Designed Protein-Protein Association

See allHide authors and affiliations

Science  11 Jan 2008:
Vol. 319, Issue 5860, pp. 206-209
DOI: 10.1126/science.1150421


The analysis of natural contact interfaces between protein subunits and between proteins has disclosed some general rules governing their association. We have applied these rules to produce a number of novel assemblies, demonstrating that a given protein can be engineered to form contacts at various points of its surface. Symmetry plays an important role because it defines the multiplicity of a designed contact and therefore the number of required mutations. Some of the proteins needed only a single side-chain alteration in order to associate to a higher-order complex. The mobility of the buried side chains has to be taken into account. Four assemblies have been structurally elucidated. Comparisons between the designed contacts and the results will provide useful guidelines for the development of future architectures.

In a cell, permanently associated proteins guarantee mechanical integrity, whereas transient associations are indispensable for metabolism and the regulation thereof. Our study focused on permanent contacts, many of which have been established at atomic resolution (1). Extensive analyses (28) showed that these contacts are diversified and that only a few rather general rules can be extracted: The contact area should be larger than about 600 Å2, and the contacting surfaces should be complementary and predominantly nonpolar. Moreover, the contribution of hydrogen bonds and salt bridges at the contact rim is negligible. It is rather easy to destroy a contact by introducing a bulky side chain that is sterically incompatible. It is also comparatively simple to produce weak contacts randomly, as demonstrated by many examples using surface mutations to induce proteins to crystallize (9, 10). However, the creation of a desired permanent contact is a more difficult task (11, 12). Whereas establishing tight contacts by screening is well known from the antigen-antibody system that has also been applied in engineered complexes (13), the designed production of a tight novel contact has to our knowledge only been tried once, but the system failed to crystallize (14). However, in a number of cases, a given contact was modified by mutations (1518). Here, we report the production and analysis of designed permanent homo oligomers of five proteins.

The importance of symmetry for the association of identical protein subunits was recognized some time ago (19). At the lowest symmetry level, a single contact patch has to provide the whole binding energy, which usually involves a large number of residues. As a consequence, the creation of such a contact requires numerous mutations. Because these contacts also generally form infinite helices that cannot be crystallized (Fig. 1A), we did not try to construct such an assembly. In contrast, a C2-symmetric dimer (20) has a contact multiplicity of 2 and therefore requires only half the number of mutations. Moreover, it is usually globular and thus crystallizable (Fig. 1A). The same applies to higher symmetries. Associating two C2-symmetric dimers along their molecular axes results in a D2 tetramer with a contact multiplicity of 4 (20), whereas C4 tetramer units give rise to D4 octamers with a multiplicity of 8 (fig. S4). Obviously, the higher the symmetry, the fewer mutations are required, which explains the abundance of symmetry in natural homo oligomers.

Fig. 1.

Design of protein assemblies (24). The proteins in (C) to (H) are depicted as thick-lined Cα plots at various scales with mutated residues as colored spheres. (A) Sketch of an asymmetric interface between patches a and b, which, in general, gives rise to an infinite helix (top). A C2-symmetric interface also between patches a and b doubles the numbers of contacts and forms a globular complex (bottom). Along the same lines, the reported D2, D4, and D8 oligomers have 4-, 8-, and 16-fold contacts, respectively (fig. S4). (B) Side-chain mobility of the C4-symmetric Rua, color-coded from 0° (blue) to 90° (red) angular spread in the torsion angles χ1 and χ2 (24). The C- and N-terminal domains are at the top and bottom, respectively. (C) Pga-A and -B designed in crystal contact a-a(25). (D) Pga-C and-D designed in crystal contact f-f (25). (E) Oas-A and-B planned as a D2 tetramer at a rotation angle of 86° around a common molecular twofold axis (vertical). (F) Oas-C designed as a D2 tetramer at an alternative rotation angle of 29°. (G) Designed D2 tetramer of Uro-A around a common molecular twofold axis (vertical). The designed contact is between the NAD+-binding domains (residues 142 to 343), which are given in lighter hues. (H) Designed octameric Rua-D with a head-head contact.

Whereas the symmetry concerns the framework of a design, the actual production of homo oligomers was based on the exchange of side chains on the surface. We refrained from touching any main chains, because the corresponding structural changes are much more difficult to predict. Because contacts rigidify the involved side chains, the side-chain mobility describes part of the entropy reduction required for associations (21, 22). Therefore, if possible, mobile side chains have to be identified and eliminated. Because the mobility of a side chain depends on its environment, its derivation from a single crystal structure is difficult. However, for the enzyme l-rhamnulose-1-phosphate aldolase (Rua) (23)10 high-resolution crystal structures with different packing arrangements are available (24). Within this group, we compared the side-chain torsion angles χ1 and χ2 for each relevant residue, derived the angular spread in the 10 structures, and defined it as the side-chain mobility (24). The resulting mobility distribution over the surface contained surprisingly few hot spots (Fig. 1B). Unfortunately, it was not available when we designed our contacts, but we were able to use it for the assessment of an unexpected contact (see below).

The production of homo oligomers was tried at increasing levels of symmetry. In our attempt to make dimers of monomeric 6-phospho-β-galactosidase (Pga) (25), we enriched two crystal contacts across local twofold axes (Fig. 1, C and D) with nonpolar residues. In order to design tetramers from dimeric O-acetylserine sulfhydrylase (Oas) (26) and urocanase (Uro) (27), we aligned the dimers along their molecular axes and performed a one-dimensional search along the relative rotation angle (fig. S4), which yielded two suitable orientations for Oas and one for Uro (Fig. 1, E to G). The same was carried out for Rua (23) and its molecular fourfold axis (Fig. 1H and fig. S4B). The alignment along an axis allows three arrangements of the contacting partners, which we designated head-head, tail-tail, and head-tail (equals tail-head), defining an arbitrary end as the head. With Oas and Uro, we performed only a tail-tail association, whereas all three types of association were assayed with Rua. The D8-symmetric tail-tail association of an engineered fragment of the mycobacterial porin MspA (Myp) (28) was fully unexpected.

Guided by one of the crystal contacts of Pga, we produced the two mutants Pga-A and Pga-B (24) that yielded about 3% dimer, as shown by size-exclusion chromatography (SEC) and dynamic light scattering (DLS) (Fig. 2A and Table 1). The Pga-A dimer was stable in rechromatography and in DLS. Crystallization attempts of the dimer fraction failed. The single dimer peak in the rechromatography (Fig. 2A) demonstrated that the obtained dimer was stable. Its structure is certainly similar to that shown in Fig. 1C because the contact is defined by the applied mutations. Following the other crystal contact, we constructed the two mutants Pga-C and Pga-D (24) that resulted in about 50% dimer (Fig. 2B). Both dimers were stable in rechromatography and in DLS (Table 1). However, neither of them yielded crystals, although a total of 24 mg of dimer was used in crystallization screens. Again, the dimer structure is most likely close to that of Fig. 1D.

Fig. 2.

SEC of modified and native proteins (24). Each peak is labeled with its apparent mass (kD). Initial runs are given in solid lines, rechromatographed fractions in dashed lines, and standards in dotted lines. (A) Pga-A with monomer and covalent-dimer standard (24). (B) Pga-D with monomer and covalent-dimer standard (24). (C) Oas-A with rechromatographed tetramer and dimer. (D) Uro-A with wild-type standard. (E) Rua-A and Rua-E (dash-dot line) with wild-type standard. The Rua-B and Rua-C distributions were identical to that of Rua-A. (F) Hexadecameric Myp-A with octameric Myp-B as standard (24).

Table 1.

Protein-protein associations (24). Double prime entries indicate the same entry as in the line above; dash entries indicate that a small number of experiments were unsuccessful; none means no success in a very large number of experiments.

View this table:

On the basis of the molecular twofold axis of native Oas, we designed a D2-symmetric tetramer by axis alignment and rotation angle adjustment. We chose a tail-tail association and constructed the three mutants Oas-A, -B, and -C (24) for association at rotation angles of 86° and 29° (Fig. 1, E and F). The obtained tetramer fractions ranged around 6% (Fig. 2C). For all three mutants, rechromatography resulted in pure tetramers, indicating that they were stable, and adopted structures similar to those of Fig. 1, E and F, as defined by the positions of the mutations. We produced a total of 35 mg of tetramers, which we used in various crystallization screens. These attempts yielded a number of initial crystals. Among them, one Oas-C crystal form grew to sizes suitable for an x-ray analysis. Unfortunately, it contained only dimers (Table 1 and tables S1 and S2).

With Uro we followed the same approach as with Oas. We chose a tail-tail association at the nicotinamide adenine dinucleotide (NAD+) domains and found a suitable rotation angle of 65° (Fig. 1G). We introduced two small nonpolar side chains in place of two polar side chains and adjusted one nonpolar side chain in order to improve the fit. The resulting mutant, Uro-A, formed a tetramer fraction of 80% or more (Fig. 2D). The purified tetramer was stable in SEC and DLS (Table 1). It was crystallized, and the structure was established at high resolution (24). The actual tetramer followed the design in general terms (Fig. 3A). An analysis showed that the two observed contacts between the four NAD+ domains were nearly identical, which confirmed that they were highly specific. However, the contact that was actually produced deviated from that which was designed by a relative shift of about 1.5 Å between the contact partners, and this shift was incompatible with the designed D2 symmetry (Fig. 1G).

Fig. 3.

Established oligomer structures (24). All mutations are marked by purple spheres. (A) Crystal structure of C2-symmetric Uro-A showing the twofold molecular symmetry axis (red) and four local twofold axes relating the cores (darker colors) and the NAD+ domains (light colors) to their counterparts. The interface between core and NAD+ domains was broken in the lower left and upper right chains. (B) D4-symmetric octamer Rua-A. (C) C2-symmetric octamer Rua-B. (D) Negatively stained electron micrograph of Rua-E showing the fiber association and a Rua-A octamer (B) at the scale defined by the box edge. (E) Native mycobacterial porin (28). The encircled membrane-immersed part was deleted, giving rise to Myp-A. (F) D8-symmetric association of two Myp-A molecules (top and bottom ring). The positions of the 52-residue deletions are marked by red spheres (fig. S1D).

Apparently, the designed contact was so strong that this incompatibility caused a disruption of the associated proteins themselves rather than preventing contact formation. The observed new contact caused two of the four NAD+ domains to be lifted off the protein cores, one in each dimer. This was unexpected because the native contact area between core and NAD+ domain was much larger than the newly formed interface. As a consequence, the designed global D2 symmetry broke up into four local C2 symmetries, the axes of which were tilted by about 10° and displaced by about 2 Å in relation to the D2 design (Fig. 3A). However, the tetramer kept a global C2 symmetry defined by the twofold axis pointing toward the viewer.

Rua is a C4-symmetric tetramer that is associated via its C-terminal domains, whereas its N-terminal domains dangle like tails into the solvent (Fig. 1H). We designed the D4-symmetric tail-tail complexes Rua-A, Rua-B, and Rua-C with inter-digitating tails and very large interfaces (Table 1 and table S1). Given the contact multiplicity of 8 for the design, single point mutations in Rua-A and -B and a double mutation in Rua-C sufficed to yield 100% octamer in SEC and DLS (Fig. 2E). Moreover, all three mutants crystallized. The high-resolution structures showed that Rua-A formed the designed octamer (Fig. 3B), whereas Rua-B formed an octamer with displaced four-fold axes (Fig. 3C) and Rua-C dissociated into tetramers during crystallization (24). Surprisingly, the observed contact area of the Rua-B octamer was only 60% of the planned one, and its symmetry was merely C2 (Table 1). However, it turned out that its average side-chain mobility was 32° (angular spread) compared to 43° of the respective area of the designed contact (Fig. 1B). We therefore suggest that the unexpected contact is explained by a lesser reduction of the side-chain mobility, which amounts to a lesser entropy reduction on association.

Following the successful tail-tail association, we designed the D4-symmetric head-head octamer Rua-D (Fig. 1H), which showed an octameric fraction of 60% confirmed by SEC and DLS (Table 1). However, the octamer fraction yielded crystals that contained only tetramers. In the next step, we combined the tail-tail mutant Rua-A with the head-head mutant Rua-D, resulting in the head-tail mutant Rua-E. With its contact surfaces at both ends of the fourfold axis, Rua-E was designed to form polymers. The polymerization was confirmed by SEC (Fig. 2E). Moreover, electron micrographs demonstrated that these polymers formed the expected fibers (Fig. 3D).

In a separate project, our aim was to produce a soluble carrier for large nonpolar molecules, such as organic catalysts, by removing the membrane-immersed part of the mycobacterial porin MspA (28). A 52-residue deletion yielded Myp-A (Fig. 3, E and F). To our surprise, Myp-A formed a soluble hexadecameric complex instead of the expected octameric ring. The hexadecamer was identified by SEC of Myp-A in comparison with the octameric mutant Myp-B (24) (Fig. 2F). Myp-A was crystallized and structurally elucidated, showing a D8-symmetric double ring associated at the position of the deletion (Fig. 3F).

Reviewing our data, we find that the introduction of large nonpolar side chains is an efficient means of constructing permanent contacts. Our experience with Pga-A, -B, -C, and -D was that only a few of them were necessary to produce contacts with very low dissociation and association rates at 4°C under physiological conditions. However, such side chains tend to reduce the protein yield (24), presumably because of folding problems. Moreover, these easily obtained contacts were most likely not well defined because extensive crystallization attempts with Pga-C and Pga-D dimers never resulted in crystals (24), which is uncommon for rigid proteins but a well-known problem with, for instance, fusion proteins (29). We conclude that it is relatively easy to form complexes by introducing large nonpolar side chains, such as phenylalanines or tryptophans, but rather difficult to construct a rigid one.

In light of our observations with Pga, we introduced smaller nonpolar residues such as valines and isoleucines into dimeric Oas and dimeric Uro. The mutants of Oas were tetramers in solution and yielded small crystals. However, it was only possible to establish a crystal structure containing a native dimer. Along the same lines, the Rua-C and Rua-D octamers found at pH = 7.0 in SEC and DLS yielded crystals at this pH that contained only native tetramers (24)(Table 1).

Moreover, a second crystal form of Rua-A grown at pH = 4.9 in a low-polarity solvent contained only tetramers, although Rua-A formed 100% octamers in this pH range in SEC and DLS. The same applies for a second crystal form of Rua-B grown at pH = 5.0 in a low-polarity solvent. In both cases, the tetramers observed in the crystals were confirmed by SEC analyses at conditions close to those of the respective crystal forms. We conclude that the produced Rua octamers were strong enough in a mild physiological environment but failed to withstand harsher crystallization conditions in, for example, high salt concentrations, extreme pH values, or low-polarity solvents. Achieving stability in a broad range of unphysiological conditions is a more difficult task.

Symmetry is an important factor in protein association because it enhances the multiplicity of a single point mutation (Fig. 1A). The highly symmetric Rua octamers with a contact multiplicity of 8 form complexes after merely one or two mutations, and the 16-fold contact of Myp-A required no designed mutations at all. On the other hand, high multiplicity is hazardous for evolving organisms that need to avoid detrimental mutations. This may explain the rare occurrence of C4 tetramers such as Rua compared to the ubiquitous D2 tetramers, because C4 can associate with eightfold contacts whereas the contacts in an association of D2 symmetric partners are fourfold at most (fig. S4).

To what extent did our constructs follow the design? The answer depends on the details available. For all reported mutants (table S1), the designs (Fig. 1, C to H) were confirmed by the comparatively rough SEC and DLS data. Uro-A fitted the design at low but deviated at high resolution. The same applies for the Pga and Oas complexes, where the mutations most likely form the contacts. The available high resolution of Uro-A revealed that the novel contact deviated slightly from the design but was so strong that it caused the opening of a large, surprisingly weak interface within the protein partners (Fig. 3A). Rua-A followed the design except for a 3 Å shift that did not change the symmetry (Fig. 3B), whereas Rua-B formed an unexpected C2 symmetric complex with a much smaller contact area (Fig. 3C). Because this Rua-B complex can be explained by the influence of the side-chain mobility, the mobility is probably an important factor in contact design. Our experiments demonstrate that the production of a particular contact is quite feasible, whereas high precision seems difficult to achieve.

Supporting Online Material

Materials and Methods

Figs. S1 to S4

Tables S1 and S2


References and Notes

View Abstract

Stay Connected to Science

Navigate This Article