Functional modulation and directed assembly of an enzyme through designed non-natural post-translation modification

Designed phenyl azide incorporation combined with bioorthogonal Click chemistry to regulate enzyme activity, or promote its stable assembly on graphene.


Introduction
Post-translational modication (PTM) is central to biology by expanding and modulating the function of a large number of proteins. 1 Many of these events essentially permanently covalently modify a protein, ranging from attachment of small moieties (e.g. methylation, 2 cofactors 3 ) to larger events such as proteolysis and glycosylation. Each of these factors can impact signicantly on protein structure and function thus inuencing and even enabling inherent protein activity. The presence of sequence and/or structural motifs in combination with of a wide variety of subsidiary machinery, mostly enzymes, is required to achieve exquisite specicity both in terms of the target protein and the spatial position in the modied target. The complexity of these systems can be a signicant hindrance with respect to their transfer to new proteins where such PTMs are not inherent. Furthermore, PTM is largely restricted to a select chemical set preexisting in nature. Precise modication with non-natural adducts may be a more appropriate and useful means to expand and modulate protein function, 4,5 including for use in non-biological contexts.
Covalent modication with non-natural adducts is traditionally achieved using chemistry inherent to the natural amino acid repertoire, mostly amine, carboxyl and thiol groups. The main problem is lack of specicity as such chemistry is normally distributed across the surface of a protein and is ubiquitous in the proteome. A more general and powerful approach is the introduction of new chemically reactive handles not present in the native 20 amino acid set through the use of an expanded genetic code 6 (see ref. 7-10 for recent reviews). The unique reactivity of a chosen non-natural amino acid (nAA) together with the ability to select both the target protein and the residue within the target means that specicity comparable to natural PTM events can be achieved. 10,11 The use of highly specialized caged nAAs has been very effective in controlling activity [12][13][14][15] but is restricted to certain protein types and chemistry, and can add signicant bulk to the amino acid side chain during the production and folding of the nascent polypeptide and its subsequent folded form.
The incorporation of phenyl azide chemistry into proteins through the use of p-azidophenylalanine (azF) 17 is an attractive alternative. As well as being only one atom bigger than the natural amino acid tyrosine and having been successfully incorporated into a wide variety of proteins, 7-9 the phenyl azide moiety opens up different routes to non-natural PTM (nnPTM): photochemical transformations and Click chemistry adduct addition. 9 The use of phenyl azide photochemistry to control protein activity has recently been demonstrated. [18][19][20][21][22] Azidealkyne cycloaddition is fast becoming a useful approach for orthogonal biomolecule conjugation but has largely been used in a passive way, for example, to label proteins. 7,[23][24][25] Given the versatility in terms of the array of adducts available coupled with the inherent bioorthogonality and biocompatibility, 26 it is surprising Click chemistry has not been used more extensively as a general direct modulator of protein activity. Additionally, useful adducts can be placed at strategic positions to expand and facilitate protein function in a manner akin to co-factors. This includes attachment of entities to facilitate interfacing and assembly with secondary systems or materials. [27][28][29][30][31] As in many natural biomolecular assemblies, a dened and optimal protein-material interface is critical for maximal communication between the individual elements. Interfacing proteins with carbon sp 2 materials such as graphene is gaining signicant interest as it forms the basis for constructing hybrid bio-transistors in which events at even the single protein molecule level can be used to gate conductance through graphene. 27,32 Here, we show that the activity and assembly onto pristine graphene of the antibiotic resistance protein TEM b-lactamase 33,34 can be directly controlled by Click chemistry. By genetically encoding phenyl azide chemistry at designed positions in TEM and using Cu-free biocompatible strain promoted azide-alkyne cycloaddition (SPAAC; Fig. 1) 35,36 different adducts can be attached (1, 2 and 3; Fig. 1) that either reduce or restore activity on modi-cation, and dene assembly of TEM on graphene.

Results and discussion
The three main adducts with a core dibenzylcyclooctyne (DBCO) reactive handle (Fig. 1) chosen have very distinct properties: 1 is an amine derivative of DBCO that has hydrogen donor and acceptor groups opening up the potential to form H-bonds between residues not normally close enough to each other in the protein structure; 2 is a large, planar and hydrophobic rhodamine dye (Texas Red) that has proved useful in labelling proteins for uorescent imaging 23 but could also act as an effective bulk spatial and steric blocking element if required; 3 is a pyrene derivative that will aid protein interfacing with extended carbon sp 2 materials (e.g. graphene) through p-p stacking. 29,37 Both 1 and 2 are "off-the-shelf" products while 3 can be generated by a simple succinimidyl ester reaction. Thus, all 3 adducts can easily be accessed by the wider bioscience community without any synthetic chemistry knowledge. The azF dependent production of active enzyme is shown in ESI Fig. 1. †

Modulating enzyme activity
The in silico design process was performed using a combination of ROSETTA 38 and molecular dynamics 39 as outlined in the Supporting Methods. † Y105, a partially surface exposed residue ($110Å 2 z 50% relative surface exposure), was chosen as a site for negative modulation through nnPTM due to its location close to the catalytically important SDN loop and its role in forming a partial lid over substrate binding cle ( Fig. 1; ESI Fig. 2 †). Y105 is thought to be especially important in dening the size of the substrate binding cle (therefore substrate specicity) 40 and local dynamics of residues critical for activity. 41 In silico modelling predicted exchanging the hydroxyl group for an azide should not have a major effect on enzyme structure around the locality of residue 105 (Fig. 2a). In fact ampicillin hydrolysis was slightly enhanced (higher k cat ) for TEM Y105azF but had a slightly lower affinity (higher K M ) making overall catalytic efficiency similar to wt TEM ( Fig. 3 and ESI Table 1 †). TEM Y105azF was receptive to modication by SPAAC as indicated by the estimated labelling efficiency of $80% with 2.
As predicted, TEM Y105azF was largely inhibited on modication with either 1 or 2 ( Fig. 3) despite neither having any signicant inherent inhibitory effect on wild-type TEM (ESI Table 1 †). Both catalysis and substrate binding were disrupted as evident by the decrease in k cat and increase in K M (ESI Table  1 †) for ampicillin resulting in an overall decrease in catalytic efficiency of $70% and 85% when modied with 1 or 2 respectively compared to unmodied protein. This is similar to the most deactivating site-directed mutants of Y105. 40 Fig. 1 Modification of TEM b-lactamase. The structure of TEM blactamase (top left) with the inhibitor imipenem (yellow sticks to emphasise the substrate binding pocket) highlights the residues targeted for replacement with azF (red spheres) together with the key catalytic residues (cyan sticks) and the U-loop (green). The protein structures were generated using PyMol. 16 A representation of the SPAAC reaction using the activated alkyne present in the core DBCO moiety is shown (top right). The R groups attached to the DBCO used in this study are shown (bottom half).
Modelling of the covalent complex (vide infra) between TEM Y105azF and 1 suggested that modication blocks access at one end of the substrate binding site, with the amine of 1 forming a hydrogen bond with the backbone carbonyl group of G238 (ESI Fig. 3 †). The enzyme kinetics suggests that simple steric blocking of substrate binding is unlikely to be the sole mechanism of action as catalysis and substrate binding are equally affected, similar to that observed for classical mixed inhibition models. The relatively small difference in catalytic efficiency of TEM Y105azF modied with either 1 or 2 indicates that the most signicant effect is occurring close to the linkage site, as suggested by the covalent complex model (ESI Fig. 3 †). However, the bulkier group of 2 is exerting a slightly greater effect with regards to k cat rather than K M (Fig. 3 and ESI Table  1 †) suggesting that long range interactions made by the rhodamine dye moiety may be forcing the structure around residue 105 to adapt less catalytically procient form. Currently, the separation of modied from unmodied protein has proved unsuccessful, so the observed activity may be contributed by unmodied protein only (estimated to be $20%; vide supra). Therefore, the inhibition levels observed here represent the lowest currently achievable and we cannot rule out modied species are essentially inactive.
P174 is another largely surface exposed residue ($80Å 2 z 60% relative surface accessibility) that is relatively distant from the active site (Fig. 1). P174 however contributes towards one of the turns that comprise the U loop ( Fig. 1 and ESI Fig. 4 †), a region critical for substrate binding, substrate specicity and catalysis. 42,43 It was hypothesized in the design process that mutating P174 to a non-cyclic amino acid would result in local conformational changes around the turn region. Molecular modelling supported this hypothesis with the structure of the turn slightly shied compared to wt TEM ( Fig. 2b and ESI  Fig. 5 †). The side chain of azF174 lies approximately perpendicular to the native proline, with the azido moiety tting into a shallow pocket and making a local polar interaction network with R42 and T265 (Fig. 2c). These new interactions may act as the driver of local conformation changes associated with the P174azF mutation, which in turn alters the local bonding network that subtly shis the conformation of both the active site and substrate binding residues, including S70 (ESI Fig. 5 †). Based on the TEM P174azF model it was further postulated that on modication with 1, the formation of the triazole link would break the interaction network with R42 and T265 so reconstituting the original loop structure and activating the protein.
The amine group also has the potential to form new long range hydrogen bonds in spatially local regions ( Fig. 3 and ESI Fig. 4 †).
As suggested by the model, replacement of P174 with azF results in a signicant change in activity. The overall catalytic efficiency of TEM P174azF was 60% lower compared to wt TEM (Fig. 3). The major effect was on k cat which was 3 fold lower compared to wt TEM (ESI Table 1 †), supporting the evidence from the model that the likely effect of azF incorporation at residue 174 is to disrupt interactions of residues associated with catalysis. This was offset by a slightly increased ampicillin affinity.
TEM P174azF was accessible to Click modication, but efficiency was lower (estimated $33% using the absorbance properties of 2) compared to TEM Y105azF . It is unknown how local protein microenvironment dictates SPAAC efficiency but dynamics, relative exposure to aqueous solvent and the character of shallow "pockets" have been suggested as possible determinants. 23 The molecular model suggests that interaction of the azide moiety with other residues, which may play a role  through, for example, altering the relative populations of resonance structures sampled by the azide moiety (R-N]N + ]N À / R-N À -N +^N ), and/or its accessibility and available orientations to the incoming DBCO. The effect of the SPAAC nnPTM on the ability of TEM P174azF to hydrolyse ampicillin varied depending on the DBCO adduct. Modication with 2 exerted a similar but somewhat smaller inhibitory effect ($40% drop) as observed for TEM T105azF (Fig. 3 and ESI Table 1 †). Modication with 1 resulted in a signicant increase in overall activity restoring apparent catalytic efficiency essentially to wild-type levels (Fig. 3). This should be considered as a lower estimate of activation as $70% of the protein may be unmodied (vide supra). With both adducts, the most signicant contribution was the change in k cat of TEM P174azF compared to unmodied protein (ESI Table 1 †); modication with 1 increased k cat by 210% while addition of 2 reduced k cat by almost half.
To understand how 1 exerts its benecial effect on TEM P174azF , the nnPTM product was modelled. Using the TEM P174azF model as a starting point, the triazole linkage between azF and DBCO of 1 was generated in silico by parameterizing the azF-DBCO complex to calculate optimized geometries and the electrostatic potentials. The model was subjected to molecular dynamics for a total of 5 ns. The model of nnPTM product, termed TEM P174azF +1 suggested that the general structure was closer to wt TEM than the original TEM P174azF model ( Fig. 3c and ESI Fig. 5 †). The Uloop structure around residue 174 and the local bonding networks for TEM P174azF +1 were largely comparable with wt TEM. The orientation of residue 174 differs signicantly to accommodate 1 (ESI Fig. 5 †). The amine group of 1 bends back and appears to come within hydrogen bonding distance ($2.6Å) of the sidechain carbonyl group of the adjacent residue N175 (Fig. 3c). The loss of the azide group on formation of the trizole removes the interactions with R43 and T238. However, the reason for 1 acting as an activator and 2 as an inhibitor may be down to the reduced bulk and ability of 1 to form local H-bonds.

Protein-graphene interfacing
In some instances, it would be attractive to select a residue intimately associated with active site regions so that secondary non-related events can be coupled to activity without signicant effect on function. Linking proteins to carbon sp 2 materials such as graphene is especially attractive; local charge and electrostatic changes in the protein can be used to "gate" the electronic properties of the sp 2 material 27,44 so generating sensing systems with ultimate single molecule resolution. The pyrene moiety of 3 (Fig. 1) is especially attractive as an interfacing agent as it allows dened coupling of a protein to the sp 2 material through p stacking. 37 This has been demonstrated previously using optimally placed cysteine residues as the reactive handle for attachment to pyrene-coated single walled carbon nanotubes. 27 However, this required removal by mutagenesis of native cysteine residues. Here, we demonstrate that dened enzyme interfacing on a clean graphene face is feasible though the use of designed azF placement.
W165 is another partially surface exposed residue ($100Å 2 z 50% relative surface accessibility) that fulls the required criteria (Fig. 1). It resides in the U-loop and splits two functionally important residues; E166 is a conserved catalytic residue involved in proton shuttling 34 and R164 forms buried ionic interactions critical for stabilizing the U loop and thus the positioning of E166. 45 The approach used to model TEM P174azF modied with 1 was applied to the W165azF mutation to ascertain the potential position of the adduct group and its effect on structure (Fig. 4a). Modelling suggested that the overall structure of the protein would be largely unperturbed and the adduct would point outwards into the solvent. Mutating W165 to azF did slightly reduce overall catalytic efficiency towards ampicillin (1.5 fold) but crucially, modication with 3 was feasible and had little effect on overall activity (ESI Fig. 6 †).
By coupling 3 to TEM W165azF prior to surface assembly we can directly interface the protein with graphene in a dened manner without surface coating with pyrene, as has been previously used. 27,46 Intermittent-contact mode AFM (tapping mode) imaging indicated that TEM W165azF modied with pyrene binds stably to graphene surfaces ( Fig. 4b and c; ESI Fig. 7 & Movie 1 †). The protein molecules are not signicantly disturbed by multiple scans. The average apparent heights were 3 nm, which is close to the predicted height of the protein bound in the designed orientation ($3 nm; ESI Fig. 7 †). The average apparent lateral dimension was larger than predicted (10 nm versus 5 nm), as a result of tip convolution effects. In comparison, wt TEM did not stably bind to the surface, with tip contamination commonly observed aer multiple scans (ESI Fig. 8 †).

Conclusions
SPAAC is a powerful approach for precise and designed protein PTM that supplement protein activity using a non-intrusive, non-caged reaction handle. We have demonstrated here that SPAAC can be used beyond simple passive labelling of proteins and can actively modulate activity through different mechanisms without the need to synthesize specialized molecules. It also facilitates the interfacing of proteins in a designed and dened manner with useful active materials such as graphene. This negates the need to remove inherent chemistry through mutagenesis allowing the focus to be on placing the reactive handle for optimal coupling between the protein and material. In silico modelling with both nAA and adducts can greatly aid the design process by which useful variants are generated and their affects rationalised, which will in turn lead to more accurate designs.

Experimental
In silico protein modelling A detailed description of the in silico modelling of TEM containing the azF mutation and modication with the DBCOamine adduct (1) are provided in the ESI. † Briey, geometry optimised structure les, force eld parameters for azF and azF-DBCO amine adduct complexes were generated. The published crystal structure of TEM b-lactamase to 1.8Å resolution (PDB code 1BTL 45 ) was used as a starting point for Monte Carlo simulations within the ROSETTA soware package. 38 The lowest energy model was used as the starting point for molecular dynamics using GROMACS. 39 The modelling was run on the Raven cluster as part of the Advanced Research Computing @ Cardiff facility.

Enzyme activity assays
The wt and mutant TEM variants were generated as outlined in the ESI. † The proteins were expressed in E. coli using a bespoke plasmid based on pBAD called pBADKAN, and puried as described in the ESI. † The TEM-dependent kinetics of ampicillin hydrolysis (3 235 ¼ 1500 M À1 cm À1 ) were determined spectrophotometrically using 1 cm path length QS quartz cuvette (Hellma). Hydrolysis assays were carried out in a 1 mL reaction volume. Puried enzyme was diluted to a nal concentration of 250 ng mL À1 in 50 mM sodium phosphate buffer, pH 8 at room temperature. Reactions were started by addition of ampicillin and hydrolysis was measured by the decrease in absorbance at 235 nm. Ampicillin concentrations ranged from 50 mM to 800 mM. Kinetic parameters were calculated using initial rate of hydrolysis at each substrate concentration and then tting to the Michaelis-Menten equation using GraphPad Prism. TEM b-lactamase activity using the colorimetric substrate nitrocen was performed essentially as described previously. 47,48 Click modication SPAAC reactions were performed on pure protein using a vefold molar excess of dibenzylcyclooctyne (DBCO) reagent to protein. DBCO-amine (1) and DBCO-Fluor 585 (2) were obtained from Click chemistry tools, while DBCO-pyrene was synthesized from DBCO-amine and 1-pyrenebutanoic acid, succinimidyl ester via nucleophilic substitution. Reactions were le overnight at room temperature in PBS. SPAAC reactions with 2 were analysed by SDS-PAGE and subsequent imaging of the uorescent dye on a transilluminator. Labelling efficiencies were calculated as described in the ESI. †

Protein deposition on graphene and AFM imaging
Protein samples were deposited onto graphene on a copper foil substrate. Monolayer graphene was deposited on the copper substrate by chemical vapor deposition (CVD) using methane as carbon source. A detailed description of the growth procedure is provided in the ESI. † The foil pieces ($4 mm 2 ) were immersed in PBS containing 1 nM protein and incubated at room temperature for 10 min. Following incubation, the foil pieces were rinsed with high purity deionised water and dried with N 2 gas. AFM imaging was performed in using a Veeco Nanoscope IIa (Bruker) in tapping mode. The same area was scanned up to 10 times with a total of 6 different regions imaged to check the stability of the protein molecules on the surface of graphene. A video of the surface scans is provided as ESI. † Although the images show some dri, it is clear that the proteins are stably bound to the surface.