Signaling of High Mobility Group Box 1 (HMGB1) through Toll-like Receptor 4 in Macrophages Requires CD14

INTRODUCTION The toll-like receptors (TLRs) are a major family of pattern recognition receptors (PRRs) that reside in cell membranes, both at the cell surface and in endosomes, that recognize and respond to a variety of bacterial products, called pathogen-associated molecular patterns (PAMPs) (1). Some members of the TLR family, notably TLR2, TLR4 and TLR9, also recognize multiple endogenous damage-associated molecular patterns (DAMPs) such as high mobility group box 1 (HMGB1), heat shock proteins (HSPs), heparan sulfate and mammalian DNA, which are released after cellular stress or injury and can drive sterile inflammatory responses (2–5). Whereas the molecular bases for TLR recognition of many microbial molecules are well characterized, the mechanisms by which TLRs detect DAMPs are less clear. HMGB1 is an archetypal DAMP that was originally identified as a nuclear protein involved in binding DNA and stabilizing DNA interactions with transcription factors to regulate gene transcription (6). Although the cytokinelike properties of HMGB1 were initially described in models of sepsis, HMGB1 has more recently been shown to be a mediator of inflammation in models of sterile injury (7–9) and chronic inflammation (10). Whereas HMGB1 triggers signaling through a wide range of receptors, it is the capacity of HMGB1 to trigger TLR4 signaling that is thought to define its cytokinelike and cytokine-inducing activities (11,12). Recent studies show that only HMGB1 in which cysteine 106 is maintained in the thiol state, and also in whcih cysteines 23 and 45 form a disulfide bond, is capable of activating TLR4 signaling (13–15). It is unknown whether HMGB1 recognition by the TLR4/ myeloid differentiation protein 2 (MD2) complex shares similarities with other prototypical activators of TLR4 signaling. Optimal activation of TLR4 by bacterial lipopolysaccharide (LPS) involves the formation of a signaling complex that includes the coreceptor molecules MD2 and CD14, as well as intracellular signaling molecules including myeloid differentiation primary response protein 88 (MyD88) and TIR domain-containing adapter-inducing interferon-β (TRIF) (16,17). This interaction facilitates an intracellular signaling cascade that culminates in the translocation of the tranSignaling of High Mobility Group Box 1 (HMGB1) through Toll-like Receptor 4 in Macrophages Requires CD14


INTRODUCTION
The toll-like receptors (TLRs) are a major family of pattern recognition receptors (PRRs) that reside in cell membranes, both at the cell surface and in endosomes, that recognize and respond to a variety of bacterial products, called pathogen-associated molecular patterns (PAMPs) (1). Some members of the TLR family, notably TLR2, TLR4 and TLR9, also recognize multiple endogenous damage-associated molecular patterns (DAMPs) such as high mobility group box 1 (HMGB1), heat shock proteins (HSPs), heparan sulfate and mammalian DNA, which are released after cellular stress or injury and can drive sterile inflammatory responses (2)(3)(4)(5). Whereas the molecular bases for TLR recognition of many microbial molecules are well characterized, the mechanisms by which TLRs detect DAMPs are less clear.
HMGB1 is an archetypal DAMP that was originally identified as a nuclear protein involved in binding DNA and stabilizing DNA interactions with transcription factors to regulate gene transcription (6). Although the cytokinelike properties of HMGB1 were initially described in models of sepsis, HMGB1 has more recently been shown to be a mediator of inflammation in models of sterile injury (7)(8)(9) and chronic inflammation (10). Whereas HMGB1 triggers signaling through a wide range of receptors, it is the capacity of HMGB1 to trigger TLR4 signaling that is thought to define its cytokinelike and cytokine-inducing activities (11,12). Recent studies show that only HMGB1 in which cysteine 106 is maintained in the thiol state, and also in whcih cysteines 23 and 45 form a disulfide bond, is capable of activating TLR4 signaling (13)(14)(15). It is unknown whether HMGB1 recognition by the TLR4/ myeloid differentiation protein 2 (MD2) complex shares similarities with other prototypical activators of TLR4 signaling.
Optimal activation of TLR4 by bacterial lipopolysaccharide (LPS) involves the formation of a signaling complex that includes the coreceptor molecules MD2 and CD14, as well as intracellular signaling molecules including myeloid differentiation primary response protein 88 (MyD88) and TIR domain-containing adapter-inducing interferon-β (TRIF) (16,17). This interaction facilitates an intracellular signaling cascade that culminates in the translocation of the tran-

Signaling of High Mobility Group Box 1 (HMGB1) through Toll-like Receptor 4 in Macrophages Requires CD14
scription factor nuclear factor (NF)-κB to the nucleus (16). LPS responsiveness is enhanced by dimerization of TLR4 molecules and mobilization of the signaling complex to a portion of the plasma membrane known as a lipid raft (18). Lipid rafts are defined as glycosphingolipidenriched domains within the cell membrane that form detergent-resistant membrane fractions (19). These fractions have light buoyancy density on sucrose gradients and are rich in both cholesterol and glycosphingolipids (20). Glycosylphosphatidyl inositol-anchored proteins such as CD14 were the first group of proteins reported to be enriched in lipid rafts (20). These lipid rafts, or membrane rafts, are believed to be small, dynamic domains that compartmentalize cellular processes and facilitate cellular signaling (19,20).
Although LPS can bind to CD14, this interaction alone is not sufficient to induce proinflammatory signaling (21,22). CD14 is thought to shuttle LPS to TLR4coupled MD2 (16). This interaction may in turn serve to activate the TLR4 transmembrane signaling apparatus (18). Recruitment of signaling molecules to the lipid rafts may also lead to internalization of both TLR4 and LPS, a process that may be required for an adequate inflammatory response to LPS (23). The importance of this process is demonstrated by the attenuation of LPS-dependent TLR4 activation on disruption of the raft complex (18).
Whereas HMGB1 was demonstrated to display TLR4-dependent activity, it is unclear what role CD14 and lipid rafts play in this process. Here we report that CD14 is required for activation of TLR4dependent signaling by HMGB1. Furthermore, this occurs in association with CD14, TLR4 and MD2 accumulation within lipid rafts. Thus, recognition of cytokinelike HMGB1 by macrophages requires CD14.

Animal Care
All experimental protocols were approved by the Institutional Animal Use and Care Committee of the University of Pittsburgh. Experimental procedures were carried out in accordance with all regulations regarding the care and use of experimental animals as published by the National Institutes of Health. Animals used in all experiments were maintained in laminar flow cages in a specific pathogenfree atmosphere at the University of Pittsburgh (Pittsburgh, PA, USA). Male C57BL/6 (WT) mice, 8-12 wks old, were purchased from The Jackson Laboratory (Bar Harbor, ME, USA). Animals were allowed access to rodent chow and water ad libitum. TLR4 -/mice were generated in our laboratory on a C57BL/6 background and backcrossed at least six times (24). Briefly, mice harboring a floxed allele of TLR4 (TLR4 lox[FRT-neo-FRT]/+ ) were generated initially by Ozgene (Bentley, Bentley DC, Western Australia) and were mated with an EIIa-Cre transgenic mouse (B6.FVBTg [EIIa-cre]C5379Lmgd/J) to generate a null allele of TLR4 (TLR4 del/+ or TLR4 +/-). The transgenic EIIa-Cre mouse expresses Cre recombinase in almost all tissues, including those of preimplantation embryos, and has been used previously to mediate recombination between loxP sites in germ cells (25). TLR4 +/progeny were confirmed by polymerase chain reaction genotyping by using multiple primer pairs and interbred to generate TLR4 -/mice, which were unresponsive to LPS (data not shown). TLR4 -/mice showed no differences in overall health or viability compared with WT littermates. CD14 -/mice (a gift from Mason Freeman, Massachusetts General Hospital, Boston, MA, USA) also on a C57BL/6 background were back-crossed at least six times and bred in our facility (26). MyD88-competent (MyD88 +/+ ) and MyD88 -/mice on a C57BL/6 background were provided by Ruslan Medzhitov (Howard Hughes Medical Institute, Yale University, New Haven, CT, USA). TRIF-deficient (Trif LPS2/LPS2 , synonymous TRIF -/-) mice on a C57BL/6 background were provided by Bruce Beutler (Scripps Institute, La Jolla, CA, USA). MyD88 -/and TRIF -/strains were backcrossed at least 10 times into a C57BL/6 background and bred in our facility (27).
MyD88-deficient mice and their MyD88 +/+ WT controls were provided water supplemented with trimethoprim (4 mg/mL) and sulfamethoxazole (40 mg/mL) until 8 wks of age. Antibiotics were stopped for at least 2 wks before mice were used in experiments.

Reagents
All tissue culture plates and flasks were purchased from BD Biosciences (San Diego, CA, USA). Dulbecco's modified Eagle medium (DMEM) and RPMI 1640 and fetal bovine serum (FBS) was from Lonza BioWhittaker (Walkersville, MD, USA). OPTI-MEM was purchased from Invitrogen/ Life Technologies (Carlsbad, CA, USA). Ultrapure LPS (Escherichia coli 0111:B4) was purchased from List Biological Laboratories (Vandell Way, CA, USA). Polymixin B was from Sigma-Aldrich (St. Louis, MO, USA). Rabbit anti-mouse TLR4 was acquired from Santa Cruz Biotechnology (H-80; Santa Cruz, CA, USA). Mouse anti-mouse flotillin-1 and rat anti-mouse CD14 were purchased from BD Biosciences. Rabbit anti-mouse MD2 was obtained from Abcam (Cambridge, MA, USA). Pam3CSK4 (TLR2/1 ligand) and polyI:C (TLR3 ligand) were from InvivoGen (San Diego, CA, USA). Pam3CSK4 and poly I:C were tested for LPS contamination by Limulus amoebocyte lysate assay (Associates of Cape Cod, East Falmouth, MA, USA; or BioWhittaker), which determined that there was no detectable LPS (minimum sensitivity of 0.125 EU/mL). Concentrations of all ligands and inhibitors were used, as indicated in the corresponding figure legends.

Recombinant Human HMGB1
Recombinant rat HMGB1, expressed in E. coli and purified to homogeneity as previously described (7,28), was a gift from Kevin Tracey (The Feinstein Institute for Medical Research, Manhasset, NY, USA). HMGB1 was tested for LPS contamination by Limulus amoebocyte lysate assay, which determined that there was no detectable LPS (minimum sensitivity of 0.125 EU/mL).

Cell Culture and Transfection
RAW264.7 cells and HEK293 cells (American Type Culture Collection, Manassas, VA, USA) were maintained in DMEM containing 10% FBS, 2 mmol/L L-glutamine and 100 U/mL penicillin and streptomycin. Cultures were maintained at 37ºC in a humidified atmosphere of 5% CO 2 and 95% air. HEK293 cells were stably transfected with the following: (a) empty expression vector; (b) TLR4, (c) TLR4 with MD2 or (d) TLR4, MD2 and CD14. The vectors for TLR4, MD2 and CD14 were a gift from Jeffrey L Platt (University of Michigan, Ann Arbor, MI, USA), as previously described (29). Subsequent cell lines were maintained in DMEM supplemented with 10% fetal bovine serum and the appropriate selection antibiotics. Control cell lines were transfected with empty expression vectors and maintained in selection conditions as described above. Adenovirus vectors for CD14 were constructed using plasmid vectors generously provided by Regine Landmann (University Hospital, Basel, Switzerland). Recombinant adenoviruses were generated by cotransfection of Ψ5 and pAdlox-CD14 linearized with SwaI into the adenovirus packaging cell line CRE8, which expressed Cre recombinase. Recombinant adenoviruses were propagated in CRE8 cells, purified by a ViraBind Adenovirus Purification kit (Cell Biolabs, San Diego, CA, USA) according to manufacturer directions. Subsequent dialysis proceeded according to standard protocols, with the purified product being stored at -80°C. Adenoviral vectors were infected into RAW264.7 cells at a multiplicity of infection of 50. Control adenoviral vectors (AdΨ5) were also infected into similar groups of cells obtained at the same time. Infection was allowed to occur for 4 h before cell treatment. Cells were collected for each assay as detailed above.

Isolation and Culture of Murine Peritoneal Macrophages
Peritoneal macrophages were isolated as previously described (30). Thioglycollateelicited peritoneal macrophages were collected 4 d after intraperitoneal injection of 1.5 mL 4% Brewer thioglycollate medium (Sigma-Aldrich) by peritoneal washout with RPMI 1640. Peritoneal washout fluid was subsequently strained, pelleted and treated with ACK lysing buffer (Lonza Walkersville, Walkersville, MD, USA) to remove any red blood cells. Extracted cells were counted and resuspended in RPMI 1640 containing 10% FBS, 2 mmol/L L-glutamine and 100 U/mL penicillin and streptomycin. Macrophages were plated at 7 × 10 5 cells/ well in a volume of 100 µL/well into 96-well cell culture plates, and cells were allowed to recover overnight at 37ºC in a humidified atmosphere of 5% CO 2 and 95% air. Cell media were changed the following day, and cells were treated with a total volume of 100 µL/well of medium with indicated experimental treatment. At designed time points after treatment, supernatants were collected and stored at -80ºC until analyzed for cytokines and chemokines. All cell culture experiments were performed in triplicate.

NF-κB Luciferase Reporter Assays
Activation of NF-κB was measured by using an NF-κB dual luciferase reporter assay as previously described (29). Briefly, HEK293 cell lines stably expressing TLR4, TLR4 and MD2, or TLR4, MD2 and CD14, were seeded into 24-well tissue culture plates (3 × 10 4 cells/ well) in 0.1 mL DMEM containing 10% fetal bovine serum with L-glutamine. Cells were allowed to adhere at 37ºC overnight and were then transfected with 0.05 µg/ well of pTK Renilla-luciferase (Promega, Madison, WI, USA) and 0.025 µg/well of NF-κB firefly luciferase (a gift from Carlos Paya from Institute Pasteur, Paris, France) using Lipofectamine™-2000 (Invitrogen/ Life Technologies). After transfection, cells were washed once with phosphatebuffered saline (PBS) and cultured for 24 h at 37ºC overnight in 100 µL DMEM containing 10% fetal bovine serum and Lglutamine. After various treatments, media were removed, and cells were washed once with PBS. The cells were then lysed by using 100 µL passive lysis buffer (Promega) in a dual-luciferase re-porter assay system from Promega with rocking at room temperature for 15 min. Renilla and firefly luciferase were then assayed simultaneously by using a Synergy Mx Monochromator-Based Multi-Mode Microplate Reader (Biotek, Winooski, VT, USA). Activation of NF-κB is reported as a ratio of firefly luciferase to the constitutively expressed Renilla luciferase internal control and is the mean of triplicate wells.

Sucrose Gradient Fractionation
RAW264.7 cells were grown on 10-cm tissue culture plates and stimulated with LPS or HMGB1 for the indicated times. Cells were then washed twice with icecold PBS and lysed in ice-cold MBS buffer [25 mmol/L 2-(N-morpholino)ethanesulfonic acid solution (MES) (pH 6.5), 150 mmol/L NaCl, 1% Triton X-100, 1 mmol/L Na 3 VO 4 and protease inhibitors] and incubated on ice for 30 min. Lysates were separated by density centrifugation in a discontinuous sucrose gradient (40%, 36% and 5% sucrose in MBS). Samples were spun at 250,000g in a Beckman SW41 swinging bucket rotor for 18 h at 4°C, by using slow acceleration and no brake. Fractions were removed and stored at -80°C until needed. Samples were then separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) to identify proteins within each fraction. Lipid raft laden fractions were identified via immunoblot for flotillin-1 via Western blot analysis.

Western Blot
Samples were separated by 8-12% SDS-PAGE and transferred onto a nitrocellulose membrane. The membrane was blocked for 1 h in Tris-buffered saline (TBS)-Tween-20 (0.1%) with 5% milk, followed by immunostaining with optimized dilutions of primary antibody in 1% milk in TBS-Tween-20 (0.1%) overnight at 4°C. Membranes were washed three times for 10 min in TBS-Tween-20 (0.1%), and antibody binding was detected with horseradish peroxidase-conjugated secondary antibodies in a standard enhanced chemiluminescence reaction according to manu-facturer's instructions (Pierce) and exposed to Kodak X-Omat film (Kodak, Rochester, NY, USA).

Imaging Cytometry
Measurements of pERK (Abcam), pp38 and pJNK (Cell Signaling, Danvers, MA) were performed in WT, TLR4 -/or CD14 -/peritoneal exudate cells using Cellomics Arrayscan VTi (ThermoFisher Scientific, Pittsburgh, PA, USA) as previously described (31,32). Cells were plated on 384-well plates and stimulated with either HMGB1 or LPS for time points up to 4 h; then cells were permeabilized, stained with appropriate antibodies and imaged.

Biolayer Interferometry
Affinity of target proteins to HMGB1 protein was measured by using a FortéBio Octet QK platform and default settings for the sample stage orbital rate (1,000 rpm) at 30°C. Amine reactive biosensors were equilibrated in a 100 mmol/L MES. Biosensors were activated with a 1:50 0.4 mol/L 1-ethyl-3-(3-dimethylamino) propyl carbodiimide, hydrochloride (EDC)/ N-hydroxysuccinimide (NHS):MES solution. The amine biosensors were loaded with a solution of 25 µg/mL HMGB1 and then introduced into solutions containing nanomolar concentrations of sTLR4, sCD14 or MD2. Recombinant antibodies to HMGB1 (RD, N-term, 5H and GLIOA5) were also assessed for HMGB1 affinity. Spastic paraplegia 4 was used as a positive binding control. The binding reaction was quenched by 1 mmol/L glycine, and PBS was used to equilibrate the protein. Dissociation was assessed by washing biosensors in PBS, and affinity to HMGB1 was calculated using FortéBio software (Menlo Park, CA, USA), controlling for the buffer only sample.

Statistical Analysis
Data are presented as mean ± standard error of mean (SEM). Experimental results are analyzed for their significance by the Student t test by using SigmaStat (SPSS, Chicago, IL, USA). Significance was established at the 95% confidence level (P < 0.05).
All supplementary materials are available online at www.molmed.org.
TLR4-mediated signaling of HMGB1 is unknown (13). We first confirmed that nonreduced, nonoxidized recombinant HMGB1 would stimulate TNF-α release in RAW264.7 murine macrophages in a TLR4-dependent manner. To do this, RAW264.7 cells were stimulated for 18 h with PBS (control-no stimulation) or 0.1, 0.5 or 1 ng/mL LPS; and TNF-α levels in supernatants were measured by ELISA. Even the lowest concentration of LPS used was able to stimulate RAW264.7 to produce TNF-α after 18 h ( Figure 1A). TNF-α induced by 1 ng/mL LPS was completely abrogated by concurrent treatment with 10 µg/mL polymyxin B (PMB). Similarly, HMGB1 at concentrations from 0.5 to 10 µg/mL induced significantly increased levels of TNF-α in supernatants after 18 h in a concentration-dependent manner. The concentrations of TNF-α were not significantly decreased with concurrent treatment with PMB, confirming that LPS contamination did not account for the TNF-α release induced by the HMGB1 used for these studies (see Figure 1A). To confirm that the induction of TNF-α release was TLR4 dependent, we isolated peritoneal macrophages from WT (C57BL/6 mice) and TLR4 -/mice and stimulated the cells ex vivo with either LPS (1 ng/mL) or HMGB1 (2, 5 or 10 µg/mL) for 18 h. As expected, LPS induced TNF-α in peritoneal macrophages from WT mice but not in TLR4 -/mice ( Figure 1B). HMGB1 also stimulated TNF-α release in peritoneal macrophages from WT mice, even at the lowest concentration of 2 µg/mL (see Figure 1B). Similarly to LPS, HMGB1 did not induce TNF-α release in TLR4 -/peritoneal macrophages, confirming that HMGB1induced TNF-α increase occurs via TLR4dependent signaling (see Figure 1B).

HMGB1-Dependent TLR4 Activation Requires CD14 In Vitro
To assess the role of CD14 in TLR4dependent recognition of HMGB1, we measured NF-κB activation by reporter assay in HEK293 cells transfected with (a) TLR4, (b) TLR4 with MD2 or (c) TLR4, MD2 and CD14 and stimulated the cells with LPS (100 ng/mL), HMGB1 (5 and 10 µg/mL) or TNF-α (10 ng/mL as a positive control) for 18 h. As expected, all cells responded to the positive control, TNF-α, to exhibit significant NF-κB activation. Only HEK293 cells transfected with the whole TLR4/MD2/CD14 complex significantly induced NF-κB activation in response to LPS compared with controls ( Figure 2A). Similarly, HMGB1 only significantly activated NF-κB in HEK293 cells transfected with TLR4/MD2/CD14, at concentrations of 5 and 10 µg/mL (see Figure 2A). In HEK293 cells transfected with TLR4 alone, HMGB1 as well as LPS failed to induce significant NF-κB activation, even in the presence of MD2. Moreover, to verify the importance of CD14 in signaling and to determine if TLR4 and CD14 alone were sufficient to allow downstream signaling, the HEK/TLR4 and HEK/TLR4/MD2 cell lines were exposed to HMGB1 with or without transfection with an adenoviral vector carrying a CD14 cDNA (AdCD14). The cell line expressing TLR4 and MD2 responded to HMGB1 if transfected with AdCD14 (Figure 2B). These findings demonstrate that CD14 enhances HMGB1-dependent TLR4 activation in HEK cells.
To determine if inflammatory mediators induced by HMGB1 in macrophages other than TNF-α were also TLR4 and CD14 dependent, we also measured macrophage production of three chemokines: MCP-1 (CCL2), IP-10 and MIP-1α (CCL3). As shown in Figure 3B, both LPS and HMGB1 induced significant increases in the supernatant levels of these three chemokines at 18 h from WT macrophages. However, no increase was observed in either TLR4 -/or CD14 -/macrophages (see Figure 3B).

HMGB1 Stimulates Clustering of the TLR4/MD2/CD14 Complex in the Lipid Raft
Having demonstrated that CD14 is essential in HMGB1/TLR4 signaling, we next investigated the role of CD14-enriched lipid rafts in the recognition of TLR4/MD2. To determine whether TLR4 was present in the lipid raft of RAW264.7 cells at baseline, untreated cell lysates were subjected to sucrose gradient fractionation. After fractionation, lysates were drawn off in 12 separate 1-mL aliquots and protein was subsequently precipitated. An equal volume of each fraction was separated by SDS-PAGE, and Western blot was performed by using an antibody to flotillin-1 to identify raft-enriched fractions. The membrane was then probed by using an antibody to TLR4. In resting cells, no TLR4 was de- tected in the raft fractions ( Figure 4A). Similarly, RAW264.7 cells were then stimulated with 1 µg/mL HMGB1 for 15-90 min, and the cell lysates were subsequently subjected to sucrose gradient fractionation. Maximal clustering of TLR4 in the lipid raft was seen after 60 min of stimulation with HMGB1, with clustering occurring as early as 15 min after treatment (data not shown). Colocalization of TLR4 and the raft protein GM1 was also assessed by confocal microscopy at multiple time points (Figure 4B). Minimal colocalization was observed at baseline; however, increased colocalization was observed at both 15 and 60 min, which diminished by 90 min of stimulation with HMGB1. These findings suggest that TLR4 migrates into the lipid raft transiently on stimulation with HMGB1, as shown by the close association with the raft protein GM1.
The adaptor protein MD2 closely associates with TLR4 and enhances TLR4dependent signaling after stimulation with LPS or HMGB1 (13,16,22). To determine if HMGB1 similarly causes the migration of MD2 to the lipid raft, RAW264.7 cells were subjected to sucrose gradient fractionation both before and after stimulation with HMGB1. As expected, MD2 was limited to nonraft fractions at baseline ( Figure 5). After stimulation with HMGB1, however, there was migration of MD2 to the raft complex ( Figure 5). CD14 is a glycosylphosphatidyl inositol-linked protein found in relative high abundance within the lipid raft (18). We confirmed this observation in unstimulated RAW264.7 cells. After stimulation with 1 µg/mL HMGB1, there may be a slight increase in CD14 protein within the raft fraction ( Figure 5).

HMGB1-Induced Mitogen-Activated Protein Kinase Activation Depends on TLR4 and CD14
To determine if signaling pathways downstream of MyD88 and TRIF were ac- tivated in a CD14-dependent manner, we measured the levels of nuclear phosphorylated mitogen-activated protein (MAP) kinases (pp38, pERK and pJNK) in HMGB1-or LPS-treated peritoneal macrophages by imaging cytometry. Peritoneal macrophages from WT, TLR4 -/and CD14 -/mice were treated with HMGB1 (5 µg/mL) or LPS (10 ng/mL) in Opti-MEM for 30 min, 1 h and 4 h. After HMGB1 or LPS treatment, activation of pp38 ( Figure 7A), pERK ( Figure 7B) and pJNK ( Figure 7C) were markedly increased compared with nontreated macrophages in WT by 30 min. This increase persisted to 4 h for pERK and pJNK and 2 h for pp38. The activation of all three MAP kinases was markedly diminished in both TLR4 -/and CD14 -/macrophages in HMGB1-and LPS-treated cells. Taken together, these studies of macrophage activation and signaling indicate that the cytokinelike activation of macrophages by HMGB1 requires CD14.

DISCUSSION
The capacity of HMGB1 with cysteine 106 in the thiol state to activate TLR4 is now well established (9,13). TLR4/MD2 recognition of LPS is facilitated by CD14, which is thought to transfer LPS to the TLR4/MD2 complex (22,34). The goal of the study reported here was to determine if TLR4-dependent activation of macrophages by HMGB1 also requires CD14. We provide strong evidence in both cell lines, as well as primary macrophages, that CD14 is necessary for the recognition of HMGB1 by TLR4. Furthermore, our findings suggest that CD14 could play a role in the formation of a signaling complex within lipid rafts. We also extend previous findings to show that HMGB1-TLR4 signaling involves not only MyD88 but also TRIF and extends to MCP-1, IP-10 and MIP-1α in addition to TNF-α.
The explanation for inconsistencies in the capacity of HMGB1 to trigger specific cellular receptors was recently explained by a series of reports that defined the structural requirements for cytokinelike and chemokinelike activities of HMGB1. Yang et al. (9,13) demonstrated that cysteine 106 in the B-box domains of HMGB1 must be in the nonreduced and nonoxidized state to trigger signaling through TLR4 by binding to MD2. Recently, these observations were carried even further to show that not only must cysteine 106 be in the thiol state, but the other two cysteines at positions 23 and 45 must form a disulfide bond (35,36). If all of these cysteines are in the thiol state, HMGB1 stimulates the production of and interacts directly with CXCL12 to activate the chemokine receptor CXCR4 (15). For all of our experiments, we used HMGB1 in the C23-C45 disulfide, C106thiol conformation (a gift from Kevin Tracey).
Whether any of these in vivo conditions also require CD14 recognition of HMGB1 is uncertain. We also recently found that CD14 -/mice are protected from I/Rinduced liver damage to a similar level to that observed in TLR4 -/mice.
Most of our knowledge of TLR4 structure and function comes from experiments involving its prototypic ligand LPS. We know that LPS does not interact directly with TLR4, but instead binds first to CD14, which enables subsequent binding of LPS with MD2 (16,50). Because neither CD14 nor MD2 contain transmembrane domains, signaling depends on activation of TLR4 when LPS triggers a conformational change in the TLR/MD2 complex (16,34). Members of this signaling apparatus accumulate within the lipid raft on stimulation with LPS, facilitating efficient signal transduction and promoting a subsequent proinflammatory response (7,51). In addition, multiple intracellular signaling proteins may also migrate into the raft scaffolding following stimulation with LPS (16).
Here we show that recombinant HMGB1 also requires CD14 to trigger signaling through TLR4. We demonstrated that on stimulation with HMGB1, both TLR4 and MD2 migrate to the lipid raft and closely associate with the raft protein GM1. CD14 was found within the lipid raft domain before stimulation to a large degree, but increased after HMGB1 stimulation. These findings suggest a requirement for both MD2 and CD14 within the raft and suggest that some interaction between these coreceptors may take place on the cell surface on stimulation with HMGB1. Because CD14 resides largely within the lipid rafts at baseline, it seems reasonable to suggest that this interaction may take place within the raft. Furthermore, the requirement for CD14 may be to facilitate the association of other proteins within the lipid raft complex (something worthy of additional study). In addition to showing that HMGB1-induced TNF-α production is TLR4 and CD14 dependent, we show that other mediators known to be produced after TLR4 ligation are also produced in a TLR4-and CD14-dependent manner by HMGB1 stimulation. HMGB1 induced robust production of MCP-1, IP-10 and MIP-1α in a TLR4-and CD14-dependent manner. Furthermore, we also demonstrated that HMGB1-induced TNF-α is TRIF and MyD88 dependent, consistent with what is known regarding LPS downstream signaling. Therefore, it is not surprising that downstream-signaling events including MAP kinases in response to HMGB1 were markedly diminished in the absence of TLR4 and CD14.

CONCLUSION
We demonstrate that CD14 is critically important in efficient HMGB1-dependent TLR4 activation. A detailed mechanistic understanding of HMGB1-TLR4 interaction and the downstream signaling pathways in the pathophysiology of inflammation could have significant implications for designing therapeutics to limit HMGB1mediated inflammation.