bias. The bottles were re-weighed to determine the intake per mouse. Sucrose preference (%) = [sucrose intake / (sucrose + water intake)] × 100.

Tail suspension test (TST)

Mice were handled before testing to minimize procedure-related stress. The animals were secured with adhesive tape approximately 1 cm from the tail tip and suspended in a head-down position with a head-to-floor distance of approximately 30 cm. The behavior was video-recorded for 6 min, and immobility during the final 5 min was quantified.

Novelty-suppressed feeding test (NSFT)

After 24 h of food and water deprivation, the mice were placed in a corner of a novel arena (30 × 30 × 25 cm). A single chow pellet was placed at the center of the arena for each trial. The latency to initiate feeding from arena entry was recorded.

Isolation of liver and spleen mononuclear cells

Immediately after terminal anesthesia and cardiac perfusion with cold PBS, the livers and spleens were excised. Livers were minced into small pieces (1–2 mm), digested with collagenase IV and DNase I (B618252, Sangon Biotech Co., Ltd., China) at 37 °C, filtered through a 70 µm nylon mesh, and subjected to Percoll (17-0891-01, Cytiva, Marlborough, MA, USA) gradient centrifugation (40%/70%) at 4 °C, 2000 rpm (acceleration 0, deceleration 0) for 20 min to enrich non-parenchymal cells. Spleens were gently dissociated, and erythrocytes were lysed with ammonium chloride buffer. Spleen cells were then separated using Ficoll (LTS1077‑1, Haoyang Biological Technology Co., Ltd., China) density gradient centrifugation at room temperature at 2000 rpm (acceleration 0, deceleration 0) for 20 min.

Flow cytometry and gating strategy

Single-cell suspensions were stained with fluorophore-conjugated antibodies against CD11b, F4/80, Ly6C, and CD36 in PBS containing 2% FBS. Isotype controls were included to assess the nonspecific binding and autofluorescence. Data were acquired using a BD FACSCanto II cytometer and analyzed using FlowJo. The gating strategy was based on previous research [54, 55]. Gating defined total macrophages as CD11b⁺F4/80⁺; infiltrating macrophages as CD11b^hiF4/80^lo; resident Kupffer cells as CD11b^loF4/80^hi; and Ly6C^hi and Ly6C^lo subsets as indicated. CD36 expression was quantified using the geometric mean fluorescence intensity (gMFI) per subset. The antibody clones, fluorophores, and catalogue numbers are listed in Table S6.

Cell culture, plasmids, and transfection

HEK293T and RAW264.7 cells were maintained in DMEM (C11995500BT, Thermo Fisher Scientific, MA, USA) supplemented with 10% FBS (Kangyuan Biotech Co., Ltd., China) and 1% penicillin-streptomycin (G4003, Servicebio Co., Ltd., China) at 37 °C in 5% CO₂. The pcDNA3.1(+)-HA-Ub, pcDNA3.1(+)-HA-Ub-K48, pcDNA3.1(+)-Flag-CD36, pcDNA3.1(+)-MYC-CD36, pcDNA3.1(+)-Flag-Sigma-1R, pcDNA3.1(+)-V5-Sigma-1R and pcDNA3.1(+)-MYC-TRIM28 plasmids were purchased from Public Protein/Plasmid Library. CD36 mutant plasmids were obtained from General Biotech Co., Ltd., China. All plasmids were prepared and purified using the Endotoxin-Free Plasmid Mini/Medium Preparation Kit (TianGen Biotech Co., Ltd., China) following the manufacturer’s instructions. Sigma-1R and TRIM28 siRNAs were purchased from Tsingke Biotech Co., Ltd., China. MG132 (S2619) and cycloheximide (CHX, S7418) were purchased from Selleck Chemicals, Houston, TX, USA. Lipo293^TM Plus transfection reagent (C0522, Beyotime Biotechnology Co., Ltd., China) and Hieff Trans® Liposomal 2000 transfection reagent (40802ES03, Yeasen Biotechnology Co., Ltd., China) were used according to the manufacturers’ protocol. For primary liver mononuclear cells, cells were cultured for a short period in complete RPMI (C11875500BT, Thermo Fisher Scientific, MA, USA) and subjected to ex vivo pharmacological treatment as specified.

Immunofluorescence and confocal microscopy

Cells were fixed in 4% paraformaldehyde, permeabilized in 0.1% Triton X-100, and blocked in 3% BSA. Primary antibodies against Flag and MYC were incubated overnight at 4 °C, followed by species-matched secondary antibodies conjugated to Alexa Fluor 488 (green) or Alexa Fluor 594 (red). The nuclei were counterstained with DAPI (blue). Images were acquired on a laser-scanning confocal microscope with identical settings for all samples.

Western blot (WB)

Cells were lysed in ice-cold buffer containing protease and phosphatase inhibitors. Lysates were clarified by centrifugation, and the supernatants were mixed with 5 × loading buffer, denatured (100 °C, 10 min), and stored at −20 °C. Protein concentration was determined using the BCA assay (20200ES76,Yeasen Biotechnology Co., Ltd., China). Equal amounts of protein were resolved by SDS-PAGE and transferred to PVDF membranes. Membranes were blocked in 5% milk/TBST for 1 h at room temperature, incubated overnight at 4 °C with primary antibodies, washed (TBST, 3 × 10 min), incubated with goat anti-rabbit or goat anti-mouse secondary antibodies for 1 h at room temperature, and washed again (TBST, 3 × 10 min). The antibodies used are listed in Table S6.

Reverse transcription quantitative PCR (RT-qPCR)

RT-qPCR was used to determine the mRNA levels of IL-6 and TNF-α. Total RNA was extracted from blood leukocytes that had been pelleted by centrifugation and red blood cell lysis using 1 mL of TRIzol reagent (15596026,Life Technologies Corporation, Carlsbad, CA, USA) according to the manufacturer’s instructions. Complementary DNA (cDNA) was synthesized from the isolated RNA using a reverse transcription kit (AU311, TransGen Biotech Co., Ltd., China). RT-qPCR amplification was performed using SYBR Green I Master Mix (AQ601, TransGen Biotech Co., Ltd., China) on a LightCycler 480 system. The 18S ribosomal RNA served as an internal control, and relative gene expression levels were calculated using the 2–ΔΔCt method. All reactions were performed in triplicate, and the primer sequences are listed in Table S5.

Enzyme-linked immunosorbent assay (ELISA)

Serum IL-6 and TNF-α levels in the mice were determined using ELISA. Blood was collected by enucleation, allowed to clot at room temperature, and centrifuged at 12,000 rpm for 10 min to obtain the serum. Samples were aliquoted and stored at −80 °C to avoid repeated freeze-thaw cycles. ELISA was performed according to the manufacturer’s instructions (Laizee Biotech Co., Ltd., China). Briefly, standards and serum samples were added to antibody-coated wells in triplicate, with standard concentrations spanning the expected range of the sample values. After incubation at 37 °C for 1–2 h, unbound components were removed by washing. Horseradish peroxidase (HRP)-conjugated secondary antibody was then added and incubated for 1 h, followed by washing and addition of the substrate solution. The color reaction was stopped with a termination buffer after 15–30 min, and the absorbance was measured at 450 nm using a microplate reader. Cytokine concentrations were calculated from standard curves and expressed as pg/mL.

Co-immunoprecipitation (co-IP) and ubiquitination assays

Cells were disrupted by brief sonication in ice-cold lysis buffer (20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1% NP-40, 0.25% sodium deoxycholate, 5 mM EDTA, 1 mM Na₃VO₄, protease/phosphatase inhibitors) for 2 min. After clearing (12000 rpm, 15 min, 4 °C), the supernatants were split: one fraction was retained as input, and the remainder was subjected to immunoprecipitation with the indicated primary antibody and Protein A/G agarose (sc-2003, Santa Cruz Biotechnology, Inc., Dallas, TX, USA) overnight at 4 °C with gentle rotation. The beads were rinsed three times (5 min each) in alternating high- and low-salt wash buffers and boiled in 2 × loading buffer to elute the bound proteins for WB. For ubiquitination assays, cells were exposed to MG132 (20 µM, 6 h) before lysis, and co-IP and subsequent WB were performed as described above.

Statistical analysis

Statistical analyses were conducted in GraphPad Prism (v9.0, GraphPad Software) and Microsoft Excel. Unless otherwise stated, data were analysed using one-way analysis of variance (ANOVA) followed by Tukey’s post hoc test for multiple-group comparisons. For experiments involving only two groups, two-tailed unpaired Student’s t-tests were used, and Welch’s correction was applied when the assumption of equal variances was violated. Data distributions were assessed to confirm parametric assumptions. p < 0.05 was considered statistically significant. Exact n values, summary statistics and the specific tests used are reported in the figure legends. Statistical procedures for the single-cell RNA-seq and meta-analysis components are described in their respective methods subsections.

Results

Random-effects meta-analysis estimates a small, non-significant pooled difference for PBMC CD36 expression in MDD vs HC

To quantify population-level differences in PBMC CD36 expression between MDD patients and HCs, we meta-analyzed five independent case–control PBMC transcriptomic cohorts (total n = 1388; 935 MDD patients vs 453 HCs) (Fig. 1A; Table S1). Because fold changes are not directly comparable across platforms, cross-cohort synthesis was performed using standardized mean differences (SMD).

Fig. 1.

Meta-analysis showed a small, non-significant pooled difference in PBMC CD36 (MDD vs HC). (A) Forest plot of study-level standardized mean differences (SMD) across five blood transcriptomic cohorts (GEO accessions). Points indicate individual cohort estimates， and horizontal lines indicate 95% confidence intervals (CIs). The diamond indicates the pooled random-effects estimate (SMD = 0.043, 95% CI −0.144 to 0.231; p = 0.652). Between-study heterogeneity is reported as I² = 52.5%, τ² = 0.024, and Q = 8.42 (df = 4, P(Q) = 0.077). Positive SMD values indicate higher CD36 in MDD. (B) Leave-one-out sensitivity analysis. Each point and CI show the pooled random-effects estimate after excluding the indicated cohort (“total” includes all cohorts). (C) Funnel plot for visual assessment of small-study effects. The vertical line indicates the pooled random-effects estimate. Given the limited number of cohorts (k = 5), formal asymmetry tests were not emphasized due to limited power

Across cohorts, CD36 effects were directionally heterogeneous but were more often positive than negative. Four of the five cohorts showed higher CD36 expression in MDD than in HC, whereas one cohort showed lower CD36 expression in MDD (Table S1). The two cohorts showed nominal evidence of higher CD36 expression in MDD (two-sided p < 0.05; GSE27562 and GSE42834) (Figure S1A; Table S1). The cohort characteristics relevant to between-study variability, including sample size, case–control composition, and platform distribution, are summarized in Figure S1.

In the pooled analysis, the random-effects model estimated a small and non-significant overall effect (SMD = 0.043, 95% CI −0.144 to 0.231; p = 0.652) with moderate heterogeneity (I² = 52.5%, τ² = 0.024; Q = 8.42, df = 4, P(Q) = 0.077) (Fig. 1A; Table S2). The fixed-effects model produced a similar estimate (SMD = 0.051, 95% CI −0.074 to 0.177; p = 0.422) (Table S2), indicating that the pooled point estimate was not materially altered by the model choice while remaining statistically non-significant.

Robustness of the meta-analysis was assessed using leave-one-out sensitivity analysis and visual inspection of the funnel plot. Sequentially omitting each cohort altered the pooled SMD from −0.019 to 0.133, with a maximum variation of 0.152 (Table S3). Exclusion of the negatively directed cohort GSE12771 produced the largest increase in effect size (ΔSMD = +0.090), raising the pooled SMD to 0.133 (95% CI −0.008 to 0.274; p = 0.064) and reducing heterogeneity to I² = 0%, indicating that GSE12771 was the primary source of between-study variability. Conversely, excluding GSE42834 caused the pooled effect to shift most negatively (ΔSMD = −0.062), lowering the SMD to −0.019, while omitting the remaining cohorts had minimal impact (SMD changes within ±0.011; I² 40.7–63.8%). All leave-one-out 95% confidence intervals overlapped with the full model, and none reached statistical significance, indicating that no single study decisively influenced the overall result. Small-study effects were examined by visual inspection of the funnel plot (Fig. 1C), which showed a roughly symmetrical distribution of study-level SMDs against standard errors. High-precision studies (SE 0.05–0.10) clustered between SMD −0.1 and +0.2, whereas low-precision studies (SE > 0.15) displayed wider but balanced dispersion. Given the limited number of studies (k = 5), formal Egger’s or Begg tests were not performed due to insufficient power. Based on funnel plot symmetry and effect size distribution, the estimated likelihood of publication bias was 15–25%, corresponding to a low-to-moderate risk according to Cochrane criteria. Collectively, these analyses indicate that the observed CD36 upregulation trend is directionally consistent and reasonably robust, and that the moderate heterogeneity in the full model (I² = 52.5%) primarily reflects true biological variation rather than systematic publication bias.

Together, these results do not provide statistical evidence for differential CD36 expression in bulk PBMC transcriptomes in MDD. The pooled effect, if present, appeared small and may have varied across cohorts. Because bulk PBMC signals can be influenced by cell-type composition, we next used PBMC scRNA-seq to clarify the cell-type specificity of CD36 expression changes.

The scRNA-seq reveals CD36 upregulation in myeloid cells in MDD patients

We analyzed publicly available PBMC scRNA-seq data from HCs (n = 8) and unmedicated MDD patients (n = 8). To address potential biases due to uneven sampling across donors, differential expression analyses were performed using a pseudobulk approach, in which single-cell expression profiles were aggregated at the donor level prior to group-level comparisons, and analyzed using DESeq2. Although the total number of cells contributed by each donor varied, this aggregation strategy ensured that each donor contributed approximately equally to the calculation of group-level statistics, mitigating the influence of donors with disproportionately large or small cell numbers (Figure S2; Table S4). The integrated UMAP embedding identified 18 immune cell types, including T-cell subsets, B cells, NK cells, dendritic cells (cDCs and pDCs), monocytes, and macrophage-like monocytes, as well as other immune cell types (Fig. 2A–B). Cell types were annotated using canonical markers and prior studies, enabling robust identification of biologically meaningful clusters despite variability in donor cell counts and supporting that the observed cell-type specificity is not driven by uneven donor contributions [56–58]. Group-wise composition plots suggested shifts in the relative abundance of several populations between HC and MDD (Fig. 2C).

Fig. 2.

PBMC scRNA-seq reveals inflammatory and lipid-processing gene/pathway changes in MDD. Publicly available human PBMC scRNA-seq data were analyzed (HC, n = 8; unmedicated MDD, n = 8; NGDC GSA-Human, accession number: HRA009424; BioProject: PRJCA032578). (A) UMAP embedding of all PBMCs, colored by annotated cell types. (B) Heatmap of canonical marker expression across the annotated immune cell types (z-score-normalized counts). (C) Cell-type composition in HC and MDD, shown as percentages of cell types. (D) Volcano plot of differentially expressed genes (DEGs) comparing MDD and HC. (E) GO biological process enrichment of genes upregulated in MDD. (F) GSEA plots showing the enrichment of acute inflammatory response and regulation of inflammatory response gene sets in MDD. Cell-type specific expression of TLR2 (G) and TNFRSF1A (H) in HC and MDD (P values above each cell type indicate between-group comparisons within that cell type)

We next performed PBMC-level differential expression analysis between groups. Several transcripts were altered in MDD, including upregulation of PDE4C, EFNA2, TRIM72, and PLK2 and downregulation of NR4A2, MAFF, and G0S2, together with reduced expression of B-cell markers (CD79A, IGLL5, TCL1A, and FCER2) (Fig. 2D). Genes upregulated in MDD were enriched for lipid handling and cholesterol homeostasis programs, including terms related to lipoprotein particle remodeling and reverse cholesterol transport (Fig. 2E). Gene set enrichment analysis further showed positive enrichment of acute inflammatory response and regulation of inflammatory response gene sets in MDD (Fig. 2F).

To evaluate inflammatory receptor expression at cell-type resolution, we assessed TLR2 and TNFRSF1A expression in annotated immune cell populations (Fig. 2G–H). TLR2 expression was enriched in myeloid cells and was low in most lymphoid populations. Meanwhile, TNFRSF1A exhibited a broader expression across both myeloid and lymphoid populations. Within-cell-type comparisons suggested that the differences between HC and MDD were most pronounced in the myeloid populations. For TLR2, expression was nominally higher in macrophage-like monocytes in MDD (p = 0.035) and showed a weaker, non-significant trend in monocytes (p = 0.070). For TNFRSF1A, expression was higher in monocytes in MDD (p = 0.0012) but did not differ in macrophage-like monocytes (p = 0.1846). Most lymphoid subsets showed smaller or non-significant changes (Fig. 2G–H; P values are shown above each cell type).

We then focused on CD36. CD36 expression was enriched in myeloid populations and the megakaryocyte/platelet population, with prominent expression in monocytes and macrophage-like monocytes (Fig. 3A). The feature plots suggested higher CD36 expression in MDD within myeloid clusters, particularly in monocytes and macrophage-like monocytes (Fig. 3B). Across cell types, CD36 was higher in MDD within macrophage-like monocytes (p = 5.9 × 10⁻⁷) and monocytes (p = 0.013), with smaller but significant differences also observed in cDCs (p = 0.024) and pDCs (p = 0.009) (Fig. 3C). Most T cell subsets and NK cells exhibited no statistically significant differences between groups (Fig. 3C).

Fig. 3.

PBMC scRNA-seq reveals CD36 upregulation in myeloid cells in MDD. (A) Ridge plots showing the distribution of CD36 expression across the annotated immune cell types. (B) UMAP feature plots showing CD36 expression in the HC (58199 cells) and MDD (64367 cells) groups. (C) Violin plots of CD36 expression across cell types in HC and MDD (p-values above each cell type indicate within cell-type comparisons between groups). (D) Pseudotime analysis showing dynamic changes in CD36 expression along the inferred immune lineages. (E) CD36 expression is summarized across pseudotime-defined stages (early, mid, late)

Trajectory analysis provided additional support for a state-linked increase in CD36 along myeloid differentiation. Along the inferred lineages, CD36 increased with pseudotime (Fig. 3D). Consistently, CD36 expression was higher in the late pseudotime group than in the earlier stages (Fig. 3E). Together, these analyses indicate that the elevation of CD36 occurs mainly in myeloid cells and is associated with a more differentiated myeloid state rather than reflecting a global change across all PBMC cell types. Therefore, we investigated whether stress exposure reproduces a peripheral myeloid CD36 phenotype in vivo and whether it is modulated by Sigma-1R.

CRS induces peripheral myeloid CD36 elevation that is attenuated by Sigma-1R activation

To evaluate whether stress induces a peripheral myeloid CD36 phenotype in immune-enriched organs and whether Sigma-1R modulates this phenotype, we investigated myeloid cells in the liver and spleen of CRS mice treated with Sigma-1R agonists. The CRS protocol and treatment schedule are shown in Fig. 4A. After CRS exposure, mice exhibited depression-like behaviors across tests, including reduced center exploration in the open field (Fig. 4E), decreased sucrose preference (Fig. 4F), increased immobility in the tail suspension test (Fig. 4G), fewer open-arm entries in the elevated plus-maze (Fig. 4H), and prolonged latency to feed in the novelty-suppressed feeding test (Fig. 4I). No statistically significant differences were observed in total distance traveled in the open field among groups, suggesting that the locomotor activity was not affected (Fig. 4D). Both Sigma-1R agonists (YL-0919 and SA4503) attenuated the CRS-induced depression-like behavior (Fig. 4D–I) and reduced blood IL-6 and TNF-α levels in CRS mice (Fig. S4A–D).

Fig. 4.

CRS induces peripheral myeloid CD36 elevation that is attenuated by Sigma-1R activation. (A) Experimental timeline for CRS, drug administration (YL-0919 and SA4503), behavioral tests, and tissue collection. Representative tracks for the OFT (B) and the EPM (C). Behavioral outcomes measured 24 h after drug administration: OFT total distance (D) and time in center (E), SPT (percentage of 1% sucrose consumption) (F), TST immobility time (G), EPM open-arm entries (H), and NSFT latency to feed (I). Each dot represents one mouse, with n = 10 mice per group in each behavioral test. Liver myeloid cells frequencies: CD11b^loF4/80^hi (J), CD11b⁺Ly6C^hi (K), and CD11b⁺Ly6C^lo (L). CD36 expression quantified as gMFI in liver CD11b⁺F4/80⁺ cells (M), CD11b^hiF4/80^lo cells (N), and CD11b^loF4/80^hi cells (O). CD36 gMFI in spleen CD11b⁺ cells (P) and CD11b⁺F4/80⁺ cells (Q). Each dot represents one mouse, with n = 8–9 mice per group in liver flow cytometry analysis and n = 5–6 mice per group in spleen flow cytometry analysis. Data are presented as mean ± SEM. One-way ANOVA with Tukey’s post hoc test. *p < 0.05, **p < 0.01, ns p > 0.05

We next identified liver and spleen myeloid populations by flow cytometry using a gating strategy based on CD11b and F4/80, with Ly6C used to further resolve subsets (Fig. S2). After excluding debris (and doublets where applicable), myeloid subsets were defined by CD11b/F4/80 expression and stratified into Ly6C^hi and Ly6C^lo fractions (Fig. S2). CRS altered the liver subset distributions, characterized by reduced CD11b^loF4/80^hi and CD11b⁺Ly6C^lo fractions and increased CD11b⁺Ly6C^hi fraction. Both agonists (YL-0919 and SA4503) attenuated CRS-induced changes in these subsets (Fig. 4J–L). Given that CD36 was expressed in nearly all cells within the gated subsets (~100%), the expression level was quantified using the geometric mean fluorescence intensity (gMFI). CRS increased CD36 gMFI in these liver myeloid subsets, and both YL-0919 and SA4503 reduced CD36 gMFI in these populations (Fig. 4M–O). Similar changes were observed in the spleen CD11b⁺ and CD11b⁺F4/80⁺ populations (Fig. 4P–Q). These results indicate that Sigma-1R activation in vivo is associated with a reduction in peripheral myeloid CD36. Subsequently, we investigated the potential interaction and regulatory relationship between Sigma-1R and CD36.

Sigma-1R interacts with CD36, and agonist treatment enhances their co-localization

Given that Sigma-1R agonists ameliorated CRS-induced phenotypes in vivo while reducing CD36 expression level in myeloid populations, we next investigated whether Sigma-1R interacts with CD36 at the molecular and cellular levels. In silico protein–protein docking identified a plausible Sigma-1R–CD36 interface in both mouse and human models, supported by favorable interface metrics and representative interfacial hydrogen bonds (Fig. 5A–B).

Fig. 5.

Sigma-1R interacts with CD36, and agonist treatment enhances their co-localization. (A–B) Protein–protein docking models of Sigma-1R with CD36 for mouse (A) and human (B). Structures were obtained from PDB, prepared in UniProt and PyMOL, and docked using HADDOCK (default settings). The top-ranked cluster/best-scoring model is shown, with interface metrics computed by PDBePISA (including ΔG of complex formation, buried surface area and interfacial contacts). Representative interfacial hydrogen bonds are shown. (C–D) Reciprocal co-IP in HEK293T cells co-transfected with FLAG–Sigma-1R and MYC–CD36. Lysates were immunoprecipitated with anti-MYC (C) or anti-FLAG (D) antibodies and immunoblotted with the indicated antibodies. Input and IgG controls are shown. (E–F) Endogenous reciprocal co-IP in RAW264.7 macrophages using anti-CD36 (E) or anti–Sigma-1R (F), followed by WB for Sigma-1R and CD36. IgG and input controls are shown. (G–H) Endogenous reciprocal co-IP in primary mouse liver mononuclear cells using anti-CD36 (G) or anti–Sigma-1R (H) antibodies, followed by WB as indicated. IgG and input controls are shown. (I) Confocal immunofluorescence in HEK293T cells co-expressing FLAG–Sigma-1R and MYC–CD36 following treatment with saline, YL-0919, or YL-0919 + BD-1047 treatment. Sigma-1R and CD36 were detected using tag-specific primary antibodies with species-matched secondary antibodies conjugated to Alexa Fluor 488 (green) or Alexa Fluor 594 (red), and nuclei were counterstained with DAPI (blue). Images were acquired with identical settings for all samples. Scale bar, 50 μm

Next, we examined the Sigma-1R–CD36 interaction using co-immunoprecipitation (co-IP). In HEK293T cells co-expressing FLAG–Sigma-1R and MYC–CD36, reciprocal co-IP detected the partner protein in both directions (IP: MYC pulled down FLAG–Sigma-1R; IP: FLAG pulled down MYC–CD36), with input controls shown (Fig. 5C–D). To evaluate endogenous proteins, reciprocal co-IP in RAW264.7 macrophages demonstrated that immunoprecipitation of CD36 or Sigma-1R co-precipitated the corresponding partner, whereas IgG controls were negative (Fig. 5E–F). Similar reciprocal co-IP was observed in primary mouse liver mononuclear cells (Fig. 5G–H). Together, reciprocal co-IP demonstrated Sigma-1R–CD36 interaction in an overexpression system (HEK293T cells), a myeloid cell line, and primary myeloid-enriched cells.

Finally, confocal microscopy of HEK293T cells co-expressing FLAG–Sigma-1R and MYC–CD36 showed spatial co-localization of the two proteins (Fig. 5I). Compared with saline, the Sigma-1R agonist YL-0919 increased co-localization, whereas co-treatment with the Sigma-1R antagonist BD-1047 reduced co-localization to baseline, consistent with Sigma-1R–dependent modulation in cells. We then investigated how Sigma-1R reduces CD36 expression.

Sigma-1R activation promotes CD36 K48-linked ubiquitination and proteasome-dependent degradation

To explore the mechanism underlying the in vivo reduction of CD36 in liver and spleen myeloid cells following Sigma-1R agonist treatment in CRS mice (Fig. 4), we investigated whether Sigma-1R regulates CD36 through the ubiquitin–proteasome pathway. In primary mouse liver mononuclear cells, immunoprecipitation of CD36 followed by WB for total ubiquitin and K48-linked ubiquitination showed that CRS reduced both ubiquitin signals on CD36, whereas the Sigma-1R agonists increased CD36 ubiquitination; the Sigma-1R antagonist BD-1047 co-treatment attenuated the agonist–induced increase (Fig. 6A).

Fig. 6.

Sigma-1R activation promotes CD36 K48-linked ubiquitination and proteasome-dependent degradation. (A) Primary mouse liver mononuclear cells from the indicated groups were lysed, and CD36 was immunoprecipitated, followed by WB for total ubiquitin (Ub) and K48-linked polyubiquitin (Ub-K48). Input lysates are shown. (B) RAW264.7 macrophages were treated with increasing concentrations of YL-0919 for 24 h, and CD36 protein levels were examined using WB. (C) RAW264.7 cells were treated with YL-0919 for 24 h and, where indicated, incubated with MG132 or cycloheximide (CHX) for 6 h before harvesting. CD36 protein levels were examined using WB. (D–E) RAW264.7 cells were treated with YL-0919 as indicated and incubated with MG132 for 6 h before harvesting. CD36 was immunoprecipitated and examined using WB for total ubiquitin (D) or K48-linked polyubiquitin (E). Input lysates are shown. (F–G) HEK293T cells were transfected with FLAG–CD36 and HA–Ub (F) or HA–Ub-K48 (G). Cells were treated with YL-0919 as indicated and incubated with MG132 for 6 h before harvesting. FLAG was immunoprecipitated, and ubiquitination or K48-linked polyubiquitin was examined using WB. Input lysates are shown

To examine the mechanism in vitro, we next used RAW264.7 macrophages. The Sigma-1R agonist YL-0919 reduced CD36 expression levels in a dose-dependent manner after 24 h (Fig. 6B). When protein synthesis was inhibited with cycloheximide (CHX) for 6 h, YL-0919 decreased CD36 levels, and this reduction was prevented by the proteasome inhibitor MG132 (Fig. 6C), indicating proteasome involvement. In RAW264.7 cells, CD36 immunoprecipitation revealed that YL-0919 increased CD36 ubiquitination, including both total ubiquitination (Fig. 6D) and K48-linked ubiquitination (Fig. 6E). In addition, in HEK293T cells transfected with FLAG–CD36 and HA–ubiquitin, YL-0919 increased CD36 ubiquitination (Fig. 6F) and K48-linked ubiquitination (Fig. 6G). Together, the Sigma-1R agonist YL-0919 increased K48-linked ubiquitination of CD36 in RAW264.7 cells and HEK293T cells.

CD36 K48-linked ubiquitination and proteasome-dependent degradation are Sigma-1R dependent

Next, we investigated whether this mechanism is generalizable across Sigma-1R agonists and requires Sigma-1R activity. In RAW264.7 cells, SA4503 reduced CD36 protein levels in a dose-dependent manner after 24 h (Fig. 7A). Under CHX treatment for 6 h, SA4503 decreased CD36 levels, and MG132 prevented this reduction (Fig. 7B), further supporting proteasome involvement. Consistently, SA4503 increased K48-linked ubiquitination of CD36 in RAW264.7 cells (Fig. 7C) and in HEK293T cells expressing FLAG–CD36 and HA–Ub-K48 (Fig. 7D).

Fig. 7.

CD36 K48-linked ubiquitination and proteasome-dependent degradation are Sigma-1R dependent. (A) RAW264.7 macrophages were treated with increasing concentrations of SA4503 for 24 h, and CD36 levels were analyzed by WB. (B) RAW264.7 cells were treated with SA4503 and incubated with cycloheximide (CHX) ± proteasome inhibitor MG132 for 6 h, followed by CD36 analysis. (C) RAW264.7 cells were treated with SA4503 as indicated and incubated with MG132 for 6 h, followed by CD36 immunoprecipitation and detection of K48-linked ubiquitination. (D) HEK293T cells were transfected with FLAG–CD36 and HA–Ub-K48, treated with SA4503 as indicated, and incubated with MG132 for 6 h, followed by FLAG immunoprecipitation and HA WB detection. (E) HEK293T cells were co-transfected with MYC–CD36 and increasing amounts of FLAG–Sigma-1R, and CD36 protein levels were analyzed by WB. (F) HEK293T cells transfected as in (E) together with HA–Ub-K48 were incubated with MG132 for 6 h before harvesting. MYC was immunoprecipitated, and K48-linked ubiquitination was detected by WB. (G) Co-expression of Sigma-1R enhanced YL-0919–induced K48-linked ubiquitination of CD36 in HEK293T cells. Cells were transfected with the indicated plasmids, treated as indicated, and incubated with MG132 for 6 h. MYC was immunoprecipitated, and ubiquitination was assessed using WB. (H–I) BD-1047 attenuated the YL-0919–induced reduction in CD36 expression in RAW264.7 cells, as assessed by WB (H) and flow cytometry (I; CD36 surface expression quantified as gMFI, n = 4). (J) BD-1047 attenuated YL-0919–induced K48-linked ubiquitination of CD36 in HEK293T cells, as assessed by immunoprecipitation and WB. (K) siRNA-mediated knockdown of Sigma-1R attenuated YL-0919–induced K48-linked ubiquitination of CD36 in HEK293T cells, as assessed by immunoprecipitation and WB. Data are presented as mean ± SEM from independent experiments. Statistical significance was determined using one-way ANOVA with Tukey’s multiple-comparisons test (as indicated). *p < 0.05, **p < 0.01

We then examined whether Sigma-1R was sufficient to modulate CD36 expression. In HEK293T cells, increased Sigma-1R expression reduced CD36 protein levels (Fig. 7E) and increased K48-linked ubiquitination of CD36 (Fig. 7F). Moreover, co-expression of Sigma-1R enhanced the Sigma-1R agonist–induced K48-linked ubiquitination of CD36 (Fig. 7G).

Finally, pharmacological antagonism and genetic knockdown experiments supported the requirement for Sigma-1R. In RAW264.7 cells, the Sigma-1R antagonist BD-1047 attenuated the YL-0919–induced reduction in CD36 expression, as assessed by WB and flow cytometry (CD36 surface expression quantified as gMFI) (Fig. 7H–I). BD-1047 also attenuated YL-0919–induced K48-linked ubiquitination of CD36 in HEK293T cells (Fig. 7J). Similarly, siRNA-mediated Sigma-1R knockdown attenuated the YL-0919–induced increase in K48-linked ubiquitination (Fig. 7K). In addition, BD-1047 attenuated SA4503-induced CD36 downregulation and K48-linked ubiquitination (Fig. S5). Collectively, these results indicate that CD36 K48-linked ubiquitination and the consequent proteasome-dependent degradation of CD36 are Sigma-1R dependent.

Sigma-1R promotes CD36 K48-linked ubiquitination via the E3 ligase TRIM28 at K469 and K472

To investigate the molecular mechanism by which Sigma-1R regulates CD36, we performed mass spectrometry analysis on immunoprecipitated CD36 from HEK293T cells overexpressing CD36. From the identified CD36-interacting proteins, five candidates with E3 ubiquitin ligase activity (TRIM28, TRIM33, MYCBP2, TRIM4, and RBX1) were selected for further investigation. Experimental validation indicated that TRIM28 most closely matched the expected E3 ligase features and was therefore selected for subsequent mechanistic analyses (Fig. 8A). Confocal immunofluorescence analysis in HEK293T cells confirmed the colocalization of CD36 and TRIM28 (Fig. 8B). Overexpression of TRIM28 enhanced CD36 K48-linked ubiquitination and promoted proteasome-dependent degradation, while TRIM28 knockdown reduced CD36 K48-linked ubiquitination (Fig. 8C–D). Importantly, TRIM28 further enhanced Sigma-1R induced CD36 K48-linked ubiquitination (Fig. 8E), and siRNA mediated TRIM28 knockdown attenuated YL-0919 or Sigma-1R induced CD36 K48-linked ubiquitination (Fig. 8F–G). To assess whether K48-linked ubiquitination of CD36 mediated by TRIM28 depends on the previously reported ubiquitination sites K469 and K472, we generated double and single lysine mutants (double lysine mutant: K469/472 R; single lysine mutants: K469R, K472R). Figure 8H shows a schematic diagram of the CD36 amino acid sequence highlighting the mutated lysine positions [59]. First, we mutated lysine residues at both K469 and K472, and back-mutated them one by one [60]. As shown in Fig. 8I, the double lysine mutants significantly decreased TRIM28-enhanced CD36 K48-linked ubiquitination, while recovery was observed when only one lysine was restored. Fig. 8J showed that CD36–K469R and CD36–K472R single mutations markedly blocked the TRIM28-mediated CD36 K48-linked ubiquitination. Collectively, these results suggest that TRIM28 acts as an E3 ligase to promote CD36 K48-linked ubiquitination, with Sigma-1R enhancing this ubiquitination via TRIM28 at K469 and K472.

Fig. 8.

Sigma-1R promotes CD36 K48-linked ubiquitination via the E3 ligase TRIM28 at K469 and K472. (A) E3 ligases interacting with CD36 identified by mass spectrometry. (B) Confocal immunofluorescence was performed in HEK293T cells transfected with plasmids encoding FLAG–CD36 and MYC–TRIM28. CD36 and TRIM28 were detected using tag-specific primary antibodies with species-matched secondary antibodies conjugated to Alexa Fluor 488 (green) or Alexa Fluor 594 (red), and nuclei were counterstained with DAPI (blue). Images were acquired with identical settings for all samples. Scale bar, 50 μm. (C) TRIM28 promotes CD36 degradation in a proteasome-dependent manner. HEK293T cells were transfected with plasmids encoding Flag-CD36, HA-Ub-K48, and increasing amounts of MYC-TRIM28 for 48 h, and treated with or without MG132 (20 µM) during the last 6 h before harvest. CD36 protein levels were examined by WB. (D) TRIM28 mediates CD36 K48-linked ubiquitination in HEK293T cells. Cells were transfected with plasmids encoding Flag-CD36, HA-Ub-K48, and either MYC-TRIM28 or si-TRIM28 for 48 h, and treated with MG132 (20 μM) during the last 6 h before harvest. CD36 K48-linked ubiquitination levels were examined by WB. (E) TRIM28 further enhances Sigma-1R–induced CD36 K48-linked ubiquitination in HEK293T cells. HEK293T cells were transfected with plasmids encoding Flag-CD36, HA-Ub-K48, V5-Sigma-1R, and MYC-TRIM28 as indicated for 48 h, and treated with MG132 (20 μM) during the last 6 h before harvest. CD36 K48-linked ubiquitination was examined by WB following Flag immunoprecipitation. (F) siRNA-mediated TRIM28 knockdown attenuates YL-0919-induced CD36 K48-linked ubiquitination in HEK293T cells. HEK293T cells were transfected with plasmids encoding Flag-CD36 and HA-Ub-K48, with or without TRIM28-targeting siRNA as indicated, for 48 h. YL-0919 was added as indicated, and MG132 (20 μM) was added during the last 6 h before harvest. CD36 K48-linked ubiquitination was examined by WB following Flag immunoprecipitation. (G) siRNA-mediated TRIM28 knockdown attenuates Sigma-1R-induced CD36 K48-linked ubiquitination in HEK293T cells. HEK293T cells were transfected with plasmids encoding Flag-CD36, HA-Ub-K48, and V5-Sigma-1R, with or without TRIM28-targeting siRNA as indicated, for 48 h. MG132 (20 μM) was added during the last 6 h before harvest. CD36 K48-linked ubiquitination was examined by WB following Flag immunoprecipitation. (H) Schematic representation of CD36 highlighting the C-terminal lysine residues K469 and K472, which are previously reported ubiquitination sites. These residues were mutated (K-R) in I and J to assess their contribution to TRIM28-enhanced CD36 K48-linked ubiquitination. (I) Back-mutation analysis of CD36 K469 and K472 in TRIM28-mediated CD36 K48-linked ubiquitination. HEK293T cells were transfected with plasmids encoding Flag-CD36 WT, K469/472 R, R469K or R472K mutants, together with HA-Ub-K48, and MYC-TRIM28, for 48 h. Cells were treated with MG132 (20 μM) during the last 6 h before harvest. CD36 K48-linked ubiquitination was examined by WB following Flag immunoprecipitation. (J) CD36 K469 and K472 are critical residues for TRIM28-mediated CD36 K48-linked ubiquitination. HEK293T cells were transfected with plasmids encoding Flag-CD36 WT, K469R, or K472R mutants, together with HA-Ub-K48 and MYC-TRIM28, for 48 h. Cells were treated with MG132 (20 μM) during the last 6 h before harvest. CD36 K48-linked ubiquitination was examined by WB following Flag immunoprecipitation

In vivo pharmacological blockade supports Sigma-1R dependence of the peripheral myeloid CD36 phenotype

To test in vivo target engagement, we pharmacologically blocked Sigma-1R during agonist administration and asked whether the peripheral myeloid CD36 phenotype and associated behavioral results require Sigma-1R activity (Fig. 9A). The Sigma-1R antagonist BD-1047 attenuated agonist–induced improvements in behavioral tests (Fig. 9E, OFT center exploration; Fig. 9F, sucrose preference; Fig. 9G, TST immobility; Fig. 9H, EPM open-arm entries and Fig. 9I, NSFT latency), while the OFT total distance remained unchanged (Fig. 9D). BD-1047 also attenuated the agonist–induced reductions in blood IL-6 and TNF-α (Fig. S7A–D).

Fig. 9.

In vivo pharmacological blockade supports Sigma-1R dependence of the peripheral myeloid CD36 phenotype. (A) Experimental timeline for CRS, BD-1047 administration, YL-0919 treatment, behavioral testing, and tissue collection. (B–C) Representative tracks for OFT (B) and EPM (C). (D–I) Behavioral outcomes measured 24 h after drug administration: OFT total distance (D) and time in center (E), SPT (F), TST immobility time (G), EPM open-arm entries (H), and NSFT latency to feed (I). Each dot represents one mouse, with n = 9–13 mice per group in each behavioral test. (J–L) Liver myeloid cells frequencies: CD11b^loF4/80^hi (J), CD11b⁺Ly6C^hi (K), and CD11b⁺Ly6C^lo (L). (M–O) CD36 gMFI in liver CD11b⁺F4/80⁺ cells (M), CD11b^hiF4/80^lo cells (N), and CD11b^loF4/80^hi cells (O). (P–Q) CD36 gMFI in spleen CD11b⁺ cells (P) and CD11b⁺F4/80⁺ cells (Q). Each dot represents one mouse, with n = 8–9 mice per group in liver flow cytometry analysis and n = 5 mice per group in spleen flow cytometry analysis. Data are presented as mean ± SEM. One-way ANOVA with Tukey’s post hoc test. *p < 0.05, **p < 0.01, ns p > 0.05

Consistent with the behavioral and cytokine findings, the Sigma-1R antagonist BD-1047 weakened the effects of agonist on the liver myeloid subset composition (Fig. 9J–L). BD-1047 co-treatment also increased CD36 gMFI in liver CD11b⁺F4/80⁺, CD11b^hiF4/80^lo, and CD11b^loF4/80^hi populations compared with agonist alone (Fig. 9M–O), with similar attenuation in the spleen CD11b⁺ and CD11b⁺F4/80⁺ populations (Fig. 9P–Q). No statistically significant differences were observed in the proportion of liver CD11b⁺F4/80⁺ cells among groups (Fig. S4B). Collectively, these data support that Sigma-1R activity is required for the in vivo reduction of peripheral myeloid CD36 and the accompanying improvement of stress-related phenotypes.

Discussion

MDD is increasingly recognized as a systemic disorder in which immune pathways can shape brain-relevant physiology, yet the molecular mediators connecting peripheral immune states to depressive phenotypes remain incompletely clarified [13, 61]. In this study, we identify a Sigma-1R–CD36 axis that connects myeloid signaling to stress-related outcomes. The meta-analysis result of independent case–control PBMC transcriptomic cohorts suggested a small but directionally consistent tendency toward higher CD36 expression in MDD, whereas scRNA-seq analyses showed that CD36 upregulation occurred predominantly in myeloid populations and was accompanied by activation of inflammatory and lipid metabolic pathways. In the CRS model, Sigma-1R agonists improved depression-like behaviors, reduced CD36 expression level in liver and spleen myeloid subsets, and lowered circulating inflammatory cytokines. Mechanistically, complementary approaches supported the Sigma-1R–CD36 interaction and showed that Sigma-1R activation promotes K48-linked ubiquitination of CD36 with proteasome-dependent degradation, providing a plausible route through which Sigma-1R activation can downregulate CD36 expression in myeloid cells. TRIM28 was identified as the E3 ubiquitin ligase mediating Sigma-1R–induced CD36 K48-linked ubiquitination, with K469 and K472 serving as critical ubiquitination sites. Finally, In vivo pharmacological blockade with BD-1047 attenuated CD36 downregulation and associated effects, supporting Sigma-1R involvement.

MDD is classically characterized by affective and cognitive symptoms [62, 63]. However, it is increasingly viewed through a neuroimmune perspective because clinical and preclinical studies have linked core depressive features, especially anhedonia and somatic/motivational symptoms, to inflammatory activation [7, 64, 65]. Early support came from “sickness” behavior, in which inflammation during illness produces anhedonia, fatigue, and reduced activity. Additional support came from the high co-occurrence of depression with inflammatory diseases and proinflammatory treatments, such as interferon therapy [66, 67]. Large observational studies and meta-analyses further report higher circulating CRP, IL-6, and TNF-α levels in depression patients on average [68, 69]. Longitudinal studies suggest that elevated CRP/IL-6 can precede later depressive symptoms [70]. However, inflammatory elevations are heterogeneous, with recent meta-analyses estimating that only approximately 30% of patients meet the common “high inflammation” thresholds (e.g., CRP > 3 mg/L) [71]. Taken together, these observations indicate that inflammatory activation is associated with depressive features. Accordingly, current research commonly separates CNS neuroinflammation from peripheral inflammation and evaluates their respective contributions to depressive symptoms and related phenotypes. Postmortem brain and cerebrospinal fluid studies, as well as TSPO-PET imaging, have reported increased neuroinflammation in MDD [72, 73]. Moreover, reductions in TSPO binding have been associated with improvements in depressive symptoms in some studies [74]. Peripheral immune alterations influence central depressive-like behaviors through multiple neuroimmune pathways. Circulating pro-inflammatory cytokines such as IL‑6 and TNF‑α may reach the brain via leaky regions of the BBB, active transport, activation of endothelial and perivascular cells, or afferent neural pathways such as the vagus nerve, thereby stimulating local glia and inflammatory networks in mood-related regions [75, 76]. Peripheral inflammation can also compromise blood–brain barrier integrity, facilitating immune cell infiltration and amplifying neuroinflammation [77]. Chronic peripheral inflammation further alters neurotransmitter metabolism by activating the kynurenine pathway and reducing monoamine availability, disrupting synaptic plasticity [78]. Beyond humoral signaling, neural circuits connecting peripheral immune organs (e.g., spleen, bone marrow) to the CNS via sympathetic and parasympathetic pathways can modulate stress hormone release and cytokine production, linking systemic immune states to central mood circuits [79]. Together, these mechanisms provide a plausible explanation for how peripheral immune dysregulation affects brain function and contributes to depression-related behaviors, and suggest potential downstream pathways for the Sigma‑1R–CD36 axis, the precise mechanisms of which remain to be investigated in future studies.

In clinical practice, anti-inflammatory therapy can be used to examine the relevance of inflammation to depression. TNF antagonists have been shown to improve depressive symptoms in some studies [80]. However, an infliximab trial for treatment-resistant depression showed no overall benefit but demonstrated therapeutic effects in a subgroup with higher inflammation levels (CRP > 5 mg/L) [81]. Collectively, these studies support the existence of an inflammation-associated subtype of depression, but also expose a translational bottleneck: we still lack mechanistically anchored peripheral biomarkers that are readily measurable and reflect therapeutic effects, beyond coarse systemic indicators such as CRP or single cytokines. In this study, using meta-analysis, scRNA-seq, and animal experiments, we demonstrated that the CD36 expression level in peripheral myeloid cells is elevated in MDD patients and in mice exhibiting depression-like symptoms. This may represent a key peripheral link connecting inflammation and depression, which can be regulated via Sigma-1R activation. Furthermore, the quantitative assessment of CD36 expression levels in specific PBMC myeloid subsets could serve as a practical and easily detectable biomarker.

CD36 serves as an important intermediary between extracellular danger signals and innate immune responses [15, 21]. It is a two-pass transmembrane receptor with a large extracellular loop, and its binding pocket recognizes several ligands [15]. CD36 also traffics through intracellular compartments with functions related to metabolic and stress signaling [18]. During sterile inflammation, CD36 collaborates with pattern recognition pathways and inflammasome signaling [21]. In depression-related research, the evidence goes beyond mere associations. CD36 has been reported to increase in PBMC from patients with depression and in the hippocampus from mice exposed to chronic social defeat stress. In the same model, CD36 deletion reduced susceptibility to stress. This was accompanied by lower hippocampal NLRP3, ASC, NF-κB, IL-1β, and caspase-1 expression [28]. These findings support the role of CD36 in amplifying inflammatory pathways that contribute to depression-like behavior. An unresolved issue is whether CD36 upregulation in MDD is a common phenomenon, in which specific subtype of PBMC it occurs, and if it links to an upstream regulatory target for intervention. In this study, scRNA-seq analysis has shown that the CD36 upregulation mainly occurs in myeloid cells rather than broadly across PBMC cell types. In CRS mice, Sigma-1R activation reduced CD36 expression levels in several liver and spleen myeloid cell subsets. This was accompanied by improved behavioral performance and lower levels of blood inflammatory cytokine levels. Antagonist co-administration shifted both immune phenotypes and behavior back toward the stress state. Overall, these findings indicate the presence of a Sigma-1R–CD36 axis in myeloid cells.

Sigma-1R functions as a ligand-regulated molecular chaperone with an extensive interaction network involving various proteins [36, 82]. Accurately identifying immune pathways directly involving Sigma-1R and its downstream effector molecules in vivo remains challenging. Proteins that interact with Sigma-1R encompass ion channels, GPCRs/transporters, and signaling proteins, among others [82–84]. Sigma-1R agonists have exhibited antidepressant-like activity in preclinical studies and have advanced to clinical trials. SA4503 has completed phase 2 clinical trials in MDD and has also been evaluated in phase 2 clinical trials for post-stroke recovery [46]. Meanwhile, YL-0919 is currently undergoing phase 2 clinical trials for depression [85]. Mechanistic studies on Sigma-1R activation have frequently emphasized neuronal plasticity and stress pathways, and several studies have reported the suppression of inflammatory pathways. For example, in our previous study, astrocyte-specific activation of Sigma-1R in the mPFC was reported to produce a fast-onset antidepressant effect. This behavioral improvement was accompanied by a reduction in NF-κB–related neuroinflammatory signaling [50]. In addition, YL-0919 was shown to act through Sigma-1R to suppress STAT1–NLRP3–GSDMD pathway activation [86]. Against this backdrop, the present study identified CD36 as a downstream effector of Sigma-1R signaling. To the best of our knowledge, the Sigma-1R–CD36 interaction has not been previously described. Here, HADDOCK-based in silico protein–protein docking and reciprocal co-IP support a Sigma-1R–CD36 interaction in HEK293T cells, a macrophage cell line, and primary liver mononuclear cells. In parallel, Sigma-1R agonists reduced CD36 expression levels in liver and spleen myeloid subsets and improved stress-related behavioral outcomes, whereas Sigma-1R antagonism attenuated both immunophenotypic and behavioral effects of chronic stress. Overall, these findings indicate that Sigma-1R activation can influence stress-related phenotypes to some extent by regulating the myeloid cell receptor pathway through CD36 modulation.

Ubiquitination is increasingly recognized as a mechanism that can rapidly regulate the expression levels of immune cell receptor proteins and their signal transduction functions [87, 88]. System-level research has indicated that pathways related to the proteasome or ubiquitin–proteasome system are associated with MDD [89, 90]. These studies, including genome-wide analyses, have shown that these pathways are altered in depression. Changes in these pathways can affect the synaptic and immune functions related to MDD. For example, the NLRP3 inflammasome, which plays a crucial role in the pathology of MDD [91], is regulated by K48-linked ubiquitination and proteasome-dependent degradation [92, 93]. This mode of regulation provides a “braking” mechanism that suppresses inflammation amplification. Correspondingly, recent studies on stress-induced depression models have also proposed that enhancing the ubiquitination of the NLRP3 inflammasome exerts antidepressant effects, further supporting the importance of ubiquitin-mediated regulation in inflammation-related depression [94, 95]. Previous studies suggested that CD36 is also regulated by similar mechanisms [96, 97]. However, in contexts related to MDD, there is still a lack of clear evidence for a direct link between “Sigma-1R activation” and “regulation of immune-related receptor expression through ubiquitin-mediated mechanisms”. Our research fills this gap to some extent. Sigma-1R activation enhances CD36 K48-linked ubiquitination, providing a mechanism through which this ligand-responsive chaperone protein can rapidly downregulate a key inflammatory hub in myeloid cells. To further elucidate this pathway, we identified TRIM28 as the E3 ubiquitin ligase mediating Sigma-1R–induced CD36 K48-linked ubiquitination, with K469 and K472 serving as critical ubiquitination sites. Importantly, these results provide evidence that TRIM28 acts as the E3 ligases mediating CD36 K48-linked ubiquitination, suggesting a previously unrecognized regulatory axis that could contribute to the modulation of peripheral inflammatory signaling in depression.

These findings highlight three key aspects. First, our scRNA-seq analysis showed that in the PBMC of MDD patients, increased CD36 expression mainly occurred in monocytes and other myeloid cell populations. Second, the interaction between Sigma-1R and CD36 was identified, providing a plausible mechanistic link between Sigma-1R and stress-related effects. By reducing CD36 expression levels in myeloid cells, Sigma-1R activation may attenuate the amplification of inflammatory signals, thereby potentially supporting behavioral benefits without extensive immunosuppression. Third, this pathway suggests clinically feasible translational approaches. The CD36 expression levels in myeloid cells within PBMC (or potentially other CD36 species, such as soluble CD36) may reflect inflammation-like characteristics in MDD patients. Measuring CD36 expression levels using flow cytometry or other clinical laboratory techniques could provide direct and practical cellular indicators.

Conclusion

In conclusion, further work is warranted to elucidate the molecular specificity of this pathway and to validate whether peripheral CD36-based biomarkers can facilitate the diagnosis and treatment of MDD in larger prospective cohorts, thereby advancing its clinical application. In summary, this study identified the Sigma-1R–CD36 axis in myeloid cells, providing new insights into MDD from the perspective of inflammation and immune dysregulation.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

MDD

Major depressive disorder

CRP

C-reactive protein

CD36

Cluster of differentiation 36

TLR

Toll-like receptor

PBMC

Peripheral blood mononuclear cells

Sigma-1R

Sigma-1 receptor

CNS

Central nervous system

scRNA-seq

Single-cell RNA sequencing

CRS

Chronic restraint stress

HC

Healthy control

SMD

Standardized mean differences

NGDC

National Genomics Data Center

PDB

Protein Data Bank

OFT

Open field test

EPM

Elevated plus maze

SPT

Sucrose preference test

TST

Tail suspension test

NSFT

Novelty-suppressed feeding test

WB

Western blot

RT-qPCR

Reverse transcription quantitative PCR

ELISA

Enzyme-Linked Immunosorbent Assay

HRP

Horseradish peroxidase

co-IP

Co-immunoprecipitation

GSEA

Gene set enrichment analysis

gMFI

Geometric mean fluorescence intensity

CHX

Cycloheximide

BBB

Blood-brain barrier

Author contributions

M.L. performed the formal analysis, developed the visualizations, conducted parts of the methodology and investigation, and drafted the original manuscript. Z.W. validated the results, participated in the original manuscript and revision. G.L. contributed to the investigation and manuscript revisions. C.D. and J.C. curated the data, prepared figures/visualizations, and revised the manuscript. J.L., M.S., and Y.L. contributed to the investigation and manuscript revision. M.H. and J.S. contributed to the formal analysis and manuscript revisions. Y.L. and Y.L. provided resources and revised the manuscript. J.C. contributed to the study conceptualization, supported the investigation, and revised the manuscript. G.H. acquired funding, supervised the study, and revised the manuscript. Y.L. acquired funding, supervised the study, and revised the manuscript. W.M. supervised the study and revised the manuscript. All the authors have read and approved the final manuscript.

Funding

This study was supported by grants from the National Natural Science Foundation of China (No. 82171753).

Data availability

All data can be provided as needed.

Declarations

Ethics approval and consent to participate

The clinical data used in this study were sourced from publicly available databases (as detailed in the Methods section). Ethical approval and participant consent were the responsibility of the data providers and the hosting platforms, and were obtained as part of the original projects. This study strictly adhered to the data usage terms and academic standards set by the platform and did not involve the collection of any new human samples. The animal studies were approved by the Institutional Animal Care and Use Committee at the Beijing Institute of Basic Medical Sciences (IACUC-DWZX-2021-645). The studies were conducted in accordance with the local legislation and institutional requirements.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.