Rapid Evaluation of Staple Placement in Stabilized Alpha Helices using Bacterial Surface Display

Abstract

There are a wealth of proteins involved in disease that cannot be targeted by current therapeutics because they are inside cells, inaccessible to most macromolecules, and lack small-molecule binding pockets. Stapled peptides, where two amino acids are covalently linked, form a class of macrocycles that have the potential to penetrate cell membranes and disrupt intracellular protein-protein interactions. However, their discovery relies on solid phase synthesis, greatly limiting queries into their complex design space involving amino acid sequence, staple location, and staple chemistry. Here, we use Stabilized Peptide Engineering by E. coli Display (SPEED), which utilizes non-canonical amino acids and click-chemistry for stabilization, to rapidly screen staple location and linker structure to accelerate peptide design. After using SPEED to confirm hot spots in the mdm2-p53 interaction, we evaluated different staple locations and staple chemistry to identify several novel nanomolar and sub-nanomolar antagonists. Next, we evaluated SPEED in the B cell lymphoma 2 (Bcl-2) protein family, which is responsible for regulating apoptosis. We report that novel staple locations modified in the context of BIM, a high affinity but non-specific naturally occurring peptide, improve its specificity against the highly homologous proteins in the Bcl-2 family. These compounds demonstrate the importance of screening linker location and chemistry in identifying high affinity and specific peptide antagonists. Therefore, SPEED can be used as a versatile platform to evaluate multiple design criteria for stabilized peptide engineering.

Introduction

It is estimated that ~85% of disease-associated proteins are “undruggable”: inside the cell and inaccessible to large biologics but lacking small molecule binding sites ¹. Stapled peptides, short chains of amino acids where two residues are covalently crosslinked, have been proposed as one type of therapeutic framework to fill this gap.^2–4 Covalent sidechain crosslinking has the potential to improve target affinity, facilitate cell entry, and enhance proteolytic stability.⁵ Since most protein-protein interactions are mediated through alpha-helical secondary structure, there is a natural precedent for binding peptides to be locked in this conformation. The staple forces the peptide into a conformation with enhanced alpha helicity, decreasing the entropic penalty of binding and increasing affinity.⁶ The larger size of the peptide enables a greater binding surface area, thereby creating the potential for high affinity interactions without the need for a deep hydrophobic binding pocket. The intramolecular hydrogen bonding in the peptide backbone can reduce the energy barrier to diffuse across the lipid bilayer by shedding solvating water molecules.⁷ Additionally, since most proteases encountered by a peptide in vivo recognize linear conformations, a peptide in a helical structure tends to have a longer pharmacological half-life.⁸

Despite this promise, stapled peptides have several challenges that must be overcome. These agents must be engineered with high enough membrane permeability and intracellular stability to engage sufficient levels of target based on the binding affinity and specificity.⁹ Engineering these properties is challenging, and the discovery and development of stapled peptides typically relies on solid-phase synthesis, where peptides are chemically synthesized one amino acid at a time, limiting throughput and evaluation. A peptide’s sequence and staple location have 10²⁶ possibilities even for a simple peptide of length 20, which presents both a design opportunity but also an enormous challenge. Therefore, rational design has been the common approach, with the number of evaluated sequences often on the order of dozens.^7,10–20 Previously, we have developed Stabilized Peptide Engineering by E. coli Display (SPEED) to discover high affinity MDM2 binders.²¹ In this work, we extend this approach and demonstrate its ability to accelerate stapled peptide design by identifying hotspot residues, functional staple locations, and diverse chemical linkers.

Bacteria possess several unique abilities suitable for tackling the challenges of stabilized peptide design. First, most stapling chemistries are not compatible with the 20 canonical amino acids, and other surface-based presentation approaches such as phage- and yeast surface display do not currently have high enough non-natural amino acid incorporation efficiencies to staple on the cell surface, although progress is being made.^5,22–24 Meanwhile, bacteria are able to incorporate many types of non-natural residues, whether from methionine substitution, stop codon read through, or other genetic code manipulation.²⁵ Of particular importance is azidohomoalanine residues which contain copper catalyzed click chemistry (CuAAC) suitable azides, and demonstrate exemplary efficiencies of >95% incorporation.^26–28 SPEED leverages this high incorporation efficiency to display two azides directly on the cell surface to form a stapled peptide with an intramolecularly reacted bisalkyne (Figure 1).²¹ The modularity of this reaction scheme enables the use of any bisalkyne for reaction, meaning that SPEED is equipped to engineer both the peptide sequence and staple. The bio-orthogonality of click chemistry gives bacteria an additional advantage over chemistries that use canonical amino acids, such as lactam bridge formation or cysteine alkylation, which can interfere with other proteins present on the cell surface and may impact peptide property measurement. Ribosome-based display can incorporate many types of non-natural amino acids that facilitate peptide stapling, but like phage, their small size makes it challenging to use in assays that rapidly measure stapled peptide properties like flow cytometry and fluorescent activated cell sorting (FACS).²⁹ Similarly, one-bead-one-compound approaches have enabled the measurement of ~10³ stapled peptides in parallel but rely on mass spectrometric based methods and imaging that render property measurement difficult.³⁰ This approach also relies on library members having unique masses, meaning that residues with the same mass, like leucine and isoleucine, cannot be distinguished. Furthermore, this approach is less well-suited to evaluate staple location and peptide sequence in tandem as staple location cannot be randomized efficiently using solid phase peptide synthesis. In summary, E. coli possess several attractive traits: facile genetic manipulation, efficient non-natural acid incorporation, modular bisalkyne linker chemistry, and compatibility with high-throughput screening methods.

Figure 1: Stabilized Peptide Engineering by E. coli Display (SPEED).

DNA encoding peptide is transformed into E. coli and expressed on the cell surface by incubating bacteria in an azide containing methionine analog. After click chemistry is performed directly on the cell surface, bacteria are incubated with fluorescent epitope tag antibody and protein target. Finally, bacterial cells are analyzed via flow cytometry.

In this work, we demonstrate that bacteria enable rapid measurement of affinity and specificity of peptides with different staple locations, staple types, and sequence mutations in the context of two systems, murine double minute-2 (mdm2) and B cell lymphoma 2 (Bcl-2) targeted peptides. In the first model system, the critical p53 tumor suppressor transcription factor is rendered unstable by mdm2 overexpression.³¹ Inhibition of mdm2 by a stabilized p53-like peptide (PLP) reduces the viability of cancer cells.¹⁷ In the second system, overexpression of Bcl-2 proteins leads to inhibition of apoptosis factors that prevent cancer cells from dying. Inhibition of Bcl-2 proteins by stabilized BH3 peptides regenerates cells’ ability to undergo apoptosis.³² We use SPEED to design novel stapled peptides in both these systems in the pursuit of higher affinity, greater specificity, and more structurally diverse molecules. We then translate these peptides from the cell surface to solution phase binding to confirm that the bacterial surface display captures soluble peptide properties. The results demonstrate that bacterial surface display can be used to accelerate stapled peptide engineering.

Methods

Purification of Mdm2 and Bcl-2 proteins

Mdm2-GST was expressed and purified as described previously.^21,33 Briefly, Mdm2-GST genes were ordered from IDT and cloned into the pqe80L vector with BamHI and HindIII. Protein was harvested from pLysS BL21 DE3 E. coli cells and subsequently purified with agarose glutathione beads (Thermo Fisher) in phosphate buffer at pH 7.4. After washing with more phosphate buffer, proteins were eluted with 50mM Tris pH 8.0 and 10mM reduced glutathione before purification via size exclusion chromatography in PBS with 1% glycerol and 1mM DTT. Protein was concentrated using a 10kDa concentrator and labeled with NHS-EzLink-Biotin (Thermo Fisher) and excess biotin was removed via dialysis in PBS with 10% glycerol and 1mM DTT.

Bcl-2 genes were ordered from IDT and cloned into the pqe80L vector using BamHI and HindIII with an N-terminal His tag. Briefly, LB with ampicillin was inoculated with cells from an overnight grow-out until reaching OD₆₀₀ ~1.0 and induced with IPTG at 1mM for 5 hours at 37°C. Cells were resuspended in Tris-HCl buffer at pH 8.0 with 0.5mM NaCl, 5mM imidazole, protease inhibitor, and 2mM DTT, sonicated, and centrifuged at 35,000g x 30 min at 4°C. The supernatant was then loaded onto 2mL of prewashed Ni-NTA resin and washed with 10–20mL of resuspension buffer. For protein intended for subsequent labeling with NHS-biotin or NHS-fluorophore, it was buffer exchanged on resin into PBS with 2mM DTT and 5mM imidazole. Protein was eluted in the corresponding buffer supplemented with 500mM imidazole. Proteins were assayed for purity and yield using SDS-page gel and spectroscopy. Proteins were labeled in 0.1M NaHCO₃ and NHS-biotin or NHS-fluorophore was added at a 10:1 NHS:protein ratio. Proteins were then purified using size exclusion chromatography on a S200 10/300 increase GL or S75 10/300 increase GL column in PBS/1mM DTT/1% glycerol. Fractions corresponding to Bcl-2 proteins were concentrated using a 10kDa molecular weight cut-off filter supplemented with 10% glycerol at 4°C. Degree of labeling was quantified using a fluorescent biotin quantification kit (Thermo Fisher) or quantified directly using spectroscopy. Typically, a DoL greater than 0.3 gave sufficient separation between flow cytometry signal and noise.

Bacterial surface display and on-cell click chemistry

Primers encoding peptides were purchased from IDT and incorporated into the eCPX2-pqe80L plasmid^21,34 with a 2 step PCR protocol using Q5 Hot Start Polymerase, SfiI restriction enzyme, and T4 ligase (all from NEB). An extended peptide linker containing an HA tag (final sequence YPYDVPDYAAGGGSGGGS) was incorporated into the bacterial cell surface display scaffold to normalize binding to display level with two-color labeling.³⁵ The peptide sequence was confirmed via Sanger sequencing (Michigan Advanced Genomics Core or Eurofins Genomics). Methionine auxotrophic E. coli cells (TYJV2 strain) were used in all surface display experiments and were grown overnight in M9 media containing 4mg/mL methionine and 100ug/mL ampicillin. Then, media was inoculated with the overnight culture at a 1:20 ratio for 150 minutes at 37°C. Cells were then switched to M9-amp with no methionine for metabolic depletion for 30 minutes at 37°C, followed by a 4 hour induction in M9-amp with 4mg/mL azidohomoalanine and 1mM IPTG at 22°C. At this point, bis-azide containing peptides on the surface of bacteria were reacted to form stapled peptides in 50uM CuSO₄, 250uM THPTA, 500uM (p53-like-peptides) or 100uM (Bcl-2) propargyl ether for 4 hours at 4°C. The extent of reaction was determined as previously reported and shown in Figure S1.²¹To measure the affinity of a peptide, the peptide-displaying bacteria was incubated for at least 4 hours on ice with 6–8 concentrations of biotinylated or fluorescently labeled Mdm2 or Bcl-2 protein. Cells were washed once with PBS/ 0.1% BSA, labeled with secondary antibodies (anti-HA-Alexa Fluor 488 for display level measurement and streptavidin-Alexa Fluor 647 for biotinylated protein detection) for 15 minutes on ice, washed again, and then analyzed via flow cytometry (Attune NXT or BioRad Ze5).

Mdm2 library generation and sorting

A methionine codon-free version of eCPX (except for the start codon) was used as a PCR template for generating the NNC library as described previously.²¹ After digestion and ligation, the library was transformed into electrocompetent methionine auxotrophic TYJV2 E. coli (a generous gift from J. van Deventer) achieving a library size of approximately 3 × 10⁸ members.³⁶ TYJV2 cells were used for all sorting experiments. Bis-alkyne reacted cells, in tenfold excess of the library diversity, were first labelled with 18 nM MDM2-GST-biotin in 0.2% PBS/BSA. Cells were then washed once with PBS/BSA and incubated with 500 uL MyOne C1 beads (Thermo Fisher) for 25 min at 4 °C with gentle rotation using MACSmix (Miltenyi Biotec). Magnetic beads were pulled down by a DynaMag-5 magnet (Thermo Fisher) and washed gently with 5 mL PBS/BSA. DNA was isolated from bead-bound cells using a Qiagen miniprep kit according to Ramesh et al.³⁷ Resulting DNA, generally 100–500 ng total, was transformed into fresh TYJV2 cells for additional sorting and analysis. For each linker library, 5 mL of cells were grown out, induced, and reacted as described above. Serial rounds of FACS were carried out with increasing stringency. For the first round of sorting, cells were incubated with 4 nM MDM2-GST-AF647 (collecting 2% brightest cells), second round 1 nM (collecting 0.5% brightest cells), and the third and fourth rounds incubated with 1 nM of MDM2-GST-AF647 first and then 30 nM MDM2-GST-AF488 as described previously to select for tight binders regardless of display level (roughly 1% of cells collected).²¹ Sorting was carried out in a MoFlo Astrios FACS instrument, and plasmids extracted and re-transformed into TYJV2 cells for further analysis and sorting if needed.

Mdm2 deep sequencing

Plasmids were isolated from bacterial pellets by miniprep (Qiagen). Illumina sequencing regions were added to either side of the eCPX-peptide gene by PCR amplification using Q5 DNA polymerase (New England Biolabs) following the manufacturer’s protocol and primers 1-F and 1-R (see Figure S2) PCR products were cleaned by gel extraction and re-concentrated using a ZymoClean Clean & Concentrate kit. Another PCR amplification was performed also using Q5 to add the P5 and P7 Illumina sequences for flow cell annealing as well as a unique 8 letter barcode on each end of the amplicon for demultiplexing using primers 5-(0–7) and 7-(0–7).³⁸ The second round of PCR was cleaned and re-concentrated identically. DNA concentrations were quantified using a QuBit fluorimeter, pooled, and submitted to the University of Michigan DNA Advanced Genomics core for analysis. Samples were demultiplexed by sorting samples with perfectly matched barcodes and ones that differed by up to one base pair. Samples were then analyzed with FastQC and samples with a PHRED score of less than 36 were discarded. Fastq files were then analyzed using Python scripts with Biopython and SeqIO packages. Forward and reverse reads were pairwise analyzed, discarding any sequences with differences in base pairs. The portion of the read that corresponds to the p53-based peptide was translated.

Synthesis and preparation of peptides

Bcl-2 and Mdm2 peptides were synthesized using Fmoc chemistry on a CEM Liberty Blue microwave peptide synthesis instrument as described previously or obtained through the University of Michigan Proteomics and Peptide Synthesis Core.²¹ The bis-alkyne stapling reaction was performed as described previously. Briefly, peptide in 0.1M NaHCO₃ was added to a 1:1:6:1.2 ratio of CuSO₄: THPTA: Sodium Ascorbate: bis-alkyne (propargyl ether, heptadiyne, or other) and reacted for 16 hours under gentle mixing at room temperature. Peptides were purified using reverse phase liquid chromatography on a C18 column using 0.1% TFA H₂O / acetonitrile gradient. Fractions were collected and lyophilized before analysis by mass spectrometry using ESI or MALDI through the University of Michigan Mass Spectrometry core. HPLC chromatograms, mass spectra, and tabulation of masses can be found in Figures S3–6 and Table S1. Chemical structures for all compounds in the study are tabulated in Figure S7. Circular dichroism measurements and extinction coefficient calculations are available in Figure S8 respectively.

Circular Dichroism

Mdm2 or Bcl2 peptides were dissolved in 1:1 v/v H₂O: acetonitrile at approximately 0.1mg/mL. Peptides were added to a 3mL quartz cuvette and analyzed on a Jasco J-815 CD Spectrometer at 100nm/min at 22°C. Data is reported as baseline corrected in blank solvent. We used the BeStSel webserver to calculate alpha helicity.³⁹

Biolayer Interferometry

Peptides and Bcl-2-biotin proteins were quantified using A₂₈₀ measurements and added to 0.3% BSA in PBS pH 7.4. One well was prepared for each Bcl-2-biotin protein at 100–500nM and no dependence on sensor loading concentration was observed. To obtain multiple binding curves for each peptide-protein interaction, 5 wells with concentrations varying from 10–1000nM of peptide were prepared along with 6 wells with 0.3% BSA/PBS for dissociation. An OctetRED 96 instrument with super streptavidin tips was used for all BLI experiments. The following times were used: load, 900 seconds; wash, 900 seconds; baseline, 60 seconds; association, 1200 seconds; dissociation, 3600 seconds. Data was analyzed using GraphPad Prism v10.0 using a single-phase association and dissociation for all data. Representative biolayer interferometry data can be found in Figure S9 and S10.

Results

The bacterial surface confirms hotspot residues via alanine scanning mutagenesis

Protein-protein interactions are known to be driven by a select subset of surface exposed residues known as ‘hot-spot’ residues, contributing up to 80% of the interaction strength.⁴⁰ Molecular recognition of p53-like-peptides towards mdm2 has long been known to be dominated by three hotspot residues: Phe¹⁹, Trp²³, and Leu²⁶.⁴¹ We tested the ability of the bacterial surface to identify hot spot residues via alanine scanning, where each wild type amino acid was replaced by alanine (Figure 2). The complete set of titration curves for the alanine scanning mutagenesis data is located in Figure S11. These results were compared with molecular mechanics approaches⁴² and statistical approaches⁴³ as validation for surface display of p53-like peptides. We generally found strong agreement between all three approaches; non-hotspot residues did not affect the affinity while F19A and W23A mutants did not bind at any measured concentration (where we display >10,000 relative K_d). L26A yielded a smaller decrease in binding affinity, agreeing with the fact that Leu²⁶ contributes less binding free energy than Phe¹⁹ or Trp²³. Hotspot residue identification is an important tool in the design of protein-protein interaction inhibitors as it can inform which sites are more amenable to optimization. The magnitude of binding affinity decrease can aid in the decision to preserve hotspot residues or mutate them with structurally similar groups.

Figure 2: Hotspot identification via alanine scanning of the mdm2-p53 interaction on bacteria cell surface.

(A): Select titrations of p53-like peptide alanine mutants on the bacterial surface. (B): Change in K_d from wild type p53-like peptide for each alanine mutant.

Steric hindrance governs p53-like-peptide staple location

Beyond avoiding the disruption of hotspot residues, the design of a stapled peptide requires identification of an optimal staple location. This is an important design criterion, because an inappropriate staple location can cause steric clashes with the target protein, resulting in decreased affinity, and can influence the physicochemical properties (e.g. amphiphilicity) which impact the membrane permeability.⁷ In contrast, forming new target interactions (e.g hydrogen bonds) and structurally stabilizing peptide residues are desired properties of a staple since they typically increase target affinity.^8,44–46 It is difficult to predict a priori whether a given staple location will improve or decrease binding strength beyond high-level observations with crystal structures, if available. We hypothesized that peptide display on the bacterial surface could both re-confirm the importance of stapling location from previous work and identify steric factors that might play a role in abrogating binding by comparing the unstapled and stapled p53-like peptides. We started by modifying the p53-like peptide (PLP) with alternative stapling locations: PLP(1–8) and PLP(6–13), compared to the previously published 4–11 location (Figure 3).¹⁷ Competitive inhibition curves can be found in Figure S12. We selected these positions based on the alanine scanning mutagenesis data, as neither of these staple locations replace residues that are responsible for the core interaction of p53 and mdm2. These locations are additionally of interest as staples near the protein interface that have greater potential to form novel contacts than those solely exposed to the solvent.⁴⁵ Because the mutated residues do not form key contacts with mdm2, we hypothesized that these minimally invasive substitutions in the unstapled form would have little effect on binding. In contrast, the stapled form could exhibit steric hindrance and/or new contacts that may cause changes in affinity. Indeed, affinity determination on the bacterial cell surface found that the 1–8 location was more staple-permissive than the 6–13 location, which weakened binding. Peptide alpha helicities calculated from circular dichroism spectrophotometry show that PLP(4–11) is considerably more alpha helical than the other PLP’s (60% versus 5%). However, CD measurements also showed that for all PLP’s, there were minimal changes in alpha helicity upon stapling. This suggests that the decrease in affinity for PLP(1–8) and PLP(6–13) as a result of stapling is not related to staple mediated secondary structure stabilization, further supporting the hypothesis that new steric hindrance effects drive weakened binding. Equilibrium constants from solution phase agreed qualitatively with equilibrium measurements on the bacterial cell surface, confirming that SPEED is well equipped to perform staple scanning.

Figure 3: Staple scanning p53-like peptides using the bacterial cell surface.

(A): Helix-wheel diagram of the p53-mdm2 interaction overlaid with the mdm2 crystal structure (PDB: 1YCR). K_d’s of p53-like peptide staple scan mutants on bacterial cell surface (B) or solution phase (C).

Engineering potent mdm2 binders with diverse bisalkyne linkers

We sought to explore the capabilities of SPEED to engineer structurally diverse peptides by varying the bisalkynes used in the stapling reaction. In our original work²¹, we used propargyl ether to sort a randomized library, where three critical MDM2-binding residues (F19, W23, L26) as well as the two sites for stabilization (X20, X27) were kept fixed (Figure 4). All other sites were randomized by an NNC codon scheme permitting 15 possible amino acids and no stop codons at each position, for a theoretical diversity of 3×10¹⁰. We hypothesized that by repeating this process with different bisalkynes, we would obtain potent stapled peptides with diverse sequences and linkers. We selected three new bisalkyne staples - heptadiyne, a purely aliphatic staple; a PEG2 linker with a primary amine for functionalization; and (1,3)-diethynlbenzene, a non-flexible aromatic linker; and an unreacted control. ^47,48 Randomized libraries were stabilized and then sorted by one round of magnetic sorting and four rounds of fluorescent sorting with increasing stringency for mdm2 binding. After sorting, we deep sequenced the peptides from each of the bisalkyne libraries and identified sequence patterns that emerged. Data from the (1,3)-diethynlbenzene library indicated that the sequences failed to yield major consensus groups, likely arising from an incomplete reaction due to linker inflexibility (Figure S13). In the other libraries, deep sequencing revealed a number of potential new dual cysteine motifs, potentially forming new topologies of i,i+1 and i,i+5 disulfide bonds expanding from the i,i+4 disulfide we confirmed in our previous lead molecule via nuclear magnetic resonance.²¹ Deep sequencing data additionally yields insights into the proportion and enrichment of potential disulfide motifs; if disulfides are forming, frequencies of potential disulfide motifs should increase compared to those with single cysteine residues. When we calculate the frequency of disulfide versus single cysteine residues, we see specific enrichment of potential disulfide motifs (Figure S14). This phenomenon was not observed as strongly in libraries that yielded worse enrichment ((1,3)-diethynlbenzene and the unreacted control). Next, we surveyed some of the most highly enriched sequences and performed low-throughput titrations of mdm2 to measure their binding affinity. Enrichment trajectories and all measured affinities are reported in Figures S15 and S16 respectively. We observed that in each of the bisalkyne libraries, there were multiple peptides that demonstrated improved affinity compared to the wild type sequence.

Figure 4: Engineering diverse bisalkyne stapled peptides.

(A) A randomized library of p53-like peptides was displayed on the surface of bacteria, reacted with one of three bisalkynes, and subjected to one round of magnetic and four rounds of fluorescent sorting. After sorting, cells from each library were analyzed via next generation sequencing (B). Select clones were picked based on frequency in the final library and their affinities were measured by titrating Mdm2 on the bacterial cell surface (C). Logoplots were made using Logomaker.⁴⁹

Identification of optimal staple location in BH3 domains

To utilize bacterial surface display for the selection of optimal linker location, we varied the staple location rather than the staple structure while targeting the Bcl-2 family of proteins. The B cell lymphoma 2 class of proteins was chosen for multiple reasons. First, the alpha helical domain where Bcl-2 antagonists and Bcl-2 interact is much larger than that of p53-mdm2 and is therefore more expensive to screen staple locations using solid-phase peptide synthesis.⁵⁰ Second, there are several hotspot residues, like Leu^3a and Asp^3f that would provide convenient controls for non-functional mutants.⁵¹ Finally, the Bcl-2 family is comprised of 5 highly homologous proteins that have varying levels of cellular expression and play different roles in apoptosis and resistance to chemotherapy. A standing goal of Bcl-2 inhibition is therefore to generate highly specific inhibitors.^10,52 While previous work has evaluated the impact of amino acid mutations on specificity, we sought to investigate if and how the staple location can be used to improve specificity between Bcl-2 proteins, which stabilized bacterial surface display is well suited to answer.

BIM, a naturally occurring BH3 domain with high promiscuity to all 5 Bcl-2 proteins, was selected as a scaffold for the staple scan to ensure the generalizability of staple location to different Bcl-2 proteins.³³ To investigate the effect of how staple location might impact binding affinity, every location in a 23-length BH3 domain was tested for its affinity to the Mcl-1 protein, the protein for which BIM has the highest affinity. (Figure 5) Select titrations can be found in Figure S17 The bacterial cell surface identified seven different staple locations that did not completely abrogate binding. These results were consistent with known hotspot residues: Leu^3a, Gly^3b, and Asp^3f. Mutation of these residues resulted in a complete loss of function (p8, p9, and p11).⁴⁴ We then measured the affinity of the BIM staple scan mutants to each of the other 4 Bcl-2 targets: Bfl-1, Bcl-x_L, Bcl-w, and Bcl-2. Interestingly, we found that the specificity trends of the wild type BIM peptide, which has a small degree of specificity for Mcl-1, were not the same as its stapled variants. This suggests that the staple location is playing a role in binding specificity and can serve as an additional handle for optimizing specificity.

Figure 5: Modulating affinity and specificity of B cell lymphoma 2 peptide antagonists by staple location.

(A) A helix-wheel diagram of the Bcl-2:BIM interaction overlaid with the Mcl-1 crystal structure (PDB: 2NL9) shows were the higher affinity variants were stapled. (B) The affinity and specificity of all BIM staple mutants was evaluated for all 5 Bcl-2 proteins.

Biolayer interferometry confirms bacterial surface display trends

We translated select p53-like peptides and stapled variants off the bacterial cell surface and measured their binding affinities to each of the Bcl-2 proteins to evaluate whether bacterial equilibrium measurements matched kinetic ones from biolayer interferometry. First, we observe that both PLP(1–8) and PLP(6–13) in their unstapled form bind MDM2 as strongly as PLP(4–11), confirming our hypothesis that substitution to azidohomoalanine doesn’t result in loss of function. We then measured the binding affinity of synthesized peptides in their unstapled and stapled forms and confirmed that stapling in either location weakened binding compared to their unstapled versions, to similar extents as with bacterial surface display. Overall, there was a strong correlation between BSD and BLI (pearson R² 0.82 and p < 0.0001) measurements. The slope from linear regression does not significantly differ from 1 (p=0.38) nor does the y-intercept significantly differ from 0 (p=0.13) although we observed that generally the bacterial cell surface overestimated binding affinities 3–10 fold (Figure 6).

Figure 6: Solution phase peptide affinity measurement correlates with bacterial surface.

BSD clones with measurable binding but not saturable binding are plotted as 600 nM.

Discussion

In this work, we used SPEED, which utilizes cell-surface stabilized peptides using non-natural amino acid incorporation and click chemistry (Figure 1), to measure the impact of staple chemistry and staple location on affinity peptides for MDM2 and the Bcl-2 family of proteins. We first confirmed that alanine scanning mutagenesis via bacterial surface display closely agrees with both experimental and computational approaches for hot spot identification.^42,43 The three major hotspot residues closely agreed with experimental and computational work (Figure 2). Next, we explored the landscape of stapled p53-like peptides (PLP’s) with modified linker structure and location. We hypothesized there may be potent PLP’s that have different staple locations and staple chemistries that maintain high binding affinity but allow different linker properties, such as greater charge/lipophilicity or a functional handle.^7,47 Previous work has demonstrated the importance of the linker properties and location.^5,7,8,10,44 Initial work in the optimization of stapled p53-like peptides (PLP) tested five locations of hydrocarbon staples that did not interfere with hot spot residues.¹⁷The binding affinity of these five variants spanned multiple orders of magnitude, and ultimately the authors focused on one staple location, PLP(4–11). This staple location improved affinity one hundred-fold over wild type, but it required additional mutations based on adding positive charge to have sufficient cytotoxicity. These mutations had the negative side effect of reducing the affinity fifty-fold. Chang et al. further rationally engineered ATSP-7041, a more potent version of PLP(4–11) with several mutations informed by linear phage display.⁴⁶ Aileron Therapeutics modified this molecule into ALRN-6924 which is currently in Phase 1b clinical trials for chemoprotection in breast cancer chemotherapy.⁵³ Using stabilized bacterial surface display, we tested PLP(1–8), PLP(4–11), and PLP(6–13) with triazole-based stapled peptides (Figure 3) and found that they had comparable affinities to molecules engineered with hydrocarbon staples. This approach can accelerate the measurements of mutations effect on affinity and can also serve as a tool to probe protease stability by treating the cells with a protease that simulates in vitro or in vivo conditions. In previous work, we showed that surface display experiments with protease treatment correlated with those from solution, reducing the burden of experimental measurement by synthesis and evaluation in vitro.²¹

Aside from linker location, we investigated how bacterial surface display could be used to probe the importance of the staple’s chemical properties. In many systems, it is difficult to simultaneously evaluate the contributions arising from multiple design criteria in high throughput, such as staple location, stapling chemistry, or amino acid mutations. Lau et al. investigated the effects of different triazole-based chemical linkers, and found PLP’s with an aromatic linker, 1,3-diethynlbenzene, only had K_d values of less than 1000nM with the 4–11 staple location, highlighting the complex trade-off between staple location and its chemical properties.¹² Because bacterial surface display has modularity for these components, we investigated its ability to design stapled peptides with variable sequence and staple chemistry. After generating a randomized library, reacting it with diverse bisalkyne linkers, and sorting for binding mdm2 with increasing stringency, we identified several new sequences that bind to mdm2 with high affinity (Figure 4). Importantly, these peptides each have unique sequences and staples with varying physicochemical properties such as pI, hydrophobicity, and potential intramolecular disulfide bond topology. Finally, the incorporation of the functionalizable linker (3) into high-affinity molecules could accelerate related tasks for peptides such as a fluorophore containing linker for imaging applications, a polyarginine motif for cellular penetration, or a ubiquitin ligase recruiting modality for formation of a protease targeting chimera (PROTAC).^47,48,54 These results highlight the important trade-offs between affinity, cytosolic access, and cytotoxicity as a function of sequence and staple location and the need for a method that can easily explore sequence and staple design space. SPEED is a method that can quickly assay staple location, evaluate amino acid mutations, and translate to solution phase measurements for greater coverage of design space for potent peptides such as p53-like binders.

To establish the generalizability of SPEED, we expanded the system to design stapled variants of BIM, a high affinity but non-specific inhibitor of B cell lymphoma 2 (Bcl-2) proteins that regulate apoptosis. Recent work shows that yeast surface display coupled with machine learning and sequence optimization are efficient at generating highly specific linear peptides, but rational design was necessary to generate highly specific stapled peptides.^48,52 We sought to address this challenge by measuring the affinity and specificity of stapled BIM variants. We identified several staple locations that dramatically change the specificity profile of BIM. These results recapitulate many factors that have previously been identified about BIM-based peptides as well as identify new impacts of linker location. For example, staple locations that disrupt key hotspots abrogate binding as expected. We find that Bfl-1 has the lowest affinity across all variants evaluated, which agrees with Jenson et al. where the authors had to use a PUMA-based library rather than BIM to find potent inhibitors of Bfl-1 since BIM-based peptides had low affinities.⁵⁵ Similarly, we find that these variants have very high affinities towards Mcl-1 which is consistent with the sub-nanomolar affinity of BIM and the lack of interference between the staple and those high affinity interactions.³³ Finally, we see a high degree of correlation of affinity between Bcl-xL, Bcl-w, and Bcl-2, likely resulting from high structural homology between these three proteins.³⁵ We also discovered variable specificity with different double-click linker locations. These BIM staple variants displaying altered specificity could serve as starting points for applications that rely on specific members within Bcl-2. Future work includes screening randomized libraries of Bcl-2 antagonists via SPEED using these specificity-driving staple location towards the discovery of high affinity, specificity, and efficacious Bcl-2 stapled peptide inhibitors.

Finally, we translated PLP and BIM variants off the bacterial cell surface and evaluated their binding affinities in solution using biolayer interferometry and competitive inhibition experiments. While equilibrium affinity measurements from bacterial surface display highly correlate with those from solution phase, the bacterial surface tends to overestimate solution-phase binding affinities by 3 to 10-fold in our system. The exact cause of this discrepancy is unclear. We hypothesize that the molecular crowding on the bacterial cell surface improves the conformational stability of displayed peptides and results in lower measured K_d values relative to the same sequence in solution regardless of peptide stapling. This feature is more pronounced in PLP’s than BIM variants as PLP’s tend to be less structured in phosphate buffers or trifluoroethanol (a helix inducing solvent), likely due to their shorter length (13 AA vs 23).^56,57 Therefore, synthesis and evaluation of peptides in their soluble form remains an important step in the design of new stapled peptide inhibitors. Likewise, bacterial surface display may not be able to resolve small differences in affinity, which could be a limitation if molecules need to be tuned with high precision. However, in the design of Bcl-2 inhibitors, specificities on the order of 100–1000x are needed, which is well within the abilities of surface display to measure.⁵² In conclusion, we have established the discovery of stapled peptides via bacterial surface display is a powerful method that can optimize sequence, staple location, and staple chemistry with respect to binding affinity and specificity.