Chiral discrimination of limonene by use of -cyclodextrin-coated quartz-crystal-microbalances (QCMS) and data evaluation by artificial neuronal networks
Received: 6 February 2001 / Revised: 5 April 2001 / Accepted: 19 April 2001 / Published online: 18 July 2001
(C) Springer-Verlag 2001
Abstract The enantiomeric composition of the chiral flavoring agent limonene was analyzed by means of a quartzcrystal microbalance (QCM) sensor. As chiral selectors three different modified β-cyclodextrins were investigated. The selector molecules were applied as mixtures in different polysiloxane matrices. The chiral separation factors αfor limonene obtained at 30°C by gas chromatography and by use of the QCM sensor were comparable. Evaluation of sensor data was performed by use of an artificial neuronal network (ANN); this enabled prediction of the enantiomeric composition of the gas mixtures.
Introduction
The enantiomers of chiral compounds have different properties in a chiral environment. The importance of the relationship between chirality and biological activity is well established in the odor perception of flavors and fragrances. Thus the olfactory properties of flavoring agents are often governed by their enantiomeric composition. One typical example is limonene – whereas the (–)-S enantiomer has a turpentine (mint) like odor, the (+)-R enantiomer is associated with an orange-type smell (cf.Fig.1) [1, 2]. The enantioselective analysis of limonene enantiomers by chiral gas chromatography on permethyl-β-cyclodextrin [3] revealed an enantiomeric composition of 98% for (+)-R-limonene in lemon peel and of 99.5% in grapefruit and orange peel [4]. conversely, the formation of (–)-S-limonene has been invoked in the biosynthetic pathway to (–)-menthol and (–)-carvone in mint (Mentha) species [5].
Dedicated to the memory of Professor Dr. J. F. K. Huber
C. Fietzek
Institute of Physical and Theoretical Chemistry, Center of Interface Analysis and Sensors,
University of Tübingen, Auf der Morgenstelle 8, 72076 Tübingen, Germany
T. Hermle (?) • W. Rosenstiel
Wilhelm-Schickard Institute for Computer Science, Computer Engineering, University of Tübingen, Auf dem Sand 13, 72076 Tübingen, Germany
e-mail: [email protected]
V. Schurig
Institute of Organic Chemistry, University of Tübingen, Auf der Morgenstelle 18, 72076 Tübingen, Germany
An incentive for the application of enantioselective sensors is their use in common electronic noses for quality control, because the ratio of the enantiomers can give important information about the natural or industrial origin of flavoring agents, e.g. in the screening of essential oils for artificial spiking with racemic material. Previous studies demonstrated that common quartz-crystal-microbalances (QCM) in the shear-resonance mode, and optical sensors based on reflectometric interference spectroscopy are capable of differentiating between the enantiomers of gaseous chiral substances [6, 7, 8, 9, 10, 11,12]. The differences in mass uptake or swelling of the
film, respectively, result from the thermodynamically-controlled partitioning of the enantiomers between the gas phase and an appropriate chiral stationary phase. The enantioselective materials, e.g. diamides and modified cyclodextrins, used in sensor devices were adopted as a result of their longstanding use in chiral gas chromatography [13]. Diamide phases undergo enantioselective interactions by multiple hydrogen-bonding and relatively high concentrations of the analyte enantiomers can be incorporated in the film without changing the sensing properties of the sensor. These sensors can, therefore, be used to discriminate between different enantiomers over a wide concentration range [8, 9]. Because the calibration curves for the sensors are linear, the data evaluation step can be performed by simple regression, e.g. by partial least squares (PLS) analysis [9]. With modified cyclodextrins the preferential enantioselective interactions are believed to result from partial inclusion [14]. Because the number of available recognition sites is limited, the best results for enantioselective absorption are obtained in the early domains of the Langmuir isotherms at very low analyte concentrations. In analytical gas chromatography analyte concen-
trations are very small whereas in preparative gas chromatography overloading problems can arise [15]. In agreement with this, chemical sensors coated with modified cyclodextrins can have non-linear calibration curves for higher amounts of analyte, necessitating refined modes of data processing. Yet the advantage of cyclodextrins is associated with their broad spectrum of application towards different classes of chiral compounds.
This study is aimed at utilizing well-established gaschromatographic β-cyclodextrin-containing chiral stationary phases (CSPs) as sensor materials for the chiral dis-
crimination of a typical chiral flavor compound, i.e. the enantiomers of limonene, by use of quartz microbalance sensors. For data evaluation an artificial neuronal net (ANN) was trained.
Experimental
Sensor coatings.The enantioselective films consisted of three β-CDs containing the apolar and crowded tert-butyldimethylsilyl group [16] in position 6:
1. heptakis(2,3-di-O-methyl-6-O-tert-butyldimethylsilyl)-β-cyclodextrin, denoted Me-β-CD [17];
2. heptakis(2,3-di-O-ethyl-6-O-tert-butyldimethylsilyl)-β-cyclodextrin, denoted Et-β-CD [18]; and
3 heptakis(2,3–di-O-acetyl-6-O-tert-butyldimethylsilyl)-β-cyclodextrin, denoted Ac-β-CD [19] (Fig.2).
The β-CDs were used both pure and as 50% (w/w) mixtures with the polysiloxanes: OV 1701 (polycyanopropyl phenylmethylsiloxane with approx. 7% cyanopropyl and 7% phenyl), PS086
(polyphenylmethylsiloxane with 12–15% phenyl), SE 52 (polyphenylmethylsiloxane with 5% phenyl), and SE 30 (polydimethyl siloxane with a high level of polymerization). An achiral polyether urethane (PEUT) from Thermedics, Woburn MA, USA was used as reference coating. The sensor array for the ANN evaluation consisted of ten different sensors: two PEUT, two SE 52, one SE 30,three Et-β-CD, one Me-β-CD, and one Ac-β-CD (each CD applied in a SE 52 matrix)
Analytes. R- and S-limonene were purchased from Sigma–Aldrich Deisenhofen, Germany and used after distillation. The enantiomeric purity of the samples was checked by enantioselective gas
chromatography [3] (c.f. Fig.4).
Sensors.The sensor array consisted of discrete piezoelectric quartz crystals (AT-cut) with gold electrodes on each side operating in a thickness shear mode vibration at a fundamental frequency of f o=30 MHz (quartz plate thickness 55.6 μm). Crystals were pur chased from KVG Quartz Crystal Technology, Germany. The sensor response was monitored as frequency change ?f on exposure to the test gases.
Coating of the quartz sensor was achieved by spraying the dissolved materials on to the cleaned surface by use of an air-brush pistol. The coating process was stopped when the frequency decrease was 40 kHz on each side.
Gas mixing.The enantiomers of limonene were deposited in thermo-controlled vaporizers. The analyte concentration in the gas phase is defined by Antoine’s law [20]. Defined vapor concentrations are obtained by using nitrogen as a carrier gas and by diluting with computer-driven mass-flow controllers. The accurate dosing of the enantiomers was also checked by interchanging the filled vaporizer units.
All measurements were performed at a constant temperature of 303 K and with a constant flow rate of 200 mL/ min. Up to 16 sensors were mounted in two combined flow-through brass cells with a volume of 20 mL each.
Results and discussion
Sensor signals
Coatings comprising 50% (w/w) mixtures of the cyclodextrin selectors and polysiloxanes had reasonably fast t90 times of approximately 200 s. The latter sensors could easily be evaluated using exposure times of 15 min followed by 15 min purging of the chamber.
Typical on-line signals are shown in Fig.3, where a series of limonene concentrations of 550 μg/ L
with different enantiomeric compositions is depicted for the chiral Et-β-CD–SE 52 sensor and the achiral polyetherurethane (PEUT) sensor. The difference between the responses of the chiral sensor to R- and S-limonene is more than 100 Hz,whereas no frequency changes are observed for the PEUT.The sensor signal was lower for S-limonene, in agreement with the order of elution in the gas-chromatographic experiment – S-limonene is eluted as the first peak on the chiral stationary phase Et-β-CD.
Fig.3 A series of sensor signals for limonene (550 μg L ) of different enantiomeric composition on a chiral coated sensor and an achiral polyetherurethane (PEUT) sensor. Whereas there is no difference between the responses of the achiral sensor to R- and S-limonene, for the Et-β-CD-containing sensor there is a difference of approximately 100 Hz
With regard to the nature of the polysiloxane matrices,stable behavior was observed for cyclodextrins diluted with SE 52 and SE 30 whereas cyclodextrins diluted with more polar polysiloxanes tended to crystallize from the polymer. Coating with pure β-CDs resulted in signals with t90 decreasing times of more than 1000s; these were,therefore, not evaluated further.
Comparison with enantioselective gas chromatography
To determine the chiral separation factor αchrom by GC,racemic limonene was separated on a fused silica capillary column coated with Et-β-CD in SE52 (50% w/w) at 30°C. From these measurements αchrom(ratio of the adjusted retention times of the enantiomers) was calculated to be 1.29. The chromatograms obtained are shown in Fig.4.
Fig.4 Chromatograms obtained from racemic limonene and from both single enantiomers. Conditions: column, 20 cm × 0.25 mm i.d. coated with Et-β-CD in SE 52; temp, 30 °C, carrier gas, hydrogen
Fig.5 Linear isotherms for limonene with the achiral polymer PEUT showing the same values for the adsorption of the R and S enantiomers
QCM-isotherms
The QCM-isotherms for the achiral reference PEUT and for the three β-CDs in SE52 are shown in Figs.5, 6, 7, 8.In the lower part of the figure the ratio of the R- and S-limonene signal is given. This ratio can be interpreted in terms of a chiral sensor-based separation factor αsensor. Error bars are obtained from error propagation of the experimental error. The precision for the determination of the separation factor is lower at small signals, because of the higher impact of experimental errors such as dosing, temperature, humidity, or aging. Linear isotherms with identical signals for both enantiomers are obtained for the achiral reference PEUT (cf. Fig.5). The values of αsensor approach unity, implying that no preferential adsorption of one enantiomer occurs.
With regard to the chiral stationary phases, preferential adsorption of R-limonene was observed for Me-β-CD and Et-β-Me only whereas for Ac-β-CD differences between signals for the enantiomers were negligible and within the experimental error; this last material was, therefore, unsuitable for further investigation (cf. Fig.8).
For Et-β-CD intense signals and strong bending of the isotherm were observed at low concentrations, as is depicted in Fig.6. This finding can be explained by preferential adsorption in the energetically favored recognition sites of the cyclodextrin. The αsensor values increase signif icantly towards lower concentrations, because of the greater number of unoccupied sites available in the very low concentration range. A curve-fitting procedure assuming the dual adsorption mechanism of Henry- and Langmuir-adsorption as proposed in Ref. [10] enables Interpolation to very small concentrations and, hence, comparison with chiral separation factors αchrom under GC operating concentrations. This separation factor αsensor, c=0 was calculated using the Et-β-CD isotherms in different polymers, and is given in Table 1. The values obtained vary from 1.20 to 1.30 and are in good agreement with the experimental αchrom of 1.29.
For Me-β-CD no bending, and hence no energetically favored adsorption for limonene, can be observed in Fig.7; enrichment of the R-enantiomer occurs, nevertheless.
Fig.6 The highly non-linear isotherms obtained for the Et-β-CDC-coated sensor, with especially high separation factors, αsensor.at low concentrations
Fig.7 Sensor calibration for Me-β-CD. The response of the sensor to limonene is almost linear and it can discriminate between the R and S enantiomers
Fig.8 Sensor calibration for Ac-β-CD. No differences can be found for the different enantiomers and the chiral separation factorαsensor is close to unity
Table 1 Separation factors,
αsensor (30°C), for limonene on Et-β-CD in different polysiloxane mixtures obtained by curve fitting and extrapolation to infinite dilution
Fig.9 Prediction of the concentration of limonene in the test data samples by evaluation of the sensor array using ANNs
Fig.10 Prediction of the concentrations of R- and S-limonene from the non-racemic test data samples shown in Fig.9
Fig.11 Prediction of the enantiomeric composition (%) of samples of R-limonene by evaluation of the sensor array
Prediction of the enantiomeric composition by use of artificial neuronal networks
Because of the non-linear calibration curves an artificial neuronal network (ANN) was chosen for data evaluation.The range of concentrations of the reference data set was from 0 to 1650 μg/ L
and contained signals of the both pure R- and S-limonene and the racemic (R,S) mixture.For the test data set the same concentration range was measured, but different enantiomeric compositions were chosen, in steps of 10%. The sample sequence was chosen randomly.
A backpropagation neuronal net was trained using 55 samples of the reference data set. By application of the trained neuronal net, the limonene concentration for all 29 samples of the test data set was predicted with a mean relative error of 3.72%. The corresponding true predicted plot is displayed in Fig.9. The concentrations of R- and S-limonene, respectively, were predicted from the ANN with a mean error of 6.97% as depicted in Fig.10. The larger error in the quantification of the enantiomers might be because the ANN is trained with a racemic mixture and the pure enantiomers only – new mixtures of R- and S-limonene had to be interpolated from the ANN. Another reason for the larger errors is the smaller number of sensors contributing to the prediction of the enantiomers. The enantiomeric composition of the samples was determined using the predicted concentrations of R- and S-limonene and compared with the true values (c.f. Fig.11). Prediction of different enantiomeric compositions is possible in steps of 15%.
To determine the importance of individual sensors,several strategies are known. Sophisticated methods involve the extraction of fuzzy rules or pruning algorithms.The simplest method is to inspect the weights of the trained neuronal net. According to this method the Me-β-CD and the Et-β-CD sensors seem to be the most useful.
As expected, no enantiomeric composition could be predicted by using an ANN in the absence of the chiral sensors.
Conclusion
The example of limonene demonstrates that the chiral discrimination of the enantiomers of flavoring agents is feasible by using sensors in combination with ANN data handling. The chiral separation factors αsensor obtained by use of the enantioselective sensors are comparable with αchrom
values measured by use of enantioselective gas chromatography and determination of the enantiomeric composition is possible in steps of 15%. It should be remarked that chiral discrimination of limonene in dynamic gas chromatography requires a large number of theoretical
plates in the capillary column whereas in the static sensorbased system only one single absorption/desorption step is invoked. Whereas enantioselective gas chromatography is more precise in the determination of enantiomeric composition, sensor arrays are much more easy to handle in practice.
Comparison of different β-CDs revealed that, in terms of signal evaluation and interpretation of the ANN, better chiral discrimination was achieved with Me- and Et-β-CD than with Ac-β-CD. As in enantioselective gas chromatography the use of a polymer matrix greatly improves the performance of the system.
Acknowledgment We gratefully acknowledge the funding of this work by the DFG in the framework of the Forschergruppe:Molekulare Mustererkennung mit supramolekularen Strukturen
und Polymeren. V. S. had the opportunity to meet the late Professor J. F. K. Huber at many occasions and he is indebted for his dedicated encouragement and interest in chiral aspects of analytical chemistry.
