Splenic macrophage-B cell axis drives systemic autoimmune-like pathology in Cerebral Malaria

Abstract

Cerebral malaria (CM) is a severe complication of Plasmodium infection, classically attributed to parasite sequestration and neuroinflammation. Here, we uncover a spleen-centered humoral autoimmune circuit that drives CM pathology. Proteomic analyses identify CD36 as a dominant host-derived antigen enriched in infected red blood cells (iRBCs), triggering anti-CD36 autoantibody production in patients with falciparum malaria. Although contributing to iRBCs clearance, these autoantibodies also target other CD36-expressing cells, thereby driving thrombocytopenia, endothelial injury, and macrophage activation, ultimately amplifying systemic inflammation. Mechanistically, Plasmodium infection recruits Ly6c⁺Glut1^hi macrophages to the spleen through the CCL2-CCR2 axis. These macrophages exhibit elevated proteasome activity and drive B cell activation and anti-CD36 antibody production. Targeting Ly6c⁺Glut1^hi macrophages, we develop Glutoborin, a GLUT1-directed proteasome inhibitor that preferentially suppresses their function, reduces autoantibody production, and alleviates CM-associated pathology in vivo. Together, these findings establish a spleen-centered anti-CD36 autoimmune circuit as a key driver of CM and nominate Ly6c⁺Glut1^hi macrophages as therapeutic targets.

This study identifies a pathogenic anti-CD36 autoantibody produced through aberrant activation of the splenic macrophage-B cell axis, thereby linking humoral autoimmunity to systemic inflammation and cerebral malaria pathology.

Introduction

Cerebral malaria (CM) is a severe and often fatal complication of Plasmodium falciparum infection^1,2. Traditionally, CM pathology has been attributed to the sequestration of infected erythrocytes in cerebral microvessels³. However, increasing evidence suggests that immunopathology, rather than mechanical obstruction alone, plays a central role in disease progression^4,5. Excessive cytokine release, endothelial activation, and lymphocyte infiltration disrupt the blood-brain barrier (BBB) and drive neuronal injury⁶. Although the role of T cells in CM has been extensively characterized⁷, the contribution of B cells and its mediated humoral immune response remains poorly understood.

Recent studies have highlighted striking parallels between CM and systemic autoimmune disorders, including systemic immune activation, loss of tolerance, and the production of pathogenic autoantibodies^8,9. In diseases such as systemic lupus erythematosus (SLE) and immune thrombocytopenia (ITP), autoantibodies targeting blood cells and endothelial surfaces contribute directly to vascular damage and systemic inflammation^10–12. Similarly, patients with malaria frequently develop autoantibodies against platelets and endothelial antigens^13–15, suggesting the involvement of a maladaptive B cell response in CM immunopathogenesis.

As a central hub of immunological coordination, the spleen integrates innate and adaptive immune responses and stands as a critical guardian against blood-borne pathogens¹⁶. Among splenic immune populations, macrophages function as antigen-sorting stations by capturing large pathogen-associated or damage-associated antigens via phagocytosis. These antigens are subsequently processed and released as smaller fragments through vesicular exocytosis¹⁷. Neighboring B cells can then acquire these processed antigens through receptor-mediated uptake, thereby facilitating B cell activation and antibody production under inflammatory condition¹⁸. While such macrophage-B cell crosstalk has been implicated in systemic autoimmunity, its contribution to infection-induced inflammation, particularly in CM, remains poorly characterized.

In this study, we identify an immunopathological axis in which Ly6c⁺Glut1^hi macrophages are recruited into the spleen by Ccl2⁺ red pulp macrophages and subsequently activate B cells to promote the production of anti-CD36 antibodies. While these antibodies contribute to parasite clearance, they also drive host tissue injury by exacerbating thrombocytopenia, inducing endothelial damage, and amplifying macrophage-mediated inflammation. To disrupt this axis, we develop Glutoborin, a GLUT1-specific proteasome inhibitor that preferentially impairs the pro-inflammatory function of Ly6c⁺Glut1^hi macrophage, curtails autoantibody production, and alleviates systemic inflammation. Together, our findings suggest that CM is not solely a parasite-driven disease but also involves a spleen-centered autoimmune-like process.

Results

Spleen-driven systemic inflammation governs cerebral malaria lethality

To delineate the immunopathology of cerebral malaria (CM), we employed a lethal murine Plasmodium berghei ANKA (Pb) infection model. Consistent with previous reports^19,20, infected mice developed fatal CM between days 7 and 9 post-infection (Supplementary Fig. 1a). Giemsa staining confirmed the presence of Pb-infected erythrocytes on day 7 post-infection (Supplementary Fig. 1b), and GFP expression enabled flow cytometric quantification, revealing approximately 7% parasitemia (Supplementary Fig. 1c).

Histopathological analysis and Evans blue (EB) dye extravasation assays revealed extensive cerebrovascular hemorrhage and thrombosis in Pb-infected mice (Supplementary Fig. 1d, e), accompanied by significant blood-brain barrier (BBB) disruption (Fig. 1a). Notably, inflammatory pathology extended beyond the brain, with prominent pulmonary vascular leakage, as well as dysfunction in the hepatic, renal, and cardiac tissues (Fig. 1b, c). In addition, blood glucose progressively declined from day 5 post-infection (Supplementary Fig. 1f). By day 7, infected mice exhibited pronounced thrombocytopenia and mild leukopenia (Fig. 1d).

Fig. 1. Spleen-driven systemic inflammation governs CM lethality.

a, b Representative Evans blue (EB) staining of brains (a) and lungs (b) from uninfected and Pb-infected mice on day 7. Right panel, quantification of EB staining by OD620 values (n = 3 mice, **P (brain) = 0.0092, **P (lung) = 0.0048). c Heatmap of blood biochemical markers for liver, kidney, and cardiac function in uninfected and Pb-infected mice on day 7 (n = 3 mice). AST, Aspartate aminotransferase; ALT, Alanine aminotransferase; TP, Total protein; ALB, Albumin; ALP, Alkaline phosphatase; UA, Uric acid; CREA, Creatinine; UREA, Urea nitrogen; CK, Creatine kinase; α-HBDH, Alpha-hydroxybutyrate dehydrogenase; CK-MB, Creatine Kinase-Myocardial Band; LDH, Lactate dehydrogenase. d Heatmap of CBC parameters in uninfected and Pb-infected mice on day 7 (n = 3 mice). e Representative spleen image and splenocyte counts from uninfected and Pb-infected mice (n = 4 mice, **P = 0.0052). f–j Mice underwent sham surgery (Sham) or splenectomy (SPX), followed by intraperitoneal injection with iRBCs. Samples were collected on day 7. f Survival curves of the Sham and SPX groups (n = 5 mice, **P = 0.0015). g EB staining of brain from the Sham and SPX groups. Right panel, OD620 quantification (n = 4 mice, ****P < 0.0001). h H&E staining of brain tissue from Sham and SPX groups. Arrowheads indicate hemorrhagic spots. Scale bar, 250 μm (top), 50 μm (bottom). The percentage of hemorrhagic area relative to total section area is quantified on the right (n = 3 mice, *P = 0.0219). i Platelet counts in Sham and SPX groups (n = 5 mice, ****P < 0.0001). j Expression of Cxcl10, Isg15, and Tnf in brains from uninfected, Sham and SPX mice, measured by qRT-PCR (n = 3 mice, **P (Isg15) = 0.0057, **P (Cxcl10) = 0.0055, ****P (Tnf) < 0.0001). Data are representative of two independent experiments (mean ± SEM). Significance was determined using unpaired Student’s t test and Log-rank (Mantel-Cox) test; not significant (ns), P > 0.05. Source data are provided as a Source Data file.

While the overall immune cell composition in lymph nodes remained unchanged, a reduction in total immune cell numbers was observed (Supplementary Fig. 1g–k). In contrast, Pb-infected mice exhibited splenomegaly with expansion of T and B cells (Fig. 1e and Supplementary Fig. 1l–o), implicating the spleen as a major site of immune activation. To assess its functional contribution, we performed splenectomy one week before Pb infection. Despite higher parasitemia, splenectomized mice exhibited significant protection from CM, prolonged survival and reduced cerebral pathology (Fig. 1f–h and Supplementary Fig. 1p). Moreover, both thrombocytopenia and systemic inflammation were markedly attenuated (Fig. 1i, j). These findings highlight the spleen as a pivotal orchestrator of malarial immunopathology in CM.

B cell-mediated humoral immunity exacerbates CM immunopathology

Given the marked expansion of splenic T and B cells during Pb infection, we next sought to explore their relative contributions to disease progression. To this end, we selectively depleted CD4⁺ T cells, CD8⁺ T cells and B cells using specific monoclonal antibodies, respectively (Supplementary Fig. 2a, b). Consistent with previous reports^21,22, depletion of T cells partially attenuated disease severity (Fig. 2a). Notably, B cell depletion completely abrogated disease progression and markedly reduced cerebral pathology, as evidenced by diminished EB dye extravasation and fewer cerebral hemorrhagic lesions (Fig. 2a–c). Moreover, B cell depletion ameliorated thrombocytopenia (Fig. 2d).

Fig. 2. B cell response orchestrates disease severity in CM.

a Survival curves of infected mice subjected to the indicated treatments (n = 6 mice). b Representative EB staining of brains and OD620 quantification (n = 4 mice, *P = 0.0466). c H&E staining of brain tissues. Arrowheads indicate hemorrhage or thrombi. Scale bar, 250 μm (top), 50 μm (bottom). Quantification of hemorrhagic area (n = 3 mice, *P = 0.0340). d Platelet counts in two groups of mice (n = 4 mice, ***P (Pb-infected, Iso control vs. αCD20 IgG) = 0.0002, ***P (Pb-infected vs. Pb-infected αCD20 IgG) = 0.0004). e Serum from days 5-7 post-infection (designated #1-#3) was injected into splenectomized Pb-infected mice (SPX) on day 7. f Survival curves. g Platelet counts after serum transfer (Control, n = 5 mice, #1, n = 5 mice, #2, n = 6 mice, #3, n = 6 mice, *P = 0.0130). h Parasitemia was quantified on the following day in splenectomized Pb-infected mice after passive transfer of serum (groups #1-#3) or control treatment (Control, n = 5 mice, #1, n = 5 mice, #2, n = 6 mice, #3, n = 6 mice, *P = 0.0444, **P = 0.0020). i ELISA quantification of anti-platelet IgM (left) and IgG (right) in infected vs. uninfected mice on day 7 (n = 3 mice, ****P < 0.0001). j Anti-platelet IgM levels in serum from Sham vs. SPX mice on day 7 (n = 3 mice, ****P < 0.0001). k ELISA showing purified IgM reactivity vs. isotype control (n = 3 mice, ****P < 0.0001). l Survival curves of splenectomized Pb-infected mice receiving purified IgM or isotype control (n = 5 mice). m Representative EB staining of brains and OD620 quantification (n = 3 mice, **P = 0.0030). n Platelet counts in two groups of mice (n = 5 mice, ****P < 0.0001). Data are representative of two independent experiments (mean ± SEM). Significance was determined using a two-tailed unpaired Student’s t test and Log-rank (Mantel-Cox) test. ns, P > 0.05. Source data are provided as a Source Data file.

Recognizing the central role of B cells in antibody production, we next investigated infection-induced changes in the humoral immune response. Proteomic profiling of serum immunoglobulins revealed extensive remodeling of the antibody repertoire during infection (Supplementary Fig. 2c). To assess the functional impact of this humoral shift, we transferred serum from infected donors into splenectomized mice, which are otherwise resistant to Pb-induced pathology (Fig. 2e). As shown in Fig. 2f, serum transfer restored disease susceptibility and resulted in high mortality in recipient mice. Notably, transferred serum induced severe thrombocytopenia while simultaneously reducing parasite burden (Fig. 2g, h), indicating that infection-induced antibodies exert both anti-parasitic and pathogenic effects.

In light of the severe thrombocytopenia observed following serum transfer, we hypothesized that infection induces the production of anti-platelet antibodies. To test this, serum from infected mice was transferred into uninfected recipients, resulting in rapid platelet depletion (Supplementary Fig. 2d, e). ELISA further confirmed the production of anti-platelet IgM and IgG post-infection (Fig. 2i). Notably, splenectomized mice failed to generate anti-platelet IgM after infection (Fig. 2j).

To directly assess whether anti-platelet IgM contributes to CM pathogenesis, we purified IgM from day-7 post-infection sera (Supplementary Fig. 2f, g). ELISA confirmed that the purified IgM retained anti-platelet reactivity comparable to that of the original infected sera (Fig. 2k). Passive transfer of purified IgM into splenectomized Pb-infected recipient mice (day-7) recapitulated the increased mortality phenotype (Fig. 2l), with evidence of BBB disruption accompanied by reduced platelet counts (Fig. 2m, n). Together, these findings support a direct pathogenic role of infection-induced IgM antibodies in CM.

Finally, to test whether humoral and cellular immunity can cooperate to drive pathology in a susceptible context, we performed adoptive transfer experiments in splenectomized Pb-infected recipient mice using purified day-7 IgM and/or CD8⁺CD44⁺ T cells. Transfer of high-dose IgM alone was sufficient to induce lethality, whereas transfer of CD8⁺CD44⁺ T cells alone resulted in partial lethality (Supplementary Fig. 2h). Notably, combined transfer of low-dose IgM together with CD8⁺CD44⁺ T cells markedly exacerbated mortality compared with either component alone (Supplementary Fig. 2h). Collectively, these data support a model in which humoral and cellular immune response contribute jointly to promote CM pathology.

Dual effects of anti-CD36 antibodies in parasite clearance and CM pathogenesis

To identify antigenic targets of infection-induced pathogenic antibodies, we performed proteomic profiling of immune complexes isolated from platelet lysate-serum mixtures using Protein L-coated beads. Mass spectrometry identified several platelet membrane proteins, including CD36, GPⅠb, and GPV, as potential self-antigens (Fig. 3a). Among these, CD36 emerged as a top candidate owing to its high expression not only on platelets but also on endothelial cells and activated macrophages (Supplementary Fig. 3a, b). Notably, RBCs infected with Pb showed increased CD36 expression compared to those infected with Plasmodium yoelii (Py, 17XNL is a non-lethal strain; 17XL is a highly virulent strain), implicating CD36 as a parasite-modulated host factor (Supplementary Fig. 3c, d).

Fig. 3. Anti-CD36 antibodies act as key effectors in CM pathogenesis.

a Serum from uninfected and Pb-infected mice (Day 7) was used in a platelet proteins pull-down assay. Target antigens of anti-platelet antibodies were identified by MS, highlighting CD36 and other bound proteins. b SDS-PAGE of purified CD36 protein with Coomassie staining. c ELISA of anti-CD36 IgM in the serum of infected mice (day 7) and uninfected mice (n = 3 mice, ****P < 0.0001). d Platelet counts in mice from each group (n = 3 mice, ***P = 0.0001). e ELISA of anti-CD36 antibodies in healthy controls (n = 10) vs. Pf-infected patients (n = 8, ****P < 0.0001). f Flow cytometric analysis of GFP⁺ iRBCs incubated with isotype control or purified anti-CD36 IgG in the absence or presence of complement, with or without recombinant CD36 protein. Representative histograms (left) and summary quantification (right) (n = 4 cell cultures, **P = 0.0045, ***P = 0.0004). Flow cytometry gating strategies are shown in Supplementary Fig. 9a. g Purified anti-CD36 IgG or isotype IgG was passively transferred into splenectomized Pb-infected mice on day 7 post-infection. h Platelet counts in two groups of mice (n = 5 mice, ****P < 0.0001). i EB staining of brains and OD620 quantification (n = 3 mice, ****P < 0.0001). j Survival curves of two groups of mice (n = 5 mice). k–m Mice were injected with iRBCs and randomized on day 5 post-infection to receive CD36 protein (50 μg) or BSA (50 μg) for two consecutive days before analysis. k Schematic representation of the CD36 protein treatment experiment. l Platelet counts in two groups (n = 10 mice, ***P = 0.0003). m Survival curves of two treatment groups (n = 10 mice). n Platelet counts in the anti-GPIbα IgG-treated naïve mice. o Survival curves of Pb-infected splenectomized mice subjected to the indicated treatments (n = 5 mice). Data are representative of two independent experiments (mean ± SEM). Significance was determined using two-tailed paired Student’s t test, unpaired Student’s t test and Log-rank (Mantel-Cox) test. ns, P > 0.05. Source data are provided as a Source Data file.

To further assess the presence of anti-CD36 antibodies, we purified recombinant CD36 protein and performed an ELISA assay (Fig. 3b). As shown in Fig. 3c, Pb infection elicited a robust anti-CD36 antibody response, whereas Py infection induced only minimal reactivity. In parallel, only Pb-infected mice developed thrombocytopenia (Fig. 3d), supporting a potential pathogenic role for anti-CD36 antibodies. Furthermore, we quantified anti-CD36 antibody titers in patients with Plasmodium falciparum (Pf) infection. Compared with healthy controls (HC), anti-CD36 antibody levels were significantly elevated in malaria patients (Fig. 3e).

To further validate the induction of CD36-reactive B cells during Plasmodium infection, we established a flow cytometry-based antigen-binding assay using recombinant CD36-His protein. Briefly, B cells were incubated with CD36-His on ice to enable antigen binding while minimizing internalization, followed by staining with fluorophore-conjugated anti-His or anti-CD36 antibodies for detection (Supplementary Fig. 3e). As a proof of principle, CD36 immunization served as a positive control that B cells from draining lymph nodes exhibited a distinct CD36-binding population (~ 4.28%), whereas BSA incubation and non-immunized controls showed no detectable signal (Supplementary Fig. 3f), supporting the specificity and feasibility of this assay. Using this approach, we analyzed splenic B cells from Pb-infected mice at day 7 and consistently detected a CD36-reactive population (~ 2.5-3%) using both anti-His and anti-CD36 detection strategies (Supplementary Fig. 3g, h). These data indicate that CD36-responsive autoreactive B cells are induced during Pb infection.

We next examined whether anti-CD36 antibodies contribute to parasite clearance. Given that passive transfer of day-7 serum reduced circulating iRBC levels (Fig. 2h), we generated anti-CD36 IgG by immunizing guinea pigs with recombinant CD36 protein (Supplementary Fig. 3i–k). In vitro assays showed that anti-CD36 IgG mediated robust killing of iRBCs in a complement-dependent manner, which was partially blocked by recombinant CD36 protein, confirming antigen specificity (Fig. 3f).

Passive transfer of serum from Pb-infected mice into splenectomized recipients was sufficient to induce fatal CM (Fig. 2f), suggesting a pathogenic role of anti-CD36 antibodies. To directly test this, purified anti-CD36 IgG was administered to Pb-infected splenectomized mice (Fig. 3g). Anti-CD36 IgG transfer resulted in severe thrombocytopenia, disruption of BBB integrity, and reduced survival (Fig. 3h–j). In contrast, administration of recombinant CD36 protein partially rescued platelet counts and modestly improved survival (Fig. 3k–m), indicating that soluble CD36 can neutralize pathogenic antibodies.

To determine whether thrombocytopenia alone accounts for lethality, we depleted platelets using an anti-GPⅠbα antibody. Although this treatment efficiently reduced platelet counts, it failed to induce CM-like mortality (Fig. 3n, o), demonstrating that the pathogenic effects of anti-CD36 antibodies extend beyond platelet depletion. Collectively, these findings reveal a dual role for anti-CD36 antibodies, which contribute to parasite clearance while simultaneously exacerbating cerebral malaria pathogenesis.

Anti-CD36 antibodies aggravate inflammation via the induction of endothelial injury and macrophage activation

To elucidate the mechanistic basis of anti-CD36 IgG-mediated pathology, we examined their effects on CD36-expressing endothelial cells and macrophages. In the murine endothelial cell line H5V, treatment with anti-CD36 IgG rapidly induced cell death, as evidenced by morphological alterations, elevated lactate dehydrogenase (LDH) release, and increased 7-AAD staining (Fig. 4a–e). These features are characteristic of complement-dependent cytotoxicity (CDC), indicating that anti-CD36 IgG directly triggers endothelial injury. To determine whether infection-induced antibodies exert similar effects, we purified IgM from day-7 infected sera and assessed endothelial cytotoxicity. Purified IgM induced significant endothelial cell death only in the presence of complement (Supplementary Fig. 4a, b).

Fig. 4. Anti-CD36 antibodies exacerbate CM via triggering endothelial injury and macrophage activation.

a Schematic of anti-CD36 antibodies induced endothelial damage. b Representative images of H5V endothelial cells treated with isotype control or purified anti-CD36 IgG in the presence or absence of complement. Scale bar, 100 μm. c LDH release assay of endothelial damage (n = 3 cell cultures, ***P = 0.0002). d, e Flow cytometric analysis (d) and quantification (e) of endothelial cell death (n = 3 cell cultures, ****P < 0.0001). f qRT-PCR analysis of Il1b expression in RAW264.7 cells treated with LPS (50 ng/mL) after pretreatment with isotype control or anti-CD36 IgG (n = 3 cell cultures, ****P < 0.0001). g Survival curves of mice treated with LPS (25 mg/kg) plus isotype control or anti-CD36 IgG (n = 5 mice). h qRT-PCR analysis of Il1b expression in liver (**P = 0.0016, ***P = 0.0002) and kidney (*P (LPS vs. LPS + αCD36 IgG) = 0.0209, *P (LPS vs. LPS + αCD36 IgG) = 0.0138) of mice treated with LPS (15 mg/kg) plus isotype control or anti-CD36 IgG for 12 h (n = 3 mice). i Heatmap of differentially expressed genes under four conditions: untreated, anti-CD36 IgG alone, LPS alone, and LPS with anti-CD36 IgG pretreatment. Representative enriched pathways are shown. j Schematic of anti-CD36 antibodies-induced macrophage activation. k Representative flow cytometry histograms showing surface CD36 levels. Right, quantification of CD36 MFI (n = 3 cell cultures, ***P = 0.0006). l qRT-PCR analysis of Il1b expression in RAW264.7 cells treated with LPS (50 ng/mL) after pretreatment with OA, anti-CD36 IgG or the combination (n = 3 cell cultures, *P = 0.0106, **P = 0.0084, ***P = 0.0003). Data are representative of two independent experiments (mean ± SEM). Statistical significance was determined using a two-tailed unpaired Student’s t test and Log-rank (Mantel-Cox) test. ns, P > 0.05. Source data are provided as a Source Data file.

We next examined the effects of anti-CD36 antibodies on macrophages. In contrast to endothelial cells, RAW264.7 macrophages were largely resistant to complement-dependent cytotoxicity following anti-CD36 IgG treatment, as indicated by minimal LDH release (Supplementary Fig. 4c). Instead, anti-CD36 IgG pre-treatment altered the morphology of LPS-stimulated macrophages, consistent with enhanced cellular activation (Supplementary Fig. 4d). Consistent with this, anti-CD36 IgG had minimal effects alone but significantly potentiated LPS-induced expression of pro-inflammatory cytokines, including Il1b and Il6 (Fig. 4f and Supplementary Fig. 4e). These results indicate that anti-CD36 antibody engagement sensitizes macrophages to inflammatory stimuli and drives their polarization toward a pro-inflammatory state.

Consistent with these in vitro findings, in vivo co-administration of anti-CD36 IgG and LPS resulted in significantly reduced survival and increased expression of inflammatory cytokines such as Il6 and Tnf in multiple tissues (Fig. 4g, h and Supplementary Fig. 4f, g), demonstrating that anti-CD36 antibodies amplify inflammatory injury under endotoxin challenge.

To investigate the molecular basis of this effect, we performed transcriptomic profiling of macrophages under different treatment conditions. Anti-CD36 IgG alone led to suppression of genes associated with lipid metabolism and energy homeostasis, without inducing inflammatory gene expression under basal conditions (Fig. 4i). Upon LPS stimulation, however, genes specifically upregulated in the anti-CD36 IgG plus LPS condition were enriched for inflammatory and LPS-responsive pathways, while metabolic pathways were further suppressed (Fig. 4i). Together, these findings indicate that anti-CD36 antibody engagement disrupts metabolic homeostasis and primes macrophages for amplified inflammatory responses upon subsequent stimulation.

Based on previous reports linking CD36-mediated lipid uptake to metabolic regulation²³, we hypothesized that anti-CD36 antibodies may promote macrophage activation by interfering with CD36-dependent lipid metabolism (Fig. 4j). As expected, anti-CD36 IgG induced CD36 internalization in macrophages, as demonstrated by flow cytometry under Fc receptor-blocking conditions (Fig. 4k and Supplementary Fig. 4h).

To further test this mechanism, macrophages were treated with oleic acid (OA), a known CD36 substrate, prior to LPS stimulation²⁴. OA treatment markedly suppressed LPS-induced Il1b and Il6 expression in RAW264.7 macrophages (Fig. 4l and Supplementary Fig. 4i). However, this inhibitory effect was largely abrogated in the presence of anti-CD36 IgG, which instead potentiated LPS-driven proinflammatory gene expression (Fig. 4l and Supplementary Fig. 4i), indicating that antibody-mediated disruption of CD36 impairs lipid-dependent immunoregulation. Collectively, these findings demonstrate that anti-CD36 antibodies exert cell type-specific pathogenic effects by inducing complement-dependent endothelial cell death while simultaneously promoting macrophage activation through disruption of CD36-mediated metabolic regulation.

Ly6c⁺Glut1^hi macrophages are recruited in spleen and facilitate B cell activation

To elucidate the mechanisms underlying Pb-induced splenic B cell activation, we performed single-cell RNA sequencing (scRNA-seq) of splenic immune cells. Clustering analysis identified 11 immune populations, with a marked expansion of B cells following infection (Supplementary Fig. 5a and Supplementary Table 1). Further subset analysis, combined with IFN-I signaling score assessment, revealed that activated B cells were significantly increased in both abundance and functional activity upon Pb infection (Fig. 5a, b). These findings were further validated by flow cytometry (Supplementary Fig. 5b).

Fig. 5. Ly6c⁺Glut1^hi macrophages are recruited to the spleen and promote B cell activation.

Splenic lymphocytes were isolated on day 5 post-infection, and CD45⁺ cells were subjected to single-cell RNA sequencing. a UMAP plot of five B cell subsets from uninfected (5,610) and Pb-infected (6497) mice, color-coded by cluster identity. Right, proportions of each B cell subset. b Feature plot showing the IFN-α score of B cells. c Serum anti-CD36 antibody levels measured by ELISA on day 7 post-infection in mice treated with clodronate or PBS liposomes (n = 5 mice, ****P < 0.0001). d Survival curves of infected mice treated with PBS vs. clodronate liposomes (n = 12 mice). e EB staining of brains and OD620 quantification (n = 3 mice, ***P = 0.0003). f UMAP plot of five macrophage subsets from uninfected (1,211) and Pb-infected mice (872). Right, proportion of macrophage subsets. g CellChat analysis of interactions between activated B cells and macrophage subsets. Line thickness indicates interaction strength, with values shown in the plot. h Dot plot showing the expression of genes related to inflammation, UPS and exocytosis across splenic macrophage subsets. i CellChat analysis illustrating interaction between other macrophage subsets and Ly6c⁺Glut1^hi macrophages. Line thickness indicates interaction strength, with values shown in the plot. j qRT-PCR analysis of Ccl2 expression in sorted splenic red pulp macrophages (RPM) from Sting^+/+ and Sting^−/− mice at 24 hrs post-Pb infection (n = 3 mice, *P (uninfected Sting^+/+ vs. Pb-infected Sting^−/−) = 0.0124, *P (Pb-infected, Sting^+/+ vs. Sting^−/−) = 0.0234, **P = 0.0098). k Representative flow cytometry plots for splenic Ly6c⁺Ccr2⁺ macrophages from Sting^+/+ and Sting^−/− mice at the indicated times after infection. Flow cytometry gating strategies are shown in Supplementary Fig. 9e. l Summary of Ly6c⁺Ccr2⁺ macrophage accumulation in spleen shown as frequency (**P = 0.0011) and absolute number (**P = 0.0073) at the indicated time points (n = 3 mice). Data are representative of two independent experiments (mean ± SEM). Significance was determined using a two-tailed unpaired Student’s t test and Log-rank (Mantel-Cox) test. ns, P > 0.05. Source data are provided as a Source Data file.

To identify upstream regulators of B cell activation, we performed CellChat analysis, which revealed robust ligand-receptor interactions between macrophages and activated B cells (Supplementary Fig. 5c). To determine whether splenic macrophages are essential for B cell activation, we depleted macrophages with clodronate liposomes two days prior to infection (day -2) (Supplementary Fig. 5d, e). As shown in Fig. 5c–e, depletion of macrophages markedly reduced anti-CD36 antibody levels, conferred complete survival and preserved BBB integrity.

Further characterization of splenic macrophages during infection revealed five macrophage subsets, including a marked accumulation of Ly6c⁺Glut1^hi macrophages (Supplementary Fig. 5f and Fig. 5f). Notably, these cells exhibited close crosstalk with B cells (Fig. 5g). Compared to other macrophage subsets, Ly6c⁺Glut1^hi macrophages displayed a unique transcriptional signature characterized by high expression of genes related to inflammation, proteasome activity (antigen processing) and vesicle-mediated exocytosis (antigen presentation) (Fig. 5h).

Of note, our scRNA-seq analysis indicated a potential role for Ccl2⁺ red pulp macrophages (RPMs) in Ly6c⁺Glut1^hi macrophages recruitment via the CCL2-CCR2 axis (Fig. 5i and Supplementary Fig. 5g, h). To validate this result in vivo, we employed the Sting^−/− mice, given previous studies that STING signaling regulates CCL2 expression (Supplementary Fig. 5i, j). Consistent with these findings^25,26, following Pb infection, Sting^−/− mice exhibited markedly reduced Ccl2 expression in splenic RPMs (Fig. 5j). Notably, disruption of the STING-CCL2 axis attenuated splenomegaly and decreased the accumulation of CD45⁺ immune cells in the spleen, while exerting minimal effects on peripheral blood leukocyte counts (Supplementary Fig. 5k, l). Flow cytometric analysis further demonstrated a significant reduction in both the frequency and absolute number of Ly6c⁺Ccr2⁺ macrophages in the spleens of Sting^−/− mice (Fig. 5k, l). Importantly, genetic inactivation of STING-CCL2 signaling was associated with prolonged survival following infection (Supplementary Fig. 5m). Collectively, these results identify monocyte-derived pro-inflammatory Ly6c⁺Glut1^hi macrophages as critical mediators that bridge innate immune cell recruitment and pathogenic B cell activation during Pb infection.

Glutoborin preferentially restricts Ly6c⁺Glut1^hi macrophage function

Given the critical role of Ly6c⁺Glut1^hi macrophages in anti-CD36 antibody production and CM pathogenesis, we sought to develop a strategy to selectively target this pathogenic macrophage subset. Notably, Ly6c⁺Glut1^hi macrophages exhibited elevated proteasome activity (Fig. 5h), prompting us to explore proteasome inhibition as a potential intervention. As a proof of concept, we employed bortezomib, a boronic acid-based proteasome inhibitor. In vitro, activation of RAW264.7 macrophages increased proteasome activity, which was effectively suppressed by bortezomib, accompanied by reduced macrophage activation (Supplementary Fig. 6a, b).

Based on the high levels of GLUT1 expression, an in silico screen of our in-house proteasome inhibitor library using SwissTargetPrediction identified a lead compound with a 91.33% predicted probability of GLUT1 binding and a prediction accuracy of 98.75%^27–29 (Fig. 6a). Molecular docking analysis further showed that this compound exhibited higher predicted binding affinity to GLUT1 with a docking score of − 12.52 compared to the established GLUT1 inhibitor BAY-876 with a score of − 9.44 (Fig. 6b). To validate these computational predictions, we performed surface plasmon resonance (SPR) assays using purified GLUT1 protein (Supplementary Fig. 6c). As shown in Fig. 6c, this compound bound GLUT1 with a dissociation constant of 119 nM, indicating a high-affinity and specific interaction. Given its boronic acid moiety and dual proteasome and GLUT1 targeting capacity, we designated this compound Glutoborin (Fig. 6d, e, Supplementary Fig. 6d and Supplementary note).

Fig. 6. Glutoborin preferentially inhibits Glut1⁺ macrophage.

a SwissTargetPrediction results for Glutoborin binding likelihood to proteasome and glucose transporter. b Three-dimensional molecular docking models of Glutoborin and BAY-876 bound to GLUT1. c SPR analysis of the interaction between immobilized GLUT1 and different concentrations of Glutoborin. K_D, dissociation constant. d Proteasome activity (Suc-LLVY-AMC fluorescence) in HEK293T cells treated with Bortezomib or Glutoborin (n = 3 cell cultures, ****P < 0.0001). e Chemical structures of Bortezomib and Glutoborin, with borate ion moieties highlighted. f Proteasome activity in GLUT1-knockdown 4T1 cells after Glutoborin treatment (n = 3 cell cultures, ***P = 0.0001, ****P < 0.0001). g, h RAW264.7 cells were stimulated with LPS (100 ng/mL) and mIFNγ (50 ng/mL) and then treated with Glutoborin. g Proteasome activity assessed via Suc-LLVY-AMC fluorescence (n = 3 cell cultures, ****P < 0.0001). h Expression level of Tnf was detected by qRT-PCR (n = 3 cell cultures, ****P < 0.0001). i Chemical structure of a boronic-acid-deficient Glutoborin analog (G-analog). j SPR analysis of the interaction between immobilized GLUT1 and different concentrations of G-analog. k Relative mRNA expression levels of the pro-inflammatory cytokine Tnf in RAW264.7 cells were measured by qRT-PCR under the indicated stimulation conditions (n = 3 cell cultures, ***P = 0.0001). l Gene Ontology enrichment analysis of quantitative proteomic data from macrophages treated with Glutoborin vs. Vehicle. m Heatmap of differentially expressed proteins highlighting increased abundance of proteasome subunits following Glutoborin treatment. n Immunoblot analysis of the levels of IκBα and p-IκBα in RAW264.7 cells under the indicated conditions. o Immunoprecipitation of endogenous IκBα followed by immunoblotting for ubiquitin. Data are representative of two independent experiments (mean ± SEM). Significance was determined using a two-tailed unpaired Student’s t test. ns, P > 0.05. Source data are provided as a Source Data file.

To validate GLUT1 specificity at the cellular level, we firstly used the shRNAs to knockdown the endogenous Slc2a1 in 4T1 cells (Supplementary Fig. 6e). As shown in Supplementary Fig. 6f, depletion of GLUT1 did not alter basal proteasome activity under quiescent conditions. Upon Glutoborin treatment, GLUT1-deficient cells exhibited marked resistance to Glutoborin-induced suppression of proteasome activity (Fig. 6f and Supplementary Fig. 6g), whereas cells with intact GLUT1 expression remained highly sensitive. Similar results were also observed in macrophages. LPS and IFNγ stimulation induced GLUT1 expression in RAW264.7 cells (Supplementary Fig. 6h), which were subsequently treated with Glutoborin. Glutoborin markedly inhibited proteasome activity and suppressed the expression of pro-inflammatory genes, including Tnf and Il6 (Fig. 6g, h and Supplementary Fig. 6i, j).

To determine whether these immunomodulatory effects depend on proteasome inhibition, we performed structure-function analysis using a Glutoborin analog lacking the boronic acid moiety (Fig. 6i and Supplementary note). SPR analysis showed that this analog retained comparable binding affinity for GLUT1 (Fig. 6j). However, only Glutoborin, but not the analog, suppressed the acquisition of a pro-inflammatory phenotype in RAW264.7 cells stimulated with LPS and IFNγ (Fig. 6k and Supplementary Fig. 6k).

To further define the molecular pathways targeted by Glutoborin, we performed quantitative proteomic profiling of macrophages treated with or without Glutoborin. Gene ontology analysis revealed marked suppression of NF-κB-associated programs, whereas pathways related to protein folding and protein degradation were upregulated (Fig. 6l). Notably, multiple proteasome-related components, including members of the Psma and Psmb families, were increased (Fig. 6m), supporting a proteasome-linked mechanism of action.

To further examine the regulatory effect of Glutoborin on NF-κB signaling, we assessed the dynamics of total IκBα protein and its phosphorylated form in RAW264.7 macrophages under both quiescent and inflammatory conditions. Under basal conditions, Glutoborin treatment increased phosphorylated IκBα levels compared with vehicle-treated controls (Fig. 6n). Upon co-stimulation with LPS and IFNγ, control cells exhibited a pronounced reduction in total IκBα together with robust accumulation of phosphorylated IκBα, whereas Glutoborin-pretreated cells failed to show a comparable decrease in total IκBα despite sustained high levels of phosphorylated IκBα (Fig. 6n). Furthermore, immunoprecipitation of endogenous IκBα followed by assessment of ubiquitination revealed increased levels of ubiquitinated IκBα in Glutoborin-treated cells (Fig. 6o). Collectively, these results indicate that Glutoborin limits IκBα degradation by restricting proteasome activity and thereby suppresses downstream NF-κB activation.

Glutoborin protects against CM by inactivating Ly6c⁺Glut1^hi macrophage

To evaluate the therapeutic efficacy of Glutoborin against CM, infected mice were treated with Glutoborin, G-analog or bortezomib (Fig. 7a). While bortezomib conferred partial protection, all Glutoborin-treated mice survived upon Pb infection (Fig. 7b). In contrast, the G-analog exhibited little to no protective effect, underscoring the functional importance of the boronic acid moiety (Supplementary Fig. 7a). The platelet counts in Glutoborin-treated mice were maintained within the normal physiological range and a notable reduction in BBB disruption and neuroinflammation was also observed in these mice (Fig. 7c–g). Furthermore, both total IgM levels and anti-platelet antibody titers were significantly decreased (Fig. 7h, i). Beyond its neuroprotective effects, Glutoborin treatment also restored normoglycemia and ameliorated splenomegaly (Supplementary Fig. 7b and Fig. 7j).

Fig. 7. Glutoborin restricts Ly6c⁺Glut1^hi macrophage function and protects the host against CM.

a Schematic of the treatment protocol in Pb-infected mice. b Survival curves (n = 10 mice). c Platelet counts in Vehicle- (n = 10 mice), Bortezomib- (n = 8 mice) and Glutoborin- (n = 10 mice) treated mice (*P = 0.0405, ****P < 0.0001). d, e EB staining of brains from uninfected (n = 3 mice), Vehicle- (n = 6 mice) and Glutoborin-treated (n = 6 mice) mice (d). OD620 quantification (e) (*P = 0.0270, **P = 0.0091). f H&E staining of brains. Arrowheads indicate hemorrhage. Scale bar, 250 μm (top), 50 μm (bottom). Right, hemorrhagic area/total section area (n = 3 mice, *P = 0.0161). g qRT-PCR analysis of Cxcl10 (*P = 0.0199), Isg15 (*P = 0.0216), and Tnf (*P = 0.0439) in brains (n = 3 mice). h Serum total IgM on days 4-6 by ELISA (n = 3 mice, ***P (Day 4) = 0.0004, ***P (Day 5) = 0.0004, ***P (Day 6) = 0.0003). i Anti-platelet IgM levels on day 7 (n = 3 mice, ****P < 0.0001). j Spleens and splenocyte counts from uninfected (n = 3 mice), Vehicle- (n = 4 mice), and Glutoborin- (n = 4 mice) treated mice (**P = 0.0047, ***P (uninfected vs. Pb-infected Vehicle) = 0.0003, ***P (Pb-infected, Vehicle vs. Glutoborin) = 0.0005). k UMAP of macrophage subsets from Vehicle- (1065) and Glutoborin- (851) treated mice. Right, proportion of each macrophage subset. l Proportion of B cell subsets from scRNA-seq (Vehicle, 8297, Glutoborin, 8031). m Representative flow cytometry plots of splenic Ly6c⁺Ccr2⁺ macrophages. n frequencies (*P = 0.0122, **P = 0.0046, ****P < 0.0001) and absolute numbers (**P = 0.0025, ***P = 0.0004, ****P < 0.0001) (n = 3 mice). o Feature plots showing exocytosis and IFNα response scores in macrophages. p qRT-PCR analysis of Isg15 and Ifit2 in sorted splenic Ly6c⁺Ccr2⁺ macrophages (n = 3 mice, **P = 0.0019, ***P = 0.0001, ****P < 0.0001). Data are representative of two independent experiments (mean ± SEM). Significance was determined using a two-tailed unpaired Student’s t test and Log-rank (Mantel-Cox) test. ns, P > 0.05. Source data are provided as a Source Data file.

To determine whether these protective effects were mediated through the inactivation of Ly6c⁺Glut1^hi macrophages, we performed scRNA-seq on splenic immune cells from Pb-infected mice treated with either Vehicle or Glutoborin. Unsupervised clustering analysis identified 11 distinct immune cell subsets, among which the macrophage compartment exhibited the most pronounced reduction in relative abundance following Glutoborin treatment (Supplementary Fig. 7c, d). Notably, this reduction was most prominent within the Ly6c⁺Glut1^hi macrophage subset (Fig. 7k). In parallel, Glutoborin treatment induced a marked remodeling of the B cell compartment, characterized by a decrease in activated B cells accompanied by a relative increase in follicular B cells (Fig. 7l). All of these observations were independently validated by flow cytometry (Fig. 7m, n and Supplementary Fig. 7e, f).

In addition to the reduced frequency of Ly6c⁺Glut1^hi macrophages, we observed a marked suppression of gene programs associated with vesicle-mediated exocytosis and IFN-I signaling in Glutoborin-treated mice (Fig. 7o). Consistently, CellChat analysis revealed diminished interaction between Ly6c⁺Glut1^hi macrophages and activated B cells following Glutoborin treatment (Supplementary Fig. 7g). To independently validate these findings, we flow-sorted Ly6c⁺Ccr2⁺ macrophages and subjected to qRT-PCR analysis, which confirmed that Glutoborin treatment significantly reduced the expression of Isg15 and Ifit2 in this subset (Fig. 7p).

Given the documented hepatotoxicity and cardiotoxicity of bortezomib^30,31, we further evaluated the safety of Glutoborin in vivo. Mice treated with Glutoborin exhibited significantly reduced hepatic and cardiac toxicity compared to bortezomib-treated controls, along with more rapid weight recovery (Supplementary Fig. 7h, i). Pharmacokinetic analysis in Sprague-Dawley rats demonstrated favorable bioavailability, broad tissue distribution, and moderate clearance, supporting a suitable therapeutic profile (Supplementary Fig. 7j, k). Collectively, these results indicate that Glutoborin protects against CM by impairing the pro-inflammatory function of Ly6c⁺Glut1^hi macrophage, thereby limiting pathogenic B cell activation and systemic inflammation.

Discussion

Protozoan infections are traditionally considered as drivers of acute inflammation, with disease severity largely attributed to parasite burden and excessive immune activation³². However, increasing evidence indicates that malaria is also associated with sustained immune dysregulation, autoantibody production, and immune-mediated tissue injury beyond direct parasitemia^9,33,34. Here, we show that Plasmodium berghei (Pb) infection triggers an autoimmune-like pathogenic cascade, positioning cerebral malaria (CM) as a form of infection-driven immunopathology involving loss of self-tolerance (Supplementary Fig. 8).

Mechanistically, our data suggest that CM is not solely driven by pathogen-derived immunogenicity but also involves aberrant presentation of host-derived antigens. Pb infection sustains CD36 expression on iRBCs, converting this normally tolerogenic molecule into an immunogenic target. Ly6c⁺Glut1^hi macrophages accumulate in the spleen and promote antigen processing, leading to B cell activation and the generation of anti-CD36 autoantibodies. This establishes a feed-forward immune circuit centered on a host antigen.

Anti-CD36 antibodies exhibit a dual role. While contributing to parasite control through complement-dependent cytotoxicity (CDC), they also drive pathology by inducing thrombocytopenia, endothelial injury, and macrophage activation. This combination of protective and pathogenic functions is a hallmark of autoimmune responses. Consistently, depletion of B cells or removal of the spleen markedly alleviated disease severity, highlighting a central role for humoral immunity in CM.

These findings extend current models of CM, which have primarily emphasized T cell-driven inflammation and cytokine-mediated BBB disruption^35–37. Our results support an additional mechanism in which autoantibody-mediated injury and complement activation contribute substantially to disease pathology. Thus, parasite sequestration and T cell responses alone are insufficient to explain the full spectrum of CM.

Importantly, CM pathogenesis emerges from coordinated interactions between humoral and cellular immunity. Models lacking adaptive immune components, such as nude or Rag1^−/− mice, fail to fully recapitulate disease^38–41, underscoring the requirement for immune cooperation. In immunocompetent hosts, Pb infection induces expansion of CD8⁺ effector T cells, Th1 cells, and Tfh cells, alongside robust B cell activation. Functional experiments further show that antibodies and CD8⁺ T cells act synergistically to exacerbate disease.

Beyond mechanism, our study suggests new therapeutic opportunities. Rather than targeting parasites alone, modulating immune circuits that link metabolism, antigen presentation, and autoantibody amplification may provide additional benefit. Glutoborin exemplifies this concept by targeting Ly6c⁺Glut1^hi macrophages, thereby limiting antigen presentation and dampening downstream humoral responses. This highlights metabolic regulation as a key upstream node in immune-driven pathology.

More broadly, the role of the spleen in malaria appears context-dependent. In the Plasmodium berghei model, the spleen acts as a hub for pathogenic immune amplification. However, in other settings, the spleen contributes to parasite clearance and immune protection⁴². Defining the factors that shift splenic immunity from protective to pathogenic states will be an important direction for future investigation.

Several limitations of this study should be considered. First, although anti-CD36 antibodies can be detected in patients with Plasmodium falciparum infection, our mechanistic conclusions are primarily based on the Plasmodium berghei model. Whether these antibodies play a causal role in human cerebral malaria, and how they contribute to disease heterogeneity, remains to be determined.

Second, while we identify CD36 as a dominant host-derived antigen and define a macrophage-B cell axis that drives autoantibody production, the upstream mechanisms leading to the breakdown of tolerance are not fully resolved. In particular, how infection-associated changes in antigen presentation, post-translational modification, or tissue damage confer immunogenicity to CD36 warrants further investigation.

Third, although our data support a dual role for anti-CD36 antibodies in parasite control and disease progression, the relative contributions of distinct antibody subclasses and effector mechanisms, including complement activation and Fc receptor signaling, remain to be clarified.

Finally, while targeting Ly6c⁺Glut1^hi macrophages attenuates disease in vivo, the broader impact of such interventions on host defense and long-term immunity is not yet defined, which will be important for therapeutic translation.

Methods

Animals

All animal experiments were approved by the Ethics Committee of the Peking University Health Science Center (LA2022546) and conducted in accordance with institutional and national ethical guidelines. Male wild-type (WT) C57BL/6 J mice, aged 6–8 weeks, and female guinea pigs weighing 250 g to 300 g were purchased from Beijing Vital River Laboratory Animal Technology Co., Ltd. Sting knockout (Sting^−/−) mice were generated by CRISPR/Cas9-mediated genome editing using two single-guide RNAs targeting mouse Sting1 exon 3 (CAGTAGTCCAAGTTCGTGCG) and exon 5 (AGTATGACCAGGCCAGCCCG). The Sting^−/− line used in this study carries a 4-bp deletion (ΔCACG) that introduces a frameshift and is predicted to produce a truncated STING protein with a residual length of 58 amino acids. All animals were housed in a specific pathogen-free environment and maintained on 12 h light/dark cycle.

Cell lines

HEK293T, RAW264.7 and 4T1 cell lines were obtained from American Type Culture Collection (ATCC). H5V cell line was obtained from the Cell Bank of the Shanghai Institute of Biological Sciences, Chinese Academy of Sciences (SIBCB Cell Bank). All cells were maintained in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% (v/v) heat-inactivated fetal bovine serum (FBS) and maintained in an incubator at 37 °C with 5% (v/v) CO₂.

Human samples

Human serum samples were collected from patients with Plasmodium falciparum infection at Beijing Ditan Hospital, Capital Medical University, and from healthy donors as controls. Diagnosis of Plasmodium falciparum infection was confirmed by routine clinical testing. Blood samples were obtained after written informed consent, and all procedures were approved by the Institutional Review Board/Ethics Committee of Beijing Ditan Hospital (NO. DTEC-KY2025-145-01), in accordance with the Declaration of Helsinki. Serum was separated by centrifugation, aliquoted, and stored at -80 °C until use.

Antibodies

The following antibodies were used in this study. The catalog number, clone number, and dilution factors were listed in order: Anti-CD45 (25-0451-81, 30-F11, 1:250), Anti-CD4 (35-0041-U500, GK1.5, 1:250), Anti-CD8α (35-0081-U500, 53-6.7, 1:250), Anti-B220 (50-0452-U100, RA3-6B2, 1:250), Anti-CD11b (35-0112-U100, M1/70, 1:250), Anti-F4/80 (20-4801-U100, BM8.1, 1:250), Anti-CD69 (65-0691-U025, H1.2F3, 1:250) (all from Tonbo); Anti-Ccr2 (150627, SA203G11, 1:250), Anti-Ly6c (128011, HK1.4, 1:250), Anti-CD19 (115506, 6D5, 1:250), Anti-CD21 (123407, 7E9, 1:250), Anti-CD23 (101620, B3B4, 1:250), Anti-CD36 (102611, HM36, 1:250), Anti-TER-119 (116222, TER-119, 1:250), Anti-His tag (362603, J095G46, 1:100), Anti-CD20 (152115, SA271G2) (all from Biolegend); Anti-mouse CD4-In Vivo (A2101, GK1.5), Anti-mouse CD8α-In Vivo (A2102, 2.43) (all from Selleck); Anti-CD16/CD32 (16-0161-82, 93, 1:250), Anti-mouse GPⅠbα-In Vivo (R300, Emfret Analytics), Anti-Guinea Pig IgG H&L (bs-0358D-FITC, 1:100), Anti-Guinea Pig IgG H&L (HRP) (A21070, 1:10000), Anti-GLUT1 (21829-1-AP, 1:4000), Anti-IκB α (AF1282, 1:2000), Anti-p-IκB α (2859 T, 1:1000), Anti-ubiquitin (AB120, 1:1000), Anti-STING (NBP2-24683SS, 1:1000); Anti-GFP (KM8009, 9F6, 1:5000), Anti-GAPDH (KM9002T, 1C4, 1:5000), Anti-α tubulin (RM2007, MG17, 1:5000) (all from Ray Antibody).

Plasmodium infection and drug treatment

The GFP-expressing Plasmodium berghei ANKA (Pb) strain, Plasmodium yoelii 17XL and Plasmodium yoelii 17XNL (Py) were generously provided by Prof. Jingwen Wang (School of Life Sciences, Fudan University, Shanghai, China). An experimental CM model was established as previously described⁴³. Briefly, 1 × 10⁴ Pb-infected red blood cells (iRBCs) or Py-infected red blood cells were injected intraperitoneally. Infection was monitored via Giemsa staining or flow cytometry based on GFP fluorescence. For compound treatment, mice received tail vein injection of Bortezomib (0.5 mg/kg, Selleck, S1013), Glutoborin (10 mg/kg) and its boronic acid-deficient analog (G-analog, 10 mg/kg) on days 2, 4, and 6 post-infection. To deplete CD4⁺ T cells, CD8⁺ T cells, B cells, and macrophages, mice were intravenously injected via the tail vein with anti-CD4 antibody (100 μg), anti-CD8 antibody (100 μg), anti-CD20 antibody (150 μg), and clodronate liposomes (200 μL, APExBIO, K2723) were administered by intraperitoneal injection, two days prior to infection. To deplete platelets, mice were intravenously injected via the tail vein with anti-GPⅠbα antibody (2 μg/g) on day 7 post-infection. The CD36 protein treatment group received intravenous injections on days 5 and 6 post-infection.

Giemsa staining

Peripheral blood smears were prepared from Pb-infected mice and stained with Giemsa (Solarbio, G1010). Stained slides were rinsed, air-dried, and mounted for microscopic examination to determine the percentage of iRBCs.

Complete blood count (CBC) and biochemical analysis

Blood samples for CBC were collected via the medial canthus into anticoagulant tubes and analyzed using a hematology analyzer (BH-5190VET, Guilin Urit Medical Electronic Co., Ltd). Blood for biochemical analysis was obtained by enucleation, allowed to clot, centrifuged at 1800 × g for 20 min at 4 °C, and analyzed using a biochemical analyzer (BS-200, Shenzhen Mindray Bio-Medical Electronics Co., Ltd).

Blood-brain barrier (BBB) permeability assay

Mice with neurological symptoms received tail vein injection of 2% EB (Sigma-Aldrich, E2129). After 1 h, mice were perfused with saline, and brain tissues were collected, incubated in formamide (4 °C, 72 h), centrifuged (1800 × g, 15 min), and analyzed at OD620 nm.

Splenectomy

Mice were anesthetized with tribromoethanol, and a small incision was made on the left dorsal side to expose the spleen. After ligating blood vessels, the spleen was removed, and the incision was sutured. Mice were monitored for one week before further experiments.

RNA extraction and quantitative real-time PCR (qRT-PCR)

Brain tissues were collected post-neurological symptoms onset. RNA was extracted using TRIzol reagents (Vazyme, R401-01) and reverse-transcribed according to the manufacturer’s instructions for the reverse transcription kit (Vazyme, R333-01). qRT-PCR was performed on the Biosystems 7500 Fast & 7500 Real-Time PCR System using TransStart Top Green qPCR SuperMix (TransGen Biotech), and data analysis was conducted with 7500 Software v2.3 (all primers are listed in Supplementary Table 2).

H&E staining

Mice were transcardially perfused, and brain tissue specimens were fixed in 10% (vol/vol) neutral buffered formalin, followed by paraffin embedding, sectioning, and H&E staining.

Platelet isolation and protein extraction

Platelet-rich plasma was obtained by serial centrifugation of anticoagulated mouse blood. Platelet pellets were lysed in RIPA buffer with PMSF and a protease inhibitor cocktail, incubated on ice (30 min), and centrifuged (13,000 × g, 15 min, 4 °C). The supernatant was collected for protein quantification.

Pulldown assay and mass spectrometry

Serum autoantibodies were analyzed by incubating 100-fold diluted serum with protein L beads, followed by PBS washes. Anti-platelet antibodies were identified by co-incubating platelet protein (300 µg) with serum from both Pb-infected and uninfected mice in protein lysis buffer (1:100). Subsequently, protein L beads were added for co-incubation. The precipitates were washed three times with PBS containing 0.1% (v/v) NP40. For the serum immunoglobulins, serum was collected from uninfected and Pb-infected mice at day 7 post-infection. Immunoglobulins were enriched from serum using protein L beads. The samples described above were subjected to SDS-PAGE and silver staining. Excised gel bands were subjected to in-gel tryptic digestion. The resulting peptides were dissolved in 0.1% formic acid and analyzed by LC-MS/MS using an LTQ Orbitrap Elite mass spectrometer (Thermo Fisher Scientific) equipped with a nanoelectrospray ion source (Proxeon Biosystems). Peptides were separated on a reversed-phase C18 column with a 5–32% acetonitrile gradient in 0.1% formic acid. MS/MS spectra were acquired by collision-induced dissociation, and raw data were searched against the UniProt database using the SEQUEST search engine. Protein identification or quantification values for each sample were normalized and assembled into a data matrix, with proteins as rows and experimental conditions as columns. The heatmap was generated using the R package pheatmap, where color intensity represents the relative abundance of each protein across conditions (greater intensity indicates higher relative abundance).

ELISA

Antigen-coated (10 µg/mL) 96-well plates were incubated with serially diluted serum, followed by anti-IgG (KPL, 5220-0341) or IgM (Abcam, AB97230) detection. OD450 nm was measured. For total IgM detection, a Mouse IgM ELISA Kit (NeoBioscience, EMC129.96) was used.

Lymphocyte preparation and flow cytometry

Spleen and lymph node were homogenized in PBS buffer containing 1% FBS. Red blood cells were then lysed using ACK lysis buffer, and a single-cell suspension were filtered through a 75 μm filter. Blood lymphocytes were isolated via RBC lysis and filtration. Flow cytometry was performed using fluorescently conjugated monoclonal antibodies. For fluorescence-activated cell sorting (FACS), stained cells were gated to exclude debris and sorted on a cell sorter into collection tubes containing PBS with 1% FBS for downstream assays.

GLUT1 knockdown in 4T1 cells

The full-length murine GLUT1 gene was targeted using shRNA sequences (1: GTCCTATTCCATGGTTCATTG; 2: TGAGGAGTTCTACAATCAAAC) cloned into the PLKO.1-puro backbone (Addgene). Lentiviral particles were produced by co-transfecting HEK293T cells with PLKO.1-GLUT1 shRNA, psPAX2, and pMD2.G plasmids at a 4:3:1 molar ratio using Lipofectamine 3000 (Thermo Fisher) according to the manufacturer’s instructions. Viral supernatants were collected 48 h post-transfection, filtered through a 0.45 μm syringe filter, and applied to 4T1 cells in the presence of 8 μg/mL polybrene (Sigma-Aldrich) to enhance transduction efficiency. After 24 h, the medium was replaced with fresh culture medium. Cells were selected with 4 μg/mL puromycin (Sigma-Aldrich) for 5–7 days until stable knockdown lines were established. GLUT1 knockdown efficiency was validated by Western blot. Control 4T1 cells were generated using a non-targeting shRNA in the same PLKO.1 backbone and processed in parallel.

Proteasome activity assay

To assess the effect of Glutoborin on cellular proteasome activity, HEK293T and 4T1 cells were treated with 100 nM Glutoborin or 100 nM bortezomib for 3 h. The cells were then collected and lysed using a lysis buffer containing 50 mM HEPES (pH 7.5), 5 mM EDTA, 150 mM NaCl, and 1% (v/v) Triton X-100. The supernatant was collected by centrifugation, and protein concentration was determined. For proteasome activity measurement, 20 µg of protein was added to a reaction buffer containing 25 mM HEPES, 0.5 mM EDTA, 0.05% (v/v) NP40, and 0.001% (v/v) SDS, and incubated with 50 μM Suc-LLVY-AMC or Z-LLE-AMC at 37 °C. Fluorescence intensity was measured at different time points. To assess the specificity of Glutoborin, GLUT1-knockdown 4T1 cells were treated with Glutoborin for 30 min. Cells were then collected to assess proteasome activity as described above.

Protein expression and purification

HEK293T cells expressing GLUT1-GFP protein were collected (4,000 g, 10 min) and resuspended with 1 × TBS buffer (150 mM NaCl, 20 mM HEPES, pH 7.4). Cells were lysed by sonication in the presence of PMSF (1:100) and protease inhibitor cocktail (MCE), and lysates were clarified (8000 × g, 20 min) to remove nuclei and organelles. Membranes were pelleted by ultracentrifugation (100,000 × g, 1 h) and solubilized in buffer containing 150 mM NaCl, 20 mM HEPES (pH 7.4) and 1% (w/v) n-dodecyl β-D-maltoside (DDM, Anatrace) at 4 °C for 1 h with gentle agitation. After clarification (100,000 × g, 30 min), the supernatant was incubated with Ni-NTA beads. The beads were washed with purification buffer (150 mM NaCl, 20 mM HEPES, pH 7.4 and 0.02% (w/v) DDM) supplemented with 20 mM or 40 mM imidazole, and the protein was eluted with 250 mM imidazole. Eluted protein was concentrated and further purified by size-exclusion chromatography (Superose 6 Increase 10/300 GL; GE Healthcare, USA).

The extracellular domain of CD36 was cloned into the pET28a (+) plasmid and transformed into Escherichia coli BL21. Protein expression was induced with 0.1 mM isopropyl-β-D-thiogalactopyranoside (IPTG) at 20 °C for 18 h. Bacterial cultures were harvested by centrifugation, and pellets were resuspended in lysis buffer (NTA0, protease inhibitor cocktail, 1 mM EDTA, 1 mM MgCl₂). Cells were lysed by sonication, and the lysate was centrifuged to remove debris. Soluble protein was purified using Ni-affinity chromatography with sequential elution in buffers of increasing imidazole concentration (NTA0, NTA20, NTA40, NTA80, NTA300, and NTA1000). Protein purity was assessed by Coomassie Brilliant Blue staining, and purified CD36 was concentrated using ultrafiltration. Protein concentration was determined before use in subsequent experiments.

Immunoglobulin purification

For purification of IgM from Pb-infected mice, serum was collected on day 7 post-infection, and IgM was purified by ion-exchange chromatography (Q Sepharose HP chromatography). Anti-CD36 IgG was generated in guinea pigs by subcutaneous immunization. Purified CD36 protein (300 μg) was emulsified 1:1 (v/v) with complete Freund’s adjuvant and administered at multiple sites on the dorsal surface. On days 14 and 21, guinea pigs received booster immunizations with purified CD36 protein (300 μg) emulsified with incomplete Freund’s adjuvant. Serum was collected via the peritoneal vein on day 26 post-immunization and purified by affinity chromatography (Protein A). Purification of the different immunoglobulin fractions was performed by Beijing BoaoSen Biotechnology Co., Ltd.

LDH release assay

Cytotoxicity of anti-CD36 antibodies was assessed using an LDH Cytotoxicity Assay Kit (Beyotime, C0017). Cells were treated with purified anti-CD36 IgG or isotype control antibodies (1:100) at 37 °C for 6 h in the presence of complement-active rabbit serum (non-heat-inactivated; NIS-0071, Dingguo Changsheng) or complement-inactivated rabbit serum (heat-inactivated at 56 °C for 30 min), and LDH activity was quantified by measuring absorbance at 490 nm (OD490).

7-AAD staining

Cells were incubated with purified anti-CD36 IgG, isotype control IgG, purified mIgM, or isotype control mIgM (1:100) at 37 °C for 6 h in the presence of complement-active rabbit serum (non-heat-inactivated) or complement-inactivated rabbit serum (heat-inactivated at 56 °C for 30 min). The culture supernatant and cells were then collected, stained with 7-AAD dye (Biolegend, 420404) and analyzed by flow cytometry.

Surface plasmon resonance (SPR) analysis

SPR experiments were performed using a Biacore 8k system (GE Healthcare). Purified GLUT1 protein was immobilized on a CM5 sensor chip (Cytiva, Cat# BR100399) via amine coupling in 10 mM sodium acetate buffer (pH 4.5). For initial compound screening, Glutoborin was tested at a final concentration of 50 µM. For kinetic analysis, Glutoborin was serially diluted in running buffer (PBS with 0.05% Tween-20) over a concentration range of 0.049 µM to 6.25 µM. Association and dissociation times were set at 0 s and 60 s, respectively. Sensorgrams were analyzed using Biacore Insight Evaluation Software, and equilibrium dissociation constants (K_D) were calculated by fitting the data to a 1:1 Langmuir binding model.

In vitro activation of macrophage

RAW264.7 cells were seeded in 6-well plates at a density of 1 × 10⁶ cells per well and cultured in DMEM supplemented with 10% (v/v) FBS and 1% (v/v) penicillin/streptomycin. To assess the effect of anti-CD36 antibodies on macrophage activation, RAW264.7 cells were treated with LPS (50 ng/mL, Sigma-Aldrich, L4391, Escherichia coli O111: B4) after pretreatment with oleic acid (OA, 200 μM, Sigma-Aldrich, O3008), purified anti-CD36 IgG (1:100) or the combination. RAW264.7 cells were harvested for downstream analysis. To evaluate the dynamics of proteasome activity during macrophage activation, RAW264.7 cells were stimulated with LPS (100 ng/mL) and murine interferon-γ (mIFNγ, 50 ng/mL, PeproTech, Inc., 315-05) for different time points. RAW264.7 cells were collected to assess proteasome activity as described above. To evaluate the effect of Bortezomib, Glutoborin and G-analog on macrophage activation, RAW264.7 cells were stimulated with LPS (100 ng/mL) and mIFNγ (50 ng/mL) for 12 h. After stimulation, 20 nM Bortezomib, Glutoborin or G-analog was added and incubated for an additional 12 h. RAW264.7 cells were then harvested for downstream analysis.

Passive and adoptive transfer models

The blood of mice infected with Pb at different time points was collected, and serum was isolated for subsequent transfer or purification. The serum (250 μL) or purified mIgM (high-dose, 200 μL; low-dose, 50 μL) was passively transferred into the corresponding experimental groups of mice via tail vein injection, including splenectomized mice infected with Pb or normal untreated mice, followed by downstream analysis. For the passive transfer experiment with anti-CD36 antibodies, purified anti-CD36 IgG was passively transferred into splenectomized mice infected with Pb via tail vein injection (50 μL) for subsequent experiments. For adoptive transfer of CD8⁺CD44⁺ T cells, 5 × 10⁶ cells were transferred by tail vein injection into splenectomized Pb-infected mice.

Detection of CD36-reactive B cells

Splenocytes (1 × 10⁶) from Pb-infected or uninfected mice were resuspended in staining buffer and incubated with anti-CD16/32 Fc-blocking antibody (1:100) for 10 min. Cells were washed and incubated with recombinant CD36 protein (1 μg/mL) or BSA on ice for 30 min. After washing, cells were stained on ice for 30 min with either APC-conjugated anti-CD36 antibody or PE-conjugated anti-His-tag antibody, washed again, and resuspended for acquisition. CD36-binding B cells were quantified by sequential gating on singlets, live CD45⁺ cells and CD19⁺ B cells. For assay validation, B cells from draining lymph nodes of CD36-immunized animals were included as a positive control.

Isolation of iRBCs

Blood from infected mice was collected and diluted with an equal volume of PBS, which was then carefully layered onto lymphocyte separation medium. The mixture was centrifuged at 1200 × g for 15 min (gradual acceleration to 1 and deceleration to 1). The upper layer, containing the iRBCs, was collected, washed with PBS, and resuspended. Subsequently, the cell suspension was layered onto 70% Percoll solution and centrifuged at 1200 × g for 15 min (gradual acceleration to 1 and deceleration to 1). The top layer, enriched in iRBCs, was collected, washed with PBS, and used for subsequent experiments.

Complement-dependent cytotoxicity (CDC) assay of iRBCs

1 × 10⁶ iRBCs were incubated with isotype control antibody, purified anti-CD36 IgG (1:100), or purified anti-CD36 IgG plus purified CD36 protein (25 μg) at 37 °C for 30 min in the presence of complement-active rabbit serum (non-heat-inactivated) or complement-inactivated rabbit serum (heat-inactivated at 56 °C for 30 min). The proportion of iRBCs was then analyzed by flow cytometry.

Antibody-mediated CD36 internalization assay

RAW264.7 macrophages were collected and resuspended at 1 × 10⁶ cells per 100 μL in ice-cold FACS buffer (PBS containing 1% FBS), washed twice (300 × g, 4 min, 4 °C), and incubated with anti-CD16/32 Fc-blocking antibody (1:100) for 10 min on ice. Cells were stained with purified anti-CD36 IgG (1:100) at 4 °C for 15 min, washed, and then incubated with fluorophore-conjugated secondary antibody at 4 °C for 15 min. After washing, samples were divided into a 4 °C control group and a 37 °C group incubated for 30 min to allow receptor internalization, then stopped by returning samples to ice, followed by acid stripping to remove surface-bound antibodies using 200 μL ice-cold acid wash buffer (0.5 M glacial acetic acid, 150 mM NaCl, pH 2.5) for 90 s, neutralization with FACS buffer for 2 min, and a second acid wash. Cells were washed twice, fixed, and analyzed by flow cytometry.

Immunoblot analysis

GLUT1-knockdown 4T1 cells were harvested and lysed in Co-IP lysis buffer (150 mM NaCl, 0.1 mM EDTA, 10% (v/v) glycerol, 0.5% (v/v) NP40, and protease inhibitor Cocktail). Cell lysates were subjected to SDS-PAGE followed by immunoblotting. RAW264.7 cells were pretreated with either Vehicle or Glutoborin for 2 h, followed by stimulation with LPS (100 ng/mL) and mIFN-γ (50 ng/mL) for 30 min. RAW264.7 cells were then harvested and lysed in Co-IP lysis buffer (150 mM NaCl, 0.1 mM EDTA, 10% (v/v) glycerol, 0.5% (v/v) NP40, and protease inhibitor Cocktail). Cell lysates were subjected to SDS-PAGE followed by immunoblotting. For immunoprecipitation assays, lysates were incubated with the appropriate antibodies, followed by the addition of protein A/G beads for co-incubation. After three washes with PBS containing 0.1% (v/v) NP40, immunoprecipitated proteins were analyzed by immunoblotting.

In-house compound library

The boronic acid-based proteasome inhibitor library was developed by Dr. Runtao Li’s research group, which has optimized these inhibitors over multiple iterations (CN106146542A, CN108368133A, CN114075227A).

Pharmacokinetic study

The pharmacokinetic study of Glutoborin in Sprague-Dawley (SD) rats was conducted by Beijing Yingkerui Drug Safety and Efficacy Research Co., Ltd.

RNA-seq data analysis

Total RNA was extracted from RAW264.7 cells using TRIzol reagent and used for RNA-seq library construction by Berry Genomics. Gene expression was quantified as Fragments Per Kilobase Million (FPKM), normalized with gene length and read counts. Differential gene expression analysis was performed using DESeq2 (v1.46.0, RRID: SCR_015687, Wald test). Data visualization was conducted using the pheatmap (v1.0.12, RRID: SCR_016418) and ggplot2 (v3.5.0, RRID: SCR_014601) R packages.

scRNA-seq

For scRNA-seq, CD45⁺ lymphocytes were isolated from the spleens of Pb-infected mice, Pb-infected mice treated with Glutoborin and uninfected mice using flow cytometry. Single-cell libraries were prepared and sequenced using the 10x Genomics platform by Berry Genomics.

Raw sequencing data were processed with Cell Ranger (10x Genomics, V6.0.1, RRID: SCR_017344) to generate a raw unique molecular identifier (UMI) count matrix. Data were converted into a Seurat object using the Seurat R package (V4.3.0, RRID: SCR_016341). Low-quality cells were filtered out based on UMI counts (> 6000 or < 500) and mitochondrial gene content (> 10%).

Data normalization and scaling were performed using the NormalizeData () and ScaleData () function. Merge () function was used to integrate the three samples, and the top 3,000 highly variable genes (HVGs) of the merged dataset were identified by the FindVariableFeatures (). Batch effects were corrected using the Harmony R package (V1.2.0, RRID: SCR_022206). Finally, we obtained the scaled and batch effect-corrected expression profiles of all samples for downstream analysis.

Principal component analysis (PCA) was conducted to reduce dimensions of the data, with the top thirty principal components were selected for further analysis. A total of twenty-nine clusters were identified using FindClusters () (resolution = 1.3). Marker genes for each cluster were determined using FindAllMarkers (). Cells were annotated at two stages of the analysis. First, eleven major types of immune cell were identified; second, four subpopulations of macrophage and six subpopulations of B cell were characterized. Visualization of the clusters was a two-dimensional UMAP plots. The corresponding cell proportion was calculated with the dplyr R package (V1.1.4, RRID: SCR_016708), and the barplot was generated with the ggplot2 R package (V3.5.0, RRID: SCR_014601).

Pathway activity scores were calculated using the AUCell R package (V1.20.2, RRID: SCR_021327). Cell-cell interactions were analyzed with the CellChat R package (V1.6.1, RRID: SCR_021946) based on ligand-receptor pairs from CellChatDB (https://github.com/Teichlab/cellphonedb).

GO enrichment analysis

GO enrichment analysis was performed using DAVID Bioinformatics (https://davidbioinformatics.nih.gov/tools.jsp) to identify pathway activities in distinct cell populations. Results were visualized using the ggplot2 R package (V3.5.0, RRID: SCR_014601). GO analysis was applied to RNA-seq and proteomics datasets, respectively.

Molecular docking

The X-ray crystal structure of human GLUT1 (PDB ID: 5EQH) was obtained from the Protein Data Bank. The protein structure was prepared using the Protein Preparation Wizard in Schrödinger Maestro (Schrödinger, LLC, New York, NY). Briefly, water molecules not involved in key binding interactions and other irrelevant heteroatoms were removed. Missing side chains were built, hydrogen atoms were added, and the protonation states of ionizable residues were predicted using Epik at pH 7.0 ± 2.0. The structure was then optimized and subjected to a restrained energy minimization under the OPLS4 force field to relieve any steric clashes or unusual geometries.

Candidate inhibitors (e.g., Glutoborin, BAY-876) were prepared using the LigPrep module in Maestro. For each compound, possible tautomers and protonation states were generated at pH 7.0 ± 2.0 using Epik. Low-energy conformations were retained, and partial charges and force field parameters were assigned according to OPLS4.

Molecular docking was performed with the Glide program (Schrödinger, LLC). The binding site was defined around the central cavity of GLUT1 based on the location of the original co-crystallized ligands or key residues. A grid box with a margin of approximately 10–15 Å from the key binding residues was generated. Docking was carried out using either the standard precision (SP) or extra precision (XP) mode. Default settings were employed unless otherwise stated. The resulting poses were ranked by GlideScore, and the top-scoring poses were visually inspected to ensure chemically reasonable interactions with the protein active site.

Statistical analysis and reproducibility

Data were analyzed using Prism GraphPad software v6.0, and results are presented as mean ± standard error of the mean (SEM). Statistical significance between groups was determined using two-tailed unpaired Student’s t test, paired Student’s t test, Log-rank (Mantel-Cox) test and Wilcoxon test. A P < 0.05 was considered statistically significant. All experiments were independently repeated at least twice, yielding consistent results.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

X.S., Ridong Li, and D.L. conceived and designed the study. X.S., W.W., X. Zhang, and D.L. collected and analyzed the data. X.S., W.W., L.H., X. Zhao, and D.L. performed the experiments and validation analyses. X.S., W.W., X. Zhang, L.H., X. Zhao, and D.L. prepared the figures and visualized the data. Ridong Li and D.L. developed the methodology. D.Y., W.T., X.X., Runtao Li, Y.L., J.H., R.S., and F.Y. provided materials and other resources. D.L. secured funding, managed the project, and supervised the research. X.S. and D.L. wrote the manuscript. Ridong Li and D.L. revised and edited the manuscript. All authors read and approved the final version of the manuscript.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewer for their contribution to the peer review of this work. A peer review file is available.

Funding

D.L. discloses support for the research of this work from the National Natural Science Foundation of China [grant numbers 32571053, U24A20372 and 82541009], the Beijing Natural Science Foundation [grant number F252067] and the Noncommunicable Chronic Diseases-National Science and Technology Major Project [grant number 2024ZD0520600].

Data availability

The RNA sequencing data are openly available in the Gene Expression Omnibus (GEO) database at GSE318498. The single-cell RNA sequencing data generated in this study are available in the GEO database at GSE290172 and GSE318649. The mass spectrometry data tables generated in this study, as well as the synthesis methods of Glutoborin and the G-analog together with the corresponding mass spectrometry characterization data, are provided in the Supplementary Information file. The data underlying Figures and Supplementary Figures are provided as a Source Data file. Source data are provided in this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-74882-4.