Anti-tumor effect of CTLA-4 antibody is independent of checkpoint blockade

Antibodies targeting CTLA-4 are emerging as an important class of cancer therapeutics. It is assumed that these antibodies cause tumor rejection by blocking negative signaling from the CTLA-4-B7 interactions to enhance the priming of naïve T cells in lymphoid organs. However, recent ndings have shown that the effectiveness of CTLA-4 antibody critically depends on the Fc domain and the host Fc receptors. It remains unclear if the blocking function of CTLA-4 antibody is required for its anti-tumor activity. To address this, here we have selected a non-blocking anti-CTLA-4 antibody (D138) and assessed its binding property and antitumor activity in comparison with the therapeutic CTLA-4 antibody ipilimumab. Crystal structures of CTLA-4 complexed with these antibodies show that D138 binds to a distinctly different site to that of ipilimumab on the CTLA-4 surface. D138 binding did not block the association of cells expressing CTLA-4 and B7 whereas ipilimumab did. Subsequent antitumor assay revealed that D138 was similarly effective as ipilimumab in inhibiting tumor growth in mice. This antitumor activity required Fc function for ecacy and was correlated with selective reduction of intratumor regulatory T (T reg ) cells, resulting in a signicant increase in the ratio of CD8 + over T reg cells. Overall these data clearly demonstrate that blocking CTLA-4-B7 interaction is not required for CTLA-4 antibody mediated antitumor activity, opening prospects of developing non-blocking CTLA-4 antibodies or simple binders towards other T reg surface markers for T reg -targeted immunotherapy. cytometry intratumor

these antibodies is associated with selective depletion of T reg cells within tumor lesions through Fcγ receptor-expressing leukocytes [14][15][16] . Also, in patients with advanced melanoma, response to ipilimumab correlates with the Fc receptor high a nity polymorphism (CD16a-V158F) 17 . Furthermore, it was reported that a blocking nanobody H11 without Fc has very limited antitumor activity in mice 18 . Therefore, it remains unclear if the blocking function of CTLA-4 antibodies is required for their anti-tumor activity or if a non-blocking CTLA-4 antibody would be active in inhibiting tumor growth. To address this, we have solved the crystal structure of a non-blocking CTLA-4 nanobody (D138) complexed with CTLA-4 and the crystal structure ipilimumab complexed with CTLA-4, and compared their binding properties and antitumor activities. This demonstrates that D138, a non-blocker not interfering with the hemostasis of CTLA-4 and its interactions with B7 on cell surface, is similarly effective as ipilimumab in inhibiting tumor growth in mice. These ndings open opportunities in selecting non-blocking CTLA-4 antibodies for cancer immunotherapy.

Selection of a novel non-blocking CTLA-4 antibody
In order to obtain high a nity CTLA-4 antibody, camels were immunized with human CTLA-4 and a phage library derived from peripheral blood mononuclear cells (PBMC) was generated and screened for anti-CTLA-4 nanobody or domain antibodies (dAb). One of these antibodies binds human CTLA-4 with a modest dissociation constant (Kd) of 25.3 nM (Fig. 1a, 1b) when measured by Bio-layer interferometry (BLI) and shows little interference on the binding between CTLA-4 and B7 during screening. This nanobody fused with human IgG1 Fc as seen in ipilimumab, was termed D138 and selected for further characterization (Fig. 1a). In a competitive binding ELISA assay, CTLA-4 was coated on the plates and a mixture of B7-1 or B7-2 with different concentrations of antibody was then added into the wells. The binding of B7 on the plates was subsequently quanti ed and plotted as seen in Fig. 1c. This showed ipilimumab blocked the binding of B7-1 with an IC 50 of 24.6 nM and of B7-2 with an IC 50 of 10.4 nM, while D138 had little blocking effect.
We further compared the binding properties of these antibodies by BLI assay (Fig. 1d). CTLA-4 was anchored on a sensor and loaded with ipilimumab or D138 for 200 seconds (stage I) and then washed with buffer for 60 seconds (stage II). These two stages represented the typical antibody binding curves of association and dissociation (Fig. 1d). Subsequently the sensors with preformed CTLA-4-antibody complexes were dipped into the antibody solution again to allow further association or into the antibody solution of same concentrations but with added B7-1 to check if B7-1 would bind (stage III). As expected, there was slight increase in the signal from sensor when loaded again with corresponding antibodies indicating further antibody binding onto the sensors (Fig. 1d, black curves). However, in the presence of B7-1 there was a signi cant association signal from the sensor with preformed D138/CTLA-4 complexes (Fig. 1d, blue curve) indicating signi cant binding of B7-1 onto CTLA-4 in the sensor surface. This result demonstrates that B7-1 could readily bind to the preformed CTLA-4-D138 complex, and binding of D138 would not block CTLA-4 from binding B7-1. This is different to ipillimumab where little binding of B7-1 towards preformed CTLA-4-ipilimumab complex on the sensor was observed 19 . This suggests that ipilimumab is a blocker although there were some doubts on this 20 and that D138 is a non-blocker.
Crystal structure of CTLA-4 complexed with D138-dAb or ipilimumab To elucidate the structural mechanism underlying the binding interaction between CTLA-4 and D138, we prepared recombinant dimeric CTLA-4 complexed with D138-dAb, which could be separated from CTLA-4 monomer complexed with D138-dAb by ion-exchange chromatography, and solved its crystal structure at a resolution of 3.25Å (Fig. 2a, Supplementary Table 1). The structure showed a tetramer containing a CTLA-4 homo dimer and two molecules of D138-dAb (Fig. 2b). The overall structure of CTLA-4 dimer in the complex is similar to its apo form (PDB 3OSK) 21 with a RMSD of 2.2 Å indicating D138 binding induced little conformational changes in CTLA-4. D138-dAb has a typical nanobody conformation with a long complementarity determining region (CDR3) loop of 19 residues ( Supplementary Fig. 1a, b). It binds to a concave surface area formed by the connecting loops of the front and back beta-sheets of the CTLA-4 molecule with a total buried surface area of 760 Å 2 (Fig. 2c, d, e). The binding interactions involve residues from all three CDR loops of D138. Notably, Val54 of CDR2 loop in D138 forms key hydrophobic interactions with the shallow pocket in CTLA-4 involving sidechains of Thr47, Val49, Ile67 and Tyr92 (Fig.   2d). Replacement of Thr47 and Ile67 with Ala in CTLA-4 resulted 14 and 26-fold decrease in binding a nity respectively, while replacement of Val49 and Tyr92 with Ala resulted in a complete loss of the binding a nity towards D138. There are also eleven hydrogen bonds or salt bridges formed between D138-dAb and CTLA-4 (Supplementary Table 2). Glu48 on the tip of C' strand of CTLA-4 forms a hydrogen bond with the indole ring nitrogen atom of Trp105 in the CDR3 of D138-dAb (Fig. 2c,   Supplementary Fig. 1c). Asp64 and Asp65 in C''D loop of CTLA-4 form salt bridges with Arg47 and Arg108 in CDR3 (Fig. 2c). Also Asp88 of DE loop in CTLA-4 formed a hydrogen bond with the main chain nitrogen atom of Gly55 in D138-dAb CDR2 region. Replacement of these three residues with alanine respectively dramatically decreased or abolished the binding between D138 and CTLA-4 ( Fig. 2d, Supplementary Table 3).
As the binding site of B7 on CTLA-4 is centered around the FG loop near the lower part of the front betasheet (Fig. 2e), there is no overlap between the binding surfaces of B7 and D138. Superimposition of the structures of their complexes on CTLA-4 also showed that there would be little steric clash between D138 and B7 within the CTLA-4-B7-D138 ternary complex (Fig. 2f, Supplementary Fig. 1d). This is consistent with our observation above that D138-dAb is a non-blocker and it can bind to preformed CTLA-4-B7 complex (Fig. 1d).
The crystal structure of ipilimumab Fab or ScFv domain complexed with CTLA-4 has been solved previously 10,11 , however there were some concerns if ipilimumab is really a blocker of CTLA-4-B7 interaction 20 . Here we crystallized ipilimumab-Fab/CTLA-4 complex under a different condition with a different space group and solved its structure at higher resolution (2.5Å). There were two copies of ipilimumab Fab fragment/CTLA-4 complexes within the asymmetric unit ( Supplementary Fig. 1e). The overall structure of the complex resembles the previously published structures with a buried CTLA-4 surface area of 870 Å 2 within the complex regardless the differences in crystallization conditions and resolution of the structures 10,11 . The ipilimumab binding site on CTLA-4 is located close to the lower part of the front beta-sheet and overlaps with that of B7 binding site (Fig. 2e). When superposed together, there is a clear steric clash between ipilimumab and B7 (Fig. 2f). Therefore, ipilimumab is undoubtedly a blocker of CTLA-4-B7 interactions.  (Fig. 3a). When the cells were mixed at a ratio of 1:1 directly, they readily aggregated to form clumps upon mixing due to binding interactions between cell surface expressed CTLA-4 and B7-1 as examined by uorescent microscope. Pretreatment of CTLA-4-RFP expressing cells with 100 nM ipilimumab completely blocked the aggregation formation (Fig. 3b). In contrast, pretreatment with D138 or control human IgG had no effect on the cell aggregation. This con rms that ipilimumab is a blocker of CTLA-4-B7 interactions on cell surfaces while D138 is not, which is consistent with a model of their complex on the cell surface ( Supplementary Fig. 1d).
The dose-dependent blocking effect of ipilimumab and D138 was further assessed by ow cytometry analysis. Cells expressing CTLA-4-RFP (2´10 5 cells) were pretreated with different concentrations (0-200 nM) of antibodies for 5min, then mixed with cells expressing B7-1-GFP (1´10 6 cells) and subjected to ow cytometry analysis for the tethered cells with both GFP and RFP signals. The percentage of the GFP + RFP + cells over total RFP + cells was plotted against the concentrations of the antibodies. As shown in Fig. 3c, when cells expressing B7-1-GFP were mixed with cells expressing CTLA-4-RFP, about 80% of CTLA-4 expressing cells became tethered with B7-1 expressing cells and showed strong signal for both GFP and RFP. Addition of 100 nM ipilimumab largely blocked the formation of tethered cells while no effect was observed with the addition of D138 or control IgG. The IC 50 value of ipilimumab mediated blockade of cell-cell interactions was 13.8nM (Fig. 3d) as calculated from the dose-dependent curve, which is largely consistent with IC 50 value derived from the ELISA assay (Fig. 1c). Altogether these data further con rm that D138 is a simple CTLA-4 binder and that it does not block CTLA-4-B7 binding on cell surface.
Since wild type CTLA-4 when expressed on the cell surface readily recycles between surface and endosomes 22,23 , we tested if ipilimumab or D138 binding would perturb this process. Cells stably expressing human CTLA-4 were treated with different concentrations of antibodies respectively and then checked for the amount of surface expressed CTLA-4 by ow cytometry. Interestingly we found that ipilimumab binding reduced the surface expressed CTLA-4 while D138 had no effect (Fig. 3e). This clearly demonstrates that D138 binding does not perturb the normal hemostasis of CTLA-4 on cell surface while ipilimumab does. As the binding of both antibodies towards CTLA-4 is not pH sensitive (data not shown) with their complexes unlikely affected by pH changes during endocytosis process, the different effect of D138 and ipilimumab on cell surface CTLA-4 likely arises from the difference in the binding epitopes of these two antibodies on CTLA-4.

Biological activity of D138
It is known that binding of CTLA-4 to the costimulatory molecules inhibits T-cell activation by reducing interleukin-2 (IL-2) production and IL-2 receptor expression, and by arresting T cells at the G1 phase of the cell cycle, thus an immune checkpoint blocker such as ipilimumab or tremelimumab would block CTLA-4-B7 inhibitory pathway and enhance IL-2 production 9,24 . Here we tested if D138 would have a similar effect. PMBCs from healthy donors were mixed with ipilimumab or D138 of different concentrations for one hour and then stimulated with Staphylococcal Enterotoxin B (SEB) super antigen. The concentrations of the cytokines such as IL-2, IFN-γ and TNF-α in the supernatant were measured after 5 days of culture. As expected, ipilimumab induced secretion of IL-2 dose-dependently with signi cant IL-2 production detected at the concentration of 200 nM. However, there was very limited secretion of IL-2 from D138 treated samples (Fig. 4a). There were no signi cant secretions of IFN-γ or TNF-α when treated with either ipilimumab or D138 consistent with previous studies of ipilimumab (Supplementary Fig. 2a and 2b).
As both D138 and ipilimumab could bind CTLA-4 and have the same Fc domain, we then assessed their antibody-dependent cellular cytotoxicity (ADCC) effects on the viability of co-cultured T reg cells from PBMC. This showed that both antibodies could induce T reg cell lysis leading to increased lactate dehydrogenase (LDH) release (Fig. 4b). It appeared that ipilimumab has stronger ADCC activity than D138 (Fig. 4b). When the ADCC activities of these antibodies were evaluated on engineered effector cells carrying a reporter gene with binding sites for the principal transcription factors involved in FcγRIIIA signal transduction, D138 and ipilimumab showed similar activity ( Supplementary Fig. 2c). Overall, these results showed ipilimumab could induce IL2 release from immune cells and induced ADCC activity while D138 could not acviated IL-2 release from immune cells, but could induce similar ADCC activity as ipilimumab.
Antitumor effect of D138 and ipilimumab is associated with T reg depletion within tumor lesions Subsequently, the in vivo antitumor activities of D138 and ipilimumab were assessed against MC38 colon adenocarcinoma tumor in human CTLA-4 knock-in mouse model, with treatment initiated 9 days after tumor implantation. Tumor volumes and diameters were recorded every three days. At a relatively low dose of 0.8 mg/kg, ipilimumab showed a signi cant tumor inhibition activity when compared with the control group treated with human IgG (Fig. 5a, b). Most remarkably, similar therapeutic effects of D138 was also observed with D138 treatment (0.5 mg/kg, same numbers of molecules as ipilimumab) ( Fig. 5a, b). However, the Fc mutants of both antibodies (ipilimumab-AG or D138-AG), where double mutations (D265A, P331G) in Fc (FcAG) largely abolished the binding of Fc receptors 25 , showed little antitumor activities. This indicates that the blocking function is not required for the anti-tumor activity of CTLA-4 antibody, while the Fc receptor binding ability is critical for e cient tumor inhibition.
It has been well documented that the anti-tumor activities of CTLA-4 antibodies in mice is often associated with depletion of T reg cells within the tumor micro environment [14][15][16] . Here we tested if a similar effect could be observed with D138. Although T reg depletion could occur within days after antibody injection, in order to correlate the tumor growth inhibition with the level of T reg within tumor lesions in each mouse, mice (n=10 each group) with implanted tumors were treated with four injections of antibodies or control IgG and tumor growth was followed ( Supplementary Fig. 3a). Whereas tumors grew progressively in the control IgG-treated mice and very limited growth inhibition was observed in the D138-AG and ipilimumab-AG groups, signi cant tumor growth retardation was seen with both D138 and ipilimumab-treated groups. The tumors retrieved from these two groups were smaller than those of other groups ( Supplementary Fig. 3b). These results are consistent with the anti-tumor effect seen in Fig. 5a and Fig. 5b.
The tumors were then digested and T cell populations were analyzed by ow cytometry using various T cells markers (Fig. 5c-5i and Supplementary Fig. 3). We rstly con rmed that high level of CTLA-4 expression is associated with Foxp3 + but not Foxp3cells in the CD4 + subset ( Supplementary Fig. 3c).
There were no changes in the CD4 + Foxp3cell population amongst all the treated groups ( Supplementary   Fig. 3e), but there was a signi cant decrease in the CD4 + Foxp3 + subpopulation (termed T reg cells here) in tumors treated by ipilimumab or D138 as re ected in the absolute numbers of these cells (Fig. 5d) or the percentage of T reg /T cells ( Supplementary Fig. 3d) or T reg /CD4 + cells (Fig. 5e). This was consistent with the decreased MFI of OX40 on T reg cell surface in these two groups ( Supplementary Fig. 3g). In contrast, there was a slight increase in the total T cell numbers ( Supplementary Fig. 3f), CD8 + cell numbers (Fig.   5f) and in the percentage of CD8 + /T cells (Fig. 5g) in the tumors treated with ipilimumab and D138 antibodies. This cumulatively led to a near 10-fold increase in the ratio of CD8 + over T reg cells (Fig. 5h).
To further correlate the changes in the ratio of CD8 + /T reg with tumor growth in each tumor, the ratios were plotted against the changes in size as shown in (Fig. 5i). This clearly demonstrated little changes in the ratio of CD8 + /T reg cells from the control IgG or the D138-AG treated tumors, but signi cant changes in the ipilimumab and D138 treated tumors. Interestingly, there was also a small decrease in total T reg cell numbers in ipilimumab-AG treated tumors (Fig. 5c) with some of these tumors having slightly increased CD8 + /T reg ratios (Fig. 5i), which might re ect certain role of the blocking function of ipilimumab-AG.
Nevertheless, these results show that antitumor effect of D138 and ipilimumab is associated with depletion of T reg cells within the tumor microenvironment and this selective depletion mainly depends on the function of Fc.

Discussion
It is generally accepted that anti-CTLA-4 monoclonal antibodies induce tumor rejection by blocking negative signaling from the CTLA-4-B7 interaction 7,8 , thus these antibodies are often called immunecheckpoint blockers. Most, if not all, of these antibodies in clinical trials are selected based on their blocking abilities and the Fc mediated functions were not considered to be critical to their antitumor activities 26 . However, many recent studies have shown that the anti-tumor activities of these antibodies, including ipilimumab, critically depend on the presence of Fc domain and Fc receptors on the cell surface [13][14][15]17,18 . It was also suggested by Du et al that ipilimumab functioned through depletion intratumor T reg instead of blocking CTLA-4-B7 pathway in mouse, but surprisingly they found that ipilimumab failed to block CTLA-4-B7 interactions 20 . Nevertheless, it appeared that the blocking function of CTLA-4 antibodies played a minor role in the anti-tumor activity. To dissect the blocking and Fc mediated functions of CTLA-4 antibodies, here we selected a non-blocking CTLA-4 antibody D138 (Fig. 1) and compared its binding ability and antitumor activities with those of ipilimumab. Our crystallographic and in vitro binding assays unequivocally demonstrated that D138 is a simple binder of CTLA-4 and does not impede CTLA-4-B7 interactions on cell surface while ipilimumab is a checkpoint blocker (Fig. 3).
Remarkably, in vivo assay in human CTLA-4 knock-in mouse model showed both antibodies had similar antitumor activity which were associated with a selective depletion of intratumor T reg cells and a signi cant increase in the ratio of CD8 + /T reg cells within the tumor lesions. Furthermore, both antibodies required intact Fc for tumor inhibition activity as their mutants with impaired Fc receptor binding lost antitumor activity (Fig 5a, Supplementary Fig 3a,b). Thus, the main contributing factor of the antitumor activity from both antibodies is the Fc mediated T reg depletion with blocking function of ipilimumab not essential, which is consistent with previous observations 18,20 . Although this selective intratumor T reg depletion has been shown in various mouse models, it remains less clear if CTLA-4 antibodies such as ipilimumab function through this route in human mainly due to the di culties in obtaining large numbers of tumor samples from patients before and after treatment 17,27,28 . Nevertheless, directly targeting intratumor T reg cells represents an attractive approach for clinical cancer treatment 29 . A transcriptome analysis of human cancer specimens has revealed that tumor in ltrating T reg cells were highly suppressive and expressed other speci c signature molecules such as OX40, TIGIT, PD-1, GITR and CCR8 etc 30 . Simple binders, instead of agonists or antagonists or blockers, targeting those intratumor T reg surface markers would t this purpose in inducing intratumor T reg depletion as well as CTLA-4 blocking antibodies.
However, D138 may hold some advantages over CTLA-4 blockers such as ipilimumab for cancer treatment. CTLA-4 is constitutively expressed in T reg cells, but at any given time only about 10% of total CTLA-4 is present on the cell surface with most of CTLA-4 located in the endocytosis recycle pathway within the cell 31 . It has been reported that binding of blocking antibodies on CTLA-4 on the cell surface induces CTLA-4 internalization and the detailed molecular mechanism underlying this process is not well understood 23,32 . Here we con rmed that the cell surface binding of ipilimumab induced CTLA-4 internalization with a reduction of CTLA-4 molecules on cell surface, however, to our surprise, binding of D138 on the cell surface did not induce similar reduction of CTLA-4 (Fig. 3e). As normal function of CTLA-4 in peripheral tissues is critical for maintaining the homeostasis of the immune system, de ciency in CTLA-4 expression in human is often associated with autoimmune diseases 33,34 . Also clinical use of ipilimumab is closely related to immunotherapy-related adverse events (irAE) [35][36][37] and it was recently proposed that these side effects could arise from the disturbance in the recycling of CTLA-4 in the cell 38 . Since D138, a simple binder identi ed here with strong anti-tumor activity, would neither perturb the homeostasis of CTLA-4 on cell surface, nor block the normal interactions between CTLA-4 and B7, it may induce less side effects than the CTLA-4 blocking antibodies. Therefore, the simpler could be the better, and this opens the prospects of developing simple binders towards T reg surface markers for T reg -targeted immunotherapy. A.Z. and T.X. supervised the study and wrote the manuscript along with input from F.D. and F.L.

Cell lines and culture
The MC38, HEK293T and CHO cell lines were all purchased from American Type Culture Collection (ATCC). MC38 and HEK293T cell lines were cultured in Dulbecco's Modi ed Eagle Medium (DMEM, Gibco) supplemented with 10% Fetal Bovine Serum (FBS, Gibco) and 100 mg/ml penicillin (Sigma) and 100 mg/ml streptomycin (Sigma). CHO cell line was cultured in CD CHO Medium (Invitrogen). For protein expression, HEK293T was culture in suspension in serum free medium (Gibco). All cell lines were kept at 37°C in a 5% CO2 incubator.
C57BL/6J Ctla-4 h/h female mice (6-8 weeks old) were purchased from Shanghai Model Organisms Center and maintained in a special pathogen-free environment and in individual ventilation cages. It is considered that there are no known contaminants in the dietary materials that could in uence the tumor growth. The protocol and any amendment(s) or procedures involving the care and use of animals in this study were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of Shanghai Model Organisms Center prior to conduct. During the study, the care and use of animals were conducted in accordance with the regulations of the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC).

Generation of camel nanobodies against CTLA-4
Anti-CTLA-4 nanobodies were generated and prepared using similar procedure as previously described 39 . Camels (Camelus bactrianus) were immunized with injections of human CTLA-4 Fc fusion protein.
Peripheral blood lymphocytes (100 ml) were isolated one week after the last immunization. The DNA sequences corresponding to the variable domains of the heavy-chain antibodies were ampli ed with speci c primers and used to create a nanobody phage display library. Enrichment screening against human CTLA-4 was performed in 96-well plates coated with 10 μg of protein per well. High-a nity bacteriophages were obtained after 4 rounds of screening. Ninety-six individual colonies were randomly selected and ampli ed in culture with positive colonies sequenced. The interested nanobodies such as D138-dAb were fused with human IgG1 Fc and expressed in HEK293T cells. Ipilimumab variants, CTLA-4 fused with mouse Fc or human Fc and B7-fusion proteins were similarly expressed in HEK293T cells and puri ed by a protein G column.

Binding a nity measurement of D138 with CTLA-4 variants
The measurement was performed on Octet K2 through Bio-layer interferometry (BLI). The recombinant anti-human CTLA-4 antibody D138 was diluted to 10 mg/ml and anchored on the Anti-Human Fc Capture (AHC) biosensor for 120 seconds with a threshold 2 nm thickness. Human CTLA4-mFc or CTLA-4-His variants was then serially diluted (200 nM, 100 nM, 50 nM, 25 nM, 12.5 nM), loaded for 120 seconds, dissociated for 300 seconds and regenerated for 5 seconds with 10 mM glycine-HCl (pH 1.7). Binding rate (K on ) and dissociation rate (K off ) were calculated using a simple one-to-one Languir binding model (Octet K2 Data Analysis Software Version 9.0). The equilibrium dissociation constant (Kd) was calculated as the ratio K off /K on .

Evaluation of the effect of antibodies on CTLA-4 and B7 binding
For the ELISA assay to evaluate if CTLA-4 antibodies would block the interaction between CTLA-4 and B7, plates were coated with CTLA-4-human Fc fusion protein at 3 μg/ml with 3% BSA in PBS for 2 hours at room temperature. Mixtures containing antibodies (D138 or ipilimumab) of different concentrations and 50ng/ml B7-1-or B7-2-mouse Fc fusion protein in PBS plus 0.05% Tween-20 and 1% BSA were applied to the plate and incubated for additional 2 hours at 37 °C. The bound B7-1 or B7-2 was detected with the horseradish peroxidase (HRP)-conjugated goat anti-mouse IgG1, which was subsequently measured using tetramethylbenzidine (TMB) substrate with absorbance at 450 nm. The mean value of triplicate wells (±SD) was then plotted against the concentration of antibodies with IC 50 tted (Fig. 1c).
A BLI assay was also developed to assess the effect of antibody on CTLA-4-B7 binding. Human CTLA-4 fused mouse Fc (CTLA-4-mFc) protein was diluted to 10mg/ml, anchored on the Anti-Mouse Fc Capture (AMC) biosensors for 150 seconds, loaded with ipilimumab (50 nM) or D138 (50 nM) for 200 seconds (stage I) and then washed with PBST for 60 seconds (stage II). Subsequently the biosensors were dipped into the same antibody solution or a antibody solution of same concentration with added 200 nM B7-1-Fc for 200 seconds (stage III). The biding response sensorgram were followed (Fig. 1d).

Crystallization and structural study of CTLA-4 complexed with antibodies
The Fab fragment of ipilimumab was produced in HEK293T mammalian expression system 40  Recombinant glycosylated human CTLA-4 ectodomain (amino acids 3-125) with a C-terminal His-tag and recombinant D138-dAb were co-expressed in HEK293T cells using a similar procedure as ipilimumab. The secreted protein in the medium was rst puri ed by a Ni-NTA column and the eluted samples were then further puri ed by a SP ion exchange column (Fig. 2a). This separated CTLA-4 dimer complexed with D138-dAb from its monomer-D138-dAb complex. Puri ed CTLA-4 dimer complex was concentrated tõ 20 mg/ml and the initial crystallization was screened similarly as above and crystals suitable for data collection were obtained at room temperature from 12% PEG 3350, 3% tascimate, pH 5.0.
Crystals were cryo-protected with 20% glycerol in the mother liquor and ash-cooled in liquid nitrogen. Diffraction data were indexed and processed with iMos m and scaled with Aimless from the CCP4 suite 41 . The initial phases were obtained by molecular replacement using Phaser 42 with CTLA-4 (PDB 3OSK), nanobody model (PDB 5JDS) and ipilimumab Fab (PDB 5TRU), respectively. The models were subsequently manually rebuilt using Coot 43 and re ne to good geometry using Refmac 44 . The atomic coordinates and structure factors have been deposited in the Protein Data Bank (PDB) (Supplementary Table S1). The binding interface between antibody and CTLA-4 was analyzed by PISA 45  uorescence microscope analysis (Fig. 3b) or at 1:5 for ow cytometry analysis (Fig. 3c). In the ow cytometry analysis, the tethered cells with both GFP and RFP signal were counted and calculated against cells with RFP signal (Fig. 3d). The data was imported into Prism 7 (GraphPad) and analyzed using a four-parameter logistic (4PL) nonlinear regression curve.

Flow Cytometry analysis of T cell populations
For analysis of T cell population within each tumor, C57BL/6J mice were inoculated with MC38 cells as above and treated with four injections antibodies. The tumor-bearing mice (n=10) were killed one day after the fourth immunotherapy ( Supplementary Fig 2a). Tumor samples, spleen and lymph nodes were extracted from the mice. Each tumor was cut into small pieces, digested with Collagenase IV (0.5 mg/ml,  Table 3    Data represent mean±S.E.M. and statistical analysis were performed by two-way repeated measures