Structural Basis for Treating Tumor Necrosis Factor α (TNFα)-associated Diseases with the Therapeutic Antibody Infliximab*

Background: Although infliximab has high efficacy in treating TNFα-associated diseases, the epitope on TNFα remains unclear. Results: The crystal structure of the TNFα in complex with the infliximab Fab is reported at a resolution of 2.6 Å. Conclusion: TNFα E-F loop plays a crucial role in the interaction. Significance: The structure may lead to understanding the mechanism of mAb anti-TNFα. Monoclonal antibody (mAb) drugs have been widely used for treating tumor necrosis factor α (TNFα)-related diseases for over 10 years. Although their action has been hypothesized to depend in part on their ability to bind precursor cell surface TNFα, the precise mechanism and the epitope bound on TNFα remain unclear. In the present work, we report the crystal structure of the infliximab Fab fragment in complex with TNFα at a resolution of 2.6 Å. The key features of the TNFα E-F loop region in this complex distinguish the interaction between infliximab and TNFα from other TNF-receptor structures, revealing the mechanism of TNFα inhibition by overlapping with the TNFα-receptor interface and indicating the crucial role of the E-F loop in the action of this therapeutic antibody. This structure also indicates the formation of an aggregated network for the activation of complement-dependent cytolysis and antibody-dependent cell-mediated cytotoxicity, which result in development of granulomatous infections through TNFα blockage. These results provide the first experimental model for the interaction of TNFα with therapeutic antibodies and offer useful information for antibody optimization by understanding the precise molecular mechanism of TNFα inhibition.

Monoclonal antibody (mAb) drugs have been widely used for treating tumor necrosis factor ␣ (TNF␣)-related diseases for over 10 years. Although their action has been hypothesized to depend in part on their ability to bind precursor cell surface TNF␣, the precise mechanism and the epitope bound on TNF␣ remain unclear. In the present work, we report the crystal structure of the infliximab Fab fragment in complex with TNF␣ at a resolution of 2.6 Å. The key features of the TNF␣ E-F loop region in this complex distinguish the interaction between infliximab and TNF␣ from other TNF-receptor structures, revealing the mechanism of TNF␣ inhibition by overlapping with the TNF␣-receptor interface and indicating the crucial role of the E-F loop in the action of this therapeutic antibody. This structure also indicates the formation of an aggregated network for the activation of complement-dependent cytolysis and antibody-dependent cell-mediated cytotoxicity, which result in development of granulomatous infections through TNF␣ blockage. These results provide the first experimental model for the interaction of TNF␣ with therapeutic antibodies and offer useful information for antibody optimization by understanding the precise molecular mechanism of TNF␣ inhibition.
Tumor necrosis factor ␣ (TNF␣) is an inflammatory cytokine that plays a central role in acute inflammation and is responsible for a diverse range of signaling events within cells that triggers necrosis or apoptosis (1)(2)(3)(4). TNF␣ is mainly produced in activated macrophages and natural killer cells, whereas lower expression is found in a variety of other cells, including fibroblasts, smooth muscle cells, and tumor cells (5). Human TNF␣ is translated as a 26-kDa membrane-associated form and is then cleaved in the extracellular domain through the action of matrix metalloproteases to release a mature soluble 17-kDa protein (6). TNF␤ (also known as lymphotoxin) is another important TNF member, and its primary sequence shares high sequence and structural similarities with TNF␣ (7,8). Both TNF␣ and TNF␤ affect a number of normal and neoplastic cell processes.
The correct functioning of TNF requires effective communication with TNF receptors (TNFRs). 4 Currently, two structurally distinct TNFRs, named TNFR1 and TNFR2, have been identified; both bind with the released soluble form and membrane-associated form of TNF␣, respectively (9,10). The binding of TNF␣ to TNFR1 has been shown to induce apoptosis and lead to activation of transcription factors involved in cell survival and inflammatory responses as well as to initiate the pathways that lead to caspase activation through the TNFR-associated death domain and FAS-associated death domain proteins (11)(12)(13). This physiologic relevance suggests that sequestering TNF␣ could be used to treat human autoimmune diseases (14), and a number of anti-TNF␣ agents (drugs and mAbs) have been developed to treat patients with TNF␣-associated diseases such as Crohn disease, psoriatic arthritis, rheumatoid arthritis, ankylosing spondylitis, and persistent uveitis (15).
Therapeutic mAbs have high efficacy in treating TNF␣-associated diseases. Currently, three versions of therapeutic mAbs, i.e. etanercept (Enbrel), infliximab (Remicade), and adalimumab (Humira), have been approved by the United States Food and Drug Administration. Among them, infliximab is a chimeric antibody composed of a complement-fixing human IgG1 constant region (75%) and a murine-derived antigenbinding variable region (25%) (16). Infliximab was developed in 1993 and was first approved for treating Crohn disease. Its use has since been extended to the treatment of ankylosing spondylitis, psoriatic arthritis, rheumatoid arthritis, and various inflammatory skin diseases (17). Infliximab is known for its ability to neutralize the biological activity of TNF␣ by binding to the soluble (free floating in the blood) and transmembrane (located on the outer membranes of T cells and similar immune cells) forms of TNF␣ with high affinity, preventing it from binding to cellular receptors and inducing the lysis of cells that produce TNF␣ (18,19). Infliximab affects the TNF␣-mediated signaling pathways of cell proliferation, apoptosis, and cytokine suppression (20). Although the binding avidity or affinity between TNF␣ and infliximab is reportedly variable because of the different measurement methods used, the high binding avidity/affinity results in the formation of stable TNF␣-infliximab complexes (21)(22)(23). Interestingly, although TNF␣ shares high sequence and structural similarities with TNF␤, there is no evidence to show that infliximab can neutralize TNF␤ (24), which indicates the high specificity of infliximab in interacting with TNF␣.
Although crystallographic studies on TNF␣-TNFR2 and TNF␤-TNFR1 complexes in past decades provided the breakthrough for understanding how TNF functions through communicating with receptors (8,25,26), the experimental structure of TNF␣ in complex with the therapeutic antibodies remains exclusive, and the precise mechanism and the epitope on TNF␣ is still unclear (27). In this work, the crystal structure of TNF␣ in complex with the infliximab Fab fragment is reported at a resolution of 2.6 Å. The crystal structure of the TNF␣-infliximab Fab together with the structures of TNF␤-TNFR1 and TNF␣-TNFR2 complexes rationalizes the inhibition of TNF␣-receptor interaction by overlap between the mAb-and TNFR-binding sites on the TNF␣. Moreover, the distinct features of the E-F loop on TNF␣ in the TNF␣-infliximab Fab complex suggest the molecular basis for the specific binding of infliximab to TNF␣ but not TNF␤. The structure of the TNF␣-infliximab Fab complex also indicates the formation of an aggregated network for the inhibition of membrane-associated TNF␣ function and, therefore, activation of complement-dependent cytolysis and antibody-dependent cell-mediated cytotoxicity, which result in the reported risk of developing granulomatous infection of TNF␣ blockages. These results lead to a better understanding of the mechanism of mAbs used for treating TNF␣-associated diseases and provide a new focus for the design of future drugs that target TNF␣ with high efficacy and specificity and with fewer adverse effects.
The transformed cells were grown in Luria-Bertani (LB) medium at 37°C until the OD 600 reached 1.5, and protein expression was induced with 0.5 mM isopropyl 1-thio-␤-D-galactopyranoside for 4 h. The bacterial cells were incubated in a lysis buffer (PBS) containing 1 mg/ml lysozyme, 1 mM PMSF, and 1% Triton X-100 for 20 min on ice followed by sonication. The cell lysate was cleared by centrifugation (10,000 ϫ g) and filtration (0.45 m). Solid ammonium sulfate was added to the supernatant to a final concentration of 35%, immediately mixed, and incubated on a roller at 4°C for 2 h. The solution was then cleared by centrifugation (10,000 ϫ g). After discarding the precipitate, solid ammonium sulfate was continuously added to the supernatant to a final concentration of 60%, immediately mixed, and incubated on a roller at 4°C for 4 h. The precipitated protein was pelleted by centrifugation (10,000 ϫ g), and the supernatant was discarded. The precipitate was dissolved in PBS (20 mM phosphate, pH 8.0 and 150 mM NaCl) and separated by gel filtration using Superdex 75 (GE Healthcare). After desalting to 20 mM Tris-HCl, pH 8.0, the target fraction was further purified with a 20-column volume linear NaCl gradient elution from high performance Q-Sepharose (GE Healthcare). The purity was confirmed to be Ͼ95% using sodium dodecyl sulfate-polyacrylamide electrophoresis (SDS-PAGE) analysis. The bioactivity was measured in a cytotoxicity assay using the TNF-susceptible murine L-929 cell line in the presence of the metabolic inhibitor actinomycin D (28).
Infliximab (29) was cloned, expressed, and purified following reported procedures. Briefly, the EcoRV and XbaI sites were added to the 5Ј-end of the heavy chain variable region gene (V H ), and an NheI site was added to the 3Ј-end. The PCR product was cloned into the pGEM-T vector, and its sequence was confirmed by DNA sequencing. V H was excised through EcoRV and NheI digestion and then inserted into the EcoRV/NheI sites of the pAH4604 vector containing the human ␥-1 constant region gene (C H ). The resultant pAH4604-V H vector was cleaved with XbaI and BamHI, and then the 3.3-kb fragment containing the chimeric rodent/human antibody heavy chain gene was cloned into the pcDNA3.1(Ϫ) vector (Invitrogen), which was digested with the same restriction enzymes, yielding the chimeric heavy chain expression vector pcDNA3.1(Ϫ)-V H C H . The human chain constant cDNA (C L ) was obtained as a 0.3-kb PCR product derived from pAG4622. The light chain variable region gene (V L ) of infliximab was fused to the 5Ј-end of the C L using the overlapping PCR method. The resultant chimeric light chain gene (V L C L ) with a HindIII site upstream of the start codon and an EcoRI site downstream of the stop codon was cloned into the pGEM-T vector, and its sequence was verified. V L C L was excised through HindIII and EcoRI digestion and ligated into the pcDNA3.1 Zeo(ϩ) vector (Invitrogen) cleaved with the same restriction enzymes, yielding the chimeric light chain expression vector pcDNA3.1 Zeo(ϩ)V L C L . The chimeric light and heavy chain expression vectors were co-transfected into Chinese hamster ovary K1 cells using Lipofectamine 2000 reagent (Invitrogen). Stable transfectants were isolated by limiting dilution in the presence of 600 g/ml G418 and 300 g/ml Zeocin. The culture supernates from individual cell clones were analyzed for antibody production using a sandwich enzyme-linked immunosorbent assay. The assay used goat anti-human IgG Fc (Kirkegaard and Perry Laboratories, Inc., Gaithersburg, MD) as the capture antibodies and goat anti-human -horseradish peroxidase (HRP) (Southern Biotechnology Associates, Birmingham, AL) as the detecting antibodies. Purified human IgG1/ (Sigma) was used as the standard control. The clones that produced the highest amount of recombinant antibodies were selected and grown in serum-free medium. The recombinant antibodies were purified using protein A affinity chromatography from the serum-free culture supernatant. The antibody concentrations were determined by absorbance at 280 nm, and the purity was confirmed using SDS-PAGE analysis. Bioactivity was measured in a cytotoxicity antagonist assay using the TNF-susceptible murine L-929 cell line in the presence of the metabolic inhibitor actinomycin D and TNF␣. The Fab fragment of infliximab for crystallographic investigation was obtained through papain digestion of infliximab. The digested protein sample was loaded onto a protein A-Sepharose 4 FF column (GE Healthcare), and the Fab fragment eluted in the flow-through was separated from the Fc fragment and further purified using ion-exchange chromatography using a Q-Sepharose FF column (GE Healthcare). The protein sample was concentrated to ϳ10 mg/ml and then exchanged to a stock buffer containing 10 mM Tris-HCl, pH 8.0 and 100 mM NaCl.
TNF␣ and infliximab Fab were mixed at a molar ratio of 1:1 and incubated for 10 h at 4°C to form the complex before crystallization. The mixed protein was further purified using Superdex 200 gel filtration columns (GE Healthcare) following the procedure suggested by the manufacturer. The fractions were analyzed by SDS-PAGE, and the purity was Ͼ95%. The purified protein was then concentrated to 30 mg/ml in 20 mM Tris-HCl, pH 7.4 and 150 mM NaCl for crystallization.
Crystallization-Crystallization of the infliximab Fab/TNF␣ complex was performed at 290 K using the hanging drop vapor diffusion method. The crystals grew in drops consisting of 1.5 l of protein and 1.5 l of reservoir solution against 200 l of reservoir solution. The initial crystal appeared after 3 days of growth in 1.4 M sodium/potassium phosphate, pH 8.2 with poor diffraction quality after initial screening. Sodium citrate (300 mM) was added to the original solution, and crystals with good diffraction quality were obtained after 5 days of growth. The crystals were soaked in a cryoprotectant solution consisting of the reservoir solution and 25% (v/v) glycol and then flash frozen in liquid nitrogen for x-ray diffraction.
X-ray Data Collection, Processing, and Structure Determination-The initial data set for the TNF␣-infliximab Fab complex was collected at the BL17A beamline (Photon Factory, Japan) at a resolution of 3.1 Å. The optimized crystals with good diffraction quality were diffracted to 2.6-Å resolution at 100 K in the Beijing Synchrotron Radiation Facility 3W1A and Shanghai Synchrotron Radiation Facility BL17U beamlines at a wavelength of 1.0000 Å with Mar165 and Mar225 charge-coupled device detectors, respectively. The data were processed, integrated, and scaled using the HKL2000 package (30). The crystals belong to space group H3 with cell parameters a ϭ b ϭ 154.0 Å, c ϭ 99.3 Å, ␣ ϭ ␤ ϭ 90°, and ␥ ϭ 120°. Only one complex molecule per asymmetric unit with a Matthews coefficient of 3.7 Å 3 /Da was present, corresponding to 63.4% solvent content (31). The statistical analysis of all data is presented in Table 1.
The infliximab Fab-TNF␣ structure was solved by molecular replacement using the crystal structures of apo-TNF␣ (Protein Data Bank code 1TNF), the light chain of the mAb cetuximab/ Erbitux/IMC-C225 (Protein Data Bank code 1YY8), and the heavy chain of humanized antibody C25 Fab fragment (Protein Data Bank code 2GCY) as the initial search models using the program PHASER (32). The structures of uncomplexed light chain from the mAb cetuximab/Erbitux/IMC-C225 and heavy chain from humanized antibody C25 Fab fragment were also used to represent the free form of infliximab Fab. The clear solutions in both the rotation and translation functions indicated the presence of one complex molecule, including one TNF␣ and one infliximab Fab molecule, in one asymmetric unit, which is consistent with the Matthews coefficient and solvent content (33). Residues that differ between infliximab and the searching model were manually rebuilt in the program Coot (34) under the guidance of the F o Ϫ F c and 2F o Ϫ F c electron density maps.
After the refinement of the model using simulated annealing, energy minimization, restrained individual B factors, and addition of 83 solvent molecules in PHENIX (35), the respective working R factor and R free dropped from 0.37 and 0.45 to 0.19 and 0.23, respectively, for all data from 50.0 to 2.6 Å. Refinement was monitored by calculating R free based on a subset containing 5% of the total reflections. Model geometry was verified using the program PROCHECK (36). Data collection and refinement statistics are detailed in Table 1. All structure figures were prepared using PyMOL (37).
Competitive Binding Assay-A 96-well plate was coated overnight at 4°C with 100 l of recombinant human TNF␣ (5 g/ml). Blocking treatment was performed at 37°C for 2 h. Before addition to the coated plate, different dilutions of infliximab were incubated with 3 g/ml biotin-labeled Yisaipu (recombinant human TNFR2-Fc fusion protein, also known as etanercept, which is biosimilar to Enbrel; a product of CPGJ Pharmaceutical, Ltd., Shanghai, China) in PBS. Preincubation mixtures were added to the coated plate. After 2 h of incubation at 37°C, the wells were washed, and an appropriate dilution of HRP-conjugated avidin was used for detection. After the addition of tetramethylbenzidine and stop solution, the absorbance was read at 450 nm with a microplate reader. The percentage of inhibition was calculated using the following formula: Percent inhibition ϭ (A 450 max Ϫ A 450 sample)/(A 450 max Ϫ A 450 blank) ϫ 100.
Kinetics and Binding Assay of TNF␣ Mutants-E-F loop replacement mutants (into GGGG and SGGSGGSGGSG) and site-directed mutants (Q67A, K112A, R138A, and Y141A) were created using PCR. The mutants were expressed and purified as described for wild-type protein. The infliximab Fab was immobilized onto the surface of a CM-5 sensor chip (GE Healthcare) via amine coupling following the manufacturer's instructions. Maximal electrostatic interaction was obtained with 10 mM sodium acetate, pH 5.0 (data not shown). Infliximab Fab immobilization levels ranging from ϳ1,000 to 1,500 resonance units were regularly obtained. For binding experiments, the BIAcore T100 (GE Healthcare) instrument was operated at 25°C, and the assay buffer was PBS (20 mM phosphate, pH 7.0 and 150 mM NaCl). The contact time (the period during which the analyte, TNF␣ mutants 4 and 11, was perfused over the chip) was limited to 300 s, and the flow rate was set at 30 l/min. For chip surface regeneration, a 10 mM glycine, pH 2.0 solution was used to dissociate the bound TNF at the end of each experiment while retaining surface integrity.
Accession Code-The coordinates and structural factors of infliximab Fab in complex with TNF␣ were deposited in the Protein Data Bank under accession code 4G3Y.

RESULTS
Overall Structure of the TNF␣-Infliximab Fab Complex-To elucidate the mechanism of TNF␣ inhibition through the therapeutic antibody infliximab, the Fab fragment of infliximab and functional TNF␣ were co-purified and crystallized. The crystal structure of the TNF␣-infliximab Fab complex was determined using the molecular replacement method and refined to 2.6-Å resolution with a final R work value of 19.4% (R free ϭ 23.9%) in space group H3 (Table 1). Although there is only one TNF␣infliximab Fab complex molecule in one asymmetric unit, the structure revealed a central TNF␣ trimer bound by three symmetrically arranged infliximab Fab molecules related through a crystallographic 3-fold axis (Fig. 1). This observation is analogous to the structures of TNF␣-TNFR2 (26) and TNF␤-TNFR1 (8) complexes and TNF␣ in complex with other proteins (38), indicating a 3:3 molar ratio for TNF␣ and infliximab Fab and consistent with the results of the gel filtration and analytical ultracentrifugation (data not shown here).
Only one TNF␣ molecule is present per asymmetric unit together with one bound infliximab Fab, but three TNF␣ molecules form a triangular conelike homotrimer associated through a crystallographic 3-fold axis (Fig. 1). Each TNF␣ molecule contains two packed antiparallel eight-stranded ␤-sheets, one inner and one outer, in a ␤-jelly roll topology as well as three additional N-terminal ␤-strands. The inner sheet, hidden in the trimer complex, is formed by strands BЉ-B-I-D-G in the correct spatial order, whereas the exposed outer sheet is formed by strands CЈ-C-H-E-F. Leu-29, Arg-31, Ser-52, and Tyr-56, which are crucial for TNF␣ cytotoxicity and TNFR binding affinity (2), were confirmed to have the correct conformation by cytotoxicity as described previously (28) (supplemental Fig. S1). Moreover, superimposing the TNF␣ in the TNF␣-infliximab Fab complex with wild-type TNF␣ yielded a root mean square deviation (r.m.s.d.) of 1.4 Å for the C ␣ atoms of all residues and indicated no significant overall structural difference between free TNF␣ and TNF␣ in complex except for the residues at the antibody-antigen interface.
The E-F loop of TNF␣ in the TNF␣-infliximab Fab complex that plays a crucial role in the antigen-antibody interaction is well ordered and defined by unambiguous electron density (supplemental Fig. S2). However, this region in the TNF␣-TNFR2 complex is totally unobservable (26), suggesting a flexible conformation and lack of interaction in the TNF␣-TNFR2 complex and indicating a different role of the E-F loop in antibody or receptor binding. Moreover, the E-F loop region in the TNF␣-infliximab Fab complex structure displays an extremely large r.m.s.d. value for C ␣ atoms compared with the free form, indicating that a large conformational change occurs in the loop when TNF␣ binds with the antibody, whereas a homologous loop region is completely absent in TNF␤ (Fig. 2).
is the mean of multiple observations I ih of a given reflection h. c R work ϭ ⌺ʈF p (obs)͉Ϫ͉F p (calc)ʈ/⌺͉F p (obs)͉; R free is an R factor for a selected subset (5%) of reflections that was not included in prior refinement calculations.
The infliximab Fab molecule presents a canonical immunoglobulin fold consisting of four ␤-barrel domains. The light chain is composed of residues Asp-1 to Cys-214, which fold into the V L and C L domains, and the heavy chain residues Glu-1 to Thr-226 fold into the V H and C H domains (except for the last six residues in the C terminus of the heavy chain that are missing because of a lack in density, which indicates a disordered and flexible conformation). Ala-51 L , which is located at the classical ␥-turn in the immunoglobulin family, displays a disallowed stereochemical geometry similar to its counterparts in other reported Fab structures (39). Intramolecular disulfide bonds are found in the expected positions for typical immunoglobulin Fab molecules: two between Cys-23 L /Cys-88 L and Cys-134 L / Cys-194 L and two between Cys-22 H /Cys-98 H and Cys-147 H / Cys-203 H . The complementarity-determining regions (CDRs) of the infliximab Fab have an ordinary length without unusual residues according to a Kabat sequence database search (40). The elbow angle of the infliximab Fab, defined as the angle subtended by the two pseudo 2-fold axes relating V H to V L and C H to C L , is 168°in the TNF␣-infliximab Fab complex.
Interactions between TNF␣ and Infliximab-The infliximab Fab interacts with only one TNF␣ molecule of a TNF␣ trimer in the complex structure through a large and highly complementary interface ( Figs. 2A and 3), which is consistent with the high affinity between infliximab and TNF␣. The total buried surface area between infliximab Fab and TNF␣ is 1,977 Å 2 of which TNF␣ contributes 1,035 Å 2 and the light and heavy chains of infliximab Fab contribute 450 and 600 Å 2 , respectively. This is also larger than typical protein-protein interfaces (1,560 -1,700 Å 2 ) (41). The interaction is demonstrated by a high shape complementarity value (42) of 0.72 (compared with the average shape complementarity value of 0.64 -0.68 for antibody-antigen complexes). The comparison between the interfaces of TNF␣-infliximab Fab and TNF␣-TNFR2 indicated that the interface of TNF␣-infliximab Fab overlaps with the TNFR2binding site, thus allowing infliximab to inhibit TNF␣ function.
The interface on TNF␣ is primarily composed of the C-D and E-F loop residues as well as several key residues in strands C and D that interact with Ile 28 -Trp 33 (in CDR H1), Arg 52 -Asn 57 (in CDR H2), and Tyr 102 -Ser 105 (in CDR H3) in the heavy chain of infliximab Fab. The TNF␣ G-H loop and additional residues in the C-D loop also bind to His 92 -Trp 94 (in CDR L3) and several other residues in the antibody light chain (e.g. Ser-32 in CDR L1 and Tyr-50 in CDR L2) (Fig. 3). There are over 30 pairs of interactions, including hydrogen bonds, salt bridges, and van der Waals contacts, that connect the molecules of TNF␣ and infliximab Fab in their complex; this indicates a strong and stable interaction between these two proteins and may account for their high binding affinity ( Table 2).
Several distinct differences were found by superimposing the TNF␤-TNFR1 or TNF␣-TNFR2 complex structure onto the TNF␣-infliximab Fab complex structure (Fig. 2). First, TNF␣ residues Glu 67 -His 73 and Thr 105 -Lys 112 , which are located in the C-D and E-F loops, mostly contribute to the interaction between TNF␣ and infliximab Fab. Although the overall folding of TNF␣ and its binding with infliximab Fab display nearly the same conformation with an r.m.s.d. of 1.43 Å for 157 C ␣ atoms, the E-F loop moves outward with the C-D loop that moves toward the infliximab Fab molecule to accommodate the V H and V L domains of the antibody (Fig. 4). Additionally, the G-H loop displays another slight shift, but it may not be related to the antibody interaction. In contrast, only a few residues in the C-D loop of TNF␤ bind to TNFR, and the key E-F loop region of TNF␣ is shorter in the TNF␤ amino acid sequence (Fig. 2B), which is consistent with the absence of this loop region in the TNF␤ structure. Furthermore, the E-F loop is completely miss-

FIGURE 2. Comparison of the interface between TNF␣ and receptors and infliximab Fab.
A, TNF␣ from the complex structures is represented as a colored surface with TNFR2 and the infliximab Fab interface highlighted in red at one of three interfaces on the TNF␣ trimer. The E-F loop region, which is missing in the TNF␣-TNFR2 complex because of the lack of interaction, is labeled. The TNF␤ from the TNF␤-TNFR1 complex structure is shown as a colored surface with one of the TNFR1-binding sites highlighted in red. All TNF molecules are superposed and presented in the same orientation. B, the amino acid sequence alignment of TNF␣ and TNF␤. The E-F loop, which may play a central role in antibody-antigen interaction, is framed. The numbering of residues (top) refers to that in TNF␣.
ing in the TNF␣-TNFR2 crystal structure, which indicates that the E-F loop has a flexible conformation and does not participate in the interaction between TNFR2 and TNF␣ (43). Combined with previous data that indicate that infliximab does not affect the function of TNF␤ (24), the TNF␣ E-F loop may play a central role in the specific interaction between infliximab and TNF␣ but not TNF␤.
The results of the in vitro binding assay revealed that TNF␣ E-F loop replacement mutants with the residues GGGG (named TNF␣(EF-4G)) and SGGSGGSGGSG (named TNF␣(EF-11SG)), distinctively increased the K D value 10 3 -fold over the wild-type TNF␣ (Table 3 and supplemental Fig. S3). This suggests that the E-F loop mutants can decrease the binding affinity and play essential roles in the interaction of TNF␣ with infliximab Fab. Nonetheless, the TNF␣(Q67A) and TNF␣(K112A) mutants, which are located in the C-D and E-F loops, respectively, only slightly decreased the binding affinity with infliximab Fab. This indicates that the C-D and E-F loops contribute to the interaction with the antibody through a molecular network but not via individual residues.
The G-H loop residues Asn 137 -Tyr 141 together with Thr-77 in strand D of TNF␣ interact with the infliximab Fab, consistent with the TNF␤-TNFR1 interaction (8). Arg-138 provides two ideal hydrogen bonds with Ser-91 L and His-92 L from the light chain and, thus, contributes to the interaction between TNF␣ and the antibody. Tyr-141, which extends its side chain toward the heavy chain of infliximab Fab, provides a large hydrophobic interface (via its side chain) as well as hydrogen bonding with Trp-33 H and Arg-52 H to stabilize the TNF␣-infliximab Fab complex.
In concordance with the crystallographic analysis, the results of the in vitro binding assays demonstrate that TNF␣(R138A) and TNF␣(Y141A) mutations significantly decrease the binding of TNF␣ to infliximab (Table 3). In sharp contrast, the region between A and AЈ, which participates in TNF␤-TFNR1 binding, is not involved in the TNF␣-infliximab Fab interaction. Moreover, TNF␣ residues Arg-31, Arg-32, and Tyr-87, which are crucial for binding both TNFRs (26,44), are not involved in the TNF␣-infliximab Fab interface, indicating the different binding behavior of TNF␣ with receptors and antibodies. Furthermore, the groove between two adjacent TNF␣ subunits is crucial for the interaction between TNFs and TNFRs (8,26,38). However, similar structural features are not found in the TNF␣-infliximab Fab complex structure. These data indicate that the interaction between TNFs and their receptors or antibodies are likely more complicated than previously suggested based on the structural analysis of TNF␣ with its receptors (8).  Molecular Mechanism of TNF␣ Inhibition by Infliximab-Although the molecular coordinates of TNF␣ and TNF␤ that are bound with TNFRs provide an understanding of the mechanism of TNF function, the lack of structural information on TNF␣ bound to therapeutic antibodies hinders the elucidation of the precise epitope and clear inhibition mechanism of infliximab despite the fact that infliximab therapy has been used for TNF␣-associated diseases for over 10 years. In the TNF␣-TNFR2 and TNF␤-TNFR1 structures, the cytokine-receptor interface can be conventionally characterized into upper and lower regions, which primarily focus on the D-E and A-AЉ regions, respectively. In the TNF␣-infliximab Fab structure, residues Gln 67 -His 73 and Gln 102 -Lys 112 in the TNF␣ C-D and E-F loops are largely responsible for the antibody-antigen interaction, whereas Asn 137 -Tyr 141 in the G-H loop and Thr-77 in strand D of TNF␣ complementarily contribute to this interaction.
The solvent-accessible surface contributed by these interactions covers over 60% of the total interface between TNF␣ and TNFR1, which indicates an overlap between the TNF␣ receptor-binding sites and the infliximab epitope. Moreover, although several other residues crucial for TNF␣-receptor binding do not participate in the TNF␣-infliximab Fab interface (especially the groove between two associated TNF␣ molecules in the TNF␣ trimer), the peak region of the cone of the TNF␣ trimer appears to largely contribute to the interaction with infliximab Fab.
These results may explain in part the biochemical data concerning the binding avidity or affinity of infliximab to soluble or membrane-associated TNF␣ (affinity for soluble TNF␣ is 27 pM, and avidity for membrane-associated TNF␣ is 0.45 nM) (23) compared with the binding avidity of TNFR1 to TNF␣ (0.38 nM) (43). Therefore, the binding of infliximab to TNF␣ efficiently competes with TNFRs binding to TNF␣, and the interface between TNF␣ and TNFRs is blocked with sufficient amounts of infliximab, thereby preventing TNF␣ to function further in diseases. Nonetheless, the exact TNF␣-receptor interface has still not been elucidated according to current structural investigations (8,(45)(46)(47). However, our data suggest that infliximab blocks the TNF␣-TNFR interaction by occupying part if not the same TNF␣ binding interface.

DISCUSSION
TNF␣ is an inflammatory cytokine that is predominantly produced by activated macrophages and lymphocytes, plays a central role in acute inflammation, and is responsible for a diverse range of signaling events within cells that lead to necrosis or apoptosis (1,2). Therefore, the inhibition of TNF␣ is a validated and favorable method for treating several important TNF␣-associated diseases. Currently, several receptors are known to interact with TNF␣ and, thus, play a key role in TNF␣-associated diseases. Therefore, an extensive range of TNF␣-inhibitory proteins, most of which are based on an antibody scaffold, have been developed and used with variable success as therapeutic agents to block the interaction between TNF␣ and its receptors (48).

TABLE 3 Kinetics and binding of TNF␣ mutants with infliximab Fab
Kinetics and binding of TNF␣ mutant and infliximab Fab were analyzed using a BIAcore T100. Wild-type TNF␣ and the mutants were passed over the immobilized infliximab Fab surface, and the data were globally analyzed using a simultaneous fit for both dissociation (k d ) and association (k a ). The value for K D was calculated as k d /k a . Typical error levels for the k d and k a values are less than Ϯ15%. Infliximab is a therapeutic mAb that was approved by the United States Food and Drug Administration to treat Crohn disease, ankylosing spondylitis, psoriatic arthritis, rheumatoid arthritis, and ulcerative colitis. However, because infliximab is a chimeric mAb and its use is not very well tolerated in the majority of patients, infliximab therapy leads to the production of antibodies to infliximab in a small subset of patients (49 -52). Increasing the human sequence content by grafting murine CDRs may be crucial for the integral capacity of antigen binding and should be retained during humanization (53). Therefore, structural evidence concerning the TNF␣-infliximab Fab interface could provide direct information for anti-TNF␣ antibody humanization. Together with infliximab, adalimumab is another widely used (and the first fully human) therapeutic mAb for treating TNF␣-associated diseases; it was approved by the United States Food and Drug Administration in 2008 (54). Although adalimumab has a mechanism similar to that of infliximab for treating Crohn disease, rheumatoid arthritis, psoriatic arthritis, and ankylosing spondylitis, the different binding avidities of infliximab (4.2 pM) and adalimumab (8.5 pM) for TNF␣ (23) suggest variable antigen-antibody interfaces.
Given that the binding affinity between an antibody and an antigen is one of the most important determinants for therapeutic antibody development, improving the surface complementarity of the interface between the antibody and the antigen, strengthening the interaction, and, thus, enhancing the binding affinity through mutagenesis of the paratope of the antibody are of particular interest. Although the interface between infliximab Fab and TNF␣ has high complementarity with an shape complementarity value of 0.72, the complex structure reported here still provides valuable information for enhancing their binding affinity.
First, the side chain of Gln-67 in the C-D loop of TNF␣ interacts with the acid cavity formed by the side chains of Ser-53 H and Ser-55 H with relatively long distances (Ͼ3.4 Å). Therefore, the substitution of Ser-53 H and Ser-55 H with long side chain acidic residues (e.g. aspartate) may provide more favorable interactions with Gln-67 TNF . Second, the side chain of Trp-94 L provides one van der Waals contact with TNF␣ residues, mostly with highly charged side chains (e.g. aspartate and arginine), which suggests that the substitution of Trp-94 L with a long side chain charged residue may result in more and better interactions with TNF␣. Several other substitutions, for example S91 L D, S93 L D, and N31 H Q, could potentially increase the interaction (and thus increase the binding) between infliximab and TNF␣. Nonetheless, the complexity of antibody-antigen interaction requires further testing and validation. Moreover, because the E-F loop is the most divergent portion between TNF␣ and TNF␤ (in both amino acid sequence and three-dimensional structure) and may play a central role in the specific interaction between TNF␣ and infliximab, improving the E-F loop-interacting region is crucial for increasing the mAb binding and avoiding the side effects caused by interacting with TNF␤ in host cells. Notably, although E-F is essential for the binding of infliximab to TNF␣, there is no evidence to show that this fragment is important to the biological function of TNF␣. The structure of TNF␣-TNFR2 also revealed that E-F loop does not participate in the binding to TNFRs (26). There-fore, the binding of infliximab to the TNF␣ E-F loop is not likely to directly impact the function of TNF␣ but only spatially affect the communication between TNF␣ and TNFRs.