Biophysical characterization of the inactivation of E. coli transketolase by aqueous co-solvents

Transketolase activity in organic co-solvents

Protein engineering of the cofactor-binding loops in TK was previously shown to improve the stability of TK to thermal aggregation^14,48,65. They are also known to become more structured upon formation of holo-TK⁴⁰, leading to increased thermostability compared to apo-TK^62,63,66. Therefore, to understand the effect of polar organic co-solvents upon TK stability, including any potential role of the cofactor-binding loops, both apo-TK and holo-TK were incubated with five selected polar co-solvents at a range of concentrations from 0 to 30% (v/v). In the first case, apo-TK was incubated with the organic solvents for 3 h and then incubated with the cofactor to form the holo-TK before diluting and adding the substrates. In the second case, to test holo-TK, the apo-TK was first incubated with cofactors to form the holo-TK, and then incubated with organic solvents for 3 h before diluting and adding the substrates. In both cases, the solvent incubations were shaken at 1000 rpm to ensure that the least miscible solvents, ethyl acetate (EtOAc), and n-butanol (nBuOH) were either fully dissolved or formed into emulsions at the higher concentrations (biphasic systems were only formed for EtOAc and nBuOH at concentrations of 8.3 and 7.3% v/v respectively).

The results shown in Fig. 2 compare the effect of increasing the final solvent concentrations, upon the activity retained after 3 h, for apo-TK and holo-TK. As a control, we also measured the impact of the co-solvents on the residual activity of apo-TK in the absence of any added cofactors (Fig. 3) which is due to binding of endogenous cofactors during protein expression. This confirmed that the residual enzyme activity was 286-fold less than for the co-solvent-free holo-TK, as observed previously in lysates⁶⁴. The activity from the residual holo-TK within the apo-TK preparation, also had a similar dependence on the co-solvent concentrations as the holo-TK preparations.

In nearly all cases, the increased concentration of co-solvents eventually reduced the remaining activity to zero (Fig. 2). Apo-TK was more stable than holo-TK to three of the solvents tested, but less stable than holo-TK in nBuOH, and essentially of the same stability as holo-TK in iso-propanol (iPrOH). For example, acetonitrile (AcCN) half-inactivated apo-TK and holo-TK at 20% (v/v) and 12% (v/v), respectively. Similarly, EtOAc half-inactivated holo-TK at approximately 3% (v/v), and at 6% (v/v) EtOAc for apo-TK. Nevertheless, apo-TK retained 18% residual activity at 30% (v/v) EtOAc, whereas holo-TK was already completely inactivated by 15% (v/v). The sharp decrease in activity up to 10% EtOAc and the apparent retention of activity above it for both apo and holo-TK potentially resulted from having reached the 8.3% (v/v) solubility of EtOAc in water⁶⁷. Increasing the concentration at above 8.3% (v/v) EtOAc, formed emulsions which affected the deactivation profile of apo-TK and holo-TK. Holo-TK was completely inactivated by 15% (v/v), but not for apo-TK, even at 30% (v/v).

In iPrOH, both apo-TK and holo-TK behaved similarly, with a gradual drop-in activity until there was no activity in either case at 25% (v/v) iPrOH. Tetrahydrofuran (THF) was the most effective at deactivation, with holo-TK completely inactivated at just 2% (v/v) THF and apo-TK at 4% (v/v) THF. By contrast, apo-TK was less stable to nBuOH than holo-TK with concentrations of 2% (v/v) and 7% (v/v) respectively giving rise to 50% activity. For both apo-TK and holo-TK in nBuOH the residual activity tended to zero after reaching the 7.3% (v/v) solubility limit for nBuOH.

The generally lower tolerance of holo-TK to co-solvents compared to apo-TK is counterintuitive given that holo-TK is thermodynamically more stable than apo-TK. The dependence on solvent concentration is also clearly not simple. For example, AcCN, nBuOH and iPrOH displayed an initial lag at low solvent concentrations, where the activity decreased only slightly, before reaching a critical concentration at which deactivation occurred more abruptly. Such sigmoidal profiles can indicate a cooperative transition for structural unfolding or even the dissociation of protein dimers into monomers. By contrast, THF and EtOAc titrated out the activity rapidly with increasing concentration of co-solvent, suggesting either an isotherm for solvent binding to the protein, or that any cooperative structural transition was already well underway at the lowest concentration of co-solvent tested.

As apo-TK is already known to be less thermodynamically stable than holo-TK, and yet was found typically to be more co-solvent tolerant, TK inactivation by co-solvent did not appear to relate simply to the global conformational stability (or global denaturation) of the protein. Further biophysical characterisations were undertaken to elucidate whether inactivation was due to global protein denaturation, local protein unfolding, protein aggregation, or some other effect on the protein structure that is not directly detected, such as active-site binding. Before that we determined any relationships between the properties of the co-solvents and their potency for enzyme inactivation.

Correlation of TK activity to calculated organic solvent properties

Although all the organic co-solvents in these experiments were polar, they showed considerable variability in their impact on retained TK activity, including their critical concentrations for inactivation, the sharpness of their deactivation curves, and also in their relative impacts upon apo-TK and holo-TK. The effectiveness of a polar solvent for inactivation of proteins might be expected to depend upon the polarity or hydrophobicity of the solvent, and also their ability to form hydrogen bonds in place of water.

The polar co-solvents can be categorized as aprotic (AcCN, EtOAc and THF), or protic (nBuOH and iPrOH). This simple categorization had no obvious correlation to the concentrations of co-solvents required for complete enzyme inactivation. The characteristics of each co-solvent used in terms of various theoretically calculated and experimentally determined physicochemical properties, is shown in Table 1, along with the Pearson’s R² values for linear correlations between each co-solvent property, and the molar solvent concentrations required for complete inactivation of holo-TK. Log(P) gave the best correlation for both apo-TK (R² = 0.87) and holo-TK (R² = 0.57) (Fig. 4). Other solvent properties correlated much less with the molar holo-TK inactivation concentration, such as the molecular weight (R² = 0.4), molecular volume (R² = 0.4), topological polar surface area (R² = 0.3), number of potential hydrogen bonds (R² = 0.1), dipole moment (R² = 0.2), and the dielectric constant of the co-solvent (R² = 0.4). Although the Log(P) correlation was poor, it showed the highest R² which was consistent with previous observations that correlated hydrophobicity, and not the dielectric constant, to loss of enzyme activity and stability^25,68,69. However, various mechanisms have been used to explain such previous correlations, and so we sought to further characterise the solvent-induced inactivation of TK, as summarised in Table 2 and described below.

Secondary structure of apo-TK and holo-TK in polar co-solvents

To determine the impact of co-solvents on the secondary structure content of TK, and also whether the enzyme was unfolding globally or partially, or aggregating, far-UV circular dichroism (CD) spectra were obtained at approx. 30–45 min intervals for between 3 and 22 h, for both apo-TK and holo-TK, using each volume fraction of solvent that resulted in ≈ 70% inactivation of the apo-TK enzyme (specific activity of 6 µmol mg⁻¹ min⁻¹), except for nBuOH which was increased to ensure holo-TK was also inactivated.

The time dependencies of the mean residue ellipticity at 222 nm for apo-TK and holo-TK in each co-solvent, and also without co-solvent, are shown in Fig. 5A, and the initial rates of secondary structure loss are shown in Table 2. The far-UV CD spectra of holo-TK, after incubation with co-solvents for 3 h, are also shown in Fig. 5B. It can be seen in Fig. 5A that the control samples of apo-TK and holo-TK with no added co-solvent, each showed no loss of secondary structure content over the course of at least 4 h. By contrast, relative to the control samples, the co-solvents each induced small initial losses of secondary structure at rates of between 1.0 and 2.1% h⁻¹ for apo-TK, and between 0.8 and 1.7% h⁻¹ for holo-TK, except for AcCN with holo-TK which led to a more rapid loss of 16% h⁻¹. Apo-TK and holo-TK had similar rates in any given solvent, except for in AcCN where holo-TK lost secondary structure at a rate that was 15-fold faster than for apo-TK, and conversely for nBuOH where apo-TK had a 2.7-fold higher rate than holo-TK. AcCN with holo-TK was the only sample for which aggregates were detected by CD as indicated after 4.5 h of secondary structure loss, by a sudden and simultaneous increase in the dynode voltage and mean residue ellipticity.

It can be seen from Fig. 5A that all samples, except for THF, began with the same initial secondary structure content, with a mean residue ellipticity at 222 nm of − 13,500 ± 500 deg cm² dmol⁻¹. THF by contrast appeared to begin with only 78% of the native structure (− 10,590 deg cm² dmol⁻¹), yet this only decreased to 74% after 3 h. The far-UV spectrum taken after 3 h in THF (Fig. 5B), and at all earlier time-points (not shown), indicated that the smaller ellipticity at 222 nm was due to distortion away from the spectrum expected from a predominantly α-helical protein. This was due to absorbance flattening by THF, that gave a correspondingly increased dynode voltage at 224 nm and below (not shown), to above that acceptable for the instrument, rather than resulting from any actual loss of native structure or conformational change immediately after the addition of THF. EtOAc was similarly affected by a high dynode voltage at below 222 nm leading to considerable signal scattering at below 215 nm at all time-points.

From Fig. 6A,B it can be seen that the initial rates of secondary structure loss determined by CD were found to correlate well (R² of 0.91 for apo-TK, 0.71 for holo-TK, and 0.73 combined) to the activity retained after 3 h of incubation with the co-solvents, and also to the log(P) of the co-solvents (R² of 0.94 for apo-TK), when excluding the case of AcCN for holo-TK. This indicated that the inactivation of the enzyme by co-solvent for both apo-TK and holo-TK was linked to the slow loss of secondary structure observed in most samples. Furthermore, as the secondary structure loss of only 0.8–2.1% h⁻¹ led to 70–100% inactivation in apo-TK and holo-TK, then the far-UV CD signal change was most likely due to local destabilisation and unfolding, including at one or more functionally critical elements of structure within the entire protein population, rather than global unfolding of only 2.1% of the protein population.

For AcCN with holo-TK, the significantly higher rate of unfolding, led to 54% native structure after 3 h, and indicated that AcCN had induced global unfolding, from which aggregation then occurred after 4.5 h. Holo-TK was completely inactive after 3 h, indicating that global unfolding could only account for up to 54% of the inactivation mechanism, and therefore that the local unfolding observed with apo-TK in AcCN, had additionally inactivated holo-TK.

Particle size distributions from dynamic light scattering

Dynamic light scattering (DLS) was used to measure particle size distributions of holo-TK in the presence of the co-solvents (see supplementary Fig. S1). Holo-TK in the absence of organic co-solvent gave a particle size of 7–8 nm as expected theoretically from the native homodimer structure⁷⁰. At 20% (v/v) AcCN and 8% (v/v) nBuOH, holo-TK at 0.1 mg ml⁻¹ contained DLS-detectable aggregates of 1000–4000 nm within 1 h of incubation. At 20% (v/v) iPrOH a very small aggregate peak at 400 nm was observed within 1 h, but more significantly, the monomer peak had shifted to approximately 15 nm indicating either the formation of a soluble oligomer, a shifted monomer peak due to viscosity effects, the formation of a solvent boundary layer, or otherwise significant protein unfolding. In 10% (v/v) EtOAc, only the native-like peak was detected at 8 nm indicating the retention of the homodimer, and no aggregation. For 2% (v/v) THF, the native-like peak increased slightly within 1 h to 9 nm, indicating a homodimer with the peak-shift due to altered solvent viscosity, although not inconsistent with partial local unfolding, or the formation of a significant solvent boundary layer associated with the homodimer. No aggregates were detected with THF.

It should be noted that by DLS, the % volume is proportional to the cube of the particle diameter, and hence even very small amounts of aggregates can suppress the detection of the native monomer peak. Therefore, it is very possible that the monomer, or even the 15 nm state observed for iPrOH, was present also for AcCN or nBuOH, but suppressed by the presence of larger (> 1000 nm) aggregates. In iPrOH, the smaller (400 nm) aggregate was clearly a very low-populated aggregate which allowed the 15 nm state to be observed. Overall, these results are consistent with the emergence of low levels of aggregates in nBuOH and iPrOH, more significant aggregation in AcCN as detected also by CD, and no aggregation observed by any method in EtOAc or THF. Therefore, aggregation was not well correlated to enzyme inactivation, as all solvents led to at least 70% inactivation of holo-TK. Only the significant levels of aggregate observed by CD could potentially contribute to enzyme inactivation for holo-TK, but this only occurred after 4.5 h of incubation, and for one solvent only (AcCN).

Fluorescence spectroscopy of apo-TK and holo-TK in polar co-solvents

Fluorescence tryptophan spectroscopy has been previously successfully applied to provide tertiary structural information for TK^62,63,66,71. For E. coli TK, there are 11 tryptophan residues per monomer which dominate the intrinsic fluorescence signal. These are distributed across the entire structure, with three fully buried and eight partially solvent exposed within each monomer, three located at or close to the dimer interface, and one of those being within one of the cofactor-binding loops. Thus, the intrinsic fluorescence of E. coli TK is responsive to global and local unfolding, dimer dissociation and aggregation, and can lead to both an increase or decrease in intrinsic fluorescence intensity, as observed previously where urea denaturation gave an initial increase followed by a decrease at higher urea concentrations⁶³. By contrast, thermal unfolding results in immediate aggregate formation which gives an increase in the fluorescence intensity of TK as a single cooperative transition due to a net burial of the partially exposed tryptophan residues.

Figure 7 shows the time-dependence of intrinsic fluorescence intensities, for holo-TK in 5–30% (v/v) co-solvent. EtOAc and THF both resulted in steady decreases in fluorescence over time, for all concentrations tested. EtOAc at 10% (v/v) led to a 7% decrease in fluorescence, and THF at 5% (v/v) led to a 9% decrease, each after 3 h of incubation. These results are consistent with the local unfolding without aggregation as observed by CD above. By contrast, the fluorescence intensities for AcCN, nBuOH and iPrOH, resulted in lag-phases followed by rapid increases in fluorescence. The rapid increase occurred progressively earlier with increasing co-solvent concentration, and for AcCN at least the lag phase disappeared at the highest concentrations. This lag-phase behaviour is typical of aggregation kinetics⁷², and the fluorescence increases occurred in the same samples for which aggregation was observed by DLS. For 20% (v/v) AcCN, the fluorescence intensity increased over a timescale consistent with the global unfolding and aggregation observed by CD. For AcCN, nBuOH and iPrOH, it remained possible that the fluorescence changes indicative of aggregation, were also convoluted with contributions from the local unfolding events observed by CD.

The effect of increasing co-solvent concentration on intrinsic fluorescence, as measured after 1, 2, and 3 h of incubation, is shown for apo-TK in Fig. 8 and holo-TK in Fig. 9. For apo-TK, the addition of co-solvents resulted in up to a 10% increase in the intrinsic fluorescence after 3 h in AcCN, nBuOH, iPrOH and EtOAc (Fig. 7), consistent with local rather than global unfolding. Increases were monotonic for 0–30% (v/v) AcCN, 4–30% (v/v) nBuOH, 10–30% (v/v) iPrOH, and 0–29% (v/v) for EtOAc. Below 10% (v/v) iPrOH there was no change, and at 30% (v/v) EtOAc there was a small increase. For THF, the fluorescence intensity decreased exponentially, with most of the change completed over 0–10% (v/v). The lack of obvious transitions for AcCN, EtOAc and THF, and yet clear transitions at 0–4% (v/v) nBuOH, and 10% (v/v) iPrOH, were consistent with the inactivation profiles for apo-TK in Fig. 3.

For holo-TK the addition of co-solvents resulted in similar profiles to those for apo-TK, in EtOAc and THF (Fig. 9). AcCN and iPrOH each gave no change initially and then increased sharply at above 15% (v/v) and 22% (v/v) respectively. In nBuOH, the fluorescence intensity showed a small increase that was monotonic overall, but with a small and sharp increase at 7–8%. These three sharp transitions are larger than any transitions observed in apo-TK, which didn’t aggregate, and they are consistent with the aggregation behaviour of holo-TK observed as time-dependent lag-phases in Fig. 7, and by DLS. Again, the lack of obvious transitions for EtOAc and THF, and yet clear transitions at 15% (v/v) AcCN, 7–8% (v/v) nBuOH, and 22% (v/v) iPrOH, are consistent with the inactivation profiles for holo-TK in Fig. 3. However, the aggregate formation for holo-TK tended to occur at slightly higher solvent concentrations than the holo-TK inactivation, suggesting that it occurred after inactivation, and therefore that aggregates formed from an already inactive enzyme. This was consistent with the earlier observation by CD that inactivation was primarily due to a local unfolding event (or global unfolding in the case of AcCN with holo-TK), and that in some cases for holo-TK this also led to aggregation.