Toward a Novel Drug To Target the EGF–EGFR Interaction: Design of Metabolically Stable Bicyclic Peptides

In cancer, proliferation of malignant cells is driven by overactivation of growth‐signalling mechanisms, such as the epidermal growth factor receptor (EGFR) pathway. Despite its therapeutic relevance, the EGF–EGFR interaction has remained elusive to inhibition by synthetic molecules, mostly as a result of its large size and lack of binding pockets and cavities. Designed peptides, featuring cyclic motifs and other structural constraints, have the potential to modulate such challenging protein–protein interactions (PPIs). Herein, we present the structure‐based design of a series of bicyclic constrained peptides that mimic an interface domain of EGFR and inhibit the EGF–EGFR interaction by targeting the smaller partner (i.e., EGF). This design process was guided by the integrated use of in silico methods and biophysical techniques, such as NMR spectroscopy and surface acoustic wave. The best analogues were able to reduce selectively the viability of EGFR+ human cancer cells. In addition to their efficacy, these bicyclic peptides are endowed with exceptional stability and metabolic resistance—two features that make them suitable candidates for in vivo applications.


Introduction
Protein-protein interactions (PPIs) play arelevant role in all cellular processes. The selective modulation of PPIs has emerged as an ew strategy for therapeutic intervention. PPIs are typically characterised by large,f lat, and featureless interfaces, [1] which thereby complicates the use of conventionalsmall-molecule drugs as inhibitors. [2] As an alternative, peptides have a higher degree of structuralf lexibility;t his allows them to adapt better to irregulart argets, [3] and they can be designed to display av ariety of topologies that closely mimict he structural features found in PPIs. [4] However,p eptidest end to have low serum stability and poor ability to crossp hysiological barriers. [5] In this regard,t hese limitations need to be addressed if in vivo applications of these molecules are to be pursued.
At the onset of cancer,c ells acquire the capacity to self-proliferate through the uncontrolled production of molecules that promote cell growth andd ifferentiation. [6] Among these, epidermal growth factor (EGF) has been shownt op articipate in many types of solidt umour,i ncluding heada nd neck, breast, colon,o varian, and non-small-celll ung cancer. [7,8] To perform its function, EGF binds with high affinity (K D = 1.9 nm) [9] to its membrane receptor (EGFR), [10] andt his triggers intracellular events that lead to uncontrolled cell growth, tumour invasion, and metastasis. [8] Therefore, the EGF-EGFR pathway has become am ain focus for selective chemotherapeutic intervention, and as ar esult, two classes of EGFR inhibitors have been clinically approved, namely,m onoclonal antibodies [11] (cetuximab, panitumumab), which target the extracellulard omain of EGFR,a nd small-molecule kinase inhibitors [12] (gefitinib, erlotinib), which block the intracellular phosphorylation of the receptor. Despite impressive initial response rates to these drugs, most patientse nd up developing drugr esistance, am ajor limitation that reduces the long-term efficacy of the therapy. [13] The overexpression of EGF and/or EGFR in addition to activatingm utations on the receptor have been shown to underlie the main drug-resistance mechanisms. [8] Thus, novel approaches to block the EGF-EGFR pathway and targeted combination therapies are in high demand.
In this regard, the direct inhibition of EGF is considered a promising strategy. [14] Although similarg rowth factors[ such as vasculare ndothelial growth factor (VEGF)] have been successfully addressed with this approach, [15] no current drugs are able to target EGF specifically. [7] However,w er ecently showed that peptides represent as ignificant step in this direction. [16] Herein, we present the structure-based design of bicyclic-constrained peptides that adopt aw ell-defineds tructure and mimic one of the interacting domains of EGFR. In this process, computational and biophysical methods were used as complementarya pproaches to guide the design of more stable and active analogues. In addition to studying their interaction with In cancer,p roliferation of malignant cells is driven by overactivation of growth-signalling mechanisms, such as the epidermal growth factor receptor( EGFR) pathway.D espite its therapeutic relevance, the EGF-EGFR interaction has remained elusive to inhibition by synthetic molecules, mostly as ar esult of its large size and lacko fb inding pockets and cavities. Designed peptides, featuring cyclic motifs and other structuralc onstraints, have the potential to modulate such challenging protein-protein interactions (PPIs). Herein, we presentt he structure-based design of as eries of bicyclic constrainedp eptides that mimic an interface domain of EGFR and inhibit the EGF-EGFR interaction by targeting the smaller partner (i.e.,E GF). This design processw as guided by the integrated use of in silico methods and biophysical techniques, such as NMR spectroscopy and surfacea coustic wave. The best analogues were able to reduce selectively the viability of EGFR + human cancerc ells. In addition to their efficacy,t heseb icyclic peptides are endowed with exceptional stabilitya nd metabolic resistance-two features that make them suitable candidates for in vivo applications.
EGF at the molecular level, we confirmed their potential to block the EGF-EGFR interaction in as pecific receptor-ligand assay anda lso in human carcinomac ells that constitutively overexpress EGFR. Finally,t he stabilityo ft he best designs was validatedi nb iologically aggressive media, such as serum and liver hepatocytes, which therebyc onfirmedt he feasibility of our lead compounds for furtherinv ivo application.

Results and Discussion
Design and biophysical evaluation of peptide ligands The binding of EGF to EGFR occurs through al arge (1440 2 ) and complex interface, in which various domains of the receptor are involved. Nevertheless,anumber of hotspots are found on residues 7-34 of the receptor( Figure 1A). In particular, residues 16-18 of EGFR and residues 31-33 of EGF form as hort parallel b sheet that is stabilised by backbone hydrogen bonds. Also, the a-helical motif in this EGFR domain features severalh otspots (Asp22, Leu25, Arg29) that establish key contacts with EGF.A sw er ecently reported, ac yclic peptide (cp28) comprising this EGFR region was able to recapitulate the binding of the receptor to EGF (K D = 286 mm). [16] However,t he large size and hydrophobic character of this peptide, in addition to its all-l composition, translated into poor solubility and biolog-ical stability,w hich limited its in vivo application and potential development as adrug candidate.
Using ar ational design approach, our goal was to mimic the bioactive conformationo ft he receptorw ith constrained peptides, which are endowed with superior stabilitya nd drug-like properties. [17] Initially, we designed three peptides (named cp23A,c p23B, and cp23C;F igure 1B and C), each with at urninducing d-Pro-Gly motif that replaced part of the long cp28 loop. This loop plays as tructural role in EGFR but is not directly involved in interactions with EGF.I na ddition, we conservatively mutated the Phe residues at the helix for Tyrt oi mprove the overall water solubility of the construct. Finally,p eptides cp23B and cp23C bore as econd cyclic constraint, in the form of ad isulfide bridge, whicha imed to restraint he flexibilityo f the peptide loop. In these last designs,w ei nspected the peptide sequence for side chains that would allow the introduction of ad isulfide crosslink without altering the overall conformationo ft he peptide. In cp23B, Asn6 and Leu11w ere replaced by l-homocysteine and l-Cys, respectively.A lternatively, in cp23C, Gln10 and His17 were modified by d-a nd l-configured buildingb locks (Figure1C).
The synthesis of the bicyclic scaffolds was achieved by using the native chemical ligations trategy ( Figure S1 in the Supporting Information). This approach involves Fmocbased solid-phase peptides ynthesis( SPPS) of linear peptides Figure 1. A) Detailsoft he EGF-EGFR binding interface (EGF on green surface;E GFRi ngrey cartoon, and relevant residues 7-34 are highlighted in red;adapted from PDB ID:1 IVO). B) Mainmodifications performed on the structure of the EGFR-derivedp eptide cp28. C) Sequence and cyclicconstraints of peptide analoguesc p23A-C (X = l-homocysteine). Residuesa re coloured according to their physicochemical nature: blue, polar;grey,hydrophobic;orange, aromatic; purple, basic;red, acidic;yellow,SHresidues;green, Gly.D)CDspectraofc p23A-C in phosphatebuffer (pH 6.8, 10 %TFE). E) Average CSPs induced by peptides cp23A,cp23B (reducedand oxidised),and cp23Conv arious EGF residues. featuring aC -terminal N-acylurea moiety,w hich acts as thioester precursord uring the ligation. Treatment of the unprotected peptidei ns olution with at hiol additive (4-mercaptophenol)t hen catalyses the in situ intramolecular ligation, which resultsi nt he formation of the head-to-tail cyclic peptides in under an hour at pH 7( for details, see the Experimental Section). Finally,t he thiol groups are readily oxidised to yield the desired bicyclic peptides.
The circulard ichroism (CD) spectra of monocyclic cp23A and bicyclic cp23B ( Figure 1D)s how an egative band at approximately l = 208 nm and as econd minimum at approximately l = 222 nm;this indicates aprominent content of a-helix, am otif that is crucial for the PPI. In contrast, the alternative bicyclic topology that we tested in cp23C had as trongly reducedh elicalcontent.
Severalb iophysical methods were used to assess the binding of the peptides to EGF.F irst, the effects of the interaction were evaluated in the 15 N, 1 HHSQC spectra of au niformly labelled 15 N-EGF sample. Comparison of the average chemical shift perturbations (CSPs) induced by the three analogues revealed larger perturbations for cp23B than for cp23A and cp23C ( Figure 1E). Notably,t he oxidised cp23B caused greater changes in the protein signals than its reduced form. The most significant perturbations occurred on residues on the Aa nd B loops of EGF (L8, S9, G12, H16-V19), which are involved in the EGF-EGFR interaction (Figure 2A).
To evaluatet he strength of the peptide-EGF interactions, we used as urfacea coustic wave (SAW) biosensor assay. [18] EGF was covalently immobilised on the gold surface of the chip and binding events occurring on this functionalised surface that directly affect the acoustic wave parameters of the biosensor were measured. In this assay,t he bicyclic peptidec p23B displayed the best affinity of the series (K D = 575 mm,F igure 2A and Table 1), consistentw ith the changes observed in the NMR spectra of EGF.A ss hown by these results, the helical motif is a required elementf or stabilising the entire peptide backbone. The bicyclic architecture of cp23B-but not that of cp23Cfurtherc onstrains the peptide and enables the folding of the loop in the bioactive conformation.
Having proved that we could retain the main requisites for binding in as maller,m ore rigid, and synthetically accessible scaffold, we used computational methods to enhancet he folding (in terms of resemblancet ot he EGFR bioactive conforma- Figure 2. Amino-acid sequenceso fp eptides A) cp23B, B) cp23F, and C) cp23G (X = l-homocysteine,Z= l-norvaline). For each, ar epresentative region of the 15 N, 1 HHSQC spectrum of EGF is shown, in the absence (blue) and presence( red)o fl igand (1 mm). At the bottom,t he SAWs ensorgrams over arange of concentrations and the calculated K D valuesf or the interaction with EGF are shownfor each peptide. www.chembiochem.org tion) andb inding affinity of the peptides. Compared to docking methods, which treat the protein as ar igid or semirigid body and are not convenient for flexible proteins such as EGF, molecular dynamics (MD) simulations are able to sample extensively the interacting complex in as olution environment. We subjected the cp23B-EGF complex to free MD simulation, and from regulars napshots taken during the first part of the simulation, we calculated the theoreticalm ean free energy of binding (DG = À31.4 kcal mol À1 ,a ccording to the MM/GBSA method). In the last part of the simulation, larger fluctuations occurred on the peptide backbone that caused its dissociation from EGF ( Figure S2). We then performed virtual mutationso nt he weaker points-from ad esign point of view-of the cp23B sequence. For instance, several replacement options were tested for Met22, as oxidation of its side chain often occurred during experimental manipulation. Also, we exploredt he introduction of as econd Pro residuea tt he Nterminus of the a-helix, with the aim to stabilise this structural motif. In total, we generated and evaluated in silico 28 new peptide structures (Table S1). As revealed by the calculated DG values, in most cases the Y!P mutationp roduced an 8-10 kcal mol À1 gain in interaction energy,e specially if Pro was at the 12-position. This observation confirms the stabilising effect of Pro as N-cap for the helix, [19] which translated into less backbonef luctuations and improved peptide-EGF contacts during MD ( Figure S2). Also, a conservative M22Nva mutation (as in cp23E and cp23F) was suitable for replacing the unstable side chain of this Met residue.
To test the new computational designs, we selected three of the best analogues,n amely,c p23D, cp23E, andc p23F,f or chemicals ynthesis and evaluation (Table 1). NMR spectroscopy revealed that all three peptides produce clear CSPs on the same part of the protein (L8, S9, G12, H16-V19) as cp23B, which thus confirms the same mode of binding ( Figure 2B). In the quantitative SAWexperiment, the interaction of cp23F produced well-defined association and dissociation curvest hat translatedi nto ah ighera ffinity of cp23F (K D = 279 mm)t han of cp23B.I nc ontrast, the sensorgrams for both cp23D and cp23E were of low intensity and imprecise;this indicated loss of affinity ( Figure S4). These observations suggest that the introduction of radicalm odifications on the peptides equence, as in these last designs, impairs bindingt oE GF.
Given the abovementioned results, we implemented af inal cycle of computationalm odelling for furthero ptimisation of cp23F.T ot his end, we generated as et of heterochiralm utants by introducing l-a nd d-aminoa cids in suitable positions of the sequence. Not surprisingly,m ost new mutations had ad etrimental effect on the DG values (Table S2). Nevertheless, the substitution of Gly10 for large apolar residues, such as Met (as in cp23H) or Tyr, formed new hydrophobic contacts with the Bloop of EGF (Val29, Val30), thusi mproving the DG score. The introduction of d-amino acids in position2 of the sequence also led to theoreticallys tronger interactions, and the T2r mutant( cp23G) had the lowest MM/PBSA scoreo ft he entire set (À8.1 kcal mol À1 ). As seen in the MD simulation, the guanidine group of the d-Arg forms ah ighly stable salt bridge with Asp42 in the Cloop of EGF (FigureS2) and enhances the hydrophilic character of the peptide.
Experimentally,t he affinitieso fc p23G (K D = 252 mm)a nd cp23H (K D = 115 mm), as measured by SAW, were superior to that of cp23F,a lthough not proportionally to the predicted DG values( Figure 2C and Ta ble 1). The induced shifts in the 15 N, 1 HHSQC spectra of the protein confirmed their binding to the EGFR-binding epitope of EGF.H owever,t he poor solubility of cp23H prompted us to select cp23G for furthers tudies.

Structural characterisation of the bicyclic peptides
To gain am olecular-level understanding of the differences between these analogues, we simulated their minimal-energy structures by replicae xchange MD ( Figure 3A). Relative to the helical domain in cp23B, that in cp23F is better structured and more stable, thanks to the constraining effect of the Pro residue in position 12. Nevertheless, backboner oot-mean-square deviations (RMSDs)o fa pproximately 5 (relative to the bioactive conformation,P DB ID:1 IVO) were obtained for both analogueso wing to the poor folding of the flexible loop of the peptide.T he new d-Arg mutation in cp23G partly solved this problem.I ndeed,c lose inspection of its lowest-energy structure showedh ow the hydrocarbon part of the Args ide chain fills the empty space at the central cavity of the peptide, thus preventing collapse of the flexible loop. Overall, this renders a conformation that closely mimics that of EGFR (RMSD = 2.7 ).
To confirm the computational predictions,w ei nvestigated the structural characteristics of cp23G in solution. First, we performed CD spectroscopy of the peptideo ver ar ange of temperatures ( Figure 3B). The CD spectra at 25 8Cs howed two negative bands at approximately l = 208 and 222 nm;this confirmed the existence of the a-helicalm otif. Notably,i ft he sample was heated to 84 8C, the CD spectrum was nearly identical to the former,w hich revealed ah igh conformational rigidity forc p23G, given its smalls ize. Second, we performed NMR spectroscopy measurements of the peptidei nabuffered aqueous solutiona tp H6.8. The 1 HN backbonec hemical shifts exhibited ar elatively large dispersion, which is typical of a more rigid and defined structure ( Figure S7). The secondary chemicals hifts of both 1 H a and 13 C a showedc onsecutive deviations from random-coil values, [20] negative for H a and positive for C a ,a long the Pro12-Gln20 region, whichi si ndicative of the a-helix conformation ( Figure 3C). This hypothesis was further confirmed by the presence of short-and medium-range NOEs in this part of the sequence ( Figure S7). Using these experimentalc onstraints, we modelled a3Dstructure of cp23G in solution ( Figure 3D), which was very similar to that of the previous replicae xchange simulation (RMSD = 1.8 ). Thiss tructure exhibits the features that have been key in this design process, namely,p recise folding of the peptide backbone, which is stabilised by the a-helix and the intramolecular disulfideb ridge.

PPI inhibition assays
To produce ab iological outcome, peptides bindingt oE GF must disrupt the interaction withi ts receptor.T herefore, they ChemBioChem 2018, 19,76-84 www.chembiochem.org must compete with the very strong affinities that naturally govern this process. On the other hand, such molecules benefit from acting on the extracellular space, in which higher effective concentrations can be reached. We first assessedt he capacityo fo ur peptides to inhibit the EGF-EGFR interaction in an AlphaScreen bead-based assay,w hich mimics the protein setting found in cells. For all the designs, ac oncentration-dependenti nhibitiono ft he EGF-EGFR interaction waso bserved ( Figures 4A andS 8). The cp23G peptidew as the most potent inhibitor of the series (IC 50 = 149 mm), an effect that correlates with the binding affinities obtained by SAW.
We next evaluated the effect of our peptides on the viability of A-431 cells, which represent ap articulars ubtype of skin carcinoma, characterisedb ya bnormallyh ighe xpression levels of EGFR. Cellv iability was measured by sodium 2,3-bis(2-methoxy-4-nitro-5-sulfophenyl)-5-[(phenylamino)carbonyl]-2H-tetrazolium (XTT) after 72 ho fi ncubation with EGF-containing media and ar ange of inhibitor concentrations. To validate the assay conditions, erlotinib wasu sed as ap ositivec ontrol (IC 50 = 2.9 mm). Under thesec onditions, cp23G had the greatest inhibitory effect (IC 50 = 24 mm,F igure 4B), in the same range as that of cp28 (IC 50 = 30 mm). Consistent with the previous results, cp23B and cp23F were not ablet ob lock cell proliferation fully ( Figure S9). To confirm that the inhibitory effect was spe-cific for the EGFRp athway,aparallel assay was performed on two EGFR-independent cells, namely,M CF7 (a breast cancer line that overexpresses oestrogen receptor,b ut normall evels of EGFR) and healthy fibroblasts. In both lines, lack of toxicity was observed for all the peptides tested ( Figures 4B and S9), a result that is in agreement with the principle of targeted chemotherapy.

Stability studies in biological media
One of the main challenges faced by therapeutic peptides is that they typically present short half-life in serum, rapid clearance from circulation, and poor tissue distribution. Being an all-l-peptide, our first EGF-EGFR inhibitor,c p28, was predictably labile in vivo. In fact, in human serum, cp28 showed ah alflife of 89 min, which is reasonable foran atural peptide (the half-life of al inear peptidei st ypically less than 30 min). [21] To identify the main cleavage sites on the sequence of cp28, we performed MS-MS analysiso ft he serum sample after incubating for 2h.I nterestingly,t he three main metabolites resulted from hydrolysis and/or oxidation of Arg23 and Met24 (Figure S10), ar egion in which the peptidec hain adopts ar ather extended conformation and is thus more accessible to the active site of proteolytic enzymes. On the contrary,t he new bicyclic analogues, whichp resent am ore constrained structure and no free N-or C-terminal end, were expected to present greater resistance to proteolytic degradation. To validate this hypothesis, we analysed the concentration of peptides in humans erum at 37 8Cf or 24 h. Remarkably,n od egradation was appreciated after 8h,a nd cp23F and cp23G remained more than 60 %u naltered after 24 h ( Figure 4C). In these peptides, norvaline replaces the more vulnerable Met24, and this region is further protectedf rom hydrolysis by an artificial d-Pro-Gly motif to which it is adjacent.
The greater stabilityo ft he bicyclic peptides was further confirmed in human hepatic microsomes, which represent am ore aggressivee nvironment than the serum. Indeed, liverm icrosomes contain ad iversityo fm etabolising enzymes,s uch as proteases,o xidases, dehydrogenases, and acyl-and alkyltransferases, all of which contributet op eptide metabolism,e specially in hydrophobic sequences. Experimentally,t he bicyclic cp23 analogues, and cp23G in particular,s howeds ignificantly lower clearance values than cp28 ( Figure 4D), am etrict hat correlates with the in vivo metabolic clearance. [22] The difference in stability can be explained not only by the presenceo f unnatural residues, which hamper the recognition of enzymes, but also by the lower hydrophobicity of cp23G. As ar esult, the structuralm odifications introduced in this new bicyclic design not only guarantee the target affinity and cellular efficacy-already showni nt he first cyclic inhibitor-buta dd relevant features such as increased solubility and superior biostability.

Conclusion
The EGF-EGFR interaction involves the formation of high-affinity dimeric complexes-mediated by several discontinuouse pitopes-which extend along al arge and irregular bindingi nterface. As aresult of these challengingf eatures, this protein-protein interaction (PPI) was successfully targeted with the use of only antibodies, such as cetuximab, which currently provide the main chemotherapeutic approach in severalt ypes of cancer.H owever,p oorb ioavailability and the development of acquired drug resistancer educe the long-term success of antibody-based therapies. Designed peptidesh ave emerged as a new class of therapeuticst hat combine the surfacer ecognition properties of antibodies and the pharmacological behaviour of small molecules.
Using computational algorithms and biophysical techniques, we implemented ar ational and iterative process that allowed the generation of as eries of bicyclic constrained peptides that mimic the bioactive conformation of EGFR upon binding to EGF.T he binding of these molecules to EGF was quantified by aS AW assay andw as characterised by NMR spectroscopy,a nd their inhibitory effect on the EGF-EGFR interaction wasc onfirmed in as pecific PPI-disruptiona ssay. Moreover,t hesep eptides selectivelyd ecrease the viability of human cancer cells overexpressing EGFR, with mid-micromolar IC 50 values-only one order of magnitude higher than the IC 50 value of the FDAapproved inhibitor erlotinib.A lthough weak affinities are sometimes considered ah andicap, they are not an exception for PPI inhibition and provide the opportunity to target selectively biomarkers (e.g.,E GFR) that are overexpressed in cancer cells but that are also present in most healthy tissues, thus reducingt oxicity.I ndeed,a saresult of the harmful effects of chemotherapy,t here is renewed interest in the use of weak binders,e specially as part of am ultivalent strategy. [23] All in all, the efficacy of thesep eptides in addition to their exceptional stabilityi nb iological media support their use as complementary compounds in multidrug cancer therapy.

Experimental Section
Materials: All amino acids, resins, solvents, and reagents were purchased from Bachem AG (Bubendorf, Switzerland), Iris Biotech Molecular dynamics and DG calculations: Coordinates for EGF and cp28 (residues 7-34) were extracted from the Protein Data Bank (PDB ID:1 IVO). For the bicyclic analogues, the peptide structure was manually modified, and the local clashes were relieved by performing as hort minimisation. The Amber Parm99SB force field was used;t he Leap module of the AMBER package was used to immerse the EGF-peptide complex in ap re-equilibrated octahedral . In all cases, experiments were performedint riplicate, and error bars represent the SD. ChemBioChem 2018, 19,76-84 www.chembiochem.org box of TIP3P water molecules. Chlorine or sodium ions were added to obtain an electrostatically neutral system. The initial complex structure was first subjected to am inimisation protocol consisting of 1000 steps of steepest decent method followed by 500 steps of conjugate gradient method. Thermalisation of the system was performed in the NVT ensemble during 200 ps by using at ime step of 1f sa nd increasing the temperature from 100 to 298 K, for which a force constant of 5kcal mol À1 À2 was applied to protein backbone atoms. Prior to the production run, as hort MD simulation (100 ps) in the NPT ensemble was done to equilibrate the system density to 1atm and 298 K. The simulations were performed at constant pressure (1 atm) and temperature (298 K) for 20 ns, except for some peptides that were extended to 50 ns. Low harmonic constraints (2 kcal mol À1 À2 )w ere used to reduce the protein mobility.C onstant temperature was achieved by using the Langevin thermostat with ac ollision frequency of 2ps À1 .T he SHAKE algorithm was used to keep bonds involving hydrogen atoms at their equilibrium length. The particle mesh Ewald summation method was used to deal with long-range electrostatic interactions, and ac ut-off of 10 was applied for nonbonded interactions. The simulations were conducted with the PMEMD module of the AMBER 14 program. The trajectories were analysed by using frames collected every 2psd uring the production runs.
From the collected MD trajectories, the molecular mechanics Poisson-Boltzmann surface area (MM/PBSA) and Generalised Born (MM/GBSA) methods were used to compute the free energy of binding (DG)b etween the peptides and EGF.Anumerical solution of the Poisson-Boltzmann equation, [24] implemented in AMBER, was used to calculate the polar part of the solvation free energy by using al ow (e = 1) and high-dielectric medium (e = 80) for the solute and solvent, respectively.T he solute-solvent interface was defined by the solvent-excluded molecular surface, at which as olvent probe radius of 1.4 was employed. The nonpolar contribution to the solvation was estimated from the solvent-accessible surface area [25] by using the default scaling factor in Amber (g = 0.0072 kcal mol À1 À2 ). The less computationally demanding MM/ GBSA approach was also computed by using analogous parameters. [26] The reported values of MM/GBSA and MM/PBSA were computed for as ubset of 500 equally spaced snapshots along the converged part of each simulation when both ligand and protein reached astable interaction state.
Replica exchange MD: To assess the structure of the peptides in solution, replica exchange molecular dynamics (REMD) was performed. The initial structure (identical to the one used in the MD simulation described above) was first subjected to am inimisation protocol consisting of 1000 steps of steep decent method followed by 500 steps of conjugate gradient method. The optimised structure was gradually heated to 300 Ki n2 00 ps. The final structure was chosen as the initial structure for all the 16 replicas in the REMD simulations. Te mperatures were set in ar ange from 300 to 600 Kw ith an exchange probability of 30 %. [27] Generalised Born model with an effective salt concentration of 0.2 m was deployed to mimic the solvation effect. Nonpolar solvation term was approximately represented by surface area term. [28] Integral time step was set to 1f s. Te mperature was regulated by using aB erendsen thermostat with ac oupling time constant of 1ps. SHAKE algorithm was used to constrain all the covalent bonds involving hydrogen atoms. Swaps were attempted every 2psa nd MD simulations were extended to 200 ns for each replica.
Solid-phase peptide synthesis: All bicyclic peptides were synthesised on Dawson Dbz AM resin with as ubstitution of 0.4-0.5 mmol g À1 .T he first amino acid was manually coupled. The N-Fmoc-protected amino acid (3 equiv.) was preactivated with 1-[bis(dimethylamino)methylene]-1H-1,2,3-triazolo[4,5-b]pyridinium 3-oxid hexafluorophosphate (HATU, 3equiv.) and N,N-diisopropylethylamine (DIPEA, 3equiv.) in am inimal volume of DMF and was added to the resin. The mixture was allowed to react in an orbital shaker with stirring for 30 min. Then, peptide chains were elongated on a CEM Liberty Blue microwave peptide synthesiser.F moc deprotection was performed by using 10 %( w/v)piperazine and OxymaPure (0.1 m)i nN-methylpyrrolidone (NMP)/EtOH (9:1). The N-Fmoc-protected amino acids (5 equiv., 0.2 m in DMF) were added with Oxy-maPure (1 m in DMF,5 equiv.) and N,N'-diisopropylcarbodiimide (DIC;0 .5 m in DMF,5equiv.) to the resin. The mixtures were stirred for 3min at 90 8C, except for cysteines, histidines, and arginines, which were coupled at 50 8Cf or 10 min. The N-terminal residue was introduced by using aB oc-amino acid. After chain elongation, the resin was washed extensively with DMF and 1,2-dichloroethane. 4-Nitrophenyl chloroformate was dissolved in the least amount of 1,2-dichloroethane and was added to the resin and left to agitate gently for 1h at room temperature. The resin was washed with CH 2 Cl 2 and DMF and was left to react for 30 min with a0 .5 m solution of DIPEA in anhydrous DMF.F inally,p eptides were cleaved with concomitant removal of the side-chain protecting groups by using trifluoroacetic acid (TFA)/H 2 O/triisopropylsilane (TIS) (92.5:5:2.5).
Native chemical ligation: The ligation buffer (6 m guanidinium hydrochloride, 200 mm sodium phosphate, 20 mm tris-(2-carboxyethyl)phosphine hydrochloride (TCEP), 100 mm 4-mercaptophenol, pH 7) was freshly prepared and bubbled with nitrogen. The peptide was dissolved at a2 -3 mm concentration, and the solution was stirred at RT for 4h.T hen, the mixture was acidified, extracted with methyl tert-butyl ether (2 50 mL) and loaded on aP oraPak C 18 cartridge for desalting. The guanidinium salts were washed with buffer,w hereas the peptide was eluted at the end in H 2 O/ MeCN (1:1) and freeze dried. To yield the final bicylic peptides, intramolecular disulfide bonds were formed under highly dilute conditions (20-40 mm)b ys tirring an aqueous solution of the peptides (pH 8) under air oxygen for 24 h. Peptides were purified by semipreparative HPLC with aW aters 2700 sample manager equipped with aW aters 2487 dual-wavelength absorbance detector,a Waters 600 controller,aWaters fraction collector,a nd Masslynx software by using aS unfire C18 column (150 10 mm 3.5 mm, 100 ,W aters), flow rate 6.6 mL min À1 ,s olvent A = 0.1 %t rifluoroacetic acid in water;s olvent B = 0.1 %t rifluoroacetic acid in acetonitrile.
Recombinant human EGF expression: The nucleotide sequence encoding hEGF was cloned into ap OPIN vector [29] as aS UMO fusion protein, which also included aN -terminal His-tag and a SUMO protease cleavage site. This construct produced higher ChemBioChem 2018, 19,76-84 www.chembiochem.org yields (2 mg L À1 )t han the previously reported thioredoxin fusion system. [16] Cell transformation, cell growth, protein expression, and purification were performed by following the standard protocol. [16] Uniformly labelled 15 N-EGF was obtained by growing the E. coli cells in M9 minimal medium containing 15 NH 4 Cl as the sole nitrogen source.
For the structural analysis of cp23G, the peptide was dissolved in NMR buffer supplemented with 10 %t rifluoroethanol and 10 % D 2 Of or af inal sample volume of 350 mL. Residue specific assignments were obtained from 2D total correlated spectroscopy (TOCSY), [31] whereas 2D nuclear Overhauser effect spectroscopy (NOESY) [32] permitted sequence-specific assignments. 13 Cr esonances were assigned from 2D 13 C, 1 HHSQC spectra. The TOCSYa nd NOESY mixing times were 70 and 200 ms, respectively.T he D1 relaxation delay was 1.5 s.
Simulated annealing protocol: The main NOEs among the peptide residues (Tyr16-Glu13, Arg2-Leu19, Arg21-Leu17, Gln20-Tyr16, and His15-Gln8) were deployed through as et of 500 independent simulated annealing MD simulations, for which NMR restraints were applied by using the Sander module in conjunction with the AMBER99 force field. The most-populated conformation in the REMD was used as the starting structure in each annealing simulation. The generalised Born implicit solvent model with an effective salt concentration of 0.2 m was applied. At otal of 100 ps of MD were run with a1fs time step. Al ong-range cutoff of 10.0 was used for nonbonded interactions. The upper and lower distance CaÀCa atom restraint limits were set to 10.0 and 4.5 ,r espectively,t os ample ab roader conformational according to NOE data, with af orce constant of 10.0 kcal mol À1 À2 .C hirality restraints (50.0 kcal mol À1 À2 )w ere used to avoid chirality flipping at high temperatures in the simulated annealing protocol. Temperature was increased to 700 Ki nt he former 20 ps of simulation to be then gradually cooled down to 300 Ki nt he remaining simulation time. Different seed numbers were used in each simulation. For each independent simulation, the lowest AMBER energy structure conformation was selected as representative of the conformational space explored. The 500 simulated annealing conformations were finally clustered using Ptraj module to obtain the low-energy averaged structure compatible with the NMR spectroscopy data.
SAWb iosensor: Affinity analyses were performed with aS am5 Blue biosensor (SAWI nstruments, Bonn, Germany). The chip surface was functionalised with EGF following the reported protocol, [16] and peptides were injected at ar ange of micromolar concentrations selected to cover the binding equilibrium constant. Sensorgrams were analysed by using the Origin Pro 7.5 and FitMaster software. Briefly,t he signal at the association equilibrium was represented for each ligand concentration, and nonlinear fitting of the data was performed to obtain the K D values of the interaction.
AlphaScreen assays: 10 mLo fE GF [His-tag functionalised, 3nm final concentration in phosphate buffered saline (PBS), 0.1 % bovine serum albumin (BSA), 0.1 %T ween-20, pH 7.4] and 10 mLo f peptide (at different concentrations) were added to 96-well Optiplates (PerkinElmer) and incubated at RT for 15 min. EGFR-Fc (3 nm final concentration in the same buffer) was then added and incubated for 30 min. Next, 10 mLo fP rotein A3 835 Donor beads (20 mgmL À1 final concentration, 60 min incubation) and anti-His acceptor beads (10 mL, 20 mgmL À1 final concentration) were added. After incubating for 30 min, the fluorescence emission was recorded at l = 615 nm with an EnVision Multilabel Reader.D ose-response curves were obtained with the GraphPad Prism 6.03 software by using an onlinear fit and variable slope, from which IC 50 values were calculated.
Cell viability assay [sodium 2,3-bis(2-methoxy-4-nitro-5-sulfophenyl)-5-[(phenylamino)carbonyl]-2H-tetrazolium (XTT)]: A-431 human cancer cells (2.5 10 6 EGFR cell À1 ) [33] were seeded in a9 6well Corning microplate (2 10 3 cells well À1 ). After 24 ho fi ncubation at 37 8C, cells were treated with the peptides dissolved in DMEM [high-glucose, 10 %f oetal bovine serum (FBS) containing EGF and growth factors] for 72 h( 100 mL, concentration range 0.5-150 mm). Then, the peptide solution was removed and substituted with fresh medium. Activated-XTT solution (50 mL, 0.1 mL of the activation reagent in 5.0 mL of the XTT reagent, mixed 15 min before use) was added to each well and was incubated for 4h at 37 8C. The absorbance at l = 450 nm was measured with aP owerWave X reader (Bio-Tek, INC), for which the positive control was cells treated with DMSO and the negative control was untreated cells. Each experiment was performed in triplicate, and the IC 50 value was calculated by linear-regression fitting (concentration vs. response curve) by using GraphPad Prism. The same conditions were used for studying MCF7 and J2P fibroblast viability.
Stability in human serum: Peptides at af inal concentration of 150 mm were dissolved in Hank's balanced salt solution and were incubated at 37 8Ci nthe presence of 90 %h uman serum for 24 h. Then, aliquots (50 mL) were extracted at different incubation times and were treated with MeOH (200 mL) to precipitate serum proteins. After centrifugation at 4 8Cf or 30 min, the supernatant was filtered and analysed by HPLC to calculate the percentage of intact peptide in the sample. The linear peptide VQAAIDYING (ACP) was used as positive control.