Selective Isotope Labeling and LC-Photo-CIDNP Enable NMR Spectroscopy at Low-Nanomolar Concentration

Abstract

NMR spectroscopy is a powerful tool to investigate molecular structure and dynamics. The poor sensitivity of this technique, however, limits its ability to tackle questions requiring dilute samples. Low-concentration photo-chemically induced dynamic nuclear polarization (LC-photo-CIDNP) is an optically-enhanced NMR technology capable of addressing the above challenge by increasing the detection limit of aromatic amino acids in solution up to 1,000-fold, either in isolation or within proteins. Here, we show that the absence of NMR-active nuclei close to a magnetically active site of interest (e.g., the structurally diagnostic H^α-C^α pair of amino acids) is expected to significantly increase LC-photo-CIDNP hyperpolarization. Then, we exploit the spin-diluted tryptophan isotopolog Trp-α-¹³C-β,β,2,4,5,6,7-d₇ and take advantage of the above prediction to experimentally achieve a ca 4-fold enhancement in NMR sensitivity over regular LC-photo-CIDNP. This advance enables the rapid (within seconds) detection of 20 nM concentrations or the molecule of interest, corresponding to a remarkable 3 ng detection limit. Finally, the above Trp isotopolog is amenable to incorporation within proteins and is readily detectable at 1 μM concentration in complex cell-like media, including E. coli cell-free extracts.

Graphical Abstract

Introduction

Nuclear magnetic resonance (NMR) spectroscopy is a non-invasive technique capable of yielding invaluable information on molecular structure and dynamics at atomic resolution. On the other hand, NMR is intrinsically insensitive due to small differences in population between nuclear spin eigenstates, even at the highest commercially available magnetic field of 28.2 T (1, 2). Although low sensitivity can be compensated by increased sample concentration, this is often not a feasible option due to limited solubility, small amounts of material and high aggregation propensities. The above challenges are particularly severe in the case of biological samples in aqueous media. Numerous hyperpolarization methods were developed to increase the sensitivity of liquid-state NMR by altering nuclear spin populations at thermal equilibrium. Major developments comprise Overhauser dynamic nuclear polarization (ODNP) (3–5), dissolution dynamic nuclear polarization (DDNP) (6–8), and parahydrogen-induced polarization (PHIP) (9–15), including signal amplification by reversible exchange (SABRE) (16–33).

On the other hand, a few challenges have hampered the wide applicability of the above hyperpolarization technologies, to date. These include expensive instrumentation, harsh sample conditions (e.g., high/low temperatures, freeze-thaw cycles), long hyperpolarization times, applicability to a limited number of molecules of interest, safety considerations, inability to perform in situ hyperpolarization, requirement of high sample concentration (50 μM or more) and unfeasible data collection in physiologically relevant environments.

Earlier identification of chemically induced dynamic nuclear polarization (CIDNP) (34, 35) in organic reactions led to the development of a photochemically enhanced NMR approach known as photo-CIDNP. While originally employed to assess solvent exposure (36), this technique has more recently been employed to significantly increase NMR sensitivity in solution at high applied field (37–43). Photo-CIDNP is devoid of most of the drawbacks of other approaches listed in the previous paragraph. This technique requires aromatic moieties and benefits from an O₂-free environment.

Photo-CIDNP exploits spin-selective processes between a molecule of interest and a photosensitizer dye. The phenomenon proceeds via a radical pair mechanism, as shown in Figure 1A. Photo-CIDNP can be employed for NMR sensitivity enhancement purposes (44) and, when combined with heteronuclear correlation spectroscopy, it enables highly sensitive detection of hyperpolarized ¹³C and ¹⁵N (38, 39, 45). More recently, trace amounts of commercially available glucose oxidase (GO) and catalase (CAT) were employed to remove dissolved molecular oxygen, thus preventing excited-state dye quenching and further increasing NMR sensitivity (37). The use of fluorescein (Fig. 1B) as photosensitizer, in place of flavin mononucleotide (FMN), led to extension of this technique to low sample concentration (low- to sub-μM) (40). The use of photosensitizers characterized by a long triplet-state lifetime (e.g., 20 ms for fluorescein (40, 46)) gave rise to a branch of photo-CIDNP known as low-concentration photo-CIDNP (LC-photo-CIDNP). Replacement of costly and hard-to-maintain high-power laser with LEDs (Fig. 1C) with no sensitivity losses led to enhanced ease of access, portability and safety (42).

Figure 1. Basic aspects of low-concentration photo-chemically induced dynamic nuclear polarization (LC-photo-CIDNP).

A. Schematic illustration of the radical-pair mechanism of photo-CIDNP in liquids. D and M denote the dye and molecule of interest, respectively. S₀, S₁ and T₁ refer to the ground singlet, 1^st excited singlet and 1^st excited electronic triplet states, respectively. The T-S acronym denotes triplet-singlet mixing. The frequency of T-S mixing of the ^T[D^•−M^•+] radical pair depends on the nuclear spin states of the molecule of interest M^•+, due to hyperfine coupling between unpaired electron and nuclear spins. Therefore, depending on the value of nuclear spin states, molecules can either undergo fast or slow T-S mixing. The population with nuclear spin states undergoing slow T-S mixing experiences radical-pair dissociation followed by paramagnetic relaxation, resulting in thermal nuclear-spin polarization. Conversely, the population with nuclear spin states experiencing fast T-S mixing undergoes T-S mixing followed by fast back electron transfer. Importantly, this population experiences nuclear-spin hyperpolarization. In practice, one detects the sum of the two populations, in a photo-CIDNP experiment. B. Structure of a typical small molecule of interest and photosensitizer dye employed in LC-photo-CIDNP. C. Scheme illustrating typical LC-photo-CDINP instrumentation.

LC-photo-CIDNP performs optimally at low-to-sub μM sample concentration (40, 41), and it provides fast (0.2 s) in situ nuclear hyperpolarization. The latter features enable long-term signal-averaging and multi-dimensional NMR of extremely dilute samples. This technology leads to enhanced signal-to-noise ratio (S/N) of tryptophan (Trp) and tyrosine (Tyr), either as free amino acids or within polypeptides and proteins (38, 41–43). Remarkably, high LC-photo-CIDNP enhancements are observed for C^α nuclei of aromatic amino acids, likely due to their high hyperfine coupling constants (38, 47), enabling the efficient detection of H^α-C^α pairs. The C^α chemical shifts are particularly useful because they are diagnostic of backbone secondary structure in proteins (48). To take advantage of proton’s higher sensitivity, ¹³C-to-¹H polarization transfer via reverse INEPT is typically employed (38). In addition, 2D data collection is carried out to preserve information on ¹³C^α chemical shifts (38, 41). The recently developed ¹³C RASPRINT pulse sequence is optimized for ultra-fast data collection in aqueous solvent (42).This sequence exploits minimal recycle delays, given that most nuclear-spin polarization is generated during the optical irradiation time, thus abrogating the need for lengthy inter-scan intervals (42). The ¹³C RASPRINT sequence is particularly useful for the ultra-rapid detection of H^α-C^α pairs at low sample concentration, down to ca. 500 nM (42). More recent studies showed that LC-photo-CIDNP hyperpolarization can be further enhanced in the presence of low-μM concentrations of reductive radical quenchers including ascorbic acid, a.k.a. vitamin C, leading to detection of 200 nM aromatic amino acids (43). Despite the above advances, it is highly desirable to render liquid-state NMR spectroscopy even more sensitive so that it approaches detection limits of other leading techniques, e.g., fluorescence and mass spectrometry.

In this work, we provide theoretical and experimental evidence illustrating that nearby NMR-active nuclei (e.g., ¹H’s, ¹³C and ¹⁵N) significantly decrease the extent of observable ¹³C LC-photo-CIDNP polarization. Upon replacement of these nuclei with either NMR-inactive (¹²C) or low-gyromagnetic-ratio (²H) nuclei by selective isotope labeling, we were able to achieve an unprecedented improvement in NMR detection level in solution, down to 20 nM Trp, in only a few seconds. In addition, we showed that the above isotopolog-assisted optically enhanced technology is also effective in complex cell-like media, enabling the analysis of aromatic amino acids in physiologically relevant milieu at atomic resolution and extremely high sensitivity.

RESULTS AND DISCUSSION

Theoretical predictions on the dependence of LC-photo-CIDNP-hyperpolarization on isotope substitution patterns and g-factor values.

LC-photo-CIDNP polarization depends on the g-factor of the radical form of the reduced photosensitizer dye. Fluorescein is typically employed as a dye. The g-factor of its reduced radical form in solution, however, is not known. Note that different protonation states of fluorescein in solution are expected to affect g-factor values. We carried out density functional theory (DFT) calculations to estimate the pK_a and g-factor of the fluorescein dye radical at 25 °C and pH 7.0. Water solvation effects were taken into account via the integral equation formalism polarizable continuum model (IEFPCM). Predicted g-factor values were determined for the fully protonated state of the Fl^• dye radical and for seven variably deprotonated states, as listed in Figure 2A. The pK_a values of the three titratable protons were employed to determine the population of each of the eight variably protonated species listed in Figure 2A via the Henderson-Hasselbalch equation. The resulting individual populations and corresponding g-factors were employed to predict the overall weighted-average g-factor of the fluorescein dye radical at 25 °C and pH 7.0. This g-factor value was mostly contributed by Fl^•−, and was found to be 2.003077.

Figure 2. Computational predictions.

A. Predicted pK_a and isotropic g-factor values for fluorescein radicals of different protonation states. B. Predicted photo-CIDNP polarization of three Trp isotopologs as a function of dye g-factor. Simulations were performed using g-factor of Trp^• = 2.0027, and hyperfine coupling constant of Trp and fluorescein radicals from literature sources (see details in Suppl. Inf.).

The known g-factor of Trp^•+ (47) and the weighted-average g-factor of the fluorescein dye radical, determined as described in the previous paragraph, were used as input for computational predictions of LC-photo-CIDNP polarization as a function of g-factor of the dye radical. We followed known theoretical and practical guidelines (49–51), with some modifications, as detailed below.

For freely-diffusing radical pairs, the population difference between any two nuclear-spin configurations 1 and 2 of the geminate-recombination product is

$$p_1-p_2=p\left(\sqrt{\left|{\omega }_{TS1}\right|{\tau }_d}-\sqrt{\left|{\omega }_{TS2}\right|{\tau }_d}\right),$$ [1]

where p₁, p₂ are the populations with nuclear-spin configuration 1 and 2, respectively; p is a normalization factor (see Supplementary Information), ω_TS1 and ω_TS2 are the triplet-singlet mixing frequencies of the nuclear-spin configurations, and τ_d is the average time the radical-pair components remain closely associated (beyond orbital-overlap distances) before long-term dissociation. The triplet-singlet mixing frequency of each of any possible nuclear-spin configurations is

$${\omega }_{TS}=\frac{1}{2}\left[\mathrm{\Delta }g{\mu }_B{B}_0+\sum _i{m}_i{A}_i-\sum _j{m}_j{A}_j\right],$$ [2]

where the i and j subscripts denote individual nuclei of the dye and molecule of interest, respectively. In addition, Δg is the difference between the g-factors of the radical-pair components (i.e., the dye and molecule of interest), μ_B denotes the Bohr’s magneton B₀ is the applied magnetic field, and m and A are the magnetic nuclear-spin quantum number and hyperfine coupling constant, respectively.

For different isotope labeling schemes of the molecule of interest, corresponding to different isotopologs, the third term of equation [2] is expected to vary depending on specific A_j values. Therefore, we reasoned that different isotopic substitutions ought to give rise to variable populations of geminate-pair products, resulting into different levels of photo-CIDNP hyperpolarization (52). Hence we proceeded to explore how the g-factor of dye radicals is expected to affect the theoretically achievable geminate-pair recombination products, in the case of different Trp isotopologs. We focused on three representative labeling schemes of the molecule of interest: namely, (i) uniformly ¹³C-labeled Trp (Trp-U-¹³C), (ii) a hypothetical Trp isotopolog bearing a ¹³C-¹H pair at the α carbon (C^α) and extensive deuteration at non-exchangeable sites (Trp-α-¹³C-β,β,2,4,5,6,7-d₇, Fig. 3A), and, finally, (iii) a reference Trp isotopolog enriched in ¹³C at the C^α and bearing natural-abundance nuclides at all other sites (Trp-α-¹³C, Fig. 3B). Importantly, all of the above isotopologs have a ¹³C at the α position, honoring the fact that the LC-photo-CIDNP hyperpolarization approach is particularly effective at enhancing the ¹³C^α of the Trp and Tyr amino acids, either alone or within proteins (41). Additional details about the computations, including assessment of overall probabilities of geminate recombination, proportional to nuclear-spin polarization at the Trp α carbon, are provided in the Supplementary Information.

Figure 3. Synthesis and characterization of Trp isotopologs.

A. Scheme illustrating the enzyme-cascade synthesis of Trp-α-¹³C-β,β,2,4,5,6,7-d₇. B. Scheme illustrating the enzyme-cascade synthesis of Trp-α-¹³C. C. 1D ¹H NMR characterization of Trp-α-¹³C-β,β,2,4,5,6,7-d₇. A pulse-acquire data-collection scheme with a 121.5 Hz solvent presaturation scheme during the entire 5 s recycle delay was employed. 8 scans were acquired. The sample (ca. 100 μM) was prepared in 90% H2O, 10% D2O with traces of sodium trimethylsilylpropanesulfonate (DSS) used as an internal reference. The inset shows the ¹H^α region of a spectrum including ¹³C decoupling during acquisition using bi_p5m4sp_4sp.2 composite pulse decoupling (see Exp. Section in Suppl. Inf.). D. 1D ¹H NMR characterization of HPLC-purified and lyophilized Trp-α-¹³C (ca. 200 μM). The inset shows the ¹H^α region of the ¹³C-decoupled spectrum. The same acquisition and processing parameters as in panel C were used.

The simulation results for the above three Trp isotopologs are shown in Figure 2B. Even at optimal values of g-factor of the dye radical, the two compounds bearing no deuteration are predicted to afford only limited sensitivity enhancement. On the other hand, the Trp-α-¹³C-β,β,2,4,5,6,7-d₇ deuterated isotopolog (solid circles in the plot of Fig. 2B) is expected to undergo drastic polarization enhancements that significantly vary in magnitude depending on the g-factor of the dye radical. These enhancements extend up to nearly 14.5% for dyes that bear an optimal g-factor value.

Under typical LC-photo-CIDNP conditions, fluorescein is employed as photosensitizer (41). As detailed elsewhere (40), this dye bears a conveniently long triplet-state lifetime that persists across the infrequent collisions between dye and molecule of interest, at moderate sample concentrations (low μM and lower). Therefore, despite its sub-optimal g-factor value in the plot of Figure 2B, fluorescein is the current photosensitizer of choice in LC-photo-CIDNP experiments. When fluorescein is employed as a photosensitizer at room temperature and 25 °C, the predicted probability of geminate recombination (proportional to nuclear-spin hyperpolarization (52)) of uniformly ¹³C-labeled Trp (Trp-U-¹³C) is the lowest (ca. 5%). On the other hand, Trp-α-¹³C is predicted to experience a 4.4% more geminate-recombination than Trp-U-¹³C. More strikingly, the selectively labeled isotopolog Trp-α-¹³C-β,β,2,4,5,6,7-d₇, which includes extensive deuteration at C^β and aromatic carbon sites, is predicted to experience 28.2% more geminate recombination than Trp-U-¹³C.

Trp analogs bearing variable isotopic substitution patterns were prepared and characterized.

Encouraged by the computational predictions of Figure 2B, we proceeded to prepare some relevant isotopologs and experimentally test their performance in terms of achievable LC-photo-CIDNP hyperpolarization.

Previous selective labeling schemes were based on supplementing bacterial growth media with [4-¹³C]-δ-aminolevulinic acid (53) or with different singly labeled ¹³C-glucose isotopologs (54). These methods, however, suffer from low isotope-incorporation yields and are not easily customizable to the desired labeling patterns. Here, we employed an alternative recently developed enzyme-cascade approach (55) to readily achieve high levels of isotope enrichment starting from affordable and readily available precursors. This approach is particularly convenient due to its modular nature (55). For instance, it enables the facile synthesis of a wide variety of desired isotopologs by simply changing the labeling pattern of the starting materials (i.e., glycine and indole) and solvent (water).

The synthetic schemes of Figure 3A and 3B detail the route that led to the generation of Trp-α-¹³C-β,β,2,4,5,6,7-d₇ and Trp-α-¹³C with 44.7% and 31.8% yield, respectively. Given that the amounts of purified products were more than adequate for our purposes, no further optimization was performed. The identity of purified Trp-α-¹³C-β,β,2,4,5,6,7-d₇ and Trp-α-¹³C was confirmed by ESI MS and ¹H NMR. The pulse-acquire ¹H NMR spectra of the two isotopologs are shown in Figure 3C and 3D and confirm the presence of the expected ¹³C-¹H scalar coupling. The latter disappears upon ¹³C decoupling. In addition, no ¹H aromatic resonances are observed for Trp-α-¹³C-β,β,2,4,5,6,7-d₇, consistent with expectations.

The Trp-α-¹³C-β,β,2,4,5,6,7-d₇ isotopolog leads to large LC-photo-CIDNP sensitivity enhancements.

The two Trp isotopologs described in the previous section and the commercially available uniformly labeled Trp-U-¹³C,¹⁵N were initially analyzed under dark (i.e., LED-off) conditions. The ¹³C RASPRINT pulse sequence (42) was employed, except that the typically short recycle delays (50 ms) (42) were replaced by long ones (5 s) to allow for nearly complete longitudinal relaxation. This pulse sequence enables the detection of ¹³C-bound protons and is routinely employed in LC-photo-CIDNP (42).

Figure 4A, which focuses on H^α protons under dark conditions, shows that the selectively labeled Trp-α-¹³C and Trp-α-¹³C-β,β,2,4,5,6,7-d₇ isotopologs display a larger S/N than uniformly labeled Trp (Trp-U-¹³C,¹⁵N). While Figure 4A provides a useful qualitative assessment, the different sample concentrations prevent quantitative comparisons. A more quantitative analysis of NMR behavior under dark conditions is provided in Figure 4B. The plot in this figure examines both intensities at the resonance maximum and areas, and normalizes the data so that they are independent of sample concentration. Here we note that, relative to Trp-U-¹³C,¹⁵N, the Trp-α-¹³C isotopolog displays 20% increases in resonance area and intensity at the maximum. Conversely, Trp-α-¹³C-β,β,2,4,5,6,7-d₇ displays a 42% increase in area and an even more dramatic 192% increase in intensity at the maximum. The latter enhancement results from narrower ¹H linewidths, as confirmed by T₂ measurements (Supplementary Information). In summary, perhaps not surprisingly, detection of Trp H^α resonances under dark conditions is rendered more facile by the extensive deuteration of the Trp-α-¹³C-β,β,2,4,5,6,7-d₇ isotopolog. It is worth noting that H^α resonances are important in biology because they are diagnostic of structure, especially when analyzed in the context of protein backbone.

Figure 4. LED-off and LED-on data show enhancements due to both dark and light contributions.

A. NMR spectra showing the H^α resonance of different Trp isotopologs. Data were collected with the 1D ¹³C RASPRINT pulse sequence under dark (i.e., LED-off) conditions at the concentrations listed above the spectra. A 5 s recycle delay was employed. B. Quantitative analysis of H^α-resonance areas and intensities-at-the-maximum relative to Trp-U-¹³C,¹⁵N, after normalization to eliminate differences due to variable concentrations. C. LC-photo-CIDNP spectra acquired with the 1D ¹³C RASPRINT pulse sequence under light (LED-on) conditions for three Trp isotopologs of equal concentration. D. Quantitative analysis of H^α resonance area and intensity-at-the-maximum under light (i.e., LED-on) conditions relative to Trp-U-¹³C,¹⁵N. Data in panels B and D are shown as avg ± S.E. (n=2). All spectra were acquired at 24 °C with 64 scans, 4 dummy scans, and an optical irradiation time of 0.2 s.

Next, we performed 1D LC-photo-CIDNP experiments on all three Trp isotopologs under light (i.e., LED-on) conditions. Data were collected at very low sample concentration (1μM). The NMR spectra under light conditions, shown in Figure 4C, display considerable increases in S/N upon going from Trp-U-¹³C,¹⁵N to Trp-α-¹³C to Trp-α-¹³C-β,β,2,4,5,6,7-d₇. Remarkably, the latter isotopolog displays a ca. 4-fold increase in S/N relative to the already intense signal of uniformly labeled Trp (Trp-U-¹³C,¹⁵N). More quantitative studies, displayed in Figure 4D, show that Trp-α-¹³C displays a 70% increase in resonance area and intensity relative to Trp-U-¹³C,¹⁵N, while Trp-α-¹³C-β,β,2,4,5,6,7-d₇ features a 120% increase in area and a 300% increase in resonance intensity.

At this juncture, it is important to consider a few simple items. First, NMR sensitivity (1, 56) is defined as S/N divided by the square root of experiment time t, i.e.,$\frac{S}{N\sqrt{t}}$. Second, the NMR S/N ratio depends on the amplitude of both the resonance of interest and the noise. Third, all data were collected with identical acquisition and processing parameters. Taking all of the above considerations into account, the results of Figure 4C and 4D clearly show that the Trp-α-¹³C-β,β,2,4,5,6,7-d₇ isotopolog displays a remarkable 4-fold increase in sensitivity relative to uniformly labeled Trp (Trp-U-¹³C,¹⁵N). Note that all experiments under light conditions must bear equivalent noise values. This result is even more significant if one notices that (a) all the Figure 4C spectra acquired under dark conditions display no detectable signal, and (b) all the Figure 4 light and dark spectra included only 32 scans. In conclusion, the overall sensitivity of the ¹³C RASPRINT experiment for the Trp-α-13C-β,β,2,4,5,6,7-d₇ isotopolog must be very high. More details on specific enhancements factors and polarization values are provided in a later section.

The Trp-α-¹³C-β,β,2,4,5,6,7-d₇ isotopolog enables optically-enhanced NMR at low nanomolar concentration.

Encouraged by the results described in the previous sections, we further explored the LC-photo-CIDNP detection limit of the Trp-α-¹³C-β,β,2,4,5,6,7-d₇ isotopolog. Towards this end, we took a few deliberate simple steps. First, to optimize detection of the target isotopolog, we modified the ¹³C RASPRINT pulse sequence by removing its constant-time evolution. Note that the absence of ¹³Cs at the β and carbonyl positions (¹³C^β and ¹³C’) of the Trp-α-¹³C-β,β,2,4,5,6,7-d₇ isotopolog renders constant-time evolution unnecessary, due to lack of ¹³C-¹³C J couplings (57). Second, we converted all the on-resonance selective pulses into hard pulses and removed the ¹³C’ off-resonance pulse. These modifications are justified by the fact that the Trp-α-¹³C-β,β,2,4,5,6,7-d₇ isotopolog only bears one ¹³C and lacks ¹³C-¹³C J couplings. Therefore, there is no need for ¹³C^α selective excitation or for selective C’ pulses to refocus ¹³C^α−¹³C’ J couplings.

Importantly, all the above modifications are beneficial because they lead to shorter experiment times and reduce signal losses arising from pulse imperfections and transverse relaxation during the pulse scheme. The resulting pulse sequence, displayed in Figure 5A, was denoted as ¹³C isotopolog-optimized RASPRINT or ¹³C iso-RASPRINT.

Figure 5. LC-photo-CIDNP in combination with selective isotope labeling enables the detection of Trp at low-nanomolar concentration.

A. Scheme illustrating the ¹³C iso-RASPRINT pulse sequence, optimized for the detection of the Trp-α-¹³C-β,β,2,4,5,6,7-d₇ isotopolog. The ¹³C iso-RASPRINT sequence differs from the ¹³C RASPRINT sequence in that on-resonance selective pulses are replaced by hard pulses and off-resonance selective pulses are removed. In addition, the constant-time evolution was removed, yet a short spin-echo was preserved to compensate for the effect of pulse field gradients. B. LC-photo-CIDNP (1D ¹³C iso-RASPRINT) spectra of 50 and 20 nM Trp-α-¹³C-β,β,2,4,5,6,7-d₇. Data were acquired under dark and light conditions, in the absence and presence of 2 μM vitamin C (VC). In each spectrum, signal-averaging was carried out across 64-scans preceded by 4 dummy scans. The optical irradiation time was 0.2 s. Data were collected at 24 °C. The spectra enclosed within the green frame show the highest degree of nuclear-spin hyperpolarization.

Then, we tested the LC-photo-CIDNP behavior of extremely dilute (50 and 20 nM) Trp-α-¹³C-β,β,2,4,5,6,7-d₇ upon data collection with the ¹³C iso-RASPRINT pulse sequence. The results (two outmost-left spectra in Fig. 5B) show that the 50 nM sample yields a detectable resonance of moderate intensity after 64 scans (Fig. 5B). On the other hand, the 20 nM sample displayed no observable signal (data not shown). Given that reductive radical quenchers including ascorbic acid, also known as vitamin C (VC) is known to protect LC-photo-CIDNP molecules of interest from photodamage (43), potentially extending detection limit, we prepared an additional set of samples containing a small concentration of VC (2 μM). The results, shown in the four righthandside panels of Figure 5B, prove that in the presence of the VC reductive radical quencher even 20 nM Trp-α-¹³C-β,β,2,4,5,6,7-d₇ can be readily detectable, within only 64 scans. This is a remarkably small concentration, corresponding to an extremely favorable detection limit of 3 ng.

Quantitative assessment of LC-photo-CIDNP enhancements and polarization values.

To more quantitatively evaluate the extent of polarization enhancement generated via LC-photo-CIDNP for each of the isotopologs analyzed in this work, we compared resonance areas under dark and light conditions according to the relation below

$$\epsilon =\frac{Are{a}_{light}}{Are{a}_{dark}}\frac{{[Trp]}_{dark}}{{[Trp]}_{light}},$$ [3]

where ε is the polarization enhancement factor, defined as the ratio between nuclear spin polarization under light and dark conditions. Note that we acquired the dark (LED-off) spectra at much higher concentrations (ca. 100 μM) than the LED-on experiments, and then adjusted for concentration differences according to equation [3]. In addition, we employed long (5 s) recycle delays to ensure nearly complete relaxation between scans. We collected the same number of scans under dark and light conditions. Clearly, if the shorter delays employed in the light experiments had been employed for data collection under dark conditions, the apparent polarization enhancement factors would have been much larger. We refrained from doing this so that we could provide the highest possible reference S/N values under dark conditions.

From percent polarization enhancements ε and from thermal-equilibrium percent polarization P_%,te, assessed from the Boltzmann distribution in the presence of a 600 MHz applied field (1), we also determined the percent polarization P_% according to

$$P_{\%}=\epsilon P_{\mathrm{\%},te}.$$ [4]

The resulting ε and P_% values of all three isotopologs examined in this work are shown in Figure 6.

Figure 6. The Trp-α-¹³C-β,β,2,4,5,6,7-d₇ isotopolog displays unprecedented LC-photo-CIDNP enhancement factors ε and percent polarization in solution.

Enhancement factors (ε) were determined from experimental data according to equation [3]. Percent polarization values were assessed from enhancement factors and thermal-equilibrium percent polarizations according to equation [4]. Thermal-equilibrium percent polarizations were estimated at room temperature at 600 MHz applied field. Sample concentrations were 1 μM and 100 μM in LED-on and LED-off experiments, respectively. All values are shown as avg ± S.E., with n=2, where n denotes the number of independent experiments.

The enhancement factor and percent polarization of the two best isotopologs are discussed next. As shown in Figure 6, the ε and P_% values of Trp-α-¹³C are 410 ± 2 and 0.496 ± 0.003%, respectively. The Trp-α-¹³C-β,β,2,4,5,6,7-d₇ displays the highest values; namely ε and P_% of 470 ± 13 and 0.568 ± 0.016%. The Trp-α-¹³C and Trp-α-¹³C-β,β,2,4,5,6,7-d₇ isotopologs performed considerably better than uniformly labeled Trp (Trp-U-¹³C,¹⁵N), with increases of 38% and 57%, respectively.

The observed sensitivity enhancements arise from both LED-on and LED-off effects.

The data of Figure 4B, collected under dark (LED-off) conditions, show that selective ¹³C labeling as well as introduction of deuteration at C^β and across the aromatic ring induces some sensitivity enhancement even in the absence of optical irradiation. Therefore, some non-photo-CIDNP-related factors must contribute to the overall observed effects.

We were able to identify two specific dark-state contributions. First, there was an increase in ¹H^α T₂ arising from either the removal of some nearby NMR-active nuclei and(or) from their replacement with low-gyromagnetic-ratio (low-γ) nuclei due to selective labeling. The results of ¹H^α T₂ measurements are presented in the Supplementary Information. Clearly, this effect accounts for the sharper resonances, and longer ¹H^α T₂ displayed by the Trp-α-¹³C-β,β,2,4,5,6,7-d₇ isotopolog even under dark conditions. Second, the elimination of many ¹³C-¹³C, multi-bond ¹H-¹³C and ¹H-¹H J couplings arising from selective labeling prevents loss of coherence via J coupling during the pulse sequence. This effect was quantified upon comparing the area/concentration ratios for the ¹H^α resonance of each isotopolog in LC-photo-CIDNP measurements carried out under dark conditions.

In addition, there were also contributions due to LC-photo-CIDNP hyperpolarization effects under light (LED-on) conditions expected from the theoretical arguments discussed in the previous sections. The contributions arising from all three factors (linewidth reduction, J-coupling effects and LC-photo-CIDNP hyperpolarization) were quantified as described in the Supplementary Text and Supplementary Figure S4. The results are summarized in Table 1 for each of the isotopologs examined in this work. For easier-evaluation purposes, the data were normalized relative to Trp-U-¹³C,¹⁵N. The product of all three separately-assessed contributions agrees well with the overall enhancement factor in sensitivity. Overall, it is intriguing to note that the sensitivity enhancements described in this work were achieved by a combination of “dark and light effects”.

Table 1.

Relative sensitivity of LC-photo-CIDNP NMR experiments (1D ¹³C RASPRINT pulse sequence) carried out on different Trp isotopologs. The role of individual light (LED-on) and dark (LED-off) contributions is listed. Uniformly labeled Trp (Trp- U-¹³C,¹⁵N) is regarded as reference compound.

topolog	Relative sensitivity partitioned into individual contributions			Product of individual contributions d	Overall relative sensitivity (assessed via 13C RASPRINT, light cond.) e
	LC-photo-CIDNP hyperpolarization a(light effect)	Elimination of J-coupling b(dark effect)	Linewidth reduction c(dark effect)
Trp-U-13C,15N	100%	100%	100%	100%	100%
Trp-α-13C	138±4%	119%	126 ± 2%	207 ± 7%	172 ± 20%
Trp-α-13C- β,β, 2,4,5,6,7-d7	157±4%	142%	190 ± 5%	424 ± 16%	397 ± 9%

^a.Determined from data in Fig. 6, including assessment of errors.

^b.Determined from data in Fig. 4B. Only one experiment was carried out.

^c.Assessed from T₂ experiments in Supplementary Fig. S2. Values are shown as avg ± S.E., with n=3, where n denotes the number of independent experiments.

^d.Errors were assessed by error propagation of the individual contributions.

^e.Three ¹³C RASPRINT experiments were carried out for each isotopolog under LED-on conditions. Relative sensitivities were assessed from resonance intensities of experiments run with identical parameters, including number of scans and duration. Relative sensitivities are shown as avg ± S.E.

LC-photo-CIDNP is effective in complex biologically relevant media.

In addition to the simple environments discussed so far, including only few purified components, LC-photo-CIDNP can also be applied to samples within extremely complex biological milieux. To demonstrate this capability in the case of the LC-photo-CIDNP technology presented here, we prepared an E. coli S30 bacterial cell extract according to known procedures (58). This extract includes most E. coli soluble proteins. The composition of the corresponding NMR sample is schematically illustrated in Figure 7A and the procedure followed to generate the cell extract is outlined in Figure 7B. Importantly, this complex biological sample included 1 μM Trp-α-¹³C-β,β,2,4,5,6,7-d₇. Figure 7C illustrates the fact that no signal was observable for this extremely dilute Trp isotopolog under dark conditions. In contrast, 1 μM Trp-α-¹³C-β,β,2,4,5,6,7-d₇ became readily detected after only 8 scans under light conditions. Remarkably, due to the nearly absent recycle delay of ¹³C iso-RASPRINT under light conditions, the entire data collection, including the hyperpolarization time, only took a total of ca. 4 s (excluding the time for the 4 dummy scans, i.e., 2 additional seconds).

Figure 7. LC-photo-CIDNP in combination with selective isotope labeling enables hypersensitive data collection in a complex cell-like medium.

A. Schematic illustration of the components of the bacterial cell-like medium, i.e., an E. Coli S30 extract, employed in this work. B. Summary of the adopted procedures to generate the E. Coli S30 extract. C. LC-photo-CIDNP spectrum of 1 μM Trp-α-¹³C-β,β,2,4,5,6,7-d7 in an E. coli S30 cell extract under dark (i.e., LED-off) and light (i.e., LED-on) conditions. Each spectrum was acquired at 24 °C with 8 scans, 4 dummy scans, and 0.2 s of optical irradiation.

Comparisons with other hyperpolarization approaches in liquids.

As a comparison, two representative examples from other leading nuclear-spin hyperpolarization approaches will be listed next. The well-known dissolution DNP (dDNP) technology enabled detection of 20 μM protein samples (59). This technology has also been used in the context of protein-ligand interactions (6, 60–65). In addition, Kouřil et al. (66) recently showed that 32% ¹³C polarization was rendered possible by bullet-DNP, an interesting variation of dDNP employing small sample-dilution factors during the dissolution process. In both the above studies, considerably higher sample concentrations were required than in the present work. Moreover, unlike LC-photo-CIDNP, the dDNP approach requires a freeze-thaw process that is not always desirable in the case of biomolecules.

Very recent studies showed that parahydrogen-induced polarization (PHIP), especially via the SABRE approach, can be employed for the detection of amino acids in solution (18, 67, 68). While this methodology is promising, identification of suitable amino-acid precursors for SABRE remains difficult, and extension of this technique to proteins has not been feasible.

CONCLUSIONS AND OUTLOOK

In this study we introduce a combination of LC-photo-CIDNP and selective isotope labeling to enable the NMR detection of low-nM biomolecular concentrations by liquid-state NMR. The use of selective isotope labeling enables an additional ca. 4-fold increase in sensitivity relative to a uniformly isotopically enriched reference molecule (Trp-U-¹³C,¹⁵N). This gain in sensitivity was crucial to access the low-nanomolar (20 nM) concentration range, providing a remarkable detection limit of only 3 ng in a few seconds. To the best of our knowledge, this is the lowest concentration achieved by high-resolution biomolecular NMR in liquids to date. Our methodology works well even in extremely complex physiologically relevant environments, including an E. coli S30 cell extract. The capability to assess Trp levels (69–71), metabolism and transport in biological media is important because of the emerging role of this amino acid and its metabolites in health and disease (72–78).

Importantly, the LC-photo-CIDNP technology, including the advances presented here, is also readily amenable to extensions to Trp-containing macromolecules. For instance, LC-photo-CIDNP of polypeptides and proteins has already been established (38, 41, 43), and it yields comparable enhancements to free Trp. Further, given that Trp analogs can be routinely incorporated into proteins via auxotroph strains (79), one can envisage that the same strains can also be employed to incorporate Trp isotopologs like Trp-α-¹³C-β,β,2,4,5,6,7-d₇ into proteins. Ultrasensitive detection of H^α-C^α pairs in proteins is highly desirable because it provides access to site-specific information on backbone secondary structure (48). Further, the ¹³C iso-RASPRINT pulse sequence can readily be run in 2D mode to resolve overlapping resonances within mixtures of aromatic amino acids or in the context of large biomolecules.

In all, the advances described in this work pave the way to a variety of biomolecular NMR applications requiring high sensitivity and fast data collection. It will be exciting to follow the forthcoming progress of this technology in the years to come.