Three main results were obtained in this study: first, it was shown that the background-free 19F nucleus of the drug favipiravir, already used for anti-viral treatment in humans, could be hyperpolarized using an easy-to-apply and inexpensive experimental setup. The applied techniques allowed to combine the light-induced signal amplification with standard NMR signal averaging, and thus achieve a good SNR despite small amounts of the substance. Second, the photo-CIDNP effect was large enough at 0.6 T to allow 19F MRI of favipiravir with a sub-mm resolution. Third, the minimal amount detectable on MRI was about 500 pmol or less per voxel. Taken together, these results provide a major step towards significantly increasing the sensitivity of MRI and making it more applicable for molecular imaging techniques. This could enable the development of new strategies to monitor the metabolism of fluorinated drugs in living organisms even when they are present only in small amounts.
Favipiravir is a very interesting molecule as it contains a pyrazine moiety. It can therefore also be hyperpolarized using para-hydrogen-based techniques. 1H hyperpolarization of favipiravir was recently reported by Jeong et al. [34]. In contrast to our approach, they used SABRE to hyperpolarize Favipiravir. Howerver, they did not investigate a transfer of the 1H hyperpolarization to the 19F nucleus. Additionally, SABRE requires dissolving the drug in an organic solvent together with an Iridium-based catalyst, both of which are toxic. Therefore, this concept cannot easily be transferred to living organisms, as it would require rapid and complete removal of these substances before being applied. However, both techniques might be used in a complementary manner to analyze the different molecular mechanisms underlying generation and transfer of hyperpolarization between different nuclei.
Although our experimental setup is optimized for imaging, the ability to acquire spectra from different nuclei simultaneously with a sufficient resolution proved to be very beneficial in optimizing signal acquisition. One can immediately check the signal strength of both nuclei as well as the SNR and the position of the resonance lines (Fig. 8). The dual-core option also allows for the compensation for Larmor frequency fluctuations (due to residual minimal magnet temperature fluctuations) without the need to adding a reference substance. The prerequisite is that at least one of the nuclei in the system (solvent/chromophore/target molecule) exhibits a sufficient SNR in each individual measurement to function as an internal locking substance. Here, the 1H nucleus of the solvent (H2O) was used to post-process both the 1H and 19F spectra by determining the phase changes, correcting them, and applying the appropriate correction to the 19F signal before the signals were averaged. Thus, non-illuminated 19F spectra could be acquired, which served as a reference for the determination of the mean SE (s. below). For more technical details see [16].
The hyperpolarized spectrum could be further resolved by extending the acquisition time of the FID four-fold. The spectrum showed a doublet (Fig. 4c) in which the difference of the two resonance lines was 8.5 Hz, giving the J-coupling between the 19F and the vicinal proton of the aromatic ring system. However, this increased resolution was not used for the standard protocol because the increased noise contribution would have required more averaging, resulting in longer total acquisition time and higher total irradiation for the chromophore, which in turn would have resulted in faster bleaching. We also refrained from acquiring non-irradiated spectra with a higher resolution, since the standard protocol already required several thousand averages in which additional chemical reactions occurred. In general, we measured favipiravir for at most one to two days when left in the magnet.
Accelerating the data acquisition for MRI required a trade-off between fast measurements with low spatial resolution and longer-duration data acquisitions needed to resolve the spatial distribution of the hyperpolarization in more detail, which is required for biomedical applications. For the mere detection and approximate localization of the substance, rapid acquisition of low-resolution images may be sufficient. These might also be used to monitor the dynamics of a target molecule in terms of both localization and metabolism [35]. However, increasing the resolution reduces the signal and increases the time required for averaging. The time scale is also determined by substance-specific parameters such as relaxation times, sequence-specific parameters such as pulse angle, gradient duration, recording principles [36], and the irradiation time. For spectroscopy imaging, we found that a repetition time of 10 s including an irradiation time of 4 s gave good results. The fast spin-echo MRI technique, widely used in routine medical MRI, also allowed to acquire of a full low-resolution image within 4 s. However, there are significant faster sequences [36]. In the future, we will also investigate these including shorter light exposure times.
Another limiting factor in fast spin-echo imaging is the hyperpolarization lifetime, which determines the number of echoes that can be acquired. While smaller image matrices can be fully acquired after only one 90° pulse, higher resolution requires segmented acquisition. This means that several 90° pulses must be applied, each preceded by a time intervall required for illumination in order to generate the hyperpolarization again. Figure 3 shows the effect of recording all phase-encoded rows of the oversampled raw data matrices in the so-called k-space after one irradiation (Fig. 3b) compared to acquiring only a quarter of the raw data matrix before generating new hyperpolarization again (Fig. 3c). The ordering of the phase-coded rows in the k-space matrix prior to Fourier-transforming to get the image also influences the image contrast. Roughly speaking, the inner part determines the brightness and the outer part determines the finer details. Here we have chosen an order starting in the middle when the signal is still high, as we were originally interested in detecting 19F with the highest SNR. Therefore, the SNR of finer details can be degraded due to the signal decay when encoding the outer parts of k-space, additionally to the gradient-induced reduction of the signal. This can be seen by comparing Fig. 3b and Fig. 3c. In the latter case, the object appears to be better resolved, albeit with a reduced SNR in the center of the object. A further increase in resolution (Fig. 3d) required more averaging and limited the acquisition to about seventeen k-space rows after each 90° pulse.
From the high-resolution image Fig. 3d, we estimated the hyperpolarization lifetime of the 19F nucleus in favipiravir to be around 200 ms (TE 4 ms, PhOS 4, turbo factor 46, i.e. 46 out of 92 phase-encoded k-space rows were acquired with sufficent SNR after one excitation requiring 184 ms). Since the focus of this study was primarily to investigate the feasibility of hyperpolarized 19F MRI, we will determine the optimized pulse parameters and the exact lifetime of 19F hyperpolarization in future experiments. However, new technical solutions may be required as the standard strategy of increasing the SNR by increasing the number of averages and/or increasing the irradiance is hampered by the inevitable chromophore bleaching and possible accumulation of by-products. We found that these effects became evident after several hundred irradiation cycles.
With the help of the complementary information from spectroscopy and imaging, the mean SE factor and the heterogeneous distribution of the degree of hyperpolarization could be estimated. In most photo-CIDNP experiments, great efforts are made to ensure homogeneous illumination [37, 38], but studying biomedical samples will inevitably lead to an inhomogeneous distribution of the degree of hyperpolarization, since these samples are usually quite heterogeneous. The estimate of the mean SE should be currently be seen as a first approximation, since we only reached about 80% of the maximum hyperpolarization as a compromise between irradiation duration and bleaching (Fig. SI 1). Another factor was that FMN concentration were much lower than favipiravir. One molecule of the chromophore thus corresponded to about ten molecules of favipiravir. It can be speculated, that the number of favipiravir molecules could be significantly reduced before the hyperpolarized signal decreases significantly, which in turn would lower the limit of detection (see below). Indications of this effect can actually be seen in Fig. SI 2. The SE per molecule could therefore be higher than estimated here.
However, the SE measured here still enabled the clear detection of hyperpolarized 19F even in high-resolution MRI data, with lower limits of < 500 pmol amounts of favipiravir per voxel. With the exception of a few spatially resolved photo-CIDNP studies [15, 16, 30] signal amplification due to photo-CIDNP is usually reported from spectroscopic measurements with concentrations of the hyperpolarizable substance between 2 and 4 mM. Since the spectroscopic signal comes from the entire sample, efforts are made to illuminate the entire sample as uniformly as possible [37, 38].
Recent developments in LED-based Low Concentration (LC)-photo-CIDNP were in the range of 1 µΜ [39] or lower. Yang et al. [24] used 500 nM of the amino acid tryptophan in their fast-pulsing LED-enhanced NMR experiments. Three years later they reported the detection of only 20 nM tryptophan when acquiring an LED-irradiated enhanced signal of a proton in a side chain in tryptophan (Trp-α-13C-β,β,2,4,5,6,7-d7 ) [19]. Most of these studies have been performed with molecules containing aromatic systems with a hydroxy group such as tyrosine, or with tryptophan, which contains also nitrogen in a second ring system. Recently, Mompeán et al. reported sub-pmol detection sensitivity using a microcoil setup at 9.4 T to acquire a photo-CIDNP-enhanced 19F signal of 0.8 µM p-fluorophenole in a detection volume of 1 µL [40].
Compared to spectroscopic measurements, MRI measurements offer the opportunity to directly detecting regions with different degrees of hyperpolarization and thus potentially reducing the detection limit without special hardware. Lower degrees of hyperpolarization can either be due to lower amounts of substance or reduced light intensity, as seen in the peripheral parts of the light cone in our experiment. In comparison to a previous 19F hyperpolarized MRI study [16] we investigated a heteronuclear system here. Favipiravir is a pyrazine carboxamide and contains two nitrogen atoms, and a hydroxy group is substituted to this heteroaromatic system. Compared to 3-fluoro-tyrosine previously studied by our, the hydroxy group and the fluorine nuclei in favipiravir have a larger distance. In addition, there are fewer 1H and 19F couplings. Therefore, the hyperpolarization in favipiravir may be less distributed within the molecule and be more efficiently transferred to the 19F nucleus, which in turn may be one potential cause for the lower SE in 3-fluoro-tyrosine. Interestingly, the structural similarity of the hetero-aromatic ring in favipiravir to the middle part of riboflavin is immediately apparent [33]. Wörner et al. published photo-CIDNP measurements for non-classical disproportionation [41]. The effect of charge redistribution within the molecule is reflected in the hyperpolarizability of these systems. Further studies to analyze potential mechanisms will be necessary.
As further difference from our 3-fluoro-tyrosine study, we found that favipiravir changed much faster when left in the magnet at 303 K for several days (which is necessary to acquire several non-illuminated 19F spectra with sufficient SNR). While 3-fluro-tyrosine remained stable for several days [16], favipiravir showed an additional 19F signal in the non-illuminated spectra just after one day, while the hyperpolarized 19F spectra remained largely constant for the first four days (Fig. SI 2f). This additional 19F signal, which is shifted to lower frequencies by approximately 1 kHz (44 ppm) relative to the native 19F favipiravir signal (see suppl. information) did not appear in the hyperpolarized spectra meaning that this substance might not to be hyperpolarizable (Fig. SI 2).
Although for technical reasons we have not yet been able to precisely determine the molecular structure of the unknown substance, there might be assumed that the additional signal originates from one of the numerous isomeric structures of favipiravir. This molecule has two protonatable nitrogen nuclei in the ring system and the OH group, which can be converted into a ketone group by rearrangement (Fig. 6). It was shown that nine isomers which can be formed under different conditions (pH values) [42]. Theoretical calculations of the transitions between the tautomeric forms were published by [43]. Due to the high sensitivity of the 19F nucleus to chemical shifts, any structural molecular change affects the position of the one fluorine signal. However, SE in the photo-CIDNP measurements are only expected for the N-heteroaromatic systems that have both a conjugated double bond system and a hydroxyl group. Thus, the detected signal, which was enhanced at neutral pH, can be unequivocally assigned to the enol form.
The results presented here show how new strategies may allow MR-based techniques to close the gap to other molecular imaging techniques. An interesting approach may be represented by combining photo-CIDNP MR spectroscopy with optical spectroscopy in the ps- and ns-time domain [44], which allows the analysis of the rapidly occurring primary radical pair reactions. Future experiments will also include 3D imaging sequences to obtaim full spatial information about the hyperpolarization distribution. However, this requires additional time-consuming optimization of the Turbo spin echo sequence in terms of irradiation timing, excitation and refocusing pulses, segmented acquisition, etc., and was therefore outside the scope of this study.
Two limiting factors of this study have to be mentioned: (a) The effects of longer illumination with longer measurement schemes lead to chromophore bleaching, which ultimately prevents the sample from being optimally hyperpolarized and thereby limits the increase in spatial resolution. Potentially phototoxic side products can also limit biomedical applications [45]. (b) The 2D measurement will be replaced in forth-coming experiments by 3D measurements, which will allow a direct estimation of the spatial distribution of the hyperpolarization. However, they are usually more time-consuming and therefore require optimization of data acquisition or the use of other fast sequences [36].
In summary, unlike other hyperpolarization techniques, photo-CIDNP offers the possibility to hyperpolarize fully biocompatible model systems. The experimental setup used here could also provide additional information about the lifetime of 19F hyperpolarization at 0.6 T similar to other nuclei [38, 46, 47]. Similarly, the hyperpolarization of hetero-nuclei such as 31P, 13C, and 15N [48–50] might be investigated because the vendor provides the appropriate coils. In particular, combining MRI and dual-core spectroscopy can provide new results about hyperpolarization transfer between different nuclei and may offer complementary results to standard high-field spectroscopy and MRI.