Neutralization of Macrophage Migration Inhibitory Factor (MIF) by Fully Human Antibodies Correlates with Their Specificity for the β-Sheet Structure of MIF*

Background: A diverse panel of fully human antibodies specific for the macrophage migration inhibitory factor (MIF) has been generated. Results: In vitro and in vivo studies revealed that antibodies specific for a β-sheet structure are potent inhibitors of MIF. Conclusion: This β-sheet structure is a promising target for anti-MIF antibody therapy. Significance: Fully human antibodies with high therapeutic potential have been identified. The macrophage migration inhibitory factor (MIF) is a proinflammatory cytokine that recently emerged as an attractive therapeutic target for a variety of diseases. A diverse panel of fully human anti-MIF antibodies was generated by selection from a phage display library and extensively analyzed in vitro. Epitope mapping studies identified antibodies specific for linear as well as structural epitopes. Experimental animal studies revealed that only those antibodies binding epitopes within amino acids 50–68 or 86–102 of the MIF molecule exerted protective effects in models of sepsis or contact hypersensitivity. Within the MIF protein, these two binding regions form a β-sheet structure that includes the MIF oxidoreductase motif. We therefore conclude that this β-sheet structure is a crucial region for MIF activity and a promising target for anti-MIF antibody therapy.

The macrophage migration inhibitory factor (MIF) 2 was described as early as 1966 (1, 2) as a soluble mediator that inhibits migration of peritoneal exudate cells. The biochemical properties and physiological role of MIF were elucidated following its cloning and recombinant expression (3,4). It is now well accepted that MIF is a pivotal regulator of innate immunity, playing a central role in inflammatory responses. MIF promotes the production of other proinflammatory mediators, such as TNF␣ (5), nitric oxide (4), and prostaglandin E 2 (6,7). One of its most striking properties is the ability to override the immunosuppressive effects of glucocorticoids (GCs). In vitro, MIF counteracts the GC-induced inhibition of cytokine secretion in monocytes (TNF␣, IL-1, IL-6, and IL-8) (8) and T-cells (IL-2 and IFN-␥) (9) and overrides dexamethasone suppression of TNF␣-induced arachidonic acid release in fibroblasts (6). In vivo studies revealed that MIF increases the mortality of endotoxemic mice treated with dexamethasone (8). MIF furthermore contributes to the maintenance of the inflammatory processes by inhibiting p53-dependent cell death and stimulating the survival of monocytes and macrophages (10). In addition to this anti-apoptotic activity, MIF exhibits pro-proliferative properties by activating ERK1/ERK2 signaling (6).
Some biological features of MIF are markedly different from those of other proinflammatory cytokines. It is constitutively expressed in both immune and non-immune cells, and its tissue distribution is almost ubiquitous. Preformed MIF is stored in cytoplasmic pools of macrophages, T-cells, and many other cells within the body, including the hypothalamic-pituitary-adrenal axis, allowing for rapid release upon stimulation without de novo synthesis (3,11,12). MIF is present in the circulation of healthy people in plasma or serum concentrations in the range of typically 1-15 ng/ml (13)(14)(15). Two distinct enzymatic activities, a tautomerase (16,17) and an oxidoreductase (18) activity, have been assigned to the MIF molecule. Both have been described as possibly responsible for certain MIF-mediated immune processes (18 -20), but no natural substrate for MIF has yet been reported.
The role of MIF in acute infections and chronic inflammatory diseases has been assessed by correlating the increased MIF levels in plasma and tissue with disease severity. The most detailed data for the up-regulation of MIF serum levels and its association with disease were described in patients with severe sepsis (14,21,22). Plasma levels of MIF correlated with disease severity and a state of shock and were significantly higher in patients who died than in those who survived. MIF concentrations significantly correlated with elevated plasma concentrations of IL-1␤, IL-6, IL-10, IL-12, and cortisol. Elevated MIF levels in patients have furthermore been determined for numerous inflammatory diseases, e.g. rheumatoid arthritis (23,24), Crohn disease (25), psoriasis (26), and multiple sclerosis (27,28). A second body of evidence for the importance of MIF in the development of certain diseases has come from studies using MIF knock-out mice. Although MIF knock-out mice do not show a severe deficit, they have a reduced susceptibility to experimental sepsis (29), arthritis (30), inflammatory bowel disease (25), and organ injury caused by systemic lupus erythematosus (31). In addition, neutralizing anti-MIF polyclonal and monoclonal antibodies have been demonstrated to have beneficial effects in animal models of experimental sepsis and septic shock (e.g. Refs. 3, 13, and 32; reviewed in Ref. 33) and in animal models of chronic inflammation and autoimmune diseases, including delayed-type hypersensitivity (34), arthritis (35), inflammatory bowel disease (25), and other disease models (reviewed in Refs. 36 and 37).
In summary, MIF has emerged in recent years as an attractive new target for treating diseases with a high unmet need, such as sepsis, autoimmune disorders, and chronic inflammation. We therefore set out to develop fully human antibodies specific for MIF and to screen for antibodies with high in vivo therapeutic potential.

EXPERIMENTAL PROCEDURES
Reagents-The cDNAs of human MIF (huMIF) and mouse MIF (moMIF) were generated from poly(A) RNA obtained from human (Clontech, Mountain View, CA) or mouse (Stratagene, San Diego, CA) liver by reverse transcription. The MIFencoding genes were amplified and cloned into the pET16b expression vector (Novagen, Madison, WI) using standard techniques. Recombinant huMIF or moMIF was expressed after transformation of the respective vector into Escherichia coli BL21 (Stratagene) as described (4). Recombinant MIF was purified after lysing the cells and removing cell debris either with a refolding step (4) or without a refolding step (38) as described previously. Both purification methods yielded equivalent products. Biotinylation of recombinant MIF was carried out using an ECL protein biotinylation kit (GE Healthcare). Biotinylated MIF-derived peptides (see Fig. 1) were synthesized by Jerini AG (Berlin, Germany).
Antibody Selection by Phage Display-The Dyax FAB310 library (39) was used for selection of MIF binders in nine different selection campaigns. In the first campaign, phage were selected on biotinylated huMIF immobilized on streptavidin beads. The second campaign used huMIF coated onto Max-iSorp ELISA plates (NUNC A/S, Roskilde, Denmark). In the third campaign, selection was carried out by alternating biotinylated moMIF and biotinylated huMIF. In six additional campaigns, selection was performed by alternating a biotinylated MIF-derived peptide (see Fig. 1) and biotinylated huMIF. 1000 clones from each campaign were randomly selected after three or four panning rounds of amplification and then tested for binding to huMIF by phage ELISA as described (39).
Antibody Production-Fab fragments were recloned into an IgG format (IgG1 and IgG4 isotypes) using standard cloning techniques. Fully human IgGs were expressed in HEK293T cells (GenHunter, Nashville, TN) after transfection of the cells using GeneJuice reagent (Novagen). Transfected cells were cultivated for 3 weeks in DMEM supplemented with 10% ultra low IgG fetal bovine serum, 1% antibiotic/antimycotic solution, 0.1% gentamycin, and 1% nonessential amino acids (all from Invitrogen, Carlsbad, CA), during which the medium was renewed twice a week, and the conditioned medium was pooled. Antibodies were purified from the cell culture supernatant by purification over protein A (GE Healthcare) and dialyzed against PBS (Invitrogen).
Epitope Mapping-Each of the six MIF-derived peptides (see Fig. 1) was diluted in coupling buffer (NUNC A/S) to give a peptide concentration of typically 5 g/ml and was added to microplates (NUNC Immobilizer TM F96 clear amino plates). huMIF and PBS were applied as controls, and the plates were incubated overnight. After the plates had been washed, the anti-MIF antibodies (4 g/ml in PBS) were added and incubated for 2 h at room temperature with gentle shaking. After further washing, HRP-labeled Fc-specific anti-human IgG (Sigma) was added. The plates were incubated and washed, and bound anti-MIF antibody was detected by adding 3,3Ј,5,5Ј-tetramethylbenzidine solution (Sigma) and stopping the reaction by adding H 2 SO 4 after 30 min. Plates were measured at 450 nm.
Inhibition of the GC-overriding Activity of MIF-The inhibition of the GC-overriding activity of MIF was tested using a modification of the assay described previously (19). THP-1 cells were purchased from American Type Culture Collection (ATCC TIB-202) and cultivated according to the supplier's recommendations. Exponentially growing THP-1 cells were centrifuged and resuspended in fresh medium to a cell density of 10 6 cells/ml. This culture was transferred into the wells of a 96-well microplate, and anti-MIF antibodies were added. The final concentration of the antibody for screening purposes was 75 g/ml (500 nM), and increasing amounts of antibody were used (typically 0.5-30 nM) to determine a dose-response curve. After overnight incubation at 37°C, dexamethasone (Sigma) was added to give a final concentration of 2 nM. O111:B4 LPS (Sigma) was added after an additional 1 h to give a final concentration of 3 ng/ml, and after an additional 6 h of incubation, the supernatant was harvested, and the IL-6 concentrations were determined using a human IL-6 CytoSet TM ELISA kit (BIO-SOURCE, Camarillo, CA). All experiments were performed in triplicates, and the mean was used for further evaluations.
Inhibition of MIF-mediated Cell Proliferation-This assay was done as described previously (6). Briefly, NIH/3T3 fibroblasts were purchased from American Type Culture Collection (ATCC CRL-1658) and cultivated according to the supplier's recommendations. Cells (1000/well) were incubated in a 96-well plate with medium containing 10% heat-inactivated calf serum (PAA, Pasching, Austria) for 3 days at 37°C. Cells were then starved overnight at 37°C by incubation in medium containing 0.5% calf serum. The 0.5% serum-containing medium was removed and replaced by fresh medium containing 10% calf serum, 75 g/ml antibody, and 5 Ci/ml [ 3 H]thymidine (GE Healthcare). After 16 h of incubation at 37°C, cells were washed twice with cold PBS, and a 5% (w/v) trichloroacetic acid solution was added. The plates were incubated for 30 min at 4°C and washed with PBS, after which 75 l of a 0.5 M NaOH solution with 0.5% SDS were added per well. These 75-l samples were added to 5 ml of Ultima Gold scintillation mixture (PerkinElmer Life Sciences) and measured in a ␤-counter. Each determination was done in triplicates, and the mean was used for further calculations.
Tautomerase Activity Inhibition Assay-The antibodies were tested for inhibition of tautomerase activity using a modification of the assay described by Bendrat et al. (17). Briefly, equal volumes of L-3,4-dihydroxyphenylalanine methyl ester (0.6 mg/ml) and potassium periodate (1 mg/ml) were mixed to obtain L-dopachrome methyl ester. Recombinant huMIF and the anti-MIF antibody were mixed in a 1:1 molar ratio and preincubated. 2 l of the preincubated mixture containing 50 ng of huMIF and 600 ng of anti-MIF antibody were added to the wells of a 96-well plate, and the reaction was started by the addition of 100 l of L-dopachrome methyl ester. The decrease in absorbance was measured at 490 nm for 3 min.
Measurement of Affinity by Surface Plasmon Resonance-A Biacore TM T200 device and evaluation software (version 1.0), sensor chips, buffer, and reagents were obtained from GE Healthcare. huMIF or moMIF was immobilized onto C1 sensor chips by amine coupling to give a loading density of 1 response unit. Anti-MIF antibodies were injected at a concentration range of typically 0.625-10 nM (diluted in HEPES-buffered saline containing EDTA and 0.5% (v/v) Surfactant P20). After each cycle, the chip was regenerated with 10 mM glycine-HCl buffer (pH 1.7). Affinity constants were calculated according to the 1:1 Langmuir model.
Endotoxic Shock Model-The experiments were performed as described (40). Briefly, 8-week-old female BALB/c mice (Charles River) were pretreated with an intraperitoneal injection of 0.5 mg of antibody and challenged interperitoneally 1 h later with a mixture of 250 ng of O111:B4 LPS and 20 mg of D-galactosamine (Sigma). Mice were bled for TNF␣ and IL-6 measurements 1 h after LPS challenge. The plasma levels of TNF␣ and IL-6 were assessed by bioassays as described (41).
E. coli Peritonitis Sepsis Model-The experiments were performed as described (13). Briefly, E. coli O111:B04 was grown overnight, and the culture was diluted in physiological saline to the required cell concentration prior to application. Female NMRI mice were challenged by intraperitoneal injection with a suspension of 6000 colony-forming units of E. coli O111:B4, 15% mucin, and 4% hemoglobin. The anti-MIF antibodies or an isotype-matching control antibody was injected intraperitoneally 2 h before bacterial challenge. Typically 20 mice were used for each group. Survival was monitored for 48 h, and Kaplan-Meier statistics were used to evaluate the survival curves.
Animal Models for Contact Hypersensitivity (CHS)-The CHS response to 2,4-dinitrofluorobenzene (DNFB) was determined using a modification of the model described previously (42). C57BL/6 mice were sensitized by treating the shaved mouse back skin with 75 l of DNFB (Sigma) solution (0.5% in 4:1 acetone/olive oil). On day 5, mice were treated with anti-MIF antibodies, isotype control antibody, or saline by intravenous injection, and 30 min later, the right ears of the mice were challenged with 20 l of 0.3% DNFB to cause a local inflammatory response. The vehicle (acetone/olive oil) was applied to the left ears as a control. The swelling response to DNFB, specified as the difference between the right and left ears, was measured 24 h after challenge using a micrometer (Mitutoyo Co., Tokyo, Japan).
Adoptive transfer experiments were performed as described (43). Briefly, C57BL/6 mice were treated intravenously with either an isotype control antibody or anti-MIF antibodies. Mice were sensitized 30 min later by painting 70 l of DNFB solution (0.5% in 4:1 acetone/olive oil) onto the skin on the back. On day 5, single cell suspensions were prepared from the spleens and regional lymph nodes of DNFB-sensitized mice and injected intravenously into naïve recipient mice. Recipients were exposed to 0.3% DNFB on the right ear and vehicle on the left ear. Ear swelling was measured 24 h later using a micrometer. H&E staining was done according to standard methods. Cryostat sections were incubated with an antibody specific for CD3 (polyclonal rabbit, DakoCytomation, Glostrup, Denmark), neutrophils (clone 7/4, Serotec, Oxford, United Kingdom), monocytes/macrophages (clone F4/80, Serotec), or an isotype control (clone LO-DNP-11, Serotec). Staining was visualized by the addition of appropriate secondary antibodies and using a fast red substrate kit (DakoCytomation).

Generation of Diverse Antibody Panel Specific for MIF-A
highly diverse panel of 145 unique MIF-specific antibodies was selected from a phage library displaying 3.5 ϫ 10 10 different Fab fragments (39). Either recombinant full-length MIF or six MIFderived overlapping peptides spanning the entire MIF sequence were used as antigens for phage panning. After three to four rounds of phage selection, all unique MIF binders identified by phage ELISA and sequencing were cloned into a mammalian expression vector, expressed as fully human IgG4 in HEK293T cells, and purified for further testing. The MIF-derived overlapping peptides used for phage panning were also used to determine the binding regions of the antibodies. A total of 74 antibodies bound to full-length huMIF but did not recognize any of the peptides and were therefore classified as specific for structural epitopes. However, another 71 antibodies recognized either one peptide or two overlapping MIF-derived peptides and were therefore classified as specific for a linear epitope. Antibodies specific for each of the nine linear binding regions indicated in Fig. 1 were identified. Thus, antibodies specific for almost every part of the primary amino acid sequence of MIF were available. This antibody panel allowed the MIF molecule to be scanned for regions important for in vivo activity in subsequent in vitro assays and in vivo disease models.
Each of the 145 IgG4 antibodies was subjected to three different functional assays to classify the antibodies according to their potential to inhibit MIF activity in vitro. The antibodies were tested for their ability to inhibit MIF-dependent cell proliferation, inhibit the GC-overriding activity of MIF, and inhibit the tautomerase activity of MIF. Antibodies that were able to reduce cell proliferation and GC-overriding activity (i.e. IL-6 secretion) significantly by Ͼ25% were classified as screening hits. A total of 15% (11 of 71) of the antibodies that recognized a linear epitope showed MIF-neutralizing properties in both cell-based assays (Fig. 1). The results suggest that the ability to neutralize MIF activity in vitro correlates with the binding region on the molecule. 27% (3 of 11) of the antibodies specific for the region comprising amino acids (aa) 46 -68 and 32% (8 of 25) of the antibodies binding to the region including aa 86 -115 were able to neutralize MIF in both cell-based assays. Antibodies binding to linear epitopes spanning aa 2-45 and 69 -85 of the MIF molecule did not show MIF-neutralizing activity in both of these assays and were therefore not classified as hits (Fig. 1). The third assay identified antibodies capable of inhibiting the tautomerase activity of MIF. MIF is able to catalyze the conversion of dopachrome to 5,6-dihydroxyindole-2-carboxylic acid (16,17), and antibodies affecting the tautomerase activity reduced the conversion rate. However, none of the antibodies specific for a linear epitope could affect this enzymatic activity. Only eight anti-MIF antibodies specific for a structural epitope were able to significantly reduce tautomerase activity, three of which also showed MIF-neutralizing properties in both cell-based assays. In total, 14 anti-MIF antibodies that were identified as recognizing a structural epitope were able to inhibit MIF-dependent cell proliferation as well as the GC-overriding activity of MIF.
To confirm the correlation between the binding region and functional activity, we selected a subset of antibodies, in which each of the nine binding regions depicted in Fig. 1 was repre-sented, for further in vitro and in vivo evaluation. The in vitro characterization comprised a dose titration for inhibition of the GC-overriding activity of MIF that allowed the maximum inhibition achievable by the antibodies (Inh max ) to be determined as well as the antibody concentration that resulted in 50% of the Inh max (IC 50 ) ( Table 1). In addition, ELISA (data not shown) and Biacore experiments (Table 1) showed that all of the antibodies that were investigated further bound similarly to immobilized recombinant moMIF and huMIF. Table 1 summarizes the results of the additional characterization studies for the antibodies that gave a significant reduction in MIF-mediated cell proliferation as well as a dose-dependent reduction in the GC-overriding activity of MIF. Based on the in vitro data, these antibodies were considered to have the most pronounced MIFneutralizing activity. The affinities (K D ) of the antibodies were found to be in the 1 nM range. The antibodies further showed IC 50 values of between 1 and 7 nM. Both data (K D and IC 50 ) demonstrated that the antibodies were highly efficient in binding and neutralizing MIF in vitro.
In Vivo Testing of Candidate Antibodies in Experimental Models of Sepsis-IgG4 specific for each of the nine binding regions depicted in Fig. 1 were tested in experimental mouse models of sepsis, i.e. an E. coli peritonitis model (13) and an endotoxic shock model (40). Antibody BaxB01, specific for the region spanning aa 50 -68, and antibody BaxG03, specific for the region spanning aa 86 -102, were able to significantly FIGURE 1. Epitope specificity of anti-MIF antibodies isolated from phage display library. The six MIF-derived and overlapping peptides depicted were used for phage panning and epitope mapping. Nine binding regions are indicated by braces with the corresponding amino acid range. The amino acid range of the binding regions marked with asterisks is an approximation, as the antibodies are specific for the central region of the respective peptide and do not bind to an overlapping region. The arrows show the percentage of antibodies specific for the region indicated that were classified as antibody hits. The absolute number of antibodies that were classified as hits and the overall number of antibodies specific for the particular region are indicated in parentheses. The amino acid sequence of MIF was taken from Swiss-Prot (P14174). The N-terminal amino acid (Met) was omitted, as it is cleaved during processing of the MIF molecule (63).

TABLE 1 Antibodies showing the most pronounced MIF-neutralizing properties in vitro
Antibodies capable of inhibiting MIF-mediated proliferation and the GC-overriding activity of MIF bound exclusively to either the center of the molecule (aa 46 -68) or the C terminus (aa 86 -115). k a is the association rate constant, and k d is the dissociation rate constant. ND, not determined.

Fully Human Anti-MIF Antibodies
reduce TNF␣ and IL-6 levels released in the endotoxic shock model (Fig. 2). BaxG03 and BaxD08, another antibody binding to the region spanning aa 86 -102, increased survival significantly in the E. coli peritonitis sepsis model (Fig. 3). Interestingly, antibodies specific for regions other than aa 50 -68 and 86 -102 failed to give significant results in any of the sepsis models. Antibodies BaxH08 (specific for aa 46 -55) and BaxF07 (specific for aa 98 -115) also failed to show significant improvement in vivo, although they showed considerable MIF-neutralizing capacity in vitro. We also tested antibodies that bound to structural epitopes, including antibodies capable of inhibiting the tautomerase activity of MIF, in the sepsis models. However, these antibodies failed to give any beneficial effect in comparison with an isotype control. Our results suggest that antibodies directed against aa 50 -68 and aa 86 -102 of the MIF molecule are particularly effective in vivo.
To investigate whether the efficacy of an antibody can be improved by enhancing the affinity, we carried out an affinity maturation of BaxG03 by generating modified versions of this antibody through complementarity-determining regions 1 and 2 and light chain shuffling (39,44). High affinity variants were selected by phage display using limiting concentrations of biotinylated antigen (45). This process allowed the isolation of an antibody showing the same epitope specificity as the parental antibody, BaxG03 (aa 86 -102), but a 10-fold higher affinity for huMIF. The affinity-matured antibody (designated BaxM159) showed superior therapeutic efficacy in experimental sepsis in the mouse endotoxic shock model (40), in which BaxB01, BaxG03, and BaxM159 were tested in parallel (Fig. 2). The most significant reduction in the proinflammatory response in relation to isotype control-treated mice was obtained after treatment with BaxM159, as mean TNF␣ levels were reduced by 96% (p Ͻ 0.001). BaxG03 led to a reduction in mean TNF␣ levels of 92% (p Ͻ 0.01), and BaxB01 to a reduction of 86% (p Ͻ 0.05) ( Fig. 2A). Similarly, in the same experiment, the most pronounced reduction in mean IL-6 levels was achieved in the BaxM159-treated group (reduction of 52%), whereas the BaxG03-and BaxB01-treated groups showed reductions of 32 and 39%, respectively (Fig. 2B). The enhanced protective effect of BaxM159 was confirmed in the live E. coli peritonitis model. Survival rates increased from 10% (control antibody) to 34% for BaxG03 (p ϭ 0.035) and 56% for BaxM159 (p ϭ 0.008) (Fig. 4). Notably, the increased survival was achieved with a lower dose of BaxM159 (300 g/mouse) than of BaxG03 (500 g/mouse).
The IgG4 isotype was applied for antibody screening, characterization, and in vivo testing in experimental models of sepsis. However, dose titrations for inhibition of the GC-overriding activity of MIF were also done using the IgG1 isotype of antibodies BaxB01 and BaxD08. BaxG03 was tested in the E. coli peritonitis sepsis model as an IgG4 and an IgG1. No differences between the IgG1 and IgG4 formats could be observed, ensuring that the in vitro and in vivo neutralizing activities of a human IgG1 and IgG4 anti-MIF antibody were identical.
In Vivo Testing of Candidate Antibodies in Models of CHS-The animal models described so far demonstrated a successful interference of the antibodies with the lethal cascade of events mediated by the release of MIF after a microbial trigger. To assess the efficacy of the fully human anti-MIF candidate antibodies identified in another model, we determined the protec-  tive potential of these antibodies in the T-cell-mediated immune response in two different animal models of CHS. The antibodies applied in this study were IgG1. First, we tested if antibodies BaxB01 (specific for aa 50 -68) and BaxM159 (specific for aa 86 -102) are able to suppress the effector phase of delayed-type hypersensitivity, i.e. accumulation of effector T-cells in DNFBsensitized mice after re-exposure of the ear to DNFB. We found that ear swelling, the primary measure in these experiments, was reduced significantly in anti-MIF antibody-treated mice compared with non-treated mice (Fig. 5A).
In adoptive transfer experiments, we examined if the anti-MIF antibodies BaxB01 and BaxM159 are able to inhibit the sensitization of the immune system and to interfere with the induction phase of CHS. We compared the CHS responses in naïve mice that received donor cells prepared from the spleens and regional lymph nodes of anti-MIF antibody-pretreated, DNFB-sensitized mice and in naïve mice that received donor cells prepared from control antibody-treated mice. We found that the CHS response in the mice that received the anti-MIF antibody was attenuated compared with the response in control mice, but it was not completely abrogated (Fig. 5B). Examination of ear biopsies showed that naïve mice injected with donor cells transferred from anti-MIF antibody-pretreated, DNFBsensitized mice developed a less pronounced inflammatory reaction. Accumulation of scattered extravascular leukocytes in the dermal/hypodermal tissue and interstitial edema were reduced compared with recipient mice injected with cells from isotype control-treated, DNFB-sensitized mice (Fig. 6). These results demonstrate that MIF is relevant in the induction phase of CHS and that neutralization of MIF affects the sensitization of the immune system.
In Vivo Protective Antibodies Show Common Three-dimensional Epitope-On the basis of the in vitro and in vivo results, we concluded that antibodies binding to the center of the primary MIF sequence (aa 50 -68) and those binding close to the C terminus (aa 86 -102) are protective in disease models. Surprisingly, competitive ELISA experiments (data not shown) demonstrated that the antibodies binding to aa 50 -68 and 86 -102 competed with each other for binding to immobilized MIF, although, on the linear sequence, the binding regions were  A, attenuated CHS reaction on the challenged ears of mice treated with anti-MIF antibody (20 mg/kg) compared with mice treated with an isotype control antibody or saline. Five mice were used per group, and reduced ear contact dermatitis was quantified by measuring ear thickness. B, adoptive transfer of CHS demonstrated that the CHS response was attenuated when donor cells were obtained from anti-MIF antibody (20 mg/kg)-pretreated, DNFB-sensitized mice compared with isotype control-pretreated mice. Six recipient mice were used per group. *, p Ͻ 0.05; **, p Ͻ 0.01 (using one-way analysis of variance). MARCH 2, 2012 • VOLUME 287 • NUMBER 10 clearly separated from each other. Analysis of the crystal structure of MIF revealed that binding regions comprising aa 50 -68 and 86 -102 are in close proximity in the three-dimensional space (Fig. 7). These regions are adjacent to each other and form a ␤-sheet structure containing the functionally important oxidoreductase motif (Cys 57 -Ala-Leu-Cys 60 ) (Fig. 7A). In the monomeric conformation, aa 50 -68 and 86 -102 assemble to form a patch on the MIF surface that mediates, at least partially, the biological properties and cytokine function of MIF. However, MIF crystallizes as a trimer (38,46,47). In a trimeric state, aa 50 -68 and 86 -102 form part of the central channel of the molecule (Fig. 7B).

DISCUSSION
We have reported herein the generation of a large and highly diverse panel of fully human anti-MIF antibodies. Phage display technology was used to select 74 antibodies recognizing a structural MIF epitope and 71 different antibodies recognizing linear epitopes. The binding specificities of the latter covered almost the entire primary sequence of the MIF molecule. In vitro char-acterization suggested a correlation between the specificity of an antibody for a certain binding region and its ability to functionally neutralize MIF activity. We found that only antibodies binding to the center of the linear MIF sequence (aa 46 -68) and antibodies with their epitope at the C terminus (aa 86 -115) were successful in inhibiting MIF activity in the cell proliferation inhibition and GC-overriding assays. These in vitro results are in line with previous reports showing the functional importance of the central region (18, 19, 48 -50), which includes the Cys 57 -Ala-Leu-Cys 60 motif that forms the catalytic center of the thiol protein's oxidoreductase activity exhibited by MIF. In vitro studies using C-terminal truncated MIF mutants demonstrated that, in addition to the center of the molecule, the C terminus is also involved in MIF biological activity. Deletion of aa 111-115 and 106 -115 at the C terminus abolished the ability of MIF to activate macrophages in a Leishmania-killing assay (51). The same report also suggested an influence of the C terminus on the redox activity of MIF, as both mutants reduced the enzymatic activity of MIF. However, to our knowledge, the  region spanning aa 86 -102 has not yet been reported to be directly involved in MIF activity.
Using our diverse panel of anti-MIF antibodies, the animal studies allowed us to link the epitope specificity of MIF-neutralizing antibodies to their therapeutic efficacy in vivo. Testing antibodies specific for each of the nine binding regions depicted in Fig. 1 in animal models of experimental sepsis and in models of CHS allowed us to narrow down the binding region of the efficacious antibodies to the linear sequences spanning aa 50 -68 and 86 -102. Obviously, in vitro activity was a prerequisite but was not sufficient to ensure in vivo activity. Only BaxB01 (specific for aa 50 -68) and BaxD08, BaxG03, and its affinity-matured version BaxM159 (all specific for aa 86 -102) significantly reduced circulating TNF␣ and IL-6 levels or significantly improved the survival rate in animal models of experimental sepsis. To prove the versatility of the fully human anti-MIF candidate antibodies identified, we determined the protective potential of our antibodies in the T-cell-mediated immune response in two different animal models of CHS. The essential role of MIF in a classic cell-mediated immune response has been described in models for the tuberculin delayed-type hypersensitivity reaction (34). In addition, MIF knock-out mice were reported to show a significant defect in the manifestation of a CHS response (52). BaxB01 and BaxM159 were able to suppress the effector phase of delayedtype hypersensitivity, confirming these previous reports. In addition, we have also shown that the antibodies interfered with the induction phase of CHS and that neutralization of MIF affected the sensitization of the immune system.
Interestingly, the two binding regions of the efficacious antibodies described herein (aa 50 -68 and 86 -102) are in close proximity in the three-dimensional structure of MIF and form a ␤-sheet structure in space (strands ␤4 and ␤5). MIF crystallizes as a trimer (38,46,47), and sedimentation equilibrium studies have shown that MIF exists predominantly as a trimeric species in solution (53). It is still not entirely clear which form of MIF, the monomeric, dimeric, or trimeric form, plays a role in the physiological function and pathological effects related to MIF activity. In the monomeric conformation, aa 50 -68 and 86 -102 assemble to form a patch on the surface of the MIF molecule (Fig. 7A). Blockage of this patch on the MIF surface may lead to neutralization of MIF function and a subsequent down-regulation of the proinflammatory activity of MIF. However, in a trimeric state, aa 50 -68 and 86 -102 form part of the central channel of the molecule (Fig. 7B). Antibodies binding to this patch might therefore disrupt the trimer formation or hamper its assembly and thus could neutralize the biological activity of MIF.
We repeatedly tested antibodies with other binding specificities for linear epitopes (i.e. antibodies recognizing aa 2-49, 69 -85, and 103-115) in animal models. For example, we included antibodies specific for regions containing the ␣-helical structures of MIF (aa 20 -31 and 70 -85) or antibodies specific for regions containing other ␤-strands (e.g. ␤2; aa 38 -44), but we never observed a beneficial therapeutic effect. Interestingly, antibodies directed to the N-terminal portion of the molecule (aa 2-22) trended in our experiments toward a beneficial therapeutic effect in the experimental sepsis models (data not shown), although in our hands, these therapeutic effects never reached statistical significance. The importance of the N-terminal region was confirmed by two studies. A recent publication (54) reports the isolation of mouse hybridoma antibodies specific for this region that were beneficial in experimental sepsis models. Antibodies against other binding regions were not described in that report. In a second study, a biologically neutralizing anti-MIF polyclonal antibody was found to bind to an epitope at the N terminus of MIF (55). The N-terminal part of MIF, comprising aa 2-22, forms a ␤-strand (␤1) that, in threedimensional space, is adjacent to the ␤-sheet structures built by aa 50 -68 and 86 -102. It can be speculated that because of the close proximity of the N-terminal ␤1-strand to the neutralizing epitopes described above, the region comprising aa 2-22 constitutes a minor although still potentially neutralizing epitope of MIF activity. Antibodies specific for the N terminus could also sterically hinder MIF trimer formation.
Of the 74 antibodies that we identified as recognizing a structural epitope, 14 were able to neutralize MIF activity in vitro in both assays applied. Antibodies with in vitro neutralizing properties, including antibodies inhibiting the tautomerase activity of MIF, were also tested in animal models. Neither the ability to reduce MIF activity in vitro in two functional assays nor the ability to inhibit tautomerase activity was sufficient to ensure in vivo protection, as all of these antibodies were unable to elicit a beneficial effect in vivo. However, inhibition of the tautomerase activity has been used as a strategy for designing small-molecule MIF inhibitors either by competing with the substrate for the tautomerase active site (56) or by covalently modifying the N-terminal proline that is essential for MIF tautomerase activity (57)(58)(59). There is evidence that tautomerase inhibitors of MIF mediate inactivation of MIF biological activity by interfering with the binding of MIF to its surface receptor CD74 rather than by inhibition of enzymatic activity (57,60).
Recently, D-dopachrome tautomerase was described as a functional homolog of MIF that exhibits inflammatory activities similar to MIF (61). aa 50 -68 and 86 -102 of MIF and D-dopachrome tautomerase have a highly similar ␤-sheet structure in the three-dimensional space (62). The pronounced in vivo efficacy of anti-MIF antibodies specific for these regions could be the result of cross-neutralization of both proteins. However, aa 50 -68 of moMIF and murine D-dopachrome tautomerase show a sequence identity of 32%, and aa 86 -102 a sequence identity of 24%. The low sequence identity does not suggest that the anti-MIF antibodies recognize both proteins, but further experiments are required to elucidate a potential cross-neutralization.
In summary, our diverse antibody panel specific for different regions of the MIF molecule made it possible for the first time to scan the MIF molecule for regions that are particularly susceptible to immunological blockage by neutralizing antibodies. Although several reports link down-regulation of MIF activity with certain epitopes or structures of MIF, an extensive comparison between functionally distinct anti-MIF antibodies has not been available to date. The in vitro and in vivo studies performed with our phage display-derived antibody panel suggest that the ␤-sheet structure formed by two regions (aa 50 -68 and 86 -102) plays an essential role in the biological activity of MIF