High‐mobility group box 1 protein orchestrates responses to tissue damage via inflammation, innate and adaptive immunity, and tissue repair

A single protein, HMGB1, directs the triggering of inflammation, innate and adaptive immune responses, and tissue healing after damage. HMGB1 is the best characterized damage‐associated molecular pattern (DAMP), proteins that are normally inside the cell but are released after cell death, and allow the immune system to distinguish between antigens that are dangerous or not. Notably, cells undergoing severe stress actively secrete HMGB1 via a dedicated secretion pathway: HMGB1 is relocated from the nucleus to the cytoplasm and then to secretory lysosomes or directly to the extracellular space. Extracellular HMGB1 (either released or secreted) triggers inflammation and adaptive immunological responses by switching among multiple oxidation states, which direct the mutually exclusive choices of different binding partners and receptors. Immune cells are first recruited to the damaged tissue and then activated; thereafter, HMGB1 supports tissue repair and healing, by coordinating the switch of macrophages to a tissue‐healing phenotype, activation and proliferation of stem cells, and neoangiogenesis. Inevitably, HMGB1 also orchestrates the support of stressed but illegitimate tissues: tumors. Concomitantly, HMGB1 enhances the immunogenicity of mutated proteins in the tumor (neoantigens), promoting anti‐tumor responses and immunological memory. Tweaking the activities of HMGB1 in inflammation, immune responses and tissue repair could bring large rewards in the therapy of multiple medical conditions, including cancer.

immune system amazes us because of its sheer cleverness and utility.
For decades, the overarching question was: how does the adaptive immune system recognize pathogens? The answer, for a long time, was that it discriminates between self and non-self, with the corollary that anything that belongs to the organism is non-antigenic and anything that does not is antigenic. That view was completely upended in the 1990s by two new concepts: pathogen-associated molecular pattern molecules (PAMPs) and damage-associated molecular pattern molecules (DAMPs). 1,2 PAMPs are entire classes of molecules that are non-antigenic but are recognized by pattern recognition receptors (PRRs) with broad and often multiple specificities; examples are common components of bacteria (like peptidoglycan, lipoproteins, and lipopolysaccharide) and of viruses (like RNA unmodified by capping at its 5′ end).
DAMPs are molecules released by dead or dying cells and particularly by cells that face unscheduled death (as opposed to cells that undergo apoptosis). Like PAMPs, they activate PRRs, and sometimes the same PRR can be activated by both PAMPs and DAMPs. Indeed, both PAMPs and DAMPs elicit the same early responses, like inflammation.
Both PAMPs and PAMPs alert antigen presenting cells of the immune system, and in particular dendritic cells (DCs), which then initiate an adaptive immune response.
Thus, the immune system does not distinguish between self and non-self, but simply whether PAMPs or DAMPs are present, and thus whether a present danger warrants a close examination of what is going on. PAMPs and DAMPs, in fact, are just telltales of danger, but from the immune system's point of view, they are adjuvants: molecules that are not antigens, but prompt the immune system to react against any antigens that it can recognize.
This new paradigm of what is recognized by the immune system is in retrospect both simple and powerful, but in the early days, it had a major flaw: PAMPs had been identified galore, but DAMPs had no qualifying members. We were fortunate enough to recognize the first DAMP, High Mobility Group Box 1 (HMGB1) protein. 3 Ironically, extracellular ATP now qualifies as a major DAMP, but in the 2000s was not considered one.
This review illustrates the role of HMGB1 as a DAMP, starting from the immune system, but framing it in the more general "system of integrity." Indeed, our major interest is tissue repair, and how the immune system supports the healing of damaged tissues. Inevitably, how the immune system recognizes or does not recognize illegitimate new tissue-canceris the other face of the same coin. HMGB1 sticks a finger in all these pies.

| HMGB1 AS A DAMP
The defining characteristic of DAMPs was that they should be released by dead cells and then elicit immune responses. We had studied HMGB1 as a nuclear protein that acts as a DNA chaperone, by binding DNA transiently and bending it reversibly. What attracted our attention to HMGB1 as a DAMP was that, against all odds, it does elicit immune responses: Kevin Tracey and collaborators reported that HMGB1 is secreted by mouse macrophages challenged with LPS, both in vitro and vivo, and is a late mediator of endotoxin lethality. 4 Macrophages do not die after secreting HMGB1, but we went on to test if dead cells could release it, which looked possible since HMGB1 is not tightly bound to DNA, unlike histones. 5 Indeed, cells lysed with detergents, but also subjected to freeze/thawing or to metabolic poisoning, released HMGB1 in the extracellular medium, whereas apoptotic cells did not. 3 Moreover, whereas lysed wildtype cells induced macrophages to secrete TNF, lysed Hmgb1 −/− cells were almost inactive. This qualified HMGB1 as protein that is released by dead cells (Figure 1) and elicits immune responses, with the added specification that apoptotic cells, which are non-immunogenic, do not release HMGB1.
The second property of a DAMP is adjuvanticity, and we showed that exogenous HMGB1 activates DCs in vitro, and in vivo promotes the production of antibodies against a soluble antigen (ovalbumin) and boosts vaccination against poorly immunogenic cancer cells. 6 Extracellular HMGB1 also controls the maturation of DCs and their migration to the closest draining lymph node and their interaction with T cells. [7][8][9] Indeed, HMGB1 activation of antigen presenting cells and their signaling to T cells is so potent that countervailing suppressive mechanisms are required, including HMGB1 binding to the immunosuppressive receptors CD24 10 and TIM-3. 11

| HMGB1 AS A SECRETED ALARMIN
Although HMGB1 is passively released by dead cells, macrophages release it without dying, 4 and therefore secrete it ( Figure 1). HMGB1 does not have a secretory leader peptide, as befits a nuclear protein, and secretion does not involve the classical endoplasmic reticulum (ER)-Golgi secretory network, but a dedicated unconventional secretory pathway. 12 HMGB1 has two nuclear localization signals (NLSs) and shuttles continually between nucleus and cytoplasm, although in normal conditions the vast majority of the protein is located in the nucleus. Acetylation of lysines or phosphorylation of serines in the NLSs 13,14 preclude the re-entry of HMGB1 into the nucleus, and the protein accumulates in the cytoplasm. In hematopoietic cells, HMGB1 is then loaded into secretory lysosomes, which can disgorge their contents outside of the cell following specific signals. 12 The mechanism for loading acetylated HMGB1 into secretory lysosomes, though, has not been elucidated. Non-hematopoietic cells can secrete HMGB1 as well, and the relocalization from nucleus to cytoplasm appears to follow the same rules; however, eventual secretion does not appear to involve secretory lysosomes, but additional pathways have not been investigated. A possibility is that HMGB1 crosses the plasma membrane directly with the help of membrane transporters. In any case, acetylated HMGB1 is a biomarker for HMGB1 secretion, as opposed to passive release by dead cells (the possible phosphorylation of secreted HMGB1 has not been investigated). Unfortunately, specific antibodies against acetylated HMGB1 have not been developed, and the differentiation between acetylated and non-acetylated HMGB1 still depends on mass spectrometry.
The fact that HMGB1 can be actively secreted as well as passively released provides a way to emit the general alarm signal in the presence of severe cell stress, but not necessarily cell damage, or in the presence of pathogens (or rather, PAMPs). An example of the former-severe cell stress-is ischemia. HMGB1 can be secreted during liver ischemia/ reperfusion, for example. 15 Reactive oxygen species (ROS) produced by hypoxic hepatocytes cause the rise of intracellular calcium, which activates calcium/calmodulin-dependent kinases (CaMKs); CaMK inhibition substantially decreased liver damage after I/R. An example of the latter-HMGB1 secretion in the presence of PAMPs-occurs when macrophages detect LPS ( Figure 2). 16 LPS binds TLR4, and signaling is initiated that activates NF-κB and phosphorylates IRFs. Both are transcription factors that, when activated, promote together the activation of interferon beta (IFNβ) transcription and eventual secretion.

IFNβ binds to interferon receptors (IFNARs) on nearby cells, leading
to activation of associated JAK kinases and STAT1 phosphorylation.
Phospho-STAT1 dimers translocate to the nucleus where they promote the transcription of interferon-responsive genes. Phospho-STAT1 dimers also recruit histone acetylases, which acetyate HMGB1 and allow its eventual secretion. A similar cascade presumably occurs when cells respond to virus infection or when IFNγ binds to its receptor.

| FORMS AND FUNCTIONS OF EXTRACELLULAR HMGB1
When outside the cell, HMGB1 has many functions and many receptors. In fact, HMGB1 has posttranslational modifications that make it work as different ligands of different receptors. Whereas acetylation does not appear to alter the binding specificities and the activities of HMGB1, the redox status of its 3 cysteines-C23 and C45 within Box A, and C106 in Box B-plays essential roles. C23 and C45 are ideally placed to form an intramolecular disulfide bond while C106 remains unpaired. The cytosol and the nucleus have a strongly negative (reducing) redox potential, and intracellular HMGB1 is largely in the reduced state. 17,18 However, the extracellular milieu is much more oxidizing in normal conditions, and even more so during inflammation, which favors the formation of the C23-C45 disulfide bond ( Figure 2). The 3 cysteines can be further oxidized to sulfonates by ROS. 18 Fully reduced HMGB1 (also called all-thiol-HMGB1) forms a heterocomplex with the chemokine CXCL12, and the HMGB1-CXCL12 heterocomplex binds CXCR4, a 7-transmembrane G-protein-coupled F I G U R E 1 HMGB1 is both a DAMP that is passively released by dead cells (right) and a soluble protein (alarmin) that is actively secreted by severely stressed cells (left). In normal conditions, HMGB1 is mainly localized in the nucleus in a non-acetylated and reduced (thiol) form. Upon tissue injury, non-modified HMGB1 leaks out from dead cells and is later converted to disulfide-HMGB1 either by spontaneous oxidation or by the encounter with ROS that are abundantly produced by inflammatory cells. Leukocytes and many other cells can secrete HMGB1: HMGB1 is first translocated to the cytoplasm after being acetylated or phosphorylated and then to the extracellular space (after being loaded into secretory lysosomes in leukocytes, or by an unknown mechanism in non-hematopoietic cells). Secreted HMGB1 can be distinguished from passively released HMGB1 (yellow in the diagram) because it is acetylated (green in the diagram). Secreted HMGB1 also becomes oxidized. The diagram on the left represents the pathway of HMGB1 secretion induced by LPS (bacterial PAMP) and interferons (IFNs, produced in response to viral PAMPs). The cascade entails two consecutive steps, IFNβ synthesis and secretion upon LPS binding to TLR4, and HMGB1 secretion upon binding of IFNβ to IFN receptors (IFNAR). HMGB1 acetylation is believed to be executed by lysine acetyltransferases (KATs) recruited by phosphorylated STAT1 dimers bound to DNA receptor (GPCR). 19  Every ligand potentially activates different pathways to a different extent, an effect known as "ligand bias." Signaling and biological effects of CXCL12 and of the HMGB1-CXCL12 complex differ, but the details are scarcely known. Both CXCL12 and the HMGB1-CXCL12 complex promote cell migration, extravasation from vessels, and tissue invasion. Most cell that can move migrate toward the HMGB1-CXCL12 complex, which is most active at concentrations close to 1 nM, while peak activity of CXCL12 is at 30 to 100 nM. 19 The disulfide form of HMGB1 does not from a complex with CXCL12, and therefore does not activate CXCR4. 18 Instead, disulfide HMGB1 binds the MD2-TLR4 complex 22 and promotes the activation of NF-κB and the transcription and secretion of multiple inflammatory mediators, 23 not unlike the best-known ligand of the MD2-TLR4 complex, LPS. However, LPS is much more potent agonist of TLR4 than disulfide HMGB1, which is only active at concentrations above 30 nM.
The contribution of the HMGB1/TLR4 axis to inflammation and immune regulation has been demonstrated in a wide range of experimental models, such as liver and lung damage, stroke, cancer, and epilepsy [24][25][26][27][28] ; its contribution to pain perception is emerging. 29,30 HMGB1 also binds to Toll-like receptors (TLRs). In complex with CpG-ODNs, HMGB1 binds to TLR9 and enhances cytokine production in macrophages, B cells, and plasmacytoid dendritic cells (DCs). 31,32 When HMGB1 is bound to nucleosomes, it activates macrophages and DCs through TLR2. 33 What is the redox state of HMGB1 in these interactions is not known.
The first receptor described for HMGB1 is the receptor for advanced glycation endproducts (RAGE), 34 a multifunctional singletransmembrane protein of the immunoglobulin superfamily. 35 RAGE recognizes multiple ligands, 36 including various S100 proteins, the β 2 integrin Mac-1, amyloid beta, fibrillar aggregates, and aspecifically carboxymethylated proteins (AGEs), in addition to HMGB1 (the disulfide form, most likely, but possibly also other forms). Upon binding of some of its ligands, RAGE is cleaved, generating a soluble form of RAGE that can function as a decoy receptor 37 and a small cytoplasmic fragment that is essential for RAGE signaling. 38 HMGB1 signaling through RAGE leads to activation of the NF-κB pathway, as well as to signal transduction through JNK, p38, and ERK MAP kinase pathways. 39 RAGE was F I G U R E 2 HMGB1 and tissue repair. Here, HMGB1 activities are illustrated during muscle injury. Reduced HMGB1 is released by damaged/necrotic muscle cells (A; inset a: normal muscle cells) and promotes leukocyte recruitment by forming a heterocomplex with CXCL12. Leukocytes are recruited (B) and the damaged tissue is highly inflamed. HMGB1 is oxidized to the disulfide form by ROS originating from infiltrating leukocytes, activates them to release proinflammatory cytokines/ chemokines, but loses the ability to form a heterocomplex with CXCL12. ROS produced by leukocytes eventually cause the terminal oxidation of HMGB1 cysteines to sulfonates; sulfonyl HMGB1 no longer has chemoattracatant nor proinflammatory activities. After inflammation resolution, macrophages with a tissue-healing phenotype release more HMGB1, which may activate stem cells (satellite cells) and promotes angiogenesis, leading to the repair of the damaged muscle (C). (D) Recapitulates the various steps in muscle damage and repair in a diagrammatic way RAGE, though, is involved in many other HMGB1-dependent signaling pathways, including the ones that lead to thrombosis, 42,43 migration of dendritic T cells to lymph nodes, 9 T-cell activation, 8 angiogenesis, 44 and the spreading of brain damage after stroke. 26

| EXTRACELLULAR HMGB1 IN TISSUE DAMAGE AND HEALING
When a tissue is damaged, HMGB1 orchestrates two key events in inflammation, leukocyte recruitment and their induction to secrete inflammatory cytokines. We observed in a model of muscle injury induced by cardiotoxin (CTX) that different redox forms of HMGB1 are present sequentially: fully reduced HMGB1 is released first and becomes disulfide HMGB1 later. 18 We proposed ( Figure 2) 45 that when released in an injured tissue by dead or severely stressed cells, HMGB1 will form a heterocomplex with the low concentrations of CXCL12 always present in extracellular fluids, and promote the production of more CXCL12 by binding the RAGE receptor on neighboring cells. A gradient of the HMGB1-CXCL12 heterocomplex will recruit leukocytes from the microcirculation. Incoming leukocytes will then be activated by disulfide HMGB1 derived by spontaneous or catalyzed oxidation of fully reduced HMGB1, and start producing cytokines, chemokines, more HMGB1, and ROS. ROS will initially convert HMGB1 into the disulfide form, potentiating the inflammatory response, but with time will further oxidize cysteines to sulfonates and inactivate HMGB1. Thus, HMGB1 would first recruit leukocytes, then would activate them, and eventually would be inactivated by them.
However, HMGB1 may also play an important role in the events that follow inflammation resolution: tissue repair and healing. 46 Indeed, HMGB1 is important in muscle regeneration after injury. 47 We envisage three reasons: HMGB1 recruits monocytes at the site of tissue damage, which eventually shift from an inflammatory to a tissue healing phenotype; it recruits local and mesenchimal stem cells

Perhaps inevitably for a protein that is central in inflammation and
injury, HMGB1 is deeply involved in tumor biology as well: inflammation is hallmark of cancer, and cancer is a full-blown though illegitimate tissue that is highly stressed, if anything because it can overgrow its supply of nutrients and is deeply dependent on neoangiogenesis. 56 Exemplary is the involvement of HMGB1 in mesothelioma, 57 an intractable tumor of mesothelial cells lining pleura and the peritoneum, which is highly associated to exposure to asbestos and other mineral fibers ( Figure 3). Asbestos causes inflammation of the mesothelium, but the exact connection between inflammation and mesothelioma had remained obscure, until we discovered that asbestos induces the necrotic cell death of mesothelial cells, the release of HMGB1 into the extracellular space and the recruitment of inflammatory cells. 58 The prolonged bio-persistence of asbestos fibers ensures that HMGB1-driven inflammation smolders chronically for many years; in fact, HMGB1 is elevated in the blood of people exposed to asbestos, and even more so in patients with diagnosed mesothelioma. 59 How exactly chronic inflammation causes the tumor transformation of mesothelial cells is not clear, but a likely possibility is that macrophages favor the survival of cells that in normal conditions would simply die out. In fact, mesothelioma contains a striking abundance of macrophages with a tissue healing phenotype, 60 and the mesothelioma cancer cells themselves contain a high level of HMGB1 and continually secrete it. 61 Most mesotheliomas appear addicted to HMGB1, since targeting HMGB1 with antibodies or small molecules prolongs the survival of mice models of mesothelioma. [61][62][63] Similar to what happens in mesothelioma, macrophages recruited by secreted HMGB1 support the growth of colon carcinoma secondary lesions in the peritoneum. 64 Melanoma and papilloma also can be initiated or supported through the HMGB1/TLR4 axis. 65 Melanoma metastatization upon UV irradiation appears to depend on inflamma-

| HMGB1 IN TUMOR IMMUNOLOGY
Any condition that causes necrotic cell death in a tumor-including necrosis due to insufficient angiogenesis, chemotherapy, or radiation therapy-will cause HMGB1 release and the recruitment of macrophages and neutrophils into the tumor, which can help the tumor recover from damage inflicted to it 68 or promote metastasis, as in melanoma. 66 Viewed from this point of view, HMGB1 is abetting cancer and is a potential anti-cancer target ( Figure 3B). However, an additional effect of HMGB1 may partially or completely counteract the help afforded by the innate immune system to the cancer tissue ( Figure 3C). Some anti-cancer therapies cause a peculiar form of cell death-immunogenic cell death (ICD)-that greatly increases the immunogenicity of the cancer cells, and therefore unleashes an adaptive immune response against the tumor and immunological memory. 69 Cancer cells exposed to radiotherapy and to some currently used chemotherapeutics (including doxorubicin, mitoxantrone, oxaliplatin, and bortezomib, but not cisplatin) undergo ICD, as demonstrated by vaccination experiments in mice and the regression of non-irradiated metastases in humans (the abscopal effect). 70,71 ICD is a form of apoptotic death, and as such, it should be silent or even suppressive from an inflammatory and immunological point of view. However, ICD is characterized by the emission of calreticulin, HMGB1, and ATP. 72 Notably, ATP is a "find-me" signal and calreticulin an "eat-me" signal for professional phagocytes, and both HMGB1 and ATP are deeply involved in inflammation at various levels. Their combination stimulates the cross presentation by DCs of neoantigens from cancer cells to the immune system and activation of CD4 T cells.  77 Since most, if not all, ICD inducers induce the production of ROS, a search for procedures that would lead to ROS production that target the ER led to the identification of hypericin-based photodynamic therapy. 77 This also suggests that ROS production that targets the ER is an upstream molecular event in ICD.
How and why HMGB1 is released during ICD is not clear, but certainly, it must be released prior to caspase 3 activation, since after that step, HMGB1 cannot be released and remains tightly associated to the apoptotic cell remnants. 3,78 Likewise, what redox form(s) of HMGB1 are involved in ICD has not been formally tested, but the ICD-triggering production of ROS and the involvement of TLR4 suggest that disulfide HMGB1 is important. Notably, the complete oxidation of HMGB1 in apoptotic cells that have progressed past the activation of caspese 3 renders HMGB1 tolerogenic rather than inflammatory. 79

| CONCLUSION: FROM EVOLUTION TO THERAPIES
HMGB1, initially identified as a chromatin protein of unknown function, 80 does function in the nucleus as a chaperone that facilitates DNA bending and nucleosome assembly. 81 These functions are probably ancestral, as all eukaryotes contain proteins structurally related to HMGB1, 82 and the yeast proteins NHP6A/B perform similar functions in chromatin. 83,84 However, during evolution, HMGB1 has acquired an additional role as a DAMP that can signal death and stress consequent to an assault to the integrity of the cell; this new role must have preceded the animal-plant divergence, as Arabidopsis contains at least one HMGB protein that performs alarmin functions upon fungal infection. 85 Notably, also a clam HMGB1 has been shown to activate innate immunity. 86 Since HMGB1 functions as alarmin are so old, we have to assume that HMGB1 became one of the pillars of the innate immune system early on, and has been reused by the adaptive immune system since its evolutionary inception in teleost fish. Notably, the ancestral HMGB gene has acquired several paralogs in mammals, 82 and possibly these closely related proteins (HMGB2 to 4) have related but subtly different functions when compared with HMGB1.
During such a long evolutionary time, HMGB1 and its family members have apparently been used in every biological process where any cell has an interest into the death or the possible death of a neighboring cell, for any reason from being alerted to pathogens, to supporting or rescuing the ailing cell, to the process of replacing it. Some of these processes are associated to pathologies, such as infection or degenerative diseases that cause cell death, but some may be completely physiological and related to the homeostasis of the organism: hundreds of millions of cells are replaced each day in our body. Perversely, some of these functions may not be completely aligned to each other, for example, when immune cells from the innate arm abet tumors, and immune cells from the adaptive arm try to eliminate them. The true step forward will come when we will be able to tweak the different functions of HMGB1 in order to favor one outcome over the opposite one: accelerating tissue healing must not accelerate tumor growth.
Given the complexity of HMGB1 biology, a lot of work remains to be done, but the rewards in terms of innovative therapeutic approaches in a variety of clinical settings are fully worth the effort.