Structural Insights into Neonatal Fc Receptor-based Recycling Mechanisms

Background: The neonatal Fc receptor (FcRn) regulates the serum half-life of human IgGs and serum albumin (SA). Results: Structures of human FcRn bound to its two ligands shed new lights on the corresponding interactions. Conclusion: Molecular mechanisms for IgG/SA recycling and increased IgG/FcRn binding affinity are presented. Significance: Understanding the molecular basis of the IgG/SA/FcRn interaction is crucial to develop biologics exhibiting beneficial pharmacokinetics properties. We report the three-dimensional structure of human neonatal Fc receptor (FcRn) bound concurrently to its two known ligands. More particularly, we solved the crystal structure of the complex between human FcRn, wild-type human serum albumin (HSA), and a human Fc engineered for improved pharmacokinetics properties (Fc-YTE). The crystal structure of human FcRn bound to wild-type HSA alone is also presented. HSA domain III exhibits an extensive interface of contact with FcRn, whereas domain I plays a lesser role. A molecular explanation for the HSA recycling mechanism is provided with the identification of FcRn His161 as the only potential direct contributor to the corresponding pH-dependent process. At last, this study also allows an accurate structural definition of residues considered for decades as important to the human IgG/FcRn interaction and reveals Fc His310 as a significant contributor to pH-dependent binding. Finally, we explain various structural mechanisms by which several Fc mutations (including YTE) result in increased human IgG binding to FcRn. Our study provides an unprecedented relevant understanding of the molecular basis of human Fc interaction with human FcRn.

The neonatal Fc receptor (FcRn) 2 plays a central role in the regulation of IgG and serum albumin (SA) levels in mammals. FcRn comprises ␤2-microglobulin and a membrane-anchored ␣-chain related to the ␣-chain of the class I major histocompatibility complex (1)(2)(3). It interacts in a pH-dependent fashion with SA or IgG molecules so that the strength of the corresponding interactions ranges from highest at acidic pH to lowest around neutral pH (4 -8). This remarkable property is critical to recycling IgG and SA from within acidic endosomes back to the general circulation and allows their rescue from a degradative pathway (7)(8)(9)(10), thereby prolonging their serum half-life.
Efforts to probe or optimize the corresponding interfaces have yielded several noteworthy results. For instance, a particular set of substitutions known as YTE (M252Y/S254T/T256E) resulted in an ϳ10-fold pH-dependent increase in the binding of human(ized) IgGs to FcRn. Such a modified humanized IgG exhibited an ϳ2-4-and 4-fold increase in its serum half-life in human and cynomolgus monkeys, respectively (11,12). Other studies with human serum albumin (HSA) used modeling and crystallographic approaches coupled with mutagenesis to identify several important HSA residues, and pinpointed the crucial role played by its domain III (13,14).
Here, we set out to solve the x-ray crystal structure of human FcRn bound to wild-type HSA alone, as well as to both wildtype HSA and Fc-YTE. We report diffracting crystals for each complex and have solved the three-dimensional structure of both. Of note, the only HSA⅐FcRn complex three-dimensional structure described so far involves a quadruple HSA mutant whose affinity for FcRn is increased by ϳ300-fold (14). The crystal structures of human FcRn alone, or rat FcRn alone or in complex with rat Fc have also been determined (2,3,15,16). However, structures of human FcRn bound to a human Fc or to wild-type HSA have yet to be elucidated. Our study not only provides these data for the first time, but also allows a view of FcRn bound concurrently to both of its known ligands. This work allows a more complete and relevant understanding of the molecular basis by which HSA and human IgG interact with human FcRn and are recycled in a pH-dependent manner. This fills an important knowledge gap in the field of IgG and wild-type HSA recycling in human, the understanding of which has long been limited to structural modeling, mutational approaches, and assumptions drawn from studies using rodent molecules. We also provide a detailed description of the molecular mechanisms by which YTE and other known mutations increase IgG binding affinity to FcRn. Our data should prove very useful in providing clues to further optimize IgG/FcRn and HSA/FcRn interactions and develop the next generation of IgG-or HSA-based therapeutics. These pharmacokinetically improved molecules could potentially provide many benefits, such as the ability to decrease their administration frequency or dosing requirements.

EXPERIMENTAL PROCEDURES
Conventions and Reagents-All antibody amino acid positions mentioned in the text are identified according to the EU numbering convention (17). Human embryonic kidney (HEK) 293 cells were obtained from ATCC.
Generation and Expression of Human IgG and IgG-YTE Mutants-Human IgG1 and IgG1-YTE (MedImmune) were cloned into a proprietary mammalian expression vector encoding a hybrid C/pelB signal sequence, human cytomegalovirus major immediate early (hCMVie) enhancer, promoter, and 5Ј-untranslated region. Generation of variants at position 310 was carried out by site-directed mutagenesis using a QuikChange XL mutagenesis kit (Agilent Technologies, Santa Clara, CA) according to the manufacturer's instructions. HEK 293 cells were transiently transfected with the various IgG constructs using 293Fectin (Invitrogen) and standard protocols. The conditioned media was harvested at 144 h post-transfection.
Expression and Purification of FcRn-Recombinant human FcRn was expressed in HEK 293 cells as previously described (12). FcRn-containing conditioned media was collected at 72, 144, and 216 h post-transfection and pooled. After adjusting to pH 6.0 with 1 M HCl, the media was passed over a IgG-Sepharose column (GE Healthcare) previously equilibrated with 50 mM NaOAc, pH 5.2. Following washes to baseline with the same buffer, FcRn was eluted using 50 mM sodium phosphate, pH 7.8. Fractions containing FcRn were pooled and loaded onto a 5-ml HiTrap Q HP column (GE Healthcare) previously equilibrated with 50 mM Tris-HCl, pH 8.0, and eluted in a 0 -0.5 M NaCl gradient. Purified FcRn was dialyzed against 50 mM NaOAc, pH 5.2, 100 mM NaCl overnight at 4°C and concentrated to ϳ4 mg/ml. This procedure allowed us to recover over 95% homogeneous FcRn (Fig. 1, a and b).
Generation, Expression, and Purification of HSA Wild-type and Variants-HSA wild-type used for preparing the binary complex was produced recombinantly in-house by cloning the corresponding DNA sequence (corresponding to residues 25-609 of entry number P02768 of the Swiss-Prot database) into a proprietary mammalian expression vector encoding a hybrid C/pelB signal sequence, human cytomegalovirus major immediate early (hCMVie) enhancer, promoter, and 5Ј-untranslated region. HEK 293 cells were transiently transfected with this HSA construct using 293Fectin (Invitrogen) and standard protocols. The conditioned media was harvested at 144 h post-transfection. Recombinant, yeast-produced wild-type HSA used in the preparation of the ternary complex was purchased directly from Albumin Biosciences, Inc. (Huntsville, AL).
HEK 293-and yeast-produced HSA were then further purified using the same protocol. Briefly, the purchased protein (diluted to ϳ5 mg/ml) or HEK 293-conditioned media was titrated into 2.5 M urea for 2 h at 37°C in an effort to remove bound lipid molecules. The resulting protein solution was then directly applied to a 5-ml HiTrap Q HP column (GE Healthcare) previously equilibrated with 50 mM Tris-HCl, pH 8.0. After washing with the same buffer, HSA was eluted in a 0 -1 M NaCl gradient. The purified protein was dialyzed against 50 mM NaOAc, pH 5.2, 100 mM NaCl overnight at 4°C, and concentrated to ϳ4 mg/ml. This procedure allowed us to obtain over 95% homogeneous HSA (Fig. 1, a and b).
HSA variants at position 510 were generated from the corresponding previously described cloned wild-type sequence. Generation of point mutations was carried out by site-directed mutagenesis using a QuikChange XL mutagenesis kit (Agilent Technologies) according to the manufacturer's instructions. Although HSA has a predicted molecular mass of ϳ65 kDa, it migrated at ϳ60 and 50 kDa under R and NR conditions, respectively. This difference is likely attributable to the presence of 17 internal disulfide bonds, the differential reduction of which alters the compactness (and thus the migration) of the molecule. Fc-YTE migrated at ϳ30 and 50 kDa under R and NR conditions, respectively, because of the presence of an expected interchain disulfide bond in its hinge region. Fc-YTE under R conditions seemed to run slightly slower than the expected ϳ25 kDa. We attribute this difference to the presence of carbohydrates at a canonical N-glycosylation site (Asn 297 ). Under NR conditions, this glycosylation effect is likely counterbalanced by the presence of one inter-and four intra-chain disulfide bonds that increase the compactness of the molecule, thereby accounting for an overall migration at the expected ϳ50 kDa. FcRn migrated as expected as two bands corresponding to its ␣and ␤2-microglobulin chains at ϳ30 and 10 kDa, respectively (under both R and NR conditions).
HEK 293 cells were transiently transfected with the various HSA constructs using 293Fectin (Invitrogen) and standard protocols. The conditioned media was harvested at 144 h post-transfection.
Determination of HSA and IgG pH Dependence of Binding toward Human FcRn-Comparison of the interaction of soluble human FcRn with immobilized HSA wild-type and variants at acidic and neutral pH was carried out using a BIAcore 3000 (Biacore International AB, Uppsala, Sweden) in MES (50 mM MES, 0.15 M NaCl, 3 mM EDTA, 0.005% (v/v) P20, pH 6.0) or HBS-EP (10 mM HEPES, 0.15 M NaCl, 3 mM EDTA, 0.005% (v/v) P20, pH 7.4) buffer. HSA was captured directly from the corresponding conditioned media at a typical surface density of ϳ800 RU using an anti-HSA antibody (Abcam, Cambridge, MA) coupled to the dextran matrix of a CM5 sensor chip (GE Healthcare) at a surface density of ϳ12,000 RU. Human FcRn was used at concentrations of 1 and 10 M at pH 6.0 and 7.4, respectively. In all cases, FcRn was also flowed over an uncoated cell, and the sensorgrams from these blank runs were subtracted from those obtained with HSA-coupled cells.
Comparison of the interaction of soluble human FcRn with immobilized human IgG, IgG-YTE, and variants thereof at acidic and neutral pH was carried out essentially as described above. In particular, IgGs were captured directly from the corresponding conditioned media at a typical surface density of ϳ800 -1,600 RU using a mouse anti-human light chain antibody (Abcam) coupled to the dextran matrix of a CM5 sensor chip (GE Healthcare) at a surface density of ϳ5,200 RU.
Preparation of HSA⅐FcRn and HSA⅐FcRn⅐Fc-YTE Complexes-Previously purified HSA and FcRn were mixed at a, respectively, 1:1.1 molar ratio to generate the binary complex. The ternary complex was formed by mixing the previously purified HSA, FcRn, and Fc-YTE at, respectively, 1.1:1.1:0.5 molar ratio because of the presence of two potential FcRn binding sites on the Fc moiety. Further purification and buffer exchange for the binary and ternary complexes were carried out using Superdex TM S75 10/300 GL and S200 10/300 GL columns (GE Healthcare), respectively, pre-equilibrated with 50 mM NaOAc, pH 5.2, 100 mM NaCl. Representative chromatograms of HSA, FcRn, Fc-YTE, and HSA⅐FcRn and HSA⅐FcRn⅐Fc-YTE complexes are shown in Fig. 2, a and b. The purified binary and ternary complexes were then concentrated to 10 and 4 mg/ml, respectively, and subjected to crystallization trials as described in the following section.
Crystallization-Sitting drop crystallization experiments were initially set up in 96-well Intelli-plates (Art Robbins Instruments, Sunnyvale, CA) using a Phoenix crystallization robot (Art Robbins Instruments). Favorable conditions were first identified using the following commercially available crystallization screens: Peg/Ion (Hampton Research, Aliso Viejo, CA) and JCSGϩ Screen (Molecular Dimensions, Apopka, FL). The well and drop compartments of the 96-well plates were first filled with 50 and 0.3 l, respectively, of the various screen solutions. 0.3 l of HSA⅐FcRn or HSA⅐FcRn⅐Fc-YTE complexes at concentrations of 10 and 4 mg/ml, respectively, in 50 mM NaOAc, pH 5.2, 100 mM NaCl was then added to the drop compartment. The resulting mixture was allowed to equilibrate against the well solution at room temperature.
Based on initial screening results for the binary complex, the E9 screen solution from Peg/Ion (4% Tacsimate, pH 4.0, 12% PEG 3350) was selected for further optimization. This step included a previously described seeding procedure (19). More particularly, each well was filled with 45 and 5 l of the E9 and various Additive solutions (Hampton Research), respectively, and thoroughly mixed. For each condition, a sitting drop was formed by mixing 0.3 l of the corresponding well solution with 0.3 l of the HSA⅐FcRn complex at 10 mg/ml in 50 mM NaOAc, pH 5.2, 100 mM NaCl. 0.05 l of seed-containing E9 solution was then added. The best condition was found when Additive B9 (2.0 M NaSCN) was used. Further optimization in hanging drops followed using varying drop volumes (1 to 6 l) and ratios of protein/reservoir solution in the drop (1:1 to 5:1).
Based on initial screening results for the ternary complex, the same E9 screen solution from Peg/Ion (4% Tacsimate, pH 4.0, 12% PEG 3350) was selected for further optimization. This step included the same seeding procedure as for the binary complex. More particularly, each well was filled with 45 and 5 l of the E9 and various Additive solutions (Hampton Research), respectively, and thoroughly mixed. For each condition, a sitting drop was formed by mixing 0.3 l of the corresponding well solution with 0.3 l of the HSA⅐FcRn⅐Fc-YTE complex at 4 mg/ml in 50 mM NaOAc, pH 5.2, 100 mM NaCl. 0.05 l of seed-containing E9 solution was then added. The best condition was found when Additive A9 (0.1 M YCl 3 ) was used.
X-ray Data Collection and Processing-One diffraction dataset from a single crystal was collected at a wavelength of 1.000 Å and temperature of 100 K for the binary complex, at the IMCA-CAT 17ID Beamline of the Advanced Photon Source (APS) of the Argonne National Laboratory (University of Chicago, Chicago, IL) equipped with a PILATUS 6M detector (Paul Sherrer Institute, Villigen, Switzerland). Cryoprotection was achieved by soaking the selected crystals in 0.16 M NaSCN, 2.88% Tacsimate, pH 4.0, 8.64% PEG 3350, 20% glycerol. Crystals were then flash-cooled in liquid nitrogen. We recorded 400 frames using an oscillation range of 0.5°, a crystal to detector distance of 450 mm, and an exposure time of 1 s.
One diffraction dataset from a single crystal was collected at a wavelength of 1.000 Å and temperature of 100 K for the ternary complex, at the X06DA Beamline of the Swiss Light Source (SLS) equipped with a PILATUS 2M detector. Cryoprotection was achieved by soaking the selected crystal in 8 mM YCl 3 , 2.88% Tacsimate, pH 4.0, 8.64% PEG 3350, 20% glycerol. The crystal was then flash-cooled in liquid nitrogen. We recorded 444 frames using an oscillation range of 0.5°, a crys-tal to detector distance of 300 mm, and an exposure time of 0.3 s. All diffraction images were processed using the HKL 2000 package (20).
Determination of HSA/FcRn Three-dimensional Structure-The models of HSA and human FcRn were built from PDB entries 1AO6 (21) and 1EXU (15), respectively. The unit cell with parameters a ϭ 136.5, b ϭ 115.9, c ϭ 186.2 Å, and ␤ ϭ 104.5°contained four nearly identical HSA⅐FcRn complexes. Two pairs of complexes (1-3 and 2-4) were related through pseudo-translation along the c axis (fractional 0.49). The other two pairs (1-2 and 3-4) were related through a pure non-crystallographic 2-fold rotation (179.8°). The four copies of human FcRn could be easily placed using MolRep (22). However, HSA had to be broken into separate parts to find an appropriate solution. More precisely, the four FcRn molecules were first entered as a fixed model and a solution for four copies of HSA domain I (DI, amino acids 3-203) was sought. The corresponding solution, which included four copies of each FcRn and HSA DI, was then fixed and a solution for four copies of HSA domain II (DII, amino acids 208 -396) was sought and found. HSA domain III (DIII, amino acids 402-585) had to be further broken into two parts corresponding to subdomains IIIa and IIIb (DIIIa, amino acids 402-491 and DIIIb, amino acids 520 -585) to find the corresponding solution. The model then underwent refinement using Refmac5 (22) with automatically generated global NCS restraints. Manual re-building was carried out with the "O" program (23). The refinement converged to a final R/R free of 0.23/0.26. Of note, residues between HSA DI/II and DII/III (amino acids 204 -207 and 397-401, respectively) as well as the connecting peptide between HSA DIIIa and DIIIb (amino acids 492-519) were built manually during the iterative refinement/rebuilding procedure. Data and refinement statistics are presented in Table 1. The final root mean square (r.m.s.) deviations between the equivalent C ␣ atoms of the complexes in the asymmetric unit were 0.13, 0.13, and 0.11 Å for FcRn ␣-chain, HSA, and ␤2-microglobulin, respectively. No residues  MARCH 14, 2014 • VOLUME 289 • NUMBER 11 in the final model were found in the disallowed region of the Ramachandran plot (90.3% in most favored region, 9.4% in additionally allowed region, 0.3% in generously allowed region). The refined model of a representative region fitted into electron density is presented in Fig. 3a.

Structure of Human FcRn Bound to HSA and a Human Fc Fragment
Determination of the HSA/FcRn/Fc-YTE Three-dimensional Structure-HSA/FcRn (this study) and Fc-YTE (PDB entry 3FJT) (18) were used as models. The unit cell with parameters a ϭ b ϭ 153.2, c ϭ 146.0 Å contained one FcRn, one HSA, and one Fc-YTE polypeptide (half-Fc). Using Phaser (24) from the CCP4 program suite, we found one solution for each component of the complex with a final LLG of 3997.34 and R-factor of 0.48. The model then underwent rigid-and jelly-body refinement using Refmac5 (22). The refinement converged to a final R/R free of 0.28/0.30. Data and refinement statistics are presented in Table 1. No residues in the final model were found in the disallowed region of the Ramachandran plot (89.5% in most favored region, 10.3% in additionally allowed region, 0.2% in generously allowed region). The refined model of a representative region fitted into electron density is presented on Fig. 3b.

RESULTS AND DISCUSSION
Crystallization and Structure Determination-Diffraction quality crystals for the binary complex grew in ϳ3-4 days in hanging drops where 1-5 l of the HSA⅐FcRn complex at a concentration of 10 mg/ml in 50 mM NaOAc, pH 5.2, 100 mM NaCl was mixed with 1 l of 0.2 M NaSCN, 3.6% Tacsimate, pH 4.0, 10.8% PEG 3350 and left to equilibrate against 300 l of 0.2 M NaSCN, 3.6% Tacsimate, pH 4.0, 10.8% PEG 3350 (Fig. 4a). The symmetry of these crystals belonged to monoclinic space group P2 1 according to Pointless (Collaborative Computational Project 4, CCP4 suite) (25), and reached a diffraction limit down to 3.0-Å resolution. The asymmetric unit contained four HSA⅐FcRn complexes. This corresponds to a crystal volume per protein weight (V M ) of 2.6 Å 3 /Da and a solvent content of 62%. SDS-PAGE analysis of these crystals confirmed that they indeed contained the expected complex formed by HSA and FcRn (Fig. 1b).
Diffraction quality crystals for the ternary complex grew in ϳ4 weeks in sitting drops where 0.3 l of the HSA⅐FcRn⅐Fc-YTE complex at a concentration of 4 mg/ml in 50 mM NaOAc, pH 5.2, 100 mM NaCl was mixed with 0.3 l of 0.01 M YCl 3 , 3.6% Tacsimate, pH 4.0, 10.8% PEG 3350 and left to equilibrate against 50 l of 0.01 M YCl 3 , 3.6% Tacsimate, pH 4.0, 10.8% PEG 3350 (Fig. 4b). The symmetry of these crystals belonged to tetragonal space group P4 1 2 1 2 according to Pointless (25) and reached a diffraction limit down to 3.8-Å resolution. The asymmetric unit contained one HSA⅐FcRn⅐half-Fc-YTE complex. This corresponds to a V M of 3.0 Å 3 /Da and a solvent content of 60%. SDS-PAGE analysis of these crystals confirmed that they  indeed contained the expected complex formed by HSA, FcRn, and Fc-YTE (Fig. 1a). The three-dimensional structures of the HSA⅐FcRn and HSA⅐FcRn⅐Fc-YTE complexes were solved to a resolution of 3.0 and 3.8 Å, respectively, using molecular replacement (see "Experimental Procedures").
Analysis of HSA Three-dimensional Structure-We found it necessary to break the HSA model into separate parts to find a solution during the molecular replacement procedure. Because we found a solution for each separate HSA domain, it is reasonable to assume that whereas their internal structure was preserved, their relative orientation changed upon binding to FcRn. We assessed such changes by superimposing 50 different HSA molecules from the Protein Data Bank (PDB) with our own using the LSQKAB program (26). The superimposition was carried out at full-length and individual domain levels. When full-length molecules were used, r.m.s deviations for C ␣ atoms varied between 2.2 and 6.6 Å (when compared with FcRn-bound HSA) and showed that DIII exhibited the largest change in its orientation relative to other domains (Fig. 5a). When superimposition was done at the domain level, deviations were considerably smaller. More precisely, HSA DII was the most conformationally conserved (r.m.s deviations for C ␣ atoms ranged from 0.6 to 1.3 Å when compared with FcRnbound HSA), whereas HSA DI exhibited intermediate variability (with r.m.s deviations for C ␣ atoms ranging from 0.9 to 1.6 Å). HSA DIII exhibited the most variability (r.m.s deviations for C ␣ atoms ranged from 1.7 to 2.1 Å), which can be accounted for the most part by a significant change in the relative position of its two subdomains (Fig. 5b). This suggests that formation of the HSA⅐FcRn complex results in changes in the relative orientation of HSA DIII and agrees well with our finding that this domain plays a major role in binding to FcRn (see below).
Analysis of HSA/FcRn Three-dimensional Structure-We present here the three-dimensional structure of wild-type HSA bound to human FcRn. A ribbon diagram of this complex is shown in Fig. 6. Overall, our complex structure is in good agreement with that involving a HSA variant (V418M/T420A/ E505G/V547A) whose affinity for human FcRn was increased by ϳ300-fold (14). In particular, the r.m.s deviation for all atoms is 0.8 Å. Of these four mutations, only HSA E505G seems to play a direct role in increasing FcRn/HSA binding affinity via two additional hydrogen bonds introduced between its carbonyl and FcRn Ser 230 side chain. This mutation also resulted in local conformational changes between HSA Ala 504 and Phe 509 when compared with the wild-type molecule of this study. HSA V418M and T420A are part of the interface but do not play an obvious role in the corresponding affinity increase. HSA V547A is not part of the interface. Therefore, we postulate their impact on FcRn binding to be mediated by long-range (V547A) or indirect (V418M and T420A) effects.
We note that no remarkable differences were detected between the conformations of human FcRn when in its unliganded (PDB entry 1EXU) (15) or bound (this study) form. FcRn binds in a crevice of the heart-shaped HSA molecule and creates direct interactions with both DI and DIII. Overall, 1800 Å 2 FIGURE 5. a, stereographic representation of the superimposition of fulllength HSA of this study (red) with HSA corresponding to PDB entry 1AO6 (32) (blue). Superimposition was carried out through the C␣ atoms of HSA DII. HSA DIII exhibits the largest change in orientation relative to other domains, with a maximum motion of 19.1 Å between the C␣ atoms of Asp 562 . HSA DI motion was more modest with a maximum distance of 5.0 Å between the C␣ atoms of Ala 172 . Pivot points for DI and DIII were found at Phe 206 and Glu 393 , respectively. b, stereographic representation of the superimposition of HSA DIII of this study (red) with HSA DIII corresponding to PDB entry 1AO6 (32) (blue). The superimposition was carried out through the C␣ atoms of HSA DIIIa. Significant variability in the relative orientation of HSA DIIIa and DIIIb was seen, with a maximum distance of 8.5 Å between the C␣ atoms of Asp 562 . The pivot point was found at Glu 501 . This and all other figures were made using PyMOL (DeLano Scientific, Palo Alto, CA). of surface area is buried at the interface. Of that, ϳ500 and ϳ200 Å 2 correspond to HSA DI/FcRn ␣-chain and HSA DIII/ ␤2-microglobulin interfaces, respectively. The majority (ϳ1100 Å 2 ) rests between HSA DIII and the FcRn ␣-chain. A prominent feature of FcRn is a groove made by two long ␣-helices (Tyr 60 -Leu 82 and Pro 132 -Glu 175 ) backed by a sevenstranded antiparallel ␤-sheet. A kink in the second ␣-helix (between Gln 142 and Ala 147 ) results in the separation of the groove into a short and long part. HSA DI is positioned above the longer portion of the groove and interacts with both ␣-helices. The interface mostly comprises charged or polar amino acids that are involved in direct electrostatic interactions, including two hydrogen bonds between HSA Asn 111 /O␦1 and FcRn Ser 58 /O␥ and between HSA Arg 81 /N⑀ and FcRn Thr 153 / O␥1 (3.6 Å distance cutoff). Other electrostatic interactions involve FcRn Ser 58 /Lys 63 /Asn 149 and HSA Asp 89 /Asp 108 / Asn 111 (Fig. 7). In addition, FcRn His 161 contributes to this network via a weak hydrogen bond (ϳ4 Å) between its N⑀2 nitrogen and the carbonyl of HSA Glu 82 . This histidine is one of two at the HSA/FcRn interface and likely plays a direct role in the corresponding pH dependence of binding (see below  (Fig.  8a). Unlike that with HSA DI, this interface exhibits a very prominent hydrophobic core centered on FcRn Phe 44 /Val 57 / Trp 59 /Trp 61 and HSA Gln 417 /Pro 421 /Leu 460 /Leu 463 /Val 469 . This is in good agreement with isothermal titration calorimetry showing that HSA binds FcRn with a large positive change in entropy (6). This area does not contain any histidine residues, and therefore does not contribute to pH dependence. The interface between HSA DIIIb (including the IIIa/IIIb connecting peptide) and FcRn is less extensive than that involving DIIIa and comprises four hydrogen bonds (3.6 Å distance cutoff). These involve FcRn Arg 42 /Trp 53 /Asn 173 and HSA Tyr 497 / Val 498 /Thr 506 /His 510 . Other electrostatic interactions involve FcRn Ser 230 /Asp 231 and HSA Glu 505 (Fig. 8b). The interface is centered on FcRn Trp 53 , which contributes its N⑀1 atom to hydrogen bonds with the carbonyls of HSA Thr 506 . It also contains His 510 in a position to form a strong (2.6 Å) hydrogen bond with FcRn Asn 173 /O␦1. This seems to suggest a direct role of this residue in the pH dependence of the corresponding interaction. To test this hypothesis, we substituted His 510 by all possible amino acids (except cysteine), and assessed binding of the resulting variants to human FcRn at both acid and neutral pH (Fig. 9). Binding at pH 6.0 ranged from similar to significantly decreased when compared with HSA wild-type, which highlighted the overall importance of this residue to the corresponding interaction. All variants, however, maintained a pHdependent binding profile and none bound human FcRn at pH 7.4. Therefore, we conclude that HSA His 510 is not directly and majorly involved in the pH dependence of binding. We suggest that the latter likely relies on the only other interface histidine, namely FcRn His 161 .
␤2-Microglobulin plays a relatively minor role in this complex. In particular, the guanidinium group of ␤2-microglobulin Arg 12 is involved in a weak bidentate interaction with HSA Glu 505 /O⑀1 and O⑀2, whereas ␤2-microglobulin Glu 50 interacts with the carbonyl of HSA Phe 502 (Fig. 10).
HSA DIII has previously been identified as critical to the interaction with FcRn. Indeed, it was reported that DIII alone exhibits significant binding to human FcRn, whereas DI and DII individually or together show no measurable interaction (6). Andersen et al. (13) and Schmidt et al. (14) also found that DIII plays a crucial role in HSA pH-dependent binding to FcRn, although it was also suggested that DI and DII could also bring a moderate contribution (13). Our findings agree well with these data and reaffirm the crucial role played by HSA DIII, along with a significant contribution of DI. Andersen et al. (13) also built a full-length HSA/FcRn model and ran a docking procedure with bias toward solutions involving FcRn His 161 /His 166 and HSA His 464 /His 510 /His 535 . We have found here, however, that FcRn His 166 and HSA His 464 /His 535 are not part of the corresponding interface and therefore do not play a direct role in this pH-dependent interaction. This led these authors to inaccurate conclusions, in particular regarding the existence of important interactions between FcRn His 166 and HSA Glu 505 , FcRn Glu 54 and HSA Glu 510 , FcRn Lys 150 /Glu 151 and HSA Glu 501 /Lys 500 , and FcRn His 161 and HSA Glu 531 . Indeed, our study shows that the corresponding residues are, respectively, ϳ18, 10, 25, and 30 Å apart, and thus unable to engage in such interactions. Interestingly, the assumption that FcRn His 166 is at the core of the interface seemed to be at least partially based on the inability of the H166A mutant to bind HSA (27). We propose that this mutation only acts in a indirect fashion, maybe by disrupting a network of internal interactions within FcRn (e.g. with Glu 54 ) and leading to deleterious changes on the FcRn side of the complex.
Analysis of HSA⅐FcRn⅐Fc-YTE Three-dimensional Structure-We present here the three-dimensional structure of human FcRn bound concurrently to its two known ligands. The ribbon diagram of the HSA⅐FcRn⅐Fc-YTE ternary complex is shown in Fig. 11. The HSA⅐FcRn portion superimposes well with that of the binary complex, with a r.m.s deviation of C␣ atoms of ϳ0.2 Å. The Fc-YTE moiety also superimposes well with that of the same molecule in an unbound state (PDB entry 3FJT) (18) with a r.m.s deviation of C␣ atoms of ϳ0.1 Å.
The FcRn molecule binds Fc between its CH 2 and CH 3 domains. The interface exhibits very strong hydrophobic prop- erties as it involves FcRn Tyr 88 , Leu 112 , Phe 117 , Trp 131 , Pro 132 , and Leu 135 , and Fc-YTE Leu 251 , Tyr 252 , Ile 253 , Leu 309 , and Leu 314 . Six hydrogen bonds and salt bridges between Fc-YTE and FcRn were identified (3.6 Å distance cutoff) and included FcRn Glu 133 /Glu 115 /Glu 116 /Gly 129 and Fc-YTE Ile 253 /Thr 254 / His 310 /Gln 311 /Asn 434 . In particular, the side chain of Fc-YTE His 310 is positioned in such a way that its N␦1 atom makes a strong hydrogen bond with FcRn Glu 115 /O⑀1 (2.8 Å distance; Fig. 12). This is in good agreement with previous studies identifying Fc His 310 as particularly important for human or murine Fc binding to murine FcRn (28 -30). Interestingly, the analysis of the human(ized) IgG/human FcRn interaction had also previously revealed the critical role played by Fc Ile 253 and His 435 (31)(32)(33). These results can be explained by Ile 253 being an important component of a large hydrophobic patch at the human IgG/FcRn interface (see above), the disruption of which likely affects formation of the complex. However, our study shows that the Fc His 435 side chain does not adopt a favorable conformation to form an intermolecular hydrogen bond and is only involved in a weak interaction involving its N⑀2 atom and the main chain carbonyl of FcRn Asp 130 (2.8 Å distance; Fig.  12). Therefore, the dramatic effects seen on human IgG⅐FcRn  complex formation upon mutating this residue are likely not caused solely by a disruption of this interaction, but rather by other yet-to-be defined indirect effects, such as local conformational rearrangements.
Finally, it has generally been assumed using rodent molecules that IgG/FcRn pH-dependent interaction relies on the titration of Fc His 310 /His 433 and FcRn His 250 /His 251 (5,34). We show here that Fc His 433 and FcRn His 248 /His 249 (human equivalent of His 250 /His 251 in mouse) are not part of the corresponding interface. Therefore, we conclude that these amino acids do not play a direct role in the corresponding pH-dependent interaction in human. Because of the central role played by His 310 (see above), it is tantalizing to identify this residue as a major contributor to this process. To test this hypothesis, we substituted His 310 by all possible amino acids (except cysteine), and assessed binding of the resulting IgG and IgG-YTE variants to human FcRn at both acid and neutral pH. Binding of all (non-YTE) IgG variants at both pH 6.0 and 7.4 was essentially not detectable (data not shown). Most IgG-YTE variants also did not significantly bind to FcRn at pH 6.0, with only a few exhibiting detectable, although dramatically decreased binding at pH 6.0 (H310E, H310K, H310M; Fig. 13). In conclusion, our data highlight the overall crucial importance of His 310 for binding FcRn at pH 6.0, and suggest that this residue plays a significant role in the corresponding pH-dependent process.
Molecular Basis for Enhancing Human IgG Interaction with FcRn-Several Fc mutations resulting in increased human IgG binding to human FcRn have been previously described and carefully characterized at a functional level, including their effect on IgG pharmacokinetic properties and biodistribution (12,33,35,36). However, the exact molecular mechanism(s) underlying the functional properties of these mutations have remained elusive. Our present study provides those for the first time.
In particular, the YTE set of substitutions (M252Y/S254T/ T256E) results in an ϳ10-fold increase in human IgG1 binding  to human FcRn, whereas studies in human and cynomolgus monkeys indicate that the serum half-life of this mutated humanized IgG1 was increased by ϳ2-4-and 4-fold, respectively, when compared with its unmutated counterpart (11,12). We find that the YTE mutations result in the creation of two salt bridges between Fc-YTE Glu 256 and ␤2-microglobulin Gln 2 (Fig. 14). The wild-type residue at this position (Thr) would be unable to engage in such interactions because of its shorter side chain. Thr 254 /O␥1 also interacts with FcRn Glu 133 /O⑀1, although it is predicted that the wild-type residue at this position (Ser) would also potentially be able to engage in a similar interaction. Therefore, we propose that Glu 256 plays the largest role in increasing the binding affinity between Fc-YTE and FcRn. This validates our earlier prediction on the YTE mechanism of action, where we suggested that its effects "are mediated by the gain of a few, yet-to-be accurately defined interactions with human FcRn at the substitutions sites" (18).
Several other mutations resulting in increased human IgG binding to human FcRn have also been described and carefully characterized. Of those, N434A was identified by alanine scanning and resulted in an ϳ3.5-fold increase in human IgG1 binding to human FcRn at pH 6.0 (32). When dosed in transgenic mice expressing human FcRn, N434A resulted in an ϳ2-3-fold increase in serum half-life when compared with the same unmutated antibody (33). T250Q/M428L was also described, which, when introduced into a human IgG1, resulted in an ϳ30-fold pH-dependent increase in IgG binding to human FcRn (35). The same mutant had a serum half-life ϳ2.5-fold longer in rhesus monkeys than the wild-type form of the same human IgG1. Finally, M428L/N434S provided an 11-fold improvement in human IgG1 affinity to FcRn at pH 6.0 along with an ϳ3-4-fold increase in serum half-life in transgenic mice expressing human FcRn (36). We find that Fc Asn 434 is involved in an unfavorable interaction of polar and hydropho- bic amino acids with FcRn Leu 135 . Upon mutation to alanine (N434A), a new hydrophobic patch is created that engages Leu 135 in favorable contacts. Substitution to serine (N434S) allows the creation of additional hydrogen bonds with either or both carbonyls of FcRn Gly 129 and Trp 131 . Finally, Fc Thr 250 and Met 428 are not in direct contact with FcRn and no additional direct intermolecular interaction seems to be introduced when mutated to Gln and Leu, respectively. Therefore, these two mutations likely increase human IgG binding to human FcRn in an indirect fashion, maybe through subtle favorable conformational changes at the interface or local rearrangement of the water molecules network.
Our HSA/FcRn and HSA/FcRn/Fc-YTE structures provide valuable information in the field of IgG and HSA recycling. In addition to providing new insights on these well known molecules, they are expected to guide the design and development of novel improved variants exhibiting beneficial pharmacokinetics properties. This is particularly true of the HSA/FcRn interaction where much of the interface remains to be probed. We also hope our results will provide new engineering clues and lead to a revival of studies aimed at optimizing the IgG/FcRn interaction, whose interface has been exhaustively mutated by a plethora of rational design and library-based approaches.