The Photocatalyzed Thiol‐ene reaction: A New Tag to Yield Fast, Selective and reversible Paramagnetic Tagging of Proteins

Abstract

Paramagnetic restraints have been used in biomolecular NMR for the last three decades to elucidate and refine biomolecular structures, but also to characterize protein‐ligand interactions. A common technique to generate such restraints in proteins, which do not naturally contain a (paramagnetic) metal, consists in the attachment to the protein of a lanthanide‐binding‐tag (LBT). In order to design such LBTs, it is important to consider the efficiency and stability of the conjugation, the geometry of the complex (conformational exchanges and coordination) and the chemical inertness of the ligand. Here we describe a photo‐catalyzed thiol‐ene reaction for the cysteine‐selective paramagnetic tagging of proteins. As a model, we designed an LBT with a vinyl‐pyridine moiety which was used to attach our tag to the protein GB1 in fast and irreversible fashion. Our tag T1 yields magnetic susceptibility tensors of significant size with different lanthanides and has been characterized using NMR and relaxometry measurements.

A fast and irreversible tagging of the GB1 T53C mutant, through the creation of a thioether bond between the tag alkene and the free thiol of the cysteine, is obtained using a photocatalyzed thiol–ene reaction. The paramagnetic effects on the protein NMR spectra are measured and analyzed, and the tag coordination is studied through relaxometry experiments.

1. Introduction

Nuclear magnetic resonance (NMR) is a powerful technique used for the study of biomacromolecules in solution. Indeed, it offers the opportunity to investigate structure, behavior, internal motions and mechanism of action of proteins, as well as their interactions with small molecules and other biomolecules at the atomic level.1, 2

Structure calculation by NMR mainly relies on the collection of short‐range distance restraints (up to ∼5–6 Å) provided by the time‐consuming and troublesome analysis of NOESY spectra. The use of long‐range paramagnetic distance restraints (up to ∼40 Å), such as pseudo‐contact shifts (PCS), induced by a paramagnetic ion, has been widely proposed to help in de novo structure determination by NMR3, 4, 5, 6, 7, 8 or in the refinement of pre‐existing X‐ray structures.9, 10, 11 In this way, a more reliable model describing the protein in solution can be obtained from the crystal structure .12, 13

Furthermore, PCS restraints, together with residual dipolar coupling (RDC) restraints that originate from the same paramagnetic susceptibility tensor, have been exploited in the analysis of the internal dynamics of multi‐domain proteins14, 15, 16, 17, 18, 19, 20, 21 and in the investigation of the interaction of proteins with their partners or ligands.22, 23, 24

One approach to incorporate paramagnetic metal ions into metallo‐proteins is via the exchange of the naturally occurring diamagnetic metal ion.25, 26, 27 As a more widely applicable alternative, any protein, not necessarily binding metal ions, can be made paramagnetic by attaching metal binding peptides or organic small synthetic ligands chelating paramagnetic metal ions.28, 29, 30, 31, 32, 33 The idea is to specifically attach the metal at a chosen position to monitor protein dynamics or protein‐protein and protein‐ligand interactions.34

To generate such restraints, one often relies on chemically synthesized lanthanide‐binding tags (LBTs).28 An important feature of LTBs is their ability to react quickly and selectively with a given amino acid in the protein. In addition, it is desirable that the metal coordination is rigid and avoids the presence of stereoisomers upon protein conjugation, which may complicate NMR data analysis. Apart from non‐natural amino acids, paramagnetic tags generally target cysteines. The standard strategy is the conjugation by formation of a disulfide bond: this technique has been widely used but it is hampered by the stability of the bond formed.35, 36 Thus, in recent years, much effort has been concentrated on the design of paramagnetic tags that are stable under reducing conditions once attached to the protein,37, 38 and which could also be used in cells.39 For this purpose, copper‐catalyzed azide−alkyne “click” cycloadditions, (which, however, require the introduction of non‐natural amino acids i. e. p‐azido‐phenylalanine40 in the protein sequence), and thiol‐ene coupling (TEC)41 (which relies on existing or mutated cysteine residues) have emerged as useful alternative tools in protein chemical conjugation. In particular, thiol‐ene coupling creates, through a radical mechanism, a stable thioether C−S bond between the free thiol of a cysteine and a double bonded carbon in an irreversible fashion. This reaction can be spontaneous but may also be photo‐catalyzed.42 There are a few examples of the use of this reaction for the paramagnetic tagging of proteins43 but the reactions are slow (overnight). Here, we describe a new vinyl‐pyridine‐based paramagnetic tag and its conjugation to proteins via the presented photo‐catalyzed thiol‐ene reaction to demonstrate that this reaction increases the applicability of paramagnetic tagging by making it suitable for proteins that are not stable for extended periods of time in vitro.

To design LBTs, it is important to consider the geometry of the complex and the chemical inertness of the ligand.44 ((2S,2′S,2′′S,2′′′S)‐1,1′,1′′,1′′′‐(1,4,7,10‐tetraazacyclododecane‐1,4,7,10‐tetrayl)tetrakis(propan‐2‐ol)), (S)‐THP has been proposed as a simple yet enantiomerically pure45 platform for the design of LBTs.46 Hence, we have designed a THP‐like LBT (T1) to be conjugated to proteins via TEC through a vinyl‐pyridine group, as the single point of attachment.

We have tested the new tag with GB1 as a model protein with a single engineered cysteine residue. The effect of the designed LBT, chelating different paramagnetic metal ions (Lu³⁺, Yb³⁺ and Dy³⁺), on the resonances of the protein has been investigated by solution NMR. Relaxometry measurements were also performed to obtain information on the lanthanide coordination sites in T1(Ln).

2. Results and Discussion

2.1. Synthesis of T1

In order to synthesize T1, a picolinic‐acid‐like intermediate bearing the double‐bond for conjugation and a leaving group for cyclen alkylation was needed. A was designated as a key intermediate in the synthesis. Derivative A was synthesized from commercial chelidamic acid (Figure 1). The carboxylic acids of chelidamic acid were protected as methyl esters and position 4 was brominated with tetrabutylammonium bromide (TBAB) and P₂O₅.

Figure 1.

Synthesis of intermediate A a. H₂SO₄, MeOH, 18 h, 98 %. b. P₂O₅, TBAB, Tol, 3 h, 97 %. c. NaBH₄, MeOH, 75 % d. trivinyl‐boroxin, K₃PO₄, JohnPhos, Pd(dba)₂, dioxane, 2 h, 68 % e. MsCl, DIPEA, DCM, 95 %

The reduction of a single methyl ester was easily achieved thanks to the peculiar reactivity of pyridine 2,6 diesters. The double bond in position 4 was then introduced via a Suzuki‐like coupling47 and the benzylic hydroxyl was subsequently replaced by a chloride as a good leaving group, giving A with a yield of 50 % calculated over 5 steps.

The intermediate A was then added to an excess of cyclen to isolate the mono functionalized cyclen derivative B (Figure 2). B was subsequently alkylated using (S)‐propylene oxide and the methyl ester was deprotected to yield C. Lanthanide‐chelation was achieved quantitatively, by refluxing overnight C in the presence of LnCl₃ salts in a H₂O/MeCN mixture.

Figure 2.

Synthesis of T1 a. DCM, 3 days b. (S)‐propylene oxide, MeOH, 4 days, quant. c. LiOH, THF/H₂O, 18 h, quant. d. LnCl₃ nH₂O, H₂O/MeCN

2.2. Conjugation to GB1 T53C

In order to test our LBT we chose the GB1 T53C mutant as a model system. GB1 is a small globular protein which is stable under many conditions and was engineered to bear one cysteine. Vinyl picolinic acids have been shown to react through TEC with cysteines overnight by Su et al.,43, 48 but this reaction is quite slow (overnight). To reduce the reaction time, we used a photo‐activated radical initiator, 2,2‐dimethoxy‐2‐phenylacetophenone (DPAP), which has been shown to catalyze TEC when activated with UV at 365 nm.

To a solution of GB1, 5 equiv. of tag and 5 equiv. of DPAP were added. The mixture was irradiated by a UV lamp at 365 nm for 1 hour after which the sample was analyzed by NMR. We used T1 loaded with Lu³⁺, Yb³⁺ and Dy³⁺. The protein tagging was evaluated recording 2D ¹H‐¹⁵N HSQC spectra which showed after 1 hour a conjugation of 80 % (Figure 3). The reaction yield was evaluated from the relative intensity of the signals of free and conjugated protein. In each 2D ¹H‐¹⁵N HSQC, only one set of new peaks was observed, indicating that no side‐reaction or protein degradation took place. We can assume that the reaction rate depends on the accessibility of the cysteine and on the stability of the corresponding thiyl species and may require protein‐specific optimization. We can indeed expect that buried cysteines might not react under these conditions. For example, attempts to functionalize another protein, the small N‐terminal domain of PEX1449, 50 were apparently hampered by the reduced solvent accessibility of the target cysteine on the protein surface and steric hindrance of the tag (see Supporting Information for more details). More radical initiator/UV wavelength couples could also be explored.

Figure 3.

Tagging of GB1 with T1 using the thiol‐ene reaction.

2.3. Evaluation of Paramagnetic Effects

The tagging of the T53C mono‐cysteine variant of the GB1 protein with paramagnetic [T1(Yb)] and [T1(Dy)] resulted in PCS in the 2D ¹H−¹⁵N HSQC spectrum (Figure 4). As expected, the two paramagnetic metal ions provide shifts in opposite directions.25 In order to differentiate between the chemical shift perturbation stemming from the tagging itself and the paramagnetic contribution to the shifts (PCS), we also tagged GB1 with the diamagnetic [T1(Lu)]. The NMR signals, which shifted after the addition of the diamagnetic compound correspond to residues located close to the tagging‐site (c.f. S.I. Figure S2). We note, that for each cross‐peak in the 2D ¹H−¹⁵N HSQC spectrum of diamagnetic samples only a single cross‐peak is observed for the paramagnetic species. We therefore conclude that T1 is indeed present as a single stereoisomer or a single conformation, giving rise to a single set of paramagnetically shifted peaks.

Figure 4.

2D ¹H−¹⁵N HSQC of GB1 T53C tagged with [T1(Lu)] (black), [T1(Yb)] (blue) and [T1(Dy)] (red), acquired at 700 MHz and 298 K. In the insert, in the top‐left corner, a zoom on residue G41 is shown. Interestingly, [T1(Yb)] and [T1(Dy)] provide shifts in the same directions for W43, because this residue is located at the interface between the negative and positive PCSs iso‐surfaces.

The PCS is described as a function of both the distance and orientation of the electron‐nuclear spin vector and the magnitude of the anisotropy tensor ($\Delta \chi$ ),51, 52 according to the equation

$${\displaystyle {\delta }^{\mathrm{p}\mathrm{c}}=\frac{1}{12\pi r^3}\left[{\Delta \chi }_{ax}\left(3{\mathrm{c}\mathrm{o}\mathrm{s}}^2\theta -1\right)+\frac{3}{2}{\Delta \chi }_{rh}{\mathrm{s}\mathrm{i}\mathrm{n}}^2\theta \mathrm{c}\mathrm{o}\mathrm{s}2\phi \right]}$$

where r, $\theta$ and $\phi$ are the spherical coordinates of the nucleus in the frame in which the anisotropy tensor is diagonal and has its origin at the metal position. ${\Delta \chi }_{ax}$ and ${\Delta \chi }_{rh}$ are the axial and rhombic components of the tensor, defined as

$${\displaystyle {\Delta \chi }_{ax}={\chi }_{zz}-\frac{{\chi }_{xx}+{\chi }_{yy}}{2}}$$

$${\displaystyle {\Delta \chi }_{rh}={\chi }_{xx}-{\chi }_{yy}}$$

The program FANTEN53 was used to obtain the best fit $\Delta \chi$ tensor (consisting of 5 parameters: ${\Delta \chi }_{ax}$ , ${\Delta \chi }_{rh}$ , the three Euler angles defining the frame in which the tensor is diagonal) and the distance of the metal ion to the nuclear spins observed.

First, the resonances of 19 signals in the spectra of the protein tagged with [T1(Yb)] were unambiguously assigned, and the PCS were evaluated. The anisotropy tensor for the Yb³⁺ tagged protein was determined from the best fit of these 19 H^N PCS to the X‐ray structure (PDB 1IGD),54 using the program FANTEN. During these tensor calculations, the position of the metal was obtained. New NMR signals could then be assigned, taking advantage of the PCS values predicted by the program for the other nuclei, so that a total of 36 H^N PCSs were obtained for the [T1(Yb)] tagged protein.

The assignment of the spectrum of the protein tagged with [T1(Dy)] was obtained by comparison with the assigned spectra of the [T1(Yb)] and [T1(Lu)] tagged proteins (Figure 4); in this way 32 PCS were obtained for [T1(Dy)].

The PCS originating from [T1(Yb)] and [T1(Dy)] were first analyzed separately (Figure S3). The metal positions obtained from the two sets of data were similar (the metal is found at 8.6 and 8.1 Å, from the Cβ of the cysteine residue, for [T1(Dy)] and [T1(Yb)], respectively), as well as the direction of the main axes of the best fit tensors (Figure 5A and 5B).

Figure 5.

Graphical representation of the magnetic susceptibility anisotropy tensor orientations. The x, y and z axes (corresponding in turn to the directions with the smallest, intermediate and largest magnetic susceptibility) are represented as red, yellow and blue arrows, respectively. The panels A and C show the orientation of Yb tensors when only its own PCS dataset was considered, and when the datasets of Yb and Dy were both taken into account in FANTEN, respectively. The panels B and D show the orientation of the Dy tensors when only its PCS‐data‐set was considered, and when data sets of Yb and Dy where both taken into account in FANTEN, respectively.

Then, the two sets of PCSs were evaluated jointly by constraining both metals to reside in the same position. The final position of the metal, obtained by FANTEN (8.3 Å from the Cβ of the cysteine residue), is very close to the positions found from the sets of data analyzed separately, and compatible with the expected covalent structure of the tag. The orientations and magnitudes of the two tensors were almost unaffected with respect to the values obtained from the separate fits, thus showing high consistency between the two sets of paramagnetic data (Table 1 and Figure 5). The two tensors are almost coaxial, with angles of 11.5° between the two Z‐axes, 5° between the two X‐axes and 12° between the two Y‐axes.

Table 1.

Tensor parameters calculated with the program FANTEN using the PCS values measured with [T1(Yb)] and [T1(Dy)] implemented separately or jointly in the evaluation of the metal position.

Metal	PCS source	Q factor	Δχax (10−32 m3)	Δχrh (10−32 m3)
Yb	Yb	0.061	1.91±0.02	0.68±0.25
Dy	Dy	0.072	−6.51±0.02	2.54±0.89
Yb	Yb & Dy	0.064	1.83±0.03	−0.59±0.28
Dy	Yb & Dy	0.072	−6.81±0.21	3.06±0.89

The agreement between experimental and back‐calculated PCS obtained from the fit of both sets of data is very good, with Q factors of 0.064 and 0.072, for [T1(Yb)] and [T1(Dy)], respectively (Figure S4). The PCS iso‐surfaces for the two metals are represented in Figure 6.

Figure 6.

Graphical representation of PCS iso‐surfaces of 1 (blue) and −1 (red) ppm obtained for [T1(Yb)] (A) and [T1(Dy)] (B) using the program FANTEN.

It has been shown that some LBTs can be immobilized on the protein surface by an electrostatic interaction between the lanthanide cage and a carboxylate belonging to an Asp or Glu residue.43, 55 This interaction may decrease the tag mobility and therefore increase the magnitude of the magnetic susceptibility anisotropies. In the case of GB1 T53C, there are two carboxylates at distances of 9 and 11 Å from the cysteine (E42, E56), respectively. We therefore expected to see such a stabilizing interaction taking place between our tag and our protein. However, we find that the axial anisotropies of the two tensors are 4–5‐fold reduced compared to the values expected for the anisotropies of the magnetic susceptibility tensors of the Yb³⁺ and Dy³⁺ ions usually calculated from rigid systems containing these paramagnetic ions.56, 57 This likely indicates that some motional averaging of the magnetic susceptibility tensors occurred, so that the best fit tensors result from the averaging of the magnetic susceptibility anisotropy tensors due to tag mobility.

Interestingly, the position of the metal calculated with FANTEN is relatively far from the protein surface. This observation corroborates the initial design of the tag to not interact with the protein surface, thus being very mobile, in agreement with the small tensor anisotropies observed. We suppose that in T1 the lanthanide ions have a coordination number of 9 and have no further coordination site with which to interact with the carboxylate positioned on the protein surface.

The sizable mobility of the tag may be attributed to the two rotatable bonds, after the pyridine moiety, that tether the tag to the protein surface, which are absent in similar tags reported in the literature (such as DO3MA58). An additional methyl group on the pyridine moiety (i. e. in position 3 or 5) could have possibly restrained the rearrangement of the pyridine and stabilize the tag. Moreover, a different attachment site on the protein surface could have affected differently the tensor parameter values because of different possible protein/tag interaction (steric hindrance and/or hydrophobic interactions) which might rigidify the tag.

2.4. Relaxometry

To characterize the coordination of the lanthanides in T1, Fast Field Cycling (FFC) relaxometry experiments were performed.59, 60, 61, 62 The ¹H nuclear magnetic relaxation dispersion (NMRD) profiles of the [T1(Gd)] complex in water solution at 10, 25 and 37 °C are shown in Figure 7. The profiles are characterized by dispersions somewhat smoother than predicted by the Lorentzian spectral density function; however, any attempt to reproduce them by considering inner sphere and outer‐sphere contributions failed (with the diffusion coefficients constrained to values in the range expected for water solutions) unless the distance of closest approach between paramagnetic ion and diffusive water molecules was larger than 8 Å. Therefore, no sizable contribution from outer‐sphere relaxation is apparent, and the profiles were fit by including two protons at a distance r ₁ and other two protons at a distance r ₂ from the Gd³⁺ ion (r ₁ and r ₂ were left free to be adjusted in the best fit analysis).

Figure 7.

NMRD Profiles of [T1(Gd)] measured at different temperatures.

The best fit parameters are reported in Table 2 and the corresponding profiles are shown in Figure 7. Due to the low sensitivity to the correlation time τ_v for electron relaxation, the values of τ_v were kept fixed to values typically observed in gadolinium complexes.63, 64 Although there exists some covariance among the different parameters, the analysis indicates that the lifetime of the two protons at r ₁=3.05 Å, τ _M(1), is as long as several microseconds (ca. 4 μs at 25 °C). The analysis also shows contributions from fast exchanging second‐sphere water molecules, with a lifetime on the picosecond timescale. The number of these second‐sphere water molecules is totally covariant with the metal‐proton distance r ₂; if one water molecule is considered, r ₂ is about 3.5 Å.

Table 2.

Best fit parameters obtained from the NMRD profiles of [T1(Gd)], shown in Figure 7. Contributions from two protons in the first‐coordination and second‐coordination spheres were considered.

	37 °C	25 °C	10 °C
τR [ps]	35	57	112
Δt [cm‐1]	0.023
τv [ps]	15	20	25
r 1 [Å]	3.05
τ M(1) [μs]	3.4	3.8	4.6
r 2 [Å]	3.48
τ M(2) [ps]	8.8	18	51

Fits of equivalent quality, however, can be obtained also for increasing values of r ₁ (and decreasing values of r ₂). When the condition r ₁=r ₂=3.3 Å is met, almost indistinguishable best fit profiles are calculated, the complex reorientation time τ_R, the transient ZFS Δ_t and the lifetimes τ _M(1) and τ _M(2) being somewhat increased (τ_R=84 ps, Δ_t=0.027 cm⁻¹, τ _M(1)=5.3 μs and τ _M(2)=28 ps, at 25 °C).

The fast exchanging second‐sphere water protons may be related to one or more water molecules hydrogen bonded in positions allowing for a large mobility.

The presence of two protons exchanging in the microsecond time scale at 3.05 Å, i. e. at the distance expected for the protons of a water molecule regularly coordinated to the Gd³⁺ ion, or at a somewhat larger distance (such as 3.3 Å), could be related to one water molecule hydrogen‐bonded to the oxygen atoms of hydroxyl groups or of the carboxylate group. Its lifetime is much longer (several microseconds) than the lifetime of the water molecule coordinated to the Gd³⁺ ion in DO3A‐like or DOTA‐like complexes (on the sub‐microsecond timescale), likely because of the bipositive charge of this complex which disfavours water exchange with respect to neutral or negative complexes.65, 66 A long lifetime of this water molecule indicates a relatively “compact” and stable environment surrounding the Gd³⁺ ion, which may prevent the replacement of the coordinated water by negatively charged groups present on the protein surface. This is in agreement with the observed tag mobility that affects the magnitude of the PCS‐determined tensor. Alternatively, these slowly exchanging protons may be the hydrogen atoms of the three hydroxyl groups coordinated to the Gd³⁺ ion.

3. Conclusions

In this work, we present the synthesis and the performance of an enantiopure paramagnetic tag for NMR spectroscopy, designed to efficiently conjugate proteins under mild conditions using UV irradiation. In particular, we have demonstrated that UV‐catalyzed TEC is applicable to the paramagnetic tagging of GB1 T53C with a high yield and short reaction times. The creation of the stable thioether bond required much shorter reactions times than previously described methods.43, 67

The conjugation involves the protein's cysteine side‐chain and forms thioether bonds without affecting the paramagnetic properties of the tag. Being easily obtained, and in high yields, the formation of non‐reducible thioether bonds is convenient and makes this strategy a method of choice in the field of paramagnetic tagging, opening the way to development of more novel and rigid paramagnetic tags. Those tags could be applied to the study of larger systems, where the PCSs analysis could be complemented by the use of residual dipolar couplings and paramagnetic relaxation enhancements measurements.68, 69

Experimental Section

Organic Synthesis

All the reagents were purchased from Sigma‐Alrich, except cyclen which was purchased from Chematech. Small‐molecules NMR experiments were recorded on a Bruker AVANCE II 500 MHz (¹H Larmor frequency) at 298 K. HPLC was performed with an Agilent 1200 Sries with ZORBAX 300SB‐18 analytica and semi‐preparative. Mass spectrometry was performed on a Thermo‐Fisher LTQ‐XL ESI.

Formation of Lanthanide Complexes

C (70 mg) and LnCl₃ nH₂O (100 mg) were dissolved in 6 mL H₂O/MeCN (50/50). The pH was adjusted to 7 and the mixture was refluxed. Chelation was typically quantitative overnight, as shown by LC/MS analysis. Reaction mixtures were purified via semi‐preparative HPLC.

Protein Conjugation

GB1 T53C was expressed and purified as reported in the supplementary information. Prior to conjugation, the protein was buffer‐exchanged into NaPi 20 mM, pH 7.5, then concentrated to 180 μM. 5 equivalents each of DPAP and [T1(Ln)] were added to the protein, as well as 10 % D2O. The mixture was transferred into a 5‐mm NMR tube and argon was gently bubbled through the solution for 5 minutes. The tube was placed under a UV Lamp (UVGL‐55 Mineralight 26 W) at 365 nm for 1 h, after which the sample was measured using NMR spectroscopy and the spectra analyzed. Excess of small molecules were then washed away by buffer exchange and NMR spectra were recorded again.

NMR Measurements and PCS Analysis

All the experiments were acquired on a Bruker AVANCE NEO NMR spectrometer operating at 700 MHz (¹H Larmor frequency) at 298 K equipped with a 5 mm TCI 3 channels HCN cryo‐probehead. All the spectra were processed with the Bruker TopSpin 4.0.7 software package and analysed with the program CARA70 (ETH Zürich).

The spectra were collected using a protein concentration of ∼180 μM in buffered solution (20 mM sodium phosphate pH 7.5).

The assignment of GB1 was taken from the literature.71

The PCS values were calculated from the difference in the value of chemical shift of each amino acid peak between the paramagnetic [T1(Yb) or T1(Dy)] and diamagnetic [T1(Lu)] 2D ¹H‐¹⁵N HSQC spectra acquired.

The fitting of the PCS tensor was carried out using the program FANTEN.53

Relaxometry Measuremnts

¹H nuclear magnetic relaxation dispersion (NMRD) profiles of the [T1(Gd)] complex in water solution were obtained by measuring the water proton relaxivity as a function of the applied magnetic field at the temperature of 10, 25 and 37 °C. The profiles were measured with a SPINMASTER2000 fast field cycling relaxometer (Stelar, Mede (PV), Italy) operating in the 0.01–40 MHz ¹H Larmor frequency range. The measurements are affected by an error below 1%, when fitted to a monoexponential decay/recovery of the magnetization in the field cycling experiment.

Conflict of interest

The authors declare no conflict of interest.

Supporting information

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials are peer reviewed and may be re‐organized for online delivery, but are not copy‐edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors.

Supplementary

M. Denis, C. Softley, S. Giuntini, M. Gentili, E. Ravera, G. Parigi, M. Fragai, G. Popowicz, M. Sattler, C. Luchinat, L. Cerofolini, C. Nativi, ChemPhysChem 2020, 21, 863.