A cross-disciplinary perspective on the innate immune responses to bacterial lipopolysaccharide

Abstract

The study of innate immunity to bacteria has focused heavily on the mechanisms by which mammalian cells detect lipopolysaccharide (LPS), the conserved surface component of gram-negative bacteria. While Toll-like Receptor 4 (TLR4) is responsible for all the host transcriptional responses to LPS, recent discoveries have revealed the existence of several TLR4-independent responses to LPS. These discoveries not only broaden our view of the means by which mammalian cells interact with bacteria, but also highlight new selective pressures that may have promoted the evolution of bacterial immune evasion strategies. In this review, we highlight past and recent discoveries on host LPS sensing mechanisms and discuss bacterial countermeasures that promote infection. By looking at both sides of the host-pathogen interaction equation, we hope to provide comprehensive insights into host defense mechanisms and bacterial pathogenesis.

Introduction

Precise spatial and temporal regulation of the innate immune response is essential for the host to defend against pathogen infection. However, when such responses become over-amplified or dysregulated, the resulting disruption of homeostasis may lead to the development of immunopathology. In line with this, a deleterious outcome that correlates with bacterial infection is sepsis, which is a clinical syndrome with high mortality rate (Angus et al., 2001). Specifically, during gram-negative bacterial infection, lipopolysaccharide (LPS), the major cell wall component of these microbes, plays a dominant role in the induction of sepsis and hence is referred to as endotoxin. On the molecular level, a hallmark of sepsis is the uncontrolled release of pro-inflammatory cytokines (Cohen, 2002). However, for more than a century, the mechanism of sepsis induction has remained elusive due to the obscure nature of the cellular receptor for LPS. In the past two decades, a series of elegantly designed studies unraveled the host surface molecules responsible for LPS sensing. These molecules include LPS binding protein (LBP), CD14, MD-2 and TLR4 (Gioannini and Weiss, 2007). Among them, TLR4 represents central mediator of innate and adaptive immune responses to LPS (Medzhitov, 2009; Poltorak et al., 1998a). The observation that TLR4-deficient cells lose all the canonical responses to LPS, such as expression of pro-inflammatory cytokines and interferons (IFN), has established the dogma that TLR4 accounts exclusively for all the host responses to LPS (Beutler, 2000; Beutler et al., 2001). Strikingly, recent studies have revealed the existence of host responses to LPS that do not require TLR4 (Hagar et al., 2013; Kayagaki et al., 2013; Zanoni et al., 2009; Zanoni et al., 2011). These new discoveries allow the field to move away from the TLR4-centric view of LPS signaling and begin focusing on questions of how the diverse LPS response pathways in mammalian cells are integrated. Moreover, these discoveries reveal opportunities to uncover novel immune-evasion strategies used by successful bacterial pathogens. In this review, we adopt a host-pathogen interaction perspective by highlighting both sides of this equation. We first focus on the identification and operation of the LPS response pathways that are induced by bacteria, thus painting a picture that would lead to a more thorough understanding of gram-negative bacterial infection induced sepsis. We then focus on the means by which nature has permitted some bacteria to evade these LPS response pathways. This latter discussion reveals a much more comprehensive view of host-microbe interactions, and highlights the fact that during the evolutionary ‘arms race’ with the host, bacterial pathogens have developed sophisticated mechanisms to subvert host defense.

Identification of proteins that function as sensors of extracellular LPS

Crucial to the innate immune system is its rapid responsiveness to the presence of infectious agents. In the case of gram-negative bacterial infection, mammalian cells have evolved the capability to sense picomolar (pM) concentrations of LPS (Gioannini and Weiss, 2007). Such a level of sensitivity is thought to facilitate microbial detection and control of infection as rapidly as possible. Tremendous efforts have been devoted to elucidating the host molecules that participate in LPS sensing over the last 25 years.

The isolation of LBP from acute phase rabbit serum represented a landmark in our knowledge of host LPS sensing (Tobias et al., 1986). LBP exhibits high affinity to LPS and hence mediates the extraction and disaggregation of LPS from the cell wall of gram-negative bacteria. Furthermore, LBP functions as an opsonin to facilitate the binding of bacteria or LPS-coated red blood cells to macrophages, which can subsequently induce tumor necrosis factor alpha (TNFα) secretion (Schumann et al., 1990). Thus, it has been proposed that LBP delivers LPS to a putative receptor expressed on the macrophage surface. The first candidate for this putative receptor was CD14. Non-genetic evidence in support of this idea came from the use of monoclonal antibodies against CD14 that down-regulated the adhesion of LBP-LPS complexes to macrophages (Wright et al., 1990). Additionally, a chemically modified LPS variant has been shown to crosslink CD14 on the surface of human THP-1 monocytes (Tobias et al., 1993). Originally described as a marker for myeloid lineage cells, CD14 is anchored on the macrophage surface by its C-terminal glycosylphosphatidylinositol (GPI) lipid modification, without any transmembrane segments (Haziot et al., 1988). Interestingly, a soluble form of CD14 (sCD14) that lacks its GPI membrane anchor exists in plasma and plays a role in sensitizing CD14-negative cells (e.g. epithelial and endothelial cells) towards LPS stimulation (Frey et al., 1992; Pugin et al., 1993). Genetic proof of the importance of CD14 in LPS detection derives from studies using mouse lacking this gene. CD14 deficient mice are resistant to doses of LPS that are lethal to wild type mice (Haziot et al., 1996), and CD14-deficient cells are defective for cytokine expression induced by low concentrations of LPS. However, these mice also suggested that CD14 was not solely responsible for LPS responses, as CD14 deficient cells were still able to respond to high doses of LPS (Haziot et al., 1996; Perera et al., 1997). This observation, coupled to that fact that CD14 does not possess a transmembrane domain, suggested that additional protein(s) might mediate signal transduction from the cell surface to the cytoplasm.

The identification of TLR4 by a genetic approach solved this puzzle. Long before the dawn of molecular biology, it has been reported that mouse strains such as C3H/HeJ and C57BL/10ScCr were resistant to the lethal effect of LPS treatment (Coutinho et al., 1977; Heppner and Weiss, 1965). Consequently, the genomic region containing the mutation that renders C3H/HeJ mice hyporesponsive to LPS was named the Lps locus (Poltorak et al., 1998b). Shortly after the discovery of the human TLR4 gene (Medzhitov et al., 1997), Craig Gerald noted that the human TLR4 locus was syntenic with the mouse LPS locus (Gerard, 1998), suggesting that TLR4 was involved in LPS detection and signaling. Positional cloning of the functional gene in the LPS locus revealed that TLR4 was indeed required for LPS responses in mice (Poltorak et al., 1998a). Reverse genetics approaches adopted by Shizuo Akira and colleagues to generate TLR4 knockout mice substantiated the central role of TLR4 in LPS detection (Hoshino et al., 1999). Indeed, TLR4-deficient mice are resistant to traditional murine models of LPS-induced sepsis and TLR4-deficient macrophages are unable to induce the expression of any genes in response to LPS.

While mouse genetics defined the critical role of TLR4 in LPS signaling, attempts to reconstitute the TLR4-LPS signaling pathway in vitro revealed that an additional accessory molecule(s) was required. For example, expression of TLR4 in the human HEK293 cells or mouse Ba/F3 cells did not confer LPS responsiveness to these cell lines (Shimazu et al., 1999), suggesting that another molecule is required to permit TLR4 to signal in response to LPS. Bioinformatic analysis provided clues to the identity of this unknown molecule. In the early 90s, it was discovered that radioprotective protein 105 (RP105) could convey signals that are important for B cell activation and resistance to radioactive stress (Miyake et al., 1994). Subsequently, a secreted protein called MD-1 was found to interact with the leucine-rich repeat (LRR) domain of RP105 and increase the expression of this protein. Based on the N-terminal sequence homology between RP105 and TLR4, it was speculated that a protein similar to MD-1 might interact with TLR4, facilitating the function of this receptor. Indeed, computational analyses identified a protein homologous to MD-1, which was therefore named MD-2 (Shimazu et al., 1999). MD-2 expression conferred LPS responsiveness to the Ba/F3 cell line stably expressing TLR4, and MD-2 physically interacted with the extracellular domain of TLR4. Perhaps most interestingly, MD-2 binds to LPS directly (Shimazu et al., 1999; Viriyakosol et al., 2001) and does so with high affinity. Derivation of MD-2 knockout mice solidified its role in LPS signaling, as these mice phenocopy TLR4 deficiency is all assays of LPS-induced gene expression (Meng et al., 2011; Nagai et al., 2002b). Taken together, CD14 and MD-2 were revealed as critical molecules required for LPS sensing, whereas TLR4 was responsible for inducing LPS-induced signal transduction.

Recognition of LPS by the TLR4/MD-2 Complex

Knowledge from charactering the composition of LPS provides a solid basis for our understanding of how it is recognized by the TLR4/MD-2 complex. Serving as the major outer membrane component of gram-negative bacteria, LPS is not a simple, homogenous compound. Rather, it is a mixture of lipid and carbohydrates with high degree of structural complexity. In general, LPS is comprised of three components: the O-antigen, the core oligosaccharide and lipid A. Among these, lipid A accounts for most of the immunostimulatory effects of LPS (Rietschel et al., 1994). In the case of Escherichia coli LPS, the lipid A backbone consists of two phosphorylated glucosamine molecules, to which six acylated lipid chains with a carbon length of 12-16 are attached covalently. The phosphorylation status, the number and the carbon length of acyl chains determine the agonistic activity of lipid A to the TLR4/MD-2 complex (Raetz et al., 2007). As will be discussed in more detail in this review, alteration in lipid A structure is a virulence strategy exploited by bacteria to avoid host detection. Interestingly, the recognition by TLR4/MD-2 of lipid IVa, the precursor of lipid A that only contains four acyl chains, exhibits species-specificity. For instance, lipid IVa acts as an antagonist on human macrophages whereas it is a weak agonist on mouse macrophages (Golenbock et al., 1991).

A growing body of structural studies revealed the underlying mechanism of LPS recognition by TLR4/MD-2. In the absence of LPS, MD-2 forms a heterodimer with the horseshoe-like extracellular domain of TLR4 (Kim et al., 2007; Park et al., 2009). Such binding is mediated largely by hydrophilic interactions. The MD-2 binding interface on TLR4 is comprised of the A and B patches. The A patch contains mostly negative charged residues, whereas the B patch is mainly positively charged. Respectively, the TLR4-binding surface of MD-2 consists of residues that are charge-complimentary to TLR4. MD-2 belongs to a small β-cup fold protein family, with the LPS-binding pocket formed by two antiparallel-β sheets (Ohto et al., 2007). The high affinity binding between LPS and MD-2 is achieved by the optimized distribution of the hydrophobic residues inside the pocket. In particular, five of the six acyl chains in lipid A fit inside the MD-2 cavity; whereas the sixth acyl chain extends on the surface of MD-2. Such partial exposure of the sixth acyl chain is essential for the hydrophobic interaction with the C-terminal portion of the TLR4 extracellular domain from another TLR4/MD-2 complex, thereby forming the M-shaped TLR4/MD-2/LPS dimer (in a 2:2:2 format) (Park et al., 2009). LPS-induced dimerization of the two TLR4/MD-2 complexes in the extracellular space likely facilitates the dimerization of the TLR4 intracellular TIR domain that subsequently initiates TLR signaling.

Orchestrated action of LBP, CD14, TLR4 and MD-2 in sensing extracellular LPS

Following the discovery and functional characterization of LBP, CD14, TLR4 and MD-2, fascinating questions emerged as how these molecules coordinate with each other to elicit host LPS response. To this end, the Weiss group engineered an auxotrophic Neisseria strain to produce ¹⁴C labeled LPS via metabolic labeling (Giardina et al., 2001). Due to the absence of the O-antigen, the Neisseria LPS is also referred to as lipooligosaccharide. The application of radio-labeled LPS ensures the sensitive and quantitative detection of intermolecular LPS transfer. Results from several elegantly-designed studies demonstrated that in the presence of albumin, LBP can bind to LPS-rich surfaces (e.g. bacterial outer membranes), which somehow alters the organization in the membrane to permit the binding of LPS by CD14, thus creating monomeric complexes between these two molecules (Gioannini and Weiss, 2007; Prohinar et al., 2007). CD14 then transfers LPS to the TLR4/MD-2 complex, thereby leading to cell activation (Gioannini et al., 2004; Gioannini et al., 2005) and (Figure 1). Remarkably, it has been estimated this LPS-transfer cascade can convert a single E. coli bacterium into enough LPS-CD14 complexes that can activate a thousand macrophages (Gioannini and Weiss, 2007), massively enhancing the host response.

Figure 1. Diverse Host Responses to LPS.

Macrophage are capable of sensing LPS in the extracellular milieu and the cytosol. Detection of extracellular LPS is achieved by sequential actions of LBP, CD14 and the TLR4/MD-2 complex. TLR4 induces distinct host transcriptional responses at different cellular localizations, which is determined by the sorting adaptors TIRAP and TRAM. The LPS receptor CD14 also acts independently of TLR4 to promote endocytosis, calcium fluxes and NFAT activatyion, the latter two activities occurring only in DCs. Finally, an unknown intracellular sensor induces caspase-11-dependent responses to intracellular LPS.

TLR4-dependent signaling responses to LPS

A general working mechanism for the TLR-mediated signal transduction is that, upon ligand detection, the cytosolic TIR domain of TLRs recruits one or more cytoplasmic TIR domain-containing proteins through TIR-TIR interactions. MyD88, TIRAP, TRIF and TRAM are the best characterized TIR domain-containing proteins that serve as adaptors in TLR signaling. Upon TLR activation, these adaptors bridge TLRs to downstream kinases of the IRAK and MAPK families. Together with TRAF family E3 ligases, the TLR signaling cascade activates several transcriptional regulators, such as NF-κB, AP-1 and interferon regulatory factors (IRF), which collectively drive the expression of genes involved in the host immune responses (Akira et al., 2006). Specifically, the adaptor pair MyD88 and TIRAP elicits the production of pro-inflammatory cytokines by activating the NF-κB and MAPK-dependent pathways (Horng et al., 2002; Yamamoto et al., 2002), whereas the TRAM and TRIF pair activates the kinases TBK1 and IKKε to promote type I IFN expression (Yamamoto et al., 2003a; Yamamoto et al., 2003b). Thus, recruitment of these adaptors ensures differential signaling outcomes by different TLRs. Among all the TLRs, TLR4 is unique in a sense that it can use all of the four adaptors, leading to the production of both pro-inflammatory cytokines and type I IFN. These observations raised the question of how TLR4 differentially propagates the MyD88 and TRIF-dependent pathways to mediate two distinct signaling outcomes.

The answer to this question was provided by the studies deciphering the cell biological properties of these aforementioned adaptor proteins (Figure 1). At steady state, the TIRAP localizes to the plasma membrane and endosomes via interactions with various acidic phosphoinositides (and phosphatidylserine) that are found on these compartments (Kagan and Medzhitov, 2006; Bonham et al., 2014). These protein-lipid interactions are mediated by an N-terminal lipid binding domain of promiscuous specificity. The promiscuity of TIRAP's lipid binding domain allows it to survey multiple organelles for the presence of activated TLRs (Kagan and Medzhitov, 2006; Bonham et al., 2014). In the case of TLR4, enrichment of phosphatidylinositol-4,5-bisphosphate [PI(4,5)P2] in plasma membrane sub-domains such as lipid rafts and actin-rich membrane ruffles recruits TIRAP to these locations (Kagan and Medzhitov, 2006). Similarly, the plasma membrane and endosomal localization of TRAM is determined by an N-terminal bipartite localization domain, which is comprised of a myristoylation site and a PI-binding motif (Kagan et al., 2008; Rowe et al., 2006). Upon sensing LPS, TIRAP is thought to interact with TLR4 at PI(4,5)P2-rich plasma membrane sub-domains and recruit cytosolic MyD88, which eventually leads to the assembly of myddosome, a protein complex consists of TIRAP, MyD88 and IRAK family kinases (Bonham et al., 2014). This complex promotes pro-inflammatory cytokine production. Subsequently, TLR4 is internalized into endosomes where it engages TRAM, which further recruits TRIF to trigger type I IFN production (Kagan et al., 2008). Importantly, the plasma membrane signaling of TLR4 could be uncoupled from its endosomal signaling, as dynasore (a small molecule that blocks endocytosis) inhibited the TRIF-dependent pathway, whereas it exerted no effect on the MyD88-dependent pathway initiated from the plasma membrane (Kagan et al., 2008). Consequently, TIRAP and TRAM could be categorized as “sorting adaptors” which function to define both the subcellular sites of TLR signaling and the nature of the signaling pathways activated (Figure 1) (Kagan, 2012).

The cell biological study of TLR4 and its associated adaptors has not only explained the operation of the LPS signaling pathways, but has revealed several emerging concepts in the operation of innate immune signaling in general. For example, TLR4 is not the only receptor that separates its subcellular site of ligand binding from the site of signal transduction. Recent work revealed that TLR2, TLR9 and the cytosolic RIG-I like Receptors (RLRs) also follow this cell biological “rule” (Barbalat et al., 2009; Sasai et al., 2010; Dixit et al., 2010). Second, sorting adaptors such as TIRAP and TRAM are localized at the site of signal transduction before any microbe is detected, and thus define the subcellular sites of innate immune signal transduction. The family of sorting adaptors in the innate immune system now extends beyond those used by TLR4, and includes the RLR adaptor MAVS (Kagan, 2012). Finally, as will be discussed below, cell biological analysis of TLR4 revealed the existence of cellular responses to LPS that do not actually involve this receptor (Figure 1).

TLR4-independent responses to extracellular LPS

The traditional view of CD14 is that it functions to deliver LPS to the TLR4/MD-2 complex, thus acting to sensitize mammalian cells to low concentrations of LPS. Recent studies have revealed a more complex role of this LPS-binding protein. Biochemical analysis indicated that CD14 is required for the recruitment of TLR4 to lipid rafts upon LPS binding (Triantafilou et al., 2002), an observation suggested that CD14 might have more functions than simply delivering LPS to TLR4/MD-2. Further supporting evidence for this notion came from studies demonstrating that CD14 deficiency does not ablate MyD88 and TRIF signaling to similar extents. TRIF signaling appears to be most dependent on CD14, as opposed to MyD88 signaling (Jiang et al., 2005; Perera et al., 1997). An explanation for this finding came from studies that established a crucial role for CD14 in promoting the endocytosis of TLR4, a process that is necessary for TRIF-dependent signal transduction (Zanoni et al., 2011).

Perhaps the most notable feature of CD14-dependent TLR4 endocytosis is that this activity does not require the signaling functions of TLR4. Two pieces of evidence are in support of this finding. First, a TLR4 mutant that lacks all signaling functions retains the ability to be internalized in response to LPS treatment (Zanoni et al., 2011). Second, cells deficient in both of the signaling adaptors MyD88 and TRIF also retain the ability to internalize TLR4 in response to LPS (Zanoni et al., 2011). Thus, CD14 controls a cellular response to LPS that does not depend on TLR4 signal transduction (Figure 1). Characterization of the CD14-dependent endocytosis pathway reveals the involvement of additional regulatory components, such as phospholipase Cγ2 (PLCγ2), intracellular calcium, the tyrosine kinase Syk and the ITAM containing adaptors DAP12 and FcεRγ (Chiang et al., 2012; Zanoni et al., 2011). That this newly defined cellular response to LPS proceeds independently of TLR4, but is required by TLR4 to induce TRIF-dependent signaling, highlights the interconnections between the signaling pathways activated when mammalian cells encounter gram-negative bacteria. It remains to be determined whether the CD14-dependent recruitment of TLR4 to lipid rafts is required for receptor internalization, what other proteins are internalized into cells via the CD14-dependent endocytosis pathway, and how cargo for this pathway is determined.

The role of CD14 in promoting endocytosis in response to LPS occurs in all cell types examined that co-express TLR4 and CD14, including macrophages, dendritic cells (DCs), fibroblasts and B-cells engineered to express both genes (Zanoni et al., 2011). Remarkably, recent studies revealed another TLR4-independent response to LPS that occurs in DCs specifically (Figure 1). In DCs, LPS upregulated several genes that are not expressed in macrophages, such as IL-2, nur77 and mPGES-1, the latter being an enzyme that promotes the synthesis of prostaglandin E2 (Zanoni et al., 2012). The products of these genes regulate processes that distinguish DC and macrophages functions in the activation of naïve T-cells. For example, mPGES-1 controls edema formation, which is needed to promote efficient delivery of antigens to the lymph nodes that drain infected tissues (Zanoni et al., 2012). Importantly, Nur77 expression is important for DC homeostasis, as it regulates the lifespan of LPS-activated DCs. The expression of these genes is dependent on NF-κB, as well as the calcium-regulated transcription factors of the NFAT family (Zanoni et al., 2009). Whereas TLR4 can induce NF-κB activation, it cannot activate NFAT or induce calcium influxes. These latter two activities are triggered by CD14, and proceed normally in cells derived from TLR4 knockout mice (Figure 1). Thus, a common theme emerges whereby CD14 activates an interconnected set of TLR4-dependent and –independent responses that are required to promote the expression of genes that control innate and adaptive immunity.

TLR4-independent responses to intracellular LPS

In this section, we continue the exploration of TLR4-independent responses to LPS, but we now move to the intracellular environment of mammalian cells. The study of intracellular LPS responses has been almost non-existent, as extensive characterization of cellular responses induced by extracellular LPS has dominated the past two decades. Nevertheless, scattered evidence suggested the existence of TLR4-independent host responses to LPS (Vogel et al. 1980; Haziot et al., 2001; Gerth et al., 2005). Two recent studies revealed that intracellular LPS can trigger the Caspase-11-dependent formation of non-canonical inflammasomes by a process that does not require TLR4 (Hagar et al., 2013; Kayagaki et al., 2013). This LPS response pathway differs from the TLR4-independent responses discussed above in that it does not depend on CD14. These reports move us even further from the TLR4-centric view of LPS signaling that has so dominated the last 15 years (Beutler, 2000; Beutler et al., 2001), and suggests the existence of another sensor of LPS that is yet to be identified (Figure 1).

Inflammasomes are macromolecular protein complexes in the cytosol of macrophages, the formation of which is triggered by various pathogens and stress stimuli (Lamkanfi and Dixit, 2012). The assembly of canonical inflammasomes results in the activation of Caspase-1, which subsequently triggers two distinct signaling outcomes: First, the processing and secretion of pro-inflammatory cytokines, such as IL-1. Second, the induction of an inflammatory cell death called pyroptosis. In comparison, the activation of non-canonical inflammasomes is triggered by Caspase-11. Caspase-11 deficient mice (not Caspase-1 deficient mice) are resistant to LPS-induced sepsis, implying that Caspase-11 also participates in host signaling response to LPS (Kayagaki et al., 2011; Wang et al., 1996). However, the upstream signal leading to Caspase-11 activation remains elusive (Figure 1).

Based on the cumulative observations that various gram-negative bacteria activate Caspase-11-dependent inflammasome assembly, it has been speculated that a molecule unique to gram-negative bacteria might promote Caspase-11 activation (Aachoui et al., 2013). Following the previous finding that the B subunit of cholera toxin (CTB) and LPS could trigger the activation of Caspase-11 dependent inflammasomes (Kayagaki et al., 2011), CTB was found to act as a vehicle for delivering LPS inside the cell. Evidence supporting this idea came from the observation that LPS serotypes that could not bind to CTB were unable to induce Caspase-11 dependent cellular responses (Kayagaki et al., 2013). Such observations were further validated by direct transfection or electroporation of LPS into macrophages, which led to Caspase-11 activation (Kayagaki et al., 2013; Hagar et al., 2013). Alternative means of LPS delivery into cells also promoted Caspase-11 activation, such as coating the gram-positive bacteria Listeria monocytogenes with LPS (Hagar et al., 2013). Altogether, these results suggest that LPS is the molecule that induces Caspase-11 activation. However, how LPS “naturally” enter the cytosol during bacterial infection and triggers Caspase-11 activation is unknown (Aachoui et al., 2013). Furthermore, studies to identify upstream activators of Caspase-11 have so far been unsuccessful, but it is clear that neither CD14, MD-1, MD-2, RP105 nor TLR4 are required. Finally, it is noteworthy that Caspase-11 activation requires transcriptional priming by TLR ligands (Kayagaki et al., 2011; Kayagaki et al., 2013). During the onset of sepsis, LPS serves dual roles: it first primes Caspase-11 expression and subsequently activates cellular responses dependent on Caspase-11 activity.

While we do not yet know the identity of this new LPS receptor, studies on the types of LPS that can activate Caspase-11 suggest a similar mechanism of recognition to that of MD-2. For example, tetra-acylated LPS, which is known to activate the TLR4/MD-2 pathways poorly, is a weak activator of Caspase-11 (Hagar et al., 2013; Kayagaki et al., 2013). These results suggest a common feature of LPS signaling induced by MD-2/TLR4 and by the Caspase-11 activator. In human cells, Caspase-4 and Caspase-5 are homologues to mouse Caspase-11. It would be interesting to determine whether human macropahges responds to intracellular LPS in a manner similar to mouse macrophages, and if the answer is positive, which one of these aforementioned Caspases is required for such response.

Available data also raises the question of exactly where within cells this new LPS receptor resides and signals from. Since LPS must be delivered into cells to activate Caspase-11, it is unlikely that the sensor would be surveying the extracellular environment. Rather, this sensor is likely present within the cell. While the simplest interpretation of the current data is that the receptor is present in the cytosol, the experiments using CTB suggest an alternative. CTB is one of the best characterized protein toxins, and is well-known to enter cells and deliver its catalytically active A subunit (CTA) to the lumen of the endoplasmic reticulum (ER) (Chinnapen et al., 2007). However, cell biological analysis indicates that CTB cannot enter the cytosol (Chinnapen et al., 2007). These observations therefore raise the question of how CTB can deliver LPS to the cytosol if it cannot access the cytosol itself. It is possible that, like CTA, CTB delivers LPS to the lumen of the ER where it then releases LPS to somehow access the cytosol. An alternative is that the receptor may be an ER-resident protein, which can detect LPS in the ER lumen. A final possibility is that the LPS receptor may reside in the cytosol after all, and that in macrophages CTB does access this compartment.

Strategies taken by pathogenic bacteria to specifically evade TLR signal transduction

It is now clear that there are several LPS sensors that operate in mammalian cells to detect gram-negative bacteria (Figure 1). The collective actions of these receptors promote the cellular responses to LPS that are important for fighting infection, and the development of symptoms associated with sepsis. Based on these diverse yet interconnected LPS response pathways, it could be considered surprising that mammals ever suffer from bacterial infections. However, these potent antimicrobial mechanisms likely provided selective pressure for bacteria to evolve immune evasion strategies that evade many of our defenses. In the following sections, we provide examples of strategies used by gram-negative bacteria to evade our formidable immune defenses, with a focus on the means by which the LPS response pathways are either evaded or disrupted. We highlight bacterial strategies that likely evolved to specifically evade recognition by the host, and provide examples of bacterial survival strategies that result in an evasion of TLR signaling “accidentally” (Figure 2). This latter set of examples is intriguing, as it underscores the numerous ways in which studies of the cell biology of infection can be superimposed on studies of cell biology of innate immunity.

Figure 2. Bacterial strategies to avoid detection by the host innate immune system.

A. Evasion of host LPS sensors. Many gram-negative bacteria modify their LPS structures to evade detection by surface localized TLR4/MD-2 complex as well as the putative intracellular LPS sensor. B. Interference of host TLR signaling by bacterial type III effectors. The Yersina effector YopJ blocks activation of NF-κB and MAPK pathways by acetylating members of IKK and MKK families. The Shigella effector OspF inhibits MAPK pathway by irreversibly removing the phosphate group from the phosphothreonine residue of members of the MAPK family. C. Accidental avoidance of TLR signaling by bacterial effectors that modulate membrane trafficking. The Salmonella effector SopB dephosphorylates PI(4,5)P2 to promote invasion, which indirectly disrupts the proper localization of TIRAP, thereby abrogating TLR4 signaling from the plasma membrane. To establish a niche for intracellular replication, almost all vacuolar pathogens deliver effector proteins by various secretion systems to avoid fusion with lysosomes. An indirect outcome of such pathogen-induced detour of host membrane trafficking is the avoidance of detection by endosomal TLRs, such as TLR9.

Direct evasion of innate immune detection by alterations in the structure of LPS

As LPS is sensed directly by the TLR4/MD-2 complex, altering the structure of LPS, especially the lipid A component, is a common strategy employed by extracellular and intracellular pathogens to avoid host detection (Figure 2A). Yersinia pestis, the causative agent of plague, provides an example of how extracellular pathogens evade TLR signaling by modifying the lipid A structure. In the natural environment, Y. pestis is a rodent pathogen that can survive in the gut of fleas (Perry and Fetherston, 1997). Human infection is usually caused by flea bites. During its infectious cycle, Y. pestis adapts to different temperatures in different host species: 21-25°C in fleas and 37 °C in human or rodents. Mass spectrometry demonstrated that Y. pestis mainly produces hexa-acylated lipid A at 25°C, whereas it synthesizes tetra-acylated lipid A at 37°C (Kawahara et al., 2002). The production of hexa-acylated lipid A is correlated with the resistance to the cationic antimicrobial peptides produced by the flea innate immune system (Rebeil et al., 2004). On the contrary, the tetra-acylated lipid A is structurally similar to lipid IVa, which is non-stimulatory to the human TLR4/MD-2 complex, and is partially stimulatory to the murine counterpart, thus facilitating immune evasion by this bacterium (Montminy et al., 2006). Notably, an engineered Y. pestis strain constitutively expressing hexa-acylated LPS at 37°C stimulates potent pro-inflammatory signals in reporter cell lines and fails to establish systematic infection in wild type mice (Montminy et al., 2006). Furthermore, the protective effect is provided by the TLR4 pathway, as TLR4 knockout mice succumb when infected with this engineered strain. Besides Y. pestis, several other extracellular pathogens modify lipid A structure to avoid host detection, including Helicobacter pylori (Moran et al., 1997) and Porphyromonas gingivalis (Coats et al., 2009; Rangarajan et al., 2008).

The above-described discovery of an as-yet-unidentified intracellular LPS sensor suggests that intracellular pathogens may also be under pressure to alter the structure of their LPS. Indeed, several intracellular pathogens alter their LPS structure during infections. For example, upon entering host cells, Salmonella enterica serovar Typhimurium mediates deacylation and palmitoylation of its lipid A component by a process dependent on the bacterial two component PhoP-PhoQ system (Guo et al., 1997). These modifications decrease the stimulatory effect of S. typhimurium LPS in inducing inflammation as well as increase the bacterial resistance to antimicrobial peptides, thereby contributing to the intracellular survival of S. typhimurium (Guo et al., 1997; Guo et al., 1998). Francisella tularensis, the etiological agent of tularemia, is well-known for producing LPS with low immunostimulatory effect. When added to mononuclear cells, the amount of TNFα production induced by F. tularensis LPS is about 10³-fold weaker in comparison to that induced by E. coli LPS (Sandstrom et al., 1992). Detailed analyses revealed that F. tularensis is capable of synthesizing mono-phosphorylated and tetra-acylated lipid A (Phillips et al., 2004; Vinogradov et al., 2002). Furthermore, three of the acylated lipid side chains contain 18 carbons whereas the remaining one contains 16. These features largely account for the immunologically silent properties of F. tularensis LPS, which is pivotal for this bacterium to survive inside the host cell. Recently, it has been reported that Shigella flexneri can remodel its lipid A structure specifically during its intracellular growth phase (Paciello et al., 2013). Mass spectrometry analysis demonstrated that the majority of S. flexneri LPS purified from infected epithelial cells contained tri-or tetra-acylated lipid A. On the contrary, bacteria grown in broth produced hexa-acylated lipid A. The hypo-acylated intracellular LPS possesses weak immunostimulatory potency in that it triggers significantly lower amount of cytokine production via the TLR4 signaling pathway. Further, during S. flexneri infection, hypo-acylated LPS dampened inflammasome activation in macrophages, as indicated by the reduced release of IL-1β (Paciello et al., 2013). Overall, as could be summarized from numerous studies on the modification of LPS structure by extracellular and intracellular bacteria, an emerging theme is that alteration of LPS structure, in particular the lipid A component, represents a common virulence strategy adopted by bacterial pathogens to evade host innate immune detection. It should be noted however, that alterations in acyl chain length or number have never been found to completely prevent TLR4 signaling. These modifications only make the bacterial LPS a weak activator. Perhaps for this reason, additional bacterial immune evasion strategies have evolved.

Disruption of TLR4 signaling by bacterial Type III secretion systems.

Bacterial pathogens utilize virulence proteins and secretion systems to disrupt host signal transduction pathways (Figure 2B). Seven secretion systems have been identified in bacteria (Tseng et al., 2009). Among these, the type III secretion system (T3SS) is utilized by various gram-negative bacteria to deliver virulence proteins into the host cytosol. These virulence proteins are commonly referred to as ‘effectors’ in that they manipulate various host functions to promote bacterial invasion and survival inside the host.

YopJ is one of the first identified Yersinia type III effectors that disrupts host innate immune response by inhibiting TNFα production in infected macrophages (Palmer et al., 1998). Biochemical analyses revealed that YopJ interferes with both NF-κB and MAPK pathways by interacting with members of the IKK and MKK families (Figure 2B), which regulate signaling by TLRs and other innate immune receptors (Orth et al., 1999; Orth et al., 2000). Mass spectrometry demonstrated that YopJ is a serine/threonine acetyl transferase that mediates the attachment of an acetyl group to the cognate serine and threonine residues in the activation loop of IKKs and MKKs (Mukherjee et al., 2006). These acetylated substrates could no longer be phosphorylated by their upstream protein kinase, which in turn abrogates the signal transduction cascades induced by upstream receptors. Therefore, YopJ represents a type III effector that is dedicated to paralyzing the host innate signaling pathways. Functional and structural homologues of YopJ exist in other pathogenic bacteria (Orth et al., 2000), suggesting that the means by which this effector disrupts innate immune signaling represents a common immune evasion strategy.

Unlike YopJ, which modifies upstream molecules of the NF-κB and MAPK pathway, the Shigella type III effector OspF targets members of MAPK family by a posttranslation modification named eliminylation (Brennan and Barford, 2009). This modification cleaves the C-OP bond (whereas canonical phosphatases attacks the CO-P bond) on the phosphothreonine residue located in the aforementioned activation loop of p38 and ERK1/2, thereby irreversibly inactivating these MAPKs (Li et al., 2007; Figure 2B). Biochemically, the OspF-mediated eliminylation is distinct from dephosphorylation in that the removal of a hydroxyl group from the cognate threonine prevents it from being phosphorylated by the upstream kinases, which explains the irreversible nature of this modification. Thus, OspF is the founding member of a new bacterial phosphorylthreonine lyase family. The effect of OspF on host gene expression is profound (Arbibe et al., 2007). On one hand, an obvious outcome of OspF disruption of MAPK activity is the selective inactivation of MAPK-dependent genes. On the other hand, although it does not block NF-κB activation, OspF represses the induction of some NF-κB-dependent genes such as IL-8. Detailed analysis revealed the mechanism of this unexpected observation: OspF-mediated inactivation of ERK1/2 prevents the phosphorylation of histone H3, which negatively impinges on the chromatin reorganization events, eventually rendering the promoter binding sites of the NF-κB-dependent genes less accessible to NF-κB (Arbibe et al., 2007). In conclusion, by targeting MAPK kinases, OspF selectively modulates host gene expression to counteract host innate immune response and promote S. flexneri infection.

An apparent advantage of targeting MAPKs and IKKs is that bacterial pathogens could counteract the detection by multiple TLRs (and other receptors), as these proteins are the converging points downstream of various immune receptors and therefore serve as the master regulators of the innate immune response (Figure 2B). Compared to the ample examples of bacterial interference with signaling hub function, our knowledge of bacterial effector targeting individual TLRs is limited. This could partially be due to the limitation of the assays available to monitor the particular biochemical or cell biological activities of TLRs. In retrospect, our understanding of host innate immune receptors usually precedes that of pathogen immune evasion strategies. For instance, the discovery of the antiviral protein MAVS, which is the critical sorting adaptor in the RLR signaling pathways, led to the identification of an immune evasion mechanism used by hepatitis C virus (HCV). HCV encodes a protease called NS3/4A, which cleaves MAVS from mitochondria and mitochondria-associated membranes (Horner et al., 2011; Li et al., 2005), thus inactivating the ability of this adaptor to induce antiviral responses. Such a discovery could not have been made prior to the identification of MAVS, and the importance of its localization to mitochondria and other membranes. Analogously, we anticipate that an increased understanding of TLR function at the receptor proximal level will uncover additional bacterial immune evasion strategies that manipulate specific TLR functions.

Accidental evasion of TLR-mediated signaling responses by bacterial pathogens

The examples listed above highlighted bacterial immune evasion strategies that probably evolved to disrupt TLR-mediated recognition or signaling. In this next section, we examine how some strategies exist that likely evolved to serve some other need in the host-pathogen encounter, but resulted in the ability of the pathogen to evade TLR-mediated host defenses. In this regard, these strategies of immune evasion evolved by accident (Figure 2C).

Precise communication between different compartments of the cell depends on various membrane trafficking events. These principles also govern the operation of the innate immune signaling. As previously discussed, the compartmentalization of TLR4 signaling components determines the distinct outcome of signaling events (Kagan et al., 2008). Additionally, compartmentalization of nucleic acid sensing TLRs in endolysosomal networks represents an important mechanism for maintaining self tolerance, as the intracellular localization of these TLRs prevents the recognition of self nucleic acids by limiting receptor accessibility (Barton et al., 2006). The protease-dependent activation of endosomal TLRs serves as an additional layer of regulation of these receptors (Ewald et al., 2008). For instance, the activation of TLR9, the endosomal sensor for unmethylated CpG DNA motifs, requires the collective action of several endosomal proteases (Ewald et al., 2008; Park et al., 2008). Therefore, membrane trafficking events regulate the activity of the TLR network from several distinct aspects.

Several bacterial pathogens exist that alter membrane trafficking pathways within eukaryotic cells. The ability to alter membrane trafficking is most widely used to either promote invasion of the bacteria into normally non-phagocytic cells or to prevent delivery of the phagosomes in which they reside to lysosomes. These activities are common in bacterial pathogens that infect cells that do not express endosomal TLRs, such as epithelial cells, or even phagocytic protists. As such, the ability to either promote invasion or to block delivery to lysosomes cannot evolve to prevent delivery to a TLR-containing compartment. Rather, these aforementioned pathogenic tactics are likely evolved to facilitate the establishment of a niche permissive for bacterial intracellular growth. S. typhimurium offers examples of means by which it evade TLR signaling from the plasma membrane and from endosomes, but these evasion strategies probably did not evolve to do so.

A hallmark of S. typhimurium pathogenicity is its ability to invade non-phagocytic cells. Featured by the drastic rearrangement of host actin cytoskeleton, the invasion process is dependent on the orchestrated actions of Salmonella pathogenicity island 1 (SPI-1) effectors (Zhou and Galan, 2001). One of these effectors, SopB, is a PI phosphatase that dephosphorylates PI(4,5)P2 during infection (Hernandez et al., 2004; Terebiznik et al., 2002). Furthermore, overexpression of this effector leads to rapid depletion of PI(4,5)P2 from plasma membrane sub-domains, which changes the physical properties of the plasma membrane (Terebiznik et al., 2002) . Therefore, a direct cell biological consequence mediated by SopB during S. typhimurium infection is to promote membrane fission, which facilitates the invasion process (Terebiznik et al., 2002). Because the TLR sorting adaptor TIRAP is localized to the plasma membrane via interactions with PI(4,5)P2, it is expected that injection of SopB into mammalian cells would release TIRAP from the membrane and diminish the ability of TLR4 to promote inflammatory responses during S. typhimurium infection. Consistent with this idea, ectopic expression of SopB abrogated the plasma membrane localization of TIRAP and further disrupted the TLR4-induced NF-кB activation, by a process dependent on the PI phosphatase activity of SopB (Kagan and Medzhitov, 2006). Thus, by altering the plasma membrane PI composition, an effector protein designed to promote bacterial invasion accidentally interferes with the proper localization of the sorting adaptor TIRAP, which likely renders plasma membrane TLR4 signaling less robust (Figure 2C).

Upon entry into the host cell, S. typhimurium employs a different T3SS called SPI-2, which injects into cells effectors to establish the Salmonella containing vacuole (SCV) for its intracellular replication (Figueira and Holden, 2012). Although the SCV acidifies during maturation, this compartment is devoid of mannose-6-phosphate receptor (MPR) and its associated proteases, such as cathepsins (Drecktrah et al., 2007; Hang et al., 2006). The SPI-2 effector protein SifA is responsible for interfering with MPR trafficking. SifA associates with a host protein SKIP, which further sequesters the small GTPase Rab9 by forming a ‘hetero-trimer’ (McGourty et al., 2012). Importantly, Rab9 regulates the trafficking of MPR from the Golgi apparatus to the late endosomes under normal conditions (Barbero et al., 2002). Thus, by disrupting Rab9 function, SifA prevents the recruitment of MPR and associated cathepsins to SCV. Since cathepsins are required for the activation of TLR9, the SifA-induced alteration in membrane trafficking likely protects S. typhimurium from TLR9 detection (Figure 2C).

Strategically similar to S. typhimurium, Legionella pneumophila, the causative agent of Legionnaire's disease, manipulates host trafficking pathways to promote its survival inside professional phagocytes. Shortly after being internalized, L. pneumophila employs a type IV secretion system (T4SS) to establish the Legionella containing vacuole (LCV), an ER-like compartment that evades lysosome fusion (Isberg et al., 2009). A central characteristic of LCV maturation is the recruitment of small GTPases, such as Arf1 and Rab1, which are regulators of trafficking events between the ER and the Golgi apparatus (Kagan et al., 2004; Nagai et al., 2002a). In line with this, L. pneumophila devotes a cohort of effectors to manipulate almost every aspect of Rab1 function with distinct biochemical mechanisms which include: nucleotide exchange, nucleotide hydrolysis, reversible AMPylation and PCylation (Sherwood and Roy, 2013). A direct consequence of manipulating Rab1 function is the recruitment of host ER-derived material to the surface of LCV, which contributes to the evasion of lysosome fusion; whereas an indirect consequence of this event is the avoidance of TLR9 detection (Figure 2C). Consistent with this notion, when compared to heat-killed L. pneumophila, live L. pneumophila induced significantly less IL-12 p40 production in both macrophages and DCs (Newton et al., 2007). Furthermore, the IL-12 p40 production in response to this bacterium could be suppressed by ODN1826, a TLR9 inhibitor (Newton et al., 2007). These results confirm that L. pneumophila is capable of avoiding TLR9 detection during infection. Of note, humans are a terminal host of L. pneumophilae; the natural host of this bacterium is various amoeba species that thrive in soil or water environments, which do not possess TLRs (Swanson and Hammer, 2000). Thus, the selective pressure on L. pneumophila to avoid lysosomal degradation by its natural hosts inadvertently equipped this bacterium with the ability to avoid host intracellular TLR detection.

Conclusions and Future Perspectives

In the review, we have outlined the historical and recent discoveries of how mammalian cells detect gram-negative bacteria, with a focus on LPS sensing and microbial strategies of immune evasion. Standing from the host and bacteria interface, we are fascinated by the diverse signaling pathways and the multi-angle approaches employed by the host to detect bacteria. We also note that these comprehensive approaches of detecting bacteria likely provide selective pressure for the evolution of immune evasion strategies that promote infection.

From the host perspective, the investigation of host LPS sensing in the past two decades have been fruitful. We have obtained comprehensive knowledge with regards to the signaling networks, transcriptional regulation as well as host responses associated with TLR4. Furthermore, by studying the cell biology of TLR4, new regulatory mechanisms of innate immunity have been uncovered. First, the localization of ligand detection is usually separated from the subcellular sites of signal transduction. This observation places protein trafficking pathways at the very center of the regulation of innate immunity. Second, the subcellular sites of signal transduction often determine the nature of the signaling pathways activated. These principles apply not only to TLRs, but also to other host molecules essential for innate immune response to pathogens. Perhaps most excitedly, recent advances have demonstrated that TLR4 is not only receptor capable of activating cellular responses to LPS. For example, the surface molecule CD14 (not TLR4) is capable of activating endocytosis and the NFAT transcription factors in response to LPS. The latest discovery of the Caspase-11-dependent cell death induced by intracellular LPS further adds to the growing list of such non-transcriptional host responses to LPS. Accordingly, many new questions are emerging from these novel discoveries, some of which were highlighted in this review.

From the pathogen perspective, we have learned that successful bacterial pathogens have developed tactics to dismantle host innate immune defenses during their long co-evolution with the host. First, bacteria can modify their LPS structure to avoid detection by the host. With the discovery of the intracellular LPS sensing mechanism, we now know that such driving forces for LPS variation comes not only from the cell surface, but also from within the cell. Second, bacterial pathogens utilize secretion systems and effectors to disrupt innate immune signaling. The known targets of these virulence factors are the signaling hubs of the host innate immune signaling, such as NF-κB, or MAPK pathways. By doing so, pathogens could protect themselves from detection by multiple receptors. However, the studies on the Shigella effector OspF exemplifies the idea that bacteria do not shut down all cellular responses to LPS, but blocks a subset of responses. It is conceivable that limited induction of host inflammatory responses is either beneficial to promote bacterial dissemination in the intestinal environment or required for bacterial pathogens to compete with commensal bacteria at the site of infection. In conclusion, future studies on host and bacterial interactions in the aspect of innate immunity will continue to reveal novel host detection mechanisms and bacterial strategies to modulate host innate immune response.