Characterising side chains in large proteins by protonless ¹³C-detected NMR spectroscopy

Abstract

Side chains cover protein surfaces and are fundamental to processes as diverse as substrate recognition, protein folding and enzyme catalysis. However, characterisation of side-chain motions has so far been restricted to small proteins and methyl-bearing side chains. Here we present a class of methods, based on ¹³C-detected NMR spectroscopy, to more generally quantify motions and interactions of side chains in medium-to-large proteins. A single, uniformly isotopically labelled sample is sufficient to characterise the side chains of six different amino acid types. Side-chain conformational dynamics on the millisecond time-scale can be quantified by incorporating chemical exchange saturation transfer (CEST) into the presented methods, whilst long-range ¹³C-¹³C scalar couplings reporting on nanosecond to millisecond motions can be quantified in proteins as large as 80 kDa. The presented class of methods promises characterisation of side-chain behaviour at a level that has so far been reserved for the protein backbone.

Analysis of side-chain motions by NMR has so far been restricted to small proteins and methyl-bearing side chains. Here, the authors present NMR methods based on ¹³C direct detection of highly deuterated protein samples that yield sharp and well-resolved signals and allow the characterisation of side-chain conformational dynamics of six different amino acid types in medium-to-large proteins.

Introduction

Proteins are dynamic entities whose molecular function is intrinsically related to their structure and dynamic sampling, both in the immediate vicinity of active sites and in regulatory sites^1,2. Proteins are often viewed as representations of their backbone and most experimental studies of protein dynamics to date have primarily focussed on the protein backbone, with much less attention paid to side chains^3–9. Whilst knowledge of the behaviour of the protein backbone is essential in order to understand many aspects of protein function, bringing side chains into focus is crucial. It is the side chains that give the amino acids in proteins their unique chemical diversity, for example, side chains form critical parts of many active sites of enzymes and the side chains presented on the protein surface are key to substrate recognition and binding events. In many cases, the dynamics and interactions of side chains are as, if not more, important to the biological function than the overall backbone conformation. In addition, the motions of side chains are often decorrelated from the backbone^10–12. In order to understand the mechanisms of enzymes and characterise macromolecular interactions and regulation, it is imperative to be able to specifically characterise the side-chain structure and movements.

Nuclear magnetic resonance (NMR) spectroscopy is uniquely situated to generally characterise the interactions and conformational sampling of side chains in proteins. However, most experimental studies of protein motions using NMR spectroscopy to date have focussed on the backbone or methyl groups, and detailed information regarding side-chain behaviour has been restricted to small proteins^13,14 and methyl-bearing side chains^15–17. The limitations of currently available approaches are mainly due to line broadening and signal loss as a result of efficient relaxation pathways, as well as insufficient resolution and signal overlap in the NMR correlation maps used to extract parameters reporting on dynamics and interactions.

Here, we present a class of NMR methods that allows a more general characterisation of side chains in medium-to-large proteins. The presented class of methods is anchored in ¹³C-direct detection¹⁸ NMR spectroscopy of per-deuterated proteins. The slow relaxation of the aliphatic ¹³C nuclei in deuterated proteins results in sharp NMR signals, whilst the correlation of two ¹³C chemical shifts results in high-resolution and well-resolved two-dimensional NMR spectra. It is shown that the slow transverse ¹³C relaxation rates enable a large range of NMR experiments to be performed to characterise the structure, interactions and dynamics of side chains in medium-to-large (< 82 kDa) proteins. The presented experiments include a quantification of long-range ¹³C–¹³C scalar couplings¹³ reporting on the sampling of side-chain dihedral angles and chemical exchange saturation transfer (CEST) experiments¹⁹ to characterise low-populated states.

Results and discussion

Side-chain ¹³C–¹³C correlation spectra

The BMRB²⁰ databank provides a large database of assigned NMR chemical shifts, including the side-chain ¹³C chemical shift. Analysis of these data reveals six side chains (Fig. 1; Supplementary Fig. 1) with a ‘terminal’ ¹³C (¹³C_t), which is directly bonded to just one other ¹³C and which has a chemical shift distribution that is isolated from its directly bonded penultimate ¹³C (¹³C_p). This distinct chemical shift profile means that the terminal ¹³C_t can be selectively excited in an NMR experiment using frequency-selective pulses. Selective excitation enables spectral filtering and observation of specific residue types in multidimensional NMR spectra, even when using uniformly isotopically labelled proteins.

Fig. 1.

Examples of aliphatic ¹³C chemical shift distributions. Distribution of the assigned ¹³C chemical shifts of (a) arginine and isoleucine (b) residues extracted from the BMRB²⁰ databank. Terminal ¹³C_t, directly coupled to just one other ¹³C are coloured blue and the bonded, penultimate ¹³C_p, are  coloured red. For clarity, only one of the two ¹³C_t–¹³C_p pairs have been highlighted in isoleucine (See Supplementary Fig. 1 for other sites)

The core element of the presented side-chain-specific NMR experiments, Fig. 2a, correlates the terminal carbon ¹³C_t with its directly bonded penultimate ¹³C_p within a protein side chain. The terminal ¹³C_t includes ¹³C^δ in the arginine side chains, ¹³C^δ1 and ¹³C^γ2 in isoleucine, ¹³C^ε in lysine, ¹³C^δ in proline, ¹³C^γ2 in threonine and ¹³C^γ in valine residues. The fact that the magnetisation of interest starts and is detected on the terminal ¹³C_t has some key advantages. The homonuclear coupling pattern for ¹³C_t is a simple doublet, which allows for the magnetisation of interest to be transferred completely between ¹³C_t and ¹³C_p using simple INEPTs and also facilitates virtual decoupling in the ¹³C_t detected dimension.

Fig. 2.

¹³C–¹³C side-chain correlation spectra of per-deuterated proteins. a Schematic representation of the NMR pulse sequence used to obtain ¹³C–¹³C side-chain correlation spectra. The flow of the magnetisation between ¹³C_t (blue) and ¹³C_p (red) is shown above the sequence with colour gradients. The following delays are used: Δ = 1/(4J_CC) ≈ 7.1 ms, T = 1/(2J_CC) ≈ 14.1 ms, where J_CC is the one-bond ¹³C–¹³C scalar coupling constant. Rectangular pulses are high-power and not selective, bell-shaped pulses are frequency selective (90°: white outlined, 180°: black). Deuterium, ²H, is decoupled throughout the sequence and frequency discrimination is obtained by states–TPPI of phase ϕ²¹. b Schematic representation of post-processing to obtain the decoupled spectrum. c Arginine ¹³C^δ–¹³C^γ correlation of the 18-kDa protein T4L L99A, obtained on a 1.4 mM sample at a static field of 14.1 T at 278 K in 37 min

In the core experiment, Fig. 2a and Supplementary Fig. 2, magnetisation is initially transferred from ¹³C_t to ¹³C_p, then labelled with the chemical shift of ¹³C_p during t₁ and finally transferred back to ¹³C_t for detection. The doublet peak splitting observed for ¹³C_t because of the evolution of the single one-bond ¹³C_t–¹³C_p scalar coupling during acquisition (t₂) can conveniently be resolved by recording two sub-spectra, in-phase (IP) and anti-phase (AP)²². Figure 2b shows how subsequent post-processing is used to virtually ‘decouple’ the spectra in the direct ¹³C_t dimension, so that single peaks are observed in multidimensional correlation spectra. One-bond ¹³C–¹³C couplings in indirect dimensions (t₁) are removed using constant-time evolutions²³, and evolutions of the ²H–¹³C couplings are efficiently suppressed using ²H decoupling (see Supplementary Fig. 2).

Application of the method to observe ¹³C–¹³C correlations in the 18-kDa L99A mutant of T4 lysozyme (T4L L99A) is shown for arginine ¹³C^δ–¹³C^γ in Fig. 2c, and for other side-chain correlations in Supplementary Figs. 3 and 4. Of importance is that (1) the NMR correlation spectrum in Fig. 2c is obtained in 37 min at 278 K, where the rotational correlation time of the 18-kDa T4L L99A mimics that of an ~30-kDa protein at room temperature, (2) six different ¹³C–¹³C correlation maps can be obtained in less than 12 min (see Supplementary Figs. 3, 4) and (3) a single uniformly [²H,¹³C] isotopically labelled sample could be used to characterise six side chains. The lack of directly bonded protons and absence of efficient relaxation pathways for aliphatic ¹³C in highly deuterated proteins dramatically reduces the ¹³C relaxation rates. For T4L L99A, the isoleucine ¹³C^δ1 non-selective longitudinal relaxation rates, R₁, range between 0.12 and 0.24 s⁻¹ at 278 K and a field of 14.1 T (see Supplementary Table 1 and Supplementary Fig. 5). This necessitates longer recycling delays and fewer scans per unit time, and the lower gyromagnetic ratio of ¹³C compared with ¹H leads to an intrinsic lower signal-to-noise. However, reduced relaxation rates also lead to small transverse relaxation rates, between 2.7 and 8.8 s⁻¹ for isoleucine ¹³C^δ1 in T4L L99A at 278 K and 14.1 T (see Supplementary Table 2 and Supplementary Fig. 7), which in turn lead to very sharp signals. For example, with the ¹³C-detected method, it became possible to observe sites in arginine side chains of T4L L99A that were not detectable in a ¹H-detected equivalent experiment, even when using a highly optimised isotope-labelling scheme (Supplementary Fig. 6). Another striking advantage of ¹³C–¹³C correlation spectra compared with ¹H–¹³C spectra is the substantially better chemical shift dispersion in the directly detected dimension, resulting in significantly better peak separation (see Supplementary Fig. 6 for a comparison). Moreover, as compared with ¹H–¹³C spectra, the ¹³C–¹³C correlation maps directly provide the chemical shift of two aliphatic ¹³C that are both known to report on the structure and sampling of side chains^24,25.

Direct-detected ¹³C spectra of a 82-kDa protein

The excellent spectra obtained on the medium-sized T4L L99A protein at low temperature and the favourable ¹³C transverse relaxation rates show that ¹³C-direct detection is ideally suited for characterising side chains in large proteins. Side-chain ¹³C–¹³C correlation maps were recorded for the significantly larger 82-kDa malate synthase G (MSG) protein²⁶. Figure 3a and c shows the isoleucine ¹³C^δ1–¹³C^γ1 and valine ¹³C^γ–¹³C^β correlation maps of MSG, respectively, where excellent chemical shift dispersion and resolution result in nearly no overlap of peaks even in this large and uniformly [²H,¹³C] isotopically labelled system. The slow transverse relaxation rate of aliphatic ¹³C in per-deuterated proteins means that the experiment shown in Fig. 2a easily can be extended to obtain numerous other NMR parameters reporting on motions, conformations and interactions. TOtal Correlated SpectroscopY (TOCSY) NMR experiments are typically used to aid side-chain chemical shift assignment²⁸ of proteins and a couple of representative examples are shown in Fig. 3b for MSG, with details of the sequence given in Supplementary Fig. 8.

Fig. 3.

¹³C–¹³C side-chain correlation spectra of a large protein. a Isoleucine ¹³C^δ1–¹³C^γ1 and c valine ¹³C^γ–¹³C^β correlation spectra of a 0.5 mM sample of the 82-kDa protein MSG. Each of the two spectra in a and c were recorded in 7.5 h at 18.8 T using a standard helium-cooled Bruker TCI radio-frequency probe. b Correlating the ¹³C^δ1–¹³C^γ1 of isoleucine with the rest of the side chain ¹³C for chemical shift assignment. Representative TOCSY strips from a three-dimensional ¹³C-detected CC-TOCSY experiment (see Supplementary Fig. 8) for I229 and I309 (marked with black arrows in a. Chemical shift assignments in a and c were obtained from (refs. ^26,27) combined with the 3D-CC-TOCSY. Peaks labelled Ile in c are isoleucine ¹³C^γ2–¹³C^β correlations. All experimental parameters are given in Supplementary Materials and Methods

Characterisation of millisecond dynamics

A full characterisation of the role of side chains requires an appreciation of their motions within the protein. Of substantial importance is that the method detailed above can be extended to allow quantification of side-chain motions across a wide range of timescales. Although many experiments will be possible using the new scheme, two examples, which together report on side-chain motions on timescales from nanoseconds to milliseconds, were chosen here to highlight the versatile applicability.

Protein dynamics and conformational exchange on the millisecond timescale have been shown to be important for the function of many proteins²⁹ and the chemical exchange saturation transfers (CEST)^19,30 NMR experiment, amongst others, allows a quantification of exchange events on this timescale. CEST experiments have recently been adapted for side chains in small proteins¹⁴, in a ¹H-detected manner. Integrating CEST with the ¹³C-detected method described above (see Supplementary Fig. 9) allows quantification of side-chain conformational exchange in medium-to-large proteins between the ground state (G) and an excited state (E). Examples of CEST profiles are shown in Fig. 4 and Supplementary Fig. 9, where the conformational exchange of T4L L99A at 278 K is quantified by a ¹³C-detected CEST experiment. The calculated overall exchange rate, k_ex = 128 ± 18 s⁻¹, and the population of the excited state, p_E = 1.15 ± 0.11% (see Supplementary Fig. 10), agree well with previous studies¹¹ showing that reliable parameters are derived from ¹³C-detected CEST experiments. The chemical shifts of aliphatic ¹³C within the protein side chains report on the rotameric sampling of the side chain²⁴, and the results of the CEST experiments in Fig. 4 also report on the rotameric sampling in the excited state. Specifically, from the chemical shifts obtained for V103 in the ground and excited states, ϖ_G and ϖ_E, respectively, it can be calculated²⁴ that the χ₁ angle of V103 changes from a predominantly trans conformation (91% trans, 9% gauche-plus) in the ground state to a mainly gauche-minus conformation in the excited state (26% gauche-plus, 16% trans, 58% gauche-minus), which is in agreement with the existing structures of T4L L99A in the ground and the excited state¹¹. Access to the ¹³C chemical shifts in the excited state via the ¹³C-detected CEST experiment also reveals that the V103 side chain is more dynamic in the excited state with a broader distribution of side-chain rotamers.

Fig. 4.

Quantifying side-chain motions on the millisecond timescale. CEST profiles to quantify millisecond chemical exchange of V103 ¹³C^γ2 in T4L L99A at 278 K are shown. The circles represent experimentally obtained normalised intensities, and the lines are the result of a least-squares fit. Uncertainties in the reported ϖ_G and ϖ_E are obtained using the covariance method in the least-squares fit (see Materials and Methods for full details)

Measuring long-range ¹³C–¹³C scalar couplings

The second application involves the measurement of three-bond ¹³C^δ–¹³C^α scalar couplings¹³, ³J_CαCδ, reporting on the conformational sampling of the side-chain χ₂ dihedral angle on a timescale from nanoseconds to milliseconds. The three-bond scalar coupling ³J_CαCδ relates to the side-chain χ₂ dihedral angle via a Karplus relationship, where large coupling constants are observed when the χ₂ angle is in a trans conformation, and small couplings are observed when χ₂ is in a gauche conformation. Intermediate values of the coupling constant are observed when the side chain is dynamic about the χ₂ dihedral angle. Extending the ¹³C–¹³C correlation experiment to measure these couplings (see Supplementary Fig. 10) showed that the majority of the arginine side chains in T4L L99A are predominantly in a trans conformation around the χ₂ angle (see Supplementary Table 3) in agreement with crystallographic data. For the two arginine side chains of T4L L99A shown in Fig. 5a, the intermediate value of the coupling constant for R14 shows that this side chain is flexible and dynamic around the χ₂ angle, while the high value observed for the R148 side chain shows a rigid trans conformation. This is in good agreement with previous characterisations of the dynamics of the arginine side chains of T4L L99A^6,31,32, as well as with the structure of T4L L99A, where R14 is on the surface and R148 is engaged in a salt bridge with D10. Long-range ¹³C^δ–¹³C^α scalar couplings were also measured for isoleucine residues in the 82 -kDa MSG protein (Fig. 5b; Supplementary Table 4). Of the 44 isoleucine residues in MSG, ¹³C^δ–¹³C^α scalar couplings were obtained for 32, allowing a quantification of the dynamics of these side chains. Access to both the ¹³C^δ1 chemical shift and the ¹³C^δ1–¹³C^α scalar coupling allows a full characterisation of the rotameric sampling about the χ₂ angle. For example, a large ¹³C^δ1 chemical shift and a small ¹³C^δ1–¹³C^α coupling for I200 in MSG show that this residue is restrained in a rare gauche-plus state (see Supplementary Table 4)²⁵, in agreement with the crystal structure of MSG.

Fig. 5.

Quantifying long-range scalar ¹³C–¹³C couplings. a Two examples of strips used to quantify long-range ¹³C^δ–¹³C^α scalar couplings in the arginine side chains of T4L L99A at 18.8 T and 278 K. The R14 side chain is flexible, while the R148 is rigid in agreement with relaxation measurements⁶. b Strips from a three-dimensional experiment to quantify long-range ¹³C^δ1–¹³C^α scalar couplings in isoleucine side chains of the 82-kDa MSG at 14.1 T and 310 K. The I238 side chain is dynamic around the χ₂ angle, as seen by the intermediate value of the ³J_CαCδ1 coupling constant, whereas I42 is more rigid in a trans conformation. Positive contours are depicted as red, while negative contours are in blue. Values for ³J_CαCδ1 in a and b are mean ± s.d.

A strategy to characterise side chains in proteins using ¹³C-direct-detected NMR has been developed. Applications of ¹³C-detection to macromolecules has so far mainly been employed to investigate intrinsically disordered regions^18,22, where many experiments are now available to exploit the improved resolution of ¹³C compared with ¹H. Existing methods to characterise side chains in medium-to-large proteins have been limited to methyl-bearing residues and require specific labelling. The presented method employs ¹³C-detection and capitalises on the distinct chemical shift profiles of side chains to specifically observe different residues in a single uniformly labelled sample. As such, it allows a more general characterisation of side chains for investigations in large proteins (~82 kDa) and for a quantification of the dynamics of these side chains. The slow transverse ¹³C relaxations rates in fully deuterated proteins enables a large range of NMR experiments to characterise side-chain structure, interactions and dynamics. A few examples of experiments integrated with the side-chain ¹³C–¹³C-direct-detected method are detailed above. Combined with recent developments of hardware aimed at ¹³C-direct-detected NMR spectroscopy, we envisage that the new method will be particularly useful for experimentally quantifying functional side chains, at atomic resolution, in medium-to-large proteins. Forthcoming applications of this suite of methods promise a burgeoning appreciation of the role side chains play in orchestrating protein function.

Methods

Sample preparations

Isotopically labelled T4 lysozyme mutant C54T/C97A/L99A (T4L L99A) was expressed and purified as described previously³³. Briefly, a codon-optimised form of the gene containing the mutations L99A, C54A and C97A in a kanamycin-resistant, pET-29b vector was transformed into BL21 (DE3) chemically competent E. coli cells. A single colony was used to inoculate 5-mL culture, which was grown overnight at 37 °C. This was used to inoculate 50 mL of minimal M9 media made with ²H₂O and supplemented with 1 g L⁻¹ [¹H,¹⁵N]-ammonium chloride as the sole nitrogen source. For the uniformly labelled [²H,¹³C,¹⁵N] sample, 2 g L⁻¹ [²H, ¹³C]-glucose was added as the sole carbon source. For the [¹H¹³C-Lys,Arg; ²H¹²C] isotopically labelled sample, 2 g L⁻¹ [²H, ¹²C]-glucose was added to the media, and 0.15 g L⁻¹ [¹H¹³C]-L-lysine, and 0.15 g L⁻¹ [¹H,¹³C]-L-arginine was later added 1 h prior to induction. This culture was grown overnight at 37 °C and used to inoculate a final 2-L culture. The final culture was grown to OD₆₀₀ ≈ 0.6 before a 16-h induction with 1 mM IPTG at 18 °C. The T4L L99A protein was initially purified by ion-exchange chromatography. After lysis by sonication in (50 mM NaPO₄ (pH 6.5), 2 mM EDTA, 5 mg of DNAse1 (Sigma) and 1 cOmplete^TM Mini Protease Inhibitor Cocktail tablet (Sigma) per 50-mL lysate), the soluble fraction was loaded onto a 5-mL HiTrap SP Sepharose FF column (GE Healthcare) (50 mM NaPO₄ (pH 6.5), 2 mM EDTA). Protein was eluted from the column using a gradient of 1 M NaCl and T4L99A eluted at ~300 mM NaCl. Pooled fractions were further purified by size-exclusion chromatography using a Superdex S75 gel filtration column (GE Healthcare) (50 mM NaPO₄ (pH 5.5), 2 mM EDTA and 25 mM NaCl). Pooled fractions were buffer exchanged into the final NMR buffer (50 mM NaPO₄ (pH 5.5), 2 mM EDTA, 25 mM NaCl and 2 mM NaN₃). The NMR sample contained 1.4 mM protein in 95%/5% ¹H₂O/²H₂O for the uniformly [²H¹³C¹⁵N] labelled sample and 1.5 mM protein in 100% ²H₂O for the [¹H¹³C-Lys,Arg; ²H¹²C] labelled sample.

Isotopically labelled MSG was produced as described previously²⁶. The gene with a C-terminal hexahistidine tag in an kanamycin-resistant pET-28a vector was transformed into BL21 (DE3) chemically competent E. coli cells. A single colony was used to inoculate 5-mL culture, which was grown overnight at 37 °C. This was used to inoculate 50 mL of M9 minimal media made with ²H₂O and supplemented with 1 g L⁻¹ [¹H,¹⁵N]-ammonium chloride and 2 g L⁻¹ of [²H,¹³C]-glucose as the sole nitrogen and carbon sources, respectively. The pre-culture was used to inoculate 1 L of M9 media, which was grown at 37 °C to OD₆₀₀≈0.45 before a 1-in-2 dilution to make the final 2-L culture volume. This final culture was grown to OD₆₀₀ ≈ 0.45 before a > 16-h induction with 1 mM IPTG at 16 °C. The MSG protein was initially purified by affinity chromatography. After lysis by sonication (20 mM Tris, 300 mM NaCl, 20 mM 2-mercaptoethanol, 10 mg of DNase1 (Sigma), 10 mg of hen egg lysozyme (Sigma) and 1 cOmplete^TM Mini Protease Inhibitor Cocktail tablet (Sigma) per 50-mL lysate), the soluble fraction was loaded onto a HisTrap 5-mL HP column (GE Healthcare) (20 mM Tris, 300 mM NaCl and 20 mM 2-mercaptoethanol, pH 7.8). Protein was eluted from the column using a gradient of 250 mM imidazole in the same buffer and the MSG protein eluted at ~90 mM imidazole. Protein in pooled fractions was unfolded (20 mM Tris, 100 mM NaCl, 10 mM 2-mercaptoethanol and 6 M guanidium chloride, pH 7.8) for 1 h at room temperature to allow full exchange of amide protons. Protein was refolded by rapid dilution (~1:50) into ice-cold refolding buffer (20 mM Tris, 100 mM NaCl, 10 mM 2-mercaptoethanol, 5 mM MgSO₄, 10% sucrose and 6 cOmplete^TM Mini Protease Inhibitor Cocktail tablets (Sigma) per litre buffer, pH 7.8). The protein was incubated, stirring at room temperature for 2 h before purification by affinity chromatography as described above. The pooled fractions were further purified by size-exclusion chromatography using a Superdex 200 5/150 gel filtration column (GE Healthcare) (20 mM NaPO₄, 5 mM dithiothreitol) and the MSG eluted at ~75 mL. Overall, 20 mM MgCl₂, 0.05% NaN₃ and 10% ²H₂O was added to the samples to make up the final NMR buffer.

Database extractions

The ¹³C chemical shift distributions for arginine, isoleucine, valine, threonine, lysine and proline shown in Supplementary Fig. 1 were extracted from the BMRB database²⁰ using the PACSY database³⁴ and the provided API (https://github.com/uwbmrb/BMRB-API), which was embedded in an in-house written python script.

NMR spectroscopy

All ¹³C-detected experiments were performed on a Bruker Avance II 600-MHz spectrometer using a ¹³C-optimised TXO coldprobe or a Bruker Avance HD 800-MHz spectrometer using an HCN inverse TCI coldprobe equipped with cooled ¹H and ¹³C preamplifiers.

NMR experiments on T4L L99A

The two-dimensional ¹³C–¹³C correlation spectra in Fig. 2c, Supplementary Figs. 3 and 4 were all recorded at a static magnetic field of 14.1 T and a temperature of 278 K. The pulse sequence used was the one shown in Fig. 2a and Supplementary Fig. 2. Deuterium ²H decoupling was achieved using a constant-wave (CW) field applied at 1.1 kHz, which in our hands gave significantly better decoupling than using composite decoupling schemes, e.g. WALTZ16³⁵. During the INEPT transfers, the ²H decoupling field was centred on the frequency of the ²H bound to ¹³C_t, while for indirect chemical shift evolution, the ²H decoupling field was centred on the frequency of the ²H bound to ¹³C_p. For example, for arginine ¹³C^δ–¹³C^γ, the ²H decoupling field was centred at 1.55 ppm (²H^γ) during the indirect chemical shift evolution and at 3.10 ppm elsewhere. Frequency-selective 90° (180°) pulses were applied with RE-BURP (E-BURP)³⁶ shapes. Spectra shown to the left (right) in Supplementary Figs. 2 and 3 were recorded with 10 (33) complex points in the indirect dimension, 4 scans per transient and a recovery delay of 4 s, leading to a total acquisition time of 12 min (38 min) per spectrum.

The ¹³C-detected CEST experiments were recorded at 14.1 T using the pulse sequence shown in Supplementary Fig. 9. CEST experiments to characterise valine residues were recorded with 22 complex points in the ¹³C^β dimension (sweep width of 800 Hz) and 50 CEST saturation points between −600 and 600 Hz. Eight scans were recorded per transient and a recovery delay of 4 s was used, leading to a total acquisition time of 40 h. CEST experiments to characterise arginine and threonine residues were recorded with 30 complex points in the ¹³C^γ dimension (sweep width of 1136 Hz) and 42 CEST saturation points between −300 and 300 Hz. Sixteen scans were recorded per transient and a recovery delay of 4.4 s was used, leading to a total acquisition time of 98 h. The CEST field, ω_CEST, and its inhomogeneity was obtained, as described by Guenneugues et al.³⁷, using a 20 mM sample of [U-²H,¹³C]-isoleucine. Uncertainties in I/I₀ were estimated from duplicate measurements.

The 3D long-range ¹³C–¹³C scalar coupling experiment shown in Fig. 5a was recorded at 18.8 T and obtained using the pulse sequence shown in Supplementary Fig. 11. The spectrum was recorded with 42 complex points in the indirect ¹³C^γ1 constant-time period (t₂), 24 complex points in the indirect t₁ period, 8 scans per transient and a recovery delay of 4 s, leading to a total acquisition time of 75 h. Further experimental details are given in the legend of Supplementary Fig. 11.

NMR experiments on MSG

The two-dimensional ¹³C–¹³C correlation spectra in Fig. 1e and S8 were recorded at a static magnetic field of 18.8 T and a temperature of 310 K. The pulse sequence used was the one shown in Fig. 2a and Supplementary Fig. 2. The two spectra in Fig. 3 were recorded with 43 complex points in the indirect constant-time dimension, 16 scans per transient and a recovery delay of 10 s, leading to a total acquisition time of 7.5 h per spectrum.

The 3D ¹³C–¹³C-TOCSY spectrum shown in Fig. 3b was recorded using the pulse sequence described in Supplementary Fig. 8. The spectrum was recorded with 54 complex points in the indirect ¹³C^γ1 constant-time period (t₂), 24 complex points in the indirect t₁ period, 4 scans per transient and a recovery delay of 10 s, leading to a total acquisition time of 120 h. Further experimental details are given in the legend to Supplementary Fig. 8.

The 3D long-range ¹³C–¹³C scalar coupling experiment shown in Fig. 5b was obtained using the pulse sequence shown in Supplementary Fig. 11. The spectrum was recorded with 42 complex points in the indirect ¹³C^γ1 constant-time period (t₂), 27 complex points in the indirect t₁ period, 8 scans per transient and a recovery delay of 7.8 s, leading to a total acquisition time of 160 h. Further experimental details are given in the legend to Supplementary Fig. 11.

Chemical shift assignments

The side-chain chemical shift assignment of T4L L99A was obtained based on the assignments published previously^6,11 combined with ¹H and ¹³C-detected CC-TOCSY experiments and the ¹H and ¹³C-detected long-range ³J_CαCδ experiments. The stereospecific assignment of the valine ¹³C^γ was taken from ref. ³⁸.

The chemical shift assignment of the MSG valine ¹³C^γ–¹³C^β correlation spectrum was transferred from a previously published assignment^26,27, that in turn was based on ¹H-detected experiments. The chemical shift assignment of the isoleucine ¹³C^δ1–¹³C^γ1 spectrum was obtained by a combination of a previous assignment^26,27 and the ¹³C-detected TOCSY experiment (Fig. 3b; Supplementary Fig. 8) and the ¹³C-detected long-range ³J_CαCδ experiment (Fig. 5b; Supplementary Fig. 11).

Data analysis

All NMR spectra were processed using nmrPipe³⁹ and initially analysed using NMRFAM-Sparky⁴⁰ or CCPN⁴¹. Peak intensities in experiments reporting on relaxation rates and CEST, Fig. 4 and Supplementary Figs. 5, 7, 9 and 10, were obtained using FuDA^42,43.

Long-range scalar couplings were derived by first obtaining the peak heights of the peaks corresponding to the diagonal peak, I_d = I(¹³C^δ–¹³C^γ–¹³C^δ) and the cross-peak I_c = I(¹³C^α–¹³C^γ–¹³C^δ) using the inbuilt tools of NMRFAM-Sparky and CCPN. The long-range scalar coupling was then calculated according to⁴ I_c/I_d = tan²(2πJ_CCT_J), where J_CC is the three-bond scalar coupling and T_J is the coupling evolution delay in the pulse sequence in Supplementary Fig. 11. As described previously¹³, passive couplings cancel by taking the ratio of I_c and I_d, such that this ratio only reports on the long-range coupling. For the ¹H-detected experiments, the error was determined as the root-mean-square deviation of two experiments. For the other long-range scalar coupling experiments, errors in the obtained I_c and I_d were estimated from RMSD of the spectral region, where no peaks were observed. The uncertainty of the calculated ³J_CC was estimated using a Monte-Carlo procedure to propagate the errors from the intensities.

Carbon ¹³C-detected CEST experiments were analysed using in-house written software in python using the LMFIT⁴⁴ module for least-squares minimisation of the target function

$${\chi }^2\left(\mathbf{p}\right)=\sum _i{\left({Ĩ}_{\mathrm{obs},i}-{Ĩ}_{\mathrm{calc},i}\left(\mathbf{p}\right)\right)}^2∕{\sigma }^2$$ 1

where ${Ĩ}_{\mathrm{obs},i}$ and σ are the experimentally obtained normalised intensities $\left(Ĩ=I∕I_0\right)$ and their uncertainty, respectively. The sum is over the different CEST saturation points and CEST field strengths, ω_CEST. ${Ĩ}_{\mathrm{calc},i}$ are the calculated normalised intentities calculated as a function of the model parameters p. Briefly, calculated intensities were obtained by evolving the spin system according to the Liouvillian described previously^14,19,45,46, also taking into account the one-bond ¹³C–¹³C scalar coupling, which generally leads to a simple line broadening of the CEST profiles. An inhomogeneity of the saturation field of 5% was used. The model parameters consist of the chemical shifts of the ground and excited states, ϖ_G and ϖ_E, respectively, the longitudinal relaxation rate that was assumed to be identical in the ground and excited state, R_1G = R_1E = R₁ and the transverse relaxation rate in the ground and excited states, R_1G, R_2E. For the analysis of R125, it was imposed that R_2G = R_2E, since the data did not contain enough information to determine R_2E; for other fits, both R_2G and R_2E were determined.

The overall exchange range, k_ex, and the population of the excited state, p_E, were treated as global parameters. Uncertainties of the obtained residue-specific parameters were estimated using the covariance method⁴⁷, as implemented in LMFIT. The uncertainties of k_ex and p_E were estimated by performing a grid search (see Supplementary Fig. 10).

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Author contributions

R.B.P. produced isotope-labelled samples. R.B.P. and D.F.H. designed, performed and analysed all experiments. R.B.P. and D.F.H. wrote the paper.

Data availability

The data that support the findings of this study and the pulse programmes (Bruker) are available from the corresponding author upon reasonable request. The source data underlying Figs. 1, 4 and Supplementary Figs. 1, 5, 7 and 9 are provided as a Source Data file.

Code availability

The pulse programmes (sequences) are available for Bruker spectrometers from the corresponding author upon request.

Competing interests

The authors declare no competing interests.

Supplementary information

Supplementary Information accompanies this paper at 10.1038/s41467-019-09743-4.