Lupus autoantibodies initiate neuroinflammation sustained by continuous HMGB1:RAGE signaling and reversed by increased LAIR-1 expression

Abstract

Cognitive impairment is a frequent manifestation of neuropsychiatric systemic lupus erythematosus, present in up to 80% of patients and leading to a diminished quality of life. In the present study, we used a model of lupus-like cognitive impairment that is initiated when antibodies that crossreact with excitatory neuronal receptors penetrate the hippocampus, causing immediate, self-limited, excitotoxic death of hippocampal neurons, which is then followed by a significant loss of dendritic complexity in surviving neurons. This injury creates a maladaptive equilibrium that is sustained in mice for at least 1 year. We identified a feedforward loop of microglial activation and microglia-dependent synapse elimination dependent on neuronal secretion of high mobility group box 1 protein (HMGB1) which binds the receptor for advanced glycation end products (RAGE) and leads to microglial secretion of C1q, upregulation of interleukin-10 with consequent downregulation of leukocyte-associated immunoglobulin-like receptor 1 (LAIR-1), an inhibitory receptor for C1q. Treatment with a centrally acting angiotensin-converting enzyme inhibitor or with an angiotensin-receptor blocker restored a healthy equilibrium, microglial quiescence and intact spatial memory.

Extended Data Fig. 1 |. Pathology is sustained for at least 12 months.

a) Decreased dendritic complexity in 12 month old (m.o.) Balb/c DNRAb⁺ compared with DNRAb⁻ mice (mean +/− SEM; n = 4–5 mice per group; n = 55–59 neurons analyzed per group; two-tailed linear mixed model test with Tukey adjustment; p = 0.016). b) Decreased dendritic spine density in 12 m.o. Balb/c DNRAb⁺ compared with DNRAb⁻ mice (median (solid line) with quartiles (dash); n = 4 mice per group; n = 15–18 neurons analyzed per group; two-tailed Mann-Whitney test; p < 0.0001). c) Representative sections of microglia in CA1 stratum radiatum stained for Iba1 (red) and CD68 (white) in 12 m.o. DNRAb⁺ and DNRAb⁻ Balb/c mice (n = 3 mice per group). d) Increased activation score in 12 m.o. Balb/c DNRAb⁺ microglia compared to 12 m.o. Balb/c DNRAb⁻ counterparts based on morphology and CD68 expression (median (solid line) with quartiles (dash); n = 3 mice per group; n = 110–169 microglia scored per group; two-tailed Mann-Whitney test; p < 0.0001).

Extended Data Fig. 2 |. Ex vivo adult microglia and neonatal cultures respond similarly to HMGB1.

Increased mRNA expression for a) Tnf (p = 0.0019); b) Il1b (p = 0.0030); c) C1qa (p = 0.1302); and d) Ifnb1 (p = 0.0453) in B6 microglia stimulated with 1 μg/ml HMGB1 compared with unstimulated microglia cultured in medium for 6 hours (mean +/− SD; ex vivo microglia isolated from 3–4 B6 mice; two-tailed unpaired t-test).

Extended Data Fig. 3 |. Acute neuronal loss is not affected by loss of RAGE or microglial LAIR-1.

a) Decreased CA1 neurons in WT B6 (p = 0.0004) and RAGE KO mDNRAb⁺ (p < 0.0001) mice compared to their mDNRAb⁻ counterparts (median (solid line) with quartiles (dash); n = 3–4 mice per group; n = 72–96 sections per group; two-tailed Kruskal-Wallis test). b) Decreased CA1 neurons in LAIR-1 cKO DNRAb⁺ mice compared to DNRAb⁻ (median (solid line) with quartiles (dash); n = 3 mice per group; n = 67–98 sections per group; two-tailed Mann-Whitney test; p = 0.0394).

Extended Data Fig. 4 |. Single-cell RNA-seq clustering and quality control.

a) UMAP plot colored by clustering. b) UMAP plot split by mouse of origin, colored by cell type. No evidence of strong batch effects in the UMAP space. c) Violin plot of various QC metrics of interest, with similar distributions observed in each mouse. QC metrics include the score returned by Azimuth (Azimuth Score), number of genes per cell (nGene), number of UMI per cell (nUMI), percent UMI coming from mitochondrial reads (Percent Mitochondrial), percent UMI mapping to ribosomal proteins (Percent Ribosomal Protein), and doublet scores (scds). d) Feature plots of genes associated with a known microglia activation signature. Subclustering within microglia subtypes is largely driven by these variables. e) Feature plots of number of genes per cell (nGene) and number of UMIs per cell (nUMI). Subclustering within microglia subtypes is largely driven by these variables.

Extended Data Fig. 5 |. Concordance score of Ms4a7+ microglia with microglial gene signatures.

a) Higher DAM signature gene set score in Ms4a7⁺ compared with Homeostatic microglia (Keren-Shaul et al. (2017) median (solid line) with quartiles (dash); n = 3 mice per group; n = 2515–15285 cells/cluster; two-tailed Mann-Whitney test; p < 0.0001). b) Lower Homeostatic signature gene set score in Homeostatic compared with Ms4a7⁺ cluster (Keren-Shaul et al.; median (solid line) with quartiles (dash); n = 3 mice per group; n = 2515–15285 cells/cluster; two-tailed Mann-Whitney test; p < 0.0001). c) Higher NPSLE signature gene set score in Ms4a7⁺ compared with Homeostatic microglia (Makinde et al.; median (solid line) with quartiles (dash); n = 3 mice per group; n = 2515–15285 cells/cluster; two-tailed Mann-Whitney test; p < 0.0001). d) Higher MGnD signature gene set score in Ms4a7⁺ compared with Homeostatic microglia (Krasemann et al.; median (solid line) with quartiles (dash); n = 3 mice per group; n = 2515–15285 cells/cluster; two-tailed Mann-Whitney test; p < 0.0001).

Fig. 1 |. Microglia are activated by neuronal HMGB1.

a, Representative sections of microglia in CA1 stratum radiatum stained for Iba1 (red) and CD68 (white) in Camk2a-cre⁻Hmgb1^fl/fl B6.H2^d DNRAb⁺ mice (WT; n = 5, top) and Camk2a-cre⁺Hmgb1^fl/fl B6.H2^d DNRAb⁺ mice (HMGB1 cKO; n = 5, bottom). b, Decreased activation score in microglia from Camk2a-cre⁺Hmgb1^fl/fl B6.H2^d (HMGB1 cKO) DNRAb⁺ mice compared with Camk2a-cre⁻Hmgb1^fl/fl B6.H2^d (WT) DNRAb⁺ mice based on morphology and CD68 expression (median (solid line) with quartiles (dash); n = 5 mice per group; n = 7–18 microglia scored per mouse; two-tailed Mann–Whitney U-test: P = 0.0079). c, Relative mRNA expression of Tnf, Il1b and C1qa increased with HMGB1 stimulation in cultured WT (B6) microglia stimulated with 0, 500 or 1,000 ng ml⁻¹ of HMGB1 for 4.5 h in serum-free medium (mean ± s.d.; microglia cultured from four independent litters; two-tailed, one-way, repeated-measure ANOVA on log transformed data: P < 0.0001 (Tnf), P < 0.0001 (Il1b) and P = 0.0238 (C1qa)). d, Secretion of TNF and IL-1β increased with HMGB1 stimulation in cultured WT (B6) microglia stimulated with and without 1 μg ml⁻¹ of HMGB1 for 24 h in serum-free medium (mean ± s.d.; microglia cultured from five independent litters; two-tailed, paired Student’s t-test: P < 0.0001 (TNF) and P = 0.0073 (IL-1β)). e, Relative mRNA expression of Ifnb1 increased with HMGB1 stimulation in cultured WT (B6) microglia stimulated with 0, 100 or 1,000 ng ml⁻¹ of HMGB1 for 4.5 h in serum-free medium (mean ± s.d.; microglia cultured from three independent litters; two-tailed, one-way, repeated-measure ANOVA on log transformed data: P = 0.0141). f, Secretion of IFN-β increased with HMGB1 stimulation in cultured WT (B6) microglia stimulated with and without 1 μg ml⁻¹ of HMGB1 for 24 h in serum-free medium (mean ± s.d.; microglia cultured from five independent litters; two-tailed, paired Student’s t-test: P = 0.0037).

Fig. 2 |. DNRAbs induce microglial activation through RAGE.

a, Loss of RAGE decreased the relative mRNA expression of Tnf, Il1b, C1qa and Ifnb1 with HMGB1 stimulation compared with WT (B6) microglia. WT and RAGE KO microglia were stimulated with 0, 100 or 1,000 ng ml⁻¹ of HMGB1 for 4.5 h in serum-free medium (mean ± s.d.; microglia were cultured from three independent litters; two-tailed, two-way, repeated-measure ANOVA on log transformed data: P < 0.0001 (Tnf), P < 0.0001 (Il1b), P = 0.0069 (C1qa) and P = 0.0004 (Ifnb1)). b, Representative sections of microglia in CA1 stratum radiatum stained for Iba1 (red) and CD68 (white) in WT (B6) and RAGE KO mDNRAb⁻ (left) and mDNRAb⁺(right) mice (n = 5 mice per group). c, Increased activation score in microglia in DNRAb⁺ WT B6 (P = 0.0003) but not RAGE KO mice (P > 0.9999) compared with DNRAb⁻ counterparts based on morphology and CD68 expression (median (solid line) with quartiles (dash); n = 5 mice per group; n = 10–24 microglia scored per mouse; two-tailed Kruskal–Wallis test). d, Representative tracings of CA1 pyramidal neurons from mDNRAb⁻ and mDNRAb⁺ in WT B6 and RAGE KO mice. e, Analysis of dendritic complexity showed a decrease in WT B6 mDNRAb⁺ compared with mDNRAb⁻ groups (mean ± s.e.m.; n = 7 mice per group; n = 25 neurons analyzed per group; two-tailed, linear mixed model test with Tukey’s adjustment: P = 0.002). f, Analysis of dendritic complexity showed no difference in RAGE KO mDNRAb⁺ compared with mDNRAb⁻ groups (mean ± s.e.m.; n = 6–7 mice per group; n = 24 neurons analyzed per group; two-sided, linear mixed model test with Tukey’s adjustment: P = 0.619).

Fig. 3 |. Treatment with ACE inhibitors requires LAIR-1 for efficacy.

a, Decreased Lair1 mRNA expression in microglia isolated from DNRAb⁺ compared with DNRAb⁻ B6.H2^d mice (mean ± s.d.; n = 8 mice per group, pooled from two independent experiments; two-tailed, unpaired Student’s t-test: P = 0.0294). b, Lair1 mRNA expression in microglia isolated from DNRAb⁺ B6.H2^d mice increased in mice treated with captopril (5 mg kg⁻¹ i.p.) compared with those treated with enalapril (5 mg kg⁻¹ i.p.) daily for 2 weeks (mean ± s.d.; n = 4 mice per group; two-tailed, unpaired Student’s t-test: P = 0.0005). c, Representative sections of microglia in CA1 stratum radiatum stained for Iba1 (red) and CD68 (white) in DNRAb⁺ WT (B6.H2^d) and LAIR-1 cKO DNRAb⁻ mice treated with saline (left) or captopril (5 mg kg⁻¹, right; n = 3–5 mice per group). d, Decreased activation score in DNRAb⁺ microglia in WT B6.H2^d (P = 0.0383) but not LAIR-1 cKO (P = 0.5699) mice treated with captopril compared with saline-treated counterparts based on morphology and CD68 expression (median (solid line) with quartiles (dash); n = 3–5 mice per group; n = 12–78 microglia scored per mouse; two-tailed Kruskal–Wallis test). e, Representative tracings of CA1 pyramidal neurons from DNRAb⁻ and DNRAb⁺ LAIR-1 cKO mice treated with saline or captopril. f, Captopril treatment had no effect on dendritic complexity in LAIR-1 cKO DNRAb⁻ (P > 0.99) and DNRAb⁺ (P = 0.88) mice treated with either saline or captopril (5 mg kg⁻¹; mean ± s.e.m.; n = 2–3 mice per group; n = 19–27 neurons analyzed per group; two-tailed, linear mixed model test with Tukey’s adjustment).

Fig. 4 |. Hippocampal microglial scRNA-seq shows that ACE inhibitors mitigate DNRAb phenotype.

a, UMAP plot of single-cell microglia data, with each cell colored by microglia cluster: cycling (pink), homeostatic (blue), IFN responsive (orange), Ms4a7⁺ (green), S100a4⁺ (yellow) and Tmem119⁻ (light green). b, Cell-type composition plots. The percentage of total cells is represented in each cluster in DNRAb⁻, DNRAb⁺ and captopril-treated DNRAb⁺ B6.H2^d mice (mean ± s.e.m.; n = 3 mice per group; two-tailed ANOVA, false discovery rate (FDR)-adjusted P values shown: P = 0.01477 (homeostatic), P = 0.026844 (Ms4a7⁺), P = 0.488584 (cycling), P = 0.488584 (IFN responsive), P = 0.01477 (S100a7⁺) and P = 0.034437 (Tmem119⁻)). c, Stacked bar plots of the cell-type composition in each condition. d, Volcano plots of the differentially expressed genes comparing between different conditions (columns) in Ms4a7⁺ and homeostatic microglia (rows). The y axis is log₁₀(P value), the x axis is the log(fold-change) (log(FC)) reported by EdgeR. Genes with FDR < 0.05 are colored red and genes of interest are labeled. e, Expression of genes of interest in different conditions in Ms4a7⁺ microglia. Pseudobulk expression levels are calculated on a per-mouse basis (y axis, log₂(transcripts per million) (log₂(TPM))), stratified by condition (mean ± s.d.; two-tailed FDR; P values are uncorrected).

Fig. 5 |. ARB replicates effects of ACE inhibitors on neurons and microglia.

a, Increased Agtr1a mRNA expression in microglia stimulated with 1 μg ml⁻¹ of HMGB1 compared with microglia cultured in medium for 6 h (mean ± s.d.; ex vivo microglia isolated from three to four B6 mice; one-tailed, unpaired Student’s t-test: P = 0.0380). b, Representative sections of microglia in CA1 stratum radiatum stained for Iba1 (red) and CD68 (white) in DNRAb⁺ (top) and DNRAb⁻ (bottom) Balb/c mice treated with saline (left) or telmisartan (right, 1 mg kg⁻¹; n = 2–3 mice per group). c, Decreased activation score in DNRAb⁺ microglia in Balb/c mice treated with telmisartan compared with saline-treated counterparts (P = 0.0046) based on morphology and CD68 expression (median (solid line) with quartiles (dash); n = 2–3 mice per group; n = 10–15 microglia scored per mouse; two-tailed Kruskal–Wallis test). d, Representative tracings of CA1 pyramidal neurons from DNRAb⁻ and DNRAb⁺ Balb/c mice treated with saline or telmisartan. e, Telmisartan treatment rescued dendritic complexity in DNRAb⁺ Balb/c mice compared with saline-treated counterparts (mean ± s.e.m.; n = 5–6 mice per group; n = 2–18 neurons analyzed per mouse; two-tailed, linear mixed model test with Tukey’s adjustment: P = 0.01).

Fig. 6 |. IL-10 is induced by HMGB1 and suppresses Lair1 expression.

a, Il10 expression is higher in B6.H2^d DNRAb⁺ mice than in DNRAb⁻ mice in Ms4a7⁺ hippocampal microglia (P < 0.0001). Pseudobulk expression levels are calculated on a per-mouse basis (y axis, log₂(TPM)), stratified by condition (mean ± s.d.; two-tailed FDR; P values uncorrected). b, Increased Il10 mRNA expression in microglia stimulated with 1 μg ml⁻¹ of HMGB1 compared with unstimulated microglia cultured in medium for 6 h (mean ± s.d.; ex vivo microglia isolated from four WT (B6) mice; two-tailed, paired Student’s t-test: P = 0.0008). c, Decreased Lair1 mRNA expression in microglia stimulated with 20 ng ml⁻¹ of IL-10 compared with unstimulated microglia cultured in medium for 6 h (mean ± s.d.; ex vivo microglia isolated from 6 WT (B6) mice; two-tailed, paired Student’s t-test: P = 0.0123).

Fig. 7 |. Outcomes of DNRAb-mediated neuronal damage.

Exposure of hippocampal CA1 pyramidal neurons to DNRAbs results in DNRAb binding to NMDARs, mediating excitotoxic death in 20–30% of neurons. A maladaptive equilibrium begins as a microglial response to apoptotic neuronal debris and progresses as stressed neurons secrete HMGB1, which activate microglia by binding RAGE. Activated microglia secrete proinflammatory cytokines, type I IFN, C1q and IL-10 in response to HMGB1. IL-10 suppresses LAIR-1 expression on microglia. The secreted HMGB1 acts as a bridge by binding both synaptic proteins and C1q, which tags synapses for microglia-dependent synapse loss, resulting in a loss of neuronal dendrite branching and spine density. Captopril or telmisartan treatment decreases the proinflammatory ATII/AT1R ligation by inhibiting ACE, which converts ATI to ATII, or blocking AT1R, respectively. This allows for microglial LAIR-1 upregulation and quiescence, the return to a healthy homeostasis and regrowth of dendritic branches and spines. ACE_i, ACE inhibitor. Image created with Biorender.com.