Vanillin - American Chemical Society

May 21, 2004 - (1) De Niro, M. J.; Epstein, S. Science 1977, 197, 261-263. (2) Monson, D. K.; Hayes, J. M. Geochim. Cosmochim. Acta 1982, 46, 139-. 14...
0 downloads 0 Views 108KB Size
Anal. Chem. 2004, 76, 3818-3825

NMR Approach to the Quantification of Nonstatistical 13C Distribution in Natural Products: Vanillin Eve Tenailleau,*,† Pierre Lancelin,‡ Richard J. Robins,† and Serge Akoka†

1- LAIEM -CNRS UMR 6006, Faculte´ des Sciences et Techniques, Universite´ de Nantes, 2 rue de la Houssinie` re BP 92208, 44322 Nantes cedex 3, France, and CRL-Rhodia, 85 rue de Fre` res Perret-BP 62, 69192 Saint-Fons cedex, France

Quantitative 13C NMR conditions have been established that permit the precise determination of site-specific 13C/ 12C ratios at low or natural abundance. Spectral acquisition parameters have been optimized in order to obtain minimum intensity distortions over the spectral width and in relation to the major sources of inaccuracy: the relaxation times, the decoupling pulse and power, and nuclear Overhauser effects. A major reduction in experimental time resulting from a study of the relaxation times and variance analysis has been achieved. The influence of 1H decoupling conditions on peak areas was shown to be critical in that different relative peak areas are obtained according to the decoupling power. The efficiency with which the quantitative 13C NMR method can determine site-specific 13C/12C ratios in natural products has been tested for 12 independent samples of vanillin from different sources. Discriminatory analysis performed in the space defined by the site-specific carbon isotope ratios allows natural vanillin and that from different synthetic origins to be unambiguously distinguished. The growing importance of 13C as a general marker of molecular origin, both in the medical domain and in a broad range of biological studies, makes it increasingly pertinent to refine methodology for the quantitative determination of 13C/12C ratios. In particular, the challenge is to obtain precise values for 13C levels close to or at natural abundance. This is especially desirable for whole-body studies in human metabolism, when tracer costs can become prohibitively high, or where intramolecular 13C distribution is to be determined at natural abundance. Isotope 13C/12C ratios can be obtained by isotope ratio mass spectrometry (IRMS) with great accuracy, but this technique provides only the overall mean. Frequently, these are insufficient and critical site-specific data are lost. Site-specific isotope ratios of carbon at very low abundance have been determined by IRMS following time-consuming chemical degradations into smaller molecules and the isolation of molecular fragments.1,2 However, such processes are seldom entirely satisfactory, and it is very * To whom correspondence should be addressed. E-mail: eve.tenailleau@ chimbio.univ-nantes.fr. Fax: (33) 251125712. † Universite´ de Nantes. ‡ CRL-Rhodia. (1) De Niro, M. J.; Epstein, S. Science 1977, 197, 261-263.

3818 Analytical Chemistry, Vol. 76, No. 13, July 1, 2004

difficult to avoid undesirable fractionation effects.3 On-line pyrolysis techniques show promise for small molecules4 but are inadequate for larger chemical species. In contrast, the direct determination of site-specific isotope ratios of carbon at natural abundance by 13C NMR spectroscopy has the potential to avoid these problems and is a priori very attractive.3 Due to the wide range of chemical shifts and the narrow line width in 13C NMR, peak overlap is unusual, allowing the simultaneous determination of all carbon positions present. There is virtually no limit to the size of the molecule studied, since relaxation times decrease with molecular size. Nevertheless, to measure natural or very low-enrichment 13C/12C isotope ratios with a suitable degree of accuracy has remained a challenging problem. Quantitative 13C NMR spectrometry suffers from a restricted range of deviation in 13C/12C isotope ratios (50‰ on the δ scale),5 from a wide range of chemical shifts, from long longitudinal relaxation times, and from site-specific nuclear Overhauser effects (NOEs).3 Moreover, accuracy is very sensitive to the modulation of the spectrum by imperfect broad-band decoupling. All these sources of error were highlighted by the pioneering work of Caer et al. in 1991.3 However, largely due to improved equipment, by 1999 Zhang et al. were able to show the consistency of IRMS and NMR for the small molecule, glycerol.6 In this paper, we report the development of a generic approach to obtaining accurate 13C/12C ratios. Each parameter liable to introduce inaccuracy has been optimized to minimize error. Particular attention has been paid to the relaxation times, decoupling pulse and power, and NOEs. While it is always tempting to develop novel methodology on model systems, we have opted to study the aromatic phenylpropanoid derivative, vanillin. This compound has major importance in the food and flavor industry for its organoleptic qualities. Furthermore, the vanillin molecule presents several typical technical difficulties for 13C NMR. It has a wide range of chemical shifts in carbon (135 ppm) and in proton (6 ppm), long relaxation times (2) Monson, D. K.; Hayes, J. M. Geochim. Cosmochim. Acta 1982, 46, 139149. (3) Caer, V.; Trierweiler, M.; Martin, G. J.; Martin, M. L. Anal. Chem. 1991, 63, 2306-2313. (4) Yamada, K.; Tanaka, M.; Nikagawa, F.; Yoshida, N. Rapid Commun. Mass. Spectrom. 2002, 16, 1059-1064. (5) Craig, H. Science 1961, 133, 1833-1834. (6) Zhang, B.-L.; Trierweiler, M.; Jouitteau, C.; Martin, G. J. Anal. Chem. 1999, 71, 2301-2306. 10.1021/ac0496998 CCC: $27.50

© 2004 American Chemical Society Published on Web 05/21/2004

Figure 1. Vanillin molecule with carbon sites numbered in decreasing chemical shift (a) and 13C NMR spectrum of vanillin in acetone recorded at 125.76 MHz with an interpulse delay of 131 s and a decoupling power of 0.8 W (b). ac ) acetone.

(up to 19 s), and a wide range of NOE factor. Hence, methodology sufficiently robust for the determination of the site-specific 13C/ 12C ratios of vanillin should be easily applied to 13C quantitative measurements in other molecules. Having shown that satisfactory conditions for determining the 13C/12C ratios at all sites of vanillin can be established, the efficacy of the method has been rigorously tested by comparing samples of vanillin from different sources. The challenge here was to determine at what level of precision it is possible to distinguish different metabolic backgrounds for the same molecule on the basis of the variation in the 13C/12C ratios. It has been shown that acceptable analysis times can be obtained without loss of accuracy. EXPERIMENTAL SECTION Material. Twelve independent samples of vanillin were obtained from the following sources: extract from pods (2); hemisynthetic, produced from lignin (3); synthetic, produced from guaiacol (3); biotechnological, produced from natural ferulic acid by bacterial biotransformation (4). Sample Preparation. A total of 500 mg of each sample was dissolved in 1 mL of acetone (acetone-h6 + acetone-d6; 1:1). NMR Experiments. The quantitative 13C NMR spectra were recorded using a Bruker DRX 500 spectrometer fitted with a 5-mmi.d. dual probe 13C/1H carefully tuned at the recording frequency of 125.76 MHz. The experiments were performed at 298 K. The experimental parameters for 13C NMR spectral acquisition were the following: pulse width 4 µS (90°), interpulse delay D ) 131 or 21 s, scan number 92, spectral width 27 500 Hz (220 ppm), sampling period 1 s. Four spectra were recorded for each measurement. Inverse-gated decoupling techniques were applied in order to avoid any NOE. WALTZ-16 was used as decoupling sequence with decoupling pulses of 65 or 35 µs corresponding to a decoupler power of 0.8 or 2 W, respectively. The offset of the decoupler was placed at 5 ppm in the proton chemical shift scale. The T1 values were determined by using an inversion recovery sequence, with 12 inversion-time values of τ ranging from 50 ms to 100 s and by using the T1 processing software of the spectrometer.

Data Processing. The free induction decay was submitted to an exponential multiplication inducing a line broadening of 2 Hz. The curve fitting was carried out in accordance with a Lorentzian mathematical model using Perch Software (Perch NMR Software, University of Kuopio, Finland). Isotope parameters were submitted to a statistical treatment, which consisted of a variance analysis and a principal component analysis using ADE-4 Software (CNRS, Lyon University, France). Principal Component Analysis (PCA). Data treatment by PCA was used to process table-crossing vanillin samples (as individuals) and specific isotopic deviations (as variables). PCA in general acts to reduce the dimensionality of a problem and to transform interdependent coordinates into significant and independent ones. The first principal component accounts for as much of the variability in the data as possible, and each succeeding component accounts for as much of the remaining variability as possible and must be completely uncorrelated to the previous (orthogonal) component. Hence, an analysis in terms of principal components can show linear interdependence in data.The principal components were computed via the covariance matrix (centered data) since all measurements of the variables were in the same units.7 Isotope Ratio Mass Spectrometry Experiments. The overall carbon isotope ratios of the samples were measured using a Finnigan Delta E mass spectrometer equipped with a Carlo Erba microanalyzer. All values obtained had a standard deviation of e0.3‰. Isotope Analysis. Specific carbon isotopic abundance was calculated using

Ai ) (fi/Fi)Ac

(1)

where fi is the molar fraction of the isotopomer monolabeled in position i, Fi is thepopulation of site i in the case of a random distribution of the stable isotopes, and Ac is the overall 13C (7) De Lagarde, J. In Initiation a` l’analyse des donne´ es; Dunod: Paris, 1995; pp 101-112.

Analytical Chemistry, Vol. 76, No. 13, July 1, 2004

3819

Table 1. 13C NMR Chemical Shifts and Longitudinal Relaxation Times (T1) for a Solution of Vanillin in Acetone (500 mg/mL) at 298 K site i

chemical shift (ppm) T1 (s)

1

2

3

4

5

6

7

8

190.4 2.8

152.4 11.7

147.8 18.7

129.5 13.4

126.1 2.0

114.9 2.4

109.8 2.37

55.2 2.9

abundance measured by IRMS

∑S

fi ) Si/

(2)

i

i

∑P

Fi ) Pi/

(3)

i

i

where Si is the area of the peak corresponding to the molecular site i and Pi is the number of equivalent carbons for the molecular site i. In the following, we will call the term fi/Fi the reduced molar fraction. Specific carbon isotopic deviations were calculated using

δi ) (Ri/RPDB - 1) × 1000

(4)

where RPDB is the isotopic ratio of Pee Dee Belemnite, international standard for IRMS8

Ri ) 13C/12C ) Ai/(1 - Ai): specific carbon isotopic ratio (5) Total Reduced Molar Fractions. The vanillin molecule contains eight different atoms of carbon (Figure 1). Each site presents only one equivalent carbon, therefore

Pi ) 1

∑P ) 8

and

i

so

Fi ) Pi/

i

∑P ) 1/8 i

i

Corrective coefficients were applied to peak areas to compensate for intensity losses due to satellite lines, which are assigned to the bilabeled isotopomers. In accordance with the 13C natural mean abundance of 1.1%, areas were multiplied by (1+ n0.011), where n was the number of carbons directly connected.6 Partial Reduced Molar Fractions. Partial reduced molar fractions were calculated using only sites 1, 5, 6, 7, and 8.

∑S

fi ) Si/

i

with

i ∈ [1,5,6,7,8]

i

∑P ) 5 i

i

and then

∑P ) 1/5

Fi ) Pi/

i

i

Similar corrective coefficients for total reduced molar fractions were applied to peak areas as described above. RESULTS AND DISCUSSION Measurement Conditions. Acetone was chosen as solvent because it allows a high solubility of vanillin in order to obtain a maximum signal-to-noise ratio. It also enables easy recovery of the samples. (8) Craig, H. Geochim. Cosmochim. Acta 1957, 12, 133-149.

3820

Analytical Chemistry, Vol. 76, No. 13, July 1, 2004

Flip angle and interpulse delay are not the best measurement conditions in terms of signal-to-noise ratio. However, Cookson and Smith9 have shown that, for high accuracy, an optimal signal-tonoise ratio is always achieved by using near 90° pulse angles. The interpulse delay D was selected on the basis of the longitudinal relaxation times. The T1 values are listed in Table 1. To obtain quantitative measurements for all eight sites with an accuracy of 1‰, a repetition time (D) equal to 7 times the longest T1 was chosen: 7 × 18.7 ) 130.9 s. The value of 1 s for sampling period may seem short. However, it was chosen in order to limit the heating effect of decoupling and to avoid NOEs. Furthermore, taking into account the line broadening of 2 Hz, it is sufficient to avoid any truncation of the FID. Measurement conditions were therefore matched to quantitative analysis. Specific Isotopic Abundances Ai and Specific Isotopic Deviations δi. Total molar fractions from 13C NMR spectra acquired with D ) 131 s and a decoupling power of 0.8 W were calculated for each sample. Table 2 shows the inter- and intrasample variation of the total molar fractions for samples of vanillin of biotechnological origin. As can be seen, the successive acquisitions from a single sample give a value that varies less than 0.2% of the mean. Intersample repeatability within a batch from the same source is also good, varying no more than 0.2%. It is thus apparent that the precision of the method is quite sufficient to differentiate the 13C/12C ratios, first between the different carbon atoms within a given molecule and, second, between molecules of the same chemical species obtained from different sources. Using the overall mean A0 determined by IRMS, it is possible to calculate absolute Αi values for each carbon position in the vanillin molecule. These are given in Table 3 for the mean of the four origins of vanillin. For each source, the ∆value (Αimax - Αimin) is ∼26‰ with a precision of ∼3‰, indicating that changes in the 13C/12C ratios as low as 3 parts in 1000 can be determined. The intramolecular 13C/12C ratios are found to be highly nonstatistical for all sources of vanillin examined, as previously reported.4 The biological interpretation of this has been discussed previously.10 Two sitess1 and 8sshow notably wide dispersion ranges, indicating these could be especially useful markers of metabolic events related to (bio)synthesis. Specific isotopic deviations were calculated for the 12 vanillin samples from total molar fractions from 13C NMR spectra acquired with D ) 131 s and a decoupling power of 0.8 W and from overall 13C deviations measured by IRMS. These are given in Table 4 and plotted in Figure 2 for the mean of the four origins of vanillin. These results show that isotopic deviations are quite originspecific and our values for deviations are consistent with those (9) Cookson, D. J.; Smith, B. E. Anal. Chem. 1982, 54, 2593-2596. (10) Schmidt, H.-L.; Werner, R.; Eisenreich, W. Phytochem. Rev. 2003, 2, 6185.

Table 2. Total Reduced Molar Fractions for the Eight Sites of Vanillin Determined from with D ) 131 s and a Decoupling Power of 0.8 Wa

13C

NMR Spectra Acquired

site i

B1a

B1b

B1c

B1d

mean

SD

B1

B2

B3

B4

mean

SD

1 2 3 4 5 6 7 8

1.002 1.010 0.999 1.013 0.995 0.996 0.998 0.988

1.006 1.008 1.001 1.010 0.991 0.998 0.996 0.991

1.004 1.006 1.000 1.012 0.994 1.000 0.996 0.988

1.002 1.009 1.002 1.010 0.992 0.998 0.997 0.990

1.003 1.008 1.000 1.011 0.993 0.998 0.997 0.989

0.002 0.001 0.001 0.001 0.002 0.002 0.001 0.002

1.003 1.008 1.000 1.011 0.993 0.998 0.997 0.989

1.002 1.007 1.002 1.012 0.995 0.996 0.995 0.991

1.005 1.007 1.003 1.011 0.992 0.996 0.997 0.989

1.003 1.004 1.004 1.011 0.994 0.998 0.997 0.989

1.004 1.008 1.001 1.011 0.993 0.997 0.996 0.990

0.002 0.002 0.002 0.001 0.001 0.001 0.001 0.001

a Results show the intrasample variation for one biotechnological sample (B1a-B1d) and the intersample variation over the four samples of biotechnological origin (B1-B4).

Table 3. Specific Isotopic Abundance (in %) [( Standard Deviation (n ) 3 and n ) 4) or Range (n ) 2)] of the Eight Sites of Vanillin Calculated from 13C NMR Spectra Acquired with D ) 131 s and a Decoupling Power of 0.8 Wa

Table 4. Specific Isotopic Deviations (in ‰) [( Standard Deviation (n ) 3 and n ) 4) or Range (n ) 2)] of the Wight Sites of Vanillin Calculated from 13C NMR Spectra Acquired with D ) 131 s and a Decoupling Power of 0.8 Wa

Αi for source (%)

A1 A2 A3 A4 A5 A6 A7 A8 Aar Α0

δi for source (‰)

ex-guaiacol (n ) 3)

ex-lignin (n ) 3)

biotechnology (n ) 4)

expods (n ) 2)

1.096 ( 0.002 1.076 ( 0.002 1.073 ( 0.002 1.081 ( 0.002 1.078 ( 0.002 1.076 ( 0.003 1.080 ( 0.001 1.074 ( 0.001 1.077 ( 0.003 1.079 ( 0.001

1.088 ( 0.003 1.075 ( 0.002 1.078 ( 0.005 1.086 ( 0.003 1.070 ( 0.004 1.080 ( 0.002 1.074 ( 0.003 1.099 ( 0.002 1.077 ( 0.005 1.081 ( 0.001

1.075 ( 0.002 1.078 ( 0.002 1.074 ( 0.002 1.084 ( 0.002 1.064 ( 0.002 1.068 ( 0.002 1.067 ( 0.002 1.060 ( 0.002 1.072 ( 0.007 1.071 ( 0.001

1.097 ( 0.002 1.091 ( 0.002 1.089 ( 0.003 1.101 ( 0.004 1.073 ( 0.002 1.087 ( 0.001 1.078 ( 0.003 1.093 ( 0.004 1.087 ( 0.009 1.089 ( 0.001

a Results have been averaged over n samples for each origin. Line Aar gives the average value over the aromatic carbons. Line Α0 gives the overall mean determined by IRMS. The Ai standard deviations over the four measurements made on the same sample were typically 0.002%. n is the number of samples analyzed for each source, four NMR spectra being averaged for each sample (see Table 2 for detailed example of sample B1).

obtained by IRMS after vanillin degradation.11 Notwithstanding, they differ from those determined by NMR by Caer et al. in 1991.3 These authors found that Ai values for natural, ex-lignin and exguaiacol vanillin were very close and, in particular, that the aromatic carbons showed the higher values. In contrast, in the present study, δi for the three origins differ significantly (Figure 2) and Ai or δi of the aromatic carbons tend to be lower than those for CHO and CH3 for natural and ex-lignin vanillin. These disparities with previous 13C NMR measurements3 can be explained by differences in the acquisition conditions. In ref 3, Caer et al. recorded spectra on an old AM 400-MHz spectrometer, whereas we used a more modern DRX 500 MHz instrument. In the case of Caer’s work, the bigger dispersion obtained with WALTZ-16 required the use of broad-band noise decoupling.12 Therefore, spectral acquisitions had to be carried out in three domains in order to obtain good decoupling results. Since the WALTZ-16 sequence is now well implemented on the NMR spectrometer, our experiments gave access to all 13C chemical (11) Krueger, D. A.; Krueger, H. W. J. Agric. Food Chem. 1985, 33, 323-325. (12) Shaka, A. J.; Keeler, J.; Frenkiel, T.; Freeman, R. J. Magn. Reson. 1983, 52, 335-338.

δ1 δ2 δ3 δ4 δ5 δ6 δ7 δ8 δar δ0

ex-guaiacol (n ) 3)

ex-lignin (n ) 3)

biotechnology (n ) 4)

expods (n ) 2)

-13.5 ( 2 -31.6 ( 3 -34.7 ( 2 -27.8 ( 2 -30.4 ( 4 -32.3 ( 3 -28.4 ( 3 -33.9 ( 2 -30.8 ( 3 -29.2 ( 0.4

-20.8 ( 4 -32.6 ( 3 -30.7 ( 4 -23.2 ( 2 -37.4 ( 2 -28.4 ( 2 -34.3 ( 1 -11.1 ( 4 -31.1 ( 5 -27.3 ( 0.7

-33.1 ( 1 -30.0 ( 1 -34.2 ( 2 -25.2 ( 1 -42.9 ( 1 -39.4 ( 2 -40.0 ( 1 -46.4 ( 1 -35.3 ( 6 -36.4 ( 0.4

-12.6 ( 2 -18.2 ( 2 -19.8 ( 3 -8.8 ( 3 -34.1 ( 2 -22.4 ( 1 -30.9 ( 2 -16.9 ( 3 -22.4 ( 9 -20.5 ( 0.4

a Results have been averaged over n samples for each origin. Line δar gives the average value over the aromatic carbons. Line δ0 gives the overall mean determined by IRMS. n is the number of samples analyzed for each source, four NMR spectra being averaged for each sample (see Table 2 for detailed example of sample B1).

Figure 2. Specific isotopic deviations of the eight sites of vanillin calculated from 13C NMR spectra acquired with D ) 131 s and a decoupling power of 0.8 W. Results have been averaged over each origin and standard deviations (ex-guaiacol, n ) 3; ex-lignin, n ) 3; biotechnology, n ) 4) or range (expod, n ) 2) are plotted as error bars. 0 ex-guaiacol, ∆ ex-lignin, O biotechnology, and + expod.

shifts within a single spectrum, provided that the offset frequency, the irradiation length, and the power of the decoupling pulses had been calibrated. Moreover, Caer et al.3 did not use corrective coefficients. On the other hand, our values for deviations are quite consistent with the results of Bensaid et al.13 and of Krueger and Krueger11 obtained by IRMS after vanillin degradation. It would Analytical Chemistry, Vol. 76, No. 13, July 1, 2004

3821

Table 5. Partial Reduced Molar Fractions [( Standard Deviation (n ) 3 and n ) 4) or Range (n ) 2)] for the 12 Samples of Vanillin from 13C NMR Spectra Acquired with a Decoupling Power of 0.8 W and D ) 21 or 131 sa

fi/Fi partial 1 5 6 7 8

ex-guaiacol (n ) 3) D1 ) 21 s D1 ) 131 s 1.016 ( 0.002 0.994 ( 0.002 0.995 ( 0.002 1.000 ( 0.002 0.994 ( 0.002

1.014 ( 0.002 0.997 ( 0.002 0.995 ( 0.002 0.999 ( 0.001 0.994 ( 0.001

ex-lignin (n ) 3) D1 ) 21 s D1 ) 131 s 1.007 ( 0.002 0.987 ( 0.002 1.000 ( 0.003 0.991 ( 0.004 1.016 ( 0.005

1.006 ( 0.004 0.989 ( 0.002 0.998 ( 0.003 0.992 ( 0.002 1.016 ( 0.001

biotechnology (n ) 4) D1 ) 21 s D1 ) 131 s 1.009 ( 0.003 0.996 ( 0.002 1.001 ( 0.002 1.000 ( 0.003 0.994 ( 0.002

1.008 ( 0.002 0.997 ( 0.002 1.001 ( 0.002 1.000 ( 0.002 0.994 ( 0.002

expods (n ) 2) D1 ) 21 s D1 ) 131 s 1.011 ( 0.002 0.987 ( 0.001 1.001 ( 0.001 0.991 ( 0.002 1.070 ( 0.002

1.011 ( 0.002 0.989 ( 0.003 1.001 ( 0.001 0.993 ( 0.002 1.007 ( 0.002

a Results have been averaged over each origin. The (f /F ) standard deviations over the four measurements made on the same sample were i i typically 0.002 (0.2%). n is the number of samples analyzed for each source, four NMR spectra being averaged for each sample.

Figure 3. Representation of the PC plane of principal component analysis performed on isotopic deviations δi13C. The 4 measurements from each of the 12 samples are represented in the plane of the two main axes (CP1, CP2) and the relative weightings are indicated in parentheses.

appear, therefore, that distortions in the 13C NMR spectra, especially related to the decoupling conditions, can be crucial in introducing artifactual δi values, notably in differentially influencing the apparent 13C/12C ratio for carbons with a differing number of bound hydrogen atoms. The importance of decoupling power is studied in detail below. To what extent, then, can these δi values be related to the origin of the different vanillin samples? It is apparent from Table 4 and Figure 2 that deviations show considerable fluctuations depending on origin, with the most negative values being found for vanillin produced by biotechnology from natural ferulic acid. Taking all 48 analyses (12 vanillin samples and 4 spectra per vanillin sample) as the data set, a PCA was performed in the space of the eight specific isotopic deviations δ1-δ8, calculated from each of the 13C NMR spectra acquired with D ) 131 s and a decoupling power of 0.8 W. The ensuing distribution of the samples is represented in Figure 3 in the plane of the two principal components (CP1, CP2) with the relative weightings indicated in parentheses. The 12 vanillin samples are seen to group in 4 regions of the plane, and each region can be seen to represent the samples obtained from one (bio)synthetic source. Thus, for example, all significant correlation between δi values for the four samples of biotechnological origin falls within the same region of the PC plane (Figure 3). Similarly the three ex-lignin vanillin (13) Bensaid, F. F.; Wietzerbin, K.; Martin, G. J. J. Agric. Food Chem. 2002, 50, 6271-6275.

3822 Analytical Chemistry, Vol. 76, No. 13, July 1, 2004

samples, the 2 natural vanillin samples and the 3 samples ex-guaiacol are grouped. Furthermore, each source of vanillin is isolated, indicating there has to be effective discrimination between the 12 samples to classify them according to their origin. A variance analysis reveals sites 1 and 8 as the most discriminating. Significant contributions came only from site 8 (52%) for CP1 and from site 1 and 8 (34 and 25%, respectively) for CP2. These results are consistent with the data presented in Figure 2, where the biggest differences between origins are in sites 1 and 8. Partial Reduced Molar Fractions fi/Fi Obtained with D ) 21 s. The PCA analysis based on all eight δi values indicated that sites 2-4 contribute only 12% or less toward CP1 and CP2. Yet these relatively undiscriminating sites present the longest relaxation times, bringing experimental time to 3 h 30 min (D ) 7 × 18.7 s). Thus, by omitting sites 2-4, the maximum T1 value becomes that of site 8 (2.95 s) and D becomes 21 s instead of 131 s. Under these conditions, measurements are no longer quantitative for sites 2-4. As a result, deviations δi, which are calculated from overall mean 13C abundance (measured by IRMS) and from total molar fractions, can no longer be obtained. However, in many applications of 13C analysis, it is desirable to focus only on certain target sites that are pertinent to the study. In the example of vanillin, it is valuable to define the extent to which discriminatory efficiency is retained during accelerated acquisition conditions. To this end, partial reduced molar fractions were calculated for the 12 vanillin samples from 13C NMR spectra acquired with D ) 21 s and a decoupling power of 0.8 W. Results were averaged over each origin, and values are given in Table 5 (with those obtained with D ) 131 s) and plotted in Figure 4. Mean values and standard deviations for both acquisition conditions show no significant differences in the mean and in the standard deviations (p < 0.001). Hence, it can be concluded that no loss of intra- or extramolecular reproducibility has resulted from the use of D ) 21 s. Stability Over Time of Partial Molar Fractions. Stability during a Short Time. Table 6 shows the typical intrasample variations in the partial molar fractions for one sample of vanillin (hemisynthetic H1). As can be seen, the successive acquisitions from a single sample give a value that varies less than 0.3% of the mean. The same values were obtained on other samples (data not shown). Intersample repeatability within a batch from the same source is also good, varying not more than 0.3%. It is thus apparent that the stability of the method is quite sufficient to differentiate the 13C/12C ratios, between the different carbon atoms within a

Figure 4. Partial reduced molar fractions of the sites 1, 5, 6, 7, and 8 of vanillin calculated from 13C NMR spectra acquired with D ) 21 s and a decoupling power of 0.8 W (a). Results have been averaged over each origin and standard deviations (ex-guaiacol, n ) 3; ex-lignin, n ) 3; biotechnology, n ) 4) or range (expod, n ) 2) are plotted as error bars. 0 ex-guaiacol, ∆ ex-lignin, O biotechnology, and + ex-pod. Table 6. Stability over Time of Partial Reduced Molar Fractions for Five Sites of Vanillin Determined by 13C NMR Spectra Acquired with D ) 21 s and a Decoupling Power of 0.8 Wa short time stability intrasample on H1

long time stability

intersamples on H1, H2, H3

intrasample on H1

site i

mean

SD

mean

SD

mean

SD

1 5 6 7 8

1.006 0.987 1.001 0.995 1.011

0.001 0.002 0.001 0.002 0.003

1.007 0.989 0.997 0.994 1.014

0.001 0.001 0.003 0.002 0.002

1.008 0.988 1.000 0.994 1.010

0.001 0.002 0.002 0.001 0.003

a Results show the intrasample stability over a short time for the sample H1 (four successive measurements of the same hemisynthetic vanillin sample H1), the intersample stability over a short time for the three hemisynthetic origin samples H1, H2, and H3, and the intrasample stability during a long time for H1 (five measurements performed over 15 months of the same hemisynthetic vanillin sample H1). Note that the five measurements performed over 15 months were recorded on 15 December 2001, 16 December 2002, 20 December 2002, 20 February 2003, and 14 March 2003.

given molecule and between molecules of the same chemical species obtained from different sources. Stability during a Long Time. The NMR measurements presented for all the vanillin samples discussed were acquired over a period of 15 months. To monitor any drift with time, a control measurement was performed on the same sample (hemisynthetic H1) before each series of measurements. The data obtained for this sample have therefore been analyzed to evaluate measurement stability with time over a 15-month period. Site-specific standard deviations have been calculated for the partial molar fractions obtained from the five sets (each of 4 repetitions) of 13C NMR spectra. Results are gathered in Table 6. Standard deviation values remain inferior to 3 per 1000. The results show that the repetition of measurements over a 15-month period produces a scatter within the sample set of results that is no larger than that obtained with the measurements of several samples at the same time. Therefore, it can be concluded that the acquisition conditions are robust and stable in the long term.

Comparison of Partial Reduced Molar Fractions fi/Fi Obtained from Spectra Performed with a Decoupling Power of 0.8 or 2 W. A significant variation in fi/Fi depending on the decoupling power used was noted during initial experiments. The influence of this parameter was therefore investigated in detail. Several decoupling techniques, based on composite rectangular pulses, have been proposed such as MLEV, WALTZ, GARP, and MPF.14-17 All these techniques have been previously evaluated in relation to their capability to achieve consistent decoupling efficiency over a wide frequency range. In our case, the most important criterion is not the effective decoupling bandwidth but rather the homogeneity within the bandwidth. Because of the accuracy needed, only a very small variation with frequency can be tolerated and experience showed that WALTZ-16 gave the best results from this point of view (data not shown). More recently, adiabatic decoupling, which inverts spins of different offsets adiabatically and successively while sweeping the frequency from one side of the offset to the other, turns out to be one of the best decoupling schemes.18 However, adiabatic decoupling was not evaluated for the present application, as it is not feasible with our spectrometer hardware. In decoupling sequences such as WALTZ-16, the rf amplitude used plays an important role. When it is increased, the pulse duration is reduced, which decreases the sensitivity to offresonance effects. However, the amount of sample heating that occurs during decoupling is primarily determined by the average decoupling power, which is proportional to the square of the rf amplitude. To avoid sample heating is a major goal in optimizing decoupling schemes. When the rf amplitude is too high, B0 homogeneity is disturbed and the line shape is modified. However, when rf amplitude is too low, coupling is not totally removed, which induces an unresolved doublet. In both cases, peak areas are affected because peaks are distorted in relation to the Lorentzian model, from which their surface areas are determined. An evaluation of the influence of decoupler power was carried out with a D ) 21 s and a decoupling power of 0.8 or 2 W. As predicted from the above argument, the relative peak areas measured from 13C NMR spectra, hence the partial molar fractions, are significantly influenced by decoupling conditions (Table 7). Critically, the values for the aromatic carbons are now higher than for the 1 and 8 positions, as graphically shown in Figure 5a. Sites 1 and 8, however, still show the greatest dispersion and results remain origin-specific. Since evolutions are proportional, partial reduced molar fraction behavior can be compared to Ai (eqs 1 and 4). The 2-W decoupling power provides more powerful pulses, hence allowing more homogeneous spectral width decoupling. Under these conditions, acquisitions are less dependent on offresonance phenomena, which explains why the results are more consistent with those of Caer et al.,3 where decoupling parameters were specially adapted for each frequency domain. Thus, for an individual sample, different relative areas are obtained for each peak according to the decoupling conditions. (14) Levitt, M. H.; Freeman, R. J. Magn. Reson. 1981, 43, 502-507. (15) Shaka, A. J.; Keeler, J.; Freeman, R. J. Magn. Reson. 1983, 53, 313-340. (16) Shaka, A. J.; Barker, P. B.; Freeman, R. J. Magn. Reson. 1985, 64, 547552. (17) Fujiwara, T.; Anai, T.; Kurihara, N.; Nagayama, K. J. Magn. Reson. 1993, 104, 103-105. (18) Zhang, J. W. S.; Gorenstein, D. G. J. Magn. Reson. 1996, 123, 181.

Analytical Chemistry, Vol. 76, No. 13, July 1, 2004

3823

Table 7. Partial Reduced Molar Fractions [( Standard Deviation (n ) 3 and n ) 4) or Range (n ) 2)] for the 12 Samples of Vanillin from 13C NMR Spectra Acquired with a Decoupling Power of 2 W and D ) 21 sa fi/Fi partial

ex-guaiacol (n ) 3)

ex-lignin (n ) 3)

biotechnolog (n ) 4)

expods (n ) 2)

1 5 6 7 8

0.996 ( 0.002 1.004 ( 0.003 1.003 ( 0.001 1.011 ( 0.002 0.986 ( 0.002

0.986 ( 0.002 1.000 ( 0.001 1.002 ( 0.002 1.004 ( 0.001 1.006 ( 0.001

0.989 ( 0.001 1.005 ( 0.001 1.007 ( 0.002 1.008 ( 0.001 0.987 ( 0.001

0.991 ( 0.001 0.997 ( 0.002 1.006 ( 0.001 1.004 ( 0.002 0.999 ( 0.001

a Results have been averaged over each origin. The (f /F ) standard deviations over the four measurements made on the same sample were i i typically 0.002. n is the number of samples analyzed for each source, four NMR spectra being averaged for each sample.

Figure 5. Partial reduced molar fractions of the sites 1, 5, 6, 7, and 8 of vanillin calculated from 13C NMR spectra acquired with D ) 21 s and a decoupling power of 2 W. Results have been averaged over each origin and standard deviations (ex-guaiacol, n ) 3; ex-lignin, n ) 3; biotechnology, n ) 4) or range (expod, n ) 2) are plotted as error bars (a). 0 ex-guaiacol, ∆ ex-lignin, O biotechnology, and + expod. Bidimensional representations of the principal component analysis performed on partial reduced molar fractions fi/Fi of the sites 1, 5, 6, 7, and 8 of vanillin calculated from 13C NMR spectra acquired with D ) 21 s and a decoupling power of 2 W (b).

However, differences in reduced partial molar fraction between origins are retained. A PCA was performed in the space defined by the five partial reduced molar fractions fi/Fi for sites 1, 5, 6, 7, and 8 calculated from the individual 13C NMR spectra of the 12 samples of vanillin acquired with D ) 21 s and a decoupling power of 2 W. The classification obtained by this analysis is shown in Figure 5b, where each of the 48 analyses (12 samples and 4 spectra per sample) are represented in the plane of the two main axes (CP1, CP2) and the relative weightings are indicated in parentheses. As can be seen, the 12 samples again group according to their source, correlations for all 48 analyses falling into discrete and isolated areas on the CP1/CP2 plane. Thus, by PCA, it is shown that effective separation of the different vanillin origins can be achieved using only five of the eight potential sites. A variance analysis reveals that major contributions are due only to sites 8 and 1 (78 and 10%, respectively) for CP1 and from sites 1 and 5 (59 and 37% respectively) for CP2. Other sites contributed less than 7% each. These results are consistent with those of Figure 5a. When only four of the five partial reduced molar fractions fi/Fi were taken into account in the PCA processing, it was no longer possible to correctly discriminate the four origins (data not shown). Hence, it can be concluded that, provided the samples are examined under identical conditions and that the five partial reduced molar fractions fi/Fi (for sites 1, 5, 6, 7, and 8) were included in the PCA processing, 13C NMR remains discriminating. 3824 Analytical Chemistry, Vol. 76, No. 13, July 1, 2004

CONCLUSIONS It is thus demonstrated that quantitative 13C NMR is an effective technique for the determination of 13C/12C intramolecular distribution at low or natural abundance, provided acquisition conditions are rigorously controlled. The method can be time-consuming, in that carbon atoms with long T1 values impose long interpulse delay times. However, lengthy and time-consuming sample preparation or partial degradation is avoided, which, coupled with the precision of the technique, makes it efficient for determining 13C/12C ratios. For targeted applications, in which interest is focused on only selected carbon atoms, it is shown that reduced interpulse delay times can be used without loss of precision. Thus, in the case of vanillin, a reduction from 210 to 30 min was achieved. It is further shown that the decoupling conditions used influence considerably the site-specific fi/Fi values obtained. However, provided acquisition conditions are consistent, it is found that fi/Fi values can be directly compared for the same chemical species obtained from different (bio)synthetic sources. Thus, the 12 samples studied are effectively grouped within the 4 sources since the specific isotope ratios due to the way in which the vanillin was made are significantly influenced by the synthetic route. The extent to which this capacity to differentiate origin on the basis of the fi/Fi values is currently being investigated with vanillin samples from a range of others origins. Although these conditions can effectively be used to determine 13C distribution within a given molecule and can be used to

compare accurately differences in distribution between different samples of the given molecule, they cannot provide absolute values for δi, the site-specific 13C/12C deviations in the molecule. This parameter is accessible through a combination of 13C NMR, to determine total reduced molar fractions, and IRMS, by which the mean overall δ value is determined (see Table 3). To resolve this final problem in developing a fully quantitative 13C NMR protocol to determine 13C/12C ratios, it is necessary to reduce further the imprecision introduced by frequency-dependent phenomena, notably unequal decoupling on spectral width. Adiabatic pulses that allow a homogeneous effect on the overall frequency range could provide a suitable solution.16 Furthermore, the use of an internal reference will avoid IRMS measurements (19) Robins, R.; Billault, I.; Duan, J.-R.; Guiet, S.; Pionnier, S.; Zhang, B.-L. Phytochem. Rev. 2003, 87-102.

and should allow δi to be calculated from partial molar fractions. Thus, affiliations between substrates and products, which cannot be deduced from molar fractions, will be accessible as is currently the case for quantitative 2H NMR.19 ACKNOWLEDGMENT The contribution of the Scientific Council of the Region of the Pays-de-la-Loire to the purchase of a 500-MHz spectrometer is gratefully acknowledged. We thank Carol Wrigglesworth (Scientific English, Nantes, France) for linguistic assistance.

Received for review February 24, 2004. Accepted April 15, 2004. AC0496998

Analytical Chemistry, Vol. 76, No. 13, July 1, 2004

3825