Toll-Like Receptors in Health and Disease: Complex Questions Remain¹

The history of these receptors is now well known. In brief, Drosophila Toll regulates aspects of embryonic fly development (dorsoventral patterning), but deficiencies in Toll also render the adult fly vulnerable to fungal infections. In humans, Medzhitov et al. (1) showed that constitutively active human TLR mutants activated NF-κB and pathways that lead to the development of adaptive immunity. A key breakthrough was provided in studies showing that mice resistant to endotoxic shock contained natural mutations in TLR4 (2, 80). From there, an explosion of research has identified, thus far, 10 human TLRs that respond to a diverse range of agonists (Table I).

Much interest has focused on TLR4, facilitated by mice with naturally occurring mutations in this gene. The most important role for TLR4 is its ability to enable responses to LPS. In humans, a number of polymorphic alleles of TLR4 have been identified. One such relatively common TLR4 polymorphism may be associated with an increased risk of septic shock (3) and a decreased risk of atherosclerosis (4), but is not associated with increased risk of meningococcal disease (5, 6), although other rare alleles of this gene may predispose to this infection (6).

TLR4 is involved in signaling in response to a wide variety of exogenous and endogenous molecules (Table I). Whether all these putative activators will stand the test of time remains to be seen; for example, evidence is emerging that HSP70-mediated TLR signaling might be due to LPS contamination of protein preparations (7). There are no known structural similarities between LPS, the F protein of respiratory syncytial virus, and the anti-tumor agent, Taxol, yet all use TLR4 for signal transduction. This suggests that the original notion of recognition of common molecular patterns (PAMPs) may have to be broadened.

Other molecules, such as the immunomodulatory protein, TNF-stimulated gene 6, have selective sequences that enable interactions with molecules with differing structures such as proteins and glycosaminoglycans (8). Therefore, theoretically, the existence of specific extracellular domains in the TLR regulating interactions with multiple agonists is conceivable, but domains that conclusively mediate binding of putative TLR4 ligands have yet to be identified. Consideration of the underlying biology also allows generation of other hypotheses where specific ligand recognition becomes more complex, and where TLRs act not as receptors, but as integrators of cell signaling. Cells of the innate immune system evolved to respond to pathogens that exist as particulate matter (as opposed to soluble mediators), and the bacterial cell wall components tested in the lab as “pure” agonists, such as LPS and peptidoglycan, also typically form aggregates or micelles. Therefore, the interaction of agonist and cell normally takes place over a large contact area, perhaps with focal areas of active signaling involving multimeric protein complexes in lipid rafts, containing a variety of pattern recognition receptors and other proteins (9, 10). Thus, it has been hypothesized that direct contact of the TLRs with the agonist is not always required (10), but rather that TLR extracellular domains may play a role in stabilization of signaling complexes. The subsequent nature of the cellular response could then result from the integrated signaling of multiple proteins of many families, with potential for these helper molecules to vary between cell types and agonists (10).

In support of these arguments, HSP60, a protein released by damaged bacteria and human cells alike, signals in a TLR4-dependent manner, yet it binds to a separate receptor on the cell surface (11). There is also still controversy as to whether the principal TLR4 activator, LPS, binds to TLR4. Most certainly, there is a close interaction of LPS with CD14, MD-2, CD11b/CD18, and TLR4, suggesting presentation of LPS to TLR4 by these proteins (12), but others have not found evidence of a direct interaction of LPS with TLR4 (10). Of these accessory molecules, MD-2 plays a critical role in TLR4 expression and signaling, and may mediate interactions between LPS and TLR4 (its essential role is reviewed in Ref. 13). Additionally, mutagenesis of MD-2 identifies residues that may be proportionally more important to LPS or Taxol signaling (14). LPS responses are modified by ancillary molecules such as CD14 and CD11b/CD18, as evidenced by a highly selective diminution of LPS-induced gene expression in macrophages from mice deficient in these molecules (15). Scavenger receptors also co-operate in the recognition of LPS and other microbial agonists. Scavenger receptor-A (SR-A) and macrophage receptor with collagen structure each bind and are regulated by LPS (16, 17), and participate in microbial recognition leading to phagocytosis (16, 18). Scavenger receptors may also modulate TLR4-mediated responses to LPS, since SR-A-mediated internalization of LPS leads to its degradation but does not induce cellular activation (19), and SR-A knockout mice are more sensitive to endotoxemia (20). Finally, by analogy with signaling of Drosophila Toll where, in response to infection, proteases generate a true TLR ligand, there is some evidence for protease-dependent signaling in human TLR4 responses (21, 22), suggesting that LPS signaling may involve complex cell surface events. Thus, the nature of TLR4 remains uncertain: is it a true receptor, or an integrator of signaling but not of binding, or does it possess both capabilities? For many stimuli, it is probably more appropriate to refer to TLRs as Toll-like responders rather than receptors.

The concept of TLR signaling within macromolecular complexes allows the development of other hypotheses. The generation of constitutively active TLR4 homodimers and TLR2 heterodimers can be achieved by the overexpression of extracellularly mutated TLR proteins or fusion proteins of intracellular TLR and extracellular integrin or CD4 domains (1, 23, 24), and TLR4 cross-linking can also cause signaling (25). That the elimination of the leucine-rich repeats (LRRs) that comprise much of the extracellular domain of TLR4 enables constitutive signaling suggests that the LRRs may even act to prevent multimerization of TLRs under resting conditions. It is also possible that some TLRs, which may exist in vivo as preformed dimers, could signal continuously even in the absence of ligand. In such a scenario, while they are distributed around the membrane and are separated from each other, they would fail to achieve a local signaling threshold required for activation of the cell. Additionally, local mechanisms that deactivate or inhibit signaling (e.g., association with suppressor proteins) may prevent responses in the absence of stimulus, or negatively feedback on the cell to repress signaling. In this model, the assembly of signaling complexes may enhance rate of signaling by stabilizing receptors long enough to permit interactions with signaling systems effectively (“kinetic proofreading”, see Ref. 26), by permitting new signaling complexes to interact, or by segregating signaling activators from signaling suppressor systems.

TLR2, like TLR4, mediates responses to a wide range of molecules, and requires helper proteins such as CD14 for at least some of these responses. There is evidence for a direct interaction of the receptor with agonists such as peptidoglycan (27), but it is a reasonable speculation that TLR2 signaling may also be at least in part dependent upon lipid raft or membrane complex formation, and that it may not in all circumstances truly bind the putative “ligand”. Recent work has highlighted the role of the coreceptor, Dectin-1, in TLR2 signaling (28, 29).

Additional questions remain regarding the site of activation of TLRs. TLR2 and TLR4 activation can take place at the cell surface, and leukocytes that are responsive to LPS typically show cell surface TLR4 expression (30). However, many TLR agonists are not exposed on the surface of microbes. TLR3 responds to double-stranded RNA (31), which is produced and is likely to be a functional agonist intracellularly (32). Some PAMPs may require phagocytosis and intracellular processing to signal via TLRs, and TLR2/TLR6 heterodimers can become associated with phagolysosomes, allowing interaction with their constituents (33). TLR2 is also responsible for signaling in response to some mycobacterial cell wall constituents, another intracellular pathogen (33). Other proteins may be involved in intracellular responses to PAMPS, including NOD1 and NOD2, proteins containing LRR and CARD motifs, which have been implicated in intracellular signaling in response to bacterial constituents (34, 35). TLR9 activation by CpG-motif-containing oligonucleotides requires internalization of CpGs in endosomes, which themselves need to mature and become acidified to enable signaling, suggesting that a specific mechanism is required to enable agonist/TLR interaction (36). To complicate matters, not all CpG oligonucleotides cause the same pattern of cellular activation (37), suggesting that TLR9 function may be modified in different cell types, perhaps by heterodimerization with another TLR. Thus, the mechanisms of agonist recognition and engagement with cell signaling may show considerable variation between TLRs, and for each TLR perhaps also show variations between cell types.

TLR signaling is an important component of our response to microbes, but not necessarily the only contributor to the final cellular output (10). Within such multicomponent pathways, some elements may be shared. For example, phosphoinositide-3 kinase (PI-3K) activation is a feature of both TLR signaling and activation of FcγRs during the uptake of opsonized microorganisms (38). Additional complexity of response is provided by the ability of single microbes to activate multiple TLRs, as illustrated by Streptococcus pneumoniae, which may activate TLR4 via a virulence factor, pneumolysin (39), TLR2/6 and 2/1 heterodimers via peptidoglycan and lipoteichoic acid, and TLR9 via CpG motifs, and these responses may occur at various membrane and intracellular locations in a coordinated pattern. Many TLR responses are associated with new gene transcription, e.g., for cytokine synthesis. Other responses in leukocytes, such as induction of IL-1β generation through activation of the inflammasome (40) or respiratory burst (41), point to signaling that does not depend upon gene transcription. Responses of some cells to TLR activators may be additionally dependent upon autocrine or paracrine loops involving cytokine (e.g., TNF-α, IFNs, IL-1) or proinflammatory mediator (e.g., PAF) generation (30, 42).

TLRs share a common activation pathway mediated through their Toll-IL-1R domain (TIR) signaling domains, resulting in activation of NF-κB (Fig. 1). The adapter proteins MyD88 and TIR-containing adapter protein (TIRAP)/MyD88 adapter-like associate via TIR-mediated interactions with the TLR, and initiate the formation of a signaling complex involving IL-1R-associated kinases (IRAKs) and TNFR-associated factor 6 (TRAF6) (though IRAK-independent signaling via TRAF6 has been described for TLR3 (43)). Interspecies differences may affect our understanding of these signaling pathways, as illustrated by IRAK4^−/− mice showing impaired responses to a broad range of viral and bacterial type stimuli (44), yet IRAK4 deficiency in humans resulting in susceptibility to a more limited range of pyogenic organisms, including S. aureus and S. pneumoniae, but not viruses or opportunistic infections (45, 46). Dissociation of the early signaling complex within the cytoplasm allows the interaction with proteins such as the MAP3K, TAK1, which plays a role in both IKK activation and signaling into the mitogen-activated protein kinase (MAPK) cascades (47). Activation of the IKK complex triggers the activation of NF-κB by well-characterized pathways involving ubiquitination and degradation of IκBα (reviewed in Ref. 48).

Individual TLRs also cause specific signaling, in both broad and subtle patterns. Despite the prevalent notion that TIR domains are highly conserved, the homology of these regions between TIR-containing receptors is actually relatively low, and although there are more focused areas of conservation in key regions involved in signaling, these TIR domains provide ample room for TLR-type specific interactions with downstream signal components (49). To date, much of our knowledge of these signaling pathways is derived from studies of knockout mice (natural or engineered). Other work has exploited lessons learned from the C3H/HeJ mouse, mutating the proline residue in the BB loop of the TIR domain to generate dominant negative (DN) constructs that can be used to block TLR signaling, often in comparison with inhibitory peptides whose sequence includes the BB loop of the targeted TIR domain, coupled to a cell-permeable sequence such as that derived from Antennapedia protein. However, studies of DN constructs have not always given the same results as studies of the relevant knockout mouse; for example, recent studies of the TIRAP knockout mouse have shown it to be involved in both TLR2- and TLR4-mediated activation of NF-κB (50), whereas other studies associated it more selectively with TLR4 signaling (51, 52). Resolution of these studies is complicated by differences between species, and an incomplete understanding of redundancy and compensatory adaptation in knockout mice, together with uncertainty over the specificity and exact modes of action of DN signaling constructs or inhibitory peptides.

One selective TLR signaling pathway is the MyD88-independent activation of IFN regulatory factor 3 (IRF-3), that has been implicated in generation of type 1 IFNs. Although the induction of IFN-β was first demonstrated to distinguish TLR4 from TLR2 signaling (52), both TLR3 and TLR9 activation also result in the production of IFN-β. Secreted IFN-β can act in an autocrine fashion to induce a variety of IFN-dependent genes such as the chemokine CXCL10 and inducible NO synthase. Data suggested that both LPS-induced IFN-β, as well as STAT1 tyrosine phosphorylation, were MyD88-independent, but required coupling of the adapter protein TIRAP to TLR4 (51, 52). However, recent evidence from knockout mice demonstrates that TIRAP plays a role in the activation of NF-κB by TLR2 (50). Very recently, a new adapter protein, Toll-IL-1domain-containing adapter molecule-1/TIR domain-containing adapter inducing IFN-β (53) has been identified that is associated with type 1 IFN generation mediated via TLR3, although its role in LPS-induced activation of IFN-β remains uncertain. It seems likely that additional adapters will be described in the future that provide further mechanisms facilitating selective TLR signaling. Other IFN-associated transcription factors, e.g., IRF-1 and IRF-2, have also been implicated in TLR4 signaling. Mice deficient in these factors show various patterns of susceptibility to specific infections and impaired generation of IL-12 and IFN-γ in response to LPS (54).

The growing family of adapter proteins does not necessarily explain the more subtle differences in signaling between TLRs: for example, TLR2 and TLR4 exhibit both shared and different actions on human neutrophils that are most likely to be caused by relatively small differences in activation of proinflammatory signaling pathways perhaps including MAPKs, protein kinase Cs, and PI-3K (41). The processes that enable these smaller differences in signaling output remain to be clarified—although once again, a mechanism where TLR responses are integrated with, and modified by, signaling derived from an activated membrane complex of many proteins offers possible additional explanations for selective TLR responses (10).

Additional complexity is provided by the fact that TLR signaling results in contrasting outputs in different cell types. For neutrophils, TLR4 signaling acts both directly and indirectly via monocytes to prevent constitutive apoptosis (41), yet TLR4 activates a Janus kinase/caspase-dependent pathway that induces endothelial cell apoptosis (55), and in monocytes, TLR2 activates both prosurvival and proapoptosis signals, the latter via Fas-associated death domain protein and caspases (56). IRF-1 and IRF-2 are also involved in LPS-induced apoptosis, with IRF-1 conferring resistance to LPS-induced apoptosis in macrophages (57), while IRF-2 knockout mice show increased LPS-induced apoptosis of Kupffer cells (58).

Thus, the coordinated outputs of activation of multiple TLRs, in association with the activation of many other receptors involved in host defense, results in the development of innate immunity and links with the development of adaptive immunity. Cytokine induction enables specific defense: IRF-3 mediates antiviral responses via IFN-β, while IL-12 is essential for defense against intracellular pathogens. The killing of phagocytosed microorganisms by macrophages involves TLR-mediated up-regulation of inducible NO synthase, particularly in rodent macrophages which have a high-output system for NO production, by shared and TLR-specific pathways (52, 59).

Down-regulation of signaling is also an area of considerable interest. There is evidence that TLRs can desensitize their own signaling, and also that TLR4 and TLR2 can heterologously desensitize each other (60). This may in part be achieved by the increased expression of inhibitory molecules, such as the inhibitor of signaling, IRAK-M (61), or the association of Tollip with the signaling complex, preventing IRAK activation (62). Unsurprisingly for a kinase-regulated system, investigations of IL-1 signaling suggest that relatively specific phosphatases may participate in signaling deactivation (63). Suppressor of cytokine signaling 1, a member of the suppressor of cytokine signaling family that suppresses the JAK-STAT signaling cascade activated by proteins such as IFN-γ, also reduces NF-κB activation by LPS and appears to be involved in TLR4 homologous desensitization (64). Finally, in monocytes pretreated with LPS to induce a state of “in vitro tolerance”, MyD88 fails to associate with TLR4 in response to LPS in the manner seen in normally responsive cells (65).

Microorganisms have also learned to down-regulate TLR activation, presumably to promote pathogen survival. The vaccinia virus genome encodes proteins that prevent NF-κB activation by TLRs including TLR3, by inhibiting the interactions of IRAKs and TRAF6 (66). Bacteria of Yersinia spp. secrete a protease, YopJ, that inhibits NF-κB activation and impairs TLR-mediated cell survival signals, but not proapoptotic signals, resulting in TLR-mediated induction of macrophage apoptosis (67). These bacteria additionally use virulence factors to induce a TLR2-mediated, IL-10 dependent, anti-inflammatory response (68). Mycobacteria have mechanisms to inhibit effective Ag presentation, since lipoarabinomannan binding to dendritic cell (DC) lectins inhibits TLR-mediated cell maturation (69).

Reduction of excess inflammation, for example in the context of septic shock, by down-regulating TLR responses is a feasible therapeutic goal. TLR4^−/− mice are resistant to the effects of systemic endotoxin. LPS or lipid A from Rhodobacter sphaeroides, and the synthetic lipodisaccharide, E5564, prevent activation of human TLR4 by LPS, offering the possibility of new treatments for septic shock (70, 71). Microbial targeting of intracellular signaling illustrates other approaches that could be adapted to become viable pharmaceuticals. There are numerous other inflammatory diseases where TLR activation may be important. In diseases such as atherosclerosis or some arthritides, involvement of infectious agents in disease initiation has been proposed, but intervening at disease initiation would require a degree of prescience we do not yet possess. Nonetheless, in these diseases TLR signaling may be important in disease progression, if infectious agents show persistence, or if endogenous damage signals are acting in a TLR-dependent manner. The association of genetically impaired TLR4 function with a decreased risk of atherosclerosis supports such a hope (4).

TLR activators are extremely potent adjuvants, acting on DC to drive typical Th1 type responses during Ag presentation. Contrary to expectations, TLR4^−/− mice also showed a role for TLR4 in driving Th2 cytokine generation through modulation of DC costimulatory molecule expression (72), and LPS stimulation amplified Th2 responses in a mouse model of allergic airway inflammation (73). Some data also suggest that TLR2 activation can favor Th2 phenotypes by induction of IL-12 p40 homodimers and a reduced capacity for production of bioactive IL-12 p70 (74, 75). Nonetheless, the general trend for TLR activation to favor Th1 suggests that activation of these receptors in the context of allergen presentation may be exploited to augment allergen desensitization therapies, or perhaps anti-tumor responses. Imidazoliquones, synthetic compounds that may have a therapeutic range against a variety of viruses and parasites, activate TLR7 and TLR8 (76), and their Th1-inducing activities also have potential adjuvant roles in human therapeutics (77). In the context of allergen desensitization, interest has focused on the potent Th1-driving abilities of CpG motifs acting via TLR9, and direct conjugation of CpG oligonucleotides to allergen may be a potent stimulus to cause allergen desensitization while minimizing potential CpG side effects (reviewed in Ref. 78). Similarly, the finding that vaccine-enhanced RSV disease can be mitigated by administering the vaccine together with a nontoxic TLR4 agonist, monophosphoryl lipid A, also provides a potential example of TLR4 manipulation to prevent disease (79).

While the study of TLRs is still in its infancy, the field has made rapid strides to explain many aspects of our ability to respond to the pathogenic world around us. Many questions remain regarding the agonist recognition systems and signaling pathways of these molecules, but their exploitation is likely to generate new therapies for inflammatory diseases.