Residue‐Specific Signatures of Structural Water Identified by Dissolution Dynamic Nuclear Polarization with UV‐Generated Radicals

ABSTRACT

Conserved structural water molecules stabilize protein folds and modulate their function, yet remain difficult to observe in solution because exchange‐based readouts favor solvent‐exposed sites. Here, we introduce a protocol to detect structural water molecules under native conditions, using long‐lived hyperpolarized water (HyperW) enabled by UV‐induced, nonpersistent radicals. In the model protein chymotrypsin inhibitor 2, HyperW‐enhanced two‐dimensional NMR correlation spectra across pD 5.5–8.4 reveal strong exchange‐driven enhancements at solvent‐exposed residues. By contrast, a distinct group of four residues shows hyperpolarized amide signals, which disappear when through‐space polarization transfer via nuclear Overhauser effect (NOE) is suppressed using a CLEANEX‐PM experiment. The CLEANEX‐negative/HyperW‐positive signature, together with the spatial proximity of these residues to crystallographically conserved water molecules, supports NOE‐mediated transfer from long‐residence internal water, not distinguishable by standard NMR methods. The combined observables establish HyperW NMR as a residue‐specific reporter of structural hydration and hydration‐coupled dynamics under native conditions, providing a route to link conserved water observed in crystals to their roles in solution.

Structural water molecules in proteins are notoriously hard to detect in solution. Hyperpolarized water with UV radicals as polarizing agents provides the sensitivity and resolution to detect residue‐specific signatures of structural water with NMR spectroscopy.

Water molecules are integral components of proteins, shaping their structure and function [1]. They stabilize secondary and tertiary structure [2, 3, 4] and participate in electron [5, 6] and proton [7, 8, 9] transfer reactions. While solid‐state methods like X‐ray crystallography and, more recently, cryo‐electron microscopy can localize water molecules at high resolution [10, 11], identifying crystallographically conserved and internally bound structural water in liquid solution remains challenging [12]. Nuclear magnetic resonance (NMR) is widely used to study biomolecules in solution, using flexible pulse‐sequence design to probe water–protein interactions. Previous efforts to detect structural water molecules by NMR have focused on nuclear Overhauser spectroscopy (NOESY and ROESY) [13, 14, 15] and magnetic relaxation dispersion [16]. While seminal NOESY/ROESY studies established that the sign and behavior of cross‐relaxation can be used to distinguish dipolar transfer through space from chemical exchange in protein hydration experiments (e.g., BPTI and related systems), practical application can remain challenging: weak or transient hydration contacts are often sensitivity‐limited, and exchange contributions can mask residue selectivity in rapid multidimensional readouts under native‐like conditions. The key limitations of NMR methods are their intrinsically low sensitivity and the difficulty of unambiguously discriminating between structural, surface, or bulk water [17, 18]. Hyperpolarized water (HyperW) produced by dissolution dynamic nuclear polarization (dDNP) yields liquid‐state proton polarization that makes the bulk water signal orders of magnitude stronger than in thermally polarized experiments [19, 20].

From this bulk reservoir, hyperpolarization is transferred to biomolecules via chemical and magnetic exchange mechanisms, enabling studies of ordered [21, 22] and intrinsically disordered [23, 24, 25] proteins, RNAs [26, 27], as well as folding processes and equilibria [28, 29]. Recently, small‐molecule mechanistic studies have focused on magnetization transfer via direct and exchange‐relayed NOEs [29, 30, 31].

Standard experimental practice for HyperW is the use of stable organic radicals as polarization agents and pressurized systems for ultrafast sample transfer between polarizer and detection spectrometer [32]. hyperpolarization decays rapidly in the liquid state because of radical‐driven paramagnetic relaxation. Moreover, high proton concentrations have challenged the resolution of HyperW‐enhanced NMR, especially in two‐ and three‐dimensional experiments [33]. A recent advance generates HyperW using UV‐induced, nonpersistent radicals formed in frozen pyruvic acid mixtures. These radicals are stable only at cryogenic temperatures and recombine to diamagnetic species upon dissolution above 190 K, generating a radical‐free liquid‐state sample after dissolution. The latter enables longer‐lived, liquid‐state polarization and simpler, even manual, transfers [34, 35, 36]. Here, we use and extend this concept to a robust, protein‐compatible workflow that delivers multidimensional HyperW NMR on a standard 400 MHz spectrometer, without pressurized transfer hardware, opening up native‐condition hydration readouts. Using UV‐HyperW‐enhanced SOFAST HMQC, we track the hydration dynamics of a model protein by systematically varying the pD of the hyperpolarized water, spanning exchange timescales from slow amide–water exchange at low pD to fast exchange at high pD. Comparing with thermally polarized experiments in deuterated buffers, we identify a subset of resonances that do not arise from chemical exchange, corroborated by CLEANEX‐PM NMR, and show how these signals result from magnetization transfer through structural water molecules. Rather than suggesting new mechanistic principles for separating exchange from through‐space transfer, we introduce a radical‐free HyperW workflow and a combined HyperW+CLEANEX residue signature that enables rapid multidimensional hydration readouts under the solution conditions used here.

Our UV‐HyperW approach using thermally unstable pyruvic acid radicals combines high polarization, long liquid‐state lifetimes, and a simplified workflow without pressurized transfers. The radicals, vital for solid‐state dynamic nuclear polarization, were generated by irradiating frozen beads ($10·6$ $\mu \mathrm{L}$) of 20% pyruvic acid (12 $\mu \mathrm{L}$), 30% deuterated glycerol (18 $\mu \mathrm{L}$, 99.8% D) and 50% water (30 $\mu \mathrm{L}$) at 280–450 nm (Figure 1a) [35, 36]. UV irradiation for 10 min produced radical concentrations >45 mM, allowing fast and efficient hyperpolarization to $\ge 65\%$ within 60 min in a 6.7 T magnetic field (Figure S1) [36]. The 60 $\mu \mathrm{L}$ DNP sample was rapidly dissolved by automatic injection of 4.5 mL deoxygenated (argon‐purged), deuterated buffer (99.9% D₂O) heated to ${170}^{\circ }\mathrm{C}$ to produce radical‐free, hyperpolarized water containing 0.7% HDO (360 mM protons). The UV‐HyperW was collected in a syringe and transferred manually to the 400 MHz NMR spectrometer, where it was injected into the NMR tube in 10 s. After this, the liquid‐state polarization of HDO was higher than 30% (enhancement of $>10000$) with relaxation times of more than 60 s at ${40}^{\circ }\mathrm{C}$ in neat HyperW (Figure S2). The high proton liquid‐state polarization was a consequence of the immediate recombination of the UV radicals into diamagnetic species at the moment of dissolution. Moreover, the absence of paramagnetic species in the hyperpolarized water led to long relaxation times and simplified the experimental demand, as it could be performed without the need for pressurized transfer systems with magnetic rails. HyperW experiments using stable nitroxide radicals rely on considerably larger DNP samples (4%–8% final HDO content) to compensate polarization losses due to paramagnetic relaxation, leading to decreased relaxation times. Such samples require specialized pressurized sample transfers for polarization preservation [31, 33, 37]. When our UV‐HyperW was injected into protein solutions, the $T_1$ time of HDO was between 30 and 40 s (Figure S3). The observed decrease and variability in relaxation times across protein samples were primarily due to differences in the molecular oxygen content of the samples, as it is a significant driver of $T_1$ relaxation in the absence of other paramagnetic impurities such as metal ions [38]. Throughout the whole workflow, particular emphasis was therefore placed on avoiding molecular oxygen by purging all collection, transfer, and detection vessels with argon. Elevated proton concentrations (1.5x–2x compared to neat HyperW) in the protein samples and fast chemical exchange with the protein may also have contributed to the decrease of $T_1$(HDO).

FIGURE 1.

Overview of the UV‐HyperW method. (a) Thermally unstable radicals were generated at liquid nitrogen temperatures in frozen solutions of pyruvic acid by UV irradiation. (b) Hyperpolarization setup: UV‐irradiated DNP samples (red) contained 20% pyruvic acid, 30% d⁸‐glycerol, and 50% H₂O with 50 mM radical concentrations after UV irradiation. Samples were polarized with 188 GHz microwave irradiation in a SpinAligner (left) with a field strength of 6.7 T operating at 1.2 K. After dissolution with deuterated buffer the hyperpolarized water (HyperW, green) was transferred to the NMR system (400 MHz, right) in a syringe and injected into the protein solution. The final NMR sample (blue) contained 200 $\mu \mathrm{M}$ chymotrypsin inhibitor 2 (CI2) in a 0.5% HDO and 99.5% D₂O buffer with a final proton concentration of 250 mM. All experiments were recorded at a temperature of 313 K.

The protein we investigated here was chymotrypsin inhibitor 2 (CI2), which serves as a well‐characterized model system in protein science, extensively used to study protein folding and stability [39, 40, 41]. Crystal structures of CI2 contain many water molecules, eight of which are conserved in wild‐type (wt) CI2 (Figure 2) and across different mutant variants, including a double mutant (dm, L49I/I57V) with a stabilized folded state [42, 43]. Four of the water molecules are situated close to the hydrophobic core of the protein, and prior NMR and computational studies suggest that they stabilize the protein's irregular core sheet and rigidify the inhibitory loop [44, 45, 46].

FIGURE 2.

(a) Crystal structure (PDB: 7a1h) of CI2 with conserved water molecules (red spheres) [43, 46]. (b) Possible hydrogen‐bond contacts (r(O–N) < 3.5 Å) with amide functionalities in CI2.

To probe residue‐resolved hydration in CI2, we injected 400 $\mu \mathrm{L}$ hyperpolarized water (0.7% HDO) into 100 $\mu \mathrm{L}$ solutions of uniformly and ‐labeled wt CI2 (SI 1.2) in deuterated buffers ranging from pD 5.5 to 8.4. The final protein solution had a concentration of 250 $\mu \mathrm{M}$ and a temperature of $40\pm 1^{\circ }\mathrm{C}$ (SI 1.5.1). After injection, we measured the polarization (Table S1), and applied a proton‐saturation sequence to eliminate hyperpolarized signals resulting from decay products of the pyruvic acid radicals in the amide spectral region. We then recorded – SOFAST HMQC experiments [47] with 100 ms scan delay (acquisition time + relaxation delay) in 24 s. The resulting spectra (Figures 3a–d and S4) revealed a subset of well‐resolved amide‐proton resonances selectively enhanced by the HyperW. These resonances had average linewidths of 22–30 Hz for and 33–45 Hz for compared to 20–23 and 28–30 Hz in thermally polarized SOFAST HMQC experiments and 20 and 27 Hz in reference FHSQC experiments, respectively (Figure S5). While the linewidth reflects relaxation and field inhomogeneity, the linewidth in the indirect dimension additionally includes broadening arising from the progressive change in starting HyperW polarization between $t_1$ increments during the 2D acquisition [33]. The conservation of resolution is a result of our robust UV‐HyperW method, in which turbulence and bubble formation are limited, paramagnetic relaxation does not contribute to the proton linewidths, and nitrogen signal broadening due to HyperW decay on the experimental timescale is minimal as a result of the long HyperW decay times of 30–40 s. When we performed SOFAST HMQC experiments with 500 ms scan delay (101 s) instead of 100 ms (24 s), the linewidth increased by a factor $>2$ (Figure S6), as HyperW decay became more relevant.

FIGURE 3.

Comparison of HyperW SOFAST HMQC experiments of wild‐type CI2 performed at pD (a) 5.5 in acetate buffer, (b) 6.3 in MES buffer, (c) 7.4 in HEPES buffer, and (d) 8.4 in TAPS buffer at 40. SOFAST HMQC experiments were recorded with 96 indirect t1 increments in 24 s with a two‐step phase cycle and 100 ms scan delay. Spectra were normalized to the maximum signal in the spectral region and a cutoff below 15 % intensity. (d–g) Mapping of HyperW‐enhanced amide resonances onto the CI2 crystal structure (PDB: 7a1h). The sphere radius represents the relative calculated solvent‐accessible surface area (h–k) for water (1.4 Å sphere) [48].

At pD 5.5 (Figure 3a), the strongest signals belonged to K2 (N‐terminus) and M40 (inhibitory loop), both highly solvent‐exposed as identified by solvent‐accessible surface area (SASA) calculations (Figure 3i) [49]. Additional signals arose from nearby T3 and G35, from K53, and from V60 and G64 at the C‐terminus. Notably, two core $\beta$‐sheet residues (R48 and F50) were also detected. Beyond backbone amide signals, three strong arginine side‐chain resonances were also detected (Figure S4). At pD 6.3 (Figure 3b), the majority of amide signals increased. New peaks appeared from T39, E41, and Y42, all neighbors of M40 in the inhibitory loop, and from R62 in the C‐terminal region, while the relative amplitude of K2 decreased. At pD 7.4 (Figure 3c), both trends continued: T39, E41, and Y42 intensified, and additional inhibitory loop residues I37 and V38 emerged. Conversely, M40 weakened substantially, K2 was no longer detectable, and signals from R48, F50, and G64 disappeared. New side‐chain resonances from asparagine and glutamine residues were also observed. At pD 8.4 (Figure 3d), V60 was no longer observed, while new resonances appeared from less accessible residues, including E7, S12, E14, and I29. The correlation between solvent accessibility and observed hyperpolarized residues, which is mediated by exchange, is consistent with HyperW experiments in other folded proteins [22, 28] and with standard H/D exchange experiments of CI2 [50].

After decay to thermal equilibrium, we remeasured the SOFAST HMQC experiments. In the 99% deuterated buffer, this experiment yielded signals only of very slow exchanging amide protons (Figure S7). The expected trend of decreasing signals at increasing pD resulted in only the three highly protected residues I20, V47, and L49 showing up at pD 8.4, consistent with the CI2 literature [51]. Only four specific amides were detectable at thermal equilibrium and enhanced in HyperW experiments: R48, F50, R62, and G64.

In conjunction with our HyperW experiments, we performed CLEANEX‐PM experiments [52]. CLEANEX‐PM detects proton signals arising solely from chemical exchange with water at thermal equilibrium, thus complementing HyperW. This is achieved through the application of phase‐modulated spinlocking, which suppresses through‐space polarization transfer via NOE during the mixing time. Figure 4 (and Figure S8) compares thermal (gray) and HyperW‐enhanced (orange/green) SOFAST HMQC with CLEANEX‐PM (purple) spectra using a mixing time of 100 ms at pD 5.5 (a) and 6.3 (e). The amide resonances R48, F50, R62 (weak at pD 5.5), and G64 were detected in both thermal and HyperW‐enhanced spectra but were absent in CLEANEX‐PM (up to 2 s mixing, Figure S9), excluding fast chemical exchange as their origin.

FIGURE 4.

Comparison of SOFAST HMQC in (a) deuterated acetate and (e) MES buffer recorded after injection of hyperpolarized water (orange/green) and after relaxation to thermal equilibrium (gray) to CLEANEX‐PM (purple). Residues R48, F50, R62, and G64 were enhanced in the HyperW experiment and also detected at thermal equilibrium, but not with the CLEANEX‐PM experiment.

To verify this pattern, we examined the L49I/I57V double mutant of CI2, in which two residues in close proximity of the group of four are modified to create synergistic stabilization of the protein. At pD 6.3, the same HyperW‐positive and CLEANEX‐negative signature was observed for these four amides despite shifted resonance positions (Figure S10). We also recorded 3D NOESY HSQC [53] and water‐selective (WS) NOESY experiments (Figure S11) [54] of wt and dm CI2 to compare our approach with prior efforts to detect structural water. The NOESY spectra contained significantly more signals than detected by HyperW, including all exchange‐dominated (CLEANEX‐positive) signals as well as the four highlighted amides (Figure S12). Exchange‐dominated resonances showed markedly increased build‐up rates in WS NOESY at higher pH, while nine resonances (I29, I30, L32, T36, R43, R48, F50, R62, and G64) showed unchanged signal build‐up, corroborating the NOE mechanism but not unambiguously distinguishing the HyperW‐positive/CLEANEX‐negative signals. In comparisons of HyperW‐enhanced SOFAST HMQC with CLEANEX‐PM experiments in other proteins [22, 28, 29], such signals have been attributed to exchange‐relayed NOE transfer through labile protons in the biomolecules.

Because thermal signals for this group of four residues were detectable in our deuterated samples, enhancement factors could be determined directly as $\epsilon =I_{\mathrm{HyperW}}/I_{\mathrm{thermal}}$. For a scan delay of 100 ms, the enhancements at pD 5.5 and pD 6.3 (given as $\epsilon (\mathrm{pD}5.5),\epsilon (\mathrm{pD}6.3)$) were: R48 (5, 48), F50 (6, 11), R62 (3, 17), and G64 (4, 16). At pD 5.5, increasing the scan delay to 500 ms kept the same residue pattern but increased $\epsilon$ to 8 (R48), 14 (F50), 4 (R62), and 8 (G64), respectively. Since water polarization remained stable across measurements (Table S1), these differences reflect the effects of exchange and NOE on the signals. To rationalize these findings, we considered four possible transfer pathways between protons in HyperW, structural water (SW), exchangeable protein moieties (XH) and backbone amide moieties (NH):

$$\begin{array}{c}\hfill {\mathrm{H}}_{\mathrm{HyperW}}\stackrel{\mathrm{NOE}}{\to }{\mathrm{H}}_{\mathrm{NH}}\end{array}$$ (1)

$$\begin{array}{c}\hfill {\mathrm{H}}_{\mathrm{HyperW}}\stackrel{k_{\mathrm{ex}}}{\to }{\mathrm{H}}_{\mathrm{XH}}\stackrel{\mathrm{NOE}}{\to }{\mathrm{H}}_{\mathrm{NH}}\end{array}$$ (2)

$$\begin{array}{c}\hfill {\mathrm{H}}_{\mathrm{HyperW}}\stackrel{k_{\mathrm{ex}}}{\to }{\mathrm{H}}_{\mathrm{SW}}\stackrel{\mathrm{NOE}}{\to }{\mathrm{H}}_{\mathrm{NH}}\end{array}$$ (3)

$${\mathrm{H}}_{\mathrm{HyperW}}\stackrel{\mathrm{NOE}}{\to }{\mathrm{H}}_{\mathrm{SW}}\stackrel{\mathrm{NOE}}{\to }{\mathrm{H}}_{\mathrm{NH}}$$ (4)

Cases (1) and (4) are unlikely: direct NOE from bulk water is inefficient due to the short correlation time of free water molecules [14, 55], and a two‐step NOE through structural water molecules would be even less efficient. Although significant direct NOE transfer across larger distances has been proposed in fully protonated environments [56], the dilution conditions in our experiments make this highly unlikely. Case (2) would require nearby labile side‐chain protons, but in CI2 the nearest such group (R48) lies at 4.4 Å, too distant for efficient transfer (Figure S13). By contrast, Case (3) is supported by structural evidence: all four amide groups are located within 3.5 Å of crystallographically conserved water molecules (Figure 2). Structural water molecules exchange with bulk on timescales from nanoseconds to hundreds of microseconds [57, 58], which is sufficient to mediate polarization transfer through a combination of exchange of the structural water molecules and efficient intramolecular NOE under our experimental conditions. The dependence of enhancement factors on the pD level, and therefore on the speed of exchange, supports this claim further. We therefore conclude that the most consistent mechanism is an exchange‐relayed NOE via structural water molecules.

Accordingly, HyperW provides highly sensitive, residue‐specific, solution‐state detection of structural water molecules in a folded protein. In CI2, a group of four amides proximal to conserved water molecules exhibits HyperW enhancement without CLEANEX‐PM detectability, consistent with NOE‐mediated transfer, while pD‐dependent enhancements at exposed sites reflect chemical exchange. Classic NOESY/ROESY studies established that dipolar transfer and exchange can be distinguished mechanistically (e.g., via sign behavior), and these experiments remain foundational hydration probes. For CI2, hydration cross peaks were detectable in our conventional 3D HSQC‐NOESY and water‐selective NOESY experiments [53, 54]. These required long multidimensional acquisitions and did not offer a rapid, residue‐selective separation of internal long‐residence waters from exchange‐dominated surface contacts. By contrast, the HyperW‐positive/CLEANEX‐negative criterion isolates the four amides linked to conserved internal waters in a single rapid 2D readout after injection. The practical advance here is that HyperW supplies a transient, highly polarized water reservoir and enables rapid acquisition of 2D protein correlation spectra after injection. When paired with CLEANEX‐PM, which suppresses through‐space transfer during the mixing period, this yields an operationally simple, residue‐specific filter (“HyperW‐positive/ CLEANEX‐negative”) that highlights amides influenced by internal, long‐residence water under the conditions used here. Importantly, UV‐HyperW improves practical sensitivity by substantially increasing the available water magnetization that drives water‐to‐protein transfer and slowing the decay of the hyperpolarization, avoiding spectral broadening. HyperW therefore complements traditional NMR hydration probes by reporting on structural hydration under native conditions in a workflow that is enabled by radical‐free HyperW and simple transfers, and that can be implemented on any NMR instrumentation used for standard protein studies.

Conflicts of Interest

Andrea Capozzi works for Polarize ApS. Polarize ApS commercializes dDNP‐related instrumentation.

Supporting information

Supporting File 1: anie71460‐sup‐0001‐SuppMat.pdf.

Data Availability Statement

The data that support the findings of this study are openly available in [Zenodo] at https://doi.org/10.5281/zenodo.17457036, reference number [17457036].