Pharmacological modulation of nucleic acid sensors — therapeutic potential and persisting obstacles

Abstract

Nucleic acid sensors, primarily TLR and RLR family members, as well as cGAS–STING signalling, play a critical role in the preservation of cellular and organismal homeostasis. Accordingly, deregulated nucleic acid sensing contributes to the origin of a diverse range of disorders, including infectious diseases, as well as cardiovascular, autoimmune and neoplastic conditions. Accumulating evidence indicates that normalizing aberrant nucleic acid sensing can mediate robust therapeutic effects. However, targeting nucleic acid sensors with pharmacological agents, such as STING agonists, presents multiple obstacles, including drug-, target-, disease- and host-related issues. Here, we discuss preclinical and clinical data supporting the potential of this therapeutic paradigm and highlight key limitations and possible strategies to overcome them.

Introduction

Protecting the intracellular microenvironment from infectious pathogens is an ancient strategy for the preservation of cellular homeostasis. Accordingly, both prokaryotic and eukaryotic cells are equipped with mechanisms for the detection of exogenous nucleic acids, based on either nucleotide sequence (for example, prokaryotic CRISPR–Cas systems) or sequence-independent features (for example, eukaryotic pattern recognition receptors (PRRs))^1,2 (Box 1). With the acquisition of multicellularity, early eukaryotes faced the need for preservation of organismal (over cellular) homeostasis. Thus, most (if not all) mechanisms of adaptation to cellular stress, including pathogen invasion, became able to initiate and coordinate systemic responses aimed at the maintenance of organismal homeostasis³. In addition, the molecular machinery for nucleic acid sensing underwent considerable expansion and diversification, and acquired the ability to detect endogenous nucleic acids localizing in non-physiological subcellular compartments, folding with non-conventional secondary structures or bearing specific chemical alterations⁴. Thus, modern mammals encode various compartmentalized systems that respond to DNA and RNA molecules of potential pathogenic significance (that is, potentially associated with cellular or organismal dysfunction)². Such PRRs act by initiating a set of highly overlapping signal transduction pathways aimed at the restoration of cellular functions, in the context of preserved systemic homeostasis. This integrated response involves alterations of the intracellular microenvironment coupled with the secretion of bioactive factors that generally convey a danger signal as they mediate immunomodulatory effects, such as cytokines and damage-associated molecular patterns^5,6 (Fig. 1).

Fig. 1: Nucleic acid sensing at the interface between cellular and organismal homeostasis.

When sufficient amounts of nucleic acids exhibiting structural features potentially recognized as non-self are available for physical engagement with their sensors (Box 1), the latter initiate signalling pathways that generally culminate in the activation of an adaptative response aimed at the restoration of cellular homeostasis (step 1) and the emission of bioactive factors (step 2), including cytokines and damage-associated molecular patterns (DAMPs). Depending on their concentration, stability and potential for systemic distribution within the bloodstream, these molecules can elicit biological responses in neighbouring cells of different types (step 3), as well as in distant cells (step 4), ultimately favouring the preservation of microenvironmental and organismal homeostasis. CR, cytokine receptor; PRR, pattern recognition receptor.

The detection of potentially pathogenic nucleic acids in mammalian cells involves mainly (1) specific members of the Toll-like receptor (TLR) family, including the RNA sensors TLR3, TLR7 and TLR8, as well as the DNA sensor TLR9 (ref.⁷), (2) so-called RIG-I-like receptors (RLRs), a group of RNA-specific PRRs named after RIG-I (retinoic acid-inducible gene I protein; also known as DDX58)⁸, and (3) cyclic GMP–AMP synthase (cGAS), a sensor of cytosolic double-stranded DNA (dsDNA) that signals via stimulator of interferon genes (STING; also known as TMEM173)⁹. Of note, although mammalian TLRs were named after their homologue in Drosophila melanogaster (Toll, the first member of the family to be characterized)¹⁰, the latter does not act as a bona fide PRR but mediates immunological effects downstream of a proteolytic cascade driven by recognition of microbial products by circulating or transmembrane receptors¹¹ (Box 2). Accumulating evidence indicates that defects in TLR, RLR and cGAS signalling contribute to multiple human disorders beyond infectious diseases, encompassing cardiovascular, autoimmune and neoplastic conditions¹². Consistent with this notion, normalizing aberrant nucleic acid sensing or the consequences thereof (for example, deregulated cytokine secretion) mediates beneficial effects in numerous preclinical models of disease¹².

Nucleic acid sensors (NASs) have been standing out as attractive targets for the development of novel drugs for decades¹³. Nonetheless, very few pharmacological agents targeting NASs are currently available for use in humans. Existing drugs include (1) imiquimod, a synthetic TLR7 agonist licensed by the US Food and Drug Administration (FDA) for the topical treatment of genital warts, superficial basal cell carcinoma and actinic keratosis¹⁴ and (2) an oligonucleotide rich in CpG regions (acting as a TLR9 agonist) currently approved as an adjuvant to the hepatitis B virus (HBV) vaccine HEPLISAV-B¹⁵. In addition, bacillus Calmette–Guérin, an inactivated variant of Mycobacterium bovis, is used as a vaccine against tuberculosis, as well as for the treatment of non-invasive bladder carcinoma¹⁶, and Cadi-05, a preparation from Mycobacterium indicus pranii, is approved as an adjuvant for the treatment of leprosy¹⁷. However, while these agents are generally considered as mixed TLR2, TLR4 and TLR9 agonists, the largest fraction of their biological effects originates from TLR2 and TLR4 signalling¹⁶ and will therefore not be discussed further in this Review.

Here, we discuss preclinical and clinical data demonstrating the vast therapeutic potential of NAS-targeting molecules, identify the obstacles preventing the complete realization of such potential and formulate strategies to foster the clinical development of this therapeutic paradigm. The promise of targeting signalling pathways elicited by free nucleotides (for example, ATP) or their products (for example, adenosine) for therapeutic purposes was recently discussed elsewhere¹⁸ and hence is not covered in this Review. Similarly, discussion of the pharmacological modulation of signal transducers acting downstream of NASs is beyond the scope of this Review.

Box 1 Natural agonism of the nucleic acid-sensing machinery

Arguably, self from non-self discrimination is the most important function of the nucleic acid-sensing machinery, allowing the initiation of rapid responses against potentially pathogenic DNA and RNA molecules in the context of tolerance for their physiological counterparts (and hence the avoidance of autoimmune reactions). In mammals, this process is governed by three general principles: (1) availability, which is a compound function of nucleic acid initial concentration, rate of degradation by endogenous nucleases and shielding by nucleic acid-binding proteins; (2) localization, referring to the subcellular compartment where nucleic acids and their sensors are localized; and (3) structure, referring not only to secondary nucleic acid conformations but also to precise chemical modifications of the (deoxy)ribose backbone or bases. Altogether, these parameters determine the likelihood for any given nucleic acid sensor to physically encounter a sufficient amount of its agonist in a physicochemical state compatible with recognition and hence initiate robust downstream signalling. Of note, DNA in mammalian cells is mostly found in a double-stranded configuration, whereas RNA almost invariably exists in a single-stranded configuration, and both nucleic acids are associated with a large number of proteins. In the context of compatible availability and localization, the most common structural RNA features detected as non-self are (1) long double-stranded RNA (dsRNA; at least 30 base pair's (bp)), (2) dsRNA of less than 30 bp with blunt ends and a 5′ triphosphate group, (3) dsRNA with a 5′ 7-methylguanosine cap coupled to a stretch of single-stranded RNA (ssRNA) of five or more nucleotides, (4) ssRNA with a 5′ triphosphate group and secondary structures (for ssRNA molecules longer than 47 nucleotides), (5) long ssRNA with large and stable secondary structures and (6) ssRNA containing poly(U) stretches or rich in AU or GU repeats. The structural DNA features that potentially initiate nucleic acid signalling are (1) double-stranded DNA (dsDNA) of more than 40–80 bp, (2) dsDNA of at least 12 bp with adjacent single-stranded DNA containing unpaired guanines (G-rich Y-form DNA), (3) unmethylated CpG dsDNA, (4) single-stranded DNA rich in AT repeats and (5) DNA–RNA hybrids. Importantly, many of these structural features also characterize nucleic acids in their fully physiological state (for example, DNA–RNA hybrids are required for transcription, dsDNA molecules of more than 40–80 bp are common during DNA replication at the lagging strand, mitochondrial DNA is poorly methylated), reinforcing the importance of availability and localization for the discrimination of self from non-self nucleic acids⁴. In addition, a variety of eukaryotic enzymes process endogenous nucleic acids to prevent unwarranted nucleic acid sensor signalling. For instance, three-prime repair exonuclease 1 (TREX1) efficiently degrades cytosolic dsDNA and limits cyclic GMP–AMP synthase (cGAS) signalling¹⁵⁹, while dsRNA-specific adenosine deaminase (ADAR) catalyses A-to-I RNA editing, limiting retinoic acid-inducible gene I protein (RIG-I) and melanoma differentiation associated protein 5 (MDA5) activation²⁸⁴.

Box 2 History of TLRs: fly versus mammalian

Viral interference was identified as early as in 1957 (refs^285,286), and nucleic acids were mechanistically implicated only a few years later^287,288. Nonetheless, the precise molecular mechanisms underlying viral interference remained obscure until Toll from Drosophila melanogaster was attributed a key role in innate immune responses to fungal infection¹⁰, and the first human homologue of Toll was cloned²⁸⁹. The genome of D. melanogaster encodes nine different members of the Toll family, which share homology with mammalian Toll- like receptors (TLRs) and (less so) with mammalian IL-1 receptors. At odds with its mammalian counterparts, however, Toll (the best characterized member of the Toll family in flies) is not activated on binding to microbial components or endogenous damage-associated molecular patterns but rather responds to cleavage products of the circulating factor Spatzle. Importantly, Spatzle activation can be initiated by the recognition of microbial products (for example, fungal β-glucans, bacterial peptidoglycan), as well as by signals involved in embryonic patterning. Thus, while mammalian TLRs mainly act as bona fide pattern recognition receptors, Toll participates in innate immune responses downstream of the actual pattern recognition receptors, de facto resembling mammalian cytokine receptors. Moreover, fly (but not mammalian) TLRs are critical mediators of physiological embryogenesis. Toll signalling in D. melanogaster resembles TLR signalling in mammals as it involves a homologue of mammalian MYD88 that recruits a multiprotein complex at the cytosolic tail of Toll to initiate the derepression of a dimeric, nuclear factor-κB-like transcription factor (Dorsal/Dif) (Box 3). However, mammalian (but not fly) TLRs can also drive the activation of various interferon regulatory factors, including IRF3, IRF5 and IRF7, as well as mitogen-activated protein kinase (MAPK) signal transduction pathways culminating in activator protein 1 (AP-1)-dependent transcription (Box 3). In summary, it appears that TLRs in higher eukaryotes lost (at least part of) their critical functions in embryogenesis as they acquired the ability to orchestrate a large panel of transcriptional programmes involved in innate immunity^11,290.

TLR agonists

The signalling cascades elicited by nucleic acid-sensing TLRs (Box 2), primarily TLR3, TLR7, TLR8 and TLR9, play a major role in the control of infectious and neoplastic disorders. Accordingly, a large number of agonists for one or more such TLRs are commercially available, and multiple agents have recently entered clinical development (Table 1). Despite such a potential, the development of pharmacological agonists of nucleic acid-sensing TLRs faces several obstacles, as discussed later in this Review.

Table 1 Selected pharmacological nucleic acid sensor activators in clinical development

TLR3 agonists

TLR3 is an endosomal TLR primarily expressed by B cells, specific dendritic cell (DC) subsets and epithelial and endothelial cells¹⁹. Viral double-stranded RNA (dsRNA) molecules are the prototypic ligands for TLR3 (ref.²⁰) (Box 1), which relies on Toll-like receptor adaptor molecule 2 (TICAM2) and TICAM1 (also known as TRIF) for signalling²¹. Depending on the experimental setting, the Tlr3^−/− genotype has been associated with detrimental, neutral or beneficial effects on antiviral responses²². Despite these apparently contradictory findings, which mainly reflect the use of different experimental end points, and the potentially divergent effects of short-term versus long-term NAS signalling³, robust preclinical evidence suggests that TLR3 agonists can be harnessed as immunostimulants in multiple settings²³. Moreover, autosomal dominant mutations in TLR3 have recently been associated with severe influenza pneumonitis in children²⁴, further reinforcing the relevance of TLR3 for antiviral responses in humans.

Commonly used TLR3 agonists include the synthetic dsRNA analogue polyinosinic:polycytidylic acid (poly(I:C)) and two derivatives: (1) poly(I:C) stabilized with poly(l-lysine) and carboxymethylcellulose (poly-ICLC; Hiltonol) and (2) poly(I:C) containing uridylic acid in a 12:1 molar ratio in the poly(C) strand (poly(I:C₁₂U), also known as rintatolimod and marketed as Ampligen). Poly(I:C) was demonstrated to drive robust interferon responses and to favour cross-priming in both mouse and human experimental systems^25,26, largely reflecting the elevated levels of TLR3 in CD8α⁺ DCs²⁷. Abundant preclinical data demonstrate that poly(I:C) administered locally is a potent adjuvant for viral prophylaxis and therapeutic anticancer vaccination²⁸. However, poly(I:C) is unstable, and results from early clinical trials testing poly(I:C) in patients with leukaemia unveiled side effects including shock, renal failure and hypersensitivity reactions in the context of limited therapeutic efficacy²⁹. Thus, the clinical development of poly(I:C) was terminated (source http://www.clinicaltrials.gov).

Poly-ICLC

Poly-ICLC was initially synthesized to circumvent the stability issues of poly(I:C)³⁰, and entered clinical testing in the late 1970s³¹, initially for oncological indications, including brain tumours³². Since then, the number of oncological indications in which poly-ICLC has been clinically tested (most often in combination with chemotherapeutics, radiation therapy or cancer vaccines) has grown to include a variety of haematological^33,34 and solid^35,36,37 malignancies, although most efforts remained focused on brain tumours^38,39. In most of these studies, poly-ICLC was well tolerated and mediated robust immunostimulatory effects that, at least in some patients, were associated with clinical benefits³⁷. Today, official sources list more than 50 clinical trials testing poly-ICLC in combination with a variety of (immuno)therapeutics in patients with cancer (source http://www.clinicaltrials.gov) (Table 1). Of note, results from a randomized, placebo-controlled, double-blinded clinical trial (NCT02071095) suggest that poly-ICLC may also be useful for the induction of (at least transient) antiviral responses in HIV-1-positive individuals⁴⁰. However, no clinical trial is currently open to pursue this therapeutic paradigm (source http://www.clinicaltrials.gov).

Poly(I:C₁₂U)

The first studies evaluating poly(I:C₁₂U) as an anticancer agent were performed in the mid-1980s⁴¹, but the development of poly(I:C₁₂U) was soon redirected towards antiviral applications⁴², largely on the basis of preclinical data demonstrating that poly(I:C₁₂U) was able to reduce the concentration of antiretroviral agents required for HIV-1 control in vitro⁴³. Although additional data in support of the ability of poly(I:C₁₂U) to prevent the decrease in CD4⁺ levels associated with HIV-1 infection were reported^44,45, development was discontinued and no clinical trials testing this therapeutic paradigm are currently open (source http://www.clinicaltrials.gov). In addition, poly(I:C₁₂U) attracted attention as a therapeutic option for chronic fatigue syndrome⁴⁶. Promising results from early clinical studies fostered the initiation of a phase III trial based on intravenous poly(I:C₁₂U) administration (NCT00215800). In this setting, poly(I:C₁₂U) ameliorated physical symptoms but did not reduce the levels of biochemical markers of disease, and side effects were frequent⁴⁷. One trial testing intravenous poly(I:C₁₂U) administration in patients with chronic fatigue syndrome is currently open to expanded access (NCT00215813; source http://www.clinicaltrials.gov). In addition, a few clinical trials are currently testing intradermal poly(I:C₁₂U) in combination with immunotherapy in patients with breast cancer (NCT01355393, NCT03599453), ovarian cancer (NCT01312389, NCT02432378, NCT03734692), colorectal cancer (NCT03403634) and prostate cancer (NCT03899987) (source http://www.clinicaltrials.gov) (Table 1), a renewed interest fostered by promising preclinical data on the ability of poly(I:C₁₂U) to act as an adjuvant to cancer vaccines in mice⁴⁸.

Other agonists

ARNAX is a novel synthetic DNA–dsRNA hybrid molecule that also promotes robust cross-priming by DCs by acting as a TLR3 agonist⁴⁹. Use of ARNAX in combination with a cancer vaccine and a programmed cell death 1 ligand 1 (PD-L1)-targeting immune checkpoint blocker (ICB) has been proposed as a strategy to overcome resistance to agents targeting programmed cell death 1 (PD-1; also known as PDCD1), at least in mice⁴⁹. To the best of our knowledge, ARNAX has not yet entered clinical development (source http://www.clinicaltrials.gov).

TLR7 and TLR8 agonists

TLR7 and TLR8 are expressed in the endosomal compartment of DCs, monocytes, macrophages and mast cells⁵⁰. Of note, TLR7 is preferentially expressed by plasmacytoid DCs and B cells, while TLR8 is mostly found in monocytes, neutrophils, and monocyte-derived DCs⁵¹. TLR7 and TLR8 recognize single-stranded RNA (ssRNA) molecules containing GU-rich or poly(U) sequences^52,53 (Box 1), resulting in the secretion of multiple cytokines, including type I interferon, downstream of MYD88 signalling⁵⁴ (Box 3). On the basis of the key role of type I interferon in viral interference, TLR7 has long been considered (together with TLR3 and TLR9) as an ideal target for the development of novel antiviral agents⁵⁵, which culminated in the development of synthetic imidazoquinolines⁵⁶. Although most of these compounds failed to allow robust viral interference and mediate direct protective effects on non-infected cells, some imidazoquinolines demonstrated considerable efficacy as immunostimulants²³.

Box 3 Principles of TLR signalling

Toll-like receptors (TLRs) are an evolutionarily conserved family of pattern recognition receptors acting in plasmatic and endosomal compartments. The human genome encodes six TLRs expressed on the cell surface (TLR1, TLR2, TLR4, TLR5, TLR6 and TLR10) as well as four endosomal TLRs (TLR3, TLR7, TLR8 and TLR9). Three additional TLRs have been identified in the mouse and rat genomes (that is, TLR11, TLR12 and TLR13). With the exception of TLR3 and TLR10, all human TLRs can (either optionally or obligatorily) signal via a supramolecular complex containing MYD88, interleukin-1 receptor-associated kinase 1 (IRAK1), IRAK2 and IRAK4 (so-called myddosome). Conversely, TLR3 obligatorily signals via Toll-like receptor adaptor molecule 2 (TICAM2) to activate TICAM1 (also known as TRIF), which can also be optionally engaged by TLR4. Both the myddosome and TRIF recruit TNF receptor-associated factor 6 (TRAF6) to stimulate the formation and activation of a supramolecular complex containing transforming growth factor-β-activated kinase 1 (TAK1; also known as MAP3K7) and multiple TAK1-binding proteins. TAK1 not only derepresses canonical nuclear factor-κB (NF-κB) transcription via the IκB kinase (IKK) complex (see Box 4 for additional details) but also initiates multiple mitogen-activated protein kinase (MAPK) signalling pathways culminating in the activation of transcriptional programmes mediated by JUN and cyclic AMP-responsive element-binding protein 1 (CREB1). At least in some cell types, TRAF6 can also activate interferon regulatory factor 5 (IRF5)-dependent transcriptional programmes (culminating in the secretion of proinflammatory cytokines) and can drive a TANK-binding kinase 1 (TBK1)-dependent, fully post-translational mechanism culminating in the activation of caspase 8 and apoptotic cell death. At odds with the myddosome, TRIF also recruits TRAF3 to promote the assembly and activation of multiprotein complexes containing inhibitor of nuclear factor-κB kinase subunit-ε (IKBKE), TBK1 and optionally TRAF family member-associated NF-κB activator (TANK). In this context, TBK1 catalyses the activating phosphorylation of IRF3 or IRF7, which allows their dimerization, nuclear import and transactivation of IRF3- or IRF7-dependent genes (prototypic target IFNB1). Of note, all human TLRs are activated by ligand-induced homodimerization or heterodimerization, irrespective of specificity. The latter ranges from conserved bacterial components such as lipopolysaccharide (a TLR4 agonist) and flagellin (a TLR5 agonist) to atypical nucleic acids such as double-stranded RNA (a TLR3 agonist) and CpG-rich DNA (a TLR9 agonist). Finally, some TLRs also recognize endogenous proteins that act as danger signals, such as high mobility group box 1 (HMGB1, a TLR4 agonist)^82,291.

Box 4 Principles of cGAS–STING signalling

On binding to cytosolic double-stranded DNA, cyclic GMP–AMP synthase (cGAS) dimerizes and hence acquires the ability to convert ATP and GTP into cyclic G(2′,5′)pA(3′,5′)p, a second messenger also known as cyclic GMP–AMP (cGAMP). cGAMP (as well as other cyclic dinucleotides of bacterial origin) promotes the dimerization-dependent activation of stimulator of interferon genes (STING; also known as TMEM173) at the cytosolic surface of the endoplasmic reticulum and the endoplasmic reticulum–Golgi intermediate compartment, culminating in the recruitment of one or more signal transducers with kinase activity, including (1) TANK-binding kinase 1 (TBK1), (2) NF-κB-inducing kinase (NIK; also known as MAP3K1) and (3) the heterotrimeric complex IκB kinase (IKK). TBK1 drives interferon regulatory factor 3 (IRF3)-dependent transcriptional programmes (see Box 3 for additional details). NIK stimulates the catalytic activity of conserved IKKα (also known as CHUK) dimers, resulting in the IKKα-dependent phosphorylation of p105 (also known as NFKB2) bound to RELB. On phosphorylation, p105 is subjected to partial proteasomal degradation, resulting in the activation of non-canonical nuclear factor-κB (NF-κB) transcriptional programmes by p52–RELB heterodimers (prototypic target CXCL13). On IKK-dependent phosphorylation, IκB (also known as NFKBIA) is degraded by the proteasome, resulting in the derepression of heterodimers composed of RELA (also known as p65) and p50 (also known as NFKB1) and canonical NF-κB signalling (prototypic target TNF). cGAS–STING signalling is regulated by a variety of mechanisms, including (1) the elimination (by secretion or degradation) of cGAS and STING activators, (2) a variety of post-translational modifications (for example, phosphorylation, ubiquitylation, sumoylation, glutamylation) that regulate enzymatic activity, dimerization potential or degradation (by the proteasome or autophagy) and (3) transcriptional programmes initiating negative-feedback or positive-feedforward loops. Importantly, the net biological outcome of cGAS–STING signalling appears to exhibit an elevated degree of context dependency, ranging from the activation of robust immune responses against pathogens or malignant cells to the establishment of subacute inflammation fostering metastatic dissemination. Both cell-intrinsic and cell-extrinsic (microenvironmental) factors underlie such heterogeneity^9,292.

Imidazoquinolines as antivirals

Imiquimod (also known as R-837) is a prototypic imidazoquinoline that originally attracted attention for its antiviral effects in animal models of cytomegalovirus and herpes simplex virus 2 (HSV-2) infection⁵⁷, and is now approved for the topical treatment of genital warts, superficial basal cell carcinoma and actinic keratosis⁵⁸. Intriguingly, the efficacy of Aldara (5% imiquimod cream) may reflect both TLR7-dependent and TLR7-independent pathways⁵⁹, underlying the key importance of formulation and drug–excipient interactions in the clinical activity of TLR agonists (and most likely many other drugs). Resiquimod (also known as R-848) is a mixed TLR7 and TLR8 agonist that was also conceived for antiviral applications, reflecting its ability to favour IL-12 secretion and limit viral replication in monocytes isolated from HIV-1-infected individuals⁶⁰. However, despite promising efficacy results in randomized phase II clinical trials enrolling patients with HSV-2 infection⁶¹ or hepatitis C virus (HCV) infection⁶², the safety profile of resiquimod was not superior to that of recombinant interferon-α (IFNα; the historical gold standard treatment for HCV infection)⁶². Thus, the clinical development of imiquimod and resiquimod as antiviral agents has been discontinued (source http://www.clinicaltrials.gov).

Imidazoquinolines as anticancer agents

As the development of imidazoquinolines as antivirals was abandoned, efforts have been refocused on anticancer applications, at least partially due to the enthusiasm raised by the approval of imiquimod for the treatment of some skin tumours⁵⁸. The type I interferon-dependent anticancer effects of systemic imiquimod were first reported in 1992 (ref.⁶³), and since then imiquimod (both topical and systemic) has been demonstrated to mediate therapeutic effects, alone or in combination with other drugs, in a variety of preclinical tumour models, including models of mammary, lung and prostate carcinoma^64,65. The anticancer activity of imiquimod has been consistently linked to the CC-chemokine ligand 2 (CCL2)-dependent recruitment of plasmacytoid DCs to the tumour bed, culminating in abundant type I interferon secretion linked to activation of CD4⁺ T helper cells producing interferon-γ (IFNγ), tumour necrosis factor (TNF) and IL-17, and intradermal γδ T cells, at least in mice^66,67. Despite these promising findings, multiple clinical trials testing imiquimod as an off-label intervention in cancer patients, generally in combination with standard chemotherapeutic regimens, failed to demonstrate long-term efficacy^68,69, with some notable exceptions, such as the combination of imiquimod and monobenzone in patients with melanoma⁷⁰. It is tempting to speculate that such a restricted efficacy spectrum reflects the ability of some tumours to establish a robustly immunosuppressive microenvironment⁷¹. Supporting this possibility, intratumoural imiquimod exerted superior effects against thymoma cells established in Ido1^−/− mice, which lack one enzyme with major immunosuppressive effects⁷². Similarly, imiquimod has been favourably combined with an ICB targeting cytotoxic T lymphocyte-associated protein 4 (CTLA4), which reverses local immunosuppression, in a patient with nivolumab-resistant melanoma⁷³. The ability of imiquimod to mediate anticancer effects as an off-label intervention is currently being investigated in no fewer than 45 clinical trials, most of which involve topical imiquimod application (source http://www.clinicaltrials.gov) (Table 1).

In a phase I clinical trial, topically applied resiquimod demonstrated clinical activity in patients with cutaneous T cell lymphoma, a skin malignancy for which limited therapeutic options are available⁷⁴. Specifically, 75% of patients with cutaneous T cell lymphoma being treated with resiquimod experienced a reduction in the number or size of cutaneous lesions, with nearly 30% of them achieving complete eradication of the lesions⁷⁴. Although data on long-term disease outcome are still missing, these findings point to the possibility that resiquimod may mediate stand-alone therapeutic effects in patients with cutaneous T cell lymphoma. That said, the current development of resiquimod as an anticancer agent is mostly focused on its ability to act as an immunological adjuvant. Indeed, accumulating evidence indicates that topically applied resiquimod can be conveniently combined with peptide-based vaccines, such as recombinant cancer/testis antigen 1B (CTAG1B; also known as NY-ESO-1), to reactivate clinically relevant CD8⁺ cytotoxic T lymphocyte (CTL) responses against melanoma⁷⁵ or can be used to reverse immunosuppression in models of cervical carcinoma linked to human papilloma virus 16 infection⁷⁶. Nonetheless, only two clinical trials testing resiquimod as an immunological adjuvant for cancer therapy are currently active (NCT00960752, NCT02126579) (Table 1).

Recently, motolimod (also known as VTX-2337), another imidazoquinoline specific for TLR8 (ref.⁷⁷), has attracted some attention. In preclinical studies, motolimod administration to unfractionated white blood cells improved the ability of natural killer cells (NK cells) to mediate antibody-dependent cellular cytotoxicity (ADCC) towards cancer cell lines and freshly isolated head and neck cancer cells⁷⁸. Moreover, motolimod promoted IL-12, TNF and IFNγ secretion by DCs, thus favouring the activation of tumour-specific CTL responses⁷⁸. On the basis of these findings, subcutaneously administered motolimod is currently being tested in combination with ICB specific for PD-1 and/or cetuximab in patients with head and neck cancer (NCT02124850, NCT03906526), as well as together with durvalumab plus chemotherapy in women with platinum-resistant ovarian cancer (NCT02431559) (source http://www.clinicaltrials.gov) (Table 1).

Other agonists

Results from the first-in-concept human trial testing the safety, tolerability and immunogenicity of a novel mixed TLR7, TLR8 and RIG-I agonist (CV8102) administered alone or with fractional doses of a licensed rabies vaccine as a model antigen (NCT02238756) have recently been released⁷⁹. This study identified the safe and efficient dose range for CV8102 (ref.⁷⁹), initiating further clinical testing in oncological indications. CV8102 is currently being tested in combination with a therapeutic vaccine in patients with hepatocellular carcinoma (NCT03203005), and together with a PD-1 blocker in individuals with various types of cancer, including melanoma (source http://www.clinicaltrials.gov) (Table 1).

TLR7 agonists have also been investigated for their ability to limit asthmatic and allergic reactions. The rationale of this approach resides in the fact that allergic responses involve the polarization of CD4⁺ T cells towards a type 2 T helper cell (T_H2) cell profile, which (at least theoretically) can be counteracted by agents promoting T_H1 cell polarization (including TLR7 agonists)⁸⁰. In line with this model, the TLR7 agonist S-28463 reduced airway resistance, leukocyte infiltration and IgE levels in mouse models of allergic sensitization to ovalbumin⁸⁰. Preliminary clinical data confirmed that the intranasal administration of the TLR7 ligand GSK2245035 is well tolerated and reduces reactivity in patients with allergic rhinitis⁸¹. However, after the completion of a series of clinical trials with largely unreported results (NCT01480271, NCT01607372, NCT01788813, NCT02446613, NCT02833974), the sponsor cancelled the last ongoing study (NCT03707678) (source http://www.clinicaltrials.gov). Additional preclinical investigation is required to further understand the potential of TLR7 agonists for asthma treatment.

TLR9 agonists

TLR9 is an endosomal TLR abundantly expressed by a variety of immune cells, including monocytes, plasmacytoid DCs, B cells and neutrophils as well as CD4⁺, CD8⁺ and γδ T cells (in both humans and mice)⁸². Moreover, TLR9 can be found in the pulmonary and intestinal epithelium as well as in keratinocytes⁸². The prototypical ligands for TLR9 are unmethylated CpG-rich oligodeoxynucleotides (ODNs), which are common in prokaryotic and viral genomes (but not in their mammalian counterparts, as CpG dinucleotides are often methylated therein)⁸³ (Box 1). Immune cells from Tlr9^−/− mice are hyporesponsive to CpG-rich ODNs as well as to cytokine stimulation, resulting in increased resistance to  otherwise lethal CpG-rich ODN challenges⁸⁴, but exacerbated sensitivity to pathogen infection^85,86. Consistent with this, prophylactic treatment of mice with synthetic TLR9 ligands conveys transient protection against a wide range of viral, bacterial and parasitic pathogens, including Ebola virus, Bacillus anthracis, Listeria monocytogenes and Francisella tularensis⁸⁷. Moreover, the supernatant of human peripheral blood mononuclear cellss exposed to CpG-rich ODNs allows robust viral interference against HCV, largely as a consequence of type I interferon signalling and downstream expression of 2′-5′-oligoadenylate synthetase 1 (OAS1; which activates cellular RNase L to cleave viral genomes)⁵⁵. Altogether, these findings pointed to TLR9 as a potential target for the development of potent antiviral agents acting directly (via viral interference) as well as indirectly (on immunostimulation).

CpG-rich ODNs and IMOs as antivirals

Initial efforts focused on immunomodulatory oligonucleotides (IMOs), synthetic TLR9 agonists that incorporate cytosine or guanine analogues. As compared with CpG-rich ODNs, IMOs display increased stability, species-independent activity and a clear structure–activity relationship⁸⁸. In line with this notion, IMOs were shown to promote IL-12 and IFNγ production in mouse, monkey and human cellular systems, as well as expression of activation markers such as CD86 and CD69 by plasmacytoid DCs and B cells, respectively⁸⁸. Moreover, administration of IMOs to wild-type mice induced activation of Janus kinase–signal transducer and activator of transcription (JAK–STAT) signalling downstream of TLR9, type I interferon signalling and IFNγ signalling, as demonstrated in mice lacking TLR9 and mice lacking type I interferon receptors or IFNγ⁸⁹. Such activity is relevant for chronic HCV infection, a setting in which patients often become resistant to treatment with recombinant IFNα⁹⁰. Indeed, the liver of mice pretreated with IFNα remained responsive to transactivation of interferon-stimulated genes (ISGs) by IMOs, and the ISG-activatory activity of IMOs compared favourably with that of IFNα, correlating with increased hepatic infiltration by immune cells⁸⁹. On the basis of these findings, one specific IMO (tilsotolimod, also known as IMO-2125) has been evaluated in clinical trials for toxicity in HCV-infected patients (NCT00728936, NCT00990938). Preliminary results from these studies suggest that the use of tilsotolimod as a stand-alone intervention is associated with a limited incidence of serious adverse events, which is greatly exacerbated in the presence of ribavirin (source http://www.clinicaltrials.gov). To the best of our knowledge, the clinical development of IMOs as antiviral agents has now been discontinued.

TLR9 agonists have also attracted attention as potential therapeutics for chronic HIV infection. Initial work with the mouse Friend retrovirus model demonstrated that CpG-rich ODNs are an effective post-exposure treatment, which is linked to considerable reductions in circulating and splenic viral load in the context of T_H1 cell cytokine secretion and activation of virus-specific CTL responses⁹¹. Surprisingly, prophylactic administration of CpG-rich ODNs not only failed to prevent infection but aggravated viral load in this model⁹¹. Similar observations were made in rhesus monkeys treated intravaginally with CpG-rich ODNs and then exposed to simian immunodeficiency virus (SIV) via the same route⁹². Along similar lines, TLR9 activation in mouse splenic cells engineered to express the HIV-1 genome appears to favour (rather than inhibit) viral replication⁹³, and a similar activation reportedly occurs in human latently infected mast cells⁹⁴. Conversely, CpG-rich ODNs suppress the replication of both CC-chemokine receptor 5 (CCR5)-tropic and CXC-chemokine receptor 4 (CXCR4)-tropic HIV-1 in human lymphoid tissue ex vivo⁹⁵. Although the precise molecular mechanisms explaining these apparently paradoxical observations remain to be elucidated, it is tempting to speculate that cellular context plays a major role in this setting. Results from early clinical studies investigating lefitolimod (a CpG-rich ODN with a covalently closed dumbbell shape, also known as MGN1703) in HIV-1-positive patients indicate that this approach is safe and induces robust virus-specific humoral and cellular immunity⁹⁶, as well as signatures of type I interferon responses in the colonic epithelium⁹⁷. However, long-term clinical benefits appear to be limited, as only one of 12 participants achieved prolonged control of viraemia⁹⁶. Only one clinical study testing lefitolimod as an antiviral agent (in combination with HIV-1-specific antibodies) is currently open according to official sources (NCT03837756). Moreover, lefitolimod is being investigated in combination with chemotherapy or immunotherapy in cancer patients (NCT02077868, NCT02668770) (Table 1).

CpG-rich ODNs and IMOs as vaccine adjuvants

The development of TLR9 agonists as immunological adjuvants to prophylactic vaccines against viral infection culminated with the approval of one CpG-rich ODN, namely CpG-1018 (also known as 1018 ISS), for use in combination with recombinant hepatitis B surface antigen (HBsAg) as an HBV vaccine for humans in November 2017 (ref.¹⁵). Agatolimod (also known as CpG-7909 or PF-3512676) is another synthetic CpG-rich ODN tested as an adjuvant for prophylactic HBV vaccination. Despite promising preclinical and clinical findings, however, agatolimod failed clinical development for this indication⁹⁸, for unclear reasons. In this setting, improved CpG-rich ODN formulations, including ODNs wrapped with non-agonistic ligands for DC receptors such as C-type lectin domain containing 7A (CLEC7A), have been shown to mediate superior immunostimulatory effects⁹⁹. Recombinant HBsAg, hepatitis B core antigen (HBcAg) and CpG-rich ODNs have also been investigated as therapeutic vaccines against HBV, at least in preclinical models¹⁰⁰. In this setting, vaccination promoted robust humoral responses against both HBsAg and HBcAg as well as potent CD8⁺ and CD4⁺ T cell responses that limited the circulating manifestation of infection in diseased mice, with no evident liver toxicity¹⁰⁰. However, neither of these latter approaches has yet entered clinical development to our knowledge (source http://www.clinicaltrials.gov).

CpG ODNs and IMOs as anticancer agents

TLR9 is often expressed in the tumour microenvironment, not only by immune cells but also by (at least some) malignant cells and the tumour endothelium. Early efforts to use TLR9 agonists for cancer therapy were therefore focused on the activation of regulated cell death in malignant cells or the inhibition of neoangiogenesis^101,102. However, it soon became clear that the greatest promise of TLR9 agonists for cancer therapy resided in their ability to promote type I interferon secretion, at least theoretically, setting the stage for the repolarization of the tumour microenvironment from an immunosuppressive state characterized by high levels of myeloid-derived suppressor cells and M2-like tumour-associated macrophages to an immunostimulatory one characterized by elevated levels of mature DCs and M1-like tumour-associated macrophages¹⁰³. Abundant preclinical data suggest that TLR9 agonists are generally insufficient to mediate robust anticancer effects but can be conveniently combined with a variety of immunotherapeutics, especially when administered intratumourally^104,105,106.

Results from clinical trials testing CpG-rich ODNs or IMOs in patients with haematological^107,108 and solid tumours^109,110,111, generally in combination with standard-of-care therapy, demonstrated that this therapeutic approach is associated with acceptable safety and tolerability, as well as preliminary evidence for activity. However, subsequent randomized studies generally restricted the initial enthusiasm, often owing to modest efficacy¹¹² and accrued toxicity^113,114. In particular, agatolimod failed to increase progression-free and overall survival in a phase III study on patients with lung cancer receiving standard-of-care chemotherapy^113,114, despite promising results in a similar phase II setting¹¹⁵. IMOs were often administered subcutaneously in these studies, which may explain the relative lack of efficacy. Further supporting this possibility, CpG-rich ODNs delivered locally to patients with early-stage melanoma conferred durable disease control in two recent randomized phase II studies¹¹⁶. Of note, CpG-rich ODNs and IMOs often drive CTL responses characterized by elevated PD-1 expression¹¹⁷, suggesting that combinatorial regimens involving ICBs could be particularly efficient in this setting. Data in support of this possibility are emerging¹¹⁸. In particular, intratumoural administration of SD-101, a synthetic CpG-rich ODN, in the context of pembrolizumab therapy resulted in improved immune activation in patients with melanoma¹¹⁸. Similarly, SD-101 could be favourably combined with low-dose radiation therapy in patients with low-grade B cell lymphoma, resulting in a high rate of objective responses at non-irradiated tumour sites¹¹⁹, an immunological phenomenon commonly known as abscopal response¹²⁰.

Several CpG-rich ODNs and IMOs are currently being tested in clinical trials for their ability to enhance the efficacy of chemotherapy, radiation therapy or ICB-based immunotherapy in patients with a variety of tumours. These oncological indications include (but are not limited to) head and neck cancer (NCT02521870), colorectal cancer (NCT02077868), melanoma (NCT02521870, NCT02644967, NCT03445533), lymphoma (NCT02927964, NCT03322384, NCT03410901), breast carcinoma (NCT01042379) and mixed solid tumours (NCT03007732, NCT03052205, NCT03831295) (Table 1). Of note, most of these trials involve SD-101, while the clinical development of agatolimod as an anticancer agent has been discontinued (source http://www.clinicaltrials.gov).

Other applications

Similarly to TLR7 agonists, TLR9 agonists have also attracted attention as potential agents for the management of allergic reactions¹²¹. Supporting the rationale of using TLR9 agonists to limit T_H2 cell immune responses by boosting their T_H1 cell counterparts, a CpG-rich ODN efficiently suppressed IL-33-driven airway hyperreactivity downstream of type I interferon-dependent activation of NK cells and consequent secretion of IFNγ (a potent T_H1 cell polarizer) in mice¹²². However, a few studies testing this therapeutic paradigm in patients with moderate to severe allergic asthma (NCT01673672, NCT02087644) were prematurely withdrawn or terminated as they did not meet the primary end point (source http://www.clinicaltrials.gov).

RLR agonists

RLRs are a cytosolic family of DexD/H-box RNA helicases that respond to viral RNA molecules (Box 1) by activating type I interferon secretion downstream of mitochondrial antiviral signalling protein (MAVS) and TANK-binding kinase 1 (TBK1) activation^8,123 (Box 5). RIG-I is arguably the best characterized member of the RLR family, and was first described only 15 years ago¹²⁴. Thus, mammalian RLR research is a considerably younger field than mammalian TLR research. Nonetheless, RLRs are promising targets for the development of antiviral and anticancer agents.

Box 5 Principles of RLR signalling

Retinoic acid-inducible gene I protein (RIG-I; also known as DDX58) and melanoma differentiation associated protein 5 (MDA5; also known as IFIH1) are the prototypic RIG-I-like receptors (RLRs) encoded by the mammalian genome. Both RIG-I and MDA5 detect abnormal RNA molecules in the cytosol. However, while RIG-I preferentially binds short double-stranded RNA (dsRNA) molecules (less than 300 base pairs (bp)) with blunt ends and a 5′ triphosphate group, MDA5 predominantly detects long RNA molecules (more than 1,000 bp) with no end specificity. On RNA binding via a helicase domain, RIG-I undergoes extensive conformational rearrangement that allows ATP binding, exposure of caspase recruitment domains (CARDs) and oligomerization. Conversely, MDA5 CARDs are constitutively exposed and oligomerize as multiple MDA5 molecules assemble to form a filament on dsRNA. The oligomerization of RIG-I and MDA5 is favoured by (but does not strictly rely on) unanchored lysine 63-linked polyubiquitin chains. DExH-box helicase 58 (DHX58; another RLR, also known as LGP2) can also recognize cytosolic dsRNA and hence modulate RIG-I and MDA5 signalling, but fails to act as an independent RLR because it lacks CARDs. Indeed, exposed CARDs allow RIG-I and MDA5 oligomers to bind mitochondrial antiviral signalling protein (MAVS) on the outer surface of mitochondria or peroxisomes, resulting in the nucleation of MAVS helical filaments that initiate IκB kinase (IKK) and TANK-binding kinase 1 (TBK1) signalling (see Box 3 for additional details). Intriguingly, once initiated, MAVS polymerization self-propagates via a bona fide prion-like mechanism. Moreover, the biological output of MAVS signalling appears to be dependent on subcellular localization. Thus, whereas peroxisomal MAVS is responsible for rapid type I interferon-independent responses to cytosolic RNA, potentially mediated by interferon regulatory factor 1 (IRF1)–IRF3 heterodimers, its mitochondrial counterpart underlies delayed, type I interferon-dependent responses initiated by IRF3 homodimers. Whether RIG-I and MDA5 exhibit some degree of specificity for peroxisomal versus mitochondrial MAVS remains to be elucidated. Active RIG-I (but not MDA5) can also bind PYD and CARD domain-containing protein (PYCARD; also known as ASC) and caspase 1 to form a supramolecular platform for the proteolytic maturation of IL-1β and IL-18. This is particularly important as, in the absence of additional stimuli, the RLR-initiated nuclear factor-κB-dependent transactivation of IL1B and IL18 would generate only biologically inactive precursors^8,123.

RIG-I agonists

Antiviral agents

The vast majority of currently available RLR agonists targeting RIG-I were conceived for antiviral applications¹²⁵, largely reflecting the key role of RLRs in natural antiviral defences. SB 9200 (also known as inarigivir soproxil or GS-9992) is an orally available prodrug of a dinucleotide agonist of RIG-I and nucleotide-binding oligomerization domain-containing protein 2 (NOD2), a cytosolic PRR involved in the elimination of invading pathogens^126,127. SB 9200 not only favours RIG-I and NOD2 signalling, culminating with type I and type III interferon secretion¹²⁶ and IL-1β release downstream of inflammasome activation¹²⁸, respectively, but also locks RIG-I and NOD2 on pregenomic viral RNA, impeding the binding of the viral polymerase and synthesis of the DNA minus strand¹²⁶. Consistent with this dual activity, SB 9200 mediated robust antiviral effects in woodchucks chronically infected with woodchuck hepatitis virus, a model of chronic HBV infection, both as a stand-alone agent and combined with entecavir^126,129. In this setting, hepatic toxicity was limited, and the therapeutic efficacy of SB 9200 was comparable to that of antivirals currently used for the management of HBV infection (for example, tenofovir)^126,129 SB 9200 has also been reported to be efficacious in models of chronic HCV infection¹³⁰. Two clinical trials are currently testing the activity of SB 9200 (alone or combined with tenofovir) in patients with chronic HBV infection (NCT02751996, NCT03434353) (Table 1). To the best of our knowledge, the results from two additional trials that official sources list as completed have not yet been disseminated (source http://www.clinicaltrials.gov).

5′-triphosphate RNA (5′-pppRNA) is a natural agonist of RIG-I (ref.¹³¹) (Box 1). Human lung carcinoma A549 cells receiving 5′-pppRNA in vitro reportedly acquire resistance to infection not only by RNA viruses such as dengue virus and chikungunya virus^132,133 but also by DNA viruses such as vesicular stomatitis virus and vaccinia virus¹³³, reflecting the establishment of potent, type I interferon-dependent viral interference. Unfortunately, 5′-pppRNA is unstable and unable to cross the plasma membrane, and hence is not a good candidate for clinical development per se. In the attempt to resolve this issue, short stem–loop RNA (SLR) molecules that present a single duplex terminus and a triphosphorylated 5′ end (and hence retain strong RIG-I-binding capacity) have been developed¹³⁴. SLRs activate RIG-I more potently than longer and unstructured 5′-pppRNA, both in vitro and in vivo, when delivered with a transfection reagent¹³⁴. According to official sources, SLRs have not yet entered clinical development (source http://www.clinicaltrials.gov).

Anticancer agents

RIG-I has also attracted attention as a potential target for the development of novel anticancer agents¹³⁵. One particularly interesting approach in this setting is the use of 5′-triphosphate small interfering RNAs, which simultaneously downregulate any protein of choice and drive RIG-I activation¹³⁶. Recent data indicate that 5′-triphosphate small interfering RNAs specific for ATP-binding cassette subfamily B member 1 (ABCB1; also known as MDR1) mediate cytotoxic effects in chemoresistant leukaemia cells as they (1) increase sensitivity to doxorubicin and (2) drive potent type I interferon responses¹³⁶. That said, only a few RIG-I agonists have entered clinical development for oncological indications, namely RTG100 (NCT03065023) and MK-4621 (NCT03739138) (Table 1). However, NCT03065023 was prematurely terminated, reflecting the sponsor’s decision to establish a new treatment protocol before continuation, and NCT03739138 has been recruiting for only a few months (source http://www.clinicaltrials.gov). Some cytotoxic chemotherapeutics favour the release of nuclear non-coding RNA molecules in the cytosol, de facto triggering therapeutically relevant RIG-I activation¹³⁷. These latter findings suggest that commonly used therapies may activate NASs via poorly defined mechanisms that require urgent elucidation.

Other RLR agonists

Nucleic acid band 2 (NAB2) is a dsRNA molecule isolated from yeast that acts as a mixed TLR3 and melanoma differentiation associated protein 5 (MDA5) agonist¹³⁸. Liposomally delivered NAB2 was superior to poly(I:C) in promoting type I interferon secretion by mouse bone marrow DCs, resulting in improved cross-priming of antigen-specific CTLs¹³⁹. Similarly, liposomal delivery of NAB2 also stimulated higher type I interferon secretion in vivo as compared with poly(I:C)¹³⁹, correlating with the ability to promote secretion of a variety of proinflammatory cytokines, including type I interferon and IL-12, by human monocyte-derived DCs^138,139. Moreover, NAB2 complexed with a cationic agent effectively acted as an adjuvant to a prophylactic cancer vaccine based on vaccinia virus Ankara encoding a tumour-associated antigen¹³⁸. BO-112 is a nanoplexed formulation of poly(I:C) complexed with polyethylenimine that on intratumoural administration to mouse melanomas drives a particularly immunogenic variant of cell death downstream of MDA5 signalling^140,141, resulting in the activation of systemic immunity against non-treated, distant lesions¹⁴². One clinical trial has just been started to test the activity of BO-112 in adults with aggressive solid tumours (NCT02828098; source http://www.clinicaltrials.gov) (Table 1).

Despite there being only two MDA5 agonists under development, MDA5 stands out as a promising target for the creation of novel drugs with antiviral and anticancer applications. Indeed, a homozygous missense mutation in human IFIH1 (encoding MDA5) increased susceptibility to infection by common viruses, including seasonal rhinoviruses¹⁴³. Moreover, MDA5 stands at the core of a signalling pathway that is particularly relevant for cancer therapy¹⁴⁴. Thus, the inappropriate activation of MDA5 and protein kinase R (PKR; another RNA sensor, also known as EIF2AK2)¹⁴⁵ by endogenous RNA molecules is normally prevented by A-to-I editing, the major catalytic function of adenosine deaminase RNA specific (ADAR)¹⁴⁶. Notably, ADAR deficiency has been associated with decreased cancer cell proliferation, increased release of type I interferon and sensitization to ICB-based immunotherapy in mouse models of ICB-resistant cancers¹⁴⁴, and some subsets of cancer cells exhibit bona fide non-oncogene addition to ADAR¹⁴⁷. These results indicate that both ADAR inhibitors and MDA5 agonists may constitute optimal combinatorial partners for ICB-based immunotherapy for the management of ICB-resistant tumours. MDA5 also senses highly unstable native mitochondrial dsRNA molecules, which are generally degraded by SUV3-like RNA helicase (SUPV3L1) and polyribonucleotide nucleotidyltransferase 1 (PNPT1)¹⁴⁸. Thus, mitochondrial dsRNA molecules stand out as interesting templates for the development of novel MDA5 agonists.

In 2017, a high-throughput screening based on ISG bioluminescent reporters identified the benzobisthiazole compound KIN1000 as a potent RLR inducer, driving the design and synthesis of hundreds of analogues that were tested for potency on the basis of interferon regulatory factor 3 (IRF3) nuclear translocation¹⁴⁹. This approach led to the selection of KIN1148, which was further developed as an immunological adjuvant. Consistent with its ability to drive type I interferon responses, KIN1148 allowed prophylactic vaccination of mice against a species-adapted variant of a pandemic human influenza virus¹⁴⁹. The actual RLR underlying the adjuvant-like activity of KIN1148, however, remains to be elucidated. Consistent with this lack of mechanistic knowledge, KIN1148 has not yet entered clinical development (source http://www.clinicaltrials.gov).

cGAS–STING agonists

cGAS was first characterized in 2013 as a sensor for cytosolic dsDNA (Box 1) that signals via STING^150,151 (Box 5). In the same year, multiple mechanistic details of cGAS–STING signalling were revealed^152,153,154, and the cGAS–STING pathway was also attributed key roles in antiviral defences^155,156. One year later, type I interferon downstream of STING signalling was shown to underlie the ability of DCs to optimally cross-prime tumour-specific CTLs^157,158, identifying the cGAS–STING pathway as a potential target for both antiviral and anticancer interventions. Consistent with this notion, Tmem173^−/− mice are unable to mount therapeutic responses to ICB-based immunotherapy¹⁵⁷, and STING-deficient cancer cells implanted in immunocompetent animals are impaired in their ability to elicit abscopal responses on irradiation¹⁵⁹. Considerable efforts are currently being dedicated to the development of STING (rather than cGAS) agonists (see later), at least in part reflecting the notion that cGAS naturally recognizes dsDNA molecules that require a delivery platform to penetrate the cell membrane (Box 1), while natural STING agonists are small cyclic dinucleotides (CDNs) that can be easily mimicked by small molecules¹⁵¹. Importantly, accumulating evidence suggests that STING agonists lose efficacy when administered locally at high doses, especially in oncological settings¹⁶⁰, at least in part reflecting the cytotoxic effects of robust STING activation in T lymphocytes^161,162. This calls for the development of sustainable strategies for the delivery of STING agonists, potentially based on the systemic administration of prodrugs¹⁶³.

DMXAA

5,6-Dimethylxanthone-4-acetic acid (DMXAA; also known as vadimezan or ASA404) is a potent STING agonist¹⁶⁴ originally developed as a vasculature disrupting agent¹⁶⁵. Consistent with its ability to promote the secretion of type I interferon and TNF, which was recognized long before the identification of STING¹⁶⁶, DMXAA has been shown to mediate robust antiviral activity in models of HBV and HSV infection^167,168 and stand-alone therapeutic efficacy in preclinical models of acute myeloid leukaemia and mammary carcinoma^169,170. Results from multiple phase I and phase II clinical trials testing DXMAA in patients with a variety of tumours in the 2000s demonstrated that DXMAA can be safely administered in biologically active doses in most settings, including combinatorial regimens with cytotoxic chemotherapy^171,172. However, none of these and subsequent clinical studies, including a randomized phase III trial testing DXMAA in combination with standard-of-care chemotherapy in patients with advanced non-small-cell lung cancer¹⁷³, could demonstrate robust therapeutic benefits, and the clinical development of DXMAA was discontinued (source http://www.clinicaltrials.gov). The reasons underlying the dismal efficacy of DXMAA in patients with cancer remain to be elucidated. However, structural studies demonstrated that DXMAA efficiently binds mouse, but not human, STING^174,175. Thus, the lack of STING-dependent immunostimulation may be one reason underlying the clinical failure of this agent.

Cyclic dinucleotides

Cyclic GMP–AMP (cGAMP) and other CDNs of bacterial origins are natural STING agonists^151,176. Of note, some of these CDNs, such as cyclic diguanylate monophosphate have been successfully used in preclinical mouse tumour models to mediate stand-alone therapeutic activity¹⁷⁷ or combinatorial therapeutic activity¹⁷⁸. Similarly, cGAMP and cyclic diadenylate monophosphate reactivated latently infected cells and enhanced SIV-specific responses ex vivo in cynomolgus macaques with naturally controlled SIV infection¹⁷⁹. However, natural CDNs, including cGAMP, are highly sensitive to hydrolysis by ectonucleotide pyrophosphatase/phosphodiesterase 1 (ENPP1)¹⁸⁰, resulting in suboptimal therapeutic activity. Moreover, human (but not mouse) STING exists in different polymorphic variants (that is, wild type, REF, HAQ, R232H, AQ, R293Q, R71H, G230A, and HQ) with differential sensitivity to canonical CDNs¹⁸¹, as well as in multiple splicing isoforms with altered stability¹⁸².

Second-generation STING agonists

The considerations noted earlier drove the development of second-generation STING agonists with increased potency, largely based on rational design, including (1) 2′-3′-cG^SA^SMP, a hydrolysis-resistant biphosphothioate analogue of cGAMP¹⁸⁰ and (2) ADU-S100 (also known as MIW815 or ML RR-S2 CDA), a hydrolysis-resistant molecule that efficiently activates mouse as well as human STING (in all its isoforms)¹⁸³. Intratumoural ADU-S100 administration mediated robust therapeutic effects in mouse models of melanoma, mammary carcinoma and colorectal cancer, coupled with the establishment of long-term immunological memory¹⁸³. ADU-S100 has also been successfully combined with a variety of immunotherapeutic agents in preclinical tumour models, including ICBs and recombinant cytokines¹⁸⁴. Along similar lines, a variant of ADU-S100 lacking the mixed linkage (an internal stabilizing bond) was successfully used to boost the efficacy of a cancer cell vaccine producing colony stimulating factor 2 (CSF2; also known as GM-CSF) against experimental mouse melanomas, which was further boosted by PD-1 neutralization¹⁸⁵. These findings provided sufficient rationale to initiate the clinical development of ADU-S100, which is currently being tested in combination with CTLA4- or PD-1-blocking antibodies in patients with advanced lymphoma or solid tumours (NCT02675439, NCT03172936, NCT03937141) (Table 1). Importantly, accumulating evidence indicates that the intratumoural administration of STING agonists can mediate cytostatic or cytotoxic effects on CD8⁺ CTLs^161,162 and B cells¹⁸⁶. As these findings reduced (at least to some degree) initial expectations for STING agonists as anticancer agents, results from the aforementioned clinical trials are urgently awaited.

Amidobenzimidazoles

The lack of STING agonists that are compatible with systemic delivery prompted the initiation of a high-throughput screening aimed at the identification of cGAMP competitors for STING binding¹⁸⁷. In this setting, two amidobenzimidazole compounds were selected and used for crystallographic studies, which revealed that each STING monomer binds one amidobenzimidazole molecule at (or very near to) the cGAMP-binding pocket. On the basis of these findings, a dimeric amidobenzimidazole molecule was developed and optimized for increased stability, leading to the generation of compound 3 (ref.¹⁸⁷). Importantly, subcutaneous administration of compound 3 caused production of type I interferon, TNF and IL-6 in wild-type (but not Tmem173^−/−) mice, and intravenous delivery mediated robust anticancer effects in a mouse model of colorectal cancer, which were abolished by the depletion of CD8⁺ CTLs¹⁸⁷. These findings identify a promising lead for the development of non-nucleotide STING agonists amenable to intravenous administration.

Antagonizing NASs

Dysregulated cytokine signalling plays a major role in several autoimmune and inflammatory disorders, including (but not limited to) systemic lupus erythematosus (SLE), rheumatoid arthritis, multiple sclerosis, autoimmune encephalitis and psoriasis. Consistent with this notion, these conditions are generally managed with broad-spectrum anti-inflammatory agents such as glucocorticoids or with drugs that specifically target pathogenic cytokines, such as the TNF blocker etanercept or the IL-17A blocker secukinumab¹⁸⁸. Accumulating preclinical data (discussed later) indicate that the overproduction of some cytokines involved in autoimmune and inflammatory disorders originates from deregulated nucleic acid sensing, suggesting that antagonists of NASs may be valuable therapeutic tools for these indications (Table 2). In addition, deregulated cytokine signalling downstream of NASs is involved in the origin of diseases previously thought to have no immunological causative component, such as some neurodegenerative and cardiovascular conditions¹⁸⁹, further expanding the therapeutic potential of NAS antagonists.

Table 2 Potential indications for antagonists of nucleic acid sensors

TLR antagonists

Autoimmune disorders

The involvement of TLR3, TLR7, TLR8 and TLR9 in autoimmune and inflammatory conditions was first hypothesized on the basis of genome-wide association studies¹⁹⁰. Consistent with this notion, MRL/lpr mice (an experimental model of SLE) exhibit nephritic kidney degeneration coupled with extensive infiltration by TLR3⁺ cells, and it was suggested that disease progression can be accelerated by IL-6, IL-12 and type I interferon production downstream of systemic poly(I:C) administration¹⁹¹. However, these findings could not be reproduced by independent investigators¹⁹². Moreover, when experimental SLE is established in a Tlr3^−/− genetic background, the course of the disease remains unchanged¹⁹³. Along similar lines, the Tlr8^−/− and Tlr9^−/− genotypes both fail to ameliorate (and rather aggravate) disease progression in MRL/lpr mice^193,194,195. This contradicts the suggestion that TLR8 underlies deregulated type I interferon secretion in experimental models of SLE¹⁹⁶ and the finding that TLR9 is required for B cells to produce autoantibodies involved in the pathogenesis of SLE, at least in mice^193,195,197. Moreover, the levels of TLR9-expressing B cells, plasma cells and monocytes are increased in patients with SLE, correlating with the overproduction of pathogenic autoantibodies^198,199. Although the precise reasons underlying these apparently controversial observations remain to be elucidated, it is tempting to speculate that the elevated redundancy of TLR signalling (Box 3) may play a considerable role. In support of this hypothesis, a chemical inhibitor of the common TLR effector TBK1 (that is, compound II)²⁰⁰, as well as a mutation in unc-93 homologue B1 (UNC93B1), a TLR signalling regulator that compromises endosomal TLR signalling²⁰¹, mediates protective effects in experimental models of SLE.

Notably, TLR7 seems to play a non-redundant role in the pathogenesis of SLE, at least in mice. MPL/lpr mice backcrossed into Tlr7^−/− mice have been reported to develop SLE with significantly reduced renal disease and pathogenic autoantibodies^197,202. Moreover, male BXSB mice with a Y-linked autoimmune accelerator locus (Yaa) develop spontaneous SLE-like disease due to duplication of a four-megabase gene segment containing Tlr7 transposed to the Y chromosome²⁰³. The pathogenic role of TLR7 in this latter model was demonstrated by gene dosage experiments showing that Tlr7 copy number directly correlates with disease severity in Yaa male mice²⁰⁴. Consistent with this, class R inhibitory ODNs mediated therapeutic effects in MRL/lpr mice downstream of mixed TLR3 and TLR9 inhibition²⁰⁵. Altogether, these observations suggest that TLR7 antagonists may be valuable therapeutic tools for the management of SLE. Alternating 2′-O-ribose methylation in the sense strand has been proposed as a robust chemical approach for the generation of TLR7 antagonists²⁰⁶. However, to the best of our knowledge, the clinical development of TLR7 antagonists for the management of SLE stands at an impasse (source http://www.clinicaltrials.gov).

Conversely, this approach has been pursued in clinical development for other autoimmune disorders. Experimental evidence based on knockout mice indicate indeed that TLR7 contributes to the cause of rheumatoid arthritis^207,208. At odds with the SLE setting, however, TLR3 and TLR8 also appear to mediate non-redundant pathological functions in rheumatoid arthritis. In line with this notion, Tlr3^−/− and Ticam1^−/− mice are protected from experimental rheumatoid arthritis as compared to their wild-type counterparts²⁰⁸ Moreover, topical resiquimod or imiquimod administration has been harnessed to provoke a systemic autoimmune disorder with rheumatoid arthritis features or a psoriasis-like condition in mice, respectively^67,209. Finally, the synovial lining and sublining of patients with rheumatoid arthritis express high levels of several TLRs with pathogenic potential^210,211.

Consistent with a pathological role for TLRs in various autoimmune conditions, TLR inhibitors, including ODN-1411, have been shown to limit deregulated cytokine secretion in human models of rheumatoid arthritis, in vitro²¹², and to decelerate disease progression in mouse models of psoriasis²¹³. Moreover, experimental autoimmune encephalomyelitis (a model of multiple sclerosis) progresses more slowly in Tlr9^−/− mice than in wild-type mice²¹⁴. Along similar lines, silencing regulator of G-protein signalling 1 (RGS1) suppressed endosomal TLR signalling in a rat model of collagen-induced arthritis, culminating in reduced disease progression²¹⁵. These observations spurred considerable interest in the possibility to antagonize nucleic acid-sensing TLRs for therapeutic purposes, culminating in the initiation of multiple clinical trials in patients with psoriasis (source http://www.clinicaltrials.gov). In particular, IMO-8400 (a mixed TLR7, TLR8 and TLR9 antagonist) demonstrated clinical activity in a phase IIa, randomized, placebo-controlled trial in patients with moderate-to-severe plaque psoriasis²¹⁶ (NCT01899729). Additional IMO-8400 testing has been done in patients with dermatomyositis (NCT02612857), as well as in individuals with non-Hodgkin lymphoma expressing a hyperactive variant of MYD88 (NCT02092909, NCT02252146, NCT02612857). Although official sources list all these latter studies as completed, only the results of NCT02252146 have been disseminated, demonstrating the safety of IMO-8400 in patients with non-Hodgkin lymphoma. IMO-3100 (a mixed TLR7 and TLR8 antagonist) was also recently tested in patients with moderate-to-severe plaque psoriasis (NCT01622348), and subcutaneous administration resulted in a mean reduction in the thickness of index lesion in the absence of major toxicity. Another mixed antagonist of TLR7 and TLR8 is currently under clinical development for autoimmune disorders (source https://www.bms.com/researchers-and-partners/in-the-pipeline.html) but no further information is available on the molecule and precise indication.

Cardiovascular disorders

Deregulated TLR signalling may also be involved in the development and progression of atherosclerosis, a possibility that began to be investigated following the correlation between a variety of pathogens (for example, Chlamydia pneumoniae, Helicobacter pylori, HCV and HIV-1) and increased risk of cardiovascular disorders^217,218,219. Apoe^−/− mice (a genetic model of atherosclerosis) experience a reduction in atherosclerotic lesions coupled with decreased macrophage infiltration when backcrossed on Myd88^−/− mice^220,221. Although such a protective effect was initially attributed to the inhibition of TLR2 and TLR4 signalling (which is not initiated by nucleic acids)^222,223, more recent genetic data suggest that TLR9 plays a similar detrimental role²²⁴. Consistent with this, the mixed TLR7 and TLR9 antagonist ODN-2088 ameliorated neointimal thickening and decreased macrophage infiltration in a mouse model of postinterventional remodelling²²⁵, an effect that appears to depend on TLR9 inhibition²²⁶. Indeed Tlr7^−/−Apoe^−/− mice exhibit accelerated disease progression as compared with their Apoe^−/− counterparts²²⁶, and Tlr9^−/− mice are protected against experimental cardiomyopathy driven by mitochondrial DNA spillage²²⁷. Besides identifying TLR9 as a target for the development of antagonists with cardioprotective effects, these observations raise the interesting possibility that TLR7 agonists, rather than antagonists, may be useful for the management of atherosclerosis. To the best of our knowledge, no antagonist of nucleic acid-sensing TLRs is currently under clinical development for the treatment of cardiovascular conditions (source http://www.clinicaltrials.gov).

Neurodegenerative disorders

Suggesting the involvement of TLRs in the pathogenesis of neurodegenerative disorders, Myd88^−/− mice exhibit decreased amyloid accumulation in the central nervous system as compared with their wild-type littermates in a model of Alzheimer disease²²⁸, and deregulated TLR7 signalling has been shown to cause an Alzheimer disease-like neurodegenerative disorder in mice²²⁹. Similarly, neuronal degeneration is aggravated by CpG-rich ODNs and limited by deletion of Tlr9 in a toxin-induced model of Parkinson disease ²³⁰. In addition, TLR9 levels are elevated in the striatum of post-mortem parkinsonian brains (as compared with age-matched post-mortem brains)²³¹ Conversely, TLR9 activation by CpG-rich ODNs mediated beneficial effects on amyloid accumulation and cognitive decline in a transgenic model of Alzheimer disease²³². The reasons underlying this apparent paradox remain unknown but may be related to an intrinsic difference in the inflammatory component of Parkinson disease and Alzheimer disease. Irrespective of this conundrum, the TLR9 antagonist COV08-0064 (also known as MP-3964) limited neurodegeneration in mice exposed to the Parkinson disease-inducing toxin 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine²³⁰, suggesting that TLR9 may be a target for the development of neuroprotective drugs. This therapeutic paradigm has not yet entered clinical development (source http://www.clinicaltrials.gov).

Taken together, these observations suggest that despite a large body of preclinical evidence in support of an aetiological role for multiple TLRs in several human diseases, the development of TLR antagonists proceeds at a slow pace.

RLR and cGAS–STING antagonists

Recently, pharmaceutical companies have begun to explore therapeutic RLR and cGAS–STING antagonists²³³, reflecting the aetiological contribution of deregulated NAS signalling to multiple disorders. For instance, ADAR — the main endogenous inhibitor of MDA5 signalling — is required for embryonic development and adult life owing to its ability to prevent systemic autoimmune reactions with lethal consequences and to favour the establishment of thymic tolerance^234,235. Consistent with this notion, ADAR mutations have been associated with a variety of autoimmune conditions potentially leading to immunodeficiency, including SLE²³⁶, Evans syndrome²³⁷ and Aicardi–Goutières syndrome²³⁸, three different settings in which pharmacological inhibitors of MDA5 may mediate therapeutic effects.

Along similar lines, mutations in the three-prime repair exonuclease 1 gene (TREX1), which codes for the main endogenous inhibitor of cGAS–STING signalling^159,239, have been associated with multiple autoimmune disorders, including some SLE variants²⁴⁰, and Aicardi–Goutières syndrome²³⁹. Moreover, deregulated cGAS–STING signalling has been involved in the pathogenesis of a variety of inflammatory conditions in preclinical models, including age-related macular degeneration²⁴¹, myocardial infarction²⁴², sepsis²⁴³, non-alcoholic steatohepatitis^244,245, inflammation-driven carcinogenesis²⁴⁶ and the Sjögren syndrome²⁴⁷. Altogether, these observations identify multiple diseases that would benefit from the development of cGAS or STING inhibitors. Only a few such compounds have been synthesized so far, including (1) a series of small molecules (including C-176 and C-178) that display robust covalent binding to STING and therapeutic activity in Trex1^−/− mice²⁴⁸, (2) astin C, a cyclopeptide isolated from the medicinal plant Aster tataricus that exhibits anti-inflammatory activity in both mouse Trex1^−/− and human TREX1^−/− models and sensitizes mice to HSV-1 infection²⁴⁹ and (3) X6, a small cGAS inhibitor that was superior to the antimalarial drug hydroxychloroquine in the attenuation of disease symptoms in Trex1^−/− mice²⁵⁰. Of note, recent data suggest that aspirin also acts as a bona fide cGAS inhibitor²⁵¹. This latter observation not only adds another entry to the ever-increasing list of aspirin targets but also raises the generally underappreciated possibility that common drugs may modulate NASs via unclear mechanisms, and that such pharmacological activities deserve attentive scrutiny.

In summary, although the development of RLR and cGAS–STING antagonists has just started, the signal transduction cascades emanating from these NASs stand out as particularly promising targets for the development of novel anti-inflammatory agents.

Persisting obstacles

Despite intensive efforts from numerous pharmaceutical companies, very few pharmacological NAS modulators are available for use in humans, and the development of many others was discontinued at preclinical or clinical stages, as discussed earlier. A variety of drug-related, NAS-related, disease-related and host-related obstacles can be invoked to explain the failure of several NAS modulators to enter the clinic. All these factors should be carefully taken into consideration for the successful development of novel NAS modulators for clinical use (Fig. 2).

Fig. 2: Revised pharmacological audit trail for modulators of NASs.

To develop novel pharmacological modulators of nucleic acid sensors (NASs) for clinical use, it will be critical to identify in advance (1) specific therapeutic paradigms in which alterations of NAS signalling provide an aetiological contribution to disease, (2) patient subsets among whom such defects have a major impact on disease progression, (3) cell populations in which deregulated NAS signalling drives pathogenesis (diseased cells, which do not necessarily overlap with cell populations normally associated with the clinical condition) and (4) the precise nature of NAS signalling alterations and how they impact on the biology of diseased cells as well as on the biology of neighbouring cells not manifesting NAS defects (bystander cells). This may allow the identification (ID) of a precise target for pharmacological interventions, which should be interrogated for (1) single-nucleotide polymorphisms (SNPs) of biological relevance (in both the target itself and in downstream signal transducers or effectors), (2) crosstalk/redundancy with other NASs, (3) sensitivity to standard-of-care (SOC) therapy and (4) influence of sex on biological outcome. This integrated analysis may allow the identification of a drug candidate, which will have to be developed while (1) co-developing a delivery platform as specific for diseased cells as possible, (2) considering potential biological effects from the drug vehicle, (3) identifying a suitable delivery route and (4) analysing drug–drug interactions, especially if SOC therapies influence the biological functions of the drug target, its signal transducers or downstream effectors. Finally, the ability of the candidate drug to normalize NAS signalling in diseased cells, and hence restore normal biological functions in diseased or bystander cells, will have to be assessed. In the absence of efficacy, the target, candidate drug or delivery platforms will have to be reconsidered. Conversely, a clinical response potentially linked to prolonged patient survival may become evident. In the absence of either or both, the complete therapeutic paradigm or patient selection should be entirely re-evaluated. PD, pharmacodynamics; PK, pharmacokinetics.

Drug-related obstacles

Some NAS agonists are second messengers or short-lived nucleic acid species, such as cGAMP (the natural STING ligand, which is rapidly degraded by ENPP1)¹⁵¹ and 5′-pppRNA (the natural RIG-I ligand, which is particularly unstable)¹³⁴. As a result, the use of natural NAS ligands or very similar synthetic compounds is associated with suboptimal pharmacological efficacy owing to their limited stability in vivo^134,180. However, this issue can be resolved by the development of stabilized molecules, such as ENPP1-insensitive cGAMP derivatives¹⁸⁰, SLR molecules¹³⁴ and compound 3 (ref.¹⁸⁷).

Similarly, the natural ligands of cytosolic NASs (that is, RLRs and the cGAS–STING system) are unable to cross the plasma membrane owing to their electrical charge¹³⁴, which further complicates their use. Although multiple transfection agents are available to circumvent this issue in vitro, only a few of these are compatible with in vivo applications²⁵², and even fewer have been tested in the clinic, including the linear polyethylenimine derivative JetPEI (source http://www.clinicaltrials.gov). These issues call for the development of small RLR and cGAS–STING modulators that efficiently penetrate the plasma membrane. Recent data demonstrating that aspirin can act as a cGAS inhibitor²⁵¹ suggest that acetylation may be harnessed for the development of effective cGAS inhibitors. Furthermore, these data support the potential of drug repurposing approaches to uncover unsuspected modulators of cytosolic NASs with desirable physicochemical properties. As an alternative, efforts should be dedicated towards the development of efficient transfection reagents that are fully compatible with clinical applications. This latter approach may also be helpful for targeting NAS modulators to specific disease sites or cell populations. Indeed, multiple NAS agonists can be delivered only locally (for example, intratumourally or intradermally), reflecting a high potential for on-target toxicity⁶⁷ (see later). In support of this concept, phospholipid micelles loaded with magnetic nanoparticles have been successfully used to promote the accumulation of a cancer vaccine with poly(I:C) and imiquimod as adjuvants in experimental melanomas and tumour-draining lymph nodes in mice, resulting in therapeutic effects that could be further enhanced by PD-1 blockage²⁵³.

Finally, it will be important to carefully consider the immunomodulatory effects of excipients, which (at least theoretically) may aggravate or compromise the effects of NAS modulators. As a notable example, 5% imiquimod cream (in one of the FDA-approved formulations) has been shown to mediate both TLR7-dependent and TLR7-independent immunostimulatory effects, part of which originated from the vehicle⁵⁹. This latter observation implies that higher doses of imiquimod may have been required to achieve therapeutic effects in clinical settings if another excipient had been used in the drug formulation, potentially increasing TLR7-dependent toxic effects.

NAS-related obstacles

NASs are generally expressed by a wide variety of cell types, including immune cells, epithelial cells and endothelial cells, as well as (at least some) cancer cells, reflecting their key role in the control of pathogenic infections²³. Such an expression pattern has been harnessed for the development of therapeutic paradigms based on NAS modulators, such as the use of TLR3 agonists for the induction of cancer cell apoptosis²⁵⁴. However, the same pattern poses specificity issues linked to on-target toxicity at non-diseased sites, including some cases of moderate cytokine release syndrome observed on intranasal administration of an experimental TLR7 agonist²⁵⁵. Moreover, the detection of potentially pathogenic nucleic acids is such a central biological process that several of the underlying molecular cascades exhibit a considerable degree of functional redundancy and crosstalk³ (Fig. 3). In line with this notion, although correlative data point to the implication of specific NASs in the pathogenesis of some inflammatory conditions, pharmacological or genetic inhibition of the NAS in question fails to mediate robust therapeutic effects, such as in the case of TLR3 and SLE^191,193 (see above). This issue has both experimental and therapeutic implications. On the one hand, the elevated redundancy of the system (that is, the ability of multiple NASs to detect similar nucleic acids) implies that conclusively excluding the implication of nucleic acid sensing from the pathogenesis of a specific condition may require the simultaneous inactivation of several NASs, which may not always be feasible experimentally. On the other hand, the intimate crosstalk between signalling cascades elicited by multiple NASs implies that efforts to develop NAS antagonists may have to be refocused on targets that are shared by multiple NAS-driven pathways, such as MYD88, TBK1 or IRF3, or their shared biological effectors (for example, type I interferon, IL-6). Consistent with this notion, numerous agents that neutralize NAS-related cytokines, such as the IL-6 blocker tocilizumab and the TNF blockers etanercept and infliximab, are currently available for clinical use¹⁸⁸. Conversely, a phase III trial testing anifrolumab, a monoclonal antibody that inhibits type I interferon signalling, in patients with SLE (NCT02446912) failed to meet the primary end points²⁵⁶, hence stalling the development of interferon blockers for this specific indication. The precise reasons underlying these unexpected and disappointing results remain to be elucidated. That said, it is tempting to link them to the elevated number of potentially pathogenic cytokines secreted downstream of NAS signalling.

Fig. 3: Molecular crosstalk in mammalian nucleic acid sensing.

Mammalian nucleic acid sensors (NASs) activate signalling pathways that exhibit considerable degrees of redundancy (that is, multiple NASs can be activated by the same molecular species) and crosstalk, not only as they involve shared signal transducers such as mitogen-activated protein kinase (MAPK), IκB kinase (IKK), NF-κB-inducing kinase (NIK) and TANK-binding kinase 1 (TBK1) but also as they activate transcriptional programmes that (at least in some settings) culminate in the transactivation of shared gene sets (not shown). This implies that currently available pharmacological NAS agonists (in red) and antagonists (in blue) may exhibit poor biological specificity (agonists),or limited biological efficacy (antagonists) (Boxes 2,3,5). 5′-pppRNA, 5′-triphosphate RNA; ABZI, amidobenzimidazole; AP-1, activator protein 1; CDN, cyclic dinucleotide; cGAS, cyclic GMP–AMP synthase; CMA, 10-carboxymethyl-9-acridanone; CREB1, cyclic AMP-responsive element-binding protein 1; DMXAA, 5,6-dimethylxanthone-4-acetic acid; IMO, immunomodulatory oligonucleotide; INH-ODN, inhibitory oligodeoxynucleotide; IRF, interferon regulatory factor; MDA5, melanoma differentiation associated protein 5; NAB2, nucleic acid band 2; NF-κB, nuclear factor-κB; RIG-I, retinoic acid-inducible gene I protein; SLR, short stem–loop RNA; STING, stimulator of interferon genes; TLR, Toll-like receptor.

The existence of NAS-coding gene polymorphisms also complicates the development of pharmacological NAS modulators. At least for some NASs, not all polymorphic variants are equally sensitive to a given drug or equally proficient at activating downstream signalling cascades. Thus, the five variants of human STING exhibit differential sensitivity to canonical CDNs¹⁸¹, and a large number of polymorphisms affecting nucleic acid-sensing TLRs or their signal transducers have been associated with differential functionality^257,258. Further complicating the situation, various polymorphisms have also been identified in the promoter regions of NAS-coding genes²⁵⁹. Altogether, these latter observations point to a great degree of interindividual variability in NAS responsiveness to natural and pharmacological modulators. Developing drugs that have comparable activity on the most common NAS variants may be feasible, at least in some settings¹⁸³. Conversely, genetically testing patients for polymorphic variants in NASs, their transducers and effectors may be challenging, especially considering that the functional impact of multiple polymorphisms remains to be elucidated. An alternative approach may consist of testing expected biological outcome (for example, cytokine secretion) on patient samples (for example, circulating monocytes, freshly isolated cancer biopsy specimens). However, standardizing this approach to generate a predictive biomarker for response also seems difficult.

Finally, we believe that the development of NAS modulators has been hampered by the lack of precise knowledge of the kinetics of NAS signalling, as well as by intrinsic biological differences between human NASs and their murine counterparts. Thus, in some settings, such as the tumour microenvironment, acute and robust secretion of NAS-related cytokines (for example, type I interferon) is beneficial and supports the generation of antitumour immunity, while the establishment of a mild, chronic response mediates detrimental effects³. Moreover, multiple agents that drive robust TLR signalling in human cells fail to do so in mouse systems²⁶⁰. Additional investigation is required to clarify the ideal kinetics of NAS modulation in each disease and to determine the suitability of murine platforms for preclinical development in this setting. Of note, recently uncovered biological aspects of NAS signalling (for example, the cGAS-inhibitory activity of acetylation)²⁵¹ offer a promising approach for the development of improved screening platforms for the identification of novel NAS modulators. To the best of our knowledge, however, the actual potential for discovery of such screening strategies remains to be demonstrated.

Disease-related obstacles

The development of clinical NAS modulators has often been complicated by disease-related obstacles, including the underappreciated and underinvestigated ability of standard-of-care treatments to modulate NAS signalling. These treatments include (but most likely are not limited to) aspirin, which directly inhibits cGAS²⁵¹; radiation therapy, which triggers both cGAS and RIG-I signalling^137,159; and various chemotherapeutics (for example, doxorubicin, cisplatin, etoposide), which initiate RIG-I and/or TLR3 signalling^137,261. As NAS modulators are generally administered in combination with other drugs, the development of safe and effective therapeutic regimens calls for the careful evaluation of the effects of such drugs on NAS signalling. An additional level of complexity is provided by the fact that, at least in some disease settings, including infectious and malignant disorders, the aetiological agent (that is, pathogens, cancer cells) is naturally provided with or acquires mechanisms for evading NAS signalling^141,262. This implies that some NAS modulators may be unable to mediate the expected therapeutic effects, depending on the specific molecular alteration(s) involved in such an escape mechanism. For instance, a fraction of patients with breast cancer exhibit reduced IRF7 levels (linked to poor disease outcome)²⁶³, suggesting that these tumours may be poorly sensitive to activation of NASs that culminates in IRF7-dependent transcription (Boxes 2,3,5). Finally, while dysregulated NAS signalling has long been associated with infectious, autoimmune and malignant conditions, the notion that NASs may be implicated in the pathogenesis of other conditions has just begun to emerge. Thus, excessive type I interferon production downstream of cGAS–STING signalling play a key role in the cardiac maladaptation to myocardial infarction, suggesting that cGAS or STING antagonists may mediate therapeutic effects in this setting²⁴². This possibility, however, remains to be experimentally validated. Similar considerations apply to RIG-I and cerebral stroke²⁶⁴ and (we suspect) many other disorders with a hitherto poorly characterized inflammatory component. Elucidating the precise contribution of NAS signalling to human disease may foster the development of clinically useful NAS modulators.

Host-related obstacles

An underappreciation of multiple host factors may have been an obstacle in the development of NAS modulators, as has been the case for other classes of compounds, such as autophagy modulators²⁶⁵. One such factor is represented by variants of genes that cannot be directly linked to the molecular machinery for nucleic acid sensing. For instance, specific splicing variants of WDFY family member 4 (WDFY4), a protein mostly known for its role in antigen cross-presentation²⁶⁶, have been reported to boost signalling by multiple NASs, including TLR3, TLR9 and MDA5, potentially explaining the impact of WDFY4 polymorphisms on susceptibility to SLE²⁶⁷ and dermatomyositis²⁶⁸. The existence of polymorphic variants in (at least theoretically) NAS-unrelated genes that influence NAS signalling adds yet another layer of complexity to the development of predictive biomarkers based on genetic profile. Sex has also been suggested to influence NAS signalling, via poorly characterized mechanisms. For instance, in a rat model of spontaneous hypertension, the difference in disease severity between males and females was attributed to differential levels of circulating mitochondrial DNA, resulting in exacerbated TLR9 signalling in the former²⁶⁹. Similarly, sex hormones have been proposed to affect the physiopathology of non-alcoholic steatohepatitis by altering MYD88 expression levels²⁷⁰. Although increasing attention is being given to the elucidation of sex-related differences in multiple biological processes, additional work is required to precisely characterize the impact of sex on NAS signalling.

Concluding remarks

Despite a large body of preclinical and clinical evidence demonstrating that NASs are promising targets for the development of novel pharmacological agents (Table 1), this therapeutic paradigm remains largely unrealized, at least in part due to the elevated degree of redundancy and molecular crosstalk that characterize not only NASs and their signal transduction cascades but also  their ultimate biological effectors (that is, cytokines), their receptors and the molecular pathways they activate. Thus, numerous obstacles need to be considered for the development of clinically useful NAS modulators (Fig. 2). Initial efforts were focused on agonistic (over antagonistic) molecules, largely reflecting early findings on the beneficial role of NAS signalling in the host response to viral infection. Conversely, NAS antagonists have attracted attention from pharmaceutical companies only recently²³³, following the realization that the pathogenesis of multiple human disorders besides autoimmune and inflammatory conditions involves deregulated NAS signalling²⁴². These observations explain, at least in part, the imbalance in the number of NAS agonists and antagonists under development. Over the past two decades, the development of agonists has been considerably refocused from antiviral to anticancer applications. We surmise that such a shift originated not only from the discovery of highly efficacious therapies that allow the lifelong control of previously deadly viruses (for example, HIV-1)²⁷¹ but also from the concomitant explosion of immunotherapy as a new approach to the clinical management of cancer⁷¹. In this setting (that is, cancer immunotherapy), current efforts focus on using NAS agonists as immunological adjuvants (as opposed to stand-alone therapeutic agents), as a large body of preclinical and clinical literature demonstrate that the robust immunosuppressive circuitries established by progressing tumours most often cannot be circumvented by single interventions⁷¹.

One avenue that must be pursued to foster the development of NAS modulators with clinical applications relates to a large number of proteins that de facto interact with potentially pathogenic nucleic acids but have been relatively poorly investigated as therapeutic targets, including (but not limited to) (1) nucleases such as TREX1 (ref.¹⁵⁹), RNase H²⁷² and RNase L²⁷³, (2) polymerases such as RNA polymerase III (ref.²⁷⁴), (3) components of the DNA repair machinery such as DNA-dependent protein kinase (DNA-PK)²⁷⁵, breast cancer type 1 susceptibility protein (BRCA1)²⁷⁶, double-strand break repair protein MRE11 (ref.²⁷⁷), and double-strand break repair protein RAD50 (ref.²⁷⁷), and (4) additional DNA-binding proteins involved in innate immunity, such as DEAD-box helicase 41 (DDX41)²⁷⁸, absent in melanoma 2 (AIM2)²⁷⁹, interferon-γ-inducible protein 16 (IFI16)²⁸⁰, PKR¹⁴⁷ and polyglutamine binding protein 1 (PQBP1)²⁸¹. In particular, it will be important to determine the pathophysiological role of each of these proteins in various disease settings and understand whether they are amenable to therapeutic targeting. Moreover, it will be interesting to investigate the possibility of targeting common NAS signal transducers, such as MYD88, TBK1 and the IκB kinase complex, as a potential means to circumvent the elevated degree of redundancy and crosstalk of the system. Of note, it has recently been demonstrated that supramolecular platforms commonly used to drive cytokine secretion in response to nucleic acid sensing, such as the myddosome (Box 3), can be engineered to mediate user-defined biological outcomes²⁸². These findings suggest that nucleic acid sensing involves a large degree of modularity that may be harnessed for therapeutic applications. This possibility awaits urgent experimental verification.

In summary, multiple components of the molecular machinery for nucleic acid sensing continue to attract attention as potential drug targets, although the clinical development of multiple NAS modulators has been discontinued. We surmise that the identification of pharmacological NAS modulators with increased stability and hence that are amenable to systemic delivery, combined with the acquisition of precise mechanistic knowledge of the role of NASs in healthy and diseased tissues, will be instrumental for the successful clinical translation of this therapeutic paradigm. Optimized screening strategies amenable to high-throughput applications, such as the luminescence-based approach that was recently harnessed for the identification of novel cGAS inhibitors²⁸³, will be essential in this context, as they will provide various molecular scaffolds for targeted medicinal chemistry. Additional work is required to harness the full therapeutic potential of NAS modulators.