Convergent Solutions to Binding at a Protein-Protein Interface

See allHide authors and affiliations

Science  18 Feb 2000:
Vol. 287, Issue 5456, pp. 1279-1283
DOI: 10.1126/science.287.5456.1279


The hinge region on the Fc fragment of human immunoglobulin G interacts with at least four different natural protein scaffolds that bind at a common site between the CH2 and CH3domains. This “consensus” site was also dominant for binding of random peptides selected in vitro for high affinity (dissociation constant, about 25 nanomolar) by bacteriophage display. Thus, this site appears to be preferred owing to its intrinsic physiochemical properties, and not for biological function alone. A 2.7 angstrom crystal structure of a selected 13–amino acid peptide in complex with Fc demonstrated that the peptide adopts a compact structure radically different from that of the other Fc binding proteins. Nevertheless, the specific Fc binding interactions of the peptide strongly mimic those of the other proteins. Juxtaposition of the available Fc-complex crystal structures showed that the convergent binding surface is highly accessible, adaptive, and hydrophobic and contains relatively few sites for polar interactions. These are all properties that may promote cross-reactive binding, which is common to protein-protein interactions and especially hormone-receptor complexes.

Protein-protein interactions are central to the control of many biological functions, but we do not yet understand what general features (if any) of a protein surface are most important for binding. Nature often provides convergent solutions to biological problems, and study of this “consensus” information can provide insight into the essential requirements for function. There are many examples of hormones that bind multiple receptors, or receptors that bind multiple hormones (1). Structural analysis of these complexes shows that the common protein will use virtually the same set of contact residues for binding many partners (2). One of the most striking and well-characterized examples of this principle is a consensus binding site found on the constant fragment (Fc) of immunoglobulin G (IgG), which interacts with four different proteins, each having radically different folds (Fig. 1) (3). We wondered if this common site was selected because its location is optimal for biological function or because the intrinsic physical properties of the site make it optimal for binding many different proteins.

Figure 1

Ribbon diagrams of an IgG-Fc subunit (blue) in complex with (A) domain B1 of Protein A, (B) domain C2 of Protein G, (C) rheumatoid factor, and (D) neonatal Fc-receptor (3). All four proteins bind to an overlapping region at the CH2/CH3domain interface.

To address this question, we performed an in vitro selection (4) to isolate peptides that bound Fc without the constraint that the peptides function in vivo. A library of cyclic peptides was constructed that consisted of 4 × 109different peptides of the form XiCXjCXk(where C is cysteine, X is a random amino acid, and i +j + k = 18) (5). Peptides from this library were expressed polyvalently on the surface of M13 bacteriophage as NH2-terminal fusions to the gene VIII protein and selected for binding to immobilized Fc (6).

In principle, peptides could have been selected to bind to potentially any region of the Fc because of the unbiased nature of the library. However, after several rounds of selection, the library became dominated by a single peptide, Fc-I (ETQRCTWHMGELVWCEREHN) (7). Repetition of the selection experiment again gave Fc-I and also a related peptide, Fc-II (KEASCSYWLGELVWCVAGVE). The Fc-II peptide shared the cysteine spacing and the internal GELVW sequence seen in Fc-I. Apparently, these two peptides bound Fc with an affinity high enough to be selected over any other Fc binding peptides present in the starting pool. Both peptides were synthesized and found to compete with Protein A (Z-domain) (8) for binding to Fc with inhibition constants (K i's) of about 5 μM (9, 10), implying that these peptides bind to an overlapping site on Fc coinciding with the Protein A binding site.

The gene sequence for Fc-II was transferred to gene III of M13 bacteriophage (11) and improved by monovalent phage display. Five residue blocks were randomly mutated in six separate libraries to exhaustively cover the noncysteine positions in the peptide sequence (12). Preferred residues from selection of these libraries for binding to Fc were then recombined to give three more libraries spanning the peptide sequence (13). Selection patterns from these libraries suggested a 13-residue core Fc binding sequence (DCAWHLGELVWCT). The corresponding peptide (Fc-III) was synthesized and found to inhibit binding of Protein A (Z-domain) to Fc with aK i of 25 nM (14). Thus, although Fc-III is seven residues shorter than Fc-II, it binds 200 times more tightly. Despite its smaller size, the binding affinity of Fc-III to Fc was only about twofold weaker than that of the domains from Protein A and Protein G, which are each about four times larger and bind with K d's of around 10 nM (15).

To understand whether binding of Fc-III occurs through interactions similar to those of the natural Fc-binding domains, we determined the x-ray crystal structure of the Fc-III peptide in complex with Fc at 2.7 Å resolution (Table 1) (16). We found that the peptide adopts a β-hairpin conformation unrelated to any known Fc binding motif (Fig. 2). A β-bulge conformation at Leu-9 in Fc-III accommodates the noneven number of residues separating the two cysteines, and 8 of the 12 amino acid side chains make extensive interactions with Fc. The remaining side chains (the disulfide bridged cysteines, a tryptophan, and a leucine) create a small hydrophobic core on the back side of the peptide that is facilitated by the β bulge and type II β turn. Although the resolution of the structure is not sufficient to conclusively identify hydrogen bonding interactions, it appears that eight hydrogen bonds are formed between the peptide and Fc, including those involved in intermolecular salt bridges.

Figure 2

Crystal structure of Fc-III (DCAWHLGELVWCT-NH2), in complex with IgG-Fc. (A) Ribbon diagrams of two Fc-III peptides in complex with the Fc dimer; (B) close-up view of the peptide interacting with the surface of IgG-Fc. This structure has been deposited in the PDB under accession number 1DN2.

Table 1

Data collection and refinement statistics.

View this table:

The Fc-III peptide targets the consensus binding site of the natural Fc binding proteins (Fig. 3). Although much smaller in size, the Fc-III peptide covers almost as much area (650 Å2) on the surface of Fc as do the fourfold larger IgG-Fc binding domains from Protein A and Protein G and the 15-fold larger rheumatoid factor (each cover about 740 Å2) (17).

Figure 3

Molecular surface representation of the consensus binding site on IgG-Fc [coordinates from Deisenhofer (3)]. (A) Fc with superimposed binding interfaces of Protein A, Protein G, and rheumatoid factor. (The 4.5 Å crystal structure of the Fc-receptor/Fc complex was excluded because of its low resolution.) Atoms are colored blue, yellow, or red depending on whether they are involved in one, two, or three of the interfaces, respectively (17). (B) As in (A) with the interface of the Fc binding peptide (Fc-III) superimposed in green. Fc-III interacts with many of the atoms that are found in the interfaces of the other three Fc binding proteins.

Despite the lack of structural similarity between the peptide and the natural Fc binding proteins, detailed inspection of the interactions formed by all of these molecules with Fc reveals a number of common features in the consensus binding region (Fig. 4). For example, the phenyl ring of Phe-14 (18) in Protein A (Fig. 4E, 3) occupies the same position as the indole ring of Trp-11 of Fc-III (Fig. 4B, 3). The indole nitrogen of Trp-43 in Protein G (Fig. 4C, 2) makes the same hydrogen bond with Asn-433 on Fc as does the main-chain amide from Thr-13 of Fc-III (Fig. 4B, 2). Tyr-98H and Asp-31H (Fig. 4D, 6 and 9) from the rheumatoid factor heavy chain make identical hydrophobic and polar interactions, respectively, as do Val-10 and Glu-8 on Fc-III (Fig. 4B, 6 and 9), and Lys-28 from Protein G (Fig. 4C, 8) makes the same salt-bridge that His-5 does from the Fc-III (Fig. 4B, 8). In a striking example of repeated convergent evolution at the atomic level, the backbone amide of Val-10 in Fc-III (Fig. 4B, 5), Glu-27 in Protein G (Fig. 4C, 5), the backbone amide of Tyr-98H in rheumatoid factor (Fig. 4D, 5), and Gln-11 in Protein A (Fig. 4E, 5) all make the same buried hydrogen bond with the backbone amide proton of Ile-253 on Fc.

Figure 4

Topology of the consensus binding site on Fc. (A) Conserved interaction sites. The predominantly hydrophobic consensus region is shaded. Hydrogen bonding sites are shown with diagonal lines and salt bridging locations are denoted by open circles. Nitrogen and oxygen are colored blue or red, respectively, and carbon and sulfur atoms are colored green. Hydrogens are not shown. (B to E) Comparison of the Fc binding interactions of (B) the selected peptide, Fc-III (DCAWHLGELVWCT-NH2), (C) domain C2 from Protein G, (D) rheumatoid factor, and (E) domain B1 of Protein A. Numbers indicate the following conserved interactions: (1) salt-bridges with His-433, (2) hydrogen bonding to Asn-434, (3) hydrophobic packing onto His-435, (4) burial of the hydrophobic “knob” formed by Ile-253 and Ser-254, (5) hydrogen bonding to main chain (N-H) of Ile-253, (6) hydrophobic packing onto Met-252 and Tyr-436, (7) hydrogen bonding to Ser-254, (8) salt-bridges with Glu-380, and (9) salt-bridges with Arg-255. For clarity, only interfacial atoms are shown, and only nitrogen and oxygen atoms involved in conserved polar interactions are colored blue or red, respectively. The remaining contact atoms are colored yellow and green. The dynamic adaptability of this site can viewed in a movie on Science Online

With its slightly smaller contact surface, the peptide mimics two to six polar interactions and many analogous nonpolar contacts formed by each of the other Fc binding domains. Charge-charge interactions are distributed at the top and bottom of the binding site, hydrophobic contacts line both sides, and a strip of hydrogen bonding interactions runs directly through the center (Fig. 4). Interactions that are present in all of the binding interfaces are mediated by a shared set of contacts with atoms on six amino acid side chains—Met-252, Ile-253, Ser-254, Asn-434, His-435, and Tyr-436—as well as shared contacts with adjacent atoms on the peptide backbone. These atoms form a contiguous 525 Å2 patch of solvent-accessible surface area on Fc.

To assess whether the consensus region on the surface of Fc has unusual properties, we generated a comparison set of several million random surface patches of similar size on five Fc crystal structures (3, 19). These hypothetical binding sites were then evaluated according to multiple criteria and compared with the consensus site (Fig. 5, A and B). The consensus binding region was distinguished by a high degree of solvent accessibility and a predominantly nonpolar character, suggesting that burial of exposed hydrophobic surface area is an important driving force behind binding at this site. This result is consistent with statistical surveys of distinct protein interactions, which suggest an important energetic role for hydrophobic burial in protein association (20). In addition, the low hydrogen bonding ability of the surface in this region (19% of the surface is capable of hydrogen bonding compared with 37% on average) indicates that this site places proportionately fewer specific geometric constraints on binding partners, because fewer complementary polar interactions are required for binding (Fig. 5B).

Figure 5

The consensus binding site compared with other IgG-Fc surface patches. Atoms in the consensus region are more accessible (A) and less polar (B) than atoms in most hypothetical binding sites of similar size. Arrows indicate the surface properties of the consensus region averaged over four complexes (excluding Fc-receptor) and in uncomplexed IgG-Fc (19). The comparison set consisted of 2.5 million 525 Å2 surface patches of random globular shape distributed over the entire Fc dimer in five crystal structures. The solvent-accessible surface fraction in (A) is defined as the fraction of the maximum potential solvent-accessible surface area on protein atoms not buried by secondary or tertiary packing interactions. Because patches are of uniform size, those with high accessibility will have their accessible surface area concentrated onto fewer atoms than will patches with lower accessibility. In (B), atoms capable of acting as hydrogen bond donors or acceptors were classified as polar, and all other atoms were deemed nonpolar. (C) Fc dimer with symmetry related subunits shown in blue and yellow. Atoms found in patches with the highest average accessibility and lowest average polarity are colored red (21). The open circle identifies the location of the consensus binding site on one of the subunits. The consensus site on the other subunit is obscured on the back side of the dimer.

However, the consensus binding site is not the only exposed, less-polar region on the surface of the Fc dimer. We used patch analysis to search for regions with high solvent accessibility and below-average polarity (21) and found that the consensus binding site is part of a larger exposed hydrophobic region that extends about halfway across the CH2 domain and includes residues near the 309-helix and the 280's loop (Fig. 5C). Residues on the tip of the CH2 domain that would contact CH1 in an intact IgG molecule were also identified. Given that other equally accessible and nonpolar sites exist on Fc, other properties, such as shape or adaptability, must also contribute to making the consensus site the preferred locus for binding.

The consensus region on Fc undergoes a series of conformational changes in order to complement the distinct surface of each binding domain. Much of this adaptability is manifested in positional adjustments of Met-252 and its immediate neighbor Met-423. These residues adapt to form a pocket for Val-10 on Fc-III (Fig. 4B) and for Lys-31 on Protein G (Fig. 4C), or to present a much flatter surface for interaction with Tyr-98H in rheumatoid factor (Fig. 4D) and Phe-6 in Protein A (Fig. 4E). Similarly, Ile-253, His-433, and Asn-434 adopt different rotameric conformations depending on which protein is bound (Fig. 4, B to E).

Additional adaptability arises from changes in the relative orientation of the CH2 and CH3 domains. The location of the consensus binding site on Fc at a domain hinge is reminiscent of many cytokine receptor binding sites, which also show substantial structural adaptability and bind multiple ligands. The intrinsic adaptability of hinge region binding sites and the flexibility of the loop structures that typically constitute these sites may be important factors in facilitating promiscuity. Thus, the location of the consensus site at a domain hinge probably contributes to making this particular exposed nonpolar region the preferred locus for binding.

Mutagenesis experiments confirm an energetic role for the consensus patch in binding. Alanine replacement of Ile-253 or His-435 on Fc significantly disrupts binding (ΔΔG > 2.5 kcal/mol, where G is the Gibbs energy) of Protein A (Z-domain) (22, 23), and replacement of Asn-434, His-435, or Tyr-436 disrupts binding (ΔΔG > 1.5 kcal/mol) of the selected peptide. On the complementary side of these interfaces, alanine substitutions of Gln-11, Phe-14, or Ile-32 on the Z-domain (23) and Val-10 or Trp-11 on the selected peptide (24) all result in ΔΔG > 2.0 kcal/mol reductions in affinity.

The consensus binding site on Fc is thus an adaptive, exposed, nonpolar, and energetically important region on the surface of Fc that is primed for interaction with a variety of distinct molecules. In recent years, numerous peptides selected to bind other protein receptors have been found to target and block the binding sites of natural ligands (25), suggesting that most receptors contain preferred binding sites on their surfaces. These sites have evolved to bind natural ligands but are competent for binding other ligands as well. Our results show that a peptide can target such a site by mimicking the specific interactions of the natural binding domains while presenting the interacting groups from an entirely different structural scaffold. Further investigation of the properties that characterize attractive sites on protein surfaces should improve our understanding of binding and assist in the development of therapeutic ligands.

  • * To whom correspondence should be addressed. E-mail: jaw{at}


View Abstract

Navigate This Article