Ultra-high dispersion NMR reveals new levels of detail

The full-text may be used and/or reproduced, and given to third parties in any format or medium, without prior permission or charge, for personal research or study, educational, or not-for-pro t purposes provided that: • a full bibliographic reference is made to the original source • a link is made to the metadata record in DRO • the full-text is not changed in any way The full-text must not be sold in any format or medium without the formal permission of the copyright holders. Please consult the full DRO policy for further details.

Every day very large numbers of 1 H NMR spectra are acquired both in academia and industry: 1 H NMR is the main workhorse of chemical structure determination. One common problem with 1 H NMR spectra is that signals overlap. The low dispersion of proton signals and the extra space needed to accommodate signal multiplicity oen conspire to make such spectra hard to interpret. One remedy is to use two-dimensional NMR methods. Experiments such as 1 H- 13 C HSQC 1 that exploit the large dispersion of the carbon dimension provide the best resolving power for most molecules. Unfortunately, many samples remain difficult to analyse even with these methods. Typical problems that are frequently encountered are those in which structural similarity between different parts of a molecule is present. Carbohydrates, complexes, lipids, aromatic systems and oligomers are typical examples. Naturally, the presence of multiple species makes the problem even worse. Some cases lead to such difficulties that analysis is oen not even attempted.
There is little that can be done to deal with the poor chemical shi range of protons, but, fortunately, signal dispersion can be improved by collapsing proton multiplets into singlets using "pure shi" methods. 2,3 This approach can be highly efficient: to achieve similar resolution by the usual means of increasing the magnetic eld would require elds that are way beyond current engineering capabilities. 4,5 However, in spite of the great performance advantage of pure shi methods, their potential remains largely untapped; they are as yet relatively little used. They are two reasons for this. The rst is that there is a mismatch between what a chemist considers a typical sample and the samples that have been used to demonstrate pure shi methods. Most samples run by NMR services catering to chemists are not particularly concentrated, tend to be impure (only end products tend to be pure), oen contain unknowns, and are sometimes unstable. In contrast, samples used for demonstrating the performance of new NMR techniques tend to be concentrated, pure, of known composition, and stable. The second reason is the common assumption that pure shi experiments are complex and time-consuming, and not suitable for routine service use.
The purpose of this paper is to show, as for example in Fig. 1-3 below, that this assumption is both incorrect and unfortunate: many problems that could be solved using pure shi methods go unnoticed. It is precisely in routine NMR facility use that such methods can make a major impact on research projects that require frequent NMR analysis. 7 Furthermore, it is not true that all pure shi experiments require excessively long times. The sensitivity of current hardware means that most samples can be run within modest time allocations. As will be shown here, impressive results can be produced routinely even with samples of uncertain composition, oen dilute, sometimes unstable, and always complex. The target audience is that of researchers dealing with small-to medium-sized molecules.
The examples presented here are all drawn from an NMR service that caters to typical synthetic chemists. All the samples formed part of real investigations, and had to be analysed within the time-constrained schedule of such a service. In all cases, sample composition turned out to be more complex than initially anticipated, and in some cases, there was no indication that pure shi methods were necessary. Three experiments were used, a one-dimensional pure shi 1 H, a pure shi COSY and a modied real-time pure shi 1 H-13 C HSQC experiment. Onedimensional spectra were produced using a Zangger-Sterk 5 (ZS) pulse sequence, taking the precaution of reducing the bandwidth to be decoupled to cover only signals of interest. It has to be kept in mind that sensitivity deceases as decoupling bandwidth increases. The recently reported PSYCHE method 6 could have been used to advantage, had it been available before the manuscript was in production, as it has higher sensitivity and does not require a compromise to be made between sensitivity and decoupling bandwidth. The COSY experiment used a single quantum constant-time sequence instead of the previously reported multiple-quantum ltered experiment, 8 as it is more sensitive. PSYCHE-TOCSY 9 could also been used following similar arguments to the 1D experiment. Both constant-time COSY and PSYCHE-TOCSY are less sensitive than their traditional counterparts, but this is normally not a problem for the target audience provided that large constant times in the rst case and long adiabatic pulses in the second are avoided. Combining these experiments with covariance processing 10 further simplies spectra, as in the examples presented, but this is optional. For HSQC, a variant of a previous version 11 was used differing only in that, in this version, adiabatic pulses were used to improve the off-resonance performance of the carbon pulses. Unlike other pure shi experiments, this one is more sensitive than its conventional counterpart. All pulse sequences can be found in the ESI. † Other pure shi experiments exist that require far longer acquisition times than their conventional analogues. In some situations those will be necessary. The proposed set of experiments can then be used to assess the severity of the problem and whether the aforementioned ones are worth running.   2 Conventional (a) and pure shift (b) HSQC spectra of a metallophoto-catalyst whose structure is depicted in (a). The number and nature of species present in solution was of interest because the activity of the catalyst depends on the replacement of the labile group (-R, originally a triflate), with weak chelators. It was also of interest because degradation products compromise the performance of the catalyst. Replacement of the labile group is characterised by patterns of peaks that shadow one another, due to small chemical shift changes. On the other hand, degradation products produce signals that deviate from such patterns. The shadowing pattern can be seen in both experiments, but the pure shift version (b) reveals more species that its conventional counterpart (a). Each spectrum took 28 min to acquire, while the carbon spectrum took 3 h. The total catalyst concentration was 7.5 mM, made up of multiple species. Fig. 3 An example of how the proposed combination of experiments was used to solve a problem involving a dilute sample containing unexpected components. A researcher was analysing the 1 H, COSY and HSQC spectra of a sample containing the expected mercury complex shown in (c). The researcher was only suspicious of the 7.85 ppm signal but a colleague, suspecting that this was not the only problem, acquired a pure shift 1 H spectrum. This revealed the presence of more signals than could be attributed to the expected product. Some signals were doubled or even tripled, indicating that the sample contained at least three species. This led to the acquisition of pure shift COSY and HSQC experiments. These revealed that other signals that appear as singlets in the one-dimensional pure shift spectrum were mixtures of overlapping signals. Notice that their correlation patterns are quite similar even in the HSQC spectra, thus indicating structural relatedness. An extended analysis indicated the presence of the uncoordinated ligand and two complexes. The free ligand was actually present but in exchange with at least one of the complexes. It was this dynamic process that prevented the researcher from detecting the presence the free ligand. Note that the combination of pure shift 1 H, COSY and HSQC made the analysis possible. The pure shift experiments took a total of 1.5 h of spectrometer time although only one should have sufficed. Notice how a little more than one hour of spectrometer time made an important impact on this particular investigation. The total concentration was 1.6 mM but three species are present. The cross-peak labelled with an asterisk is a folded peak. Note that the labels of the axes in (c) are correct, F 1 and F 1 , since covariance NMR derives the frequency of both dimensions from t 1 data. Valsartan, an angiotensin II receptor antagonist, is a good example of why it is helpful to acquire pure shi spectra routinely. The drug was being investigated to determine whether its conformations in solution could predict the existence of polymorphs in solid formulations. Part of the investigation focused on the spin system shown in Fig. 1a. The spin system is duplicated due to the presence of different conformers, but this duplication went unnoticed even when a combination of conventional experiments was used ( 1 H, 1 H-1 H COSY, 13 C{ 1 H} and 1 H-13 C HSQC). In contrast, such duplication became clear when pure shi variants were used (same gure). 20 min was all that was needed to break the problem using the pure shi experiments.
Time restrictions are a fact of life for NMR facilities, and users oen struggle to solve the problem at hand under typical time constraints. The problem is aggravated when samples are suspected of being unstable, as in the next example. Here, the object of the investigation was the photo-catalyst of Fig. 2a.
There was an interest in determining how many different complexes existed in solution, as the efficiency of the catalyst depends on the nature and stability of such complexes. All the complexes were expected to have the same structure, apart from one labile group (-R). This group was initially a triate, but was expected to be replaced with other species present in solution. This type of problem is very difficult because all the species are very similar, and produce signals that are very close to one another. The utility of the catalyst also depended on its stability, so the need to identify possible degradation products further complicated matters. Cases such as the present one can be analysed using 1 H-13 C HSQC experiments, because signals from similar complexes produce patterns only slightly displaced from one another while degradation products tend to produce at least some signals with signicant deviations. The problem is that conventional 1 H-13 C HSQC can only be successful if the carbon dimension is very well digitised, which is very timeconsuming. In this particular case, adequate digitisation would have required 4 to 5 h of spectrometer time. This was not available because little time was le aer three hours of spectrometer time had already been consumed acquiring a 13 C{ 1 H} spectrum that did not provide much insight. In contrast, a pure shi HSQC spectrum provided the necessary resolving power within 28 min, showing that at least ve very similar species were present. Notice how major peaks are shadowed by minor ones, producing the repetitive pattern noted earlier (see Fig. 2b.) It is remarkable that this was accomplished in spite of using standard equipment, the short time available, and the low concentration of the sample, 7.5 mM shared among the different species. This was possible because while the resolving power of the pure shi HSQC experiment increases only linearly with experiment duration in the carbon domain, the large resolution improvement afforded by pure shi acquisition comes at no cost in experiment time. A variety of similar examples from our NMR service can be seen in the ESI † section.
Although in the previous case a single experiment was enough, it is generally better to use the three proposed experiments together. A case that typies this situation is that of Fig. 3. Here another researcher was making a biomedical imaging probe 12 that required formation of the mercury complex shown. The unexpected form of the signal at 7.85 ppm led to the measurement of a 1D pure shi 1 H spectrum. Unexpectedly, this revealed that some signals were doubled or tripled, indicating that the sample contained at least three species (see for example the 6.8-6.9 ppm cluster of peaks). The task was then to determine the identities of these species. It was necessary to correlate and group those extra signals, but this was not possible using conventional COSY. The double pure shi COSY spectrum revealed that some signals that appear as singlets in the pure shi 1 H spectrum actually originate from mixtures of signals that show similar correlation patterns. This becomes clear when all three complementary pure shi experiments are used in conjunction. Further analysis indicated the presence of the free ligand and two complexes, meaning that the sample was unsuitable for its intended use. However, the experiments also indicated that the problem could be solved by repeating the reaction with a higher mercury to ligand ratio. The reaction was attempted and indeed a single product was obtained. The take-home message here is not that pure shi experiments revealed some hidden signals, but that they allowed analysis at a level of detail that made it possible to identify an unsuspected problem with the synthesis, and subsequently optimise the synthesis to avoid the problem. Again, this was done with a sample that contained less than 1 mg mL À1 of material (i.e. sub-mM concentrations) using a conventional (not cryogenically-cooled) probe. All the experiments took the same time to run as their classic counterparts, with the exception of the pure shi 1 H 1D spectrum, which took 13 min. The pure shi HSQC was le to run for 55 min, but 30 would have been sufficient, and the constant-time COSY took 25 min. To put these experiments in context, a standard 13 C{ 1 H} spectrum was acquired. Chemists typically rely on such experiments to provide resolving power, yet aer 4 h the resulting spectrum (included in the ESI †) was barely usable.

Conclusions
In all the cases presented, pure shi methods revealed a level of detail that was unobtainable using conventional NMR experiments. In all cases, the pure shi experiments tted well within the typical schedule of the NMR service, and indeed in some cases required less time than conventional experiments. The potential of pure shi experiments as enablers of better and more efficient investigations is only going to grow in the near future, and there is much to be gained by early adoption. Used in conjunction with emerging time-saving approaches such as non-uniform sampling 13 and with high sensitivity probes, 14,15 such experiments are poised to deliver a minor revolution in NMR spectroscopy.
1, and EP/L018500/1). Dr Kiraly thanks the European Commission for a Marie Curie Intra-European Fellowship, RTACQ4PSNMR, no. PIEF-GA-2013-625058. The authors also thank the Universities of Durham and Manchester for support.