Isothiocyanate-mediated cyclization of phage-displayed peptides enables discovery of macrocyclic binders

Abstract

Cyclic peptides exhibit advantages in binding protein targets with high affinity and competency in inhibiting protein-protein interactions. Cyclic peptide phage display with more than a billion variants is an invaluable tool in drug discovery. However, achieving efficient peptide cyclization on phages remains a challenge because of the limited availability of reaction sites, which also restrict scaffold diversity. Here, we report an isothiocyanate-derived cross-linker featuring dual reactive groups: a bromide that covalently attaches to cysteine thiols and a thiocyanogen that selectively forms a thiourea bridge with either the N-terminal amino group or ε-amines of lysine, depending on pH. This strategy enables pH-modulated cyclization. At pH 6.5, head–to–side chain cyclization occurs, and at pH 9.5, side chain–to–side chain ligation is performed. Both processes simultaneously generate thiourea scaffolds. To demonstrate the versatility and biocompatibility of this approach, we constructed cyclic peptide libraries using both cyclization methods and successfully selected binders for several targets, including cyclophilin D, murine double minute 2, and Keap1, with dissociation constants ranging from micromolar to nanomolar. Given the broad pharmacological potential of the thiourea moiety, this phage display library opens previously unidentified chemical space with high scaffold diversity and the integration of a proven pharmacophore for the development of cyclic peptide therapeutics.

The isothiocyanate-derived phage display generates cyclic peptide libraries with thiourea scaffolds for drug development.

INTRODUCTION

Macrocyclic peptides have emerged as an important class for the development of molecular recognition tools and lead compounds because of their unique biological properties that combine the advantages of small molecules and biological macromolecules, especially for targeting untamable proteins of interest such as protein-protein interaction (PPI) interfaces (1–5). Structure-guided cyclic peptide design with the aid of emerging machine learning algorithms has been gradually adopted (6–8). However, its performance in reconciling convex or polar epitopes at the interaction interface and heavy reliance on the accuracy of initial structures needs to be improved (9–11).

Alternatively, the phage display approach experimentally selects binders exclusively on the basis of their affinity to the structure of protein targets, and its functionality could be extended by covalently installing different chemical groups. As such, phage display for cyclic peptides can readily reach high-throughput screening with a diversity of up to 10¹² by manipulating genetic coding for amino acid components and chemical modifications (12). Besides the thiol ligation between two cysteine residues, cyclization and modifications via reactive N-terminal Cys, N-terminal amine, or Lys residues have been developed for expanding topology space and eliminating instability under reducing conditions (13–16). Symmetric cystine cross-linkers are another option for generating stable cyclic peptides for therapeutic purposes, such as 1,3,5-tris (bromomethyl)benzene (TBMB) and its derivatives bicycle toxin conjugates and tumor-targeted immune cell agonists (17–20). However, the life cycle of bacteriophage is compromised because of the loss of the intrinsic disulfide bond of phage coat protein III (pIII) and reduced infection efficiency (~100-fold) (21). Other strategies involve the cross-linking of amino groups, by using an asymmetric cross-linker with an amino reactive unit such as 2-((alkylthio)(aryl)methylene)malononitrile, 4-(bromomethyl)benzaldehyde, and ortho-phthalaldehyde, to generate variously functional macrocyclic peptides with asymmetric backbones (22–25). Embedding electrophilic unnatural amino acids (O2beY, etc.) is another effective strategy to generate cyclic peptides with an asymmetric skeleton (21, 26, 27). Beyond the cyclization, the introduction of functional groups or nonpeptide moieties into genetically encoded peptides can bring both structural and functional novelties (28–31). Thiourea is one of the most stable sulfur carbonyl compounds in nature, and it also has pronounced chemical and biological activity as well as biocompatibility. The exploration of thiourea derivatives in lead compounds is a promising approach to enhancing biological binding, leading to a variety of thiourea-containing approved drugs, including thioacetazone, enzalutamide, and carbimazole (32–35). As early as 1936, Todrick and Walker showed that allyl isothiocyanate in alkaline conditions blocked the N-terminal amino of cysteine, leading to the formation of the thiourea without influencing the sulfhydryl group (36). Inspired by this characteristic, the most famous tool, Edman degradation, was pioneered for the determination of the sequence of proteins or peptides (37). Then, functional molecules derived from isothiocyanates, including fluorescein isothiocyanate, have frequently appeared in the detection of various biological pathogens with the coupling of the primary amino (38, 39). However, it was not until 2016 that Xuan et al. first incorporated the isothiocyanate group into protein based on the genetically encoded system of prokaryotic cells to accomplish intramolecular cross-linking with adjacent Lys residues and stably intermolecular coupling of Zprotein/Afb-Trx via thiourea bridge. Advances in these chemoselective modifications enable macrocyclization by linking the isothiocyanates to the fixed N-terminal amines or Lys residues in the linear peptides (40).

In the current work, we presented a one-pot strategy for the incorporation of thiourea into the cyclic backbone and developed a phage display platform for thiourea-containing cyclic peptides. A cross-linker, 1-(bromomethyl)-4-isothiocyanatobenzene (pNCSBnBr) (fig. S1), with an isothiocyanate group and a thiol-reactive unit, was used for the one-step strategy to generate cyclic peptides by isothiocyanate addition and cysteine alkylation. Notably, the cyclization is pH modulated, head–to–side chain cyclization and side chain–to–side chain ligations were achieved at pH 6.5 and 9.5, respectively, and the structural diversity of the backbone has been further amplified (Fig. 1A). With this unique method, we identified ligands binding to cyclophilin D (Cyp D), MDM2, and Keap1 with high affinity. Given the pharmacological advantages of thiourea, we provide a previously unknown avenue for developing the therapeutic thiourea-containing cyclic peptides (Fig. 1B).

Fig. 1. The construction of thiourea-containing cyclic peptides using one-pot cyclization.

(A) The cyclic modification of a bifunctional cross-linker derived from isothiocyanate with an isothiocyanate group and a thiol-reactive unit enables the generation of two types of thiourea-containing skeletons of cyclic peptides under aqueous buffer of different pH. (B) Phage panning process using streptavidin magnetic beads for the selection of functional cyclic peptides featuring thiourea skeletons.

RESULTS

Validation of peptide cyclization under different physiological environments

Previous reports have established that isothiocyanate addition is limited by the degree of deprotonation of amino groups so that ε-amines of lysine only participate in the addition reaction in alkaline buffers (40). The distinction in pK_a (where K_a is the acid dissociation constant) between the N-terminal amino group and ε-amines inevitably leads to their different nucleophilicity under acid-base environments (41–43). Before implementing the selection of cyclic peptides on the phage surface, two types of linear peptides were designed to explore the selectivity of this reaction. A model peptide flanked by an N-terminal amino group and a fixed sulfhydryl group as cyclization units was first designed (H₂N-AVGSLQRGC-CONH₂, pep 1). The cross-linker pNCSBnBr with a bifunctional group (2 mM) was added to cyclize the model peptide (0.5 mM) at phosphate-buffered saline (PBS), pH 6.5, containing acetonitrile, followed by high-performance liquid chromatography (HPLC) and mass spectrometry (MS) to track the transformation process in changing conditions (Fig. 2A). Initially, 0 to 20% acetonitrile was added for the dissolution of the cross-linker, but it was found that the yields of the target products increased with the gradual decrease of acetonitrile and lastly reached the highest level with 93% in pure aqueous solution. As explored, the N-terminal amino and thiol groups of the linear peptide were smoothly transformed, and the target cyclic peptide was observed at PBS, pH 6.5, over 12 hours at 37°C. To explore the generality of the cross-link strategy, a peptide sequence with varying size (H₂N─ AYGSLQRDGC ─CONH₂, pep 3) was prepared for the cyclization study, confirming the feasibility of cross-linking in either weakly acidic or alkaline buffer. The cross-reactivity caused by ε-amines of lysine was subsequently evaluated with isothiocyanate addition. Subjecting the pep 2 with a fixed Lys and blocked N-terminal amino to the alkaline protocol, real-time HPLC/MS also revealed the absolute conversion of linear precursors at PBS, pH 9.5, over 3 hours at 37°C (44). Specifically, the ε-amines of the model pep 2 were not affected by isothiocyanate at PBS, pH 6.5, demonstrating the selectivity of different types of amino groups (Fig. 2B). This selectivity is consistent with the extra nucleophilicity of the N-terminal amino group, which has been clearly analyzed in previous studies of protein modification (23, 43). On the basis of MS analysis, the side products observed in the HPLC seem to be the result of the isothiocyanate hydrolysis (45–47) and concurrent coupling of multiple linker molecules (figs. S11 to S14). In addition, we carried out conjugation reactions between four isothiocyanate derivatives and two linear peptides: AcNH-KAVGTLSTC-CONH₂ and KLVSSFSSF. The modification outcomes were analyzed by HPLC (figs. S16 and S17), and the derivatives displayed distinct coupling behaviors. In brief, isothiocyanate derivatives with low polarity exhibited limited accessibility to amine groups in aqueous solution. Effective coupling occurred only when the benzylic bromide moiety of the linker first reacted with the cysteine residue of the linear peptide, thereby generating a proximity effect that facilitated subsequent amine coupling by the isothiocyanate, ultimately leading to cyclization. This property imposes a limitation on the accessible modification sites, such that only amine groups located in proximity to cysteine residues are capable of participating in the reaction. These results provide prospective support for the exploration of surface modification of phage proteins in the selection stage.

Fig. 2. One-step strategy for selectively forming the cyclic versions of model peptides.

(A) Schematic representation of head–to–side chain cyclization strategy for accessing cyclic pep 1 by isothiocyanate addition with the N-terminal amino and cysteine alkylation. Real-time HPLC and MS represented the increased yield of the cyclized products along with the decrease of acetonitrile in the solution. (B) Discriminative representation for handing pep 2 under conditions of different pH by side chain–to–side chain cyclization strategy. The conversion of linear precursors was observed at PBS, pH 9.5, and the highest purity was obtained in the buffer without acetonitrile. Only generating the alkylated linear peptide (*) and isothiocyanate may also undergo hydrolytic reaction with the pH shifted to 6.5.

Isothiocyanate modification on pIII and biocompatibility exploration

The successful cyclization of the model peptides was unable to support the application of this strategy into biological systems. Therefore, pep 1 with a flexible chain AGSGGSG was fused to the N terminus of the N1-N2 domain of the disulfide-free pIII for selective cyclization of pNCSBnBr. The fusion protein was purified by adding 2 mM tris(2-carboxyethyl)phosphine hydrochloride (TCEP) to the buffer to release the sulfhydryl group, and then TCEP was removed by size exclusion due to the hindrance of the reducing agent to the cyclization. The cross-linker (0.5 mM) was added to pep 1–D1-D2 (0.1 mM) at PBS, pH 6.5, over 12 hours at 37°C. Compared with fusion proteins without pNCSBnBr treatment, the cyclic target product was observed to have a substantial mass shift by MS (Fig. 3A). Given the minimal molecular weight shift after the cyclization of the fusion protein, we incorporated a tobacco etch virus (TEV) protease cleavage site between the linear peptide ANSQNSC and carrier protein for precise characterization. MS analysis confirmed the expected molecular weight of the cyclic peptide following chemical modification and enzymatic digestion, validating the effectiveness of our cyclization strategy (figs. S18 and S19). Extension of the cyclization and enzymatic cleavage assays to wild-type phage further supports this conclusion (fig. S20). Two-dimensional nuclear magnetic resonance (2D-NMR) spectra intuitively proved the formation of the thiourea skeleton after cyclization modification. The 1D- and 2D-NMR spectra of cyclic peptide ANSQNSC were recorded (table S1 and figs. S21 to S27), enabling detailed characterization of the key residues (A, C, Y, and F), and, most importantly, the identification of the key cyclic reaction site between the thiourea linker, the Ala NH₂ group, and the Cβ of the Cys residue. A total correlation spectroscopy (TOCSY) experiment showed correlations from the linker NH at δ 9.73 to their aromatic proton (δ 7.28, 7.29), the Ala NH (δ 7.68), Hα (δ 4.82), and CH₃ atoms (δ 1.32), respectively, providing evidence for the presence of the thiourea linker motif between Ala and the linker. A heteronuclear multiple bond correlation (HMBC) from benzylic hydrogen (CH₂) of the linker to Cys-Cβ undoubtedly confirmed the connectivity of the thiourea linker with the Cys residue. To reflect the efficiency of isothiocyanate modification on the phage display platform, a pull-down experiment was performed for a given phage (Fig. 3B). The N-biotin was prepared by coupling a biotin to the cross-linker (fig. S10). Meanwhile, peptide sequences were fused to the N terminus of the pIII with a flexible chain AGSGGSG encoded between the peptide and pIII. After expression with TG1 strain, the bacteriophage reduced with TCEP was purified and diluted using PBS, pH 6.5. Different concentrations of N-biotin were added to several equal phage solutions and reacted over 12 hours at 37°C. The biotinylated phage was pulled down using magnetic beads, followed by calculating the optimal yield. When the amount of N-biotin treating the phage was up to 400 μM, ~10¹¹ of the phage was captured. Notably, the titer of the control group (~10⁴), the phage unlabeled with biotin, decreased significantly after the pull-down. After the efficiency of the isothiocyanate addition was confirmed, further biocompatibility assay was performed by evaluating changes in the life cycle of the phages with the head–to–side chain cyclization strategy. Toward this end, a peptide library, N(12)-library, was created starting from pSEX81, featuring N-terminal random sequences (H2N-AX12C), a flexible chain, and pIII. Among them, the presence of the first fixed alanine is to facilitate the removal of the signal peptide after random sequence expression. A collection with a capacity of 10⁹ was generated by electroporation, and the phage library was amplified after helper phage infection and stored frozen after TCEP reduction. Equal volumes of phage solution with a titer of 10¹¹ were pipetted and treated with a series of concentration gradients of cross-linker at pH 6.5 over 12 hours at 37°C. The measured results showed no negative effects on the phage infectivity when the concentrations of pNCSBnBr were up to 600 μM (Fig. 3B). In addition, the biocompatibility of isothiocyanate modifications in alkaline conditions was also evaluated. A second phage cyclic peptide library, K(7)-library, with a random sequence (GGSGGKX7C) was constructed in a similar protocol at pH 9.5 over 3 hours at 37°C. The corresponding phage titers were measured to determine the infectious activity of the phage. As the titer indicated, there was a 200-fold decrease in phage colonies with concentrations of 250 μM (Fig. 3B). We found that the higher the buffer pH, the faster the phage’s infective ability decreased. At the same cross-linker concentration, resulting from the efficiency of the reaction being greatly increased, the binding of the linker could be completed in a shorter time. Additional titer experiments revealed distinct behavioral patterns between the two strategies: (i) The acidic condition demonstrated superior biocompatibility, albeit with a slower reaction rate, and (ii) the alkaline condition achieved faster modification but exhibited toxicity to phage (fig. S28). Therefore, the biocompatibility of the pNCSBnBr-dominated cyclization strategy was attested with the appropriate context, which is consistent with the long history of isothiocyanates as marker tools in biological systems.

Fig. 3. Validation of the application of isothiocyanate modification in biological environments.

(A) The mass shift from treating model peptide fused D1-D2 domain of the disulfide-free pIII with isothiocyanate-induced cyclization strategy at pH 6.5. (B) Comparison of phage recoveries of the N(12)-library and the K(7)-library after pull-down assays using biotin-conjugated isothiocyanate reagent (N-biotin). (C) Schematic representation of phage titers. For the N(12)-library, a series of measured phage titers were obtained after treatment with polyethylene glycol (PEG)–NaCl, TCEP, and cross-linker at different concentrations, respectively, using a weak acid protocol. For the K(7)-library, an alkaline cyclization procedure was used to detect the change in the life cycle of the phages in titers with the addition of concentration gradients of the cross-linker.

The selection and identification of binders of Cyp D

The proof of biocompatibility of pNCSBnBr led us to pursue the selection targeting the Cyp D. As a unique mitochondrial cyclophilin in the cyclophilin family, Cyp D plays a catalytic role in the cis-trans isomerization of peptidyl-prolyl with the peptidyl-prolyl-isomerase domain and regulates the opening or closing of the mitochondrial permeability transition pore on the inner mitochondrial membrane under oxidative stress (48, 49). The instability of the mitochondrial permeability transition pore is associated with a series of diseases such as ischemia-reperfusion injury, neurodegenerative disorders, and diabetes, leading to the discovery of Cyp D inhibitors as a potential clinical treatment strategy (50, 51). Approximately 10¹¹ phage particles were quantitatively pipetted from the N(12)-library and used for panning cyclic peptides targeting Cyp D after incubation with pNCSBnBr (0.4 mM) at pH 6.5 over 12 hours at 37°C and one round of purification. Three rounds of panning with streptavidin magnetic beads and real-time detection of the phage titers revealed a 31-fold increase in phage enrichment over the first round, while no variation was observed in the negative control (fig. S30). Next-generation sequencing was then applied to dissect the peptide sequences of promising binders (figs. S30 and S31). The largest proportion (7.4%) of sequences translated from the dataset was shown (Fig. 4A). The most highly enriched sequence family has been systematically cataloged (fig. S31). For identifying the binding ability of the resulting peptides from washing, linear peptide D1 was synthesized by SPSS and cyclized to CD1 with an isothiocyanate strategy. The CD1 was proven to have an affinity of 0.74 ± 0.03 μM in biolayer interferometry (BLI) and surface plasmon resonance (SPR) (Fig. 4D and fig. S32). Correspondingly, the linear precursor D1 exhibited a dissociation constant (K_d) value of 27.74 ± 2.65 μM (Fig. 4D and fig. S33). This more than 10-fold gap just emphasized the effectiveness of our cyclization strategy and the criticality of conformational constraints by thiourea to the binder. Negligible binding of the cyclic peptide ligands CD1 with unrelated proteins (Keap1) revealed not obvious interactions (fig. S34). The targeting interface of CD1 toward Cyp D was investigated using a 2D ¹H-¹⁵N heteronuclear single-quantum coherence (HSQC) NMR titration assay (fig. S35). We prepared a ¹⁵N-labeled protein solution titrated with increasing concentrations of CD1. The resulting chemical shift perturbations (CSPs) revealed that the S1 active site and the S2 pocket of the target protein were involved in binding the cyclic peptide ligand (fig. S35B). Residues Asn¹⁰², Trp¹²¹, Arg⁵⁵, Gln⁶³, and Thr⁷³ within S1 and S2 sites contributed key intermolecular interactions (fig. S35C). Notably, these pockets constitute the core region responsible for the prolyl-isomerase activity of Cyp D, suggesting that CD1 has the potential to inhibit the enzymatic function. In vitro inhibition experiments were thus performed to investigate the influence of CD1 to the cis-trans isomerization of peptidyl-prolyl of Cyp D. A median inhibitory concentration (IC₅₀) of 4.14 ± 0.39 μM was confirmed (Fig. 4E).

Fig. 4. The panning of two types of thiourea-containing cyclic peptides toward several therapeutically relevant proteins.

(A) Active sequence featuring thiourea resolved by next-generation sequencing (NGS) from selecting the cyclic N(12)-library targeting Cyp D. (B) Enriched thiourea-containing binders of Keap1 and MDM2 derived from the K(7)-library. (C) Schematic representation of the NGS analysis with enriched sequences after the three-round selection of the cyclic K(7)-library targeting Keap1. The diversity plot revealed the proportion of enrichment for each order of magnitude, and the height and width of the square columns, respectively, represented the proportion and the diversity of enriched sequences in the total library. The sequences and the most enriched peptide family are listed. Relative abundance of unique amino acid residues after the three-round selection. (D) The affinity of thiourea-containing binders and linear precursors binding to Cyp D was determined using BLI. (E) CD1 showed micromolar inhibition for Cyp D.

The discovery and validation of a cyclic peptide ligand from pH 9.5 against Keap1

Encouraged by the initial success in the development of bioactive cyclic peptides produced under the condition of pH 6.5, further exploration was performed using the cyclic K(7)-library from alkaline conditions. After subjecting the cyclization reaction to the above K(7)-library at PBS pH 9.5 over 3 hours at 37°C, the incubation buffer was exchanged with tris-buffered saline, pH 7.5, to ensure the stability of the screening process. Keap1 was selected as the first target, which was used frequently to discover PPI inhibitors in previous reports. These inhibitors block the interaction between Keap1 and Nrf2 and contribute potential therapeutic value for multiple malignant diseases (52, 53). To demonstrate the practical value of the second cyclic peptide library, streptavidin binding protein was prepared during the production of PPI inhibitors, against which multiple rounds of panning were sequentially processed. Following the monitoring of the number of phage particles, we were surprised to observe that there was a substantial increase in the output colonies, about 140 times after the second round of selection, and lastly reached a 210-fold increase in the third round (fig. S36A). We present the sequence family showing the most notable enrichment (Fig. 4C). What puzzled us was that the most enriched sequence K1 did not occur in the theoretical motif “ETGE” that appeared in previous reports (Fig. 4C) (27). Notably, peptide hits lacking the E(S/T)GE motif have been reported to exhibit potent binding to Keap1. (54). The corresponding cyclic peptide AcNH-KSLRSLQSC-CONH₂ (CK1) was synthesized for affinity determination. Consistent with the selection results, CK1 indeed exhibited a strong binding with the K_d of 46.12 ± 2.59 nM (Fig. 5A), while only the micromolar level of the linear format K1 further demonstrated the necessity of the thiourea bridge to maintain the affinity (fig. S31). To ensure the accuracy of the analytical method, other measurement tools, surface plasmon resonance (SPR) and fluorescence polarization (FP), were used to reveal the same level of results (fig. S36B). Despite lacking the canonical active motif, CK1 retained potent affinity for Keap1. The binding of Nrf2_(69–84) peptide to Keap1 was inhibited by CK1 at the inhibition constant (K_i) of 317 ± 38.6 nM (Fig. 5E). The reason for this phenomenon could be that the generation of the skeleton leads to a drastic transformation of the conformational constraints. Therefore, the dynamic simulation was performed to resolve the binding mode (Fig. 5B). The data at the microscopic level indicated that CK1 interacts with a large number of residues in the Kelch domain, including Arg³⁸⁰, Arg⁴¹⁵, and Arg⁴⁸³, which were also present in the KEAP1-Nrf2 complex (55). The theoretical support in consistency with the interactive mode suggests that the authenticity of lacking the traditional motif did not decrease the potential of the CK1 to be a potent binder. Last, a more precise demonstration of the specific binding was provided using the mutagenesis study. On the basis of the information presented in the simulation study, several critical residues of the Kelch domain (Arg³⁸⁰, Arg⁴¹⁵, and Arg⁴⁸³) that may provide the interaction force were mutated to the conserved ALA, respectively. The affinity of the target protein with CK1 was predictably reduced after the mutation of key residues was executed, and the affinity was markedly decreased more than 10-fold (Fig. 5C). These results strongly suggested that CK1 was indeed restricted to a few amino acid residues in the Keap1 Kelch domain. Ala scans were also performed to identify the binding characteristics of ligands. The mutations of Ser at the fifth and eighth amino acids result in >300- and >100-fold reductions in binding affinity, respectively (Fig. 5D). Furthermore, the degradation test showed good stability of the thiourea-containing cyclic peptide CK1 over 12 hours in calf serum (fig. S41). Together, the isothiocyanate-induced cyclization strategy is bio-orthogonal and triggers the coupling of lysine residues in an alkaline environment, resulting in the generation of previously unidentified cyclic peptide libraries with a thiourea bridge and the development of potent active ligands. This pioneering study provides a reference for the development of drugs.

Fig. 5. Schematic representation of the predictive binding pattern of CK1 with Keap1 and the affinity identification of CK1 with Keap1 or Keap1 mutant.

(A) Chemical structure of the thiourea-containing cyclic peptide CK1 and the affinity of thiourea-containing binders with Keap1 by BLI. (B) The tridimensional diagram of the CK1-bound Keap1 from MD, which indicated that CK1 interacts with several residues (Arg³⁸⁰, Arg⁴¹⁵, and Arg⁴⁸³) in the Kelch domain. The affinity of thiourea-containing binders CK1 with Keap1 by BLI. (C) There was a substantial decrease in the binding capacity of Ala-mutant Keap1 binding CK1. (D) The mutations of Ser result in >300-fold reductions in binding affinity. (E) The affinity of thiourea-containing binders CK1 with Keap1. FP assay showed that the binding of Nrf2_(69–84) peptide to Keap1 was inhibited by CK1.

The selection of ligands from pH 9.5 against MDM2

The reliability of the cyclization reaction relying on the alkaline environment was identified by targeting MDM2. MDM2 is a well-known E3 ubiquitin-protein ligase involved in the ubiquitination and degradation of a variety of proteins, whose inhibitors are mainly used in clinical cancer treatment (56, 57). The cyclic K(7)-library was prepared for panning with the same process as KEAP1, but the increase in phage titer was only maintained at fourfold after three rounds (fig. S42). The resolved major sequence, M1, was synthesized and cyclized to the cyclic peptide CM1 (fig. S43A), which was characterized for MDM2 binding using BLI. Consistent with the panning, CM1 showed only a single micromole affinity (5.22 ± 1.70 μM) toward MDM2 (fig. S43B). The contribution of the isothiocyanate-derived skeleton to MDM2 binding was determined by the unobvious signal of the linear precursor M1 (fig. S44). Further verification was provided by the mutagenesis, by which the reduction of the K_d value was up to 13-fold for CM1 binding to MDM2_ala (fig. S43D). These characterization results were in line with expectations, proving that our cyclization strategy for generating the previously unidentified skeleton in peptides was reasonable and feasible, and provides a solid foundation for clinical study against MDM2.

DISCUSSION

Exploration of previously unidentified approaches for cyclization beyond the natural peptide bond and disulfide bond and, simultaneously, the introduction of pharmacophores will expand conformational space and potential pharmacological profile for cyclic peptides. However, this inevitably increases the chemical complexity of cross-linkers and cyclization processes. Here, we developed a one-pot cyclization strategy using a simply synthesized cross-linker (pNCSBnBr) for readily loading thiourea on the skeleton of cyclic peptides. This strategy not only realizes the accessibility of thiourea-containing peptide macrocycles but also is compatible with a general display system for the generation of genetically coded libraries featuring diverse skeletons by modifying N-terminal amino or ε-amines to explore the functional ligands of several proteins. The introduction of thiourea afforded the discovery of thiourea-containing cyclic peptides with high activity and micromolar inhibitors modulating the conformations of ligands at the protein interface. Moreover, the reliance of the isothiocyanate moiety on the proximity effect further limits its accessible modification sites on phage coat proteins, thereby favoring the selective and precise cyclization. As a result, the overall impact on the phage life cycle was slight in our strategy.

Note that the isothiocyanate derivatives exhibited cross-selectivity under different conditions, resulting in diverse skeletons being formed at the modification stage. One example was that head–to–side chain cyclization was carried out in PBS, pH 6.5, under which the isothiocyanate was only selectively coupled with the N-terminal amino because of appropriate deprotonation, and the ε-amines of lysine could not be attacked, thus showing the selectivity of the reaction. Another case was using ε-amines of lysine to participate in the isothiocyanate addition in PBS, pH 9.5, to yield side chain–to–side chain cyclic peptides. Under alkaline conditions, the ε-amines of lysine with complete deprotonation in the protein or peptide could contribute to the cyclization process with the loss of selectivity.

The isothiocyanate-derived cross-linker was successfully applied for the cyclization of the linear peptide on pIII. Pull-down assay and biocompatibility assay, as critical preconditions for biological selection, confirmed that our cyclization strategy was suitable for panning on phage platforms. Therefore, two genetically coded cyclic peptide libraries were reasonably derived. The N(12)-library, produced in PBS, pH 6.5, was used to identify binders/inhibitors of Cyp D. The cyclic peptide CD1 inhibited the cis-trans isomerization of peptidyl-prolyl of Cyp D with an IC₅₀ of 4.14 ± 0.39 μM in vitro. The critical role of the thiourea-containing skeleton was verified by the substantially reduced interaction of linear precursor D1 with protein. Note that the cyclic peptide inhibitor CD1 of Cyp D (AHVTPGFMRLQGSC, IC₅₀: 4.14 μM) occurs similarly in amino acid sequence to the previously reported short peptide inhibitor (WACLQ, IC₅₀: 0.37 μM) of Cyp D, which proves our hit to be effective (58). In contrast to a previously reported nonpeptide inhibitor natural ligand, cyclosporine, CD1 was found to be inferior in hindering the isomerization ability of Cyp D. We hypothesized that this is due to the large size of the cyclic peptides. Randomized sequences more consistent with the peptidyl-prolyl Isomerase (PPIase) active site of cyclophilins would aid in developing cyclic peptides with lower nanomolar inhibition in future studies.

For the cyclic K(7)-library from the alkaline protocol, the efficient binder of the Keap1 was first selected. During the three-round panning cycle, the increase in phage recovery was more than 100-fold enrichment after the panning of the second round, suggesting that there may be a nanomolar binder enriched in the phage collection. The relative abundance of unique amino acids was clearly changed with localization, visually revealing the most accessible cyclic peptides. Consistent with the sequence analysis, CK1, without the “ETGE” motif thought to contribute important energy constraints in the Keap1/Nrf2 peptide complex, was indeed able to efficiently target Keap1 with the same level of potency (K_d = 46 nM) and mode as the discovered cyclic peptide inhibitors. Evidence for these conclusions is provided by molecular dynamics (MD) and protein mutations. The K_d from 46 nM decreases to 17.9 μM, changing the affinity more than 400-fold. The “ETGE” motif dominated the acidic moiety of cyclic peptides, while our cyclization strategy enriched the sequence with serine as the main moiety. The thiourea scaffold provided a key constraint within the interaction interface, leading to distinct active motifs. Therefore, CK1 with a divergent thiourea scaffold contributes an alternative structure of peptide for disrupting the PPI of the Keap1/Nrf2. In addition, the micromolar binder targets MDM2, suggesting that a specific design is required to adapt a diverse set of target proteins. Both targets have been frequently used in the screening of cyclic peptides with diverse cyclization scaffolds. For the Keap1, cyclization strategies incorporating electrophilic unnatural amino acids have yielded low nanomolar (K_d, 40 to 100 nM) cyclic peptides containing the canonical ETGE motif (27). It is important to note that we are not the first to report high-affinity peptides lacking the canonical ETGE motif. For instance, noncanonical disulfide-constrained bicyclic peptides without the ETGE motif have also been shown to bind Keap1 with very high affinity (54). These observations collectively demonstrate that the chemical nature of the cyclization scaffold profoundly influences the selection of peptide sequences with distinct binding profiles. For MDM2, the thiourea-containing cyclic peptide ligands exhibit a unique amino acid sequence that differs from previously reported peptide inhibitors (22, 56). Although the binding affinity is moderate, this result clearly illustrates that our cyclization strategy can generate previously unidentified recognition modes for targeting conserved binding pockets. In addition, the Cys-Cys cross-linked analogs (CCK1, CCK2, CCD1, and CCM1) of the four cyclic peptide binders (CK1, CK2, CD1, and CM1) were synthesized via reaction with DBMB. The binding values evaluated by BLI revealed a substantial decrease in potency, with measured affinities typically exceeding 100 μM (CCK1, >100 μM; CCK2, >500 μM; CCD1, >500 μM; CCM1, not determined). This result underscores the necessity of the thiourea backbone for high-affinity binding (fig. S46). This finding suggests that the selection of thiourea-bridged cyclic peptides may offer unique perspectives for developing ligands against therapeutically relevant targets, potentially expanding the ligand toolbox for challenging PPIs.

When the system was converted to alkaline, a more powerful ligand targeting Keap1 was obtained with the K_d value reaching low nanomolar, even with the reduced cyclization selectivity. The reason for this phenomenon is thought to be that the deprotonation of amino groups was more complete at the solution of pH 9.5, resulting in the isothiocyanate-induced more absolute cyclization on the phage surface. This is consistent with observations from the biocompatibility assay that the lower concentration and shorter time were required for the cross-linker under alkaline conditions. The cyclic peptide display platform with the incorporation of isothiocyanates still faces some critical challenges. First, thiocyanates could not distinguish between ε-amines and N-terminal amino groups in the alkaline buffer. Second, our cross-linker has a certain reactivity to the carboxylic acids that occurs for isothiocyanates, which has been mentioned in previous reports, leading to a negative impact on the selectivity of the cyclization.

This one-step peptide cyclization of unprotected peptides using isothiocyanate is reported, and its simplicity, feasibility, and operability will greatly facilitate the hit cyclic peptide synthesis after decoding the DNA sequence and subsequent development. Future progress will be directed toward the exploration of more flexibly constricted and structurally stable thiourea-containing bicyclic peptides. In addition, the construction of libraries with a mixture of multiple sizes (7 to 20 amino acids) would be a promising way to select tighter binders.

In summary, an efficient phage display platform for thiourea-containing cyclic peptides was established. The reliance on the diversity of the skeleton of the platform was on the cross-selectivity of isothiocyanate, resulting in the derivation of two types of cyclic peptide libraries. The libraries have been validated by the discovery of potent ligands for therapeutically relevant targets. Submicromolar binders of Cyp D are identified using the peptide library of head–to–side chain cyclization produced in a weakly acidic environment. When the thiourea skeleton was converted to side chain–to–side chain, better affinity to Keap1 was achieved, with the K_d value reaching low nanomolar. Collectively, we present a previously unknown strategy that combines phage display with a one-pot reaction approach to developing thiourea-containing cyclic peptide ligands. The versatility of this approach holds substantial potential for advancing drug discovery efforts targeting challenging proteins.

MATERIALS AND METHODS

Chemicals and materials

Fmoc-protected amino acids, Rink amide MBHA resin, and Wang resin were purchased from CS Bio Ltd. (Menlo Park, CA). Peptides were purified using a ZORBAX SB-C18 column (Agilent Technologies, Santa Clara, CA). TCEP was obtained from MedChemExpress (Monmouth Junction, NJ). MS was performed using a Q-Exactive liquid chromatography–MS (LC-MS) system (Thermo Fisher Scientific, Waltham, MA) and a GCMS-QP2010 SE (Shimadzu, Kyoto, Japan). Peptide binding was verified using an Octet R Series instrument (Sartorius, Göttingen, Germany). Absorbance measurements were recorded using a microplate reader (Tecan, Männedorf, Switzerland).

The synthesis of peptides

The appropriate amount of resin was weighed into the reactor and allowed to swell for 15 min. Fmoc groups were removed from the resin using piperidine, followed by washing with N,N-dimethylformamide (DMF; 3 × 20 ml). Fmoc─Cys(Trt)─OH (3 equiv) was preactivated in DMF, then N-hydroxybenzotriazole (HOBt; 3.5 equiv), O-(benzotriazol-1-yl)-N,N,N′,N′-tetramethyluronium hexafluorophosphate (HBTU; 3.5 equiv), and N,N-diisopropylethylamine (DIEA; 3.5 equiv) were added. The activated amino acid solution was transferred to the resin, and the mixture was stirred for 3 hours. After draining and washing the resin, another equivalent of Fmoc─Cys(Trt)─OH was added and reacted for an additional 3 hours. The resin was washed with DMF (5 × 10 ml) to remove unreacted reagents. Capping efficiency was assessed via a ninhydrin test: A small amount of resin was added to a DMF/ninhydrin solution and heated for 5 min to monitor the color change. Fmoc deprotection for the next coupling cycle was performed using piperidine (20 ml) for 15 min, and the process was repeated. This coupling-deprotection cycle was repeated until the final amino acid was coupled. The peptide was cleaved from the resin by treatment with a mixture of trifluoroacetic acid (TFA), triisopropylsilane, and water (95:2.5:2.5, v/v/v; 20 ml) for 3 hours with stirring. The resin was removed by filtration, and the crude peptide was precipitated using cold diethyl ether (Et₂O; 30 ml) and collected by centrifugation (6000 rpm, 7 min). The precipitate was washed sequentially with ice-cold Et₂O (20, 15, and 10 ml), followed by centrifugation (6000 rpm, 7 min) after each wash. The crude peptide was purified by reversed-phase HPLC using an Eclipse XDB-C18 column (99.4 mm by 250 mm, 5 μm) on an Agilent 1260 Infinity system. A linear gradient of acetonitrile (CH₃CN) and water (0.05% TFA) from 10:90 to 100:0 was used at a flow rate of 4 ml/min, with detection at 220 nm. Analytical HPLC was performed using a ZORBAX SB-C18 column (4.6 mm by 250 mm, 5 μm) on an Agilent system at 1 ml/min using a gradient of CH₃CN (solvent A) and H₂O with 0.05% TFA (solvent B) from 10 to 90% A over 30 min. Peptide identity was confirmed by LC-MS using a Q-Exactive system (Thermo Fisher Scientific).

Peptide cyclization

PBS, pH 6.5: Linear peptides (1 equiv) were dissolved in PBS (pH 6.5) to a final concentration of 0.5 mM. 1-(Bromomethyl)-4-isothiocyanatobenzene (1 equiv) was added to initiate cyclization, and the reaction was stirred at 37°C for 12 hours. Real-time monitoring of the reaction was performed by HPLC using a gradient of CH₃CN (0.05% TFA) and water (10:90 to 90:10), and mass analysis was carried out on a Thermo Fisher Scientific Q-Exactive LC-MS system.

PBS, pH 9.5: Linear peptides bearing N-terminal acetylation were dissolved in PBS (pH 9.5) at a concentration of 0.5 mM. The cross-linker was added to initiate cyclization, and the reaction proceeded at 37°C for 2 hours. The progress and identity of the cyclized products were confirmed by HPLC and MS.

Isothiocyanate modification on pIII

The pep 1, a flexible chain AGSGGSG, and the D1-D2 domain were amplified using

Pep 1–fw: 5′-CATGCTGCCATGGCTGCCGTGGGCAGCCTGCAGCGCGGTTGTGCTGGTTCAGGTGGATCCGGAGATATCAGAGCTGAAACTGTTG-3′, and L1-rv: 5′-CATGCTCGAGTTAGTGGTGGTGGTGGTGGTGGCCAGCATTGACAGGAGG-3′.

1. The fusion protein was reduced by adding 2 mM TCEP, which was subsequently removed by size exclusion chromatography, as the reducing agent interferes with cyclization. The cross-linker (0.5 mM) was then added to pep 1–D1-D2 (0.1 mM) in PBS buffer (pH 6.5), and the reaction mixture was incubated at 37°C for 12 hours. The product was analyzed by matrix-assisted laser desorption/ionization–time-of-flight–MS.

2. The peptide ANSQNSC and a TEV protease recognition site were inserted into the D1-D2 domain of pIII using primers TEV_fw3 and L4-rv. The fused gene was cloned into the pSEX81 vector. Protein expression was induced with 1 mM isopropyl β-d-1-thiogalactopyranoside (IPTG), and the resulting fusion protein was purified using Ni-NTA affinity chromatography. The fusion protein was reduced by adding 2 mM TCEP, and the reducing agent was removed using a 3-kDa Amicon Ultra centrifugal filter. The cross-linker (0.4 mM) was then added to pep 1–D1-D2 (0.1 mM) in PBS buffer (pH 6.5), and the solution was incubated at 37°C for 12 hours. Subsequently, the buffer was exchanged to 50 mM tris-HCl, 50 mM NaCl, 1 mM EDTA, and 1 mM dithiothreitol (DTT) (pH 8.0) using size exclusion chromatography. The protein (40 μM) was digested with TEV protease (1.2 to 4.8 U/μl) at 30°C for 1 hour. Digestion was monitored by HPLC and MS.

NMR titration

Two-dimensional NMR spectra of Cyp D were acquired at 25°C on a Bruker 800-MHz NMR spectrometer. NMR data were processed and analyzed using TopSpin 4.4 (Bruker BioSpin). ¹⁵N-labeled Cyp D was dissolved in 90:10 H₂O/D₂O containing 100 mM NaCl, 100 mM Na₂HPO₄, 1 mM EDTA, and 2 mM DTT at pH 7.0. A series of ¹H-¹⁵N HSQC spectra were recorded on 0.3 mM Cyp D samples with the gradual addition of various concentrations of CD1. CSPs for the amide ¹H and ¹⁵N resonances were calculated using the equation: Δδ = $\sqrt{{\left(\mathrm{\Delta }{\mathrm{\delta }}_{\mathrm{H}}\right)}^2+{\left(\frac{{\mathrm{\Delta }\mathrm{\delta }}_{\mathrm{N}}}{5}\right)}^2}$, where Δδ_H and Δδ_N are the differences in chemical shifts [in parts per million (ppm)] between the free and bound forms of Cyp D.

1D- and 2D-NMR spectroscopy studies

For sample preparation, 4.6 mg of purified cyclic peptide was dissolved in 450 ml of DMSO-d₆ and transferred into a standard 5-mm NMR tube. NMR data were collected at 298 K using a Bruker AVANCE NEO 800-MHz spectrometer with PA QXI probe at the Lanzhou magnetic resonance center. The chemical shifts are referenced to the solvent peak of DMSO-d₆ at 2.50 ppm for ¹H NMR and 39.7 ppm for ¹³C NMR. The TOCSY spectrum was recorded with 2048 by 256 data points, using 80 ms for scalar coupling transfer, with a spectral width of 11 ppm by 11 ppm using the TopSpin-supplied mlevphpp (homonuclear Hartman-Hahn transfer using MLEV17 sequence) pulse sequence. The TOCSY spectrum was accumulated with 128 scans. The ¹H-¹³C HSQC and HMBC spectra were acquired using standard Bruker pulse programs hsqcetgpsi2 and hmbcgplpndqf, with data matrices of 1024 by 256 and 4096 by 128, respectively. The spectral widths were set to 15 ppm (¹H) by 165 ppm (¹³C) for HSQC and 11 ppm (¹H) by 200 ppm (¹³C) for HMBC. HSQC and HMBC spectra were accumulated with 64 and 512 scans, respectively.

Pull-down experiment

A pull-down assay was carried out using the biotinylated cross-linker pNCSBnBr-biotin (N-biotin). The peptide sequence AVGSLQRGC or KVGSLQRGC (pep 1) was genetically fused to the N terminus of pIII by polymerase chain reaction (PCR) amplification of the pSEX81 plasmid using the following primers:

Pep 1–fw: 5′-CATGCTGCCATGGCTGCCGTGGGCAGCCTGCAGCGCGGTTGTGCTGGTTCAGGTGGATCCAAA-3′, and L3-rv: 5′-CATGTTTGGATCCACCTGAACCAGC-3′.

The flexible linker AGSGGSG was encoded between the peptide and the pIII. The resulting plasmids were introduced into Escherichia coli TG1 cells via electroporation and initially cultured in 1 ml of LB medium. TG1 cells (1 ml) harboring the pSEX81_pep 1 (AVGSLQRGC) plasmid were then expanded into 100 ml of 2× YT medium supplemented with ampicillin (50 μg/ml) and incubated at 37°C with shaking at 220 rpm. Following incubation, the cultures were centrifuged, and the phage-containing supernatant was transferred to new centrifuge bottles. Ice-cold polyethylene glycol (PEG)–NaCl solution was added to the supernatant, followed by incubation on ice for 30 min to precipitate the phages. Phages were pelleted by centrifugation, and the supernatant was carefully decanted. The bottles were inverted on filter paper for 2 min to remove residual liquid. The phage pellet was resuspended in 10 ml of PBS (pH 6.5) and transferred to a 50-ml centrifuge tube. Residual cell and phage debris were removed by centrifugation, and the resulting supernatant was transferred to a fresh 50-ml Greiner tube. TCEP (1 mM) was added to the phage solution, which was then incubated at 37°C for 1 hour. PEG-NaCl solution was again added, followed by a second 30-min precipitation on ice. The phages were pelleted, and the supernatant was removed. The pellet was washed three times with ice-cold PBS. The phage samples were incubated with N-biotin (400 μM) at 37°C for 12 hours, concentrated to 1 ml, and washed three times with PBS. Various concentrations of N-biotin were added to equal phage aliquots and incubated at 37°C for 12 hours. The phages were pulled down using magnetic beads, and the optimal yield was determined.

Construction of phage library

At pH 6.5, an N(12)-library was constructed from the pSEX81 plasmid, featuring N-terminal random sequences (H₂N-AX₁₂C), a flexible linker, and the pIII. Library inserts were introduced by PCR amplification of pSEX81 using the following primers:

AX12C-fw: 5′-CATGCTGCCATGGCTGCCNNKNNKNNKNNKNNKNNKNNKNNKNNKNNKNNKNNKNNKTGTGCTGGTTCAGGTGGATCCAAA-3′, and L3-rv: 5′-CATGTTTGGATCCACCTGAACCAGC-3′.

At pH 9.5, a K(7)-library was generated from pSEX81, encoding random sequences (GGSGGKX₇C) at the N terminus. Library inserts were introduced by PCR amplification of pSEX81 using the following primers:

KX7C-fw: 5′-CATGCTGCCATGGCTGCCGGTGGCTCAGGCGGTAAGNNKNNKNNKNNKNNKNNKNNKTGTGCTGGTTCAGGTGGATCCAAA-3′, and L3-rv: 5′-CATGTTTGGATCCACCTGAACCAGC-3′.

The primers were annealed by heating to 95°C for 5 min followed by rapid cooling on ice. DNA was synthesized via extension using the Klenow fragment (4 hours at 37°C; 15 min at 65°C). The PCR product and pSEX81 phagemid were digested with the restriction enzymes Nco I and Bam HI, and the resulting DNA fragments were ligated using T4 DNA ligase. The ligated products were electroporated into competent E. coli cells. Transformed cells were incubated at 37°C for 1 hour with gentle shaking before colony counting.

Phage library production and infectivity studies

Clones were selected from culture plates containing ampicillin (50 μg/ml) and grown in 1 liter of 2× YT medium at 37°C with shaking at 220 rpm. One hour after infection with helper phage, IPTG (1 mM) was added to induce peptide expression from the phagemid, and incubation was continued overnight at 37°C with shaking. Following expression, cells were removed by centrifugation at 8000 rpm for 20 min at 4°C. The resulting supernatant was transferred to a new centrifuge bottle, and ice-cold PEG-NaCl solution was added to precipitate the phage. After 10 min on ice, the phage pellet was collected by centrifugation at 9000 rpm for 20 min at 4°C. The supernatant was discarded, and the pellet was resuspended in 10 ml of PBS (pH 7.4) in two portions, followed by centrifugation at 12,000 rpm for 3 min at 4°C to remove residual bacterial debris. From the purified phage solution, 40 μl was retained and stored at 4°C. For reduction, TCEP was added to a final concentration of 1 mM, and the mixture was incubated at 37°C for 1 hour. The phage was again precipitated with ice-cold PEG-NaCl, collected by centrifugation (9000 rpm, 20 min, 4°C), and used in subsequent modification steps. For the N(12)-library, phage particles were incubated with various concentrations of pNCSBnBr in PBS (pH 6.5) at 37°C for 12 hours. For the K(7)-library, phages were similarly treated at pH 9.5 for 3 hours. After modification, phages were purified by PEG-NaCl precipitation and stored at 4°C.

Infectivity studies

For the N(12)-library, 40-μl aliquots of the purified, TCEP-treated (1 mM), and pNCSBnBr-modified samples (100, 200, 400, and 600 μM) were stored at each step for subsequent evaluation of phage infectivity. Similarly, for the K(7)-library, 40 μl of the purified, TCEP-treated (1 mM), and pNCSBnBr-treated samples (50, 100, and 250 μM) were collected. Diluents of each sample were prepared via 10-fold serial dilution. E. coli TG1 cells at an optical density of 0.4 to 0.5 (OD₆₀₀) were mixed with 10 μl of each dilution (10⁻⁷, 10⁻⁹, 10⁻¹¹, and 10⁻¹³), followed by incubation at 37°C with shaking at 220 rpm for 1 hour. The infected cultures were plated on LB agar supplemented with ampicillin. Colonies were counted after overnight incubation to assess infectivity.

Phage panning

For the selection experiments, 15 μg of target protein was initially immobilized onto magnetic beads. After 10 min of incubation at room temperature, the beads were washed three times with binding buffer. The beads were then resuspended in 300 μl of binding buffer, and 150 μl of blocking buffer was added to block nonspecific sites. Simultaneously, the phage library was blocked under the same conditions by incubation at room temperature for 2 hours. The blocked phage library was divided into two portions and added to the experimental and control groups, respectively, followed by incubation at room temperature for 30 min. The magnetic beads were collected using a magnetic stand and washed eight times with washing buffer and twice with binding buffer to minimize nonspecific binding. Bound phages were eluted by incubation with 100 μl of elution buffer [0.2 M glycine-HCl, pH 2.2, containing BSA (1 mg/ml)] for 15 min at room temperature. The eluent was immediately transferred to tubes containing neutralization buffer (1 M tris-HCl, pH 9.0) and mixed thoroughly. An aliquot (10 μl) of the neutralized eluent from both the experimental and control groups was used for phage titration. The remaining phage was amplified by infecting E. coli TG1 cells. Each target was subjected to three rounds of biopanning. Several individual clones were randomly selected from the plates and cultured for subsequent DNA sequencing.

Next-generation sequencing analysis

The amplified phagemids were used as templates to generate the phage libraries. DNA was amplified using primers containing sample-specific barcodes. A 50-μl PCR mixture, containing 100 ng of phage vector and 100 nM of each primer, was subjected to 20 cycles of amplification (98°C for 10 s, 49°C for 6 s, and 72°C for 20 s). The PCR products were separated on 3% agarose gels, and the desired bands were excised and purified. High-throughput sequencing was performed by Wuhan BioBank. Sequencing data were parsed and analyzed using MATLAB scripts.

BLI assay

Target proteins (Cyp D, Keap1, and MDM2) were biotinylated in buffer solution. SA biosensors (Gator Bio) were used for immobilization (59, 60). Before functionalization, the sensors were immersed in PBS (pH 7.4) for 10 min to remove the protective layer. They were then incubated in 1 μM protein solutions for 15 min and rinsed with buffer for 10 min. All immobilization steps were performed in PBS (pH 7.4), which served as the running buffer. To eliminate residual unbound protein, the sensor surfaces were treated with eight to ten 10- to 12-s injections of 20 mM phosphoric acid before initiating binding studies. Peptide solutions at varying concentrations were applied to the functionalized sensors, with rinsing in buffer following each association step. All binding experiments were carried out at 25°C in freshly prepared, filtered, and degassed PBS (pH 7.4). Before each binding cycle, the instrument was primed with buffer three to four times to equilibrate the protein-functionalized surfaces. Peptides were injected over the immobilized proteins for 4 to 6 min at a flow rate of 30 μl/min, and dissociation was monitored for 10 to 15 min. Between binding cycles, sensor surfaces were regenerated by injecting 20 mM phosphoric acid for 10 to 12 s. To correct for baseline drift due to peptide dissociation, blank buffer was injected. For strong protein-peptide interactions, 1 M GalNAc was used for regeneration. Reference sensors lacking immobilized protein were used to subtract nonspecific binding. Sensorgrams were fitted using a 1:1 binding model.

SPR assay

The Biacore X100 analysis system was used to characterize protein-ligand interactions. Biotinylated molecules were reversibly captured using a Biotin CAPture Kit (GE Healthcare). The biotinylated target protein was first injected for immobilization onto the chip surface. A series of ligand samples at varying concentrations was then prepared and loaded sequentially into the sample pool. The entire detection procedure lasted 6 hours. After completion of the assay, the results were analyzed and exported using the instrument’s data analysis software. The chip was regenerated using a commercial regeneration buffer before reuse.

Fluorescence polarization assay

The affinity of FAM-CK1 for Keap1 was measured by fluorescence polarization assay. Working solutions were prepared by diluting the FAM-CK1 stock with 1× PBS to a final concentration of 30 nM. The Keap1 solution was diluted in PBS to final concentrations of 0, 5, 10, 20, 40, 50, 100, 200, 400, and 800 nM. Subsequently, 160 μl of reaction solution was added to a 96-well plate and incubated at room temperature for 10 min. Fluorescence anisotropy was measured using a TECAN microplate reader with excitation and emission wavelengths of 485 and 535 nm. Each sample was analyzed in triplicate.

Molecular simulation

The Desmond package from Schrödinger was used to simulate the interaction. The molecular docking generated the initial static protein-ligand complexes, while MD simulations traced atomic movements over time, offering insights into the ligand-binding orientation in a physiological environment. Maestro’s Protein Preparation Wizard optimized the initial protein-ligand complex, and the system was assembled using the System Builder tool. The system was solvated using the TIP5P water model within an orthorhombic box. The OPLS 2005 force field was used during the production run. Neutralization was accomplished by adding NaCl to a concentration of 0.15 M. Equilibration included 1 ns in the NVT ensemble (300 K, 1 atm), followed by an additional 1 ns in the NPT ensemble.

Supplementary Materials

This PDF file includes:

Supplementary Text S1 to S3

Figs. S1 to S64

Tables S1 and S2