The ‘Cytokine Storm’: molecular mechanisms and therapeutic prospects

Abstract

Cytokine storm syndrome has generally been described as a collection of clinical manifestations resulting from an overactivated immune system. Cytokine storms (CSs) are associated with various pathologies, as observed in infectious diseases, certain acquired or inherited immunodeficiencies and autoinflammatory diseases, or following therapeutic interventions. Despite the role of CS in tissue damage and multiorgan failure, a systematic understanding of its underlying molecular mechanisms is lacking. Recent studies demonstrate a positive feedback loop between cytokine release and cell death pathways; certain cytokines, PAMPs, and DAMPs, can activate inflammatory cell death, leading to further cytokine secretion. Here, we discuss recent progress in innate immunity and inflammatory cell death, providing insights into the cellular and molecular mechanisms of CSs and therapeutics that might quell ensuing life-threatening effects.

The cytokine storm and cell death

The term ‘cytokine storm’ (CS) has been increasingly used in both the scientific literature and public media to denote an out-of-control inflammatory response, although a specific molecular definition to delineate CS from normal inflammation and to describe the amounts and types of cytokines involved has remained elusive. CS was first used in the early 1990s to describe the effects of graft-versus-host disease (GvHD) [1, 2] and later in the infectious disease setting in the early 2000s in studies of cytomegalovirus (CMV) [3], Epstein–Barr virus (EBV)-associated hemophagocytic lymphohistiocystosis (HLH) [4], group A streptococcus (GAS) [5], influenza virus [6], variola virus [7] and severe acute respiratory syndrome coronavirus (SARS-CoV) [8]. The CS was also described in cases of human H5N1 avian influenza virus infection during the 2005 ‘bird flu’ outbreak [9]. Since then, the term CS has been steadily gaining scientific and public interest, with the ongoing coronavirus disease 2019 (COVID-19) pandemic, caused by the SARS-CoV-2 virus, further fueling attention.

Although the CS has been best studied in infectious diseases and sepsis, it has also been described in non-infectious settings. For example, it is known as macrophage activation syndrome (MAS) when associated with rheumatic diseases such as systemic lupus erythematosus (SLE) and systemic juvenile idiopathic arthritis (sJIA) [10]. The CS also has a role in hyperinflammatory diseases such as inherited and non-inherited forms of HLH [11]. Furthermore, CSs have been reported in response to therapeutic interventions, such as GvHD during hematopoietic stem cell transplantation (HSCT) or hyperinflammation following the administration of immunotherapeutic agents, including antibodies and chimeric antigen receptor (CAR) T cells [12–14]. However, despite the clinically important incidence of CSs under diverse conditions, little is known about the mechanisms underlying how they are initiated or perpetuated. Understanding the physiological basis of CSs is crucial to providing new potential insights into fine-tuning immune responses to treating these life-threatening conditions.

Many of the clinical features of the CS, which include fever, hepatosplenomegaly, progressive liver failure with coagulopathy, cytopenias, and hyperferritinemia [15] (Table 1), can be explained by the effects of common pro-inflammatory cytokines, such as IL-1, IL-6, interferon-γ (IFN-γ), tumor necrosis factor (TNF), and IL-18 [16]. Amounts of these cytokines are increased in the circulation of most patients with CS syndrome (CSS; also known as cytokine shock) [15]. Increasing clinical evidence has described links between the pathogenesis of a CS and programmed cell death (PCD) processes. For example, during CSS in patients with sepsis, there is marked cell death of immune cells such as B and T cells in lymphoid organs [17]. Blocking lymphocyte cell death improves survival in a mouse model of sepsis, which suggests that cell death may play a pathogenic role during CSS [18, 19]. Indeed, autopsies of patients with sepsis showed that the death of immune cells was one of the underlying causes of mortality [20]. The cell death involved in CSS was historically characterized as apoptosis, but despite the favorable results of using apoptosis inhibitors in animal models of CSS (such as sepsis), translation of these therapies to the clinic has met with limited success [21]. These limitations suggest that PCD pathways other than apoptosis might also be involved in driving certain pathological features of a CS.

Table 1 |.

Clinical manifestations in cytokine shock syndromes[16]

Constitutional Symptoms	Nervous System	Gastro-intestinal System	Vascular/Lymphatic System	Rheumatologic System
• Fever • Headache • Fatigue • Anorexia	• Confusion • Delirium • Ataxia • Seizures • Anosmia	• Nausea • Vomiting • Diarrhea • Abdominal pain	• Anemia, cytopenia • Coagulopathy • Increased ferritin, cytokines, CRP and D-dimer • Hemorrhage, stroke • Lymphadenopathy • Endothelial damage	• Vasculitis • Arthritis • Arthralgia
Heart	Skin	Lungs	Liver	Kidney
• Hypotension • Arrhythmias • Cardiomyopathy • Ischemia • Cardiogenic shock	• Urticaria • Rash • Edema • Vesicles	• Pneumonitis • Pulmonary edema • Dyspnea • Hypoxemia • ARDS	• Hepatomegaly • Elevated bilirubin • Liver failure • Increased AST, ALT and LDH	• Acute kidney injury • Proteinuria • Hematuria • Kidney failure

Recent studies have elucidated the molecular mechanisms underlying synergies between three innate immune-mediated PCD pathways — inflammasome-dependent pyroptosis, apoptosis, and necroptosis; together, these pathways provide a framework for molecularly defining the CS. We refer to the crosstalk between these cytokine-mediated inflammatory PCD pathways as PANoptosis—a PCD activated by pathogens such as influenza A virus (IAV), cytokines, or by sterile triggers such as cancer cells, and regulated by the PANoptosome (a molecular scaffold enabling key molecules from the pyroptosis, apoptosis, and necroptosis pathways to contemporaneously engage) [22]. Studies have shown that cytokines are intricately linked to cell death mechanisms and involved in a positive feedback loop whereby cytokine release causes inflammatory cell death that facilitates further pathogenic cytokine release through membrane pores and cell lysis, culminating in a CS to drive severe, life-threatening damage to host tissues and organs [23]. Indeed, systemic inflammation, tissue damage, multiorgan failure and mortality in CSS are prevented by combined treatments with TNF and IFN-γ neutralizing antibodies, blocking cytokine-mediated inflammatory cell death in murine models of CSs (such as poly(I:C)induced HLH, LPS-induced sepsis, and SARS-CoV-2 infection) [23]. The identification of cell death-associated molecules, primarily caspase-8, Z-DNA-binding protein 1 (ZBP1), transforming growth factor-β-activated kinase (TAK1), and receptor-interacting serine/threonine protein kinase 1 (RIPK1), as master switches of inflammasome activation and of PCD pathways [24-31], has further established the concept of cytokine-mediated inflammatory cell death in infection, inflammatory disease, and cancer [32–34]. Understanding this cytokine-cell death positive feedback loop through which cytokines may cause inflammatory cell death to drive further cytokine release, CSs, organ damage, and lethality, is crucial for more effectively treating and possibly preventing CSS during diseases.

In this review, we discuss the mechanisms that initiate and perpetuate the CS. We briefly describe the stimuli that can trigger a CS. Next, we detail a few landmark studies that have elucidated molecular pathways of inflammatory cell death in association with the CS, systemic inflammation, tissue damage, and multiorgan failure. Finally, we discuss therapeutic interventions harboring the potential to mitigate CS-mediated life-threatening inflammation. An improved mechanistic understanding of the CS can inform the development of novel, candidate therapeutic strategies to treat numerous fatal diseases associated with CS such as sepsis, HLH, and the ongoing COVID-19 pandemic.

Stimuli that trigger the CS

Multiple stimuli can result in CSS with largely similar clinical manifestations (Table 2). Based on the source of the stimulus, a CS can be broadly classified into pathogen-induced CS (sepsis), autoinflammatory or monogenic CS, or therapeutic intervention-induced CS. Understanding the underlying mechanisms linking these stimuli to the CS is important to identify the molecular pathways involved.

Table 2 |.

Different forms of CSS: contributing cells and driving cytokines in humans and murine models

Cytokine Storm Syndromes (CSS)	Cause(s)	Contributing cells	Primary driving cytokines	References
Pathogen-induced
Bacterial sepsis	GAS, Staphylococcus aureus, Francisella tularensis, Klebsiella pneumoniae Yersinia pestis	Heterogenous	Multiple cytokines, particularly TNF and IFN-γ	[37–41]
Viral sepsis	IAV, Ebola virus, SARS-CoV, MERS-CoV, SARS-CoV-2	Heterogenous	Multiple cytokines, particularly TNF and IFN-γ	[45]
Autoinflammatory and monogenic
pHLH	Mutations in PRF1, UNC13D, STX11, STXBP2	CD8+ T cells	IFN-γ	[49–55]
sHLH	Epstein-Barr virus, CMV	CD8+ T cells, myeloid cells	IFN-γ, IL-1	[3, 4]
CAPS	NLRP3 mutation	Myeloid cells	IL-1β	[140]
Familial Mediterranean fever	MEFV mutation	Myeloid cells	IL-1β	[171]
NLRC4-MAS	NLRC4 mutation	Myeloid cells	IL-18	[142]
Therapeutic intervention-induced
CAR T-cell therapy	Infusion of CAR T cells	CAR T cells Myeloid cells Tumor cells	IL-6, IL-1	[139]
Blinatumomab treatment	Infusion of T-cell receptor-engaging antibodies: anti-CD19 and anti-CD3	T cells Myeloid cells Tumor cells	IL-6	[13]
Allogeneic stem cell transplantation	N/A	T cells	N/A	[172]

CAPS, cryopyrin-associated periodic syndrome; CAR, chimeric antigen receptor; CMV, cytomegalovirus; GAS, group A streptococcus; IAV, influenza A virus; MAS, macrophage activation syndrome; pHLH, primary hemophagocytic lymphohistiocystosis; sHLH, secondary hemophagocytic lymphohistiocystosis.

Pathogen-induced CS.

Infection-associated CS or sepsis is the leading cause of death globally for patients in the intensive care unit (ICU) [35]. According to the World Health Organization, sepsis kills around 11 million people each year [36], and this life-threatening condition has been primarily studied in the context of bacterial and viral respiratory infections (Table 2).

For example, bacterial infections with Streptococcus pyogenes, Staphylococcus aureus, Francisella tularensis, Klebsiella pneumoniae and Yersinia pestis result in a systemic inflammatory response in humans and mice [37–41]. These bacteria usually trigger NF-κB signaling and the activation of interferon regulatory factor 3 (IRF3) and IRF7 in mammalian immune cells by binding to pattern recognition receptors (PRRs) [42]. They can also drive inflammasome activation, particularly in innate immune cells [43]. Together, this signaling pathway induces the production of several pro-inflammatory cytokines, including IFN-γ, TNF, IL6, IL-1β and IL-18 in immune cells [43] (FIG. 1). However, only a subset of patients with sepsis develops CSS, while others develop immunoparalysis. Patients who survive the initial CS but subsequently die may be those who do not recover from immunoparalysis [44].

Figure 1 |. Signaling pathways producing pro-inflammatory cytokines.

a) The cartoon depicts the recognition of pathogen-associated molecular patterns (PAMPs) or damage-associated molecular patterns (DAMPs) by pattern recognition receptors (PRRs) such as the Toll-like receptors (TLRs), NOD-like receptors (NLRs), C-type lectin receptors (CLRs), RIG-I–like receptors (RLRs), or the cGAS-STING axis, engaging NF-κB, MAPK, and interferon regulatory factor 3 (IRF3)–IRF7 activation for the transcription of pro-inflammatory cytokine genes [42, 43, 89]. Fungal ligands bind the CLR dectin 1 and signal through the dectin 1–SYK pathway to activate NF-kB signaling. TLR signaling through MYD88 activates the NF-kB pathway, whereas TRIF-dependent TLR signaling primarily involves IRF3 and IRF7 activation. The depiction is based on results from human and murine studies. b) Inflammasome sensors can interact directly with their target ligand or respond to a variety of physiological changes. This leads to the recruitment of apoptosis-associated speck-like protein containing a caspase activation and recruitment domain (ASC) and caspase 1 (CASP1) to form an inflammasome complex. CASP1 then undergoes proximity-induced autocatalytic cleavage to take on its active form. Activated CASP1 cleaves gasdermin D (GSDMD) to release the N-terminal domain, which oligomerizes in the plasma membrane and forms pores, thereby inducing cell death through pyroptosis. Activated CASP1 also cleaves pro–interleukin (IL)-1β and pro–IL-18 into their active forms, which are released through the GSDMD pores [43]. The depiction is based on results from human and murine studies. MAVS, mitochondria antiviral signaling; MDA5, melanoma differentiation-associated protein 5.

In the case of viral infections, sepsis has been widely studied in the context of IAV, Ebola virus, and coronaviruses. IAV remains a major threat to public health and has caused four pandemics in the 20^th century, most notably in 1918–1919, with the largest known outbreak of any infectious disease to date. An exaggerated immune response contributed to the lethality of the 1918 IAV pandemic. In addition, human-infecting coronaviruses (CoVs), including SARS-CoV, MERS-CoV, and SARS-CoV-2, trigger robust inflammatory responses that can contribute to pathology in humans and mice [45]. The pro-inflammatory genes that are upregulated in innate immune cells in patients with severe or critical COVID-19 are located primarily downstream of the NF-κB and type I IFN signaling pathways [46, 47]. Peripheral blood mononuclear cells (PBMCs) obtained from healthy donors and subsequently infected with SARS-CoV-2 have shown increased production of pro-inflammatory cytokines, including TNF, IL-6, IFN-γ, and IL1β in vitro [23]. Several studies analyzing cytokine profiles from patients with severe COVID-19 have suggested a direct correlation between a CS and lung injury, multiple organ failure, and unfavorable prognosis [48] (Box 1). In particular, TNF and IFΝ-γ are highly upregulated in the serum of patients with severe COVID-19, and as discussed later, TNF- and IFN-γ–induced cell death contribute to systemic inflammation, tissue damage, acute respiratory distress syndrome (ARDS), multiorgan failure, and mortality in COVID-19 [23].

BOX-1: The Cytokine Storm in COVID-19.

Although CSS-inducing agents differ in fundamental respects, the interactions of PAMPs or DAMPs with PRRs evoke similar cytokine profiles. Studies have now compared the cytokine profiles in patients with COVID-19 with those experiencing other types of CSS, leading some to suggest that CSS does not occur in COVID-19 [222–224]. A meta-analysis comparing inflammatory cytokine profiles reported lower amounts of pro-inflammatory cytokines, particularly IL-6, in the serum of patients with COVID-19 compared to patients with sepsis or CAR T cell-induced CSS [222]. Also, RNA sequencing analysis of peripheral blood mononuclear cells (PBMCs) from patients with influenza A virus (IAV) infection or with COVID-19 identified lower cytokine expression in PBMCs of patients with COVID-19 [223]. However, a major limitation of these studies is that the authors draw conclusions from comparisons between patients with COVID-19 and other CSS-associated diseases, but not healthy subjects [222, 223]. By contrast, another similar study found increased amounts of pro-inflammatory cytokines, particularly TNF and IL-1, being expressed from PBMCs taken from patients with COVID-19 compared to patients with IAV infection [225]. Additionally, cytokines have not been measured simultaneously on the same platform across diseases with CSS. Moreover, analyzing cytokines at just one time point may not be indicative of the full disease course. There are several other factors that influence the analysis of cytokine measurement, including time and compartment of sample collection, underlying diseases or polymorphisms, as well as previous pathogen exposure and immunosuppression. Indeed, a longitudinal analysis of the host response to SARS-CoV-2 and IAV in ferrets found increased amounts of IL-6, IL-1 and other chemokines in the cells obtained from nasal wash of SARS-CoV-2–infected ferrets compared to those in IAV-infected ferrets 7 days after infection, supporting the idea that a CS does occur in COVID-19 [224]. So far, most studies have focused on direct measurements of cytokines in the peripheral blood compartment and have failed to interrogate the cytokine profile in local tissues such as the lung. All of these points should be taken into consideration when drawing conclusions about the occurrence of a CS in COVID-19 patients.

Autoinflammatory or monogenic CS.

Monogenic abnormalities leading to autoinflammation can also cause CSS. Primary HLH (pHLH) is a classic example of monogenic CSS (Table 2). pHLH, also known as familial HLH, is a rare hereditary disease of immune dysregulation caused by one of several underlying genetic mutations that primarily affect T cell and NK cell activity in infants and young children [49–51]. Mutations in the PRF1, UNC13D, STX11 and STXBP2 genes --crucial for granule-dependent cytolytic functions of T and NK cells-- predispose children to pHLH [52–55]. Dysregulation of the perforin-dependent cytolytic pathway by which antigen-presenting cells (APCs) are eliminated by cytotoxic lymphocytes (CTLs) results in excessive production of pro-inflammatory cytokines, particularly IFN-γ, and the overactivation of NK cells and CTLs [53]. However, disease onset, severity and incidence in pHLH are highly variable, depending on the genes involved and the types of mutations present [56].

Additionally, HLH can develop without the aforementioned germline mutations as a result of infection, certain autoinflammatory conditions. or malignancies -- referred to as secondary HLH (sHLH), or acquired HLH[4]. Although various infectious agents have been linked to sHLH, viruses --particularly EBV-- remain the most common triggers [4]. However, there is currently no clear distinction between pHLH and sHLH, as children with pHLH often present with coinfections [57]. The crucial role of T cells in the development of HLH is exemplified by a mouse model in which IFN-γ produced by CD8⁺ T cells contributed to hematological abnormalities in perforin-deficient mice infected with lymphocytic choriomeningitis virus (LCMV) [58–60]. However, despite their marked deficiency of T cells and NK cells, patients with severe combined immunodeficiency can still develop HLH, suggesting that the pathogenesis of HLH may not always involve T cells and NK cells [61].

Germline mutations in genes regulating the innate immune system and inflammasome signaling are also associated with CSS. Mutation or dysregulation of PRRs or molecules involved in inflammatory signaling and cell death pathways can lead to the secretion of large quantities of cytokines, contributing to the pathology of several diseases such as Familial Mediterranean fever (Table 2). The specific mechanisms through which genes involved in cell death pathways play roles in driving CSS (e.g. those encoding inflammasome components, ZBP1, caspase-8, FADD, RIPK3 and RIPK1) are discussed below.

Therapeutic intervention-induced CS.

Therapy induced-cytokine release and cell death are crucial to eliminating cancer; however, they can also cause a CS. CAR T cell therapy has been a major innovation in cancer immunotherapy. Although the majority of patients achieve clinical remission following treatment with CAR T cells, a sizable proportion of patients develop CSS, with manifestations ranging from fever and hypotension, to hypoxia and organ failure [12]. It is commonly thought that CSS in these patients occurs due to the activation of bystander cells, particularly myeloid cells, by CAR T cells in the tumor environment [12]. Subsequent inflammatory cell death and tumor lysis induce a CS through the release of inflammatory cytokines [62]. Pro-inflammatory cytokines, including IL-6, IL-1 and IFN-γ, are consistently found to be elevated in the serum of patients treated with CAR T cells who develop a CS [62]. Also, in a mouse model of CAR T cell–induced CS where mice were injected with Raji tumor cells and subsequently treated with 1928z CAR T cells, the overall amounts of pro-inflammatory cytokines (including IL-6, CCL2, G-CSF, IL-3, IFN-γ, GM-CSF and IL-2) strongly correlated with CSS severity and mortality [63]. As not all patients treated with CAR T cells develop a CS, additional factors, such as CAR design, co-morbidity, and underlying genetic polymorphisms, are likely to play a part in the generation of a CS. Additionally, to date, there have been no reported cases of CSS using NK-cell CAR therapy in cancer patients [64].

CSs can also occur in response to other T cell–engaging immunotherapies. One example is the treatment of patients with relapsed or refractory acute lymphoblastic leukemia with blinatumomab-- a bi-specific antibody that binds to CD19⁺ and CD3⁺ T cells [13]. Similarly to CAR T cells, this antibody induces T cells and macrophages to produce pro-inflammatory cytokines [13, 14]. Another earlier example of a T cell–engaging immunotherapy is the superagonist anti-CD28 monoclonal antibody TGN1412, which directly activates CD4⁺ effector memory T cells and which induced rapid CSS within minutes after infusion in six healthy volunteers [65]. TGN1412 stimulation of T cells results specifically in the production of IL-2 and IFN-γ, suggesting that these two cytokines are associated with TGN1412-induced CSS [66]. However, preclinical studies failed to predict the CS in human volunteers. Preclinical studies had been performed in vitro in human PBMCs, as well as in vivo in rodents using CD28SA (surrogate antibody to TG1412 and in cynomolgus macaques using TGN1412 [67]. In rodents, the activation of Tregs by CD28SA prevented the CS: deletion of regulatory T cells (Tregs) in mice resulted in a significant increase in the amount of circulating pro-inflammatory cytokines relative to controls [68]. In macaques with a CS, the failure to respond to TGN1412 was deemed to be due to lack of CD28 expression on CD4⁺ effector memory T cells in macaques, presumably preventing TGN1412 from activating these cells and inducing cytokine production [66]. Therefore, these preclinical models were flawed in predicting the clinical safety of TGN1412. Other antibody-based therapies have also lead to a CS. For example, infusion of the anti-CD20 monoclonal antibody rituximab has induced a CS in patients with B-cell chronic lymphocytic leukemia [69]. The targeted killing of CD20⁺ B cells induced by this antibody, including via apoptosis, had led to cytokine release [70]. Collectively, T cell-engaging immunotherapies and antibody-based therapies are known to induce CS in humans, and careful evaluation using optimal preclinical models is imperative in the mechanistic study of these treatments and to be able to assess their safety before putative translation to the clinic.

Molecular mechanisms of a CS

Although many different stimuli can induce CSS, in all cases, the interactions of pathogen-associated or damage-associated molecular patterns (PAMPs or DAMPs) with PRRs of the innate immune system, activate downstream molecular pathways that are conserved across stimuli and mammalian species [43]. The host innate immune response to infection or sterile insults provides the first line of defense against damage and activates major signaling pathways for the production of inflammatory cytokines and chemokines (FIG. 1). These cytokines are crucial for clearing infections and maintaining cellular homeostasis at physiological concentrations, but the dysregulated release of pro-inflammatory cytokines can result in a CS. Thus, characterizing the pathways that can result in dysregulated cytokine release is essential to robustly identifying molecular targets that might lead to novel candidate therapeutic strategies to treat CSS.

Cytokines can induce inflammatory cell death.

Among the numerous pro-inflammatory cytokines that are elevated during a CS, IL-1, IL-6, TNF and IFN-γ are of paramount importance [71]. Of these, TNF and IFN-γ have been widely studied in the context of cell death and independently induce apoptosis or necroptosis in a context-dependent manner [72, 73]. Intraperitoneal injection of TNF and IFN-γ induces mortality in mice [23, 74], mirroring the major symptoms of CSS observed in patients with COVID-19 [23]. TNF and IFN-γ are also known to cause cell death in multiple cell types, contributing to various pathological conditions such as neurological disorders (multiple sclerosis and dementia), liver damage, chronic obstructive pulmonary disease (COPD), fibrosis, and osteoporosis [23, 74]. Mechanistically, the combination of TNF and IFN-γ induces STAT1- and IRF1-dependent nitric oxide (NO) production, which in turn activates caspase-8 and inflammatory cell death, with characteristics of pyroptosis, apoptosis, and necroptosis, in murine macrophages [23] (FIG. 2a). Whereas pyroptosis is executed by gasdermin family members, apoptosis, which can also be inflammatory and lead to cytokine release, is executed by caspase-3 and caspase-7 following the activation of upstream initiator caspases (caspase-8, caspase-10 or caspase-9), and necroptosis is executed by RIPK3–mediated oligomerization of mixed lineage kinase domainlike (MLKL) in multiple cell types [75]. Activation of these cell death pathways results in the release of more cytokines and alarmins, which function as ‗danger signals’ to alert surrounding immune and stromal cells to the presence of cell/tissue damage, and to further fuel the cytokine release, culminating in CS [23, 43]. The mechanistic descriptions of how innate immunemediated inflammatory cell death in response to cytokines, PAMPs, or DAMPs, occurs and induces cytokine release (fueling the cytokine-cell death positive feedback loop), are detailed below.

Figure 2 |. Regulation of inflammatory cell death by cytokines, cell death proteins, and infection.

a) Schematic representation of inflammatory cell death induced by the combination of TNF and IFN-γ. Activation of JAK–STAT1 signaling by TNF and IFN-γ induces the upregulation of interferon regulatory factor 1 (IRF1), which in turn triggers upregulation of inducible nitric oxide synthase (iNOS) and the production of nitric oxide (NO). NO then leads to caspase 8 (CASP8) activation. Activated CASP8 induces gasdermin E (GSDME)-mediated pyroptosis, caspase-3 and caspase-7 (CASP3 and CASP7)-driven apoptosis and RIPK3 and mixed lineage kinase domain-like pseudokinase (MLKL)-mediated necroptosis [23]. The depiction is based on results from human and murine studies. b) TNF stimulation induces rapid assembly of a multiprotein signaling complex containing TNF receptor type 1 (TNFR1)-associated DEATH domain protein (TRADD) and receptor interacting protein kinase 1 (RIPK1), which subsequently recruits E3 ubiquitin ligases such as cIAP1/2 and LUBAC [99, 100]. The K63-linked ubiquitin chains generated by cIAP1/2 induce transforming growth factor β-activated kinase 1 (TAK1) recruitment and RIPK1 phosphorylation. LUBAC further conjugates the RIPK1 complex with M1linked linear ubiquitin chains [100]. The deubiquitinase A20 mediates the cleavage of K63-linked ubiquitin chains on RIPK1. Furthermore, A20 can add K48-linked ubiquitin chains to RIPK1, thereby targeting it for proteasomal degradation [101]. The phosphorylation of RIPK1 by IKKα– IKKβ, TBK1–IKKε and TAK1 maintains its prosurvival signaling [88]. Inhibition of RIPK1 phosphorylation dissociates RIPK1 from the TNFR1 complex and triggers its assembly with Fas-associated death domain protein (FADD) and CASP8, which triggers gasdermin D (GSDMD) and GSDME-mediated pyroptosis and CASP3–CASP7–mediated apoptosis [83]. Inhibition of CASP8 induces RIPK1 to phosphorylate RIPK3, thereby triggering MLKL-mediated necroptosis. The depiction is based on results from human and murine studies. c) Influenza A virus (IAV) Z-RNA is recognized by the Zα2 domain of Z-DNA binding protein 1 (ZBP1) [25, 93]. Caspase-6 (CASP6) promotes the interaction between ZBP1 and RIPK3 [92]. ZBP1 triggers NLRP3-dependent GSDMD cleavage, CASP8-dependent CASP3 and CASP7 activation, and MLKL phosphorylation to induce inflammatory cell death during IAV infection [25]. The depiction is based on results from murine studies. ASC, apoptosis-associated speck-like protein containing a caspase activation and recruitment domain; CARD, caspase recruitment domain; DD, death domain; DED, death effector domain; FADD, fas-associated death domain protein; LRR, leucine-rich repeat; RHIM, receptor interacting protein homotypic interaction motif.

The role of RHIM domain proteins in cell death and the pathogenesis of a CS.

RIPK1, RIPK3, ZBP1, and TIR-domain-containing adapter-inducing interferon-β (TRIF), are four mammalian proteins that harbor RIP homotypic interaction motif (RHIM) domains; they are key components of signaling pathways involved in innate immune-mediated cell death in response to cytokines, PAMPs, or DAMPs, as well as in inflammation and subsequent cytokine release in CSS [76–80]. The function of these proteins in CSS has been studied in mice and humans: RIPK1 is a master regulator of cytokine signaling as well as of cell survival and death [43]. For example, RIPK1 controls cytokine release in a murine model of human neutrophilic dermatosis, in which a point mutation in the protein tyrosine phosphatase, non-receptor type 6 (Ptpn6^spin) causes inflammatory skin disease; irradiated mice transplanted with fetal liver cells from Ptpn6^spinRipk1^−/− mice are protected from disease, as are Ptpn6^spinIl1a^−/− mice, highlighting the important role of RIPK1 in mediating IL-1α release, driving this pathology [81]. Additionally, in response to TNF and engagement of related death receptors, RIPK1 undergoes posttranslational modifications such as ubiquitylation and phosphorylation, thereby promoting the formation of a pro-inflammatory and cell survival-mediating protein complex [82] (FIG. 2b). However, in response to defects in NF-κB signaling required for survival, RIPK1 may be subsequently released from the TNF receptor into the cytosol to nucleate other cell death molecules such as FADD, CASP8, or RIPK3, thus activating inflammatory cell death pathways [83] (FIG. 2b). Indeed, the scaffold functions of RIPK1 are known to be crucial for inhibiting systemic inflammation in mice by limiting cell death [76, 77], and loss of RIPK1 in mice leads to postnatal lethality marked by symptoms of CSS, such as systemic inflammation, severe anemia, cytopenia, and neutrophilia [77]. The lethal phenotype caused by RIPK1 deficiency is rescued in mice by co-deletion of the pro-apoptotic caspase-8 and the pro-necroptotic RIPK3, but not by individual deletions of caspase-8, RIPK3 or the downstream necroptosis effector MLKL, suggesting that there is coordination between apoptosis and necroptosis in driving the CS and lethality in the absence of RIPK1 [76, 77]. In addition, loss of TNF signaling in Ripk1^−/−Tnfr1^−/− mice does not rescue postnatal lethality mediated by RIPK1 deficiency [77]. Whereas Ripk1^–/– mice die postnatally, mice expressing a mutant form of RIPK1 that cannot undergo caspasemediated cleavage (Ripk1^D325A/D325A mice) die perinatally from systemic inflammation, showing that caspase-mediated RIPK1 cleavage is important to preventing necroptosis and allowing embryogenesis [84]. These mice can still be rescued by the combined loss of caspase-8 and RIPK3, suggesting that apoptosis and necroptosis drive lethality [84]. Moreover, heterozygous missense mutations in RIPK1 -- D324N, D324H, D324V and D324Y, which are resistant to caspase cleavage -- are associated with early-onset periodic fever syndrome and severe intermittent lymphadenopathy in humans [84, 85]. Impaired cleavage of RIPK1 D324 variants by caspase-8 sensitizes patient PBMCs to TNF-induced RIPK1 activation, which drives apoptosis and necroptosis and leads to the overproduction of inflammatory cytokines [85].

Similarly, mutation of the conserved RHIM in RIPK1 (mRHIM) causes perinatal lethality characterized by epidermal hyperplasia, dermatitis, and increased influx of neutrophils in the peritoneum, concomitant with increased cytokine concentrations in the serum of Ripk1^RHIM/RHIM mice relative to wildtype mice [78, 79]. Furthermore, mice expressing RIPK1 RHIM only in keratinocytes (Ripk1^RHIM/E-KO) develop skin inflammation compared to mice expressing wildtype RIPK1 (Ripk1^WT/E-KO) [78]. The perinatal lethality or skin inflammation in mice expressing RIPK1 mRHIM is prevented by loss of RIPK3, MLKL, or ZBP1, which suggests that the RIPK1 RHIM mutation may primarily drive the spontaneous activation of the RIPK3–MLKL-mediated necroptosis pathway in mice [78, 79]. Immunoprecipitation assays in primary mouse embryonic fibroblasts from Ripk1^RHIM/RHIM mice show enhanced ZBP1 binding with RIPK3 compared with their interaction in wildtype fibroblasts, suggesting that the RIPK1 RHIM typically prevents ZBP1 from binding and activating RIPK3 [78]. In humans, RIPK1 deficiency caused by rare homozygous mutations is associated with severe lymphopenia, recurrent infections, early-onset inflammatory bowel disease (IBD), and progressive polyarthritis [80]. Relative to healthy patients, IL-1β concentrations are increased in the supernatant of PHA-stimulated whole blood cells, or from LPS-stimulated CD14⁺ monocytes obtained from patients with homozygous mutations in RIPK1, leading to complete RIPK1 deficiency; these results mirror the increased inflammasome activity that is observed in RIPK1-deficient human monocyte-like THP-1 cells stimulated with LPS [80]. Increased cell death mediated by overt caspase-8 and RIPK3 activation in the absence of RIPK1 may promote NLRP3 inflammasome activation and cytokine release in primary murine macrophages, given that Pam3Cys-induced IL-1β release is prevented in Casp8^LysMCre bone marrow derived macrophages (BMDM) treated with nec-1 [86]. Furthermore, LPS-stimulated RIPK1^–/– THP-1 cells undergo increased necroptosis compared with RIPK1^+/+ THP-1 cells [80], further highlighting the importance of RIPK1 in inflammatory cell death.

In addition to their importance in sterile systemic inflammation, RHIM proteins also play a role in activating cell death pathways in microbial infections associated with CSS. Certain bacteria induce inflammatory cell death through their PAMPs or toxic effectors, such α-hemolysin and Hla, that are secreted into host cells. For example, S. aureus releases α-hemolysin and Hla, which activate necroptosis and pyroptosis via RIPK3 and NLRP3 signaling, resulting in increased inflammatory cytokine release and severe lung pathology in mice [87]. Similarly, Yersinia pestis, a causative agent for plague that has historically led to major pandemics, inactivates TAK1 with its effector protein YopJ, inducing cell death and inflammatory cytokine release in primary murine macrophages [28]. TAK1 is a prosurvival protein kinase that activates NF-κB and MAPK signaling pathways downstream of TLRs, the IL-1 receptor, and the TNF receptor, inducing the expression of a number of pro-inflammatory and pro-survival genes, many of which suppress the activation of caspase-8 and RIPK1 in mammals [88] (FIG. 2b). Upon inhibition of NF-κB signaling, a cytosolic complex containing RIPK1 is formed to drive caspase-8–mediated inflammatory cell death in various cell types, such as mammalian macrophages [83] (FIG. 2b). Moreover, pharmacological inhibition and genetic deletion of TAK1 have led to spontaneous caspase-1 cleavage in an NLRP3-dependent manner and cell death in murine macrophages [26, 27]. Genetic deletion of RIPK1 has enhanced MLKL phosphorylation (necroptosis) but abrogated caspase-1 and GSDMD cleavage (pyroptosis), and activation of caspase-3 and caspase-7 (apoptosis) during infection with Yersinia sp. [28]. However, the combined loss of caspase-8, RIPK3 and caspase-1 has completely protected against Yersinia-induced cell death and cytokine release in murine macrophages [28], highlighting the essential role of these molecules in the cytokine-cell death positive feedback loop.

During IAV infection, the TLR–TRIF axis, RIG-I signaling, and NLRP3 inflammasome, induce the production of type I IFN and pro-inflammatory cytokines, such as IL-1β and IL-18, in immune and non-immune cells [89–91]. Type I IFN subsequently upregulates the expression of the Z-nucleic acid sensor ZBP1 via IRF1. The Zα2 domain of ZBP1 senses replicating IAV Z-RNAs to assemble a signaling complex that leads to inflammatory cell death (FIG. 2c), resulting in cytokine release and ultimately, lung damage in mice [25]. Caspase-6 promotes binding of RIPK3 to ZBP1 to promote IAV-induced activation of the NLRP3 inflammasome and cell death in murine macrophages [92]. ZBP1 also drives inflammatory cell death, cytokine release, skin and colon inflammation, as well as perinatal lethality in mice under sterile inflammatory conditions [93–96]. Altogether, these findings are relevant in that they highlight the important role of ZBP1 in contributing to a CS under different disease conditions.

Whereas almost all pathogens are capable of inducing similar sets of cytokines, it is still unclear why only some bacteria or viruses induce a CS. One possibility is that certain pathogens possess virulence factors that act on RHIM-containing proteins to inhibit cell death, and thus prevent the positive feedback loop that leads to a CS [97, 98]. Indeed, one of the effectors of enteropathogenic Escherichia coli, EspL, induces proteolytic cleavage of RHIM proteins including RIPK1, RIPK3, ZBP1 and TRIF, possibly limiting necroptosis and inflammation in mice [97]. Similarly, numerous viruses have evolved mechanisms to inhibit cell death and prolong the survival of infected host cells [98].

Collectively, these studies suggest that RHIM-containing proteins can provide a crucial mechanistic link between the occurrence of a CS and cell death. Indeed, RIPK1 is a key molecule in the cytokine-cell death positive feedback loop that limits ZBP1- and RIPK3-mediated inflammatory cell death and cytokine release, thus preventing systemic inflammation and multiorgan damage, as observed in patients with CSS.

The role of TNF in cell death and the pathogenesis of a CS.

In addition to the RHIM-containing proteins, cytokines themselves and their corresponding receptors and signaling pathways are important regulators of cell death and in the CS. TNF receptor signaling is one of the best studied examples of this regulation [15]. TNF signals through two distinct receptors, TNFR1 and TNFR2, and signaling through TNFR1 promotes various biological processes including cell death [99]. Downstream of TNF receptor engagement, components of the linear ubiquitin chain assembly complex (LUBAC) — SHARPIN, HOIP and HOIL-1 — are required for optimal NF-κB signaling and cytokine secretion in mammals [100] (FIG. 2b). In the same pathway, the ubiquitin-editing enzyme TNFAIP3 (also known as A20) is crucial for inhibiting NFκB signaling [101]. However, deficiency in LUBAC components in mice (Hoil1^–/–) results in severe inflammation or lethality, mediated by TNF-dependent apoptosis and necroptosis [102]. The severe skin inflammation observed in mice lacking SHARPIN (Sharpin^cpdm mice) is prevented in Sharpin^cpdmTnf^–/– and Sharpin^cpdmCasp8^–/+Ripk3^–/– mice [99, 100]. MYD88 is required for the production of TNF in Sharpin^cpdm mice, and Sharpin^cpdmMyD88^–/– mice are also protected from systemic inflammation and skin disease [103]. The importance of innate immune-mediated inflammatory cell death mechanisms downstream of cytokine signaling in driving this CS-mediated pathology is further highlighted by the finding that Sharpin^cpdmNlrp3^–/–, Sharpin^cpdmCasp1/11^–/– or Sharpin^cpdmIl1b^–/– mice show delayed onset of inflammation compared with Sharpin^cpdm mice [104, 105]. By contrast, Hoil1^–/–Casp8^–/–Ripk3^–/– mice experience embryonic lethality [102], possibly owing to impaired fetal hematopoiesis that might be driven by deregulated cytokine production in the presence of RIPK1. Indeed, Hoil1^–/–Casp8^–/–Ripk3^–/– Ripk1^–/– mice are protected from embryonic lethality [102]. Similar to these mouse models, deficiency of HOIL-1 and HOIP in humans is associated with autoimmunity, inflammation, and hyperactive cytokine release [106, 107]. Patients with mutations or polymorphisms in HOIL1 or HOIP present with systemic inflammation and immunosuppression and are susceptible to recurrent opportunistic infections [106, 107]. In addition, monocytes derived from HOIL1deficient patients display hyperproduction of IL-6 upon IL-1β stimulation [106]. Also, A20^–/– (Tnfaip3^–/–) mice exhibit widespread tissue inflammation and perinatal lethality [108]. While most A20^–/– mice die within a day of birth, A20^–/–Ripk3^–/– mice survive 2–3 weeks before succumbing to inflammatory death [109], suggesting the potential involvement of other cell death pathways beyond necroptosis. Indeed, presentation of a spontaneous erosive polyarthritis phenotype characterized by synovial and periarticular inflammation with high numbers of infiltrating mononuclear cells is reduced in A20^myel-KONlrp3^–/– mice, implying a possible pathogenic role for A20 in pyroptosis [110]. Moreover, polymorphisms in human TNFAIP3 are correlated with increased risk of the autoimmune diseases SLE and psoriasis, as well as with inflammatory pathology associated with cystic fibrosis [111–113]. Overall, the components of LUBAC and A20 prevent systemic inflammation by limiting the cell death induced by TNF signaling, and the systemic inflammation and cell death observed in the absence of LUBAC and A20 can be rescued with the loss of PANoptosome components.

The physiological role of TNF signaling in triggering inflammatory cell death and CSS is further supported by in vivo studies in mice that lack caspase-8 or FADD. Mice lacking either of these molecules in intestinal epithelial cells (Casp8^ΔIEC and Fadd^ΔIEC) exhibit increased infiltration of immune cells, mostly myeloid cells, into the colonic mucosae and expression of proinflammatory cytokines such as IL-6, IL-1β, IFN-γ and TNF, leading to spontaneous intestinal inflammation [114, 115]. Intestinal epithelial cells of Fadd^ΔIEC mice are characterized by caspase-8 activation that sensitizes them to TNF signaling-induced necroptosis, which potentially drives acute inflammation and lethality in Fadd^ΔIEC mice [114, 115]. Indeed, Fadd^ΔIECTnf^–/– mice show mild focal epithelial lesions and immune cell infiltration in the colonic mucosa compared with control mice [114, 115]. The pathogenic role of TNF is further supported by the lethality observed in Casp8^ΔIEC mice injected intravenously with TNF, which is not lethal to wildtype mice [114]. Moreover, Fadd^ΔIECRipk3^–/– mice or Casp8^ΔIEC mice injected with nec-1 show improved intestinal pathology compared with Fadd^ΔIEC or Casp8^ΔIEC mice [114, 115]. However, Fadd^ΔIECTnf^–/– mice still exhibit inflammation when compared with wildtype mice, suggesting that TNF-independent mechanisms might also contribute to FADD deficiencyinduced pathology in mouse intestinal epithelial cells [115]. Likewise, ablation of RIPK3 in mice lacking FADD in keratinocytes (Fadd^E-KORipk3^–/–) fully prevents the development of skin lesions in Fadd^E-KO mice [116]. Although Casp8^–/–Ripk3^–/–, Fadd^–/–Ripk3^–/–, Casp8^–/–Mlkl^–/– and Fadd^–/– Mlkl^–/– mice are protected from embryonic lethality [117, 118], loss of MLKL does not prevent systemic and intestinal inflammation in mice that express catalytically inactive caspase-8 (Casp8^C362A/C362AMlkl^–/–) [30]. The systemic inflammation observed in Casp8^C362A/C362AMlkl^–/– mice is characterized by increased concentrations of various pro-inflammatory cytokines, including IL1β and IL-18 [30], which might imply pathological roles of the inflammasome and pyroptosis in these mice. Indeed, Casp8C362A/C362AMlkl–/–Casp1–/– or Casp8C362A/C362AMlkl–/–Asc–/– mice are rescued from lethality and systemic inflammation, indicating that caspase-8 functions as a master regulator of pyroptosis, apoptosis, and necroptosis [29, 30]. Children with early-onset IBD exhibit increased expression of RIPK3 and MLKL, but reduced caspase-8 expression in the inflamed ileum and colon relative to healthy controls [119]. Similarly, patients with inherited caspase-8 (CASP8) deficiency present with early onset intestinal inflammation [120]. These studies provide correlative evidence of a link between inflammatory cell death and IBD in the clinical setting [119, 120]. Overall, inflammatory cytokine signaling pathways are intricately linked with cell death pathways that can contribute to a CS.

The above discussion makes it evident that inflammatory cell death, induced by the modulation of cell death regulators, the sensing of PAMPs, DAMPs, or cytokines, can induce further cytokine secretion leading to physiological disorders such as systemic and local inflammation in the skin and intestine. A further example of the important role of cell death in driving a CS is provided by mutation of the proline-serine-threonine phosphatase-interacting protein 2 gene in mice (Pstpip2^cmo), which causes inflammatory lesions in the bones and various degrees of skin and paw inflammation that closely resemble the human disorder known as chronic recurrent multifocal osteomyelitis. Specifically, whereas neutrophils and IL-1β are crucial for the initiation of the disease in Pstpip2^cmo mice, joint inflammation and cytokine release are inhibited in Pstpip2^cmoRipk3^–/–Casp8^–/–Casp1/11^–/– mice [33, 121]. Overall, these studies may serve to provide a framework and basis for the mechanistic definition of a CS. Although the particular triggers, pathways and molecules involved can differ, each of these distinct responses may initiate a positive feedback loop between cytokine release and cell death pathways that can lead to a CS and thus, potentially to multiple organ failure and unfavorable disease prognosis [16].

Therapeutics to treat a CS

Thus far, treatment of a CS has been largely aimed at maintaining critical organ function by limiting the collateral damage caused by an activated immune system. Although general immunosuppression was initially considered as a potential strategy in CSS, a deeper mechanistic understanding of the pathways involved has highlighted the failures of general immunosuppression and suggested that more targeted therapies may be a superior approach. Corticosteroids, such as glucocorticoids and dexamethasone, act as anti-inflammatory and immunomodulatory agents by inhibiting the synthesis of pro-inflammatory cytokines and driving lymphocyte cell death [122]. However, in IAV-mediated severe pneumonia, a meta-analysis found that patients treated with glucocorticoids might present worse clinical outcomes than those not receiving glucocorticoid therapy -- including an increased risk of secondary bacterial infection and death [123]. In the case of COVID-19, dexamethasone has been shown to be effective only in patients requiring mechanical ventilation or supplemental oxygen in the randomized open-label phase2/3 RECOVERY trial in hospitalized patients with COVID-19 (primary outcome: all-cause mortality based on hazard ratio from Cox regression) (NCT04381936)^I [124]; this suggests that patient selection is crucial. Corticosteroids are generally avoided as frontline therapy for the treatment of CSS in patients receiving CAR T cell therapy. However, they are recommended for the treatment of neurologic adverse effects of T cell-engaging therapies, as well as in patients presenting with MAS and HLH [11, 125]. Overall, these findings suggest that while corticosteroids have been widely used across disease indications in the past, more targeted therapies should be considered to treat CSs.

To develop targeted therapeutics, a fundamental understanding of the biological processes and mechanistic details underlying a CS will be crucial. In this review, we discussed the connections between cytokine release and inflammatory cell death, and the cellular and molecular mechanisms of this positive feedback loop in a CS. These mechanistic insights should aid the development of novel therapeutic strategies for the numerous fatal conditions associated with a CS. Based on these insights, we review here existing CS blockers and therapeutics that are under development to target various aspects of the CS loop (Table 3).

Table 3 |.

Targeted therapeutics in cytokine storm-associated diseases in humans

TARGETED PATHWAY TREATMENTS
	Therapeutics	Anti-IL-1	Anti-IL-18	Anti-IL-6	Anti-TNF	Anti-IFN-γ	JAK inhibitors	Caspase inhibitors
CSS	Sepsis	[173]			[35, 174]
	COVID-19	[175, 176]		[177, 178]	[179–181]	[182]	[157, 183]
	sJIA	[184, 185]	[145]	[186]			[187]
	SLE						[188]
	Adult Still’s disease	[189]	[188]
	HLH			[190]		[131, 191]	[192]
	NLRC4-MAS		[145]
	Castleman’s disease			[193]
	GvHD			[194, 195]			[196]
	CAR T-cell therapy	[197]		[198]
	Blinatumomab
Autoinflammatory	Rheumatoid arthritis	[199]		[200]	[201]		[202]
	Giant cell arteritis	[203]		[149]
	Gout	[204–206]
	Crohn’s disease				[207, 208]	[209]	[210]
	Ulcerative colitis				[211]		[212]
	Psoriasis	[213]			[214]		[215, 216]
	CAPS	[217, 218]
	FMF	[219]
	NASH							[220]
	Refractory asthma				[221]
	Drugs	Anakinra Canakinumab Rilonacept	Tadekinig-α	Tocilizumab Siltuximab Sarilumab	Adalimumab Certolizumab Etanercept Golimumab Infliximab	Emapalumab	Baricitinib Tofacitinib Upadacitinib Ruxolitinib	Emricasan

Green: FDA-approved drugs (for any indication) with efficacy in the given setting

Blue: Mixed responses or ongoing clinical trials

White: N/A

Numbers represent references.

Anti-cytokine therapy.

Cytokines themselves play key roles in driving much of CSS pathology; therefore, treatments aimed at dampening pro-inflammatory signaling could improve clinical outcomes. Suppressing cytokine signaling by using recombinant proteins, inhibiting cytokine production, or by neutralizing cytokines directly with monoclonal antibodies, has shown promising effects in various clinical settings (Table 3).

Anti-TNF Antibodies

One of the central cytokines involved in CSS that has been carefully studied over the years is the pro-inflammatory cytokine TNF. IAV-infected mice treated with a neutralizing antibody against TNF exhibit reduced pulmonary immune cell infiltration and less severe pathology, as well as improved survival relative to controls [126, 127]. The pathological roles of TNF have also been demonstrated in a co-infection model, where the administration of staphylococcal enterotoxin B (SEB) to IAV-infected mice led to enhanced cytokine production and lethality compared to SEB or IAV alone, and the neutralization of TNF by dimeric TNFR Fc fragment antibodies prolonged survival by two days compared with PBS treatment [128]. These findings would be particularly relevant to patients with IAV infection who have a concurrent bacterial infection. Case studies have shown that patients on anti-TNF antibody therapy to treat certain rheumatic diseases, such as rheumatoid arthritis and SLE, who become infected with SARSCoV-2 tend to be less likely to be hospitalized, which suggests that prophylactic anti-cytokine therapy might be beneficial to treat certain cases of COVID-19 [129]. Administration of neutralizing antibodies against TNF has also improved survival in mouse models of poly(I:C)induced HLH and LPS-induced sepsis [23]. Unfortunately, this success has not translated to humans, with most anti-TNF antibody trials showing no clinical benefit for treating sepsis [35]. One of the key reasons behind the failure of anti-TNF antibody therapies in CSS is likely patient heterogeneity, not only in terms of site and source of infection but also with respect to genetic background and comorbidities, but further research is required to robustly address this issue.

Anti–IFN-γ Antibodies

Among several cytokines that are increased in the serum of patients with HLH, IFN-γ seems to be a particularly promising target based on results from mouse models. Neutralization of IFN-γ markedly prevents the development of HLH-like pathology in perforin- and Rab27a-deficient mice (Pfp^–/– and Rab27a^–/– mice) infected with LCMV-WE [60, 130], although the mechanisms by which excess IFN-γ leads to the characteristic histological and clinical features of HLH in these mice have not been defined. Despite this, recently, the US FDA approved anti–IFN-γ antibody emapalumab to treat patients with relapsed/refractory HLH, based on favorable results in a single arm, open label phase 2/3 trial of 34 pHLH patients treated with emapalumab in combination with dexamethasone and/or other HLH-directed therapies as needed (primary endpoint: overall response) (NCT01818492; NCT02069899)^II [131]. Also, neutralizing antibodies against IFN-γ have improved animal survival in a mouse model of LPS-induced sepsis [23]. Furthermore, anti–IFN-γ antibody treatment in LPS-challenged mice overexpressing human IL-6 (IL6-Tg) in an experimental model of MAS associated with sJIA, improved survival and reduced the amounts of several pro-inflammatory cytokines relative to IgG treated IL6-Tg mice [132]. Moreover, administration of IFN-γ has reduced hemophagocytosis in the liver of LCMV-infected Pfp^–/– and Rab27a^–/– mice compared with IgG treated mice, indicating that HLH can potentially be associated with the phagocytosis-promoting effect of IFN-γ [133]; however, more work is needed to thoroughly investigate this therapy option in patients with HLH.

Anti–IL-1β antibodies and IL-1 receptor antagonists

Cell death through pyroptosis leads to IL-1β release. Monoclonal antibodies against IL-1β, such as canakinumab, or recombinant human IL-1Ra, such as anakinra, are beneficial in individuals with autoinflammatory cryopyrinopathies. Case reports have revealed that patients can be effectively treated with recombinant IL-1Ra (receptor antagonist) in the case of CSS associated with sJIA, adult-onset Still’s disease, and SLE [134–136]. Currently, a phase 1 randomized placebo-controlled trial is underway to evaluate the efficacy of anakinra for the treatment of sHLH by comparing the number of acquired infections or deaths in the treatment and placebo groups (primary outcome) (NCT02780583)^III [137]. Canakinumab and anakinra are also being studied for treatment of COVID-19-induced ARDS and CSS [138]; for instance, a phase 3 randomized placebo controlled study is evaluating canakinumab in hospitalized patients with COVID-19 with the primary outcome being the number of patients with clinical response (NCT04362813)^IV, and a phase 2/3 2-arm study of anakinra vs standard of care in patients with CSS secondary to SARS-CoV-2 infection with a primary outcome of treatment success (NCT04443881)^V. In addition, anakinra has abolished CSs and neurotoxicity in humanized mice with high leukemia burden (recapitulating key features of a CS); this approach might potentially offer a therapeutic strategy to tackle neurotoxicity associated with CAR T-cell therapy [139]. Indeed, the long-term safety profile of anakinra has been well-established in patients with rheumatoid arthritis and severe cryopyrin-associated autoinflammatory syndrome (CAPS) [140]. IL-1Ras like anakinra might also be beneficial in treating diseases associated with IL-1α release, such as neutrophilic dermatosis [81], and efficacy has been observed in certain case reports [141].

Anti–IL-18 treatments

Pyroptosis executed by GSDMD pore formation triggers IL-18 release [43]. Patients with a CS caused by MAS have high amounts of IL-18 in their serum [142]. The pro-inflammatory effects of IL-18 are counterbalanced by its high-affinity endogenous antagonist, IL-18 binding protein (IL-18BP) [143]. Repeated challenge of mice deficient in IL-18BP (Il18bp^–/–) with a toll-like receptor 9 agonist, mirrors clinical manifestations of MAS, such as pancytopenia and hemophagocytosis in bone marrow smears; blocking IL-18 signaling using a monoclonal antibody against the IL-18 receptor rescued this phenotype [144]. In line with this mouse work, a case study reported that administration of tadekinig-α, a recombinant IL-18BP, reduced MAS flare frequency and severity in a patient with resistant sJIA and recurrent MAS [145]. However, further studies are needed to fully understand the utility of this candidate therapy in the clinical setting, and clinical trials are currently ongoing.

Anti–IL-6 receptor antibodies

Similar to IL-1β and TNF, upregulation of IL-6 involves the NF-κB signaling pathway, and IL-6 binding to its receptor induces robust pro-inflammatory signaling [146]. Tocilizumab, a humanized monoclonal antibody against IL-6R, has been reported to be effective in severe rheumatoid arthritis, sJIA, multicentric Castleman’s disease and CAR T cell-induced CSS [147–150]. In humanized mice with high leukemia burden, monocyte-derived IL-1β and IL-6 drive CSS and neurotoxicity induced by CAR T cells, potentially suggesting a role for anti–IL-1 and anti–IL-6 antibody therapy in combating these side effects [63, 139]. Anti–IL-6 antibody therapies have also been tested with mixed clinical responses in COVID-19. Whereas the phase 3 randomized placebo-controlled COVACTA (NCT04320615)^VI and Kevzara (NCT04327388)^VII trials in patients with severe COVID-19–associated pneumonia using tocilizumab and sarilumab, respectively, (anti-IL-6R antibodies) did not meet their primary or key secondary endpoints of improved clinical status and reduced patient mortality [151], the phase 3 randomized placebo-controlled EMPACTA (NCT04372186)^VIII trial of tocilizumab met its primary endpoint of a reduction in the number of patients requiring mechanical ventilation by day 28, but failed to improve mortality. In addition, anti-IL-6R antibody treatment in the phase 4 randomized, embedded, multifactorial adaptive platform REMAP-CAP trial (NCT02735707)^IX met its primary endpoint and improved survival in critically ill patients with COVID-19 [152]. The improved results in the REMAP-CAP study, however, might be due to the additional organ support that was provided for patients in the intensive care unit.

Combination therapies

The efficacy of targeting a single cytokine as described may be limited due to the continued production and signaling of other functionally redundant cytokines that contribute to immune activation and end-organ damage in CSS. Indeed, synergy between TNF and IFN-γ occurs upstream of inflammatory cell death in murine and human macrophages, and the consequent signaling cascade and cell death can lead to further release of other cytokines and alarmins [23]. Moreover, the combined administration of neutralizing antibodies against TNF and IFN-γ has improved survival more than single-cytokine blockade of IFN-γ or TNF in mouse models of poly(I:C)-induced HLH and LPS-induced sepsis [23]. The combined neutralization of TNF and IFN-γ has also protected mice from lethality during SARS-CoV-2 infection [23]. Another key example of the potential utility of cytokine combination therapy highlights IL-6 and IL-1 targeting: tocilizumab (anti-IL-6R antibody) fails to protect mice from CAR T cell-mediated neurotoxicity, whereas anakinra (recombinant human IL-1Ra) is uniquely associated with rescue from lethal neurotoxicity [139]; this suggests that anti–IL-6 and anti–IL-1 antibody combination therapies might be effective against CAR T cell-induced CSS and/or neurotoxicity, although this warrants further investigation.

Small molecules.

Several small molecules are available for inhibiting cytokine production and cytokine signaling, scavenging reactive oxygen species (ROS) and preventing cell death. These molecules have the potential to interrupt the CS feedback loop, and some, such as JAK inhibitors, have shown promising effects in both pre-clinical and clinical settings for the treatment of CSS (Table 3). Furthermore, as cytokines exhibit high transcriptional regulation, studies to characterize transcriptional regulators and their mechanisms of action in the production of specific cytokines involved in CSS have the potential to identify whether these molecules might be considered as candidate drugs and targets in therapeutic discovery.

JAK inhibitors

TNF and IFN-γ activate JAK–STAT signaling pathways that can lead to cell death through NO production [23]. JAKs also mediate signaling pathways of multiple other cytokines, including IFNs, IL-2, IL-3, IL-4, IL-6, IL-7, IL-9, IL-12, IL-10, IL-15, IL-21, GM-CSF and G-CSF [153, 154]. JAK–STAT pathways are required for many innate and acquired immune responses, and mutations of JAKs and STATs can predispose affected individuals to certain autoimmune diseases such as rheumatoid arthritis, vascular inflammatory diseases, and insulin resistance [154]. JAK inhibitors are available in an oral form and have good bioavailability, which makes these inhibitors potentially preferable to injectable medications, such as the anti-cytokine antibodies, for chronic and prolonged use. Because JAKs function as common downstream mediators in the signaling pathways of several cytokines, JAK inhibitors offer a broad spectrum of activity and are an appealing strategy to break the CS loop for the treatment of CSS. For example, ruxolitinib, a JAK inhibitor, can reduce tissue immunopathology and disease manifestations in murine models of pHLH and sHLH [155, 156]. Additionally, JAK inhibitors might be beneficial in the fight against COVID-19, as the JAK inhibitor baricitinib, when combined with the broad-spectrum antiviral drug remdesivir, was found to be superior to remdesivir alone in the phase 3 randomized controlled ACTT-2 trial in hospitalized patients with COVID-19 in reducing recovery time, based on a stratified log-rank test (primary outcome) and accelerated improvement in clinical status (NCT04401579)^X [157].

Antioxidants

Several studies suggest that ROS and NO contribute to CSS pathology [158]. For example, ROS produced by alveolar phagocytic cells in mice during IAV infection can induce immune cell infiltration into the lungs, thereby driving significant lung immunopathology; depletion of superoxide by injection of superoxide dismutase protects mice against lethal IAV infection, indicating the potential of antioxidant therapy in treating this infection [159]. Edaravone is an US FDA-approved antioxidant for the treatment of patients with amyotrophic lateral sclerosis, a fatal neurodegenerative disease [160], which might be effective in CSS, although to our knowledge, this remains to be tested.

Activated macrophages express high amounts of inducible nitric oxide synthase (iNOS)-- an enzyme that catalyzes the production of NO [23]. Excess NO is harmful to tissues and causes vasodilation and hypotension -- common clinical features of CSS requiring vasopressor administration [161]. Thus, NOS inhibitors reduce hypotension, organ dysfunction, and mortality in murine models of sepsis, including LPS- and TNF-induced shock [162]. However, a randomized, placebo-controlled phase 3 trial of the NOS inhibitor 546C88 in patients with sepsis found that this treatment positively impacted blood pressure and vascular resistance but also increased mortality compared with the control group, leading to premature termination of the study [163]. Although conjectural, one of the possible reasons for this failure might be the nonselective nature of the NOS inhibitor used in the study. NO can be generated by three isoforms of NOS — iNOS, endothelial NOS and neuronal NOS. iNOS can be induced by pathogens and cytokines and contributes to the pathology of various diseases [161]. In animals, most studies using selective NOS inhibitors have shown better protective effects compared with the use of nonselective NOS inhibitors. Indeed, the transgenic overexpression of eNOS in mice (eNOS Tg) improves survival during LPS shock, underscoring the need for NOS specificity in appropriately determining clinical outcomes during sepsis [164]. However, in contrast to the benefits of specific iNOS inhibitors, iNOS-deficient mice (Nos2^–/–) are not protected against sepsis [165]. The conflicting results for iNOS inhibitors versus iNOS deficiency might be explained by the concentrations of NO produced in cells and tissues, although this remains to be tested. Reducing iNOS activity might be more beneficial than complete inhibition, with the assumption that residual NO might exert essential protective functions. Overall, these dichotomous roles of NO have proved to be the biggest challenge in the development of potential antioxidant therapies to treat CSS.

Caspase inhibitors

One of the major clinical manifestations of CSS is tissue damage [16]; therefore, protecting organs against the damage triggered by the CS is a relevant therapeutic strategy. TNF and IFN-γ activate caspases-3, 7, 8 and 9 in mouse macrophages, and inhibition of caspase-8 suppresses TNF- and IFN-γ-induced cell death in mouse macrophages and mortality in mice subjected to LPS-induced sepsis [23]. Caspase inhibition can improve organ function and/or survival in various mouse models of ischemia-reperfusion injury, burns, endotoxemia, and sepsis [166]. Administration of the pan-caspase inhibitor zVAD inhibits cell death in thymocytes to improve survival of mice and rats in a cecal ligation and puncture models of sepsis [19, 166]. Emricasan is also an orally available pan-caspase inhibitor which has been tested in a completed randomized, placebo-controlled phase 2 trial in patients with nonalcoholic steatohepatitis cirrhosis (primary outcome: measure of change in alanine aminotransferase from baseline to day 28 (NCT02077374))^XI [167], and emricasan might be considered for clinical trials in other diseases associated with CSS. However, while many have attempted to develop caspase inhibitors, caspases are historically difficult to target [168], suggesting that focusing on molecules further upstream in the cell death signaling pathway might be a superior strategy.

Concluding remarks

Innate immune cells such as macrophages, dendritic cells, and NK cells, release pro-inflammatory cytokines in response to infectious and sterile insults. The initial inflammatory response is crucial to clear the infection; however, a dysregulated inflammatory response can cause a CS, potentially leading to tissue and organ damage, and mortality. CS can also occur in response to some therapeutic interventions and underlying conditions during cancer and autoinflammation. Here, we reviewed the molecular mechanisms connecting cytokine release, cytokine signaling, and innate immune-mediated cell death in a positive feedback loop to molecularly define a CS (FIG. 3). An improved understanding of this feedback mechanism, including the specific roles of cytokines involved, how immune cells trigger cell death, and the molecules that constitute the cell death signalosomes, is fundamental to developing candidate therapeutic strategies to ideally and precisely target a CS in a given individual (see Outstanding Questions).

Key Figure, Figure 3 |. Schematic representation of the mechanism of a cytokine storm.

During pathogenic infections, autoinflammatory diseases, and other cytokine storm (CS)-inducing conditions, innate immune cells from humans and mice become activated and release multiple pro-inflammatory cytokines, including TNF and IFN-γ [16]. The synergistic response of TNF and IFN-γ can induce inflammatory cell death, which can result in more cytokines being produced and which might ultimately lead to a CS, with potential tissue and organ damage, and mortality in humans and mice [23]. Some components of Figure 3 were derived from BioRender.com.

Outstanding Questions.

• Currently, a cytokine storm (CS) is often treated in a disease-specific manner that depends on the underlying trigger. However, there may be advantages to finding targets that are independent of the trigger to provide a more global strategy for CS management. What therapeutic targets can be used to prevent CSs regardless of the underlying cause?

• What is the therapeutic window for anti-cytokine therapies in a CS? How does changing the timing of the treatment impact the course of the disease?

• TNF and IFN-γ drive cell death, and blocking these cytokines can provide benefits in murine models of CS syndrome, but not a full cure for all cases. What is the contribution that cytokines other than TNF and IFN-γ could have in mediating cell death and a CS?

• How does cell death affect the release of IL-6? Can neutralization of TNF and IFN-γ inhibit the production of IL-6, and prevent a therapeutic intervention-induced CS? Can neutralization of TNF and IFN-γ be used to treat or prevent global CS-associated diseases?

• The absence of caspase-8 leads to cell death in vivo, but the caspase-8 node is also essential for inducing various forms of inflammatory cell death in response to PAMPs or DAMPs. How does the same molecule regulate two contrasting phenotypes?

• Since dexamethasone dampens NF-κB signaling necessary to produce cytokines and JAK inhibitors prevent IFN signaling, can the combination of dexamethasone with a JAK inhibitor mimic conditions similar to neutralization of TNF and IFN-γ to prevent a CS?

Based on the molecular mechanisms linking inflammatory cell death and the CS, it is clear that TNF and IFN-γ are of paramount importance; they are up-regulated in almost all cases of CSs [16], and their synergism induces inflammatory cell death pathways in myeloid cells [23]. Neutralization of these two cytokines can provide superior protection in pre-clinical mouse models of CSS [23], suggesting that combined re-purposing of these therapies might potentially be beneficial in patients with CSS, ideally including those presenting with COVID-19. However, determining the therapeutic window of anti-cytokine therapies is challenging. The paradigm of “hit hard and hit early,” popularized for the treatment of AIDS [169], might also work in the treatment of a CS. Indeed, early release of TNF is a crucial driver of lethality caused by SEB in mice, whereas late release does not play much of a role [170]. Various anti-cytokine therapy results from clinical trials have failed to meet the high expectations raised in preclinical models, most likely owing to the short half-life of most cytokines and the narrow therapeutic windows that have been possible. Further research should also focus on the identification of drugs targeting the upstream innate immune sensors and cell death signaling pathway molecules that can be used across CSS regardless of underlying causes. Because of the interconnection between inflammatory cell death and the CS, such identification might be achieved by looking outside the classical CSS repertoire and considering therapeutic strategies that target cell death.

Highlights.

• A Cytokine storm (CS) is a clinically relevant condition that has been associated with several life-threatening diseases.

• Although CS syndrome-inducing agents can fundamentally differ, the interactions of PAMPs or DAMPs with PRRs evoke similar cytokine profiles in mammals.

• Cytokines are intricately linked to cell death mechanisms in mammals and are involved in a positive feedback loop whereby cytokine signaling or PAMP/DAMP sensing causes inflammatory cell death that facilitates further pathogenic cytokine release, culminating in a CS to drive severe, life-threatening damage to host tissues and organs.

• Neutralization of TNF and IFN-γ can prevent cell death and inhibit the occurrence of a CS in mouse models of SARS-CoV-2 infection, hemophagocytic lymphohistiocystosis, and sepsis.

• A Combination of FDA approved drugs might be more beneficial than single treatment approaches for the clinical management of CSs.

Glossary

Acute respiratory distress syndrome (ARDS)

life-threatening lung injury caused by low oxygen concentrations in the blood as a result of fluid build-up in alveoli.

Alarmins

Damage-associated molecular patterns such as HMGB1 or IL-1α; released by damaged or necrotic cells.

Autoinflammation

Caused by the overactivation of the innate immune system with little or no evidence of a specific adaptive immunity, such as auto-reactive T cells or autoantibodies.

Chimeric antigen receptor (CAR) T cells

T cells that have been engineered with a specific receptor protein to enable targeting of a specific protein. They typically combine antigen-binding and T-cell activating functions into a single receptor.

Chronic obstructive pulmonary disease (COPD)

chronic inflammatory lung disease causing obstructed airflow from the lungs.

Chronic recurrent multifocal osteomyelitis

rare idiopathic autoinflammatory bone disease that mostly affects children and adolescents.

Cryopyrin-associated autoinflammatory syndrome (CAPS)

rare hereditary inflammatory disorder related to a gain-of-function mutation in the protein cryopyrin (NLRP3). CAPS encompasses a continuum of three phenotypes: familial cold autoinflammatory syndrome, Muckle-Wells syndrome, and neonatal-onset multisystem inflammatory disease.

Cryopyrinopathies

group of rare autoinflammatory diseases: includes familial cold autoinflammatory syndrome, Muckle-Wells syndrome and chronic infantile neurologic cutaneous articular syndrome (neonatal-onset multisystemic inflammatory disease).

Early-onset periodic fever syndrome

disease characterized by recurrent spontaneous attacks of multi-systemic inflammation in children.

Gasdermin

conserved family of proteins among vertebrates comprised of six members in humans, gasdermin A, -B, -C, -D, -E (DFNA5) and DFNB59.

Hemophagocytic lymphohistiocystosis

Hemophagocytosis is the process by which activated macrophages or phagocytes engulf erythrocytes, leukocytes, platelets and their precursors. Hemophagocytic lymphohistiocystosis is a life-threatening hyper-inflammatory condition occurring when histiocytes and lymphocytes become overactive and attack the body.

Linear ubiquitin chain assembly complex (LUBAC)

E3 ubiquitin ligase complex composed of HOIL-1, HOIP, and SHARPIN that generates linear polyubiquitin chains and stabilizes TNF-R1 signaling to regulate the NF-κB pathway.

Lymphadenopathy

enlargement or swelling of one or more lymph nodes from several potential causes, including infection, autoimmunity, and malignancy.

Macrophage activation syndromes (MAS)

severe and potentially fatal complication of several rheumatic diseases characterized by uncontrolled activation of macrophages and T lymphocytes.

Multicentric Castleman’s disease

rare disorder involving an overgrowth of cells in multiple lymph nodes throughout the body; associated with human herpes virus type 8 (HHV-8) and HIV-1.

Necroptosis

form of regulated cell death; executed by RIPK3 and MLKL upon caspase inhibition.

PANoptosis

unique, physiologically relevant inflammatory programmed cell death pathway regulated by the PANoptosome; it has key features of pyroptosis, apoptosis, and necroptosis.

PANoptosome

complex assembled under certain conditions to provide a molecular scaffold for contemporaneous engagement of key molecules involved in pyroptosis, apoptosis, and necroptosis.

Pathogen-associated or damage-associated molecular patterns (PAMPs or DAMPs)

PAMPs are small molecular motifs conserved within a class of microbes. DAMPs are produced or released by damaged and dying cells. Both PAMPs and DAMPs are recognized by pattern recognition receptors in host cells to activate innate immune responses.

Pattern recognition receptors (PRRs)

Germline-encoded molecules recognizing conserved molecular structures of microbial origin called PAMPs. They also sense endogenous molecules produced by, or released from damaged and dying cells called DAMPs.

RIP homotypic interaction motif (RHIM)

short non-globular sequence stretch that is necessary for the recruitment of RIPK1 and RIPK3 by ZBP1. This motif is also found in TRIF.

Ripk1^D325A/D325A

Mice with this mutation are resistant to the caspase-mediated cleavage of RIPK1 that inhibits necroptosis; mice are hypersusceptible to inflammation and are embryonically lethal.

Severe combined immunodeficiency

group of rare disorders caused by mutations in different genes involved in the development and function of immune cells; patients suffer recurring infections with bacteria, fungi, and viruses.

Systemic juvenile idiopathic arthritis (sJIA)

rare autoinflammatory disease leading to inflammation in the joints and other parts of the body. It affects only 10% to 15% of children with JIA.