MDA-MB231, DU-145, CT26, 4T1 and Jurkat cells were purchased from ATCC. MDA-MB231 and DU-145 cells were cultured in DMEM medium with 10% Fetal Bovine Serum (FBS), 100 units/mL penicillin and 100 μg/mL streptomycin. CT26, 4 T1, and Jurkat cells were cultured in RPMI1640 medium with 10% FBS, 100 units/mL penicillin and 100 μg/mL streptomycin. All cells were grown at 37 °C in a humidified atmosphere containing 5% CO2.

The recombinant human PD-L1 extracellular domain (ECD) protein (cat# FCL0784B, G&P Biosciences, Santa Clara, CA) was coated on two wells of a 96-well plate. On the first well, PD-L1 was incubated with PD-1 protein, followed by incubation with the Ph.D.™-12 Phage Display Peptide Library (NEB, Ipswich, MA). The unbound phages were transferred to the second well, which was coated with PD-L1. The bounded phages were eluted from the second well and amplified. In each biopanning, approximately 10¹¹ pfu phages were loaded, and eluted phages were tittered and amplified for the next round of selection.

Ninety-six-well plates were coated with 100 ng of PD-L1 protein (G&P Biosciences, human PD-L1 ECD, cat# FCL0784B. mouse PD-L1, cat# FCL3502B) and later blocked with 2% BSA for 2 h at room temperature. Various concentrations of peptides were loaded into the wells and incubated for 1 h at room temperature. Biotinylated PD-1 (G&P Biosciences, human PD-1 ECD, cat# FCL0761B; mouse PD-1, cat# FCL1846) was added and incubated for 1 h. Streptavidin-HRP (R&D systems) and chromogenic substrate were added into the wells. OD₄₅₀ was then recorded and referenced to OD₅₄₀.

Binding affinities of the PD-L1 specific peptides for human PD-L1 protein were determined by SPR (BI4500, Biosensing Instrument). PD-L1 protein was diluted to 10 μg/mL with sodium acetate buffer (pH 5.0, GE Healthcare, PA) and covalently coated onto a CM5 sensor chip (CM Dextran Sensor Chip, Biosensing Instrument) using the standard Amine Coupling Kit (GE Healthcare, PA). Approximately 6500 RU of PD-L1 protein were immobilized onto the chip. A second channel was used as a reference. HBS-EP+ buffer (GE Healthcare) was employed at a flow rate of 60 μL/min. A series of concentrations of each peptide (15, 30, 60, 125, 250, 500, 1000, 5000 and 10,000 nM) were prepared in HBS-EP+ running buffer to obtain the equilibrium dissociation constant (K_D) values of the peptides. The CM5 sensor chip was regenerated with 10 mM NaOH for 20 s. The results were analyzed using the software of Bi data analysis software [11].

Binding of the peptides to PD-L1-positive cancer cells (MDA-MB-231 and DU-145) and PD-L1-negative cancer cells (MCF-7) was evaluated as we described before with modifications [18]. The cells were treated with the non-enzymatic cell dissociation solution (MP Biomedicals, Santa Ana, CA) and diluted to a density of 1 × 10⁶ cells/mL in Opti-MEM. The suspended cells were incubated with various concentrations of 5-FAM-labeled anti-PD-L1 peptides or Cy5-labeled PD-L1 antibody for 1 h at 37 °C with gentle rotation. After washing, the cells were analyzed using a FACSCalibur flow cytometer (BD Biosciences, Franklin Lakes, NJ).

3D spheroids of MDA-MB-231 cells were generated using the Spheroid Formation Extracellular Matrix (ECM) as per the company’s protocol (Amsbio, Cambridge, MA). Briefly, 3000 tumor cells were mixed with 50 μL Spheroid Formation ECM and loaded into a Corning™ 96-well Ultra-low Attachment Treated Spheroid Microplate (Corning, Pittsburgh, PA). The plate was centrifuged at 200 g for 3 min at 4 °C. The cells were then incubated at 37 °C until the diameter reached approximately 700 μm. The Cy5-labeled CLP002 peptide and anti-PD-L1 antibody were incubated with the spheroids for 2 or 6 h. After washing, penetration of the peptide and antibody inside the tumor spheroids was determined using confocal microscopy.

Proliferation and apoptosis of Jurkat T cells were assessed as described [19, 20]. Briefly, 3 × 10⁴ Jurkat T cells were cultured alone or co-cultured with 1.5 × 10⁵ DU-145 cancer cells in a 24-well plate for the proliferation assay. For the apoptosis assay, 6 × 10⁴ Jurkat T cells were cultured alone or co-cultured with 3 × 10⁵ DU-145 cancer cells in a 6-well plate. After incubation with the anti-PD-L1 peptides (5 μM) or anti-PD-L1 antibody (1 μM) for 24 h, Jurkat T cells were harvested from the supernatant. Proliferation of Jurkat cells was determined using the CellTiter-Glo luminescent cell viability assay (Promega, WI), and apoptosis of the cells were determined using the Dead Cell Apoptosis Kit with Annexin V Alexa Fluor® 488 and Propidium Iodide (Thermo-Fisher Scientific, Pittsburgh, PA) as we described before [21].

The crystal structure of human PD-L1 protein (PDB ID: 5C3T) and its binding residues to PD-1 were reported previously [22]. Structures of the peptides were generated using BIOVIA Draw (BIOVIA) and then aligned to the PD-L1 structure using Autodock Vina. Illustrations of the PD-L1 protein and peptide complex were generated using Pymol (Delano Scientific).

The animal protocol was approved by the University of Missouri-Kansas City, Institutional Animal Care and Use Committee (IACUC). Five-week old male and female Balb/c mice were purchased from Charles Rivers Laboratories (Wilmington, Massachusetts) and housed in a temperature and humidity controlled room on a 12-h light-dark cycle. Approximately 5 × 10⁵ CT26 cells were subcutaneously injected into the right flank. The mice were randomly divided into five different groups (10 mice/group, 50% female, 50% male). The mice were intraperitoneally injected with 2 mg/kg peptide daily when the tumor size reached 50–100 mm³. The anti-mouse PD-L1 antibody (10F.9G2, BioXcell) was administrated as a positive control at a dose of 10 mg/kg every two days. The tumor size was assessed with a caliper and calculated with the formula 0.5 × length×width². ELISA kits of PD-L1 (Cat# DY1019–05), IFNγ (Cat# DY485–05), and IL-6 (Cat# DY406–05) were used to measure the expressions of PD-L1, IFNγ, and IL-6, respectively, in tumors as per the company’s instructions (R&D Systems, Minneapolis, MN).

Tumor tissues were fixed in 10% formalin, embedded in paraffin, sectioned, and mounted on glass slides by Truman Medical Centers Anatomic Pathology Core (Kansas City, MO). The slides were heated in Tris buffer (pH 9.0) for 45 min to recover antigen. After deparaffinization and rehydration, the sections were stained with the anti-mouse CD8 alpha antibody (Abcam, ab209775) overnight at 4 °C. The slides were incubated with a biotinylated goat anti-rabbit secondary antibody, followed by the DAB chromogen mixture. Four sections (2 male and 2 female) were imaged in each group. For each section, 3 regions were randomly selected for imaging.

Data are expressed as the mean ± standard deviation (SD). The difference between any two groups was determined by one-way analysis of variance (ANOVA) with Tukey’s post hoc test. For the tumor volume, the difference between any two groups was determined by two-way ANOVA with Tukey’s post hoc test. P < 0.05 was considered statistically significant.

The aim of this study is to discover small peptides not only specifically bind to PD-L1 but also block the interaction between PD-L1 and PD-1. We developed a novel biopanning strategy to select peptides that specifically bind to PD-L1. As Fig. 1a shows, after five rounds of biopanning, the number of eluted phages increased dramatically, indicating significant enrichment of PD-L1-specific phages in the elution. Totally 57 single phage colonies were randomly selected for sequencing, and 4 peptide sequences were discovered (Fig. 1b). The CLP002 peptide and CLP003 peptide have 21 and 32 repeats, respectively, while the CLP001 peptide and CLP004 peptide have 1 and 3 repeats, respectively.Fig1Fig. 1

Discovery of anti-PD-L1 peptides using a novel biopanning procedure. a The number of recovered phages from each round of biopanning. b Sequences of the discovered anti-PD-L1 peptides. c Binding affinities of the selected peptides towards human PD-L1 protein and albumin were measured using SPR. d Binding curves of the peptides on PD-L1-positive human cancer cells (MDA-MB-231 and DU-145) and PD-L1 deficient human cancer cells MCF-7. Binding curves of the anti-PD-L1 antibody were measured on DU-145 and MCF-7 cells. Results are represented as the mean ± SD (n = 3)

Binding affinities of the discovered peptides against the recombinant human PD-L1 ECD protein were evaluated using SPR. The PD-L1 protein was immobilized on a CM5 golden chip by the direct amine coupling method. As illustrated in Fig. 1b, the K_D values of CLP001, CLP002, CLP003 and CLP004 for human PD-L1 were 534, 366, 117, and 544 nM, respectively. The peptides are considerable competitive against the PD-1/PD-L1 interaction (K_D = ~ 4 μM) [23]. Even though the binding affinity of peptides are generally lower than that of antibodies, the very high affinity of antibodies may result in on-target, off-tumor toxicity in healthy tissues that express low levels of PD-L1. In a recent study, scientists constructed several chimeric antigen receptor (CAR) T cells with different affinities to ICAM-1. CAR-T cells with micromolar affinity to ICAM-1 showed better antitumor efficacy and safety than CAR-T cells with nanomolar affinity. CAR-T cells with nanomolar affinity lyse healthy cells that express a low level of ICAM-1. By contrast, CAR-T cells with micromolar affinity only attack tumor cells with high levels of ICAM-1 but not healthy cells with low levels of ICAM-1, leading to less toxicity [24].

Having shown the high affinity of the peptides to PD-L1, we next examined the specificity of these peptides. We first measured nonspecific binding of the peptides against bovine serum albumin (BSA) using SPR. As revealed in Fig. 1c, the response curves of BSA to the peptides did not change with gradient concentrations of the peptides, indicating negligible binding between the peptides and BSA. By contrast, the response curves of the peptides to the PD-L1 protein were correlated with the peptide concentrations. We further examined their specificity to PD-L1-positive human cancer cells DU-145. MCF-7 human cancer cells are PD-L1 deficient and used as a negative control in this study [25, 26]. As Fig. 1d revealed, all peptides and PD-L1 antibody (29E.2A3, BioXcell, West Lebanon, NH) exhibited high binding affinity to PD-L1-positive cancer cells (DU-145) but low affinity to PD-L1-deficient human cancer cells MCF-7. These results clearly suggest that the anti-PD-L1 peptides specifically bind to recombinant human PD-L1 protein as well as PD-L1 overexpressing human cancer cells.

We next determined whether the anti-PD-L1 peptides block the human PD-1/PD-L1 interaction. An anti-human PD-L1 antibody (R&D Systems, cat# AF156) was used as a positive control to calibrate this assay. As shown in Fig. 2a, the anti-human PD-L1 antibody blocked the PD-1/PD-L1 interaction with a half maximal inhibitory concentration (IC₅₀) of 36.76 nM, which is consistent with the report from the company. IC₅₀ of the antibody against PD-L1 overexpressing DU-145 cancer cells is 38.11 nM, which is comparable to the blocking effect on PD-L1 protein (Fig. 2b). We next examined the blocking efficiency of the anti-PD-L1 peptides at 10 μM (Fig. 2c). CLP002 showed the highest blocking efficiency, whereas CLP001 did not block the PD-1/PD-L1 interaction. We also determined IC₅₀ of each peptide using recombinant human PD-L1 protein and DU-145 cancer cells. As revealed in Fig. 2d-f, CLP002 exerted the best blocking effect (85%) with an IC₅₀ of 2.17 μM, when the plate was coated with the human PD-L1 protein. The blocking effect was 80% with an IC₅₀ of 1.43 μM, when the plate was coated with DU-145 cells. IC₅₀ of the CLP003 peptide was 2.22 μM with 60% blocking efficiency against the human PD-L1 protein, and the IC₅₀ was 3.05 μM with a 56% blocking efficiency against DU-145 cancer cells.Fig2Fig. 2

Blockade of the PD-1/PD-L1 interaction by the anti-PD-L1 peptides and antibody. a Blocking profile of the anti-human PD-L1 antibody (R&D, AF156) against human PD-L1 protein. b Blocking profile of the anti-human PD-L1 antibody (R&D, AF156) against DU-145 cells. c Blocking efficiency of the anti-PD-L1 peptides (10 μM) and the anti-human PD-L1 antibody (1 μM) against human PD-L1 protein. d IC₅₀ and blocking efficiency of the peptides and antibody against human PD-L1 protein and human cancer cell line DU-145. e Blocking profiles of the peptides against human PD-L1 protein. f Blocking profiles of the peptides against DU-145 cells. g Blocking efficiency of the peptides and an anti-mouse PD-L1 antibody (BioXcell, 10F.9G2) at 10 μM against a mouse PD-L1 protein. h IC₅₀ and blocking efficiency of the peptides against mouse PD-L1 protein and mouse cancer cell line 4 T1. i Blocking profiles of the peptides against mouse PD-L1 protein. j Blocking profiles of the peptides against mouse cancer cell line 4 T1. Results are represented as the mean ± SD (n = 3)

We performed molecular docking studies to simulate the interactions between the anti-PD-L1 peptides and the human PD-L1 extracellular domain protein (PDB ID# 5C3T) using Autodock Vina integrated into PyRx [29]. Illustrations of the PD-L1/peptide complexes were generated using Pymol (Fig. 3). The PD-L1 residues responsible for the PD-1/PD-L1 interaction were previously reported and highlighted in yellow [22]. The binding residues of CLP002 and CLP003 on PD-L1 are highly overlapped with that of PD-1 (Fig. 3b and Fig. 3c). As illustrated in Fig. 3a, the CLP001 peptide does not bind to the PD-1/PD-L1 interaction residues, which explains the fact that the CLP001 peptide binds to PD-L1 but does not block the PD-L1/PD-1 interaction (Fig. 2d-e). Similarly, there is only a small overlap between the CLP004/PD-L1 binding area and the PD-L1/PD-1 interaction residues (Fig. 3d). This is also in accordance with the poor blocking efficacy of the CLP004 peptide in Fig. 2.Fig3Fig. 3

Molecular docking for the interaction between the anti-PD-L1 peptides and human PD-L1 protein (PDB ID: 5C3T). a Modeling of the interaction between CLP001 and PD-L1. b Modeling of the interaction between CLP002 and PD-L1. c Modeling of the interaction between CLP003 and PD-L1. d Modeling of the interaction between CLP004 and PD-L1. The PD-L1 residues responsible for peptide binding are highlighted in green. The binding residues for human PD-1 protein is highlighted in yellow. The overlapping PD-L1 residues for binding both anti-PD-L1 peptide and PD-1 protein are highlighted in pink

In the tumor microenvironment, PD-L1-overexpressing tumor cells inhibit T cell activation and promote T cell apoptosis, leading to exhausted phenotype and impaired effector function of the T cells [20]. The PD-1/PD-L1 interaction also suppresses T cell proliferation and inhibits the secretion of inflammatory cytokines [30]. We, therefore, co-cultured Jurkat T cells with PD-L1-overexpressing DU-145 cancer cells to investigate whether the anti-PD-L1 peptides reverse the inhibitory effect of DU-145 cancer cells on Jurkat T cells.

As revealed in Fig. 4a, DU-145 cells significantly inhibited T cell proliferation through the PD-1/PD-L1 interaction. Treatment of the co-cultured cells with the CLP002 peptide restored Jurkat T cell proliferation, which is consistent with previous reports [19, 31–33]. For example, Freeman et al. reported that human PD-L1 protein suppressed the proliferation of T cells in a dose-dependent manner. By contrast, the PD-L1 protein did not inhibit the proliferation of PD-1 knockout T cells, suggesting that the inhibitory effect of DU-145 cells is mediated by the PD-1/PD-L1 interaction [19]. In another study, PD-L1 expression on myeloid-derived suppressor cells (MDSCs) was found selectively upregulated by hypoxia-inducible factor-1 α (HIF-1α) under hypoxia, leading to suppression of T cells. Using HIF-1α or PD-L1 inhibitors reversed MDSC-mediated T cell suppression under hypoxia [31].Fig4Fig. 4

The CLP002 peptide restores T cell proliferation and prevents T cell apoptosis. Jurkat T cells were co-cultured with DU-145 cells and then incubated with the anti-PD-L1 peptides or antibody for 24 h. The CLP002 peptide and antibody restore Jurkat T cell proliferation (a) and reduces Jurkat T cell apoptosis (b-c) in the presence of PD-L1 overexpressing DU-145 cells. Results are represented as the mean ± SD (n = 3). (** p < 0.01; *** p < 0.001)

We hypothesize that low-molecular-weight peptides have better tumor penetration than antibodies, which may lead to improved therapeutic efficacy. A 3D tumor spheroid model of MDA-MB-231 cells was developed to compare tumor penetration of the CLP002 peptide and the anti-PD-L1 antibody (29E.2A3, BioXcell, West Lebanon, NH). Cy5-labeled peptide and antibody were incubated with the tumor spheroids (~ 700 μm in diameter) for 2 and 6 h, followed by confocal microscopy analysis to evaluate tumor penetration. As illustrated in Fig. 5a-b, the CLP002 peptide exhibited better tumor penetration than the antibody. Fluorescence of the Cy5-labeled CLP002 peptide was detected as deep as approximately 250 μm from the periphery of the spheroids. By contrast, Cy5-labeled antibody was only detected on the periphery of the spheroids, suggesting very limited tumor penetration.Fig5Fig. 5

3D spheroid penetration of the CLP002 peptide and anti-PD-L1 antibody. 3D tumor spheroids of MDA-MB-231 cells were generated to compare the tumor penetration capability of the CLP002 peptide and the anti-PD-L1 antibody (BioXcell, 29E.2A3). Cy5-labeled peptide and antibody were incubated with the tumor spheroids (~ 700 μm in diameter) for 2 and 6 h, followed by confocal microscopy analysis to evaluate tumor penetration. a Representative Z-stacked confocal images of the spheroids with a z-step of 50 μm. The scale bar represents 200 μm. b The depth of penetration is quantified by mean fluorescence intensity. Results are represented as the mean ± SD (n = 3)

We evaluated the antitumor activity of the anti-PD-L1 peptides using the CT26 colorectal tumor-bearing mouse model (Fig. 6a), which has been widely used to evaluate the activity of PD-1/PD-L1 inhibitors [11, 34]. Once the average tumor volume reached 50–100 mm³, the peptides (2 mg/kg) were administrated intraperitoneally daily, as described in a previous study [11]. The anti-mouse PD-L1 antibody (BioXcell, 10F.9G2) was administered intraperitoneally every other day at 10 mg/kg as reported [35]. As Fig. 6b to d showed, CLP002, CLP003, and the antibody effectively suppressed tumor growth. As shown in Fig. 6e, the tumor weights of the PD-L1 antibody, CLP002 and CLP003 group were significantly smaller than the saline group. In general, CLP002 exerted a better tumor inhibitory effect than CLP003, which was similar to the antibody. It is noteworthy to mention that the peptides were screened against the human PD-L1 protein, which, in fact, would compromise the anti-tumor activity of the peptides in a mouse model. We are, therefore, cautiously optimistic with the antitumor activity of the peptides in human cancer cells.Fig6Fig. 6

Anti-tumor activity of the anti-PD-L1 peptides and antibody. a CT26 tumor-bearing Balb/C mice (n = 10, 5 male and 5 female) were intraperitoneally injected with the anti-PD-L1 peptides (2 mg/Kg) daily for a total of 10 injections and the anti-mouse PD-L1 antibody (10 mg/Kg) every other day for a total of 5 injections. b Tumor volume measured over time. Tumor volume results were represented as the mean ± SE (n = 10). c Tumor growth curves of individual mice in each group. Image d and weight e of tumors harvested at day 14. The results were represented as the mean ± SD (n = 10). The expressions of IFNγ f, PD-L1 g and IL-6 h in harvested tumors were measured using ELISA. i The numbers of CD8⁺ T cells in each specimen were quantitated after immunohistochemical staining. Results were represented as the mean ± SD (n = 4). j Representative images of tumor specimen stained with anti-CD8 antibody. The scale bar represents 200 μm. (* p < 0.05; ** p < 0.01; *** p < 0.001)