Exploring conformational preferences of proteins: ionic liquid effects on the energy landscape of avidin

In this work we experimentally investigate solvent and temperature induced conformational transitions of proteins and examine the role of ion–protein interactions in determining the conformational preferences of avidin, a homotetrameric glycoprotein, in choline-based ionic liquid (IL) solutions. Avidin was modified by surface cationisation and the addition of anionic surfactants, and the structural, thermal, and conformational stabilities of native and modified avidin were examined using dynamic light scattering, differential scanning calorimetry, and thermogravimetric analysis experiments. The protein-surfactant nanoconjugates showed higher thermostability behaviour compared to unmodified avidin, demonstrating distinct conformational ensembles. Small-angle X-ray scattering data showed that with increasing IL concentration, avidin became more compact, interpreted in the context of molecular confinement. To experimentally determine the detailed effects of IL on the energy landscape of avidin, differential scanning fluorimetry and variable temperature circular dichroism spectroscopy were performed. We show that different IL solutions can influence avidin conformation and thermal stability, and we provide insight into the effects of ILs on the folding pathways and thermodynamics of proteins. To further study the effects of ILs on avidin binding and correlate thermostability with conformational heterogeneity, we conducted a binding study. We found the ILs examined inhibited ligand binding in native avidin while enhancing binding in the modified protein, indicating ILs can influence the conformational stability of the distinct proteins differently. Significantly, this work presents a systematic strategy to explore protein conformational space and experimentally detect and characterise ‘invisible’ rare conformations using ILs.


Introduction
Proteins are inherently dynamic and can transiently sample an ensemble of conformational states. 1 The heterogeneity of the conformational ensemble (for example, compact globular states, localised secondary structure, and coil-like states) reects the broad distribution of conformers, the different shapes and stabilities, a protein can adopt in its immediate environment. 1,2 The energy landscape of a protein encodes the relative stabilities of its different conformational states and the energy barriers that separate the conformations. 2 The populations of the conformations follow statistical thermodynamic distributions, and the heights of the energy barriers separating the conformations dene the timescale of conformational transitions. [3][4][5] We can visualise that for a ligand-binding protein in solution, if the free energy barriers are low relative to the Boltzmann energy (k B T), thermal uctuations can lead to signicant populations of multiple conformers. 2,6 A ligand can then interact with the lowest energy conformer, or with one of a number of higher energy conformers that are populated in solution. Upon a population shi, the relative populations of conformers that already pre-exist in solution redistribute. 1 However, experimental techniques are limited when determining conformational diversity, particularly since structural characterisation of conformations, other than the most prevalent one, in solution is intensely challenging. 4,[7][8][9] Despite this, studying protein conformational diversity, key for understanding protein function, continues to be extensively investigated using a wide range of theoretical models from protein physics to Darwinian selection. 2,[10][11][12] The goal to determine which selective pressures can induce a stable protein conformation is further complicated by the large conformational space a protein can sample, preferential interactions available, and environmental conditions. 13 This begs the question: Can we tune environmental factors to select for more stable protein conformations?
Avidin, a homotetrameric glycoprotein naturally occurring in egg whites, is an ideal model system for studying protein susceptibility to stress-induced conformational changes, due to its high structural and thermal stability to extremely harsh environmental conditions. [14][15][16][17][18][19][20] The overall fold of each monomer in avidin is a classical b-barrel formed by eight antiparallel b-strands. 17 The ligand binding site is a pocket, located at the ends of each b-barrel which contains both polar and aromatic residues. 18,19 Avidin is composed of 72 basic residues and 48 acidic residues per molecule, resulting in an isoelectric point of 10.5, below which avidin has a net positive charge. 20 At a pH below 4.5, the net positive charge on avidin increases as the glutamate, and then the aspartate, residues become protonated. 16 Above a pH of 10.5, the lysine residues are deprotonated, resulting in a net negative charge on avidin. 16,20 When biotin is bound to avidin the interaction promotes remarkably high stability to the avidin-biotin complex (K d $ 10 À15 ), 14,15 and the high pI and carbohydrate content of avidin allows it to bind non-specically to ligands and molecules other than biotin. Additionally, the ability to chemically modify the surface of avidin has been exploited for a wide range of biochemical, biophysical, and medical applications, as well as in computational studies examining the conformational changes of avidin induced by ligand binding. 14 It can be envisioned, that the native unbound state of avidin exists as a dynamic ensemble of rapidly interconverting conformations, which can be described by a relatively rugged energy landscape with many local energy minima and energy barriers. 4,5 A biochemical modication of avidin would change the relative free energies of individual conformations as well as the energy differences between conformations. Thus, we chose to use avidin to investigate the conformational transitions, stability against thermal denaturation, and conformational preferences of proteins under specic environmental conditions.
In recent years, room-temperature ionic liquids (ILs) have been explored as potential solvents, providing unique environments for proteins, enhancing solubility, stability, and catalytic activity of proteins and enzymes. 21 Given the wide range of ions available, by choosing specic ion combinations, ILs can be tailored to give unique properties (for example, immiscibility, low nucleophilicity, acidity, and solvation strength), and alter protein conformations in a tuneable sense. 21,22 For example, protic ILs have been shown to stabilise the secondary structure of proteins above denaturing temperatures, while aprotic ILs can have destabilising effects. 22,23 Additionally, biocompatible ILs, for example choline-based ILs, can improve the long-term stability and solubility of proteins, although electrostatic binding of nucleophilic anions with protein surfaces can disrupt internal packing interactions, crucial for the structural stability of proteins. 1,22 Since ionic interactions play a major role in understanding the physicochemical and biological phenomena involved in protein folding and unfolding processes, a great deal of work has focused on studying the versatile properties of ILs, and the interactions responsible for stabilising and destabilising proteins. 21,22,24 Yet, to date, no experimental study has investigated the role of ion-protein interactions in determining the conformational preferences of proteins in ILs.
Inspired by the high tunability of ILs, for the rst time, we aimed to determine whether we can use ILs to stabilise and experimentally nd specic, rarely populated protein conformations from a heterogeneous conformational ensemble, where environmental factors result in the interplay of different conformations. This would provide us with a unique opportunity to experimentally determine the conformational energy landscape of proteins, and nd 'invisible' conformers, accessible by conformational transitions from the native state. We chose seven biocompatible IL and buffer solutions (of a comparable pH range to the IL solutions), to investigate and uncover the conformational space of the robust and reliable avidin system and determine the favoured conformations under particular environmental conditions. Choline amino acid and fatty acid ILs were chosen as these are widely used and well characterised in the literature, with toxicities much lower than traditional ILs, such as imidazolium-based ILs. 21,22 Additionally, amino acids are particularly attractive as potential anions, due to their inherent biocompatibility, and since the presence of additional hydroxyl, carboxylic acid and amide groups on the side chain have been shown to improve the thermal stability of proteins. 22 We also note that to our knowledge, fatty-acid-based ILs have yet to be examined for improving the conformational stability of proteins, and thus are of particular interest to us. We took advantage of our ability to uniquely modify avidin and study the effects of ILs on the conformations, stability, thermostability, and binding behaviour of unmodied and modied avidin using a multi-technique approach. First, the conformational and thermal stability between unmodied and modied avidin were characterised and differences in the conformational states were determined using dynamic light scattering (DLS), differential scanning calorimetry (DSC), and thermogravimetric analysis (TGA) experiments. Then, we examined the effects of different ILs at varying concentrations on the structure and thermal stability of the proteins using small angle X-ray scattering (SAXS), differential scanning uorimetry (DSF) experiments, thermal shi assays (TSA), thermodynamic data analysis, and variable temperature circular dichroism (CD) spectroscopy experiments. Finally, in order to further study the effects of ILs on avidin binding and correlate thermostability with conformational heterogeneity and preferences we conducted a binding study. This novel systematic approach enables us to probe the conformational landscape of the distinct protein systems and build a comprehensive picture of the heterogeneity of avidin in the presence of different ILs.

Results and discussion
In order to characterize the conformational ensembles of unmodied and modied avidin and assess the thermal behaviour of the different avidin conformers, the hydrodynamic radius (R h ), decomposition temperature and any reversible phase transitions were measured using DLS, TGA and DSC, respectively. The hydrodynamic radius (R h ), the effective radius of an ion, including the bound solvent molecules on its surface, was measured by DLS for aqueous native avidin as 26.4 AE 0.3 A, in reasonable agreement with the literature (Table 1). 26 Upon cationisation, a slight increase in R h (27.9 AE 0.2 A) was observed, which then increased signicantly once the surfactants were added, with aqueous modied avidin R h measuring 49.5 AE 0.4 A. Freeze-dried modied and unmodied avidin were also characterised by DSC and TGA (see ESI Fig. S1 †). A melting point for modied avidin was observed at 23.7 C, in reasonable agreement with the literature for this modication procedure on other proteins, 27 while for unmodied avidin no phase transitions were detected in the range of À80 to 100 C (Table 1 and ESI Fig. S1 †). By TGA, the decomposition temperature observed for modied avidin (381 C) was much higher than that of unmodied avidin (263 C). This provides evidence that there is a difference in the conformational ensemble of unmodied and modied avidin, respectively, shown by the higher degradation temperature and the presence of a melting transition observed for the conformational states of modied avidin.
Previously, such modications have been shown to cause changes in the relative prevalence of b-sheet, a-helix and unordered secondary structures, 28 specically an increase in a- helix and decrease in b-sheet content. We interpret the inuence of modication as a shi from b-strand to a-helical conformers of avidin, attributed to the binding of surfactant to avidin, which could alter the conformational exibility of the exposed loops connecting the b-strands of avidin, and thus the accessibility of avidin to aqueous solvent. 29,30 Given that disordered proteins have been shown to be more heat stable and soluble compared to their folded counterparts, 2 it is not surprising that modied avidin exhibits a raised decomposition temperature. Thus, it can be suggested that the process of modication altered the conformational ensemble of avidin to favour a thermally stable, rarely populated conformation.
Having established that our experiments are starting from different conformational ensembles, we further examined the structure of the unmodied and modied avidin conformers and the effects of IL composition and concentration on changes in the compaction of avidin. By performing SAXS experiments (see ESI Fig. S2 †) we found that unmodied avidin in IL and in aqueous buffers are comparable in secondary structural compactness, as indicated by the radius of gyration (R g ) values ( Table 2). Most notably, at higher IL concentrations, R g was reduced and the degree of compaction of avidin increased. It can be suggested that with an increase in IL concentration, the coordination of ions to the a-helices and b-sheets increased proportionally, reducing the number of exposed charged surfaces, and compacting local protein folds (Fig. 1). 29 [Met] environments, the R g value of unmodied avidin decreased (albeit higher initial R g values) with increasing concentration. This consistent trend could suggest a conformational shi from conformer of higher conformational exibility to compact conformers, and preferences that restrict the conformational search by chain compaction. 4,32 For the modied avidin, two distinct structures were observed in the SAXS data with different radii. By control experiments with surfactant-only solutions (no avidin present, see ESI Fig. S3 †) and comparison to the DLS data for modied avidin, it was established that the larger structures (R g $ 45-50 A) were modied avidin, while the smaller structures (R g $ 15-19 A) were free surfactant micelles, formed by excess surfactant dissociating from the modied structure under aqueous conditions. 33 Since the hydrophilic head of the surfactant molecules can interact with the ions in solution and form micelles, the effect of molecular connement and proportion of ions available that could compact the avidin structure likely changed. 34 Table 2 shows that for the modied avidin at 10 wt% IL, almost all samples, except in [Cho][Ger], exhibited reduced R g and increased compactness, similar to the compaction effect observed for unmodied avidin. Specically, preferential hydrophobic interaction between the long hydrocarbon tail of the geranate anions and the hydrocarbon component of the surfactant molecules, could result in reduced interactions between the IL and modied avidin, partly explaining the lower observed compaction in [Cho] [Ger]. At 50 wt% IL, features representing the modied protein structures were poorly dened and could not be tted (see ESI Fig. S4 †), while the micelle peaks became more dened. The likely conclusion is that the ions of the ILs were in competition with the surfactant molecules, potentially resulting in ion exchange within the nanoconjugates. This disrupted the nanoconjugates sufficiently that they were no longer detected by SAXS and limited the effective protective effects of the ions on the hydrophobic surfaces of the protein. While the conformational sampling of modied avidin was affected, 6,34 it is clear that the avidin could be driven toward the compact conformations by ILs. Signicantly, in order to understand the conformational preference of avidin, it is essential to obtain information about the structure and stability behaviour of the different conformers in highly-dened environments.
Next, we aimed to further study the conformational heterogeneity of avidin and connect the observed macromolecular connement with thermal stability behaviour of the different conformers generated. Control differential scanning uorimetry (DSF) experiments revealed that avidin was most stable at pH 7.1 (0.05 M citrate/phosphate buffer), with T m of 88.7 C. At pH 5.1 (0.05 M citrate/phosphate buffer), the T m was slightly reduced (86.4 C), alluding to the contribution of pH, linked to net surface charge and the stability of avidin in solution. 35 Similarly, at pH 9.1 (0.05 M tris buffer), the T m was further reduced (82.9 C). Accounting for differences in conditions and method used, measured T m values were in reasonable agreement with the literature. 36 The behaviour of avidin in alkaline pH (in the range pH 7.1-9.1), is in agreement with the effects of molecular connement of avidin by the solvent molecules, via solvent-protein interactions, leading to conformational redistribution and a population shi toward a thermally less stable avidin conformer. 4,29,32 T m values were then measured in aqueous IL solutions at 10 wt%. We found that [Cho][Asp] increased avidin thermostability by the greatest amount, with T m exceeding that of the pH 7.1 control by 4.5 C. In 10 wt% [Cho]Cl solution, with pH 4.66, the T m was 87.6 C, above that of the pH 5.1 control. Most notably, [Cho][Met] and [Cho][Phe], both with pH > 10.7, gave T m of 85.8 C and 86.6 C, respectively. These values were relatively higher than the pH 9.1 control (82.9 C), emphasising that the presence of IL can mediate, and in some cases enhance the thermostability of avidin under specic conditions.
To further examine the idea that molecular connement by the IL inuences the thermal stability and conformations of avidin, 'buffered' samples of the IL solutions ( and the T m values increased by +2.3 C and +0.9 C, respectively. Unsurprisingly, the increase in T m (+2.8 C) was larger in [Cho]Cl, where the pH change was much greater, increasing from 4.7 to 6.9. The observation that adding buffer can increase the T m even when the pH change is small, as exemplied in the case of [Cho][Met], indicated more favourable conditions for thermostability in this complex mixture. It can be suggested that in buffered solutions, compared to neat IL, a different number of solvent molecules can associate on the avidin surface, and form larger patches, increasing the conformational stability of avidin and the T m , as we observe in our work. 37,38 We illustrate the barrier to denaturation for avidin in the different ILs with plots of the fraction of the protein denatured against temperature as determined from the DSF data (Fig. 2, see ESI Fig. S5 and S6 †), with T m representing the temperature at which half the protein is denatured. Advantageously, this method highlights not only the value of T m in each case, but also the temperature range over which denaturation occurs and approximate the rate of denaturation. 27 Using the DSF data we were able to calculate the thermodynamic properties of the denaturation process (ESI Method S1 and Table S1 †). The thermodynamic data revealed a key trend: the enthalpy and entropy change (DH m and DS m , respectively) of avidin denaturation were both higher when citrate-phosphate buffer was added to the aqueous IL solution. This effect was most prominent in [Cho][Met]; in the aqueous IL solution (in the absence of buffer) DH m increased from 161 kJ mol À1 to 205 kJ mol À1 upon the addition of buffer to the solution, and DS m also increased from 1880 J K À1 mol À1 to 2330 J K À1 mol À1 . As discussed earlier, the addition of buffer also resulted in an increase in T m for avidin in all IL solutions, indicating that despite the increase in entropic driving force for denaturation in this more complex, higher energy system, the balance between enthalpy and entropy of denaturation overall resulted in a higher thermal barrier to denaturation. These ndings support the concept of molecular connement, involving the strong, direct associations between the molecules of the buffered ILs and charged or polar groups of avidin. Specically, for the unbuffered IL solutions, where the effects of molecular connement were less pronounced, enhanced conformational motions enabled avidin to jump out of wells and explore different conformational environments. However, in the case of the buffered IL solutions, a larger number of molecules were in contact with avidin compared to in aqueous IL solutions. Consequently, in buffered ILs the greater molecular connement imparted resulted in reduced exibility and avidin sampled a sufficiently low energy conformation, so stable that it became trapped in an energy minimum that exceeded the available thermal energy. We highlight that in [Cho]Cl the increase in enthalpy and entropy change was the smallest (DH m from 229 kJ mol À1 to 246 kJ mol À1 and DS m from 2620 J K À1 mol À1 to 2730 J K À1 mol À1 ); however, the addition of buffer provided the greatest increase in thermal stability in this IL. This suggests that in this system the local conformations of avidin became trapped in the buffered IL by energy minima that were relatively more stable than kT. Thus, it appears that connement is strongly dependent on the solvent used and is linked to conformational states. We note that while the results obtained for avidin in buffered IL imply that increased molecular connement can enhance conformational stability, 38 additional work should be conducted to assess the effect of internal structure in increasing thermal stability.
Notably, DSF could not be used to reliably extract information about protein stability for modied avidin or in [Cho] [Hex] or [Cho][Ger] due to high background uorescence caused by hydrophobic interactions between the SYPRO orange dye and the hydrocarbon chains of the IL anions. Nonetheless, the DSF  Table S1 for thermodynamic properties and Method S1 for calculation methods †). data demonstrated that the aqueous environment was important for the thermostability of the avidin conformers, and distinct conformers with different thermostabilities were present in each IL solution.
Thus far we have provided extensive evidence that the conformers, with different populations, change in response to the different IL solutions. 39 We consider that the variability in interaction energy between ions and proteins provides the possibility that some conformations may be strongly preferred while others are inhibited. 40 Hence, we conducted temperature variable CD experiments, in order to probe the energy landscape of avidin and study the role of ion-protein interactions in determining the conformational preferences of avidin in ILs. These were combined with thermodynamic calculations to study the pathways and the conformational transitions and preferences of unmodied and modied avidin in low and high concentrations of ILs. It should be noted that while variable temperature CD spectroscopy is a well-established technique in structural biology for analysing proteins in different solutions, 9,17 the technique is limited when studying ILs, as UV analysis showed that ILs absorb in the far UV (190-260 nm) region and distort the protein sample signal (see ESI  (Fig. 3). Specically, for both unmodi-ed and modied avidin, we observed temperature dependant conformational changes, with a high proportion of b-strand conformations at lower temperatures, shiing to an increasing a-helical population at high temperatures. In 10 wt% [Cho]Cl, unmodied avidin exhibited a positive peak at 228 nm and a negative peak at approximately 218 nm, corresponding to the native avidin b-sheet secondary structure. 15 For modied avidin, the former is not visible, while the latter is signicantly increased in breadth and intensity, likely due to an increase in a-helical structure, with additional negative peaks at 208 nm and 222 nm. 41 We suggest that with increasing temperature, an increase in non-specic interactions between the IL solution molecules and avidin surfaces leads to exposed protein surfaces and disrupted secondary structure, which manifests as the loss of secondary structure in the range of 80 < T m < 90 C (Fig. 3).
Given that the structure of avidin facilitates the entry of biotin into its binding pocket, the thermodynamically stable conformation of avidin is proposed to be a more exible and more open conformation. 42 In line with this, the energy landscape of unbound avidin in IL can be described as rugged, with avidin largely lacking specicity for a single compact conformation, resulting in a heterogenous ensemble of conformations of different stabilities and disordered regions (Fig. 4). 4 Smoothing the landscape can be described by the drive of avidin to remove hydrophobic surfaces from contact with the IL, which would reduce the free surface energy. 5 We observed higher thermostability of unmodied avidin in   Each minimum corresponds to a conformational state of avidin in IL solution, with ions excluded from schematic. The deeper the minima, the more stable the conformation of avidin, and the deepest minimum is associated with avidin in its native, initial state. The interactions between the IL and avidin lead to a rugged surface, with non-native local minima in which avidin conformers can get trapped. As represented by black arrows, avidin can explore conformational space until becoming trapped in a local minimum surrounded by high energy barriers it cannot overcome. the IL ions facilitate conformational transitions via energetically favourable interactions, or by stabilising partially folded states. The protein uctuates between multiple potential energy minima, overcoming free energy barriers between them, toward a thermally stable conformation. 5 In the second case, the IL ions promote conformations susceptible to denaturation, usually due to weak non-specic interactions, and the protein becomes rigid and kinetically trapped by a barrier and is separated from a more thermodynamically favourable and thermally stable conformation. 39,43 In both cases, the extent to which the avidin conformers resemble the native state depends on the initial secondary structure of avidin and the experimental conditions, including temperature, ionic strength, pH, and solution concentration. Notably, in different IL solutions, the native contacts of avidin entered in competition with a vast number of alternative inter-molecular interactions that increased the roughness of the avidin energy landscape. A rougher surface, with a larger number of relatively shallow energy minima, as well as a rough surface within the deeper energy minima, would cause more hindrance for avidin to explore alternative conformational states, which are less kinetically accessible (Fig. 4). 44 Thus, our ndings emphasise that not only the depth and the presence of energy minima, but also the roughness of the energy surface dictates the stability of avidin.
By performing thermodynamic calculations, we can begin to correlate the conformational preferences to the structure of avidin, and gain insight into how the ions associate with avidin in relation to the bulk solution. It should be noted that due to T m approaching the maximum temperature that could be measured without the solvent boiling (95 C), the calculation uncertainties in DH m and DS m were relatively large. Nonetheless, while the denatured state was approximated from a small number of data points, the trend was consistent between the samples and the margin of error was consistent.
It is expected that with increasing IL concentration, the increase in conformational diversity and disorder of the native, initial state would be such that the entropy gain associated with denaturation would decrease. 45 Accordingly, our calculations for unmodied avidin showed that DH m and DS m decreased with increasing IL concentration ( Cl. This may also be due to molecular connement effects discussed earlier, involving the local and long-range contribution of the ionprotein interactions, as the motions of the most exible domains of avidin and the IL molecules are restricted. We found that the values of DH m for modied avidin were 40-70% of the values observed for unmodied avidin, indicating that overall the thermal barrier to denaturation for modied avidin was lower. In this context, compared to unmodied avidin, the intrinsically more highly disordered modied avidin has an energy surface that contains many local energy minima separated by relatively low energy barriers, ensuring transition between different conformational states. Specically, for modied avidin, the values of DH m and DS m increase with increasing IL concentrations, with more pronounced changes observed in [Cho]Cl than [Cho][Hex], supporting that higher concentration is linked to greater entropy and larger free energy, where avidin can adopt a wide range of different conformations with different orientations and arrangements of its helices. It can be suggested that the signicant increase in entropy change for avidin denaturation with increasing IL concentration in [Cho]Cl (1800 J K À1 mol À1 for 50 wt% [Cho]Cl compared to 760 J K À1 mol À1 for 30 wt%), came from the greater exibility of the helical loop of modied avidin in 50 wt% [Cho] Cl. Furthermore, reduction in T m of over 50 C was observed in this case, compared to the modied protein in either concentration of [Cho][Hex], implying avidin could explore conformational space, sampling a wide conformational area and overcoming energy barriers, until becoming trapped in a freeenergy minimum, away from the initial native conformation. This observation highlights that not all avidin systems are created equal, with some ensembles being more heterogeneous than others, and different ensembles exhibiting preferences for particular conformations with different thermostabilities.
Binding experiments were conducted for unmodied and modied avidin in various solvents with azo-dye HABA, a known, weak ligand for avidin, in order to study the effects of different ILs and possible changes in binding affinity via conformational shis. 46 Solutions containing 10 wt% of each of our choline-ILs were tested, as well as 0.1 M solutions of pH 5.1 citrate-phosphate buffer, pH 6.8 phosphate buffer and pH 9.1 and 11 tris buffer. HABA-avidin binding was monitored by UV-Visible (UV-Vis) spectroscopy, with the concentration of the HABA-avidin complex determined from a peak at 500 nm (3 500 ¼ 34 500 M À1 cm À1 ; see ESI Fig. S7 † for example of raw UV-Vis data). The approximate values of the binding dissociation constant, K d , and the number of available binding sites, n, were determined by applying non-linear regression and the Hill-Langmuir equation to plots of the saturation constant (the ratio of concentration of bound HABA to avidin molecules) against the known concentration of HABA added (Table 4 and Fig. 5). 47 In order to systematically examine whether avidin undergoes substantial, environmentally dependent conformational transitions, converting among conformers under different pH conditions, we rst analysed buffer environments as controls.
In buffer solution with pH 6.8 (0.1 M phosphate), at very low HABA concentrations the amount of bound HABA increased rapidly, plateauing with maximal value in this range of approximately 84% of binding sites occupied (Fig. 5). In reasonable agreement with the literature, the observed K d of avidin was 1.13 Â 10 À5 mol dm À3 and n was a reasonable approximation to 4, the number of binding sites on each avidin molecule. 19,48 Binding in buffer solution with pH 5.1 (0.1 M citrate-phosphate), was nearly as efficient as in pH 6.8, indicating good structural preservation of the binding sites. At pH 9.1 (0.1 M tris buffer) this decreased by approximately 10%, with maximal binding site occupancy in this range at 71%, and in pH 11 (sodium hydroxide solution) no binding at any HABA : avidin ratio was observed. It should be noted that this is caused at least in part by ligand effects. HABA has two tautomeric structures, an azo-form and a hydrazone form, only the latter of which is known to bind to avidin. 49 The pK a of the phenolic proton in HABA is approximately 8.2. At pHs above this, HABA will exist primarily as a dianion in the azo-form and avidin-HABA binding is reduced, attributed to the additional energy required to convert the dianion to the hydrazone form for binding. 50,51 This is consistent with our observation of decreased, but still substantial binding in pH 9.1 control buffer solution. It should be noted that the pK a of the hydrazone proton is also anticipated to be around 10, so at high pHs the binding hydrazone tautomer is unable to form and binding cannot occur. 52 For unmodied avidin, we observed no binding in 10 wt%

[Cho][Met] and [Cho][Phe]
, for which the solution pHs were known to be >10.5. This would be inuenced by the pH effect described above, and hence it was not possible to discern the effect of the ions in these ILs on protein-ligand interactions. A total lack of binding was also observed in 10 wt% [Cho][Ger] and [Cho][Hex], respectively, the pH of both of which were 8-8.5 (see Table 2) which could decrease binding but would not be inhibitory. However, it is well established in the literature that ligand binding in avidin is dependent on hydrophobic interactions of residues in the binding site. 49 In both [Cho][Hex] and [Cho][Ger], the anion hydrocarbon chain would be reasonably hydrophobic. It is therefore possible that interactions between the anion and the avidin binding site affected the conformation of avidin such that it could no longer bind HABA. This is further supported by the ndings of the CD experiments, where in [Cho][Hex] avidin conformational change was evidenced by the increase in T m (Fig. 3).
While it is unlikely that major structural changes occurred in the avidin binding site, it is probable that as ions interacted with specic residues, local disturbances indirectly affected the binding site. 29 For example, the ions could disrupt the structuring water molecules surrounding the protein surface and raise the entropy of exibility of the protein, leading to lower binding between avidin and ligand. 53

In [Cho][DHP] and [Cho]
[Asp], the binding curves were of similar amplitude and curvature to buffer, with maximal binding site occupancy of 78% and 81%, respectively, achieved in the range tested; yet, in [Cho]Cl the binding curve slope was signicantly shallower, with maximal occupancy just 53% and K d increased by more than a factor of 10 to 14 Â 10 À5 mol dm À3 , indicating weaker binding (Table 4 and Fig. 5). It is worth considering that despite  signicantly weaker binding observed, unmodied avidin in [Cho]Cl showed higher thermostability, independent of [Cho]Cl concentration. Thus, the specic interactions between the ions and protein residues, whether transient or permanent, affected the electrostatic energies of the protein and could steer it to alternative conformations. 5,29 On the basis of this, we propose that the available conformational space for avidin in 10 wt% [Cho]Cl becomes less restricted as molecular connement is reduced, and lower binding corresponds with more conformationally heterogenous ensembles. Modied avidin in pH 6.8 phosphate buffer showed weaker binding (K d ¼ 18 AE 5 Â 10 À5 mol dm À3 ) and n ¼ 0.5 AE 0.1, indicating that a large proportion of binding sites were no longer active following modication (Fig. 5). Values of n in IL solutions were slightly reduced but comparable to buffer, indicating the ILs were not able to signicantly alter the number of active binding sites. Likely, the presence of excess surfactant (veried by SAXS data), precluded the ligand from binding to avidin. 33,54 To overcome this, one mechanism proposed is for the IL to effectively compete with existing hydrogen-binding interactions and separate the avidin and surfactant. 28,34 However, due to the strong interactions involved, this was not possible for the ILs examined. We focus our analysis on modi-ed avidin in 10 wt% [Cho][DHP] and [Cho]Cl, for a direct comparison with the data obtained for unmodied avidin.
1) was slightly stronger than but comparable to the pH 6.8 buffer. Given that ILs decreased binding strength of unmodied avidin, but slightly increased it for modied avidin, it can be suggested that the ions are likely essential mitigators, coordinating to the charged surface residues of avidin to counteract adverse modication effects and enhance binding strength. 55 It is worth noting that the K d in 10 wt% [Cho]Cl was the same for the unmodied and modied avidin (within the range of uncertainty), while in all other solutions the K d of modied avidin was signicantly lower. Additional evidence for a similar IL-mitigating temperature dependence is also demonstrated by the decrease in thermostability observed for modied avidin in 30 wt% and 50 wt% [Cho]Cl compared to in other solutions examined. The reduced thermostability is attributed to a shi in the conformational landscape of modied avidin. It can be suggested that when the binding site was not exible enough, there was a trade-off between avidin thermostability and conformational stability, in order to preserve the structure of the binding site and maintain high ligand-protein affinity; however this hypothesis requires further investigation. 3,46 Given that avidin can bind to HABA and biotin, it can likely explore a larger conformational space compared to a protein that can only bind to a single ligand. 4,56 In order to demonstrate that ILs can cause a population shi in the conformational landscape of avidin, the biotin ligand, a stronger ligand for avidin than HABA (K d ¼ 10 À15 compared to 6 Â 10 À6 , respectively), was used to study possible competitive inhibition for substrate binding. 48 By performing a titration of the bound complex with biotin, the amount of HABA displaced was monitored by UV-Vis spectroscopy, and the inhibition constant, K i , for biotin with the avidin-HABA complex was calculated ( Fig. 5 and Table 5). 46,48 As expected, the effect of IL on binding with a strong ligand was less pronounced than with a weak ligand (for example, a factor of 5 difference between buffer and [Cho]Cl in biotin binding compared to factor of 13 for HABA).
The  (Fig. 6), demonstrating that biotin binding of unmodied avidin was weaker in these ILs. 48,57 It is also worth noting that, despite HABA not occupying all binding sites at the start of the displacement study, HABA was displaced from binding sites, at a constant rate, in each IL. This shows biotin bound to both the empty and HABA-occupied sites simultaneously and proportionally under all conditions tested. Therefore, in all cases where dynamic conformations were induced by ILs, the ability of avidin to bind biotin at both empty and occupied sites, with little distinction between the two, was retained, regardless of any excess ligand required.
Our   inhibition by biotin was less effective in the former case than the latter. These results demonstrate that the ILs did not directly inhibit the binding site; and we propose that the interactions between the ions and protein facilitated avidin to span a large conformational space and participate in diverse interaction scenarios, for example, allowing sampling of rare conformers and enhanced affinity for HABA relative to biotin. 32 Cl solution (30 mg mL À1 ) was prepared and added dropwise to a gravity column packed with 150 mL resin wetted-bed (approximately 5 equivalents) at a ow rate of 5 mL min À1 , forming [Cho]OH in situ which was added dropwise to a round bottom ask containing the amino acid (phenylalanine or methionine) and water, which was cooled in an ice-bath with stirring. The column was washed through with 300 mL of ultrapure water and the collection vessel was stirred at 3 C and ambient pressure for 24 hours in the dark. Water was removed by rotary evaporation at 30 C and a 9 : 1 acetonitrile/methanol mixture was added to precipitate any unreacted amino acid. The mixture was ltered and the solvents were removed by further rotary evaporation at 30 C. The resulting ILs were dried under reduced pressure (2 mbar) and 35

Protein modication
Surface modication was performed as reported previously. 62 Briey, cationisation was performed by coupling acidic residues on avidin (2 mg) to DMPA (0.9 g, excess) in dilute aqueous solution at pH 5.8 using EDC (0.4 g). The ltered protein solution was dialysed against ultrapure water for 24 hours to remove excess reactants and byproducts and protein concentration was determined by UV-Vis spectroscopy on a Shimadzu UV-2600 (Shimadzu Corporation, Kyoto, Japan). The cationised protein solution was added dropwise to a M n 690 glycolic acid ethoxylate lauryl ether solution, with a 5 : 1 surfactant: positive surface residue ratio (580 equivalents per avidin molecule). Aer further stirring (2 hours) and dialysis (48 hours), the solution was centrifuged at 4000 rpm for 30 minutes, the supernatant decanted and lyophilised and the residue thermally annealed at 60 C until an off-white, viscous liquid formed.
DSC was performed on a TA Q2000 (TA Instruments, New Castle, DE, USA) to demonstrate the conjugation of the protein and surfactant. Samples were heated at 10 C min À1 between À80 C and 100 C and cycled 4 times to erase the thermal history and achieve reliability. Decomposition temperatures were found by TGA on a NETZSCH STA 449 F5 Jupiter (NETZSCH Group, Selb, Germany). Samples in alumina pans were heated from 25 to 500 C at 10 C min À1 under 40 mL min À1 of N 2 and decomposition onset temperature (T d ) determined using Proteus Analysis soware. DLS measurements were collected using a Malvern Zetasizer mV (Malvern Panalytical Ltd, Malvern, UK). Aqueous solutions with a protein concentration of 0.1-1.0 mg mL À1 were measured in disposable cuvettes at 25 C distributions were formed from the accumulation of ten 30 second experiments.
Small angle X-ray scattering Samples of appropriate IL and concentration were prepared from pure IL and ultrapure water and transported under cold conditions (2-8 C) to Diamond Light Source (Oxford, UK). SAXS experiments were carried out on either the B21 or I22 beamlines. Protein samples were 5 mg mL À1 for unmodied avidin and 3.5 mg mL À1 for modied avidin on B21 beamline and 8 mg mL À1 for all protein on I22 beamline. Data was averaged, background subtracted and analysed using ScÅtter soware by SIBYLS beamline, Advanced Light Source (Berkeley, CA, USA). SasView soware (http://www.sasview.org/) was used to more accurately determine the R g of the protein in each sample by tting the data using set parameters and models. The SLD values of the protein, modied protein, surfactant and solvent were calculated from the molecular formulae and density using the online SLD calculator by NIST (Gaithersburg, MD, USA) (https://sld-calculator.appspot.com/). Plots were tted over at least 400 data points, with ts optimised by minimising Chi squared for realistic parameter values.

Differential scanning uorimetry
Tris buffer (0.15 M, pH 9.1) was prepared using tris base and adjusted with HCl. Citrate-phosphate buffers (0.15 M, pH 5.1 and 7.2) were prepared using a 2 : 1 ratio of disodium phosphate and citric acid monohydrate and adjusted with NaOH.
Protein-IL and protein-buffer samples (45 mL) were prepared with 2.0 mg mL À1 avidin and pipetted into a 96-well plate. SYPRO orange aliquots were diluted 25 times with ultrapure water and 5 mL added to each sample, for a nal SYPRO orange concentration of 20X, and the plate was sealed with adhesive lm. Samples were run on an Eppendorf Mastercycler ep realplex quantitative cycler (Eppendorf, Hamburg, Germany) with a 25-99 C temperature ramp, 0.5 C resolution and 4 C min À1 heating rate, followed by a 30 s hold at 99 C. Fluorescence was measured at 570 nm, the emission maximum of SYPRO orange. Spectra for protein-free background samples of each IL were subtracted from the sample spectra. The data was smoothed using Origin Savitzky-Golay method with 7 points and polynomial order 5 (OriginLab Corporation, Northhampton, MA, USA). T m , DS m and DH m were calculated from the raw uorescence data (see ESI Fig. S5 and S6 †) by assuming a two-state denaturation process for avidin (see ESI Method S1 † for detailed method).

Circular dichroism
Temperature controlled CD spectra were recorded at 2 C intervals between 25-95 C on an Applied Photophysics Chirascan Spectropolarimeter (Applied Photophysics Limited, Leatherhead, UK) at Diamond Light Source on beamline B23 with 1 point per 1 nm, 1 s per point from 190-270 nm for [Cho] Cl and 210-260 nm for [Cho] [Hex]. Background spectra for the IL solutions were recorded at room temperature and subtracted from the temperature variable data. Background normalisation was achieved by setting the average ellipticity of the 263-270 nm region to zero. Data was smoothed and T m , DS m and DH m were calculated as for DSF.

HABA binding
Excess HABA was added to buffer/water, the solution shaken for 2 hours then ltered using a 0.22 mm polyethersulfone lter to remove undissolved HABA. HABA concentration was determined by UV-Vis spectroscopy with extinction co-efficient 3 350 ¼ 20 500 M À1 cm À1 . 63 IL was added to aqueous HABA solutions to give 10 wt% IL solutions of known HABA concentration. Modied and unmodied avidin (800 mL, 0.2 mg mL À1 ) samples were prepared in buffer or 10 wt% IL and known amounts of HABA in the corresponding solution were added, UV-Vis spectra recorded, then additions repeated until approaching saturation. Degree of binding was calculated from the absorbance at 500 nm and 3 500 ¼ 34 500 M À1 cm À1 for bound HABA. Reference samples (equal amount HABA solution in the buffer or IL) were recorded simultaneously and the spectra subtracted. Values of dissociation constant K d and number of binding sites n were calculated from plots of HABA concentration vs. saturation function r, with r ¼ ½HABA bound ½Avidin and tting using the enzyme kinetics non-linear regression on origin.

Competitive biotin/HABA binding
Bound HABA-avidin samples with initial HABA : binding site ratio at least 10 : 1 were titrated with a 0.5 mM biotin solution and the decrease in absorbance at 500 nm measured by UV-Vis Spectroscopy. Biotin concentration was plotted against r to calculate the IC 50 of biotin, the half maximal inhibitory concentration and hence the K i (inhibition constant) for avidinbiotin binding in the presence of HABA.

Conflicts of interest
There are no conicts to declare.