Dual strategies for peptidoglycan discrimination by peptidoglycan recognition proteins (PGRPs)

Abstract

The innate immune system constitutes the first line of defense against microorganisms in both vertebrates and invertebrates. Although much progress has been made toward identifying key receptors and understanding their role in host defense, far less is known about how these receptors recognize microbial ligands. Such studies have been severely hampered by the need to purify ligands from microbial sources and a reliance on biological assays, rather than direct binding, to monitor recognition. We used synthetic peptidoglycan (PGN) derivatives, combined with microcalorimetry, to define the binding specificities of human and insect peptidogycan recognition proteins (PGRPs). We demonstrate that these innate immune receptors use dual strategies to distinguish between PGNs from different bacteria: one based on the composition of the PGN peptide stem and another that senses the peptide bridge crosslinking the stems. To pinpoint the site of PGRPs that mediates discrimination, we engineered structure-based variants having altered PGN-binding properties. The plasticity of the PGRP-binding site revealed by these mutants suggests an intrinsic capacity of the innate immune system to rapidly evolve specificities to meet new microbial challenges.

The innate immune system recognizes invading microbes by means of conserved pattern recognition receptors that bind unique products of microbial metabolism not produced by the host (pathogen-associated molecular patterns) (1, 2). Examples of microbial ligands recognized by pattern recognition receptors such as Toll-like receptors, peptidoglycan recognition proteins (PGRPs), and NOD proteins include lipopolysaccharide of Gram-negative bacteria, lipoteichoic acid of Gram-positive bacteria, nonmethylated CpG sequences, flagellin, and peptidoglycan (PGN) of Gram-negative and -positive bacteria. Cellular activation by pattern recognition receptors results in acute inflammatory responses involving cytokine and chemokine production, direct local attack against the invading pathogen, and induction of the adaptive component of the immune system. In humans, overactivation of inflammatory responses can lead to septic shock, which accounts for 100,000 deaths annually in the United States alone. By sitting at the intersection of the pathways of microbial recognition, inflammation, and cell death, the innate immune system offers emerging opportunities for the development of therapeutics to modulate immune responses (3).

PGRPs, a newly discovered class of pattern recognition receptors, are highly conserved from insects to mammals (4–7). By detecting PGN from both Gram-negative and -positive bacteria, PGRPs are important contributors to host defense against microbial infections (2, 4). PGNs are polymers of alternating N-acetylglucosamine (GlcNAc) and N-acetylmuramic acid (MurNAc) in β(1→4) linkage, crosslinked by short peptide stems composed of alternating l- and d-amino acids (8, 9) (Fig. 1A). Whereas the carbohydrate backbone is conserved among all bacteria, except for de-N-acetylated or O-acetylated variants, considerable diversity exists in the peptide moiety (8, 9). According to the residue at position 3 of the stems, PGNs are classified into two major groups: l-lysine type (Lys-type) and meso-diaminopimelic acid type (Dap-type). Dap-type PGN peptides are usually directly crosslinked, whereas Lys-type PGN peptides are interconnected by a peptide bridge that varies in length and amino acid composition in different bacteria (Fig. 1A). Moreover, bacteria differ widely in the extent of crosslinking (5–75%), thereby introducing additional variability in PGN structure (8, 9).

Fig. 1.

Structure of PGN and PGN derivatives. (A) General structure of natural PGN. R¹, H (Lys) Gram-positive PGN; R¹, COOH (Dap) Gram-negative PGN; R², H or crosslink. (B) The monomeric muramyl pentapeptides, MPP (R¹, H, compound 1) and MPP-Dap (R¹, COOH, compound 2), are represented in a single figure. (C) The crosslinked Lys-type PGN analog, CL-PGN (compound 3). (D) The natural Dap-type PGN fragment, TCT, which contains an anhydro bond between C6 and C1 of MurNAc.

In Drosophila, PGRPs activate two distinct signaling pathways that induce production of antimicrobial peptides: the Toll receptor pathway, which is primarily triggered by Lys-type PGNs from Gram-positive bacteria, and the Imd pathway, which is mainly activated by Dap-type PGNs from Gram-negative bacteria (2). Mammalian PGRPs, located in neutrophil and eosinophil granules, participate in the intracellular killing of both Gram-positive and -negative bacteria (4, 10, 11).

A major difficulty in identifying molecular determinants recognized by pattern recognition receptors such as PGRPs, or in simply establishing which general category of PAMP is recognized, arises from the usual practice of purifying these products from bacterial cell walls, such that contamination with other cell wall components has often led to contradictory conclusions. For example, although human NOD1 was initially believed to detect lipopolysaccharide (12), and Drosophila PGRP-LC to sense both lipopoysaccharide and PGN (13), it now appears that these receptors recognize PGN alone (14, 15). Similarly, mammalian Toll-like receptor 2 is no longer thought to detect lipopolysaccharide (16, 17), but rather lipoteichoic acids, zymosan, and PGN (18), although PGN recognition by Toll-like receptor 2 has been disputed (19). Further complicating the analysis of receptor–ligand interactions is the exclusive reliance on biological assays to monitor recognition (1, 2), because these assays measure immunostimulatory capacity, not direct binding. To circumvent these difficulties, we synthesized muramyl pentapeptides that contain either Lys (1) or Dap (2) as the third amino acid (Fig. 1B) and a crosslinked Lys-type PGN (3) (Fig. 1C). We used these compounds, in conjunction with isothermal titration calorimery (ITC), to elucidate the intrinsic PGN-binding specificities of human and insect PGRPs, the strategies these pattern recognition receptors employ to discriminate PGNs from different microbes, and the structural basis for this discrimination.

Results

Analysis of PGN Analog Binding to Human and Insect PGRPs. Use of classical fluorenylmethoxycarbonyl chemistry and solid phase synthetic techniques enabled the assembly of compounds 1–3 (Fig. 1 B and C). Having synthesized these compounds, attention was focused on determining thermodynamic parameters of their binding to human and Drosophila PGRPs. C-terminal PGN-binding domain of PGRP-Iα (PGRP-IαC) recognizes MurNAc-l-Ala-d-isoGln-l-Lys-d-Ala-d-Ala (MPP) (Fig. 1B, 1), a muramyl pentapeptide representing the conserved core of Lys-type PGN from Gram-positive bacteria (Fig. 1A), with a K_b of 45 × 10³ M^–1. Despite the low affinity, there are measurable signal changes upon successive injections of ligand into the protein (Fig. 2 A and B) that allow for accurate determination of K_b (20, 21). Furthermore, the fit of the ITC data to a single-site model returns an n value of 1.0 PGRP-IαC per MPP, which indicates that we have accurately determined the starting concentrations of both protein and ligand solutions and that PGRP-IαC and MPP are each fully active. The solution state stoichiometry obtained by ITC is consistent with the single PGN-binding site per PGRP-IαC monomer observed by x-ray crystallography (22). Truncation of the peptide stem of MPP at position 2 reduces affinity ≈60-fold based on the binding of GlcNAc-MurNAc-l-Ala-d-Ala (GMDP) (Table 1). Injection of volumes, up to 10-μl aliquots, of 10-fold molar excess of Dap into 0.09 mM PGRP-IαC resulted in heats similar to those of dilution (Fig. 5A, which is published as supporting information on the PNAS web site) with no detectable change in the incremental heat per mole of added Dap (Fig. 5B), indicating no binding. Under these experimental conditions and those used throughout this study, interactions with K_b > 100 M^–1 should be detectable (21). These results and those presented below reveal that PGN recognition by PGRPs, although of low affinity (K_b < 10⁵ M^–1), is nevertheless highly selective (see Discussion).

Fig. 2.

Calorimetric titrations of human PGRPs with monovalent and bivalent PGN ligands at 275–277 K. (A) Raw data obtained from 50 automatic injections of 2-μl aliquots of 12.3 mM MPP solution into 0.264 mM human PGRP-IαC solution. (B) Nonlinear least-squares fit (solid line) of the incremental heat per mole of added ligand (open squares) for the titration in A. The equilibrium binding constant obtained from this titration is K_b = 45.0 ± 1.0 × 10³ M^–1 with n = 1.01 ± 0.01. (C) Raw data obtained from 45 automatic injections, first one of 1-μl aliquot, and the remaining of 3-μl aliquots of 1.42 mM CL-PGN solution into 0.042 mM human PGRP-IαC solution. (D) Nonlinear least-squares fit (solid line) of the incremental heat per mole of added ligand (open squares) for the titration in C; K_b = 74.2 ± 4.6 × 10³ M^–1 with n = 2.02 ± 0.03.

Table 1. Binding constants (×10³ M^–1) of PGN derivatives to human and Drosophila PGRPs at 275 K.

Protein	MPP	MPP-Dap	TCT	CL-PGN*
h-PGRP-IαC	45.0 (±1.0)†	NB	20.2 (±0.8)†	74.2 (±4.6)
h-PGRP-S	6.3 (±0.4)	47.4 (±6.0)	23.4 (±1.9)	NB
d-PGRP-LCx	NB†	63.5 (±4.7)	19.8 (±0.8)‡	NB
d-PGRP-LCa	17.8 (±1.5)†	NB	8.7 (±1.6)	NB
h-PGRP-S (G68N, W69F)	85.1 (±0.4)	3.6 (±0.9)‡	45.4 (±2.1)	85.2 (±8.1)
d-PGRP-LCx (G393N, W394F)	23.8 (±1.3)‡	NB‡	18.8 (±1.8)‡	NB‡
d-PGRP-LCa (Q412N, K413F)	36.3 (±1.7)‡	72.2 (±14.0)	33.8 (±1.9)	83.0 (±9.6)

^*For CL-PGN, n ranged from 1.95 to 2.04 with uncertainties from 2.6% to 4.5%; for all other PGN derivatives, n ranged from 0.99 to 1.08 with uncertainties from 0.4% to 12.9%. K_b values × 10³ M^-1. Values in parentheses represent uncertainties of fit. h, human; d, Drosophila; NB, no binding detectable (K_b < 100 M^-1); CL-PGN, crosslinked PGN (3, Fig. 1C); Dap, meso-diaminopimelic acid; GMDP, GlcNAc-MurNAc-l-Ala-d-isoGln; MPP, MurNAc-l-Ala-d-isoGln-(2S,6R)-l-Lys-d-Ala-d-Ala (1, Fig. 1B); MPP-Dap, MurNAc-l-Ala-d-isoGln-(2S,6R)-Dap-d-Ala-d-Ala (2, Fig. 1B); TCT, GlcNAc-MurNAc(1,6-anhydro)-l-Ala-d-isoGlu-(2S,6R)-Dap-d-Ala (Fig. 1D).

^†K_b at 277 K.

^‡K_b at 276 K.

In addition to MPP, PGRP-IαC recognizes 3, an analog of crosslinked Lys-type PGN (CL-PGN) composed of MPP connected to the muramyl tetrapeptide MurNAc-l-Ala-d-isoGln-l-Lys-d-Ala via a pentaglycine bridge (Fig. 1C), with K_b = 74 × 10³ M^–1 and stoichiometry of 2.0 PGRP-IαC per CL-PGN (Fig. 2 C and D). The simplest model that best fits the binding isotherm comprises two independent, identical sites of each CL-PGN engaging two PGRP-IαC, indicating accommodation of the connecting peptide. At the same time, a strict requirement for l-Lys at position 3 of the peptide stem is implicit in the inability of PGRP-IαC to bind the muramyl pentapeptide MurNAc-l-Ala-d-isoGln-(2S,6R)-Dap-d-Ala-d-Ala (MPP-Dap) (Fig. 1B, 2), which corresponds to the core of Dap-type PGNs from Bacillus and Gram-negative bacteria (Table 1). By contrast, human PGRP-S preferentially recognizes MPP-Dap (K_b = 47 × 10³ M^–1) over MPP (6.3 × 10³ M^–1), and fails to bind CL-PGN (Fig. 5 C and D). The ability of PGRP-S to recognize both MPP and MPP-Dap, albeit with somewhat different affinities, is consistent with the finding that mice deficient in PGRP-S exhibit increased susceptibility to i.p. infections with B. subtilis (Dap-type PGN) and Micrococcus luteus (Lys-type PGN) (11). Notably, M. luteus PGN is only ≈25% crosslinked (8, 9), which should not preclude recognition by PGRP-S. The observed binding of human PGRP-S to MPP and MPP-Dap also agrees with recent data showing that this PGRP inhibits the in vitro growth of both Staphylococcus aureus (Lys-type PGN) and Escherichia coli (Dap-type PGN) (23).

The Drosophila PGRP-LC receptor, which is mainly triggered by Gram-negative bacteria, exists on the cell surface as three splice isoforms (PGRP-LCa, -LCx, and -LCy), each comprising a unique PGN-binding extracellular domain linked to identical transmembrane and cytoplasmic domains (13). These isoforms are believed to form homo- or heterodimers, via their membrane proximal cytoplasmic domain (24), with distinct PGN binding characteristics (15, 25). To investigate the specificities of PGRP-LCx and PGRP-LCa, we expressed their extracellular domains in soluble form. PGRP-LCx binds MPP-Dap with a K_b of 64 × 10³ M^–1 (Fig. 3 A and B) but fails to recognize MPP (Table 1). Conversely, PGRP-LCa binds MPP (K_b = 18 × 10³ M^–1; Fig. 3 C and D) but not MPP-Dap. Neither PGRP recognizes CL-PGN. These findings implicate PGRP-LCx as the isoform responsible for detecting Gram-negative bacteria while excluding PGRP-LCa from such a role.

Fig. 3.

Calorimetric titrations of Drosophila PGRP receptors with PGN ligands and tracheal cytotoxin at 275–278 K. (A) Raw data obtained from 40 automatic injections, first one of 1-μl aliquot, and the remaining of 4-μl aliquots of 1.98 mM MPP-Dap solution into 0.086 mM Drosophila PGRP-LCx solution. (B) Nonlinear least-squares fit (solid line) of the incremental heat per mole of added ligand (open squares) for the titration in A. The equilibrium binding constant obtained from this titration is K_b = 63.5 ± 4.7 × 10³ M^–1 with n = 1.01 ± 0.02. (C) Raw data obtained from 50 automatic injections of 1-μl aliquots of 12.46 mM MPP solution into 0.032 mM Drosophila PGRP-LCa solution. (D) Nonlinear least-squares fit (solid line) of the incremental heat per mole of added ligand (open squares) for the titration in C; K_b = 17.8 ± 1.4 × 10³ M^–1 with n = 1.01 ± 0.13. (E) Raw data obtained from 130 automatic injections of 2-μl aliquots of 2.89 mM TCT solution into 0.037 mM Drosophila PGRP-LCx solution. (F) Nonlinear least-squares fit (solid line) of the incremental heat per mole of added ligand (open squares) for the titration in E; K_b = 19.8 ± 0.8 × 10³ M^–1 with n = 1.04 ± 0.06. (G) Raw data obtained from 30 automatic injections of 2-μl aliquots of 9.7 mM MPP solution into 0.023 mM Drosophila PGRP-LCx (G393N, W394F) solution. (H) Nonlinear least-squares fit (solid line) of the incremental heat per mole of added ligand (open squares) for the titration in G; K_b = 23.8 ± 1.3 × 10³ M^–1 with n = 1.06 ± 0.09.

PGRP-LCx exhibits a similar overall specificity profile as PGRP-S except in its greater capacity to distinguish Dap-type from Lys-type PGN. On the other hand, PGRP-LCa, despite its resemblance to PGRP-IαC in binding MPP but not MPP-Dap, differs from PGRP-IαC in not recognizing CL-PGN. Thus, the four PGRPs analyzed here represent three qualitatively distinct specificity profiles after accounting for the similarity of PGRP-S and PGRP-LCx.

These results demonstrate that PGRPs use dual strategies, one based on Lys- or Dap-type specificity and another that relies on sensing the PGN crossbridge, to achieve exquisitely selective PGN recognition and discrimination. PGRP-S, PGRP-LCx, and PGRP-LCa distinguish PGN ligands on the basis of both criteria, whereas PGRP-IαC, although remarkably specific for Lys-type PGN, is insensitive to crosslinking. Importantly, the identity of the amino acid at position 3 of the stem, coupled with differences in the type and amount of crosslinking between stems, account for almost all variability in PGNs from different bacteria (8, 9).

The result that PGRP binding to PGN analogs highly depends on the composition of the peptide stem helps explain the ability of PGRPs to distinguish PGNs from different bacteria, as measured in biological assays (1, 2, 26–29). In addition, the biological relevance of our finding that certain PGRPs detect PGN crosslinking is supported by very recent data showing that M. luteus PGN, which is only ≈25% crosslinked (8, 9), is capable of triggering the Drosophila Imd pathway through the PGRP-LC receptor, but that Staphylococcus aureus PGN, which is highly (≈75%) crosslinked, is inactive (30). Moreover, S. aureus PGN became as stimulatory as M. luteus PGN after enzymatic digestion of its pentaglycine crossbridges.

Interaction of Human and Insect PGRPs with Tracheal Cytotoxin. We also measured PGRP binding to GlcNAc-MurNAc(1,6-anhydro)-l-Ala-d-isoGlu-(2S,6R)-Dap-d-Ala (TCT) (Fig. 1D), a natural monomeric fragment of Dap-type PGN containing an anhydro form of MurNAc (Fig. 3 E and F) (31). TCT is the factor responsible for tissue damage in whooping cough and gonorrhea infections (31, 32). Acting through the PGRP-LC receptor, TCT is also a potent activator of the Drosophila Imd pathway (15). Surprisingly, TCT binds all four PGRPs with similar affinities (Table 1), ranging from 8.7 × 10³ M^–1 (PGRP-LCa) to 23 × 10³ M^–1 (PGRP-S). The reported inability of PGRP-LCa to bind TCT in pull-down assays (25, 33) may be explained by the lower sensitivity of this nonquantitative detection method compared to ITC.

Because PGRP-IαC and PGRP-LCa both bind TCT, yet show no detectable reactivity toward MPP-Dap, we hypothesize that the 1,6-anhydro bond of TCT, which locks MurNAc into the ¹C₄ conformation, alters the interaction of the peptide stem with PGRPs in a manner preventing discrimination against Dap at position 3. Promiscuous binding of TCT to PGRPs, or PGRP-like molecules, could contribute to the diverse biological effects of this PGN fragment, which include triggering normal light organ morphogenesis in the squid (34).

Structural Basis for PGN Discrimination by PGRPs. To identify the site (or sites) of PGRPs responsible for discriminating between l-Lys and Dap at peptide position 3, and between crosslinked and noncrosslinked PGN, we engineered structure-based variants of human and Drosophila PGRPs. In the crystal structure of PGRP-IαC bound to the muramyl tripeptide MurNAc-l-Ala-d-isoGln-l-Lys (MTP) (22), the ligand is bound in a deep groove, where the side chain of l-Lys packs against Asn-236 and Phe-237 at one extremity (Fig. 4A). Sequence variability at these two positions among >40 PGRPs suggests that they may account for the ability of these proteins to distinguish PGNs from different microbes. To test this hypothesis, we asked whether the specificity profile of PGRP-S could be converted to that of PGRP-IαC by mutating Gly-68 and Trp-69 of PGRP-S to the corresponding residues of PGRP-IαC (Asn and Phe, respectively). In sharp contrast to wild-type PGRP-S, which binds MPP ≈7-fold less tightly than PGRP-IαC, the PGRP-S double mutant binds MPP ≈2-fold better than PGRP-IαC(K_b = 85 × 10³ M^–1 versus 45 × 10³ M^–1) (Table 1). The PGRP-S mutant also resembles PGRP-IαC in binding MPP-Dap with ≈25-fold lower affinity than MPP, whereas the wild-type PGRP-S exhibits a preferential recognition of MPP-Dap. Unlike wild-type PGRP-S, for which no interaction with CL-PGN could be detected (Fig. 5 C and D), the mutant binds this ligand with K_b = 85 × 10³ M^–1 (Table 1), an affinity similar to that of PGRP-IαC (74 × 10³ M^–1; Fig.2 C and D). At the same time, the PGRP-S mutant retains the capacity to recognize TCT, with ≈2-fold higher affinity than wild-type PGRP-S or PGRP-IαC. These results demonstrate that a single site in the PGN-binding cleft of PGRPs, corresponding to residues 236 and 237 in PGRP-IαC (Fig. 4A), mediates discrimination between l-Lys and Dap at peptide position 3 and between crosslinked and noncrosslinked Lys-type PGN.

Fig. 4.

PGN-binding site of PGRPs. (A) Crystal structure of human PGRP-IαC in complex with MTP, showing interactions at the binding site (22). MTP is drawn in stick representation. Carbon atoms are cyan, nitrogen atoms are dark blue, and oxygen atoms are red. PGRP-IαC is yellow; residues making hydrogen bonds (dashed lines) with MTP are green. In purple are Asn-236 and Phe-237, drawn in ball-and-stick representation, which form van der Waals contacts with the side chain of l-lysine. MurNAc, N-acetylmuramic acid; ALA, l-alanine; IDG, l-isoglutamine; LYS, l-lysine. (B) View of the PGN-binding site of human PGRP-S (35). The orientation is the same as in A. In green are residues of PGRP-S corresponding to MTP-contacting residues in the PGRP-IαC–MTP complex. Gly-68 and Trp-69 (purple) are predicted to contact the side chain of Lys-type PGN or Dap-type PGN. Gly-68 is represented by its carbonyl oxygen. (C) Structure-based sequence alignment of specificity-determining residues of mammalian and insect PGRPs. Residues corresponding to Asn-236 and Phe-237 of human PGRP-IαC in A are highlighted in yellow. Mammals: Bt, Bos taurus; Cd, Camelus dromedarius; Hs, Homo sapiens; Mm, Mus musculus; Rn, Rattus norvegicus; Ss, Sus scrofa. For human and mouse PGRP-Iα and PGRP-Iβ, C and N indicate the C-terminal and N-terminal PGRP domains, respectively. Insects: Ag, Anopheles gambiae; Bm, Bombyx mori; Ce, Calpodes ethlius; Dm, Drosophila melanogaster; Ms, Manduca sexta; Tn, Trichoplusia ni. Sequence alignments were performed by using the program clustalw at ExPASy (www.expasy.ch). The figure was generated by using espript (http://espript.ibcp.fr/ESPript/ESPript).

Further support for this conclusion is provided by mutagenesis of Drosophila PGRPs. The double-mutant PGRP-LCx (Gly393Asn and Trp394Phe), which bears the same substitutions as the PGRP-S mutant, follows the expected pattern of Lys-type versus Dap-type discrimination by recognizing MPP (K_b = 24 × 10³ M^–1; Fig. 3 G and H), but not MPP-Dap (Table 1), a pattern inverse from that of wild type. However, the mutations do not confer on PGRP-LCx the same ability to bind CL-PGN we observed for PGRP-IαC and PGRP-S (Gly68Asn and Trp69Phe), presumably due to additional structural differences between the Drosophila and human proteins. As a result, PGRP-LCx (Gly393Asn and Trp394Phe) closely resembles PGRP-LCa in its specificity profile. By contrast, PGRP-LCa (Gln412Asn and Lys413Phe), a highly promiscuous variant, has acquired the ability to bind both CL-PGN (83 × 10³ M^–1) and MPP-Dap (72 × 10³ M^–1) while retaining MPP and TCT recognition (Table 1). Taken together, these results underscore the plasticity of the PGN-binding site of PGRPs.

The structures of the PGRP-IαC–MTP complex (22) and PGRP-S (35) provide an explanation for the effect of the mutations on recognition of Lys- versus Dap-type PGN. In the PGRP-IαC–MTP complex (Fig. 4A), the side chain of Asn-236 protrudes from the wall of the binding groove, making van der Waals contacts with the side chain of l-Lys, especially atoms Cε and Nζ. Attachment of a carboxy group to the Cε atom, which distinguishes Dap from l-Lys, would create steric clashes with Asn-236, decreasing affinity. The corresponding Gly-68 of PGRP-S (Fig. 4B), because it lacks a side chain, does not interfere with binding. Accordingly, mutation of Gly-68 (or Gly-393 of PGRP-LCx) to Asn should reduce, or abolish, recognition of Dap-type PGN ligands, as observed for MPP-Dap (Table 1). Less evident is the structural basis for the effects of mutations on discrimination between crosslinked and noncrosslinked Lys-type PGN, which will require crystallographic analysis of the interaction of PGRPs with the crossbridge.

Discussion

The binding of PGN derivatives to PGRP receptors, although highly selective, is of low affinity (K_b < 10⁵ M^–1). Such low-affinity, high-specificity recognition systems are gaining increasing importance in serving key biological functions, including T cell receptor recognition of peptide/MHC ligands (36–38), cell surface carbohydrate–protein interactions (39–41), and cell–cell adhesion (42). Moreover, it may be that even seemingly small (<10-fold) differences in the affinity of PGRPs for monovalent PGN ligands, such as we measured in several cases, are amplified by multiple PGRP–PGN interactions to establish specificity effects at the cellular level, as described for carbohydrate-binding proteins (43, 44). Indeed, the polymeric nature of natural PGN should facilitate multivalent binding of PGRPs. On the protein side, dimerization or oligomerization of PGRPs, as observed for PGRP-LC (15, 24, 25) and PGRP-IαC (45), may further enhance specificity at the cell surface or in solution. In this regard, cadherins have been shown to mediate highly specific intercellular adhesion through amplification of small affinity differences between low-affinity cadherin dimers as a result of multiple interactions (42).

Our results provide a basis for predicting the PGN-binding specificity of PGRPs that have not been characterized experimentally. Fig. 4C presents a structure-based sequence alignment of mammalian and insect PGRPs in the region encompassing PGRP-IαC residues 236 and 237, which we have shown by site-directed mutagenesis to mediate discrimination between l-Lys and Dap at position 3 of the PGN peptide stem and between crosslinked and noncrosslinked PGN. Group I PGRPs, like PGRP-IαC, contain Asn-Phe at positions 236 and 237 (or the homologous combinations Asp-Phe, Asn-Tyr, Asn-Trp, Gln-Tyr, or Gln-Tyr). These PGRPs are likely to exhibit a specificity profile similar to that of PGRP-IαC (i.e., recognition of MPP and CL-PGN but not MPP-Dap). Group II PGRPs, like PGRP-S and PGRP-LCx, contain Gly-Trp (or Gly-Tyr or Gly-Phe) and likely bind MPP-Dap better than MPP and do not bind CL-PGN. Group III PGRPs, which thus far has only one member (PGRP-LCa), contain Gln-Lys and bind only MPP. Group IV PGRPs, whose specificity we cannot currently assign, contain other sequences at these positions. It is notable that ≈65% (31 of 46) of the known PGRPs can be classified into Groups I or II.

The overall validity of this classification is supported by biological data. Thus, Drosophila PGRP-SA, which we predict should recognize Lys-type but not Dap-type PGN, is, in fact, the main recognition element for Gram-positive bacteria in insects (2). Similarly, our classification predicts that Drosophila PGRP-LE, like PGRP-LCx, should bind Dap-type PGN. Indeed, it has been demonstrated that PGRP-LE functions synergistically with PGRP-LC in conferring resistance to E. coli and other bacteria having Dap-type PGN (46).

Adaptive immunity relies on highly diverse receptors (antibodies and T cell receptors) generated through somatic gene recombination, whereas innate immunity is mediated by germline-encoded, nonrearranging receptors of restricted diversity (1, 2). Although more structurally conserved than other microbial components (e.g., proteins), ligands such as PGN and LPS present sufficient heterogeneity to pose a significant challenge to specific recognition by pattern recognition receptors. Moreover, the innate immune system must be adaptive enough to counter new microbial challenges. In the case of PGRPs, we have shown that two amino acid mutations suffice to alter binding specificity from Lys-type to Dap-type PGN (and vice versa) or from noncrosslinked to crosslinked forms of this cell wall component. Because these variables together account for most of the known diversity in PGN structure (8, 9), it appears the binding sites of individual PGRPs are poised to facilitate rapid evolution of new specificities to respond to changes in the microbial environment while using existing signaling or effector pathways. Indeed, the relative ease with which the PGN-binding characteristics of PGRPs can be manipulated raises the intriguing possibility of rewiring the innate immune system by, for instance, genetically reprogramming the Drosophila PGRP-LC–Imd and PGRP-SA–Toll pathways to respond to Gram-positive and Gram-negative bacteria, respectively, thereby inverting the existing activation pattern. In addition, detailed knowledge of the interaction of innate immune receptors with microbial ligands should provide a rational basis for the use of microbial cell wall molecules as adjuvants for vaccines and modulators of inflammation (3).

Methods

PGRP Production. Procedures for expressing human PGRP-S (residues 1–175) and the C-terminal PGN-binding domain of human PGRP-Iα (PGRP-IαC; residues 177–341) by in vitro folding from E. coli inclusion bodies have been described in refs. 22 and 35. The PGN-binding domains of Drosophila PGRP-LCx (residues 325–500) and PGRP-LCa (residues 343–520) were prepared similarly to the human proteins (see Supporting Materials and Methods, which is published as supporting information on the PNAS web site). Folded proteins were purified by using a MonoS (PGRP-S, -LCx, and -LCa) or MonoQ (PGRP-IαC) ion exchange column (Amersham Pharmacia Biosciences) followed by a Superdex 75 gel-filtration column (Amersham Pharmacia Biosciences). Site-directed mutagenesis of human PGRP-S and Drosophila PGRP-LCx was performed by using a QuikChange mutagenesis kit (Stratagene). Mutant proteins were expressed and purified similarly to wild type. All purified PGRPs were >95% pure by SDS/PAGE and behaved as monomers in size exclusion chromatography. N-terminal sequencing and MALDI mass spectrometry confirmed the identity of wild-type and mutant proteins.

PGN Derivatives. The target branched glycopeptide 3 (Fig. 1C) was assembled by polymer-supported synthesis with a hyperacid sensitive Sieber Amide resin (Supporting Information; see also Fig. 6, which is published as supporting information on the PNAS web site). Briefly, the resin-bound compound 6 was obtained through a series of steps by using standard coupling chemistry, Fmoc-protected amino acids, and a suitably protected MurNAc derivative. Upon removal of the ivDde protecting group on the lysine side chain at position three of 6, the pentaglycine bridge of the branched glycopeptide 8 was added. After incorporation of the second peptide stem, the partially deprotected 9 was cleaved from the solid support and the anomeric allyl moiety removed from MurNAc. Finally, purification by size exclusion resulted in the target compound 3 (Fig. 1C) as a mixture of α/β anomers. Compounds 1 and 2 (Fig. 1B) were prepared by a similar approach with either Fmoc-l-Lys(Mtt) or Fmoc-Dap(Boc, ^tBu) (47). GlcNAc-MurNAc-l-Ala-d-Ala and Dap were purchased from Sigma. Tracheal cytotoxin was purified from Bordetella purtussis culture supernatants as described in ref. 31.

ITC Measurements and Analysis. Thermodynamic parameters for the binding of PGRPs to PGN derivatives were determined by using a Microcal VP-ITC titration calorimeter. PGRPs were dialyzed exhaustively, and PGN derivative solutions were prepared in the final dialysate. Concentrations of PGN derivatives were based on dry weights. Protein concentrations were determined by absorbance at 280 nm with molar extinction coefficients (ε, M^–1 cm^–1) of 28,800 (PGRP-IαC), 39,800 (PGRP-S), 34,300 [PGRP-S (G68N, W69F)], 43,560 (PGRP-LCx), 38,060 [PGRP-LCx (G393N, W394F)], 32,560 (PGRP-LCa), and 32,560 [PGRP-LCa (Q412N, K413F)] (48). Solutions were prepared in 1 mM sodium phosphate buffer (pH 7.2 ± 0.1) (PGRP-IαC and PGRP-S), 1 mM Mes (pH 5.8 ± 0.1) (PGRP-IαC, PGRP-S, PGRP-LCx, and PGRP-LCa), or 1 mM Hepes (pH 6.8 ± 0.1) [PGRP-S (G68N, W69F), PGRP-LCx (G393N, W394F), and PGRP-LCa (Q412N, K413F)]. Buffer conditions were chosen for maintaining solubility at high protein concentrations necessary for ITC; for PGRP-IαC, the binding of PGN derivatives is pH-independent between pH 5.8 and 7.2. Aliquots (1–10 μl) of the PGN derivative solution (0.85–32.4 mM) were added to 1.41 ml of PGRP solution (0.022–0.264 mM) via a 250-μl microsyringe stirrer at 310 rpm.

Titration data were analyzed by using a single-site fitting model. A computerized nonlinear least square fitting method was used to determine the change in enthalpy (ΔH°_b), equilibrium binding constant (K_b), and molar stoichiometry (n). The shape of the titration curve is determined by the unitless quantity c, defined as the product of the initial macromolecule concentration ([PGRP]₀) and K_b, c = K_b[PGRP]₀. The c values for the present titrations (0.6 < c < 11.9) were within the permitted range (0.001 < c) for accurate K_b determinations (20, 21). Data acquisition and analysis were performed by using the software package origin.