Increasing the Affinity of a Human IgG1 for the Neonatal Fc Receptor: Biological Consequences¹

The protection of IgGs from degradation as well as the maternofetal transfer of IgG from mother to young are essential, tightly controlled processes in mammals. The neonatal FcR (FcRn)³ plays a central role in both of these events (1). Much is known about the role of FcRn in rodents, in which it has been shown to mediate transcytosis of IgGs across the epithelium of the neonatal intestine (2), yolk sac (3), and mammary gland (4), as well as to maintain constant levels of serum IgG (5, 6). In humans, FcRn is also believed to play a role in IgG transport. This has been shown both for placenta (7, 8) and adult epithelia (9, 10). A role for FcRn in IgG homeostasis in humans has been hypothesized (5, 11, 12), but to date, there is no direct evidence to invoke its role in such a process.

FcRn is a heterodimer formed by the association of an L (β₂-microglobulin) and H (α-chain) (13) chain. The α-chain, composed of three domains (α1, α2, and α3), is anchored to the membrane via a transmembrane region. The rat, murine, and human H chain sequences (7, 13, 14) as well as the rat and human FcRn structures (15, 16, 17) are homologous to the class I major histocompatibility complex molecules. Crystallographic studies of the extracellular domain of unliganded human FcRn (17) and of the extracellular domain of rat FcRn free and bound to rat IgG (15, 16) have indicated that human and rat FcRn are structurally similar. The most notable feature of the interaction of murine and human FcRn with murine and human IgG is its pH dependency: the Fc portion of IgGs binds FcRn with a high affinity at pH 6.0 and is released at pH 7.2 (2, 18). This suggests, along with colocalization studies (19), that for cells bathed in near neutral pH, the IgG/FcRn complex forms within acidified endosomes. More precisely, FcRn acts as a salvage receptor, binding and transporting pinocytosed IgGs in intact form both within and across cells, and rescuing them from a default degradative pathway. Although the molecular mechanisms responsible for salvaging IgGs are still largely unknown, it is thought that unbound IgGs are directed toward proteolysis in lysosomes, whereas bound Igs are recycled to the surface of the cells and released. This control takes place within the endothelial cells located diffusely throughout adult tissues (5, 20).

Fc residues critical to the mouse Fc-mouse FcRn interaction have been identified by site-directed mutagenesis. In particular, I253, H310, H433, N434, and H435 (EU numbering) (21) are major components of the functional epitope (3, 8, 22) in which H310 and H435 are responsible for the pH dependence of binding (23). I253, H310, and H435 were found to be critical for the interaction of human Fc with murine FcRn (12). Likewise, studies of the human Fc-human FcRn complex have demonstrated the crucial role played by I253, S254, H435, and Y436 (8, 24). Functional epitopes on the rat FcRn have been also identified by site-directed mutagenesis in both the α-chain and β₂-microglobulin chain (25, 26), demonstrating which residues on the FcRn are critical for the interaction with both mouse and rat IgG.

There seems to be a correlation between decrease (3) and increase (11) in the affinity of murine hinge-Fc fragments for murine FcRn at pH 6.0 and their serum t_1/2 in mice. The same correlation exists when one considers the interaction of human hinge-Fc fragments with mouse FcRn (8, 12). This observation has obvious relevance for the generation of therapeutic Abs with increased serum persistence. Although it has been proposed that a tighter IgG-FcRn interaction at pH 7.5 can result in a decreased IgG serum persistence (27), this phenomenon has not been investigated in detail. Furthermore, most of the studies published to date have been limited to the analysis of the interaction of rodent FcRn with rodent or human IgGs (3, 11, 12, 16, 22, 23, 26, 28, 29, 30). Recent studies involving the binding of human IgGs to human FcRn have provided important information, such as the role of the receptor in the maternofetal transfer of γ-globulin (8), the chemical nature of the interface, and general characteristics of the interaction (17), as well as the relative contribution of individual residues to the energetics of complex formation (24). Human Fc mutants with all solvent-exposed residues along the C_H2 and C_H3 domains substituted by alanine were produced (24). This led to the identification of >12 variants with increased affinity to the human receptor, most by a factor of 1.5- to 2-fold. Two combination mutants showed increased affinity by up to 12-fold (24).

We have now further characterized the human Fc-human FcRn and human Fc-mouse FcRn interactions. Using rationally designed libraries and phage display, we provide evidence that a human IgG1 can be engineered for large (>30 times) increases in affinity toward both human and murine FcRn. As a step toward a better understanding of these interactions, we investigated the binding of different IgG1 variants at neutral pH. We show that IgG1 mutants exhibiting an increase in affinity toward murine FcRn by a factor of >5-fold at pH 6.0 also show significantly better binding at pH 7.4 when compared with the wild-type molecule. In contrast, IgG1 mutants exhibiting an affinity increase toward human FcRn by a factor of up to 10-fold at pH 6.0 all bind very poorly at pH 7.4. When injected in equal amounts into different BALB/c mice, three IgG1 variants with a 1.4- to 19-fold affinity increase to murine FcRn at pH 6.0 exhibit lower than wild-type serum concentrations (reported in ng_{variant/wild type}/ml_serum). Those mutants are not proteolyzed more efficiently than a wild-type control in serum or whole blood, nor are they seen to accumulate in different tissues or be excreted at a detectable level in urine. We hypothesize that retention of binding at neutral pH offsets the advantage of improved binding at pH 6.0, leading to the intracellular sequestration of the corresponding IgG molecules in FcRn-containing tissues.

Our results suggest that the pH dependency of binding of IgG1 to FcRn plays a critical role in determining the biological effect of an affinity-improving substitution.

All chemicals were of analytical grade. Restriction enzymes and DNA-modifying enzymes were purchased from New England Biolabs (Beverly, MA). TG1 and CJ236 Escherichia coli were obtained from APBiotech (Piscataway, NJ) and Bio-Rad (Richmond, CA), respectively.

Recombinant mouse and human FcRn were produced as described (8, 29). Human FcRn was also obtained following isolation from human placenta cDNA (Clontech, Palo Alto, CA) of the genes for human β₂-microglobulin (21) and codons −23–267 of the human α-chain (7) using standard PCR protocols. L and H chains along with their native signal sequence (7, 21) were cloned in pFastBac DUAL and pFastBac1 bacmids, respectively, and viral stocks produced in Spodoptera frugiperda cells (Sf9) according to the manufacturer’s instructions (Invitrogen, Carlsbad, CA). High-Five cells were infected at a multiplicity of infection of 3 with the baculoviruses encoding α- and β₂-chains using commercially available protocols (Invitrogen). Supernatant of infected insect cells was then adjusted to pH 6.0 with hydrochloric acid and applied to a 10 ml IgG Sepharose 6 Fast Flow column (APBiotech). The resin was washed with 200 ml 50 mM MES, pH 6.0, and FcRn eluted with 0.1 M Tris-Cl, pH 8.0. Purified FcRn (>95% homogeneity, as judged by SDS-PAGE) was dialyzed against PBS, flash frozen, and stored at −70°C.

A human hinge-Fc gene spanning aa residues 226–478 (Kabat numbering) (21), derived from MEDI-493 human IgG1 (31), was cloned into the pCANTAB5E phagemid vector (APBiotech) as an SfiI/NotI fragment. Four libraries were generated by introducing random mutations at positions 251, 252, 254, 255, 256 (library 1); 308, 309, 311, 312, 314 (library 2); 385, 386, 387, 389 (library 3); and 428, 433, 434, 436 (library 4), using a Kunkel-based strategy (32). Briefly, four distinct hinge-Fc templates were generated using PCR by overlap extension (33), each containing one TAA stop codon at position 252 (library 1), 310 (library 2), 384 (library 3), or 429 (library 4), so that only mutagenized phagemids will give rise to Fc-displaying phage. Each TAA-containing ssDNA (TAAssDNA) was then prepared as follows: a single CJ236 E. coli colony harboring one of the four relevant TAA-containing phagemid was grown in 10 ml 2× YT medium supplemented with 10 μg/ml chloramphenicol and 100 μg/ml ampicillin. At OD₆₀₀ = 1, VCSM13 helper phage (Stratagene, La Jolla, CA) was added to a final concentration of 10¹⁰ PFU/ml. After 2 h, the culture was transferred to 500 ml of 2× YT medium supplemented with 0.25 μg/ml uridine, 10 μg/ml chloramphenicol, 30 μg/ml kanamycin, and 100 μg/ml ampicillin, and grown overnight at 37°C. Phage were precipitated with PEG6000 using standard protocols (34) and purified using the Qiaprep Spin M13 Kit (Qiagen, Valencia, CA), according to the manufacturer’s instructions. A total of 10–30 μg of each uracil-containg TAAssDNA template was then combined with 0.6 μg of the following phosphorylated oligonucleotides (MWG Biotech, High Point, NC) in 50 mM Tris-HCl, 10 mM MgCl₂, pH 7.5, in a final volume of 250 μl (randomized regions are underlined): library 1, 5′-CATGTGACCTCAGGSNNSNNSNNGATSNNSNNGGTGTCCTTGGGTTTTGGGGGG-3′; library 2, 5′-GCACTTGTACTCCTTGCCATTSNNCCASNNSNNGTGSNNSNNGGTGAGGACGC-3′; library 3, 5′-GGTCTTGTAGTTSNNCTCSNNSNNSNNATTGCTCTCCC-3′; library 4, 5′-GGCTCTTCTGCGTSNNGTGSNNSNNCAGAGCCTCATGSNNCACGGAGCATGAG-3′.

Each mixture was incubated at 90°C for 2 min, 50°C for 3 min, and 20°C for 5 min. Synthesis of the heteroduplex DNA was conducted by adding 30 U of both T4 DNA ligase and T7 DNA polymerase in presence of 0.4 mM ATP, 1 mM dNTPs, and 6 mM DTT. The mixture was incubated for 4 h at 20°C. Heteroduplex DNA was affinity purified and desalted using the Qiaquick DNA purification kit (Qiagen). A total of 1–5 μg of heteroduplex DNA was then electroporated into 300 μl of electrocompetent TG1 E. coli cells in a 2.5 kV field using 200 Ω resistance and 25 μF capacitance until a library size of 1 × 10⁸ (library 1 and 2) or 1 × 10⁷ (library 3 and 4) was reached. Cells were grown in 500 ml of 2× YT medium containing 100 μg/ml ampicillin and 10¹⁰ PFU/ml of VCSM13 helper phage overnight at 37°C. Phage were precipitated with PEG6000, as previously described (34).

Phage were resuspended in 5 ml 20 mM MES, pH 6.0/5% skimmed milk/0.05% Tween 20 and added (100 μl of 5 × 10¹² PFU/ml/well) to 20 wells of a Maxisorp immunoplate (Nunc, Rochester, NY) previously coated with 1 μg of murine FcRn and blocked with 5% skimmed milk. After incubation for 2 h at 37°C, wells were washed 10–30 times with 20 mM MES, pH 6.0/0.2% Tween 20/0.3 M NaCl and phage eluted by incubation in 100 μl PBS, pH 7.4/well for 30 min at 37°C. Phage were used to reinfect exponentially growing E. coli TG1, as described (35). Three to six rounds of panning were conducted. Typically, 20–40 clones were characterized after three (library 1, 2, and 4) and six (library 3) rounds of panning by dideoxynucleotide sequencing (36) using a ABI3000 genomic analyzer (Applied Biosystems, Foster City, CA).

Representative Fc mutations were incorporated into the human IgG1 MEDI-493 (31). The H chain was subjected to site-directed mutagenesis using a Quick Change Mutagenesis kit (Stratagene), according to the manufacturer’s instructions, and sequences were verified using a ABI3000 sequencer. The different constructs were expressed transiently in human embryonic kidney 293 cells using a CMV immediate-early promoter and dicistronic operon in which the γ1 chain is cosecreted with the κ-chain (31). IgG1s were purified from the conditioned medium directly on 1 ml HiTrap protein A columns, according to the manufacturer’s instructions (APBiotech).

The interaction of soluble murine and human FcRn with immobilized human IgG1 variants was monitored by SPR detection using a BIAcore 3000 instrument (Pharmacia Biosensor, Uppsala, Sweden). No aggregated FcRn that could interfere with affinity measurements (37, 38) was detected by gel filtration. Protein concentrations were calculated by the bicinchoninic acid method for both human and murine FcRn or using the 1% extinction coefficient at 280 nm of 1.5 for IgG1 wild type and variants. IgGs were coupled to the dextran matrix of a CM5 sensor chip (Pharmacia Biosensor) using an amine coupling kit, as described (39). The protein concentrations ranged from 3 to 5 μg/ml in 10 mM sodium acetate, pH 5.0. The activation period was set for 7 min at a flow rate of 10 μl/min, and the immobilization period was set to between 10 and 20 min at a flow rate of 10 μl/min. Excess reactive esters were quenched by injection of 70 μl of 1.0 M ethanolamine hydrochloride, pH 8.5. This typically resulted in the immobilization of between 500 and 4000 resonance units (RU). Human and murine FcRn were buffer exchanged against 50 mM PBS, pH 6.0, 7.4, or 8.5, containing 0.05% Tween 20. Dilutions were made in the same buffers. All binding experiments were performed at 25°C, with concentrations ranging from 2.86 μM to 1 nM at a flow rate of 5–10 μl/min; data were collected for 25–50 min, and three 1-min pulses of PBS, pH 7.4, were used to regenerate the surfaces. FcRn was also flowed over an uncoated cell, and the sensorgrams from these blank runs were subtracted from those obtained with IgG1-coupled chips. Runs were analyzed using the software BIAevaluation 3.1 (Pharmacia Biosensor). Association constants (K_a) were determined from Scatchard analysis by measuring the concentration of free reactants and complex at equilibrium after correction for nonspecific binding (40). Errors were estimated as the SD for two or three independent determinations and were <20%.

Analyses were conducted in BALB/c mice (Harlan, Indianapolis, IN) for MEDI-493 wild type, G385D/Q386P/N389S, M252Y/S254T/T256E, and H433K/N434F/Y436H, and were approved by MedImmune’s Review Board. Each animal (10/group) was injected i.m. with 2.5 μg of protein in a volume of 100 μl of PBS. Mice were bled from the retro-orbital sinus for each time point. Data were collected every 24 h until day 5 postinjection and every 48 h thereafter until day 12. Urine and serum were also collected at 6 and 24 h post-i.m. injection of 30 μg of each IgG1 (two animals/group). Concentrations were determined by an anti-human IgG ELISA. Briefly, individual wells of a 96-well Maxisorp Immunoplate (Nunc) were coated with 50 ng of a goat anti-MEDI-493 (anti-idiotype) Ab (MedImmune). The plates were blocked with 0.5% (w/v) BSA (Sigma-Aldrich, St. Louis, MO), incubated with samples or standards (0.1–20 ng/ml), then with a HRP conjugate of a goat anti-human Fc polyclonal Ab (Sigma-Aldrich). Peroxidase activity was detected with 3,3′,5,5′-tetramethylbenzidine (Kirkegaard & Perry Laboratories, Gaithersburg, MD), and the reaction was quenched with 0.18 M H₂SO₄. The absorbance at 450 nm was measured with a V_max kinetic microplate reader running SoftMaxPro 3.1.1 software (Molecular Devices, Sunnyvale, CA). The t_1/2 of the elimination phase (β phase), which takes place after the rapid redistribution phase, were determined using a one-phase exponential decay model provided by GraphPad Prism (GraphPad Software, San Diego, CA), using data points between day 1 and 12 postinjection.

Analyses were conducted in BALB/c mice for MEDI-493 wild type, M252Y/S254T/T256E, and H433K/N434F/Y436H. Using two mice per group, each animal was injected i.v. or i.m. with 30 μg of Ig in a volume of 100 μl of PBS or with 100 μl of PBS. After 24 h, animals were killed. Lungs, spleen, liver, intestine, heart, kidneys, bladder, as well as muscle at the site of i.m. injection were harvested.

Organs were removed into 1 ml PBS, ground, and lysed by sonication, and lysates were cleared by centrifugation. Immunoprecipitation was performed for 2 h at room temperature using 50 μl of a 50% suspension of goat anti-human IgG-Sepharose 4B conjugate (Zymed, South San Francisco, CA) previously washed three times with 1 ml triethylamine 100 mM, pH 11.0. Beads were washed three times with 1 ml PBS/0.1% Tween 20, pH 7.2, and elution was conducted in 50 μl 100 mM triethylamine, pH 11.0. Eluates were boiled for 5 min and loaded on a SDS-PAGE (4–20% gels; Invitrogen) in presence of 50 mM DTT. Electrotransfer to nitrocellulose filters was made as described (34). Detection was conducted with a HRP conjugate of a goat anti-human IgG (anti-Fab; Sigma-Aldrich) and the ECL Western blotting detection reagents (APBiotech).

Organs were removed into neutral buffered 10% Formalin solution (Sigma-Aldrich). Immunohistochemistry experiments were conducted at Molecular Histology (Gaithersburg, MD) on paraffin-embedded samples using a HRP conjugate of a goat anti-human IgG (anti-whole molecule; Sigma-Aldrich) and standard protocols (DAKO Handbook on Immunochemical Staining Methods; DAKO, Carpenteria, CA).

Analyses were conducted in BALB/c mice. Briefly, MEDI-493 wild type, M252Y/S254T/T256E, and H433K/N434F/Y436H were radiolabeled with Na¹²⁵I by the chloramine-T procedure (41) at Lofstrand Labs (Gaithersburg, MD) to a specific radioactivity of ∼5 × 10⁶ dpm/μg. Using two mice per group, each animal was injected i.m. with 2.5 μg of iodinated Ig in a volume of 100 μl of PBS. After 24 h, animals were killed. Lungs, spleen, liver, intestine, and heart were harvested and transferred into neutral buffered 10% Formalin solution (Sigma-Aldrich). The radioactivity present was quantitated by phosphor imaging analysis after 3-wk exposure (42) using a Fuji BAS 5000 running Image Gauge V3.45.

Libraries of human hinge-Fc fragments (residues 226–478, Kabat numbering) (21) derived from the human IgG1 MEDI-493 (31) were expressed using the pCANTAB5E phagemid vector. As previously noted (11), functional Fc-hinge homodimers linked to the gene III coat protein are formed and displayed at the surface of M13 by virtue of the leakiness of the amber stop codon located between hinge-Fc:E-Tag and gene III.

Four libraries (1, 2, 3, 4) were constructed by Kunkel-based mutagenesis, which contained Fc variants with residues randomized at contact positions between Fc and murine or human FcRn. Library 1 targeted positions 251, 252, and 254–256 (EU numbering) (21); library 2 positions 308, 309, 311, 312, and 314; library 3 positions 385, 386, 387, and 389; and library 4 positions 428, 433, 434, and 436. D312 and N389 are not predicted to contact murine or human FcRn, but due to their solvent-exposed nature as well as close proximity to the FcRn binding site, these residues were also randomized. Residues I253, H310, and H435 are highly conserved across species (21), play a major role in the energetics and pH specificity of binding to murine and human FcRn (8, 24, 29), and therefore were fixed during mutagenesis to retain the crucial functional epitope. To limit the size of the libraries and to avoid cumbersome sampling procedures, W313, E388, H429, E430, A431, and L432, which are not predicted to contact rat or human FcRn (16, 23), were also fixed. Libraries 1 and 2 consist of 1 × 10⁸ clones each, representing a significant proportion of the 3.4 × 10⁷ (32⁵) possible codon permutations from the NNS randomization strategy. Likewise, libraries 3 and 4 of 1 × 10⁷ clones each represented a significant proportion of the 1.0 × 10⁶ (32⁴) possible codon permutations. After three rounds of panning, a strong consensus sequence had emerged for libraries 1, 2, and 4 (Fig. 1), but not library 3. After six rounds of panning, a consensus sequence had appeared for library 3 (Fig. 1). Clones isolated from library 1 have predominantly L at position 251, Y at 252, T or S at 254, R at 255, and broader variability at 256 (Fig. 1,B). Clones isolated from library 2 have predominantly T at position 308, P at 309, S at 311, D at 312, and L at 314 (Fig. 1,B). Clones isolated from library 3 have more variability at every position, but predominantly R at position 385, T at 386, R at 387, and P at 389 (Fig. 1,B). Clones isolated from library 4 have predominantly M at position 428, Y at 434, and broader specificity at other positions, with H, R, K, and S being frequently found at 433, and H, Y, R, and T at 436 (Fig. 1 B).

The ability of murine and human FcRn to bind the Fc variants was investigated using SPR (BIAcore). Representative mutations identified after panning libraries 1 through 4 (Fig. 1,A) were introduced into the Fc portion of a human IgG1 (Table I). Injection of different concentrations of human or murine FcRn over the immobilized IgG1 variants gave concentration-dependent binding at pH 6.0. A typical resonance profile for equilibrium binding is shown for the M252Y/S254T/T256E mutant (Fig. 2, A and B). In all cases, equilibrium (or near-equilibrium) binding levels were reached within 50 min. The Scatchard plots for the binding of the M252Y/S254T/T256E mutant to murine and human FcRn are shown in Fig. 2, C and D. The plots were all linear, and apparent K_a were calculated from the relevant slopes. Measurements were conducted in duplicate or triplicate and confirmed that the immobilized IgGs retained their original binding activity.

Because there are two nonequivalent binding sites on human IgG for mouse/human FcRn (17, 43), the receptor was used in solution to avoid avidity effects. Consistent with this, systematically higher affinities are observed when FcRn, rather than IgG, is immobilized on the biosensor chip (17, 29, 44, 45). Under our experimental conditions, mainly interactions corresponding to the higher affinity association (i.e., single-liganded receptor) are measured, accounting for the linearity of the Scatchard plots.

The K_d for the interaction of wild-type human IgG1 with murine and human FcRn at pH 6.0 (269 and 2527 nM, respectively) agree well with the values determined by others (8). Our data (Table I) indicate that the I253A mutation virtually abolishes binding to human and murine FcRn, as reported previously (12, 24). This is not the result of misfolding of the Ab, as this mutant retains the same sp. act. as the wild-type molecule (MEDI-493) in a microneutralization assay (31) (data not shown).

We generated human IgG1 mutants with increased binding affinity toward both murine and human FcRn at pH 6.0 (Table I). Overall, improvements in complex stability were less marked for the human IgG1-human FcRn pair than for the human IgG1-murine FcRn pair. The largest observed increases in affinity for single Fc region mutants toward murine and human FcRn compared with wild-type IgG1 were 30 (N434F/Y436H)- and 11 (M252Y/S254T/T256E)-fold, respectively. However, ranking of the most critical positions remains unchanged when comparing human and murine FcRn: the largest increases in IgG1-murine FcRn complex stability (ΔΔG ≥ 1.3 kcal/mol) occurred for mutations at positions 252, 254, and 256 (M252Y/S254T/T256E and M252W) and 433, 434, and 436 (H433K/N434F/Y436H and N434F/Y436H). Likewise, the same mutations were found to have the most profound impact on the IgG1-human FcRn interaction and also resulted in the largest increases in complex stability (ΔΔG ≥ 1.0 kcal/mol). Substitutions at positions 308, 309, 311, 385, 386, 387, and 389 had little or no effect on the stability of the complexes involving human or murine FcRn (ΔΔG ≤ 0.4 kcal/mol). Therefore, residues at the center of the Fc-FcRn combining site contribute significantly more to improvement in complex stability than residues at the periphery (Fig. 3). Two combination variants were tested in which two Fc regions were simultaneously substituted. When tested against human FcRn, M252Y/S254T/T256E/H433K/N434F/Y436H shows an additive effect and exhibits binding that is better than M252Y/S254T/T256E and H433K/N434F/Y436H, but does not bind significantly better than H433K/N434F/Y436H to murine FcRn. M252Y/S254T/T256E/G385R/Q386T/P387R/N389P does not show any additive effect, as binding to both murine and human FcRn is not significantly different from M252Y/S254T/T256E.

Phage-derived IgG1 mutants exhibiting a significant increase in affinity toward murine FcRn at pH 6.0 (ΔΔG ≥ 1.0 kcal/mol) showed a parallel significant increase in binding to the mouse receptor at pH 7.4 (SPR signal_{pH 7}.₄/SPR signal_{pH 6.0} ∼0.5 at equilibrium; Fig. 4, C and D). IgG1 mutants exhibiting a small increase in affinity toward murine FcRn at pH 6.0 (ΔΔG < 0.4 kcal/mol) bound poorly to the mouse receptor at pH 7.4 (Fig. 4,B). Raising the pH to 8.5 resulted in a dramatically lower binding of the mutants to murine FcRn when compared with at pH 6.0 and 7.4 (Fig. 4, A–D). In contrast, IgG1 mutants exhibiting a large affinity increase toward human FcRn at pH 6.0 (ΔΔG up to 1.4 kcal/mol) only showed minimal binding to the human receptor at pH 7.4 (SPR signal_{pH 7}.₄/SPR signal_{pH 6.0} < 0.1 at equilibrium; Fig. 4, G and H). Only M252Y/S254T/T256E/H433K/N434F/Y436H (ΔΔG = 2.4 kcal/mol at pH 6.0) bound significantly to human FcRn at pH 7.4 (SPR signal_{pH 7.4}/SPR signal_{pH 6.0} = 0.45 at equilibrium; unpublished observations). None of the variants tested bound at a detectable level to human FcRn at pH 8.5.

To correlate magnitude of the increase in affinity of the Fc-FcRn interaction with biological consequences, three IgG1 mutants were selected for pharmacokinetic studies. Those display large (M252Y/S254T/T256E and H433K/N434F/Y436H) to small (G385D/Q386P/N389S) increases in affinity toward both murine and human FcRn. The pharmacokinetics of MEDI-493 wild type, G385D/Q386P/N389S, M252Y/S254T/T256E, and H433K/N434F/Y436H were analyzed in BALB/c mice. Those molecules were injected in equal amounts into different mice, and their serum concentration (reported in ng_{variant/wild type}/ml_serum) was determined by ELISA (see Materials and Methods). We report an inverse correlation between the binding affinity of the IgG1 variants to murine FcRn at pH 6.0 and their serum concentration (Fig. 5). M252Y/S254T/T256E and G385D/Q386P/N389S show significantly lower than wild-type serum concentrations starting at day 1 post-i.m. injection of 2.5 μg/animal of IgG1, whereas H433K/N434F/Y436H is virtually undetectable at the same time point (Fig. 5). Similar clearance results were obtained after i.v. injection of 2.5 μg/animal of IgG1 (data not shown). The same trends are observed after i.v. or i.m. injection of a >10-fold higher amount (30 μg/animal) of H433K/N434F/Y436H and M252Y/S254T/T256E as early as 6 h postinjection (Fig. 6). This is not the result of an enhanced degradation of the variants by serum or blood cell-associated proteases or of denaturation, as they have the same ex vivo stability in mouse serum and whole blood as the wild-type IgG1 for 24 h at 37°C, as determined by an anti-human IgG ELISA requiring the simultaneous presence of both a functional idiotope and Fc region (see Materials and Methods, data not shown). The same assay did not detect MEDI-493 wild type, M252Y/S254T/T256E, or H433K/N434F/Y436H in urine. Although the IgG1 variants we tested exhibit lower serum concentration when compared with wild type, β phase t_1/2 are not significantly different for MEDI-493 wild type, G385D/Q386P/N389S, and M252Y/S254T/T256E (Table II). The relationship, if any, between serum concentrations and β phase t_1/2 is at present unclear.

Presence of MEDI-493 wild type, M252Y/S254T/T256E, and H433K/N434F/Y436H was investigated by immunoprecipitation of different organs and tissue (lungs, spleen, intestine, heart, kidney, bladder, and site of i.m. injection) using an anti-human IgG conjugate. Muscle tissue at the site of injection revealed two polypeptide chains with apparent molecular mass ∼50 and ∼30 kDa, corresponding to the H and L chains, respectively, of MEDI-493 wild type, M252Y/S254T/T256E, and H433K/N434F/Y436H (Fig. 7). No significant difference could be observed between the three IgGs. Those molecules could not be detected in lungs, spleen, intestine, heart, kidney, or bladder after i.m. or i.v. injection. Those results indicate that no enhanced retention or degradation of the mutants is taking place in muscle tissue. In agreement with those data, immunohistochemistry on lungs, spleen, liver, intestine, heart, kidney, and bladder did not reveal any significant amounts of M252Y/S254T/T256E and H433K/N434F/Y436H (unpublished observations).

To more accurately determine the concentrations of MEDI-493 wild type, M252Y/S254T/T256E, and H433K/N434F/Y436H in different organs, a more sensitive detection assay was used in which IgG1s were iodinated and i.m. injected in BALB/c mice. Different tissues were then submitted to a phosphor imaging analysis. It revealed that MEDI-493 wild type is found at the highest levels in lungs, spleen, heart, and liver. Intestine exhibits lower levels. M252Y/S254T/T256E and H433K/N434F/Y436H follow the same trend, but at significantly lower concentrations overall than MEDI-493 wild type (Table III, Fig. 8). Those results indicate that no significant enhanced retention of the mutants is taking place in the organs tested. No conclusion pertaining to the degradation rate of the mutants in those organs can be formulated because the lower-than-wild-type organ concentrations might just reflect the overall lower serum concentrations.

In this study, we have demonstrated that the stability of the complex between murine or human FcRn with a human IgG1 can be increased by >30-fold by screening libraries of randomized Fc residues. Four libraries were designed that targeted Fc residues in contact with, or in close proximity of, the α1/α2 domains of murine and human FcRn (1, 15, 16, 23). These Fc libraries introduced substitutions in three loops at the CH₂-CH₃ junction (strands B and E of CH₂, and G of CH₃) and in one region between strands C and D of CH₃.

Engineering of the Fc-FcRn interaction indicates that major improvements in murine FcRn-human IgG1 and human FcRn-human IgG1 complex stability (0.6 kcal/mol ≤ ΔΔG ≤ 2.4 kcal/mol) occur on substituting residues located in a band across the Fc-FcRn interface (M252, S254, T256, H433, N434, and Y436). Substitutions of residues at the periphery (V308, L309, Q311, G385, Q386, P387, and N389) resulted in little or no increase in the free energy of complex stabilization (ΔΔG ≤ 0.4 kcal/mol). V308T/L309P/Q311S, G385R/Q386T/P387R/N389P, and, to a lesser extent, G385D/Q386P/N389S were strongly selected for (Fig. 1, A and B), but did not exhibit a marked increase in affinity toward both murine and human FcRn (ΔΔG ≤ 0.4 kcal/mol). This suggests that the two peripheral FcRn-contacting regions are not suitable targets for the engineering of IgGs with significant increases in affinity for FcRn. This might reflect the hot spot nature of the central region. In this situation, the energy of stabilization arises primarily from a few residues located at the center of a protein-protein interface (40). In fact, central residues I253 and H435 have already been shown to be critical to formation of the complex between human Fc and murine FcRn (12), and I253, S254, H435, and Y436 play a crucial role in the interaction of human IgG1 with human FcRn (8, 24). Thus, proximity of a pre-existing hot spot might be an important criteria to consider when engineering mutants for improved binding activity. However, H310 has been shown to be part of the energetic epitope for both human and murine FcRn, and yet mutations in its vicinity did not yield any variant with improvement in affinity above 1.7-fold. More likely is that additional factors around the substituted residue, such as the local chemical microenvironment, protein conformational changes, as well as solvent reorganization, have to be taken into account. Alternatively, the observed pattern might reflect other factors, such as library design, incomplete library sampling, and inefficient selection of more favorable residues at the peripheral positions due to a decrease in Fc stability or less efficient display at the surface of M13.

Interestingly, the variant with the highest affinity to human FcRn was obtained by combining the M252Y/S254T/T256E and H433K/N434F/Y436H mutations and exhibited a 57-fold increase in affinity relative to the wild-type IgG1. In comparison, both single substitutions resulted in only an 11- and 6.5-fold improvement in binding to human FcRn, respectively, as compared with wild-type IgG1. In contrast, the combination of M252Y/S254T/T256E with G385R/Q386T/P387R/N389P did not result in further affinity improvement toward human FcRn relative to M252Y/S254T/T256E. This confirms the dominance of the central over the peripheral residues in the energetics of the complex formation.

Efficient binding of human Fc to murine FcRn at pH 6.0 requires the presence of several wild-type Fc residues. For example, leucine is very conserved at 251, arginine at 255, aspartic acid at 312, leucine at 314, and methionine at 428 (Fig. 1). Retention of those wild-type residues indicates they are the most favorable for efficient binding to murine FcRn. These amino acids are well conserved in human and murine IgGs (21). In particular, D312 is not predicted to contact human or murine FcRn, and conservation of this residue in the phage-derived clones suggests that it has an important, but indirect role in the energetics of association with murine FcRn. Alternative explanations include biases in the library or incomplete sampling.

Another specificity trend is observed when one considers positions 308, 309, and 311 in which threonine, proline, and serine, respectively, are very strongly favored over the corresponding wild-type residues (Fig. 1). Surprisingly, this does not correlate with the magnitude of increase in affinity, as V308T/L309P/Q311S binds less than 2-fold better than the wild-type IgG1 to both human and murine FcRn (Table I). There appears to be a selection for residues with an aromatic side chain at position 252 (Y, F, and W). Engineering of the interaction of murine Fc fragments with murine FcRn has already revealed that a leucine at this position is the most favorable residue (11). No clones were found in our experiments that contained a leucine at 252. This difference might reflect the distinct local environment present in both murine and human Fc.

The three-dimensional structure of human and rodent FcRn is very similar (17). However, there are a number of differences at the interface of contact with Fc, the most notable of which is a substitution of L135 (human) for D137 (mouse). Our selection results indicate a strong trend toward conservation of H and selection of K and R at position 433 as well as toward selection of H and R at position 436. This underlines the importance of a positively charged residue at those locations. Interestingly, histidine at position 436 is highly conserved in murine IgG1 sequences, in which it is thought to interact through a salt bridge with murine FcRn D137 (23). In contrast, tyrosine at position 436 is conserved in human IgG1 sequences, in which it is believed to make hydrophobic interactions with human FcRn L135. Selection of H436 after selection of random human Fc libraries on murine FcRn probably recreates the important interaction with D137. This is supported by the 30-fold increase in the affinity of N434F/Y436H for murine FcRn compared with wild type. Elucidation of the molecular mechanisms by which human FcRn not only accommodates a histidine at position 436, but also has enhanced binding to N434F/Y436H will require a more detailed analysis, such as the construction of additional mutants and x-ray crystallography.

Surprisingly, efforts to use human FcRn for screening the libraries were not successful due to the lack of consensus sequences after seven rounds of panning. A likely explanation is that the selection conditions we tested were detrimental to its stability and/or function. As a result, libraries were panned against murine FcRn, with a binding step at pH 6.0 and an elution step at pH 7.4. This strategy allowed direct isolation of clones that bound more tightly at pH 6.0 than at pH 7.4. However, there was a marked difference in the effect of pH on the binding of the IgG1 variants to either human or murine FcRn. Increases in the stability of the complex with murine FcRn at pH 6.0 correlate with increases in the stability of the complex at pH 7.4 (Fig. 4, A–D). In contrast, most IgG1 mutants bound very poorly to human FcRn at pH 7.4 (Fig. 4 E–H). In all cases, binding was dramatically decreased when pH was raised to 8.5, strongly suggesting the mutations we identified led to a change in the pKa of H310 and/or H435, which are responsible for the pH dependency of binding (23). This has clear implications for the pharmacokinetics of the IgG1 variants, as higher affinity to FcRn at pH 7.4 might adversely affect release into the serum and offset the benefit of the enhanced binding at pH 6.0. Our pharmacokinetic and immunohistochemistry studies in mice support this hypothesis because serum levels after i.m. or i.v. injection were consistently lower for G385D/Q386P/N389S, M252Y/S254T/T256E, and H433K/N434F/Y436H when compared with the wild-type molecule. Those mutants are also present at lower levels in different tissues when compared with wild type. No evidence of enhanced degradation in serum, blood, or muscle tissue at the site of injection was found. Likewise, there was no detectable enhanced excretion in urine. It is possible to hypothesize that because of their more efficient binding at pH 7.4, the IgG1 mutants are sequestered throughout the body in FcRn-containing tissues such as the endothelium of small arterioles and capillaries (20). Retention of the IgGs in cells may lead to their degradation. This may not be an issue in humans because the majority of the IgG1 mutants we tested do not bind well to human FcRn at pH 7.4 when compared at pH 6.0. If this hypothesis is correct, introduction of the mutations we identified into the Fc fragment of therapeutic Abs may increase their serum persistence.

Enhancements of the affinity of our human IgG variants for both human and murine FcRn are anticipated from further engineering, such as the randomization of new sets of residues and the selection of the best combinations of mutations by DNA shuffling (46). Those studies will yield important information pertaining to the molecular basis of protein-protein interactions. Provided retention of the pH dependency of binding can be conserved, it might be possible to generate IgG mutants exhibiting enhanced efficacy due to long serum t_1/2. Those molecules could be a valuable addition to the therapeutic Ab field and have potential utility in domains as diverse as cancer, respiratory diseases, neonatal medicine, and diagnostic.

We thank Fran Palmer-Hill, Mary Bladen, James Johnson, and Jing Tan for expert technical assistance.