ANALYST, JANUARY 1993, VOL. 118 73 Use of Rapid Scan Correlation Nuclear Magnetic Resonance Spectroscopy as a Quantitative Analytical Method Herve Barjat, Peter S. Belton and Brian J. Goodfellow* AFRC Institute of Food Research, Norwich Research Park, Colney, Norwich, Norfolk, UK NR4 7UA Rapid scan correlation (RSC) nuclear magnetic resonance (NMR) spectroscopy combines the advantages of rapidity of Fourier transform methods with advantages in dynamic range of continuous wave methods. A systematic investigation of the technique for quantitative analysis has been carried out. It is concluded, provided suitable precautions are taken, that RSCNMR can be useful as a quantitative method. Application is illustrated by the determination of ethanol in aqueous media and wines over the range 0.01-15% v/v.Keywords: Rapid scan correlation nuclear magnetic resonance spectroscopy; ethanol; quantitative analysis Although there are a wide range of analytical applications of high resolution nuclear magnetic resonance (NMR) spectro- scopy,l its use for quantitative analysis in a quality control environment, as opposed to a research environment, has been limited. There are two main reasons for this: the first is the high cost of modern Fourier transform (FT) spectrometers and the second is related to the way in which an FT system collects data. In FTNMR all the frequencies of the spectrum contribute to each point of the time domain signal. This results in a multiplex advantage,Z which gives higher signal-to-noise ratio (S/N) spectra but also means that the intensity of one very large signal may dominate all the others.Hence small signals are insufficiently digitized to appear in the spectrum. This problem is generally referred to as the dynamic range problem. It is often encountered with a dilute solute in the presence of a solvent. The problem can be solved by isotropic dilution of the solvent, which may not be practical, or by a solvent suppression technique. Although the latter approach is often used it is not without its own problems.3 The cost and dynamic range problems of FTNMR can be overcome by using continuous wave (CW) NMR. Continuous wave NMR spectrometers are relatively inexpensive and by scanning only the region of the spectrum of interest, intense peaks can be avoided. Major problems with CWNMR, however, are the poor signal-to-noise ratio and speed of data acquisition.If these drawbacks could be overcome, CWNMR would be an attractive technique for routine quantitative use. Rapid scan correlation (RSC) NMR is a technique that combines the speed advantage of FTNMR with the low cost and partial scanning capability of CWNMR. The principles of RSCNMR were first outlined in 1974.4.5 In normal CWNMR the spectrum is acquired by sweeping the magnetic field while irradiating the sample at a constant radiofrequency (r.f.). If the field is swept too fast, distortion of the peaks in the spectrum occurs. This distortion, or ringing, can only be avoided if the scan rate is substantially smaller than l/T1 (spin-lattice relaxation time) or l/T2 (spin-spin relaxation time) of the sample under study.This means that 1 scan at high resolution can take about 10 min to acquire. In order to co-add scans and improve signal-to-noise ratios, the scan rate must be fast. A fast scan, however, results in the distorted spectrum shown in Fig. 1 (A). In order to interpret and analyse this spectrum the distortion needs to be removed. The origins of effects that give rise to ringing are well understood and an expression for it can be obtained which is of the forrn"35 ~ ( t ) = e-Iht* (1) where h is the sweep rate in rad s-1 and t is the time after peak maximum. The ringing is removed from the spectrum by * To whom correspondence should be addressed. cross-correlating eqn. (1) with the inverse FT of the rapidly scanned spectrum. Cross-correlation involves multiplying the complex conjugate of one function by another.After cross- correlation the time domain function is Fourier transformed to produce a spectrum in the frequency domain with the ringing remo:red [Fig. l(B)]. Hence, by co-adding a number of rapidly scanned spectra and then cross-correlating it is possible to obtain undistorted spectra with excellent S/N. The above process, i.e., rapid scan followed by correlation, takes place on a timescale comparable to that of FTNMR. Rapid scan correlation NMR, therefore, has the speed advantage of FTNMR and partial scanning advantage (avoiding dynamic range problems) of CWNMR. It does, however, lack the full multiplex advantage of FT techniques. In order to test the potential of this method as a quantitative analytical technique, the proton NMR spectra of ethanol in water over a wide concentration range were examined under different acquisition conditions.This system was chosen as being typical of a real analytical problem in the food industry and as typifying the problems of large solvent and small solute signals. Experimental A Hitachi (Nissei Sangyo) R-1200 rapid scan NMR spec- trometer operating at 60 MHz for protons was used for all B 1 - b Fig. 1 correlation Spcctrum of 5% ethylbenzene in CCI?: A, before and B, after74 ANALYST, JANUARY 1993, VOL. 118 experiments. The permanent magnet operated at 35 "C. All samples were equilibrated at 35 "C in a heated chamber, provided with the spectrometer, before measurement. Resol- ution was adjusted using a tetramethylsilane (TMS) sample.The ethanol calibration solutions (2.85-14.25% v/v) were prepared using 95% ethanol (Hayman) by pipetting the relevant amounts of ethanol into calibrated flasks. Solutions of concentration 2.85, 4.75, 7.60, 11.40 and 14.25% v/v were obtained. The lower concentration standards were prepared by dilution of the above stock solutions to give solutions of 9.5 X 10-3, 2.85 x 10-2, 4.75 x 10-2, 6.65 x 10-2, 9.50 x 10-2 and 19.0 x 10-2% v/v concentration. The wine and beer samples were obtained from a local retail source. The samples were stored at + 1 "C and shaken before being pipetted directly into a 5 mm NMR tube. Independent determinations of the ethanol contents of the wine and beer samples were carried out by Lincolne, Sutton and Wood, Analytical Consulting Chemists, Norwich. In order to adjust acquisition parameters for a whole spectrum, a sweep width of 10 pprn (600 Hz) was selected with a scan rate of 10 s per scan and a filter width of 200 Hz.After correlation to produce the undistorted spectrum the phase was adjusted and r.f. power and receiver gain were optimized to produce the best signal. In order to determine the best conditions for partial spectral acquisition, a variety of experimental conditions were tried. Details of these are given under Results and Discussion. When a number of scans are co-added to improve the S/N, it is important that any field drift is eliminated or accounted for as this could produce unacceptable line broadening. This can be effected by adding a deuteriated solvent to the sample and using a deuterium lock. However, with the R-1200 spec- trometer it is possible to use a technique known as 'field cure' in which a reference peak is selected and each subsequent spectrum is shifted by the computer software so that the reference peaks are in alignment.In samples with suitably large signals one of the spectral peaks was chosen. With dilute ethanol samples there was insufficient signal and 0.12% v/v acetone was added to act as a reference for these samples. Results and Discussion The spectrum of ethanol consists of a triplet due to the CH3 group at 1.7 pprn and a quartet due to the CH2 group at 3.7 ppm. The signal from the hydroxyl proton is merged with the water peak. The peak at 1.7 pprn was chosen for detailed study as it is the most remote from water and has contributions from the largest number of protons.In order to test the quantitative response of the spectrometer, the effects of a number of different experimental variables were examined. The first of these was the effects of the irradiating r.f. power. The acquisition conditions for this experiment were: sweep width , 100 Hz; sweep rate, 15 s per scan; and filter, 200 Hz. In general, higher powers result in increased signal intensity , but excessive power can lead to line broadening and loss of signal intensity owing to saturation effects; hence the highest r.f. power which avoids saturation should be chosen. This consideration is complicated, however, by the interfering effects of the water signal, because, although the water peak maximum was not irradiated, the weak signal in the wings of the Lorenztian line was inevitably irradiated and at high power gave rise to a very uneven baseline, making quantification difficult.At the other extreme of r.f. power level it was found that the lowest settings gave insufficient signal for detection in samples with less than 0.475% v/v ethanol. The choice of irradiating power must therefore be a compromise between over-all signal intensity and baseline quality. However, the latter problem may be improved by use of baseline fitting routines, which were not available to us in these experiments. In order to obtain the best S/N it is desirable to adjust the amplification of the signal so that the analogue-to-digital (ND) converter is filled. On the other hand it is desirable for calibration purposes to leave instrumental parameters unc- hanged.It was found that a single amplification factor was inadequate to cover the whole range of ethanol concentrations investigated (0.01-15% v/v) and hence the samples were examined in the ranges 0.01-3 and 1-15% v/v. These represent two ranges of practical interest, for example, 'alcohol-free' drinks and the normal range of ethanol in wines and beers. The sweep rate and filter width are interdependent parameters. The fastest acquisition requires the maximum sweep rate, but if the spectrum is to be undistorted then an increase in sweep rate must be accompanied by a correspond- ing increase in filter width, which will degrade the S/N. In Fig. 2 the effects of filter width from 1 Hz [Fig.2 (F)] to 200 Hz [Fig. 2 (A)] are shown for a spectrum of 1% v/v ethanol with a sweep rate of 5 s per scan (this is equivalent to the 5 s taken to scan a 600 Hz spectrum and is equivalent to a scan rate of 8.3 ms Hz-1). There is little difference between the 100 and 200 Hz filter widths but clear broadening at 5 Hz width. Below this value further signal distortion is evident, but signals are still measurable. It may, therefore, be possible to improve S/N in the spectrum at the cost of signal distortion while still retaining sufficiently recognizable data to be useful. Probably a better strategy would be t o capture the data using fairly broad filter settings and using digital filtering methods in the software. When the filter width is kept constant and the scan rate is increased there appears to be some spectral distortion at the highest scan rates (Fig.3). In principle this should not occur as even in Fig. 3 (D) the scan rate is only equivalent to 200 Hz s-1, which is similar to the 200 Hz filter width chosen. Filter effects are not apparently the only difficulty as at the higher scan rates there also appears to be baseline distortion. The origins of these effects are not clear but the results clearly indicate that lower scan rates are likely to give better signals. In practice the choice of low scan rates over a limited spectral range is not such a severe problem because the time taken for a 60 Hz (1 ppm) scan for the fastest rate is 0.3 s and for the slowest 1.5 s. Even for a 256 scan acquisition, therefore, the time difference is only a few minutes.An important feature of the RSCNMR method is that many signals can be co-added to improve signal-to-noise ratios. Fig. 2 of A, 200; B, 100; C , 5 ; D, 4; E, 3; and F, 1 Hz Methyl triplct of cthanol in watcr (1% v/v) with filter settingsANALYST, JANUARY 1993, VOL. 118 75 - 6 Fig. 3 A , 15; B, 10; C. 5 ; and D, 3 s per scan Mcthyl triplct of ethanol in water (1% v/v) with sweep rates of 8 z, in 0 0 0 10 20 30 40 N: Fig. 4 field drift compensation Plot of S/N versus square root of thc number of scans without This is only useful, however, if there is no magnetic field drift between scans. In order to test this, a series of spectra were obtained on a 0.95% v/v ethanol solution at a high scan rate over the complete 600 Hz range. The plot of SIN versus the number of scans is shown in Fig.4. Clearly the expected N112 dependence is not observed. This is due to field drift effects causing the spectral lines to widen as the number of scans increases. However, when the field cure system is used a good linear relationship is obtained (Fig. 5 ) even though the spectral range has been reduced to 300 Hz to exaggerate any field drift effect. Rapid scan correlation NMR is, therefore, able to deliver the expected S/N enhancement on co-adding scans. Good quantitative results depend on precise peak area measurements. Ten integrations were carried out for the 14.25 100 150 i- I $ 1 50 0 0 0 0 0 I . 0 I I I 0 2 4 6 8 N: Plot of S/N versus square root of the number of scans with field Fig.5 drift compensation 500 Gi 400 4d .- t 3 > 5 300 c .- e m 200 2? - Y (0 a" 100 0 2 4 6 8 10 12 14 16 Ethanol in water (% v/v) Fig. 6 Calibration graph for ethanol in water over the concentration rangc 2.85-14.25% v/v. The gradient of the line is 32.95 with an intercept of 13.02. The correlation coefficient is 0.999 and 7.6% v/v ethanol standards to check the variation. It should be noted at this point that the phasing of the spectrum is critical for good integration; hence the phase of the signal was carefully adjusted before data acquisition. For the 14.25% v/v sample the mean was 480.9 with a standard deviation of k2.92 and for the 7.6% v/v sample the values were 269 and k2.4, respectively. The 15-3% v/v calibration was carried out with the power level set so that the lowest concentration sample gave a suitable integration value (~100).The amplitude was set so that the signal from the most concentrated sample just filled the A/D converter. Sixteen scans were co-added with a sweep rate of 15 s per scan. Fig. 6 shows the calibration obtained. Each spectrum in the calibration took about 60 s to acquire, resulting in the whole calibration taking less than 15 min. After the calibration had been obtained, the wine and lager samples were run. The results are presented in Table 1. The calculated results show good agreement with those given by the manufacturers and demonstrate that the tech- nique is quantitative in real systems of interest. The reproducibility of the calibration graph was tested by running the calibration over a number of days.Seven calibrations were carried out giving mean values of 33.64 k 1.32 for the gradient, 8.35 k 11.33 for the intercept and 0.9992 ? 0.0005 for the correlation coefficient of the line. Although the variations from day to day are fairly small, the fluctuation of the intercept means the calibration needs to be run each day. However, as the whole calibration takes less than 15 min to acquire, this is practical. By using the calibrations obtained over the 11 d period, the commercial samples (see Table 1) gave mean values of 12.0 k 0.20 (C6tes76 ANALYST. JANUARY 1993, VOL. 118 Table 1 Results for ethanol content in wine and becr samples Calculated Peak area concentration of Concentration (arbitrary ethanol given on label Sample units) (Yo v/v) (Yo v/v) C6tes du RhBne 41 1 12.1 12.0 Muscadet 413 12.1 12.0 Nierstcincr 31s 9.2 9.0 Lager 192 5.4 5.0 Table 2 Comparison of results obtained with different techniques for ethanol content in wine and beer samples.All values in % v/v Distillation/ specific Label Sample RSCNMR GC gravity value CBtes du RhBne 12.4 12.0 12.1 12.5 Muscadet 12.5 12.1 12.1 12.0 Niersteiner 9.3 9.2 9.2 9.5 Lager 5.2 5.1 5.1 5 .O 6.0 5.0 4.0 3.0 2.0 1.0 0 -1.0 6 (PPm) Fig. 7 shows result of increased amplification and partial scanning Spectrum of 0.475% v/v ethanol in water. Expanded region du RhGne), 11.7 k 0.28 (Muscadet), 8.9 k 0.20 (Niersteiner) and 5.1 k 0.20 (lager). In order to obtain a comparison between RSCNMR and two standard chemical methods, new wine and beer samples were analysed by RSCNMR, distillation/specific gravity and gas chromatography (GC); the last two analyses were carried out in a commercial analytical laboratory under standard condi- tions.The results are shown in Table 2.1- It can be seen from Table 2 that the various methods give results that are in good agreement with the amount of ethanol givcn by the manufacturers. It should be noted here that the ethanol content given by the manufacturers in Table 2 differs from that given in Table 1. The brands of drinks purchased were identical for the two sets of experiments; however, they were purchased at different times. The different ethanol contents must, therefore, result from different batches being purchased. The errors involved for each method are: RSCNM R, rt0.22% v/v; distillation/specific gravity (relative density), +0.35% v/v for the wine samples and rf-0.14% v/v for the beer.The different errors obtained for the distillation/ -1 The reference method (Commission regulation, EEC. number: 2676/90) is distillation followed by density measurement at 20 "C using a pycnometer. Specific gravity measurements may also be used and the results converted to YO v/v using tables given. The repeatability of this method is 0.10% v/v and the rcproducibility 0.19% v/v (using trained staff). I I 2.0 1 .o (PPm) Fig. 8 D, 0.0665; and E , 0.095% v/v Spectra of ethanol in water at A, 0.0095; B, 0.0285; C, 0.0475; specific gravity method for the wine and beer samples reflected the different number of repeat measurements taken (14 for the beer and 6 for the wine).The error for the GC method was not given. These results indicate that even at this early stage in the development of RSCNMR the results obtained are at least as good, within experimental error, as those obtained by standard methods used for ethanol determi- nation for a range of wines and beers. The low level (0.01-0.3% v/v) ethanol calibration required more careful adjustment and set up. Fig. 7 shows the single scan spectrum of 0.475% v/v ethanol in water, with the power and amplification adjusted to optimize for the water signal. The methyl signal at 1.2 ppm cannot be seen. However, by partially scanning the methyl region and increasing the irradiation power and the amplification the signal can clearly be observed (see expansion in Fig.7). This clearly illustrates the advantages of partial scanning. The low concentration ethanol solutions gave, even under optimized conditions, very weak signals. The spectra for five of these solutions are shown in Fig. 8. In order to obtain a signal the irradiation power level was near the maximum. The problem with these high power levels is that the tail of the water peak extends into the methyl region and gives a distorted baseline. This also gives problems with peak height and peak area measurements. Shimming, for low concentra- tion samples, is also important to reduce spinning side bands, which can interfere with peak measurement. For these low level ethanol concentrations the signals were too small to be used as a reference for the field cure function. Acetone was added (0.12%) and used as a reference peak.In order to obtain sufficient S/N, 64 scans were co-added with high power and a 45 s per scan sweep rate. The intensity of the ethanol triplet at these concentrations was too weak for adequate integrations; hence peak heights (central peak of triplet) were measured. The resulting calibration graph gave a line (grad-ANALYST, JANUARY 1993, VOL. 118 77 ient 280.0, intercept -0.45 and correlation coefficient 0.999) with excellent linear correlation. Each spectrum in this set took 17 min to acquire, hence this is not a rapid calibration. However, it does demonstrate the sensitivity of the instrument. A 0.0095% v/v ethanol sample with 256 scans co-added gave an S/N of 10 for the central peak. Conclusions The results obtained suggest that RSCNMR has considerable potential as a quantitative analytical method. The spec- trometry is similar to that in conventional NMR techniques but has the advantages of high speed and relatively low cost. A wide range of concentrations can be determined and further developments in data handling arc possible. There arc problems with spectrometer drift, and the origins of these are not yet fully understood. However, even within the existing limitations the accuracy and speed of the instrumentation make it suitable for a wide range of analytical applications. The authors thank Nissei Sangyo Co. Ltd. for the loan of the R-1200 spectrometer and Dr. P. Gadsby for advice and assistance. References Analytical NMR, eds. Field, L. O., and Sternhell, S., Wiley, Chichester, UK, 1989. Marshall, A. G., and Verdun, F. R.. Fourier Transforms in NMR, Optical and Mass Spectrometry, Elsevier, Amsterdam, 1990, p. 98. Hore, P. J . , Methods Enzymol., 1989, 176, 64. Gupta, R. K., Ferretti, J. A., and Becker, E. D., J . Magn. Reson., 1974, 13, 275. Dadok, J., and Sprecher, R. F., J . Mugn. Reson., 1974,13,243. Paper 2103688F Received July 13, 1992 Accepted September 30, 1992