Genetically-encoded discovery of proteolytically stable bicyclic inhibitors for morphogen NODAL†

In this manuscript, we developed a two-fold symmetric linchpin (TSL) that converts readily available phage-displayed peptides libraries made of 20 common amino acids to genetically-encoded libraries of bicyclic peptides displayed on phage. TSL combines an aldehyde-reactive group and two thiol-reactive groups; it bridges two side chains of cysteine [C] with an N-terminal aldehyde group derived from the N-terminal serine [S], yielding a novel bicyclic topology that lacks a free N-terminus. Phage display libraries of SX1CX2X3X4X5X6X7C sequences, where X is any amino acid but Cys, were converted to a library of bicyclic TSL-[S]X1[C]X2X3X4X5X6X7[C] peptides in 45 ± 15% yield. Using this library and protein morphogen NODAL as a target, we discovered bicyclic macrocycles that specifically antagonize NODAL-induced signaling in cancer cells. At a 10 μM concentration, two discovered bicyclic peptides completely suppressed NODAL-induced phosphorylation of SMAD2 in P19 embryonic carcinoma cells. The TSL-[S]Y[C]KRAHKN[C] bicycle inhibited NODAL-induced proliferation of NODAL-TYK-nu ovarian carcinoma cells with apparent IC50 of 1 μM. The same bicycle at 10 μM concentration did not affect the growth of the control TYK-nu cells. TSL-bicycles remained stable over the course of the 72 hour-long assays in a serum-rich cell-culture medium. We further observed general stability in mouse serum and in a mixture of proteases (Pronase™) for 21 diverse bicyclic macrocycles of different ring sizes, amino acid sequences, and cross-linker geometries. TSL-constrained peptides to expand the previously reported repertoire of phage-displayed bicyclic architectures formed by cross-linking Cys side chains. We anticipate that it will aid the discovery of proteolytically stable bicyclic inhibitors for a variety of protein targets.


INTRODUCTION
23 out of the 60 FDA approved peptide drugs are peptide macrocycles. 1 Macrocyclization of peptides increases binding affinity, 2 improves permeability through the cell membrane [3][4] and stability towards enzymatic hydrolysis [5][6] compared to linear peptides. [7][8] The large surface area of macrocycles has been critical for identifying molecules that bind extended protein surfaces and inhibit protein-protein interactions. 4 Introduction of a bridgehead into macrocyclic topologies to form so-called bicyclic peptides could further decrease conformational flexibility and increase stability or binding potency. [9][10] Bioactive bicyclic peptides that have been reported thus far originate from natural products, 11 computational approaches, 12 cyclization of known bioactive peptides, [13][14][15] or screening of combinatorial libraries. [16][17] To fuel the last method, synthesis on the solid support can yield libraries of 10 2 -10 6 diversity, [16][17] and DNA-encoded libraries [18][19][20] or late-stage chemical diversification of biosynthesized peptides displayed on mRNA 21 or phage [22][23][24][25][26] can give rise to libraries with 10 9 -10 12 diversity. 27 Development of new approaches for chemical diversification of mRNA and phage display libraries [27][28] make it possible to screen and discover new macrocyclic and bicyclic topologies with value-added properties. In this manuscript, we develop a methodology for late-stage chemical diversification of phage display peptide libraries to create previously unexplored genetically-encoded libraries of new, "terminus-to-side chain" bicyclic topology.
There are currently two strategies for synthesizing phage display bicyclic libraries: Both of them employ cross-linking of Cys side chains with electrophiles ( Figures 1A). They can be divided into two sub-categories: (i) Cross-linking of three Cys residues with a C3-symmetric electrophile to yield bicycles displayed on phage 5 ( Figure 1A); (ii) Cross-linking of four Cys residues with C2-symmetric electrophiles to yield a mixture of three regioisomeric bicycles displayed on phage 29 ( Figure 1A). Bicyclic libraries have also been synthesized in mRNA display libraries using strategy (i) or via incorporation of two pairs of orthogonally reactive unnatural amino acids (UAAs) into mRNA display libraries. [30][31] Incorporation of UAAs 4 into phage display libraries has been reported and UAAs have been used to generate a phage display macrocyclic libraries. [32][33][34][35] In this manuscript, we sought to devise the modification approach that uses peptide libraries made of 20 common amino acids: Bypassing the complexity of UAA incorporation avoids biases that might result from the incorporation of such UAAs in the phage library. 36 We combined well-established modification of phage display libraries with N-terminal Ser and Cys-side chains to generate a novel genetically-encoded bicycles topology. This topology does not display a free N-terminus and unlike previous reported strategy(ii), this strategy yield a single regioisomer. ( Figure 1B).
Aldehyde is a versatile bio-orthogonal handle. In proteins, aldehyde can be incorporated by periodate oxidation of N-terminal Ser. [37][38] This method has been used for modifications of peptides, 39 for PEGylation of clinically relevant growth factors and cytokines to improve their stability in preclinical studies, 40 and for synthesis of antibody-drug conjugates. [41][42] Libraries with N-terminal Ser have been previously converted to peptide-aldehydes and modified by oxime and hydrazine, [43][44] benzamidoxime, 45 or Wittig reaction 46 and used for selection of diverse chemically-modified peptide ligands. [43][44][45][46][47][48][49][50][51] Our report provides the first example of using this versatile technology for bicyclization. To this end, we devised a macrocyclization approach that employed an N-terminal aldehyde residue. To demonstrate the value of such a library in discovering new bioactive bicycles, we employed this library to discover inhibitors of protein NODAL and antagonists of NODAL-induced signaling.
The extracellular embryonic morphogen NODAL belongs to the transforming growth factor-beta (TGFβ) superfamily. It is a stem-cell-associated factor that has emerged as a putative target for the treatment of cancer. 52-62 NODAL is normally restricted to embryogenesis, wherein it maintains pluripotency in the epiblast and governs the formation of the body axis and left-right asymmetry. [63][64][65] After development, NODAL is relatively restricted to reproductive cell types and is not detectable in most normal adult tissues. [63][64]66 However, NODAL expression re-emerges in a large number of divergent cancers. [52][53][54][55][56][57][58][59][60][61][62] In almost every cancer studied thus far, the acquisition of NODAL expression is associated with increased tumorigenesis, invasion, and metastasis. It supports self-renewal in pancreatic and breast cancer stem cells and is enriched in breast, 67 prostate, 58 melanoma, 56 pancreatic, 68 colon, 54 and ovarian 50,60 cancer cells with stem-cell properties. NODAL exerts its function by binding to and activating the cell surface receptors Alk4 and Alk7 in cooperation with the co-receptors Cripto-1 (FDGF1) or Cryptic (CFC1) to form a ligand-receptor complex that leads to the phosphorylation of Smad2/3 and the transcription of target genes, including NODAL itself. The only available inhibitor of NODAL to date, monoclonal anti-NODAL antibody 3D1, 69 has demonstrated success in preclinical models of melanoma and is currently undergoing further preclinical evaluation. No small-molecule or macrocyclic inhibitors of NODAL are presently available. In this manuscript, we employed bicyclic phage libraries to discover the first in class bicyclic ligands for NODAL protein. These ligands antagonize NODAL-induced signaling and specifically suppress NODAL-promoted proliferation of cancer cells. Evaluation of these antagonists benefited from the unique topology of the macro-bicycles that masked the N-terminus and equipped these macro-bicycles with multi-day stability in serum-rich cell culture media.

RESULTS AND DISCUSSION
Optimization of bicyclization on unprotected synthetic peptides. The chemical linkers TSL-1, TSL-3, and TSL-6 containing aminooxy and benzyl chloride functional groups ( Figure 2A) were synthesized (Scheme S1) and tested for their ability to modify a series of unprotected peptides of structure SXnCXmC where X is any amino acids except Cys and n+m ranges from 3 to 12. To mimic the conditions that would be suitable for modification of phage displayed library of peptides, we used model peptides at the micromolar concentration in aqueous buffers and treated them with super-stoichiometric reagents ( Figure   2B). Figures 2C-2D described monitoring of the reaction progress between a representative model peptide SICRFFCGGG (200 µM) and NaIO4 (2.4 mM) to form the N-terminal oxoaldehyde. Quenching the excess of NaIO4 with an excess of methionine, decreasing the pH of the reaction, and addition of 1 mM TSL-6 6 formed the oxime ( Figure 2B). At pH ranging from 2.0 to 3.5, the rate constant of this ligation was k = 0.81-0.93 M -1 s -1 . (Figures 2C-2D). In these conditions, oxime ligation was complete within 1 hour. Increasing the pH to 4.5 decreased the rate (k = 0.37 M -1 s -1 ) and led to only partial completion in 1 hour ( Figure 2D). Little to no oxime was formed at a higher pH ( Figure 2D). We note that aniline can catalyze oxime reactions 51,70 ; however, we avoided aniline and other nucleophilic catalysts to prevent the formation of byproducts with TSL-6. 71 The addition of 1 mM TCEP to the ligated product reduced the disulfide linkage. Raising the pH to 10 led to bicyclization of peptides in 3 hours.
This reaction successfully modified 28 unique peptides of different spacing between the Ser and Cys residues with an average isolated yield of 40% ( Figure 2E). The yields for 33 bicycles are summarized in Table S1, products modified with TSL-6, TSL-1 and TSL-3 are denoted as #b, #c and #d respectively.
Additional characterization of TSLs modified peptides can be found in Schemes S2-S35. The 28 unique model peptides were selected either at random (1a-3a, 6a-9a, and 25a-28a) or chosen in the phage display selection (4a-5a and 14a-21a). Two peptides (12a-13a) were adapted from a previous publication. 72 Table S2 further highlights the various physicochemical properties of these peptides. We compared the yields of this reaction to modification of peptides with other reagents such as pentafluorophenyl sulfide (PFS), 1,3,5-tris(bromomethyl)benzene (TBMB), and α,α′-dibromo-m-xylene (DBMB). PFS cyclized peptides had an average yield of 35.5% (Scheme S36). TBMB cyclized peptides had an average yield of 35% (Schemes S37-S38). DBMB cyclized peptides had an average yield of 31% (Table S3). In conclusion, the aqueous biocompatible modification of peptides with TSL effectively produced bicyclic peptides with yields compatible with other reagents used in peptide cyclization or bicyclization. Although oxime linker is known to be reversible, we observed these bicycles to be stable in aqueous ammonium acetate (pH 4.7), PBS (pH 7.4), and Tris (pH 8.5) buffers for a month at room temperature ( Figure S1).
Modification of phage display libraries. The bicyclization approach described above was compatible with the modification of the phage display peptide libraries. To quantify the efficiency of bicyclization 7 reaction in phage libraries, we used biotinylation and phage capture steps similar to approaches employed in previous publications ( Figure 3). [43][44][45][46] Previously, formation and reactivity of aldehyde in phage libraries was quantified by exposing the library to an aldehyde-reactive aminooxybiotin (AOB) and counting the number of biotinylated particles captured by streptavidin paramagnetic particles ("AOB capture," Figure   S2C). 51 Using the reported oxidation conditions, we exposed a phage displaying SX1C X2X3X4X5X6X7C library of 10 8 peptides to an ice-cold solution of NaIO4 (60 μM in PBS) for 9 min, quenched the oxidation by 0.5 mM methionine for 20 min and used AOB capture to confirm that 93±11% of the library was converted to aldehyde. Reacting with 1 mM solution of linchpin TSL-6 at pH 3.5 for 1 hour consumed most of the aldehyde functionalities ( Figure S2E). After removing excess TSL-6 by size exclusion spin column, we exposed the phage to a biotin-thiol reagent (BSH, Figure 3E) and captured the biotinylated clones by streptavidin paramagnetic particles. This "BSH capture" confirmed that 52±4% of the library contained thiol-reactive benzyl chloride groups ( Figure 3B). Exposure of phage to TCEP and then pH 10 buffer completed bicyclization as evidenced by the decrease in BSH capture. In the control condition, incubation of the TSL-6 ligated library at pH 10 in the absence of TCEP did not lead to any decrease in BSH capture, indicating that the number of benzyl chloride groups on phage remained unchanged in the absence of TCEP. We estimated 41±13% of library to be converted to TSL6-bicyclic library ( Figure 3D).
Detailed calculation of the conversion percentage can be found in Figure S3. Similar monitoring of the modification of the SX1C X2X3X4X5X6X7C library with TSL-1 and TSL-3 ( Figure S4) and the SX1C X2X3X4C phage with TSL-6 ( Figure S5) demonstrated a generality of this approach.
The ligation condition showed minor effects on the infectivity of the phage ( Figure S3). To confirm that the chemical modification did not compromise the integrity of the phage DNA, we performed PCR (Figure S6) of the library and deep sequenced the PCR amplicons to monitor the sequence diversity of the library before and after chemical modification. If chemical modification significantly damaged the DNA, we anticipated observing a change in library composition. As the composition of the library before and 8 after the modification remained the same (Figures S7-S9), we concluded that the modification did not impact the diversity of the phage library and did not impact the integrity of the phage DNA. These studies collectively demonstrate the construction of a library of 10 8 bicycles that offer the potential for discovering bicyclic ligands for any target using canonical selection approaches.
Selection of bicycles that bind to NODAL. We applied a TSL-6-modified phage display library SX1C X2X3X4X5X6X7C to discover a ligand for the morphogen NODAL. We performed three rounds of phage selection using His6-tagged NODAL protein as the bait. In between rounds of selection, we raised the stringency by increasing the number of washes and reducing the amount of immobilized NODAL protein ( Figure 4A). In round 3, we also performed two control selections; in first control, we panned the unmodified R3 library against the NODAL protein (R3-UN) and in second control, we panned the TSL-6modified R3 library against unrelated His6-tagged protein (R3-TG). Phage recovery increased in R3 when compared to R1 and R2. This recovery was ablated by 20-fold when the unmodified round 3 library was panned against NODAL (R3-UN) and when the TSL-6-modified library was panned against unrelated protein (R3-TG) ( Figure 4B). Deep sequencing the output of all selection round and the control experiments identified families of sequences that exhibited high normalized abundance in R3 and low normalized abundance in R1, R2, and control experiments R3-UN, and R3-TG ( Figures 4B and S10). From these families, we selected six representative sequences for further validation (14a-19a; Figures 4B and S10).

Validation of NODAL bicyclic inhibitors.
The bicycles 14b-19b were chemically synthesized and tested for their ability to antagonize NODAL-induced signaling in a previously established P19 cell line. [75][76] Stimulations of the P19 cells with rhNODAL at 100 ng/mL for 1 hour led to the phosphorylation of SMAD2 ( Figure 4D, column 3). This phosphorylation was inhibited by ALK4/7 kinase inhibitor SB431542 ( Figure 4D, column 4), as previously reported. 58,[73][74] Bicyclic peptides 14b-19b at 100 µM were able to inhibit rhNODAL-induced phosphorylation of SMAD2 ( Figure S11A). At the concentration of 10 µM, bicyclic peptides 18b and 19b inhibited phosphorylation of SMAD2 ( Figure 4D, columns 9 9 and 10), whereas bicyclic peptides 14b-17b exhibited no activity ( Figure 4D columns 5-8and Figure   S11B). As 19b exhibited robust and reproducible inhibition of phosphorylation ( Figure S11B), we further tested the ability of 19b to suppress the NODAL-induced proliferation of ovarian cancer cells. We transfected ovarian cancer cells (Tky-nu) with a plasmid vector containing human NODAL and used a GFP transfected Tky-nu cell line as an isotype control. Tky-nu-NODAL and Tky-nu-GFP cell lines were cultured in the presence and absence of 19b for 72 hours. Treatment of Tky-nu-GFP cell with 10 µM 19b had no effect on the proliferation, whereas the viability of Tky-nu-NODAL cells was reduced to 23% when compared to untreated Tky-nu-NODAL cells ( Figure 4E). The response to 19b was dose-dependent with apparent IC50 between 0.1 and 1 µM 19b ( Figure 4E and S13B). The discovery of 19b served as a promising starting point for developing more potent NODAL antagonists.
Proteolytic stability of bicycles. Intrinsic proteolytic stability of the bicyclic scaffold was critical to the evaluation of the NODAL antagonist in the aforementioned cell-based assays. Specifically, we found that 53% of the bicyclic peptide antagonist 19b remained intact after 72 hours of incubation at 37 °C in a serum-rich culture medium ( Figure S14). We followed up on this observation and tested the stability of a panel of bicyclic scaffolds in various proteolytic degradation conditions ( Figure 5). We exposed the bicycles for 5 hours at 37 °C to Pronase TM : a mixture of endo-and exo-proteases known to cleave proteins into individual amino acids. The analysis of 21 other TSLs bicycles ( Figure 5A) highlighted that 25-90% of the bicycles remain intact after 5 hours of exposure to Pronase TM (Figures S15-S22). In these conditions all the tested linear and monocyclic disulfides degraded to <1%. When ten of these bicycles were exposed to fresh mouse serum at 37 °C, on average, 72% of the starting amount was intact after 5 hours ( Figures   5A and S22-S23). Monocyclic peptides formed by modifying peptides by (α-α'-dibromo-m-xylene) DBMB [75][76] have the same topology as one of the rings in TSL-modified bicycles. We observed that on average 13% of the DBMB macrocycles remained intact after 5-hour treatment by Pronase TM , compared to 62% from the TSLs-modified set ( Figures 5B and S24-S27, the values represent average from the set of n=14 sequences modified by DBMB or TSLs). Figure 5C represents an example of bicycle 6b that remained 83±6.9% intact after 5 h of incubation in Pronase TM ; the DBMB macrocycle 6g and the disulfide precursor 6a degraded to <1% under the same conditions. We tested the stability of two sequences modified with the PFS cross-linker developed by the Pentelute Lab. 77 In Pronase TM , macrocycle PFS-STCQGECGGG and bicycle TSL-3-STCQGECGGG exhibited similar stability, whereas macrocycle PFS-SICRFFCGGG exhibited lower stability than bicycle TSL-3-SICRFFCGGG (Figures 5B, S28, and   S29). Due to differences in the shape of the cross-linkers resulting in different conformations of peptides, the results were difficult to interpret and we did not expand on this comparison further. In general, it is not trivial to quantify the advantages of any peptide cross-linker in comparison to all other available crosslinkers to-date; however, a comparison of a set of n=14 peptides modified with closely related DBMB and TSL linkers indeed suggests that bicyclization yields a significant improvement in stability.

Molecular Dynamics Simulation of Bicycle Structures.
In testing the stability of a large, diverse set of bicycles, we observed preliminary linker-dependent and sequence dependent trends in degradation. For example, Pronase TM degradation of peptide SWDYRECYLEC modified with TSL-1 or TSL6 linker yielded minor but statistically significant differences: 82±13% and 68±14% intact bicycles after 5 hours ( Figure 5A). To explore these differences, we employed molecular dynamics (MD) simulation of the conformational ensemble of these bicycles. The penultimate amino acids in TSL-1-SWDYRECYLEC and TSL-6-SWDYRECYLEC bicycles yielded different Ramachandran plots describing the dihedral angles for -WDYR-sequences in the first ring. On the other hand, the dihedral angle populations for the -CYLEC-sequence in the second ring were similar ( Figure S30). The MD simulation suggested that conformations of two rings are decoupled from one another. Differences in degradation for two bicycles, thus, might originate from the enhanced flexibility in one of the rings. Similar decoupling was observed in in TSL-1-SHCVWWDC and TSL-6-SHCVWWDC bicycles. The penultimate amino acid, His, exhibited different clustering of the dihedral angles. On the other hand, -VWWD-sequence in the second ring had similar backbone conformations in both bicycles ( Figure S30). These studies provide important starting point for understanding ground-state conformational ensemble of these molecules.

CONCLUSION
In conclusion, two-fold symmetric tridentate linchpins that contain aldehyde and two thiol-reactive groups enable a robust one-pot bicyclization of peptides SXnCXmC. Such libraries can be used to discover productive antagonists of protein-protein interactions. The bicycles show good stability in digestive conditions. Although the 28 bicyclic peptide sequences tested do not exhaustively sample all possible combinations, the tested peptides included all the potentially problematic amino acids (Lys, Arg, His, Tyr, Trp, Asp/Glu, Ser/Thr). Proteolytic stability of bicyclic architecture sans a free N-terminus is significantly improved when compare to closely-related DBMB-cross-linked monocycles. As the strategy is compatible with phage display libraries containing the SXnCXmC motif, we anticipate that other peptide libraries that contain this motif will be amenable to such late-stage functionalization. We noted that many genetically encoded libraries do not contain N-terminal Ser and instead have an N-terminal Met or Met analogs encoded by AUC starting codon. However, it is possible to introduce an N-terminal Ser into these systems by expressing a library with N-terminal TEV-cleavable sequence: H-MENLYFQ\S (where \ denoted as the cleavage site). A conceptually similar approach has been recently demonstrated by Jianmin Gao and co-workers who expressed ENLYFQ\C in phage libraries and used TEV cleavage to expose the N-terminal Cys. 78 Finally, the lower symmetry of the TSL-style linkers allows their diversification with any chemotype of C2-symmetry. 71 It offers a significant expansion of the bicyclization repertoire beyond traditional architectures produced from three-fold symmetric cross-linkers.

Notes
Conflict of interest statement: Patent application describing this invention was filed by TEC Edmonton in July 2018. R.D. is a shareholder of 48Hour Discovery Inc., the company licensing the technology.