Celastrol Exerts Neuroprotective Effect via Directly Binding to Hmgb1 Protein in Cerebral Ischemic Reperfusion

Dandan Liu China Academy of Chinese Medical Sciences Institute of Chinese Materia Medica Piao Luo China Academy of Chinese Medical Sciences Institute of Chinese Materia Medica Liwei Gu China Academy of Chinese Medical Sciences Institute of Chinese Materia Medica Qian Zhang China Academy of Chinese Medical Sciences Institute of Chinese Materia Medica Peng Gao China Academy of Chinese Medical Sciences Institute of Chinese Materia Medica Yongping Zhu China Academy of Chinese Medical Sciences Institute of Chinese Materia Medica Xiao Chen China Pharmaceutical University Junzhe Zhang China Academy of Chinese Medical Sciences Institute of Chinese Materia Medica Nan Ma China Academy of Chinese Medical Sciences Institute of Chinese Materia Medica Jigang Wang (  jgwang@icmm.ac.cn ) China Academy of Chinese Medical Sciences Institute of Chinese Materia Medica https://orcid.org/0000-0002-0575-0105


Background
Tripterygium wilfordii Hook. f. (TwHF)-based prescription is widely used in the treatment of autoimmune diseases, tumor and so on in China for centuries [1,2]. Out of many bioactive constituents isolated from Tripterygium wilfordii, celastrol has attracted closed attention for more than 70 years in terms of possible medicinal property [3]. Celastrol exhibits a diversity of pharmacological effects in a wide range of disorders, such as cancer, diabetic, obesity [4], neurodegenerative diseases [3]. An abundance of existing researches have proved neuroprotective effects of celastrol in neurodegenerative diseases through antioxidant and attenuating neuro-in ammation [5]. Besides, celastrol excellently relieved acute ischemic stroke-induced injury by promoting microglia/macrophage M2 polarization [6], reducing the expressions of p-JNK, p-c-Jun and NF-κB [7], and inhibiting high mobility group box 1 (HMGB1)/NF-κB signaling pathway to exhibit anti-in ammatory and antioxidant actions in transient global cerebral ischemic rats [8]. However, there are few studies about whether celastrol has neuroprotection effect for cerebral ischemic-reperfusion (I/R) injury and speci c binding protein targets.
Neuro-in ammatory processes have been implicated in the pathophysiology of multiple stages of cerebral I/R injury, and targeting neuro-in ammation has always been an attractive treatment in stroke [9]. It has taken several periods for the intensive study of HMGB1 in in ammation related diseases, which is currently one of the crucial pro-in ammatory alarmin of stroke. The non-histone DNA binding protein HMGB1 is primarily located in the cell nucleus and behaves different biological functions according to cellular locations, binding receptors and redox states. HMGB1 shifts to the cytoplasm and extracellular space by activated immune cells or passively released by necrotic or damaged cells, with dynamic redox states due to distinct posttranslational modi cations [10] and activated in ammatory immune reaction [11]. Outside of the cell, HMGB1 serves as a damage-associated molecular pattern (DAMP) or alarmin to mediate in ammation through receptors including advanced glycation end products (RAGE), toll like receptor 2 and 4 (TLR2, TLR4) [12]. HMGB1 constitutes three domains: A box, B box (positively charged) domains and carboxyl terminus (negatively charged) acidic tail. The three cysteines located at positions 23, 45 (A box) and 106 (B box) mainly determine the redox state and physiological functions of HMGB1.
The fully reduced HMGB1 only possesses chemotaxis by binding to CXCL12, and stimulates immune cell in ltration through the CXCR4 receptor in a collaborative way. An intramolecular disul de bond of HMGB1 in cysteines C23 and C45 with critical C106 in a reduced state has pro-in ammatory activity like cytokines via the TLR4/MD-2 complex, induces nuclear NF-κB translocation and produces tumor necrosis factor (TNF) in macrophages. Meanwhile, chemotactic and cytokine activities disappeared after all the cysteines oxidized (sulfonyl HMGB1) [13,14]. Therefore, disul de HMGB1 isoform is a biomarker of in ammation and blocking extracellular disul de HMGB1 isoform maybe a potential direction for the treatment in ammation and immune diseases, including stroke.
Quantitative chemical proteomics technology based on small molecule compound probe and chemical labeling has been widespread used for seeking targets and elucidating the mechanism of natural and traditional medicines [15], including Artemisinin [16], andrographolide [17], curcumin [18], aspirin [19].
With the help of activity-based celastrol probe (cel-p), tandem mass tags (TMT) labeling, liquid chromatography-tandem mass spectrometry (LC-MS/MS) and cellular thermal shift assay (CETSA), we elucidated the neuroprotective mechanism and targets of celastrol on stroke I/R injury, and uncovered that celastrol directly bond with HMGB1 to inactivate its cytokine activity and targeted HSP70 and NF-κB to exert anti-in ammatory activity.

Animals
All experiments were carried out for the sake of minimizing the number and suffering of animals. All The Sprague-Dawley rats within 12h of birth (Vital River Laboratories, Beijing, China) were used for primary cortical neurons isolation. Male Sprague-Dawley rats (260-280g, Vital River Laboratories, Beijing, China) used for middle cerebral artery occlusion (MCAO) experiment were housed in standard breeding environment without restriction to diet and drinking.
Click chemistry, pull down and LC-MS/MS reagents: NaVc; CuSO4; TAMRA Azide and Biotin-azide were obtained from Sigma, USA. High capacity neutravidin agarose resin; sequencing grade modi ed Trypsin; TMT 10 plex reagent set; TEAB and Pierce™ Quantitative Fluorometric Peptide Assay Kit were purchased from Thermo Fisher Scienti c, USA. Oasis HLB Extraction Cartridge was obtained from Waters. THPTA was obtained from Click Chemistry Tools.
Rat primary cortical neurons isolation and RAW 264.7 cell culture The neonatal Sprague-Dawley rats within 12h of birth were used for primary cortical neurons isolation as previously established with minor revise [20]. Brie y, the cortex of newborn rats was sterile separated in pre-cooling (4°C) DMEM/F-12 (1:1). The minced cortex tissue was digested with 0.2mg/ml DNase and 2mg/ml papain, and inactivated by adding 10% volume FBS. The cell suspension was washed twice with DMEM/F-12 (1:1) and re-suspended in DMEM/F-12 (1:1) containing 10% FBS and 1 × PS. The cells suspension passed through 300 mesh sieves were seeded on L-polylysine pre-coated ori ce plate or dish and incubated at 37°C incubator with 5% (v/v) CO 2 . 4-6h later, the DMEM/F-12 (1:1) was replaced with complete Neurobasal-A medium replenished with 1× PS, 1× Glutamax-I and 1× B27. The culture medium was change half every 2-3 days, and all the experiments were carried out on the seventh day unless otherwise stated.
RAW 264.7 cell was cultured in DMEM containing 10% FBS, 1× PS and maintained in a cell incubator. Cells were passage every 2-3 days and TNF-α was testing in cell within 20 passages.

Oxygen glucose deprivation (OGD) insult
The transient OGD model was constructed to simulate cerebral I/R injury in cultured primary neurons as previously described [21]. Brie y, the Neurobasal-A medium was displaced with deoxygenated, sugar free DMEM, and the cells were incubated in a hypoxia chamber (STEMCELL Technologies, Canada ) lled with 95% N 2 + 5% CO 2 for 4 h in 37°C incubator and returned to normal culture condition according to the experimental requirement. In contrast, control cells were cultured in normal culture conditions. After reaching the established time, cell viability evaluation was determined by CCK8 assay or cells were collected for other experiments. For CCK8 assay, absorbance was measured using a multimode plate reader (PerkinElmer, USA) at 450 nm.
Proteome reactivity pro les of primary neuron treated with cel-p Fluorescence labeling pro ling of cel-p binding proteins was conducted in living primary neuron with or without celastrol competitor and OGD model refer to previous operation [22]. Similarly, increasing concentrations of cel-p (0-1.6 μM) or cel-p (0.8μM) + competitor (celastrol 2 ×, 4 ×, 6 ×, 8 ×) were added into the 6 well plates with or without OGD interfere and incubated for 4 h in cell incubator. Then supernatant of cell lysate were collected and BCA method was used for protein concentration quanti cation. The click chemistry reaction was conduct with NaVc (100 mM stock solution, nal concentration 1 mM), THPTA (100 mM stock solution, nal concentration 100 μM), CuSO4 (100 mM stock solution, nal concentration 1 mM) and TAMRA Azide (5 mM stock solution, nal concentration 50 μM) in equal amounts (100 μg) of extracted protein for 2h at room temperature. The protein was precipitated with 1 ml pre-cooling (-20°C) acetone, and re-dissolved with 30 μl 1× SDS loading buffer. 15 μl of sample was separated with 10% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) gel, and the labeling pro les were visualized with in-gel uorescence scanning in laser scanner (Azure Sapphire RGBNIR, USA) and then the gel was stained with CBB.
Cellular imaging of cel-p Cellular imaging experiment was conduct with uorescence microscopy as described previously to verify the utility of cel-p for imaging of potential cellular targets [23]. To track the cellular distribution of cel-p, living primary neurons were incubated with 0.8 μM cel-p for 0-6 h. The cells were xed with 4% paraformaldehyde solution for 10 min and 0.2% Triton-X 100 permeated 15 min. Click chemistry reaction was carry out (regents and concentration referenced 2.3.3) for 2h and washed thrice to remove excessive agents. The nuclei were stained with Hoechst for 10min. The images were obtained with confocal uorescence microscopy (Leica TCS SP8 SR, Germany).

Pull down, TMT labeling and targets validation
Pull-down, TMT labeling and LC-MS/MS experiments were carried out to identify the interacting cellular targets of celastrol and primary neuron according to previous description with certain modi cation [24].
Primary neuron cells cultured in 100 mm dishes were divided into the following groups and continue incubated for 4 h : Control (less than 1% DMSO), cel-p (4 μM), and celastrol 8 × (cel-p 4 μM + celastrol 32 μM Target engagement assay of celastrol with HMGB1 was performed the same with previous article with minor modify [25]. The primary neurons in 100 mm dishes were collected with PBS containing protease inhibitor. The cells were subjected to freezing and thawing cycles in liquid nitrogen and repeated mechanical crushing to obtain cell lysate supernatant by centrifugation. Then equivalent supernatant (1 mg) was treated with either DMSO or celastrol (20 μM) 1h at room temperature with gently shaking. The treated supernatant was divided into ten equal parts and heated according to designated temperatures. The cooled samples were centrifugation again to obtain supernatant and conducted Western blot analysis.
Western blot analysis Supernatants of neuron lysate or rat cerebral cortex tissues lysate were obtained with RIPA lysate and protease inhibitor. For the cytoplasm and nuclear protein extraction, the nuclear-cytosol extraction kit was used on the basis of instructions. Fully reduced and disul de bond HMGB1 isoforms were detected in rat primary cortex neuron and condensed culture supernatant with non-reducing PAGE gel, and samples were collected avoid from reducing agent (β-mercaptoethanol or DTT). After OGD and celastrol treatments as detailed above, the culture supernatant was collected at 0, 4, 8, 12, 24, 48 h and centrifuged to discard cell debris. Then the supernatant was concentrated 20 folds with Amicon Ultra-4 50kDa and Amicon Ultra-4 10kDa. Protein content was identi ed using the BCA assay, and the denatured sample was separated with 10%, 12% or 15% SDS-PAGE gel.
The bands were visualized with enzyme-linked chemiluminescence in the detection system (Azure C400, USA).

Immuno uorescence staining
For animal Immuno uorescence, after series dewaxing and dehydration, rat cerebral cortex para n slice were incubated in antigen retrieval for 10 min in 95°C and permeabilized 15 min in 0.2% Triton X-100. The slices were blocked 1h in 5%BSA, incubated with primary antibodies against HMGB1, HSP70, or NF-κB at 4 °C overnight and 2 h with secondary uorescence antibodies (goat anti-rabbit, 1:500; goat anti-mouse, 1:500, Abcam) avoiding from light. After 10 min of Hoechst staining, the slices were photographed with laser scanning confocal microscope.
For cell Immuno uorescence, the treated cells were washed with PBS, xed in 4% paraformaldehyde solution and permeabilized in 0.2% Triton-X 100. The rest procedures were the same with animal Immuno uorescence.

Expression and puri cation of HMGB1 A box and B box
Recombinant human HMGB1 box A (residues 1-89) and box B (residues 90-175) were cloned in a modi ed pET-24d vector (Novagen, Madison, WI) expressing a protein with an N-terminal 6-His tag. The E. coli BL21 was transformed with pET24d-HMGB1 A box and pET24d-HMGB1 B box, cultured in LB medium containing 50g/ml kanamycin at 37℃ to an absorbance of 0.8 at 600 nm, and expression was induced with 0.4 mM Isopropyl-D-1-thiogalactopyranoside (IPTG) for 12h at 16℃ before being harvested by centrifugation. Cell pellets were suspended in lysis buffer (20 mM  For activity assay of celastrol and HMGB1 protein, recombinant human HMGB1 protein and celastrol combined HMGB1 activity were measured by stimulating TNF-α in RAW 264.7 cells for 24h. For con rming whether celastrol blocked the binding of receptors TLR4 and RAGE to B box, the semi in vivo precipitation of B box complex experiment was conduct as previously with minor revise [26]. 100 μg B box proteins were rst reacted with or without equivalent amount (equimolar with B box, 10 -2 μmol) and ve folds amount of celastrol (5 ×10 -2 μmol) in Ni-beads column for 1 h. Then 500 μg lysate of primary neurons was add into the Ni-beads column and reacted 2 h at 4a, and eluted to precipitate B box complex. The denatured complex sample was subjected to immunoblot analysis with antibodies against TLR4 and RAGE. The same B box with equivalent amount of celastrol elution did not incubate with cell lysate was used for TNF-α analysis in RAW 264.7 cells to research whether celastrol reduced the ability of B box inducd TNF-α secretion.

Induction of MCAO and neurological defect assessment in rats
By inserting an lament to occlude the right middle cerebral artery (MCA) of male Sprague-Dawley rats (250-280g), we established cerebral I/R model in vivo as described previously [27]. Brie y, the rats fasted overnight were anesthetized with continuous supply of 3% iso urane + 95% oxygen mixture. A 4-0 mono lament was inserting into the internal carotid artery through a tiny incision in the external carotid artery to block the cerebral blood ow of MCA. 90 min later, the lament was withdrawn to resume blood stream providing. The Sham group did the same steps without inserting the lament. The neurological de cits of rats were assessed at 24, 48, 72 h after reperfusion respectively by an experimenter who was blinded to the group information. Zea-Longa ve-point scores was used to assess neurologic de cits according to previous description [27]. The animals without symptoms of neurological impairment or dying after surgery were rejected. 72h after reperfusion, infarct volume was determined by 2, 3, 5-triphenyltetrazolium chloride (TTC) stating and calculated with Image J software as previously [27]. Nissl-stained cells in the rat cerebral cortex were observed at 200 × magni cations with a light microscope to assess Nissl body damage.

Statistical analysis
Data were presented as the mean ± SEM. Raw data were statistically analyzed with Graph Pad Prism 5.0. The density of Westren blot bands was quanti ed using Image J software. The data were analyzed using one-way ANOVA. Fisher's least-signi cant difference post hoc test was used to test the differences between two groups. P value less than 0.05 was considered statistically signi cant.

Results
Celastrol and cel-p showed similar neuroprotective effect in vitro Cel-p was synthesized with an alkynyl handle to the carboxyl terminal of celastrol (Fig. 1A). We rst detected the cytotoxicity of celastrol and cel-p in living primary neuron. As shown in Fig. 1B, the IC 50 results suggested that cel-p closely mimicked the original compound in biological activity (celastrol IC 50 = 2.2 μM, cel-p IC 50 = 2.0 μM). For cell viability assay, primary neurons were incubated in 96 well plates for 7 d and then experienced OGD model to evaluate the neuroprotective effect of celastrol and cel-p. As shown in Fig. 1C, celastrol exhibited obvious neuroprotective effect on the OGD model of primary neurons. The optimal dose of celastrol was 0.1-0.8 μM, and the optimal administration time was 48 h after OGD model. Cel-p showed similar neuroprotective effect with celastrol (Fig. 1D). Therefore, celastrol remained biological activity after introducing biorthogonal reaction groups and cel-p could be used to instead the celastrol for subsequent research.
Celastrol signi cantly decreased pathological changes of MCAO rats Previous researches indicated that celastrol reduced neurological de cit, brain water content and infarct volume in rat permanent cerebral ischemic model [6,7]. Therefore, we mainly evaluated whether celastrol exhibited neuroprotective effect for rat cerebral I/R injury. At 72 h after MCAO model, celastrol (1mg/kg, i.p.) signi cantly decreased the infarct volume ( Fig. 2A, B), improved behavior indexes (Fig. 2C) and reduced cortex pathological changes of Nissl staining (Fig. 2D, E) of MCAO rats compared with Model group. The neuroprotective effect of celastrol was similar to positive control edaravone injection (6mg/kg, i.v.).

Cel-p possessed high bioconjugation e ciency
We evaluated the labeling pro les of cel-p in living primary neurons. As shown in Fig. 3A, cel-p showed strong labeling e ciency and concentration dependent labeled living primary neurons protein and produced obvious visible bands at as low as 0.8 μM probe concentration in 4 h of incubation time. As shown in Fig. 3B, the labeling pro les of cel-p (0.8μM) became weaker in the presence of competitor celastrol (1.6-6.4μM), suggesting that cel-p bond similar intracellular targets with celastrol. Cellular imaging experiments with click chemistry reaction were performed to study the cellular location of cel-p in living cells. As exhibited in Fig. 3C, cel-p mainly localized in the cell cytoplasm within 2 h and gradually entered into the cell nucleus, indicating that the cel-p was able to label cytoplasm and nuclear proteins after 4h of labeling. These data demonstrated that cel-p possessed high bioconjugation e ciency under in situ condition and was a suitable substitute of celastrol for subsequent chemical proteomics procedures.
Celastrol directly targeted HMBG1 and did not affect the expression of HMGB1 Next, we proceeded to identify cellular targets of celastrol by quantitative chemical proteomics technology. The protein with enrichment ratio R (cel-p/cel-p+8× cel) > 1.5, p < 0.05 was set as signi cant hits, and the protein information labeled by cel-p under this standard was analyzed. The identi ed protein hits were systematically analyzed and displayed by corresponding volcano plots after cel-p (4 μM) with or without celastrol 8 × (32 μM) treatment for 5h. A total of 1405 proteins were identi ed by cel-p target recognition experiment, of which 120 were highly reliable. A complete list of identi ed proteins was provided in Table S1. On the basis of these criteria, HMGB1 was identi ed as a one of the direct binding proteins of celastrol, which has fairly high credibility and is currently one of the crucial pro-in ammatory alarmin of stroke (Fig. 4A). Primary neurons pull down, Western blot and Immuno uorescence assays veri ed that celastrol 8× could completely compete the binding of cel-p to HMGB1 protein, further demonstrated that HMGB1 was a direct binding protein of celastrol ( Fig. 4B-D). The CETSA results also proved the direct binding of celastrol and HMGB1 protein, so as to decrease the protein degradation with increasing temperature compared with Control group (Fig. 4E-F).
As shown in Fig. 4C-D, the HMGB1 protein in celastrol 8 × (3.2 μM) group was almost invisible. In order to con rm whether high concentration of celastrol or cel-p affected the expression of HMGB1 in primary neurons, we treated normal living primary neuron with high concentration of celastrol (20 μM and 40 μM) for 5 h to evaluate its effect on HMGB1 expression with Western blot. In OGD model, we also examined the effect of cel-p (4μM) or cel-p + celastrol 8 × on HMGB1 expression in neuron cells lysate or living neuron cells. As shown in Fig. 4G, high concentration of celastrol or cel-p did not decrease the expression of HMGB1. We speculated that high concentration of celastrol occupied the binding sites of HMGB1, which led to the failure of HMGB1 protein to bind to HMGB1 antibody, rather than the degradation of HMGB1 by high concentration of celastrol or cel-p. In additon, we con rmed that celastrol had no effect on HMGB1 expression in normal cells in a certain dose and time range (Fig. 4H-K). Unexpectedly, we did not observe the time-dependent increasing secretion of HMGB1 in OGD model. On the contrary, the expression of HMGB1 in Model and M + cel groups both decreased in a time-dependent manner ( Fig. 4L-N). In conclusion, celastrol directly targeted HMGB1 and did not affect the expression of HMGB1.
Celastrol played neuroprotective effect through HSP70 and NF-κB p65 As mentioned above, we did not nd that celastrol affected the expression of HMGB1 in primary neurons with or without OGD insult. Celastrol induced HSP70 response and suppressed NF-κB activation to inhibit in ammatory responses and regulate innate immunity response in previous researches [28 -30]. Therefore, we established OGD model in vitro and MCAO model in vivo to mimic cerebral I/R injury and tested whether celastrol affected the distribution changes of HMGB1 in the cytoplasm and nucleus and the expression changes of protein HSP70 and NF-κB p65. We found that the expression of HMGB1 in the cytoplasm of Model and M + cel group signi cantly increased, while the expression of HMGB1 in the nucleus obviously decreased (Fig. 5A-F).In summary, HMGB1 overall expression was barely affected by celastrol, and celastrol could hardly affect the distribution of HMGB1 in the cytoplasm and nucleus 48h after OGD injury. In contrast, celastrol signi cantly increased both of the overall and nucleus expression of HSP70 and decreased the overall and nucleus expression of NF-κB p65, which was consistent with previous studies [28 -30] (Fig. 5A-F). The Immuno uorescence results were in line with Western blot results in vitro (Fig. 5J). Similar Western blot and Immuno uorescence results were also observed in MCAO rats ( Fig. 5H-N). Compared with Model group, celastrol (1mg/kg) remarkable increased the expression of HSP70, down regulated the expression of NF-κB, and had no effect on the expression of HMGB1 (Fig. 5 H-N) in rats MCAO model.
Celastrol did not affect the secretion and redox state of HMGB1 HMGB1 in the concentrated supernatant of primary neuron OGD Model group increased gradually in a time-dependent manner, and reached peak at 24h (Fig. 6 A). The secreted HMGB1 in Model and M + cel (0.8 μM) group in 48h was almost the same (Fig. 6B), which indicated that celastrol did not affect the secretion of HMGB1 after OGD injury. HMGB1 in concentrated supernatant was actively secreted in response to OGD injury at 48h. Celastrol hardly affected the redox state of HMGB1, presented as the disul de bond form HMGB1 showed a faster mobility in non-denaturing page gel and celastrol did not affect its mobility (Fig. 6C). In addition, the disul de bond but not fully reduced form HMGB1 occupied the vast majority form of Model and M + cel group in primary neurons injured by OGD compared with Control group (Fig. 6D). According to the present results, celastrol could not affect the secretion and expression of HMGB1 protein, nor affect the redox state of HMGB1. As shown in Fig. 6E, HMGB1 includes two DNA binding domains (A, B box) and the acidic C-terminal tail. Three cysteine residues (Cys23, 45 and 106) in A, B box mainly determine the redox state of HMGB1. The disul de bond between Cys23 and Cys45 and reduced Cys106 is indispensable for the binding of HMGB1 to TLR4 and cytokine-inducing activity [10]. Hence, we focused on whether celastrol bond with HMGB1 to weaken its cytokine activity.

Celastrol directly bond to HMGB1, HMGB1 A box and B box
Celastrol exhibited strong combining ability with recombinant human HMGB1 protein in a concentration dependent manner (Fig. 7A-B). Celastrol almost competed the binding of IAA-yne to HMGB1 or DTT reduced HMGB1 (Fig. 7C-D), which indicated that celastrol could occupy the Cys106 of disul de HMGB1 or Cys23, 45, 106 of reduced HMGB1. This result was consistent with previous studies that celastrol was proposed to exert cellular effects by forming covalent adducts with cysteine residues of proteins [31][32][33].
The binding ability of celastrol to HMGB1 protein was stronger than GA and Metformin, which were recognized HMGB1 inhibitors by binding to A, B box and C-terminal acidic tail of HMGB1 respectively ( Fig.  7E-F). Celastrol also almost completely blocked the binding of IAA-yne to cysteines group of recombinant human HMGB1 A, B box. The binding of cel-p to A and B box could not be blocked by IAA (Fig. 7 G), which indicated that celastrol could bind to other sites of HMGB1 in addition to the cysteines 23, 45 and 106.

Celastrol remarkably blocked the cytokine activity of HMGB1 and B box
Celastrol did not affect the expression and redox state of HMGB1 after OGD injury. In addition, previous research indicated that only disul de bond form HMGB1 possessed cytokine activity [10]. Therefore, we mainly explored whether celastrol in uenced the disul de bond form HMGB1 induced TNF-α increase in RAW 264.7 cells. According to the description of the manual, EC 50 of HMGB1 for stimulating RAW 264.7 cells TNF-α production was 0.7855-0.8342μg/ml, so we chose 0.8 μg/ml HMGB1 to induce the secretion of TNF-α. As shown in Fig. 8A, 0.8μg/ml HMGB1 obviously increased the TNF-α secretion compared with control group, and 0.1μM or 0.05μM celastrol apparently decreased the TNF-α secretion in RAW 264.7 cells. 0.02μg/ml and 0.2μg/ml recombinant human HMGB1 B box obviously increased the TNF-α secretion compared with control group, and 1× (10 -6 μmol and 10 -5 μmol) celastrol apparently decreased the TNF-α secretion in RAW 264.7 cells (Fig. 8B). According to the published literature, TLR4 is the only receptor of HMGB1 for producing cytokine by binding to the B box cysteine 106 [34]. Then we utilized recombinant human B box, B box-celastrol complex to verify that 1× and 5× celastrol obviously blocked the binding of receptors TLR4 and RAGE to B box (Fig. 8C). In conclusion, celastrol remarkably blocked the cytokine activity of HMGB1 and B box by directly binding to them to block the combination of in ammatory receptors with them.

Discussions
Our studies demonstrated that: (1) celastrol exhibited neuroprotective effect for ischemic stroke in vitro and in vivo; (2) celastrol directly bond the HMGB1; (3) celastrol did not affect the expression of HMGB1, increased the HSP70 and decreased the NF-κB expression to play anti-in ammatory effect in vitro and in vivo; (4) celastrol bond the 106 cysteine of disul de bond HMGB1 or 23, 45, 106 cysteines of fully reduced HMGB1; (5) celastrol scavenged the overproduced TNF-α induced by disul de bond HMGB1 and B box; and (6) celastrol played anti-in ammatory effect by binding to the B box to block the combination of TLR4 and RAGE with HMGB1 B box . Taken together, our ndings suggested that the neuroprotective action of celastrol for ischemic stroke was due to its inhibition of neuroin ammation effect through up regulating HSP70 and decreasing NF-κB expression and directly binding with HMGB1 protein. The speci c experimental process was shown in Fig. 9.
With the maturation of Mass-spectrometry based proteomics technology, stable isotopes label has been widely used in the research of biomarkers for diseases and drugs targets through quantitative measurement of relative or absolute protein amounts in healthy versus disease states [35]. TMT and iTRAQ are commercially available and widespread application isobaric tag for they allow multiplexing of up to 10 samples with high-resolution instruments and a range of sample types applicable [36]. TMT isobaric labeling could simultaneous identi cation and quanti cation of the complex protein mixtures components with the key work ow of sample denaturation, digestion, isobaric tagging of tryptic peptides, fractionation, mass spectrometric analysis and data processing [37,38]. First, we veri ed that celastrol remained biological activity after introducing biorthogonal reaction groups. Celastrol exhibited neuroprotective effect for ischemic stroke in vitro and in vivo. With the aiding of cel-p, TMT label and LC-MS/MS technologies, we identi ed 1405 proteins in cel-p target recognition experiment, of which 120 were highly reliable. HMGB1 was identi ed as a direct binding protein of celastrol with fairly high credibility (Fig. 4).
CETSA is a label-free biophysical technique for studies of target engagement in cells and tissues based on ligand binding affects protein stability and cellular studies of proteins redox modulations [39].
Generally speaking, many proteins unfold after heating and precipitate rapidly in cells, and drug binding protein can stabilize the protein and reduce its degradation with increasing temperature compared with no treated cells [40]. CETSA based on immunoassays (such as Western blot, proximity ligation assays, mass spectrometry) detection is a hot technique to validate ligand binding of drugs to proteins in lysates, cells and tissues, which is based on measuring the protein melting curves changes after different heating steps and quantifying the amount of remaining soluble protein [40,41]. We investigated whether celastrol combined with HMGB1 and stabilized HMGB1 in cell lysate samples subjected to temperatures from 37 to 82 °C. The results showed that compared with DMSO treated control group, the celastrol treated group signi cantly stabilized the HMGB1 and decreased its degradation with temperature increase (Fig. 4E-F).
HMGB1 is highly expressed in the nucleus of multiple cell types, and the redox state of intracellular and extracellular HMGB1 is dynamic mainly related with the 23, 45, and 106 cysteines. Disul de bond form of HMGB1 (C106 in thiol and C23, C45 form disul de bond) is requisite for TLR4/MD-2 interaction to induce TNF release and NF-κB activation [10]. HMGB1 can be released by passive or active secretion via multiple pathways. Passive release of HMGB1 occurs rapidly during primary necrosis with fully reduced or disul de forms, or nuclear retention and passive release during cell apoptosis secondary necrosis with mainly fully oxidized form (sulfonyl HMGB1). Active secretion happen late stage in pyroptosis with posttranslational modi cation and mainly disul de form [12]. Previous studies showed that celastrol signi cantly suppressed HMGB1/NF-κB pathway to alleviate in ammatory pain and exhibit neuroprotective effect in transient global cerebral I/R, and inhibited HMGB1 expression to decrease myocardial I/R injury [8, 42,43]. Different from previous studies, the peak expression of HMGB1 was not detected in the Model group, and celastrol did not affect the expression of HMGB1. The results of this experiment may be related to the type of cells we selected. Because mammalian neurons are terminally differentiated, postmitotic cells, and the isolated rat cortex primary neurons have little ability of division and proliferation in vitro in the absence of inducer [44]. The neuron cell culture medium supernatant Western blot results proved that primary neurons suffering from OGD injury mainly actively secreted disul de bond form of HMGB1, and the formation of disul de bond could hardly be prevented by celastrol (Fig. 6). Plasma HMGB1 rapidly increases and acts as a pro-in ammatory cytokine to activate microglia, aggravate excite-toxicity induced neuronal death and aggravate brain injury during the acute damaging phase of ischemic insult [45,46]. Early HMGB1 translocation and release occurs mostly in injured neurons and acts as a proin ammatory cytokine by interacting with receptors of RAGE, TLR2 and TLR4 [47]. High levels of HMGB1 in the serum and cerebrospinal uid (CSF) are related with severity of animal ischemic brain damage. In addition, HMGB1 in the serum of lipopolysaccharide (LPS) administered MCAO animals was up-regulated and mainly disul de bond type [48]. Blockade of HMGB1 with antagonists has been veri ed an effective treating style for animal stroke model, including GA [49], HMGB1 A box, anti-HMGB1 monoclonal antibody [50]. Previous studies demonstrated that peripheral disul de HMGB1 produced more obvious pro-nociceptive activity than all-thiol HMGB1 via activating TLR4 other than RAGE [51]. Here, we con rmed that celastrol did not affect the secretion, redox state and expression of HMGB1 both in normal and OGD insult neuron cells. Apart from binding with the cysteines, celastrol also occupied other sites of HMGB1. Considering that only disul de bond form HMGB1 has cytokine property, we focused on the effect of celastrol on disul de bond form HMGB1. 106 cysteine is the main binding site of TLR4 receptor to HMGB1 to play cytokine activity. In our research, celastrol directly bond to HMGB1 A, B box and blocked the binding of HMGB1 B box to its receptors TLR4 and RAGE, resulting in in ammatory activity loss. Celastrol disrupted the TNF-α inducing capacity of HMGB1 and B box in RAW264.7 cells. Therefore, celastrol directly bond to HMGB1 to make it lose in ammatory activity, rather than reducing its secretion or changing its redox activity. In addition, celastrol played antiin ammatory effect in cerebral ischemic injury by targeting HSP70 and NF-κB.
Although celastrol and its numerous derivatives exhibit potential therapeutic effects against various diseases, none of them have been approved for clinical use due to their toxic effects, low solubility and narrow therapeutic dose range [52]. Therefore, how to solve the toxicity of celastrol and improve its e cacy is the next focus direction. In addition, a growing body of evidence supports the idea that in ammation plays different roles in different stages of stroke [53]. HMGB1 shows different activities according to redox modi cations, and may play a more complex role in ischemic stroke to be explored. In addition to cytokine activity, HMGB1 also exerts bene cial effects in axonal regeneration, endothelial activation, angiogenesis, neurovascular repair and remodeling [11,54]. Therefore, it is necessary to consider carefully about promoting or inhibiting HMGB1 in different stages of stroke.

Conclusion
In summary, we performed a proteome-wide investigation of direct cellular protein binding targets of celastrol in primary neuron, and identi ed 120 targets with fairly high credibility through quantitative chemical proteomics approach. The present study demonstrated the neuroprotective effect of celastrol against cerebral I/R injury through targeting HSP70 and NF-κB and disrupting the cytokine activity of disul de bond HMGB1 in rat primary cortical neurons OGD model and adult rats MCAO model, which may provide a potential therapeutic direction for ischemic stroke therapy. By directly binding to the B box, celastrol blocked the binding of TLR4 and RAGE receptors with B box to exhibit anti-in ammatory activity.
To the best of our knowledge, this is the rst study to evaluate the direct binding of celastrol to protein HMGB1. We hope that the data and ndings from the present study can provide useful guidance for the clinical use of celastrol in future.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. SupplementaryTableS1.xlsx