Two-Helix Supramolecular Proteomimetic Binders Assembled via PNA-Assisted Disulfide Crosslinking

Peptidic motifs folded in a defined conformation are able to inhibit protein-protein interactions (PPIs) covering large inter-faces and as such they are biomedical molecules of interest. Mimicry of such natural structures with synthetically tractable constructs often requires complex scaffolding and extensive optimization to preserve the fidelity of binding to the target. Here, we present a novel proteomimetic strategy based on a 2-helix binding motif that is brought together by hybridization of peptide nucleic acids (PNA) and stabilized by a rationally positioned intermolecular disulfide crosslink. Using a solid phase synthesis approach (SPPS), the building blocks are easily accessible and such supramolecular peptide-PNA helical hybrids could be further coiled using precise templated chemistry. The elaboration of the structural design afforded high affinity SARS CoV-2 RBD (receptor binding domain) binders without interfer-ence with the underlying peptide sequence, creating a basis for a new architecture of supramolecular proteomimetics.


Introduction
Proteomimetics are macromolecular structures with a defined tertiary fold that allows them to serve the function of their template protein. [1] Much of the appeal of such structures is generated by their ability to answer the complex proteinprotein interaction challenge, particularly their suitability to disrupt contacts with large surface and multiple-shallow peptide binding sites. [2] A growing number of architectures are being explored with variety of tertiary fold stabilization strategies, for example, through employing (partially) artificial backbones with predictable folding patterns. [3,4] In some cases, other biopolymers, such as nucleic acids, are introduced to take advantage of programmable nature of the constraint provided by the Watson-Crick base pairing. [5,6] This approach has been successfully demonstrated for peptidomimetics, where the therapeutic peptide bioactive conformation could be stabilized via short complimentary PNA sequences [5,[7][8][9] or for conformational switches where nucleic acid hybridization constraints were used to modulate or inform of binding events. [10][11][12][13] We recently showed that even more complex structures could be assembled, for which we used PNAs to constrain peptide loops and display them in a suprabody scaffold, in analogy to presentation of CDRs (complementarity determining region) of variable loops in antibodies. [14] It is also well established that ligands paired through hybridization can interact synergistically with a target. [15][16][17][18][19][20] Libraries of such hybridization-driven hybrids could be screened to identify the optimal binding geometry. [21,22] However, in those approaches the peptides were not purpose-fully subjected to a higher order structuring, effectively amounting to limited pre-organization and not taking full advantage of proteomimetic approach.
Important structural motifs ubiquitously mediating PPI are α-helices. [23] As such, they are a straightforward target of mimicry efforts with various classes of potential inhibitors and a notable example of miniature proteins. [24] Among those, there has been considerable interest in minimal size mimetics of Z domain scaffold 'affibodies' [25][26][27][28] (3-helix bundles) [29][30][31] that offer superb capabilities to engage biomolecules for the purposes of diagnostic and therapeutic interventions [32] and as a general multiuse tool in biotechnology. [33,34] The interaction with the target is mediated by several residues located on the same face of two helices. As such, the bulk of minimization efforts has been focused on truncation of the third non-participatory helix ( Figure 1a) and stabilization of the subsequent 2-helix structure via i) introduction of point mutations on both binding and nonbinding faces of the helices [25,35] ii) optimization of the interhelical hinge region, [27] and iii) replacing non-covalent contacts via permanent covalent surrogate. [36][37][38] However, all current approaches rely on helix-turn-helix architecture, which although is the most straightforward natural binding motif for antiparallel aligned helical dimers, [39] often requires substantial sequence optimization in order to compensate for the misalignment in internal knob-into-hole hydrophobic packing. [40] As a broadly applicable solution bis-aromatic linkers were proposed to cross link relevant helix dimers with incorporated reactive side chains in the place of an existing salt bridge, in addition to core optimization. [41][42][43] An alternative approach would be to maintain a structural but otherwise modular element, which would constrain the peptides in the required fold. In one example, the spontaneous formation of a triple helix was used to promote the formation of a coiled helix bundle, however the overall fold was found to be highly concentration dependent, with significant contribution of higher order oligomers. [44] In this work, we propose a simpler supramolecular solution, in which the helical peptides are directionally brought together by short 6-mer PNA tags. Their spontaneous hybridization in aqueous medium removes the needs for a covalent loop, while helix packing and fold arrangement is enforced by a precisely positioned disulfide crosslink, obviating the need for further sequence optimization ( Figure 1b).
We focused on a sequence from the known miniprotein binder of RBD (receptor binding domain) of the SARS-CoV-2 spike named LCB1. It is a small 56 residue (6.8 kDa) de novo designed 3-helical scaffold that was found to bind its target with sub-nM affinity. [45] The availability of refined cryo-electron microscopy maps (PDB: 7JZU) of the complex opened up the possibility to rationally position the disulfide crosslink, while starting from an optimized, high affinity binding interface ensured higher chances of mimicry success. [46]

Results and Discussion
Using insights from the structure of LCB1 in complex with RBD, we identified a suitable position for the PNA modules within the turn region. We hypothesized that breaking the linear sequence at Gly20, while removing the third helix starting from Gly39, would place the hybridized duplex away from the native interaction interface and thus should not sterically impact the association of the proteomimetic with its target (Figure 1c). Next, we inspected the side chains in the central core (a and d heptad positions) to establish optimal disulfide placement, which revealed several suitable pairs ( Figure 1d). We focused on sites distant from the primary supramolecular linkage, as it was shown that placing a crosslink closer to the original loop position likely abrogates binding. [26] In our first attempt, we decided to mutate the hydrophobic core Ile9 and Ile32 (which Cβs are sufficiently close 5.35 < 5.5 Å) [47] to Cys, which upon oxidation should be able to form a stabilizing disulfide bond. [48] In this way, two helical peptides, referred to as R (red) and B (blue) were generated, together with their 6-mer PNA containing hybrids R PNA and B PNA (Figure 2a). In order to form helical dimers, the compounds were incubated in a suitable aqueous alkaline oxidation buffer with DMSO [49] (DMSO / 40 mm NH 4 HCO 3 , pH 8.0, 1 : 2), which ensured the solubility of the substrate hybrids while preserving the high fidelity of PNA hybridization. [50] Expectedly, when mixed in equimolar ratio, B and R peptides (devoid of PNA) yielded mixture of all possible products ( Figure 2a); in contrast, the nucleic acid containing hybrids formed almost exclusively the desired heterodimer as verified by LC-MS analysis (Figure 2b), consistent with hybridization-driven templated reaction. [51][52][53][54] Mixing the substrates at 50 μm and incubation for 24 h resulted in a low apparent conversion (< 25 %), with mostly unreacted starting materials therefore additional conditions and additives were screened (Figure 2c). We have previously shown that the use of a similar 4-mer PNA tag allows for efficient templated chemistry (native chemical ligation) using substrates at 10 μm, [9] nonetheless due to the slower kinetics of the air/DMSO disulfide formation we achieved higher product formation (> 80 %, Figure 2d) chiefly by increasing the substrates concentration to 100 μm, while a slight elevation of temperature to 37°C allowed for more reproducible results as well as desirable faster initial period. Use of additives such as bridging oxidant [55] (100 μm Cu(II) salt) or general redox buffer [56] (1 mm glutathione disulfide -GSSG) although resulted in speeding up the initial product formation, overall led to lower conversion. The use of a 6-mer tag ensured stability of the PNA duplex at this temperature, since an identical DNA sequence should have a T m > 45°C, and it is well established that the PNA/PNA complementarity is inherently stronger. [57][58][59] Moreover, upon covalent cross-linking of the adduct, the thermal stability of the PNA duplex becomes independent of concentration, since the hybridization is intramolecular, and further increases due to the high effective concentration. [7,9] The heterodimeric products were purified via RP-HPLC and their affinity towards the immobilized target protein -RBD (see Figure S1 for sequence) was measured via SPR (Figure 3), together with the appropriate controls without Cys mutations (denoted as R(À C) and B(À C); see Figure S2 for full sequences and legend). Direct co-injection of peptides constituting the interacting helices (up to 2 μm) was not sufficient to observe any affinity towards the target protein (see Figure S3 for all sensorgrams and kinetic parameters). However, when paired  (7)). Next, we evaluated the effect of the  introduced Cys mutations by mixing B PNA and R PNA without prolonged incubation (no disulfide formation), which revealed a small gain in affinity induced by the Ile replacement. Pleasingly, the construct with preformed disulfide (B-R) PNA exhibited the highest affinity of K D = 85 nm, suggesting that the heterodimeric peptide-PNA indeed folded into the desired species in solution. Correct pre-organization of the interface was evidenced by a notable increase in k a > 10 5 m À 1 s À 1 (Figure 3 chart) , past the boundary where an interaction starts to be limited by diffusion rather than conformational change. [60] Interestingly, replacing native hydrophobic side chains pointing towards the now removed third helix/core (Ile5, Leu16, Val28, Leu31) with polar mutants, although usually beneficial, [25] in our hands abolished the observed RBD affinity, suggesting a possible role of the existing core arrangement in generating the appropriate knobs-into-holes helical alignment ( Figure S4).
Although our first attempt at execution of the planned architecture resulted in a relatively tight binder, we found that the native LCB1 miniprotein interface is not fully recapitulated, with the original three helix bundle affinity for RBD of K D = 11.5 nm (see Supporting Information Note explaining the difference between the measured here and reported LCB1 affinity). Furthermore, the (B-R) PNA heterodimer showed a desirable fast on-rate but the introduction of the disulfide crosslink caused a noticeable 3-fold drop in off-rate, which is a preferred parameter for a potential inhibitory therapeutic. [61] Such a change may be driven by loss of PNA duplex mobility due to the rigidification of the binding interface that induced a potential conformational clash or loss of favorable contacts. [62] As a control informing on relative impact of the PNA we also prepared 2-helix (B(À C)R(À C)) PEG3 in which the heterodimer is linked covalently via extended PEG 3 chain. To our surprise, this compound performed better than the supramolecular version R(À C)B(À C) PNA (Figure 4a) supporting the hypothesis that the position of the PNA chain is non-optimal. Therefore, in the next step we decided to investigate the influence of the PNA attachment method by varying the length of a flexible polyethylene glycol (PEG) linker between the helical peptide and nucleic acid components. Upon the introduction of a single 9atom PEG chain the affinity of the disulfide heterodimer did not change substantially (Figure 4b). On the other hand, further extension resulted in a marked improvement of the affinity up to K D = 11 nm.
We speculate that a PEG 2 linker provided enough distance for the optimal arrangement of the PNA duplex and/or the alignment of coiled helices. Finally, we explored a different disulfide position within the core but more distant from the PNA tags. Thus, replacing Ile5 and Phe35 with Cys residues allowed for an additional increase of the complex stability to K D = 6.4 nm, importantly driven by slowing the dissociation rate to k d = 3.9 × 10 À 4 s À 1 and surpassing in this way the native acetylated miniprotein, with the complex displaying more than 60 times longer half-life.

Conclusion
Summarizing, we embarked on research to demonstrate a new supramolecular proteomimetic design, trying to recapitulate the interface of dimeric coiled helices. Starting from a high affinity 3-helix bundle miniprotein, we applied the standard third helix truncation and added PNA chains to drive the insolution assembly and templated crosslinking chemistry. Rationally positioning of the disulfide bond allowed us to form moderately strong RBD binders via conformation stabilization. In a key finding, we realized that it was necessary to spatially separate the supramolecular duplex from the binding interface, and elongation of the linker further improved the affinity of the hybrids. Together with improved locking of the coiled helices through the disulfide position change, the optimization efforts led to the full recreation of the native level of binding strength but with more favorable dissociation profile.
These results point to a possible platform architecture for scanning libraries for the purpose of affinity maturation or, owing the recent advances in affinity selection-mass spectrometry, [63] combinatorial discovery of new peptidic binding motifs from highly diverse pools exploring other crosslinking chemistries than disulfide formation. However, prior to this pursuit, certain limitations of the current design must be addressed, namely poor solubility of hybrids (< 10 μm), likely due to the use of unmodified PNAs. [64,65] The efforts to explore these exciting avenues are currently on the way in our laboratory.

Experimental Section
SPPS synthesis, purification of the constructs and heterodimer formation: All peptide-PNA hybrids were synthesized using standard solid-phase peptide synthesis (Fmoc chemistry with HATU activation [14] on an Intavis AG Multipep RS instrument) with 5-20 mg of novaPEG Rink amide resin (EMD Millipore, standard loading 0.3-0.4 mmol/g). All materials but the PNA monomers [66] were purchased from commercial vendors and used as delivered. The peptides (peptide-nucleic acid hybrids) were cleaved from the resin with a cocktail (300 μL/mg beads) made of TFA/ethane-1,2dithiol (EDT)/water/triisopropylsilane (TIPS) (95 : 2.5 : 2.5 : 1, v/v/v/v) for 2.5 h at room temperature, followed by standard diethyl ether precipitation. All compounds were purified by a reverse phase semipreparative HPLC Agilent Technologies 1260 Infinity with ZORBAX 300SBÀ C18 column (9.4 × 250 mm) using linear gradients of acidified MeCN (B) in water (A) (mobile phase A = 0.1 % TFA in water, B = 0.1 % TFA in acetonitrile, v/v) at a flow rate of 3 min/mL. The fractions were analyzed by MALDI-TOF-MS (Bruker Daltonics Autoflex TOF/TOF using a 40 mg/mL solution of 2,5-dihydroxybenzoic acid (DHB) as matrix), combined and lyophilized. The disulfide formation was performed by 16-24 h incubation of 100 μm of purified substrates in a 2 : 1 mixture of 40 mm NH 4 HCO 3 pH 8 and DMSO at 37°C, and further purified using abovementioned method.

SPR measurements:
All PNA containing hybrids were subjected to annealing process (5 min at 95°C) before any manipulations. SPR measurements were performed using a Biacore T200 instrument (GE Healthcare) with the RBD-His protein immobilized on the sensor chip either via 1) covalent attachment to a CM-5 dextran functionalized surface using EDC and N-hydroxysuccinimide according to the manufacturer's built-in protocol or 2) capture on a NiNTA chip, both up to 900 RU surface density. For the reference flow cell, the surface was capped with ethanolamine or left untreated for CM5 and NiNTA chips, respectively. The following parameters were used to perform kinetic and affinity analysis: flow rate 20 μL/min, regeneration by 6 s PBSÀ P + pH 2.0 treatment or standard 350 mm EDTA treatment for CM5 and NiNTA chip respectively. All analyses were performed using 1 × PBSÀ P + (28995084, Cytiva) as the running buffer. SPR data was analyzed with the Biacore Evaluation Software (GE Healthcare), in which it was fitted to a 1 : 1 Langmuir binding model.