Natural product mixture analysis by matrix-assisted DOSY using Brij surfactants in mixed solvents

The assignment of NMR signals to speci ﬁ c components in a mixture is a challenging task. Di ﬀ usion-Ordered Spectroscopy (DOSY) has provided important progress in this area, allowing the signals originating from individual components of di ﬀ erent molecular sizes to be distinguished. However, when the sizes of the compounds are similar and/or the spectra are overlapped, signal assignment can easily become intractable. The use of a co-solute in a matrix-assisted DOSY (MAD) experiment can be a useful solution, improving di ﬀ usional (and sometimes spectral) resolution by exploiting selective binding to the matrix. MAD has generated promising results in the study of several types of mixtures, including those of functional and structural isomers. The challenge is to apply MAD to molecules with high structural similarity, for example in natural product mixtures. Various surfactants, including SDS, AOT and CTAB have previously been shown to be e ﬀ ective in MAD analysis. Here we present an important addition, the Brij family of nonionic surfactants. We demonstrate the use of Brij micelles in mixed solvents with a variety of mixtures relevant to natural products.


Introduction
NMR is arguably the most powerful technique for the elucidation of chemical structure, but the spectra of mixtures are oen very overlapped, which greatly complicates assignment. Diffusion-Ordered Spectroscopy (DOSY) 1,2 provides a 2D spectrum in which the second dimension disperses signals as a function of diffusion coefficient, allowing signals originating from components of different molecular sizes to be distinguished, 1 and hence is very useful in mixture analysis. A DOSY experiment encodes diffusion information into the attenuation of individual signal amplitudes in a series of spectra, 3 and is typically performed using modern versions of the pulsed eld gradient stimulated echo (PFGSTE) sequence, such as the Oneshot sequences. 4,5 Differentiation between signals is, however, naturally only possible when species diffuse at different rates. Many types of mixture, notably natural products, contain chemically cognate species of very similar sizes, and here the standard DOSY experiment is at best of limited value. However, by manipulating the matrix in which the solutes diffuse, differential binding can be exploited to facilitate separation of signals in a "matrix-assisted DOSY" (MAD) experiment. 6 In a successful MAD experiment, the analytes interact to different extents with a slowly-diffusing co-solute (e.g. micellar surfactant) in fast exchange. The average solute diffusion coef-cient, D av , is then given by Lindman's law: where D u and p u and D b and p b are the diffusion coefficients and the fractions of unbound and bound solute molecules respectively. 7 MAD has been used in the resolution of a variety of organic mixtures, including mixtures of isomers, exploiting selective cosolute binding that depends on such factors as polarity, amphiphilicity and stereochemistry, resulting in altered diffusion coefficients and allowing the separation of the individual component spectra. SDS and AOT (micelles and reverse micellar aggregates, respectively) have been shown to be useful tools in the separation of signals of dihydroxybenzenes, 6 monomethoxyphenols 8,9 (in both cases positional isomers), and medium chained alcohols 9 in MAD experiments. SDS micelles have also been used with a mixed solvent system, allowing complete identication of each component in avonoid mixtures. 10 Polymers like polyvinylpyrrolidone (PVP) 11 and polyethylene glycol (PEG) 12 have also been used as efficient matrices in MAD. b-Cyclodextrin 13,14 has been used as a chiral matrix, allowing the resolution of a mixture of two native epimers of naringin. 13 Microemulsions have been used to separate the component signals in commercial drug formulations. 15 A simple change of solvent can also be effective. 16 Even solids like silica (SiO 2 ) powder and ODS -(octadecylsilane) coated chromatographic stationary phases can be used as matrices, and have been shown to allow the separation of signals in a mixed, homologous series of aromatic compounds. 17 When a solid matrix is used, an HRMAS-DOSY 18 experiment may be needed to provide adequate signal resolution. Satisfactory resolution of signals has been demonstrated in such experiments, but one complication is of course the necessity for sufficient resolution in the chemical shi dimension, which is more difficult to achieve in heterogeneous media. 19 Here, choosing a liquid phase that is matched in magnetic susceptibility to the solid can allow standard liquidstate NMR methods to be applied. 20 Silica-based methods predominantly separate solute signals according to their hydrogen-bonding affinity, primarily distinguishing between solutes with different functional groups. 21 Flavonoids are an important family of natural compounds with interesting biological properties, most importantly as antioxidants; 22 over 4000 types of avonoid compounds have been identied in vascular plants. 23 Monoterpenes are also important compounds, and are the main constituents in the majority of plant essential oils. They give plants their unique odoriferous properties and are volatile because of their low boiling points. 24 HPLC is oen used to separate the components of plant extracts, with the fractions then being identied and characterised using mass spectrometry and/or NMR. 25 The use of DOSY (and MAD) can enable identication directly in an intact mixture, or in a fraction containing inseparable components. Most MAD analyses have hitherto been performed in aqueous solution, in which many natural product compounds of interest, including monoterpenes, have low solubility. It is therefore of interest to explore alternative solvent systems in which the solubilities of these compounds are higher.
Various surfactants, including SDS, AOT and CTAB, have already been used in MAD analysis, 6,8,9 but not the Brij family of nonionic surfactants. These consist of a hydrophilic head containing a variable number of polyoxyethylene groups, and a hydrophobic polymethylene tail. 26 Brij surfactants have been used as carriers in drug delivery 27 due to their low toxicity, 28 and in solubilization of drugs that have low water solubility. 26 Previous studies have shown that surfactants like SDS can form micelles in binary solvent systems, 29,30 but there are no such reports for the Brij family. Here, we demonstrate the use of Brij 78 and 98 (Scheme 1; Table 1) in mixed water-DMSO solution in MAD experiments primarily aimed at molecules with low solubility in water.   The CMC behavior of various PEO-block copolymers has been investigated in water-ethanol mixed solvents. [33][34][35] These studies have shown that the CMC increased aer addition of ethanol. The addition of DMSO-d 6 to D 2 O seems to show the same tendency, providing better solvent conditions for the Brij 78 and 98 compared to pure D 2 O and disfavoring the formation of micelles (self-assembly). Further thermodynamic studies would be useful to understand this observation better.

Results and discussion
To demonstrate the use of DOSY and MAD for mixtures containing monoterpenes and avonoids, a test system of 6    compounds (Scheme 2) was chosen for analysis. Standard DOSY spectra of a mixture of the monoterpenes (Scheme 2, compounds 1-3) in 20% DMSO-d 6 -80% D 2 O (v/v) shows a separation of L-(À)-carvone (3) signals from those of the other two compounds, carvacrol (1) and thymol (2) (Fig. 2a). This is unexpected given the very similar molecular weights, but may reect both the difference in shape and the reduced propensity to hydrogen bonding, and shows the potential of DOSY to separate the signals even of quite similar species.
Adding Brij 78, the monoterpenes 1 and 2 show strong association with the micelles, while compound 3 interacts only weakly. The strength of interaction with Brij 78 is clearly different for compounds 1 and 2, and the signals separate in the diffusion domain (Fig. 2b), making it possible to identify all of the three compounds. On addition of Brij 98, resolution of 1 and 2 is less pronounced and when using SDS in 20% DMSO-d 6 -80% D 2 O (v/v) as co-solute, no resolution is observed between 1 and 2 (results not shown). The effect of Brij 78 concentration was investigated using samples containing 3.0 mM of each monoterpene (Fig. 3) with best resolution achieved at 5.0 mM surfactant.
For the avonoid mixture (Scheme 2, compounds 4, 5 and 6), the standard DOSY spectrum in 50% DMSO-d 6 -50% D 2 O (v/v) shows all signals having very similar diffusion coefficients (Fig. 4a). On adding Brij 98, the signals of the three avonoids become well resolved in the diffusion dimension, allowing individual characterization (Fig. 4b) and improving on earlier results obtained with SDS. 10 The effect of Brij 98 concentration was also investigated using samples containing 5.0 mM of each of the avonoids (Fig. 5). The best resolution was obtained with 10 mM of Brij 98; the CMC in this mixed solvent was measured as 8.0 mM (Fig. 1).
Comparing the chemical shis of the Brij signals before and aer addition of interacting compounds can give some insight into the nature of the association between solute and micelle. Changes for particular resonances suggest that the local environment surrounding the corresponding hydrogen atom has changed, evidencing the presence of the solute near that hydrogen. 36,37 For Brij 78, the peaks corresponding to the protons on the rst ethoxylate group and others close by (3.54, 3.41 and 1.54 ppm, respectively) experienced the greatest upeld shi in the presence of L-(À)-carvone (1), thymol (2) and carvacrol (3). The remaining ethylene oxide chain showed little change in chemical shi and the nonpolar chain showed no change ( Table 2).
For the monoterpenes, the strength of binding increases in the order carvone < thymol < carvacrol, the same order as that of estimated octanol-water partition coefficient (log P): 2.70, 3.34 and 3.82, respectively (estimated using Molinspiration Chem-informatics® soware tool) 38 as a measure of molecular hydrophobicity. This is consistent with the hypothesis of incorporation into a micellar core. However, considering the hydrocarbon chain as the core of the micelle and the ethoxylate chain as the shell region, in the Brij 78 case it would appear that the monoterpenes are preferentially bound at the core/shell interface.
Brij 98 exhibits a different pattern to Brij 78, with all protons in the chain exhibiting changes in chemical shi in the presence of quercetin (4), setin (5) and (+)-catechin (6). The hydrocarbon chain closest to the EO chain (H3 and H4) experiences the greatest upeld shi upon addition of avonoids. The EO chain (H1) and the nonpolar chain (H8) also experience chemical shi changes, but to a smaller degree ( Table 2). As with the monoterpenes, the avonoids appear to be concentrated at the core/shell interface, but their effects spread throughout the EO and hydrocarbon chains, presumably because of their signicantly larger size.

Conclusions
We demonstrate efficient methodologies for NMR analysis of monoterpene and avonoid mixtures, using for the rst time nonionic Brij surfactants in binary solvent mixtures in MAD experiments. In the particular cases studied, Brij 78 and 98 showed better discrimination than SDS, the surfactant currently most commonly used in MAD.

Experimental section
NMR experiments were carried out non-spinning on a Varian Inova 400 MHz spectrometer. Temperature control was set to 25 C. The Oneshot pulse sequence 4,39 was used for DOSY experiments, using a diffusion-encoding pulse width (d) of 3 ms, a diffusion delay (D) of 200 ms and nominal gradient strengths ranging from 5.0 to 27.0 G cm À1 with 32 gradient amplitudes increased in equal steps of gradient squared. TSP was used as internal reference for reference deconvolution 40 in all the samples. All processing was done in the DOSY Toolbox, 41 and in all DOSY processing the data were corrected for the effect of non-uniform pulsed eld gradients. 42 Experiments carried out to determine the Brij critical micelle concentrations (CMCs) were performed with a diffusionencoding pulse width (d) of 4.5 ms and a diffusion delay (D) of 400 ms, using Brij solutions prepared by dilution from a stock solution.