The application of high resolution di ﬀ usion NMR for the characterisation and quanti ﬁ cation of small molecules in saliva/dentifrice slurries

The application of DOSY (Di ﬀ usion Ordered Spectroscopy) NMR as a technique for the virtual separation of toothpaste adjuvants in model saliva is reported for the ﬁ rst time. In addition, the scope and limitations of DOSY NMR are considered using the DOSY Tool Box processing software, as is the quanti ﬁ cation of the adjuvants and components of saliva by quantitative NMR (qNMR). These techniques represent a new and powerful tool for the evaluation of complex mixtures of natural products with a view to identifying biomarkers for disease within the oral cavity.


Introduction
Within the eld of oral care research Nuclear Magnetic Resonance (NMR) techniques have barely been considered as a technique for characterising the small molecule components of saliva or those anti-microbial adjuvants in a dentifrice slurry. The main references to NMR in the oral sciences focus on solid state (MAS) 19 F and 31 P NMR spectroscopy and the potential application of advanced 2D and pseudo 2D NMR techniques have largely been ignored. 1 The direct application of Diffusion Ordered Spectroscopy (DOSY) NMR and quantitative NMR (qNMR) for the characterisation and quantication of small molecules, natural products and biomarkers in saliva/dentifrice slurry models has until now gone unreported.
To augment the effectiveness of toothpaste a number of actives are added to improve the health of both hard and so tissues found in the oral cavity. These fall into three broad categories: anti-caries actives (e.g. sodium uoride, sodium monouorophosphate and amine uoride), sensitivity actives (e.g. potassium nitrate, potassium chloride, strontium acetate, stannous uoride and bioactive glass), gum health actives, comprising mainly antimicrobial agents such as triclosan, zinc citrate, stannous, essential oils, isopropylmethylphenol plus a wide range of other natural products. [2][3][4][5] Despite a considerable number of publications on dental plaque pathogenicity and biolm formation, there are relatively few specic studies looking at the biological mode of action/activity of natural products. 6,7 Many of the actives evaluated are well-known secondary plant metabolites of the polyphenol class; however, the majority of agents are non-specic anti-microbial agents, having an impact on a variety of oral microorganisms in a range of ways. The antimicrobial effects of such agents are generally measured as the minimum bactericidal concentration (MBC) which kills particular microbes, or the minimum inhibitory concentration (MIC) which prevents bacterial growth. However, in a ow through environment such as the mouth. the longevity and retentivity in the oral cavity (oen termed substantivity) of agents are also crucial to their efficacy. For example during toothbrushing, antimicrobial agents may be present at greater than the MBC or MIC for short periods. However, at sub-MIC concentrations, agents may have a range of more subtle, but still important effects, such as reducing microbial (re)growth, inhibiting key metabolic processes such as acid production, protease activities or polysaccharide synthesis interfering with bacterial adhesion. 8,9 Understanding where agents reside in the oral cavity, for how long, and to which components they bind are therefore crucial to improving oral care product efficacy. Such understanding could then lead to more rational approaches to optimising product formulation and active delivery. The reported difficulties when attempting to evaluate the efficacy of a natural product are predicated on the complexity of the oen crude mixtures of natural products available and the technical challenges of time-consuming purication steps. 10 The importance of basic chemical characterisation of small/natural extract molecules cannot be overstated, particularly where the synergistic effects of more than one component are being investigated. 6 Recent improvements in high resolution NMR instrumentation, coupled with the exceptional work conducted in the improvement of exible processing soware has made DOSY NMR an increasingly valuable tool in complex mixture analysis, 11-13 including biouids and drug preparations, simultaneously identifying and in part quantifying their components. [14][15][16] The variation in isolation and derivatisation methods has led to oen inconsistent data and, despite the excellent review by Koo et al. 6 there is clearly scope for the evaluation of DOSY NMR as a non-destructive analytical method, which requires little standardisation. The NMR experiments are efficient, with data being acquired in just over one hour from sample collection. This, coupled with the ability to discriminate NMR signals on the basis of size (hydrodynamic radius), can obviate the need for 1D NMR spectral "binning" of multiple chemical shi regions when attempting to identify principal components. Through internal standardisation, approximate data regarding the mass of a component can be achieved. The aims of this communication are primarily to consider the scope and limitation of DOSY NMR as tools for the characterisation of natural product adjuvants in saliva models and saliva/dentifrice slurries and to demonstrate the potential for internally standardised qNMR to be used to quantify individual component concentrations, without the need for protracted separation and purication steps. For the purpose of evaluating the application of DOSY NMR/qNMR the organic antimicrobial preservative 4-isopropyl-3-methyl phenol (IPMP) was initially used as a reference compound in different matrices: these included a simple surfactant solution, model saliva and saliva/dentifrice slurry.
Whilst MS methods have been used to analyse complex mixtures, the sample preparation and purication processes inuence how the molecules of interest interact with other components in solution. 17 MS n methods have been shown to be useful in these scenarios but the technique is still destructive. Use of conventional 2D techniques for molecular identication have been used to verify ambiguous signals and should be used in tandem with a technique such as DOSY. Whilst it is possible to use for example HSQC to separate 1H overlaps and aid in signal assignment, the possible interactions between the various components meant we were just as interested in characterising their environment as much as identifying components individual molecules. This will prove of particular value for our current research, which looks at the interactions of such molecules with proteins found in saliva in gingivitis. All of the above mentioned techniques were used in the cases of ambiguity. The interaction between SDS and the IPMP, whilst predictable, illustrates the merit of this technique in a simple system such as that discussed. The application of DOSY to identify these environments and also individual molecules has been previously reported for other complex mixtures and this work aimed to extend this methodology further.

Materials and methods
Being phenolic and having similar chemical and physical motifs to other naturally occurring phenolic compounds, IPMP provided a convenient model for DOSY NMR/qNMR method development. Using IPMP, DOSY NMR was used to observe the inuence of pH, sodium dodecyl sulphate (SDS) and salt concentration on the diffusion characteristics of different adjuvants and also their impact on the accuracy of the qNMR data at different concentrations of IPMP. It is worthy of note that the proportional change in diffusion for TSP was not signicantly greater than the observed change in diffusion for other internal standards owing to increased viscosity of saliva. Interactions with proteins seem fast on the NMR timescale and therefore not particularly strong. We acknowledge, however, that this may be a limitation in this study.

Simple surfactant solution
Unless otherwise stated, all chemicals were obtained from Sigma-Aldrich. Measurement of solutes was carried out on a Fisherbrand MH-214 balance. PBS Buffer solution was prepared (100 mM NaCl) and corrected to pH 6, 7, 8 & 9. The SDS was added to a concentration of 0.75% w/w and IPMP was dissolved directly in the buffer and made up to the concentration range 0.005%, 0.01%, 0.025%, 0.05% and 0.1% w/w. Each of these sample concentrations was prepared separately ve times and 3 qNMR experiments run on each to validate the precision of concentration calculations. Deuterated sodium trimethylsilylpropionate (TSP) was used as an internal reference standard at 14.4 mM (DOSY) and 7.4 mM (qNMR) concentrations.

Model saliva
Fresh model saliva solutions were prepared using the recipe as stated by Klimek et al. 18 and used on the same day of preparation. 18 The solutions were kept at 25 C and out of direct sunlight.

Model saliva/dentifrice slurry
The slurry was prepared by stirring 5 g of Aquafresh Ultimate Toothpaste (GSK, Brentford, UK) slurried in a solution of model saliva (8 ml) prepared as per Section 2.2. Five samples were prepared at each concentration (0.005%, 0.01%, 0.025%, 0.05% and 0.1% w/w wrt IPMP). The slurry was centrifuged at 4500 rpm (3089 grams) for 30 minutes to remove the solid-components of toothpaste such as silicas and titanium dioxide, and the supernatants only used for analysis. NMR experiments were undertaken in triplicate for qNMR on the supernatant.

Qualitative/DOSY NMR analysis
A Bruker Avance III 600 MHz NMR spectrometer with 5 mm TXI Probe and temperature control unit was used for all 1 H NMR experiments. 5 mm Bruker Single Use NMR tubes (Product Code Z117777) were used. All spectra were acquired on Topspin 3.0 (Bruker, Germany) and 64 000 complex data points were acquired over a sweep width of 10.3112 ppm using a stimulated echo bipolar pulsed eld gradient with 1 spoil gradient and 3-9-19 WATERGATE sequence (STEBPGP1S19). This was used to obtain the diffusion series with d ¼ 2.4 ms and D ¼ 100 ms. The relaxation delay was set to 7 s and the WATERGATE pulse duration was 1000 ms with 64 linear gradient steps, from 2-95% gradient intensity, each consisting of 16 scans.
Each sample was allowed to equilibrate within the NMR spectrometer for 5 minutes. All NMR experiments were carried out at 25 C. A sine bell shaped window function phase was applied over all data points prior to Fourier transformation (16 384 points) using Topspin 3.0 (Bruker, Germany). Diffusion data were processed using DOSY Toolbox, (Mathias Nilsson, Manchester University) and TSP was used for Lorentzian reference deconvolution. Individual peaks were tted exponentially aer a 2 nd order polynomial baseline correction was employed. 15 Errors in diffusion coefficient were calculated based on the standard deviation for each diffusion curve and are in line with the estimated error as reported for a similar mixture of ca. 0.1 Â 10 À10 m 2 s À1 . 19 The residual sum of squares for each of the diffusion curves was less than 5 Â 10 À3 in all cases. 1 H-1 H correlation spectroscopy (COSY) was used in addition to predicted M r data from the diffusion correlation to correctly assign the 1 H signals relating to the mixture of components for the commercial toothpaste/articial saliva samples. A 2D homonuclear shi correlation pulse sequence using gradient pulses for selection and multiple quantum ltering was used (cosygpmfqf). Size of F1 FID was 2048 with a sweep width of 7.7692 ppm and a dwell time of 161 ms and relaxation delay of 1.861 s to give an acquisition time of 0.3297 s. A sine bell shaped window function was applied over all data points prior to Fourier transformation, phasing and baseline correction using Topspin 3.0 (Bruker, Germany).
Matrix matched samples were spiked with reference standards to conrm the identity of the components where there was ambiguity. All reference standards were obtained from Sigma-Aldrich Ltd and used without further purication.

Quantitative NMR (qNMR)
The quantication of components in natural product mixtures through comparison with the internal standard TSP has already been reported. 10,11 A Bruker Avance III 600 MHz NMR spectrometer with 5 mm TXI Probe and temperature control unit was used for all experiments. 5 mm Single Use NMR tubes (Bruker, Product Code Z117777) were used. All spectra were acquired on Topspin 3.0 and 65 536 complex data points were acquired over a sweep width of 12.9909 ppm using a 90 pulse angle and an acquisition time of 4.2030 s. A sine bell shaped window function phase shied by 90 was applied over all data points prior to Fourier transformation, phasing and baseline correction using Topspin 3.0. The chemical shi of all data was referenced to the TSP reference 0 ppm. All spectra were acquired at 25 C. Three replicates of the qNMR experiments described were carried out for each of the ve IPMP samples and the 5 different concentrations. The average integral of all IPMP signals was used for quantication and all IPMP signals had a s/ n ratio greater than 200 making them acceptable for qNMR processing. Quantitative data obtained using this method has been shown to compare well to traditional LC-MS and LC-UV techniques. 20

DOSY NMR
3.1.1. Internal calibration. For globular molecules of similar density, it has been shown that the diffusion coefficient is proportional to cube root of the relative molecular mass. However, this assumes an even, spherical distribution of density and the Stokes-Einstein equation, on which this principle is based, assumes free diffusion, necessitating a coefficient be determined to innite dilution. As the diffusion coefficient varies with concentration and viscosity change, internal references can be used to eliminate the complications of these effects when determining molecular mass. This technique has been reported by Li, et al. 21 and is reviewed by Macchioni, Ciancaleoni, Zuccaccia & Zuccaccia 22 and allows the determination of the molecular mass of an unknown component by using internal reference standards. In the case of IPMP in PBS, IPMP (170 g mol À1 ), water (18 g mol À1 ) and TSP (150 g mol À1 ) were used to generate a calibration graph based on eqn (1), where D ¼ diffusion coefficient (m 2 s À1 ) and M r is relative molecular mass.
where the viscosity and density remain constant between samples, corrections to a and b can be avoided, however diffusion correlation was determined for each of the samples as such variables cannot be taken for granted. The correlation of log D and M r for IPMP in simple solution (r 2 ¼ 1) values for a and b constants of À0.37 and + 1.49 respectively. The value of a agrees well with the literature value but the value of b is substantially lower owing to the relatively high viscosity of water when compared to common organic solvents. This has been reported previously by our group. 10 Fig. 1 shows a generally good correlation of mass vs. diffusion for all components with the exception of IPMP in the presence of sodium dodecyl sulphate (SDS). As reported by Nilsson et al. 23 association of lipophilic molecules to SDS micelles substantially decreases the apparent diffusion of these molecules and can actually be used to resolve compounds of similar mass in carefully controlled matrix assisted diffusion experiments. The deviation from linearity indicated the association of IPMP with a molecule/aggregate of substantially higher M r .
3.1.2. Characterisation in simple buffer. The control solution of IPMP in PBS shows uniform diffusion for those signals correlating to IPMP and the TSP control (Fig. 2). Standard deviation of the diffusion coefficients ranges from 0.1-0.2 Â 10 À10 m 2 s À1 . It should be noted that the lipophilic nature of IPMP demanded a solution of higher pH than would normally be encountered in saliva in order to ensure complete dissolution. This is reected in some of the qNMR data discussed in Section 4. The DOSY spectrum for IPMP in the presence of SDS has been overlaid in Fig. 2 and highlights the substantial shi in diffusion coefficient for IPMP when compared to TSP. TSP is used as both a chemical shi standard and diffusion standard to enable compensation for changes in viscosity and therefore the disproportionately low diffusion coefficient for IPMP suggests association with SDS micelles. The change in diffusion rate cannot be ascribed to an increase in overall viscosity as correction for this with TSP still results in a statistically significant reduction of the diffusion coefficient for IPMP.

Characterisation of saliva.
The impact of articial saliva, as a complex mixture of salts and protein, on diffusion was analysed and the pseudo 2D DOSY plot (Fig. 3) provided similar diffusion coefficients for IPMP signals to those found for SDS alone. The standard deviations for the determined diffusion coefficients and the signal resolution are similar to those observed in the absence of articial saliva. Despite a high ionic strength, which has the potential to signicantly attenuate proton signal strengths, it was possible to obtain a good quality DOSY map of the components IPMP, SDS and TSP in articial saliva. TSP has a similar diffusion value in both cases, implying that the overall viscosity changes due to the introduction of articial saliva are relatively small.
3.1.4. Characterisation of dentifrice slurry. Having characterised IPMP in the presence and absence of a surfactant and established TSP as an internal reference for viscosity and veried a correlation between log M r and log D for different systems, samples of commercial toothpaste (n ¼ 3) were analysed to validate the capacity of DOSY NMR for virtual separation of components of dentifrice slurry. Fig. 4, 5 and 6 show aliphatic, carbohydrate and aromatic regions of the averaged pseudo 2D DOSY plots. By determining the constants a and b for the Fig. 1 Correlation of log D values of components to log M r in A phosphate buffered saline (pH 9),glycerol and sodium dodecyl sulphate (pH 8) : glycerol, sodium dodecyl sulphate and artificial saliva, and X toothpaste and artificial saliva (pH 8). IPMP was dosed in all cases aside from the toothpaste. R 2 value for all range from 0.95-1. Ellipsoid indicates IPMP outliers due to binding to sodium dodecyl sulphate micelles. Decreasing trend line gradient is shown with viscosity. log D/log M r correlation it was possible to directly extract approximate mass data for lipophilic components with reasonable accuracy (AE4%). These predicted mass values coupled with chemical shi and coupling data from 1 H and 1 H 1 H COSY experiments were used to assign the signals in the DOSY plots as shown.

qNMR
Quantitative NMR (qNMR) was rst described in 1963 and is increasingly regarded as a powerful non-destructive technique with traceability to SI units, providing advantages over many other analytical methods. 24,25 The development of affordable high eld NMR instruments coupled with more elegant pulse sequences has prompted a recent urry of activity in the eld of NMR analysis, including human and plant metabolomic studies. 26,27 A recent review of the applications of qNMR over the last eight years highlights its growing importance in the context of metrology and supports the statement that qNMR can be regarded as a primary method of purity analysis for organic compounds. 28 Using TSP as the internal standard, not only for  diffusion but also for quantication, enables the calculation of unknown analytes by comparing the integral of known and unknown signals. Despite growing acceptance of this technique, there are acquisition and processing factors which can contribute to uncertainty when calculating concentrations of analytes. When dealing with complex mixtures with oen high ionic strength (Q-factor), accuracy can be substantially reduced 29 and it is important that the matrix in which the analytes appear is properly characterised. A key part of the current work is to consider the impact of high ionic strength, viscous matrices and the direct impact on the ease with which qNMR experiments can be efficiently and accurately undertaken. Fig. 7 shows the effect of pH and co-solvents/surfactants on the accuracy with which IPMP can be quantied. The impact  of viscosity and high salt content from the articial saliva on the validity of this technique to quantify IPMP is shown in Fig. 8. The direct application of qNMR in dentifrice slurry is evaluated in Fig. 9. The range of concentrations was deliberately reduced in size to highlight the limits of quantication using this method with potential applications in real-world saliva samples. The impact of pH on the correlation graphs is shown in all cases.

Virtual separation of the components
Whilst the virtual separation of small molecules in mixtures using DOSY NMR is established, 14 we report the analysis of dentifrice slurry using DOSY NMR for the rst time. Use of TSP as an internal diffusion standard enables the association of the IPMP to SDS micelles to be clearly observed (Fig. 2) and for simple systems where there is no competition it is possible to extract information as to what extent association is occurring between host and guest based on eqn (2), discussed in the review by Fielding. 16 Where D obs ¼ observed diffusion coefficient, X I ¼ mole fraction of free IPMP, X IS ¼ mole fraction of bound IPMP, D I ¼ diffusion of free IPMP, D IS ¼ diffusion of Bound IPMP.  Using the drop in diffusion coefficient for TSP upon addition of SDS as a marker of increased viscosity, it is possible to correct the unbound Diffusion coefficient and calculate the approximate mole fraction for bound and unbound fractions of IPMP. The extent to which the diffusion coefficient is reduced as a consequence of association with SDS micelles (M r $ 18 000), suggests that 60% IPMP is bound to SDS and 40% unbound. It should be stated that the lack of additional signals for IPMP in the presence of SDS suggests fast exchange on the NMR timeframe. This nding is in good agreement with the measured exchange rates of sparingly soluble uorescent probes in SDS micelles, which has been previously measured on the microsecond time frame. 13 Whilst the inclusion of articial saliva does little to impact the separation of the organic components, the limitation of DOSY NMR becomes more apparent when commercial toothpaste is included. It has previously been reported 10,14 that it is possible to correlate M r with D and this appears to be independent of viscosity. From Fig. 1 it is possible to observe a trend for Citrate, TSP, and glycerol and obtain a and b constant values of À0.803 and 2.332 respectively. This enabled the prediction of M r from log D values and this correlation coupled with 1 H 1 H COSY data enabled the identication of saccharin, sorbitol and urea in the dentifrice slurry (Fig. 5). A linear trend for the log D vs. log M r exists, and this can be observed in this case for the suitably hydrophilic components.
However, the lipophilic components, which are associated with SDS micelles, appear to deviate from the linear trend, rendering prediction of M r with any accuracy difficult. The attempt to extrapolate approximate M r values for unknown components using a log D vs. log M r would appear limited to those components that do not associate with SDS micelles. Some of the components, which appeared micellised, could be identied through 1 H 1 H COSY and included eugenol (Fig. 6) and (À) menthol.
Recent work on the virtual separation of regioisomers of methoxyphenol using SDS implies that differences in diffusion can be ascribed to different relative affinity for the SDS micelles, log P values of the compounds in Fig. 10 were evaluated using ChemOffice Chemdraw Ultra 12. 22 It could be hypothesised that a greater log P could result in greater affinity for the non-polar component of SDS micelles and result in a reduced observed diffusion coefficient (as originally observed for IPMP). The variation in diffusion coefficient between these three molecules should, therefore correlate approximately to their individual lLog P values, however, the diffusion coefficients observed by this group actually show a correlation (R 2 ¼ 0.98 of log D to log M r ) for these three molecule series. Whilst further work will be necessary to determine the factors which contribute the most to the observed change in diffusion coefficient of a lipophilic molecule when in the presence of a surfactant, this preliminary data suggests it may still be possible to predict M r values based on log D data in the presence of surfactants. In mixtures such as these, not only are a small series of known hydrophilic diffusion standards necessary for the establishment of a log D vs. log M r correlation, but also that a series of lipophilic diffusion standards may be necessary for a reliable extrapolation of M r values for unknown compounds based on log D data. It should be stressed that further studies will need to be conducted to look at the effects of residency time in the micelles and the potential impact of any complexed lipophilic components on shape and thus apparent diffusion.

Reliability of qNMR
One of the initial hypothesis was that qNMR was a valid technique for the accurate quantication of organic species in a  complex, viscous mixture. Reference to Fig. 7 indicates that pH has a substantial impact on the accuracy of the technique and that proper dispersion of the analyte is essential for correlation. This is reected in the discrepancy between data shown in Fig. 7 compared to samples containing SDS. Even at pH 9, the IPMP phenoxide is insoluble at high concentrations >0.05% w/w. It should be mentioned that other groups have focused on instrumental uncertainties (type B) 30 to account for the observed precision, however for this work, they are largely dwarfed by the inherent error in the weighing balance and the type A uncertainties, so their relevance in this case was questionable. In light of the lipophilicity of the IPMP, SDS was essential for its dispersion and for correlation of the calculated concentration to the amounts added. The accuracy of the qNMR method was 98-99% based on the weighed amounts of IPMP for each % concentration for an SDS/glycerol mixture in PBS, but in the presence of ZnCl 2 , the accuracy fell to 85% at higher concentrations of IPMP with an increase in error to AE1 mM, suggesting insoluble Zn 2+ salt formation over time. The exchange of Zn 2+ with Na + could also affect the size and shape of the SDS micelle, this may result in the promotion of longer rods, rather than spheroids; however, if this is a uniform occurrence throughout the solution it should have equal impact on lipophilic species. Surprisingly the qNMR correlation for IPMP in articial saliva (Fig. 9) and dentifrice slurry (Fig. 8) showed a good correlation at all concentrations. The accuracy was affected by pH with a lower pH favouring accuracy of 98-99% for IPMP in the dentifrice slurry and a higher pH favouring accuracy of 96-98% in the articial saliva, SDS, glycerol model. There was no statistically signicant difference in the error margins for the qNMR calculation of concentration of IPMP for either of the two systems at low concentration.

Conclusion
We have demonstrated that DOSY NMR is a useful tool for the characterisation and quantication of natural products as represented by IPMP in systems of variable viscosity and in the presence of surfactants and high salt concentration. Whilst it is important that multiple diffusion standards should be used when attempting to identify unknown components based on predicted M r from log D data, the use of qNMR to determine accurately the concentration of dissolved organic shows considerable promise for analysing complex mixtures of natural products. Through the non-destructive identication and quantication of these components, some of the anti-plaque or antimicrobial effects of polyphenolic and terpinoid derivatives could potentially be investigated directly in situ and without the need for separation and the potential loss of key components, which may contribute to this activity. Consistent and accurate quantitative data is possible for these systems and this will facilitate the investigation of the combined impact of individual components with the potential to assist in the development of a combination of natural products with enhanced anti-plaque activity.