The FG Loop of PD-1 Serves as a "Hotspot" for Therapeutic Monoclonal Antibodies in Tumor Immune Checkpoint Therapy.

Programmed cell death 1 (PD-1)/PD-1 ligand-1 (PD-L1)-blocking monoclonal antibodies (mAbs) have taken center stage for tumor immune checkpoint therapy. Identification of the "hotspots" on PD-1 for mAbs will help to develop next-generation oral deliverable agents with long-lasting efficacy. Here, we identified two PD-1-targeting mAbs, GY-5 and GY-14, with PD-1/PD-L1-blocking efficacy. Complex structural information revealed that both mAbs mainly bind to the FG loop of PD-1, which also contributes multiple interactions with PD-L1. The FG loop adopts substantially varied conformations upon binding to different mAbs, providing a novel targetable region for the development of PD-1-specific biologics and small chemical molecules. Glycosylation modifications of PD-1 could be observed in three of the four potential N-linked glycosylation sites. However, the binding of GY-5 and GY-14 to PD-1 was not affected by glycosylation. These findings broaden our understanding of the mechanism of anti-PD-1 mAbs and provide insight into the development of agents targeting PD-1.

Introduction

Immune checkpoint therapy (ICT) that targets co-inhibitory or co-stimulatory molecules to modulate anti-tumor T cell reactivity has achieved clinical success since the approval of the cytotoxic T-lymphocyte-associated protein 4 (CTLA-4)-targeting drug ipilimumab by the US Food and Drug Administration (FDA) in 2011 (Callahan et al., 2016, Tan and Gao, 2015). Programmed cell death 1 (PD-1), a member of the CD28-B7 superfamily, is an important co-inhibitory molecule in the modulation of T cell activity (Ishida et al., 1992). PD-1 ligand 1 (PD-L1) and then PD-1 ligand 2 (PD-L2) were subsequently identified as the ligands of PD-1 (Freeman et al., 2000, Latchman et al., 2001, Nishimura et al., 1999). Interruption of the PD-1/PD-L1 interaction with a monoclonal antibody (mAb) to re-stimulate tumor-specific T cell reactivity has been proved to be a promising strategy for treating multiple tumors in clinical applications (Motzer et al., 2015, Robert et al., 2015, Topalian et al., 2012).

The clinical success of anti-PD-1 or anti-PD-L1 mAbs in tumor therapy has initiated an era of anti-tumor drug development to modulate tumor-specific immune responses by targeting immune checkpoint molecules, either co-stimulatory (e.g., 4-1BB) or co-inhibitory (e.g., CTLA-4, PD-1) molecules, to treat tumors (He et al., 2017, Tan and Gao, 2015, Tan et al., 2016). Six mAbs targeting PD-1 or PD-L1 have been approved by the US FDA since 2014 (Tan et al., 2016). However, clinical responsiveness and benefits from these therapies are still limited owing to the disadvantage of the monotherapeutics in highly heterogeneous tumors. Therefore a combination of mAbs targeting varied immune checkpoints or other immune therapeutic strategies, e.g., oncolytic virus and chimeric antigen receptor-engineered T cells, are under evaluation at both the basic research and clinical levels to improve the clinical responsiveness and benefits of the tumor ICT. Looking for mAb replacement like small molecules for convenient drug delivery is another focal point for the field.

The interaction of PD-1 and PD-L1 plays pivotal roles in immune suppression within the tumor microenvironment (Tan and Gao, 2015, Tan et al., 2016). Upregulated PD-L1 expression on tumor cells is correlated with tumor progression and, hence, is a valuable indication for unfavorable prognosis (Iwai et al., 2002, Gridelli et al., 2017). On the other hand, higher PD-L1 levels in tumor tissue indicate better responsiveness to PD-1/PD-L1 blockade treatment, to a certain extent, in ovarian, kidney, pancreatic, and gastric cancers (Powles et al., 2017, Massard et al., 2016, Apolo et al., 2017, Balar et al., 2016). Inducible PD-1 expression on T lymphocytes (tumor-infiltrating lymphocytes in particular) can lead to the tolerance of tumor-specific T cells to tumors (Tumeh et al., 2014). Moreover, PD-1 has also been found to be expressed in tumor cells, and tumor-cell-intrinsic PD-1 can promote tumorigenesis by modulating downstream mammalian target of rapamycin signaling (Kleffel et al., 2015). Forced expression of PD-1 or PD-L1 on T cells or tumor cells underlies the rationale that blockade of the PD-1 signaling would restore tumor-specific T cell function to eliminate tumor cells (Curiel et al., 2003, Hirano et al., 2005).

Recently, the complex structures of FDA-approved mAbs targeting PD-1 or PD-L1 were determined, providing critical information for our understanding of mAb-based PD-1/PD-L1 blockage for ICT (Tan et al., 2018, Liu et al., 2017, Tan et al., 2016, Tan et al., 2017, Lee et al., 2016, Na et al., 2017). The binding and blocking mechanisms of the two PD-1-targeting mAbs, nivolumab (Opdivo, Bristol-Myers Squibb) and pembrolizumab (Keytruda, Merck) are reported to be quite different (Tan et al., 2016). Nivolumab binds to the N-terminal loop of PD-1, which is outside the Ig-like domain, and competes for the binding of PD-L1 with its light chain (L chain) (Lee et al., 2016, Tan et al., 2017). On the other side, pembrolizumab binds to the C’D loop of PD-1, involving mainly the heavy chain (H chain) of the mAb, and competes with the PD-1/PD-L1 interaction with both its H and L chains (Na et al., 2017). Although the binding regions of nivolumab and pembrolizumab on PD-1 are different, the binding of nivolumab to PD-1 would abrogate the additional binding of pembrolizumab, indicating the competitive binding profiles of these two mAbs (Lee et al., 2016, Tan et al., 2017, Na et al., 2017). However, whether there are “hotspots” for mAb-based anti-PD-1 checkpoint blockage therapy or additional novel “hotspot” regions within PD-1 for therapeutic mAb development remains unknown.

Here, we report the screening of therapeutic mAbs targeting PD-1 and the structural basis of two of these mAbs for PD-1/PD-L1 blockage. We found that the FG loop of PD-1 was targeted by both mAbs, indicating that the PD-1 FG loop may serve as a novel “hotspot” for mAb-based PD-1 ICT. Moreover, the dependency of glycosylation modifications of PD-1 to the binding of these two mAbs was also investigated. Our findings will aid in the future development of biologics or small chemical molecules by targeting PD-1.

Results

Tumor Suppression Efficacy and Humanization of PD-1-Targeting mAbs

To investigate the hotspots on PD-1 for mAbs, B6/C57 mice were vaccinated with human PD-1 protein expressed by HEK293T cells. Thirty one cell hybrid clones that yield PD-1-specific mAbs were obtained after hybridization. Among these mAbs, 23 were found to interrupt the interaction between PD-1 and PD-L1 through a flow cytometry-based assay in which PD-L1 was expressed on 293T cells (Figure S1). Evaluation of the efficiency of enhancing T cell reactivity with enzyme-linked immunospot assays revealed that 11 of the mAbs elevated the T cell responses against influenza A virus M1 peptide pool (Figure S2). Sequences of eight mAb clones were obtained, and the representative GY-5 and GY-14 mAbs were selected for further investigations for their tumor suppressive efficacy and blocking mechanistic studies (Table S1).

Tumor-bearing mouse models were used to investigate anti-tumor activity of chimeric mAbs of GY-5 and GY-14 of human IgG4 subtype. The anti-tumor efficacies of GY-5 and GY-14 were evaluated in the human non-small-cell lung cancer cell line HCC-827-bearing NOD prkdc−/− IL−2Rg−/− (NCG) mouse model with pre-established human immunity by inoculation with 1 × 107 peripheral blood mononuclear cells from healthy donors (Figure 1A). An Ebola virus GP-protein-specific mAb, 13C6 (Audet et al., 2014), was used as a negative control. The mAbs were injected intraperitoneally. twice a week for six doses, and tumor volumes were monitored twice a week. The results revealed that both GY-5 and GY-14 showed significant tumor suppression 2 weeks after their first dose compared with 13C6 (Student's t test, p < 0.05 or <0.01) (Figure 1).

Figure 1

Tumor Suppression Efficacy of the GY-5 and GY-14 mAbs

(A) Flowchart of the animal study. 1×107 PBMCs from healthy donors were inoculated intravenously into each NCG mouse, and 5×106 HCC-827 cells were inoculated subcutaneously 3 days later. Chimeric GY-5 and GY-14 mAbs were injected intraperitoneally (i.p.) every 3 or 4 days from day 7 (D0) after tumor inoculation. The size of the tumor was monitored every 3 or 4 days after injection of GY-5 or GY-14. An Ebola virus-specific mAb, 13C6, was injected and used as a negative control.

(B and C) Mice bearing subcutaneous HCC-827 tumors for 7 days were treated i.p. with GY-5 at a dose of 200 μg/mouse or control IgG, with five mice for each treatment group (B). Mice bearing HCC-827 tumors for 7 days were treated i.p. with GY-14 at a dose of 200 μg/mouse or control IgG, with eight mice for each treatment group, which was independent of the experiment with GY-5 (C). The data with dot shows the average tumor volume of the group, whereas the SE is presented as longitudinal bars. *p < 0.05, **p < 0.001. The red line represents the average tumor size of the control IgG treatment group.

(D and E) Tumor sizes of individual mice are presented for GY-5 treatment group (D) and GY-14 treatment group (E) , with each black line showing the changes of the tumor size of a mouse. The red lines represent tumor size of each mouse of control IgG treatment group. The data presented here are representative of two independent experiments with similar results.

See also Figures S1 and S2 and Table S1.

GY-5 and GY-14 were subsequently humanized via the “CDR grafting” method, and the PD-1/PD-L1 blocking efficiency and binding affinity of the humanized mAbs were evaluated (Figures S3 and S4). Humanized GY-5 and GY-14 could efficiently block the binding of PD-1 to PD-L1s expressed on 293T cells (Figure 2A). The binding affinities (KD) of both chimeric and humanized GY-5 or GY-14 for PD-1 were then analyzed using surface plasmon resonance (SPR) (Figures 2B–2D). We found that the binding affinity of humanized GY-5 and GY-14 (KD = 9.62 and 2.21 nM, respectively) for PD-1 was similar to that of the chimeric GY-5 and GY-14 mAbs (KD = 1.22 and 1.54 nM, respectively) (Table S2). Therefore the humanized GY-5 and GY-14 could serve as promising PD-1-targeting therapeutics for tumor ICT.

Figure 2

Blocking and Binding Characteristics of Humanized GY-5 and GY-14

(A) The blocking of the binding of PD-1-mFc to PD-L1s expressed on the surface of 293T cells by humanized GY-5 and GY-14 mAbs. The PD-L1-expressing HEK293T cells were stained with PBS and used as negative controls, whereas the staining with PD-1-mFc proteins was used as a positive control. The blocking efficacies of GY-5 and GY-14 were analyzed by staining of a protein complex of mAbs and PD-1-mFc at a molar ratio of 2:1 and a final concentration of 10 μg/mL. The absence of the PD-1-mFc-staining-positive subpopulation indicates the blockage of the PD-1/PD-L1 interaction.

(B–D) SPR-assay-based characterization of the binding of chimeric and humanized GY-5 (B) or GY-14 (C) to PD-1 was accomplished using a single-cycle BIAcore T100 system. The binding kinetics of nivolumab to PD-1 was evaluated and served as a positive control (D). The equilibrium dissociation constant (KD) of the binding is labeled accordingly. The data presented here are representatives of two independent experiments with similar results.

See also Figures S3 and S4 and Table S2.

Structural Basis of GY-5 and GY-14 Binding to PD-1 for PD-1/PD-L1 Blockage

The complex structures of GY-5/PD-1 and GY-14/PD-1 were determined at a resolution of 2.6 and 1.4 Å, respectively, which enabled us to analyze the binding and blocking mechanisms of these two mAbs (Table S3). Overall, GY-5 and GY-14 bind to PD-1 with similar binding orientations, and both mAbs mainly bind to the FG loop of PD-1 (Figures 3A and 3B). The binding of GY-5 mainly involves the CDR2 and CDR3 of its H chain, and CDR1 and CDR3 of its L chain, with a buried surface of 1,730.1 Å2 (Figure 3A and Table S4). On the other side, GY-14 utilizes all six CDRs of both the H and L chains to contact PD-1, with a buried surface of 1,602.6 Å2 (Figure 3B and Table S5). The previously reported complex structures of PD-1/PD-L1, nivolumab/PD-1, and pembrolizumab/PD-1 enabled us to comprehensively compare the binding surface of these mAbs and the ligand (Figure 3C). Nivolumab mainly binds to the N-terminal loop of PD-1, with partial contacts with the FG loop, whereas pembrolizumab mainly binds to the C’D loop. Both GY-5 and GY-14 mainly bind to the FG loop of PD-1, with the binding surfaces of GY-5 and GY-14 being more proximal to that of nivolumab.

Figure 3

The Complex Structure of GY-5 or GY-14 with PD-1

(A and B) The overall structure of the GY-5 and PD-1 complex (A) or GY-14 and PD-1 complex (B). The V fragments of GY-4 and GY-5 are shown as cartoons, with the heavy chain (VH) and light chain (VL) depicted in different colors, whereas PD-1 is shown in surface representation, and the glycan modifications at N116 are shown as smuggle spheres (light blue). The CDR1, CDR2, and CDR3 loops of the heavy chain (HCDR1, HCDR2, and HCDR3) are colored in blue, yellow, and red, respectively. The CDR1, CDR2, and CDR3 loops of the light chain (LCDR1, LCDR2, and LCDR3) are colored in pink, orange, and green, respectively. The FG loop of the PD-1 molecule is highlighted in hot pink.

(C) Comparison of the binding surface of PD-L1 and complex structures of known PD-1-targeting mAbs on PD-1. The binding surface of PD-L1 and the mAbs targeting PD-1, GY-4, GY-5, nivolumab, and pembrolizumab on PD-1 are presented in orange, green, cyan, yellow, and purple as shown. The PD-1s in the complex with each of the partners were extracted and presented.

See also Table S3.

Detailed analysis shows that the FG loop of PD-1 contributes major hydrogen bond interactions with GY-5 and GY-14 (Figure 4). Specifically, amino acids of the FG loop (P130, K131, Q133, and I134) formed multiple hydrogen bond interactions with HCDR3 (E97, D100, Y103, and Y104), LCDR1 (H31 and D33), and LCDR3 (Y101) of GY-5 (Figure 4A). Similarly, amino acids of the FG loop (L128, P130, K131, A132, Q133, and I134) formed multiple hydrogen bond interactions with HCDR1 (E33), HCDR3 (E99, M101, N102, T103, W105, and Y106), LCDR2 (H31), and LCDR3 (Y101) of GY-14 (Figure 4B).

Figure 4

Binding Details of GY-5 and GY-14 to the FG Loop of PD-1

The amino acids forming hydrogen bond interactions between GY-5 and the FG loop of PD-1 (A) or GY-14 and the FG loop (B) are presented as sticks. The LCDR3 and LCDR1 of GY-5 are colored in green and purple, respectively, whereas the HCDR3 of GY-5 and the FG loop of PD-1 are presented in wheat and light blue. The CDRs of GY-14 are colored the same as that of GY-5. The hydrogen bonds are shown as black dashed lines. See also Tables S4 and S5.

We next analyzed the blocking mechanisms of GY-5 and GY-14 to the PD-1/PD-L1 interaction by superimposition of the structure of the previously reported PD-1/PD-L1 complex (PDB code: 4ZQK) with the GY-5/PD-1 complex or GY-14/PD-1 complex, individually (Figure 5). These analyses revealed that the binding of GY-5 and GY-14 induced stereospecific hindrance involving both their H chains to interrupt the binding of PD-L1 to PD-1 (Figures 5A and 5B). The amino acids of the FG loop (L128-Q132) display competitive binding to PD-L1 by both GY-5 and GY-14 (Figures 5C and 5D). However, the competitive binding surfaces of GY-5 and GY-14 are substantially different from each other. These findings suggest that GY-5 and GY-14 bind to PD-1 with similar binding and blocking modes, which is distinct from that of nivolumab and pembrolizumab. In addition, the overwhelming binding affinity of GY-5 or GY-14 (KD = 1.22 and 1.54 nM, respectively) for PD-1 over PD-L1 (KD = 0.7–8.3 μM) also ensures the binding priority of the mAbs (Tan et al., 2017). Taken together, the blockade binding mechanism of GY-5 and GY-14 lies in both the overwhelming binding affinity and H-chain-induced stereo-hindrance to the binding of PD-L1 to PD-1.

Figure 5

Structural Basis of the Competitive Binding of GY-5 and GY-14 with PD-L1

(A and B) Superimposition of the GY-5/PD-1 complex structure (A) or GY-14/PD-1 complex structure (B) with the PD-1/PD-L1 complex structure (PDB code: 4ZQK). PD-L1 is shown in ribbon format in gray, whereas PD-1 is depicted in surface format in light blue. The VH of GY-5 is depicted as a ribbon in light pink and the VL in cyan, whereas the VH of GY-14 is colored in lemon and VL in wheat.

(C and D) Competitive binding surfaces of GY-5 (C) and GY-14 (D) with PD-L1 on PD-1. The residues in contact with PD-L1 alone are colored in gray, whereas the overlapping residues bound by both PD-L1 and GY-5 (C) or GY-14 (D) are colored in marine. The residues in contact with VH and VL of GY-5 alone are colored in light pink and cyan, respectively. The amino acids in PD-1 which contact with the mAbs or PD-L1 were labeled.

Glycosylation-Independent Binding of GY-5 and GY-14

PD-1 has four potential N-linked glycosylation sites (N49, N58, N74, and N116) in its IgV domain. We previously reported the N-linked glycosylation modifications at N58, which consist of two N-acetylglucosamines (NAG) and one fucose, for which the protein was prepared from mammalian cells (Tan et al., 2017). In the present structure of PD-1 from the GY-5/PD-1 complex, which was expressed in insect cells, N-linked glycan modifications were visible in three of the four potential N-linked glycosylation sites: N49, N58, and N116 (Figure 6A). The glycosylation modification at N58 is similar to that observed previously, whereas only a NAG was visible at N49 and N116 (Figures 6B–6D). Considering the flexibility of the glycan chains, N49 and N116 may have more complicated glycan modifications, which is also possible for N74.

Figure 6

Glycosylation Modifications of PD-1 and Glycosylation-Independent Binding of GY-5 and GY-14

(A) Structure of PD-1 with glycans depicted as sticks in purple, and the FG loop is colored in red. Four potential glycosylation sites, N49, N58, N74, and N116, are shown as sticks in light teal.

(B–D) The 2 Fo-Fc electron density maps of N-linked glycans contoured at 1.0 sigma at N49, N58, and N116 are represented in blue. Only one N-acetylglucosamine (NAG) was visible at N49 and N116 (B and D). (C) Three glycans consisting of two NAGs and one L-fucose (FUC) can be clearly observed at N58.

(E) SPR assay characterization of the binding of GY-5 and GY-14 to PD-1 proteins obtained from different expression systems, which enabled varied levels of glycosylation, using a BIAcore T100 system. The refolded PD-1 protein (L25-R147), which was expressed in E. coli and refolded in vitro, and PD-1 protein obtained from insect cells were analyzed for binding affinity with GY-5 and GY-14, with the mAbs immobilized on the chip. The binding affinity (KD) is labeled accordingly. The data presented here are a representative of three independent experiments with similar results.

See also Figure S5 and Table S2.

The glycosylation may play a role in the folding and function of PD-1 and may further affect the binding of these mAbs (Pinho and Reis, 2015). Therefore we analyzed the binding of GY-5 and GY-14 to PD-1 proteins obtained from the insect cell expression system, which enables partial glycosylation of PD-1 compared with the HEK293T cell expression system, or PD-1 proteins refolded from inclusion bodies expressed in E. coli cells as previsouly described (Li et al., 2005, Tan et al., 2017), which have no glycosylation modifications at all. The binding characteristics of these two mAbs were further investigated with SPR analysis. Similar to the glycosylation-independent binding of nivolumab to PD-1, the binding affinity of GY-5 and GY-14 to PD-1 proteins obtained from insect cells (KD = 1.64 nM and 0.52 nM, respectively) or E. coli (KD = 3.52 and 0.34 nM, respectively) showed no substantial differences from those of PD-1 proteins from 293T cells (Figures 2B, 2C, and 6A and, Table S2). These results indicate that the binding of both GY-5 and GY-14 to PD-1 is independent of PD-1 glycosylation.

The FG Loop of PD-1 Serves as a Novel “Hotspot” for PD-1-Targeting mAbs

To investigate the conformational variations of PD-1 upon binding to different mAbs, the PD-1 structures extracted from the GY-5/PD-1 and GY-14/PD-1 complexes and the other two structurally known nivolumab/PD-1 and pembrolizumab/PD-1 complexes were superimposed. The fold motif of the extracellular PD-1 consists of two β-sheets with multiple strands, together with multiple connecting loops (Figure 7A). Superimposition of the structure of PD-1 from the GY-5/PD-1 complex and GY-14/PD-1 complex yields a root-mean-square deviation of 0.509 Å for 85 Cα pairs, demonstrating the conformational conservation of the PD-1 proteins upon binding to different mAbs. Although the β-sheet cores of the PD-1s are highly conserved upon binding to different mAbs, the loops connecting the β-strands exhibit substantial conformational variations (Figure 7B). Three loops of PD-1 are targeted in the interaction with therapeutic mAbs, the C’D loop (pembrolizumab), the N-terminal loop (nivolumab), and the FG loop (GY-5 and GY-14) (Figure 7B). The C’D and the N-terminal loops are visible only upon binding to pembrolizumab and nivolumab, respectively, suggesting the high flexibility of these loops. Taken together, the loops of PD-1 are more prone to be targeted and serve as “hotspots” for therapeutic mAb binding.

Figure 7

Comparative Binding of PD-1-Targeting mAbs

(A) The location of the loops on PD-1, with the N-terminal loop colored in red, BC loop in green, and FG loop in blue. The invisible C’D loop is depicted as dashed lines in purple.

(B) Superimposition of apo-PD-1 (gray) and the PD-1s extracted from the complex structures of PD-1/PD-L1 (orange) (PDB code: 4ZQK), PD-1/nivolumab (yellow) (PDB code: 5WT9), PD-1/pembrolizumab (light pink) (PDB code: 5JXE), GY-5/PD-1 (green), and GY-14/PD-1 (cyan). The loops that contributed major binding to the mAbs are highlighted in dashed circles.

(C) Comparison of the FG loop of the PD-1s from the complex structures. The FG loop of PD-1 shifted 10.3 Å upon the binding to nivolumab or GY-5.

Partial contacts to the N-terminal loop of PD-1 by GY-5 could be observed, and the influence of this N-terminal loop to the binding affinity of GY-5 was evaluated with an N-terminally truncated PD-1 protein (N32-R147) using SPR (Figure S5). No substantial difference in the binding affinity of GY-5 for the N-terminally truncated PD-1 (N32-R147) (KD = 11.5 nM) or PD-1 with the N-terminal loop (L25-R147) (KD = 3.52 nM) was observed. This finding suggests that although the N-terminal loop of PD-1 provides contacts with GY-5, the binding affinity of GY-5 to PD-1 is not affected by the N-terminal loop, which is distinct from the N-terminal loop-dependent binding of nivolumab (Tan et al., 2017).

We further compared the FG loops derived from multiple complex structures to investigate the conformational changes of the FG loop upon binding to different counterparts (Figure 7C). The FG loop of PD-1 contributed partial interaction to the binding of nivolumab, whereas no direct interaction was involved when PD-1 bound to pembrolizumab. The invisibility of the FG loop upon binding to pembrolizumab suggests that this loop has high conformational flexibility. The FG loop of PD-1 upon binding to GY-14 exhibits similar conformation to PD-1 bound to PD-L1 and that of the apo-PD-1. However, a substantial shift of 10.3 Å was observed in the FG loop upon binding to nivolumab or GY-5. Taken together, the FG loop of PD-1 adopts substantially different conformations to bind to varied counterparts, either serving as a major target for mAb binding (e.g., GY-5 and GY-14) or providing partial contacts with mAbs (e.g., nivolumab) or its ligand.

Discussion

In the present study, we report two PD-1-specific mAbs, GY-5 and GY-14, with potent tumor suppressive efficacy. PD-1 contains a front β-sheet face comprising the CC’FG strands and a back β-sheet face comprising the AA’BDE strands. The binding of PD-1 to PD-L1 involves the front β-sheet faces of both molecules, with additional contributions of the FG loop. Structural analysis revealed that both GY-5 and GY-14 mainly bind to the FG loop of PD-1. Together with the previously reported complex structures of nivolumab/PD-1 and pembrolizumab/PD-1, the binding of the four mAbs targeting PD-1 exhibits “loop-dominated” binding characteristics, which is different from PD-L1 binding. The highly flexible loops of PD-1 adopt different conformations when binding to these mAbs. The binding of nivolumab mainly involves the N-terminal loop, whereas pembrolizumab mainly binds to the C’D loop. Both the N-terminal and the C’D loops of PD-1 are away from the binding interface of PD-1/PD-L1. In contrast, both GY-5 and GY-14 mainly bind to the FG loop of PD-1, which shows varied conformations upon binding to different mAbs or its ligand, PD-L1. In contrast to the loops targeted by nivolumab or pembrolizumab, which are away from the PD-1/PD-L1 binding interface, the FG loop of PD-1 plays critical roles in the interaction with PD-L1 (Lin et al., 2008, Lazar-Molnar et al., 2008). Therefore the blocking of the PD-1/PD-L1 interaction by GY-5 and GY-14 relies on occupancy of the FG loop of PD-1 with overwhelming binding affinity compared with the PD-1/PD-L1 interaction. Although the N-terminal loop of PD-1 exhibits partial contacts with GY-5, the overall binding affinity of GY-5 to PD-1 is not affected, as determined through SPR assays with an N-terminally loop-truncated PD-1. The FG loop-dominated binding of GY-5 and GY-14 indicates a completely different binding mode from that of nivolumab and pembrolizumab, suggesting that GY-5 and GY-14 are potential therapeutic mAbs in addition to the two commercially available mAbs. Taken together, the FG loop of PD-1 may serve as an important region for the development of PD-1-targeting biologics or small chemical molecules.

Li et al. recently reported that glycosylation of PD-L1 in tumor cells is essential for interaction with PD-1, and an mAb targeting glycosylated PD-L1 would promote PD-L1 internalization and degradation (Li et al., 2018). A wide range of alterations in the glycoproteins on tumor cells can occur, which may correlate with the development and progression of multiple tumors (Pinho and Reis, 2015). Disordered glycosylation modification of the proteins in tumor cells was usually correlated with dysregulated protein folding, trafficking, and protein-protein interactions (Xu and Ng, 2015). In addition to the expression in T cells, PD-1 is also expressed in tumor cells and macrophages, and the expression in these cells may also correlate with the treatment efficacy of PD-1/PD-L1-blocking mAb-based tumor ICT (Kleffel et al., 2015, Gordon et al., 2017, Huang et al., 2009). Therefore investigations of PD-1 glycosylation and the glycan dependency of the interaction of PD-1 with these mAbs would be valuable for our understanding of immune checkpoint blockade therapy (Tan et al., 2017).

PD-1 has four potential N-linked glycosylation sites, and mutational analysis indicates glycosylation modifications at each of these sites (Tan et al., 2017). In the present study, glycosylation modifications were structurally visible at three of the four glycosylation sites: N49, N58, and N116. Considering the flexibility of the glycans, the possibility of glycosylation at N74 cannot be excluded, and more complicated glycan structures may exist at the other three N-linked glycosylation sites. Structural analysis revealed that all these glycosylation sites are located away from PD-1/PD-L1-binding face, suggesting that the glycosylation modifications have no direct influence on the PD-1/PD-L1 interaction. Previous studies reveal that the binding of nivolumab is independent of PD-1 glycosylation, as determined through the analysis of binding affinity to glycosylation-site-mutated PD-1s or PD-1 proteins obtained from different expression systems that enable different levels of glycosylation of PD-1 (Tan et al., 2017). Although pembrolizumab binds to different regions compared with nivolumab, the structural analysis reveals that the N-linked glycosylation sites are also located away from the pembrolizumab/PD-1 interface (Na et al., 2017). The structural analysis and binding assays of GY-5 and GY-14 to PD-1 proteins obtained from different expression systems demonstrated that the binding of GY-5 and GY-14 is also independent of PD-1 glycosylation. Therefore the FG loop-targeting mAbs may be promising therapeutics independent of dysregulated glycosylation modifications of PD-1 in both immune and tumor cells.

Taken together, we identified two PD-1/PD-L1 blocking mAbs targeting PD-1 with tumor suppressive efficacy. These two mAbs mainly bind to the FG loop of PD-1, which is distinct from the other structurally clear anti-PD-1 mAbs. Glycosylation modifications could be observed at three N-linked glycosylation sites, but the glycosylation modifications were not involved in the binding of these two mAbs to PD-1. These findings have broadened our understanding of mAb-based ICT and will aid in the future development of therapeutics by targeting PD-1.

Limitations of the Study

We do not know the functional anti-tumor advantage of the two mAbs identified in the present study over commercially available PD-1-targeting mAbs. Moreover, there are more PD-1-specific mAbs under clinical investigations and the binding regions of these mAbs on PD-1 may be different from the currently known binding epitopes. Especially, the N58 glycosylation is near the PD-1/PD-L1-binding interface, and future studies should take considerations whether this glycosylation would affect the binding of the mAb.

Methods

All methods can be found in the accompanying Transparent Methods supplemental file.