Peptidoglycan Recognition Proteins Are a New Class of Human Bactericidal Proteins*

Skin and mucous membranes come in contact with external environment and protect tissues from infections by producing antimicrobial peptides. We report that human peptidoglycan recognition proteins 3 and 4 (PGLYRP3 and PGLYRP4) are secreted as 89-115-kDa disulfide-linked homo- and heterodimers and are bactericidal against several pathogenic and nonpathogenic transient, but not normal flora, Gram-positive bacteria. PGLYRP3 and PGLYRP4 are also bacteriostatic toward all other tested bacteria, which include Gram-negative bacteria and normal flora Gram-positive bacteria. PGLYRP3 and PGLYRP4 are also active in vivo and protect mice against experimental lung infection. In contrast to antimicrobial peptides, PGLYRPs kill bacteria by interacting with their cell wall peptidoglycan, rather than permeabilizing their membranes. PGLYRP3 and PGLYRP4 are expressed in the skin, eyes, salivary glands, throat, tongue, esophagus, stomach, and intestine. Thus, we have identified the function of mammalian PGLYRP3 and PGLYRP4, and show that they are a new class of bactericidal and bacteriostatic proteins that have different structures, mechanism of actions, and expression patterns than antimicrobial peptides.

Mammalian PGLYRPs were initially thought of as pattern recognition receptors similar to insect PGRPs (18,19). However, mammalian PGLYRP1 is present in granulocyte granules and likely participates in killing of phagocytized bacteria (31)(32)(33), and PGLYRP2 is an N-acetylmuramoyl-L-alanine amidase (34 -36), an enzyme that cleaves the stem peptide from the glycan chain of peptidoglycan and is constitutively produced in the liver and secreted into the bloodstream (18,35,36). Because the function of mammalian PGLYRP3 and PGLYRP4 has been unknown, here we have identified their function by testing the hypothesis that human PGLYRP3 and PGLYRP4 (and also PGLYRP1) have bactericidal activity and are expressed in tissues that come in contact with the external environment.

EXPERIMENTAL PROCEDURES
PGLYRPs-Blasticidin-resistant stable transfectants of Drosophila S2 cells expressing human PGLYRP1, PGLYRP3, and PGLYRP4 (Gen-Bank TM accession numbers AF076483, AY035376, and AY035377), or both PGLYRP3 and PGLYRP4 (PGLYRP3:4) in an inducible pMT/BiP/ V5-His vector (Invitrogen) were generated and integration of the constructs into host cell genome was confirmed by PCR. The transfectants were grown in serum-free medium (Drosophila SFM, Invitrogen) and expression of PGLYRPs was induced with 500 M CuSO 4 . All PGLYRP proteins were secreted and purified from culture supernatants by nickel-agarose (His-Bind Kit, Novagen) affinity chromatography under native conditions as recommended by the manufacturer following dialysis of the supernatants against binding buffer, except that all buffers included 4 mM CaCl 2 and 10% glycerol, and elution was done with 300 mM imidazole. PGLYRP-containing fractions were dialyzed against 10 mM Tris, pH 7.6, with 150 mM NaCl, 5 mM CaCl 2 , and 10% glycerol. The proteins were kept at Ϫ80°C and their purity is shown in Fig. 1, A and B. For Ca 2ϩ removal, proteins were treated with 8 mM EGTA for 10 min at 20°C and dialyzed against the above buffer without CaCl 2 or mocktreated and dialyzed with CaCl 2 . De-glycosylation was performed with 0.15 units of N-glycosidase/g of protein (glycopeptidase F from Flavobacterium meningosepticum, Sigma) for 1 h at 37°C.
Human PGLYRP1, PGLYRP2, PGLYRP3, and PGLYRP4 in pcDNA3.1 vector were transiently expressed in COS-7 cells (35) and used to confirm formation of PGLYRP1, PGLYRP3, and PGLYRP4 homodimers and PGLYRP3:4 heterodimers and de-N-glycosylation (not shown). PGLYRPs were also subcloned into the inducible pET-32 bacterial expression vector with N-terminal thioredoxin, His 6 , and S tags (Novagen), and recombinant PGLYRP proteins were expressed in Escherichia coli as inclusion bodies, which were isolated using the Bug-Buster kit (Novagen) and differential centrifugation as recommended by the manufacturer. The inclusion bodies were dissolved in 6 M urea and recombinant proteins were purified by nickel-agarose (His-Bind Kit, Novagen) in 6 M urea as recommended by the manufacturer. The purification yielded single bands on Coomassie Blue-stained gels that corresponded to the bands detected on Western blots with anti-tag antibodies and were used to confirm the specificity of anti-PGLYRP antibodies (in addition to PGLYRPs expressed in S2 cells, Fig. 1C).
Antimicrobial Assays-Microbicidal activity of PGLYRPs (purified from S2 cell supernatants), magainin ([Ala 8,13,18 ] (39). Microorganisms, at 0.5 to 2.5 ϫ 10 6 /ml (25 ϫ 10 6 /ml in experiments on membrane permeabilization), were incubated with PGLYRPs or bovine serum albumin (Sigma, or recombinant human IgG1 Fc fragment, expressed in S2 cells, with similar results, not shown) as a control, at the concentrations and the lengths of time indicated under "Results," at 37°C with shaking. The samples were diluted in 0.9% NaCl and plated on LB, BHI, or YPD agar plates for colony counts. For inhibition of killing, PGLYRPs were incubated with 200 g/ml of B. subtilis or S. aureus peptidoglycan (40,41), dextran (74 kDa from Leuconostoc mesenteroides, Sigma), or other products for 15 min at 20°C before addition of bacteria. Spheroplasts were obtained by treatment of 2-25 ϫ 10 6 /ml S. aureus with 10 g/ml lysostaphin or B. subtilis with 10 g/ml of lysozyme (hen egg from Sigma) and 10 g/ml of muramidase (mutanolysin from Sigma) in "protoplast" medium (Schaeffer medium with 0.75 M sucrose) at 37°C, and viability was determined by overlaying bacteria in 0.7% agar onto 1.5% agar plates, both in CYS medium with 0.5 M sucrose (42,43). The effectiveness and kinetics of S. aureus and B. subtilis spheroplast formation by treatment with lysostaphin or lysozyme plus mutanolysin, respectively, was confirmed by phase-contrast microscopy and by demonstrating a decrease in optical density at 660 and 540 nm.
We performed several controls to assure that the assay for bacterial killing by colony counts accurately measured the killing of bacteria, rather than PGLYRP-mediated bacterial aggregation: (a) we did not see any significant bacterial agglutination by microscopy at the concentrations of bacteria and PGLYRPs that were used in the killing assays; (b) agglutination is concentration-dependent and usually occurs only at equivalent ratios of bacteria to agglutinins, and is not prominent at excessive concentrations of either agglutinins or bacteria; however, killing of bacteria by PGLYRPs was similar at a wide range (1000-fold) of bacterial concentrations; (c) breaking any potential aggregates by high speed vortexing or gentle sonication did not change the results; and (d) bacteria could be rescued from killing by PGLYRPs by diluting into a medium with 0.75 M sucrose and plating into semisolid agar, if there was just aggregation of bacteria, the aggregates would stay in the agar and there would be no prevention of killing. Bacterial aggregation was also not responsible for the lower bacterial counts recovered from the lungs in in vivo experiments, because bacterial counts in the homogenized nasal tissues were not significantly different in PGLYRP3-treated and untreated mice 1 h after infection. The V5-His tag also had no influence on bacterial killing, because PGLYRP1 with the V5-His tag (purified on nickel-agarose) and PGLYRP1-Fc fusion protein without the V5-His tag (human IgG1 Fc fragment fused to the C-terminal of PGLYRP1, and purified from culture supernatants of S2 stable transfectants on protein A-agarose) both had the same bactericidal activity (not shown).
Intranasal Lung Infection-Female BALB/c mice (6/group), 6 -8 weeks old (from Harlan Sprague-Dawley), were anesthetized with ketamine/xylazine, 4.5 g of PGLYRP3, PGLYRP4, heat-inactivated (100°C, 5 min) PGLYRP3, or recombinant Fc IgG fragment in 15 l or buffer alone was instilled intranasally, 15 min later the mice were challenged intranasally with 3 ϫ 10 8 S. aureus in 10 l of saline, and at 4 h postchallenge of the numbers of S. aureus in homogenized lungs were determined by colony counts. The results were expressed as geometric means and the significance of the difference between PGLYRP-treated and buffer-treated groups was determined using the Mann-Whitney U test. The experiments were approved by the Institutional Animal Care Committee.
Antibodies-Anti-PGLYRP antibodies were obtained by immunizing rabbits with peptides corresponding to the following amino acids: CLD-PQHPVMPRKV for PGLYRP3, and CSQRLRELQAHHVHNNSG for PGLYRP4, coupled to KLH, followed by affinity purification on Sul-foLink gel (Pierce) with corresponding peptides linked through the N-terminal Cys, elution with Tris glycine buffer, pH 2.5, and dialysis against phosphate-buffered saline, pH 7.2. The antibodies were specific for each PGLYRP, as demonstrated by their reactivity on Western blots with only one PGLYRP from which the peptide was derived, and no reactivity with the other PGLYRPs (Fig. 1C). An antibody to a different peptide, which did not react with human PGLYRP3 or PGLYRP4, prepared and purified by the same method, was used as a negative control. V5 mouse monoclonal antibody was from Invitrogen and peroxidaselabeled S-protein from Novagen.
Immunohistochemistry-Paraffin 5-m sections of skin, eye, submandibular salivary gland, parotid salivary gland, throat, tongue, esophagus, stomach, small intestine, and colon, from normal human donors (male and female, 21-65 years old), were prepared by BioChain. Following standard deparaffinization, re-hydration, and quenching of endogenous peroxidase by 30 min incubation in 0.3% H 2 O 2 , the slides were incubated with 0.6 -1.2 g/ml of anti-PGLYRP3, anti-PGLYRP4, or control antibody (obtained and tested for specificity as described above) overnight, followed by biotinylated second antibody and Vectastain Elite ABC kit (Vector) with 3,3Ј-diaminobenzidine as a substrate (which generates a brown reaction product) and counterstaining with hematoxylin (blue). Lysozyme was detected in the intestine as described (45) using the Vectastain ABC kit.
RNA, Real-time RT-PCR, and Northern Blots-RNA from normal human liver, skin, esophagus, and tongue were obtained from Clontech and Stratagene. Corneas were isolated from normal human eyes preserved in RNAlater (Ambion), and obtained from the National Disease Research Interchange or Central Florida Lions Eye and Tissue Bank. RNA was extracted from cornea or cultured cells using TRIzol reagent (Invitrogen). Quantitative real time reverse transcriptase-polymerase chain reaction (RT-PCR) was done using TaqMan reagents and the ABI Prism 7000 Sequence Detection System as recommended by the manufacturer (Applied Biosystems). Briefly, first-strand cDNA was synthesized from 1 g of RNA using TaqMan Reverse Transcription Reagents (Applied Biosystems) and random hexamers (Invitrogen) for 10 min at 25°C, followed by reverse transcription for 60 min at 37°C, and inactivation for 5 min at 95°C. The comparative C T method was used with 18 S RNA as an endogenous control. Each sample was assayed in duplicate with TaqMan Universal PCR Master Mix (Applied Biosystems), primer concentrations of 0.6 M (experimental) and 0.2 M (18 S), and 0.1 M probe concentrations. The cycling conditions were: uracil-DNA glycosylase incubation at 50°C for 2 min, AmpliTaq Gold DNA polymerase activation at 95°C for 10 min, and 40 two-step cycles of 95°C for 15 s and 60°C for 60 s. Each experiment (including reverse transcription) was repeated at least three times. The forward and reverse primers and the probes were as follows: PGLYRP1, GCAGCACTACCACATGAA-GACACT, CGAGCCCGTCTTCTCCAAT, CTGGTGCGACGTGG-GCTACAACTTC; PGLYRP2, CTGGATCCTACTCGGATTGCT-ACT, GCAGAAGCTGTGTGTCTGGTCTT, CCTGGCTGAGCTG-GAGCAGAAAGTG; PGLYRP3, ACCTGCCTGGACCCTCAAC, CC-AAGCAGATCGTTTGATGATG, TGCCCAGGAAGGTTTGCC-CCA; PGLYRP4, TCCAGTTATGTTCAGCCACTTCTT, AAGCCT-TCTTCAGGCTTGTCTTC, AGAACTGCCTGGCCCCTCGGC; and 18 S, GCCGCTAGAGGTGAAATTCTTG, CATTCTTGGCAAATG-CTTTCG, ACCGGCGCAAGACGGACCAG. Northern blots were done as described (18) with the following modifications: 10 g of RNA/ lane was used, the probes were the following fragments amplified with the following primers: PGLYRP2 forward, ACAATGGCCCAGGGTGT-CCTCT and reverse, GTACCCCTCAACCACGGCACAC; PGLYRP3 forward, GTCCCAGCCCTGCTGCCTTATC and reverse, CCTGGG-GAGGAGGTGGCTCTTA; PGLYRP4 forward, TCCAGAAGGGC-CACCTGTCATC and reverse, GACAAGGTTCGGGCCACATCAC; and the ␤-actin probe was from Clontech. The purified fragments were labeled with 32 P using the RadPrime DNA Labeling System (Invitrogen).
Keratinocytes and Other Cells-Human epidermal keratinocytes, isolated from neonatal foreskin, and human umbilical vein endothelial cells (both from Cascade Biologics) were grown in EpiLife medium with Human Keratinocyte Growth Supplement or Medium 200 with Low Serum Growth Supplement, respectively (Cascade Biologics), and used between the 3 rd and 5 th passage. Before each experiment, the cells were maintained in medium without the supplement for 18 -24 h. The human U373 astrocytoma cell line was cultured in RPMI 1640 medium with 10% fetal calf serum (HyClone). Human peripheral blood monocytes from normal healthy donors were obtained and cultured as described previously (46). Cells were stimulated for 4 h with 2 ϫ 10 8 /ml of heat-killed (70°C, 30 min) B. subtilis or E. cloacae.
Other Methods-Soluble S. aureus peptidoglycan or insoluble S. aureus or B. subtilis peptidoglycan, S. aureus wall teichoic acid, and E. faecalis lipoteichoic acid (LTA) were purified as described (40,41). B. subtilis and S. aureus LTA, and ultrapure E. coli lipopolysaccharide (LPS) were from InvivoGen. ReLPS (LPS from Salmonella minnesota Re mutant), dextran sulfate (M r 50,000), poly(IC), ␤-glucan (laminarin from Laminaria digitata), mannan, protein A, protein G, and micro-granular cellulose were from Sigma. Assays for hydrolysis of soluble and insoluble S. aureus and B. subtilis peptidoglycans and heat-killed S. aureus and B. subtilis bacteria by PGLYRPs were performed as described (35). Binding of PGLYRPs to bacteria was performed at 4°C for 2 h as described (18) with 1 g of PGLYRPs and 15 g of bacteria in 500 l, the buffer that was used for the killing assay was used as the binding buffer, and the same buffer with 0.3 M NaCl as the washing buffer. Inhibition of PGLYRPs binding to bacteria was tested by preincubating 1 g of PGLYRPs with 100 g of inhibitors for 30 min at 4°C before adding bacteria.

RESULTS
Bactericidal Activity of PGLYRP1, PGLYRP3, and PGLYRP4-Human PGLYRP1, PGLYRP3, and PGLYRP4 were secreted into the medium primarily as disulfide-linked homodimers, and cells that were cotransfected with human PGLYRP3 and PGLYRP4 exclusively secreted disulfide-linked PGLYRP3:4 heterodimers (Fig. 1A). Co-expression of both PGLYRP3 and PGLYRP4 in the same cells was required for the formation of heterodimers, because mixing of PGLYRP3 and PGLYRP4 individually expressed in separate cells or mixing of cells separately transfected with PGLYRP3 and PGLYRP4 did not result in formation of heterodimers (not shown). These dimers formed both in insect S2 (Fig. 1A) and in mammalian COS-7 (not shown) cells transfected with PGLYRPs.
Purified human PGLYRP1, PGLYRP3, PGLYRP4, and PGLYRP3:4 had strong bactericidal activity for Gram-positive bacteria and different PGLYRPs had different spectra of activity (Fig. 2). B. subtilis was highly sensitive to killing by PGLYRP3, PGLYRP4, and PGLYRP3:4 (bacterial viability decreased by 5 logs in 1 h) and was less sensitive to PGLYRP1. B. cereus and B. licheniformis were most sensitive to killing by PGLYRP3:4 (viability decreased by 5 logs in 1-3 h). L. acidophilus was highly sensitive to killing by PGLYRP1, PGLYRP3, and PGLYRP4 (viability decreased by 5 logs in 1 h) and much less sensitive to PGLYRP3:4. All these bacilli are nonpathogenic Gram-positive transient (but not normal) flora.
Pathogenic Gram-positive bacteria were also killed by PGLYRPs. L. monocytogenes was highly sensitive to killing by PGLYRP3 and PGLYRP4 (viability decreased by 5 logs in 1-2 h), less sensitive to PGLYRP3:4, and not sensitive to PGLYRP1. S. aureus was highly sensitive to killing by PGLYRP3 (viability decreased by 5 logs in 2 h), and somewhat less sensitive to PGLYRP4, PGLYRP3:4, and PGLYRP1. To further evaluate the sensitivity of clinically relevant S. aureus isolates, we tested the sensitivity of seven S. aureus strains to PGLYRPs. All seven S. aureus strains were most sensitive to PGLYRP3 and PGLYRP4, but the extent of their sensitivity varied. Under the conditions of Fig. 2  , and there was no correlation with the sensitivity to PGLYRPs and antibiotics (methicillin, MRSA), presence of capsule, and ␣-toxin production. These results indicate that S. aureus may acquire some degree of resistance to PGLYRPs (although we have not found a completely resistant S. aureus strain). This increased resistance may be related to the ability of S. aureus to colonize areas of the body that express PGLYRPs (e.g. nasal carrier rate for S. aureus in some populations, such as in hospitals, can be as high as 30%).
By contrast, normal flora Gram-positive bacteria S. epidermidis and M. luteus (normal skin flora), E. faecalis (normal intestinal flora), and S. MARCH 3, 2006 • VOLUME 281 • NUMBER 9 JOURNAL OF BIOLOGICAL CHEMISTRY 5897 agalactiae (often present in vagina and genitourinary tract) were much less sensitive to killing by PGLYRPs (Fig. 2). S. pyogenes, which is a pathogenic Gram-positive bacterium with a high carrier rate in the throat, was also much less sensitive to killing by PGLYRPs (1 log decrease in viability, compared with 5 log for S. aureus and other sensitive bacteria, Fig. 2). Gram-negative bacteria, E. coli (Fig. 1), P. vulgaris, and E. cloacae (not shown), and a yeast, C. albicans (not shown), were not killed by any of the PGLYRPs. PGLYRPs were bacteriostatic for all bacteria, but not for C. albicans.
PGLYRPs Are Active in Vivo and Require Ca 2ϩ and N-Glycosylation-Because PGLYRP3 and PGLYRP4 are secreted from cells and expressed on the skin and mucous membranes, we tested whether PGLYRP3 was active in vivo at its likely site of action. We tested whether S. aureus lung infection in a mouse model could be prevented by sequentially applying PGLYRP3 and bacteria into the nose, the normal portal of entry of bacteria. Instilling 4.5 g of PGLYRP3 intranasally 15 min before intranasal challenge with 3 ϫ 10 8 S. aureus reduced 33 times the number of S. aureus recovered from the lungs 4 h later (Fig. 3B, p ϭ 0.038). Instilling PGLYRP4 reduced the number of S. aureus in the lungs 12 times, whereas instilling heat-inactivated PGLYRP3 or an unrelated protein (recombinant Fc IgG fragment) had no effect (not shown). These results demonstrate that PGLYRP3 and PGLYRP4 are active in vivo in preventing S. aureus lung infection in mice.
The bactericidal activity of PGLYRPs required Ca 2ϩ . The PGLYRP proteins were bactericidal when purified and assayed for bactericidal activity in the presence of Ca 2ϩ (Figs. 2, 3, and 5), but not when purified and assayed without Ca 2ϩ , and Ca 2ϩ could not be replaced by Mg 2ϩ (not shown). Also, the bactericidal activity of PGLYRPs was abolished by a Ca 2ϩ chelator, EGTA (Fig. 3C).
All four PGLYRPs were N-glycosylated, as their relative molecular mass decreased by 4 -10 kDa following treatment with N-glycosidase (PGLYRP3 and PGLYRP4, Bactericidal Activity of PGLYRPs Depends on Interaction with Cell Wall Peptidoglycan and Not Membrane Permeabilization-PGLYRPs have a peptidoglycan-binding groove that is specific for muramyl tripeptide (35,47) and, therefore, their bactericidal action likely involves mass was because of deglycosylation rather than proteolytic degradation. C, specificity of anti-PGLYRP3 and anti-PGLYRP4 antibodies. Rabbit polyclonal antibodies, generated against PGLYRP3-derived or PGLYRP4-derived peptides react specifically on a Western blot (WB) (reducing conditions) with human PGLYRP3 or PGLYRP4, respectively, expressed in S2 cells, and do not cross-react with other human PGLYRPs. PGLYRP1 was also expressed in S2 cells and PGLYRP2 was expressed in E. coli with an S, rather than V5, tag. The reactivity of all four recombinant PGLYRPs with anti-V5 tag or anti-S tag antibodies is shown in the right panels. 500 ng/lane of each protein was used. The bands detected on Western blots with anti-tag and anti-PGLYRP antibodies corresponded to the bands on Coomassie Blue-stained gels. -kDa dimers and their apparent molecular mass is reduced 4 -10 kDa by N-glycosidase, whereas similar treatment of bovine serum albumin, which was not N-glycosylated, did not diminish its molecular mass, which confirms that the decrease in interaction with peptidoglycan. Indeed, killing of bacteria was greatly inhibited in the presence of peptidoglycan added to the bacteria/PG-LYRP mixture (Fig. 3E), whereas LTA only partially inhibited killing and other bacterial cell wall components (wall teichoic acid, protein A, protein G, and LPS) did not inhibit the bacterial killing (not shown). These results indicate that killing of bacteria involves binding of PGLYRPs to peptidoglycan.
We next determined whether binding of PGLYRPs to bacteria correlates with the bacteriostatic or bactericidal activity of these proteins, which bacteria bind PGLYRPs, and which bacterial components are involved in this binding. PGLYRP3, PGLYRP4 (Fig. 4A), and PGLYRP3:4 and PGLYRP1 (not shown) bound to a variety of both Gram-positive and Gram-negative bacteria and a fungus, C. albicans (and to a much lower extent to Saccharomyces cerevisiae). Gram-positive bacteria usually bound more PGLYRPs than Gram-negative bacteria. PGLYRPs did not bind to microgranular cellulose. The binding of PGLYRP3, PGLYRP4 (Fig. 4B), and PGLYRP3:4 and PGLYRP1 (not shown) to both Gram-positive and Gram-negative bacteria was com- pletely inhibited by exogenous peptidoglycan, and partially inhibited by LTA. Binding to Gram-negative bacteria was also partially inhibited by smooth LPS. Binding to bacteria was not inhibited by wall teichoic acid, ReLPS, dextran sulfate, poly(IC), ␤-glucan, mannan (Fig. 4B), and protein A and protein G (not shown). These results indicate that binding of PGLYRPs to bacteria is specific and has the highest affinity for pepti- doglycan, low affinity for LTA and smooth LPS, and that this binding is not based on simple charge interaction, because highly negatively charged polymers (dextran sulfate and poly(IC)) did not inhibit the binding of PGLYRP to bacteria. Binding of PGLYRPs to both Grampositive and Gram-negative bacteria is consistent with the bacteriostatic effect of PGLYRPs for these bacteria. These results indicate that peptidoglycan is the preferred ligand for human PGLYRPs in bacteria, that LTA and LPS can also bind PGLYRPs with lower affinity, and that binding to bacteria is required for antibacterial activity of PGLYRPs. Because PGLYRPs are bactericidal for only some bacteria and because exogenous peptidoglycan inhibits their bactericidal activity, binding of PGLYRPs to peptidoglycan in the bacterial cell wall is likely required for the cidal effect, but may involve additional post-binding events.
We analyzed the mechanism for killing of bacteria by PGLYRPs by comparing it with the mechanism of bacterial killing by enzymes that lyse peptidoglycan, antibiotics that inhibit peptidoglycan synthesis, and peptides that permeabilize bacterial membranes. The function of peptidoglycan in the bacterial cell wall is to counteract the high osmotic pressure of bacterial protoplast. Lysostaphin (peptidoglycan-lytic enzyme) rapidly kills S. aureus because of loss of peptidoglycan integrity and osmotic lysis (Fig. 5A, no sucrose). This osmotic lysis results in rapid membrane permeabilization, and the kinetics of membrane permeabi-  lization correlates with the kinetics of bacterial killing, because killing is caused by osmotic lysis of bacteria (Fig. 5A, no sucrose). Penicillin kills bacteria by inhibiting peptidoglycan synthesis; however, penicillin does not cause early permeabilization of bacterial membranes (Fig. 5B, no sucrose), because penicillin-treated bacteria only die when they start to grow and are unable to synthesize peptidoglycan. Therefore, in penicillin-killed bacteria membrane permeabilization is substantially delayed until the bacteria have grown without synthesizing peptidoglycan. Vertebrate antimicrobial peptides, such as magainin (Fig. 5C), kill bacteria by traversing cell wall and permeabilizing cell membrane (1-4). Therefore, killing of bacteria by magainin is characterized by early membrane permeabilization, and kinetics of membrane permeabilization correlate with the kinetics of bacterial killing assayed by colony counts (Fig. 5C). The kinetics of bacterial killing and membrane permeabilization by PGLYRPs resembled those by penicillin: PGLYRPs also rapidly killed bacteria (measured by colony counts), but did not cause permeabilization of their membranes (Fig. 5, D and E).
When bacteria lose their cell wall peptidoglycan because of digestion with an enzyme or because of inhibition of peptidoglycan synthesis by antibiotics, such as penicillin, they can be kept alive in the protoplast medium containing 0.75 M sucrose, and when they are diluted and plated in the protoplast medium, they can rebuild their cell walls. Consistent with this model, killing of S. aureus by lysostaphin, penicillin, and also by PGLYRPs was prevented in the protoplast medium containing 0.75 M sucrose (Fig. 5), and, as expected, none of these agents caused membrane permeabilization, which confirms that they all act on the cell wall and not on the cell membrane (note that treatment with lysostaphin in 0.75 M sucrose does not result in permeabilization of the cell membrane, because sucrose protects the cells from osmotic lysis). Lysostaphin still effectively digested peptidoglycan in 0.75 M sucrose (as shown by microscopy and immediate cell death upon transfer of bacteria into medium without sucrose), and PGLYRPs still bound to peptidoglycan in 0.75 M sucrose (as shown in a binding assay). By contrast, magainin induced rapid membrane permeabilization and bacterial killing in 0.75 M sucrose (Fig. 5C), confirming that magainin targets and permeabilizes bacterial cytoplasmic membrane.
Because PGLYRPs are large proteins that may not easily traverse the bacterial cell wall and reach the cell membrane, we next wanted to exclude the possibility that the lack of membrane permeabilization by PGLYRPs was because of their limited access to the cell membrane. However, increasing access to the cell membrane by digestion of the cell wall peptidoglycan with lysostaphin increased membrane permeabilization only by magainin, but did not result in any membrane permeabilization by PGLYRPs or penicillin (Fig. 5). PGLYRPs still bound to lysostaphin-treated bacterial cells (not shown), because lysostaphin only partially removes peptidoglycan (creates spheroplasts, rather than true protoplasts). PGLYRPs also had no toxicity to mammalian (COS-7) or insect (S2) cells (not shown), confirming that they do not target bacterial or eukaryotic membranes.
We next considered whether bactericidal action of PGLYRPs was because of their possible bacteriolytic activity or hydrolytic activity toward peptidoglycan. All PGLYRPs have an amidase homology domain, but out of four mammalian PGLYRPs only PGLYRP2 has an N-acetylmuraloyl-L-alanine amidase activity, i.e. it hydrolyzes the amide bond between the lactic group of N-acetylmuramic acid and L-Ala of the peptidoglycan stem peptide (34,35). However, mammalian PGLYRP2 only digest isolated peptidoglycan and does not efficiently digest intact bacteria or peptidoglycan in intact bacterial cell walls (34,35). PGLYRP1, PGLYRP3, and PGLYRP4 lack the Zn 2ϩ -binding amino acids, which are required for amidase activity of PGLYRP2 (34,35), and PGLYRP1, PGLYRP3, and PGLYRP4 expressed in COS-7 cells do not have any detectable amidase activity (35). However, because PGLYRPs expressed in COS-7 cells were obtained in the absence of Ca 2ϩ and did not have bactericidal activity (35), we re-tested whether PGLYRP1, PGLYRP3, PGLYRP3:4, and PGLYRP4 proteins, which were purified from S2 cell supernatants in the presence of Ca 2ϩ and were bactericidal, had peptidoglycan-lytic activity. We did not detect any amidase activity or any other peptidoglycan-lytic or bacteriolytic activity of these bactericidal PGLYRPs using polymeric soluble S. aureus peptidoglycan, polymeric insoluble S. aureus or B. subtilis peptidoglycan, or intact killed S. aureus or B. subtilis bacteria within the time frame and conditions under which they killed 99% of bacteria (not shown).
These results are consistent with the results on the kinetics of bacterial killing and membrane permeabilization: peptidoglycan-lytic lysostaphin caused rapid killing of S. aureus because of osmotic lysis that correlated with membrane permeabilization of osmotically lysing bacteria, both of which were fully prevented in protoplast medium with 0.75 M sucrose (Fig. 5A). By contrast, killing of S. aureus by PGLYRPs was not accompanied by early membrane permeabilization (Fig. 5, C and D), which would have been observed if bacteria were being killed because of digestion of peptidoglycan and osmotic lysis.
To confirm the above results with another PGLYRP-sensitive bacterial species, we performed similar experiments (to those shown in Fig. 5) on the kinetics of bacterial killing and membrane permeabilization using B. subtilis treated with muramidases (mutanolysin and lysozyme, instead of lysostaphin, because lysostaphin only hydrolyzes S. aureus, but not B. subtilis peptidoglycan), penicillin, magainin, and PGLYRPs. The results were similar to those shown in Fig. 5 with S. aureus: peptidoglycan-lytic enzymes caused rapid killing of B. subtilis because of osmotic lysis that correlated with membrane permeabilization of osmotically lysing bacteria, both of which were prevented in the protoplast medium with 0.75 M sucrose; penicillin caused rapid killing of B. subtilis that was prevented in 0.75 M sucrose and was not accompanied by early membrane permeabilization; magainin caused rapid killing and membrane permeabilization that was not prevented in 0.75 M sucrose; and PGLYRPs caused rapid killing that, similarly to penicillin, was not accompanied by early membrane permeabilization (results not shown). Therefore, these results with B. subtilis and those in Fig. 5 with S. aureus both show that the mechanism of bacterial killing by PGLYRPs is different from the mechanism of killing by peptidoglycan-lytic enzymes and membrane-permeabilizing antibacterial peptides, and that it resembles the effect of antibiotics that inhibit peptidoglycan synthesis.
Many antibiotics inhibit peptidoglycan synthesis by interacting with peptidoglycan-synthesizing enzymes, because they either irreversibly bind to the active site of peptidoglycan-synthesizing enzymes (␤-lactams) or are analogs of peptidoglycan fragments or precursors. Because PGLYRPs bind to peptidoglycan (15-19, 31, 34, 35, 47), they are unlikely to also bind to the peptidoglycan-synthesizing enzymes or to serve as analogs of peptidoglycan fragments. However, the avid binding of PGLYRP to peptidoglycan or peptidoglycan biosynthetic precursors (e.g. GlcNAc-MurNAc-pentapeptide) could inhibit access to peptidoglycan or its precursors of enzymes that participate in cell wall synthesis, especially peptidoglycan hydrolases that need to digest the existing peptidoglycan to allow incorporation of the newly synthesized peptidoglycan into the existing cell wall, or transglycosylases and transpeptidases that link newly synthesized peptidoglycan to the existing peptidoglycan in the cell wall. This mechanism would be similar to the effects of antibiotics such as vancomycin, mersacidin, or actagardine, which inhibit peptidoglycan synthesis by binding to peptidoglycan or its metabolic precursors, whose action is inhibited by exogenous peptidoglycan or its fragments, and that work at higher concentrations than the antibiotics that inactivate peptidoglycan-synthesizing enzymes (48,49). Our results support this hypothesis, because PGLYRPs inhibited initial digestion of peptidoglycan by lysostaphin, which caused inhibition of early cell permeabilization by peptidoglycan-lytic enzymes (compare membrane permeabilization by 1 g/ml of lysostaphin alone to lysostaphin plus PGLYRPs in Fig. 5, B, D, and E, no sucrose). This inhibition is likely because of competition of PGLYRPs and peptidoglycan-lytic enzymes for binding to peptidoglycan. However, despite this initial inhibition of lysostaphin-mediated peptidoglycan digestion and lysostaphin-induced osmotic lysis (permeabilization) by PGLYRPs, the eventual killing of bacteria (measured by colony count) by a combination of lysostaphin and PGLYRPs was enhanced compared with the killing by each agent alone (Fig. 5, A, D, and E). Moreover, killing of bacteria by a combination of lysostaphin and PGLYRPs was not prevented by 0.75 M sucrose, in contrast to complete prevention of bacterial killing by each agent alone in 0.75 M sucrose (Fig. 5, A, D, and E), although under these conditions there was still no early permeabilization of bacterial membranes, confirming that neither PGLYRPs nor lysostaphin directly permeabilize cell membrane. These results further support the action of both lysostaphin and PGLYRPs on the cell wall. Moreover, the synergistic effect of lysostaphin and PGLYRPs also indicates that lysostaphin and PGLYRPs have different mechanisms of action (peptidoglycan-   A, B, D, and E) and pars plana (PP in A,  B, G, and H), and to a much lower extent in the flat superficial layer of corneal epithelium (J and K). Immunoperoxidase/hematoxylin staining with anti-PGLYRP3 (A, D, G, and J) or anti-PGLYRP4 (B, E, H, and K) or control (C, F, I, and L) antibody is shown. Eye pigment granules are visible in control antibody-stained panels in the iris (IR), ciliary body pars plana epithelium, and to a much lower extent in the ciliary process of ciliary body, and immunoperoxidase staining for PGLYRP3 and PGLYRP4 is superimposed over the natural pigment. Bar ϭ 50 m. lytic and non-lytic mechanism); bacteria are unable to re-synthesize their peptidoglycan after digestion with lysostaphin and after treatment with PGLYRPs. Altogether these results indicate that PGLYRP1, PGLYRP3, and PGLYRP4 kill bacteria by interacting with their cell wall peptidoglycan, but they do not hydrolyze peptidoglycan and do not permeabilize cell membranes.
PGLYRP3 and PGLYRP4 Are Expressed in the Skin, Eyes, Oral Cavity, and Gastrointestinal Tract-pglyrp3 and pglyrp4 genes are located in the epidermal differentiation complex gene cluster on chromosome 1q21 (18), which encodes genes coordinately expressed in specialized epithelial cells, such as keratinocytes. Both human PGLYRP3 and PGLYRP4 were highly expressed in epidermal keratinocytes (Fig. 6, A-D, but not in the devitalized keratin layer), eccrine sweat glands and ducts (Fig. 6, A-E), hair follicles (Fig. 6, F and G), sebaceous glands (Fig.  6, F and G), the ciliary body epithelium of the eye (Fig. 7, A-I), and to a much lower extent in the superficial layer of corneal epithelium (Fig. 7, J-L). PGLYRP4, but not PGLYRP3, was highly expressed in the mucous cells in the submandibular salivary gland and secretory glands in the throat (Fig. 8, which secrete mucus), but neither protein was expressed in serous cells of the submandibular salivary gland or throat (Fig. 8), or in the parotid salivary gland (not shown), which has only serous cells (which secrete enzymes). Both PGLYRP3 and PGLYRP4 were highly expressed in mature epithelial cells in the tongue and esophagus (Fig. 9), in gastric glands in the stomach (PGLYRP3 mainly in parietal cells, which secrete acid, and PGLYRP4 mainly in neck mucous cells, which secrete glycoproteins, Fig. 10), and in apical columnar absorptive epithelial cells in the small intestine and colon (Fig. 11), but not in Paneth cells (which produce antimicrobial peptides and lysozyme, and which were identified at the bottom of the crypts by their morphology and immunoreactivity with anti-lysozyme antibodies), goblet cells (which secrete mucus), endocrine cells, or lymphoid nodules.
Human tissues that expressed PGLYRP3 and PGLYRP4 proteins also expressed PGLYRP3 and PGLYRP4 mRNA (Fig. 12). The expression of PGLYRP3 and PGLYRP4 mRNA in keratinocytes was increased following exposure to bacteria (Fig. 12, A and C). In contrast, exposure of   human primary endothelial cells and other non-epithelial cells (U373 astrocytoma cells or blood monocytes) to bacteria or other stimuli (cytokines interleukin-1␤ and tumor necrosis factor-␣, or insulin-like growth factor and transforming growth factor-␣) did not induce expression of the PGLYRP3 and PGLYRP4 mRNAs, despite high induction of interleukin-6 mRNA (not shown). PGLYRP3 and PGLYRP4 are thus selectively expressed in epithelial cells.

DISCUSSION
Here we have identified the function of mammalian PGLYRP3 and PGLYRP4, and we demonstrate that, together with PGLYRP1, they form a new class of bactericidal proteins that have a different structure, mechanism of action, and expression than the currently known mammalian antimicrobial peptides. These PGLYRPs are bactericidal both in vitro and in vivo for pathogenic and nonpathogenic Gram-positive bacteria, but not for normal flora bacteria, and they are bacteriostatic for most Gram-positive and Gram-negative bacteria. Thus, our results suggest that normal flora bacteria have developed resistance to bactericidal mechanisms constitutively present in the skin, eyes, and mucous membranes (such as PGLYRP3 and PGLYRP4), and can colonize these areas. The bacteriostatic effect of PGLYRPs on normal flora bacteria makes perfect sense for host physiology: normal flora bacteria are not killed, but their overgrowth is limited. Nonpathogenic not normal flora bacteria (transient flora, such as Bacillus and Lactobacillus) are uniformly sensitive to killing by PGLYRPs. L. monocytogenes, a pathogen that does not infect skin and mucous membranes, is also highly sensitive. S. aureus, a pathogen that often infects skin, has intermediate sensitivity to killing by PGLYRPs, with some strains highly sensitive and some less sensitive, whereas S. pyogenes, a pathogen that frequently infects throat and skin and has a high carrier rate in the throat, is more resistant to killing by PGLYRPs. These results demonstrate how normal flora adapt to their environment and how successful pathogens evade the immune system at the site of infection.
We have demonstrated that PGLYRP1, PGLYRP3, and PGLYRP4 are secreted and form disulfide-linked homo-or heterodimers, which have different spectra of bactericidal activity. Secretion of PGLYRPs is consistent with the recent crystallographic model of PGLYRP3, which shows that PGLYRP3 is not a transmembrane protein (47), as proposed before (18,19). Human PGLYRP3 and PGLYRP4 both have two PGRP domains, and each PGRP domain has one binding site for peptidoglycan (47), thus, PGLYRP3 and PGLYRP4 monomers and dimers have 2 and 4 peptidoglycan binding sites, respectively. However, because these PGRP domains are not identical (have 37-42% identity), the fine binding specificity or affinity of each PGRP domain in each PGLYRP molecule is likely different. The diversification of the specificities of PGLYRP is then further increased by formation of PGLYRP3:4 heterodimers, so the host can fine tune the specificities of PGLYRPs by expressing PGLYRP3 and PGLYRP4 either in the same or separate cells, to form hetero-or homodimers, respectively. In addition, PGLYRPs have hydrophobic domains on the opposite side of the molecule from the peptidoglycanbinding groove, which were previously hypothesized to interact with signal transduction molecules (29). In mammalian PGLYRPs, however, these hydrophobic domains are likely to play a role in their interaction with bacteria. FIGURE 12. Human PGLYRP3 and PGLYRP4 mRNA expression is increased by exposure to bacteria. PGLYRP3 and PGLYRP4 mRNA are constitutively expressed in esophagus, skin, tongue, cornea, and cultured keratinocytes, but not in liver (negative control), in contrast to PGLYRP2 mRNA, which is constitutively expressed in liver, but not in the esophagus, skin, tongue, cornea, and cultured keratinocytes, determined by Northern blot (A) and real-time RT-PCR (B and C). Exposure of cultured keratinocytes to 2 ϫ 10 8 bacteria/ml for 4 h increases expression of PGLYRP3 and PGLYRP4 mRNA and induces expression of PGLYRP2 mRNA (A and C). The results in B and C are mean Ϯ S.E. of three to four experiments, calculated by the comparative C T method relative to endogenous 18 S RNA, and expressed as ratios to the lowest amount of RNA out of all groups, which was assigned a value of 1.
PGLYRP3 and PGLYRP4 proteins are selectively expressed in tissues that come in contact with the environment: in the skin epidermis, hair follicles, sebaceous glands, and eccrine sweat glands; in the ciliary body of the eye (which produces aqueous humor that fills anterior and posterior chambers of the eye, the spaces between cornea and the iris, and the iris and lens); in the corneal epithelium of the eye; in the mucussecreting cells of the main salivary (submandibular) gland and in mucus-secreting glands in the throat (both mucus-secreting glands selectively express PGLYRP4, but not PGLYRP3); in the tongue and esophagus in squamous epithelial cells; in the stomach in acid-secreting Parietal cells (PGLYRP3) and glycoprotein-secreting neck mucous cells (PGLYRP4); and in the small and large intestine in the columnar absorptive cells, but not in mucus-secreting goblet cells and not in the crypts in Paneth cells, which produce antimicrobial peptides, such as defensins, phospholipase A 2 , and lysozyme (1)(2)(3)(4). Mouse orthologues of human PGLYRP3 and PGLYRP4 are also expressed in the intestinal tract and salivary glands (50). Therefore, several types of tissues and cells that express PGLYRP3 and PGLYRP4 are different from the tissues and cells that typically express antimicrobial peptides (1)(2)(3)(4).
PGLYRP3 and PGLYRP4 also differ from antimicrobial peptides in their size and mechanism of action. They are much larger than antimicrobial peptides (from 89 to 115 kDa disulfide-linked dimers) and, unlike vertebrate antimicrobial peptides, they do not kill bacteria by permeabilizing their membranes, but rather by interacting with their cell wall peptidoglycan. Because the affinity of monomeric PGLYRP1 for peptidoglycan is 13 nM (31), the avidity of binding of PGLYRP3, PGLYRP4, or PGLYRP3:4 dimers, which have four peptidoglycan binding sites, is expected to be very high and, because of their large size, PGLYRPs likely interfere with the normal function of peptidoglycan and cell growth.
The requirement for Ca 2ϩ for bactericidal activity of human PGLYRPs may explain why previously mouse (31,33) or human (51) recombinant PGLYRP1, purified without Ca 2ϩ , was only bacteriostatic, but not bactericidal. The bovine PGLYRP1 orthologue purified from PMN granules was bactericidal (32), and thus probably retained Ca 2ϩ that was bound to it in vivo.
In conclusion, mammalian PGLYRPs function as both recognition and effector molecules: three PGLYRPs (PGLYRP3, PGLYRP4, and PGLYRP1) are bactericidal, and are produced in epithelial cells, body secretions, or polymorphonuclear leukocytes, where they are poised to kill bacteria. They form a new class of bactericidal proteins that have a different structure, mechanism of action, and expression than currently known mammalian antimicrobial peptides. The fourth mammalian PGLYRP, PGLYRP2, is a peptidoglycan-lytic enzyme (34 -36). By contrast, several insect PGRPs function as bacterial recognition molecules that trigger production of antimicrobial peptides, melanin, or reactive oxygen species, but there are no insect PGRPs with known direct microbicidal activity. Thus, direct bactericidal activity of PGRPs either evolved in vertebrates or mammals, or is yet to be discovered in insects.