The Cytosolic DNA-Sensing cGAS–STING Pathway in Cancer

Cancer has long been described as a wound that does not heal. This important paradigm was first proposed by Rudolph Virchow and highlights the parallels between cancer formation and inflammation (1, 2). In normal physiologic response to injury, several mechanisms regulate the timely termination of wound-healing processes (3). However, mounting evidence now paints a complex landscape in which unresolved inflammation is a potent driver of tumorigenesis (4–6). The cytosolic DNA-sensing cGAS–STING pathway has emerged as a potential mechanism to drive inflammation-mediated tumorigenesis (7). Indeed, chronic activation of cGAS–STING and its downstream effector programs, such as TBK1, have been linked with persistent inflammation and cancer progression (7, 8).

Cyclic GMP-AMP synthase (cGAS) is a cytosolic DNA sensor that serves to mount an immune response against the invasion of microbial pathogens such as viruses (9). Activation of cGAS, in turn, stimulates the adapter protein STING to trigger interferon (IFN) signaling (10). The existence of homologs for cGAS and STING across both eukaryotes and prokaryotes suggests that DNA sensing is an evolutionarily conserved mechanism against pathogenic infections (11–15). Beyond the antimicrobial function of cGAS and STING, recent evidence has expanded their roles in cancer, including other cellular functions such as regulation of DNA repair and autophagy. In this review, we discuss the dichotomous roles of cGAS–STING in tumorigenesis and the profound implications of this pathway for therapeutic approaches against cancer.

cGAS is activated by interacting with double-stranded DNA (dsDNA) in a sequence-independent manner (9, 14, 16). The DNA ligands bind with cGAS in a minimal 2:2 complex to induce conformational changes that allow cGAS to catalyze ATP and GTP into 2′,3′-cyclic GMP-AMP (cGAMP), a cyclic dinucleotide comprising both 2′–5′ and 3′–5′ phosphodiester linkages (17–19). Interestingly, longer DNA is more potent in activating cGAS and promotes liquid-like droplet formation, in which cGAS and dsDNA are spatially concentrated for efficient cGAMP synthesis (20–22). This liquid–liquid phase transition depends on the concentrations of cGAS and DNA, suggesting that a minimal threshold of DNA content must be surpassed to activate cGAS, such as in viral infections. The second messenger cGAMP then activates STING at the endoplasmic reticulum (ER), in which STING undergoes a higher-order oligomerization to form tetramers (23, 24) and translocates from the ER to ER-Golgi intermediate compartments (Fig. 1). At the Golgi, palymitoylation of STING has been proposed to recruit TANK binding kinase 1 (TBK1) and interferon regulatory factor 3 (IRF3; refs. 25, 26). Recent structural studies revealed that tetramerization of STING serves as a signaling platform to recruit and activate TBK1 dimers (26). In turn, TBK1 transphosphorylates the C-terminal domains of STING to recruit IRF3 for activation (26, 27), at which point IRF3 translocates to the nucleus and exerts its transcriptional function in expressing immune-stimulated genes (ISG) and type 1 IFNs (9, 19). In parallel, STING also activates IKK to mediate the induction of NFκB-driven inflammatory genes. Following activation, STING is trafficked to endolysosomes for degradation (28).

The cGAS–STING pathway serves as a crucial bridge between detection of viral pathogens and immune host defense mechanisms. Growing evidence supports the notion that cGAS is an indiscriminate immune sensor for virus-derived dsDNA, which in turn stimulates downstream IFN signaling. Several reports have demonstrated the protective role of cGAS against a broad range of DNA viruses, which includes vaccinia virus, Kaposi sarcoma–associated herpesvirus (KSHV), murine gammaherpesvirus 68, and herpes simplex 1 (29–31). Furthermore, cGAS also participates in the restriction of retroviruses, such as HIV (32). DNA intermediates generated from reverse transcription of the HIV genome may be recognized by cGAS to stimulate downstream STING–TBK1 signaling (33). This unique property of cGAS–STING has important therapeutic implications, as cancers frequently exhibit derepression of endogenous retroviral elements, which is accentuated by epigenetic modifying therapies (34–36).

Beyond canonical RNA sensors that are extensively reviewed elsewhere (37–39), the cGAS–STING DNA-sensing pathway may have indirect roles in facilitating immune responses against RNA viruses. Indeed, cGAS-deficient mice were more vulnerable to West Nile viral infections (30). However, the molecular details as to how cGAS–STING restricts RNA viruses remain poorly understood. One plausible mechanism may involve the inadvertent release of organelle-enclosed DNA as a pathologic consequence of viral infections. Indeed, Dengue RNA viruses were shown to induce the cytoplasmic exposure of mitochondrial DNA within infected cells to elicit cGAS–STING-dependent IFN signaling, illustrating a mechanism by which cGAS can act as an indirect immune sensor against the presence of dsRNA (40). In cancer, such a mechanism may further sensitize responses to dsRNA, as ISG-expressing tumors were revealed to upregulate RIG-I in the setting of chronic STING activation (41).

Despite the crucial role of the cGAS–STING–IFN pathway in antiviral defense, viral pathogens have acquired evasive strategies to escape host immune surveillance to drive pathologic states, such as chronic infection and cancer (42, 43). For instance, the HPV18 and human adenovirus 5 DNA tumor viruses encode viral oncoproteins E7 and E1A, respectively, that antagonize STING (44). The ability of viral oncoproteins to suppress the DNA-sensing pathway suggests that such oncogenic viruses have evolved dual functions to both promote malignancy and suppress innate immune signaling. Other cancer-related viruses such as KSHV and hepatitis B express IRF1, tegument protein ORF52, and viral polymerases that potently disrupt the cGAS–STING pathway (29, 45, 46). Interestingly, Eaglesham and colleagues revealed that viral poxins modulate STING signaling by degrading 2′3-cGAMP (42). It remains to be determined if such a mechanism exists in tumor-causing viruses to evade immune recognition. This wide repertoire of viral antagonists against cGAS–STING highlights the importance of evolutionary pressures for DNA tumor viruses to counter this DNA-sensing pathway and downstream antitumor programs, which will be described in later sections.

Unlike normal cells, cancer cells are often replete with cytosolic dsDNA. This DNA can be derived from multiple sources, including genomic, mitochondrial, and exogenous origins. The effects of ectopic cytosolic dsDNA in cancer are still poorly understood, yet evidence suggests that it may have both antitumorigenic and protumorigenic effects in a manner that is dependent on the specific context as well as stage of tumor progression. It is likely that during the early steps of transformation, cytosolic DNA leads to immune surveillance as well as cancer cell–intrinsic senescence. On the other hand, loss of key cell-cycle and immune-checkpoint effectors may enable cytosolic dsDNA to activate chronic inflammatory signaling associated with prosurvival and metastatic programs. Recent evidence suggests that chromosomal instability (CIN) is a primary source of cytosolic dsDNA. CIN is a hallmark of human cancer, and it is often associated with tumor progression, therapeutic resistance, and distant metastasis. Cancer cells harboring unstable genomes are prone to chromosome missegregation during mitosis. One consequence of such segregation defects is the generation of micronuclei, a reservoir of nuclear-derived genomic substrates, in a cell cycle–dependent manner (47). Micronuclear envelopes are prone to rupture and expose their genomic contents into the cytosol, which in turn triggers the cGAS–cGAMP–STING pathway (refs. 48, 49; Fig. 1). The regulators of cGAS–STING activation, such as nucleases (DNAses) and the process of cellular compartmentalization itself, are reviewed elsewhere (50). STING mediates the transcriptional activity of a broad repertoire of molecular programs, which include inflammation, senescence, autophagy, and metastasis. Following a similar mechanism, acute genomic stressors induced by radiation, cisplatin, and intrinsic DNA damage generate cytosolic DNA to activate cGAS–STING in cancer cells (7, 48, 49, 51). It is possible that the ability of CIN to promote cGAS–STING-mediated IFN responses explains why complex aneuploidy patterns that result from chromosome missegregation are relatively uncommon during the early steps of tumorigenesis and expand once the tumor develops tolerance for a chronic inflammatory milieu that is devoid of type I IFN (48, 49, 52–55).

Apart from the nuclear compartment, the mitochondria may serve as another genomic source to stimulate the cytosolic dsDNA-sensing pathway in cancer (Fig. 1). Malignant cells undergoing oxidative stress and mitochondrial dysfunction release mitochondrial DNA (mtDNA) in the cytosol (56). The molecular details that allow mtDNA to escape from the mitochondria require further investigation, but may involve permeabilization of the outer and inner mitochondrial membranes (57). These findings are congruent with the rationale behind several anticancer therapies targeting permeabilization of mitochondria, which results in the release of potent biomolecules to trigger cell death (58, 59). It is possible that these pharmacologic compounds induce the cytoplasmic leakage of mtDNA to promote cGAS–STING-mediated antitumor responses. As a strategy to avoid the DNA-sensing machinery, tumors harboring mitochondrial dysfunction have been shown to actively silence STING, which effectively ablates downstream IFN signaling (56).

On the other hand, tumors also acquire mtDNA from the extracellular milieu to engage the cGAS dsDNA-sensing cascade. Indeed, tumors deficient in mtDNA reconstituted their mtDNA pool by exosomal transfer from host cells to restore mitochondrial function and enhance metastatic potential (60, 61). Together with recent reports of the cGAS–STING pathway in driving malignant behaviors (62), these observations raise an interesting possibility that, once tolerant to cytosolic dsDNA signaling, cancer cells may seek extratumoral mtDNA to sustain cGAS–STING-driven tumorigenesis in addition to supplementing mitochondrial functions.

Interestingly, mtDNA and genomic DNA may not be synonymous with respect to their ability to activate cGAS. Recent evidence suggests that nucleosome-bound chromatin has a higher binding affinity for cGAS, but reduces the catalytic activity of cGAS in comparison with unchromatinized DNA (49, 63). These findings suggest that stearic hindrance and constrained architecture of cytosolic chromatin may limit cGAS-catalyzed generation of 2′3-cGAMP, which aligns with the possibility that genetic structures devoid of histones such as mtDNA are more efficient in activating cGAS. However, this concept remains controversial, as a separate study suggests that cytoplasmic chromatin increases cGAS activity (51).

Other sources of genomic substrates, such as apoptotic-derived DNA, exosomes, and transposable elements, may also elicit cGAS–STING activation in tumors (Fig. 1). For instance, horizontal transfer of DNA from dead HRAS^V12;c-MYC rat cancer cells engendered tumorigenic phenotypes in p53-deficient mouse fibroblasts (64). In addition, the uptake of stromal-derived and tumor-derived exosomes by cancer cells can also promote malignant behaviors (65, 66). The precise composition of the biomolecules within exosomes that are responsible for driving such effects is poorly understood. However, recent reports have linked IFN signaling with exosomal DNA detection by cGAS, as well as chronic cGAS–STING stimulation with metastasis (62, 67, 68). These findings raise an interesting prospect that delivery of exosome-derived DNA to cancer cells may trigger the cGAS–STING pathway to upregulate protumorigenic programs. Finally, dysregulation of chromatin remodeling genes and IFN signaling has been shown to induce transcriptional derepression of retroviral coding sequences residing in 3′ untranslated regions of IFN-stimulated genes in cancer cells (34). The resulting accumulation of dsRNA and reverse-transcribed dsDNA stimulates the RIG-I RNA and cGAS DNA-sensing pathways, respectively. Importantly, these immune pathways may further promote IFN responses, which in turn maintain transcriptional derepression, facilitating a positive feedback loop to enhance RNA and DNA recognition. With the advent of pharmacologic inhibitors targeting epigenetic modulators for cancer therapy, these observations raise an important question if long-term treatment with such agents might exacerbate tumor progression through chronic innate immune stimulation.

In particular cellular contexts, STING activation is not fully dependent on the cytosolic DNA sensor cGAS (Fig. 2). Beyond the classic cGAS–cGAMP–STING axis, Dunphy and colleagues have revealed an alternative mechanism in which the DNA-repair proteins ATM and PARP1 cooperate with the DNA binding protein IFI16 to promote noncanonical STING signaling in response to etoposide-induced DNA damage (69). In a similar vein, STING regulates cell-cycle progression in a cGAS-independent manner in specific tumor models, such as in HCT116 colorectal carcinomas (70). These results have important implications, as tumors lacking cGAS expression may still sustain active STING through other DNA binding partners. These observations open an opportunity to explore how other DNA sensors and adapter proteins may converge on the STING signaling platform and the molecular properties that dictate such engagements in cancer.

Beyond its function as an innate cytosolic sensor for dsDNA, cGAS may have noncanonical roles in the nucleus (48, 49). Contrary to its dominant 2′3-cGAMP production in the cytosol, cGAS is enriched in the nuclear compartment on long interspersed nuclear elements and centromeric satellite repeats (71). The mechanism in which cGAS gains nuclear entry remains unknown, but has been proposed to occur as a result of nuclear membrane disassembly in mitosis. It is worth noting that nuclear-localized cGAS exhibits limited responses to endogenous DNA compared with exogenous DNA, suggesting unknown regulators exist to dampen immune responses against self-DNA. Perhaps, the inhibitory effect of chromatin on cGAS catalytic activity is an evolutionary strategy to avoid immune recognition of self-DNA by preventing mitotic cells from ectopically activating cGAS after nuclear-envelope breakdown. This observation also raises the exciting possibility for cGAS to adopt alternative functions in the nucleus. For instance, the unique behavior of cGAS in binding nuclear elements highlights an interesting hypothesis that nuclear cGAS may exert functional roles in epigenetic or chromatin architecture modulation. Furthermore, nuclear cGAS has been shown to facilitate tumorigenesis through the inhibition of homologous recombination (HR) DNA repair in response to genotoxic stress–induced DNA damage (refs. 72, 73; Fig. 2). In normal cells, the ability of nuclear cGAS to impair HR might derive its evolutionary origin from selective advantage for such cells to counter DNA integration by proviruses (74–76).

The nuclear function of cGAS has important implications in tumor formation and cancer therapy. Several cellular safeguards exist to inhibit the propagation of preneoplastic cells harboring unresolved DNA damage (77, 78). In this setting, it is likely that nuclear cGAS will trigger cell-cycle arrest and apoptotic programs. Indeed, radiation exposure to bone marrow–derived macrophages with functional cGAS exhibited increased mitotic cell death compared with cGAS^−/− cells (72). On the other hand, tumors with high tolerance to DNA damage can provide a permissive environment for nuclear cGAS to optimally exert its tumorigenic effects in response to genotoxic stress. For example, chronic radiation exposure in such cells generates double-strand breaks and induces nuclear translocation of cGAS. Subsequently, nuclear cGAS may suppress the HR–DNA repair machinery to license further accumulation of DNA damage in the cancer genome, establishing a feed-forward cycle of chromosome instability. The extent of DNA damage is limited, as increasing chromosome instability beyond a tolerable threshold reduces tumor fitness (79). This notion highlights an intriguing role for nuclear cGAS in sensitizing tumors to DNA-damaging agents, independently of its role in type 1 IFN signaling.

Innate cytosolic DNA sensing plays a crucial role in mounting antitumor responses in both tumor cell–autonomous and non–cell-autonomous manners. This tumor surveillance mechanism has been well characterized and is mediated by infiltrating immune cells, such as natural killer (NK) cells and T cells, through IFN signaling (80, 81). Activation of cGAS–STING in cancer cells may serve as a barrier to early neoplastic progression through the upregulation of a battery of inflammatory genes, such as type 1 IFNs (Fig. 3). Importantly, activation of this pathway also mediates the secretion of proinflammatory cytokines, chemokines, proteases, and growth factors that are collectively termed senescence-associated secretory phenotype (SASP), which play an important role in restricting tumorigenesis (51, 53, 82). The resulting immune-stimulatory factors from this cascade can either attenuate tumor growth in a cancer cell–autonomous manner or recruit immune cells for tumor clearance (51, 52, 70). Indeed, suppression of STING in melanoma and pancreatic cancer cells led to reduced immune infiltration, which allowed increased tumor growth in vivo (83, 84).

Apart from cancer-specific cGAS–STING activation, the host may also harness this inflammatory pathway for tumor surveillance (Fig. 3). Antigen-presenting cells (APC), such as dendritic cells (DC) and macrophages, are thought to clear necrotic malignant cells (85, 86). Through an unestablished mechanism, tumor DNA is thought to be transferred and released into the cytosol of DCs and macrophages (87). The accumulation of tumor DNA, in turn, activates STING–IRF3-induced IFN signaling to enforce tumor-antigen presentation on DCs and, as such, cross-prime CD8⁺ T cells for antitumor immunity (87–89). Alternatively, tumors secrete cGAMP into the extracellular space, which is imported into host immune cells through the folate transporter SLC19A1 (90). Tumor-derived cGAMP subsequently activates host STING–IRF3 and induces NK-mediated tumor killing (81). Although DCs are suggested as primary responders to tumor cGAMP (88), it is possible that other cell types such as fibroblasts also possess such function. Further studies will be required to unveil the full repertoire of immune cells that detect tumor cGAMP and how tumor and host cGAS–STING might cooperate to facilitate tumor suppression.

Tumors with high levels of CIN face a unique challenge in which their chronic release of cytosolic DNA poses a risk to activate cGAS–STING-mediated IFN signaling and limit tumor growth. To circumvent such suppressive effects, cancer cells have adopted strategies to inhibit this pathway and drive protumorigenic programs. Cancer cells may silence the cytosolic DNA-sensing pathway to evade immune surveillance. Decreased protein expression of cGAS and STING has been shown in a small number of late-stage tumors (91–93). This is congruent with a pan-cancer analysis that revealed a small subset of tumors (~1%–25%) harboring increased methylation in the promoters of cGAS and STING compared with matched normal tissues (94). In this same report, however, reduced, rather than increased, promoter methylation was observed in the majority of tumors, including aggressive histologies such as pancreatic and thyroid cancers, suggesting that loss of cGAS and STING to facilitate immune evasion is not a common feature of cancer (95).

The majority of tumors retain some level of cGAS and STING proteins, which suggests that direct silencing of these genes is not the dominant mechanism for immune evasion. Tumors may escape immune detection by co-opting STING-dependent DNA sensing. For example, HER2–AKT activation in melanoma and colorectal adenocarcinoma cells was shown to selectively abrogate the selective activation of the TBK1–IRF3 signaling downstream of STING with the potential of enabling other types of signaling downstream of STING such as NFκB (96). Furthermore, cancer cells treated with STING agonist markedly increased PD-L1 expression and proinflammatory cytokines (97, 98). Although the mechanism remains to be determined, Lim and colleagues have demonstrated that RelA/NFκB signaling stabilizes PD-L1 protein expression to coordinate evasion of T-cell immune surveillance and promote tumor growth (99), which raises a unique prospect for the STING-driven RelA/NFκB pathway in cancer cells to modulate coinhibitory checkpoint molecules for immune evasion. In addition, STING-dependent DNA sensing may play a crucial role in shaping an immune-suppressive tumor microenvironment. In human squamous cell carcinomas, STING signaling abrogated tumor immunogenicity by recruiting regulatory T cells (100). In this same vein, exposure to radiation in MC38 colon tumors altered the immune landscape by mobilizing myeloid suppressor cells in a host STING-dependent manner (101). Additional studies will be required to further delineate the stepwise mechanism by which tumor and host STING activity facilitates a potent immune-suppressive environment. Collectively, these observations highlight the importance of cGAS–STING activation in not only eliciting tumor immunogenicity, but also driving malignant behaviors, such as immune evasion. Such dichotomous roles for this DNA-sensing pathway will be further explored below.

Recent studies have revealed an important link between the cGAS–STING cascade and cellular senescence, a state of irreversible cell-cycle arrest. In response to senescence-inducing genotoxic stress, such as radiation and oncogene activation, growth-arrested cells generate cytosolic DNA substrates, which alerts the cytoplasmic DNA sensor cGAS (51, 52, 70). This, in turn, activates STING-mediated induction of SASP in an NFκB-dependent manner (refs. 51, 53, 82; Fig. 3). Components of the SASP promote inflammation and reinforce cell-cycle arrest (102, 103). Indeed, oncogenic RAS activation has been shown to induce senescence and SASP in a cGAS–STING-dependent manner (51, 52). The fact that SASP is driven by NFκB, and not by the IRF3–IFN pathway, suggests nonredundant roles of downstream STING effectors in regulating senescence (51).

Transcriptional data of a subset of human tumors revealed that poor patient survival was associated with reduced cGAS and STING expression (53, 93). In line with these findings, experimental evidence suggests that loss of cGAS or STING results in compromised senescence and SASP responses, accelerated spontaneous immortalization, and increased tumor growth (51–53, 70). It is important to note that high expression of STING has also been shown to correlate with poor prognosis in a subset of patients with colorectal cancer (104), reflecting the potential role for STING in promoting tumor growth and immune evasion. The incongruent relationship between cGAS–STING RNA levels and patient outcome across different cancers may reflect technical limitations of bulk RNA sequencing in these studies, i.e., tumor cell–intrinsiccGAS–STING expression is masked by contaminating stromal cells, or represent biological differences in which cGAS–STING exerts tumor-promoting or tumor-suppressing roles in particular tumor types.

Paradoxically, and in the absence of functional downstream cell-cycle effectors such as p53 and p21, among others, chronic stimulation of cGAS–STING signaling exerts tumorigenic effects by establishing an immune-suppressive microenvironment leading to metastasis and resistance to chemotherapeutic agents (5). Unlike acute STING-driven SASP, chronic SASP-related inflammation is correlated with malignant behaviors, such as evasion of oncogene-induced senescence and immune suppression (51, 105). Similarly, chromosomally unstable tumors chronically activate STING-dependent noncanonical NFκB signaling to drive secretion of proinflammatory cytokines and metastasis (ref. 62; Fig. 4). The molecular basis underlying the protumorigenic role of SASP remains to be fully elucidated, yet it may involve particular oncogenic triggers and alternative splicing events that change the composition of SASP to include immune-suppressive and proinflammatory cytokines (105, 106). Importantly, the release of oncogenic SASP-related factors is not limited to cancer cells. Neighboring growth-arrested fibroblasts can release soluble factors that elicit proliferation of premalignant human keratinocytes. This interesting cross-talk between normal tissue and cancer cells raises the possibility that the host STING-mediated SASP phenotypes may have non–cell-autonomous roles in driving tumorigenesis (107). Taken together, these observations suggest that the consequences of cGAS–STING inflammatory signaling are context dependent. Acute activation of STING in early neoplastic cells reinforces cell-cycle arrest through SASP. As cancer cells tolerate long-term cGAS–STING signaling and lose downstream cell-cycle regulators, the inflammatory processes adopt protumorigenic roles that enable senescence evasion and aggressive tumor growth.

It is worth noting that precise spatial control may exist for cGAS activation in cancer. Using human monocyte cell lines, Barnett and colleagues have revealed that the N-terminus of inactive cGAS anchors to phosphatidylinositol 4,5-bisphosphate PI(4,5)2 at the plasma membrane, which restricts cGAS access to self-DNA (108). Upon activation, cGAS is released into the cytosol to recognize endogenous dsDNA and initiate interferon responses. The interaction of cGAS with PI(4,5)2 raises an interesting prospect for PI3K in regulating cGAS activation, and if high PI3K/AKT signaling licenses chronic activation of cGAS–STING signaling for tumorigenesis. Further work will be required to delineate the molecular requirements that determine the fate of downstream STING signaling to either promote or suppress tumor formation.

Emerging evidence now supports a link between autophagy and the DNA-sensing adapter STING. Autophagy is a primordial function of STING that precedes STING's role in interferon signaling (109, 110). Gui and colleagues have shown that autophagy constitutes an IFN-independent pathway in response to viral infection (110). The paradoxical role of autophagy in tumor progression has been widely studied. In normal cells, autophagy modulates cytoprotective effects to maintain homeostasis by mediating functions that include clearance of damaged organelles, misfolded proteins, and oxygen radicals (111). The removal of such cytotoxic elements impairs the acquisition of genetic abnormalities and mutations that would lead to cancer formation. For instance, mouse hepatocytes with defects in ATG5 and ATG7 exhibit higher levels of oxidative stress and genomic DNA damage that favor their transformation to liver adenomas (112). However, aggressive tumors can co-opt autophagy to drive prosurvival and metastatic programs (111, 113, 114). The increased metabolic and energy demand to sustain malignancy may force cancer cells to adopt autophagy as a protective mechanism for survival. It is likely that the dual role of autophagy in cancer may be context dependent based on factors such as genotype, tumor type, and microenvironment.

Similar to autophagy, STING has been shown to adopt dichotomous roles in tumor progression (Figs. 3 and 4). It is tempting to postulate that such duality of STING functions emerged as a consequence of its role as a regulator of autophagy. In line with this notion, Nassour and colleagues demonstrated STING-driven macroautophagy, a subtype of autophagy involving autophagolysosomes, is a critical step in preventing the proliferation of cells undergoing replicative crisis (109). Attenuation of either cGAS–STING or autophagy licensed RB–p53-deficient cells to bypass telomeric crisis and tolerate acquisition of both chromosomal copy-number and structural alterations. It is worth noting that this consequence of cGAS–STING activation may depend on the genetic context, as functional p53, alternatively, triggers autophagy-independent senescent programs well before cells reach the replicative crisis stage (70, 115). The molecular mechanism underlying the STING–autophagy–cell death axis remains to be determined, but potential pathways may involve regulation of mitophagy or calcium signaling (116, 117). Beyond cellular senescence, these observations illuminate a novel role for STING-mediated autophagy as an additional barrier against early neoplastic progression in normal cells.

Alternatively, the ER stress response and autophagy can contribute to the development of advanced cancers by enabling cancer cells to survive in stressful environments (118). Disseminating cancer cells were shown to exert their dormant and immune-evasive potentials through the ER stress response, in which this imbalance of ER homeostasis may help fuel metastasis (refs. 113, 114; Fig. 4). Taken together with the protumorigenic role for cGAS–STING in chromosomally unstable tumors (62), it is possible that, in certain cellular contexts, the self-DNA sensing pathway may act in concert with autophagy–ER stress programs to support tumor progression. In line with this hypothesis, ER stress induced the upregulation of STING expression and resistance to EGFR tyrosine kinase inhibitors in non–small cell lung cancers (119). Apart from the tumor cell–intrinsic role of ER stress in driving malignancy, the tumor microenvironment may exploit such metabolic programs to facilitate immune suppression (118). The tumor milieu was proposed to restrict proper uptake of nutrients such as glucose in intratumoral T cells, which leads to ER stress and subsequent activation of the IRE1α–XBP1 unfolded protein response (UPR) pathway. Chronic activation of the UPR reprograms T cells to adopt a dysfunctional state with decreased antitumor functions. Future studies will be required to define both the molecular hierarchy between cGAS–STING and ER stress and how these pathways interact in driving tumorigenesis.

Recent work has demonstrated an important link between CIN and tumor metastasis (62). Ongoing chromosome segregation errors generate cytosolic dsDNA that is sensed by the cGAS–STING pathway (49, 62). Tumors harboring unstable genomes engage in STING-dependent noncanonical NFκB and inflammatory responses that favor invasion and metastasis (Fig. 4). In parallel, tumors with CIN also suppress antiviral type 1 IFN production, which may explain the susceptibility of IFN-defective malignant cells to viral oncolysis (120, 121). Apart from the role of CIN in driving malignant behaviors by intrinsic cGAS–STING activation in tumors, there is growing evidence that metastasis may be driven in a tumor cell nonautonomous manner. In metastatic human breast tumors, cancer cells communicate with adjacent astrocytes through cGAMP signaling (122). cGAMP generated by tumor cGAS is exported to astrocytes via gap junctions, which in turn activate astrocyte STING and initiate the release of inflammatory cytokines to promote tumor progression and metastatic cancer cell survival in the brain. This bidirectional cross-talk mediated by cGAMP transfer between tumors and normal tissues is reminiscent of mechanisms observed in antiviral defense, suggesting a potential role in which tumor-derived cGAMP may allow cancer cells to interact with their environment and exert their tumorigenic effects (123).

A crucial unanswered question is how tumors alter the downstream circuitry of STING to adopt metastatic behaviors. One possible mechanism may involve precise control of STING expression levels. In support of this hypothesis, a recent study revealed that the magnitude of STING signaling determined the induction of apoptotic programs in macrophages and T lymphocytes (124), suggesting that modulation of STING activity may select for distinct downstream effector programs. Further investigation is warranted to uncover the molecular requirements and context that dictate metastasis promoting or suppressive outcomes downstream of the cGAS–STING cascade.

The crucial role of the cGAS–STING pathway as an activator of both adaptive and innate immune responses highlights the clinical relevance of DNA-damaging therapies. A growing body of evidence has shown that therapeutic agents such as radiation, PARP inhibitors (PARPi), and etoposide help generate cytosolic DNA and invoke STING-dependent IFN production for antitumor immunity (49, 69, 125, 126). For example, radiation treatment has been linked with cGAS–STING to suppress tumor growth (48, 88). Furthermore, PARP inhibition has been shown to elicit cGAS–STING signaling in tumors and host immune cells to promote tumor clearance, proinflammatory signaling, and increased immune infiltration (125–128). It is important to note that the particular stimulus of DNA damage may elicit markedly different programs to activate STING signaling. In contrast to PARP inhibition and radiation, cytosolic DNA induced by etoposide can trigger STING-mediated secretion of IFNs in a cGAS-independent manner (69). Taken together, these observations align with the notion that tumors with inherent defects in DNA-repair pathways are permissive to cytoplasmic exposure of DNA, which in turn may induce immunogenicity in a cGAS–STING–IRF3-dependent manner (129). Such synergy between DNA damage and cGAS–STING activation raises an interesting prospect for combinatorial strategies targeting these pathways to improve immunologic clearance of tumors.

On the other hand, even in the therapeutic context, persistent stimulation of the cGAS–STING pathway can lead to resistance. STING-induced IFN activation by prolonged radiation exposure promotes an immunosuppressive environment by recruiting myeloid-derived suppressor cells to MC38 adenocarcinoma cells (101). Other immune cell types may also help diminish the therapeutic efficacy of radiation. Contrary to the classic role of DCs in IR-tumor control, these cells may concurrently adopt protumorigenic behaviors. An intricate balance between STING-dependent canonical and noncanonical NFκB pathways coexists in DCs to modulate antitumor immunity after radiation (130). Defective RelB in DCs with functional STING–IRF3–RelA signaling led to increased IFN production and regression of transplanted murine adenocarcinoma tumors. These results suggest that concurrent activation of STING-dependent noncanonical NFκB pathway mitigates optimal radiotherapy effects. As described earlier, chronic stimulation of cGAS–STING favors noncanonical NFκB to mediate metastatic behaviors within breast tumors (62). Therefore, it is possible that prolonged activation of STING enables immune cells to participate in a similar mechanism to adopt protumorigenic functions. Taken together, the role of noncanonical NFκB in mediating malignant phenotypes in both cancer and immune cells provides a rationale for this pathway as a candidate pharmacologic target. Future work will be required to refine our understanding of how the STING–noncanonical NFκB cascade may drive metastasis and cancer-drug resistance.

The ability to harness the host's immune system in targeting cancer has unleashed a paradigm shift in cancer treatment. Therapeutic antibodies that block immune-checkpoint proteins, such as CTLA4 and PD-1/PD-L1, have led to durable responses in a subset of patients with cancer (131–133). However, a substantial proportion of patients either do not receive durable clinical benefit from initial therapy with either CTLA4 or PD-1/PD-L1 inhibitors or relapse after a partial response (134–140). The limitations of immune-checkpoint therapy highlight the need to search for other modulators that may enhance tumor immunogenicity.

The role of cGAS–STING signaling in modulating antitumor responses has sparked the development of pharmacologic agonists for STING. Inspired by the discovery that 2′3′-cGAMP activates human STING to initiate robust downstream IFN signaling (9, 17–19, 141, 142), the majority of STING-activating agents are synthetic analogues of 2′3′-cGAMP, which includes chemical modifications that increase STING-induced antitumor efficacy by rendering synthetic cGAMP resistant to hydrolysis (142, 143). Direct pharmacologic activation of STING has been shown to restrict tumor growth and enhance immunogenicity in several cancer-bearing mouse models. In immune-competent mouse models of colorectal cancer and melanoma, intratumoral injection of cGAMP activates STING-driven IFN responses in DCs, which in turn presents tumor-associated antigens on major histocompatibility complexes to activate CD8⁺ T cells for antitumor killing (143–145). In a subset of tumors such as B-cell malignancies, STING may serve as a therapeutic vulnerability, as STING agonists have been shown to induce apoptosis, allowing the release of tumor antigens to further cross-prime antitumor T cells (146). Beyond T cell–driven tumor regression, STING agonists have also been shown to exert critical roles in ensuring the full execution of cell-mediated immunity. Autologous tumor rechallenge in mice previously treated with DMXAA, a mouse-selective STING agonist, impaired reformation of primary tumors (143). Most importantly, STING agonists may serve as promising adjuvants. cGAMP administration in tumor-bearing mice potentiated the therapeutic effects of immune-checkpoint inhibitors and radiotherapy (87, 88, 145). This synergy may be a consequence of STING activation–induced expression of immunosuppressive molecules, such as PD-L1, in tumors (97). These therapeutic benefits were absent in STING-deficient mice, which highlights the crucial role of host STING signaling in enforcing tumor immunogenicity in nonimmunogenic tumors.

The promising avenue of STING as a pharmacologic target for immunotherapy has accelerated the development of human STING agonists for clinical trials. Careful design of such analogues is necessary for clinical efficacy, as allelic variants of human STING are poorly responsive to 3′–5′ phosphodiester-linked cGAMP (98, 141). Two clinical phase I trials are ongoing for the intratumoral delivery of STING agonists (ADU-S100 and MK-1454) in solid human tumors and lymphomas (147, 148). Dose escalations were well tolerated, and evidence of CD8⁺ tumor infiltrations at injected tumor lesions was observed, including in patients with prior treatment with checkpoint inhibitors (148). However, preliminary results have revealed only a modest clinical response with no significant activity seen upon single-agent administration and only partial responses when STING agonists were administered concurrently with anti–PD-1 therapy in advanced tumors. These partial responses paralleled those seen upon treatment with immune checkpoint blockade alone. Thus, it remains to be determined why such limited responses were observed, but the intratumoral approach with the ADU-S100 and MK-1454 STING agonists may have restricted optimal antitumor activity at noninjected tumor lesions. In an attempt to overcome this challenge and expand the pharmacologic toolbox for immunotherapy, Ramanjulu and colleagues have recently discovered amidobenzimidazole as a novel STING agonist, in which intravenous delivery of this small molecule elicited robust tumor regression in immune-competent mice (149). The safety profile of systemic administration of STING agonists will remain an important clinical consideration, and the effect of such drugs on the patient's hemodynamic status will need to be closely monitored.

Despite the excitement in harnessing STING agonists for cancer treatment, the recent link between cGAS–STING and metastasis provides a rationale for STING inhibition in late-stage cancers (62). Although acute STING activation poses a barrier to early tumorigenesis, prolonged activation of the cGAS–STING signaling axis may inadvertently suppress innate antitumor immunity and drive aggressive behaviors. This may explain the lack of durable clinical benefit from STING agonists in late-stage human tumors. It is likely that constitutive cGAS–STING activation in chromosomally unstable tumors induces preexisting resistance to STING agonists, such that tumors have already evolved to eschew the deleterious effects of cytosolic dsDNA by the time therapy is initiated. Preexisting resistance to constitutive cGAS activity in cancer raises the promise for STING antagonists, such as the recently discovered class of covalent STING inhibitors (150), for the treatment of chromosomally unstable and metastatic cancers. Furthermore, the role of cGAS in activating STING and inhibiting homology-directed repair to drive tumorigenesis highlights the unique prospect for cGAS inhibitors as an orthogonal approach to abrogate cGAS–STING signaling in late-stage cancers (151–153).

The clinical and molecular guidelines for STING agonists and antagonists remain to be determined. Differences in CIN states between primary and metastatic lesions suggest that active CIN may be one factor that helps predict personalized treatments. Indeed, increased CIN, chronic cGAS–STING activation, and poor patient survival were correlated with metastatic and not primary tumors (62), which reinforces the exciting prospect for STING inhibitors in advanced disease. Moreover, precise modulation of STING may be required for optimal therapeutic benefit. A recent preclinical study demonstrated that persistent STING activation only mildly reduced tumor growth and abrogated the development of T cell–driven adaptive immunity (154). In a similar vein, chronic engagement of cGAS–STING led to increased carcinogen-induced tumor formation (7). These results suggest that a delicate balance must be maintained for effective antitumor responses, in which surpassing a particular threshold of STING activity may license progression of malignancy. Therefore, careful patient selection based on genomic and phenotypic markers of CIN might provide insight into which patients may benefit from STING pathway inhibition or activation.

Beyond the canonical role of cGAS–STING in antiviral immunity, recent evidence has emerged that expands the functional roles of cGAS–STING to cancer. Tumors are able to co-opt the cGAS–STING pathway to either suppress or promote malignancy by regulating a diverse array of molecular programs.

However, important questions remain unanswered in our quest to understand the molecular requirements and contexts that define how tumors regulate each pathway to drive tumorigenesis in a cGAS–STING-dependent manner. For instance, it remains unclear how aggressive tumors with a functional cGAS–STING axis suppress type 1 IFN and upregulate alternative programs, such as the canonical and noncanonical NFκB pathways, to adopt metastatic behaviors. One possibility may involve modulation of the intensity of cGAS–STING activation in determining the switch between tumor-suppressive or tumor-promoting functions (124). Yet, the underlying mechanisms that define such outcomes remain poorly understood and require further studies.

The rapid progress in resolving the molecular details of the cGAS–STING pathway has led to a better understanding of how to harness this signaling cascade in cancer therapy. The diversity of signaling pathways downstream of STING as well as the potential for their independent functions beyond the linear IFN-specific role opens up novel and exciting therapeutic opportunities. The presence of nuclear cGAS raises an important question of how the innate DNA sensor tolerates self-DNA (53), which likely involves suppression of nuclear cGAS activity through an undetermined mechanism (71). Further work will be required to define the molecular context in which cGAS and STING act in concert or independently.

A thorough understanding of the temporal and spatial controls in which cGAS–STING exerts a tumor-promoting or suppressive role will improve our ability to appropriately target this pathway and ensure adequate patient selection. In early and chromosomally stable tumors, cGAS–STING activation might represent a potent therapeutic strategy. On the other hand, inhibition of this pathway may elicit a vulnerability in late-stage aggressive cancers that have likely acquired adaptive resistance, and, as such, pose as a barrier to metastatic spread. Currently, clinical trials are harnessing STING agonists to induce tumor immunogenicity. However, hyperactivation of STING signaling may inadvertently worsen clinical outcomes if tumors have already co-opted the STING pathway to drive malignant programs and to suppress its antitumor functions. It is likely that the particular tumor stage, genotype, CIN state, basal level of cGAS–STING activation, and microenvironment will dictate therapeutic responses to STING agonists or antagonists. Therefore, a better understanding of the determinants of response and careful selection of patients will be required to identify the subset of patients who will benefit from pharmacologic therapy that either activates or inhibits this important pathway.

S.F. Bakhoum is a paid consultant for Volastra Therapeutics Inc., served as an ad hoc consultant for Sanofi, has ownership interest (including patents) in Volastra Therapeutics Inc., and is a member of the scientific advisory board and board of directors of Volastra Therapeutics Inc. No potential conflicts of interest were disclosed by the other author.

S.F. Bakhoum is supported by the Office of the Director, NIH, under award number DP5OD026395 High-Risk High-Reward Program, the Department of Defense Breast Cancer Research Breakthrough Award W81XWH-16-1-0315 (Project: BC151244), the Burroughs Wellcome Fund Career Award for Medical Scientists, the Parker Institute for Immunotherapy at MSKCC, the Josie Robertson Foundation, and the MSKCC core grant P30-CA008748.