HMGB1-mediated endoplasmic reticulum stress in monocyte-derived dendritic cells regulates CD4+ T cell immune responses in myasthenia gravis Authors: Qing Zhang, Zhongya Gu, Yaxuan Li, Mengge Yang, Xue Ma, Xuanxuan Pan, Zhuajin Bi, Jing Lin, Mengcui Gui, Furong Wang, Min Zhang, Zhijun Li, Bitao Bu Published: 2026 ABSTRACT AbstractIntroductionMyasthenia gravis (MG) is an autoimmune disorder characterized by immune dysregulation at the neuromuscular junction. Monocyte-derived dendritic cells (moDCs) are increasingly recognized as key drivers of MG pathogenesis; however, the mechanisms that govern their dysfunction remain incompletely understood.MethodsThis study integrated human genetics, patient immunophenotyping, and therapeutic evaluation in the experimental autoimmune myasthenia gravis (EAMG) model. Monocytes were isolated and differentiated into moDCs to assess the effects of HMGB1 and its inhibitor, 18α-glycyrrhetinic acid (18α-GA), on the expression profiles, phenotypes, and functions of moDCs. Therapeutic efficacy was further assessed in EAMG using 18α-GA, HMGB1 neutralizing antibody, and adoptive transfer of 18α-GA-induced tolerogenic moDCs.ResultsMendelian randomization (MR) analyses revealed bidirectional causality between MG and moDCs. MG patients exhibited elevated plasma HMGB1 levels and mature phenotypes of moDCs. Mechanistically, HMGB1 activated the IRE1α/XBP1 and NF-κB pathways in moDCs through the engagement of TLR4, thereby inducing endoplasmic reticulum (ER) stress, promoting the activation of moDCs, and driving an imbalance in CD4+ T cell immune responses. Through a multitargeted mechanism of action, 18α-GA effectively inhibited both the IRE1α/XBP1 and NF-κB pathways in moDCs, thereby blunting ER stress and driving the differentiation of moDCs toward tolerogenic phenotypes. The minimal effective dose of 18α-GA alleviated EAMG by counteracting HMGB1-driven, moDC-mediated CD4+ T cell dysfunction, and showed better therapeutic performance compared with HMGB1 neutralizing antibody. Moreover, the adoptive transfer of 18α-GA-induced tolerogenic moDCs was effective in treating EAMG.ConclusionsHMGB1-mediated ER stress reprogrammed moDCs via the IRE1α/XBP1 and NF-κB pathways to promote pathogenic CD4+ T cell responses in MG. Targeting the HMGB1-ER stress axis in moDCs could restore immune balance and represent a promising therapeutic strategy for MG.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12974-026-03908-x.Keywords: Myasthenia gravis, HMGB1, Endoplasmic reticulum stress, Monocyte-derived dendritic cells, CD4+ T cells Introduction Myasthenia gravis (MG) is an autoimmune disease characterized by muscle weakness and fatigue, which significantly impairs the daily lives of patients [1, 2]. MG occurs due to complex immune dysregulation, with autoantibodies and the complement cascade wrongly attacking the neuromuscular junctions (NMJs) [3]. Accumulating evidence suggests that dendritic cells (DCs) play a pivotal role in regulating T cell-mediated adaptive immunity and are involved in the immunopathogenesis of MG [4]. DCs, the most potent professional antigen-presenting cells in the immune system, exhibit unparalleled efficiency in antigen uptake, phagocytosis, and proteolytic processing during the pathogenesis of autoimmune diseases [5, 6]. DCs in the thymus and peripheral blood specifically recognize and present AChR (acetylcholine receptor) proteins, thereby driving the production of pathogenic autoantibodies in MG [4, 7]. Mature DCs exhibit a 100 to 1000-fold greater capacity to activate naïve T cells as compared with macrophages or B cells [8]. Monocyte-derived dendritic cells (moDCs), also known as inflammatory DCs, are essential for infections and autoimmune diseases [9, 10]. The regulatory networks governing the heterogeneity and functional plasticity of moDCs in MG remain poorly defined, warranting further mechanistic exploration. High mobility group box 1 protein (HMGB1) is a nuclear DNA-binding protein that is involved in various intracellular biological processes, including DNA replication, repair, transcription, and translation [11]. Endoplasmic reticulum (ER) stress constitutes an adaptive protective cellular stress response that functions to attenuate the accumulation of unfolded proteins in the cytosol, thereby averting their pathological aggregation [12]. Excess ER stress induced stronger activation of moDCs and then programmed the moDCs to direct distinct CD4+ T cell fates [9, 13]. HMGB1 promoted the maturation and activation of DCs and mediated the tissue damage by inducing the release of inflammatory cytokines from DCs [14]. Elevated plasma HMGB1 levels were correlated with mature moDCs in patients with graft vs. host disease and pulmonary fibrosis [15, 16]. However, the regulatory effects of HMGB1 on moDCs remain unclear in MG. In this study, we explored immune cell signatures using MR (Mendelian randomization) analysis and identified the phenotype and molecular profiles of circulating DCs in MG patients. Transcriptome analysis and in vitro functional experiments were conducted to investigate the molecular networks of human moDCs and their influence on CD4+ T cell responses. Animal experiments explored the efficacy of HMGB1 inhibitor 18α-glycyrrhetinic acid (18α-GA) and its regulation of moDCs, offering novel insights into MG pathogenesis and treatment. Materials and methods MR analysis Summary statistics of single-nucleotide polymorphisms (SNPs) for immune cells and MG were retrieved from IEU Open GWAS project (https://gwas.mrcieu.ac.uk/datasets/). The summarized data of 731 immunophenotypes from 3,757 European individuals are referenced under accession numbers GCST90001391 to GCST90002121 [17]. The GWAS ID for MG is GCST90018656, including 81 cases and 178,630 controls from East Asians. The inverse-variance weighted method was employed as the primary MR analysis. SNPs significantly associated with exposures or the outcome were selected as instrumental variables with a significance threshold of P value < 5 × 10− 6. We conducted pleiotropy, heterogeneity tests, and sensitivity analysis to ensure the robustness of our findings. Participants and sample collection A total of 32 MG patients (who were admitted to the Department of Neurology, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology from December 2023 to September 2025) and 20 healthy controls (HCs) were enrolled in the study. Participant demographics are detailed in Table 1. The inclusion criteria included the following: (1) clinically confirmed MG diagnosis exhibiting typical fatigable weakness with seropositivity for AChR antibody (AChR-Ab) [18]; (2) immunotherapy-naïve patients or those with discontinued corticosteroid/immunosuppressive therapies ≥ 6 months before enrollment [19]; (3) participants aged between 18 and 75 years. Exclusion criteria were malignancies (except thymoma), other autoimmune diseases, or severe infectious diseases. Table 1Characteristics of subjects enrolled in the studyAll casesHCs (n = 20)MG (n = 32) P Age, y, mean ± SD55.2 ± 13.5651.27 ± 15.380.354Sex, n (%) Female11 (55.0%)19 (59.4%)0.756 Male9 (45.0%)13 (40.6%)Symptoms at onset, n (%) Ocular MG26 (81.3%)− Generalized MG6 (18.7%)Disease duration, mo, median [IQR]3.00 [1.00–18.00]−MGFA classification, n (%) I10 (31.3%)− II- III22 (68.7%)Imaging diagnosis for thymus, n (%) Normal19 (59.4%)− Hyperplasia5 (15.6%) Thymoma8 (25.0%)MG subgroups, n (%) EOMG12 (37.5%)− LOMG20 (62.5%)AChR-Ab (nmol/L), median [IQR]7.93 [3.79–9.95]−Repetitive nerve stimulation, n (%) Positive23 (71.9%)− Negative9 (28.1%)Treatment, n (%) Pyridostigmine10 (31.3%)− Drug-free22 (68.7%)Abbreviations: MG myasthenia gravis, HCs healthy controls, EOMG early-onset MG (onset > 18 years and ≤ 50 years), LOMG late-onset MG (onset > 50 years), AChR-Ab acetylcholine receptor antibody, MGFA Myasthenia Gravis Foundation of America Characteristics of subjects enrolled in the study Abbreviations: MG myasthenia gravis, HCs healthy controls, EOMG early-onset MG (onset > 18 years and ≤ 50 years), LOMG late-onset MG (onset > 50 years), AChR-Ab acetylcholine receptor antibody, MGFA Myasthenia Gravis Foundation of America Peripheral blood mononuclear cells (PBMCs) were obtained via gradient centrifugation using Lymphoprep (Catalog No. 07801, STEMCELL Technologies, Canada). Isolated PBMCs were either used immediately or cryopreserved in liquid nitrogen until further use [20]. Thawed PBMCs were rested overnight as previously described [21]. Plasma and sera were separated from the blood samples and stored at − 80 °C until analysis. EAMG model Female C57BL/6 mice (6–8 weeks old, weighing 16–20 g, Guangdong Yaokang Biotechnology Co., Ltd., Guangzhou, China) were housed under specific pathogen-free conditions. To induce the EAMG model, the Torpedo AchRα 146–162 peptide (LGIWTYDGTKVSISPES, Qiangyao Biotechnology Co., Ltd., Shanghai, China) was utilized [22]. On day 1, each EAMG mouse received 50 µg of Torpedo AchRα 146–162 peptide emulsified in 100 µL of complete Freund’s adjuvant (CFA; Cat. No. GC207003, Servicebio, Wuhan, China) and 100 µL PBS containing 1 mg of Mycobacterium tuberculosis strain H37Ra (Cat. No. 231141, Difco Laboratories, Detroit, USA) subcutaneously (sc). On day 29, the EAMG mice received 50 µg of Torpedo AchRα 146–162 peptide emulsified in 100 µL of incomplete Freund’s adjuvant (IFA; Cat. No. GC207004, Servicebio, Wuhan, China) and 100 µL PBS. Mice in the Control group were immunized with the same mixed emulsion lacking the peptide. To determine the minimum effective dose (MED), EAMG mice were administered 18α-GA (Cat. No. HY-N0375, MedChemExpress, Princeton, USA) at 0, 0.25, 0.5, 1, or 2 mg/kg intraperitoneally (ip) every other day (q2d) starting on day 30. 18α-GA was dissolved in the vehicle (5% Tween-80, 10% DMSO, 40% PEG300, and 45% saline). On day 58, mouse sera were collected before sacrifice for liver and kidney function and electrolytes using Automatic Biochemical Analyzer (Chemray240 and Chemray800, Shenzhen Ruiduo Life Analysis Science Co., Ltd., Shenzhen, China). The MED was defined as the lowest dose that resulted in statistically significant reductions in clinical scores and Torpedo AchRα 146–162-specific IgG levels, as well as a significant increase in grip strength, compared with the Control group at the final observation. Clinical scores of EAMG were evaluated according to previously reported methods [22]. Blood samples were collected at 0.5, 2, 4, 8, 12, 24, and 48 h after the last administration of 0.5 mg/kg 18α-GA. The high-performance liquid chromatography (HPLC) system (Thermo U3000, Thermo Fisher Scientific, Waltham, USA) was used for the chromatographic separation of the 18α-GA. The pharmacokinetic parameters were analyzed using DAS 2.0 (Mathematical Pharmacology Professional Committee of China, Shanghai, China). This dose-finding and safety evaluation study was performed only once, with 8 mice in each group. The following two therapeutic studies were each independently repeated twice using separate animal cohorts, with 8 mice per group in each cohort. Each cohort was conducted on different dates using the same predefined protocol. Mice were randomized in each cohort, and all endpoint analyses were conducted in a blinded manner with unknown group assignments. Body weights, grip strength, clinical scores, and AchRα 146–162-specific IgG levels were recorded at prespecified time points, while terminal tissues were harvested at endpoints. Four additional groups were set up for the intervention-control study: Control, EAMG, EAMG + 18α-GA, and EAMG + HMGB1 neutralizing antibody (Cat. No. ARG66714, Arigo, Hsinchu, China). All interventions were administered from day 30, and the detailed protocols are presented in Fig. S1A. In the Control group, mice received vehicle (ip, q2d) plus PBS intravenously (iv) twice weekly (q2w); in the EAMG group, mice received vehicle plus isotype control IgG (Cat. No. ARG66747, Arigo, Hsinchu, China) (5 mg/kg, iv, q2w); in the EAMG + 18α-GA group, mice received 18α-GA (0.5 mg/kg, ip, q2d) plus isotype control IgG; and in the EAMG + HMGB1 neutralizing antibody group, mice received vehicle plus HMGB1 neutralizing antibody (5 mg/kg, iv, q2w) on the same schedule. To investigate the therapeutic efficacy of tolerogenic moDCs induced by 18α-GA (18α-GA moDCs) against EAMG, mice were randomly assigned to the EAMG group and the EAMG + 18α-GA moDCs group. The detailed experimental design is illustrated in Fig. S1B. Tolerogenic moDCs (8 × 10⁵ cells/mouse) suspended in 0.1 mL PBS were adoptively transferred into mice in the treatment group from day 30 (iv, once weekly after the model establishment [23]). Mice in the EAMG group were administered an equal volume of PBS (0.1 mL) as the vehicle control. All mice were sacrificed on day 58 for sample collection. Electromyography (EMG) detection of EAMG Mice were anesthetized with sodium pentobarbital, then fixed and connected to an EMG instrument (NDI-200PC, Shanghai Haishen Medical Electronic Instrument Co. Ltd., Shanghai, China). The stimulating electrode was inserted into the sciatic notch, the negative electrode into the abdominal wall, and the recording electrode into the gastrocnemius muscle of the ipsilateral leg. The evoked muscle action responses were recorded after applying 8 supramaximal stimuli at 3 Hz to the sciatic nerve. Each mouse underwent 2 rounds of repetitive nerve stimulation (RNS), and the decrement percentage was calculated by comparing the first and fifth evoked responses as described previously [24]. DCs evaluated by flow cytometry Cell viability was determined by measuring Fixable Viability Stain 780 (Catalog No. 565388, BD Pharmingen, San Diego, USA), followed by all surface and intracellular staining. To detect phenotypes of human peripheral blood DCs, surface proteins were stained with the: CD14 Alexa Fluor 700 (Clone M5E2, Catalog No. 557923)/CD16 FITC (Clone 3G8, Catalog No. 555406) (BD Pharmingen, San Diego, USA) and CD141 PE (Clone M80, Catalog No. 344103)/CD209 PerCP-Cy5.5 (Clone 9E9A8, Catalog No. 330109)/CD11c PE-Cy7 (Clone Bu15, Catalog No. 337215)/Lineage (Lin; CD3/CD19/CD56 APC (Clone OKT3/HIB19/5.1H11, Catalog No. 317317/302211/362503))/CD123 BV421 (Clone 6H6, Catalog No. 306017)/HLA-DR BV 510 (Clone L243, Catalog No. 307645)/CD1c BV605 (Clone L161, Catalog No. 331537)/CD86 BV650 (Clone IT2.2, Catalog No. 305427)/CD83 BV785 (Clone HB15e, Catalog No.305337) (BioLegend, San Diego, USA). Human blood DCs were classified into four subsets according to a previous report [25]: pDCs (plasmacytoid DCs, Lin−CD14−CD16−HLA-DR+CD11c−CD123+), cDC1s (classical type 1 DCs, Lin−CD14−CD16−HLA-DR+CD11c+CD141+), cDC2s (classical type 2 DCs, Lin−CD14−CD16−HLA-DR+CD11c+CD1c+), and moDCs (Lin−CD14+CD209+HLA-DR+CD11c). The corresponding gating strategy is depicted in Fig. S2. Mouse splenic DCs were performed surface staining with: CD8a FITC (Clone 53-6.7, Catalog No. 100705)/CD11c PE (Clone N418, Catalog No. 117307)/CD86 PerCP (Clone GL-1, Catalog No.105025)/CD64 PE-Cy7 (Clone W18349C, Catalog No. 164413)/Lin (CD3/CD19/NK1.1) APC (Clone 17A2/6D5/PK136, Catalog No. 100235/115511/108709)/B220 Alexa Fluor 700(Clone RA3-6B2, Catalog No. 103231)/CD83 BV421 (Clone HB15e, Catalog No. 305323)/CD11b BV510 (Clone M1/70, Catalog No. 101245)/MHC-II BV605 (Clone M5/114.15.2, Catalog No. 107639)/CD45 BV785 (Clone 30-F11, Catalog No. 103149) (BioLegend, San Diego, USA), and CD209a BV650 (Clone 5H10, Catalog No. 740621, BD Pharmingen, San Diego, USA). Mouse DCs are conventionally classified into pDCs ( Lin−CD45+CD64−MHC-II+CD11cdimB220+), cDC1s (Lin−CD45+CD64−MHC-II+CD11c+B220−CD8a+), cDC2s (Lin−CD45+CD64−MHC-II+CD11c+B220−CD11b+), and moDCs (Lin−CD45+CD64+CD209a+MHC-II+CD11c) by referring to the previous study [26], with the detailed gating strategy illustrated in Fig. S3. For ex vivo induction of IFN-α in pDCs, 1 × 106 SMCs were stimulated with 1 µM CpG ODN 2216 (Catalog No. tlrl-2216, InvivoGen, San Diego, CA, USA) and 1 mL GolgiPlug (Catalog No. 555029, BD Pharmingen, San Diego, USA) as previously described [27]. SMCs were then stained antibodies against surface markers, including CD11c PE, CD64 PE-Cy7, Lin (CD3/CD19/NK1.1) APC, B220 Alexa Fluor 700, MHC-II BV605, and CD45 BV785. Fixation and permeabilization were performed using the Cytofix/Cytoperm Soln kit (Catalog No. 554715, BD Pharmingen, San Diego, USA). Intercellular proteins were stained with the IFN-α FITC (Clone RMMA-1, Catalog No. 22100-3, PBL Assay Science, Piscataway, USA). MoDC differentiation Human monocytes were positively isolated from fresh PBMCs with CD14 Microbeads (Catalog No. 130-050-201, Miltenyi Biotec, Bergisch Gladbach, Germany). Purity of CD14+ monocytes after sorting was above 90% (Fig. S4A). The sorted cells were cultured in complete RPMI-1640 medium and stimulated with 40 ng/mL human recombinant GM-CSF (Cat. No. AF-300-03-50UG, PeproTech, New Jersey, USA) and 40 ng/mL human recombinant IL-4 (Cat. No. 200-04-50UG, PeproTech, New Jersey, USA) [4]. On day 6, immature human moDCs were treated with different concentrations of HMGB1 (Cat. No. 1690-HMB-050, R&D Systems, Minnesota, USA). Apoptosis assay was performed by Annexin V-FITC/PI-PE apoptosis detection kit (Catalog No. 40302ES20, Yeasen, Shanghai, China). 4µ8C (Catalog No. HY-19707, MedChemExpress, Princeton, USA) was utilized to inhibit IRE1 endoribonuclease (RNase) activity. On day 6, moDCs were pretreated with the TLR4 inhibitor TAK-242 and/or the RAGE inhibitor FPS-ZM1 (Catalog No. HY-11109/HY-19370, MedChemExpress, Princeton, NJ, USA) for 1 h. MoDCs were then stimulated with PBS (Control), HMGB1 (1 µg/mL), 4µ8C (20 µM), or HMGB1 plus 4µ8C in fresh culture medium for 48 h. Splenic monocytes and mononuclear cells (SMCs) from mice were isolated via gradient centrifugation using the extraction kit (Catalog No. TBD2011MPK, TBD Science, Tianjin, China). Red blood cells were removed by red blood cell lysis buffer (Cat. No. G2015-500ML, Servicebio, Wuhan, China). The harvested monocytes were then cultured in RPMI-1640 medium in the presence of 20 ng/mL murine recombinant GM-CSF (Cat. No.AF-315-03-20UG, PeproTech, New Jersey, USA) and 20 ng/mL murine recombinant IL-4 (Cat. No. 214-14-20UG, PeproTech, New Jersey, USA) [28]. On day 6, immature mouse moDCs were divided into six groups and incubated for 48 h: Control (PBS), HMGB1 (1 µg/mL), 18α-GA (40 µM) [29], HMGB1 neutralizing antibody (10 µg/mL), HMGB1 + 18α-GA, and HMGB1 + HMGB1 neutralizing antibody. The 10:1 ratio of anti-HMGB1 antibody to HMGB1 adopted in this study is within the effective range reported in prior studies [30, 31], and 10 µg/mL HMGB1 neutralizing antibody has been validated to achieve sufficient neutralization in immune cells [31, 32]. Co-cultivation of human moDCs with naïve CD4+ T cells Selection of human naïve CD4+ T cells was performed from fresh PBMCs with Microbeads (Catalog No. 130-094-131, Miltenyi Biotec, Bergisch Gladbach, Germany). Purity of human naïve CD4+ T cells, defined as CD4+ CD45RA+ cells (CD45RA APC, Clone HI100, Catalog No.304112, BioLegend, San Diego, USA), is shown in Fig. S4B. Naïve CD4+ T cells and CD4+ CD25− T effector (Teff) cells were labeled with CFSE dye (Cat. No. C0051, Beyotime, Shanghai, China). To evaluate the effect of direct antigen presentation by moDCs on CD4+ T cell proliferation, 1 × 10⁴ moDCs were co-cultured with 5 × 10⁴ CFSE-labeled naïve CD4+ T cells at a DC: T cell ratio of 1:5 for 72 h as previously described [33]. CD4+CD25+ Treg cells and Teff cells were isolated using MicroBeads (Cat. No. 130-109-557, Miltenyi Biotec, Bergisch Gladbach, Germany) (Fig. S4C-D). For the Treg suppression assay, co-culture was performed as described previously [34]. Briefly, 5 × 10⁴ CFSE-labeled Teff cells were co-cultured with 1 × 10⁵ moDCs in the absence or presence of 5 × 10⁴ Treg cells at a Treg: Teff cell ratio of 1:1 for 5 days. The human AChRα 195–212 peptide (sequence: DPTYLDITYHFVMQRLPL, 18 aa) was supplemented into the above co-culture systems at 1 µg/mL to activate autoreactive CD4+ T cells targeting AChR in MG patients specifically [4, 35]. The proliferation of naïve CD4+ T cells and Teff cells was determined by flow cytometry based on the dilution of CFSE. To assess CD4+ T cell differentiation, unlabeled naïve CD4+ T cells (2 × 10⁵) were co-cultured with moDCs (4 × 10⁴) at a DC: T cell ratio of 1:5 for 48 h [36]. Cell surface staining of human CD4+ T subsets was performed using CD4 FITC (Clone L200, Catalog No.550628) (BD Pharmingen, San Diego, USA) and CD25 PE (Clone BC96, Catalog No.302605) (BioLegend, San Diego, USA). For intracellular staining, CD4+ T cells were incubated for 4 h at 37 °C with Leuko Act Cktl with GolgiPlug (Catalog No. 550583, BD Pharmingen, San Diego, USA). Then, the cells were permeabilized using Cytofix/Cytoperm Soln kit and stained with the IFN-γ PE-cy7 (Clone 4 S.B3, Catalog No. 557844, BD Pharmingen, San Diego, USA), and IL-4 BV510 (Clone MP4-25D2, Catalog No. 500835)/IL-17A BV605 (Clone BL168, Catalog No. 512325)/FOXP3 BV650 (Clone 206D, Catalog No. 320133) (BioLegend, San Diego, USA). Mouse CD4+ T Cells evaluated by flow cytometry Mouse SMCs were isolated, and CD4+ T cells were surface stained with combinations of CD4 FITC (Clone GK1.5, Catalog No.100405)/CD3 PerCP-Cy5.5 (Clone 17A2, Catalog No.100217)/CD25 APC (Clone PC61, Catalog No.102011)/CD45 BV785 (Clone 30-F11, Catalog No. 103149) (BioLegend, San Diego, USA). Then, the cells were treated with Leuko Act Cktl, supplemented with GolgiPlug, and permeabilized using Cytofix/Cytoperm Soln kit. Intracellular cytokine staining of mouse CD4+ T cells was performed with the IFN-γ BV650 (Clone XMG1.2, Catalog No. 505831)/IL-4 PE (Clone 11B11, Catalog No. 504103) (BioLegend, San Diego, USA), and IL-17A eFluor 450 (Clone eBio17B7, Catalog No. 48-7177-80)/FOXP3 PE-eFluor 610 (Clone FJK-16s, Catalog No. 61-5773-82) (Thermo Fisher Scientific, Waltham, USA). Selection of mouse splenic naïve CD4+ T cells was performed from fresh SMCs with Microbeads (Catalog No. 130-104-453, Miltenyi Biotec, Bergisch Gladbach, Germany). The sort purity of mouse naïve CD4+ T cells, defined as CD3ε+CD4+ cells (CD3ε PE-Cy7, Clone 500A2, Catalog No.152313, BioLegend, San Diego, USA), was above 90% (Fig. S4E). On day 6, immature mouse moDCs were then incubated with 40 µM 18α-GA for 48 h to test its putative tolerogenic effect. In the presence of 1 µg/mL Torpedo AChRα 146–162 peptide, 18α-GA moDC were co-incubated with CFSE-labeled naïve CD4+ T cells to assess CD4+ T cell proliferation. Flow cytometry data were analyzed using CytoFLEX flow cytometer (Beckman Coulter, Fullerton, USA) and Flowjo 10.0 (Tree Star, Ashland, USA). Gating was performed with fluorescence minus one (FMO) controls as references, and the Δ median fluorescence intensity (MFI) values were calculated by subtracting the baseline fluorescence derived from FMO controls. Immunofluorescence The cell-climbing slices were placed in the 24-well plates prior to the moDC differentiation culture. On day 8, cells on slides were stained with 1:1000 Fluor 488-phalloidin (Cat. No. ab176753, Abcam, Cambridge, USA) at room temperature for 60 min to visualize the cytoskeleton. The slides were sealed with DAPI-containing anti-fluorescence quenched blocking agent (Cat. No. G1407-25ML, Servicebio, Wuhan, China). Frozen 8-µm sections of mouse gastrocnemius muscle were fixed in Fixative solution (Cat. No. G1101-500ML, Servicebio, Wuhan, China) for 5–10 min. After blocking with 5% BSA at room temperature for 1 h, sections were incubated overnight at 4 °C with rabbit anti-C5b-9 antibody (1:1000; Cat. No. ab55811, Abcam, Cambridge, USA). Following rinsing, sections were incubated in the dark for 1 h with YSFluor 594-conjugated donkey anti-rabbit IgG (1:1000; Cat. No. 34212ES60, Yeasen, Shanghai, China) and CF488A-labeled α-bungarotoxin (1:500; Cat. No. 00005, Biotium, Fremont, USA). Immunofluorescence images were obtained using an OLYMPUS BX53 fluorescence microscope (Olympus, Tokyo, Japan). Antibody ELISAs and cytokine multiplex arrays Human plasma HMGB1 (Catalog No. RX105888H, Ruixin Biotechnology Co., Ltd., Quanzhou, China) and mouse plasma HMGB1 (Catalog No. MU30043, Bioswamp, Wuhan, China) were measured by ELISA Kits. Human IL-6/IL-10/IL-12 p70/TNF-β/IFN-γ/IL-4/IL-17A were detected from culture supernatant using the LEGENDPlex multiplex cytokine assay (Human B cell Panel) (Catalog No. 740527, BioLegend, San Diego, USA) according to the manufacturer’s instructions. Mouse IL-6/IL-10/IL-12 p70 (Catalog No. SEKM-0007/SEKM-0010/SEKM-0013, Solaibao, Beijing, China) and TNF-β (Catalog No. EK0515, Boster, Wuhan, China) ELISA Kits were used to evaluate the cytokine secretion in the supernatants of mouse moDCs. To measure anti-Torpedo AchRα 146–162 IgG levels, 96-well plates were coated overnight at 4 °C with the peptide (5 µg/mL in 100 µL). The following day, plates were washed with PBS-T (PBS containing 0.05% Tween 20) and blocked with 200 mL of blocking buffer (Catalog No.70-E0512, Multi Sciences, Hangzhou, China) for 1 h. Plasma samples from EAMG mice (diluted 1:100 in PBS-T) were added (100 µL per well) and incubated for 2 h. After washing, biotinylated rabbit anti-mouse (1:2000; Catalog No. BA1001, Boster, Wuhan, China) was added and incubated for 1 h, followed by HRP Conjugated Streptavidin (1:5000; Catalog No. BA1088, Boster, Wuhan, China) for 30 min. After washing, a TMB substrate solution (Catalog No. PR1200, Solaibao, Beijing, China) was added to produce a color reaction. The absorbance was detected by Microplate Reader (Thermo Fisher, Waltham, USA) at 450 nm. Haemolytic complement activity (CH50) and supernatant soluble C5b-9 (sC5b-9) detection Serum samples were then incubated with 10 µM 18α-GA or 2.5 µg/mL HMGB1 neutralizing antibody in the presence or absence of 0.25 µg/mL HMGB1 stimulation. For the combined treatment groups, sera were preincubated with 18α-GA or HMGB1 neutralizing antibody at 37 °C for 30 min before the addition of HMGB1, followed by a further 30 min incubation. Sera were incubated with hemolysin (Catalog No. BM351Y, Bersee, Beijing, China)-sensitized 2% sheep erythrocytes at 37 °C for 30 min. Haemolytic activity was determined based on the released hemoglobin in the supernatant, measured spectrophotometrically at 540 nm [37]. For sC5b-9 measurement, a parallel serum incubation system without hemolysin or erythrocytes was established under identical grouping and treatment conditions. After incubation, samples were centrifuged, and supernatants were collected for sC5b-9 quantification using human (Catalog No. SEKM-0538, Solaibao, Beijing, China) and mouse (Catalog No. ELK11157, ELK Biotechnology, Wuhan, China) ELISA kits. RNA-sequencing Total RNA from the human moDCs with or without HMGB1 stimulation was extracted by TRIzol (Catalog No. 15596026, Invitrogen, Carlsbad, USA). Oligo (dT) magnetic beads were used for mRNA enrichment, followed by sequencing on the BGISEQ platform (Huada Gene Technology Co. Ltd., Wuhan, China). Sample quality was assessed using Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA). Differentially expressed genes (DEGs) were defined as differential transcripts with |log2FC| > 1 and adjusted P value < 0.05. The data obtained were analyzed using BGI Dr. Tom online system (https://biosys.bgi.com). Volcano maps, Gene Ontology (GO), and Kyoto Encyclopedia of Genes and Genomes (KEGG) bioinformatics analysis were used to visualize the DEGs. protein-protein interaction (PPI) analysis to investigate the functional link between the key DEGs. Electron Microscopy (EM) For scanning EM, a minimum of 1 × 10⁵ human moDCs were collected. The cells were fixed with electron microscopy fixation fluid (Cat. No. G1102, Servicebio, Wuhan, China) at room temperature for 1–2 h in the dark, and then stored at 4 °C until imaged using SEM SU8100 (Hitachi, Tokyo, Japan) as previously reported [38]. For transmission EM, the gastrocnemius muscles of the mice were prepared using the same electron microscopy fixation fluid mentioned above. After fixation, the muscle tissues were subjected to dehydration, infiltration, resin embedding, and cut into ultra-thin sections according to the reference [39]. Subsequently, the ultra-thin sections were stained with uranyl acetate (Cat. No. L885990, Macklin, China), lead citrate (Cat. No. 6159-44-0, YaJi Biological, Shanghai, China), and imaged using TEM HT7700 (Hitachi, Tokyo, Japan). RT-PCR and RT-qPCR Total RNA of human moDCs was extracted using TRIzol, and cDNA was synthesized via PrimeScript RT Reagent Kit (Catalog No. RR037B, Takara, Shiga, Japan). For the determination of XBP-1 splicing, the cDNA product was amplified by RT-PCR via 2 × Hieff PCR Master Mix (Catalog No. 10102ES03, Yeasen, Shanghai, China). The following primers were used to amplify the XBP1 cDNA: forward primer 5′-AAACAGAGTAGCAGCTCAGACTGC-3′ and reverse primer 5′-CCTTCTGGGTAGACCTCTGGGAG-3′, which produce uXBP1 of 473 bp and sXBP1 of 447 bp. The PCR products were digested with the restriction enzyme Pst I (Catalog No. D6566, Beyotime, Shanghai, China) and further run on a 2% agarose gel, resulting in uXBP1 fragments of 290 and 183 bp [40]. The frequency of XBP1 splicing was calculated via densitometry as follows: intensity of sXBP1 band / ([intensity of sXBP1 band] + [intensity of uXBP1 band]) ×100%. The original electropherograms for RT-PCR analysis are shown in Fig. S5. For RT-qPCR, cDNA was amplified by SYBR-Green Kit (Catalog No. 11201ES08, Yeasen, Shanghai, China) and performed by CFX Connect Real-Time System (BioRad Laboratories, Hercules, USA). Expression was illustrated by the 2−(ΔΔCt) method. Primer details for RT-qPCR are presented in Table S1. Human GAPDH and mouse Ppia were employed as reference genes for normalization of the expression data. Western blot Mouse moDCs were lysed in RIPA buffer containing PMSF (100:1, Catalog No. P0013B, Catalog No. ST2573, Beyotime, Beijing, China). Proteins were electrotransfered onto PVDF membranes and incubated with primary antibodies: rabbit anti-BIP (1:3000, Catalog No.11587-1-AP)/rabbit anti-GAPDH (1:3000, Catalog No. 10494-1-AP)/rabbit anti-NF-κB (1:3000, Catalog No. 10745-1-AP)/rabbit anti-phospho-NF-κB (Ser468)(1:1000, Catalog No. 82335-1-RR)/rabbit anti-IKBα (1:2000, Catalog No. 10268-1-AP)/rabbit anti-phospho-IKBα (Ser32/36)(1:1000, Catalog No. 82349-1-RR)(Proteintech, Chicago, USA), rabbit anti-sXBP1 (1:1000, Catalog No. 12782T, CST, Danvers, USA), and rabbit anti-IRE1α (1:1000, Catalog No. bs-8680R)/rabbit anti-phospho-IRE1α (Ser 726) (1:500, Catalog No. bs-4308R)(Bioss, Beijing, China). Afterward, the membranes were incubated with HRP-conjugated goat anti-rabbit secondary antibody (1:5000, Catalog No. SA00001-2, Proteintech, Chicago, USA). The chemiluminescence signal was detected using ECL Super Kit (Catalog No. RM02867, ABclonal, Wuhan, China). Immunoblots were quantified by densitometric measurement using BLT GV6000Pro software (Biolight Biotechnology Co., Ltd., Guangzhou, China). The original Western blots are provided in Fig. S6. Molecular docking and surface plasmon resonance (SPR) Molecular docking was conducted to assess the binding affinity of 18α-GA to target proteins using Autodock tools 1.5.6 (Scripps Research Institute, La Jolla, USA), with structures obtained from Protein Data Bank (https://www.rcsb.org/) and PubChem databases (https://pubchem.ncbi.nlm.nih.gov/). After preprocessing and docking simulation, the optimal conformations were visualized and analyzed to characterize intermolecular interactions and binding stability. SPR assays were conducted on a Biacore T200 instrument (GE Healthcare, Uppsala, Sweden) to verify the specific binding of 18α-GA to IκBα and BIP. CM5 chips were activated and immobilized with target proteins, and then injected with gradient concentrations of 18α-GA. Real-time sensorgrams were analyzed to determine the affinity parameters. Statistical analysis Prior to analysis, the data were checked for completeness, normality, and outliers. Categorical variables are presented as n (%), while continuous variables are represented as mean ± SD or median [IQR]. Data analysis was performed by Student’s t tests, Mann-Whitney U tests, chi-squared tests, or one-way ANOVA where appropriate. Correlation analyses were evaluated by Spearman’s rank test. Bonferroni’s correction was applied for multiple comparisons. P value < 0.05 was considered as statistically significant. As the treatment-by-cohort interaction was not significant, data from the independent animal cohorts were pooled for statistical comparison [41]. Immunofluorescence images, gel electrophoresis of the RT-PCR, and western blot bands were quantified using ImageJ 1.5 (NIH, Bethesda, USA). Statistical analyses were conducted using Origin 2021 (OriginLab, Northampton, USA), SPSS 24.0 (IBM, Armonk, USA), GraphPad Prism 8.01 (GraphPad Prism, San Diego, USA), and R 4.4.1 (R Foundation for Statistical Computing, Vienna, Austria). Further details about statistical analysis, including sample size (n), are indicated in the figure legends. Results MR analysis between immunophenotypes and MG CD11c+HLA-DR++monocyte (a characteristic phenotype of moDCs) absolute count (AC), the percentage of CD4−CD8−/leukocyte, MFI of CD45 on CD14+ monocyte, CD25 on activated secreting Treg, CD38 on IgD+, CD127 on CD4+, CD28 on CD28+CD45RA+CD8br, FSC-A on plasmacytoid DC, FSC-A on T cell correlated with higher MG risk. The percentages of CD39+ secreting Treg/secreting Treg, CD39+ secreting Treg/CD4 Treg, naive CD8br/CD8br, central memory (CM) CD4CD8−/T cell, CD14+CD16+ monocyte/monocyte, CD28+CD4−CD8−/T cell, MFI of CD19 on switched memory (sw mem) B cell, IgD on IgD+CD38− unswitched memory (unsw mem) B cell, CD4 on CM CD4+ showed protective effects (Fig. 1A). Reverse MR analysis assessed causality between MG and positive immunophenotypes, finding significant causality for CD11c+HLA-DR++ monocyte AC, the percentages of naive CD8br/CD8br, and CD4−CD8−/leukocyte (Fig. 1B). The forest plot, scatter plot, funnel plot, and leave-one-out analyses illustrating the causal effect of CD11c+HLA-DR++ monocyte AC on MG are presented in Fig. 1C-F, whereas the analyses addressing reverse causality are shown in Fig. 1G-J. Mendelian randomization (MR) analysis of 731 immunophenotypes and MG (myasthenia gravis). A IVW (inverse-variance weighted) results of forward-direction MR analysis between 731 immunophenotypes and MG. B IVW results of reverse-direction MR between positive immunophenotypes and MG. C-J The forest plot, scatter plot, funnel plot, and leave-one-out analyses illustrate the causal association between CD11c+HLA-DR++ monocyte AC (absolute count) and MG on forward-direction and reverse-direction analysis. Circulating DC subsets and plasma HMGB1 between MG patients and HCs MG patients exhibited evident proliferation of moDCs, as evidenced by their significantly higher AC and percentage compared to those of HCs. The percentage of total classical type DCs (cDCs) was also higher in MG, although the difference was less remarkable than that of moDCs (Fig. 2A-B). The concentration of HMGB1 was higher in MG patients and positively associated with the percentage of moDCs (Fig. 2C-D). MoDCs from MG patients displayed more mature phenotypes than those from HCs, as evidenced by the higher ΔMFI values of CD83, CD86, and HLA-DR (Fig. 2E-F). There was a significant positive correlation between the plasma HMGB1 and mature markers of moDCs in MG (Fig. 2G). Additionally, receiver operating characteristic curve analysis demonstrated that plasma HMGB1 and the percentage of peripheral moDCs had clinical diagnostic values (Fig. 2H). Identification of peripheral blood DC subsets in myasthenia gravis (MG) patients. A-C The absolute counts and percentages of blood DC subsets, and levels of plasma HMGB1 between naïve MG (32 biological replicates, 2 technical replicates) and healthy controls (HCs) (20 biological replicates, 2 technical replicates) (two-tailed Mann-Whitney U test). D Heatmap of correlation analysis between the proportions of DC subsets and plasma HMGB1 in MG (32 biological replicates, 2 technical replicates, two-tailed Spearman’s correlation test). E-F Representative flow histograms showing the △median fluorescence intensity (MFI) values of CD83, CD86, and HLA-DR on monocyte-derived dendritic cells (moDCs) and a dot plot representing the difference between naïve MG (n = 32 biological replicates, 2 technical replicates) and HCs (20 biological replicates, 2 technical replicates) (two-tailed Mann-Whitney U test). G Heatmap of correlation analysis between the △MFI values of CD83, CD86, and HLA-DR on moDCs and plasma HMGB1 in MG (32 biological replicates, 2 technical replicates, two-tailed Spearman’s correlation test). H Receiver operating characteristic (ROC) curves of peripheral moDCs and plasma HMGB1 for predicting MG. Data are expressed as box-and-whisker plots depicting the smallest observation, lower quartile, median, upper quartile, and largest observation. *P < 0.05, **P < 0.01, ***P < 0.001 HMGB1 exposure altered differentiation of human moDCs To examine the contribution of HMGB1 to moDC differentiation, we cultured moDCs in the presence of HMGB1 at different concentration gradients (Fig. 3A). The purity of human moDCs was more than 80% (Fig. 3B). HMGB1 induced necrosis and apoptosis of moDCs only at a high concentration (10 µg/mL for 48 h) in vitro (Fig. 3C-D). At low concentrations (0.25–1 µg/mL, 48 h), HMGB1 promoted the maturation and activation of moDCs in a concentration-dependent manner, as reflected by increased expression of surface maturation markers (CD83, CD86, and HLA-DR), elevated secretion of proinflammatory cytokines (IL-6 and IL-12 p70), and reduced release of anti-inflammatory cytokines (IL-10 and TGF-β) (Fig. 3E-G). Following 48 h of treatment with 0.25–1 µg/mL HMGB1, mRNA expression of the tolerogenic regulator AIRE was significantly downregulated in a dose-dependent manner (Fig. 3H). MoDCs stimulated with 0.5 ug/mL of HMGB1 at 48 h revealed prominent dendrite outgrowths characterized by significantly larger perimeters and surface areas compared to controls, indicating enhanced morphological maturation (Fig. 3I-J). HMGB1 exposure altered the phenotypes of monocyte-derived dendritic cells (moDCs) in myasthenia gravis (MG) patients. A HMGB1 treatment schematic in the moDCs culture. B Representative gating strategy for human moDCs. C-D Representative scatter plots and bar charts showing the apoptosis assay of moDCs treated with different concentrations of HMGB1 for 48 h (8 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). E-F Representative flow histograms and a bar chart showing the △median fluorescence intensity (MFI) values of CD83, CD86, and HLA-DR on moDCs treated with different concentrations of HMGB1 for 48 h (8 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). G IL-6, IL-10, IL-12 p70, and TGF-β levels in cell supernatants from moDCs treated with different concentrations of HMGB1 for 48 h (8 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). H The mRNA expression levels of AIRE in moDCs treated with different concentrations of HMGB1 for 48 h (8 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). I-J Representative fluorescence images and quantification of cell perimeter and area in moDCs stimulated with 0.5 µg/mL HMGB1 for 48 h (n = 258 cells) or left unstimulated (n = 671 cells) (8 biological replicates, 3 fields/technical replicate, two-tailed Student’s t test). Data are presented as mean ± SD. *P < 0.05, **P < 0.01, ***P < 0.001 HMGB1 induced ER stress in moDCs Compared to the control moDCs, treatment with 0.5 µg/mL HMGB1 for 48 h resulted in significant transcriptional changes, with 121 genes being upregulated and 78 genes being downregulated (Fig. 4A). GO term enrichment analysis revealed that response to ER stress was among the most significantly enriched pathways (Fig. 4B). Subsequent PPI network analysis, integrated with the above ER stress-associated GO annotation dataset, is presented in Fig. 4C. We further explored the specific impacts of HMGB1 on the three canonical ER stress signaling pathways in moDCs and found that the low concentration (1 µg/mL for 48 h) of HMGB1 primarily activated the IRE1α/XBP1 pathway, whereas the high concentration (10 µg/mL for 48 h) of HMGB1 stimulation predominantly induced the PERK/CHOP pathway. In contrast, no significant alterations were observed in the ATF6 pathway. HMGB1 promoted NF-κB pathway activation characterized by increased NF-κB transcription and decreased IκBα transcription in moDCs, with a more pronounced activation observed upon high concentration stimulation. HMGB1 also promoted apoptosis of moDCs by upregulating the expression of pro-apoptotic gene BAX and downregulating the anti-apoptotic gene BCL2 at high concentration (Fig. 4D). As an ER stress sensor, IRE1 possesses kinase and RNase activities and mediates unconventional splicing of XBP1 mRNA [42]. Both low- and high-dose HMGB1 enhanced XBP1 splicing in moDCs versus controls, while no obvious difference existed between the two treatment groups (Fig. 4E-F). HMGB1 induced endoplasmic reticulum (ER) stress in monocyte-derived dendritic cells (moDCs). A Volcano plot showing the differentially expressed genes (DEGs) between moDCs with or without HMGB1 stimulation. B GO terms of biological process analysis of the DEGs. C Protein-protein interaction network for DEGs. D The mRNA expression levels of IRE1α, PERK, ATF6, BIP, CHOP, sXBP1, IκBα, NF-κB, BCL2, and BAX in moDCs among the Control, 1 µg/mL HMGB1, and 10 µg/mL HMGB1 groups (8 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). E-F Splicing of XBP1 mRNA in moDCs examined by RT-PCR among the above three groups (3 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). G Representative electron microscopy images showed the morphology of the endoplasmic reticulum (yellow arrows) of moDCs. H The mRNA expression levels of IRE1α, sXBP1, IκBα, and NF-κB in moDCs among the Control, HMGB1, 4µ8C, and HMGB1 + 4µ8C groups (8 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). I-J Splicing of XBP1 mRNA in moDCs examined by RT-PCR among the above four groups (3 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). K-N Representative flow histograms and bar charts showing the △median fluorescence intensity (MFI) values of CD83, CD86, and HLA-DR on moDCs (8 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). O-R IL-6, IL-10, IL-12 p70, and TGF-β levels in cell supernatants from moDCs (8 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). Data are presented as mean ± SD. *P < 0.05, **P < 0.01, ***P < 0.001 Electron microscopy revealed that moDCs stimulated with HMGB1 (1 µg/mL, 48 h) exhibited prominent ER expansion and luminal swelling, which is indicative of ER stress and ER lumenal protein accumulation. Notably, the IRE1α RNase inhibitor 4µ8C abrogated HMGB1-induced ER abnormalities and effectively promoted the recovery of ER morphology (Fig. 4G). HMGB1 significantly increased IRE1α, sXBP1, IKKα, and NF-κB expression in moDCs, and 4µ8C fully blocked these effects, verifying that NF-κB activation relies on the IRE1α-XBP1 axis (Fig. 4H). XBP1 splicing assays revealed that 4µ8C effectively abrogated the HMGB1-mediated XBP1 splicing in moDCs (Fig. 4I-J). To clarify which primary receptor mediates HMGB1-induced dysfunction in moDCs, we pretreated moDCs with the selective TLR4 inhibitor TAK-242 (5 µM, 1 h) and RAGE inhibitor FPS-ZM1 (30 nM, 1 h), respectively. We further confirmed that HMGB1 facilitated moDC activation primarily through TLR4, but not RAGE. In moDCs, inhibition of TLR4 using TAK-242 downregulated CD83, CD86 and HLA-DR expression, reduced the release of IL-6 and IL-12 p70, and elevated the levels of IL-10 and TGF-β (Fig. 4K-R). HMGB1 modulated moDC-mediated CD4+ T cell immune responses via TLR4/IRE1α pathway We further examined whether 4µ8C (20 µM, 48 h) and/or TAK-242 (5 µM, 1 h) could restore the immune tolerance of moDCs in the presence of HMGB1 (1 µg/mL, 48 h), as per the experimental design in Fig. 5A. HMGB1 promoted naive CD4+ T cell proliferation in co-culture systems of moDCs and naïve CD4+ T cells, which was largely abrogated by blocking TLR4 and/or IRE1α (Fig. 5B-C). HMGB1 drove naïve CD4+ T cell differentiation toward Th1 and Th17 cells while inhibiting their polarization into Th2 and Treg cells. However, moDCs treated with 4µ8C and/or TAK-242 effectively restored the balance among Th1, Th2, Th17, and Treg cell populations (Fig. 5D). Multiplex cytokine assays on cell culture supernatants demonstrated that moDCs stimulated with HMGB1 induced naive CD4+ T cells to produce pro-inflammatory cytokines (IFN-γ and IL-17A) while reducing secretion of the anti-inflammatory cytokine IL-4. Inhibition of TLR4 and/or IRE1α signaling fully reversed these cytokine perturbations (Fig. 5E-G). HMGB1-treated moDCs markedly impaired the suppressive function of Treg cells, thereby enhancing the proliferation of Teff cells. Inhibition of TLR4 and/or IRE1α in moDCs restored Treg suppressive capacity and suppressed Teff cell proliferation (Fig. 5H-I). Notably, combined blockade of both TLR4 and IRE1α signaling using TAK-242 + 4µ8C exerted the strongest effect on all moDC-mediated CD4+ T cell immune responses described above. HMGB1 modulated monocyte-derived dendritic cells (moDCs)-mediated CD4+ T cell responses. A-C CFSE-labeled naive CD4+ T cell proliferation was assessed in the presence of the human AchRα 195–212 peptide, with comparisons among the Control, HMGB1, HMGB1 + TAK-242, HMGB1 + 4µ8C, and HMGB1 + TAK-242 + 4µ8C groups (8 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test compared with the HMGB1 group). D Representative flow cytometry plots and summary dot plots showing the percentages of Th1, Th2, Th17, and Treg cells among the above five groups (8 biological replicates, 3 technical replicates, one-way ANOVA compared with the HMGB1 group followed by Bonferroni’s multiple-comparison test). E-G The levels of IFN-γ, IL-4, and IL-17A in naïve CD4+ T cell supernatants and moDC-naïve CD4+ T cell co-culture supernatants (8 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). H-I Proliferation of AChRα 195–212-specific Teff cells in a moDC-dependent Treg suppression assay (8 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test compared with the HMGB1 group). Data are presented as mean ± SD. *P < 0.05, **P < 0.01, ***P < 0.001 18α-GA suppressed the effects of HMGB1 on mouse moDCs in vitro Subsequently, the inhibitory effects of 18α-GA and HMGB1 neutralizing antibody on mouse splenic moDCs were compared, as outlined in Fig. 6A. The purity of mouse moDCs was confirmed by flow cytometry (Fig. 6B). 18α-GA were capable of inducing immune tolerance in mouse moDCs, as evidenced by the downregulated expression of CD83, CD86, and HLA-DR. The immunomodulatory effect of 18α-GA on moDCs is more pronounced than that of the HMGB1 neutralizing antibody (Fig. 6C-D). Western blot and RT-PCR analysis revealed that HMGB1 stimulation significantly activated the IRE1α-XBP1 and NF-κB pathways in moDCs, as evidenced by elevated protein levels of p-IRE1α, BIP, sXBP1, p-IκBα, p-NF-κB, and enhanced XBP1 splicing. These effects were markedly reversed by 18α-GA (Fig. 6E-L). Importantly, 18α-GA exerted its inhibitory effects even in the absence of exogenous HMGB1 stimulation, indicating that it can directly inhibit the IRE1α/XBP1/NF-κB signaling axis. Molecular docking showed that 18α-GA formed stable complexes with HMGB1, IκBα, and BIP, with optimal binding energies of −8.6, −9.1, and −8.4 kcal/mol, respectively (Fig. 6M-O). SPR assays confirmed direct binding of 18α-GA to IκBα and BIP, with equilibrium dissociation constants (Kd) of 3.61 µM and 24.5 µM, respectively (Fig. 6P-Q), supporting that 18α-GA induced tolerogenic moDCs by multitarget inhibition of the IRE1α/XBP1 and NF‑κB signaling pathways. 18α-GA inhibited IRE1α/XBP1 and NF-κB pathways on mouse monocyte-derived dendritic cells (moDCs). A Flow diagram of cultured moDCs from the experimental autoimmune myasthenia gravis (EAMG) mice. B Representative gating strategy for mouse moDCs. C-D Representative flow histograms showing the △median fluorescence intensity (MFI) values of CD83, CD86, and MHC-II on moDCs and a bar chart representing the difference among the Control, HMGB1, 18α-GA, HMGB1 neutralizing antibody, HMGB1 + 18α-GA, and HMGB1 + HMGB1 neutralizing antibody groups (3 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). E-J Western blot and quantification of p-IRE1α/IRE1α, BIP, sXBP1, p-NF-κB/NF-κB, and p-IKBα/IKBα among the above six groups (3 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). K-L XBP1 mRNA splicing examined by RT-PCR among the above six groups (3 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). M-O Molecular docking results of 18α-GA with HMGB1, IκBα, and BIP. P-Q Surface plasmon resonance (SPR) analysis of 18α-GA interactions with IκBα and BIP. Data are presented as mean ± SD. *P < 0.05, **P < 0.01, ***P < 0.001 Efficacy, safety, and pharmacokinetic profiles of 18α-GA in EAMG 18α-GA attenuated myasthenic symptoms in EAMG mice when administered at a dose range of 0.5–2 mg/kg, resulting in less weight loss, lower clinical scores, higher grip strength, and reduced Torpedo AchRα 146–162 IgG levels compared with untreated EAMG mice (Fig. S7A-D). Glycyrrhizic acid (GL) and its metabolic products may induce the characteristic adverse effect of pseudoaldosteronism, manifested by increased potassium excretion, water-sodium retention, weight gain, and metabolic alkalosis [43, 44]. No obvious liver or renal dysfunction, nor electrolyte disturbances, were detected in mouse serum samples (Fig. S7E-L). Therefore, 0.5 mg/kg (ip, q2d) was selected to examine the pharmacokinetics based on MED. 18α-GA had a long plasma half-life of ~ 69.315 h with an AUC (0–∞) of 1671.881 µg/L·h (Fig. S7M-N). We further compared the therapeutic effect of 18α-GA (0.5 mg/kg, ip, q2d) and HMGB1 neutralizing antibody (5 mg/kg, iv, q2w) on EAMG mice. Compared to the EAMG group, both the 18α-GA group and the HMGB1 neutralizing antibody group displayed less weight loss, lower clinical scores, higher grip strength, and reduced AChRα 146–162 IgG levels (Fig. 7A-E). Spleen and inguinal lymph node weights, indicators of immune activation in EAMG, were significantly reduced in mice treated with 18α-GA or HMGB1 neutralizing antibody (Fig. 7F-H). Ultrastructural analysis of the NMJs revealed that EAMG mice had reduced or absent postsynaptic membrane ruffling and widened synaptic clefts. These abnormalities were mitigated by 18α-GA and HMGB1 neutralizing antibody (Fig. 7I). RNS demonstrated that EAMG mice that received 18α-GA and HMGB1 neutralizing antibody had reduced attenuation (Fig. 7J-K). We detected elevated plasma HMGB1 levels in EAMG mice, which could be inhibited by 18α-GA and HMGB1 neutralizing antibody (Fig. 7L). Compared with controls, EAMG mice showed robust C5b-9 signals co-localized with α-BTX-labeled AChR, indicating prominent complement activation at NMJs. HMGB1 neutralizing antibody and 18α-GA effectively inhibited complement activation and C5b-9 deposition at NMJs in EAMG mice (Fig. 7M). Specifically, 18α-GA showed better therapeutic effects against EAMG than HMGB1 neutralizing antibody, supported by the better improvements in above multiple key pathogenic and clinical parameters. 18α-GA exerted potent therapeutic effects in experimental autoimmune myasthenia gravis (EAMG). A Representative mice in the Control group and the EAMG group. B-E Comparative analyses of the body weight, clinical scores, grip strength, and levels of Torpedo AChRα 146–162 IgG among the Control, EAMG, 18α-GA, and HMGB1 neutralizing antibody groups (16 biological replicates per group in 2 cohorts, 2 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test compared with the Control group at each time point). F-H Representative photographs and weights of the spleen and inguinal lymph nodes among the above four groups (16 biological replicates per group in 2 cohorts, 2 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). I Representative electron microscopy images of the neuromuscular junctions (NMJs) among the above four groups (yellow arrows indicate the disruption at NMJs; Pre, presynaptic membrane; Post, postsynaptic membrane). J-K The attenuation rates of the gastrocnemius muscles detected by low-frequency RNS (repetitive nerve stimulation) (16 biological replicates per group in 2 cohorts, 2 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). L Concentration of plasma HMGB1 detected by ELISA (16 biological replicates per group in 2 cohorts, 2 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). M Representative immunofluorescence images of complement deposition at NMJs among the above four groups stained with α-BTX (green), DAPI (blue), and C5b-9 (red). Data are presented as mean ± SD. *P < 0.05, **P < 0.01, ***P < 0.001 18α-GA restored DC homeostasis and reversed Th1/Th2/Th17/Treg imbalance in EAMG EAMG mice showed significant increases in the percentages of splenic moDCs, cDC1s, and cDC2s, with moDCs exhibiting the most prominent expansion, while pDCs and total cDCs remained unchanged. Both HMGB1 neutralization and 18α-GA treatment reduced the expansion of moDCs and cDC2s (Fig. 8A). Compared with the Control group, EAMG mice exhibited significantly elevated expression of maturation markers (CD83, CD86, and HLA-DR) on cDC1s, cDC2s, and especially moDCs. In contrast, pDCs showed no marked differences in CD83, CD86, or MHC-II expression among the groups, and the IFN-α secretion by pDCs also remained unchanged (Fig. 8E). Notably, both HMGB1 neutralizing antibody and 18α-GA treatment markedly attenuated the upregulation of CD83, CD86, and MHC-II on cDC1s, cDC2s, and moDCs, with the most pronounced effect observed in moDCs (Fig. 8B-F). Correlation analysis further showed that HMGB1 was positively associated with the expression of CD83, CD86, and MHC-II on moDCs, supporting a close link between HMGB1 and moDC maturation (Fig. 8G). EAMG mice had elevated splenic Th1/Th17 proportions and reduced Th2/Treg populations, while 18α-GA and HMGB1 neutralizing antibody restored the balance of these CD4+ T cell subsets (Fig. 8H-I). Specifically, 18α-GA exhibited more potent immunosuppressive effects on DC and CD4+ T cell subsets in vivo compared with HMGB1 neutralizing antibody. 18α-GA restored DC and CD4+ T cell homeostasis in experimental autoimmune myasthenia gravis (EAMG). A-C The percentages of splenocyte DC subsets among the Control, EAMG, 18α-GA, and HMGB1 neutralizing antibody groups (16 biological replicates per group in 2 cohorts, 2 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). B-E Representative flow histograms showing the △median fluorescence intensity (MFI) values of CD83, CD86, and MHC-II on splenic DC subsets and bar charts representing the difference among the above four groups (16 biological replicates per group in 2 cohorts, 2 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). F Percentage of IFN-α secretion in pDCs among the above four groups (16 biological replicates per group in 2 cohorts, 2 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). G Heatmap of correlation analysis between the △MFI values of CD83, CD86, and HLA-DR on moDCs in EAMG (48 biological replicates, 2 technical replicates, two-tailed Spearman’s correlation test). H-I Representative flow cytometry plots and summary dot plots showing the percentages of splenic Th1, Th2, Th17, and Treg cells among the above four groups (16 biological replicates per group in 2 cohorts, 2 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test) Adoptive transfer of 18α-GA moDCs alleviated EAMG and reversed Th1/Th2/Th17/Treg imbalance Given the immature phenotypes of moDCs induced by 18α-GA (Fig. 6C), we further assessed their cytokine production and biological functions. 18α-GA treatment decreased IL-6 and IL-12p70 release but enhanced secretion of IL-10 and TGF-β in moDCs compared with controls (Fig. 9A). AIRE expression trended upward in 18α-GA-treated moDCs, though the change was not significant (Fig. 9B). CFSE staining demonstrated that moDCs exposed to 18α-GA substantially inhibited antigen-specific proliferation of naive CD4+ T cells (Fig. 9C-D). 18α-GA-induced tolerogenic moDCs (18α-GA moDCs) ameliorated experimental autoimmune myasthenia gravis (EAMG). A IL-6, IL-10, IL-12 p70, and TGF-β levels in cell supernatants from control moDCs and 18α-GA moDCs (3 biological replicates, 3 technical replicates, two-tailed Student’s t test). B The mRNA expression levels of AIRE between control moDCs and 18α-GA moDCs (3 biological replicates, 3 technical replicates, two-tailed Student’s t test). C-D Proliferation of naive CD4+ T cells labeled with CFSE after co-culture of control moDCs and 18α-GA moDCs (3 biological replicates, 3 technical replicates, two-tailed Student’s t test). F-H Comparison of the body weight, clinical scores, grip strength, and levels of Torpedo AChRα 146–162 IgG in EAMG mice treated with 18α-GA moDCs or not (16 biological replicates per group in 2 cohorts, 2 technical replicates, two-tailed Student’s t test). I Representative immunofluorescence images of complement deposition at neuromuscular junctions (NMJs) between the above two groups stained with α-BTX (green), DAPI (blue), and C5b-9 (red). J Representative electron microscopy images of the NMJs between the above two groups (yellow arrows indicate the disruption at neuromuscular junctions; Pre, presynaptic membrane; Post, postsynaptic membrane). K-O Comparison of the attenuation rates of RNS, weights of the spleen and inguinal lymph nodes between the above two groups (16 biological replicates per group in 2 cohorts, 2 technical replicates, two-tailed Student’s t test). P-Q Summary dot plots and representative flow cytometry plots comparing the percentages of splenic Th1, Th2, Th17, and Treg cells between the above two groups (16 biological replicates per group in 2 cohorts, 2 technical replicates, two-tailed Student’s t test). Data are presented as mean ± SD. *P < 0.05, **P < 0.01, ***P < 0.001 Adoptive transfer of 18α-GA moDCs to EAMG mice exerted a significant therapeutic effect, characterized by increased body weight, reduced clinical scores, enhanced grip strength, and AChRα 146–162 IgG levels (Fig. 9E-H). Immunofluorescence results indicated diminished C5b-9 deposition at AChR-enriched neuromuscular endplates in 18α-GA moDC-treated mice relative to EAMG controls (Fig. 9I). Ultrastructural observation further revealed that 18α-GA moDC treatment alleviated synaptic damage and preserved neuromuscular junction morphology (Fig. 9J). In addition, RNS recordings demonstrated restored neuromuscular transmission function following 18α-GA moDC treatment (Fig. 9K-L). At the systemic immune level, adoptive transfer of 18α-GA moDCs effectively mitigated EAMG-induced splenomegaly and lymphadenopathy (Fig. 9M-O). Mechanistically, 18α-GA moDCs reshaped CD4+ T cell immune homeostasis in vivo by suppressing the differentiation of Th1 and Th17 cells and facilitating Th2 and Treg polarization (Fig. 9P-Q). Discussion This study revealed the dysfunctional immune responses of moDCs in MG and was the first to demonstrate that 18α-GA can alleviate EAMG by specifically targeting the HMGB1-moDCs-CD4+ T cell axis. These findings not only provide a novel perspective on the pathological mechanisms of MG but also lay a solid foundation for the development of targeted immunomodulatory therapies. Our results revealed that the percentages of total cDCs were also higher in MG patients, although the difference was less remarkable than that of moDCs. MoDCs are a highly versatile population and serve pivotal roles in the host defence against infection and inflammation [45]. Monocytes can transform into moDCs under both in vitro and in vivo conditions [46]. MoDCs are vital in antigen processing and presentation to CD4+ T cells [47]. cDC1s specialize in cross-presenting exogenous antigens to CD8+ T cells, cDC2s primarily present antigens to CD4+ T cells, and pDCs are mainly responsible for producing IFN-I during viral infections [10]. Our ex vivo experiments on SMCs, conducted under the same culture conditions as moDCs, revealed that HMGB1 significantly induced the maturation of cDC1s and especially cDC2s. Both 18α-GA and HMGB1 neutralizing antibody efficiently reversed these effects (Fig. S8A-D). By comparison, pDCs showed negligible phenotypic changes and unaltered IFN-α secretion (Fig. S8E-H). Collectively, our results demonstrated that moDCs and cDC2s acted as major targets of the HMGB1-mediated inflammatory axis in MG, with moDCs playing a predominant role. The poor response of pDCs is likely attributed to context-dependent or bidirectional regulation by HMGB1 [48, 49]. The cross-talk between moDCs, cDCs, and pDCs ensures a coordinated and robust immune response [50, 51] and needs future studies to elucidate this interplay in MG. Our results revealed that low concentrations of HMGB1 predominantly activated the IRE1α/XBP1 pathway, while high concentrations primarily triggered the PERK/CHOP pathway and induced apoptosis of moDCs. HMGB1 induced endothelial progenitor cell apoptosis via the PERK/CHOP pathway, thereby triggering apoptosis [52]. These findings are consistent with previous studies showing that the PERK/CHOP axis exerts pro-apoptotic effects across multiple cell types and is functionally linked to activation of the NF-κB signaling pathway [53, 54]. HMGB1 activated the TLR4-XBP1-ER stress-related NF-κB signaling, which promoted neuroinflammation and contributed to depressive-like behaviors [55]. This study further confirmed that TLR4, but not RAGE, served as the primary mediator of the HMGB1-IRE1α/XBP1/NF-κB signaling axis in moDCs. ER stress induced DC maturation and activation through the IRE1α/XBP1 pathway [56]. Previous studies indicated that HMGB1 regulated DC function through TLR4-dependent signaling, integrating DC maturation, cytokine secretion, and migration/adhesion [14, 57]. The IRE1α/XBP1 and NF-κB pathways engage in bidirectional crosstalk, and their core function lies in maintaining ER homeostasis and participating in the pathogenesis of cancer, inflammation, and infection [58, 59]. Aberrant activation of these two pathways establishes a positive feedback loop that amplifies pathogenic inflammatory immune responses, with their crosstalk nodes emerging as promising therapeutic targets [60, 61]. 18α-GA exhibited anti-inflammatory and anticancer activities by binding directly to affect HMGB1 chemotactic and mitogenic activity and inhibiting its release [62, 63]. Both 18α-GA and 18β-glycyrrhetinic acid (18β-GA) are hydrolysis products of GL, however, 18α-GA exhibits pronounced anti-inflammatory, antioxidant, and immunomodulatory activities [64]. GL and GA metabolites elicit pseudohyperaldosteronism by inhibiting renal 11β-HSD2 activity, thereby impairing the conversion of cortisol to cortisone and consequently leading to inappropriate activation of the mineralocorticoid receptor by cortisol [44, 65]. As a selective 11β-HSD1 inhibitor, 18α-GA exhibits weaker inhibition of 11β-HSD2 compared to the conventional 18β-epimer and is therefore mechanistically predicted to have a lower adverse reaction profile [66]. In the present study, after 4 weeks of continuous administration at doses ranging from 0.5 to 2 mg/kg for mice, 18α-GA did not cause significant abnormalities in biochemical analysis, suggesting favorable long-term safety within this dose window. GL can be considered to block the cascade of ER stress and suppress NF-κB activation [67, 68]. GL and 18α-GA, characterized by a lipophilic triterpenoid structure, traverse cell membranes via passive diffusion to enter the cytoplasm and exert diverse biological functions [69, 70]. Previous studies have demonstrated that 18α-GA can suppress phosphorylation of IKBα and inhibit NF-κB activation to exert anti-inflammatory effects [71]. Molecular docking and SPR assays confirmed the specific binding of 18α-GA to IκBα and BIP, suggesting that 18α-GA could induce tolerogenic moDCs by interacting with HMGB1 to attenuate its immunostimulatory activity, binding to IκBα to suppress sustained NF-κB activation, and complexing with BIP to dampen amplification of the ER stress-IRE1α/XBP1 axis. 18α-GA attenuated the maturation of DCs and suppressed T cell-mediated immune responses, thereby demonstrating promising therapeutic potential for autoimmune diseases [72]. Our results indicated that targeting HMGB1 with 18α-GA might be a novel immunotherapeutic strategy for autoimmune diseases by driving the generation of tolerogenic moDCs. MG-associated tissue injury arises principally from local complement activation at NMJs, while circulating sC5b-9 is not a fully validated biomarker and cannot precisely mirror endplate pathology [73]. Therefore, in our in vivo study, we focused on C5b-9 deposition at the NMJs of EAMG mice rather than serum sC5b-9. In the ex vivo serum assay, HMGB1 significantly enhanced complement activation, as evidenced by an increase in CH50 and a decrease in supernatant sC5b-9 levels. These effects were effectively attenuated by both 18α-GA and an HMGB1 neutralizing antibody (Fig. S8I-L). HMGB1 has been demonstrated to directly bind to C1q and activate the classical complement pathway in an antibody‑independent manner, thereby inducing the generation of C3a and the deposition of C5b‑9 [74, 75]. The inverse relationship between elevated CH50 levels and reduced sC5b-9 indicated a greater complement consumption for assembly of the terminal complement complex. Collectively, these findings demonstrated that HMGB1 bridged inflammatory signals and complement-dependent tissue injury in MG, suggesting that 18α-GA could exert therapeutic effects not only by modulating DC function but also by suppressing complement activation. Nevertheless, this exploratory study has several limitations. Firstly, the human sample size is relatively small from a single institution. Secondly, 18α-GA exhibits a multi-target profile, which precludes concluding that its effects are solely mediated by the HMGB1-IRE1α/XBP1 axis. Thus, future studies should investigate other known targets of 18α-GA to clarify the relative contributions of this axis versus other pathways. Thirdly, future studies should employ CD11c-Cre-mediated conditional gene knockout mice and bone marrow chimeras/adoptive cell transfer to elucidate moDC-specific IRE1α/XBP1/NF-κB regulatory networks and validate the therapeutic target of moDCs for MG. Conclusion This study confirmed that HMGB1 induced the dysregulation of moDCs via the IRE1α/XBP1 and NF-κB pathways and established the theoretical basis for 18α-GA as a novel therapeutic agent for MG (Fig. S9). Subsequent studies should further elucidate the complexity of the HMGB1 regulatory network and explore moDCs-based immunotherapeutic strategies to enhance clinical efficacy. Supplementary Information Supplementary Material 1: Fig. S1. Flowchart of animal experiments. (A) Protocol for the therapeutic intervention of experimental autoimmune myasthenia gravis (EAMG) using 18α-GA and HMGB1 neutralizing antibody. (B) Protocol for adoptive transfer therapy of EAMG with 18α-GA-induced tolerogenic moDCs. Supplementary Material 2: Fig. S2. Gating strategy for human blood DC subsets. Supplementary Material 3: Fig. S3. Gating strategy for mouse splenic DC subsets. Supplementary Material 4: Fig. S4. Sorting efficiency of the cell subsets. (A-E) Diagrams of the sorting strategy for human monocytes (CD14+ cells), human naïve CD4+ T (CD4+CD45RA+) cells, human Treg (CD4+CD25+) cells, human Teff (CD4+CD25−) cells, and mouse naïve CD4+ T (CD3ε+CD4+) cells. Supplementary Material 5: Fig. S5. Original RT-PCR electropherograms for Fig. 4E, Fig. 4I, and Fig 6K. Supplementary Material 6: Fig. S6. Original Western blots for Fig. 6E. Supplementary Material 7: Fig. S7. Treatment effect of 18α-GA on EAMG (experimental autoimmune myasthenia gravis) at different doses. (A-D) Comparative analyses of body weight, clinical scores, grip strength, and levels of Torpedo AChRα 146–162 IgG among EAMG mice received 0, 0.25, 0.5, 1, and 2 mg/kg 18α-GA (8 biological replicates, 2 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test compared with the EAMG group at each time point). (E-L) Changes of serum alanine aminotransferase (ALT), aspartate aminotransferase (AST), albumin (ALB), blood urea nitrogen (BUN), creatinine (CR), sodium (Na), potassium (K), and chloride (CL) in mice among the above five groups (8 biological replicates, 2 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). (M-N) High-performance liquid chromatography (HPLC) chromatogram and pharmacokinetic curve of 18α-GA. Data are presented as mean ± SD. *P < 0.05, **P < 0.01, ***P < 0.001. Supplementary Material 8: Fig. S8. Effects of 18α-GA on HMGB1-induced DC activation and complement activation. (A-F) Representative flow cytometry histograms and bar charts showing △median fluorescence intensity (MFI) values among the Control, HMGB1, 18α-GA, HMGB1 neutralizing antibody, HMGB1 + 18α-GA, and HMGB1 + HMGB1 neutralizing antibody groups (3 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). (G-H) Representative flow cytometry plots and a bar chart representing the percentage of IFN-α secretion in pDCs among the above six groups (3 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). (I-L) Hemolytic activity in sera and soluble C5b-9 (sC5b-9) levels in parallel serum supernatants from MG patients and EAMG mice among the above six groups (3 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test compared with the HMGB1 group). Data are presented as mean ± SD. *P < 0.05, **P < 0.01, ***P < 0.001. Supplementary Material 9: Fig. S9. Mechanisms for 18α-GA alleviated myasthenia gravis (MG) via tolerogenic monocyte-derived dendritic cells (moDCs). Elevated HMGB1 (high-mobility group box 1) promotes MG by activating the IRE1α/XBP1/NF-κB axis to induce dysfunctional moDCs, which in turn enhance naïve CD4+ T cell proliferation and skew CD4+ T cell subset polarization. Notably, 18α-GA specifically abrogates HMGB1-mediated effects on moDCs via multiple targets, highlighting its potential as a novel therapeutic strategy for MG. Supplementary Material 10: Table S1. Primers used in the RT-qPCR to amplify the coding region. Acknowledgements We thank the sample donors for taking part in this study and the picture materials provided by Figdraw (https://www.figdraw.com). Authors’ contributions Min Zhang, Zhijun Li, and Bitao Bu proposed the conceptualization, administered the project, and supervised the study. Min Zhang and Bitao Bu acquired funding. Qing Zhang, Zhongya Gu, and Yaxuan Li curated the data, conducted formal analysis, carried out the investigation, and validated the methodology for the study. Mengge Yang, Xue Ma, Xuanxuan Pan, and Zhuajin Bi provided resources and utilized software for the study. Jing Lin, Mengcui Gui, and Furong Wang validated the findings and visualized the data. Qing Zhang and Zhongya Gu wrote the original manuscript. All authors read, critically reviewed, and approved the manuscript. Funding This study was supported by grants from the National Natural Science Foundation of China (No. 82271478 and No. 82371411). Data availability The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. Declarations Ethics approval and consent to partcipate The human component of this study, which involved the collection of clinical samples for experimental analyses, was conducted in accordance with the principles of the Declaration of Helsinki and approved by the Ethics Committee of Tongji Hospital (Protocol ID: TJ-IRB20220748). Written informed consent was obtained from all patients and HCs before sample collection. For the animal experiments, all procedures were approved by the Animal Ethics Committee of Tongji Hospital (Protocol ID: TJH-202404033). Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Qing Zhang and Zhongya Gu contributed equally to this work. Contributor Information Min Zhang, Email: zhang_min_3464@126.com. Zhijun Li, Email: lizhijun@tjh.tjmu.edu.cn. Bitao Bu, Email: bubitao@tjh.tjmu.edu.cn. References Associated Data Supplementary Materials Supplementary Material 1: Fig. S1. Flowchart of animal experiments. (A) Protocol for the therapeutic intervention of experimental autoimmune myasthenia gravis (EAMG) using 18α-GA and HMGB1 neutralizing antibody. (B) Protocol for adoptive transfer therapy of EAMG with 18α-GA-induced tolerogenic moDCs. Supplementary Material 2: Fig. S2. Gating strategy for human blood DC subsets. Supplementary Material 3: Fig. S3. Gating strategy for mouse splenic DC subsets. Supplementary Material 4: Fig. S4. Sorting efficiency of the cell subsets. (A-E) Diagrams of the sorting strategy for human monocytes (CD14+ cells), human naïve CD4+ T (CD4+CD45RA+) cells, human Treg (CD4+CD25+) cells, human Teff (CD4+CD25−) cells, and mouse naïve CD4+ T (CD3ε+CD4+) cells. Supplementary Material 5: Fig. S5. Original RT-PCR electropherograms for Fig. 4E, Fig. 4I, and Fig 6K. Supplementary Material 6: Fig. S6. Original Western blots for Fig. 6E. Supplementary Material 7: Fig. S7. Treatment effect of 18α-GA on EAMG (experimental autoimmune myasthenia gravis) at different doses. (A-D) Comparative analyses of body weight, clinical scores, grip strength, and levels of Torpedo AChRα 146–162 IgG among EAMG mice received 0, 0.25, 0.5, 1, and 2 mg/kg 18α-GA (8 biological replicates, 2 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test compared with the EAMG group at each time point). (E-L) Changes of serum alanine aminotransferase (ALT), aspartate aminotransferase (AST), albumin (ALB), blood urea nitrogen (BUN), creatinine (CR), sodium (Na), potassium (K), and chloride (CL) in mice among the above five groups (8 biological replicates, 2 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). (M-N) High-performance liquid chromatography (HPLC) chromatogram and pharmacokinetic curve of 18α-GA. Data are presented as mean ± SD. *P < 0.05, **P < 0.01, ***P < 0.001. Supplementary Material 8: Fig. S8. Effects of 18α-GA on HMGB1-induced DC activation and complement activation. (A-F) Representative flow cytometry histograms and bar charts showing △median fluorescence intensity (MFI) values among the Control, HMGB1, 18α-GA, HMGB1 neutralizing antibody, HMGB1 + 18α-GA, and HMGB1 + HMGB1 neutralizing antibody groups (3 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). (G-H) Representative flow cytometry plots and a bar chart representing the percentage of IFN-α secretion in pDCs among the above six groups (3 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test). (I-L) Hemolytic activity in sera and soluble C5b-9 (sC5b-9) levels in parallel serum supernatants from MG patients and EAMG mice among the above six groups (3 biological replicates, 3 technical replicates, one-way ANOVA followed by Bonferroni’s multiple-comparison test compared with the HMGB1 group). Data are presented as mean ± SD. *P < 0.05, **P < 0.01, ***P < 0.001. Supplementary Material 9: Fig. S9. Mechanisms for 18α-GA alleviated myasthenia gravis (MG) via tolerogenic monocyte-derived dendritic cells (moDCs). Elevated HMGB1 (high-mobility group box 1) promotes MG by activating the IRE1α/XBP1/NF-κB axis to induce dysfunctional moDCs, which in turn enhance naïve CD4+ T cell proliferation and skew CD4+ T cell subset polarization. Notably, 18α-GA specifically abrogates HMGB1-mediated effects on moDCs via multiple targets, highlighting its potential as a novel therapeutic strategy for MG. Supplementary Material 10: Table S1. Primers used in the RT-qPCR to amplify the coding region. Data Availability Statement The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.