Development of a Novel Backbone Cyclic Peptide Inhibitor of the Innate Immune TLR/IL1R Signaling Protein MyD88

MyD88 is a cytoplasmic adaptor protein that plays a central role in signaling downstream of the TLRs and the IL1R superfamily. We previously demonstrated that MyD88 plays a critical role in EAE, the murine model of multiple sclerosis, and showed that the MyD88 BB-loop decoy peptide RDVLPGT ameliorates EAE. We now designed and screened a library of backbone cyclized peptides based on the linear BB loop peptide, to identify a metabolically stable inhibitor of MyD88 that retains the binding properties of the linear peptide. We identified a novel cyclic peptide protein mimetic that inhibits inflammatory responses to TLR ligands, and NFκB activation in response to IL-1 activation. The inhibitor, c(MyD 4-4), is metabolically stable in comparison to the linear peptide, blocks MyD88 in a specific manner, and inhibits MyD88 function by preventing MyD88 dimerization. Finally, treatment of mice with c(MyD 4-4) reduced the severity of clinical disease in the murine EAE model of multiple sclerosis. Thus, modulation of MyD88-dependent signaling using c(MyD 4-4) is a potential therapeutic strategy to lower innate immune inflammation in autoimmune CNS disease.

alpha helix termed the "BB loop" 18,19 . The heptameric peptide RDVLPGT is the target site in the BB loop 20 and three amino acids (Arginine, Aspartic acid and Proline) are highly conserved across all TIR containing proteins (the absence of Proline in TLR3 may explain the MyD88-independent signaling of this receptor) 21 . The MyD88 BB loop governs dimerization of MyD88, an essential step for initiation of oligomerization, myddosome formation, and downstream inflammatory signaling 22,23 . The heptameric BB loop peptide competitively inhibits MyD88 homodimerization 20 and ameliorates inflammatory disease in several animal models 16,[24][25][26][27] , suggesting that targeting the BB loop is a feasible way to inhibit MyD88 signaling in vivo [28][29][30] .
Linear peptides are rapidly degraded by intestinal and plasma peptidases and are not orally bioavailable due to metabolizing enzymes in the intestinal lumen [31][32][33] . Small molecule peptidomimetics are usually metabolically stable, however the naturally selected properties of the peptide that confer target specificity and binding properties can only be approximated 34,35 . We developed a metabolically stable version of the MyD88 decoy peptide via backbone cyclization 36 . Backbone cyclization of peptides takes advantage of the naturally selected structure of the peptide while providing a ring element that decreases conformational freedom and sensitivity to metabolizing enzymes. We show below that the cyclic MyD88 BB loop peptide is more bioactive than the linear peptide and is resistant to degradation in the intestine and plasma. The cyclic peptide inhibitor blocks human and mouse TLR2 and TLR4 stimulation, inhibits MyD88 dimerization, and reduces clinical disease in EAE.

Results
Backbone cyclized derivatives of the MyD88 BB loop linear peptide RDVLPGT. MyD88 function depends on dimerization through the TIR domain 37 . The RDVLPGT peptide corresponds to the region between the βB strand and the αB helix 18 (the "BB loop") and competitively inhibits MyD88 dimerization and function 20 . We screened a library of backbone cyclized peptides to identify a metabolically stable derivative of the BB loop peptide that retains the binding and functional properties of the parent peptide. The library was designed using two non-natural building blocks as previously reported 38 . The method, called cycloscan 39 , is based on the concept of screening focused backbone cyclic libraries with spatial diversity that maintain the original side chains of the active region in the protein (RDVLPGT), Backbone cyclization combines N-alkylation with cyclization. In the case of the MyD88 BB-loop decoy peptide RDVLPGT, the tertiary amide bond preceding proline served as one of the anchor sites and was therefore replaced by an N-Alkylated glycine building unit (AGBU) to allow backbone cyclization to the second AGBU at the N-terminus (Fig. 1). Tryptophan was added to the N-terminal AGBU for quantification purposes (Fig. 1). Following the above procedure 16 backbone cyclic peptides were prepared with various ring size and position of the urea bridge in the ring (Table S1). To screen the library, human THP-1 macrophages were pre-treated with the cyclic peptides and then stimulated with Pam3CSK4, a synthetic TLR2/1 lipopeptide agonist. The two compounds that inhibited production of hTNFa strongest, without causing cytotoxicity, were composed of either 4 carbons on each arm (c(MyD 4-4)) or 6 carbons on each arm (c(MyD6x6)). These

Metabolic stability of c(MyD 4-4).
Metabolic instability is one of the major drawbacks of linear peptides 31,32 . Peptide metabolism in the plasma is essential for regulation of important physiological processes of hormones, antibodies and other enzymes 40 . Additionally, peptide metabolizing enzymes in the intestinal lumen degrade dietary proteins to tri/dipeptides and single amino-acids 33 . We compared the stability of the linear RDVLPGT inhibitory peptide (MyDI) to the cyclic c(MyD 4-4) (Fig. 2). Both linear and cyclic MyDI peptides were incubated with either rat plasma or brush border membrane vesicles (BBMVs) derived from the lumen of rat intestines 41,42 . The linear peptide was rapidly degraded in the presence of plasma or BBMVs. In contrast, c(MyD 4-4) was highly stable in plasma and at 240 minutes >90% of the starting material was detected ( Fig. 2A). Similarly, roughly 80% of the c(MyD 4-4) was recovered at the end of the 2 h incubation with BBMVs, arguing that cyclization stabilizes the peptide in the harsh enzymatic environment of the intestinal lumen (Fig. 2B). c(MyD 4-4) blocks human and mouse macrophage TLR2 and TLR4 stimulation. We next compared the activity of the c(MyD 4-4) compound to the linear BB loop peptide, and to a commercially available small molecule inhibitor of MyD88, ST2825 43 . Incubation with c(MyD 4-4) blocked human macrophage cytokine production in response to TLR2 stimulation (Pam3CSK4) significantly more than the linear MyDI peptide (Fig. 3A), and was also more inhibitory than the MyDI peptide when mouse macrophages carrying an NFκB-luciferase reporter gene were activated by Pam3CSK4 or LPS, a TLR4 ligand (Fig. 3B,C). Surprisingly, ST2825 did not block human macrophage cytokine production when macrophages were stimulated with Pam3CSK4 or LPS (Fig. 3D,E). ST2825 was highly cytotoxic to RAW264.7 NFκB-luc cells (Fig. 3F-H), precluding comparison to c(MyD 4-4). Thus, c(MyD 4-4) is a metabolically stable cyclic peptide inhibitor of TLR2/4 driven inflammatory signaling.

c(MyD 4-4) and the linear MyDI peptide bind to the same region of the MyD88 TIR domain. The
BB loop is a highly conserved region of the MyD88 TIR domain. In particular, the proline at position P200 of MyD88 is highly conserved through evolution and is present in TIR domains of adaptor proteins and all TLRs except TLR3 21 . Since backbone cyclization opened the pyrrolidine ring of the proline side chain, the cyclization could disrupt the structure of the peptide and alter its binding properties compared to the linear peptide. We therefore tested by competition if c(MyD 4-4) binds to the same region of the TIR domain as the linear BB loop peptide. To this end, we produced recombinant MyD88 TIR domain (aa 150-296 of the MyD88 protein) as a fusion protein to SUMO 44 and attached it to ELISA plate wells (Fig. S2A). Biotinylated linear BB loop peptide (Biotin-MyDI) bound the TIR domain in a concentration dependent manner (Fig. S2B) that was inhibited by

c(MyD 4-4) blocks MyD88 dimerization.
MyD88 dimerization is mediated through the BB loop region 45,46 and inhibition of dimerization is considered to be the mechanism by which the MyDI linear peptide blocks TLR signaling. We therefore tested if the c(MyD 4-4) has a similar effect on MyD88 dimerization in living cells. To test dimerization, we co-transfected HEK 293 cells with HA-MyD88 and Flag-MyD88. MyD88 dimerization was determined by the detection of Flag-MyD88 following immunoprecipitation of HA-MyD88. Transfection and co-transfection was confirmed by Western Blot analysis (Fig. 4B). 48 hr after co-transfection cells were incubated with different concentrations of c(MyD 4-4), and stimulated for 30 min with IL-1β to activate MyD88. Cells were then lysed, and MyD88 was immunoprecipitated using anti-HA antibody followed by immunoblot analysis with anti-Flag antibody. MyD88 homodimerization was strongly inhibited by c(MyD 4-4) in a concentration dependent manner (Fig. 4C,D).  determined by tracking the translocation of p65 from the cytoplasm to the nucleus in response to each signal. c(MyD 4-4) blocked p65 nuclear translocation in response to IL-1β stimulation, but had no effect on NFκB translocation in response to TNFα stimulation (Fig. 5A,B), suggesting that the cyclic compound inhibition is specific, similar to the activity of the linear parent peptide 16 .

c(MyD 4-4) inhibits T cell IFNγ and IL-17 secretion and ameliorates EAE.
MyD88-dependent signaling plays a major role in autoimmunity 11,14,47,48 , and we recently reported that inhibition of MyD88 can lower disease scores in the EAE mouse model of multiple sclerosis 16 . To investigate the in vivo effects of c(MyD 4-4) treatment on the EAE disease profile, we immunized mice with MOG 35-55 peptide emulsified in CFA, and administered Pertussis toxin (PTX) on day zero and at 48 hrs. Mice were treated i.p. three times a week with 4 mg/kg c(MyD 4-4) or PBS. To determine the effect of c(MyD 4-4) on the developing autoimmune response, we harvested draining lymph nodes (DLN) from immunized mice eleven days following immunization. DLN of treated mice were smaller and contained fewer cells (3 × 10 6 ± 0.5 × 10 6 ) than DLN of control mice (50 × 10 6 ± 7 × 10 6 ). DLN cells from control and c(MyD 4-4) treated mice were activated ex-vivo with increasing doses of MOG  , and cytokine secretion was analyzed 72 h after activation and normalized for cell number. As shown in Fig. 6A,B, T cells from DLNs of mice treated with c(MyD 4-4) secreted significantly less IFNγ and IL-17 than control treated mice. Finally, groups of treated vs. control mice were followed for EAE disease activity; treatment with c(MyD 4-4) led to significantly reduced EAE disease severity (Fig. 6C).

Discussion
We developed a novel inhibitor of MyD88 directed to the central BB loop region of the TIR domain that is responsible for TIR-TIR interactions. Following MyD88 dimerization mediated by the TIR domain, the MyD88 death domain (DD) interacts with IRAKs through DD-DD interactions leading to assembly of the myddosome complex and activation of inflammatory responses 22 . TIR-TIR interactions are not as stable as DD-DD interactions 49 , suggesting that targeting TIR-TIR binding is a more feasible means of blocking myddosome assembly. The BB loop peptide RDVLPGT is highly conserved and maps to the area of interaction between MyD88 monomers 43 . The linear peptide disrupts MyD88 dimerization and inhibits hyperthermia in response to IL-1 in vivo 50 . Blocking MyD88 strongly influences adaptive immunity by inducing a shift away from Th1/Th17 type responses 16 resulting in protection in several autoimmune/inflammatory disease models [6][7][8][9]16 . Despite excellent target specificity, pharmacologic limitations of peptides such as their high sensitivity to proteases restrict their potential as drug candidates. Indeed, we show that the RDVLPGT peptide is rapidly degraded in the presence of plasma or BBMVs. Several groups have identified small molecules that mimic the structural elements and the bioactive conformation of the linear MyDI peptide, either through rational design 43,51 or screening of molecular libraries 52,53 . These small molecules overcome the pharmacologic limitations of peptides at the cost of losing the naturally selected target binding properties. Peptide backbone cyclization, in contrast, preserves target specificity while conferring resistance to endo and exopeptidases rigidity 54,55 . Backbone cyclization can also improve biological activity by reducing conformational entropy.
c (MyD 4-4) was selected from a library of backbone cyclized peptides based on the RDVLPGT BB loop peptide. The bridge is anchored on the nitrogen of the proline residue replacing the pyrrolidine ring. Proline is unique in its cyclic structure among the ribosomally encoded amino acids since it contains a secondary amine formed by the pyrrolidine ring. The peptide bond preceding proline is therefore a tertiary amide bond that is exploited in proteins due to its ability to adopt the cis conformer leading to turns in the protein structure 56 . Furthermore, the conformational influence exerted by proline in peptides is very similar to that of N-alkylated amino acids 57 , and proline containing bioactive peptides can therefore be modified by the synthetic incorporation of N-alkylated amino acids 58 . The MyD88 P200 proline of the BB loop is highly conserved in TIR domains of TLRs and IL-1R family members and serves to confer specificity to TIR-TIR domains as evidenced by the lack of P200 in the TIR domain of TLR3, the only TLR that does not interact with MyD88 21 . Thus, the conserved proline residue in MyD88 BB-loop decoy peptide was replaced by an N-Alkylated glycine building unit that allows backbone cyclization. The creation of a library of backbone cyclized MyD88 inhibitory peptides enabled screening of compounds with spatial diversity that maintained the original side chains of the BB loop peptide.
The c(MyD 4-4) peptide is metabolically stable in human plasma and even in the presence of intestinal enzymes (BBMVs), conditions that lead to rapid degradation of the linear peptide. Confirmation that c(MyD 4-4) binds to the same site on the MyD88 TIR domain as the linear RDVLPGT peptide was especially important in light of the chemical modification to the conserved P200 side chain. Furthermore, c(MyD 4-4) blocks MyD88 function through the same mechanism of action as the linear peptide, i.e. prevention of MyD88 dimerization. Importantly, c(MyD 4-4) inhibits MyD88 dimerization at a 10-fold lower concentration compared to the linear peptide. Functionally, c(MyD 4-4) blocks human and mouse macrophage activation by TLR2 and TLR4 agonists better than the linear peptide, and better than the commercially available small molecule inhibitor ST2825. Therefore, the backbone cyclized c (MyD 4-4) is a metabolically stable inhibitor of MyD88 function with preserved target specificity and improved biological activity.
The phenotype of patients carrying spontaneous MyD88 mutations highlights the dominant role of the TIR domain in controlling myddosome assembly and subsequent inflammatory signaling. To homodimerize, residues in the BB loop of one MyD88 TIR domain interact with the βD and βE strands and the αE helix of the partner TIR domain 21,49 . The R196C polymorphism that affects the Arg of the BB loop (the first Arg of the MyDI peptide), disrupts TIR homodimerization leading to loss of NF-κB signaling and childhood susceptibility to pyogenic bacterial infection 15 . The mutagenesis study of Ohnishi H et al. 17 further demonstrated that R196A reduces MyD88 interaction with MAL, the adaptor protein that bridges between TLR4 and MyD88, leading to NFκB loss of function. The L252P gain of function mutation described in hematological malignancies is located in the βD strand 59 that interacts with the BB loop, and constitutively activates NF-κB complex activation through formation of spontaneous myddosome clusters 60 . R288A is an additional mutation identified by Ohnishi et al. 17 that decreases NFκB activation via decreased affinity between MyD88 and MAL; R288A is located in the αE strand that mediates BB loop attachment. Taken together, these studies suggest that the BB loop region and the domains that interact with it are ideal targets for calibrating NFκB activation during inflammatory responses.
Innate immune inflammatory signals drive Th1/Th17 mediated autoimmune diseases 47 . In the absence of MyD88, mice are mostly resistant to the induction of autoimmune diseases such as collagen induced arthritis and EAE 14,61 . In fact, MyD88-dependent signaling is required for previously activated encephalitogenic T cells to adoptively transfer EAE 11 , suggesting that MyD88 is necessary not only to prime autoimmune T cells but also to reactivate them. The most obvious requirement for MyD88 is in antigen presenting cells, in order to enable them to respond to environmental signals (such as TLR ligands) and educate T cells toward Th1/Th17 differentiation. However, MyD88 is expressed in many immune and non-immune cell types, and can contribute to autoimmunity at multiple points during disease development. Our finding that c(MyD 4-4) inhibits differentiation of autoimmune Th1/Th17 cells and ameliorates disease in the EAE model establishes a proof of concept for the therapeutic potential of MyD88 inhibition in this disease. Further refinement of the pharmacologic properties of this MyD88 inhibitor will enable its testing in multiple conditions where innate immune MyD88-dependent signals promote disease pathogenesis 62 .

Peptides.
MyD88 inhibitor peptide (MyDI, RDVLPGT), or the scrambled version of the peptide (MyDI-sc, PTDLVRG), were synthesized in the Institute of Chemistry, Hebrew University, Jerusalem, Israel, by standard Fmoc chemistry protocols 38 using Rink amide methylbenzhydrileamine (MBHA) resin (loading, 0.66 mmol/gr) as the solid support, and purified by HPLC. Peptides below 95% purity were excluded from further examination. Linear and scrambled peptides were dissolved in water for further examination. A library of backbone cyclized peptides based on the RDVLPGT sequence was synthesized using a non-commercial N-Fmoc-[N-(Alloc) x-alkyl] glycine building block (AGBU) in place of the proline residue. AGBU synthesis and cyclization were performed according to previously described procedures 39  Peptide Stability. The linear peptide and c(MyD 4-4) (10 µg/ml) were mixed with fresh plasma from male Wistar rats (Harlan, Israel) and incubated at 37 °C for 240 min. Triplicate samples were taken at time 0 and after 5, 30, 60, 120, 180 and 240 min. Rat brush border membrane vesicles (BBMV) were prepared by Ca 2+ precipitation from the combined duodenum, jejunum, and upper ileum of male rats as described 41,42 . Briefly, intestines were washed with ice cold saline and separated from mucus. The intestinal mucosa was separated from the luminal surface and placed immediately into buffer containing 50 nM KCl and 10 mM Tris-HCl (pH 7.5, 4 °C). The samples were then homogenated (Polytron PT 1200, Kinematica AG, Switzerland) and 10 mM CaCl 2 was added. The homogenate was placed on a shaker for 30 min at 4 °C and then centrifuged 10,000 g for 10 min. The supernatant was separated and centrifuged at 48,000 g for 30 min and an additional two purification steps were undertaken by suspending the pellet in 300 mM mannitol and 10 mM Hepes/Tris (pH 7.5) and centrifuging at 24,000 g/h. The quality of the BBMV purification was tested using the brush border membrane enzyme markers gamma-glutamyl transpeptidase (GGT), leucine amino peptidase (LAP) and alkaline phosphatase (Sigma-Aldrich, St Louis, MO). Peptides were mixed with purified BBMVs in MES buffer (2- (N-morpholino) To determine the concentration of the compounds, experimental samples were diluted 1:1 with ice-cold acetonitrile and centrifuged (11000 RPM, 10 min). The supernatant was separated and evaporated (Vacuum Evaporation System, Labconco, Kansas City, MO). Samples were reconstituted with acetonitrile: water 70:30 and then centrifuged (11000 RPM, 10 min). The amount of the compounds was determined using high performance liquid chromatography mass spectrometry (HPLC-MS) Waters 2695 Separation Module, equipped with Micromass ZQ detector. The resulting solution (100 μl) was injected into the HPLC system Immunoprecipitation and Western blot. Plasmid pCMV-HA-MyD88 (full length) was a gift from Bruce Beutler (Addgene plasmid # 12287) 63 and plasmid pCMV-Flag-1 MyD88 (full length) was obtained from Dundee university (Dundee, U.K). Plasmids were co-transfected by TurboFect reagent (ThermoFisher Scientific, MA, USA) to Human embryonic kidney (HEK) 293 T cells. After 48 hr MyD88 inhibitors or controls were added for three hr and in the last 20 minutes cells were incubated with 30 ng/ml IL-1β (ProSpec, Rehovot, Israel) in order to enhance the co-immunoprecipitation signal. Fig. S3G shows that co-immunoprecipitation (co-IP) samples from cells treated with IL-1β produced stronger bands than from cells not untreated samples, as shown by others 64 . Cells were lysed in Ripa buffer in the presence of protease inhibitors and incubated on ice for 25 min. Western blot was used to determine expression of the transgenes prior to immunoprecipitation. In brief, whole cell lysates were separated by gel electrophoresis and transferred to nitrocellulose membranes. Membranes were blocked with 5% skim milk for 1 hr and HA was detected using polyclonal mouse anti-HA (Novus Biologicals, Littleton, CO) followed by goat anti-mouse IgG-HRP (R&D systems, Minneapolis, MN). Flag was detected using polyclonal rabbit anti-Flag (R&D systems, Minneapolis, MN) followed by goat anti-rabbit IgG-HRP (Abcam, Cambridge, UK). Membranes were exposed to chemiluminescent substrate in the presence of hydrogen peroxide, using the E-ECL-chemiluminescense detection kit (Biological Industries, Israel), and images were captured using a Bio-Rad imaging system (Bio-Rad, Hercules, CA). For immunoprecipitation, cell lysates were incubated with Pierce Anti-HA magnetic beads (ThermoFisher Scientific, MA, USA) for 30 minutes at room temperature with mixing. After washing the beads, the supernatant containing the target antigen was eluted with 0.1 M glycine pH 2 at room temperature on a rotator for 8 minutes. Beads were separated magnetically, and the supernatants were separated by gel electrophoresis and Western blotting as described above.

Induction of EAE and treatment.
Mice were immunized s.c. in the flank with 100 µg MOG 35-55 emulsified in CFA supplemented with 300 µg M. tuberculosis (Mt) H37RA (BD Difco, NJ, USA). Pertussis Toxin (PTX, List Biological Laboratories, CA, USA) was injected i.p. at the time of immunization and 48 h later. In some experiments, animals were sacrificed prior to onset of clinical symptoms in order to analyze the draining lymph node response. EAE was scored on a scale of 0-6: 0, no impairment; 1, limp tail; 2, limp tail and hind limb paresis; 3, ≥1 hind limb paralysis; 4, full hind limb and hind body paralysis; 5, hind body paralysis and front limb paresis; 6, death.
Draining lymph node (DLN) cell activation. Mice immunized s.c. with 100 µg MOG 35-55 /CFA supplemented with 300 µg M. tuberculosis (Mt) H37RA (BD Difco), injected i.p. with Pertussis Toxin (PTX, List Biological Laboratories) and treated with c(MyD 4-4) or control were sacrificed eleven days after immunization, and single cell suspensions from the popliteal, inguinal and axillary LNs were prepared. Cells were cultured in 96 well plates (0.5 × 10 6 per well) for 72 h with or without increasing concentrations of MOG  peptide.
Statistical analysis. The 2-Tailed t test was used for statistical evaluation of all the results except the two way analysis of variance "ANOVA" test that was used for the EAE model. Values are shown for data that reached a significance of P ≤ 0.05 (*), P ≤ 0.01 (**), P ≤ 0.005, (***), P ≤ 0.001 (****). Bars show mean and standard deviation (s.d.) and in Fig. 2a-b and Fig. 6c standard error of the mean (s.e.m.) (Prism v.5, GraphPad Software Inc. San Diego, USA).