Main

During microbial infection or tissue damage, DNA and RNA potently activate the innate and subsequent adaptive immune responses1,2. In mammals, TLR3, TLR7 and TLR9 recognize, respectively, double-stranded RNA, single-stranded and short double-stranded RNAs, and hypomethylated DNA1,2,3, whereas the RIG-I-like receptors (RLRs), namely retinoic acid-inducible gene I (RIG-I) and melanoma differentiation-associated gene-5 (MDA5) are best known as RNA-sensing receptors in the cytosol4,5. In addition, cytosolic DNA-sensing receptors, which include DNA-dependent activator of IRFs (DAI) and absent in melanoma 2 (AIM2), also trigger the innate and adaptive immune systems6,7,8. It has recently been shown that RLRs also participate in the cytosolic DNA-sensing system3,9,10,11. The hallmark of innate immune responses activated by these receptors is the induction of type-I interferons (IFNs), proinflammatory cytokines and chemokines1, except that by AIM2, which is a critical component of the inflammasome that typically promotes the secretion of interleukin (IL)-1β (ref. 7). So far, no universal or shared mechanism of action for these nucleic-acid-receptor classes is presumed to operate in their activation.

To gain new insights into the nature of the cytosolic DNA-sensing systems, we performed an unbiased biochemical screen to identify proteins involved in DNA recognition on the basis of their direct binding to a B-form DNA, poly(dA-dT)·poly(dT-dA) (B-DNA; ref. 12); the most prominently recovered proteins were HMGB1, HMGB2 and HMGB3 (Supplementary Fig. 1a). HMGB proteins are highly expressed in the nucleus, where they regulate chromatin structure and transcription, but they are also present in the cytosol and in extracellular fluids13,14. HMGB1 was recently shown to participate in the activation of several immune receptors, including TLRs13,15,16,17,18.

To examine the direct interaction of HMGBs with nucleic acids, recombinant HMGBs were purified and assayed in vitro for their binding to biotin-conjugated B-DNA. Both HMGB1 and HMGB2 were precipitated by immobilized B-DNA, which was inhibited in a dose-dependent manner by non-conjugated B-DNA (Supplementary Fig. 1b). HMGB1–B-DNA binding was, however, quite inefficiently inhibited by the addition of calf thymus-derived or bacteria-derived DNA, each of which only weakly activates the cytosolic-sensing pathways8,12 (Supplementary Fig. 1b). The addition of double-stranded RNA (polyinosinic-polycytidylic acid; poly(I:C)) or single-stranded RNA (poly(U)), but not imiquimod (R837), a non-nucleic-acid agonist for TLR7, efficiently inhibited HMGB1–B-DNA binding (Supplementary Fig. 1b). In contrast, HMGB2–B-DNA binding was not affected by either poly(I:C) or poly(U) (Supplementary Fig. 1b). Thus, HMGB1 binds to these immunogenic RNAs but HMGB2 does not. Furthermore, a TLR9 agonist CpG-B oligodeoxynucleotide (ODN; refs 19, 20) and antagonist base-free phosphorothioate deoxyribose homopolymer (PS; ref. 21) were the most potent in inhibiting HMGB1–B-DNA binding, which is consistent with a previous report16 (Supplementary Fig. 1b). In contrast, a weak TLR9 agonist base-free natural phosphodiester deoxyribose homopolymer (PD; ref. 21) showed little, if any, inhibition (Supplementary Fig. 1b). Binding of HMGB1 to biotin-conjugated poly(U), a TLR7 agonist, was also strongly inhibited by the addition of free CpG-B ODN and PS, but not by R837 (Supplementary Fig. 1c; see below). Finally, we also found that HMGB3, expressed in certain cell types22,23, binds both DNA and RNA (Supplementary Fig. 1d). These results indicate a correlation between the affinity of a type of nucleic acid to HMGB and its immunogenicity.

To study the contribution of HMGBs to the nucleic-acid-mediated activation of innate immune responses, we first examined cells derived from gene-targeted mice for HMGB1 or HMBG2. Mouse embryonic fibroblasts (MEFs) from HMGB1-deficient mice (Hmgb1-/-) showed a significant defect in messenger RNA induction for type-I IFNs, IL-6 and RANTES in response to cytosolically delivered B-DNA or poly(I:C) at all doses examined, whereas the response to lipopolysaccharide (LPS) remained unaffected (Fig. 1a and Supplementary Fig. 2a, b). Similar results were obtained with conventional dendritic cells (cDCs) differentiated by culturing Hmgb1-/- fetal liver with granulocyte–macrophage colony-stimulating factor (GM-CSF) (Supplementary Fig. 2c). Cytokine gene induction by Hmgb2-/- MEFs was defective when stimulated by B-DNA and not poly(I:C), which is consistent with the interaction of HMGB2 with DNA only (Fig. 1b and Supplementary Figs 1b and 2d). Accordingly, in Hmgb1-/- MEFs expressing a small interfering RNA (siRNA) that specifically targets HMGB2, type-I IFN gene induction on stimulation with B-DNA, but not with poly(I:C), was decreased further; the residual response is presumed to have been due to HMGB3 (Supplementary Fig. 2e, f).

Figure 1: The contribution of HMGBs to cytosolic DNA-mediated or RNA-mediated activation of innate immune responses.
figure 1

a, Hypoinduction of IFN-β in Hmgb1-/- cells on cytosolic delivery of DNA or RNA. Hmgb1+/+ or Hmgb1-/- MEFs were stimulated with B-DNA or poly(I:C) for 6 h, or stimulated with LPS (200 ng ml-1) for 2 h. The induction levels of Ifnb1 mRNA were determined by quantitative real-time RT–PCR (qRT–PCR). Asterisk, P < 0.01 compared with Hmgb1+/+ MEFs. Data in all panels are presented as means and s.d. (n = 3). ND, not detected. b, Hypoinduction of IFN-β in Hmgb2-/- MEFs on cytosolic delivery of DNA but not RNA. Asterisk, P < 0.001 compared with Hmgb2+/+ MEFs. c, Impaired innate immune responses to cytosolic nucleic acids in MEFs deficient in all HMGBs. MEFs transduced with retrovirus expressing siRNA targeting all HMGBs (HMGB-si) or control siRNA (Ctrl-si) were lipofected with B-DNA or poly(I:C), and then mRNA expression levels of the indicated genes were measured by qRT–PCR. Asterisk, P < 0.01 compared with Ctrl-si-MEFs. d, Impaired induction of IFN-β by cytosolic delivery of nucleic acids obtained from various sources at 6 h, but not by stimulation with LPS (200 ng ml-1) for 2 h in MEFs deficient in all HMGBs. Asterisk, P < 0.01 compared with Ctrl-si-MEFs.

PowerPoint slide

Full size image

We then examined MEFs in which the expression of all three HMGB proteins was suppressed by use of pan-HMGB siRNA vector (Supplementary Fig. 2g, h). As shown in Fig. 1c, the mRNA induction for type-I IFNs and other cytokines on stimulation with B-DNA or poly(I:C) was inhibited more strongly in MEFs expressing the pan-HMGB siRNA (HMGB-si-MEFs) than in the Hmgb1-/- and Hmgb2-/- cells tested above. Similar results were obtained on stimulation of HMGB-si-MEFs with IFN-stimulatory DNA (ISD; refs 2, 24), viral or bacterial DNA, and single-stranded RNA bearing 5′-triphosphates25,26 (Fig. 1d and Supplementary Fig. 2i). In contrast, the induction of cytokine genes remained unaffected in HMGB-si-MEFs stimulated by LPS, indicating that the inhibition is selective to stimulations with nucleic acids (Fig. 1d and Supplementary Fig. 2i–k). Moreover, mRNA induction of the genes activated by a variety of cytokines occurred normally in these cells, indicating that gene transcription is not generally affected by the absence of HMGBs (Supplementary Fig. 2l, m). Essentially the same observations were also made in RAW264.7 macrophage and NIH/3T3 fibroblast cell lines (Supplementary Fig. 2n–q). In macrophages, cytosolic DNA induces the formation of the inflammasome by the activation of AIM2, which triggers the secretion of IL-1β (ref. 7). The involvement of HMGBs in the DNA–AIM2–inflammasome pathway was underscored by the observation that B-DNA-induced secretion of IL-1β is significantly impaired in Hmgb1-/- fetal liver-derived macrophages and RAW264.7 cells expressing the pan-HMGB siRNA (Supplementary Fig. 3a–c). These results suggest that all three HMGBs are required for the full-blown activation of innate immune responses by cytosolic nucleic acids.

We next examined whether the signalling pathways activated through the cytosolic receptors are affected by the absence of HMGBs. The activation of the IRF3 and NF-κB transcription factors was measured in HMGB-si-MEFs or MEFs expressing Renilla luciferase-siRNA (Ctrl-si-MEFs) stimulated by B-DNA or poly(I:C). As shown in Fig. 2a, the formation of dimerized IRF3, a hallmark of IRF3 activation, was strongly suppressed in HMGB-si-MEFs but robust in Ctrl-si-MEFs on stimulation by B-DNA or poly(I:C). Similarly, the activation of extracellular signal-regulated kinase (ERK) (data not shown) and NF-κB was also suppressed in HMGB-si-MEFs (Fig. 2b). Thus, these results indicate that in the context of the nucleic-acid-mediated activation of innate immune responses, HMGBs function in the cytosolic receptor–IRF3/NF-κB signalling pathways.

Figure 2: A requirement for HMGBs in cytosolic nucleic-acid-receptor-mediated activation of signalling pathways and in antiviral innate immune responses.
figure 2

a, b, Activation status of IRF3 (a) and NF-κB (b). MEFs expressing siRNA targeting all HMGBs (HMGB-si) or control siRNA (Ctrl-si) were lipofected with B-DNA or poly(I:C). Dimerization of IRF3 was assessed by native PAGE followed by immunoblot analysis. Activation of NF-κB was analysed by electrophoretic mobility-shift assay. c, Induction of type-I IFNs by virus infection. MEFs expressing siRNA targeting all HMGBs (HMGB-si) or control siRNA (Ctrl-si) were infected with VSV or HSV-1. Ifnb1, Ifna1 and Ifna4 mRNA expression levels were monitored by qRT–PCR. Data in c are presented as means and s.d. (n = 3). Asterisk, P < 0.01; double asterisk, P < 0.05 compared with cells expressing Ctrl-si. ND, not detected.

PowerPoint slide

Full size image

Because cytosolic nucleic-acid-sensing systems are crucial to antiviral immunity1,4, we next examined the contribution of HMGBs to the virus-induced activation of innate immune responses. The gene induction of type-I IFNs and replication of vesicular stomatitis virus (VSV) or herpes simplex virus type 1 (HSV-1) in infected HMGB-si-MEFs and Ctrl-si-MEFs were measured. Ifnb1 mRNA induced by these viruses was much lower in HMGB-si-MEFs than in Ctrl-si-MEFs (Fig. 2c). However, the suppression of Ifna (Ifna1 and Ifna4) mRNA induction was more resistant to HMGB downregulation than Ifnb1 mRNA (Fig. 2c), which is presumably due to the operation of one or more mechanisms that do not involve the nucleic acid–HMGB pathway27. Accordingly, virus replication was increased in HMGB-si-MEFs (Supplementary Fig. 4a). Similar observations were made for RAW264.7 cells expressing the pan-HMGB siRNA (Supplementary Fig. 4b, c). These data, together with those for virus-derived ‘naked’ DNAs (Fig. 1d), support the hypothesis that HMGB sensing of virus-derived nucleic acids is critical for the effective avocation of the antiviral innate immune responses.

Given that HMGBs (particularly HMGB1) bind to all TLR agonistic nucleic acids (Supplementary Fig. 1b–d), we next examined the roles of HMGBs in the activation of TLRs. A requirement for HMGB1 in TLR9 signalling in cooperation with the multivalent receptor RAGE (receptor for advanced glycation end-products) was reported previously16,18. As shown in Fig. 3a, the mRNA induction of proinflammatory cytokines was impaired, albeit partly, in Hmgb1-/- cDCs on stimulation with TLR3 (poly(I:C)) or TLR9 (CpG-B ODN), whereas mRNA levels were unaffected on stimulation with TLR4 (LPS). The residual induction is likely to be mediated by other HMGBs. Indeed, when the pan-HMGB siRNA was expressed in RAW264.7 cells, mRNA induction by poly(I:C) or CpG-B ODN was more severely inhibited (Supplementary Fig. 5). We also cultured Hmgb1-/- fetal liver with Flt3-ligand to enrich for plasmacytoid dendritic cells (pDCs), and then stimulated the cells with ligands for TLR7 and TLR9 to drive the pDC-mediated high-level production of type-I IFNs. Ifnb1 and Ifna4 mRNA induction by TLR9 agonist CpG-A (D19) ODN20 was inhibited in Hmgb1-/- pDCs and CpG-A ODN-stimulated RAW264.7 cells expressing the pan-HMGB siRNA (Fig. 3b and Supplementary Fig. 5). In addition, Ifnb1 mRNA expression was severely inhibited when Hmgb1-/- pDCs were stimulated by the TLR7 agonist poly(U), yet when stimulated by R837, which does not bind to HMGBs (Supplementary Fig. 1b), the expression levels were the same as those in wild-type pDCs (Fig. 3b). The deficiency in HMGB1 had a more profound effect on TLR7 stimulation by poly(U) than on TLR9 stimulation by CpG-A ODN (Fig. 3b), possibly because poly(U) is not bound by HMGB2 (Supplementary Fig. 1b). We speculate that, like TLR9 signalling, TLR3 and TLR7 signalling may also involve RAGE-bound HMGB1 (ref. 18). Thus, these results together underscore the role of HMGBs in the activation of all nucleic-acid-sensing TLRs.

Figure 3: A requirement for HMGB1 in the nucleic-acid-mediated activation of TLRs.
figure 3

Hmgb1+/+ or Hmgb1-/- cDCs (a) and pDCs (b) were stimulated by the indicated TLR ligands, and then mRNA expression levels for various cytokines were measured by qRT–PCR. Data in all panels are presented as means and s.d. (n = 3). Asterisk, P < 0.01 compared with wild-type cells. ND, not detected.

PowerPoint slide

Full size image

Because CpG-B ODN and the antagonist PS bind to HMGBs with extraordinarily high affinities (Supplementary Fig. 1b, c), we tested whether these compounds interfere with innate immune responses evoked by immunogenic nucleic acids of lower affinities; we reasoned that a functional consequence of the pretreatment of HMGBs with these high-binding-affinity nucleotide analogues would ‘mask’ HMGBs, thereby selectively interfering with the activation of innate immune responses by other nucleic acids. Indeed, wild-type MEFs were hyporesponsive to stimulation with cytosolic B-DNA or poly(I:C) when pretreated with CpG-B ODN, which itself does not evoke a detectable cytokine response in this cell type, whereas their response to LPS remained normal (Fig. 4a). As expected, the suppressive effect of CpG-B ODN was weaker if added after stimulation with B-DNA (Supplementary Fig. 6a). Furthermore, even in TLR9-deficient pDCs and cDCs, a profound hyporesponse was observed on pretreatment with CpG-B ODN or PS followed by stimulation by TLR3 or TLR7 with their cognate nucleic acid ligands (Fig. 4b). In contrast, activation of TLR7 by R837 was not affected by pretreatment with CpG-B ODN or PS (Fig. 4b and Supplementary Fig. 6b). Consistent with this, when B-DNA was used as an adjuvant the induction of ovalbumin-specific CD8 T-cell responses was severely impaired by the pretreatment of TLR9-deficient mice with CpG-B (M. Miyajima and T.T., unpublished observation). These results therefore lend further support to the universal role of HMGBs in the nucleic-acid-mediated innate immune responses, although strictly the possibility is not ruled out that the CpG-B ODN-mediated or PS-mediated interference with nucleic acid signalling described here may involve additional mechanisms.

Figure 4: Interference of HMGB-nucleic-acid-induced activation of innate immune responses by use of high-binding-affinity nucleotide analogues.
figure 4

a, MEFs were stimulated by cytosolic delivery of B-DNA, poly(I:C) or LPS with or without pretreatment for 30 min with 1 μM CpG-B ODN. Ifnb1 mRNA expression levels were then monitored by qRT–PCR. Data in all panels are means and s.d. (n = 3). Asterisk, P < 0.01, pretreated cells versus non-pretreated cells. ND, not detected. b, Bone marrow-derived Tlr9-/- pDCs were either lipofected with 1 μg ml-1 poly(U) or stimulated with 25 μg ml-1 R837 for 8 h, in the absence of pretreatment or after pretreatment with 5 μM PS or 1 μM CpG-B ODN for 30 min. The mRNA expression levels of the indicated genes were measured by qRT–PCR. Asterisk, P < 0.01, pretreated cells versus non-pretreated cells. c, A schematic view of the hierarchy in the nucleic-acid-mediated activation of immune responses. All immunogenic nucleic acids bind HMGBs (promiscuous sensing), which is required for subsequent recognition by specific pattern recognition receptors (discriminative sensing) to activate the innate immune responses.

PowerPoint slide

Full size image

We propose, and our data support, that HMGBs serve as universal sentinels for nucleic acids that are required for the full-blown, nucleic-acid-induced activation of innate immune responses mediated by the more discriminative pattern recognition receptors (Fig. 4c). What is the biological significance of the HMGB’s function that we describe here? Because HMGBs are broadly found in multicellular organisms including plants28, one may speculate that HMGBs are a remnant of a primitive sensing system for foreign nucleic acids, which has evolved along with and in support of discriminative TLRs and cytosolic receptors of the innate immune signalling system. In this regard, it is worth noting that HMGB1 associates with TLR9 (refs 16, 18) and RLRs (Z.W. and T.T., unpublished observation), suggesting that nucleic-acid-bound HMGBs have evolved to function as co-ligands for discriminative nucleic-acid receptors; in other words, the binding of nucleic acids to HMGBs is a precondition for the more efficient, subsequent recognition by and activation of TLR, RLR and cytosolic nucleic-acid receptors (Fig. 4c).

Nevertheless, it remains unclear exactly where HMGBs bind to nucleic acids and how the HMGB–nucleic acid complexes activate their respective receptor signalling cascades. In vivo visualization data lend support to the concept that HMGB1 functions together with RIG-I on the endosomal membrane, through which nucleic acids are released into the cytosol by low-pH-dependent membrane fusion29,30 (Supplementary Fig. 7). In view of a previous report showing the interaction between HMGB1 and TLR9 in the endosomal compartment16, we conjecture that endosome membrane-associated signalling may be universal to nucleic-acid-mediated innate immune responses. In addition to their role in nucleic-acid sensing described here, HMGBs serve additional functions including gene transcription, all of which may be viewed as a long-term evolutionary consequence of their ability to bind nucleic acids. Nevertheless, further study is required for a more complete understanding of the precise role and mechanism of action of these proteins. Finally, our findings have a number of interesting therapeutic implications, such as in the use of HMGB inhibitory compounds in the suppression of nucleic-acid-mediated pathological immune responses.

Methods Summary

Pull-down assay

MEFs were treated with poly(dA-dT)·(dT-dA) (B-DNA; 10 μg ml-1) for 4 h before mass spectrometry analysis. After stimulation, cells were homogenized with a Dounce homogenizer (Wheaton Science) in homogenization buffer (20 mM HEPES pH 7.4, 20% glycerol, 50 mM KCl, 2 mM MgCl2, 1 mM phenylmethylsulphonyl fluoride (PMSF), 10 μg ml-1 aprotinin, 10 μg ml-1 leupeptin). Cytosolic protein extract was prepared from the homogenate by centrifugation at 19,000g for 30 min, and then incubated with 1.4 μg ml-1 B-DNA, the 5′ terminus of which had been conjugated with biotin, and streptavidin (SA)-conjugated magnetic beads (Invitrogen) at 4 °C for 15 min. The ‘pulled-down’ sample was treated with DNase I (Invitrogen) in reaction buffer (20 mM Tris-HCl pH 8.4, 20 mM MgCl2, 50 mM KCl) and the resultant supernatant was subjected to silver staining (Invitrogen) or mass spectrometry analysis. The in vitro pull-down assay was performed as essentially described previously8. Recombinant HMGB1, HMGB2 and HMGB3 proteins were first incubated for 30 min at 15–20 °C with or without competitors. The supernatants were then mixed for 30 min at 4 °C with biotin-conjugated B-DNA after preincubation with SA-conjugated magnetic beads in binding buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 mM EDTA, 1% Nonidet P40, 100 μg ml-1 leupeptin, 1 mM PMSF, 1 mM Na3VO4). The mixture was then washed extensively with the same buffer, separated by SDS–PAGE and immunoblotted with anti-HMGB1, anti-HMGB2 and anti-HMGB3 antibodies.

Online Methods

Mice, cells and reagents

Mice used for this study were on the genetic background of C57BL/6, except Tlr9-/- mice, which were on the Balb/c background. The generation of Tlr9-/- , Hmgb1-/- and Hmgb2-/- mice were described previously12,31,32. MEFs, RAW264.7, NIH/3T3 and HEK293T cells, bone marrow-derived cDCs and pDCs from Tlr9-/- mice were maintained as described previously8,33. Hmgb1-/- macrophages, cDCs and pDCs were generated from fetal liver haematopietic progenitors (lineage-marker-negative cells purified with the MACS Lineage depletion kit from Miltenyi Biotec) by culturing fetal liver cells for 2 days in the presence of SCF (20 ng ml-1), IL-3 (10 ng ml-1) and IL-6 (10 ng ml-1) followed by a 6-day culture in the presence of 20 ng ml-1 macrophage colony-stimulating factor (for macrophages), 20 ng ml-1 GM-CSF (for cDCs) or 100 ng ml-1 human Flt3-ligand (for pDCs). SCF, IL-3 and IL-6 were purchased from Peprotech. IFN-γ and TNF-α were purchased from R&D Systems. IFN-β was provided by TORAY Industries. B-DNA and calf thymus genomic DNA were purchased from Sigma. Biotin-conjugated poly(dA-dT)·poly(dT-dA) was purchased from Hokkaido System Science. Other oligo DNAs including ISD24, CpG ODNs, fluorescein isothiocyanate-conjugated base-free phosphorothioate deoxyribose homopolymer (PS; 20-mer) and fluorescein isothiocyanate-conjugated base-free natural deoxyribose homopolymer (PD; 20-mer) were purchased from Fasmac. E. coli DNA and R837 were purchased from InvivoGen. Poly(U) and LPS were purchased from Sigma. Poly(I:C) was purchased from GE Healthcare Biosciences. B-DNA, poly(I:C) and other nucleic acid ligands were used at a concentration of 10 μg ml-1 as described previously8, unless otherwise mentioned. The complex formation of CpG-A ODN with DOTAP (Roche) was performed as described previously33. MitoTracker Deep Red 633 was purchased from Invitrogen. Antibodies against the following proteins were used: HMGB1 and HMGB2 (Abcam), HMGB3 (TransGenic Inc.), IRF3 (ZM3; Zymed), β-actin (AC-15; Sigma-Aldrich), NF-κB p65 (C20; Santa Cruz Biotechnology) and phospho-STAT1 (58D6; Cell Signaling).

Plasmid constructions and protein purification

Murine HMGB cDNAs were obtained by RT–PCR on total RNA from MEFs, and then cloned into pGEX4T3 vector (GE Healthcare Biosciences) at the SalI and NotI sites. The glutathione S-transferase (GST)-tagged HMGB proteins were purified by using glutathione-Sepharose beads (GE Healthcare Biosciences). HMGB proteins and GST protein were separated by treatment with thrombin proteinase (Novagen). Other expression vectors are listed in the Supplementary Information.

Immunoblot analysis

Cell lysis and immunoblot analysis were performed as described previously8. IRF3 dimer was assessed by native PAGE, followed by immunoblot analysis with anti-mouse IRF3 antibody as described previously8. The quantification of IRF3 dimer was performed by the NIH Image application. Similar results were obtained in three independent experiments.

RNA analysis

RNA extraction and reverse transcription were performed as described previously8. qRT–PCR analysis was performed with a Lightcycler480 and SYBR Green system (Roche Molecular Biochemicals). All data are presented as relative expression units after normalization to GAPDH. Data are presented as means and s.d. for triplicate determinations. All data were reproduced in two or more additional independent experiments. Primer sequences for murine GAPDH, IL-6, RANTES, IκB-α, IFN-α1, IFN-α4 and IFN-β have been described previously8. Other primer sequences are listed in the Supplementary Information.

Statistical analysis

Differences between control and experimental groups were evaluated with Student’s t-test.

Enzyme-linked immunosorbent assay (ELISA)

Murine IFN-β, IL-6 or IL-1β was measured by ELISA. IFN-β ELISA kit was purchased from PBL Biomedical Laboratories. IL-6 and IL-1β ELISA kits were obtained from R&D Systems. All data were reproduced in two additional independent experiments.

RNA interference

siRNA vectors were constructed by inserting oligonucleotides into EcoRI and HindIII sites of the pSUPER.retro.puro retrovirus vector9. The siRNA targeting sequences for murine HMGB1/2/3 (pan-HMGB siRNA), HMGB2 and Renilla luciferase (control) are 5′-GTATGAGAAGGATATTGCT-3′, 5′-GCGTTACGAGAAACCAGTT-3′ and 5′-GTAGCGCGGTGTATTATACA-3′, respectively. Retroviruses were prepared as described previously9. Transduced cells were selected by puromycin (Sigma) at 2 μg ml-1 (MEFs) or 4 μg ml-1 (RAW264.7 cells) for 48 h.

Electrophoretic mobility-shift assay

Electrophoretic mobility-shift assay was performed as described previously8. An oligonucleotide probe containing a consensus NF-κB binding sequence was used8. The presence of p65 in the NF-κB–DNA binding complex was also confirmed by detection of a supershifted band with an anti-p65 antibody (data not shown).

Viral infection

Cells were infected for 12 h with 1.0 multiplicity of infection of HSV-1 or VSV, as described previously8. To measure the yield of HSV-1 or VSV, a plaque-forming assay was performed as described previously8. All data were reproduced in two additional independent experiments. Virus preparation was described previously8.