Female C57BL/6 mice (6–8 weeks old; 16–20 g) were purchased from the Jinan Pengyue Experimental Animal Breeding Center. They were housed under controlled conditions at the Animal Center of Binzhou Medical University Hospital (Binzhou, China), including a 12-hour light/dark cycle, standard mouse chow, and a stable temperature and humidity environment. This study was approved by the Institutional Animal Care and Use Committee of Binzhou Medical University Hospital (20241108-14).

A total of 36 eight-week-old untreated mice were randomly divided into three groups. After a one-week acclimatization period, 12 mice were randomly selected for the control group (Ctrl), while the remaining mice were used to establish the EAT model after the same acclimatization period. EAT was induced by multi-site subcutaneous (s.c.) immunization. Mice were immunized with 100 μg of pTg (Sigma, T1126) emulsified in Freund’s adjuvant—complete (Sigma, F5881-6X) in the first week and incomplete (Sigma, F5506-6X) from the second to the eighth week—to establish the EAT model, with twelve mice randomly selected as the model group (EAT) (23). Following EAT induction, mice were randomly allocated to the EAT+GL group and received glycyrrhizin (50 mg/kg; MCE, HY-N0184), a functional inhibitor of Hmgb1, via intraperitoneal (i.p.) injection once daily for 7 consecutive days (n = 12) and administered at a fixed time window each day (24). The experimental timeline is illustrated in Figure 1A.

To assess anxiety-like behavior and exploratory activity, an open-field test was performed in a square arena (45 × 45 cm) with 40-cm-high opaque walls. Each mouse (n = 12 per group) was placed in the same corner at the beginning of the test and allowed to explore freely for 5 min while behavior was recorded from above. Videos were analyzed using the JL Behv behavioral video analysis system (Shanghai Jiliang Software Technology Co., Ltd., China). Total distance traveled and time spent in the central ROI were quantified as indices of locomotor activity and anxiety-like behavior. The central ROI was defined as a 25 × 25 cm square in the center of the arena, and all ROIs were kept identical across groups.

Spatial working memory was assessed using a Y-maze consisting of three identical arms positioned 120° apart. Mice (n = 12 per group) were placed at the end of one arm and allowed to explore freely for 10 min while behavior was recorded from above and analyzed using Any-maze. An arm entry was counted when the mouse’s center point crossed into an arm. Spontaneous alternation was defined as consecutive entries into three different arms. The alternation rate (%) was calculated as: Alternation rate (%) = (number of correct alternations/(total arm entries − 2)) × 100.

Recognition memory was assessed using the novel object recognition (NOR) task in the same arena (45 × 45 cm). Mice were habituated to the empty arena for 2 consecutive days. On day 3 (training), mice were allowed to explore two identical objects for 10 min. After 24 h (test), one familiar object was replaced with a novel object, and mice explored for 10 min. Object exploration time was quantified using the JL Behv system. Exploration was defined as the mouse’s nose oriented toward the object within 2 cm for at least 0.5 s; climbing or sitting on the object was not counted as exploration. The discrimination index (DI) was calculated as: DI = (Time with novel − Time with familiar)/(Time with novel + Time with familiar), reflecting preference for novelty.

Spatial learning and memory were assessed using the Morris water maze. The circular pool (diameter 1.5 m; height 0.5 m) was filled with water maintained at 20 ± 1 °C and rendered opaque with white nontoxic dye. A hidden platform (10 cm diameter) was submerged 1 cm below the water surface in one quadrant. Mice were trained for 5 consecutive days to locate the platform, and swim paths were tracked automatically using Any-maze. Escape latency, path length, and swim speed were recorded during training. On day 6, a probe trial was performed with the platform removed, and the number of crossings over the former platform location and/or time spent in the target quadrant within 60 s were recorded to evaluate memory retention.

Mice were deeply anesthetized with isoflurane (induction 3% in oxygen; maintenance 1.5% via a nose cone) delivered using an isoflurane vaporizer. Oxygen flow was set to 0.5–1.0 L/min. Depth of anesthesia was confirmed by loss of righting reflex and absence of pedal withdrawal reflex prior to perfusion, and animals were kept warm on a heating pad throughout the procedure. Mice were then transcardially perfused via the left ventricle: for experiments requiring fresh tissues, perfusion was performed with ice-cold sterile PBS to remove intravascular blood and tissues were collected immediately for immune cell isolation, protein extraction, and RNA analysis; for experiments requiring fixed tissues, mice were perfused with PBS followed by 4% paraformaldehyde for in situ fixation, and tissues were subsequently processed for histology and immunostaining. Death was confirmed by cessation of respiration and heartbeat after perfusion.

After behavioral testing, thyroids were dissected, fixed in 4% paraformaldehyde, dehydrated, paraffin-embedded, and sectioned at 4–5 μm. Sections were stained with hematoxylin and eosin (H&E; Solebo Biotech, Beijing, China) to assess inflammation. Lymphocytic infiltration was scored blinded as 0 (normal), 1+ (1–10%), 2+ (10–30%), 3+ (30–40%), and 4+ (≥50%). Follicular lumen and architecture were examined under a light microscope.

Mice were anesthetized and transcardially perfused with cold PBS followed by 4% PFA. Tissues were cryoprotected in 30% sucrose for 2 days and sectioned at 20–40 µm. Sections were permeabilized with 0.4% Triton X-100 in PBS, blocked with 5% BSA for 1 h, and incubated overnight at 4 °C with primary antibodies against GFAP-labeled astrocytes (Cell Signaling, D1F4Q), Iba1 is a classic and widely used marker for microglia. (Abcam, ab178847), Hmgb1 (Abcam, ab190377), CD4 (Santa Cruz, sc-13573), IL-17A (Abcam, ab79056), CD68 (Abcam, ab53444), C3 (Proteintech, 66157-1-IG), and AQP4 (Santa Cruz, sc-32739). Sections were then incubated with appropriate secondary antibodies for 1 h at room temperature, and nuclei were counterstained with DAPI. Images were captured using an Olympus FV10 confocal microscope. Microglial morphology was analyzed by Sholl analysis in Fiji-ImageJ; One quantification pattern was based on DAPI, GFAP, and AQP4 immunostaining, as previously described. Briefly, in Fiji, the capillary was identified by the fluorescence-deficient lumen surrounded by DAPI- and GFAP-enriched signals, and a donut-shaped ring ROI extending 5 pixels outward from the vessel boundary was generated. AQP4 polarization was then calculated as the ratio of AQP4 fluorescence intensity within this donut ROI to the global AQP4 signal in the field: AQP4 polarization = donut-shaped area AQP4/global AQP4 (25).

Proteins were extracted from cortex and hippocampus using RIPA buffer, and concentrations were determined by BCA assay. Samples were separated by SDS-PAGE and transferred to 0.45 μm PVDF membranes. Membranes were blocked with 5% non-fat milk in TBST for 2 h, then incubated overnight at 4 °C with primary antibodies against Hmgb1 (Abcam, ab190377), IL-17A (Abcam, ab79056), and RAGE (Cell Signaling Technology, 42544; all 1:1000), with GAPDH as loading control. After incubation with HRP-conjugated secondary antibodies (1:10,000; ZSGB-BIO), bands were visualized using ECL (Meilunbio, MA0186-1) on a chemiluminescence imaging system (LI-COR Biosciences).

Total RNA from the hippocampus and cortex of mice was extracted using Trizol reagent (Sangon Biotech, B511311). cDNA was synthesized according to the manufacturer’s instructions using the Evo M-MLV Mix Kit with gDNA Clean (Hunan Accurate Biotechnology, AG11728). RNA purity was measured by spectrophotometry. PCR reactions were set up using SYBR^® Premix Ex Taq™II and amplified on the CFX96 Touch™ Real-Time PCR Detection System (Bio-Rad Laboratories, USA). Data were normalized to GAPDH expression and analyzed using the 2-ΔΔCt method.

Blood samples were collected via orbital extraction under anesthesia. After standing at room temperature for 30 minutes, the samples were centrifuged at 3000 rpm for 20 minutes to obtain serum. Following euthanasia, brain tissues were rapidly harvested, homogenized, and centrifuged to collect the supernatant. Levels of mouse thyroid stimulating hormone (TSH), thyroglobulin antibody (TgAb), free thyroxine (FT4), Estradiol (E2), IL-1β (Interleukin 1 Beta), TNF-α (Tumor Necrosis Factor Alpha), IL-6 (Interleukin 6) and High mobility group proteinB1 (Hmgb1) were measured using mouse ELISA kits (Huamei Biotechnology, Bioswamp, Elabscience Biotechnology, Wuhan). All procedures were performed according to the manufacturer’s instructions.

Single-cell suspensions from brain tissue were prepared as previously described with minor modifications (26). Briefly, mice were anesthetized and transcardially perfused with ice-cold sterile PBS to remove intravascular residual cells. Brains were collected and placed in pre-chilled RPMI-1640 (Lonza). Tissue was mechanically dissociated and enzymatically digested with collagenase/dispase (1 mg/mL) and DNase I (10 μg/mL; Roche Diagnostics) at 37 °C for 45 min with gentle shaking (~100 rpm). The resulting cell suspension was passed through a 70-μm cell strainer and subjected to a 70%–30% Percoll density gradient to remove myelin; leukocytes were collected from the interphase and washed. Cells were blocked with anti-mouse Fc block (BioLegend) and stained with the following fluorochrome-conjugated antibodies: CD45-eFluor 450 (eBioscience, Cat# 48-0451-82), CD11b-APC (BioLegend, Cat# 101212), TMEM119-PE-Cy7 (eBioscience, Cat# 25-6119-82), and P2RY12-PE (BioLegend, Cat# 848003). Zombie Aqua (BioLegend, Cat# 423102) was used to exclude dead cells. Samples from all groups (Ctrl, EAT, and EAT+GL) were processed and stained in parallel within the same batch to minimize experimental variability. Data were acquired on a BriCyte E6 flow cytometer and analyzed in FlowJo; ≥3×10^5 events were recorded per sample. The gating strategy sequentially included FSC/SSC to exclude debris, singlets, live cells, CD45⁺ cells, and CD11b⁺ cells. CD11b⁺ cells were then separated into CD45^int and CD45^high populations based on CD45 expression, and TMEM119 and P2RY12 were assessed within each population to validate microglial identity. Positive gates were set using tissue-matched fluorescence-minus-one (FMO) and unstained controls (27, 28).

Fiji-ImageJ software (Version 2.9.0, NIH, Bethesda, MD, USA) was used to analyze histological and Western blot results. Statistical analyses were performed using GraphPad Prism 10.0 software. The Shapiro–Wilk test was first applied to assess data normality. For comparisons involving only two groups, an unpaired t-test was conducted. For other datasets, one-way analysis of variance (ANOVA) followed by Tukey’s multiple comparisons test was used. For the Morris water maze test, two-way repeated measures ANOVA was applied to compare the three groups. All data are presented as mean ± standard error (SE), and a p-value < 0.05 was considered statistically significant.

Histological comparison of thyroid tissues from control (Ctrl) and experimental autoimmune thyroiditis (EAT) mice was performed both under basal conditions and after EAT induction. H&E staining confirmed the successful establishment of the EAT model, as evidenced by significant lymphocytic infiltration in the thyroid gland (Figure 1B). Concurrently, ELISA was performed to quantify anti-thyroglobulin antibodies (TgAb), thyroid-stimulating hormone (TSH), and free thyroxine (FT4) levels in both serum and brain tissue, confirming successful model induction while maintaining normal thyroid hormone levels (Figure 1C). Western blot analysis of the cerebral cortex and hippocampus revealed elevated Hmgb1 protein levels under disease conditions, suggesting a potential role for Hmgb1 in EAT-associated neuroinflammation (Figure 1D). Immunofluorescence (IF) staining further confirmed Hmgb1 expression in cells of the cerebral cortex and hippocampus. In Ctrl mice, Hmgb1 staining was predominantly nuclear, whereas in EAT mice, pronounced nucleocytoplasmic translocation was observed in the cortex as well as in hippocampal subregions CA1, CA3, and the dentate gyrus (DG), as indicated by yellow arrows (Figure 1E). Neither pTg injection nor glycyrrhizic acid administration induced any detectable change in circulating E2 levels (Supplementary Figure S1A). To assess systemic and CNS Hmgb1 dynamics, Hmgb1 levels were quantified by ELISA in serum, cerebrospinal fluid (CSF), and the soluble fraction of brain tissue homogenates. ELISA revealed that Hmgb1 levels were significantly elevated in EAT mice across serum, CSF, and brain soluble extracts, whereas glycyrrhizic acid (GL) treatment markedly reduced these increases (Supplementary Figures S1B, S2).

To evaluate the functional role of Hmgb1 in EAT-induced neurobehavioral alterations, glycyrrhizic acid (50 mg/kg) was administered to EAT model mice. A battery of behavioral assays—including the Morris water maze, Y-maze, novel object recognition, and open field test—was used to assess cognitive performance and anxiety-like behaviors (Figures 2A–D). Concurrently, the expression of inflammatory cytokines in the hippocampus and cortex was analyzed. In the Morris water maze task, EAT mice displayed impaired spatial learning, evidenced by prolonged latency to locate the hidden platform during training. This cognitive deficit was significantly attenuated by Hmgb1 inhibition, as reflected by shorter escape latencies in the EAT+GL group on days 4 and 5 (Figure 2Ea). In the probe trial, the EAT+GL group spent more time in the target quadrant (Figure 2Eb) and showed increased platform crossing frequency (Figure 2Ec) compared to untreated EAT mice, indicating enhanced spatial memory retention. The Y-maze test revealed deficits in working memory in EAT mice, demonstrated by a reduced spontaneous alternation rate, which was significantly improved following Hmgb1 inhibition (Figures 2B, F). Similarly, the novel object recognition test showed that EAT mice exhibited impaired recognition memory, spending less time with the novel object. Treatment with glycyrrhizic acid restored novel object preference to levels comparable to control animals (Figures 2C, G). Anxiety-like behavior was assessed using the open field test. EAT mice spent significantly less time in the central zone, indicative of increased anxiety levels. This phenotype was reversed in the EAT+GL group, which displayed increased time in the center, consistent with reduced anxiety (Figures 2D, H). At the molecular level, EAT mice exhibited a significant upregulation of pro-inflammatory cytokines—including IL-1β, IL-6, and TNF-α—in both the cortex and hippocampus. These increases were markedly attenuated in the EAT+GL group. In contrast, the anti-inflammatory cytokine IL-10 showed no significant changes after Hmgb1 inhibition (Figures 2I, J). ELISA measurements further confirmed significantly elevated protein levels of IL-1β, IL-6 and TNF-α in EAT mice, whereas these increases were markedly reduced in the EAT+GL group (Supplementary Figure S1C). Collectively, these findings demonstrate that inhibition of Hmgb1 ameliorates cognitive deficits and anxiety-like behaviors in EAT mice, likely through suppression of neuroinflammation.

3o investigate the effects of EAT on microglial activation and assess the impact of Hmgb1 inhibition, IF staining was performed in the cortex and hippocampus of EAT mice (Figures 3A, B). Quantification revealed a significant increase in Iba1 microglia in EAT mice compared to controls, which was reversed following treatment with the Hmgb1 inhibitor glycyrrhizin (Figure 3D). Co-labeling with anti-Iba1 and anti-Hmgb1 antibodies revealed a shift in Hmgb1 localization from the nucleus to the cytoplasm in microglia, indicating active release of Hmgb1 into the extracellular environment (Figures 3A–C). This suggests that microglia may serve as a source of extracellular Hmgb1 under EAT conditions. To further evaluate microglial morphology, Sholl analysis was conducted to assess process complexity and activation status (Figures 3Ea, Fa). EAT mice exhibited hallmark features of microglial activation, including reduced branch intersections (Figures 3Eb, Fb), increased soma diameter (Figures 3Ec, Fc), and decreased total branch number (Figures 3Ed, Fd). Hmgb1 inhibition attenuated these morphological changes, indicating suppression of microglial overactivation in both cortical and hippocampal regions (Supplementary Figure S3).

To investigate the effect of Hmgb1 on neuroinflammation, we examined microglial activation in the hippocampus, cortex, and cerebellum of EAT mice. Immunofluorescence staining for CD68, a marker of activated microglia, revealed a marked increase in CD68 microglia in EAT mice compared with controls. Notably, treatment with an Hmgb1 inhibitor significantly reduced the proportion of microglia co-expressing CD68 in the hippocampus and cortex, indicating that Hmgb1 plays a role in microglial activation in EAT mice. (Figure 4) (Supplementary Figure S4).

To further characterize the role of astrocytes in HT-associated neuroinflammation, we investigated the impact of EAT on hippocampal astrocyte reactivity and the modulatory effects of Hmgb1 inhibition. IF analysis revealed a marked increase in GFAP⁺ astrocyte density in the hippocampus of EAT mice, which was significantly reduced following Hmgb1 inhibition (Figures 5A, Da). Co-localization of GFAP and Hmgb1 demonstrated nuclear-to-cytoplasmic translocation of Hmgb1 in astrocytes, suggesting that astrocytes may serve as an additional source of extracellular Hmgb1 during HT-related neuroinflammation (Figure 5Db). To assess astrocyte phenotypic changes, IF staining for GFAP and complement component 3 (C3) was performed. Compared to controls, EAT mice showed elevated C3 expression, indicative of a shift toward the neurotoxic A1 astrocyte phenotype. Notably, Hmgb1 inhibition reduced C3 expression and attenuated A1 polarization (Figures 5B, Ea, b), suggesting a role for Hmgb1 in astrocyte-driven complement activation and inflammatory signaling. Using GFAP and AQP4 double staining with quantification via the Donut method (25), we observed a significant loss of AQP4 polarity in EAT mice relative to controls, indicative of glymphatic disruption. Strikingly, Hmgb1 inhibition restored AQP4 polarity, further supporting its therapeutic potential in preserving astrocytic function and CNS fluid dynamics (Figures 5C, Fa, b) (Supplementary Figure S5).

To investigate the involvement of CD4⁺ T cells and IL-17A in EAT-induced CNS injury, we performed IF and Western blot analysis. IF staining revealed a marked accumulation of CD4⁺ T cells in the cortex and hippocampus of EAT mice, with a greater density observed in the cortical region. These infiltrating CD4⁺ T cells exhibited robust IL-17A expression (Figures 6A–D). Hmgb1 inhibition significantly reduced IL-17A production by CD4⁺ T cells, as confirmed by IF co-localization (Figures 6C, D) and Western blotting (Figures 6Ea, Ec). In parallel, expression of the receptor for advanced glycation end products (RAGE)—a key surface receptor mediating Hmgb1 signaling—was elevated in EAT mice but was down regulated following Hmgb1 blockade (Figures 6Eb, di). Inhibition of Hmgb1 effectively suppressed IL-17A release in the thyroid, suggesting that Hmgb1 contributes to local Th17-mediated inflammation (Figure 6G) (Supplementary Figure S6).

To distinguish resident microglia from infiltrating peripheral myeloid cells, we performed flow cytometry on brain single-cell suspensions. Within CD45⁺CD11b⁺ myeloid cells, we separated CD45^int and CD45^high subsets based on CD45 intensity and further assessed TMEM119 and P2RY12 within each subset to validate microglial identity. Compared with controls, the EAT group showed a marked increase in the CD11b⁺CD45^high population (Ctrl: 3.12% vs EAT: 12.4%), accompanied by a decrease in the CD11b⁺CD45^int population (Ctrl: 83.5% vs EAT: 67.2%). These changes were partially restored by GL treatment (EAT+GL: CD45^high 5.08%, CD45^int 76.9%). Consistent with a microglial phenotype, the majority of cells within the CD11b⁺CD45^int gate were TMEM119⁺P2RY12⁺ across groups (Ctrl: 97.4%, EAT: 91.0%, EAT+GL: 94.5%). In contrast, TMEM119⁺P2RY12⁺ events were generally low within the CD11b⁺CD45^high gate (Ctrl: 5.65%, EAT+GL: 4.02%), indicating that this population is enriched for infiltrating myeloid cells. Notably, the TMEM119⁺P2RY12⁺ fraction within CD45^high increased to 18.0% in the EAT group, which may reflect inflammation-associated CD45 upregulation in activated microglia leading to partial overlap with the CD45^high gate. Collectively, these data indicate that EAT promotes accumulation of CD45^high infiltrating myeloid cells in the brain, whereas GL treatment reduces this infiltrating compartment and partially normalizes the distribution of microglia-associated populations (Supplementary Figure S7).