Design of a protease‐activated PD‐L1 inhibitor

Abstract

Immune checkpoint inhibitors that bind to the cell surface receptor PD‐L1 are effective anti‐cancer agents but suffer from immune‐related adverse events as PD‐L1 is expressed on both healthy and cancer cells. To mitigate toxicity, researchers are testing prodrugs that have low affinity for checkpoint targets until activated with proteases enriched in the tumor microenvironment. Here, we engineer a prodrug form of a PD‐L1 inhibitor. The inhibitor is a soluble PD‐1 mimetic that was previously engineered to have high affinity for PD‐L1. In the basal state, the binding surface of the PD‐1 mimetic is masked by fusing it to a soluble variant of its natural ligand, PD‐L1. Proteolytic cleavage of the linker that connects the mask to the inhibitor activates the molecule. To optimize the mask so that it effectively blocks binding to PD‐L1 but releases upon cleavage, we tested a set of mutants with varied affinity for the inhibitor. The top‐performing mask reduces the affinity of the prodrug for PD‐L1 120‐fold, and binding is nearly fully recovered upon cleavage. In a cell‐based assay measuring inhibition of the PD‐1:PD‐L1 interaction on the surface of cells, the IC50s of the masked inhibitors were up to 40‐fold higher than their protease‐treated counterparts. The changes in activity we observe upon protease treatment are comparable to systems currently tested in the clinic and provide evidence that natural binding partners are an excellent starting point for creating a prodrug.

1. INTRODUCTION

The programmed death pathway has dual importance in regulating healthy immune function and in cancer cells' exploitation of the pathway to evade immune detection. The programmed cell death‐1 immunoreceptor (PD‐1) is expressed on T cells. Its ligand, programmed death‐ligand 1 (PD‐L1), is found on hematopoietic cells such as dendritic cells and macrophages, and non‐hematopoietic tissues such as heart, pancreas, and skin tissue (Qin et al., 2019). Under normal conditions, PD‐1:PD‐L1 interactions protect the body from autoimmune diseases by suppressing self‐reactive T cell‐mediated immune responses (Freeman et al., 2000). Thus, as PD‐L1 is an important immune inhibitor, tumor cells overexpress PD‐L1 to escape immune detection. Additionally, overexpression of PD‐L1 on tumor cells correlates with poor disease outcomes in the clinic (Ohaegbulam et al., 2015).

Immune checkpoint inhibitors, which promote tumor killing by manipulating the normal immune response (Sharma & Allison, 2015), have expanded and revolutionized the cancer therapy landscape (Bagchi et al., 2021; Ohaegbulam et al., ²⁰¹⁵; Pardoll, ²⁰¹²; Robert, ²⁰²⁰). Therapeutics that block binding between PD‐L1 and its receptor can help restore antitumor immunity. Clinical anti‐PD‐L1 antibodies such as Atezolizumab, Avelumab, and Durvalumab have been successful across various cancers including PD‐L1‐positive triple‐negative breast cancer (Iwata et al., 2019), metastatic urothelial carcinoma (Powles et al., 2018), and non‐small‐cell lung cancer (de Sousa et al., ²⁰¹⁹; Hargadon et al., ²⁰¹⁸; Rittmeyer et al., ²⁰¹⁷). However, these PD‐L1 inhibitors and other immunotherapies are limited by immune‐related adverse effects due to cross‐reactivity to healthy tissue that releases overactive immunity at both cancerous and healthy cells (Blidner et al., 2020; Morad et al., ²⁰²¹).

One effective strategy to direct immunotherapies to the tumor site and minimize unwanted off‐target toxicity is to engineer a prodrug therapeutic. Such therapies can inactivate the drug via a tethered masking domain. Cleavage of a protease‐cleavable linker is one method to release the masking domain for tumor‐targeted drug delivery (Bleuez et al., 2022; Poreba, ²⁰²⁰). Within the scope of cancer therapy, prodrugs using protease‐cleavable linkers have been successful in reversibly occluding the binding sites of various therapeutic antibodies via a heterodimeric coiled‐coil domain (Trang et al., 2019), localizing an anti‐PD‐L1 antibody to the tumor microenvironment to reduce systemic toxicity (Assi et al., 2021; Giesen et al., ²⁰²⁰; Naing et al., ²⁰²¹), and directing the toxic interleukin‐12 (IL‐12) cytokine to immunologically “cold” tumors (Mansurov et al., 2022; Xue et al., ²⁰²²). Notably, two independent groups designed distinct IL‐12 prodrugs using portions of the endogenous receptor, IL‐12Rβ1, to control IFN‐γ release within the tumor microenvironment (Mansurov et al., 2022; Xue et al., ²⁰²²).

In a similar approach, we describe a prodrug for PD‐L1 that is masked by variants of the endogenous ligand. In the basal state, the prodrug has low affinity for PD‐L1, but when activated by cleavage, it potently blocks PD‐L1 from binding to the PD‐1 receptor. The prodrug consists of three components: an engineered soluble PD‐1 mimetic (Maute et al., 2015; Pascolutti et al., ²⁰¹⁶) (referred to as HA‐PD1 henceforth), a masking domain genetically fused to HA‐PD1 that sterically occludes the PD‐L1 binding surface on HA‐PD1, and a protease‐sensitive linker connecting the masking domain to HA‐PD1.

We chose HA‐PD1 as the basis for our prodrug because it has an exceptionally high affinity for PD‐L1, like clinical antibody‐based therapeutics, and its small size (~100 residues) could enable better tumor penetration than antibodies (Maute et al., 2015). We employed variants of the endogenous ligand, PD‐L1, for the masking domain. This domain is attractive because it naturally has affinity for PD‐1, should directly block binding of full‐length PD‐L1 to HA‐PD1, and is likely to have low immunogenicity as it is derived from a human protein. This modular three‐component system may benefit from future linker optimization to further promote tumor specificity. Furthermore, we aim to extend the approach of using an endogenous ligand as a masking domain as a general prodrug design strategy. Herein, we outline the design of various masked complexes containing mutant PD‐L1 domains to show their conditional activation via proteolytic cleavage.

2. RESULTS

2.1. Engineering a mask for a high‐affinity PD‐1 mimetic

The high‐affinity PD‐1 mimetic, HA‐PD1, is an exceptionally potent inhibitor of PD‐L1 (Maute et al., 2015; Pascolutti et al., ²⁰¹⁶). The equilibrium dissociation constant, K_D, between HA‐PD1 and PD‐L1 is 110 pM with a dissociation half‐life greater than 10 min. This high affinity presents a compelling challenge when considering how to effectively mask HA‐PD1. The masking domain's affinity to HA‐PD1 must compete with natural ligand interactions while remaining weak enough to dissociate after proteolysis. For this reason, we chose the endogenous ligand of PD‐1, PD‐L1, as our initial masking domain (Figure 1a). We hypothesized that covalently linking the mask to HA‐PD1 would effectively increase its local concentration allowing the mask to outcompete binding to naturally expressed ligand.

FIGURE 1.

Structure and mutations of masked complexes. (a) AlphaFold prediction of the wild‐type masked complex illustrating HA‐PD1 (gray) followed by a 28‐linker (yellow) containing a protease recognition site (orange) and a PD‐L1 masking domain (pink). (b) Zoomed in view of interface interactions highlighting the Y56‐mediated intramolecular bond within PD‐L1 (pink) and an intermolecular bond to HA‐PD1 (gray) via Y123. Some residues were hidden for clarity. (c) Cartoons illustrating the alanine mutations (white circles) in the masking domain (pink) of each masked complex. Note, although NoMC lacks a masking domain, it retains a cleavable linker. NoMC, unmasked complex, or no masked complex.

To screen for a functional balance between masking and release, we initially constructed three masked inhibitors. Each mask was linked to the C‐terminus of HA‐PD1 via a 28‐residue flexible linker constructed from glycine and serine residues and embedded with a 7‐residue TEV protease motif (ENLYFQ~G) in the middle (Figure 1a). The wild‐type masked complex (WTMC) is a fusion between the wild‐type N‐terminal extracellular domain of PD‐L1 and HA‐PD1. Protein energy calculations from the Robetta Alanine Scanning Server (Kortemme et al., 2004; Kortemme & Baker, ²⁰⁰²) predict that Y56A and Y123A on PD‐L1 would lower the free energy of binding between the mask and HAPD‐1 by 1.5 and 3.0 kcal/mole, respectively (Table S1). In the crystal structure of PD‐L1 complexed with HA‐PD1, Y56 forms an intramolecular hydrogen bond to E58, whereas Y123 forms a hydrogen bond across the interface to E136 on PD‐1 (Figure 1b). Both alanine mutations were combined into the double mutant masked complex, DMMC. A final construct without a masking domain (NoMC) was also designed to serve as a control (Figure 1c).

We first tested how well the masks within WTMC and DMMC prevented binding to soluble PD‐L1 using yeast surface display and flow cytometry. NoMC was also characterized as a positive control for PD‐L1 binding. The masked inhibitors were each presented on the surface of yeast in the presence and absence of TEV protease (Figure 2a). In addition to the three components of the prodrug (inhibitor, protease‐sensitive linker, and masking domain, listed from N‐ to C‐terminus), the yeast display constructs also contained a C‐terminal cMyc tag that was used with an anti‐cMyc‐FITC antibody to detect expression on the cell surface and monitor release of the masking domain following proteolysis (Figure 2a). Despite lacking a masking domain, NoMC was engineered with a protease‐sensitive linker and a C‐terminal cMyc tag to monitor cleavage efficiency in the absence of a masking domain. Binding of the prodrug to PD‐L1 was detected by mixing the cells with biotinylated PD‐L1 (bPD‐L1), followed by incubation with fluorescently labeled streptavidin (Figure 2a).

FIGURE 2.

Masked complex yeast surface expression and binding to PD‐L1. (a) Masked complexes were displayed on the surface of yeast (blue cell) through the Aga2p system. Using flow cytometry, the overall presence of the mask was monitored with an anti‐cMyc‐FITC antibody (green star). Binding of biotinylated PD‐L1 (light pink) to HA‐PD1 of the complex was measured through streptavidin‐633 (light pink star). (b) Bivariate expression and binding profiles of NoMC, DMMC, and WTMC without (top left) and with (top right) TEV incubation. To directly compare a single complex with (teal) and without (pink) protease treatment, the expression and binding profiles for WTMC and DMMC are also shown (bottom left and right, respectively). The default number (8000) of dots is shown for each condition. (c) The change in the gMFI of FITC and 633 upon protease treatment are shown for NoMC, DMMC, and WTMC, calculated using the following equation: (gMFI_+TEV − gMFI_−TEV)/(gMFI_−TEV). Additional gating details found in the Supporting Information. DMMC, double mutant masked complex, or Y56A and 123A PD‐L1 masked complex; NoMC, unmasked complex, or no masked complex; TEV, tobacco etch virus; WTMC, wild‐type PD‐L1 masked complex.

As anticipated, NoMC robustly bound bPD‐L1 with or without protease treatment and the cMyc signal was reduced to background levels with the addition of protease (Figure 2b,c). In contrast, WTMC showed a dramatic reduction in binding to PD‐L1 (Figure 2b). Following incubation with protease, there was an increase in PD‐L1 binding but binding levels did not return to those of NoMC (Figures 2b,c and S1). Interestingly, protease treatment also did not return the cMyc signal to background levels (Figures 2b,c and S1). Given the high efficiency of TEV cleavage seen with NoMC, this result suggests that the wild‐type mask remains bound to a large fraction of HA‐PD1 despite cleavage of the linker. This is reasonable, given the slow dissociation between HA‐PD1 and PD‐L1. Unlike WTMC, the DMMC masking domain was efficiently released with protease treatment; but even with the mask present, DMMC still allowed binding to bPD‐L1 (Figures 2b,c and S1). Taken together, these results showed the need to further optimize the affinity of the masking domain to both functionally mask and release.

2.2. Masked inhibitors have reduced binding affinity for PD‐L1 that can be restored with protease treatment

To further assess the affinity of the masked inhibitors for PD‐L1 and screen additional masked constructs, we performed binding measurements with soluble proteins. Five variants were expressed and purified from mammalian cells, NoMC, WTMC, DMMC, 56MC, and 123MC. 56MC is identical to WTMC but contains the Y56A mutation within the masking domain (Figure 1c). Similarly, 123MC is identical to WTMC but contains the Y123A mutation within the mask. All variants expressed well (45–75 μg protein per mL of media) and were monomeric as determined by size exclusion chromatography (Figure S2).

For the surface plasmon resonance (SPR) experiments, bPD‐L1 was immobilized on the sensor chip and single‐cycle kinetics (Karlsson et al., 2006) runs were used to simultaneously measure binding affinities and binding kinetics (Figure 3). All SPR data were collected with at least three replicates across at least two sensor chips for each condition. As expected, the unmasked inhibitor, NoMC, bound tightly to WT PD‐L1 with a K_D (400 pM; Figure 3, Table S2) similar to the reported values (Maute et al., 2015; Pascolutti et al., ²⁰¹⁶). The masked inhibitors, WTMC, 56MC, and 123MC, all demonstrated reduced affinity for PD‐L1 with K_D values of 57, 44, and 20 nM, respectively. Consistent with the yeast display experiments, DMMC masked to a lesser degree with a K_D of 3 nM for PD‐L1. Notably, the decreases in binding affinities of the masked complexes were predominantly driven by changes in the on‐rate (Figure 3a). For the unmasked inhibitor (NoMC), the association rate constant was ~2.0 × 10⁶ M⁻¹ s⁻¹, while for the three tightest masks (WTMC, 56MC, and 123MC), the rate constants varied between 1.2–4.0 × 10⁴ M⁻¹ s⁻¹ (Figure 3a, Table S2). In contrast, the dissociation rate constants were all within a factor of 2 of that measured for NoMC. Across the masked inhibitors, most of the changes in kinetics were observed during association supporting that each masking domain blocked binding to varying degrees—but, once binding had occurred, the mechanism for HA‐PD1 dissociation was likely the same across all inhibitors.

FIGURE 3.

Masks reduce affinity to PD‐L1. Biotinylated PD‐L1 was immobilized on a NeutrAvidin chip. (a) On–off rate map indicating the mean binding kinetics parameters of the various masked complexes without protease treatment. Reported error represents SD. The map was constructed using analyte concentrations from 12 nM to 1 μM for WTMC, 56MC, and 123MC; 1.2 to 100 nM for DMMC; and 0.12 to 10 nM for NoMC. (b) Representative surface plasmon resonance (SPR) sensorgrams of the masked complexes in the absence (pink) and presence (teal) of overnight TEV proteolysis. Corresponding responses were simultaneously collected with analyte concentrations from 1.2 to 100 nM. For direct comparison, across the panels are the maximum response of WTMC in the absence and presence of TEV (dotted lines). (c) Table of mean kinetic parameters for the various masked complexes before and after TEV treatment, including the fold change in KD over that of the NoMC control, and the change before and after TEV proteolysis for each complex. In the absence of TEV, experimental conditions as mentioned in (a); in the presence of TEV, conditions as mentioned in (b). Error represents SD. “n” represents the number of runs, which includes experiments collected from at least two different sensor chips for each condition; individual data included in the Supporting Information. DMMC, double mutant masked complex, or Y56A and 123A PD‐L1 masked complex; NoMC, unmasked complex, or no masked complex; SD, standard deviation; TEV, tobacco etch virus; WTMC, wild‐type PD‐L1 masked complex.

We next examined if protease treatment could restore the affinity of the masked inhibitors for PD‐L1. Analysis by SDS‐PAGE confirmed that overnight treatment with TEV protease at 37°C sufficiently cleaved the linkers in all the masked inhibitors (Figure S3). The cleaved mixture was then used as the analyte in SPR experiments. We did not purify the masks from HA‐PD1 as we expected the mixture more accurately reflected what would occur in vivo. Cleavage of the linker restored binding activity across all masked complexes to a similar degree (Figure 3b). The 30‐fold decrease in K_D observed for 56MC upon protease treatment was the largest change and a 1.4‐fold decrease for DMMC was the smallest (Figure 3c). Overall, SPR showed the dual functionality of masking and release was best observed in 56MC.

2.3. Conditional disruption of the PD‐1:PD‐L1 interaction on the cell surface

To test if the masked inhibitors conditionally blocked the binding of endogenous PD‐L1 to PD‐1 on the cell surface, we used a reporter assay that detects interactions between PD‐L1 displayed on engineered CHO cells and PD‐1 displayed on Jurkat T cells. In the assay, T‐cell receptor (TCR) signaling induces luminescence only when the interaction between PD‐1 and PD‐L1 is disrupted. We first demonstrated that the unmasked HA‐PD1 (NoMC) blocks the PD‐1:PD‐L1 interaction as effectively (IC50 = 8.7 nM) as the provided anti‐PD‐1 antibody assay control (Figure S4). We then tested the masked complexes in duplicate across two experiments. All the masked inhibitors displayed higher IC50s than NoMC (Figure 4a), with the most potent masking evident for 56MC and WTMC at ~1.8 μM (Figures 4b and S5a). Notably, the 56MC IC50 is 150‐fold greater than that of NoMC. In contrast, DMMC showed minimal masking activity, with an IC50 of 52 nM.

FIGURE 4.

Masked complexes conditionally block the PD‐1:PD‐L1 endogenous interaction on the surface of cells. In the assay, when the endogenous PD‐1:PD‐L1 interaction is disrupted, TCR activation induces luminescence. (a) The luminescence signal at various concentrations of the masked complexes with (teal) and without (pink) TEV protease treatment. Two technical replicates from a single experiment, presented as mean ± SEM are shown. (b) To directly compare the activity of the various complexes, the luminescence responses without (left) and with (right) TEV incubation are shown. (c) Data from two separate experiments were analyzed in Prism and the mean LogIC50 windows with and without TEV incubation for each masked complex are shown. The fold change of the IC50 upon protease treatment is labeled for each complex. Individual data from the two separate experiments are included in the Supporting Information. SEM, standard error of the mean; TCR, T‐cell receptor; TEV, tobacco etch virus.

As with the SPR experiments, the masked inhibitors were treated with protease overnight and the resulting mixtures were used in the blockade assay. DMMC uniquely regained full activity after protease treatment (IC50 = 10 nM). The ~20‐fold change in IC50 after protease treatment was similar for 123MC and WTMC, with WTMC showing greater autoinhibition before cleavage (Figures 4c and S5b). Consistent with the SPR results, 56MC showed the largest change in IC50 following protease treatment, decreasing 40‐fold from 1.86 μM to 48 nM.

3. DISCUSSION

To enable tumor‐targeted delivery of a PD‐L1 inhibitor, we designed a panel of conditionally masked PD‐L1 prodrugs that use variants of PD‐L1 as the masking domain. As first observed in yeast surface display experiments, wild‐type PD‐L1 served as a powerful masking domain but largely remained bound even after protease treatment. Incorporating single tyrosine‐to‐alanine mutations at the masking interface improved the dual functionality of masking and release. As observed in both SPR experiments and a cell signaling assay, the prodrug form of 56MC is autoinhibited to a similar degree as WTMC but after protease treatment more effectively binds PD‐L1 to block the interaction between PD‐1 and PD‐L1 on the cell surface.

The relative activities of the four masked inhibitors, WTMC, 56MC, 123MC, and DMMC, are largely consistent with the Rosetta energy calculations (Table S1). The calculations predicted Y123A would weaken the interaction between the mask and HA‐PD1 more than Y56A. Consistent with this prediction, stronger masking is observed with 56MC before protease treatment, but following protease treatment 123MC more potently inhibits the PD‐1/PD‐L1 interaction in the checkpoint blockade assay.

Like our masked PD‐L1 inhibitors, CytomX Therapeutics, which specializes in the design of protease‐activated masked antibodies known as Probody therapeutics, has developed a masked anti‐PD‐L1 antibody known as pacmilimab. Pacmilimab has shown promising antitumor activity with reduced toxicity in animal and human trials (Giesen et al., 2020; Naing et al., ²⁰²¹). The unmasked parent (CX‐075) and masked variant of pacmilimab exhibited a 40‐fold difference in affinity for PD‐L1 with K_D values equal to 0.25 and 10 nM, respectively (Giesen et al., 2020). In comparison, there is a 120‐fold difference in affinity for PD‐L1 between NoMC and 56MC and the dissociation constants vary over a similar range of affinities (picomolar to nanomolar) as to unmasked and masked pacmilimab. This similarity highlights the potential utility of 56MC and emphasizes the general effectiveness of using natural binding partners as masking domains. Additionally, another benefit of using native protein domains as therapeutic masks is their expected low immunogenicity.

In future studies, we aim to optimize the linker region to confer activation through tumor‐enriched proteases, such as matrix metalloproteinase‐2 (MMP‐2), and to study the toxicity and anti‐cancer activity of the molecule in animal model systems. Conveniently, both the current TEV motif (ENLYFQ~G) and the MMP‐2 motif (PLG~LAG) (Jiang et al., 2004; Trang et al., ²⁰¹⁹) are similar in length, so we anticipate the sequence can be replaced with little optimization. Additionally, we speculate the modular linker could be optimized to include other protease motifs used in successful prodrug designs (Assi et al., 2021; Mansurov et al., ²⁰²²; Xue et al., ²⁰²²).

When designing the masked PD‐L1 inhibitors, it was helpful to have a crystal structure of HA‐PD1 bound to PD‐L1. The structure made it straightforward to select domain boundaries, design the linkers, and identify mutations for tuning masking strength. Recent advances in deep‐learning‐based protein structure prediction, such as AlphaFold (Jumper et al., 2021) and RoseTTAFold (Baek et al., 2021), should further enable the exploration of native protein domains to modulate protein–protein interactions.

4. MATERIALS AND METHODS

4.1. Alanine scan

Binding free energy changes upon alanine mutation were calculated using the Computational Interface Alanine Scanning Server (Kortemme et al., 2004; Kortemme & Baker, ²⁰⁰²) at https://robetta.bakerlab.org/alascansubmit.jsp with a cleaned file of the PDB 5IUS.

4.2. Yeast display plasmid construction

To generate single clones of our masked complexes on yeast, we first constructed NoMC and then inserted different masking domains to achieve constructs for WTMC and DMMC. To do this, we first followed the Wittrup protocols (Chao et al., 2006; Gai & Wittrup, ²⁰⁰⁷) and previously described methods from the Kuhlman Lab (Hussain et al., 2018) with some modifications to create a custom pCTCON vector containing HAPD‐1. About 6 μg vector was linearized with FastDigest NheI (Thermo Fisher Scientific, FD0974), FastDigest SalI (Thermo Fisher Scientific, FD0644), and FastDigest XhoI (Thermo Fisher Scientific, FD0694) with 10X FastDigest Buffer (Thermo Fisher Scientific, B64) in a 50 μL reaction for 1 h at 37°C and then overnight at room temperature. A HA‐PD1 insert (Twist Bioscience, gene fragment) was PCR amplified with an annealing temperature of 64°C and 33 cycles using Q5 High‐Fidelity DNA Polymerase (NEB, M0491L) and the following primers (Table S3). Then, both the linearized vector and insert were concentrated to 1 μg/ μL using Pellet Paint Co‐Precipitant (Novagen through Millipore Sigma, 69049‐3, lot 2639673) following the manufacturer's protocol using 1 volume of isopropanol at step 3, and washing with 200 μL 70% ethanol a total of 3 times at step 6. The insert was co‐electroporated with linearized vector into 50 μL Saccharomyces cerevisiae strain EBY100, previously made electrocompetent as described by the Hsieh group (Benatuil et al., 2010), with a 5:1 insert:vector (~2.5 μg insert and 0.5 μg vector) using a 0.2 cm gap electroporation cuvette (USA Scientific, 9104‐5050) and the Gene Pulser II (Bio‐Rad Laboratories, 340BR) and Pulse Controller Plus (Bio‐Rad Laboratories, 339BR) set at 0.54 kV, 25 μF, and 1000 Ω. After outgrowth, plating, and colony picking, the plasmid DNA was extracted from 1 mL of cells treated with 200 μL Solution 1 and 5 μL Zymolase (modified step 1 of Zymo Research's Zymoprep™ Yeast Plasmid Miniprep II, D2004), incubated at 37°C for 1 h, then following the lysis step onward of the GeneJET Plasmid Miniprep Kit (Thermo Fisher Scientific, K0503) protocol, eluting in only 10 μL ddH2O. The plasmid was then transformed using 30 μL DH5a cells heat shocked at 42°C for 42 s, plated on LB/amp agar plates, and purified using the GeneJET Plasmid Miniprep Kit.

Subsequently, various PD‐L1 inserts (Twist Bioscience, gene fragment) were PCR amplified (Table S3) at 60.6°C using Q5 High‐Fidelity DNA Polymerase. About 6 μg NoMC vector was linearized with 1 μL each of KasI (NEB, R0544S) BamHI‐HF (NEB, R3136S) with 10X NEB Cutsmart Buffer (NEB, B7204, since discontinued) in a 50 μL reaction at 37°C overnight. Linearized vector and insert were concentrated using Pellet Paint and co‐electroporated (as described above). Transformed yeast was recovered in yeast extract peptone dextrose (YPD) media and expanded in liquid synthetic dextrose medium with casamino acids (SDCAA) at 30°C for 3 days.

4.3. Yeast display screening by flow cytometry

The expression of the masked complexes on the yeast surface (Table S4) was induced by dilution into liquid synthetic galactose medium with casamino acids (SGCAA) and cultured at 20°C for 24 h as outlined by Wittrup (Chao et al., 2006; Gai & Wittrup, ²⁰⁰⁷). 4.5 × 10⁶ yeast cells (induced and, as a control, uninduced) per well were used during the experiment. Using the Eppendorf ThermoMixer C (Thermo Fisher Scientific, 05‐412‐503) and the “DWP 1000” SmartBlock adapter (05‐414‐304), yeast cells followed three stages of incubation before flow cytometry analysis. During the first stage, cells were incubated in a deep 96‐well plate (Eppendorf through Thermo Fisher Scientific, 951033006) with 100 μL either 0.1% BSA (Thermo Fisher Scientific, BP1600‐100) in PBS (PBSF) or TEV (gift of the David Williams Lab, diluted to 1 mg/mL in PBSF) at 20°C at 1500 rpm for 1 h. Immediately after incubation, 500 μL PBSF was added to spin down using a 600 μL balance plate at 4000 rpm for 1 min at 4°C. Cells were then washed with 600 μL PBSF and incubated at 4°C at 1500 rpm for 30 min, then spun down. The long wash and pellet cycle were repeated for a total of six 30 min wash steps (or, 3 h of washing total). During the second stage, cells were incubated with either 100 μL PBSF or 20 nM biotinylated PD‐L1 (bPD‐L1, Sino Biological, 10,084‐H08H‐B, reconstituted according to the manufacturer's protocol, further diluted into PBSF) at 4°C at 1500 rpm for 1 h. Immediately after incubation, 500 μL PBSF was added to pellet cells. Cells were washed with 600 μL PBSF and incubated at 4°C at 1500 rpm for 1 min, then spun down. The quick wash and pellet cycle were repeated for a total of two 1 min wash steps. Finally, cells were incubated with either 100 μL PBSF or a 100 μL cocktail of 4 μg/mL cMyc‐FITC conjugated antibody (abcam, ab1394) and 4 μg/mL Streptavidin‐Alexa Fluor 633 conjugate (Invitrogen through Thermo Fisher Scientific, S21375) at 4°C at 1500 rpm for 30 min. Immediately after incubation, cells were pelleted. The quick wash and pellet cycle were repeated for a total of two 1 min washes. Cells were pelleted and kept on ice until they were analyzed. Flow cytometry analysis was performed on a Thermo Fisher Scientific Attune NxT flow cytometer located at the UNC Flow Cytometry core facility. Data were further analyzed using FlowJo 10.8.1 to gate on single cells. The expression and binding profiles were shown as dot plots.

To determine the relative change in cMyc‐FITC and SA‐633 upon TEV incubation, an autofluorescence gate (Autoflu) was manually created using the auto gate tool on uninduced yeast cells (Figure S1). The inverse Autoflu gate was then applied to the NoMC, WTMC, and DMMC binding profiles in the presence and absence of TEV. Each masked complex had a custom rectangle gate drawn on the population of interest (To Quantify). The change in the gMFI of FITC and 633 of the To Quantify gate were calculated using the following equation: (gMFI_+TEV − gMFI_−TEV)/(gMFI_−TEV) (Figure 2c).

4.4. Recombinant protein expression and purification

Masked complexes containing a C‐terminal 8xHis tag (Table S4) were codon optimized for mammalian systems and cloned into the expression vector PaH using the KpnI and XhoI restriction enzyme sites (Twist Bioscience, Table S5). Plasmids were transformed in 100 mL cultures of DH5α in LB media supplemented with ampicillin and purified using the NucleoBond Xtra Midi EF Midi kit for endotoxin‐free plasmid DNA (Macherey‐Nagel, 740420.50). Fusion complexes were then expressed in Expi293F mammalian cells (Thermo Fisher Scientific, A14527) by transient transfection using the ExpiFectamine293 system (Thermo Fisher Scientific, A14525) according to a modified manufacturer's protocol for a 125 mL flask; transfection and enhancement using Hyclone Cell Boost 1 (GE Healthcare/Cytiva, SH30584.02) were described previously by the Kuhlman Lab (Kudlacek et al., 2021). Cultures were harvested 5–6 days post‐enhancement, and the media were clarified by centrifugation and filtration through a 0.22 μm membrane. Clarified supernatants were concentrated with 30 kDa MWCO Amicon Ultra Centrifugal Filters (Millipore, UFC9030) and allowed to bind to Ni‐Penta Agarose 6 Fast Flow, Chemical Stable (PROTEINDEX, 11‐0228‐010) for 1 h at 4°C. After the supernatants gravity flowed through a column, the resins were washed with PBS (137 mM NaCl, 2.7 mM KCl, 8 mM Na₂HPO₄, and 2 mM KH₂PO₄) containing 25 mM imidazole, pH 7.4. Proteins were eluted with PBS, 500 mM imidazole, pH 7.4. Following elution, samples were purified by size using the HiLoad 16/600 Superdex 75 pg column (Cytiva, 28989333, lot 10085247) and the ÄKTAprime plus (Cytiva, 11001313, discontinued) running 144 mL PBS, pH 7.4 at 1 mL/m and collecting 3 mL fractions. Samples were concentrated and stored in PBS, pH 7.4 at 4°C.

4.5. Surface plasmon resonance

Experiments were conducted using a Biacore 8K instrument at 25°C in PBS containing 0.05% Tween 20, pH 7.4 (PBST). Biotinylated PD‐L1 (Sino Biological, 10084‐H08H‐B, reconstituted according to the manufacturer's protocol) was diluted to 500 ng/mL in PBST and immobilized onto a Series S NeutrAvidin (NA) Sensor Chip (Cytiva, 29407997) to yield a Rmax of ~100–200 RU. The following immobilization parameters were used: contact time 60 s; flow rate 10 μL/m. Following immobilization, affinity measurements were made with five 1:3 serial dilutions of each recombinant masked complex (starting at varying initial concentrations) in PBST using a single‐cycle kinetics method: analyte concentrations from 12 nM to 1 μM for WTMC, 56MC, 123MC; 1.2 to 100 nM for DMMC; and 0.12 to 10 nM for NoMC (Figure S6). The following parameters were used: contact time 120 s; dissociation time 1200 s; flow rate 30 μL/m; both flow cells; five concentrations per cycle. Data were collected over the course of multiple experiments using multiple sensor chips (Table S2). All data were analyzed with the Biacore Insight Evaluation software version 3.0.12 with the general single‐cycle kinetics method and a 1:1 binding kinetics fit model. In addition, data were imported to Prism version 9.4.1 to analyze data (determine the mean ± standard deviation [SD] across replicates) and create figures.

4.6. TEV proteolysis of masked complexes for SPR

Briefly, TEV was expressed from a plasmid (gift from Sharon Campbell Lab) using BL21 cells, and purified using Nickle affinity, eluted in 25 mM NaH2PO4, 200 mM NaCl, 500 mM Imidazole, and 10% Glycerol, then buffer exchanged into 25 mM NaH2PO4, 200 mM NaCl, 5 mM EDTA, 1 mM DTT, 10% glycerol, and then stored in 50% glycerol at −20°C. For all cleaved masked complexes used in SPR, 1 μM masked complex was incubated with 0.7 μM TEV diluted into PBST at 37°C for 16 h. The non‐cleaved masked complexes were diluted to 1 μM and incubated at 4°C for 16 h. Immediately before SPR, all samples were diluted to 100 nM masked complex. Furthermore, five 1:3 serial dilutions were made for the single‐cycle experiment, from 1.2 to 100 nM. All data were analyzed with the Biacore Insight Evaluation software as described above.

4.7. PD‐1:PD‐L1 neutralizing activity by PD‐1:PD‐L1 Blockade Bioassay

The activity of the masked complexes was analyzed using a PD‐1/PD‐L1 Blockade Bioassay (Promega, J1250) according to a modified manufacturer's protocol. Briefly, PD‐L1 aAPC/CHO‐K1 cells were seeded into a 96‐well assay plate (Corning through Falcon, 353296) and incubated for 16 h at 37°C with 5% CO₂. For all cleaved masked complexes, 312.5 μg/mL masked complex was incubated with 100 μg/mL TEV in 150 μL assay buffer at 37°C for 16 h. The non‐cleaved masked complexes were diluted to 312.5 μg/mL and incubated at 4°C for 16 h. Immediately prior to analysis, all samples were diluted to 156.25 μg/mL masked complex. Serial dilutions of masked complexes (0.016–156.25 μg/mL) or commercial Control Ab (Promega, J1201, Anti‐PD‐1) (0.016–25 μg/mL) were added to the plates followed by seeding PD‐1 effector cells. Note, for the masked complexes, a larger concentration range was used which required using five entire rows of the 96‐well plate rather than the recommended inner 60 wells. After co‐culture for 6 h, the Bio‐Glo™ reagent was added to each plate such that it was incubated at ambient temperature for 5 min immediately before reading the luminescence signal (which took about 6 min). The luminescence signal was measured using a CLARIOstar Plus microplate reader (BMG Labtech). The assay was performed in duplicate (technical replicates) and relative luminescence units (RLUs) were plotted against masked complexes concentration; the data were analyzed using the non‐linear fit in Prism Version 9.4.1 with all samples constrained to the same top and bottom value (89,058 and 23,041, respectively, as determined for NoMC without constraints). For replicate experiments, masked complexes (4 μM) were incubated with TEV (3 μM); the assay range was 1.3 nM to 2 μM; the top and bottom constraint values were 107,720 and 21,383, respectively, as determined for NoMC without constraints.

4.8. N‐deglycosylation of masked complexes

Masked complexes (4 μM) were first incubated with TEV (3.75 μM) at 37°C for 16 h, or without TEV at 4°C for 16 h. Then, 5 μL PNGase F (NEB, P0704L) was incubated in a 20 μL reaction for 4 h at 37°C as recommended by the manufacturer's “Non‐Denaturing Reaction Conditions” protocol. Immediately before running the gel all samples were (1) incubated with loading dye supplemented with BME to reach a final concentration of ~10% BME and (2) boiled at 99°C for 10 min. Samples were then analyzed on a 4%–20% Mini‐PROTEAN® TGX Stain‐Free gel (Bio‐Rad, 4568096) to reach ~0.5 μg Masked Complex, and ran at 200 V for 30 min. The gel was imaged after staining with Colloidal Coomassie.

AUTHOR CONTRIBUTIONS

Odessa J. Goudy: Conceptualization (equal); data curation (lead); formal analysis (equal); funding acquisition (supporting); investigation (lead); methodology (lead); supervision (supporting); validation (equal); visualization (lead); writing – original draft (equal); writing – review and editing (equal). Alice Peng: Investigation (supporting). Ashutosh Tripathy: Methodology (supporting); resources (equal); writing – review and editing (supporting). Brian Kuhlman: Conceptualization (equal); formal analysis (equal); funding acquisition (lead); project administration (lead); supervision (lead); writing – original draft (supporting); writing – review and editing (equal).

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interest.

Supporting information

FIGURE S1. Quantifying changes in yeast display fluorescence intensity.

FIGURE S2. Expression and purification of masked complexes.

FIGURE S3. TEV efficiently cleaves all masked complexes.

FIGURE S4. NoMC shows similar activity to an anti‐PD‐1 antibody.

FIGURE S5. Replicate experiment results for cell bioassay.

FIGURE S6. Representative SPR sensorgrams.

TABLE S1. Robetta alanine scan results.

TABLE S2. Comprehensive list of kinetic parameters.

TABLE S3. List of DNA and primer sequences for yeast display.

TABLE S4. Protein sequences used in all experiments.

TABLE S5. List of DNA clonal gene sequences for expression.

Goudy OJ, Peng A, Tripathy A, Kuhlman B. Design of a protease‐activated PD‐L1 inhibitor. Protein Science. 2023;32(3):e4578. 10.1002/pro.4578

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon request.