Mechanistic insights from inflammasome structures

Abstract

Inflammasomes are supramolecular complexes that form in the cytosol in response to pathogen-associated and damage-associated stimuli, as well as other danger signals that perturb cellular homoeostasis, resulting in host defence responses in the form of cytokine release and programmed cell death (pyroptosis). Inflammasome activity is closely associated with numerous human disorders, including rare genetic syndromes of autoinflammation, cardiovascular diseases, neurodegeneration and cancer. In recent years, a range of inflammasome components and their functions have been discovered, contributing to our knowledge of the overall machinery. Here, we review the latest advances in inflammasome biology from the perspective of structural and mechanistic studies. We focus on the most well-studied components of the canonical inflammasome — NAIP–NLRC4, NLRP3, NLRP1, CARD8 and caspase-1 — as well as caspase-4, caspase-5 and caspase-11 of the noncanonical inflammasome, and the inflammasome effectors GSDMD and NINJ1. These structural studies reveal important insights into how inflammasomes are assembled and regulated, and how they elicit the release of IL-1 family cytokines and induce membrane rupture in pyroptosis.

Introduction

The discovery of cytosolic complexes known as inflammasomes, which promote the release of pro-inflammatory IL-1 family cytokines from cells, was a milestone event in our understanding of fever and inflammation. This key discovery built on several previous findings, including cloning of the cDNAs for IL-1α and IL-1β in 1984, and the realization that the predicted IL-1α and IL-1β proteins are larger than those that were biochemically purified from macrophage supernatant and do not have signal sequences that would account for their secretion from cells^1,2. Soon after, the amino-terminal sequence of mature IL-1β was reported³, confirming that a cleavage event is required for its generation. In 1992, the enzyme responsible for IL-1β cleavage was first reported as the IL-1β converting enzyme (ICE)^4,5, later renamed caspase-1 — the first member of the caspase family of cysteine proteases that mediate apoptotic and inflammatory pathways^6–8. In 2002, the pursuit for the molecular machinery that activates caspase-1 led to the identification of the first inflammasome — the nucleotide-binding domain (NBD)-containing, leucine-rich repeat (LRR)-containing and pyrin domain-containing protein 1 (NLRP1) inflammasome — by the group of the late J. Tschopp⁹. At the time, the NLRP1 inflammasome was shown as a protein complex containing NLRP1, ASC (apoptosis-associated speck-like protein containing a caspase recruitment domain (CARD)), caspase-1 and caspase-5, but more recent data have revealed additional complexity (see below).

Rapid progress was made in the new field of inflammasomes in the subsequent years, with important clinical implications. It had previously been discovered that mutations in a relative of NLRP1 called cryopyrin (now known as NLRP3) caused a cluster of hereditary fever syndromes collectively referred to as cryopyrin-associated periodic syndromes (CAPS)^10,11. The Tschopp group then showed that NLRP3 assembles an inflammasome with increased activity in patients with CAPS¹² or gout¹³, prompting trials of the IL-1 receptor antagonist anakinra in these patients. Anakinra provided immediate and effective relief to patients with acute gout flares¹⁴. Patients with CAPS also responded remarkably well to anakinra¹⁵, and IL-1-blocking agents have since remained the mainstay of the clinical management of CAPS.

These early studies defining the NLRP1 and NLRP3 inflammasomes and the first demonstration of CAPS as an inflammasome-driven human disease (an ‘inflammasomopathy’) paved the way for subsequent studies that characterized new inflammasome sensor molecules (for example, neuronal apoptosis inhibitory protein (NAIP)–NBD-containing, LRR-containing and CARD-containing protein 4 (NLRC4), NLRP6, absent in melanoma 2 (AIM2), pyrin and CARD8), new inflammasome effector proteases (caspase-4, caspase-5 and caspase-11), new inflammasome substrates that orchestrate cell death and alarmin release (for example, gasdermins) or sculpt tissue immune responses (for example, IL-18 and IL-37), and, most recently, the role of ninjurin 1 (NINJ1) in mediating final membrane rupture downstream of inflammasome activation (Fig. 1). In addition, a vast literature now documents hundreds of genetic and acquired inflammasomopathies, which has fuelled the development of next-generation, biologic IL-1 blockers for clinical use (rilonacept and canakinumab) and the preclinical and clinical development of small-molecule inhibitors of individual inflammasome components.

Fig. 1 |. The canonical and noncanonical inflammasome pathways.

The figure summarizes the canonical inflammasomes comprising NLRP1, NLRP3, NLRP6, NAIP–NLRC4, AIM2 and pyrin, the noncanonical inflammasomes formed by caspase-4 and caspase-5 (human) and caspase-11 (mouse), and the adaptor and effector proteins associated with these pathways, notably the adaptor ASC and the effectors caspase-1, IL-1 family cytokines, GSDMD and NINJ1. Shown are the major protein domains involved in structural organization and function. For more detailed descriptions of each pathway, see the main text. AIM2, absent in melanoma 2; ASC, apoptosis-associated speck-like protein containing a CARD; BIR, baculoviral inhibitor of apoptosis repeat; CARD, caspase recruitment domain; DPP, dipeptidyl peptidase; dsDNA, double-stranded DNA; GSDMD, gasdermin D; HIN, haematopoietic interferon-inducible nuclear; LLPS, liquid–liquid phase separation; LPS, lipopolysaccharide; LRR, leucine-rich repeat; NACHT, nucleotide-binding and oligomerization domain; NAIP, neuronal apoptosis inhibitory protein; NBD, nucleotide-binding domain; NINJ1, ninjurin 1; NLRC, NAIP–NBD-containing, LRR-containing and CARD-containing protein; NLRP, NBD-containing, LRR-containing and PYD-containing protein; PYD, pyrin domain; UVB, ultraviolet B radiation; VbP, Val-boro-pro (DPP8 and DPP9 inhibitor).

In this Review, we summarize our current understanding of the structures of inflammasome components and how these structures relate to their functions — focusing on NAIP–NLRC4, NLRP3, NLRP1, CARD8, gasdermin D (GSDMD) and NINJ1, for which recent discoveries make such review timely. These structures provide insights into how diverse protein conformational changes determine the assembly and regulation of NAIP–NLRC4 and NLRP3 inflammasomes, how autoproteolysis of NLRP1 and CARD8 enables their unique mode of activation and regulation, and how GSDMD and NINJ1 oligomerization in the membrane offers complementary mechanisms of cytokine release and membrane disruption. We do not discuss in depth several other important inflammasome sensors, including AIM2, pyrin and NLRP6. Structural studies on AIM2 are represented by earlier studies on its mode of interaction with double-stranded DNA (dsDNA)¹⁶ and on the pyrin domain (PYD)^17–19. Although several structures exist for certain domains in pyrin²⁰, and NLRP6 has been shown to form biomolecular condensates upon interaction with dsRNA and a bacterial cell wall component²¹, further structural studies are required to elucidate their mechanisms of action. Furthermore, although we do not discuss these in detail, the observed differences between human and rodent inflammasomes require additional investigation. The biological aspects of inflammasome functions in health and disease have been reviewed elsewhere²².

What makes an inflammasome?

Inflammasomes are defined as supramolecular complexes that activate inflammatory caspases; the canonical inflammasomes contain and activate caspase-1, and the noncanonical inflammasomes contain and activate caspase-4 and caspase-5 in humans and caspase-11 in mice^23–28. Caspase-1, and now also caspase-4, have been shown to process IL-1 family cytokines (such as IL-1β and IL-18) to their mature forms^29–32. All inflammatory caspases (caspase-1, caspase-4, caspase-5 and caspase-11) process GSDMD to induce membrane pore formation, which is responsible for releasing IL-1 family cytokines from the cell^33–35 and, together with NINJ1, for inducing a highly inflammatory form of cell death known as pyroptosis^36–45.

Inflammasomes carry out intracellular surveillance as part of the innate immune system by detecting danger signals that originate from invading microorganisms or from cellular stress or damage, defending the host and restoring cellular homoeostasis through the effects of pro-inflammatory cytokines and cell death^28,46,47 (Fig. 1). Inflammasomes have been associated with numerous human diseases, from diseases of major organs, such as the heart, liver, kidney and lung, to infection, obesity, ageing and neuroinflammation^48–53.

Canonical inflammasome components

Conceptually, the core proteins of the canonical inflammasome have three main roles: a sensor protein that is triggered by infectious or sterile danger signals resulting in its oligomerization, an adaptor protein that is recruited by the sensor upon activation, and the effector caspase-1 that executes downstream responses^27,46 (Fig. 1). Inflammasome sensors are germline encoded. However, unlike other germline-encoded innate immune receptors such as Toll-like receptors (TLRs), which directly recognize pathogen-associated molecular patterns (PAMPs) or damage-associated molecular patterns (DAMPs)⁵⁴, only some inflammasome sensors, such as NAIP, AIM2 and NLRP6, directly interact with PAMPs or DAMPs. Other inflammasome sensors, such as NLRP1 and NLRP3, detect pathogen or damage signals indirectly from their resulting cellular perturbations, which may function to discriminate between harmful and harmless stimuli without necessarily distinguishing non-self from self²⁷.

Most inflammasome sensors belong to the NBD-containing and LRR-containing protein (NLR) family^23,55, which comprises multi-domain proteins characterized by a central nucleotide-binding and oligomerization domain (the NACHT domain) that can bind nucleotides (ADP or ATP) and participate in conformational changes and self-oligomerization upon activation. Twenty-two human NLRs are described, of which about nine are reported to assemble inflammasomes⁵⁵. The NLRs are classified in subfamilies according to their amino-terminal domains: NLRPs (for example, human NLRP1, NLRP2, NLRP3, NLRP6, NLRP7, NLRP9, NLRP10 and NLRP12) have an N-terminal PYD, whereas NLRBs (also known as NAIPs, for example, human NAIP) have N-terminal baculoviral inhibitor of apoptosis repeat domains. The C-terminal domain of NLRs, with the exception of NLRP10 and NLRP1, contains an LRR domain that senses ligands or cofactors, or interacts with itself; the LRR domain for NLRP1 is in between domains, whereas NLRP10 lacks the LRR domain. Additional notable non-NLR inflammasome sensors are as follows: AIM2, which contains an N-terminal PYD and a C-terminal haematopoietic interferon-inducible nuclear (HIN) domain that binds cytosolic dsDNA^56,57; pyrin (also known as marenostrin or TRIM20), which contains a PYD, a B-box-type zinc finger domain, a coiled-coil domain and a B30.2 (PRYSPRY) domain⁵⁸; and CARD8, which contains a C-terminal CARD.

Upon ligand binding, resulting in activation and oligomerization, NLRP3, NLRP6, AIM2 and pyrin all recruit the adaptor protein ASC through PYD–PYD interactions^19,58–60 (Fig. 1). The structure of the ASC PYD filament was resolved in 2014 (ref. 61), and several other PYD filament structures were revealed subsequently, the latest in 2022 (refs. 17,19,62–64), confirming their similar structural architectures. For example, the PYD of AIM2, which is autoinhibited by the HIN domain based on intramolecular electrostatic interactions¹⁸, is released by the interaction of the HIN domain with dsDNA from viral or bacterial infection¹⁶. It was modelled that HIN domains of multiple AIM2 molecules could wrap around dsDNA, positioning the linked PYDs on the HIN–dsDNA filament to recruit ASC^17,19. In addition to a PYD, ASC also has a C-terminal CARD⁶⁵ that recruits caspase-1 through CARD–CARD interactions^66,67. It is now well established that both PYD–PYD and CARD–CARD interactions generate helical assemblies to mediate homo-oligomerization and hetero-oligomerization^59,67. Thus, ASC bridges PYD-containing upstream sensors to the downstream effector caspase-1, forming an aggregate assembly known as the ‘ASC speck’ or the ‘inflammasome speck’^68–70.

NLRP1 has a C-terminal CARD that recruits ASC via CARD–CARD interactions, and ASC in turn recruits caspase-1. By contrast, CARD8 directly recruits and activates caspase-1 via CARD–CARD interactions without requiring the adaptor ASC^71,72. The NAIPs, upon activation by direct binding to bacterial ligands such as flagellin, recruit the CARD-containing adaptor protein NLRC4 through NACHT–NACHT interactions in which one NAIP molecule induces conformational change and polymerization of NLRC4 to form an inflammasome disc^73–78 (Fig. 1). The CARD-containing NLRC4 can directly recruit and activate caspase-1 through CARD–CARD interactions for cell death induction, but recruitment of ASC is required for maximal pro-IL-1β processing by NAIP–NLRC4 inflammasomes^74,75,79,80.

Activation of caspase-1

Caspase-1, which is the most well-studied of the inflammatory caspases, is the effector protease of the canonical inflammasomes (Fig. 1). Caspase-1, together with the other inflammatory caspases, consists of an N-terminal CARD followed by a CARD domain linker (CDL) that connects the CARD to the protease domain. The protease domain is composed of large (~20 kDa) and small (~10 kDa) catalytic subdomains separated by an interdomain linker (IDL). Both the CDL and IDL are susceptible to proteolysis. Caspase-1 is expressed endogenously as inactive monomers. Through recruitment by the adaptor protein ASC or NLRC4, or the sensor protein CARD8, caspase-1 monomers cluster within the inflammasome platform, which induces the dimerization of their protease domains. Caspase-1 dimerization enables caspase-1 to form the full caspase catalytic domain and gain the capacity to self-cleave. Caspase-1 cleaves itself at several sites within the IDL⁸¹ to stabilize the protease conformation and generate a dimeric species called p33/p10 (ref. 82). Caspase-1 p33/p10 is the fully active enzyme that enables the inflammasome to cleave substrates such as pro-IL-1β, pro-IL-18 and GSDMD⁸³. To terminate inflammasome signalling, the active caspase-1 p33/p10 protease self-cleaves at the CDL to release the p20/p10 protease species into the cytosol, leading to dimer dissociation and shutdown of caspase-1 activity^82,83.

Noncanonical inflammasome components

The mouse noncanonical inflammasome activates caspase-11 (Fig. 1), whereas the human noncanonical inflammasomes activate caspase-4 or capase-5 (refs. 84–86). Caspase-4, caspase-5 and caspase-11 are inflammatory caspases with a similar domain structure to caspase-1 and, like caspase-1, they are expressed in resting cells as inactive monomers that require dimerization to acquire protease activity^29,87. However, unlike the caspase-1-containing canonical inflammasomes, the noncanonical inflammasomes are lipid–protein assemblies in which the CARDs of caspase-4, caspase-5 and caspase-11 interact directly with bacterial lipopolysaccharide (LPS) or endogenous oxidized phospholipids to facilitate the activation of their latent protease function^84–86. Although the stoichiometry of ligand to caspase within this assembly is unclear, it is expected that noncanonical inflammasome complexes induce the formation of caspase dimers and, perhaps, higher-order oligomers. Dimerization of caspase-4, caspase-5 and caspase-11 confers the capacity to self-cleave at the IDL to generate p32/p10 dimers with maximal protease activity^29,87,88. Whereas caspase-11 cannot cleave pro-IL-1β, recent studies suggest that self-cleavage of caspase-4 at IDL residue D289 generates a p34/p9 protease species that directly cleaves pro-IL-1β, albeit weakly²⁹. Caspase-4 can also process pro-IL-18, as shown by multiple studies^30–32. The recent cryo-electron microscopy (cryo-EM) and crystal structures of caspase-4 in complex with pro-IL-18 have revealed a two-prong interaction mode in which the binding site outside of the active site (known as the exosite) strengthens and orients the binding site at the active site to enable proteolytic processing^89,90. Caspase-4, caspase-5 and caspase-11 cleave GSDMD to generate transmembrane pores to initiate pyroptosis^37–42,91 or pyroptosis-associated expulsion of neutrophil extracellular traps^92,93; in so doing, these caspases induce a second wave of inflammasome signalling via NLRP3 activation by GSDMD-mediated membrane damage and K⁺ efflux.

NAIP–NLRC4 inflammasomes

NAIPs are cytosolic receptors for various bacterial protein ligands, resulting in recruitment of the adaptor protein NLRC4 to form NAIP–NLRC4 inflammasomes^{74,75,94–97} (Fig. 2a). There are seven NAIP paralogues in mice (NAIP1 to NAIP7) but only one NAIP in humans. Different mouse NAIPs are specific for different bacterial ligands, among which NAIP5 and NAIP6 recognize bacterial flagellin, the major protein in the bacterial flagellum, and NAIP1 and NAIP2 recognize needle and inner rod proteins, respectively, of the type III secretion system. Although human cells have only one NAIP, they have been shown to respond to both flagellin and needle protein of the type III secretion system, perhaps depending on different splicing isoforms^95,98.

Fig. 2 |. The NAIP–NLRC4 inflammasome.

NAIP5, a typical NAIP, and flagellin, a NAIP5 ligand, are used as examples. a, Domain architectures of NAIP5 and NLRC4, with the domain colours matching those shown in part b. Domains that are shown in grey indicate that the structure of this specific region is not shown in part b. b, Formation of the NAIP5–NLRC4 inflammasome. Step 1: the unliganded NAIP5 (Protein Data Bank [PDB:7RAV]) and inactive NLRC4 [PDB:4KXF] at resting state, shown as a weak dimer complex. NAIP5 and NLRC4 can also be monomers before activation. Step 2 and step 3: the interaction between bacterial flagellin and NAIP5 induces the activation of NAIP5 [PDB:5YUD] and NLRC4 [PDB:4KXF], together with conformational changes. The grey NLCR4 molecule in step 3 represents the inactive state before its conformational changes to the active state. Step 4 to step 6: the activated NLRC4 self-propagates [PDB:6B5B], building up an inflammasome disc [PDB:3JBL]. Domains are colour coded as in part a. NAIP2 follows the same mechanism in forming an inflammasome disc. BIR, baculoviral inhibitor of apoptosis repeat; CARD, caspase recruitment domain; HD, helical domain; ID, intermediate domain; LRR, leucine-rich repeats; NAIP, neuronal apoptosis inhibitory protein; NBD, nucleotide-binding domain; NLRC, NAIP–NBD-containing, LRR-containing and CARD-containing protein; NTD, N-terminal domain; WHD, winged helix domain.

The mechanism by which NAIP–NLRC4 inflammasomes are assembled and activated has been elucidated from a series of structural and biochemical studies. When the crystal structure of NLRC4 was first solved, it showed an autoinhibited, closed conformation⁷⁶ (Fig. 2b). Subsequently, reconstitution of the complex of NAIP2 and NLRC4 with the rod protein PrgJ of Salmonella enterica subsp. enterica serovar Typhimurium suggested that each disc-shaped inflammasome contains one copy of NAIP2 and PrgJ but multiple copies of NLRC4; cryo-EM structures of 10-fold, 11-fold or 12-fold averaged discs contained mostly NLRC4 (refs. 73,77). The active conformation of NLRC4 in these structures revealed a rotation of close to 90° within the NACHT domain, between its first two subdomains (NBD and helical domain 1 (HD1)) and its last two subdomains (winged helix domain (WHD) and helical domain 2 (HD2)), together with the LRR domain, to open up the structure. This conformational change in NLRC4 seemed to be initiated by binding to PrgJ-bound NAIP2, and it was successively propagated to other NLRC4 molecules in the disc by nucleated polymerization (Fig. 2b).

The conformation of ligand-bound NAIP5 was revealed by cryo-EM structures of a flagellin-bound NAIP5 in complex with an NLRC4 mutant that does not self-associate⁹⁹, and of the wild-type flagellin–NAIP5–NLRC4 disc solved without applying symmetry¹⁰⁰. Flagellin was shown to bind in a deep pocket formed by multiple domains of NAIP5, a strategy that was suggested to limit the potential for immune evasion by the pathogen¹⁰⁰. Ligand binding seemed to open up NAIP5 into an active conformation similar to the active NLRC4 structure. However, a recent cryo-EM structure of ligand-free NAIP5 revealed an unprecedented open structure that was accessible to recruit inactive NLRC4 (ref. 78) (Fig. 2b). Ligand binding to NAIP5 induces a small rotation in the WHD of the NACHT domain that creates a steric clash with inactive NLRC4 and places a loop for optimal interaction with active NLRC4, thus both triggering and stabilizing the conversion of NLRC4 to the active state (Fig. 2b).

The NLRP3 inflammasome

NLRP3 is expressed in immune cells such as macrophages, dendritic cells and neutrophils¹⁰¹, but at a low level under resting conditions that is not sufficient to initiate NLRP3 inflammasome formation. Thus, a ‘priming’ step is needed to first upregulate the expression of NLRP3 in most cell types^28,102–104; the need for priming is shared by some inflammasomes such as NLRP6 (ref. 105) but is not required for others such as the NAIP–NLRC4 or AIM2 inflammasomes. The reason for this additional step is not clear but it is suggested that priming might help to better regulate inflammasome activation, particularly given that both NLRP3 and NLRP6 can be activated by many stimuli. The pro-IL-1β protein also often requires priming as it is usually not expressed without stimulation¹⁰⁶, in contrast to pro-IL-18, which is constitutively expressed in various cell types¹⁰⁶ but can be further upregulated under specific conditions¹⁰⁷. Acute LPS treatment has been shown to prime NLRP3 activation even in the presence of the protein synthesis inhibitor cycloheximide, which suggests that priming can occur in a non-transcriptional manner, probably involving posttranslational modifications such as deubiquitylation and phosphorylation^{104,108–113}.

After priming, NLRP3 can be activated by microbial infection and by sterile insults that trigger certain cellular perturbations^28,46. Potassium (K⁺) efflux is a pivotal upstream event for NLRP3 inflammasome activation, with high extracellular K⁺ concentration having an inhibitory effect^114–116. Widely used NLRP3 activators support the role of K⁺ efflux in inflammasome activation: nigericin is a bacterial toxin and K⁺ ionophore¹¹⁷, extracellular ATP gates the P2X7 cation channel for Ca²⁺ and Na⁺ influx and in turn activates the potassium channel TWIK2 (also known as KCNK6) for K⁺ efflux^117,118, and particulate matter induces a decrease of intracellular K⁺ concentration and increase of intracellular Ca²⁺ concentration¹¹⁶. Of note, spontaneous K⁺ efflux in response to low extracellular K⁺ concentration also activates NLRP3 (ref. 116). Nonetheless, because ionic fluxes in cells are often coupled, whether K⁺ concentration, or that of another ion, is the key inducer of the NLRP3 inflammasome remains controversial. In particular, whereas some studies have suggested that K⁺ efflux regulates Ca²⁺ signalling to activate the NLRP3 inflammasome¹¹⁹, other studies have suggested that K⁺ efflux induces NLRP3 inflammasome activation independently of Ca²⁺ signalling¹²⁰.

Crystals and particles that activate NLRP3 include monosodium urate crystals¹³, aluminium hydroxide, silica, asbestos^121,122, cholesterol crystals^123,124 and amyloid-β (ref. 125), which are associated with gouty arthritis, pulmonary and cardiovascular diseases, and Alzheimer disease. Malarial hemozoin is also an NLRP3 inflammasome activator owing to its particulate nature, and inflammasome activation may contribute to the recurrent fever attacks and neurodegeneration that are associated with malaria¹²⁶. Activation of NLRP3 by particulate matter seems to involve their phagocytosis and subsequent lysosomal damage and rupture^121,125. The importance of lysosomal damage in inflammasome activation is supported by the ability of the lysosomal damage inducer, leucyl-leucine methyl ester, to activate NLRP3 (ref. 121). However, how lysosomal damage is associated with K⁺ efflux remains unknown^116,119,121. Other factors such as reactive oxygen species, mitochondria and oxidized mitochondrial DNA have also been implicated in NLRP3 activation^127–130. In addition, several agents that activate NLRP3 independently of K⁺ efflux have been described, including the TLR agonists imiquimod and CL097, which activate NLRP3 independently of TLR signalling¹³¹.

The NLRP3 inflammasome has been heavily investigated owing to its biological significance and therapeutic relevance, which have recently been reviewed elsewhere^22,55. Here, we focus on describing the structural nature of NLRP3 — from its interaction with NEK7 to the inactive oligomer structure that we refer to as a ‘cage’ and to the active disc-like structure with ASC — which collectively begin to sketch out a series of conformational changes that occur in the NLRP3 activation pathway.

Structure of the NLRP3(ΔPYD)–NEK7 complex

The first NLRP3 structure was determined in 2019 (ref. 132) following the discovery of NEK7 as an essential factor in mouse NLRP3 activation^133–135. NEK7 is a centrosomal serine/threonine kinase; for NLRP3 activation, the kinase scaffold but not the kinase activity is required. A PYD-deleted (ΔPYD) human NLRP3 construct was shown to form a complex with human NEK7 at a sub-micromolar binding affinity¹³². The stability of NLRP3 and of the NLRP3–NEK7 complex was increased by the addition of ADP and the specific NLRP3 inhibitor MCC950 to the samples for cryo-EM structure determination¹³⁶. This ability to reconstitute a complex between inactive NLRP3 and NEK7 suggested that NEK7 is not the direct activator of NLRP3, despite its requirement to form an active NLRP3 inflammasome. Of note, although the role of NEK7 has been demonstrated in mouse cells and in mice, its function in human cells has been controversial. In particular, a recent study has suggested that in human cell lines, NEK7 is redundant for inflammasome activation when the nuclear factor-κB regulator IKKβ is activated during priming but it can synergize with IKKβ to shorten the priming requirement for NLRP3 activation¹³⁷.

The NLRP3(ΔPYD)–NEK7 structure revealed a first glimpse of inactive NLRP3, from its central NACHT domain to the C-terminal LRR domain¹³² (Fig. 3a). Whereas ADP was shown to bind the NACHT domain, the location of MCC950 binding was not determined owing to the limit in resolution. The NLRP3 NBD and HD2 (subdomains of NACHT) and the LRR domain contact the C-terminal lobe of NEK7 and assume a tight grip around it; the N-terminal lobe of NEK7 is disordered in this structure. Structure-based mutagenesis of the binding interface indicated that the LRR domain and HD2 of NLRP3 are important for NEK7 binding whereas the NBD is dispensable. This conclusion is consistent with the hypothesis that NEK7 most probably is recruited after NLRP3 has changed to its active conformation, when the NBD is rotated out of the vicinity of NEK7 (ref. 138). Thus, the NLRP3–NEK7 structure provides the molecular basis for NEK7 recruitment to the NLRP3 oligomer.

Fig. 3 |. The NLRP3 inflammasome.

a, Domain architectures of NLRP3, NEK7 and ASC, with the domain colours matching those shown in part b. Typical disease-associated mutations in NLRP3 are highlighted. b, Formation of the NLRP3 inflammasome. Step 1: upon upregulation of their expression (‘priming’), NLRP3 proteins are present in the cytosol and on the trans-Golgi network (TGN) as monomeric proteins or inactive ‘cage’ structures [PDB:7PZC]. Step 2: upon stimulation, inactive NLRP3 cages presumably undergo conformational changes. The TGN disperses into vesicles (dispersed TGN (dTGN)) containing hypothetical active NLRP3 cages. Step 3: via microtubule trafficking, the vesicles move to the microtubule organizing centre (MTOC), wherein NEK7 [PDB:6S76] proteins are located. Step 4 and step 5: NEK7 interacts with NLRP3 cages [PDB:6NPY], presumably disrupting and opening each cage into two halves. Step 6: the two halves rearrange and unite into an inflammasome disc [PDB:8EJ4 and PDB:8ERT]. Step 7 and step 8: ASC [PDB:2KN6] adaptor proteins are recruited to the NLRP3 inflammasome disc via homotopic PYD–PYD interactions [PDB:8EJ4, PDB:8ERT and PDB:3J63], resulting in a PYD–PYD filament that oligomerizes the CARDs of ASC molecules to mediate caspase-1 recruitment. Domains are colour coded as in part a. ASC, apoptosis-associated speck-like protein containing a CARD; CARD, caspase recruitment domain; FISNA, fish-specific NACHT-associated domain; HD, helical domain; LRR, leucine-rich repeats; NACHT, nucleotide-binding and oligomerization domain; NBD, nucleotide-binding domain; NLRP, NBD-containing, LRR-containing and PYD-containing protein; PYD, pyrin domain; WHD, winged helix domain.

Structure of NLRP3 ‘cages’

Structures of full-length NLRP3 as determined by cryo-EM were shown to exist as double-ring-shaped cages, with human NLRP3 forming decamers and mouse NLRP3 forming dodecamers to hexadecamers^139–141 (Fig. 3b). These NLRP3 oligomers are referred to as cages because the PYDs of the NLRP3 molecules, which are important for ASC recruitment and inflammasome assembly, seem to be shielded within the cages, possibly to prevent premature activation¹⁴⁰. In all structures, the cages are formed by interactions between the LRR domains of the two NLRP3 rings, together with interactions between the PYDs^139–141. Although the PYD–PYD interactions were not directly visualized as ordered densities, low-resolution densities were observed and the PYDs are important for cage formation. These data were interpreted as the existence of PYD filaments within the cage in one study¹⁴⁰ but were modelled as two PYD molecules in another study¹³⁹.

Surprisingly, NLRP3 cages are mainly membrane bound, as shown by fractionation experiments in NLRP3-overexpressing cells or macrophage cell lines endogenously expressing NLRP3, despite NLRP3 not being a bona fide transmembrane protein¹⁴⁰. This observation is consistent with earlier studies showing the localization of NLRP3 at various intracellular organelles before NLRP3 activation¹⁴². Immunofluorescence imaging of wild-type, immortalized bone marrow-derived macrophages revealed partial localization of NLRP3 at the trans-Golgi network (TGN), as marked by TGN38, and this localization was further increased when NLRP3 was stably expressed to reconstitute NLRP3-knockout cells¹⁴⁰. Interestingly, the TGN disperses into vesicles marked by TGN38 known as dispersed TGN (dTGN) in response to NLRP3 stimuli¹⁴³, and NLRP3 mutants that cannot form cages are defective in both TGN dispersion and NLRP3 activation¹⁴⁰.

Another study has suggested that NLRP3 inflammasome activators converge on disrupting endoplasmic reticulum–endosome membrane contact sites, causing endosomal accumulation of phosphatidylinositol 4-phosphate (PI4P) and a consequent impairment of endosome-to-TGN trafficking¹⁴⁴. In this context, endosomes can accumulate TGN38, giving them the appearance of dTGN vesicles. The TGN dispersion model of NLRP3 may, thus, be unifiable with the endosomal model in that both models involve the accumulation of PI4P-enriched vesicles that recruit NLRP3. Of note, in studies involving changes to extracellular K⁺ concentration, it was suggested that K⁺ efflux is essential for NLRP3 recruitment to vesicles and NLRP3 activation. By contrast, when NLRP3 recruitment to vesicles was forced by fusing the protein to a Golgi-localization domain, then increased extracellular K⁺ concentration had no effect on NLRP3 activation¹⁴³.

Structure of the active NLRP3 inflammasome

The first indication that NLRP3 is associated with the microtubule-organizing centre (MTOC) came from the identification of the centrosomal kinase NEK7 as an essential factor in NLRP3 activation^133–135. Indeed, the NLRP3 inflammasome has been shown to localize at the MTOC to form the caspase-1 p33/p10 active enzyme^145,146, and microtubule retrograde transport by dynein and a dynein adaptor, HDAC6, is required for this MTOC localization of NLRP3 and for inflammasome activation. As the TGN marker TGN38 was detected at the MTOC upon inflammasome activation, and NEK7 can decrease the amount of NLRP3 cages in vitro, it was proposed that NLRP3 cages are transported on vesicles to the MTOC wherein they bind NEK7 and change their structural organization^140,146.

The cryo-EM structure of an active NLRP3 inflammasome was solved recently after optimizing the reconstitution method¹³⁸ (Fig. 3b). NLRP3 and NEK7 were co-transfected and expressed in human Expi293 cells and stimulated with the NLRP3 activator nigericin before harvesting. Previous studies using an engineered NLRP3 with a bioluminescence resonance energy transfer (BRET) sensor have suggested that nigericin or other K⁺ ionophores alter NLRP3 conformation in a K⁺ efflux-dependent manner¹⁴⁷, a process that has not been achievable using recombinant proteins probably owing to the requirement of membranes. The BRET study has also revealed the importance of the fish-specific NACHT-associated (FISNA) domain of NLRP3 (ref. 147) — an additional NACHT subdomain, in front of the NBD — in mediating this conformational change, which was confirmed by the active NLRP3 inflammasome structure¹³⁸. The successful reconstitution of the active NLRP3 inflammasome required the addition of Mg²⁺ and ATPγS, presumably to keep NLRP3 in an active conformation, and of ASC PYD, to further promote NLRP3 inflammasome formation. The requirement for ASC to assemble a stable NLRP3 inflammasome is in contrast to the previously solved NAIP–NLRC4 inflammasome structure (without ASC)⁷³.

The active NLRP3–NEK7–ASC complex comprises mainly 10 or 11 individual subunits of NLRP3, with the 10-fold structure being more prevalent and solved at a higher resolution¹³⁸ (Fig. 3b). The NLRP3–NEK7 complex forms a disc-shaped structure with the NACHT domains of NLRP3 near the disc centre and the NEK7-bound LRR domains of NLRP3 at the periphery. The PYDs of NLRP3 form a PYD filament that protrudes orthogonally from the centre of the disc, which further recruits ASC through PYD–PYD interaction. The hybrid NLRP3–ASC PYD filament structure is similar to the PYD filament structures of ASC and NLRP3 separately^59,62. There is a symmetry mismatch between the disc and the filament, probably made possible by the flexible linker between the PYD and the NACHT domain of NLRP3. Large-scale conformational changes are observed between the ADP-bound inactive NLRP3 conformation¹³² and the ATP-bound active NLRP3 conformation¹³⁸, which are reflected by a rotation of approximately 85° between the FISNA–NBD–HD1 region of NLRP3 and the WHD–HD2–LRR region, and by local conformational changes in particular at the FISNA domain. Interactions at the centre of the disc are dominated by the FISNA domain, with additional contributions from NBD, HD1 and WHD¹³⁸.

Intriguingly, each NEK7 molecule binds to only one NLRP3 molecule and does not participate in NLRP3 oligomerization in the disc¹³⁸. Thus, we propose that the function of NEK7 is to break inactive NLRP3 cages to allow for reassembly. Although this disc structure provides a framework for understanding NLRP3 biology, many additional proteins can modulate NLRP3 inflammasome assembly, such as BTK¹⁴⁸, MAVS^149,150 and NLRP11 (ref. 151), and numerous posttranslational modifications of NLRP3 have been identified^25,152–154, which open up the possibility that NLRP3 cross-talks with a large number of cellular pathways to regulate its activation. Interestingly, the minimal requirement for NEK7 for NLRP3 inflammasome activation in human cell lines was shown to be attributable to IKKβ-induced recruitment of NLRP3 to the TGN during priming¹³⁷. Whether NEK7-independent inflammasome activation alters microtubule-mediated trafficking and the localization of NLRP3 inflammasome specks remains to be addressed.

NLRP1 and CARD8 inflammasomes

Although NLRP1 was the first protein to be identified as forming a caspase-1-activating complex⁹, its activation mechanism was not elucidated until much later. Like other NLRPs, human NLRP1 contains a PYD at the N-terminus. However, the PYD of NLRP1 has an autoinhibitory function rather than recruiting ASC to form the inflammasome. In support of this, mutations in NLRP1 PYD cause hyperactivation and have been associated with inflammatory skin diseases^155,156. In addition to a PYD, and NACHT and LRR domains, NLRP1 also contains a unique function-to-find domain (FIIND) followed by a CARD at the C-terminus¹⁵⁷ (Fig. 4a). The FIIND, composed of ZU5 and UPA subdomains, undergoes autoproteolytic cleavage at a Ser–Phe–Ser motif at the C-terminal end of ZU5, generating an N-terminal PYD–NACHT–LRR–ZU5 fragment that is autoinhibitory and a C-terminal UPA–CARD fragment that activates the inflammasome¹⁵⁸. This autocleavage event is required for NLRP1 function, and it is important that the N-terminal and C-terminal fragments remain associated with each other at the resting state^159–162.

Fig. 4 |. The NLRP1 inflammasome.

a, Domain architectures of NLRP1 and DPP9. Domains that are shown in grey indicate that the structure of this specific region is not shown in parts b–e. The unlabelled domains in NLRP1 represent predicted unstructured regions. Typical disease-associated mutations in NLRP1 are highlighted. The proteolysis site on the FIIND is marked as a gap. b, The structure of an NLRP1 and DPP9 ternary complex, a dimeric complex in which each half of the dimer contains ZU5 and UPA (the FIIND) of full-length NLRP1, the UPA of a processed NLRP1 C-terminal fragment and a DPP9 monomer [PDB:6X6A]. The N-terminal end of the NLRP1 C-terminal UPA inserts into the active-site tunnel of DPP9. c, The structure of DPP9 bound with the small-molecule inhibitor Val-boro-pro (VbP) in complex with NLRP1 [PDB:6X6C]. VbP occupies the active site of DPP9, which releases the NLRP1 C-terminal UPA, allowing for the activation of NLRP1. d, Clustering of NLRP1 UPA [PDB:6XKK] promotes the homotypic interactions of NLRP1 CARDs to form filaments. e. Two different orientations of an NLRP1 CARD filament [PDB:6K7V]. For parts b–e, domains are colour coded as in part a. CARD, caspase recruitment domain; CT, C-terminus; DPP, dipeptidyl peptidase; FIIND, function-to-find domain; HD, helical domain; LRR, leucine-rich repeats; NBD, nucleotide-binding domain; NLRP, NBD-containing, LRR-containing and PYD-containing protein; NT, N-terminus; PYD, pyrin domain; WHD, winged helix domain.

It is now established that proteasome-dependent ‘functional degradation’ of the N-terminal fragment is the main mechanism by which NLRP1 is activated^159,161,163. It has been observed that anthrax lethal factor cleaves the N-terminal domain of NLRP1b, a mouse NLRP1 orthologue naturally lacking a PYD, resulting in inflammasome activation^164–166. Upon cleavage, the N-terminal fragment undergoes ubiquitylation by the N-end rule ligase UBR2 (refs. 159,167), leading to its proteasomal degradation and liberation of the C-terminal fragment for inflammasome activation^159,161 (Fig. 4a). Shigella flexneri ligase IpaH7.8 directly ubiquitylates NLRP1b, also resulting in proteasomal degradation and inflammasome activation^161,163. Moreover, the prolyl dipeptidyl peptidases DPP8 and DPP9 have been shown to associate with and inhibit NLRP1 inflammasome activation, and small-molecule inhibitors of DPP8 and DPP9, such as Val-boro-pro (VbP), activate the NLRP1 inflammasome by a proteasome-dependent mechanism^{159,168–171}.

Ultraviolet B (UVB) radiation is another stimulus that is known to induce NLRP1 activation in human, but not rodent, keratinocytes^172,173, and ribosome stalling caused by UVB radiation to mediate the ribotoxic stress response has been associated with activation of the mitogen-activated protein kinase ZAKα^174,175. Putting these two observations together, it was hypothesized and demonstrated that RNA damage caused by UVB radiation or toxins activates the NLRP1 inflammasome through ZAKα; together with its downstream kinase p38, ZAKα directly phosphorylates a human-specific disordered linker region of NLRP1, leading to ubiquitylation and degradation of the NLRP1 N-terminus to activate the NLRP1 C-terminus. These data established inflammasome-driven pyroptosis as an integral component of the ribotoxic stress response^176,177. Furthermore, the p38 kinase was implicated in mediating NLRP1 inflammasome activation after viral infection¹⁷⁶.

The 3C protease from human rhinoviruses, a type of picornavirus, and other proteases from diverse picornaviruses have been shown to cleave human NLRP1, resulting in its ubiquitylation via a cullin E3 ligase machinery and N-terminal degradation^178,179. By contrast, viral dsRNA seems to activate human NLRP1 via a direct interaction independently of the proteasome¹⁸⁰. NLRP1 is also emerging as a general sensor of cellular stress, including reductive stress, in which reduced thioredoxin-1 cannot bind to NLRP1 to restrain its activity^181,182, and protein folding stress, in which distinct inhibitors of protein folding or inducers of the unfolded protein response accelerate degradation of the N-terminal fragment of NLRP1 (ref. 183).

Structure of the NLRP1–DPP9 complex

DPP8 and DPP9 inhibit both rodent and human NLRP1 inflammasomes through a direct interaction with NLRP1 FIIND¹⁷¹ (Fig. 4a). Cryo-EM structures of human or rat NLRP1 in complex with DPP9 surprisingly revealed a tripartite assembly, in which each DPP9 molecule of a DPP9 dimer binds one full-length NLRP1 and one UPA–CARD C-terminal fragment of NLRP1 (refs. 184,185) (Fig. 4b). Because these complexes were obtained from co-expression of NLRP1 and DPP9 in the absence of any stimulation, the existence of the UPA–CARD fragment suggests that it was generated either by increased degradation of the NLRP1 N-terminal fragment owing to overexpression-induced stress or by basal levels of N-terminal degradation during homoeostatic protein turnover.

In both structures, only the FIIND (ZU5 and UPA) of full-length NLRP1 and the UPA of the C-terminal fragment are visible, indicating that the remaining domains are flexibly linked^184,185. In the FIIND of full-length NLRP1, ZU5 and UPA subdomains associate extensively with each other along the longer dimensions of their β-sandwich folds^184,185, illustrating how the N-terminal and C-terminal fragments in autoprocessed NLRP1 do not dissociate (Fig. 4b) until the N-terminal fragment is degraded by the proteasome. The FIIND of full-length NLRP1 can interact with DPP9 in the absence of the C-terminal UPA–CARD fragment, whereas the recruitment of the C-terminal UPA to DPP9 requires that DPP9 is already in complex with FIIND, and UPA alone cannot stably interact with DPP9 (refs. 184,185). In the human NLRP1–DPP9 complex, the N-terminal strand of the UPA of the C-terminal fragment of NLRP1, which contains a DPP9 substrate motif, inserts into the active-site tunnel of DPP9 to reach close to the active site, which is an interaction that is crucial for the recruitment of DPP9 (ref. 185). However, when the DPP9 inhibitor VbP is present, it binds in the active-site pocket of DPP9 to displace the UPA N-terminal strand from DPP9, resulting in release of the UPA–CARD fragment¹⁸⁵ (Fig. 4c).

These structures offer a plausible mechanism to explain how the tripartite NLRP1–DPP9 assembly inhibits the C-terminal fragment of NLRP1 from assembling an inflammasome. As shown below, the UPA of NLRP1 C-terminal fragment can self-oligomerize, which promotes CARD filament formation and inflammasome activation^71,72. By contrast, DPP9-bound UPA is inhibited from oligomerization by steric hindrance (Fig. 4b), quenching the intrinsic ability of NLRP1 C-terminal fragment to form an inflammasome and avoiding inappropriate NLRP1 inflammasome activation during homoeostatic protein turnover that results in degradation of the N-terminal fragment of NLRP1. The structures imply that such inhibition requires that the concentration of full-length NLRP1 is greater than or equal to the concentration of C-terminal fragment. However, when NLRP1 is stimulated to undergo functional degradation, the amount of C-terminal fragment ultimately exceeds the amount of full-length NLRP1, leading to the accumulation of free C-terminal fragment. The C-terminal UPA and CARD then cooperatively oligomerize, which leads to CARD filament formation and inflammasome activation^71,72 (Fig. 4d,e).

Germline mutations of human NLRP1, which is expressed at high levels in human keratinocytes, lead to skin-related inflammatory diseases such as multiple self-healing palmoplantar carcinoma, familial keratosis lichenoides chronica¹⁵⁶, vitiligo^186,187 and autoinflammation with arthritis and dyskeratosis^171,188. Among these deleterious mutations, A54T, A66V, M77T (ref. 156), L155H (refs. 186,187), R726W (ref. 188), ΔF787–R843 (ref. 156) and M1184V (refs. 186) lie in the autoinhibitory N-terminal fragment, whereas P1214R (refs. 171,188) localizes at the DPP9 substrate motif of the inflammatory C-terminal fragment, causing constitutive human NLRP1 activation and downstream pyroptosis^156,171 (Fig. 4a). Thus, the structural framework of the tripartite NLRP1–DPP9 assembly also helps to explain the biology of disease-associated NLRP1 mutations.

Structure of the NLRP1 C-terminal filament

The CARD of NLRP1 interacts with ASC, which bridges NLRP1 and caspase-1 for inflammasome assembly¹⁸⁹. Previously determined CARD filament structures of the inflammasome components ASC, caspase-1 and NLRC4 (refs. 66,67) illustrated the mechanistic basis for self-oligomerization and hetero-oligomerization of CARDs. For NLRP1, an important distinguishing feature is that its CARD is ineffective in oligomerization by itself but can be promoted to oligomerize by the UPA domain^71,72. Cryo-EM structures of the NLRP1 C-terminal fragment revealed an ordered density only for the CARDs, whereas oligomerized UPA molecules tend to be flexibly linked at the outside of the filament^71,72 (Fig. 4d). Unlike any of the other CARD filament structures solved previously, the NLRP1 C-terminal filament is composed of CARD dimers with twofold symmetry, rather than monomers, despite having similar helical symmetry to the other CARD filaments⁷². This CARD dimer structure was also observed in a previous crystal structure of NLRP1 (ref. 190). The dimer composition of the NLRP1 C-terminal assembly results in a two-layered appearance whereby the dimeric partners protrude outwards from the core filament⁷² (Fig. 4d). However, the NLRP1 CARD assembly that is formed at high concentration and in the absence of UPA has a monomeric CARD filament⁷¹, and the functional significance of the CARD dimer remains unclear.

Furthermore, the structures of NLRP1 UPA–CARD and NLRP1 CARD filaments predicted that the CARD of NLRP1 recruits the CARD of ASC through charge complementarity on the cross sections of the filaments^71,72. By contrast, NLRP1 CARD was predicted not to be able to recruit the CARD of caspase-1 directly owing to lack of structural and charge compatibility at the interface. The nature of UPA oligomerization was implicated by mutations at the interface between the UPAs of full-length NLRP1 and of the C-terminal fragment, which — despite the release of the C-terminal fragment from the tripartite NLRP1–DPP9 assembly — did not activate but rather abrogated inflammasome activation¹⁸⁵. These data suggest that abrogation of inflammasome activation by the UPA mutations was owing to impaired UPA oligomerization through a mode of interaction observed in tripartite NLRP1–DPP9 (refs. 71,72,184,185).

Structure of the CARD8 inflammasome

Similar to NLRP1, human CARD8 (not present in rodents) has an autoproteolytic FIIND at its core — in addition to a disordered N-terminal region that varies between isoforms and a C-terminal CARD^162,191 — and has a similar mechanism of activation^162,170,191. The inflammatory UPA–CARD C-terminal fragment of CARD8 must be released from the inhibitory N-terminal fragment to allow for CARD8 activation¹⁷⁰. DPP8 and DPP9 directly interact with full-length CARD8 and the C-terminal fragment to suppress inflammasome activation during homoeostatic CARD8 turnover¹⁹². Small-molecule inhibitors of DPP8 and DPP9 induce pyroptosis in human monocytes and macrophages through CARD8 activation and are being investigated for potential cancer therapies^169,170. CARD8 was shown to be an inflammasome sensor for HIV-1 protease, leading to CARD8 ubiquitylation, degradation of the N-terminal fragment and activation of the C-terminal fragment in infected cells¹⁹³. In addition, anti-HIV-1 non-nucleoside reverse transcriptase inhibitors can promote premature activation of HIV-1 protease to trigger the CARD8 inflammasome, which suggests that targeting the CARD8 inflammasome could be a potent and broadly effective strategy for HIV eradication¹⁹³. Proteases from diverse coronaviruses including SARS-CoV-2 (ref. 194) and enteroviruses¹⁹⁵ can also cleave CARD8, resulting in virus-induced pyroptosis in infected CD4⁺ T cells, monocytes, macrophages or endothelial cells.

A cryo-EM structure of the CARD8–DPP9 complex has revealed regulatory mechanisms that are similar to those of NLRP1 (ref. 192). These mechanisms include the suppression of CARD8 C-terminal activity by DPP9 and full-length CARD8, the association of the N-terminal fragment with the C-terminal fragment in full-length CARD8, and the inhibition of CARD8 UPA oligomerization by the tripartite CARD8–DPP9 assembly¹⁹². However, in contrast to NLRP1, the N-terminal strand of CARD8 UPA alone does not insert deeply into the DPP9 active site and VbP does not directly displace CARD8 from DPP9 (ref. 192). Thus, it seems that VbP induces CARD8 activation mainly by proteostatic stress that upregulates CARD8 functional degradation.

Effects of inflammasome signalling

GSDMD was identified as a bona fide substrate for the inflammatory caspases through genetic screens, and the cleavage of human GSDMD at Asp275 generates an active N-terminal fragment and an autoinhibitory C-terminal fragment^37–39 (Fig. 5a,b). GSDMD had also previously been identified as a major hit for substrates of inflammatory caspases from a proteomics screen¹⁹⁶. In addition to inflammasome-mediated cleavage, GSDMD can also be cleaved and activated by caspase-8, as well as by neutrophil granule enzymes^92,197,198. GSDMD cleavage and activation are important for both the induction of pyroptosis and the release of IL-1 family cytokines from cells. The N-terminal fragment of GSDMD was shown to bind acidic lipids at the inner leaflet of plasma membranes and at mitochondrial membranes, and to mediate the formation of large transmembrane pores^40–45, providing a potential mechanistic basis for cytokine release through the pores and for membrane damage in pyroptosis^199,200. Expression of GSDMD N-terminal fragment in human cells^40–45 and treatment of bacteria with recombinant N-terminal frgament⁴² both caused cell death, indicating that the GSDMD N-terminal fragment has cytotoxic function. GSDMD belongs to a family of gasdermins and the discovery of their pore-forming activity redefined pyroptosis as gasdermin-mediated programmed lytic cell death^34,201. This definition has now been extended by the role of NINJ1 in rupturing membranes to complete pyroptosis²⁰².

Fig. 5 |. Gasdermins.

a, Domain architectures of a gasdermin and caspase-1. Domains that are shown in grey indicate that the structure of this specific region is not shown in parts b–g. b, A gasdermin molecule, represented using the structure of mouse GSDMA3 [PDB:5B5R]. The scissor icon shows the interdomain cleavage site of gasdermin. c, The structure of caspase-1 p20/p10 dimer in complex with GSDMD [PDB:6VIE and 6KMV]. The active sites are marked as scissor icons, and the exosites are marked by circles. d, The structural changes of a human GSDMD N-terminal subunit from autoinhibited state [PDB:6N9O] to pore-forming state [PDB:6VFE], which could occur during prepore to pore transition on the membrane. The pore-forming state of GSDMD N-terminal fragment looks like a stretched hand; as the ‘fingers’ stretch out to form the transmembrane β-barrel, the globular domain undergoes a 38° rotation. e, The structure of a 33-subunit GSDMD N-terminal pore viewed from two orientations [PDB:6VFE]. f, Superposition of human GSDMD (blue) and human GSDMB (purple; [PDB:8ET2 and 8GTN]) pore-forming subunits and the structure of a 24-subunit GSDMB N-terminal pore viewed from two orientations [PDB:8ET2]. g, Electrostatic surface of the GSDMD pore viewed from outside and inside [PDB:6VFE]. The GSDMD membrane-binding regions are positively charged (blue), whereas the GSDMD conduit is negatively charged (red). The cargo release route is marked by an arrow. For parts b–f, domains are colour coded as in part a. CDL, CARD linker; CT, C-terminus; IDL, interdomain linker; NT, N-terminus.

Structure of autoinhibited GSDMD

The mechanism of GSDMD autoinhibition has been indicated by crystal structures of full-length mouse GSDMA3 and full-length mouse and human GSDMD^41,203. Even though extensive loop truncations were required for crystallization of GSDMD, the structures of truncated GSDMD nonetheless revealed that, similar to the GSDMA3 structure, the N-terminal fragment is composed of a twisted β-sheet surrounded by α-helices and a largely disordered region responsible for membrane insertion, and the C-terminal fragment is composed of tightly packed α-helices (Fig. 5b). The N-terminal and C-terminal fragments interact extensively to exert autoinhibition by masking membrane-binding elements. Disease-associated, gain-of-function mutations in GSDMD often map to the C-terminal fragment at the autoinhibitory contact sites; these mutations bypass signal-dependent activation of GSDMD and lead to its constitutive activation^41,203.

GSDMD is a substrate for all inflammatory caspases. The crystal structure of caspase-1 in complex with a GSDMD IDL peptide containing the FLTD tetrapeptide motif (residues 272–275 in human GSDMD) at first suggested that caspase-1 recognizes GSDMD using the same mechanism by which caspases self-cleave²⁰⁴. However, GSDMD recognition by inflammatory caspases turned out to be more complex. Biochemical and functional characterization revealed that caspase-4 and caspase-11 first need to be fully autoprocessed to generate the p10 fragment that participates in the active enzyme. A region of the caspases away from the active site (the exosite) then interacts with GSDMD C-terminal fragment with high affinity through hydrophobic and hydrogen-bonding interactions²⁰⁵. The exosite interaction positions the interdomain cleavage site of GSDMD at the caspase catalytic site, as shown by the crystal structure of caspase-1 in complex with full-length GSDMD²⁰⁶ (Fig. 5c). This two-site mode of interaction has now also been observed in the structures of caspase-4 in complex with pro-IL-18 (refs. 89,90), which could suggest a general mechanism of substrate recognition by inflammatory caspases and could explain why inflammatory caspases seem to have fewer substrates than apoptotic caspases.

Structure of the GSDMD pore

The first gasdermin N-terminal structure in the pore form was revealed by cryo-EM studies of the GSDMA3 N-terminal pore, which showed the β-barrel nature of the pore, the large conformational changes that occur upon membrane insertion, and the mechanisms of lipid binding and oligomerization, with implications for the entire gasdermin family²⁰⁷. Recently, the human GSDMD N-terminal pore structure³³ was solved (Fig. 5d,e). In addition to fully formed pores with the large β-barrel that should traverse the membrane, these structures also revealed oligomerized rings without the β-barrel, which were dubbed ‘prepores’^41,207. The existence of a prepore structure suggests that it may represent an oligomerized intermediate structure on membrane before GSDMD insertion and pore formation, but studies using atomic force microscopy have also suggested the assembly of arc-shaped and slit-shaped membrane pores that can transform into ring-shaped pores, which implies that further oligomerization can occur even after GSDMD membrane insertion²⁰⁸.

Most GSDMA3 and GSDMD N-terminal pores reconstituted in vitro consist of 26–28 subunits and 31–34 subunits, respectively, with outer diameters of approximately 30 nm and inner diameters of approximately 20 nm. The membrane-inserted N-terminal fragment assumes the shape of a hand, with the membrane-binding α1 helix as a folded thumb, the rest of the globular region as the palm and two β-hairpins as the fingers. In autoinhibited full-length structures or prepores, the fingers of the N-terminal subunits are tucked into a fist and not extended. Thus, the main conformational change that occurs during membrane insertion involves the extension of the fingers to form the transmembrane β-barrel (Fig. 5d), whereas the globular domain faces the cytosol³³. Pore structures for GSDMB N-terminal fragments have also been determined^209,210 (Fig. 5f), and a structural comparison suggests that the main differences between the different pore structures may occur at their lipid-binding elements.

The cryo-EM structure of the GSDMD pore³³ provided a plausible explanation as to why inflammasome-activated live cells release mature IL-1β or IL-18 protein but not their pro-IL precursors²¹¹. This is important because only the mature, processed cytokines can interact with their respective receptors to activate downstream gene expression and amplify inflammation. It was noted that although pro-IL-1β and pro-IL-18 are small in size relative to the inner dimension of the GSDMD pore, they are highly negatively charged (acidic) owing to their prodomains³⁵. The GSDMD pore conduit is highly negatively charged (Fig. 5g), which suggested that it might repel negatively charged pro-IL-1β and pro-IL-18 to inhibit their release through the pore, through a mechanism known as electrostatic filtering³³. Indeed, small negatively charged dextran or protein has the slowest rate of release from GSDMD pores relative to neutral or positively charge cargoes of similar molecular weights³³. Mutating away acidic patches in the GSDMD N-terminal conduit compromised the selectivity of the pore against an acidic, negatively charged cargo^33,212. Previously, unconventional secretion of fibroblast growth factor 2 (FGF2) was shown to depend on caspase-1 activity²¹³, raising the possibility that the highly basic (positively charged) protein FGF2 can be released from cells through caspase-1-activated GSDMD pores.

Interestingly, molecular modelling and calculations of electrostatic surface potential suggested that the acidic nature of the GSDMD pore conduit is conserved in other gasdermin pores³³, which raises the question of whether there are biological reasons to deter the release of a negatively charged cargo in general while allowing the release of a neutral or positively charged cargo. Certain other IL-1 family members, for example, IL-1α and IL-37, also have an acidic prodomain and become more basic upon maturation, which suggests that the mature forms of these cytokines might also be selectively released by gasdermin pores in live cells. In this context, and for another gasdermin family member GSDME, it was shown that T helper 17 cells activate NLRP3 to result in caspase-8 and caspase-3 activation, leading to GSDME cleavage²¹⁴. GSDME pores, which also have an acidic conduit³³, then release mature IL-1α, processed by the calcium-activated cytosolic enzyme calpain, but not pro-IL-1α (ref. 214). Thus, although pro-IL-1α and IL-1α can both bind and activate the cognate receptor, in this case, only mature IL-1α is released and able to activate the receptor.

Structure of NINJ1 oligomers

The ninjurin proteins, including NINJ1 and its homologue NINJ2, are small cell-surface proteins first identified as adhesion molecules that promote axonal growth upon nerve injury^215,216. NINJ1 (but not NINJ2) was later shown to mediate plasma membrane rupture downstream of GSDMD pore formation to complete lytic cell death²⁰². The signal delivered by GSDMD to activate NINJ1 is currently unknown, but is likely to involve membrane permeabilization, membrane tension or membrane lipid redistribution, as NINJ1 is also activated by secondary necrosis, another lytic form of cell death²⁰². Both NINJ1 and NINJ2 have a predicted domain architecture comprising an N-terminal segment that is important for cell adhesion, an amphipathic helix and two transmembrane helices at the C-terminal end²⁰² (Fig. 6a).

Fig. 6 |. NINJ1.

a, Domain architecture of ninjurin 1 (NINJ1). The domain shown in grey indicates that the structure of this specific region is not shown in parts b–f. b, Structural presentations of a NINJ1 double filament [PDB:8CQR] (left) and two stacked NINJ1 subunits (right). The double filament contains six subunits each with a different shade of blue or pink. Secondary structures are labelled for the blue subunit to the right. c, Structure of a single NINJ1 subunit. d, A single filament of five NINJ1 subunits (modelled from [PDB:8CQR]), in two orientations. The α1 helix of one subunit extends over to the transmembrane helices α3 and α4 of a neighbouring subunit to make a NINJ1 chain. e, Structural representation of a NINJ1 subunit from rings or curved filaments, showing bent helices α3 and α4. f, A model of a partial NINJ1 ring built by propagating the NINJ1 segment structure, showing the concave hydrophobic side and the convex hydrophilic side. g, Two alternative models of NINJ1 activation: the membrane damage model, in which NINJ1 filaments or pores cause membrane leakage, and the membrane loss model, in which NINJ1 assembles around a patch of membrane, leading to its release. In both models, the α1 and α2 subunits of NINJ1 (shown in yellow) directly cause membrane damage. TM, transmembrane helix.

Structural studies have begun to reveal how NINJ1 may induce plasma membrane rupture. A recent cryo-EM study at a 3.8-Å resolution has shown that NINJ1 forms straight filaments, with one side being hydrophobic and the other side being hydrophilic²¹⁷. In this structure, two NINJ1 filaments are bundled in the overall structure, with the hydrophobic faces of the filaments facing each other (Fig. 6b). Within each NINJ1 subunit, the predicted amphipathic extracellular helix actually comprises two helices (α1 and α2), followed by the transmembrane helices (α3 and α4) that form a hairpin within the membrane (Fig. 6c). In each single filament, the α3–α4 hairpin of one NINJ1 subunit interacts side-by-side with that of a neighbouring subunit to form a chain assembly in the shape of a fence, within which each α1 helix crosses over to the neighbouring subunit to stabilize the assembly (Fig. 6d). The extreme N-terminal region is disordered in this structure. This structure and molecular dynamic simulations led to the hypothesis that the NINJ1 single filament with a hydrophilic side and a hydrophobic side can stably cap membrane edges or form large transmembrane pores. Thus, during lytic cell death, the extracellular helices insert into the plasma membrane to polymerize NINJ1 monomers and rupture the plasma membrane.

Two preprints reporting structural, biochemical and cell biological studies of NINJ1 now add further mechanistic insights into the ability of NINJ1 oligomers to rupture the membrane. In one of the preprints, both NINJ1 and NINJ2 structures were solved by cryo-EM as bundled double filaments at a 2.8-Å and a 3.1-Å resolution, respectively²¹⁸. These structures are similar to that previously published²¹⁷ and were proposed to have been formed during the purification process by the bound lipids as almost no protein–protein interactions exist between the filaments and some single NINJ1 filaments were also observed under cryo-EM²¹⁸. In the other preprint, curved filaments and ring-shaped oligomers of NINJ1 were observed in detergents, and when reconstituted with liposomes, NINJ1, but not NINJ2, was shown to ‘dissolve’ liposomes, leaving small heterogeneous rings and larger patches²¹⁹. Cryo-EM reconstruction using short segments from these rings and filaments (approximately six subunits), reaching a 4.3-Å resolution, revealed a similar fence-like assembly but with a few differences. At the level of individual NINJ1 subunits, instead of being straight, the α3 and α4 helices are highly bent, with Gly residues at the kinks in both helices (Fig. 6e). At the level of NINJ1 filaments, instead of being straight (Fig. 6b), there is a prominent curvature on the plane of the assembly, with the concave side being hydrophobic and the convex side being hydrophilic (Fig. 6f). The degree of this curvature can vary owing to the intrinsic plasticity of the subunit interactions, but the hydrophobic side, which seems to be denser owing to the fence-like helices, always faces the concave side²¹⁹.

These studies have begun to suggest intriguing mechanisms for NINJ1-mediated membrane rupture. The published structure²¹⁷ proposes a ‘membrane damage’ model in which the amphipathic single filaments could promote membrane rupture by capping and stabilizing membrane lesions of variable sizes to allow the release of DAMPs, or proposes that NINJ1 double filaments could open up into pores in response to osmotic pressure to form membrane lesions (Fig. 6g). The two preprints^218,219 propose a ‘membrane loss’ model in which NINJ1 oligomers could break membrane and solubilize membrane patches by encircling them, acting as membrane scaffolding proteins in nanodiscs (Fig. 6g). The latter model is supported by the hydrophobicity of the concave side of NINJ1 oligomers²¹⁹ and by the observed release of NINJ1 oligomers into the cell culture medium^218,219. In either model, the amphipathic nature of the NINJ1 oligomeric structure is key. When the rate of membrane damage exceeds that of membrane repair, or in the case of the membrane loss model when there is also insufficient membrane to patch up the damage, complete membrane rupture may ensue to release alarmins of different sizes. Although changes in membrane tension and/or lipid composition have been proposed to trigger NINJ1 activation, what exactly activates NINJ1 and how this occurs remain to be elucidated. In addition, the features of NINJ1, but not NINJ2, that determine its function in cell killing remain to be fully addressed.

Conclusions and future directions

In this Review, we present a structural and mechanistic view of inflammasomes, from how the sensor proteins are autoinhibited, regulated and activated to assemble with adaptor proteins and effector caspases to how the effector proteins (caspases, gasdermins and NINJ1) are activated to execute their functions in cytokine release and cell death. We anticipate that future structural information will further elucidate the complex regulatory network of the well-studied inflammasomes discussed here, provide a first glimpse as to the mechanisms of understudied inflammasomes, and reveal new therapeutic targets and their structural templates to tackle the large number of inflammasomopathies.

Glossary

Alarmin

Alarmins are proteins, peptides, metabolites or others that are released to the outside of a cell in response to immune activation, as a result of transmembrane pore formation, cell injury and lytic death, or degranulation. These endogenous molecules may have chemotactic and/or immune-activating properties to alert the host for defence.

Cryopyrin-associated periodic syndromes

(CAPS). A set of genetic diseases of differing severity caused by autosomal dominant mutations in NLRP3. Patients with CAPS develop spontaneous inflammation and excessive release of the cytokine IL-1β, and may also suffer from arthralgia, deafness and hives.

Microtubule-organizing centre

(MTOC). A structure in eukaryotic cells from which microtubules are nucleated and emanate, which is often synonymous with the centrosome.

Neutrophil extracellular traps

Web-like networks of extracellular DNA, histones and neutrophil granule proteins extruded by neutrophils to trap and damage pathogens.

Pyroptosis

Programmed lytic cell death mediated by cleavage of gasdermin D by caspase-1, caspase-4, caspase-5 and caspase-11, or by other gasdermin family members, that occurs downstream of inflammasome activation or other insults.

Ribotoxic stress response

Activation of stress-associated kinases such as p38 and JNK1 owing to functional ribosome defects that result in inhibition or partial inhibition of protein translation. Downstream signalling can trigger cell cycle arrest, cell death and cytokine production.

Type III secretion

A bacterial secretion system involving a needle-like protein complex used to deliver effector proteins into the host cell cytosol; often found in pathogenic Gram-negative bacteria.

Unfolded protein response

A stress response resulting from the accumulation of unfolded or misfolded proteins in the endoplasmic reticulum that helps cells to reduce the stress through translation suppression, protein degradation and production of chaperones, or that induces cell death with prolonged stress.