The regulatory isoform rPGRP-LC resolves immune activation through ESCRT-mediated receptor clearance

Abstract

The innate immune system needs to distinguish between harmful and innocuous stimuli to adapt immune activation to the level of threat. How Drosophila mounts differential immune responses to dead and live Gram-negative bacteria using the single peptidoglycan receptor PGRP-LC is unknown. We describe an alternative splice variant of PGRP-LC, termed rPGRP-LC, which selectively dampens immune activation in response to dead bacteria. rPGRP-LC mutants cannot resolve immune activation following Gram-negative infection and die prematurely. The alternative exon of rPGRP-LC encodes an adaptor module that targets rPGRP-LC to membrane micro-domains. rPGRP-LC-mediated immune resolution requires ESCRT-dependent degradation of activating and regulatory receptors. We propose that rPGRP-LC selectively responds to peptidoglycans from dead bacteria thereby tailoring the immune response to the level of threat.

Decision-making in the immune system is based on correctly assessing the level of threat. The innate immune system in particular is triggered when host pattern recognition receptors (PRRs) encounter microbe-associated molecular patterns (MAMPs) ^{1, 2}. MAMPs are typically conserved structural motifs specific to microbial non-self and absent from the host, for example bacterial cell wall constituents such as peptidoglycan or lipopolysaccharide ³. The PRR/MAMP conceptual framework however fails to explain how the immune system distinguishes between live and dead or between beneficial and pathogenic microbes. In recent years, additional immune checkpoints have been proposed that further specify immune decision-making, such as microbial viability, virulence or proliferation ^{4, 5}. MAMPs derived from peptidoglycan for example can originate from the cell wall break-down of dying bacteria but also from bacterial remodelling enzymes active during proliferation or deployment of bacterial secretion systems ⁶. The cell wall metabolites released during the latter process are anhydromuropeptides, monomeric building blocks of peptidoglycan, such as TCT (tracheal cytotoxin, or GlcNAc-1, 6-anhydro-MurNAc-L-Ala-D-Glu-mesoDAP-D-Ala). Gram-negative bacteria with high release of TCT, such as Bordetella pertussis, are highly immunogenic, and TCT reportedly contributes to the inflammatory complications of the disease ⁶. Peptidoglycan polymers on the contrary are hidden underneath the outer LPS layer in Gram-negative bacteria, and become accessible only upon bacterial death since the outer membrane is impermeable to large PGN fragments ⁷.

An intriguing question is therefore how PRRs integrate information from MAMPs with different “threat content” into a coherent immune response. In addition to being specific, the immune response also needs to be self-limiting to avoid collateral damage to the host ⁸. Multiple receptor-activated signalling pathways, including PRR signalling upstream of NF-κB activation, are fine-tuned by receptor endocytosis ⁹. PRR signalling is governed by sorting/signalling adaptors and results in functionally different outcomes depending on subcellular localization of signalling complexes. This is best exemplified by toll-like receptor 4 (TLR4) signalling, where phosphoinositide-binding adaptors such as MyD88, TRIF and TRAM dictate spatial aspects of signalling and shape immune activation and regulation ^{10, 11}.

Drosophila can engage two pathways to activate NF-κB: the Toll pathway is selectively and primarily activated by fungal and Gram-positive infections, while the IMD pathway mainly responds to Gram-negative infections. The IMD pathway is triggered by two peptidoglycan recognition receptors, plasma membrane PGRP-LC or cytosolic PGRP-LE, and activates the NF-κB-like transcription factor Relish ¹². PGRP-LC senses polymeric and monomeric breakdown products of DAP-type peptidoglycan (PGN) from Gram-negative bacteria and Gram-positive bacilli ^{13, 14, 15}. It is thought that signal relay by PGRP follows an induced-proximity model, where ligand-induced clustering of receptor tails in the cytosol recruits the signalling adaptor IMD and activates NF-κB signalling ¹⁶. PGRP-LC encodes three isoforms with distinct ectodomains that differ in their peptidoglycan-binding capacity ^{13, 17, 18}. Molecularly, detection of polymeric PGN (a breakdown product of dead bacteria) relies on homotypic clusters of PGRP-LCx receptor isoforms alone, while detection of the monomeric muropeptide TCT (a metabolite of bacterial cell wall remodelling) is dependent on LCx/LCa receptor heterodimers ^{13, 17, 19}. While LCx contains a full PGN/TCT docking grove, LCa merely binds to the TCT monomer presented by LCx ²⁰.

Intriguingly, although PGN and TCT activate the same signalling machinery, their immune activation and resolution kinetics differ. In particular, innocuous polymeric PGN induces a more efficient immune resolution than TCT ^{17, 21}. The underlying molecular mechanism remains unknown, but likely involves regulation of IMD pathway signalling. Peptidoglycan-triggered activation of the IMD pathway in Drosophila is regulated at multiple steps. Secreted enzymatic PGRPs such as PGRP-LB and PGRP-SC can hydrolyse polymeric and monomeric PGN into non-immunogenic fragments, reducing signal input into the pathway ^{21, 22, 23}. PGRP-LF is a transmembrane receptor with two PGRP domains and a non-signalling cytosolic tail that inhibits PGRP-LC activation through receptor competition. PGRP-LF interacts with the same affinity as PGRP-LCa with empty or TCT-occupied PGRP-LCx, but unlike LCa/LCx complexes, LF/LCx complexes cannot signal ^{24, 25}. A cytosolic adaptor, Pirk (also called Rudra or Pims), has been proposed to disassemble receptor-signalosomes by competing with IMD for PGRP-LC occupancy or by depleting PGRP-LC from the plasma membrane ^{26, 27, 28}. As PGRP-LF, PGRP-LB and Pirk are themselves regulated by the IMD pathway these negative regulators establish a feedback loop preventing the excessive activation of the immune system. However, based on their mode of action, none of the above regulators is able to distinguish between polymeric PGN and TCT and to tailor the immune response depending on the exposure to live or dead bacteria. In the present study, we identify and characterize a regulatory PGRP-LC isoform that carries a distinct cytosolic tail predicted to bind phosphoinositides and whose specific recruitment to polymeric peptidoglycan contributes to the resolution phase of the immune response.

Results

The PGRP-LC locus encodes an additional previously unrecognized isoform with an alternative cytosolic tail

The kinetics of IMD pathway activation in response to Gram-negative bacteria can be assessed by monitoring the expression of the antibacterial peptide gene Diptericin, a target of the NF-κB-like transcription factor Relish. Drosophila mounts differential responses to identical concentrations of live and dead Gram-negative bacteria, with dead bacteria inducing a significantly faster return to baseline (Fig. 1a). Injection of different doses of TCT triggers the IMD pathway with a kinetic similar to live bacteria, consistent with TCT acting as a MAMP that signals the presence of live bacteria (Fig. 1b). On the contrary, immune responses to different doses of whole PGN (a preparation composed of mostly polymeric PGN) mimic those to dead bacteria. The rate of resolution appears significantly faster for PGN than for TCT at any dose, suggesting a qualitative rather than quantitative difference in the resolution mechanism. Since both PGN and TCT engage IMD pathway adaptors through PGRP-LC, we hypothesized that PGRP-LC signalling might harbour a discriminatory mechanism to distinguish between PGN and TCT.

Figure 1. The PGRP-LC locus encodes cytosolic tail variant isoforms with regulatory potential.

a. w¹¹¹⁸ flies display differential Dpt mRNA expression kinetics when infected with live Ecc15 or the same bacterial pellet after heat-inactivation. Data shows mean ± SEM of n=3 repeats.

b. As in a. after injection of varying concentrations of synthetic monomeric peptidoglycan (TCT) and polymeric peptidoglycan from E.coli (PGN). The molarity of PGN is determined based on titration of its hydrolysed component parts (e.g. muramic acid, DAP) and refers to monomeric units (1 GlcNac-MurNac-tetrapeptide ~ 1000Da) in the PGN polymers (personal communication, Dominique Mengin-Lecreulx). Due to steric hindrance between PGRP domains, only every fifth monomeric unit can be bound in PGN 66, lowering the effective binding site concentration 5-fold. Data shows mean ± SEM of n=3 repeats. Resolution after PGN occurs faster than after TCT (difference in normalized slopes PGN vs TCT, 8h vs 24h by linear regression, P=0.0027779).

c. PGRP-LC genomic locus, exon structure (only protein-coding exons are numbered) and resulting polypeptides. Exon 4 (blue) encodes the known IMD-activating cytosolic tail; exon 5 (orange) encodes the cytosolic tail characterised hereafter. The coding sequence of eGFP (green) was inserted at the transcription start site in exon 4. Both exons 4 and 5 can be spliced to x, a and y ectodomains (greys), resulting in two sets of polypeptides with differing cytosolic tails and a complete range of peptidoglycan sensing domains (see also Supplementary Fig. 1). To create exon 5 deletion loci, exon 5 was removed as described in Experimental Procedures (deletion indicated by brackets), leaving the intronic sequences on either side intact. A schematic representation (not to scale) depicts domain architecture in the cytosolic tails of PGRP-LC and rPGRP-LC proteins. Exon 4 (blue) encodes the activating tail with an IMD binding domain and a RHIM motif (hatched). The domain encoded by exon 5 (orange) is characterised hereafter. Exon 6 (grey) is present in both isoforms and encodes the transmembrane (TM) domain (hatched). Note that both rescue constructs carry an N-terminal GFP tag on exon 4-containing PGRP-LC isoforms that was previously shown not to affect LC function ¹⁷.

d. Tissue expression of rLCx quantified for adult fat bodies, whole guts or midguts dissected from unchallenged w¹¹¹⁸ flies (see also Supplementary Fig. 3d).

e. LCx and rLCx show similar expression kinetics in c564-Gal4/+ flies infected with heat-killed Ecc15. Data in d and e are mean ± SEM of n≥3 repeats.

f. rLC is conserved in all sequenced Drosophilid species except D. willistoni. Protein sequences (corresponding to aa 1 to 97 in the D. melanogaster sequence) of 11 sequenced Drosophilid species were aligned using ClustalW. Orange blocks denote conserved residues defining predicted sequence elements (C4HC3 zinc finger, amphipathic α-helical coiled-coil, see also Supplementary Fig. 1).

g. Fat-body specific overexpression (driver: c564-Gal4) of rLCx selectively blunts immune responses to polymeric PGN but not to the monomer TCT. Injections as in b.

h. rLCx does not blunt IMD activation in the absence of bacterial ligands. c564-Gal4 or c564-Gal4;UAS-rLCx driver lines were crossed to UAS-PGRP-LCx, UAS-PGRP-LE or UAS-Imd. Dpt mRNA was quantified 3 days after flies were switched to 29°C for gene induction. Data are mean ± SEM of n≥3 repeats, analysed by 2-way ANOVA with Bonferroni posttests (a, g, h).

PGRP-LC is a type II transmembrane receptor, whose N-terminal cytosolic tail is identical in the three PGRP-LC isoforms LCa, LCx and LCy ^{13, 29, 30}. This cytosolic tail lacks any obvious endocytic motifs, and no ligand-induced trafficking of the receptor has been reported so far. A closer look at the PGRP-LC locus revealed a potential alternative first coding exon, which might encode a functionally distinct cytosolic tail (Fig. 1c). We termed the cytosolic tail variant ‘regulatory PGRP-LC’ (rPGRP-LC). The putative and previously unrecognized alternative exon 5 is present in the genomes of 11 out of 12 sequenced Drosophila species, with the exception of D. willistoni ³¹. Exon 5 contains both a Kozak sequence upstream of the translation start codon and a splice donor site at its 3’ end (Supplementary Fig. 1a). Similarly to the conventional exon 4, exon 5 appears to be alternatively spliced to exons encoding the transmembrane domain and the PGRP domains x, y and a (Supplementary Fig. 1b, c). This indicates that the PGRP-LC locus gives rise not to three, as previously described, but to six receptor isoforms, three PGRP-LCs and three rPGRP-LCs (Fig. 1c). Exon 4- and exon 5-containing transcripts exhibit similar tissue specificity and expression levels (Fig. 1d), and comparable expression kinetics after immune challenge (Fig. 1e), suggesting shared transcriptional control through a RELISH-dependent feedback loop.

Structure-function analysis of rPGRP-LC

To identify conserved regions in exon 5, we aligned rPGRP-LC (hereafter referred to as rLC) protein sequences from 11 Drosophila species and found two distinct conserved domains separated by a short variable linker (Fig. 1f). In silico analysis of amino acids 1-65 predicts a plant homeodomain (PHD)-type zinc finger based on a series of highly conserved cysteines arranged in a characteristic C4HC3 pattern (Fig. 1f and Supplementary Fig. 2a). PHD domains belong to a larger family of FYVE and RING zinc fingers, protein folds acting in protein-lipid interaction and ubiquitination, respectively ^{32, 33}. The rLC PHD domain clusters with known lipid-binding adaptors, predicting a role in membrane targeting (Supplementary Fig. 2b) The PHD domain is followed by a short linker region of lower conservation. Although the linker somewhat resembles a RIP homotypic interaction motif (RHIM) in species of the melanogaster subgroup, this is not substantiated by detailed computational analysis (K. Hofmann, A. Kajava, personal communication). Beyond the linker, amino acids 76 to 97 form an amphipathic α-helix that is predicted to form coiled-coils (Supplementary Fig. 2c, d). Circular dichroism analysis of the cytosolic region of rLC support these structural predictions (data not shown). Taken together, the domain architecture of the cytosolic region of rPGRP-LC might suggest that this portion mediates lipid binding, protein-protein interactions, and/or homo- or heterodimerization.

rLC specifically modulates IMD activation in response to polymeric PGN

Given the structural features of rPGRP-LC, we tested whether rLC might play a role in correctly adjusting immune responses to different ligands. Fat body-specific overexpression of rLCx potently reduced the immune response to polymeric PGN, but failed to significantly suppress TCT-induced activation of the NF-κB target gene Diptericin (Fig. 1g). This indicates that rLCx selectively modulates the response to specific ligands. Next, we tested the ability of rLC to modulate NF-κB activation triggered by ectopic expression of PGRP-LCx. In contrast to the ability of rLC to modulate the response to specific ligands, overexpression of rLCx had no effect on NF-κB activation triggered by overexpression of PGRP-LCx, PGRP-LE or IMD (Fig. 1h). This is not entirely surprising as overexpression of LC isoforms activates IMD signalling through concentration-dependent clustering of cytosolic tails ^{16, 29}. The notion that rLC fails to suppress this mode of activation suggests that rLC associates with LCx only in the presence of polymeric ligands, such as PGN. According to this scenario the polymeric nature of PGN tethers rLC to LCx. Accordingly, proper resolution of IMD signalling in response to PGN requires scaffolding of receptors by ligands with multiple receptor binding sites. This model also provides some explanation as to why overexpressed rLCx fails to modulate the response to monomeric TCT.

To study the function of the new rLC isoforms in more detail, we first made use of available mutants, ΔLC^E12 flies lacking the entire PGRP-LC locus, and ΔLC^ird7(1) flies with an insertion in exon 4 which consequently express only the rLC isoforms (Fig. 1c and Fig. 2c) ^{29, 30}. Both mutants are unable to induce antibacterial peptide gene expression or survive Gram-negative infection (Fig. 2d, e), indicating that the tail variant receptor rLC alone cannot activate the IMD pathway. We then generated a PGRP-LC locus specifically deleted for the rLC isoforms. For this we complemented the ΔLC^E12 deficiency with either an engineered transgene P[acman]-PGRP-LC containing the entire wild-type locus ¹⁷ or with P[acman]-PGRP-LC comprising the same locus deleted for exon 5 (Fig. 1c). Both rescue transgenes were inserted into the same genomic location, at 51C on the second chromosome (Fig. 2a) and crossed to either ΔLC^E12 or ΔLC^ird7(1) mutant flies (Fig. 2b). Expression of exon4-containing LC isoforms remained comparable in both engineered loci (Supplementary Fig.3a). The two ΔLC^E12 complemented lines are referred to as resc[LCwt]) and resc[LCΔex5] (extended genotypes w;P[acman]-GFP::PGRP-LC;ΔLC^E12 and w;P[acman]-GFP::PGRP-LCΔexon5;ΔLC^E12, respectively). Both resc[LCwt] and resc[LCΔex5] lines were viable, albeit with a slight but non-significant decrease in reproductive output in the rLC loss-of-function line (Supplementary Fig. 3b). This is in contrast to the loss-of-function phenotype of another regulatory receptor, PGRP-LF, which is homozygous lethal during development ²⁵, and may indicate mechanistic differences between PGRP-LF- and rLC-mediated regulation.

Figure 2. rLC is a ligand-specific regulator of IMD pathway activation.

a. Genomic rescue of PGRP-LC mutants (chromosome III) by reinsertion of Pacman constructs carrying the full PGRP-LC locus (resc[LCwt]) or the locus lacking rLC (resc[LCΔex5]) into the ϕC31 landing site at position 51C on chromosome II.

b. Genotypes and phenotypes of rescued flies used in this study. Different rescue constructs on chromosome II were crossed to different PGRP-LC mutants. See also Suppl. Fig. 3c for immune response in different genetic contexts.

c. All rLC isoforms are absent from ΔLC^E12 but transcribed in ΔLC^ird7(1). In ΔLC^ird7(1) an insertional mutation in exon 4 introduces a premature stop codon in LC but not rLC transcripts.

d. rLC alone fails to induce an immune response 8h after Ecc15 infection.

e. rLC alone cannot confer resistance to infection with Ecc15. Data represent pooled survival counts of n=3 biological repeats. Log-rank test, p<0.0001 for mutants versus wt.

f. Loss of rLC results in significantly enhanced immune responses to PGN at 8h and 24h but no change in response to TCT. Injections as in Fig. 1b. Data are mean ± SEM of n=3, analysed by 2-way ANOVA and Bonferroni posttests, comparing resc(LCwt) and resc(LCΔex5).

g. PGN-elicited responses (8h after injection) require rLCx for down-regulation; TCT-elicited responses (24h after injection) are unaffected by loss of rLCx or rLCa alone or in combination (presence of isoform is indicate by a +; see Supplementary Table 3 for full genotypes). Data represent mean ± SEM of n≥3, analysed by 1-way ANOVA and Tukey’s posttests (b) or Dunnett’s posttests (e).

In line with the rLC-overexpression phenotype, resc[LCΔex5] flies that lack rLC isoforms displayed significantly stronger IMD pathway activation after injection of polymeric PGN, while the response to TCT was unchanged (Fig. 2f). To dissect the role of rLC isoforms in more detail, we produced engineered flies carrying all possible permutations of activating isoforms LCx and LCa and regulatory isoforms rLCx and rLCa and monitored the immune response to polymeric and monomeric PGN (Fig. 2g). As previously published ^{17, 34}, LCx is necessary and sufficient to relay signal after polymeric PGN injections. Moreover, the response to PGN was significantly enhanced in flies carrying LCx and lacking rLCx, regardless of the presence of LCa and rLCa isoforms. This indicates that rLCx is necessary and sufficient to resolve IMD activation in the presence of PGN (see also Supplementary Fig. 3c). After TCT monomer injections, the heterodimeric LCx/LCa module was required to relay signal, in agreement with previous work. However, no significant increase was observed when either rLCx or rLCa or both were missing, indicating that TCT signalling is mostly resistant to negative regulation by rLC.

Additional experiments monitoring immune activation in the whole gut (IMD signalling dependent on PGRP-LC and PGRP-LE) compared to midgut (IMD signalling dependent on PGRP-LE alone) showed that the regulatory isoforms exclusively blunt PGRP-LC-activated baseline IMD signalling in response to indigenous gut bacteria and have no effect in the midgut where PGRP-LE plays a predominant role ^{17, 35} (Supplementary Fig. 3d) Moreover, steady-state immune signalling in the fat body, in the absence of infection, was not significantly affected by loss of rLC (Supplementary Fig. 3d). Collectively, our observations demonstrate that rLC isoforms function as negative regulators of PGRP-LC but not PGRP-LE activated IMD signalling. In contrast to PGRP-LF or the nonaspanins TM9SF2/4, which prevent PGRP-LC-mediated IMD activation even in the absence of stimulus ^{24, 25, 36}, regulation by rLC is ligand-dependent. Most notably, rLC isoforms seem able to differentially affect the immune response to polymeric and monomeric PGN, i.e. to differentiate between varying levels of threat.

rLC is required to resolve IMD activation during bacterial infection

In PRR/MAMP terminology, polymeric PGN is a post-mortem MAMP ⁵, which notifies efficient bacterial killing. The regulatory specificity of rLC indicates a potential role in resolving IMD activation in the presence of polymeric PGN. We therefore asked whether rLC can specifically curtail immune signalling when the immune system is facing dead bacteria. Flies lacking rLC isoforms kill bacteria as efficiently as wild-type flies, with estimated killing rates of 280 ± 60 bacteria/h (p = 1.05e-05) and 313 ± 88 bacteria/h (p = 0.000615) for resc[LCwt] and resc[LCΔex5] flies, respectively (Fig. 3a). Consistent with the notion that rLC is required for the resolution of IMD signalling following an infection, we find that flies lacking rLC isoforms were significantly delayed in resolving their immune response, despite efficient bacterial killing (Fig. 3b). The delay in resolving IMD signalling was not due to the persistence of live bacteria as the same effect was also observed after challenge with heat-killed bacteria (Fig. 3b and Supplementary Fig. 3e). Therefore, we conclude that rLC contributes to the resolution of immune activation by sensing the presence of polymeric PGN from dead bacteria.

Figure 3. The rLC regulatory isoform resolves IMD pathway activation after infection.

a. Bacterial clearance (live Ecc15-GFP) is unaffected in flies lacking rLC. Fly lysates were plated for colony counts at the indicated times. Data (n=4 biological repeats) is represented as box plots (Tukey). Linear regression analysis gives clearance rates of 217 ± 114 bacteria/h (wt, ns), 280 ± 60 bacteria/h (resc[LCwt], p = 1.05e-05) and 313 ± 88 bacteria/h (resc[LCΔex5], p= 0.000615).

b. Loss of rLC increases the amplitude and duration of IMD activation after Gram-negative challenge. resc(LCwt) and resc(LCΔex5) flies were infected by pricking with live (left panel) or heat-killed (right panel) Ecc15 or heat-killed E.coli (Supplementary Fig. 3e) and collected at indicated time points for quantification of Dpt mRNA.

c. Despite an efficient immune response flies lacking rLC die faster than wt flies when infected with live Ecc15. Data represent pooled survival counts of n=3 biological repeats. Log-rank test, p<0.0001 for resc(LCwt) versus resc(LCΔex5).

d. rLC overexpression reduces the amplitude and duration of the IMD response, as measured by Dpt mRNA expression in flies with or without fat body-specific (c564-Gal4) overexpression infected with heat-killed Ecc15. Data represent mean ± SEM of n≥3 repeats, analysed by 2-way ANOVA with Bonferroni posttests (b, e).

Despite efficient bacterial clearance, rLC mutants died significantly earlier than wild-type controls (Fig. 3c). Although this appears somewhat paradoxical, this observation is in line with previous findings indicating that deregulated IMD pathway activation can result in premature death ^{23, 25, 37}. Alternatively, we cannot exclude that even small numbers of bacteria could flare up and kill the host. While loss of rLC function results in elevated IMD signalling, we found that overexpression of rLCx in the fat body significantly blunted the immune response after challenge with heat-killed Gram-negative bacteria (Fig. 3d). Accordingly, rLCx-overexpressing flies had a significantly lower survival rate to live bacteria, possibly because they down-regulated IMD activation too early (Supplementary Fig. 3f). Of note, fat-body specific expression of rLCx did not change the base-line levels of Dpt expression per se (Fig. 3d, time-point 0). Collectively, these data identify rLCx as a novel regulatory isoform of PGRP-LC that controls IMD pathway kinetics, thereby preventing immune-induced lethality from unresolved IMD activation.

rLC interacts with Pirk through the PHD domain

We next investigated the mechanism by which rLC downregulates IMD pathway activity. First we assessed the possibility that rLC regulates the availability of activating receptors on the cell surface. To this end we evaluated the levels of endogenous GFP-tagged PGRP-LC in the presence and absence of rLC by comparing homozygous resc(LCwt) and resc(LCΔex5) flies. Intriguingly, we found that PGRP-LC significantly accumulated in rLC-deficient flies when compared to wild-type flies after challenge with dead bacteria. This suggests that rLC controls activating receptor levels (Fig. 4a, quantification on the right). Pirk, a negative regulator of the IMD pathway, has previously been reported to down-regulate the surface availability of PGRP-LC ^{26, 27, 28}. We therefore tested whether rLC interacted with Pirk, and whether this interaction is required for rLC- and Pirk-mediated down-regulation of IMD signalling. We found that Pirk co-immunoprecipitated LCx and rLCx from cellular extracts (Fig. 4b). Strikingly, the PHD domain of rLCx was indispensable for its interaction with Pirk since a PHD domain-deficient form of rLCx failed to bind to Pirk. This demonstrates that the PHD domain of rLCx is required for the binding of rLC to the negative regulator Pirk. Since Pirk can bind to different motifs in LC and rLC (RHIM and additional cytosolic motifs for LCx ^{26, 27, 28}, PHD for rLCx, this study), it is likely that Pirk mediates its effect by binding to either receptor.

Figure 4. rLC interacts with the negative regulator Pirk.

a. Endogenous GFP-PGRP-LC accumulates after infection in flies lacking rLC. resc(LCwt) and resc(LCΔex5) flies were infected as in b. and fat bodies were dissected at 0, 1, 2 and 3 h after challenge. Relish cleavage was monitored as a read-out of IMD activation.

b. Pirk interacts with the PHD domain of rLC. Indicated constructs were transfected into S2 cells and HA-tagged Pirk was purified with anti-HA antibodies. Expression and presence of co-purified rLC proteins was analysed by immunoblotting with anti-V5 antibody. Cell lysates are indicated in the left panel, and IP-ed proteins are indicated in the right panel.

c. Loss of rLC and pirk are additive. Flies deficient for either pirk or rLC or both were assessed for Dpt mRNA 8h after challenge with heat-killed Ecc15.

d. Overexpression of rLC or pirk reduces the immune response in offspring from w;c564-Gal4 or w;c564-Gal4;UAS-rLC driver lines crossed to w or UAS-pirk and challenged with heat-killed Ecc15.

e. Overexpression of rLC can compensate for loss of pirk. Flies with or without fat body-specific rLC overexpression in a pirk mutant background were tested for immune activation after infection with Ecc15 (see also Supplementary Fig. 4a, b).

f. Overexpression of Pirk does not discriminate between polymeric PGN (left panel) and the TCT monomer (right panel). Data represent mean + SEM of n≥3 repeats, analysed by 1-way ANOVA with Bonferroni (c, e) or Tukey’s posttests (d), or 2-way ANOVA with Bonferroni posttests (f).

Next, we assessed whether rLC and Pirk cooperate in regulating IMD pathway activity upon bacterial infection. Loss of either pirk or rLC alone significantly increased the immune response, and combined loss of both was linearly additive (Fig. 4c). Conversely, overexpression of either regulator reduced the immune response by about 50% and concomitant expression of rLC and pirk led to over 60% decrease, albeit with no significant additive effect (Fig. 4d and Supplementary Fig. 4a). Importantly, Pirk suppressed pathway activation stimulated by either polymeric PGN or monomeric TCT, indicating that Pirk does discriminate between live or dead bacteria. This indicates that Pirk operates differently to rLC, with rLC exclusively regulating the response to dead bacteria (Fig. 4f). Lastly, we asked whether rLC relied on Pirk for its regulatory function. Overexpressed rLCx was able to suppress IMD signalling in the absence of pirk, demonstrating that Pirk is not required for rLC-mediated regulation of IMD signalling (Fig. 4e and Supplementary Fig. 4b).

Taken together, these data indicate that despite their physical interaction, Pirk and rLC can regulate IMD activation independently of one another. Nonetheless, under physiological conditions it is highly likely that Pirk and rLC work together to suppress IMD signalling. PGN likely induces clustering of LC and rLC through interaction with the PGRP ectodomains of either receptor ¹⁹. In this scenario, optimal regulation might be achieved by multi-modular interactions between extracellular and intracellular contacts. According to this scenario, polymeric PGN binds to the extracellular portions of LC and rLC leading to their clustering. Subsequently, this allows binding of Pirk to the PhD domain of rLC and RHIM domain of LC, respectively. Together, this will enforce proper negative regulation PGRP-LC signalling.

Endocytic mechanisms regulate the IMD pathway

The surface availability and signalling capacity of transmembrane receptors is frequently regulated via endocytosis and ESCRT-mediated trafficking. Several lines of evidence suggest an involvement of endocytosis in rLC-mediated regulation of IMD signalling: (1) the cytosolic tail of rLC contains a PHD motif that is predicted to interact with membrane phospholipids, (2) rLC interacts with Pirk, a regulator previously proposed to relocalize PGRP-LC from the plasma membrane, and (3) loss of rLC leads to accumulation of PGRP-LC. To investigate the role of endocytic trafficking in the activation and the regulation of IMD signalling, we performed a small targeted screen of key endocytic trafficking components in the fat body, the main immune responsive organ of the fly (Fig. 5a, Supplementary Tables 3 and 4). While none of the tested genes prevented IMD pathway activation, interfering with the early endosomal compartment by overexpression of dominant-negative (DN) Rab5 (Rab5^DN) or Rab5 RNA interference (RNAi) increased the amplitude and duration of IMD pathway activation. In addition, knock-down of Fab1 also led to increased activation, but had no effect on the return to baseline. Lastly, disruption of the ESCRT-I complex by knock-down of vps28 (and to a lesser, non-significant degree vps23/tsg101 and the ESCRT-II component vps25), extended the duration of IMD pathway activation. Interestingly, IMD pathway activation seemed largely clathrin- and dynamin-independent, since neither overexpression of dominant-negative shibire (shi^DN), the fly homolog of dynamin, nor the knock-down of shibire or clathrin light chain (Clc) prevented Dpt induction upon Gram-negative infection.

Figure 5. Endocytic mechanisms modulate IMD pathway kinetics and rLC localization.

a. Targeted mis-expression screen for endocytic modifiers of IMD pathway activation. UAS-gene or UAS-RNAi-gene lines (see Supplementary Tables 3 and 4) were crossed to the temperature-controlled, fat body-specific driver line w;c564-Gal4;tub-Gal80^ts and adult progeny was tested for Dpt mRNA levels 8h (peak activation) and 24h (resolution) after infection. Data are mean + SEM of n≥3 repeats analysed by 2-way ANOVA with Bonferroni posttests, showing Dpt/RpL32 mRNA ratios normalized to maximal IMD activation of control line (black) at 8h after infection. Control crosses (mis-expression of PGRP-LCx) are in blue, significant hits are in red.

b. GFP-rLCx localizes to discrete cell membrane domains and to intracellular vesicles in adult fat body cells. Panels b to e show overexpression of rLCx constructs using the c564-Gal4 driver line. b is a section through the apical region of cells (facing hemolymph), b’ is a section through the cell body. Same annotation for panels c and c’. An open arrowhead indicates the location of the orthogonal section in panels a’, c’, d” and e”.

c. GFP-rLCxΔPHD distributes uniformly along the plasma membrane in fat body cells.

d. GFP-rLCx is enriched in the cortical regions of fat body cells as visualized by phalloidin staining of cortical actin. d shows green channel, d’ shows magenta channel and d” shows merged channels. Same annotation for e, e’ and e”.

e. GFP-rLCx partially overlaps with adherens junction marker β-catenin, as visualized by anti-armadillo staining. Full arrowheads indicate co-localization of pixels in green and magenta channels (apparent as white on the merged panels).

f. Hits from the screen in a. affect GFP-rLCx protein stability. c564-Gal4;UAS-GFP-rLCx flies were crossed to indicated genotypes and progeny was collected for Western Blot 3 days after induction at 29°C. Genotypes are loaded in duplicates, with each sample resulting from an independent cross.

See Suppl. Fig. 5d for quantification.

g. GFP-rLCx localizes to apical vesicles in wild-type fat bodies. A sum projection of 3 consecutive apical slices is shown in g, h and i. GFP channel is shown in white.

h. Pirk overexpression increases GFP-rLCx-positive vesicle number.

i. vps28 knock-down leads to enlarged GFP-rLCx-positive vesicles with visible lumen (arrows).

j. Quantification of GFP-rLCx vesicle distribution for indicated genotypes. See also Supplementary Fig. 5g for quantification of total cell area. Shown are box plots with 10^th and 90^th percentiles, with number of cells analysed above each box. Data were analysed by 1-way ANOVA with Dunnett’s post test. Scale bar in all images, 10 μm. Nuclear staining is DAPI (blue).

Rab GTPases shuttle between the cytosol and endomembrane compartments, controlled in part by their differential specificity for membrane phosphoinositides. By acting as molecular adaptors, Rabs give functional specificity to endomembrane compartments and regulate vesicle formation, transport and fusion in exo- and endocytosis. Rab5 in particular has a well-established role in the maturation of early endosomes in vertebrates and invertebrates ^{38, 39}. The phosphatidylinositol 3-phosphate 5-kinase Fab1 has been shown to affect receptor trafficking in the degradative pathway, but does not affect signal termination of endocytosed receptors ⁴⁰. ESCRT proteins form sequential complexes (ESCRT-0 to –III) to traffic ubiquitinated cargo to specific destinations. Frequently this promotes inactivation and removal of receptors via multivesicular bodies (MVBs), followed by lysosomal degradation ⁴¹. Our observation that inactivation of Rab5, Fab1, and ESCRT components results in deregulated IMD signalling suggests that accurate termination of IMD signalling requires the endocytic machinery. The involvement of ESCRT machinery in particular suggests that termination of IMD pathway signalling requires ubiquitin-dependent removal and lysosomal degradation of a membrane-bound receptor. Our data also indicate that bacterial detection and initiation of PGRP-LC-mediated signalling does not require endocytosis and starts at the plasma membrane.

Subcellular localization of rLCx depends on the PHD domain

Given the role of endocytosis in IMD pathway resolution, we tested whether rLCx undergoes endocytosis. To this end, we visualized the subcellular localization of rLC in the adult fat body, using overexpression of N- and C-terminally tagged versions of rLCx (Supplementary Fig. 5a). Under the same conditions, overexpression of GFP-PGRP-LCx caused cell death, precluding the analysis of its trafficking routes (data not shown.) As controls, we overexpressed known lipid-binding probes tagged with GFP (Supplementary Fig. 5b). In adult fat body cells, full-length GFP-rLCx localized to discontinuous membrane domains and to perinuclear vesicles, partially mimicking the distribution of the endosomal FYVE-GFP probe (Fig. 5b, compare to Supplementary Fig. 5b). A similar distribution occurred in hemocytes, where GFP-rLCx labelled plasma membrane and intracellular vesicles, and was particularly enriched in cell-cell contact sites (Supplementary Fig. 5c). GFP-rLCx lacking the PHD domain (GFP-rLCxΔPHD) distributed more uniformly along the plasma membrane, reminiscent of the plasma membrane localizing probe GFP-PLCδ-PH (Fig. 5c, compare to Supplementary Fig. 5b), suggesting that the PHD domain is involved in receptor localization to specific membrane sub-domains. In fat body cells, the GFP-rLCx-positive membrane domains partially overlapped with apico-lateral regions labelled by cortical actin and β-catenin, a component of adherens junctions (Fig. 5d, e). Taken together, these observations suggest that rLCx is retained in specific membrane domains in a PHD domain-dependent fashion.

Next, we tested the role of endocytosis in regulating the localisation and function of rLCx. We also included Pirk in this study. Loss of Rab5 is known to preclude formation of endosomes and block the maturation of all downstream endosomal compartments ⁴². Loss of ESCRT machinery generally leads to accumulation of ubiquitinated cargo in endomembranes or alters receptor recycling back to the plasma membrane ^{43, 44}. Intriguingly, we found that GFP-rLCx accumulated in flies in which Rab5 was knocked-down (Fig. 5f and Supplementary Fig. 5d). Likewise, depletion of ESCRT components Vps28 and Tsg101 similarly resulted in accumulation of rLCx. Conversely, overexpression of Pirk or vps28 reduced total GFP-rLCx protein. Under the same conditions, GFP-rLCx positive vesicles significantly increased in number (Fig. 5g, h and j), suggesting enhanced endocytosis. While overexpression of Pirk and Vps28 enhanced the appearance of GFP-rLCx positive vesicles, depletion of components of the ESCRT machinery led to a decrease in the numbers of such vesicles (Fig. 5g, i and j, and Supplementary Fig. 5e, f, g). These observations are in agreement with the notion that rLCx undergoes endocytic degradation via the ESCRT machinery, a process that is aided by Pirk.

Failure to endocytose rLC results in overactivation of IMD signalling

Given that PGRP-LC accumulated in rLC-deficient flies, we next tested the interdependency of rLC and ESCRT machinery in regulating levels of PGRP-LC. Due to genetic constraints, we used flies heterozygous for the rescue constructs (i.e. carrying a single copy of the rescue construct). Endogenous GFP-PGRP-LC accumulated in an rLC- and Vps28-dependent manner in the fat bodies of flies challenged with dead bacteria (Fig. 6a, b). Immediately after infection, Relish is cleaved to release the transcriptionally active fragment Rel-49. Relish cleavage was not significantly affected by lack of rLC or Vps28, indicating that early activation of the IMD response is normal (Fig. 6a,b). At the transcriptional level, accumulation of PGRP-LC was reflected by significantly delayed immune resolution when vps28 was knocked down, alone or in combination with loss of the single copy of rLC (Fig. 6c). Based on this observation, it is likely that both rLC and Vps28 act in concert to remove PGRP-LC from the plasma membrane. Indeed, the regulatory function of overexpressed rLCx was significantly impaired in a vps28 knock-down background (Fig. 6d and Supplementary Fig. 5h), suggesting that rLC acts upstream of Vps28. Interestingly, PGRP-LC was previously shown to interact with the ESCRT protein Vps28 in an S2 cell-based protein interaction network ⁴⁵.

Figure 6. rLC and ESCRT synergize to degrade activating PGRP-LC and resolve IMD pathway activation.

a. Loss of rLCx, knock-down of vps28 and a combination of both lead to accumulation of endogenous GFP-PGRP-LC during infection. Heterozygous resc(LCwt) or resc(LCΔex5) flies carrying a c564-Gal4 driver were crossed to ΔLC^E12 or vps28-RNAi,ΔLC^E12 flies. Offsprings were challenged with heat-killed Ecc15 and fat bodies were collected for Western Blot at indicated time points. Relish cleavage is shown as a read-out of IMD pathway activation. A representative blot of n=4 independent experiments is shown.

b. Quantification of endogenous GFP-PGRP-LC protein, or accumulation of Rel49 fragment, during early infection time points as in a. Data shows mean + SEM of n=4 independent Western blots as in a., analysed by Kruskal-Wallis test with Dunn’s post-test. For GFP expression, only resc(LCwt),c564-Gal4>+ versus resc(LCΔex5),c564-Gal4>vps28 RNAi is significant (p<0.05). For Rel49 accumulation, no comparison reached significance.

c. rLC and Vps28 act together in resolving IMD activation after infection. Flies as in a. were challenged with heat-killed Ecc15 and Dpt mRNA expression was monitored over time. Data represents mean + SEM of n=4 independent experiments, analysed by 2-way ANOVA and Bonferroni post-test. Note that all genotypes are single-copy rescues and therefore show a less pronounced effect of rLC.

d. rLCx acts upstream of Vps28. Regulatory capacity of overexpressed rLCx is affected in fat bodies with knock-down of vps28. IMD activation was quantified as Dpt mRNA levels in whole flies, and normalized to activation in control (w;c564-Gal4 x w¹¹¹⁸). Data represent mean + SEM of n≥3 repeats, analysed by 2-way ANOVA with Bonferroni post tests (see also Supplementary Fig. 5g).

e. dIAP2 interacts with the PHD domain of rLC. Indicated constructs were transfected into S2 cells and HA-tagged dIAP2 was purified with anti-HA antibodies. Expression and presence of co-purified rLC proteins was analysed by immunoblotting with anti-V5 antibody. Cell lysates are indicated in the left panel, and IP-ed proteins are indicated in the right panel.

As receptor ubiquitylation can serve as signal for endocytic trafficking, we tested whether rLC recruits the ubiquitin ligase DIAP2, which acts at multiple steps in the IMD pathway ^{46, 47, 48}. Indeed, co-immunoprecipitation experiments in S2 cells indicated that rLC bound to DIAP2, and that this interaction requires the PHD domain of rLC (Fig. 6e).

In summary, our data suggest that resolution of IMD signalling is controlled by rLC-mediated endocytosis and ESCRT-dependent degradation of PGRP-LC.

Discussion

PGRP-LC is the major signalling receptor that senses the presence of Gram-negative bacteria in flies. Upon ligation of PGRP-LC with PGN, it stimulates activation of the IMD pathway that ultimately drives expression of anti-microbial peptide genes. While it is clear that PGRP-LC responds to monomeric muropeptides from live bacteria, it remained elusive how this receptor contributes to the resolution phase, once bacteria are killed and release polymeric peptidoglycans. In the present study, we have uncovered additional PGRP-LC receptor isoforms with regulatory capacities that allow adequately adjusting the immune response to the level of threat. We show that rLC isoforms specifically down-regulate IMD pathway activation in response to polymeric PGN, a hallmark of efficient bacterial killing. Our data are consistent with the following mechanistic model for rLC-mediated immune resolution (Fig. 7). Upon Gram-negative infection, dividing bacteria release TCT, which efficiently activates IMD signalling by recruiting LCx/LCa dimers. Ligand binding depends on the ectodomains alone; therefore homodimers and heterodimers of activating and regulatory isoforms are equally likely to assemble. Since IMD recruitment requires dimerization of activating receptors, neither rLC:rLC homodimers nor rLC:LC heterodimers can activate the pathway. In this sense, rLC acts similarly to PGRP-LF with regards to TCT: it can form ligand-bound dimers with LC but cannot signal ²⁴. Immune activation triggers release of bactericidal effectors that rapidly kill bacteria, as well as feedback regulators of the IMD pathway, including rLC. Dead bacteria release polymeric peptidoglycan, which shifts ligand availability from TCT to PGN. PGN polymers are able to recruit multimeric complexes of activating and regulatory isoforms. While these complexes can still activate IMD signalling, the presence of rLC leads to efficient endocytosis and termination of signalling via the ESCRT pathway. The latter mechanism ensures that signalling is switched off once the balance is tipped in favour of ligands signifying dead bacteria, allowing Drosophila to terminate a successful immune response. Failure to do so results in death of the host despite bacterial clearance.

The early death of rLC deficient flies might be due to aberrant activation of apoptotic pathways downstream of IMD. Indeed, mammalian TNFR signalling, which is functionally similar to the IMD pathway, induces NF-κB and MAPK cascades from the plasma membrane, but triggers cell death from an intra-cellular compartment following endocytosis ^{49, 50}. Possibly, rLC mutants have erroneous trafficking of LC, and therefore induce apoptosis after prolonged retention of LC at the plasma membrane.

Molecularly, rLC is characterized by the presence of a cytosolic PHD domain, which is predicted to bind to phosphoinositides. We find that the PHD domain of rLC targets rLC to distinct membrane domains, but cannot exclude that this localization also relies on protein-protein interactions. Furthermore, the PHD domain is also required for the binding of rLC to the cytosolic regulator Pirk and the ubiquitin ligase DIAP2. The combined presence of a membrane localization domain and the capability to recruit downstream signalling modulators is reminiscent of the “sorting/signalling adaptor paradigm” that is emerging for mammalian PRRs ⁵¹. Sorting adaptors are cytosolic signalling components with phosphoinositide binding domains that are selectively recruited to defined subcellular locations and thereby shape the signalling output of the receptors they interact with. Examples include mammalian TLR sorting/signalling adaptor pairs TIRAP/MyD88 and the Drosophila Toll adaptor dMyD88 ^{52, 53}. Strikingly, rPGRP-LC combines features of signalling receptors and sorting adaptors into one single molecule.

Our data further show that IMD signalling is controlled by endocytic mechanisms and the ESCRT pathway. Interestingly, our screen suggests that PGRP-LC initiates IMD signalling from the membrane, as none of the screened endocytic adaptors prevented IMD activation. This is consistent with signalling downstream of the TNF receptor, a pathway homologous to IMD signalling, where recruitment of RIP1 and cIAPs to the TNFR signalling complex at the plasma membrane allows activation of NF-κB ^{49, 50}. The observation that IMD signalling is enhanced in Rab5- or Fab1-depleted fat bodies suggests that signalling continues from arrested, immature endosomes. Our screen further suggests that endocytosis of receptors into a degradative compartment is required for resolution. Failure to mature early endosomes, like defects in the formation of multivesicular bodies (MVBs), results in enhanced immune activation and defects in immune resolution.

Recent evidence from vertebrates also implicates the ESCRT machinery in suppressing spurious NF-κB activation: the TNFR superfamily member Lymphotoxin-β Receptor, which activates a signalling cascade that is functionally similar to PGRP-LC signalling, is degraded in an ESCRT-dependent manner in zebrafish and human cells (Máminska et al. Science signalling, accepted). Thus ESCRT-mediated clearance of receptors upstream of NF-κB seems remarkably well conserved.

With respect to Gram-negative bacterial detection, we uncover functional parallels in sensing and signalling by PRR/MAMP couples in Drosophila and vertebrates. In vertebrates, Gram-negative bacteria are sensed through LPS-TLR4 interactions. Like PGRP-LC, TLR4 signals from the plasma membrane and is subsequently internalised via dynamin endocytosis, followed by intracellular trafficking that relies on the ESCRT-0 component Hrs and ESCRT-I. This trafficking of TLR4 helps to desensitize cells to subsequent stimuli ⁵⁴. An intriguing question is whether down-regulation of the PGN detection machinery in flies similarly results in anergy towards subsequent bacterial infections.

From an evolutionary perspective, rLC presents an unexpected case of physical coupling of functional modules. PHD domains are extremely rare in transmembrane receptors and much more frequent in soluble, cytosolic molecules. In vertebrate immune signalling, bacterial sensing modules (e.g. TLRs), lipid-binding sorting modules (e.g. TIRAP or TRIF) and signalling modules (MyD88, TRAM) are carried on separate molecules and assembled in transient interactions ⁵¹. Drosophila MyD88 combines sorting and signalling functions in a single molecule, by-passing the need for TIRAP ⁵³. rLC takes another shortcut by combining sensing and sorting in a single receptor. The fact that Drosophila rLC has no immediate homologs in vertebrates with peptidoglycan-sensing and –signalling pathways suggests evolutionary uncoupling of sensing and sorting domains, possibly to increase the spectrum of signalling by combinatorial recruitment of adaptors to sensing receptors.

Methods

Methods and associated references are available in the online version of the paper.

Supplementary Material

Note: Supplementary Information including 5 Supplementary Figures, 4 Supplementary Tables, Supplementary Figure Legends, Supplementary Discussion and Supplementary References are available in the online version of this paper.