Deregulated inflammatory responses are involved in numerous human disorders, encompassing not only infectious and autoimmune disorders but also neurological, cardiovascular, renal, hepatic and neoplastic conditions1,2,3,4. On the one hand, disproportionate, unwarranted or unresolving inflammation can act as a bona fide disease driver, as in the case of chronic inflammatory bowel disease1. On the other hand, uncontrolled inflammatory responses may aggravate the course of conditions that originate from non-inflammatory cues, such as myocardial infarction3. Moreover, inefficient inflammatory reactions facilitate the persistence of infectious pathogens5 and enable the emergence and progression of malignant lesions in the context of failing cancer immunosurveillance6. Of note, inflammatory reactions may affect the course of specific diseases in opposing manners, largely depending on the intensity and duration of inflammation. For example, whereas indolent, chronic inflammation has been associated with oncogenesis and accelerated tumour progression in various settings7, potent inflammatory responses culminating in the engagement of adaptive immunity underlie the beneficial effects of numerous cancer therapies, including conventional chemotherapeutics8, targeted anticancer agents9 and radiotherapy10. Moreover, recent findings indicate that numerous components of the molecular cascades underlying inflammation are key for normal embryonic and postembryonic development, at least in specific settings such as neurodevelopment11. These examples highlight the crucial requirement for regulated inflammatory responses in organismal development and homeostasis.

Inflammation is generally initiated by the activation of pattern recognition receptors (PRRs) that are expressed by both immune and non-immune cells12. Importantly, PRRs can be activated not only by viral and bacterial molecules associated with infection — so-called microorganism-associated molecular patterns or pathogen-associated molecular patterns — but also by endogenous molecules that are commonly referred to as damage-associated molecular patterns (DAMPs)12. In physiological conditions, DAMPs — which include nucleic acids, small metabolites such as ATP and proteins such as calreticulin — are generally unable to drive PRR signalling because they cannot gain physical access to PRR-containing subcellular compartments13. However, cellular stress and death can be accompanied by considerable alterations in the permeability of various cellular compartments, which enables PRR activation by DAMPs and the consequent initiation of inflammatory responses12. For example, ATP functions as a DAMP only upon release into the extracellular environment when it can bind to cognate receptors expressed on myeloid cells, such as the purinergic receptors P2RY2 and P2RX7 (refs.14,15).

On the basis of these considerations, it would seem likely that mitochondria have an important role in the control of inflammatory responses, for at least three reasons16. First, mitochondria are widely considered as the evolutionary remnants of ancestral Alphaproteobacteria (the ancestors of modern Gram-negative bacteria)17, and some mitochondrial components have considerable similarity with bacterial molecules, suggesting that they might function as PRR ligands. For example, in contrast to nuclear DNA (but similarly to bacterial genomes), the mitochondrial genome is circular and not associated with histones18. Second, mitochondria have two membranes — the inner mitochondrial membrane (IMM) and the outer mitochondrial membrane (OMM) — which together offer a dual layer of control segregating mitochondrial DAMPs (mtDAMPs) from their cognate PRRs18. Third, mitochondria have a major role in the control of apoptotic and necrotic forms of regulated cell death (RCD) (Box 1), which ultimately involves irreversible mitochondrial permeabilization (and hence loss of mitochondrial compartmentalization)19. Thus, mitochondria offer a unique platform for DAMP redistribution, PRR signalling and inflammation in the context of failing adaptation to cellular stress (which is linked to RCD initiation), the ultimate goal being to elicit innate and adaptive immune responses in support of organismal homeostasis (despite the irreversible loss of cellular fitness)20. Indeed, as we discuss herein, mitochondria are master regulators of inflammatory responses, not only as they contain several bona fide DAMPs but also as they provide a physical scaffold for the activation of some PRRs21. Moreover, various cellular responses elicited by RCD-associated mitochondrial outer membrane permeabilization (MOMP), including autophagy and caspase activation, directly affect the regulation of inflammatory processes22,23.

Here we discuss the molecular mechanisms through which mtDAMPs elicit inflammation (especially, but not exclusively, intracellular mechanisms), the signal transduction cascades through which mitochondria control inflammatory processes and the relevance of mitochondria-regulated inflammation in human disease. This is particularly relevant not only as multiple fields of biomedical research have begun to realize the impact of dysregulated inflammation in disease but also as mitochondria-targeting agents are now being used for the clinical management of specific neoplasms, raising considerable expectations for further development in the context of cancer immunotherapy24.

mtDAMP signalling pathways

Findings from multiple, independent research teams have identified several signal transduction cascades through which mitochondrial dysfunction elicits inflammatory reactions, including (but not limited to) intracellular signalling through cyclic GMP–AMP synthase (cGAS) and stimulator of interferon response cGAMP interactor 1 (STING1), as activated by mitochondrial DNA (mtDNA), and through the inflammasome, as induced by mtDNA and reactive oxygen species (ROS) (Fig. 1).

Fig. 1: Main mechanisms of mitochondrial DAMP signalling.
figure 1

Various mitochondrial components and products that are released as a consequence of mitochondrial dysfunction (and potentially cell death) drive inflammatory responses upon accumulation in the cytosol or the extracellular environment. Mitochondrial DNA (mtDNA), which can exit mitochondria via BCL-2-associated X, apoptosis regulator (BAX) and BCL-2 antagonist/killer 1 (BAK1) pores or via the permeability transition pore complex (PTPC), is a potent activator of cyclic GMP–AMP synthase (cGAS), resulting in stimulator of interferon response cGAMP interactor 1 (STING1) signalling and consequent synthesis of cytokines such as interferon-β1 (IFNβ1), IL-6 and tumour necrosis factor (TNF). Cytosolic mitochondrial RNA (mtRNA) has similar effects, although they depend on retinoic acid-inducible gene I protein (RIG-I), melanoma differentiation-associated protein 5 (MDA5) and mitochondrial antiviral signalling protein (MAVS). This pathway is promoted by the BAX–BAK1-dependent release of Era-like 12S mitochondrial rRNA chaperone 1 (ERAL1), which favours MAVS stabilization at the mitochondrial surface. Moreover, upon release from dysfunctional mitochondria, both mtDNA and reactive oxygen species (ROS) can drive IL-1β and IL-18 secretion as a consequence of inflammasome signalling. Electron transport chain (ETC) functions also seem to affect inflammasome activation independently of ROS as they preserve intracellular ATP availability through phosphocreatine (PCr). ATP can be released by dying cells through lysosomal secretion and pannexin 1 (PANX1) channels, mediating both chemotactic and immunostimulatory effects on antigen-presenting cells (APCs) by binding to the purinergic receptors P2RX7 and P2RY2. Upon BAX–BAK1 oligomerization during apoptosis, diablo IAP-binding mitochondrial protein (DIABLO; best known as SMAC) release not only favours caspase activation (not shown) but also rewires NF-κB signalling from canonical to non-canonical programmes. Along similar lines, mtDNA (be it naked or complexed with the mitochondrial transcription factor TFAM) and N-formyl peptides (other mitochondrial components), which accumulate in the extracellular milieu upon regulated cell death, cause neutrophil activation via Toll-like receptor 9 (TLR9) or advanced glycosylation end product-specific receptor (AGER) and formyl peptide receptor 1 (FPR1), respectively. Finally, extracellular cardiolipin (which is normally restricted to the inner mitochondrial membrane) can be presented by dendritic cells (DCs) on the MHC class I-like molecule CD1d, resulting in the activation of γδ T cells. DAMP, damage-associated molecular pattern; ER, endoplasmic reticulum; IKK, IκB kinase; IRF, interferon regulatory factor; NIK (official name MAP3K14), mitogen-activated protein kinase kinase kinase 14; TBK1, TANK-binding kinase 1; TCR, T cell receptor.

cGAS–STING1 signalling

cGAS is a nuclear and cytosolic protein that responds to cytosolic double-stranded DNA (dsDNA) molecules by catalysing the formation of cyclic GMP–AMP (cGAMP), a second messenger that initiates an inflammatory response via STING1 (ref.25) (Box 2). Although initial studies focused on exogenous26 and nuclear27 dsDNA species as key cGAS activators, it rapidly became clear that mtDNA gaining access to the cytosol as a consequence of MOMP (Box 1) or other forms of mitochondrial dysfunction can also promote cGAS signalling, although this is tightly inhibited by apoptotic caspases28,29,30 (see later). Subsequent studies showed that whereas mtDNA, naked dsDNA and dsDNA bound to proteins that introduce a specific curvature (such as TFAM, a mitochondrial transcription factor, and high mobility group box 1 (HMGB1), a nuclear non-histone DNA-binding protein) potently activate cGAS, histone-bound nuclear dsDNA (in other words, chromatin) is a poor cGAS activator and inhibits cGAS signalling driven by naked dsDNA31,32. At least in part, this reflects the strong physical interactions between cGAS and histones, which result in structural reconfigurations that conceal the cGAS DNA-binding site, prevent cGAS dimerization and favour cGAS inactivation by mitotic kinases33,34,35. Taken together, these observations delineate a system that prevents unwarranted cGAS activation in physiological settings (for example, during mitosis when chromatin becomes exposed to cytosolic cGAS, or during programmed cell death, which often involves widespread and irreversible MOMP but is accompanied by robust caspase activation)35,36, but preserves the ability to initiate inflammatory responses in situations of danger. This is particularly (but not exclusively) true for pathogen infection, because multiple intracellular pathogens, including Mycobacterium tuberculosis, type 1 herpes simplex virus (HSV-1), dengue virus, influenza virus, encephalomyocarditis virus and severe acute respiratory syndrome coronavirus 2 drive mitochondrial dysfunction and/or MOMP, culminating in mtDNA release30,37,38,39,40,41. Furthermore, the apoptotic caspases that would otherwise inhibit cGAS signalling after MOMP are often disabled by proteins encoded by intracellular pathogens as a strategy to extend the lifespan of host cells during infection42.

Release of mtDNA in the course of MOMP generally involves the proapoptotic pore-forming proteins BCL-2-associated X, apoptosis regulator (BAX) and BCL-2 antagonist/killer 1 (BAK1)43,44 (Box 1). Mechanistically, MOMP as initiated by BAX and BAK1 rapidly impairs mitochondrial respiration owing to the loss of soluble components of the electron transport chain, such as cytochrome c, hence compromising the ability of mitochondria to preserve metabolic homeostasis and ionic equilibrium across the IMM45. In this setting, BAX–BAK1 pores in the OMM enable the extrusion of the IMM into the cytosol (driven by increasing osmotic pressure in the mitochondrial matrix), culminating in IMM breakdown and mtDNA spillage43,44. Intriguingly, it seems that a large fraction of mtDNA exiting mitochondria does not diffuse freely in the cytosol but instead remains associated with the permeabilized organelles43 (which explains, at least in part, the ability of mitophagy to inhibit cGAS signalling driven by MOMP)46,47 (see later). Recent data indicate that the relative levels of BAX and BAK1 have a key role in determining the speed of mtDNA release in the course of MOMP, largely reflecting structural and kinetic aspects of pore assembly by either of these proteins in the absence of its counterpart, with BAK1 accelerating and BAX decelerating the response in the context of negligible effects on the release of proapoptotic factors normally confined within the intermembrane space, such as diablo IAP-binding mitochondrial protein (DIABLO; best known as SMAC)48.

Of note, although MOMP was initially conceived as an ‘all-or-nothing’ phenomenon that would necessarily result in cell death45, sublethal instances of MOMP that affect only a fraction of the mitochondrial pool and do not engage RCD — known as ‘minority MOMP’ — have been described in various physiological and pathological settings49,50. In the context of sublethal stress conditions, minority MOMP, which is actively prevented by mitochondrial fusion as a consequence of BCL-2 redistribution51, not only enables inflammatory reactions driven by mtDAMPs52,53 but also favours DNA damage upon activation of DNA fragmentation factor subunit-β (DFFB)49, which can further activate inflammatory pathways.

Intriguingly, various examples of BAX-independent and BAK1-independent mtDNA release have recently been described. For example, proteolytically activated BH3-interacting domain death agonist (BID) has been shown to function as a bona fide mitochondrial pore-forming protein (rather than as an activator of BAX and BAK1) in human cells responding to Shigella infection54. Along similar lines, mild mitochondrial stress that does not result in MOMP and RCD has been associated with mtDNA release via a voltage-dependent anion channel (VDAC)-dependent mechanism37,55. This is particularly intriguing as various isoforms of VDAC are involved in the regulation (but are not required for the execution) of mitochondrial permeability transition (MPT)-driven regulated necrosis56 (Box 1), suggesting that MPT may also drive mtDNA leakage. In line with this possibility, pharmacological inhibitors of the permeability transition pore complex, which is commonly viewed as the core mediator of MPT, have been shown to limit cytosolic mtDNA accumulation and expression of interferon-stimulated genes55,57. Moreover, mutant TAR DNA-binding protein (TARDBP) has been reported to accumulate at mitochondria and initiate an MPT-like response that involves VDAC oligomerization and culminates in cGAS activation in models of amyotrophic lateral sclerosis58. As MPT generally does not cause robust caspase activation (see later), it would be tempting to suggest that mtDNA release by MPT leads to greater cGAS activation than mtDNA release by MOMP. However, the permeability transition pore complex forms pores of 1.5–3-nm diameter59, whereas dsDNA requires pores with a diameter of 3 nm or greater to diffuse60. Moreover, MPT seems to proceed efficiently with only one to nine open permeability transition pore complexes per mitochondrion, which is less than the number of BAX–BAK1 oligomers believed to underlie MOMP (more than 20)48,61. That said, both MPT and MOMP can ultimately result in the complete breakdown of mitochondrial membranes to enable mtDNA release. Molecular studies elucidating the possibility that MPT may be superior to MOMP at enabling cGAS activation by cytosolic mtDNA are urgently awaited.

In summary, despite several unknowns, an abundant literature shows the ability of mtDNA to drive potent inflammatory responses upon engagement of cGAS and STING1, especially in conditions of limited apoptotic caspase activation (for example, upon MPT, or upon MOMP in the presence of viral caspase inhibitors)42 (Fig. 1).

Inflammasome signalling

In addition to being a potent cGAS agonist, cytosolic mtDNA can also drive the activation of inflammasomes, which are supramolecular platforms for the activation of caspase 1 (CASP1), CASP4 and CASP5 (as well as CASP11, the mouse homologue of human CASP4 and CASP5)62. Initial work on inflammasomes, specifically the inflammasome that contains NLR family pyrin domain-containing 3 (NLRP3) as a sensing component, revealed that various microorganism-associated molecular patterns elicit robust CASP1 activation and consequent proteolytic maturation of IL-1β and IL-18 (refs.63,64). Subsequent work identified oxidized mtDNA released into the cytosol upon mitochondrial dysfunction as a potent NLRP3 inflammasome activator65 and also delineated a feedforward loop through which inflammasome activation facilitates mtDNA release via a ROS-dependent mechanism that connects downstream NLRP3 activation to increased upstream MPT66,67. In line with the observation that a significant fraction of mtDNA accessing the cytosol remains associated with permeabilized mitochondria43, both NLRP3 and the inflammasome adaptor PYD and CARD domain containing (PYCARD; best known as ASC) relocalize to the mitochondria-associated endoplasmic reticulum upon MOMP, via a process that depends on mitochondrial ROS68. Moreover, optimal NLRP3 inflammasome signalling seems to involve a physical interaction between NLRP3 and thioredoxin-interacting protein (TXNIP), a nuclear protein that relocalizes to mitochondria during oxidative stress69,70, as well as cardiolipin, an IMM-restricted tetraacylated phospholipid71. Taken together, these observations delineate a close association between mitochondrial dysfunction and NLRP3 inflammasome activation.

Interestingly, although various ROS-dependent mechanisms have been invoked to sustain NLRP3 inflammasome signalling downstream of mitochondrial dysfunction72, ROS inhibitors seem to disrupt inflammasome priming (the synthesis of inflammasome components), but not activation (the acquisition of proteolytic activity)73. In line with this notion, recent data suggest that oxidative phosphorylation is involved in NLRP3 signalling driven by acute exposure to bacterial lipopolysaccharide (LPS) plus ATP74 through a ROS-independent mechanism linked to preserved intracellular ATP availability via phosphocreatine75, although ROS seem to be necessary for long-term inflammasome activation upon prolonged exposure to β-amyloid. Moreover, conventional inflammasome activators such as LPS plus ATP seem to (elicit and) require mtDNA neosynthesis for optimal NLRP3 inflammasome signalling76. Indeed, NLRP3 inflammasome activation by LPS can be impaired not only by mtDNA depletion66,77 but also by the deletion of Tfam76, which is required for mtDNA replication and maintenance78, the deletion of Cmpk2 (ref.76), which encodes a rate-limiting enzyme that supplies deoxyribonucleotides for mtDNA synthesis79, or the inhibition of mtDNA synthesis with the antidiabetic agent metformin80. That said, experimental strategies that deplete mtDNA also impair oxidative phosphorylation as they limit the abundance of specific subunits of the electron transport chain, suggesting that — at least in some settings — inhibition of NLRP3 signalling downstream of mtDNA depletion may result from, or at least be aggravated by, intracellular ATP shortage. These observations suggest that the relative importance of cGAS signalling versus NLRP3 signalling induced by cytosolic mtDNA may be largely influenced by cellular bioenergetics.

Of note, NLRP3 inflammasome activation by LPS plus ATP was initially proposed to occur independently of BAX and BAK1, the key MPT regulator peptidylprolyl isomerase F (PPIF), mitochondrial antiviral signalling protein (MAVS; a signal transducer in the molecular cascade detecting foreign, altered or ectopic RNA) and the mitophagy protein parkin RBR E3 ubiquitin protein ligase (PRKN)81. Rather, this was shown to involve CASP8 and components of the necroptotic machinery, notably receptor-interacting serine/threonine kinase 3 (RIPK3)81. More recently, BAX activation and BAK1 activation have been shown to promote CASP8-dependent NLRP3 inflammasome activation as a consequence of inhibitor of apoptosis protein (IAP) degradation, and to elicit an NLRP3-independent pathway culminating in IL-1β maturation82,83, most likely upon activation of a supramolecular complex commonly referred to as the ripoptosome83. The reasons underlying these apparently discrepant findings as to the roles of BAX, BAK1 and MOMP in conventional inflammasome activation remain to be fully elucidated. At least potentially, BAX-independent and BAK1-independent inflammasome activation may involve some degree of gasdermin cleavage (which can be catalysed by inflammasome-activated CASP1 as well as by MOMP-driven CASP3 activation)84, as both gasdermin D (GSDMD) and GSDME have been shown to permeabilize mitochondria and favour mtDNA release coupled to ROS production85,86. As an alternative, the MPT has been implicated in at least some instances of NLRP3 inflammasome signalling in a recent preprint (not peer reviewed)87.

mtDNA also activates inflammasomes that use absent in melanoma 2 (AIM2) as a sensing component88, which overall resemble their NLRP3-containing counterparts in terms of their capacity to elicit CASP1-dependent IL-1β and IL-18 maturation in response to cytosolic dsDNA of various origins89, including both foreign dsDNA90 and endogenous dsDNA91. Interestingly, whereas NLRP3 inflammasomes seem to preferentially respond to oxidized DNA65, their AIM2-containing counterparts have been suggested to preferentially recognize non-oxidized DNA76. That said, both Francisella tularensis infection and non-alcoholic fatty liver disease drive mitochondrial damage coupled to AIM2 inflammasome activation upon mitochondrial ROS generation and cytosolic accumulation of oxidized mtDNA92,93. Along similar lines, Pseudomonas aeruginosa infection triggers the activation of NLR family CARD domain-containing 4 (NLRC4) inflammasomes via a process that involves ROS generation and oxidized mtDNA release by mitochondria94. Taken together, these observations highlight the importance of oxidation for robust inflammasome signalling elicited by mtDNA.

In summary, mtDNA and mitochondrial ROS function as major DAMPs for inflammasome activation at the core of a complex pathway that intersects at multiple nodes with the molecular machinery for regulation of RCD (Fig. 1).

Other inflammatory pathways

mtDNA and other mitochondrial components can also elicit inflammatory reactions via various PRRs distinct from cGAS and inflammasomes95.

Naked as well as protein-bound mtDNA molecules are potent activators of Toll-like receptor 9 (TLR9) and advanced glycosylation end product-specific receptor (AGER; also known as RAGE)96,97, two PRRs that are abundantly expressed in the endosomal compartment of myeloid cells (notably neutrophils)13. Naked mtDNA largely functions as a TLR9 agonist96, reflecting the limited methylation of CpG islands found in the mitochondrial genome98 and hence its considerable similarity to bacterial DNA (the prototypical TLR9 activator)99. Conversely, mtDNA complexed with TFAM or HMGB1, which relocalizes from the nucleus to the cytosol in the course of various stress responses (including inflammasome activation)100, has immunostimulatory effects upon binding to TLR9 or AGER97,101. Recombinant TFAM has also been shown to drive cytokine secretion by cultured human monocytes and primary microglial cells102, but the underlying mechanisms and pathophysiological relevance of this process remain to be clarified.

Importantly, activation of these inflammatory pathways most often requires mtDNA (be it naked or protein bound) to be released into the extracellular microenvironment as a consequence of RCD, where mtDNA functions as an autocrine, paracrine or endocrine immunostimulatory cue12. However, cells with increased vesicular trafficking (such as plasmacytoid dendritic cells) have been shown to engage endosomal TLR9 signalling upon mild mitochondrial dysfunction as a consequence of cytosolic mtDNA accumulation coupled to autophagic uptake and endosomal shuttling101,103,104. Along similar lines, signalling through glutamate receptors as a consequence of metabolic changes culminating in extracellular glutamate accumulation has recently been shown to promote the release of mtDNA-loaded mitochondria-derived vesicles (MDVs) by breast cancer cells independently of RCD. This is a PTEN-induced putative kinase 1 (PINK1)-dependent process that results in autocrine and/or paracrine TLR9 activation105. At odds with that study105, PRKN (a PINK1 interactor) has been proposed to actively divert mtDAMPs from inclusion in MDVs that are secreted as part of mitochondrial quality control106,107. The reasons underlying this apparent discrepancy remain to be fully elucidated. Of note, mtDAMP-containing MDVs seem to be actively produced by monocytes responding to inflammatory cues during sepsis, potentially resulting in limited neutrophil chemotaxis108. Moreover, although unrelated to inflammatory responses, the horizontal transfer of mtDNA-containing MDVs from live cancer-associated fibroblasts to tumour cells has been observed in various models of hormone-resistant breast cancer109. These studies exemplify settings in which mtDNA can be released (and potentially function as a DAMP) by cells that are not succumbing to RCD.

Another inflammatory pathway that can be stimulated by mitochondria involves the PRRs that respond to foreign, altered or ectopic RNA — known as RIG-I-like receptors (RLRs)110. In this setting, mitochondria can not only function as a source of RLR-activating RNA species as a consequence of mitochondrial dysfunction and MOMP53,111 but can also provide a cellular scaffold for optimal RLR signalling by hosting the key signal transducer MAVS on the OMM112. Specifically, mitochondria have been shown to release mitochondrial RNA (mtRNA) species that activate the RLR melanoma differentiation-associated protein 5 (MDA5; also known as IFIH1) upon depletion of polyribonucleotide nucleotidyltransferase 1 (PNPT1; a key component of the supramolecular complex responsible for mtRNA degradation), via a mechanism that involves BAX–BAK1-driven MOMP111. Moreover, mtDNA breaks have been reported to drive the BAX–BAK1-dependent release of mtRNA into the cytosol, which activates retinoic acid-inducible gene I (RIG-I; also known as DDX58) but not MDA5 (ref.53). Why some mtRNA species preferentially activate MDA5 versus RIG-I remains to be elucidated. At least potentially, this may reflect cell type-specific differences in the expression of specific PRRs or signal transducers thereof, as suggested by recent work on PNPT1 silencing in human pancreatic β-cells113. Finally, release of the mitochondrial matrix protein Era-like 12S mitochondrial rRNA chaperone 1 (ERAL1) via BAX–BAK1 pores formed in response to viral infection has recently been shown to sustain antiviral responses by promoting MAVS polymerization at the OMM, which is crucial for optimal inflammatory responses driven by RLR-activating viral RNA114. Intriguingly, MAVS resembles many other PRRs in their ability to control both cellular fate and inflammatory responses115. Thus, MAVS can promote RCD as part of its cell-intrinsic homeostatic function, but this is also connected to immune signalling via TANK-binding kinase 1 (TBK1), IκB kinase (IKK) and NLRP3 inflammasome activation116. These observations are in line with an abundant literature demonstrating that the molecular regulation of RCD may have evolved to control the kinetics and immunological manifestations of the process rather than the occurrence of cell death itself23,117,118.

At least four other mitochondrial components, namely SMAC, N-formyl peptides, cardiolipin and cytochrome c, have been shown to promote inflammatory responses. SMAC release downstream of MOMP is transduced in both proapoptotic and pro-inflammatory signalling pathways via members of the IAP family119. In addition to unleashing caspase activity, inhibition of IAPs by cytosolic SMAC (as well as by pharmacological agents commonly known as SMAC mimetics) shifts NF-κB signalling from the canonical to the non-canonical pathways upon stabilization of mitogen-activated protein kinase kinase kinase 14 (MAP3K14; best known as NIK)120,121, a process that is orchestrated by BAX–BAK1 oligomers122. Accordingly, loss of the IAP-encoding genes Birc2 and Birc3 in adult mice causes aberrant cell death and inflammation, which are fully rescued by Casp8 deletion plus pharmacological inhibition of NIK123. Moreover, SMAC mimetics have been shown to mediate anticancer effects that at least in some models involve the activation of antitumour immune responses upon macrophage repolarization124. Intriguingly, SMAC mimetics can also have direct immunomodulatory effects on T cells, for example by reprogramming CD4+ T cell differentiation from a T helper 17 (TH17)-type phenotype to a TH2-type phenotype125. Thus, whereas cytosolic mtDNA or mtRNA generally elicits a multipronged TBK1-dependent inflammatory response, SMAC-driven inflammation seems to involve mainly altered NF-κB signalling.

N-formyl peptides and cardiolipin normally reside in the mitochondrial matrix and IMM, respectively, but the structural disruption that accompanies late-stage RCD generates cellular fragments containing mitochondrial remnants that — unless taken up by professional phagocytes126 — continue to degrade in the extracellular microenvironment. Extracellular N-formyl peptides function as potent neutrophil activators upon binding to formyl peptide receptor 1 (ref.96). Extracellular cardiolipin not only promotes overexpression of the non-polymorphic MHC class I-like molecule CD1d on antigen-presenting cells (APCs)127 but also can be found associated with CD1d on the APC surface, resulting in activation of a cardiolipin-specific population of T cells expressing the unconventional γδ T cell receptor128. Whether cardiolipin is loaded on CD1d directly at the APC surface or intracellularly remains to be clarified. Extracellular cytochrome c has been proposed to have TLR4-dependent immunostimulatory effects in cultured cell models of human microglia129, but whether this pathway is activated as a response to RCD in vivo has not been specifically addressed. Along similar lines, the intra-articular injection of recombinant cytochrome c has been suggested to promote arthritis in mice via a mechanism that involves neutrophils and monocytes130, but whether cytochrome c accumulating in the synovial microenvironment downstream of RCD in vivo aetiologically contributes to arthritis remains to be formally demonstrated.

Finally, although ATP and haem are not mitochondrial components in the strictest sense, they are produced by mitochondria and mediate robust immunomodulatory effects upon release into the extracellular microenvironment131,132. Specifically, ATP released by cancer cells undergoing immunogenic cell death12, via a mechanism that involves lysosomal secretion and pannexin 1 channels133, has been shown to recruit APC precursors to the tumour microenvironment134 through a P2RY2-dependent mechanism14, and to stimulate them through P2RX7, culminating in inflammasome activation and IL-1β synthesis in support of immunosurveillance15. Extracellular haem, which functions as a mixed TLR4 and AGER agonist135,136, has been associated with multiple immunomodulatory effects, including endothelial cell and microglial activation135,137, as well as IL-1β release by macrophages downstream of NLRP3-dependent CASP1 activation plus NLRP3-independent CASP4 and CASP5 activation coupled to RCD132. The mechanistic involvement of TLR4 or AGER in this latter pathway has not been clarified.

In summary, numerous mitochondrial components and products can promote inflammation through various mechanisms (Fig. 1). Whether such a preferential position in the control of inflammatory reactions reflects the evolutionary origin of mitochondria remains to be formally established. Irrespective of this possibility, eukaryotic cells have evolved a wide range of mechanisms to control inflammatory responses elicited by mitochondria, as discussed next.

Regulation of mtDAMP signalling

As MOMP participates in multiple physiological processes, including cellular differentiation, embryonic and postembryonic development, and the maintenance of adult tissue homeostasis36,50 (Box 3), numerous safeguards have evolved to prevent unwarranted mitochondria-driven inflammation (Fig. 2). Here we describe the mechanisms through which apoptosis and autophagy suppress inflammatory responses potentially driven by mitochondria in physiological settings.

Fig. 2: Roles of apoptosis and autophagy in the inhibition of inflammatory responses elicited by mitochondria.
figure 2

Intrinsic apoptosis proceeds with the formation of BCL-2-associated X, apoptosis regulator (BAX) and BCL-2 antagonist/killer 1 (BAK1) pores at the outer mitochondrial membrane, resulting in the cytosolic accumulation of cytochrome c and consequent activation of apoptotic caspases. Caspase 9 (CASP9)-driven CASP3 activation inhibits inflammatory responses elicited by dysfunctional mitochondria by catalysing the cleavage of cyclic GMP–AMP synthase (cGAS), mitochondrial antiviral signalling protein (MAVS) and interferon regulatory factor 3 (IRF3). A similar inhibitory effect results from the disposal of dysfunctional mitochondria via PTEN-induced putative kinase 1 (PINK1)- and parkin (PRKN)-dependent mitophagy. This is promoted (at least initially) by the capacity of TANK-binding kinase 1 (TBK1) — which is activated by phosphorylation (P) following cGAS-driven stimulator of interferon response cGAMP interactor 1 (STING1) signalling and retinoic acid-inducible gene I (RIG-I)-driven MAVS signalling downstream of the cytosolic accumulation of mitochondrial DNA (mtDNA) or mitochondrial RNA (mtRNA) — to phosphorylate optineurin (OPTN) and hence improve the ability of OPTN to recruit sequestosome 1 (SQSTM1; best known as p62) to ubiquitylated proteins at the outer mitochondrial membrane such as mitofusin 2 (MFN2). Engagement of p62 culminates in the recruitment of forming autophagosomes via lipidated microtubule-associated protein 1 light chain 3 beta (MAP1LC3B; best known as LC3-II in its lipidated form). Of note, PRKN also mediates the ubiquitylation (Ub)-dependent inactivation of BAK1. Moreover, PRKN-dependent mitophagy seems to prevent mitochondrial damage-associated molecular patterns (mtDAMPs) from being incorporated in the mitochondria-derived vesicles (MDVs) that are normally released as part of mitochondrial quality control in a PINK1-dependent manner. The underlying mechanisms, however, remain to be fully elucidated. General autophagy can likewise suppress inflammatory responses driven by mitochondrial dysfunction, at least in part reflecting its ability to degrade NLR family pyrin domain-containing 3 (NLRP3) inflammasomes. ATG, autophagy-related protein; IFNβ1, interferon-β1; ROS, reactive oxygen species; TNF, tumour necrosis factor.


In the context of apoptosis, widespread MOMP results in the assembly of a cytosolic supramolecular CASP9-containing platform that culminates in the activation of executioner caspases such as CASP3, CASP6 and CASP719 (Box 1). Although executioner caspases have long been considered as the causative factors of apoptotic RCD, accumulating evidence suggests that they might instead have a major role in controlling its kinetics and immunological manifestations23,117,118. In line with this notion, pharmacological or genetic inhibition of CASP9 or executioner caspases (most often CASP3) is required for MOMP-inducing agents, including BH3 mimetics and radiotherapy, to drive robust synthesis of interferon-stimulated genes but does not quantitatively affect the cytotoxic response to these agents28,29,122,138,139. At least in part, this reflects the ability of executioner caspases to cleave and hence inactivate not only cGAS but also interferon regulatory factor 3 (IRF3), which signals downstream of STING1, and MAVS140. Moreover, executioner caspases promote the exposure of phosphatidylserine on the surface of dying cells141, which rapidly recruits phagocytes for the immunologically silent uptake and degradation of dying cells142. Finally, even in the absence of a phagocytic system, when executioner caspases are inactive, the transition between MOMP (when cells retain plasma membrane integrity and at least some metabolic functions) and cell death (which is characterized by plasma membrane permeabilization and complete metabolic shutdown) is delayed, potentially allowing an extended temporal window for the synthesis and secretion of pro-inflammatory factors downstream of MOMP143.


Sublethal (or prelethal) MOMP activates robust mitophagic responses that dispose of permeabilized or otherwise dysfunctional mitochondria via lysosomes144,145, which limits the availability of mitochondrial components, including mtDNA and ROS, for PRR signalling or release into the extracellular microenvironment upon RCD146. In line with this, pharmacological or genetic strategies for autophagy inhibition or impaired lysosomal degradation have been consistently associated with increased signalling through cGAS46, the inflammasome66 and TLR9 (ref.103) as a consequence of increased availability of mtDAMPs in the cytosol. Similar results have been obtained with experimental approaches for the selective inhibition of mitophagy (rather than autophagy in general), including deletion of the key mitophagy genes Pink1 and Prkn147,148, administration of palmitic acid149 and stimulation with tumour necrosis factor47. Furthermore, active inflammasome signalling has been shown to engage an NF-κB-dependent response that promotes the PRKN-dependent mitophagic clearance of permeabilized mitochondria77 as well as the autophagic clearance of inflammasomes themselves150, most likely as part of an adaptative pathway aimed at the restoration of cellular and inflammatory homeostasis. A similar mechanism is elicited (at least initially) by cGAS signalling, which reportedly promotes mitophagy upon activation of TBK1 or IKKα downstream of STING1 (refs.151,152). PRKN has also been shown to promote the recovery of cellular homeostasis in the context of sublethal MOMP by catalysing the ubiquitylation-dependent inactivation of BAK1 (ref.144). Conversely, mitophagy inactivation has been shown to accompany full-blown inflammatory responses driven by unrecoverable mitochondrial dysfunction, at least in part reflecting the CASP1-dependent cleavage of PRKN elicited by robust inflammasome activity153. Of note, PRKN-dependent mitophagy may also prevent the packaging of mtDAMPs into MDVs that are secreted as part of mitochondrial quality control107, which would otherwise drive inflammatory reactions in neighbouring cells106. Taken together, these observations delineate a refined system whereby autophagy (in particular, mitophagy) prevents the initiation and favours the inhibition of inflammatory responses driven by mitochondria, which require mitophagy inhibition (potentially involving CASP1-dependent PRKN degradation plus a feedforward loop linking robust inflammasome activation back to aggravated mitochondrial dysfunction)66,153 to proceed unrestricted (Fig. 3). By contrast, general autophagy is required for optimal ATP secretion (and consequent immunostimulation) in the course of immunogenic cell death, largely reflecting the ability of proficient autophagic responses to preserve intracellular ATP levels (rather than a direct effect of autophagy on mitochondria)154.

Fig. 3: The mitophagy rheostat in the control of mitochondria-driven inflammation.
figure 3

a | When only a small number of mitochondria are permeabilized, limited signalling via cyclic GMP–AMP synthase (cGAS) and the NLR family pyrin domain-containing 3 (NLRP3) inflammasome promotes mitophagy associated with the recruitment of parkin (PRKN) to dysfunctional mitochondria, TANK-binding kinase 1 (TBK1)-dependent optineurin (OPTN) phosphorylation (P) and consequent engulfment of mitochondria by forming autophagosomes. This engages a feedback mechanism that enables the restoration of cellular homeostasis in the absence of robust inflammatory responses. b | Conversely, when the functions and integrity of a large number of mitochondria are compromised, robust cGAS and NLRP3 inflammasome signalling is accompanied by mitophagy inhibition — as a consequence of caspase 1 (CASP1)-dependent cleavage of PRKN, despite OPTN phosphorylation — and increased NLRP3-dependent mitochondrial dysfunction, resulting in a feedforward loop that maximizes inflammation in the context of lost cellular homeostasis. Taken together, these mechanisms identify a rheostat that determines a threshold for recovered cellular homeostasis in the context of suppressed inflammation versus compromised cellular homeostasis in the context of acute inflammatory responses. MOMP, mitochondrial outer membrane permeabilization; mtDNA, mitochondrial DNA; ROS, reactive oxygen species; STING1, stimulator of interferon response cGAMP interactor 1.

Although these are not the only mechanisms through which eukaryotic cells fine-tune inflammatory reactions elicited by mitochondria (for example, cGAS and inflammasomes seem to inhibit each other, at least in some settings)155,156, apoptosis and autophagy exemplify molecular systems that have enabled the preservation of otherwise potentially detrimental inflammatory cues throughout the co-evolution of mitochondria and their host cells.

mtDAMP signalling in disease

Dysregulated mtDAMP signalling can be pathogenic and actively contribute to the aetiology of human disease in two opposing ways (Supplementary Table 1): inflammatory reactions driven by mtDAMPs may become disproportionate, thus fostering disorders with an (obvious or less obvious) inflammatory component; or such inflammatory reactions may be highly inefficient, ultimately enabling the emergence or persistence of infectious or neoplastic conditions.

Overactive mtDAMP signalling

Human disorders with an overtly inflammatory component that is mechanistically promoted by dysfunctional mitochondria include systemic lupus erythematosus (SLE), Crohn’s disease and multiple pulmonary and renal conditions1,4. Patients with SLE have increased circulating amounts of oxidized mtDNA as a consequence of platelet degranulation157 and impaired mitophagic responses to mitochondrial dysfunction in neutrophils (as well as autoantibodies to oxidized mtDNA), which culminate in pathogenic type I interferon responses158. Moreover, SLE is accompanied by defects in erythroid maturation that prevent the autophagic removal of mitochondria, which is a feature of normal mammalian erythropoiesis159. Therefore, mitochondria-containing erythrocytes are no longer recycled in an immunologically silent manner by the reticuloendothelial system, but instead promote potent cGAS activation and consequent type I interferon release159. In line with these findings, pharmacological inhibition of VDAC-dependent mtDNA release ameliorates lupus-like inflammation and symptoms in mice55. Finally, oxidized mtDNA is abundant within neutrophil extracellular traps, which also contribute to the aetiology of SLE via type I interferon production160.

Mutations in the autophagy-related 16-like 1 (ATG16L1) gene, which encodes a key component of the autophagy machinery22, are associated with an increased risk of Crohn’s disease161, correlating with the role of ATG16L1 in restricting accumulation of dysfunctional mitochondria and consequent inflammation in Paneth cells162. Paneth cells are functionally impaired in the ileal tissue of patients with Crohn’s disease163, a phenotype that can be recapitulated in mice by impairing mitochondrial homeostasis by deletion of the gene encoding the mitochondrial chaperone heat shock protein 1 (Hspd1)163 or of the gene encoding the mitophagy mediator prohibitin 1 (Phb1)164.

Lung biopsy samples from patients with interstitial lung disease or silicosis have increased STING1 levels, phosphorylation of its signal transducers TBK1 and IRF3, and/or higher than normal amounts of CXC-chemokine ligand 10 (CXCL10), which is elicited by type I interferon signalling165. Consistently, the sputum of patients with silicosis contains increased amounts of CXCL10 and dsDNA, which in mice exposed to silica originates from mitochondrial dysfunction165. Along similar lines, chronic obstructive pulmonary disease in humans has been associated with pulmonary mitochondrial dysfunction coupled to local ROS overgeneration166 and increased circulating levels of mtDNA and inflammatory cytokines167.

Evidence of STING1 hyperactivation has also been detected in the renal tubules of patients with acute kidney injury168. Moreover, CGAS and STING1 upregulation correlates with increased levels of inflammatory cytokines and fibrosis in patients with chronic kidney disease, who generally have low levels of TFAM expression (required for mtDNA replication and maintenance) in the kidney169. Optineurin, a component of mitophagy signalling, is often downregulated in biopsy samples from individuals with diabetic kidney disease170. Consistent with a prominent role for mitochondrial dysfunction in kidney disease, abrogation of mitophagy by Pink1 and Prkn co-deletion aggravates ischaemic acute kidney injury in mice171. Moreover, deletion of Tfam in mouse kidney tubule cells favours aberrant mtDNA packaging, cytosolic mtDNA accumulation and ultimately pathogenic cGAS-dependent inflammation169.

In addition, it is now clear that various conditions originating from non-inflammatory insults, including neurodegenerative diseases as well as hepatic and cardiovascular diseases, ultimately involve inflammatory processes that in some settings can result from mitochondrial dysfunction2,3. For example, missense mutations in PRKN and PINK1 are associated with familial Parkinson disease172, correlating with increased circulating levels of mtDNA and inflammatory cytokines147,173. Consistent with this, old Prkn–/– mice develop Parkinson disease-like symptoms together with structural and functional abnormalities of mitochondria in the brain174. Along similar lines, Alzheimer disease is accompanied by inflammatory responses driven by microglial cells that engulf mtDNA released by dying neurons175. Moreover, post-mortem brain samples from patients with Alzheimer disease show the accumulation of damaged mitochondria and signs of mitophagy inhibition, and experimental activation of mitophagy limits disease progression in a mouse model of Alzheimer disease176. Of note, defective mitophagy in Alzheimer disease may originate from alterations in the lipid profile of microglial cells (notably cholesterol accumulation)177, perhaps explaining, at least in part, the strong link between apolipoprotein E mutations and the incidence of Alzheimer disease in humans178.

Additional disorders in which mitochondrial dysfunction upstream of inflammation may have an aetiological role include cardiac maladaptation upon myocardial infarction179, septic shock180, rheumatoid arthritis47,130, sickle cell disease137, trauma96, intracerebral haemorrhage135, non-alcoholic fatty liver disease and non-alcoholic steatohepatitis93, liver failure181 and obesity182. Discussing each of these conditions in detail is beyond the scope of this Review, but these observations suggest that various additional disorders with an inflammatory component may be linked to mitochondrial dysfunction.

Inefficient mtDAMP signalling

The deregulation of inflammatory reactions elicited by mitochondrial dysfunction has broad, context-dependent effects on the pathogenesis of viral infections and cancer. Inhibition of MOMP or inflammatory responses driven by MOMP52 — reflecting, for example, the expression of viral proteins that inhibit MOMP, mtDNA, cGAS, STING1 or TBK1 (refs.41,183,184,185) or the hyperactivation of autophagy in cancer cells22 — favours viral persistence and tumour progression by impairing immunosurveillance.

For example, co-deletion of BAX and BAK1 from human cervical cancer HeLa cells not only promotes the growth of the intracellular pathogen Chlamydia trachomatis but also suppresses the ability of infected cells to secrete pro-inflammatory cytokines that would engage antibacterial immune responses upon sublethal MOMP52. HSV-1 encodes a conserved nuclease that actively depletes mtDNA during infection184, highlighting a direct mechanism by which this human pathogen avoids immune responses elicited by mtDAMPs. A similar function is mediated by non-structural protein 1 (NS1) of influenza virus41. Specifically, NS1 sequesters mtDNA molecules that access the cytosol in a MAVS-dependent manner during infection (driven by the influenza virus protein M2), thus preventing them from activating cGAS41. Accordingly, both Cgas–/– mice and Sting1–/– mice (but not Mavs–/– mice) exhibit increased viral titres upon influenza virus infection compared with their wild-type littermates, correlating with reduced pulmonary levels of type I interferon41.

Along similar lines, the HSV-1 virulence factor ICP27 inhibits STING1 and TBK1 (ref.183), the latter of which is required for the control of infection by mice, largely reflecting its ability to elicit NF-κB (rather than IRF3) responses186. Moreover, hepatitis B virus X protein promotes the ubiquitylation-dependent degradation of cGAS in infected cells185, and defects in the cGAS–STING1 pathway have been linked to increased sensitivity to hepatitis B virus infection in both human and mouse hepatocytes187. In these cases, however, the relative contribution of mtDAMP signalling to inflammatory reactions that would promote pathogen control has not been directly quantified.

Likewise, multiple cancer cell types harness autophagy46 and MOMP-driven CASP3 activation138,139 to avoid antitumour immune responses elicited by radiotherapy via mtDNA-dependent cGAS signalling. Conversely, autophagic defects that result in compromised ATP secretion in the context of immunogenic cell death not only support the resistance of cancer cells to various chemotherapeutic agents that engage antitumour immunity, such as anthracyclines and oxaliplatin15,154, correlating with inefficient recruitment and activation of myeloid cells to the tumour microenvironment134, but also promote malignant transformation in numerous immunocompetent mouse models of early oncogenesis188, a process that is under strict control by immunosurveillance189.

That said, the pathogenesis of some viral and bacterial infections involves late-stage uncontrolled inflammation, which at least in some cases originates from mitochondrial dysfunction37,190,191,192. Similarly, chronic low-grade inflammation generally contributes to oncogenesis and tumour progression7, and at least in some scenarios this is initiated by mitochondrial dysfunction193,194,195,196. Finally, SMAC release upon minority MOMP driven by Helicobacter pylori seems to have a role in both the short-term and the long-term detrimental effects of the infection as it drives pathogenic inflammation as well as DNA damage in the absence of overt RCD, ultimately promoting malignant transformation197.

Therapeutic prospects

Mitochondria have long been considered master regulators of (at least some forms of) RCD45,56, but accumulating evidence from various experimental settings now shows that the disruption of mitochondrial functions and structural integrity that accompanies RCD is also closely associated with an inflammatory response to preserve organismal homeostasis20. In keeping with this, the deregulation of inflammatory responses elicited by mitochondrial components or products has been shown to contribute to numerous human disorders, ranging from diseases that are driven by excessive inflammation to diseases that are enabled by inefficient inflammatory reactions95.

Importantly, most (if not all) of these conditions are managed clinically with therapeutic interventions that target the effector phase of inflammation, such as cytokine-neutralizing agents198, or target PRRs and their signal transducers, such as STING1 agonists199. By contrast, little attention has been given to the possibility of modulating inflammation through mitochondria-targeted agents, which may reflect the novelty of research in this area as well as a relatively small number of pharmacological interventions to target mitochondrial functions, particularly MOMP and MPT, in patients.

As it stands, only one drug that directly targets the molecular machinery of MOMP is licensed for use in humans, the selective BCL-2 inhibitor venetoclax (which is currently used for the treatment of some haematological malignancies and is under clinical evaluation for other neoplasms, including some solid tumours)24. The development of less selective inhibitors of BCL-2-like proteins200, such as navitoclax and ABT-737, has been troublesome (with the latter being discontinued) as they both mediate on-target, dose-limiting thrombocytopenia201. Although venetoclax and navitoclax would both be expected to promote inflammatory reactions that may support anticancer immunity by blocking the MOMP-inhibiting functions of BCL-2 (ref.202), they have been developed and are currently used or being tested as cytotoxic drugs against tumours that are ‘addicted’ to BCL-2-like proteins for survival9. Similar considerations apply to the use of pharmacological BAX activators to initiate MOMP, which are also in preclinical development to overcome apoptosis resistance in cancer203,204.

BAX inhibitors are being developed as cytoprotective agents for cardiovascular disorders, including chemotherapy-driven cardiotoxicity205,206. Although these agents would be expected to limit MOMP and consequently reduce inflammatory reactions driven by mitochondria, recent findings raise the intriguing possibility that (at least in BAK1-competent cancer cells) pharmacological BAX inhibitors may accelerate mtDNA release driven by MOMP (in the absence of major effects on the kinetics of caspase activation), potentially offering a window for inducing cGAS signalling before caspase-dependent cGAS cleavage and inactivation48,207. This hypothesis remains to be experimentally tested.

Cyclosporine A, a PPIF-targeting agent with prominent MPT-inhibitory activity, is also approved for use in humans56. Intriguingly, cyclosporine A is commonly used as an immunosuppressive drug for the management of autoimmune disorders and the prophylaxis of transplant rejection, largely on the basis of its ability to bind the PPIF-like cytosolic protein PPIA, ultimately resulting in lymphocyte inhibition through blocking calcineurin208. At least potentially, however, part of the immunosuppressive effects of cyclosporine A could result from MPT inhibition and the consequent suppression of inflammatory reactions driven by permeabilized mitochondria, a possibility that has not yet been formally addressed.

In conclusion, although mitochondria are clearly master regulators of inflammation, additional research is needed to address outstanding questions and surmount existing obstacles (Box 4), with the ultimate objective of targeting mitochondrial functions as a means to control inflammatory reactions in patients. As an additional layer of complexity, specific pharmacological modulators of autophagy for clinical use remain elusive209, and the development of emricasan, a pan-caspase inhibitor introduced into clinical testing for the management of cirrhosis, seems to have stalled210. Despite these and other caveats, we surmise that an improved understanding of the molecular mechanisms linking mitochondrial dysfunction to intracellular and extracellular DAMP signalling will ultimately unlock the development of mitochondria-targeting drugs for the control of inflammation.