Structural basis for cancer immunotherapy by thefirst-in-class checkpoint inhibitor ipilimumabUdupi A. Ramagopala,1,2, Weifeng Liua,b,1, Sarah C. Garrett-Thomsona,1, Jeffrey B. Bonannoa, Qingrong Yanb,3,Mohan Srinivasanc, Susan C. Wongc, Alasdair Bellc,4, Shilpa Mankikarc, Vangipuram S. Ranganc, Shrikant Deshpandec,Alan J. Kormanc,5, and Steven C. Almoa,5

aDepartment of Biochemistry, Albert Einstein College of Medicine, Bronx, NY 10461; bDepartment of Microbiology and Immunology, Albert Einstein Collegeof Medicine, Bronx, NY 10461; and cBiologics Discovery California, Bristol–Myers Squibb, Redwood City, CA 94063

Edited by James P. Allison, MD Anderson Cancer Center, University of Texas, Houston, TX, and approved April 6, 2017 (received for review November 1, 2016)

Rational modulation of the immune response with biologics repre-sents one of the most promising and active areas for the realizationof new therapeutic strategies. In particular, the use of functionblocking monoclonal antibodies targeting checkpoint inhibitors suchas CTLA-4 and PD-1 have proven to be highly effective for thesystemic activation of the human immune system to treat a widerange of cancers. Ipilimumab is a fully human antibody targetingCTLA-4 that received FDA approval for the treatment of metastaticmelanoma in 2011. Ipilimumab is the first-in-class immunotherapeuticfor blockade of CTLA-4 and significantly benefits overall survival ofpatients with metastatic melanoma. Understanding the chemical andphysical determinants recognized by these mAbs provides directinsight into the mechanisms of pathway blockade, the organizationof the antigen–antibody complexes at the cell surface, and opportu-nities to further engineer affinity and selectivity. Here, we report the3.0 Å resolution X-ray crystal structure of the complex formed byipilimumab with its human CTLA-4 target. This structure reveals thatipilimumab contacts the front β-sheet of CTLA-4 and intersects withthe CTLA-4:Β7 recognition surface, indicating that direct steric overlapbetween ipilimumab and the B7 ligands is a major mechanistic con-tributor to ipilimumab function. The crystallographically observedbinding interface was confirmed by a comprehensive cell-based bind-ing assay against a library of CTLA-4 mutants and by direct biochem-ical approaches. This structure also highlights determinants responsiblefor the selectivity exhibited by ipilimumab toward CTLA-4 relative tothe homologous and functionally related CD28.

immunotherapy | X-ray crystallography | CTLA-4 | ipilimumab | cancer

Activation of the immune system to target and eliminate ma-lignancies is recognized as one of the most promising directionsfor cancer therapy (1–4). Two broad strategies for immunotherapymay be envisaged: inhibition of negative regulators of immune re-sponsiveness (collectively known as checkpoint blockade) (2, 5–8)and activation of costimulatory pathways (8). A powerful example isprovided by antibodies targeting cytotoxic T lymphocyte-associatedantigen 4 (CTLA-4), a T-cell surface molecule, which like the ho-mologous CD28 (∼30% sequence identity) binds the B7-1 and B7-2ligands (9). While CD28 is constitutively expressed and is required,in conjunction with TCR engagement, for T-cell activation, CTLA-4is a negative regulator of T-cell function expressed after T-cell ac-tivation to terminate the response. Ipilimumab, a fully human an-tibody targeting CTLA-4 (marketed as Yervoy), demonstratedimproved overall survival in two phase-III clinical trials of metastaticmelanoma (10, 11) and received FDA approval for the treatment ofmetastatic melanoma in 2011. Ipilimumab is the first-in-class im-munotherapeutic for blockade of CTLA-4 and significantly benefitsoverall survival of patients with metastatic melanoma. Notably,combination therapies involving ipilimumab and other immuno-modulators/checkpoint antibodies, such as those targeting PD-1 ortremelimumab (anti–CTLA-4) and anti–PD-L1, can result in en-hanced activity (12–15).Multiple mechanisms have been described for CTLA-4 function.

These include negative signals from intrinsic effects of CTLA-4

expressed on activated T effector cells as a result of ligand binding.Extrinsic effects of CTLA-4 are the consequence of the ability ofCTLA-4 expressed on Tregs to remove B7 ligands from the surfaceof dendritic cells or antigen presenting cells, resulting in significantlyreduced suppression (16). This regulatory mechanism of trans-endocytosis may also be operative in activated T effector cells (17,18). CTLA-4–B7 interactions also have intrinsic effects on Treg,dampening their proliferation or activation (19).Antibodies to CTLA-4 also operate through multiple mecha-

nisms. Recent studies in murine models suggest that antibodiestargeting CTLA-4 delete intratumoral Treg cells through an Fcγreceptor (FcγR)-dependent process (20, 21). Although Treg de-pletion does not require ligand blocking, ample evidence indicatesthat inhibition of ligand binding to CTLA-4 is an important factorcontributing to the antitumor activity of anti–CTLA-4 antibodies inmurine models as well as in humans. These data include thedemonstration in murine models that targeting of the effector T-cellcompartment contributes to the antitumor activity of anti–CTLA-4,whereas exclusive targeting of the Treg cell compartment failedto elicit tumor protection—thus highlighting the importance of

Significance

Biologics represent a major class of therapeutics for the treat-ment of malignancies, autoimmune diseases, and infectiousdiseases. Ipilimumab is the first-in-class immunotherapeutic forblockade of CTLA-4 and significantly benefits overall survival ofpatients with metastatic melanoma. The X-ray crystal structureof the ipilimumab:CTLA-4 complex defines the atomic interac-tions responsible for affinity and selectivity and demonstratesthat the therapeutic action of ipilimumab is due to direct stericcompetition with the B7 ligands for binding to CTLA-4.

Author contributions: U.A.R., W.L., S.C.G.-T., J.B.B., Q.Y., M.S., S.C.W., A.B., S.M., V.S.R.,S.D., A.J.K., and S.C.A. designed research; U.A.R., W.L., S.C.G.-T., Q.Y., M.S., S.C.W., A.B.,S.M., V.S.R., and S.D. performed research; U.A.R., W.L., S.C.G.-T., J.B.B., Q.Y., M.S., S.C.W.,A.B., S.M., V.S.R., and S.D. analyzed data; and U.A.R., W.L., S.C.G.-T., M.S., A.J.K., and S.C.A.wrote the paper.

Conflict of interest statement: S.C.A., S.C.G.-T., U.A.R., W.L., and Q.Y. declare no compet-ing financial interests. A.B. is a former employee of Bristol–Myers & Squibb. A.J.K., M.S.,S.C.W., S.M., V.S.R., and S.D. are employees and stockholders of Bristol–Myers & Squibb.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.

Data deposition: The crystallography, atomic coordinates, and structure factors have beendeposited in the Protein Data Bank, www.rcsb.org/pdb/home/home.do (PDB ID code5TRU).1U.A.R., W.L., and S.C.G.-T. contributed equally to this work.2Present address: Biological Sciences Division, Poornaprajna Institute of Scientific Re-search, Bangalore 562110, India.

3Present address: Janssen Pharmaceuticals Inc., Titusville, NJ 08560.4Present address: F-Star Biotechnology Ltd., Cambridge, CB22 3AT, United Kingdom.5To whom correspondence may be addressed. Email: steve.almo@einstein.yu.edu or alan.korman@bms.com.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1617941114 PNAS | Published online May 8, 2017 | E4223–E4232

IMMUNOLO

GYAND

INFLAMMATION

PNASPL

US

Dow

nloa

ded

by g

uest

on

June

19,

202

1

http://crossmark.crossref.org/dialog/?doi=10.1073/pnas.1617941114&domain=pdfhttp://www.rcsb.org/pdb/home/home.dohttp://www.rcsb.org/pdb/explore/explore.do?structureId=5TRUmailto:steve.almo@einstein.yu.edumailto:alan.korman@bms.commailto:alan.korman@bms.comhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplementalhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplementalwww.pnas.org/cgi/doi/10.1073/pnas.1617941114

both modalities for antitumor activity (6). Blockade of CTLA-4 alsopromotes Treg suppression in vitro (6, 19).Human clinical studies with tremelimumab, which, like ipili-

mumab, blocks the interactions of CTLA-4 with its ligands,demonstrate that this antibody also has antitumor activity in ad-dition to inducing adverse events (22). Moreover, long-term sur-vival of melanoma patients treated with either ipilimumab (23) ortremelimumab have been reported (24). Notably, tremelimumabharbors the IgG2 isotype and thus cannot engage FcγRs, sup-porting a mechanism of action that relies largely on competitiveinhibition with the B7 ligands (ref. 20 and references therein). Incontrast, ipilimumab is an IgG1 that can engage human FcγR;consistent with this behavior, ipilimumab was shown to mediateantibody-dependent cell-mediated cytotoxicity (ADCC)-facilitateddepletion of Treg cells in vitro (25). However, only small numbersof patients have been analyzed for depletion of Treg at the tumorsite by ipilimumab (26, 27). Of note, FcγR polymorphisms have noimpact on the survival of ipilimumab-treated patients (28). Im-portantly, all aspects of ipilimumab-mediated CTLA-4 blockaderequire specific and high-affinity recognition of CTLA-4.CTLA-4 and CD28 are type-I integral membrane proteins

composed of a single Ig variable domain (IgV), a transmembranesegment, and a cytoplasmic tail bearing various signaling motifs.The IgV ectodomains share ∼30% sequence identity and exhibita two-layered beta sheet involving the A′GFCC′C′′ strands ofthe front sheet and the ABED strands of the back sheet. Bothmolecules exist as covalent homodimers due to a disulfide bondformed between cysteines in the stalk segments connecting theIgV and transmembrane domains. A wide range of structural andbiochemical data demonstrate that, despite this modest sequence

similarity, both CTLA-4 and CD28 use the stereochemical fea-tures of a shared proline-rich motif (MYPPPY), present in theloops joining the F and G β strands, to bind the B7-1 and B7-2ligands (9). An essential property of any CTLA-4 therapeuticantibody is the ability to specifically engage CTLA-4, whileexhibiting little or no cross-reactivity with CD28. In the case ofcancer immunotherapy, recognition of CD28 and inhibition ofligand binding could inhibit T-cell activation, which would op-pose the desired therapeutic activity. To define the nature of theinhibitory mechanism and the specificity exhibited by ipilimumabfor CTLA-4, we report the crystal structure of the complexformed by a Fab fragment of ipilimumab and human CTLA-4, aswell as complementary biochemical studies that confirm theepitope recognized by ipilimumab. This work unambiguouslydefines the ipilimumab recognition surface on CTLA-4, whichpartly overlaps the B7 ligand binding surfaces, indicating thatdirect steric competition contributes to the function of ipilimu-mab. This work also highlights the determinants responsible forthe highly selective binding to CTLA-4 and provides the foun-dation for structure-guided engineering of ipilimumab variantswith new in vitro activities (e.g., altered affinities for CTLA-4and Fc receptors) for the realization of enhanced in vivotherapeutic functions.

Results and DiscussionOverall Structure of the Human CTLA-4:Ipilimumab Complex. Thestructure of the complex between monomeric human CTLA-4(residues 1–118) and the Fab fragment derived from ipilimumabwas determined and refined to a resolution of 3.0 Å, with Rworkand Rfree of 20.3% and 26.8%, respectively (Fig. 1A, Fig. S1A, and

Fig. 1. Structure of the CTLA-4:ipilimumab complex. (A) The concave CTLA-4 front face, formed by the CC′ (beige) and FG (MYPPPY loop; dark purple) strands, isburied between CDRs of ipilimumab. CDRs, FG, and CC′ strands are colored: LCDR3 (olive) and LCDR1 (teal) make the hydrogen bonding interaction with the Gstrand. HCDR2 (pink) and HCDR1 (light gray) pack against F and C strands and participate in both hydrogen bond and hydrophobic interactions. HCDR3 (magenta)inserts into the center of the front face of CTLA-4 and exclusively participates in hydrophobic interaction. (B) Rainbow representation (N to C termini transition fromblue to red) of CTLA-4 molecules with front and back strands labeled white and black, respectively. The G-strand β bulge conserved among the antigen receptors isshown in red. (C) CTLA-4 rotated 90° around the vertical axis relative to B, highlighting the concave front surface, with approximate location of CDRs represented byarrows in respective color (as used in A). The coordinates of the ipilimumab:CTLA-4 complex have been deposited in PDB as ID code 5TRU.

E4224 | www.pnas.org/cgi/doi/10.1073/pnas.1617941114 Ramagopal et al.

Dow

nloa

ded

by g

uest

on

June

19,

202

1

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF1www.pnas.org/cgi/doi/10.1073/pnas.1617941114

Table 1). The two independent copies of the complex in theasymmetric unit exhibit similar overall organization and superposewith a Cα–root-mean-square deviation (Cα–rmsd) of 0.74 Å, cal-culated over all experimentally defined Cα atoms of CTLA-4(2–116) and the VH (2–118) and VL (2–108) domains of the ipi-limumab Fab fragment (Fig. S1; the Cα–rmsd is 1.24 Å when theconstant domains of the Fab are included). The Cα–rmsd betweenthe two molecules of CTLA-4 in the asymmetric unit is 0.65 Å.The ipilimumab Fab exhibited typical CDR structural parameters,with elbow angles for the two independent molecules in theasymmetric unit of 168° and 170°. Due to the similarity of the twocomplexes, the discussion below refers to one of the CTLA-4:Fabcomplexes in the asymmetric unit (Fig. 1 and Fig. S1B), denoted Lfor light chain, H for heavy chain, and C for CTLA-4.The binding interface is formed by residues from the C, C′, F,

and G strands (Fig. 1 B and C) of the front β-sheet of CTLA-4 andlight chain complementarity determining regions 1 and 3 (LCDR1and LCDR3) and heavy chain CDRs 1, 2, and 3 (HCDR1,HCDR2, and HCDR3) (Fig. 1). The only potential interactioninvolving LCDR2 is the 4.2 Å approach between the side-chainhydroxyl of 50Tyr and the main-chain carbonyl of 44Ser on CTLA-4[numbering is consistent with previous CTLA-4 structural reports;e.g., Protein Data Bank (PDB) ID codes 1I85 and 1I8L (29, 30)].CTLA-4 residues from the F and G strands (93Ile, 95Lys, 97Glu,106Leu, and 108Ile), the FG loop (99MYPPPY104) as well as 33Glu,35Arg, and 39Leu from the C strand and 46Val from the C′ strandform an extended interface with residues from HCDR2 (52Ser,

53Tyr, 57Asn, and 59Tyr), HCDR3 (101Trp and 102Leu), LCDR1(31Ser and 33Tyr), and LCDR3 (93Gly, 94Ser, 95Ser, and 97Trp)(Fig. 2A). The G strand from CTLA-4 is positioned along the cleftformed between the VL (LCDR1 and LCDR3) and VH (HCDR1and HCDR2) domains (Fig. 2 A and B), whereas the adjacent Fstrand packs against HCDR1 and HCDR2 (Fig. 2A). Together thesecontacts account for almost all of the hydrogen bonding interactionswithin the complex. 101Trp and 102Leu from HCDR3 contact thecenter of the concave hydrophobic patch on the front face of CTLA-4formed by 39Leu, 46Val, and 93Ile, which extends the binding in-terface toward the CC′ loop located opposite the FG loop (Fig. 2A).Overall, ∼13 hydrogen bonds and more than 90 contacts less

than 4.0 Å contribute to the CTLA-4:ipilimumab Fab interface(Table S1). The total surface area buried at the binding interface is∼1,880 Å2, with 990 Å2 contributed by the ipilimumab Fab and890 Å2 by CTLA-4, which is at the high end of observed values(i.e., 1,175–1,755 Å2) for antigen–antibody complexes (31). Theseextensive interactions are consistent with the high affinity in-teraction between CTLA-4 and the ipilimumab Fab [equilibriumdissociation constant (Kd) of 10.6 nM]. Both the heavy chain(605 Å2) and light chain (385 Å2) make significant contributionsto the binding interface. The F and G strands, which include the99MYPPPY104 loop, bury 498 Å2 of surface area, representing∼60% of the total buried surface area (890 Å2) contributed byCTLA-4. The 100YPPP103 segment buries only 56 Å2 of surfacearea upon binding ipilimumab, with the major contributors fromthe CTLA-4 99MYPPPY104 loop being Tyr-104 and Met-99, whichbury 115.0 and 85.3 Å2 of surface area, respectively. Thus, althoughthe FG loop is involved in the recognition of ipilimumab, this loopis not completely buried, as observed in the CTLA-4:B7-1 andCTLA-4:B7-2 complexes (29, 30).An important consideration is that the species crystallized in

these studies is monomeric CTLA-4, whereas the physiologicallyrelevant cell surface receptor is a disulfide-linked homodimer in-volving Cys-122, which is outside the well-ordered globular domainand not included in the construct used in this crystallographicanalysis. Structural and modeling analyses demonstrate that ipili-mumab could readily accommodate the observed interactionswithin the bona fide CTLA-4 dimer. In particular, the unique modeof CTLA-4 dimerization, involving “side-to-side” contact betweenCLTA-4 monomers, places the FG loops distal to the dimer in-terface, such that the relevant epitope in each monomer is highlysolvent accessible and appropriately positioned to recapitulate theinteractions observed between the CTLA-4 monomer and the intactipilimumab mAb (Fig. S2).

Cell-Based Evaluation of the CTLA-4:Ipilimumab Recognition Interface.To confirm the crystallographically observed binding interface, wegenerated a library of 118 dimeric CTLA-4 mutants, which weretransiently expressed in HEK293 cells and challenged with solubleipilimumab. This approach provides a gold standard for evaluatingextracellular interactions, as the members of the mutant libraryundergo all requisite co- and posttranslational modifications (e.g.,correct disulfide bond formation and glycosylation) and are pre-sented in the context of the mammalian cell surface environment(32). In the context of cell surface expression, mutation of residuesS20, R35, R40, Q76, D88, K95, E97, Y104, L106, and I108 ofhuman CTLA-4 resulted in significant loss of ipilimumab binding(Fig. 3 A and B and Fig. S3A). Of the residues identified—R35,K95, E97, Y104, L106, and I108—are observed to make directcontacts with ipilimumab in the crystal structure. Residues R40and D88 form a salt bridge that links two loops in the membraneproximal region of CTLA-4; disruption of this interaction likelycauses local or global structural perturbations, resulting in loss ofipilimumab binding. Similarly, mutation of S20 or Q76 to asparticacid also resulted in loss of ipilimumab binding, despite residingon the opposite side of CTLA-4 relative to the ligand-binding site.O- andN-linked glycosylation algorithms predict that S20 and N75

Table 1. Crystallographic data and refinement statistics

PDB ID 5TRUSource 24–ID–E, APSWavelength, Å 0.979Resolution limits, Å 34.9–3.0Space group C2221Unit cell, Å, a, b, c 95.84, 197.50, 148.12No. of observations 181,876No. of unique reflections 28,514Completeness, % 99.7 (99.9)Mean I/σI 14.6 (2.0)Rmerge on I, %* 14.3 (82.0)Rpim, %

† 5.2 (31.5)Redundancy 6.4 (6.6)BWilson, Å

2 63CC1/2 0.965 (0.881)Refinement statisticsResolution limits, Å 34.9–3.0No. of reflections, work/free 26945/1447

Protein atoms 8036Rcryst, %

‡ 0.203 (0.349)Rfree, %, 5% of data 0.268 (0.390)rmsd Bonds, Å/angles, ° 0.009/1.46Mean B, Å2§ 71Mean B, Å2, for chains L, H, l, h, C,

and c, respectively§59, 65, 80, 68, 84, 84

Fo, Fc correlation 0.93 (EDS)Ramachandran plot statistics, % 94.9 (favored), 4.5 (allowed),

and 0.6 (outlier)

Parentheses indicate statistics for the high–resolution data bin for X-rayand refinement data.*Rmerge =

Phkl

Pi jIhkl,i – j/

Phkl

Pi Ihkl,i.

†Rpim =P

hkl {1/[Nhkl – 1]}1/2 ×

Pi jIhkl,i – j/

Phkl

Pi Ihkl,i, where N is the

redundancy.‡Rcryst =

Phkl jFobshkl – Fcalchklj/

Phkl Fobshkl, where Fobs and Fcalc are ob-

served and calculated structure factors, respectively.§Average B factors were calculated with program BAVERAGE in the CCP4suite.

Ramagopal et al. PNAS | Published online May 8, 2017 | E4225

IMMUNOLO

GYAND

INFLAMMATION

PNASPL

US

Dow

nloa

ded

by g

uest

on

June

19,

202

1

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF1http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF1http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=ST1http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF2http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF3

(adjacent to Q76) are potentially glycosylated, suggesting thatmutations at residues 20 and 76 may affect installation or in-teraction with the carbohydrate moieties and potentially overallstructural stability (Fig. S3B) (33). The likelihood that mutationsat these four residues (S20, R40, N75, and D88) cause globalstructural perturbation/destabilization of CTLA-4 is supported bythe observation that they also result in the loss of binding to B7-1,B7-2, and ICOS-L, despite the observation that these residuesreside outside of the experimentally determined binding interfaces(Fig. 3 A and B and Fig. S3C).These mapping studies also highlight features of the binding

surface recognized by B7-1, B7-2, and ICOS-L. A number ofCTLA-4 mutations (R35A, R35D, K95A, K95D, E97A, E97R,Y104A, Y104D, I108A, and I108D) that strongly impair B7-1and B7-2 binding are consistent with contacts observed in theCTLA-4:B7-1 and CTLA-4:B7-2 crystal structures (29, 30) (Fig.S3C). In addition, mutations at residues E33, E48, and Y100,also present at the CTLA-4:B7-2 binding interface, result inmodest reductions in binding between CTLA-4 and B7-2. Again,mutations at residues R35, K95, E97, Y104, and I108 severelydiminished ipilimumab binding, consistent with the direct stericblockade of CTLA-4 ligand binding by ipilimumab. This conclusionis further supported by results from a competition experimentdemonstrating that ipilimumab, but not a control mAb, directlycompetes with hB7-1 for binding to beads coated with hCTLA-4(Fig. 4 A–C). In addition, mutation of a number of CTLA-4 resi-dues distal to the ligand-binding interfaces (e.g., V10, L12, S14,T69, T71, N78, Q82, G83, R85, A86, Y115, D118, and E120)exhibited impaired binding to B7-1, B7-2, and ICOS-L (Fig. S3D).Importantly, all of the CTLA-4 mutants analyzed for bindingshowed similar mCherry expression (Fig. S3E), making it unlikelythat these mutations caused severe structural instability or proteinmisfolding. It is more likely that mutations of residues on CTLA-4distal to the observed ligand binding site caused more subtle localstructural perturbations. Three of these residues (Y115, D118,and E120) contribute to the CTLA-4 dimer interface, with theremaining residues arranged in a patch that extends toward theligand binding surface (29, 30). The mechanistic underpinnings forthe binding properties of these mutants is not clear but may berelated to the previous report that CTLA-4 lacking the interchaindisulfide exists as an interconverting population of noncovalent di-mer and monomer, which fully resolves to monomer when bound

to B7-2 (34). These observations suggest a mechanism that cou-ples ligand binding with formation of the CTLA-4 dimer interface.The biological consequences of this behavior will require furtherevaluation.

Biochemical Confirmation. To further evaluate the contributions tothe binding interface, a selected subset of CTLA-4 mutants werepurified and their interactions with ipilimumab Fab were evaluatedby size exclusion chromatography (SEC) and native gel analysis. Inparticular, 95Lys, 99Met, and 104Tyr, which all contact the Fab, and105Tyr, which lies outside the crystallographically observed bindinginterface, were mutated to alanine in dimeric human CTLA-4 con-taining the native interchain disulfide. Interactions between thewild-type and mutant dimeric CTLA-4 molecules and the Fab wereexamined by SEC (Figs. 5A and 6A). The dimeric K95A and Y104Amutants did not exhibit any interaction with the Fab, whereas thedimeric M99A mutant exhibited significantly decreased associationcompared with the wild type (Fig. 6A). The Y105A mutation didnot significantly affect the interaction between dimeric CTLA-4 andthe Fab, consistent with the placement of 105Y outside the observedipilimumab recognition interface (Fig. 6A). The behavior of thewild-type dimeric CTLA-4 is consistent with the function blockingactivity of intact ipilimumab.The interactions between the Fab and the wild-type dimeric

CTLA-4 and dimeric mutants were further evaluated by nativenonreducing PAGE analysis. The ipilimumab Fab does not enterthe gel under native PAGE conditions due to the overall positivecharge of the Fab (pI of 8.8) at pH 8.0. Formation of the dimericCTLA-4:Fab complex results in a species capable of entering thegel, whereas unbound dimeric CTLA-4 results in a distinct band(Fig. 5B). As shown by the native PAGE results, the wild-typedimeric CTLA-4 interacted with Fab and formed new speciesconsistent with the formation of the complex in the gel (Fig. 5B).Dimeric K95A, M99A, and Y104A CTLA-4 mutants did notform any new species with ipilimumab Fab under native PAGEconditions (Fig. 6B). Dimeric Y105A CTLA-4 productivelyinteracted with the Fab, as demonstrated by the appearanceof a new band in the native PAGE (Fig. 6B). Importantly, asevidenced by fluorescence-monitored thermal denaturation, allof these dimeric CTLA-4 mutants are fully folded under theconditions used in these experiments (SI Materials and Methodsand Fig. S4). These results are fully consistent with the observed

Fig. 2. Critical interactions between ipilimumab and CTLA-4. (A) Surface representation of ipilimumab (heavy chain in light gray; light chain in dark gray)showing the cleft formed between the VH and VL domains. Residues on the FG (purple) and CC′ (beige) strands of CTLA-4 facing toward the Fab are shown instick representation. For clarity, side chains on the FG and CC′ strands facing away from the Fab are not shown. The G strand extends along the cleft withhydrophobic residues (104Tyr–108Ile) intercalating into the cleft between the two domains. The F and C strands stack against HCDR2 and HCDR1. 101Trp and102Leu from the tip of the HCDR3 loop contacts the center of the CTLA-4 front face and interacts with a hydrophobic patch formed by 39Leu, 46Val, and 93Ile.(B) Stick representation of the 99MYPPPY104 loop followed by G-strand residues (purple). The main-chain carbonyl and amide groups of the edge G-strandresidues not involved in interstrand interactions with the F strand participate in hydrogen bonding interactions with LCDR3 (olive) and LCDR1 (teal) residues.The 99MYPPPY104 loop (FG loop) packs against a cluster of aromatic residues, and 59Y(pink) from HCDR2 contacts the center of the FG loop and also par-ticipates in a hydrogen bonding interaction with the carbonyl oxygen of 99M.

E4226 | www.pnas.org/cgi/doi/10.1073/pnas.1617941114 Ramagopal et al.

Dow

nloa

ded

by g

uest

on

June

19,

202

1

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF3http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF3http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF3http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF3http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF3http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF3http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=STXThttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF4www.pnas.org/cgi/doi/10.1073/pnas.1617941114

crystallographic interface and support the physiological rele-vance of the CTLA-4: ipilimumab Fab complex. Interactions ofB7 ligands with wild-type CTLA-4 and mutants were confirmed bynative nonreducing PAGE analysis as well as by SEC (Figs. S5, S6,and S7).

Epitope Verification and Specificity Determinants of Ipilimumab. Thechallenge of intact ipilimumab with proteolytically generated(i.e., AspN, GluC, and Trypsin) peptide fragments derived fromthe disulfide-linked human CTLA-4–Fc fusion protein, coupledwith MS, identified a discontinuous epitope involving threedifferent linear peptides (Fig. S8). P1 stretches from26YASPGKATEVRVTVLRQA42, P2 from 43DSQVTEVCAA-TYMMGNELTFLDD65, and P3 from 96VELMYPPPYYLGIG109.From Table S1, it can be seen that residues from HCDR2 andHCDR3 interact with P1 and P2, respectively, whereas HCDR2,HCDR3, and LCDR3 interact with P3. Most of the interactionenergy between CTLA-4:B7-1 and CTLA-4:B7-2 is derived fromcontacts with 99MYPPPY104 (9), which is contained within the P3segment recognized by ipilimumab, consistent with the functionblocking activity of ipilimumab.To understand the specificity of ipilimumab toward CTLA-4, linear

peptides from both native and reduced human CD28–Fc fusion pro-tein were generated, and their interactions with ipilimumab-coupledbeads were examined. After elution from antibody-decorated beads,the mass and sequences of the bound peptides were characterizedby MALDI-TOF MS and nano-liquid chromatography (nano-LC)–MS/MS, respectively. Of particular note is the CD28 tryptic pep-tide, 96IEVMYPPPYLDNEK108, which did not bind ipilimumab(Fig. S8). In contrast, the corresponding peptide from CTLA-4,98LMYPPPYYLGIGN110, which is part of peptide P3 (Fig. S8),bound well to ipilimumab. Even though the sequence 99MYPPPY104

was present in both linear peptides from CTLA-4 and CD28, theobservation that only the CTLA-4–derived peptide binds implies thatresidues outside of this region play important roles in recognition ofthe 99MYPPPY104-containing epitope, which is consistent with thecrystallographic data as discussed above (also see Table S1).

Mechanism of CTLA-4 Recognition and Blockade by Ipilimumab. Theability of CTLA-4 and CD28 to recognize both the B7-1 and B7-2ligands is, in part, the consequence of the conserved FG loop(99MYPPPY104) shared by both receptors (Fig. 7). The in-troduction of mutations in this loop resulted in greater than 90%loss of affinity to the B7 ligands, identifying this segment as thecore of the ligand binding surface on both CTLA-4 and CD28 (9).Direct structural analyses of the CTLA-4:B7-1 and CTLA-4:B7-2complexes (29, 30) showed that the 99MYPPPY104 loop contributes∼80% of the interfacial contacts with the B7 ligands. The threeconsecutive proline residues in the FG loop assume an unusualcis–trans–cis conformation, and with the exception of 99Met, all ofthe main-chain carbonyl atoms are directed away from the ligandbinding surface (Fig. 8A). Given its chemical composition andconformation, this loop is devoid of free amide nitrogens or solvent-accessible carbonyls (except for the terminal residues 99M and 104Y)and thus lacks determinants typically associated with directionalityand specificity (i.e., hydrogen bond donors and acceptors). Thisunique and highly strained main-chain conformation provides sig-nificant geometric complementarity and hydrophobicity to supportrecognition of the concave surfaces presented by the front sheets of

Fig. 3. Mapping the ipilimumab-binding epitope on cell surface-expressedhuman CTLA-4. (A) Wild-type and mutant CTLA-4 constructs (118 total) weretransiently expressed as C-terminal mCherry fusions in HEK293 suspensioncells. Two days posttransfection, CTLA-4–expressing cells were queried witheither the ipilimumab monoclonal antibody or cells transiently expressinghB7-1-GFP, hB7-2-GFP, or hICOS-L-GFP and analyzed by flow cytometry todetermine the percentage of mCherry-positive cells (CTLA-4–expressing)bound (anti-human Alexa 488 signal reports on ipilimumab binding; GFPsignal reports on B7-1, B7-2, or ICOS-L binding). The chart highlights (coloredbackground) the average percent bound and SD from three independent

experiments for those CTLA-4 mutants that resulted in ≤50% binding to aparticular query. (B) Residues at which mutations caused a significant loss ofipilimumab binding are mapped onto the CTLA-4 crystal structure. The res-idues highlighted in red make direct contact with ipilimumab, the two blueresidues R40 and D88 form a salt bridge, and the two green residues(S20 and Q76) affected ipilimumab binding even though they are distal tothe ipilimumab recognition surface.

Ramagopal et al. PNAS | Published online May 8, 2017 | E4227

IMMUNOLO

GYAND

INFLAMMATION

PNASPL

US

Dow

nloa

ded

by g

uest

on

June

19,

202

1

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF5http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF6http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF7http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF8http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=ST1http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF8http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF8http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=ST1

the B7 ligands. The structures of the unliganded murine and humanCTLA-4 molecules demonstrate that this FG loop adopts essen-tially the same detailed conformation in both forms (35), indicatingthat CTLA-4 presents a preformed B7-recognition surface. Al-though a structure of CD28 bound to a B-7 ligand is not available,the conserved sequences and conformation of the FG loop in thestructure of a CD28:Fab complex (36), as well a series of muta-genesis experiments (9), supports a mode of interaction withB7 ligands similar to that observed for CTLA-4.Ipilimumab exploits the unique features of the FG loop and

sandwiches the front sheet of CTLA-4 between LCDR1 andLCDR3 on one side and HCDR1 and HCDR2 on the other (Fig.2A). The region between LCDR3 and HCDR2 is rich in aro-matic residues and stacks against the surface presented by the99MYPPPY104 loop (Fig. 2B). Residues in the tips of LCDR3 andLCDR1 are positioned proximal to the edge G-strand side of thefront sheet of CTLA-4 and participate in a series of hydrogenbonding interactions with main-chain atoms of the G strand (Fig.2B). HCDR1 and HCDR2 stack almost perpendicular to the C, F,and G stands of CTLA-4 and provide both hydrophobic andhydrogen bonding interactions with the solvent-exposed residueson the F and C strands as well as those located toward the terminiof the FG loop (Table S1). The HCDR3 loop contacts the center ofthe concave surface of CTLA-4 on its front sheet, with interactionsextending toward the CC′ loop (Figs. 1A and 2A). In the CTLA-4:

B7 complexes, the tip of the FG loop (99MYPPPY104) from CTLA-4contacts the front face of the B7 ligands (Fig. 8 C and D and Fig.S9 A and B), with the CC′ loop of CTLA-4 positioned away fromthe B7 surface. However, in the CTLA-4:ipilimumab complex,the –101PPP103− tip of the FG loop is only partially engaged,allowing for the association of the entire CTLA-4 front face withipilimumab (Fig. 8E and Fig. S9C). In this way, the ipilimumabinteraction surface not only sterically occludes the conserved99MYPPPY104 surface from availability to B7 ligands but also ex-tends its interaction toward the opposite side of the CTLA-4 IgVdomain (i.e., CC′ loop) (Fig. 8F). The recognition of this extendedCTLA-4 surface by ipilimumab (relative to the B7 ligands) is theconsequence of the highly twisted architecture common to the frontsheet of IgV domains of the antigen receptors, which results in theprotrusion of the long FG and CC′ loops away from the plane of thefront sheet, creating a concave surface on the CTLA-4 front sheet(Fig. 1C) (35).The differences in total buried surface area upon binding CTLA-4

(∼1,885 Å2 for ipilimumab vs. ∼1,250 Å2 for the B7 ligands), thegreater number of hydrogen bonds (∼13 for ipilimumab vs.∼5–7 for the B7 ligands), and the larger number of van derWaals contacts are consistent with the Kds exhibited by ipilimumab

Fig. 4. Ipilimumab directly competes with hB7-1 for binding to hCTLA-4 in abead-based FACS assay. (A) Schematic representation of the assay. (B) Pro-tein A beads saturated with hCTLA-4 hIgG1 were titrated with hB7-1 mIgG2aprotein. FACS analysis was used to determine the GeoMean of theFL1 channel (488) for ALL BEADS gated. Data represent the average of threeindependent experiments with error bars showing the SD. (C) Protein Abeads loaded with hCTLA-4 hIgG1 were saturated with hB7-1 mIgG2a (5 nM)and incubated with increasing concentrations of control mAb or ipilimumab.Data show the GeoMean of Fl1 normalized to the 5-nM titration point fromthat same experiment. All data show an average of three independent ex-periments with error bars showing SD.

Fig. 5. Biochemical confirmation of the ipilimumab:CTLA-4 binding in-terface, and formation of the wild-type dimeric hCTLA-4:Fab complexdemonstrated by gel filtration and native PAGE gel. (A) Gel filtration ofdimeric hCTLA-4 and Fab complex. The chromatograms for Fab, dimerichCTLA-4, and the dimeric hCTLA-4:Fab mixture are indicated as black, blue,and red curves, respectively. It should be noted that the ipilimumab Fab(molecular mass, ∼50 kDa) reproducibly exhibits aberrantly long retentiontimes relative to the smaller dimeric CTLA-4 (molecular mass, ∼25 kDa).(B) Native PAGE analysis of the dimeric hCTLA-4 and Fab complex. Theconcentration of wild-type dimeric hCTLA-4 was held constant at 10 μMwithincreasing concentration of Fab. Wild-type dimeric hCTLA-4 and Fab weremixed in molar ratios of 4:1, 2:1, 1:1, 1:2, and 1:4 from lane 1 to lane 5. Lanes6 and 7 show Fab and dimeric hCTLA-4 alone, respectively.

E4228 | www.pnas.org/cgi/doi/10.1073/pnas.1617941114 Ramagopal et al.

Dow

nloa

ded

by g

uest

on

June

19,

202

1

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=ST1http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF9http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF9http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF9www.pnas.org/cgi/doi/10.1073/pnas.1617941114

and B7 ligands for CTLA-4 (∼10 nM and ∼0.1–1 μM range, re-spectively) and the observation that ipilimumab competes effec-tively with the B7 ligands for binding CTLA-4. It is this directcompetition between ipilimumab and the B-7 ligands that, inpart, underlies the therapeutic efficacy of ipilimumab.

Ipilimumab Specificity. Although the B7 ligands interact with bothCTLA-4 and CD28, the ability of ipilimumab to discriminatebetween these two functionally distinct receptors is critical for itstherapeutic efficacy. CTLA-4 and CD28 share ∼30% sequenceidentity, including conservation of the critical core residues anddisulfide linkages that are essential for maintaining the IgV do-

main fold. Overall, the structure of the Fab-bound humanCTLA-4 monomer is similar to that in the apo–CTLA-4 homo-dimer structure (35), the CTLA-4:B7-1 complex (30), the CTLA-4:B7-2 complex (29), and the CTLA-4:Lipocalin (37) complex,with Cα–rmsds that range between 0.85 and 1.1 Å. Similarly, theCα–rmsds calculated with ipilimumab-bound human CTLA-4against human CD28 (36) and murine CTLA-4 (38) structuresare 1.51 Å and 1.27 Å, respectively.Nearly all features of the FG loop (99MYPPPY104) are con-

served among the structurally characterized CD28:CTLA-4 familymembers, including the sequences and the specific side-chainconformations; the sole exception is found in the apo–CTLA-4

Fig. 6. Gel filtration and native PAGE analysis of dimeric mutants and Fab. (A) The chromatograms for Fab, dimeric hCTLA-4 mutants, and the dimeric hCTLA-4:Fab mixture are indicated as black, blue, and red curves, respectively. (Top Left) K95A. (Top Right) Y104A. (Bottom Left) M99A. (Bottom Right) Y105A. Allpeaks of the corresponding proteins are indicated by arrows. The formation of the complex is indicated by a red arrow. (B) Distinct migration bands fordimeric hCTLA-4 mutants, Fab, and the hCTLA-4 mutant:Fab complex are diagnostic for interactions of dimeric hCTLA-4 mutants and the Fab. The con-centrations of dimeric hCTLA-4 mutants were held constant at 10 μM. (Top Left) Dimeric mutant K95A and Fab were mixed in molar ratios of 4:1, 2:1, 1:1, 1:2,and 1:4, from lane 1 to lane 5. Lanes 6 and 7 show Fab and dimeric mutant K95A alone, respectively. (Top Right) Dimeric mutant Y104A and Fab were mixed inmolar ratios of 4:1, 2:1, 1:1, 1:2, and 1:4, from lane 1 to lane 5. Lanes 6 and 7 show Fab and dimeric mutant Y104A alone, respectively. (Bottom Left) Dimericmutant M99A and Fab were mixed in molar ratios of 2:1, 1:1, and 1:2, from lane 1 to lane 3. Lanes 4 and 5 show Fab and dimeric mutant M99A alone,respectively. (Bottom Right) Dimeric mutant Y105A and Fab were mixed in molar ratios of 2:1, 1:1, and 1:2, from lane 1 to lane 3. Lanes 4 and 5 show Fab anddimeric mutant Y105A alone, respectively.

Ramagopal et al. PNAS | Published online May 8, 2017 | E4229

IMMUNOLO

GYAND

INFLAMMATION

PNASPL

US

Dow

nloa

ded

by g

uest

on

June

19,

202

1

homodimeric structure, where 100Tyr adopts a different rotamerconformation to accommodate crystal contacts (35). The confor-mation of this loop in the Fab-bound CD28 structure is also similarto that found in CTLA-4, including side-chain rotamers, with anoverall Cα–rmsd of ∼0.3 Å (Fig. 8 A and B). Based on these ob-servations, these segments are not likely to be major determinantsfor the discrimination between CTLA-4 and CD28. Although manyother CTLA-4 residues that interact with ipilimumab are conservedin CD28, there are a few important differences in sequence andconformation. 39Leu and 93Ile in CTLA-4, which interact with101Trp and 102Leu from HCDR3, are replaced in CD28 by 38His

and 93Phe, respectively, and may contribute to the ability of ipili-mumab to discriminate between CTLA-4 and CD28 (Fig. 8).As members of the antigen receptor group of the IgSF (35),

both CTLA-4 and CD28 possess β bulges in C′ and G strands,which in CTLA-4 are centered on 48Glu and 110Asn, respectively(in CD28, these bulges are centered on 46Glu and 110Asn, re-spectively) (Fig. 1B) (39). The –Gly–X–Gly– sequence of G-strandβ bulge observed in CTLA-4 is present in all VL domains andmore than 98% of VH domains of mABs (39). Notably, in CD28,the first glycine in this sequence is replaced by serine. In CD28,the single residue insertion 108Glu (between the equivalent of108Gly and 109Ile in CTLA-4), just before the G-strand β bulge,

Fig. 7. Sequence alignment of CTLA-4 and CD28. CTLA-4 residues interacting with ipilimumab are marked with blue triangles. The filled circles indicateresidues from B7-1 (crimson) and B7-2 (purple) that make contacts less than 4.0 Å. Insertion in CD28, before the β bulge on the G strand, is shown withan arrow. Notice the sequence differences in the G-strand residues just after the 99MYPPPY104 loop. The three discontinuous stretches of CTLA4 (P1,26YASPGKATEVRVTVLRQA42; P2, 43DSQVTEVCAATYMMGNELTFLDD65; and P3, 96VELMYPPPYYLGIG109) identified as part of the ipilimumab epitope, by nano-LC–MS/MS, are shown by brown lines above the sequence of CTLA-4.

Fig. 8. Mechanism of ipilimumab action. (A) Identical residues between CTLA-4 and CD28 are colored blue, and insertions in the CTLA-4 sequence are coloredred; both are mapped onto the CTLA-4 structure (yellow). (B) Identical residues between CTLA-4 and CD28 are colored blue, and insertions in theCD28 sequence are colored red; both are mapped onto the CD28 structure (yellow). Note that the MYPPPY-loop surfaces contours are similar. (C) Mode ofB7-2 (light blue) interaction with the MYPPPY-loop surface. (D) Mode of B7-1 (pink) interaction with the MYPPPY-loop surface. (E) Mode of CTLA–4 andipilimumab interactions. (F) Superposition of the CTLA-4:B7-2 and CTLA-4:ipilimumab structures, based on the CTLA-4 component in each complex, is pre-sented. Superposition of the CTLA-4:B7-1 and CTLA-4:ipilimumab complexes is shown in Fig. S9. These superpositions indicate that ipilimumab and the B7ligands compete for overlapping binding surfaces on CTLA-4.

E4230 | www.pnas.org/cgi/doi/10.1073/pnas.1617941114 Ramagopal et al.

Dow

nloa

ded

by g

uest

on

June

19,

202

1

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=SF9www.pnas.org/cgi/doi/10.1073/pnas.1617941114

exaggerates the protrusion of the G strand (Figs. 1B, 7, and 8B). Inaddition, on the G strand, immediately following the conserved99MYPPPY104 loop, all residues within the span of 105–109 aredifferent in CD28 (Figs. 7 and 9). In particular, 107Asn preceding theinsertion is involved in two hydrogen bond interactions with theF-strand backbone, which further perturbs the canonical interstrandinteraction between the F and G strands (Fig. 9 B and C), pushingthe two strands farther apart than is typical. In CTLA-4, this region isinvolved in continuous main chain–main chain and main chain–sidechain interactions with LCDR1 and LCDR3 (Fig. 2B). Superpositionof CD28 on the CTLA-4:ipilimumab complex predicts that the dis-tortion of the G strand results in not only loss of hydrogen bonds butalso a substantial steric clash with LCDR1 (Fig. 9C). The detailedstructural differences in the G strands of CTLA-4 and CD28 arelikely the major determinants responsible for the specificity ex-hibited by ipilimumab. In addition, in CTLA-4 there is an insertionof residue 47T just before the C′-strand β bulge (between theequivalent of residues 45V and 46E in CD28; Figs. 7 and 8A), whichprobably contributes to the differences in the CC′ loop conforma-tion. However, this insertion is distal to the recognition interfaceand is not predicted to significantly impact the binding or selectivity.An important consideration for normal physiology and ther-

apeutic efficacy is that an optimal biological outcome is not al-ways associated with the strongest achievable binding (i.e., veryhigh affinity between receptor and ligand). This concept is il-lustrated by numerous physiological processes, such as the needfor reversible binding interactions in oxygen delivery, replication,transcription, and translation, and these same principles are di-rectly relevant to the development of therapeutic strategies(reviewed in ref. 40). Notable examples are provided by recentreports of chimeric antigen receptor (CAR) T cells engineeredto present scFv modules spanning a range of affinities for thetarget antigen (41, 42). These studies demonstrate that scFvswith reduced affinities had increased selectivity for solid tumors,relative to normal cells, due to the requirement for higher sur-face density of target antigens to support productive engage-ment, a state often afforded by the targeted tumor cells. Forexample, CAR T cells bearing scFvs derived from nimotuxumab,but not cetuximab (43), effectively discriminated between ma-lignant cells and nonmalignant cells based on its ∼10-fold poorer

Kd for EGFR, which is expressed at significantly higher levels onthe malignant cells (41). Similar studies, which examined scFvsagainst EGFR (derived from the C10 anti-EGFR antibody) (44)and ErbB2 (derived from the 4D5 trastuzumab antibody) (45) withKds spanning ∼2–3 orders of magnitude, have reinforced thisconcept. Together, these studies demonstrate that apparent Kdsnot only control selectivity between target and nontarget cells butcan also impact CAR T-cell function (e.g., cytokine production)and suggest that on and off rates (i.e., kon and koff) will have im-portant mechanistic contributions to the ultimate therapeuticfunction. These examples and considerations underscore the needto achieve optimal, not maximal, affinities (and kinetics) to attainthe desired biological/therapeutic activity. In the case of ipilimu-mab, affinity-attenuated variants could be generated, which wouldexhibit selectivity for activated Treg cells present at tumor sites dueto their higher expression levels of CTLA-4, compared with otheractivated T cells (20, 21). Additionally, it might be possible toengineer ipilimumab variants that are more responsive to the re-duced pH often associated with the tumor microenvironment (46–48). Thus, the structure of the CTLA-4:ipilimumab complex pro-vides the foundation for the rationale design of affinity-modulatedipilimumab variants with distinct T-cell subset-targeting profilesand more selective and efficacious therapeutic properties.

Materials and MethodsDetailed methods can be found in SI Materials and Methods.

Monomeric and dimeric CTLA-4 and the monomeric IgV domain of hB7-2were refolded as previously described from inclusion bodies (29, 49). Ipili-mumab Fab fragments were prepared by papain digestion. Crystals of theCTLA-4:ipilimumab Fab complex were obtained by sitting drop vapor dif-fusion and the structure determined by molecular replacement. The crys-tallographically determined binding interface was confirmed by a high-throughput FACS analysis using a library of CTLA-4 mutants expressed onthe surface of HEK293 cells as well as by direct biochemical characterizationof ipilimumab binding to wild-type CTLA-4 and mutant CTLA-4 variants.

ACKNOWLEDGMENTS. We thank the staff of the 24-ID-E beamline, AdvancedPhoton Source, Argonne National Laboratory for assistance with X-ray diffrac-tion data collection. This work was supported by National Institutes of HealthGrants HG008325, GM094662, and GM094665 (to S.C.A.); and the Albert EinsteinCancer Center (P30CA013330).

Fig. 9. Determinants of selectivity. (A) The 99MYPPPY104 loop conformations in CTLA-4 (purple) and CD28 (yellow) are very similar. Note that only thecarbonyl moiety from 99M and the amide nitrogen from 104Y are facing toward the viewer and are available for interaction with the ligands. (B) Insertion inthe edge G strand following the conserved β bulge causes the G strand of CD28 to protrude away from the F strand, resulting in the loss of many interstrandinteractions with the F strand. (C) Possible clashes between the LCDR1 and the extended protrusion of the G strand due to insertion at 109K on CD28 (yellow).The G strand from CTLA-4 is colored purple.

Ramagopal et al. PNAS | Published online May 8, 2017 | E4231

IMMUNOLO

GYAND

INFLAMMATION

PNASPL

US

Dow

nloa

ded

by g

uest

on

June

19,

202

1

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1617941114/-/DCSupplemental/pnas.201617941SI.pdf?targetid=nameddest=STXT

1. Mellman I, Coukos G, Dranoff G (2011) Cancer immunotherapy comes of age. Nature480:480–489.

2. Korman AJ, Peggs KS, Allison JP (2006) Checkpoint blockade in cancer immunother-apy. Adv Immunol 90:297–339.

3. McGranahan N, et al. (2016) Clonal neoantigens elicit T cell immunoreactivity andsensitivity to immune checkpoint blockade. Science 351:1463–1469.

4. Couzin-Frankel J (2013) Breakthrough of the year 2013. Cancer immunotherapy.Science 342:1432–1433.

5. Pardoll DM (2012) The blockade of immune checkpoints in cancer immunotherapy.Nat Rev Cancer 12:252–264.

6. Peggs KS, Quezada SA, Chambers CA, Korman AJ, Allison JP (2009) Blockade of CTLA-4on both effector and regulatory T cell compartments contributes to the antitumoractivity of anti-CTLA-4 antibodies. J Exp Med 206:1717–1725.

7. ZouW, Wolchok JD, Chen L (2016) PD-L1 (B7-H1) and PD-1 pathway blockade for cancertherapy: Mechanisms, response biomarkers, and combinations. Sci Transl Med 8:328rv4.

8. Schwartzentruber DJ, et al. (2011) gp100 peptide vaccine and interleukin-2 in patientswith advanced melanoma. N Engl J Med 364:2119–2127.

9. Metzler WJ, et al. (1997) Solution structure of human CTLA-4 and delineation of aCD80/CD86 binding site conserved in CD28. Nat Struct Biol 4:527–531.

10. Hodi FS, et al. (2010) Improved survival with ipilimumab in patients with metastaticmelanoma. N Engl J Med 363:711–723.

11. Robert C, et al.; KEYNOTE-006 investigators (2015) Pembrolizumab versus ipilimumabin advanced melanoma. N Engl J Med 372:2521–2532.

12. Wolchok JD, et al. (2013) Nivolumab plus ipilimumab in advanced melanoma. N Engl JMed 369:122–133.

13. Larkin J, Hodi FS, Wolchok JD (2015) Combined nivolumab and ipilimumab or mon-otherapy in untreated melanoma. N Engl J Med 373:1270–1271.

14. Postow MA, et al. (2015) Nivolumab and ipilimumab versus ipilimumab in untreatedmelanoma. N Engl J Med 372:2006–2017.

15. Antonia S, et al. (2016) Safety and antitumour activity of durvalumab plus tremelimumabin non-small cell lung cancer: A multicentre, phase 1b study. Lancet Oncol 17:299–308.

16. Qureshi OS, et al. (2011) Trans-endocytosis of CD80 and CD86: A molecular basis forthe cell-extrinsic function of CTLA-4. Science 332:600–603.

17. Corse E, Allison JP (2012) Cutting edge: CTLA-4 on effector T cells inhibits in trans.J Immunol 189:1123–1127.

18. Wang CJ, et al. (2012) Cutting edge: Cell-extrinsic immune regulation by CTLA-4 ex-pressed on conventional T cells. J Immunol 189:1118–1122.

19. Quezada SA, Peggs KS, Curran MA, Allison JP (2006) CTLA4 blockade and GM-CSFcombination immunotherapy alters the intratumor balance of effector and regula-tory T cells. J Clin Invest 116:1935–1945.

20. Selby MJ, et al. (2013) Anti-CTLA-4 antibodies of IgG2a isotype enhance antitumor ac-tivity through reduction of intratumoral regulatory T cells. Cancer Immunol Res 1:32–42.

21. Simpson TR, et al. (2013) Fc-dependent depletion of tumor-infiltrating regulatory T cells co-defines the efficacy of anti-CTLA-4 therapy against melanoma. J Exp Med 210:1695–1710.

22. Ribas A, et al. (2013) Phase III randomized clinical trial comparing tremelimumab withstandard-of-care chemotherapy in patientswith advancedmelanoma. J ClinOncol 31:616–622.

23. Schadendorf D, et al. (2015) Pooled analysis of long-term survival data from phase IIand phase III trials of ipilimumab in unresectable or metastatic melanoma. J ClinOncol 33:1889–1894.

24. Eroglu Z, et al. (2015) Long term survival with cytotoxic T lymphocyte-associatedantigen 4 blockade using tremelimumab. Eur J Cancer 51:2689–2697.

25. Romano E, et al. (2015) Ipilimumab-dependent cell-mediated cytotoxicity of regula-tory T cells ex vivo by nonclassical monocytes in melanoma patients. Proc Natl Acad SciUSA 112:6140–6145.

26. Hodi FS, et al. (2008) Immunologic and clinical effects of antibody blockade of cyto-toxic T lymphocyte-associated antigen 4 in previously vaccinated cancer patients. ProcNatl Acad Sci USA 105:3005–3010.

27. Liakou CI, et al. (2008) CTLA-4 blockade increases IFNgamma-producing CD4+ICOShicells to shift the ratio of effector to regulatory T cells in cancer patients. Proc NatlAcad Sci USA 105:14987–14992.

28. Selby M, et al. (2013) Antitumor activity of concurrent blockade of immune check-point molecules CTLA-4 and PD-1 in preclinical models. J Clin Oncol 31:3061.

29. Schwartz JC, Zhang X, Fedorov AA, Nathenson SG, Almo SC (2001) Structural basis forco-stimulation by the human CTLA-4/B7-2 complex. Nature 410:604–608.

30. Stamper CC, et al. (2001) Crystal structure of the B7-1/CTLA-4 complex that inhibitshuman immune responses. Nature 410:608–611.

31. Davies DR, Cohen GH (1996) Interactions of protein antigens with antibodies. ProcNatl Acad Sci USA 93:7–12.

32. Jeon H, et al. (2014) Structure and cancer immunotherapy of the B7 family memberB7x. Cell Reports 9:1089–1098.

33. Hamby SE, Hirst JD (2008) Prediction of glycosylation sites using random forests. BMCBioinformatics 9:500.

34. Zhang X, Schwartz JC, Almo SC, Nathenson SG (2002) Expression, refolding, purifi-cation, molecular characterization, crystallization, and preliminary X-ray analysis ofthe receptor binding domain of human B7-2. Protein Expr Purif 25:105–113.

35. Yu C, et al. (2011) Rigid-body ligand recognition drives cytotoxic T-lymphocyte anti-gen 4 (CTLA-4) receptor triggering. J Biol Chem 286:6685–6696.

36. Evans EJ, et al. (2005) Crystal structure of a soluble CD28-Fab complex.Nat Immunol 6:271–279.37. Schönfeld D, et al. (2009) An engineered lipocalin specific for CTLA-4 reveals a com-

bining site with structural and conformational features similar to antibodies. ProcNatl Acad Sci USA 106:8198–8203.

38. Ostrov DA, Shi W, Schwartz JC, Almo SC, Nathenson SG (2000) Structure of murineCTLA-4 and its role in modulating T cell responsiveness. Science 290:816–819.

39. Chothia C, Novotný J, Bruccoleri R, Karplus M (1985) Domain association in immu-noglobulin molecules. The packing of variable domains. J Mol Biol 186:651–663.

40. Almo SC, Guha C (2014) Considerations for combined immune checkpoint modulationand radiation treatment. Radiat Res 182:230–238.

41. Liu X, et al. (2015) Affinity-tuned ErbB2 or EGFR chimeric antigen receptor T cells exhibit anincreased therapeutic index against tumors in mice. Cancer Res 75:3596–3607.

42. Caruso HG, et al. (2015) Tuning sensitivity of CAR to EGFR density limits recognition ofnormal tissue while maintaining potent antitumor activity. Cancer Res 75:3505–3518.

43. Talavera A, et al. (2009) Nimotuzumab, an antitumor antibody that targets the epi-dermal growth factor receptor, blocks ligand binding while permitting the activereceptor conformation. Cancer Res 69:5851–5859.

44. Heitner T, et al. (2001) Selection of cell binding and internalizing epidermal growth factorreceptor antibodies from a phage display library. J Immunol Methods 248:17–30.

45. Carter P, et al. (1992) Humanization of an anti-p185HER2 antibody for human cancertherapy. Proc Natl Acad Sci USA 89:4285–4289.

46. BelloneM, et al. (2013) The acidity of the tumormicroenvironment is amechanismof immuneescape that can be overcome by proton pump inhibitors. OncoImmunology 2:e22058.

47. Estrella V, et al. (2013) Acidity generated by the tumor microenvironment drives localinvasion. Cancer Res 73:1524–1535.

48. Kato Y, et al. (2013) Acidic extracellular microenvironment and cancer. Cancer Cell Int13:89.

49. Zhang X, Schwartz JC, Nathenson SG, Almo SC (2001) Crystallization and preliminaryX-ray analysis of the complex between human CTLA-4 and B7-2. Acta CrystallogrD Biol Crystallogr 57:898–899.

50. Otwinowski Z, Minor W (1997) Processing of X-ray diffraction data collected in os-cillation mode. Methods Enzymol 276:307–326.

51. Vagin A, Teplyakov A (2010) Molecular replacement with MOLREP. Acta Crystallogr DBiol Crystallogr 66:22–25.

52. Emsley P, Cowtan K (2004) Coot: Model-building tools for molecular graphics. ActaCrystallogr D Biol Crystallogr 60:2126–2132.

53. Murshudov GN, Vagin AA, Dodson EJ (1997) Refinement of macromolecular struc-tures by the maximum-likelihood method. Acta Crystallogr D Biol Crystallogr 53:240–255.

54. Lee B, Richards FM (1971) The interpretation of protein structures: Estimation of staticaccessibility. J Mol Biol 55:379–400.

55. Kabsch W (1976) A solution for the best rotation to relate two sets of vectors. ActaCryst A32:922–923.

56. Winn MD, et al. (2011) Overview of the CCP4 suite and current developments. ActaCryst D67:235–242.

57. DeLano WL (2005) The case for open-source software in drug discovery. Drug DiscovToday 10:213–217.

58. Fraczkiewicz R, Braun WJ (1998) Exact and efficient analytical calculation of theaccessible surface areas and their gradients for macromolecules. Comp Chem 19:319–333.

E4232 | www.pnas.org/cgi/doi/10.1073/pnas.1617941114 Ramagopal et al.

Dow

nloa

ded

by g

uest

on

June

19,

202

1

www.pnas.org/cgi/doi/10.1073/pnas.1617941114