Structural basis of checkpoint blockade by monoclonal antibodies in cancer immunotherapy

Cancer cells express tumour-specific antigens derived via genetic and epigenetic alterations, which may be targeted by T-cell-mediated immune responses. However, cancer cells can avoid immune surveillance by suppressing immunity through activation of specific inhibitory signalling pathways, referred to as immune checkpoints. In recent years, the blockade of checkpoint molecules such as PD-1, PD-L1 and CTLA-4, with monoclonal antibodies has enabled the development of breakthrough therapies in oncology, and four therapeutic antibodies targeting these checkpoint molecules have been approved by the FDA for the treatment of several types of cancer. Here, we report the crystal structures of checkpoint molecules in complex with the Fab fragments of therapeutic antibodies, including PD-1/pembrolizumab, PD-1/nivolumab, PD-L1/BMS-936559 and CTLA-4/tremelimumab. These complex structures elucidate the precise epitopes of the antibodies and the molecular mechanisms underlying checkpoint blockade, providing useful information for the improvement of monoclonal antibodies capable of attenuating checkpoint signalling for the treatment of cancer.

A s the immune system plays an important role in controlling cancer, utilizing the immune system to eliminate cancer holds great potential. Although various immunotherapeutic approaches have been shown to enhance the immune system's ability to modulate cancer, therapeutic antibodies that target regulatory pathways in T-cells to enhance antitumor immune responses, have attracted significant recent attention.
T-cell-mediated immune responses are triggered through the recognition of antigenic peptide/HLA complexes on the surface of antigen presenting cells (APCs) by T-cell receptors and are tightly regulated by antigen-independent co-receptor signals, either costimulatory or coinhibitory, providing the optimal balance between immune responses to antigens and maintenance of selftolerance under normal physiological conditions [1][2][3] . Costimulatory signals are required to enhance and sustain the function of T-cells, the most important of which is provided by the interaction of CD28, a co-receptor on T-cells, with its ligands B7-1 and B7-2 on APCs (refs 4,5). In contrast, the binding of the same B7 ligands to cytotoxic T lymphocyte-associated antigen 4 (CTLA-4), a CD28 homologue with 31% sequence identity, delivers coinhibitory signals for down-regulation of immune responses 6 . Programmed death-1 (PD-1) is also an antigenindependent co-receptor and plays a pivotal role in modulating immune responses 7 . The interaction of PD-1 with its ligands PD-L1 and PD-L2 on APCs induces inhibitory signals reducing T-cell activity 8,9 . Although both CTLA-4 and PD-1 are coinhibitory receptors, each plays a non-redundant role in the negative regulation of immune responses. While engagement of CTLA-4 by B7 ligands attenuates the early activation of naïve and memory T-cells, PD-1 modulates the function of T-cells later in peripheral tissues via interaction with PD-L1 and PD-L2 (ref. 10).
As cancer cells harbour genetic and epigenetic modifications, tumour-specific antigens are presented on the cancer cell surface and can be recognized by T-cells, therefore causing immune responses [11][12][13][14] . However, cancer cells can also evade immunological recognition and destruction through the activation of coinhibitory signalling by overproduction of immune checkpoint proteins such as PD-1 and CTLA-4 on immune effector cells and PD-L1 on cancer cells [15][16][17] . Furthermore, expression of PD-L1 on cancer cells can directly lead to the death of antigen-specific effector T-cells expressing PD-1 (ref. 18). In an inflamed tumour microenvironment, engagement of PD-1 or CTLA-4 can selflimit the antitumor immune responses and permit cancer cells to proliferate unrestrained. Advances in the understanding of the molecular mechanisms underlying the ability of cancer cells to suppress immune surveillance have devised strategies to overcome cancer-induced immune tolerance, thereby protecting the host from tumour progression. Blockade of the ligandreceptor interaction of these immune checkpoint molecules can directly enhance the function of T-cells, which represents a critical paradigm shift whereby checkpoint blockade aims at disinhibition of the activity of T-cells compared with the previous immuno-oncology concept, whereby cancer vaccines and cytokine therapies aimed at de novo activation of immune responses.
Monoclonal antibodies blocking immune checkpoints have demonstrated unprecedented therapeutic benefits in clinical trials and provided a major breakthrough in oncology [19][20][21][22][23] . While targeted therapies elicit transient clinical responses because of acquisition of cancer drug resistance usually occurring within months after an initial response, the clinical responses of checkpoint blockade therapies are often durable, with some patients free from cancer progression for many years 19,24,25 . The US Food and Drug Administration (FDA) has approved monoclonal antibodies that block CTLA-4 (ipilimumab), PD-1 (pembrolizumab and nivolumab), and PD-L1 (atezolizumab) for the treatment of melanoma, non-small-cell lung cancer, renal cell carcinoma, Hodgkin lymphoma and bladder cancer. As a number of clinical trials for antibodies targeting immune checkpoints are ongoing, additional approvals are expected for other antibodies and expanded indications in the near future 26 .
There have been a number of structural studies for immune checkpoint molecules. The structures of the extracellular portion of CTLA-4 revealed an unusual mode of CTLA-4 dimerization that places the B7 binding sites distal to the dimerization interface 27 . The crystal structures of CTLA-4 in complex with its B7 ligands showed the formation of a unique alternating network of bivalent CTLA-4 and B7-1/2 dimers, providing a structural basis for formation of unusually stable signalling complexes to regulate T-cell responsiveness within the immunological synapse 28,29 . The structures of murine PD-1 in complex with human PD-L1, murine PD-1 in complex with murine PD-L2, and human PD-1 in complex with human PD-L1 have established the structural foundations of the interaction of PD-1 with its ligands [30][31][32][33] . These structural studies for checkpoint molecules and their binding complexes have provided invaluable information for understanding the molecular mechanism of coinhibitory signals of T-cells through the communication of checkpoint molecules.
As the exact epitope and mechanism are crucial elements for antibody drugs, we report the crystal structures of checkpoint molecules in complex with the Fab fragments of therapeutic antibodies approved by the FDA or currently undergoing clinical trials, and reveal the binding modes of the complexes and the conformational changes induced by the antibody binding, thereby providing a structural basis of checkpoint blockade by monoclonal antibodies for the treatment of cancer.
The total buried surface areas of the PD-1/pembrolizumab and PD-1/nivolumab complexes are 2,126 Å 2 and 1,487 Å 2 , respectively, compared with 1,970 Å 2 for PD-1/PD-L1 (ref. 30). The epitopes of the two antibodies directly occupy part of the PD-L1 binding site (Fig. 2c). In addition, the binding of PD-1 with pembrolizumab or nivolumab induces optimal conformational changes in the BC loop (residues 57-63) and the FG loop (residues 127-134) of PD-1, which are incompatible with PD-L1 binding as they also interact with PD-L1 in distinct conformations 30 (Fig. 3a,b). The C 0 D loop (residues 81-90) was shown to be highly flexible in the NMR structure of free PD-1 (ref. 31), and disordered in the crystal structure of the PD-1/PD-L1 complex as a result of the lack of interaction 30,32,33 . However, most of the residues of the C 0 D loop in the PD-1/pembrolizumab complex are involved in the interactions with pemrolizumab with a clear electron density around them, implying that the C 0 D loop contributes to the binding affinity for pembrolizumab. This loop intrudes into the groove formed by the CDR loops of pembrolizumab (Fig. 3a,c). The electron density of the C 0 D loop in the complex structure of PD-1/nivolumab is also clearly shown, probably because of the crystal packing interaction despite the lack of interaction with nivolumab (Fig. 3b). The structural comparisons of the Fv regions of pembrolizumab and nivolumab before and after binding to PD-1 show little deviation in the conformation of the CDRs and minor adjustments in the side chains involved in the interaction with PD-1, implying that these antibody drugs maintain the CDR loops in productive binding conformations prior to interaction with PD-1 ( Supplementary  Fig. 1a,b). The high avidity of these two antibodies, which results from the bivalency of IgG, could also contribute to their tight binding to PD-1. Taken together, these structural features suggest that the mechanism by which the anti-PD-1 antibodies block the PD-1/PD-L1 interaction is through outcompeting PD-L1 for binding to PD-1.
Crystal structure of CTLA-4 in complex with tremelimumab Fab. Tremelimumab is a fully human IgG2 antibody that blocks CTLA-4. In 2015, tremelimumab was granted Fast Track Designation and Orphan Drug Designation by the FDA as a potential treatment for malignant mesothelioma. In addition, tremelimumab is being studied in combination with an anti-PD-L1, durvalumab, in multiple tumour types 37 . The CTLA-4/ tremelimumab Fab complex and free tremelimumab Fab were crystallized and their structures were determined and refined to a resolution of 2.0 Å and 2.3 Å, respectively (Fig. 6). The     Fig. 3). As observed in the structure of free CTLA-4 and that of CTLA-4 in complex with B7-1/2, the three consecutive proline residues of the MYPPPYY (residues 99-105) sequence of the FG loop of CTLA-4 in complex with tremelimumab adopt an unusual cis-trans-cis conformation and participate in interactions with the antibody 28,29,38 (Fig. 7a). The tremelimumab epitope partially occupies the B7-1/2 binding site of CTLA-4, with a larger interface area between CTLA-4 and tremelimumab (buried surface area of 1,802 Å 2 ) than the receptor-ligand interface (1,255 Å 2 for CTLA-4/B7-1 and 1,212 Å 2 for CTLA-4/B7-2) (Fig. 7b). Therefore, the binding of tremelimumab to CTLA-4 efficiently competes with B7-1/2 binding to CTLA-4, thereby blocking the function of CTLA-4 in cancer.

Discussion
Checkpoint-blocking antibodies have ensured significant advances in cancer therapy and provided a new weapon against cancer. The durability of this immunotherapy largely exceeds that of other forms of anticancer therapy and sometimes enables long-term remission where patients exhibit no clinical signs of cancer for many years. However, monotherapies of these drugs elicit the therapeutic benefit only in a fraction of patients treated. As a critical next step in the progress toward improving clinical responses, combination immunotherapies have been intensively investigated [39][40][41] . Combinations of anti-CTLA-4 or anti-PD-1/PD-L1 therapies with other immunotherapies, chemotherapies, targeted therapies, hormonal therapies and radiation therapies are being explored. The combination of anti-CTLA-4 and anti-PD-1 has been the most successful of these combinations tested so far, probably because CTLA-4 and PD-1 modulate the activity of T-cells at different stages of T-cell immunity 16,26 . Thus, dual blockade of CTLA-4 and PD-1 can be a reasonable and synergistic combination therapy based on their distinct mechanisms of action. In addition, structural studies of therapeutic antibodies to reveal their precise epitopes and therapeutic mechanisms can also facilitate a rational design of combination immunotherapies, as different epitopes and molecular mechanisms of therapeutic antibodies can lead to different therapeutic effects even when therapeutic antibodies target the same molecule 42 . Pembrolizumab and nivolumab partially share epitopes and three-dimensional (3D) space when binding to PD-1, implying that the mechanisms of antagonism of the two antibodies are highly similar (Supplementary Fig. 4). These results are consistent with earlier clinical data showing that the response rate for both drugs appears similar 26 ; thus, little synergistic therapeutic effect should be expected from a combination of only these two drugs. CTLA-4 exists as a constitutive homodimer generated by an intermolecular disulphide bond 43 (Fig. 8a). Although a single CTLA-4/tremelimumab Fab complex exists in an asymmetric unit, a crystallographic two-fold symmetry generates a homodimeric interface of CTLA-4, which is essentially identical a b c to that observed in the structures of free CTLA-4 or CTLA-4 in complex with B7-1, B7-2, and engineered lipocalin 2 (refs 28,29,38,44), implying that tremelimumab binding does not affect the homodimerization of CTLA-4 (Fig. 8b). In the crystal structures of CTLA-4 in complex with B7-1 or B7-2, the crystal packing generates an alternating periodic arrangement in which bivalent CTLA-4 homodimers connect bivalent B7-1/2 homodimers, providing a model describing the assembly of CTLA-4 and B7-1/2 at the interface between a T-cell and cancer cell 28,29 (Fig. 8c). In this model, the CTLA-4/B7 complex is expected to span a distance of B140 Å, which includes the stalk regions connecting these molecules to the opposing cell membranes. This distance is compatible with other ligandreceptor interactions at the central zone of the immunological synapse, suggesting that this oligomeric arrangement of CTLA-4/ B7 complex can facilitate the efficient use of low-abundance CTLA-4 at the T-cell surface, thereby promoting the inhibitory signalling of CTLA-4 and decreasing the local concentration of CD28 through simple steric crowding 28,29,45 . As tremelimumab is an IgG2 antibody, it is capable of binding two CTLA-4 molecules with high avidity through Y-shaped bivalent presentation; the dimension of the CTLA-4/tremelimumab complex is estimated to span 150-190 Å. This distance is incompatible with the oligomeric array of the CTLA-4/B7 complex (Fig. 8d). Therefore, in addition to simple antagonism of the interaction between CTLA-4 and B7-1/2, tremelimumab binding could prevent or disrupt the uniquely organized assembly of the CTLA-4/B7 complex within the immunological synapse. Despite the clinical success of checkpoint-blocking antibodies, biologics such as therapeutic antibodies have several shortcomings, including the high cost, limited half-life and immunogenicity caused by multiple dosages. As an alternative treatment strategy to overcome the drawbacks of passive antibody-based therapy, several small-molecule immunomodulators targeting the PD-1/PD-L1 axis have been reported [46][47][48][49][50] . While most of them are based on peptidomimetics, a recent structural study with the first nonpeptidic chemical inhibitor of the PD-1/PD-L1 interaction revealed that this compound binds to PD-L1 and induces dimerization of PD-L1, thus occluding the PD-1 binding site of PD-L1 (ref. 51). Interestingly, the complex structure of CTLA-4/tremelimumab Fab provided insight for the design of a small-molecule modulator that interferes with the CTLA-4/B7 signal pathway. Nine of ten residues in the HCDR3 loop (residues 101-110 of the heavy chain) of tremelimumab are involved in the interaction with CTLA-4 (Fig. 7a) T31  T31   Y32  Y32  I52  I52   I54  I54   F55  F55   K57  K57   H59  H59 K99 K99 S104 S104 G105 G105 S106 S106 P107 F108 F108 P107 Figure  the HCDR3 loop overlaps part of the binding site of B7-1/2 on the surface of CTLA-4 (Fig. 9a). Additionally, heavy R101 and heavy Y110 are located closely enough to be connected directly by a peptide bond (Fig. 9b). This extensive interaction and the structural features of the HCDR3 loop in the CTLA-4/ tremelimumab complex suggest that cyclic peptides based on the HCDR3 sequence may also prevent binding of B7-1/2 to CTLA-4 (Fig. 9c). As the HCDR3 loop conformation in the complex is essentially identical to that of tremelimumab Fab alone, other CDRs and framework regions of tremelimumab likely contribute to the conformation of HCDR3, implying that antibody context is important for the proper display of the tremelimumab HCDR3 (Supplementary Fig. 1c). If this is indeed the case, a cyclic peptide approach would not be so promising because of its intrinsic conformational flexibility. Nonetheless, a cyclic peptide or peptidomimetic approach based on the HCDR3 sequence in combination with the power of phage-display library technology may facilitate the discovery of small-molecule modulators of the CTLA-4/B7 interaction, providing an alternative treatment strategy to overcome the drawbacks of antibody-based drugs. In summary, the elucidation of the crystal structures of the checkpoint molecules PD-1, PD-L1 and CTLA-4 in complex with the Fab fragments of therapeutic antibodies sheds light on the molecular mechanisms underlying the therapeutic activity of these monoclonal antibodies. In addition, the precise epitopes revealed by the present complex structures should provide useful information for the improvement of current therapeutic antibodies and development of more effective immunotherapeutic combination strategies and small-molecule modulators that can attenuate checkpoint signalling.

Methods
Expression and purification of checkpoint proteins. Genes encoding the ectodomains of human PD-1 (aa 26-150), human PD-L1 (aa 18-134) and human CTLA-4 (aa 1-126) were subcloned into pET-21a (Novagen), respectively. A Cys to Ser mutation was introduced at position 93 of PD-1 to aid expression and folding. All proteins were expressed in E. coli BL21(DE3) as inclusion bodies. The cells were  Figure Table 1.
Data availability. The atomic coordinates and structure factors have been deposited in the Protein Data Bank under accession codes 5GGQ, 5GGR, 5GGS, 5GGT, 5GGU and 5GGV. All additional experimental data are available from the corresponding author on request.