Towards site-specific information on PET degrading enzymes using NMR near operational temperature

Abstract

PETases are enzymes that can break down the poly-ethylene terephthalate (PET) polymer in its constituent building blocks. This enzymatic recycling process offers a sustainable solution for producing new, high-quality plastics from previously used materials. NMR spectroscopy can help in understanding and ultimately improving these enzymes but is always confronted with the lengthy step of acquisition and interpretation of triple resonance spectra for the spectral assignment. Here, we explore whether this step can be made more efficient by recording the spectra directly at high temperature, which simultaneously corresponds to more realistic working conditions for the enzyme. Taking the inactive variant of LCC^ICCG as an example, we compare spectral quality at 30°C and 50°C, and find that the latter condition greatly improves the Signal-to-Noise (S/N) ratio of the standard triple resonance spectra. Going up to 60°C, we show that pulse sequences mainly used for the assignment of intrinsically disordered proteins (IDPs) also become feasible. As a result, we present a methodology enabling exhaustive backbone assignment based on a minimal set of triple resonance spectra acquired and analysed in less than two weeks. The assignment process hence can be completed on a time scale comparable to crystallography, bringing NMR in a favourable position to contribute to bio-structural studies on this family of highly thermostable PETases.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-22351-1.

Introduction

Among analytical techniques, biomolecular Nuclear Magnetic Resonance (NMR) stands out as an exceptionally versatile approach for probing the structure, function and dynamics of biomolecules, with particular strength in enzyme characterization^1,2. Nevertheless, recent advances in structural biology, including the rapid progress of cryo-electron microscopy (cryo-EM)³, the continued improvements in X-ray crystallography^4,5 and the rise of powerful computational modelling have reshaped how researchers obtain structural insights into proteins⁶. This paradigm shift has, in turn, posed new challenges for biomolecular NMR spectroscopy, compelling the field to refine its identity and emphasize the unique advantages it offers^7,8. Despite demanding initial stages, which often require specialized NMR facilities for tasks such as sample preparation (including stable isotope labeling), data acquisition through multiple triple- resonance experiments and complex data analysis like resonance assignment, biomolecular NMR provides an unparalleled ability to probe mechanistic details of enzymes at atomic level directly in solution⁹. In the past decades, important breakthroughs have continued to push the boundaries of what is achievable. Innovations include cell-free labeling systems¹⁰, more efficient data acquisition (Non-Uniform Sampling¹¹, SO-FAST¹²⁾, automated resonance assignment procedures^13–15 and integrative approaches that combine NMR with artificial intelligence (AI) to address long-standing challenges^16,17. Contemporary NMR spectroscopy clearly needs to take the best of these advances to open up to new horizons.

In the field of biotechnology, enzyme optimization is for many programs the key towards the success of bio-based approaches¹⁸. Computational methods based on artificial intelligence^19,20 thereby can generate huge libraries of potential candidates, and coupled to high-throughput screening methods²¹, have the potential to rapidly converge upon improved enzymes. If NMR spectroscopy wants to participate in this effort – and convince that its site-specific information is most useful for example to restrict the list of potential mutation sites for directed evolution^9,22, obtaining a fully assigned ¹H-¹⁵N spectrum as rapidly as possible seems a strict requirement.

One recent example where enzymes are at the core of a promising industrial process is bio-enzymatic recycling of plastic²³. PET hydrolases (PETases) break the ester linkages in polyethylene terephthalate (PET), a common plastic used in the manufacturing of beverage bottles, food containers or clothing fibers. In this rapidly evolving field, many newly identified PETase enzymes from diversity²⁴ have been screened and used as starting points for further engineering. It proved essential to work close to the PET glass temperature of ~ 70°C, where the plastic becomes more dynamic at the molecular level without yet crystallizing²⁵. The temperature of denaturation of the enzymes thereby has become an important factor next to their intrinsic catalytical efficacy. The initially mesophilic Ideonella sakaiensis (Is) PETase²⁶ with its T_m of 46.1°C was evolved to become the ThermoPETase (Tm 58.6°C; Son et al.²⁷), FastPETase (Tm 56.9°C; Lu et al.²⁸), DepoPETase (Tm 69.4°C; Shi et al.²⁹), DuraPETase (Tm 78.7°C; Cui et al.³⁰) and finally HotPETase (Tm 82.5°C; Bell et al.³¹). Derived from the thermophilic Leaf-branch Compost Cutinase (LCC)³², a thermostable cutinase characteristic from the type I group³³, its quadruple mutant LCC^ICCG was equally optimized based on its activity as well as its thermostability (T_m = 85.8°C for LCC and 93.3°C for LCC^ICCG at pH 8, both proteins without C-terminal His-tag)²³. The latter enzyme has shown great promise for industrial application of plastic recycling^34,35.

From an NMR standpoint, this high thermostability could be - at least in theory- a significant advantage, as spectral recording at higher temperatures should lead to faster tumbling of the enzyme and hence mimic a lower molecular weight species. Here, taking the catalytically inactive S165A variant of LCC^ICCG as an example, we explore the possibility to facilitate the resonance assignment of the thermophilic PETases by recording a set of 3D experiments directly at higher temperatures. We first measure experimentally the global tumbling time (τ_c) at different temperatures, and find a twofold reduction already when going from 30°C to 50°C. When comparing spectra at 30°C, temperature where we obtained our first assignments³⁶, and similar experiments recorded at 50°C, we find that the standard triple resonance experiments work significantly better at 50°C, with an important gain in Signal-to-Noise (S/N) ratio and hence reduction in spectral recording time. We equally find that additional experiments based on the hNcaNNH and HncaNNH pulse sequences³⁷, mostly used for assignment of highly flexible Intrinsically Disordered Proteins (IDPs) although equally applied with success to some larger enzymes after complete deuteration^38,39, become feasible without any deuteration, and this despite the 30 kDa MW range of the enzymes. Based on this observation, we achieve de novo backbone assignment of an active variant of LCC^ICCG primarily based on the resulting ¹⁵N and ¹H^N connectivities. Within 6 days of data acquisition followed by 3 days of data analysis, nearly 75% of the ¹⁵N/¹H^N resonances were assigned. Such achievement should enable NMR to follow protein design at a pace at least comparable to X-ray crystallography and alleviate the bottleneck of resonance assignment in large scale projects involving thermostable enzymes.

Materials and methods

Protein expression and purification

LCC^ICCG-S165A and its active variant were expressed and purified according to previous studies^36,40.

NMR spectroscopy

LCC^ICCG-S165A

All experiments recorded for backbone assignment at 50°C were acquired on a 800 MHz spectrometer equipped with a 5-mm cryoprobe with pulsed field z-gradients. The sample contained 580 µM of ¹⁵N-¹³C labeled protein in 25 mM Tris-HCl buffer pH 7.5 with 100 mM NaCl. The data presented here were recorded using the standard Bruker library pulse sequences: 2D ¹H-¹⁵N HSQC, HNCACB and HNCO. All spectra were processed using Topspin 4.0.8 (Bruker Biospin). Spectral analysis for backbone assignments was performed manually using POKY⁴¹. One important experimental factor we found when measuring at higher temperatures is the necessity to seal the sample. Otherwise, already after a day, evaporation set in and shim quality rapidly deteriorated. Spectral dimensions for the HNCO experiments were 3622 Hz (F₁) × 2919 Hz (F₂) × 12500 Hz (F₃) corresponding to 18 × 36 × 15.6224 ppm, with sampling durations of 15.5 ms (t₁), 10.9 ms (t₂), 160 ms (t₃). Spectra were centered at 4.7 ppm (¹H), 117.5 ppm (¹⁵N), 173 ppm (¹³CO). Spectral settings for the HNCACB experiment were 6438.5 Hz (F₁) × 2919 Hz (F₂) × 14084.25 Hz (F₂) corresponding to 32 × 36 × 50 ppm, with sampling durations of 14.9 ms (t₁), 10.9 ms (t₂), 0.009 ms (t₃). Spectra were centered at 4.7 ppm (¹H), 117.5 ppm (¹⁵N), 40 ppm (¹³Cα). Both experiments were acquired with 4 scans per increment.

Active variant of LCC^ICCG

Experiments were recorded at 60 °C on a sample containing 600 µM of ¹⁵N-¹³C labeled of the active variant of LCC^ICCG in 25 mM Tris-HCl buffer pH 7.5 with 100 mM NaCl on a 900 MHz spectrometer equipped with a 5-mm cryoprobe with pulsed field z-gradients with the parameters shown in Table 1:

Table 1.

List of experiments used for the assignment of the active variant of LCC^ICCG.

	Time domain data size (points)			Spectral width/carrier frequency (ppm)
	t1	t2	t3	F1 (1H)	F2(15N)	F3
1H,15N HSQC	3076	128		19.83/4.7	35/117
HNCACB	4000	128	128	15.42/4.7	35/117	60/42 (13C)
HNCO	4000	128	112	15.42/4.7	35/117	13/173.5 (13C)
HNcaCO	4000	128	112	15.42/4.7	35/117	13/173.5 (13C)
hNcaNNH	2048	128	128	13.88/4.7	35/117	35/117 (15N)
HncaNNH	4000	128	512	15.42/4.7	35/117	6/8 (1H)

	NS	Delay time (s)	NUS (%)	Pulse program	Exp time
1H,15N HSQC	4	1	–	hsqcf3gpph19	9 min 55 s
HNCACB	16	1	25	hncacbgpwg3d	22 h 54 min
HNCO	4	1	8	hncogpwg3d	1 h 35 min
HNcaCO	16	1	15	hncacogpwg3d	12 h 5 min
hNcaNNH	64	1	13	hncannhgp3d	1 d 22 h 46 min
HncaNNH	64	1	3.6	hncannhgp3d.2	2 d 6 h 48 min

NUS percentages were estimated based on the number of resonances expected in each dataset. Data were reconstructed using standard MDD software⁴². Spectral analysis and backbone assignments were performed manually using CcpNmr analysis software v2.5⁴³.

TRACT analysis

TRACT spectra were recorded with the pulse sequence of Lee et al., implemented as individual experiments for the α and β ¹⁵N spin states⁴⁴. A relaxation delay of 3 s was used between scans, and 4k points were recorded. A total of 24 delays ranging from 1 to 200 ms were sampled as increments in a pseudo-2D matrix and recorded with 512 scans per delay. After manual phasing and baseline correction, the integral of the [10.0-6.4] ppm region was determined and plotted as a function of the corresponding delay. The resulting curve was fitted as a mono-exponentially decaying function, and the difference between the rates of both components was interpreted in terms of τ_c as described by Robson et al.⁴⁵.

Theoretical calculus of the τ_c value as a function of temperature was performed by fixing the τ_c value at the experimentally determined value of 13 ns for 30°C, and then applying the following formula⁴⁶ :

where η_T is the viscosity of water at the temperature T (using the Celsius scale). The latter was calculated using the equation from⁴⁷

.

or, alternatively, from⁴⁸:

and does not consider that our protein sample was not in pure water but rather in 25mM Tris-HCl and 100mM NaCl buffer pH 7.5.

Results

Determination of τ_c as function of temperature

The global tumbling time τ_c (or, strictly speaking, the correlation time of the dipolar autocorrelation function of a given amide H-N vector⁴⁹⁾ is one criterium that determines the linewidth of the corresponding amide correlation peak. Other factors such as chemical exchange of the amide protons with water or the presence of dynamics in the intermediate time scale regime can equally contribute. Whereas the ratio of transverse (T₂) to longitudinal (T₁) relaxation times can be used to estimate this τ_c parameter⁵⁰, the increasing influence of intramolecular motions on the T₁ relaxation time for larger proteins⁵¹ has motivated the development of an alternative method. The TRACT (TRosy for rotAtional Correlation Times) method measures the decay of both TROSY and anti-TROSY components of the ¹⁵N doublet⁴⁴, and translates the differential rate into a value of τ_c without interference of dipole–dipole (DD) relaxation by remote protons or relaxation contributions from chemical exchange^44,45. TRACT experiments on the LCC^ICCG-S165A enzyme were recorded at various temperatures ranging from 30°C to 60°C. By integrating the envelope of the proton spectra over the amide region as a function of the relaxation delay, both (TROSY and anti-TROSY) relaxation rates were obtained using an exponential minimization (Fig. 1. A & B). The difference between both rates (ΔR₂) was interpreted using the equation given by Robson and coauthors⁴⁵ and yielded τ_c values varying from 13 to 5.7 ns upon increasing the temperature from 30°C to 60°C (Fig. 1. C). The reduction of τ_c is expected, and follows well the temperature dependence of water viscosity expected for a rigid body of this size⁴⁶ (Fig. 1.D). The small discrepancy at higher temperature could come for the buffer and salt concentrations influencing the viscosity⁵². Importantly, accelerated tumbling predicts a significant decrease in transverse relaxation rates for both amide and Cα resonances⁵³, thereby reducing the need for (partial) deuteration of the sample. We show here that because of this superior sensitivity, nearly complete assignment of the backbone and side chain spectra of LCC^ICCG-S165A can be obtained with a minimal set of experiments recorded at 50°C (Mathematical details are given in in the methods section).

Fig. 1.

Determination of the correlation time τ_c as function of temperature. (A,B) TRACT analysis of LCC^ICCG-S165A at (A) 30°C and (B) 50°C. Both panels display the single-spin-state ¹⁵N R_α and R_βrelaxation decay curves obtained using the TRACT sequence. The dots represent the experimental points, and the lines represent the results of fitting the curves to a single exponential decay. (C) Theoretical curve of the difference between ¹⁵N relaxation rates, ΔR₂ as function of τ_c following the equations of Robson and coworkers⁴⁵. (D) Theoretical curve of the temperature dependence of τ_c as predicted for a protein with its τ_c at 30°C fixed at 13 ns, based on the temperature dependence of water viscosity. Solid lines shows the τ_c values as a function of temperature as calculated with water viscosity according to Weast⁴⁷ (black) or Kestin et al.⁴⁸ (green). On C and D panels, dots correspond to experimental values at 30°C (yellow), 40°C (orange), 50°C (red) and 60°C (blue). Error bars on (C) and (D) were calculated based on and estimated uncertainty of +/- 0.5 s⁻¹ for the fits of the R_α and R_β curves.

Influence of the temperature on spectral quality

To evaluate experimentally the influence of temperature on the spectral quality of our PETase, we recorded on the same sample at 30°C and 50°C both the ¹H-¹⁵N HSQC and ¹H-¹⁵N TROSY correlation spectra using standard Bruker pulse programs (hsqcf3gpph19 and trosyf3gpph19). Already at 30°C, signal intensity improved by ~ 40% when comparing the HSQC experiment with the TROSY version, and going to 50°C clearly boosted the intensity gain by another ~ 15% (Fig. 2.A). Equally important, the number of peaks in the 2D correlation spectra remained nearly the same between spectra recorded at both temperatures (Fig. 2.B), indicating limited solvent exchange.

Fig. 2.

Selection of the best conditions and experiments for PETase assignment LCC^ICCG-S165A. (A) 1D ¹H slices through G88 (top), A59 (middle), G163 (bottom) cross peaks from ¹H-¹⁵N HSQC at 30°C (blue), ¹H-¹⁵N TROSY at 30°C (light-blue), ¹H-¹⁵N HSQC at 50°C (red) and ¹H-¹⁵N TROSY at 50°C (orange). (B) ¹H-¹⁵N HSQC at 30°C (blue) and 50°C (red). Cross peaks generally shift upfield, but their number is the same at both temperatures (C) ¹H-¹³C HSQC at 30°C (blue) and 50°C (red). The insets show how representative ¹³Cα signals are sharper at 50°C compared to 30°C. The ¹H 1D slices were taken at the ¹³C frequency displayed by the dashed lines. (D) 1D plot of the 1st FID of HNCO (top) and HNcaCO (middle) along with a positive projection of a HNCACB experiment (bottom) measured at 30°C (blue) and 50°C (red).

Further time-saving could be obtained by BEST-⁵⁴ or SOFAST-¹² HSQC sequences and should therefore be considered before setting up any triple resonance experiment. However, the reduced spin-diffusion at higher temperatures should diminish the difference between selective and non-selective relaxation after excitation of the amide proton pool, which is the underlying principle of these pulse sequences³⁶. Moreover, due to the presence of extreme ¹H chemical shifts for some amide protons of LCC^ICCG (reaching 10.9 ppm or 5.6 ppm), the use of these schemes in our particular case led to the loss of ~ 10 resonances (Figure S1), and was therefore not pursued in the present study. In accordance with the lower τ_c value, NOE based magnetization transfer proved slower at 50°C than at 30°C⁵⁵, but should simultaneously suffer less from spin diffusion (Figure S2).

In a typical backbone assignment experiment (e.g. HNCACB), the lengthy back-transfer of ¹³Cα to ¹⁵N magnetization is often a limiting step due to the rapid relaxation of the ¹³Cα transverse magnetization. While in a HMQC experiment, this loss is amplified by the short ¹H transverse relaxation time, the HSQC experiment is less impacted as the ¹H magnetization is stored along the z-axis and the ¹H T₁ is long . This is exemplified when comparing ¹H-¹³C HMQC and HSQC at 30°C where one can clearly see the increased number of visible resonances in the HSQC experiment (Figure S3). To evaluate how beneficial spectral recording at 50°C would be, we then measured ^1H-¹³CHSQC spectra with a wide ¹³C spectral window both at 30°C and 50°C (Fig. 2.C). Relative to the isolated methyl signal of M166⁵⁶, we observe a nearly 40% increase in intensity for ¹Hα-¹³Cα resonances in the spectrum recorded at 50°C compared to that at 30°C. To evaluate the effect this has on triple resonance spectra, we acquired a series of ¹H-¹³C planes from HNCO, HNcaCO and HNCACB spectra using the Bruker standard HSQC-based (hncogpwg3d, hncacogpwg3d, hncacbgpwg3d) experiments both at 30 °C and 50°C. When comparing the ¹H spectra resulting from the 1st Free Induction Decay (FID) for HNCO and HNcaCO and the positive sum of the HNCACB plane we observe significant improvement in the S/N ratio of 20–30% for HNCO and 50% for HNcaCO and HNCACB when measuring at 50°C (Fig. 2D and Figure S4). This improvement can be confirmed by the results of an automatic peak detection performed on the respective planes, that finds for both HNcaCO and HNCACB planes nearly twice the number of resonances at 50°C compared to identical dataset at 30°C. A detailed analysis of the increase of observable peaks is presented in the SI.

To obtain the resonance assignment of LCC^ICCG-S165A, we recorded a full HSQC-based HNCACB triple resonance spectrum at 50 °C and pH 7.5. Starting from our previous assignment of the ¹H-¹⁵N HSQC spectrum of LCC^ICCG-S165A at 30°C⁵⁶ which at the time required nearly 4 weeks of experimental time, we used this single spectrum to confirm the assignments at 50°C (Figure S5). A HNCO experiment was recorded to obtain the ¹³C’ resonances at 50°C. In total, 85% of the ¹H^N/¹⁵N resonances along with 91% for ¹³Cα (234 out of 258), 90% ¹³Cβ (215 out of 239 residues, excluding the 19 glycines) and 85% of ¹³C’ resonances (203 out of 239, excluding all residues preceding a proline residue) were assigned.

Raising the temperature for hNcaNNH

In our general effort to accelerate the assignment of the ¹H-¹⁵N spectrum, we explored the possibility to record a hNcaNNH experiment³⁷. While the common approach of backbone assignment consists of connecting consecutive residues through their various ¹³C resonances, these experiments have been developed to obtain sequential correlations between successive amide proton and nitrogen resonances ¹H_(i−1), ¹⁵N_(i−1), ¹H_(i), ¹⁵N_(i), ¹H_(i+1), ¹⁵N_(i+1)^37,57,58. Although recognized as true game changers in the assignment process for disordered proteins^59,60, they have not been used for non-deuterated globular proteins of the size of PETases due to their inherent low sensitivity. Indeed, the bottleneck in these experiments is the fast ¹³Ca relaxation rate that significantly deteriorates their sensitivity. To explore whether these experiments would profit from the same sensitivity gain as the above-described triple resonance spectra with a single period of ¹³Ca transverse relaxation, we recorded a hNcaNNH at 30°C and 50°C with the standard pulse sequence from the Bruker library (hncannhgp3d).

At 30 °C, no signal could be observed (Figure S6), and when raising the temperature to 50 °C the sensitivity of this experiment proved at the edge (S/N < 5 for most peaks) in a reasonable amount of experimental time (2 days 15 hours for a fully sampled dataset) (Fig. 3.A). However, the signal-to-noise ratio further increased by an average 70%, and even by a factor of two for certain residues (Figure S7) when raising the temperature to 60°C (Fig. 3.A). For many resonances where the connecting peaks were absent in the spectrum at 50°C, we obtained at 60°C the connection through their ¹⁵N_(i−1) and ¹⁵N_(i+1) resonances. (Fig. 3.B).

Fig. 3.

Connecting ¹⁵N resonances through hNcaNNH experiment. (A) Overlay of the ¹H-¹⁵N strip extracted from the hNcaNNH at 50°C (blue) and 60°C (red) for A59 and L203. ¹⁵N traces were extracted at the ¹H frequency of the selected signal. (B) ¹H-¹⁵N strips of consecutive residues at 60°C shown for residues from G113 to V118. Red resonances correspond to ¹⁵N(i-1) and ¹⁵N(i + 1) while the autocorrelation cross-peak leading to the ¹⁵N(i) was added manually in orange based on the ¹⁵N value obtained from the ¹H-¹⁵N HSQC spectra.

Rapid de novo backbone assignment

Based on the quality of the hNcaNNH acquired at 60°C, we sought to investigate whether a de novo backbone assignment primarily based on the hNcaNNH and HncaNNH experiments would be possible. Turning stretches of connecting resonances into a backbone assignment further required a HNCACB spectrum to obtain amino acid type identifying ¹³C chemical shift values (Fig. 4). Finally, we recorded HNCO and HNcaCO spectra for independent verification of the obtained assignments (Table 1). With a 500 µM sample of the active enzyme in a 5 mm tube, on a 900 MHz spectrometer equipped with a triple-resonance cryogenically cooled probe head, the complete data set was acquired within 6 days of machine time (see Methods).

Fig. 4.

De novo assignment primarily based on ¹⁵N and ¹H connectivities. (A) Example of connectivity between the HncaNNH (teal), HSQC (blue) and hNcaNNH (skyblue) experiments for residues V116-L117-V118. The ¹H-¹H and ¹⁵H-¹H strips were extracted at the ¹H^N resonance frequency of the 3 signals. The strips are only displayed in the window corresponding to the HSQC experiment (blue). The resonances of V115 and I119 respectively on the V116 and V118 strips are outside of displayed window. The orange line shows the connectivity for L117 both from V116 (i-1) and L118 (i + 1). (B) HNCACB strips displayed for all 3 residues with the Cα in red and Cβ in green.

The elevated temperature (60°C) led to improved signal-to-noise data and further allowed efficient non-uniform sampling providing very high-resolution spectra. Especially for the HncaNNH experiment, this enhanced resolution proved important to establish firm ¹H^N connectivities between resonances of neighboring residues and thereby to speed up significantly the assignment procedure. To complement the assignment strategy, the HNCACB experiment was employed to confirm the sequential connectivity of residues. This experiment enables the identification of both intra-residue (i) and preceding residue (i-1) carbon (Cα and Cβ) resonances, thereby establishing the directionality of the assignments.

As the overall goal of our work was to streamline and accelerate the backbone resonance assignment of thermostable enzymes, we tested NMRtist, an AI-driven tool that enables automated processing from peak picking to resonance assignment and ultimately to structure calculation¹⁴. NMRtist requires as input the amino acid sequence of the target protein together with peak lists derived from 2D and 3D NMR spectra. Accordingly, we prepared peak lists of ¹H-¹⁵N-HSQC and 3D HNCO, HNcoCA, HNCACB, HncaNNH, and hNcaNNH spectra. Automated peak-picking was performed with CCP-NMR v2 software, since NMRtist does not currently support peak picking for HncaNNH and hNcaNNH spectra. When providing only the carbon-connectivity-based spectra (HNCO, HNcoCA, HNCACB) along ¹H-¹⁵N-HSQC peak lists, NMRtist returned an estimated chemical shift assignment coverage of 46%. This value increased to 65% upon inclusion of the HncaNNH and hNcaNNH peak lists as input. The inclusion of HncaNNH and hNcaNNH peak lists hence significantly improves the overall assignment score, despite the fact that these spectra are underrepresented in the training data, potentially leading to reduced performance. Therefore, we proceeded simultaneously with a manual backbone resonance assignment. Within three days of spectral analysis, and without relying on the previously generated NMRtist outputs, we successfully assigned nearly 80% of backbone (¹H/¹⁵N) resonances (Figure S8). A comparison between the manual and NMRtist-derived assignments revealed inaccuracies in the latter which are likely attributable to the fact that our PETase, with its 258 amino acids, lies at the upper limit of the protein size range for which NMRtist was trained. In addition, the slightly lower percentage than the one obtained previously for the inactive LCC^ICCG variant is primarily due to additional broadening of some resonances assigned to solvent exposed residues that were already at the limit of observation at 50 °C in LCC^ICCG-S165A (T211 and G127; see Figure S5).

Discussion

The initial steps of the development of new or improved PETases have invariably involved molecular modeling and/or bioinformatics, whereby the starting 3D structures that served as input for these efforts mostly came from X-ray crystallography. Examples such as TfCut2 (PDB code: 4CG1⁶¹⁾, Cut190 (5ZNO⁶²⁾, LCC and mutants (4EBO³² and 6THS, 6THT³⁴⁾ or IsPETase (5XG0⁶³⁾, are reviewed by Liu et al.⁶⁴ and Tournier et al.²³.

Solution biomolecular NMR has only recently entered this field, and not as much as to obtain structural information but to characterize the interaction between the enzyme and small soluble substrates^56,65 or to investigate the role of the catalytic histidine with its associated pKa value⁴⁰. Recently, a solid state NMR study of the TfCut2 enzyme embedded in a trehalose glass in the presence of nanoparticles (NPs) made of ¹³C labeled PET oligomers equally added to our understanding of the enzyme/polymer interaction and the biocatalytic mechanism of the hydrolase reaction⁶⁶. However, if NMR wants to consolidate its place in this large-scale project that is the development of industrial PETases, easy, rapid and robust methods providing the assignment of the ¹H-¹⁵N and/or ¹H-¹³C spectra are required.

Our study demonstrates that recording NMR experiments at elevated temperatures can significantly accelerate backbone resonance assignment of thermostable PETases. Using the TRACT approach^44,45, we observed that the overall global tumbling time was reduced by approximately a factor of two, effectively making a 25 kDa protein behave like a ~ 10 kDa protein in the NMR tube. Furthermore, the sensitivity increase that accompanies the line narrowing by this apparent molecular weight decrease (Fig. 2) is not counteracted by opposing effects such as increased exchange with the water or other line broadening effects. Indeed, we observed very limited amide proton exchange up to 50°C/60°C, with only a couple of resonances less intense compared to 30°C. This sharply contrasts with the case of IDPs where amide proton signals at neutral to alkaline pH completely disappear at higher temperatures^67,68. An upfield shift for most of the amide protons is observed at the high temperature, due to weakened hydrogen bonding^69–71. It is not clear at this moment whether the pH dependence of our Tris buffer at higher temperature (with a predicted lowering of 0.3 units between 30°C and 50°C) contributes to this shift or to limiting the water exchange. Finally, NOE build-up curves at both temperatures (Figure S2) showed the feasibility of recording a high-resolution NOESY-HSQC spectrum even at 50 °C, and could thereby provide the experimental data for efforts to assign the ¹H-¹⁵N HSQC on the sole basis of a 3D NOESY-HSQC experiment together with the experimental (or predicted) 3D structure^14,72.

Particularly true for enzymes near or beyond 30 kDa, considered the size-limit of classical solution state NMR, assignment of the ¹H-¹⁵N HSQC spectrum remains a real bottleneck for many projects in which NMR is or could be involved. Except for intrinsically disordered proteins (IDPs), where the relevant τ_c parameter does not correlate with the MW of the protein, the size problem leads not only to an increased number of peaks but also to peak broadening and hence decreased signal-to-noise. Deuteration^73–75 together with TROSY approaches³⁶ can alleviate the latter peak broadening, but significantly increases the price per sample, with associated costs that become an important factor to consider in a program that generates many variants. Altogether, we observe in the Biomagresbank⁷⁶ an exponentially decreasing number of deposited assignments for proteins larger than 150 residues, and finally no more than ~ 100 entries for proteins of the size of LCC and its derivatives¹⁴.

Although the NMR assignment clearly will never keep up with the many PETase variants that are currently tested in academic and industrial laboratories all over the world, speeding up the process could increase its use in this rapidly evolving field. Not only do we increase the sensitivity of the standard triple resonance experiments that are the cornerstone of the assignment process by matching ¹³C frequencies (Fig. 2.D), but directly connecting consecutive amide peaks through the complementary hNcaNNH or HncaNNH spectra becomes possible (Figs. 3 and 4) and dramatically speeds up the procedure, be it manual or (semi-)automatic. High temperatures equally imply we are studying these enzymes in what we would call ‘near-working industrial conditions”³⁵ (to make the parallel with “near-physiological conditions” for disease related enzymes), and should therefore be seen as a win-win situation for solution state NMR : it not only yields higher-quality data and opens the door to experiments that would otherwise be impossible, but it also gives information about the enzyme in conditions that approximate its real-world functioning.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

V.G., E.B. and F.-X.C. recorded and analysed NMR spectra. J.G., L.P. and S.G. resources. G.L. and C.C. conceptualization, project administration, supervision, writing – original draft preparation, writing - review & editing. All authors reviewed the manuscript.

Data availability

All the parameter sets for the NMR spectra have been deposited to https://zenodo.org/records/15425383 and/or contact either glippens@insa-toulouse.fr or charlier@insa-toulouse.fr. Chemical shifts have been deposited on BMRB under the number 53331.

Declarations

Competing interests

The authors declare no competing interests.