Protein oligomerisation, commonly referred to as protein quaternary structure, is the association of specific individual polypeptide chains through defined intermolecular interactions to form a single multimeric complex (Goodsell and Olson, 2000; Nooren and Thornton, 2003; Ali and Imperiali, 2005). So prevalent is oligomerisation in nature, protein oligomers are more common than their monomeric counterparts, at least in the protein data bank (Goodsell and Olson, 2000; Ali and Imperiali, 2005). Protein oligomers can comprise purely of non-covalent intermolecular interactions or also utilise inter-subunit covalent crosslinking, predominantly via disulphide bridges. Oligomerisation is largely seen as beneficial by reducing surface residues’ (especially hydrophobics) exposure to solvent, resulting in a lower surface area to volume ratio leading to improved stability against degradation and aggregation (Larsen et al., 1998; Ali and Imperiali, 2005; Gwyther et al., 2019). Crucially, oligomerisation also leads to functional features not available in monomers; they locally concentrate multiple active sites resulting in improved activity and enabling functional cooperativity whereby synergy [communication] between each polypeptide unit can positively or negatively regulate activity or even lead to new properties (Goodsell and Olson, 2000; Gwyther et al., 2019).

Given the benefits of protein oligomerisation, protein designers and engineers have sort to address this area by constructing bespoke, artificial protein oligomeric systems (Oohora and Hayashi, 2014; Norn and Andre, 2016; Kobayashi and Arai, 2017; Ljubetic et al., 2017; Gwyther et al., 2019) ranging from simple dimers (Mou et al., 2015; Fallas et al., 2017) to higher order supramolecular structure (Thomson et al., 2014; Gonen et al., 2015; Butterfield et al., 2017). The main problem with these systems is that they lack functional synergy or even function beyond directing the assembly process. Thus, such systems tend to manifest the basic functional properties of their starting components. Recently, we have demonstrated functional synergy between designed artificial dimers (Worthy et al., 2019; Pope et al., 2020). Using computational approaches, we identified mutually compatible interfaces between various β-barrel fluorescent proteins and stabilised the interaction by genetically encoded click chemistry. Our approach allowed us to generate both symmetrical and non-symmetrical dimers together with homo and heterodimers that displayed either positive or negative functional synergy. While dimers represent the simplest protein oligomeric unit, they are the most frequently observed structural form in nature, with homo-dimers (comprising of the same polypeptide) dominating over hetero-dimer (composed of two different polypeptides) (Goodsell and Olson, 2000; Marianayagam et al., 2004; Mei et al., 2005).

The next challenge arguably involves the generation of heterodimers from structurally and functionally diverse proteins. Heterodimers are potentially rich in new functional features as they have the potential to combine drastically different and disparate functions leading to new emergent properties. Here we aim to take the next step in dimer construction and demonstrate that it is feasible to design and build intimately linked heterodimers comprised of structurally and functionally disparate proteins by linking a helical bundle protein to a β-barrel protein using a combination of computational protein design and bio-orthogonal chemistry. While bioorthogonal chemistry has been used previously to link different proteins together (Hatzakis et al., 2006; Eger et al., 2010; Hudak et al., 2012; Schoffelen et al., 2013; Torres-Kolbus et al., 2014; Kim et al., 2016; White and Bode, 2018), little effort is given to predicting compatible interfaces and residues pairs so the individual proteins generally remain functionally and structurally distinct. Furthermore, extended chemical linkers are routinely used as part of the chemical coupling process resulting in spatially separated protein units and preventing any interactions forming that normally comprise natural protein dimers (Hudak et al., 2012; Kim et al., 2012; Schoffelen et al., 2013; White and Bode, 2018). Thus, there is limited benefit above that of traditional genetic fusion approaches. Some rely on natural amino acid chemistry inherent to one partner protein to facilitate linkage (Hatzakis et al., 2006; Schoffelen et al., 2013; Torres-Kolbus et al., 2014), which can restrict coupling sites between individual units. To address these issues, we recently developed a computational design approach to facilitate the design of dimers linked by two genetically encoded compatible bio-orthogonal reaction handles (azide and alkyne) (Worthy et al., 2019; Pope et al., 2020). We focused on structurally similar proteins [fluorescent proteins with a common β-barrel architecture] whereby mutually compatible symmetrical dimer interfaces can facilitate construction. Such symmetry will not be available when structurally distinct proteins are used as proposed here and thus poses a greater challenge to predicting mutually compatible coupling sites and thus heterodimer construction.

To test our approach, we will use cytochrome b ₅₆₂ (cyt b ₅₆₂) as the helical bundle protein and the superfolder version of green fluorescent protein (sfGFP) as the β-barrel protein (Figure 1A). Cyt b ₅₆₂ is a small 4-helical bundle protein that binds haem tightly but non-covalently (Arnesano et al., 1999) while sfGFP is a directly evolved descendent of the original Aequorea victoria GFP (Pedelacq et al., 2006). We have previously linked the function of the closely related homolog of sfGFP, enhanced GFP (EGFP), with cyt b ₅₆₂ through a directed evolution domain insertion approach (Edwards et al., 2008; Arpino et al., 2012). The domain insertion approach differs significantly from heterodimerization as a single polypeptide unit contains both original proteins; the GFP primary structure is disturbed by insertion of the cyt b ₅₆₂ sequence within it at the genetic level. One variant demonstrated high energy transfer efficiency (close to 100%). Structural analysis revealed that the two original proteins are close in space within the single polypeptide unit but do not form many distinct inter-unit interactions as would be expected of a true oligomeric system. There was also little change in the inherent function of each domain compared to the starting parent protein suggesting limited synergy between the two.

Here, we used computational approaches to predict compatible interfaces between sfGFP and cyt b ₅₆₂. We then used 1-to-1 strain promoted alkyne-azide cycloaddition to covalently link the two proteins through a triazole link by genetically encoded incorporation of the azide group (via non-canonical amino acid azidophenylalanine) into cyt b ₅₆₂ and a strained cyclooctyne group (via the non-canonical amino acid strained-cyclooctyne lysine) in sfGFP. The successfully constructed heterodimers lead to enhanced sfGFP molar absorbance coefficients and were capable of energy transfer from sfGFP to cyt b ₅₆₂, with linkage site determining transfer efficiency.

While traditionally protein engineering has focused on converting oligomeric proteins into monomers, especially with regards to fluorescent proteins (Zacharias et al., 2002; Scott et al., 2018), the design and construction of artificial protein oligomers or “supramolecular” systems is currently of great interest (Norn and Andre, 2016; Yeates et al., 2016; Kobayashi and Arai, 2017; Beesley and Woolfson, 2019; Gwyther et al., 2019). Both GFP (Ding et al., 2015; Kim et al., 2015; Leibly et al., 2015; Worthy et al., 2019; Arpino and Polizzi, 2020) and cyt b ₅₆₂ (Baldwin et al., 2006; Radford et al., 2011; Brodin et al., 2012; Onoda et al., 2012; Song and Tezcan, 2014; Oohora et al., 2018; Golub et al., 2020) have been central to exemplifying the ability to generate new oligomeric supramolecular protein structures (Oohora and Hayashi, 2014). Artificial protein oligomers offer the same potential impact as oligomerisation does in nature: the generation of complex higher-order structures from a limited monomeric building block repertoire. This in turn allows new structural and thus functional space to be sampled not accessible in monomeric proteins. Whether it is O₂ binding to haemoglobin (Ciaccio et al., 2008) or enzyme catalysis and allosteric regulation (Goodsell and Olson, 2000; Ali and Imperiali, 2005), one of the main benefits of oligomerisation is synergy between individual subunits. Such linked functionality is generally hard to design and engineer into artificial complexes due to the requirement of long-range interactions that link active sites. Thus, most designed protein complexes focus on the interface region. We have shown previously that such networks can be generated between fluorescent protein homo and heterodimers using an approach like that used here, which leads to functional switching and fluorescence enhancement (Worthy et al., 2019).

Here we decided to test our ability to select compatible interfaces between disparate proteins by choosing two monomeric proteins with very different structures: a largely β-sheet protein (sfGFP) and a helical protein (cyt b ₅₆₂). The rationale for using SPAAC to covalently stabilise the heterodimer structure is that classical approaches such as disulphide bridges cannot discriminate between homo-dimers and heterodimers leading to a mixed population; the bioorthogonal nature of SPAAC means only heterodimer will form. Furthermore, the ncAAs used here have longer side chains than the short -CH₂-SH group of cysteine meaning steric clashes between monomers is less likely to inhibit covalent bond formation but still allow an intimate interaction between the individual monomer units. The triazole link is also more stable than a disulphide bridge. While our aim was not to generate a newly designed dimer interface mimicking natural protein-protein interactions, we did need to identify compatible interfaces that will at least persist for a length of time to allow covalent functionalisation. Without identifying compatible interfaces, the chances of stabilising the interface through SPAAC is minimal. Residue 50 in cyt b ₅₆₂ is largely surface exposed in a flexible extended loop and would normally be considered an ideal residue to target for covalent coupling with another protein (Supplementary Figure S1B) but SPAAC facilitated dimerisation was not possible via this residue (Supplementary Figure S1C). Previous work has shown that incorporation of azF at residue 50 is reactive (Zaki et al., 2018; Thomas et al., 2020). Lack of dimerisation is thus likely a result of the individual subunits unable to become spatially localised for long enough to promote SPAAC. This mirrors previous work with fluorescent proteins that suggested compatible interfaces are required for SPAAC facilitate dimerisation even when surface exposed residues are selected (Worthy et al., 2019)). Our computational design approach proved successful both here (Figure 1 and Supplementary Figure S4) and elsewhere (Worthy et al., 2019; Pope et al., 2020) in identifying compatible interfaces. It also provides a rationale for the inability of cyt b ₅₆₂ ^50azF to form dimers with sfGFP as residue 50 is not involved in any of the predicted protein interfaces. The step forward here was using proteins with disparate structural folds. The overall total binding energy was higher here (less negative) than for structurally homologues proteins tested previously suggesting a weaker interaction (Supplementary Figure S2 and Supplementary Table S2 versus data in references (Worthy et al., 2019; Pope et al., 2020)) but still proved useful in generating models for predicting successful coupling sites (Figure 1). The model suggests that there are intermolecular interactions between cyt b ₅₆₂ and sfGFP beyond residues involved in SPAAC, but it is not clear if they persist in the SPAAC link dimer. However, our previous structural work with SPAAC linked fluorescent proteins dimers shows that extensive intermolecular non-covalent interactions are formed at the interface, including long range interaction networks linking the active sites (Worthy et al., 2019; Pope et al., 2020). Thus, given the intimate nature of the coupling between sfGFP and cyt b ₅₆₂, additional non-covalent interactions outside the triazole linkage between the two monomers on forming the dimer are highly likely to be present.

Two heterodimers were designed based on the top ranked model (Figure 1C). In terms of sfGFP, residues 148 and 204 were selected to host the strained alkyne ncAA (SCO-K). These sites have been shown previously to be amenable to both small molecule (Reddington et al., 2012; Hartley et al., 2016) and protein (Worthy et al., 2019; Pope et al., 2020) attachment via SPAAC. H148 in sfGFP forms a H-bond with the chromophore, which directly impacts on the fluorescence properties of the protein by assisting in deprotonation of the chromophore’s phenol group. In many crystal structures, H148’s side chain is largely buried but is known to be dynamic (Seifert et al., 2003) with the “flipped out” conformation observed in some crystal structure populations, resulting in the residue becoming solvent exposed (Reddington et al., 2013; Arpino et al., 2014). Changing H148 to a larger ncAA results in breakage of a critical H-bond with the chromophore and exclusion from the core of the protein due to steric clashes (Hartley et al., 2016); the residue now amenable to chemical modification. The result is a mixed population of the protonated and anionic form of the sfGFP chromophore prior to SPAAC modification (Figure 2A). We have previously shown that we can control the relative populations of two chromophore states causing switching of the fluorescent properties in terms of excitation wavelength (Reddington et al., 2013; Hartley et al., 2016; Worthy et al., 2019). We have successfully achieved the same here through the formation of a heterodimer with promotion of the anionic form on dimerisation (Figure 2A). As mutation of H148 to a ncAA removes the H-bond to the sfGFP chromophore critical to formation of the phenolic anion, we have proposed previously that a structural water molecule replaces the imidazole group and plays the role of the H-bond acceptor that promotes ionisation of the chromophore (Worthy et al., 2019; Pope et al., 2020); the same scenario may also be the case here with a water molecule trapped at the sfGFP-b ₅₆₂ interface in a position to H-bond to the chromophore.

With respect to GFPb ^204-71, we see enhancement of the sfGFP molar absorbance (Figure 2B). Unlike H148, Q204 is surface exposed and plays little role in dictating the fluorescence properties of sfGFP, even when replaced by an ncAA (Reddington et al., 2012; Worthy et al., 2019). Water dynamics is again thought to play a major role. Dimerisation of sfGFP with itself or closely related fluorescent proteins enhances molar absorbance and structural analysis revealed ordered water molecules at the dimer interface, including waters comprising a channel through to the chromophore (Pope et al., 2020). The same may be happening with GFPb ^204-71 with the dimer interface trapping water molecules leading to a sustained water-protein bond network that improves the ability of the sfGFP chromophore to interact with light.

Bringing the sfGFP chromophore within close proximity to haem through formation of the heterodimer should result in energy transfer from sfGFP that is quenched by cyt b ₅₆₂, a feature that has been observed before for single polypeptide systems (Takeda et al., 2003; Arpino et al., 2012; Hanna et al., 2016). The extent of quenching is related to the distance between the two chromophores (Arpino et al., 2012). Quenching in a classical N- or C-terminal fusion of EGFP to cyt b ₅₆₂ was less than 65% (Arpino et al., 2012). Functional communication was present in both our heterodimers, but the efficiency of energy transfer differed depending on the SPAAC linkage positions, haem redox state and, to an extent, the excitation wavelength (Figure 3). This is despite the two residues being adjacent to each other in the structure. Linkage via sfGFP residue 148 had the highest energy transfer efficiency (>85%) with more apparent efficient energy transfer on excitation 395 nm. Linkage via sfGFP residue 204 resulted in a lower energy transfer (63%), comparable to previous “head-to-tail” fusions with cyt b ₅₆₂ where the two chromophores are not anticipated to be close in space (Arpino et al., 2012). Thus, the question arises as to why different linkage sites that are close together in sfGFP (Figure 1C) generate very different energy transfer efficiencies? Given that GFPb ^204-71 can be separated from its monomeric components by size exclusion chromatography (Supplementary Figure S6), the residual fluorescence is unlikely to be from contaminating sfGFP^204SCO monomer. We have shown previously that linkage via residue 204 in fluorescent protein dimers resulted in lower than predicted energy transfer (Pope et al., 2020). The exact reason was not clear. The calculated R ₀ (the Förster radius at which energy transfer is 50% efficient) between EGFP and cyt b ₅₆₂ is 46 Å (Takeda et al., 2001). As energy transfer is related to the donor-acceptor chromophore distance (r) and energy transfer (E) through equation E = 1/[1+(r/R ₀)⁶], we can estimate the distance between the chromophores. Previously, we showed that there was a good correlation between r from energy transfer (99%) and structure in our domain insert EGFP-cyt b ₅₆₂ construct (Figure 5). Based on energy transfer efficiency, in GFPb ^148-71 the closest interchromophore distance is predicted to be ∼27 Å while for GFPb ^204-71 it is ∼41 Å. The interchromophore distances derived from the original top ranked model suggest that it might provide a realistic representation for GFPb ^148-71 but not GFPb ^204-71 (Figure 5). As shown in Figure 5, the domain arrangement in our heterodimer model compared to the determined structure of the high energy transfer efficient domain insert protein is different; the interchromophore distance is 8 Å longer thus could account for the slightly reduced energy transfer efficiency in GFPb ^148-71. Residue 204 is close to the interface rather than directly forming the interface, which may result in the two monomers adjusting their relative placement on SPAAC compared to that of the model. There are two isomeric forms for the triazole crosslink between AzF and SCO-K (Worthy et al., 2019; Pope et al., 2020): the syn isomer that forms a turn structure and the anti form that results in an elongated linkage (Figure 5B). The anti form may dominate for GFPb ^204-71 resulting in the two monomers and thus chromophores being separated by a longer distance. Interestingly, energy transfer efficiency was oxidation state-dependent (Figures 4C,D). The ability apo-GFPb ^204-71 to bind haem suggests that the haem binding site is still accessible as predicted by our original modelling and on binding haem under non-reducing conditions exhibits a similar drop in fluorescence to that observed for holo-GFPb ^204-71 suggesting full haem occupancy (Figures 3C, 4D). On addition of reducing agent, fluorescence decreased by another third suggesting conversion from Fe³⁺ to Fe²⁺ resulted in increased energy transfer efficiency. DTT is not known to reduce the fluorescence emission spectra of sfGFP (Supplementary Figure S7 and (Reddington et al., 2015)) so the increased energy transfer efficiency may due to inherent change in the protein-bound haem such as increased molar absorbance of the α/β peaks that overlap with the sfGFP emission (Figure 4B). Even under reducing conditions, energy transfer efficiency is still ∼75% equating to an interchromophore distance of 38 Å. Thus, we cannot rule out other currently unknown events contributing to the reduced energy transfer efficiency in GFPb ^204-71, as also observed in fluorescent protein dimers linked via residue 204 (Pope et al., 2020).

To conclude, here we have shown that it is feasible to design and construct artificial heterodimers between structurally and functionally disparate proteins linked by a genetically encoded bio-orthogonal link. The in silico docking helped identify suitability compatible protein-protein interfaces that where then stabilised by a triazole link formed by SPAAC. In both heterodimer configurations tested, the ability of phenol anion chromophore form of the sfGFP to interact with light was enhanced, with energy transfer to the haem centre of cyt b ₅₆₂ demonstrating functional linkage.