摘要:

Scanning Kelvin microprobe in the tandem analysis of surface topography and chemistry Larisa-Emilia Cheran,a Hans-Dieter Liessb and Michael Thompson*a a Department of Chemistry, University of Toronto, 80 St. George St., Toronto, Ontario, M5S 3H6, Canada b Institut für Physik, Fakultät für Elektrotechnik, Universität der Bundeswehr München Werner Heisenberg-Weg, Neubiberg, 39/D-85577, Germany Received 21st December 1998, Accepted 19th May 1999 This article presents the principles of operation, performance and applications of a scanning Kelvin microprobe for combined contact potential and surface topographical measurements. The new instrument uses a miniaturized vibrating probe for exploring, point by point, the surface of a sample, through measurement of the current emerging from the local ‘capacitor’ formed between the vibrating tip and the surface. In particular, the instrument is capable of simultaneously imaging the contact potential difference and topography of a scanned surface.This tandem characterization provides unique information regarding material properties. The lateral resolution is 1 mm, the topographic resolution is in the nanometer range and contact potential sensitivity is in the millivolt range. We also present preliminary studies of graphite, silicon, mica, metal and polymer surfaces. The measurement of work function is one of the most effective methods for the investigation of surface-related processes due to the high sensitivity of this parameter to structural variations, surface modification and contamination.Accordingly, techniques for the measurement of work function offer a powerful tool for monitoring changes in all these areas. Work function can be measured both directly and indirectly. Direct measurements are based on the actual emission of electrons from the surface, that is via thermionic1 or photoelectric methods,2 or field emission.3 These methods provide excellent results for uniform, clean metal surfaces; however, it is difficult to apply these techniques to other types of material.The precision of such measurements is affected by the exponential dependence of the electron current on the work function. The result is also strongly dependent upon the intensity of the applied electric field for the formation of the electron current. Indirect methods are based on the measurement of the potential difference which appears when materials with different work function are placed in contact.Since this contact potential difference is the difference in the work function of the two materials, the work function of one material can be determined if the other is known. In this group, the Kelvin method is one of the most convenient methods. Other indirect techniques are the static capacitor,4 Oatley’s magnetron, 5 breakdown field,6 saturated diode,7 space-charge-limited diode,8 Anderson’s electron beam methods9 and, finally, specially-adapted scanning tunnelling microscopy and atomic force microscopy.10–14 The Kelvin method is based on a parallel plate capacitor model. One plate possesses a known work function and is used as a reference.The material with unknown work function represents the other plate. There is no contact between the two plates, therefore, the technique is one of few nondestructive methods, since the measured object is neither heated nor exposed to light or particles. The vertical position of one of the plates is changed sinusoidally, which gives a capacity variation of the parallel capacitor, leading in turn to a variation of charge in the capacitor.The displacement current generated this way, the Kelvin current, is proportional to the DC voltage between the plates (contact potential difference—CPD). If the DC voltage between the two plates is not zero, there will always be an AC current in the circuit, therefore the DC potential difference is “transformed” into an AC current by modulating the distance between the plates.This AC current is free from the low-frequency noise and drift, which would plague a DC measurement. Other important advantages of the method are the potential for high resolution measurement of CPD, an extremely simple experimental set-up whereby no sample preparation is necessary and the fact that different samples can be measured by simply placing them under the reference plate of the Kelvin probe.For over more than a century the Kelvin probe has been continuously improved and modified for particular applications. In terms of applications, the Kelvin probe has been used in the study of general surface chemistry,15–17 corrosion,18 stress,19 residual contamination in semiconductor technology20 and biocompatibility of materials and histological studies on living cells.21 Finally, the Kelvin probe must be modified in order to perform measurements in liquids,22 at high temperatures,23 in ion or electron emitting samples24 or in an ultrahigh vacuum environment.25 The transition from the classical, static Kelvin probe to the modern commercially-available scanning probe constitutes an impressive technical advance.However, the spatial resolution of this type of instrument, with respect to CPD, is in the range of hundreds of micrometers. Furthermore, such equipment cannot be used to generate tandem topographical image. In the present paper, we describe the design of a scanning Kelvin microprobe (SKM) that operates to a spatial resolution of 1 mm.Additionally, the instrument yields topographical images in a tandem fashion to a high resolution in the nanometric range. Together with a description of the instrument, preliminary studies on various samples are presented. Theory of the Kelvin probe Work function In any solid, liquid or gaseous material containing electrons, the thermodynamic equilibrium state is attained when the free energy F is minimized with respect to the number ni of carriers Analyst, 1999, 124, 961–970 961(F = U 2 TS is the Helmholtz free energy, where U is the total energy, T the absolute temperature and S is the entropy at a volume V).The most convenient thermodynamic quantity to use in describing the state of the system is the electrochemical potential. For the ith component of the system (such as a charged particle; electron) the electrochemical potential is defined by: �m i = (¶F/¶ni) |T,V,nj (1) The work done in bringing dn electrons having zero potential and kinetic energy from infinity up to the interface and dissolving them in the solid is mdn.This work depends on the electrical potential of the solid. Thus, a change in the electrostatic potential by an amount V shifts the electrochemical potential by 2eV. To separate the electrical aspects, the chemical potential of electrons in a solid, m , can be now defined as: m = �m 2 eV (2) The chemical potential depends only on the material and its temperature.It is helpful to remember that the electrochemical potential is identical with the energy parameter that occurs in the Fermi distribution function:2 f F ( ) exp( ) / E E E kT i i = - + 1 1 (3) Here, EF is the Fermi level, where the probability of occupancy is even (equal to 1/2; associated with the surface of the imaginary ‘electron sea’). Thermodynamic equilibrium between two macroscopic systems means that there is an equality of their electrochemical potentials.This corresponds to equality of the Fermi levels among the various systems. This concept will be especially useful in connection with discussion of the parameter, contact potential difference. The work function F is defined as the work required to extract an electron from the Fermi level EF to infinity. Electrons must overcome the potential barrier of the interface before they can escape from the surface.26 At the interface there are dipoles, e.g., water molecules that are partially or totally oriented by the electric field existing at the interface.This layer of oriented dipoles produces an electric double layer, which will contribute to the value of the internal potential of the material to which it is attached. This defines the surface potential parameter, c. Surface layers play a dominant rolor example a surface dipole layer can shift the work function by large and often poorly-controlled values.Such layers are not very well understood.19 At the surface of conductors, the electronic charge redistributes giving rise to the electrical dipole layer, which leads to the electrical potential difference between the exterior and interior of the surface. Every extracted electron influences the surface layer by repulsive electrostatic interaction with the other electrons and attractive interaction with the nuclei. When the electrons are leaving the surface they leave a charged region at the interface behind.At the surface of a semiconductor, band bending caused by surface states acts as a surface barrier. Furthermore, the electrostatic potential energy of an electron just outside a surface is very dependent on the distance from the surface. In the absence of applied electric fields the potential experienced by an electron is given8 by the image potential: V r e r ( ) = - 16 0 pe (4) where r is the distance from the surface.To summarize, there are four factors which contribute to the work function. The first is the structure of the solid. The position of the Fermi level inside conductive, semi-conductive or insulating materials governs the bulk contribution to the work function. The second factor is the work done to overcome the barrier at the surface. Outside the surface, the image force contributes to the work function and the presence of accelerating or retarding electric fields represents the fourth factor which also must be considered. In practice, the measured work function of a material is also usually influenced by a large variety of factors such as temperature, crystallographic properties, stress, surface properties, adsorbed layers, contamination etc.Contact potential difference When two materials are joined together and are in thermal and electrical equilibrium, then, from eqn. (1), the electrochemical potentials for electrons in each material must be equal: m1 = m2 (5) This is achieved by a small flow of electrons from the material with the higher Fermi level until the equilibrium state is reached.The contact potential difference (CPD) is defined as the difference between the two work functions of the solids in contact at temperature equilibrium: DF12 = F1 2 F2 (6) Measurement of the CPD thus affords a method of measuring work function differences between materials. In order to measure the CPD it is necessary to connect the materials.A direct measurement, for instance with a voltmeter, requires a circuit shortened by the measurement device. However, in a closed circuit no CPD can be measured directly (the sum of the three interfacial differences would be zero), except the case where the interfaces have different temperatures. The CPD must be measured in open circuit, i.e., using a dielectric such as vacuum or air. Instrumentation A scanning Kelvin microprobe is under continual development in our laboratory with the overall aim of the characterization of chemically-modified surfaces. The Kelvin current is used to obtain the CPD signal.The current appears because in the local capacitor formed by the tip and the sample under study, the tip is vibrating. The tip is made from a material with a known work function (tungsten). Both ‘parallel plates’ of the condenser are connected by a current measurement device (Fig. 1). When the equilibrium of the Fermi level is established between the tip and a sample, a CPD appears between the ‘plates’ (V), and the capacitor is charged.If the distance between the parallel plates is changing, the capacity is altering and, due to the constant voltage between the plates, the charge on the plates (DQ) changes too. This charging process causes a current in the measurement device, the Kelvin current. If the CPD is compensated by a variable voltage source inserted serially into the circuit (V0), there will be no current flowing in the circuit.The voltage of the source now equals the CPD value. DQ = (V + V0)DC (7) Fig. 1 Principle of CPD compensation and measurement of the Kelvin current. 962 Analyst, 1999, 124, 961–970If V = 2V0, then DQ = 0. In Lord Kelvin’s first experiment,27 this charge variation detected by an electrometer was directly compensated to null by an exterior voltage. Instead of the change of charge, a current can be measured if the capacity is variable in time, as Zissman first proposed:28 I V V C t t = + ( ) ( ) 0 d d (8) The measured current signal I is also nulled (in the same way as proposed by Kelvin), by compensating the CPD signal with the external voltage V0: if V = 2V0, then I = 0.If the tip has a sinusoidal vibration, the separation distance between plates is now: d(t) = d0 + d1 cos wt (9) where d0 is the rest position and d1 is the amplitude of vibration. The capacity C(t) is then: C t A t A d d t ( ) ( ) cos = = + e e w d 0 1 (10) The variation of the probe distance results in time variant charge dQ/dt and the Kelvin current: i t Q t t V V d A t d d t ( ) ( ) ( ) sin ( cos ) = = + + d d 0 1 0 1 2 ew w w (11) The exterior compensating voltage reduces the Kelvin current to zero in the same manner as described above.To measure the CPD on a small scale with high precision it is necessary to control closely the distance between the tip and the sample. This task is implemented by the separation of the harmonics of the Kelvin current, providing, simultaneously, a topographic image when scanning the surface.For V = DF12, the potential difference equals the (time independent) CPD between tip and sample. The Kelvin current in the time domain is (eqn. 8): i t V C t t ( ) ( ) = d d (12) In the frequency domain, the Kelvin current can be obtained through a Fourier analysis: i t I k t A d m kz kwt k k l k k K( ) sin( ) sin( ) = = - × × - = • = • Â Â w 1 12 0 2 1 2 1 DF e w (13) where: DF12 is the CPD between the tip and the sample, A is the effective area built by the tip and the sample, k the number of the harmonic and m = d1/d0 the modulation factor.Ik is the amplitude of component k and zl is a parameter depending on the modulation factor m. By dividing the first harmonic through the second harmonic, a signal sd can be obtained, which depends only on the modulation factor m: s I I z z m m d l = = = × - - 1 2 1 2 2 2 1 2 1 1 (14) The instrument has a distance control unit which holds sd constant, so the distance d0 can be forced to a constant value.The vibration amplitude d1 does not change during the measurement. The CPD information can be extracted from either of the harmonic amplitudes. The first harmonic has the best sensitivity for CPD measurement and the amplitude I1 is used for CPD information. From eqn. (13): I A d z m 1 12 0 1 2 2 1 = × × - DF e w (15) The parameters on the right side of eqn. (15) do not change during the measurement, if I1 is proportional to the CPD value.The instrument only requires the Kelvin current in order to measure the distance (topography) and the work function of a sample. Measurement configuration Fig. 2 presents a schematic diagram of the instrument. The system consists of five units; a measurement system, a currentto- voltage converter, a distance control unit, scan and signal collection units. The sample is fixed on the scanning table, which can move in the directions of the x-axis and the y-axis in order to scan the sample and along the z-axis in order to Fig. 2 Block-diagram of the instrument. Analyst, 1999, 124, 961–970 963maintain a constant distance between the tip and sample. The driving element is the Nanomover system from Melles Griot, Darmstadt, Germany. Above the sample, the tip, attached to a vibrating piezoelement, is mounted on a support. The piezocrystal, acting as an actuator for the distance control and the sensing component (the vibrating tip) are from Topometrix, CA, USA.The frequency of the vibration (2 kHz) is generated by a frequency generator or can be generated at an output of a computer data acquisition card. This signal is amplified and fed into the vibrating piezoelement. The position of the scan table is adjusted by two stepper DCmotors, which form, along the micropositioner, the components of the scanning mechanism precisely controlled by the scanning program.The whole system is operated with a PC, which functions as the communication interface between the operator and the instrument. The Kelvin current originating from the tip is converted to a voltage and amplified. This voltage is fed into 2 parallel bandpass filters in order to obtain the first and the second harmonics of the Kelvin signal. Further, the outputs of the filters are rectified, so DC-values proportional to the amplitudes of both harmonics are produced. The ratio between these two signals is calculated by an analog divider.This ratio is compared with a set point, the difference is processed by a PIcontroller. The output of the controller is recorded by the computer. The same signal is, after a further amplification, fed into the scanning table piezoelement, thus correcting the distance. When the average distance between the tip and surface decreases, the amplitude of the higher harmonic rises faster then the amplitude of the lower harmonic, causing a monotonic increase of the ratio between both amplitudes. So, by preadjusting the ratio to a constant value, a constant distance between the tip and the surface can be maintained.The output of the controller is recorded and can be used for producing a topographical image. The first and second harmonics of the Kelvin current are mainly used for distance control, because these are the harmonics with the highest amplitudes. If the first harmonic is distorted by the input of the oscillating piezoelement, higher harmonics can be used for distance control.Performance The amplitude of the first harmonic can be expressed as: I A V d m z 1 0 0 2 1 2 1 = + - × e w F (D (16) Where V0 is the automatically-controlled compensating voltage which is applied between the tip and the sample in order to compensate the amplitude of this first harmonic to zero. At this zero-current point, DF12 + V0 = 0, so: DF12 = 2V0 (17) This relation is, in principle, independent off all other factors, including e, A, w and d0.I1 can be expressed as: I1 = M·(DF12 + V0) (18) M A z d m M = - × 2 1 1 0 2 e w With being the value of the transadmittance of the current sensing unit. If the smallest detectable value for the first harmonic is I1min, then the precision s of the CPD measurement can be expressed as: s F F F = - = V I M 0 12 12 1 12 D D D min (19) The limitation of the minimum measurable signal I1min in the scanning Kelvin microprobe is set by the I/U converter. Thus, for a given DF12 the precision is only determined by M.An improvement in the precision of the CPD measurement can be achieved by raising the value of M, and this can be done by increasing the tip area (unsuitable for high lateral resolution) or reducing the tip-sample distance. A lateral resolution of 1 mm in CPD and a vertical resolution of 20 nm with respect to topography are possible with the present version of the instrument.A sensitivity of 1 mV was recorded. The actual resolution depends on many factors, for instance, on the value of the Kelvin current and on the signal-tonoise ratio of the I/U-converter. By increasing the lateral resolution, the CPD resolution decreases, because the use of a sharper tip affects the measured current by reducing its value. To compensate the drop of the current the tip must be presented closer to the surface, increasing the risk of touching the surface or discharging the voltage of the capacitor.A compromise must be found according to the particular application. In contrast, if a higher CPD resolution is required for a particular sample, this can be achieved by using a larger tip (thereby reducing lateral resolution). Measurement of insulating surfaces In the past, the Kelvin probe method was used exclusively to measure metallic or semiconductor surfaces. Since the method does not require a conductive sample surface, insulating samples can also be successfully investigated.If the dielectric constant of the sample is es, (Fig. 3) calculation of the Kelvin current is based mainly on the model of two capacitors connected in series (the previous ‘air capacitor’ and additional ‘sample capacitor’). Therefore the single capacitor model must be modified. If the distance between the tip and the surface of the conductive table is dc + d1 coswt, then the distance between the tip and the surface of the insulating sample is: d(t) = dc 2 ds + d1 cos wt (20) The index s indicates the sample.The equivalent capacity can be calculated as: C t C C d t A d A A d d d t ( ) ( ) [( / ) ] cos = + = + = - + + 1 1 1 1 1 0 1 s s s s c e e e e w e (21) Comparing eqn. (21) with eqn. (10), the modulation factor should now be defined as: m d d d s s s c = - + 1 1 ( / ) e e (22) The controlling signal sd [eqn. (14)] maintains a constant ms, which in turn holds (e/es 2 1)ds + dc constant to the value d0, namely the tip-to-sample distance for a conductive sample surface without an insulating layer.Using eqn. (22) the real distance between the tip and the conductive surface of the metallic table carrying the insulating table can be given as: Fig. 3 Measurement of an insulating sample. 964 Analyst, 1999, 124, 961–970d d d d d c s s s s = - - Ê Ë Á � � � = + - Ê Ë Á � � � 0 0 1 1 e e e e (23) This modification of the value of the Kelvin current in such a way that the distance control unit can automatically vary the distance associated with the presence of the insulating sample avoids unexpected tip-to-sample contact during the measurement and protects the tip from damage. Measurements with an SKM are normally carried out in air, so that the normal relation is es > e � e0, which means that (1 2 e/es) is positive.Compared with a region without insulating material, where dc = d0, the distance between the tip and the conductive surface of the table is increased by a value of (1 2 e/es)ds.This means when measuring an insulating sample, or an insulating area on a conductive sample, the instrument will withdraw the tip, to avoid the contact of the tip with the non-conductive layer. This property of the distance control system is an important qualitative advantage for the SKM. However, the maximum allowed thickness dsmax of the insulating layer can be calculated by setting dc = ds in eqn. (23), which gives: d d smax s = e e 0 (24) Example SKM images The examples that are described in the following text represent both CPD and topographical images of the same surface measured in air, at room temperature.In terms of recording time, with the present equipment 50 3 50 pixel images can be obtained in about 20 min of scan time if the tip is allowed to pause for 50 ms at every measurement point until local equilibrium is reached. At the present time, the color code on the Fig. 4 Surface topography (A) and CPD (B) images for the deposition of Ag onto a silicon wafer.Analyst, 1999, 124, 961–970 965right side of the following CPD images represents relative values of potential difference as indicated by the scale. We have discovered that significant discrepancies exist in the literature with respect to the ‘calibration’ of contact potential difference. Great care should be exercised in employing published data for different materials. For example, the reference surface should be ideally clean and inert—it is not possible to avoid physical or chemical adsorption except by isolation in vacuum, a requirement that introduces considerable practical difficulties.Accordingly, since our measurements are not performed in high vacuum but in air, we are not concerned with absolute values of work function or surface potentials. What is important in our measurements is a comparison of work function values for samples that have various origins, in addition to particular surface treatments.This is a particularly valuable feature for special applications in biology on which we are currently engaged. Fig. 4 shows an examination of the surface topography (A) and CPD (B) for a silicon wafer onto which a silver film had been fed (5 31025 Torr, 50 °C, layer thickness: 1000 nm). Note that the topographical image is, for the most part, very uniform as expected. However, there is an indication of the formation of a relatively narrow step at the top of the image. Interestingly, the CPD image does display bands of variable contact potential that exhibit maximum value towards the centre of the image.We attribute these to relatively small chemical changes over the same surface as the topographical image. These alterations, which apparently occur in mm bands, are very likely associated with oxidation of the silver surface (no special care was taken to store the metal surfaces in a high vacuum or inert gas atmosphere before recording the SKM images.) X-ray photoelectron spectroscopy was conducted on an analogous sputtered surface under similar conditions.Although recorded over a relative large area compared to the Fig. 5 Surface topography (A) and CPD (B) images for the masked sequential deposition of Pd then Pt onto silicon. 966 Analyst, 1999, 124, 961–970CPD image, the XP spectrum clearly shows evidence of the surface presence of oxygen (relative atomic per cent., 14.8), but no signal due to sulfur.Although this could not be confirmed from the binding energies of the Ag3d peaks, we still attribute the bands in the CPD images to areas of the surface corresponding to slightly different levels of silver oxidation, a not implausible explanation. An interesting feature of a comparison of the two images is the lack of correlation between topography and the ‘chemical’ picture. This behavior appears in other cases discussed subsequently. The pair of images in Fig. 5 relate to an experiment involving the successive deposition of layers of palladium, then platinum by sputtering onto a silicon wafer (5 31025 Torr, 50 °C, each layer thickness: 1000 nm). During the deposition of Pt a physical mask was placed at the right hand side of the represented surface. Accordingly, the topographic image (A) depicts relatively higher values at the left hand side, particularly in the bottom corner. In this case the CPD image (B) quite accurately reflects the difference in contact potential for these two metals (which is 5.12 for palladium and 5.65 for platinum).29 Furthermore, there is an apparent ‘edge’ running across both pictures, approximately at the centre in each case.In contrast to the correlation between the two images for the previous experiment, SKM analysis of a sample of highlyordered pyrolitic graphite (HOPG) shows no correlation between topographic (Fig. 6A) and CPD measurement (B). The former image shows large band widths (100 mm plus) of quite variable height, whereas the latter displays somewhat smaller features which, presumably, are associated with levels of oxidation (unrelated to topography).Figs. 7A and B indicate that the SKM technology is of great potential value for the characterization of microelectronic circuits. These images were obtained from a integrated resistor array. On a coarse level there is a general correlation between topography and CPD value.Note the appearance of quite sharp edges ‘spikes’ in both structures towards the top and bottom of Fig. 6 Surface topography (A) and CPD (B) images from highly-ordered pyrolitic graphite. Analyst, 1999, 124, 961–970 967the pictures and the occurrence of roughly two domains which are occurring in the left one-third of the images. However, a comparison of the topographic and CPD images also shows areas where there is no correlation at all. For example, the CPD value varies considerably in the centre of the image, but this is clearly not reflected in the topographic picture.Typical examples of SKM images obtained from a nonconductor are those obtained from muscovite mica (Figs. 8A and B). A sample was prepared by removing cleavage mica planes from half the area of the particular surface under investigation. This yields a step between planes at the surface. The topographical image exhibits this particularly well as can be seen by the running edge from the bottom left corner.As entirely expected the CPD image is devoid of any such information, although small islands of different contact potential are indicative of the presence of contaminating particles on the mica surface. A final interesting example of SKM study of a non-conducting surface concerns the results we obtained during a study of various samples of biomedical origin. Figs. 9A and B are both CPD images of samples of silicone polymer. In this case, the origin of this material was the surgically-removed breast implant from a patient who chose explantation consequent to the recent high-level controversy regarding the medical consequences of silicon polymer-based-breast implants. 30 Both the samples here were obtained from the same implant; however, one was isolated from the inside surface of the structure closest to the chest cavity (A) whereas the other was in contact with biological tissue towards the outside (B).First, the images demonstrate the power of the capability to represent CPD on a pseudo three-dimensional basis, that is, with spatial (xy plane) data plotted together with variation in actual CPD level (z ‘plane’).In this case the topographical images (not shown) exhibit identical smooth surfaces, whereas it is clear that the surface exposed to different tissue possesses not only an Fig. 7 Surface topography (A) and CPD (B) images obtained from a resistor array. 968 Analyst, 1999, 124, 961–970altered overall level of CPD, but also more variation than the other sample.Specifically, on average, the CPD values for the former surface are somewhat lower than the latter. This result demonstrates that the SKM offers tremendous potential for examining surface functional group chemistry, potentially at the sub-micrometre level. Conclusions The examples depicted above confirm that the scanning Kelvin microprobe represents a valuable new tool for analysis of the surfaces of conducting, semiconductor or insulator materials.Through the quantitative measurement of the CPD signal, sample property variation associated with contamination, processing and changes of chemical properties can be detected, evaluated and compared. In a number of cases the measurement of the surface topography is not the main objective of an SKM study. However, on the other hand, the instrument is capable of the discovery of underlying variations which can not be seen with an optical system.More importantly, the distance control mechanism guarantees the precision of the CPD measurement, which can define subtle variations of the work function which can not be detected by other means. Further improvements are envisaged in the attainment of higher lateral resolution, avoidance of the effects of high electric fields on very close approach to the surface and improvement of distance control stability and reliability. Efforts are underway to improve the construction of the tip and electronic control in order to enhance the measuring capabilities of the instrument.Support for this work from Materials and Manufacturing Ontario and the Natural Sciences and Engineering Research Council of Canada is very gratefully acknowledged, Further, the authors much appreciate assistance from the Science and Technology with European Partners Program of the Department Fig. 8 Surface topography (A) and CPD (B) images from a sample of mica.Analyst, 1999, 124, 961–970 969of Foreign Affairs and International Trade, Canada and the International Bureau of the GKSS Research Center, Germany which fostered the Canadian–German collaboration evident in this work. Finally, we are indebted to Dr. W Peters of Mount- Sinai Hospital, Toronto for the provision of breast implant samples and to D. C. Smith and S. Lugowski of the Institute of Biomaterials and Biomedical Engineering, at the University of Toronto, for much helpful discussion regarding the surfaces of medical implants in general.References 1 C. Herring and H. Nichols, Rev. Mod. Phys., 1949, 21, 185. 2 R. H. Fowler, Phys. Rev., 1957, 107, 1553. 3 E. W. Muller and T. T. Tsong, in Field Ion Microscopy, Elsevier, New York, 1969, p. 70. 4 T. Delchar and A. Eberhagen, J. Sci. Instrum., 1963, 40, 105. 5 C. W. Oatley, Proc. Roy. Soc., London A, 1936, 155, 218. 6 M. Green, Solid State Surface Science, ed. M. Green, Marcel Dekker, New York, 1969, vol. 1, p. 196. 7 H. Shelton, Phys. Rev.,1957, 107, 1553. 8 A. G. Knapp, Surf. Sci., 1973, 34, 289. 9 P. A. Anderson, Phys. Rev., 1935, 17, 958. 10 M. Yasutake, J. Appl. Phys., 1995, 34, 3403. 11 M. Yasutake, A. Daisuke and M. Fujihira, Thin Solid Films, 1996, 723, 279. 12 M. Nonnenmacher, M. P. O’Boyle and H. K. Wickramasinghe, Appl. Phys, Lett., 1991, 58(25), 2921. 13 M. Nonnenmacher, M. P. O’Boyle and H. K. Wickramasinghe, Ultramicroscopy,1992, 42–44, 268. 14 M. Bomisch, F. Burmeister, A. Rettenberg, J. Zimmermann, J. Boneberg and P. Leiderer, J. Phys. Chem. B, 1997, 101, 10162. 15 J. C. Tracy and J. M. Blakely, Surf. Sci., 1968, 13, 313. 16 H. C. Potter and J. M Blakely, J. Vac. Sci. Technol., 1975, 12(2), 635. 17 M. Schmidt, M. Nohlen, G. Bermes, M. Bomer and K. Wandelt, Rev. Sci. Instrum., 1997, 68(10), 3866. 18 Y. Shelgon and R. A. Oriani, J. Electrochem. Soc., 1991, 138(1), 55. 19 P. Craig and V. Radeka, Rev. Sci. Instrum., 1970, 41(2), 258. 20 W. Nabhan, B. Eqer, A. Broniatowski and G. De Rosny, Rev. Sci. Instrum., 1997, 68(8), 3108. 21 S. Yamashina and M. Shigeno, J. Electron. Microsc., 1995, 44, 462. 22 K. A. Macfaden and T. A. Holbeche, J. Sci. Instrum., 1957, 34, 101. 23 J. Nowothy, M. Sloma and W. Weppner, J. Am. Ceram. Soc., 1989, 72(4), 564. 24 B. H. Blott and T. J. Lee, J. Sci. Instrum., Ser. 2, 1969, 2, 785. 25 S. J. Danyluk, J. Sci. Instrum., 1972, 5, 478. 26 R. G. Forbes, in Scanning Tunnelling Microscopy and Related Methods, ed. R. G. Forbes, Kluwers Academic Publishers, 1990, p. 163. 27 (Lord) Kelvin, Philos. Mag., 1898, 46, 82. 28 W. A. Zissman, Rev. Sci. Instrum., 1932, 3, 367. 29 CRC Handbook of Chemistry and Physics, ed. D. Lide, CRC Press, London, 1992–1993, 12, 108. 30 B. A. Cavic, M. Thompson and D. C. Smith, Analyst, 1996, 121, 53R. Paper 8/09895F Fig. 9 CPD images of silicon-based polymer obtained from explanted breast implant. (A) represents the inside surface (closest to the chest cavity) and (B) the surface in contact with biological tissue towards the outside. 970 Analyst, 1999, 124, 961–970

ISSN:0003-2654    

DOI:10.1039/a809895f

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Chemometric techniques for exploring complex chromatograms: application of diode array detection high performance liquid chromatography electrospray ionisation mass spectrometry to chlorophyll a allomers Konstantinos D. Zissis, Samantha Dunkerley and Richard G. Brereton* School of Chemistry, University of Bristol, Cantock’s Close, Bristol, UK BS8 1TS Received 15th March 1999, Accepted 26th May 1999 Triply coupled diode array detection high performance liquid chromatography mass spectroscopy is applied to a complex mixture of at least eight chlorophyll degradation products.Derivatives are employed to determine parts of the chromatogram of composition one. Mass selection is performed on the mass spectroscopic data. Principal components analysis is performed on both the raw and simultaneously normalised/standardised data; three dimensional projections of the data are obtained and compared to conventional two dimensional graphs. Angular plots between diode array loadings characteristic of individual compounds and scores of the diode array data are described.In mass spectra, angular plots between loadings characteristic of individual compounds and the remaining diagnostic masses reveal further mass spectral structure. 1 Introduction In previous papers, we have proposed a number of approaches for exploratory data analysis of complex chromatograms, such as those arising from chlorophyll degradation products, primarily restricted to two dimensional principal components (PC) plots1–3 applied to diode array detection high performance liquid chromatography (DAD-HPLC).In recent papers,4,5 we propose that more information can be obtained by simultaneously monitoring a chromatogram using both mass spectrometry and electronic absorption spectroscopy as applied to a simple example of 2- and 3-hydroxypyridines. In this paper, we extend the methods to a more complex real world case study. In the previous case of hydroxypyridines only two co-eluting compounds were analysed, and so the mass spectrometric and diode array PC plots could be superimposed, e.g., by use of procrustes analysis.4,6,7 However chlorophyll degradation products pose different problems.Compounds with identical electronic absorption spectra (EAS) may have different allomers. For example, all compounds of structure class ‘X’ in Fig. 1 have identical EAS. However, compounds I and II have different molecular weights and so different mass spectral features.Hence, whereas I and II will cluster closely in DAD, they will not in mass spectrometry (MS). In contrast, compounds of groups ‘Y’ and ‘Z’ have very different EAS because of ring cleavage, but identical molecular weights and so many features in common with their mass spectra. Hence, it is not possible to completely overlap the MS and EAS scores plots. Another complexity is that there are several groups of compounds with different features, for example, compound groups ‘X’, ‘Y’ and ‘Z’ have distinct EAS.Two PCs are insufficient to adequately describe these groups, and so three PCs form a better model of the data. Data processing ability is improved by handling three dimensional plots. In this paper, we explore some approaches for projecting these onto two dimensions in order to obtain best discrimination. The information in such plots can be further analysed graphically to reveal more information about the system by calculating angles between characteristic wavelengths or masses.For example, the angle between wavelengths characteristic of each compound group (‘X’, ‘Y’ and ‘Z’) and the scores can be computed and plotted as a graph against time, providing Fig. 1 Structures referred to in this paper. I = Chlorophyll a, II = 132-hydroxy-chlorophyll a, III = Mg(ii)-31,32-didehydro-151hydroxy- 151-methoxy-rhodochlorin-15 acid d-lactone152methyl 173-phytylester, IV = chlorin e6-131-152-dimethyl-173-phytyl ester, V pheophytin a (A refers to the epimers, m.wt refers to molecular weight).Analyst, 1999, 124, 971–979 971information on the spectral similarity, analogous to the orthogonal projection approach. The lower the angle, the more likely the spectra belong to the test group. Using three rather than two dimensions improves this. In mass spectrometry, the loading of the mass number characteristic of an individual compound can be determined and the angle between this mass number and other masses provides further information on the most significant fragments. 2 Methods 2.1 Extraction of chlorophyll a Chlorophyll a was extracted from spinach leaves and purified by open column chromatography using a method described in an earlier paper3 and based on methods by Svec,8 Iriyama et al.,9 and Brereton et al.10 Pure chlorophyll a, dissolved in methanol, was then subjected to illumination at 4000 lx, in a water bath at 40 °C, in the presence of oxygen.3 For sunlight at sea-level, 250 lx corresponds to 1 W m22 power per unit area. After 60 min, a sample was taken, solvent removed in vacuo, the dried sample dissolved in acetone and injected to DADHPLC. 2.2 Chromatographic conditions and detection The HPLC instrument used was a Waters system (Waters Corporation, Milford, MA, USA), equipped with a 600S Controller, a 616 LC Pump and a 717 Plus Autosampler. The stationary phase was a Kromasil C18 reversed-phase column (25 3 0.46 cm id) with a particle size of 5 mm.The mobile phase consisted of a three-solvent gradient system, described in detail previously.11 All solvents used were HPLC grade (Rathburn, Walkerburn, Scotland) and the water was de-ionised and prepared using a Milli-Q filtration unit (Millipore Corporation, Milford, MA, USA). The flow rate was set at 1 mL min21 and an 8 ml injection was performed. The analysis was carried out at ambient temperature. Detection was performed using a 996 photodiode array (PDA) Detector (Waters Corporation). Electronic absorption spectra were recorded at 1 s intervals, between 350 and 800 nm, with a 2.4 nm bandwidth resolution.The MS analysis was performed using a VG Quattro Mass Spectrometer (Fisons Instruments, Altrincham, UK). Electrospray-ionisation (ESI) in the positive mode was chosen as the method of ionisation and mass spectra were recorded every 1.25 s, between 500 and 1000 mass units. The source temperature was set at 80 °C and the cone voltage applied to the source was 65 V.The cone voltage, altered accordingly, can have a dramatic effect on the fragmentation pattern of the peaks. In ESI it is common to form Na ion adducts with the compounds, meaning that masses shifted by 23 units are observed as well as the major fragmentation ions. It is also common for one or more protons to add to the molecular ions and Na adducts. The two detectors were linked together, using the configuration described elsewhere.4 After the stationary phase, a splitting device was used between the DAD and the MS detectors, to ensure the flow rate entering the two detectors simultaneously was halved. The splitter consisted of a Peek T-piece with finger tight fittings, fitted with Peek tubing of 0.50 mm id and 1/16B outside diameter. 2.3 Decoding programs and data transfer The data acquired from the HPLC-DAD instrument were decoded using a combination of two macros. The first is the 2010 DDE Assistant for Raw data macro (Version 2.10, Waters Corporation) that runs in Excel (Version 5.0a, Microsoft).This extracts the raw data into a two-column file, which is then converted into a matrix using a second in-house VBA macro. For the MS data, the MassLynx software (Version 2.1, Micromass, Altrincham, UK) which runs under Windows (3.1, Microsoft) was used. The data was decoded using a C++ program written by R. L. Erskine. All other pre-processing methods and algorithms for data handling and analysis were written either as VBA macros for Excel (Version 5.0a, Microsoft), or in MatLab (Version 4.2c.1, Mathworks, Inc.). 2.4 Selecting chromatograms, interpolation and alignment The HPLC-DAD instrument allows us to collect data throughout the entire chromatogram. However, using the 2010 DDE Assistant for Raw data macro, the 3D information extracted can be reduced to the time range of particular interest. Thus, for the HPLC-DAD data, the time selected was approximately between 33 and 38.5 min (in a total runtime of 40 min), where peaks of interest eluted.However, for the HPLC-MS data, storing all data over the entire 40 min would require enormous disk space, thus, data was acquired only between 30 and 40 min. The acquisition rate of 1.25 s was the maximum allowed, as the mass range (500 masses) was quite large. As the scan rates of the two detectors were different (1 s for HPLC-DAD versus 1.25 s for HPLC-MS), the data were interpolated to the same scale of 1 s.The interpolated data from both detectors were then aligned, and this was done by locating the main chlorophyll a peak for both detectors and marking it at the same point in time. The time scale was then changed to 0–330 s over the 5.5 min region, with chlorophyll a eluting at 233 s. Before any prior data treatment, both the HPLC-DAD and HPLC-MS data were baseline-corrected, according to the methods described in detail in an earlier paper.4 2.5 Principal components analysis Principal components analysis (PCA) is a method often used to reduce the dimensionality of measured data.12–16 For example, an I 3J data matrix, X, can be decomposed into an I3K scores matrix, T, a K 3 J loadings matrix, P, and an I 3 J matrix, E, according to the following equation: X = T · P + E where: I are the number of scans (points in time), J the number of variables (wavelengths or mass units) and K the number of principal components.The scores relate to the compounds’ concentrations, whereas the loadings relate to their spectra. The number of significant principal components should ideally correspond to the number of compounds present in a mixture. Principal components are extracted in succession, using the non-iterative partial least squares (NIPALS) algorithm, originally developed by Wold.17 The first PC is usually the most significant, with subsequent PC’s describing less information and finally representing pure noise. 2.6 Data pre-processing 2.6.1 Normalisation and standardisation. Normalisation, followed by standardisation, is a common way of preprocessing data. This ensures that both the scans (rows) and the variables (columns) of a matrix are scaled accordingly, so that they acquire equal significance. In normalisation, each value in a scan is divided by the sum of values in that row, so that the total for each scan becomes equal to one. This is performed in order that effects of concentration between successive data points in time are 972 Analyst, 1999, 124, 971–979removed.For a data matrix, X, normalisation takes place using the equation: n i j i j i j i J x x x , , , = Â1 where xi,j is a point at time i and wavelength or mass j. Standardisation then follows, by subtracting the variable mean from the values in that variable and dividing this by the standard deviation for that variable. This is nearly essential when comparing results from different sources, and particularly evident in MS, where intensities can differ by an order of magnitude.Standardisation takes place using the following equation: s i j i j j i j j i I x x x x x I , , , ( ) ( ) = - - - = Â2 1 1 2.7 Mass selection From the 500 masses recorded in the HPLC-MS data, only a small sub-set contains information descriptive of peaks. The rest of the masses are usually of very low intensity and it is impossible to distinguish between noise and peaks.In this particular study, 70 masses were retained. There are various methods for selecting diagnostic masses,4,18 but the main consideration chosen in this study was the order of significance according to their variance to mean ratio (over the 5.5 min region). The higher this ratio, the more significant the individual mass. The masses chosen were used in the calculations of the MS chromatographic profiles and derivatives for the HPLC-MS data. 2.8 Derivatives Derivative plots are a method commonly used to identify the different composition regions in a chromatogram of mixtures.3,4 Usually, four main steps are involved, as described below.The first step is to normalise the data at each point in time, so that all scans acquire equal significance. This is performed using the first equation, given in section 2.6.1. In this application, the five point Savitsky–Golay smoothed first derivative is calculated at each wavelength or mass unit using the following formula: di,j = (22nxi22,j 2 nxi21,j + nxi+1,j + 2nxi,j+2,j)/10 If a peak is pure over five points in time, the normalised spectral intensity should not vary significantly resulting in a derivative close to zero.In order for all variables to assume equal significance, the absolute values of the individual derivatives are renormalised at each wavelength or mass, using the equation: n i j i j i j i I d d d , , , = = Â1 Finally, the weighted average of the derivative at each point in time is calculated using the following formula: d d J j i j i I = = Â, 1 Points in time with a value of derivative close to zero should correspond to a pure compound, whereas points with a high value could either correspond to a mixture of compounds or noise.It is often recommended that for noisy data, the log of the derivative versus time is plotted. Then, points with the most negative values should correspond to composition 1 regions. 2.9 3D PCA projections The scores and loadings extracted from performing PCA on the various modes of treated data were rotated on a 3D space, to enhance visualisation of the pure compounds in different coordinates.This was achieved by multiplying the pure scores T, or loadings P, matrices by their corresponding rotation matrix R. The 3 3 3 rotation matrix R was obtained by multiplying individual 3 3 3 matrices Qfor each of the XY, YZ and XZ coordinates, according to the following equation: R = QXY .QXZ . QYZ The generation of one such matrix, QXY, involving rotation in the XY plane by q degrees, is described in Table 1. The overall rotation, Q, for the scores matrix, T, is given by the following formula: Q = T · R In obtaining the PC plots for both 3D scores and loadings, the first two PC’s were kept and plotted versus one another. 2.10 Angular plots The directions of the pure loadings and scores for each component should be similar, in the three dimensional PC plots.Hence, by calculating the angle between an individual wavelength in the loadings plot and the score of a composition 1 plot, it is possible to determine how well correlated the wavelength is to the pure component and so how diagnostic. Having identified the points of pure composition 1 regions by the derivatives, plots of angle versus time and mass number were obtained for the HPLC-DAD and HPLC-MS data, respectively For the HPLC-DAD data, values of angles (in degrees) were obtained between each loading corresponding to a diagnostic wavelength and each set of scores, and plotted against scan number, according to the formula: DAD –1 DAD DAD DAD DAD = cos f × Ê Ë ÁÁÁÁÁÁÁÁ � � �������� = = =    x y x y p k i k k K p k k K i k k K , , , , 1 2 1 2 1 where: DADxp,k is a value of a loading at the kth PC, characteristic of a diagnostic wavelength and DADyi,k is a value of a score at the kth PC of the 3D scores matrix.For the HPLC-MS data, values of angles were obtained between each loading corresponding to a diagnostic mass and each set of loadings and plotted against mass number, according to the equation: Table 1 Generation of a QXY matrix for a rotation of q degrees in the XY plane X Y Z X Cos (qxy) 2Sin (qxy) 0 Y Sin (qxy) Cos (qxy) 0 Z 0 0 1 Analyst, 1999, 124, 971–979 973MS –1 MS MS MS MS = cos x × Ê Ë ÁÁÁÁÁÁÁÁ � � �������� = = =    x y x y p k m k k K p k k K m k k K , , , , 1 2 1 2 1 where: MSxp,k is a value of a loading at the kth PC, characteristic of a diagnostic mass e kth PC of the 3D loadings matrix. 3 Results 3.1 Chromatographic elution profiles and derivatives The chromatograms for both HPLC-DAD, summed over the entire wavelength range (and then normalised to a maximum of 1), and for HPLC-MS, after the 70 most significant masses were selected, the data summed over these significant masses and normalised to a maximum of 1, are displayed in Fig. 2. Compounds are numbered from 1 to 8, with chlorophyll a being peak 6. The HPLC-MS elution profile is noisier than the corresponding HPLC-DAD, peaks 1 and 8 are not detected and, peak 3 is not clearly separated from peak 4 (Fig. 2b). Fig. 3 displays plots of the logarithm of derivatives versus time for both the HPLC-DAD and HPLC-MS data. Whereas all eight pure composition 1 regions can be detected using the derivatives for HPLC-DAD, only six pure regions are detected by HPLC-MS, although peak 3 is now more clearly differentiated from peak 4. 3.2 2D PC plots 3.2.1 HPLC-DAD data. The effect of data pre-processing on the appearance of the 2D plots of scores and loadings was investigated. The discussion is focused on two forms of data namely raw and simultaneously normalised and standardised data. Fig. 4 displays 2D plots for the scores and loadings of raw data, obtained using HPLC-DAD. In Fig 4a it is apparent that peaks 5 and 6 dominate in the analysis and lie in totally different directions—these correspond to the most intense peaks in the chromatogram both with different spectral characteristics.An expansion of the plot shows that peaks 2 and 7 lie in the same Fig. 2. Chromatogram of chlorophyll a degradation products, normalised to a maximum of 1, using (a) HPLC-DAD and (b) HPLC-MS. Fig. 3 Plot of logarithm of derivatives versus time for the chlorophyll a degradation sample monitored by (a) HPLC-DAD and (b) HPLC-MS. Fig. 4 2D plot for (a) the scores and (b) the loadings of PC2 versus PC1 for the raw HPLC-DAD data (expanded as appropriate). 974 Analyst, 1999, 124, 971–979b a b a direction as 6, whereas peak 4 is in a different direction. In addition, peaks 1 and 8 lie in the same direction as peak 5. The characteristics of the compounds expected in this type of chromatogram have been described in detail in a previous paper,5 from where it was suggested that peaks 6 and 7 correspond to chlorophyll a, I, and its epimer, IA, whereas peak 2 corresponds to 132-hydroxy-chlorophyll a, II.Since all three compounds exhibit identical spectra, they all lie in the same direction in a PC plot. Peaks 5 and 1 correspond to Mg(ii)- 31,32-didehydro-151hydroxy-151-methoxy-rhodochlorin-15 acid d-lactone152methyl 173-phytylester, III, and its epimer, IIIA, and lie in the same direction. Peak 8 corresponds to pheophytin a, V, in which the central Mg atom is replaced by two hydrogens. Peak 4 is likely to be chlorin e6-131-152-dimethyl- 173-phytyl ester, IV, and having been found to display a different spectrum to all other chlorophyll a allomers,5 lies in a Fig. 5 2D plot for (a) the scores and (b) the loadings of PC2 versus PC1 for the normalised and standardised HPLC-DAD data. Fig. 6 2D plot for (a) the scores and (b) the loadings of PC2 versus PC1 for the raw HPLC-MS data. Table 2 List of the main MS fragment ions for the six compounds detected by HPLC-MS Peak no.[M+H]+ , [MH2 .]+ [M+H]+ + Na [M-Phy]+, [M+H-Phy]+ [M-Phy]+ + Na [M+H-Phy]+ + Na 2, 3 909, 910 932 630 653, 654 4, 5 939, 940 962 660, 661 683, 684 6, 7 893, 894 916 614, 615 637, 638 Fig. 7 2D plot for (a) the scores and (b) the loadings of PC2 versus PC1 for the normalised and standardised HPLC-MS data. Fig. 8 Plot of angle (in degrees) versus time between the three loadings corresponding to wavelengths characteristic of different compounds (from top to bottom: 692, 665 and 652 nm) and all scores (raw HPLC-DAD data).Analyst, 1999, 124, 971–979 975b a separate direction to the other peaks, too. It is important to recognise that the molecular weights of III and IV are identical, whereas their structures are different so they exhibit different directions in the HPLC-DAD scores plot. From the plot of PC2 versus PC1 for the loadings (Fig. 4b), some information on the wavelengths characteristic of these compounds can be deduced. The wavelengths at 421 and 652 nm, which are the wavelengths of maximum absorption at the blue and red regions of III and IIIA respectively, lie at the end of inflection points in the lower part of the plot (the same direction as in the scores plot for peaks 1 and 5).The wavelengths at 430 and 665 nm (wavelengths of maximum absorption at the blue and red regions of I, IA and II, respectively) lie in the upper end of the plot, although their position is not particularly distinguishable. It has to be noted that it is normally rather difficult to obtain information on the characteristic wavelengths of these compounds, as their maxima lie within a few nm apart.Fig. 5 displays PC plots for the scores and loadings of normalised and standardised data obtained using HPLC-DAD detection. The 2D plot for the scores clearly identifies peaks 1 to 7 as points of inflection. Peak 8 corresponds to an inflection point, too, but it is not shown on the plot, as its signal is very negative and so has been removed from the graph, although included in the pre-processed data set.It is interesting to note that the pheophytin has a very different spectrum from the metallated chlorins, which is strongly reflected in the scores plot of the normalised and standardised data, but is much less evident in the raw data because the peak intensity is not very high. The remaining seven peaks fall into clusters, which illustrates the effect of data pre-processing. For example, as peaks 6 and 7 have identical spectra, they lie very near each other.In the loadings, the wavelengths now appear more spread out. The wavelength at 430 nm clearly falls in the direction corresponding to peaks 6 and 7 (in the scores PC plot), whereas the wavelength at 692 nm (characteristic wavelength of maximum absorbance for IV), falls in the direction corresponding to peak 4 (not observed in the corresponding plot for the raw data). 3.2.2 HPLC-MS data. Fig. 6 displays PC plots for the scores and loadings of raw data with ions characteristic of certain groups of compound coloured, obtained using HPLC-MS.Peaks 2–7 are identified as points of inflection, with I and IV dominating the analysis, and lying at approximately 90°. Note that the pheophytin is no longer detectable which is probably due to the mass spectral ionisation technique. The loadings give the masses characteristic for each set of compounds. Table 2 lists some of the main ions characteristic for each set of compounds.Although the masses for IV and III (coloured in blue) can be clearly distinguished from those characteristic of I and IA (coloured in red), the masses for II (coloured in green) lie in the same direction as those for IV and III, and appear hard to distinguish. Normalising and standardising the data (Fig. 7) results in totally different directions for the three sets of compounds in the 2D plot of scores. This time, the characteristic masses for each set of related compounds fall in three distinctive groups, with the masses in green being approximately at right angles to both the blue and red groups. Fig. 9 3D plot for (a) the scores and (b) the loadings of PC2 versus PC1 for the normalised and standardised HPLC-DAD data (XY: 50°, YZ: 30°, XZ: 60°). Fig. 10 Plot of angle (in degrees) versus time between the three loadings corresponding to wavelengths characteristic of different compounds (from top to bottom: 692, 665 and 652 nm) and all scores (normalised and standardised HPLC-DAD data).Fig. 11 3D plot for (a) the scores and (b) the loadings of PC2 versus PC1 for the raw HPLC-MS data (XY: 30°, YZ: 80°, XZ: 115°). 976 Analyst, 1999, 124, 971–9793.3 3D PC plots 3.3.1 HPLC-DAD data. To enhance the visualisation of the data, 3D plots for both raw and pre-processed data were obtained. Good angles were selected graphically to best illustrate the main trends. Since the same two peaks (5 and 6) dominate the analysis, little improvement was observed in the scores and loadings plots of the raw data, which are not displayed for reasons of brevity.Fig. 8 shows a plot of angles (in degrees) versus time between values for the three loadings that correspond to wavelengths characteristic of the three sets of compounds (652 nm for III and IIIA, 665 nm for I, IA and II and 692 nm for IV) and the scores. The graph obtained from the 652 nm loadings suggests 3 peaks, numbered 1, 5 and 8 at the approximate true elution times of the compounds they correspond to; note the unexpected pheophytin peak.The graph for 665 nm displays lower angles for peaks 2, 3, 6 and 7, whereas the one for 692 nm displays its lowest angle for peak 4. Clearly, the lower the angle, the more the region of the chromatogram corresponds to a compound characterised by a specific wavelength. It is important to recognise that spectra are fairly similar for all compounds, and only very small shifts and differences in intensities characterise each group. 3D plots for the normalised and standardised data obtained using HPLC-DAD are illustrated in Fig. 9. The scores clearly identify peaks 1 and 5 as one cluster, peaks 6, 7, 2 and 3 as a second cluster and peak 4 on its own. Peak 8 has been omitted from the graph, as explained previously. This time, the clustering in three groups is more obvious than when using 2D plots. In the loadings, the characteristic wavelengths of absorbance of the compounds clearly fall in the same directions as in the scores plot.Obtaining a plot of angles versus time (Fig. 10) for this type of data displays the same clustering of compounds as above. This time, peak 8 is not associated with peaks 1 and 5, suggesting that the third PC helps discriminate this compound. The graphs for wavelengths 652 and 665 nm show low values of angles for a period of approximately 30 to 40 s after the maxima of peaks 5 and 6 respectively, indicating possible peak tailing.As the data has been pre-processed, the angles have much lower values. 3.3.2 HPLC-MS data. 3D plots were obtained for the HPLCMS data, both raw and pre-processed. Fig. 11 displays two such plots for the scores and loadings of raw data. In Fig. 11a, three clear directions are obtained, approximately at 90° to one another. In contrast to the 2D plot (Fig. 6a), peak 2 is now clearly differentiated from peaks 4 and 5.In the loadings plot, the masses for I, and IA are clustered together in the centre of the plot, whereas the ones for IV and III (blue) and II (green) are clearly distinguished from one another (unlike the 2D plots). A plot of angles (in degrees) versus mass number, between the three loadings corresponding to masses most characteristic of the different compounds (m/z 916 for I and IA as an {[M+H]++Na]} adduct, m/z 932 for II as an {[M+H]++Na]} ion Fig. 12 Plot of angle (in degrees) versus mass units between the three loadings corresponding to masses characteristic of different compounds (from top to bottom: 916, 932 and 940 mu) and all loadings of raw HPLC-MS data.Analyst, 1999, 124, 971–979 977and m/z 940 for III and IV as an [MH2 . ]+ ion)}) and loadings corresponding to each of the 70 chosen masses, is given in Fig. 12. The lower the angle, the more significant the individual mass for each set of compounds; note that for clarity, angles close to 0° are placed at the top of the graph.It is evident from these graphs that several extra fragment ions that may not be immediately obvious from the raw mass spectral data can be assigned. The advantage of the angular plots using 3 PC’s can be clearly seen when compared to the equivalent plot using 2 PC’s for the raw HPLC-MS data (Fig. 13). Whereas the angular plot for I and IA (top graph) appears to be very similar to the one using 3 PC’s, there is not much discrimination between the second and third plots, as an excessive number of significant ions are obtained for the compounds with characteristic masses 932 and 940.This is because the masses corresponding to these two sets of compounds lie in the same direction in the 2D plot for the loadings of the raw HPLC-MS data (Fig. 6b). Obtaining the same plot using 3 PC’s (Fig. 11b) results in two completely different directions for the masses characteristic of these compounds, which is depicted in the 3D angular plots, where only the most significant ions for each set of compounds are obtained.The benefit of the 3D angular plots was determined by comparing these plots with those obtained using 2 PC’s. Although there was clearly a slight improvement for the HPLCDAD data and the pre-processed HPLC-MS data, too, the 2D plots for this kind of data are not shown in this paper for reasons of brevity. Fig. 14 displays PC plots for the scores and loadings of normalised and standardised HPLC-MS data.In the scores plot, peaks 4 and 5 are clearly distinguished from one another without difficulty (in contrast to the equivalent 2D plot). Additionally, their corresponding masses may be distinguished, too, in two separate clusters. Although the loadings and scores plots appear very different for the raw and pre-processed data, there is less distinction between the angular plots providing 3 PCs are used. In fact the plots for the raw data are slightly more informative than the pre-processed plots.The reason is that an angle corresponds to a correlation coefficient and not a distance, which hardly changes. 4 Conclusions This paper illustrates the use of chemometric methods for exploring complex chromatograms. Many published methods are applied to relatively simple problems, but the power of these techniques arises from analysing difficult data. The case study employed in this paper is relatively well established from previous work and provides a good example of closely eluting compounds with similar spectroscopic and chromatographic characteristics. Graphical representation of chromatographic data is important, and many of the methods described above help build up a picture of the different components in the chromatogram.Data Fig. 13 Plot of angle (in degrees) versus mass units between the three loadings corresponding to masses characteristic of different compounds (from top to bottom: 916, 932 and 940 mu) and all loadings of raw HPLC-MS data (2 PCs only). 978 Analyst, 1999, 124, 971–979b a pre-processing, especially combined normalisation and standardisation, allows detection of compounds in small quantities and also low intensity absorption and mass numbers. These can provide valuable diagnostic information. The third dimension adds to the informativeness of scores and loadings plots in many cases. Rotations and projections must be carefully chosen. In certain situations it is also possible to reduce from 4 or 5 dimensions to 2 by including extra dimensions in the rotation.It is important to recognise that PCs are abstract solutions, and therefore, a two PC plot (PC2 versus PC1) will not necessarily provide as much information as a projected three PC plot: the process of projections and rotations may be thought of as a simple form of factor analysis. Angular plots relate the position of variables between loadings and scores plots.An alternative would be to superimpose such information using a bi-plot, but the graphs become very crowded and difficult to visualise. In the case of the mass spectra, small but significant m/z values can be revealed using this procedure, providing data preprocessing (standardisation and normalisation) is performed first, yielding valuable information. It is, of course, possible to represent this information in other forms, such as via correlation coefficients. It is recommended that graphical chemometric methods are routinely employed for the exploration of complex coupled chromatographic data, and that a variety of approaches are used in complement to each other.Acknowledgements We thank EPSRC and SmithKline Beecham for financial support for this project. References 1 Y. Z. Liang, O. Kvalheim, A. Rahmani and R. G. Brereton, J. Chemom., 1993, 7, 15. 2 A. Elbergali and R. G. Brereton, Chemom. Intell. Lab. Syst., 1994, 23, 97. 3 K. Kavianpour and R. G. Brereton, Analyst, 1998, 123, 2035. 4 S. Dunkerley, J. Crosby, R. G. Brereton, K. D. Zissis and R. E. A. Escott, Analyst, 1998, 123, 2021. 5 S. Dunkerley, R. G. Brereton and J. Crosby, Chemom. Intell. Lab. Systems, 1999, 48, 99. 6 M. Kubista, Chemom. Intell. Lab. Syst., 1990, 7, 273. 7 C. Demir, P. Hindmarch and R. G. Brereton, Analyst, 1996, 121, 1443. 8 W. A. Svec, in Chlorophyll, ed. H. Scheer, CRC Press, Boca Raton, FL, 1991, p. 89. 9 K. Iriyama, N. Ogura and A. Takamiya, J. Biochem., 1974, 76, 901. 10 R. G. Brereton, A. Rahmani, Y. Z. Liang and O. M. Kvalheim, Photochem. Photobiol., 1994, 59, 99. 11 K. D. Zissis, S. Dunkerley, R. G. Brereton, K. Kavianpour and J. Crosby, Rapid Commun. Mass Spectrom., in press. 12 I. T. Jolliffe, Principal Component Analysis, Springer, New York, 1986. 13 S. Wold, K. Esbensen and P. Geladi, Chemom. Intell. Lab. Syst., 1987, 2, 37. 14 E. Malinowski, Factor Analysis in Chemistry, Wiley, New York, 2nd edn., 1991. 15 K. V. Mardia, J. T. Kent and J. Bibly, Multivariate Analysis, Academic Press, London, 1979. 16 Multivariate Pattern Recognition in Chemometrics Illustrated with Case Studies, ed. R. G. Brereton, Elsevier, Amsterdam, 1992. 17 S. Wold and E. Lyttkens, Bull. Int. Stat. Inst. Proc., 1969, 37, 1. 18 C. Demir and R. G. Brereton, Analyst, 1997, 122, 631. Paper 9/01996K Fig. 14 3D plot for (a) the scores and (b) the loadings of PC2 versus PC1 for the normalised and standardised HPLC-MS data (XY: 40°, YZ: 10°, XZ: 50°). Analyst, 1999, 124, 971–979 979

ISSN:0003-2654    

DOI:10.1039/a901996k

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Measurement uncertainty: Approaches to the evaluation of uncertainties associated with recovery† Vicki J. Barwick and Stephen L. R. Ellison Laboratory of the Government Chemist, Queens Road, Teddington, Middlesex, UK TW11 0LY Received 9th March 1999, Accepted 14th May 1999 A number of approaches for evaluating recovery and its contribution to uncertainty budgets for analytical methods are considered in detail. The recovery, R, for a particular sample is considered as comprising three elements, �R m, Rs and Rrep.These relate to the recovery for the method; the effect of sample matrix and/or analyte concentration on recovery; and how well the behaviour of spiked samples represents that of test samples. The uncertainty associated with R, u(R), will have contributions from u(�R m), u(Rs) and u(Rrep). The evaluation of these components depends on the method scope and the availability, or otherwise, of representative certified reference materials. Procedures for evaluating these parameters are considered and illustrated with worked examples.Techniques discussed include the use of certified reference materials and spiking studies, and the use of extraction profiling to predict recoveries. All the approaches discussed evaluate the recovery and its uncertainty for the analytical method as a whole. It is concluded that this is a useful approach as it reduces the amount of experimental work required. In addition, most of the required data are frequently available from method validation studies.Introduction In recent years, the subject of the evaluation of measurement uncertainty in analytical chemistry has generated a significant level of interest and discussion.1–6 It is generally acknowledged that the fitness for purpose of an analytical method cannot be assessed without some estimate of the measurement uncertainty to compare with the confidence required. Well characterised and controlled uncertainties are also fundamental to the implementation of traceability as a means of ensuring comparability of results; large uncertainties imply poor comparability.The Guide to the Expression of Uncertainty in Measurement (GUM) published by ISO7 establishes general rules for evaluating and expressing uncertainty for a wide range of measurements. The guide was interpreted for analytical chemistry by Eurachem in 1995.8 The approach described in the GUM requires the identification of all possible sources of uncertainty associated with the procedure; the estimation of their magnitude from either experimental or published data; and the combination of these individual uncertainties to give standard and expanded uncertainties for the procedure as a whole.Some applications of this approach to analytical chemistry have been published.9,10 However, the GUM principles are significantly different from the methods currently used in analytical chemistry for estimating uncertainty11–13 which generally make use of ‘whole method’ performance parameters, such as precision and recovery, obtained during in-house method validation studies or during method development and collaborative study.14–16 In earlier papers we have illustrated the use of precision and recovery data in uncertainty estimates for a range of analytical techniques.17–19 Though this establishes the principle of applying validation data to uncertainty estimation, the approach relies on estimation of uncertainties associated with recovery, including those associated with matrix change or incomplete extraction.Though recovery itself is routinely estimated during method validation, there is no general approach to the estimation of the uncertainty associated with recovery. This paper accordingly describes and illustrates some approaches to the evaluation of recovery and its uncertainty. Theoretical basis General approach In this paper, recovery R is defined as the ratio cobs/cref of observed concentration cobs to a reference value cref for the particular material tested.If known, R could be used to correct an observation to an appropriate reference scale. Were such a correction made, it is clear that any uncertainty in R will contribute to uncertainties in the declared result. R is, however, not usually obtained or considered obtainable for test samples. It is instead estimated indirectly, for example by experiments on related reference materials with a certified concentration, by comparison with an alternative definitive method, or by observing the amount of an added spike recovered from a sample matrix.In practice, measures are usually taken to ensure that the recovery is likely to be reasonably close to unity, and the assumption then made that R = 1. The main uncertainties associated with recovery arise from this assumption. To quantify the uncertainty, it is necessary to consider the degree to which a particular sample matrix under test is represented by the reference material employed and, where relevant, the extent to which spiking provides a representation of native analyte behaviour.To treat these uncertainties explicitly, it is useful to consider the recovery R for a particular sample as comprising three components: (i) �R M (so denoted because it is usually estimated as a mean of several determinations) is an estimate of the recovery obtained from, for example, the analysis of a CRM or a spiked sample.�R m may be considered as a ‘reference’ recovery, or more generally a ‘method recovery’ since it would normally be expected to apply to all determinations using the method, at least in a particular laboratory. The uncertainty in �R m is composed of the uncertainty in the nominal value (e.g., the uncertainty in the certified value of a reference material) and the † © Copyright LGC (Teddington) Ltd. 1999. Analyst, 1999, 124, 981–990 981uncertainty in the observed value (e.g., the standard deviation of the mean of replicate analyses).(ii) Rs is a correction factor to take account of differences in the recovery for a particular sample compared to the recovery observed for the material used to estimate �R m. (iii) Rrep is a correction factor to take account of the fact that a spiked sample may behave differently to a real sample with incurred analyte. These three elements are combined multiplicatively to give an estimate of the recovery for a particular sample, i.e, R = �R m 3 Rs 3 Rrep.It therefore follows that the uncertainty in R, u(R), will have contributions from and u(�R m), u(Rs) and u(Rrep). How each of these components and their uncertainties are evaluated will depend on the method scope and the availability of reference materials. In the simplest case the method scope covers a single matrix type and analyte concentration for which a representative CRM is available.However, the situation is often more complex than this. The method scope may cover a range of matrices and/or analyte concentrations and there may be no suitable CRM available. The approaches we suggest for estimating recovery and the associated uncertainty for a range of situations are summarised in Table 1. The remainder of the paper first discusses the different strategies and presents the relevant calculations, then presents experimental illustrations of selected approaches.Estimating �R m and u(�R m) using a representative CRM If a representative reference material is available, �R m is estimated by comparing the mean of replicate analyses of the CRM with the certified value: R C C m obs CRM = (1) where �C obs is the mean of the results from the replicate analysis of the CRM and CCRM is the certified value for the CRM. The uncertainty associated with the estimate of, �R m, u(�R m), is estimated by: Table 1 Summary of suggested methods for evaluating recovery Recovery component Method scope/availability of CRMs �R m Rs Rrep Single matrix and analyte concentration.Representative CRM available. Determine �R m and u(�R m) from replicate analysis of the CRM. Not applicablea Not applicableb Single matrix and analyte concentration. No representative CRM available. Determine �R m by: Analysis of a representative matrix spiked atComparison of result obtained for a typical sample with results obtained from a standard procedure; or Changing the extraction system (e.g., using a stronger solvent) to see if any more of the analyte can be recovered; or Analysing a ‘worst case’ CRM.If a CRM is available which has a matrix that is known to be more difficult to extract the analyte from compared to the sample, it can be assumed that the recovery for the sample will be no worse than the recovery observed for the CRM. Not applicablea Evaluate how representative the spike is of the native material.Possible approaches include: Monitoring the extraction of spiked and native analytes with time; Comparing spiked recovery with the recovery from a non representative CRM. Multiple matrices and/or analyte concentrations. One representative CRM available. Determine �R m and u�R m from replicate analysis of the CRM. Determine the recovery for a range of representative sample matrices spiked at representative concentrations. u(Rs) is calculated from the spread of the recovery estimates.Not applicableb Multiple matrices and/or analyte concentrations. No representative CRM available. Determine �R m and u(�R m) from analysis of representative sample matrices spiked at representative concentrations. Estimate u(Rs) from the data obtained in the calculation of �R m. Evaluate how representative the spike is of the native material. Possible approaches include: Monitoring the extraction of spiked and native analytes with time; Comparing spiked recovery with the recovery from a non representative CRM.a As the method scope covers only a single matrix type and analyte concentration the estimate of �R m and its uncertainty can be based on the analysis of a sample which is truly representative of real samples. There is therefore no need to include a correction factor to take account of differences in recovery for a particular sample, compared to the sample used in the estimation of �R m (i.e., Rs is assumed to equal 1 with negligible uncertainty).b If �R m is estimated from the analysis of a representative certified reference material it can be assumed that this will behave in a similar manner to incurred analyte in a real sample. A correction factor to take account of the fact that the recovery of the analyte from the material used to estimate �R m may not be representative of the recovery from a real sample is therefore unnecessary (i.e., Rrep is implicitly assumed to equal 1 with negligible uncertainty). 982 Analyst, 1999, 124, 981–990u R Rsn Cu CC( )( )m mobs2obs2CRMCRM= ¢D¢D ÆÙ + ÆÙ 2(2)where sobs is the standard deviation of the results from thereplicate analyses of the CRM, n is the number of replicates andu(CCRM) is the standard uncertainty in the certified value for theCRM. The contribution of Rm and its uncertainty to thecombined uncertainty for the method depends on whether therecovery is significantly different from 1, and if so, whether ornot a correction is made.This is discussed in detail later. If thereference material is only approximately representative of atypical sample, additional sources of uncertainty may need to beconsidered. These include effects of matrix or interferences intest samples which may differ from those in the CRM.Estimating Rm and u(Rm) from spiking studiesIn the absence of a suitable reference material, recovery isfrequently estimated through spiking studies, i.e., the additionof the analyte to a previously studied material.The spikedsample is prepared in such a way as to represent as closely aspossible a natural sample with incurred analyte. A number ofoptions are available. In the simplest case, a bulk sample of asuitable sample matrix known to be free from the analyte ofinterest is spiked with an appropriate concentration of theanalyte. The bulk spiked sample is then analysed in replicate.Rm is given by:RCC mobsspike= (3)where Cobs is the mean of the replicate analyses of the spikedsample and Cspike is the nominal concentration of analyte in thespiked sample.The uncertainty is estimated by:u R Rsn Cu CC( )( )m mobs2obs2spikespike= ¢D¢D ÆÙ + ÆÙ 2(4)where sobs is the standard deviation of the results from thereplicate analyses of the spiked sample, n is the number ofreplicates and u(Cspike) is the standard uncertainty in theconcentration of the spiked sample.If no blank sample matrix is available, a bulk spiked samplecan be prepared from a matrix which contains the analyte.Thespiked sample is then analysed in replicate. Rm is given by:RC CC mobs nativespike= - ÆÙ(5)where ÆÙCnative is the observed concentration of the analyte in theunspiked sample. Note that since we are concerned only withthe difference between the spiked and unspiked observations,ÆÙCnative does not have to represent the ¡¥true¡¦ value of theconcentration of the analyte in the unspiked matrix; eqn. (5)represents the change in observation divided by the change inconcentration.The uncertainty is estimated by:u R Rs n sC Cu CC( )/( ÆÙ )( )m mobs2native2obs nativespikespike= ¢D +-+ ÆÙ 22(6)where snative is the standard deviation of the mean of the resultsof repeat analyses of the unspiked matrix.If it is impractical to prepare a homogeneous bulk spikedsample for sub-sampling, then individual spiked samples can beprepared. If the spiked samples are prepared from approximatelythe same weight of a blank sample matrix, and the sameweight of the spike is added to each sample, the recovery isgiven by:Rmm mobsspike= (7)where mobs is the mean weight of the spike recovered from thesamples and mspike is the weight of the spike added to eachsample. u(Rm) is therefore estimated by:u R Rsn mu mm( )( )m mmobs2spikespikeobs = ¢D¢D+ ÆÙ 2 2(8)where smobs is the standard deviation of the results obtained fromthe spiked samples, n is the number of spiked samples analysedand u(mspike) is the uncertainty in the amount of spike added toeach sample.If the spiked samples are prepared from a sample matrixwhich contains the analyte the situation is somewhat morecomplex.The recovery for each sample, Rm(i), is given by:RC CC iiim(obs nativespike)( )( )ÆÙ=-(9)where Cobs(i) is the concentration of the analyte observed forsample i, ÆÙCnative the observed response for the unspiked sampleas before [eqn. (5)], and Cspike(i) is the concentration of the spikeadded to sample i.The mean recovery, Rm, is given by:RnC CCiiinmobs nativespike=-= 11( )( )ÆÙ(10)Therefore:RnCCCCiiiniinmobsspike+( )nativespike( )= -¢X£»£»= = 1 11 1( ) ÆÙ (11)The uncertainty is calculated using the expression of theform:7u y p qypu pyqu q [ ( , , )] [ ( )] [ ( )] = ÆÙ + ÆÙ +2222 (12)Differentiating eqn. (11) gives:= ¢D = = ¢D-= RC n CRC n CRCnC CCi i iiniiimobs spikemnative spikemspikeobs nativespike( ) ( ) ( ) ( )( )( )ÆÙ( ÆÙ )1 1 1 1112u Rn Cu Cn Cu CnC CCu Ciiiniiniiiin( ) ( )(ÆÙ )( ÆÙ )( )( )( )( )( )( )( )mspikeobsspikenativeobs nativespikespike221122211 11 11= ¢D ¢D ¢X £»£»+¢X£»£»£»¢D+ ¢D-¢D ¢X £»£»=== 2(13)Under certain experimental conditions, eqn.(13) can besimplified. Firstly, if u(Cspike(i)) < < u(Cobs(i)) and u(Cnative) theexpression becomes:u Rnu CC Cu C iiiniin( )( )(ÆÙ ) ( )( ) ( )mobsspike spikenative = ÆÙ +ÆÙ= = 1 121 122(14)Analyst, 1999, 124, 981¡V990 983This is often the case, as spiking is generally achieved by adding an aliquot of a solution or a known weight of the analyte.The uncertainties associated with such operations are usually small compared to the uncertainties associated with the observation of the amount of the analyte in a sample [i.e., u(Cobs(i)) and u( �C native)]). Furthermore, if the standard deviation of the Cspike(i) values is small compared to the meanlues, �C spike can be used in the calculation. Eqn. (14) therefore simplifies further to: u R n C u C n u C i i n ( ) ( ) (� ) ( ) m spike obs native = ¥ + ¥ =  1 1 2 2 2 1 (15) This is likely to be the case in recovery studies at a single level using similar quantities of the sample matrix in the preparation of each spiked sample.Finally, if the estimates of u(Cobs(i)) are all similar, the mean can be used. This leads to: u R C u C n u C i ( ) ( ) (� ) ( ) m spike obs native = + 1 2 2 (16) Again, this is likely to be the case when each sample is spiked at the same concentration so that all the Cobs(i) values are of similar magnitude.Estimating �R m and u(�R m) by comparison with a standard method An alternative approach to estimating �R m is by comparison with the results obtained from a standard method of known uncertainty. A representative sample is analysed, in replicate, using both the method under evaluation and the standard method. �R m is given by: R C C m method standard = (17) where �C method is the mean of the results obtained using the method under consideration and �C standard is the mean of the results obtained using the standard method.The uncertainty in the recovery, u(�R m), is therefore estimated by: u R R s n C u C C ( ) ( ) m m method 2 method 2 standard standard = ¥ ¥ Ê Ë Á � � � + Ê Ë Á � � � 2 (18) where smethod is the standard deviation of the results obtained using the method, n is the number of replicates and u(Cstandard) is the standard uncertainty associated with the standard method.Alternative approaches to estimating �R m and u(�R m) In the absence of appropriate CRMs or standard methods, and if preparing spiked samples is impractical, alternative methods of investigating the recovery are required. However, such techniques generally require an element of judgement on the part of the analyst and can often only be used as an initial indication of the uncertainty associated with method recovery. If the results of such a study indicate that the uncertainties associated with recovery are a significant contribution to the uncertainty budget, further investigation will be required to obtain a better estimate.The main techniques available include repeated extraction experiments, monitoring the progress of extraction with time, and analysis of ‘worst case’ materials. These approaches are discussed in turn below. Repeated extraction. Samples are re-extracted either under the same experimental conditions, or preferably with a more vigorous extraction system (e.g., a more polar extraction solvent).The amount of analyte extracted under the normal application of the method is compared with the total amount extracted (amount extracted initially plus the amount extracted by subsequent re-extractions). �R m is the ratio of these estimates. If re-extraction was achieved using the same conditions as the initial extraction, the difference between the true recovery and the assumed value of 1 is known to be at least 1 2 �R m. The difference could be greater, as repeated extractions under the same experimental conditions may not quantitatively recover all of the analyte from the sample. In such cases we estimate the uncertainty, u(�R m), associated with the assumed value of �R m = 1 (i.e., perfect recovery) as (1 2 �R m).If repeat extractions were carried out using a more vigorous extraction system, there is greater confidence associated with the observed difference between �R m and the assumed value of 1.This is because it is more likely that the repeat extractions will have quantitatively extracted the remainder of the analyte from the sample, thus giving greater confidence in the estimate of �R m. In such cases we estimate u(�R m) as, (1 2 �R m)/k where k is the coverage factor which will be used to calculate the expanded uncertainty. Monitoring extraction with time. For some methods, it may be possible to build up an extraction profile for the method and use it to predict how close the extraction is to completion.A procedure for doing this in supercritical fluid extraction (SFE) has been described by Bartle et al.20 The prediction relies on the extraction profile following an approximately exponential form after an initial rapid extraction. If the extraction is carried out for at least as long as the initial non-exponential period to obtain a mass of extracted analyte, m1, followed by extraction over two subsequent equal time periods to obtain masses of analyte m2 and m3, then m0, the total mass of the analyte in the sample, is given by: m m m m m 0 1 2 2 2 3 = + - (19) To estimate �R m the mass extracted during the normal application of the method is compared with the predicted total mass m0.The uncertainty associated with �R m will have contributions from the uncertainty associated with the observed mass (standard deviation of the mean of n observations) and the uncertainty associated with the prediction of m0.Analysis of a worst case CRM. If a CRM is available which has a matrix known to provide an extreme example (i.e., more difficult to extract the analyte from than test samples), the recovery observed for the CRM can provide a worst case estimate on which to base the recovery for real samples. The recovery observed from replicate analyses of the CRM is denoted RCRM. Since the CRM matrix is known to be more difficult to extract the analyte from, it is reasonable to assume that recoveries for test samples are more likely to be closer to 1 than to RCRM.It is therefore appropriate to consider RCRM as representing the lower limit of a triangular distribution.7 As a first estimate, �R m is assumed to equal 1, with an uncertainty, u(�R m), of: u R R ( ) m CRM = - 1 6 (20) Note that if there is no evidence to suggest where in the range 1 2 RCRM the recovery for test samples is likely to lie, a rectangular distribution should be assumed.u(�R m) is then estimated by (1 2 �R m)/A3. 984 Analyst, 1999, 124, 981–990Evaluating the contribution of �R m to u(R) when significance tests are used Recovery is often tested for evidence of significant difference from 1.0 (100%). In such circumstances, the contribution of �R m and its uncertainty to the overall uncertainty for the method will depend on whether it is found to be significantly different from 1, and if so, whether or not a correction is made.General rules for calculating uncertainty estimates for these different circumstances have been discussed in detail elsewhere.21 Here, we summarise them for an estimate of �R m and its uncertainty, assuming that significance is checked by comparison of a statistic t = |�R m 2 1|/u(�R m)A with a critical value tcrit. Identical principles hold for estimation of uncertainties associated with any other recovery component subjected to a significance test. Three cases arise: 1.�R m, taking into account u(�R m)A, is not significantly different from 1 so no correction to the final result is applied. Eqn. (21) applies:† u R t uR ( ) ( ) . m crit m = ¥ ¢ 1 96 (21) where u(�R m) is the required standard uncertainty associated with the estimate (1.0) of �R m. 2. �R m, taking into account u(�R m)A, is significantly different from 1 and a correction to the final result is applied. u(�R m) is again given by eqn. (21). 3. �R m, taking into account u(�R m), is significantly different from 1 but a correction to the final result is not applied.The standard uncertainty is increased to ensure that the range quoted will include the true value using eqn. (22): u R R k u R ( ) ( ) m m m = - Ê Ë Á � � � + ¢ 1 2 2 (22) where k is the coverage factor that will be used in the calculation of the expanded uncertainty. A fourth special case applies to empirical methods. An empirical method is a standardised method agreed upon for the purposes of comparative measurement within a particular field of application; the measurand is accordingly defined by the method.In such cases the recovery is arbitrarily defined as unity and the uncertainty associateerally need verification, and where a reference material is used to check the local performance, the above considerations will apply. Estimating Rs and u(Rs) from spiking studies Where the method scope covers a range of sample matrices and/ or analyte concentrations, an additional uncertainty term is required to take account of differences in the recovery of a particular sample type, compared to the material used to estimate �R m.This can be evaluated by analysing a representative range of spiked samples, covering typical matrices and analyte concentrations, in replicate. The mean recovery for each sample type is calculated. Rs is normally implicitly assumed to be equal to 1.However, there will be an uncertainty associated with this assumption, which appears in the spread of mean recoveries observed for the different spiked samples (strictly, from the between-matrix component of variance, but in practice, the dispersion of mean values usually provides a reasonable estimate). The uncertainty, u(Rs), is therefore the standard deviation of the mean recoveries for each sample type. Where Rs differs significantly from 1.0, an additional allowance should be made as in eqn.(22). Estimating Rrep and u(Rrep) Rrep is generally assumed to equal one, indicating that the recovery from a spiked sample perfectly represents the recovery observed for incurred analyte. The uncertainty u(Rrep) is a measure of the uncertainty associated with that assumption, i.e., how different Rrep might be from the assumed value of 1. The complexity of evaluating how well a spike represents the behaviour of native material varies from matrix to matrix and with the method being studied.In some cases it can be argued that a spike is a good representation of a real sample, for example in liquid samples where the analyte is simply dissolved in the matrix. In addition, if the method involves total dissolution or destruction of the matrix, for example by ashing, there may be no reason to believe that a spike would behave any differently from the incurred analyte. However, problems arise for more complex matrices and where the method involves extraction rather than total destruction or dissolution.Possible approaches to investigating the performance of spiked versus real samples include monitoring the extraction of spiked and native analytes with time, and comparison of spiked recovery with the recovery from a less representative CRM. However, these may not be appropriate in all cases. If the analyst cannot obtain any experimental evidence on the appropriateness of spiking, then judgements and/or assumptions have to be made. Ideally, Rrep should be evaluated by the analysis of a reference material (even if it is not directly comparable to the test samples) and by comparing the recovery obtained with those observed from the spiking studies.The uncertainty u(Rrep) is then estimated as: u R R k u R ( ) ( ( )) rep rep rep = - Ê Ë Á � � � + ¢ 1 2 2 (23) where k is the coverage factor which will be used to calculate the expanded uncertainty and u(Rrep)A is the uncertainty associated with the estimate of Rrep.The most straightforward approach is to spike the CRM and compare the recovery observed with that observed from the analysis of the unspiked reference material. In such cases Rrep is given by: R C C C C C rep obs(spike) obs(CRM) spike CRM obs(CRM) = - ¥ (24) where �C obs(spike) is the mean concentration observed from replicate analyses of the spiked CRM, �C obs(CRM) is the mean concentration observed from replicate analyses of the unspiked CRM, Cspike is the concentration of the spike added and CCRM is the certified concentration of the reference material.The uncertainty, u(Rrep)A, is obtained by differentiating eqn. (24) and applying eqn. (12): u R R u C C C u C u C C C C u C C u C C ( ) ( ) ( ) ( ) ( ) ( ) ( ) rep rep obs(spike) obs(spike) obs(CRM) obs(CRM) obs(spike) obs(CRM) obs(CRM) obs(spike) CRM CRM spike spike ¢= ¥ - Ê Ë Á � � � + ¥ ¥ - Ê Ë Á � � � + Ê Ë Á � � � + Ê Ë Á � � � 2 2 2 2 (25) † Note that the standard uncertainty is actually given by u(�R m).The multiplication by tcrit/1.96 effectively increases the estimate slightly to allow for small numbers of degrees of freedom, which would otherwise need to be considered in forming the combined expanded uncertainty.7 Analyst, 1999, 124, 981–990 985In this case, we are only interested in the dispersion of results obtained for the mean values �C obs(spike) and �C obs(CRM). The corresponding uncertainties are therefore estimated by the standard deviation of the mean of the observed concentrations in each case.Note that the above equation holds if the spiking study was based on the replicate analysis of a single spiked sample of the CRM. If the study was based on the analysis of a number of individual portions of the CRM (of similar weight), all spiked at a similar concentration, eqns. (24) and (25) are modified slightly: R C C C C C rep obs(spike) obs(CRM) spike CRM obs(CRM) = - ¥ (26) u R R u C C C u C u C C C C u C C u C C ( ) ( ) ( ) ( ) ( ) ( ) ( ) rep rep obs(spike) obs(spike) obs(CRM) obs(CRM) obs(spike) obs(CRM) obs(CRM) obs(spike) CRM CRM spike spike ¢= ¥ - Ê Ë Á � � � + ¥ ¥ - Ê Ë Á � � � + Ê Ë Á � � � + Ê Ë Á � � � 2 2 2 2 (27) where �u (Cobs(spike)) is the average of the uncertainties associated with each of the Cobs(spike) values divided by the square root of the number of determinations of Cobs(spike), �C spike is the average of the concentrations of the spike added to each sample and �u (Cspike) is the average of the uncertainties associated with each of the Cspike values.This approach is illustrated in the Experimental section. If there is no CRM available then the analyst will have to make a judgement based on the information available. Calculating R and u(R) The recovery for a particular sample, R, is given by R = �R m 3 Rs 3 Rrep. However, since Rs and Rrep are generally assumed to equal 1, R = �R m.The value of �R m used depends on whether or not it is significantly different from 1, and if so, whether a correction to the result for a particular sample is applied. The uncertainty associated with R, u(R) is estimated by: u R R u R R u R R u R R ( ) ( ) ( ) ( ) = ¥ Ê Ë Á � � � + Ê Ë Á � � � + Ê Ë Á � � � m m s s rep rep 2 2 2 (28) However, if Rs = Rrep = 1, eqn. (28) simplifies to: u R R u R R u R u R ( ) ( ( ) ( ) = ¥ Ê Ë Á � � � + + m m m s rep 2 2 2 (29) Experimental Polycyclic aromatic hydrocarbon (PAH) extraction studies The study was based on the analysis of a coal carbonisation site soil reference material (RM), LGC RM 6138 (Office of Reference Materials, LGC (Teddington), Middlesex, UK).The 16 analytes are listed in Table 2, together with the relevant reference values and their associated uncertainties. 20 portions of the material were analysed using the method outlined below, nine of which were spiked with a solution of the 16 target PAHs. The spiking solution had a nominal concentration of 200 mg m21 and was prepared from a 2000 mg ml21 stock solution (Supelco, Bellefonte, PA, USA).The actual (stated) concentrations of each of the analytes in the spiking solution are presented in Table 2. 0.5 ml of the spike was added to approximately 10 g reference material before the addition of the deuterated surrogate recovery standards. The spiked samples were otherwise treated in exactly the same way as the unspiked samples.The method studied is used for the determination of PAHs in soils samples.22 In normal use, air dried soil samples are soxhlet extracted with dichloromethane for six hours. Prior to extraction, 0.5 ml of a solution of deuterated PAHs (naphthalene-d8, acenaphthene-d10, phenanthrene-d10, chrysene-d12, perylened12) with a nominal concentration of 80 mg l21, are added as surrogate recovery standardard, 4-terphenyl-d14, is added to the sample and standard solutions prior to analysis by GC-MS. Calibration Table 2 Reference and spiked analyte concentrations in RM LGC 6138—Coal Carbonisation Site Soil l CRM certified values Spiking levels Analyte Reference value, CCRM mg kg21a Uncertainty/ mg kg21 Stock solution concentration/ mg ml21c Standard deviation/ mg ml21c Spiking solution concentration/ mg ml21 Amount of analyte added to sample/mg Naphthalene 32 4.0 2054 9.4 205 0.103 Acenaphthylene 7.0 1.4 1998 1.2 200 0.100 Acenaphthene 6.4 0.8 1992 2.4 199 0.100 Fluorene 15.3 1.4 1983 10.2 198 0.099 Phenanthrene 114 7.0 1974 18.9 197 0.099 Anthracene 22 3.0 1985 7.1 199 0.099 Fluoranthene 118 8.0 1990 9.2 199 0.100 Pyrene 103 6.0 1991 1.5 199 0.100 Benzo[a]anthracene 42 4.0 1987 7.5 199 0.099 Chrysene 44 5.0 1983 11.8 198 0.099 Benzo[b]fluoranthene 42 6.0 1987 8.1 199 0.099 Benzo[k]fluoranthene 21 5.0 1991 5.4 199 0.100 Benzo[a]pyrene 36 5.0 1989 0.8 199 0.099 Indeno[1,2,3-cd]pyrene 25 2.0 1992 14 199 0.100 Dibenz[a,h]anthracene 7.6 4.0 1998 3.8 200 0.100 Benzo[g,h,i]perylene 28 4.0 1993 8.4 199 0.100 a RM 6138.Robust mean value (median) of the results, on a dry sample weight basis. b RM 6138. The uncertainty quoted is the half width of the 95% confidence interval based on the robust standard deviation of the results. cValues quoted by supplier (Supelco). 986 Analyst, 1999, 124, 981–990curves are obtained for each PAH from standards prepared from a stock solution with a nominal concentration of 2000 mg ml21.Results and discussion Investigation of spiked versus native recoveries The aim of this study was to investigate how representative spiked recoveries are of extraction of the native analyte and to determine the uncertainty associated with estimating recovery through spiking. Results. The values of �R m calculated from replicate analysis of the RM using eqn. (1) are presented in Table 3, together with the corresponding estimates of �R m obtained from the analysis of the spiked samples calculated using eqn.(10). In the latter case, Cnative was taken as the mean concentration observed during the analysis of the unspiked reference material. t-tests23 were performed to compare the two estimates of recovery obtained for each analyte; the results are also included in Table 3. The results for acenaphthylene, acenaphthene, fluorene, anthracene, benzo[b]fluoranthene, indeno[1.2.3-cd]pyrene and dibenz[a,h]- anthracene indicated a significant difference (at the 95% confidence level) between the recovery estimates obtained from the analysis of the RM and those obtained from the analysis of spike samples.In the case of acenaphthylene, acenaphthene and fluorene the difference is due to the unusually high (in the case of acenaphthylene) or low estimates of recovery obtained from the analysis of the RM.Using the approach discussed previously, Rrep and u(Rrep) were calculated for each of the analytes by applying eqns. (26) and (27) respectively. The relevant data, and the resulting estimates of Rrep and u(Rrep) are presented in Table 4. The estimates of u(Cobs(spike)) and u(Cobs(CRM)) were based on the observed relative standard deviations obtained from the replicate analysis of RM 6138 (see Table 3). The values for u(Cspike) were calculated using the data given in Table 2, as described in the following section.Note that the values of u(Rrep) form a Table 3 Estimates of recovery obtained from the replicate analysis of RM 6138 Native recoverya Spike recoveryb t-test �C obs/ sobs/ result Analyte mg kg21 mg kg21 �R m �R m s(Rm) p,(v)c Naphthalene 29.5 2.1 0.920 0.876 0.13 0.33 (18) Acenaphthylene 12.3 1.7 1.915 0.863 0.063 8.3 3 1028 (11) Acenaphthene 3.30 0.25 0.471 0.825 0.043 8.3 3 10214 (18) Fluorene 8.86 0.63 0.579 0.837 0.048 1.5 3 10210 (18) Phenanthrene 101 5.7 0.884 1.041 0.29 0.15 (8) Anthracene 23.6 2.0 1.070 0.969 0.099 0.027 (18) Fluoranthene 108 7.0 0.920 0.959 0.38 0.77 (8) Pyrene 89.3 6.2 0.867 0.955 0.29 0.40 (9) Benzo[a]anthracene 37.8 2.9 0.899 0.884 0.15 0.79 (11) Chrysene 40.9 3.1 0.929 0.878 0.15 0.38 (11) Benzo[b]fluoranthene 30.9 2.6 0.737 0.824 0.11 0.043 (18) Benzo[k]fluoranthene 15.9 1.0 0.758 0.765 0.19 0.91 (9) Benzo[a]pyrene 30.4 2.7 0.845 0.825 0.13 0.68 (18) Indeno[1.2.3-cd]pyrene 25.2 2.3 0.972 0.688 0.19 8.9 3 1026 (18) Dibenz[a,h]anthracene 5.63 0.41 0.740 0.644 0.057 0.0011 (18) Benzo[g,h,i]perylene 24.0 2.6 0.856 0.818 0.10 0.38 (18) a Using eqn.(1), with 11 determinations for all analytes. b Estimates of recovery obtained from the analysis of samples of RM 6138 spiked with approximately 10 mg kg21 PAHs , calculated using eqn. (10) with 9 determinations for all analytes. c p-value for 2-tailed t-test (Null hypothesis: equal recoveries; alternative hypothesis: unequal recoveries).Where an F-test showed significant variance difference, an unequal variance t-test was applied. The figure in parentheses is the number of freedom (equal variance test) or effective degrees of freedom (unequal variance test) for the test statistic. Table 4 Estimates of Rrep and u(Rrep) obtained from the comparison of the recovery of spiked and native PAHsa �C obs(spike) �C obs(CRM) �C spike �u (Cobs(spike)) �u (Cobs(CRM)) �u (Cspike) Analyte (mg kg21) (mg kg21) (mg kg21) Rrep (mg kg21) (mg kg21) (mg kg21) u(Rrep)A |1 2 Rrep|/k u(Rrep) Naphthalene 38.39 29.46 10.20 0.952 0.896 0.630 0.0877 0.143 0.0241 0.145 Acenaphthylene 20.82 12.26 9.92 0.450 0.972 0.526 0.0695 0.0753 0.275 0.285 Acenaphthene 11.46 3.30 9.89 1.75 0.306 0.076 0.0703 0.199 0.375 0.424 Fluorene 17.10 8.86 9.85 1.45 0.399 0.190 0.0857 0.117 0.222 0.251 Phenanthrene 110.97 100.78 9.81 1.18 2.219 1.170 0.117 0.338 0.0882 0.349 Anthracene 33.10 23.55 9.86 0.906 0.883 0.598 0.0779 0.132 0.0478 0.140 Fluoranthene 117.98 108.51 9.88 1.04 2.360 2.100 0.0830 0.363 0.0208 0.364 Pyrene 98.77 89.34 9.89 1.10 2.305 1.866 0.0692 0.362 0.0494 0.365 Benzo[a]anthracene 46.49 37.77 9.87 0.983 1.240 0.873 0.0790 0.191 0.00889 0.191 Chrysene 49.50 40.87 9.85 0.945 1.155 0.921 0.0906 0.184 0.0279 0.186 Benzo[b]fluoranthene 39.06 30.94 9.87 1.12 1.042 0.792 0.0799 0.215 0.0590 0.223 Benzo[k]fluoranthene 23.47 15.91 9.89 1.01 0.548 0.315 0.0742 0.156 0.00439 0.156 Benzo[a]pyrene 38.58 30.43 9.88 0.976 1.157 0.828 0.0692 0.200 0.0120 0.200 Indeno[1.2.3-cd]pyrene 32.07 25.27 9.89 0.708 0.962 0.701 0.0980 0.147 0.146 0.208 Dibenz[a,h]anthracene 12.02 5.63 9.92 0.870 0.280 0.122 0.0724 0.128 0.0648 0.144 Benzo[g,h,i]perylene 32.06 23.96 9.90 0.956 1.069 0.772 0.0812 0.189 0.0225 0.190 a The study involved the analysis of 9 individual spiked samples which gave results Cobs(spike(i)) each with a standard deviation sobs(spike(i)).�C obs(spike) is the mean of the Cobs(spike(i)) values and �u (Cobs(spike)) is the mean of the sobs(spike(i)) values divided by A9. �C obs(CRM) is the mean of the results of 11 analyses of the RM and u(�C obs(CRM) is the standard deviation of the mean. Analyst, 1999, 124, 981–990 987major part of the estimated uncertainty, and are typically muchlarger than the reference uncertainties.Calculation of Rm and u(Rm) from spiking studiesThis study illustrates the determination of Rm and u(Rm) fromthe analysis of portions of a sample matrix containing theanalyte, spiked at an appropriate concentration of the analyte.Approximately 10 g samples of soil reference material LGCRM 6138 were spiked with 0.5 ml of a 205 mg ml21 solution ofnaphthalene in dichloromethane.Previous analyses of thereference material had a mean of 29.5 mg kg21 with a standarddeviation of the mean of 0.63 mg kg21 (n = 11). These valuescorrespond to Cnative and u(Cnative) in eqns.(9) and (13)respectively. The uncertainty in the concentration of the spikeadded has contributions from the uncertainties associated withthe concentration of the stock solution and the volumetricglassware used to prepare and add the spiking solution. Theuncertainty in the concentration of the stock solution wasquoted by the supplier (Supelco, Bellefonte, PA, USA) as 0.005as a relative standard deviation. Based on previous experiencein our laboratory, the combined uncertainties associate were estimated as 0.007 (as a relativestandard deviation).Combining these elements using root sumof squares gives an uncertainty in the concentration of the spikeadded of 0.0086 as a relative standard deviation. This value wastherefore used to calculate u(Cspike(i)) in eqn. (13). Note thatthere is also a contribution from the weight of sample taken,however previous work has shown that such uncertainties aregenerally insignificant.The additional uncertainty has thereforenot been included. Based on the results of earlier studies of themethod, precision was estimated as 0.07 as a relative standarddeviation. This value was used to calculate the u(Cobs(i)) valuesin eqn. (13). The relevant results are presented in Table 5.Using eqn. (10), Rm was calculated as 0.876. Applying eqn.(13), u(Rm) was calculated as 0.1074. However in this case,simplifications can be applied, as discussed previously.Firstly,the estimates of u(Cspike(i)) are much smaller than u(Cobs(i)) andu(Cnative); typically 0.09 mg kg21 compared to 2.7 mg kg21 and0.63 mg kg21 respectively. Eqn. (14) can therefore be applied.This gives an estimate for u(Rm) of 0.1074. In addition, since thesame amount of spiking solution was added to each sample, andthe weights of each sample were similar, the standard deviationof the Cspike(i) values (0.072 mg kg21) is small compared to themean of the Cspike(i) values (10.20 mg kg21).The mean cantherefore be used and eqn. (15) applied. This gives an estimatefor u(Rm) of 0.1074. Finally, the estimates of the uncertaintyassociated with Cobs(i) are all similar (standard deviation ofu(Cobs(i)) = 0.092) so eqn. (16) can be applied. This leads to anestimate for u(Rm) of 0.1073. This example illustrates that whenthe above assumptions hold a relatively simple calculation canbe used to obtain an estimate of u(Rm).Estimation of Rm and u(Rm) from extraction monitoringstudiesThe aim of this study was to determine whether it would bepossible to predict the total concentration of the analytes in thesample using data obtained after the usual 6 h extraction period.An estimate of the method recovery could then be obtained bycomparing the amount extracted after 6 h with the predictedtotal amount present in the sample.Two portions of RM 6138 were prepared for analysis asdescribed previously.The samples were extracted for a total of14 h (two 7 h periods on consecutive days).After 1, 2, 4 and 6soxhlet cycles, and hourly intervals thereafter, small aliquots ofthe extraction solvent were removed for analysis. After 7 h theextraction was halted and left overnight. The extraction wasthen restarted and aliquots removed at hourly intervals for afurther 7 h. For comparison, a further two samples wereextracted for 14 h, after which time an aliquot of the extractionsolvent was removed and submitted for analysis by GC-MS.Based on the concentrations observed after extracting thesamples for 4, 5 and 6 h, eqn.(19) was used to calculate the totalanalyte concentration in the sample. The results are summarisedin Table 6. The results obtained after 14 h extraction are alsoincluded for comparison.The results indicate that in some cases the predicted totalconcentration in the sample, m0, is similar to the concentrationobserved after 14 h extraction (see m0/CTOTAL column in Table6).This assumes that after 14 h all of the PAHs present havebeen extracted from the sample, thus providing a reasonableestimate of the total amount of the analytes present. However,the predictions are variable, as can be seen from the differencein the results obtained for samples 1 and 2. The differences weresignificantly greater than those observed for the duplicate 14 hextractions. Table 7 compares the concentration observed foreach sample after 6 h extraction with the predicted totalconcentrations.In theory, this could be used to obtain anestimate of Rm. However, the uncertainties associated with suchan estimate are likely to be large. The expression for estimatingthe uncertainty in m0, u(m0), was obtained by differentiatingeqn. (19) to give:u mu mmm mu mmm mmm mu m( )( )( )( )( ) ( )( )012 222 32 3222 3222 32 222=+- ÆÙ ¢D ÆÙ + ¢D--- ÆÙ ¢D ÆÙ ¢X£»£»£»£»£»£»(30)12 3 m m - Note the terms which lead to a very large estimate ofu(m0) if m2 m3.Based on previous studies of the method precision, theuncertainty associated with each of the experimental observationsused to calculate m1, m2 and m3 was estimated as 0.07 (asa relative standard deviation).The uncertainties in m2 and m3are calculated by taking the root sum of squares of theuncertainties (as standard deviations) of each of the values usedin their calculation. u(m0) was calculated for each analyte andcompared with m0 to give an indication of the relativeuncertainty in each case.The results are presented in Table 8.The results indicate that m0 could be used as a rough estimate ofthe total amount of analyte in the sample, although in somecases the uncertainties are rather large. For example, in the caseof benzo[a]anthracene in sample 1 and benzo[k]fluoranthene insample 2, the associated uncertainties are too large for practicalTable 5 Data from the analysis of samples of RM 6138 spiked withnaphthaleneSampleno. Weight/gCobs(i)/mg kg21Cspike(i) /mg kg21u(Cobs(i))/mg kg21u(Cspike(i))/mg kg211 10.02 37.21 10.25 2.605 0.08812 10.01 39.90 10.26 2.793 0.08823 10.04 40.16 10.23 2.811 0.08804 10.15 39.29 10.12 2.750 0.08705 10.20 36.36 10.07 2.545 0.08666 10.10 39.27 10.17 2.749 0.08747 9.98 38.12 10.29 2.668 0.08858 10.07 37.51 10.20 2.626 0.08779 10.03 37.73 10.24 2.641 0.0881988 Analyst, 1999, 124, 981¡V990application.This arises when m2 2 m3 is small compared to m2. Note, however, that the present calculation is crude and based on only three data points from a continuous sequence.It should therefore be possible to improve the estimates and uncertainties significantly using improved modelling and curve fitting methods. Conclusions This paper has considered various approaches for estimating analytical bias (measured as recovery) and its associated uncertainty. It is useful to consider the recovery, R, for a particular sample as being composed of three components; �R m, Rs and Rrep respectively representing a reference recovery, specific sample correction and allowance for imperfect representativeness in spiking studies.�R m and its uncertainty u(�R m) can be estimated by existing methods. The simplest and most effective involves the analysis of a relevant certified reference material. Other techniques studied (re-extraction, extraction modelling, analysis of ‘worst case’ CRMs) currently appear to lead to larger (sometimes impractically large) associated uncertainties, but can provide an initial estimate of �R m.Where a method covers many matrices and a restricted number of CRMs is available, Rs and its uncertainty can be assessed from studies on various different matrices; it is particularly important to study a representative range. u(Rs) tends to be larger than u(�R m). In the absence of relevant reference materials, analysis of spiked samples is common, but uncertainty calculations may be intricate. Treatment of the data can be substantially simplified by appropriate approximations. However, estimating the necessary additional term Rrep and its uncertainty is one of the more problematic aspects of a recovery study and leads to very substantial uncertainties.There are, therefore, methods available for characterisation of uncertainty arising from recovery, but only direct application of relevant certified reference materials or reference methods currently provides a completely general method of characterising recovery well (that is, with small uncertainty) for a particular sample type.Other general methods, particularly modelling approaches, currently lead to large uncertainties, but do appear capable of improvement. This has important implications for international efforts to improve comparability via traceability to national and international standards. Large uncertainties in nominally traceablents imply poor comparability, and there are large uncertainties associated with matrix change and analyte recovery.Given the practical impossibility of producing relevant reference materials for all matrices and analytes, it is important that these effects are characterised well to minimise uncertainties associated with the inevitable use of imperfectly matched reference materials. In Table 6 Predicted concentrations of PAHs in RM 6138 m0/ng ml21a Total extracted m0/CTOTAL after 14 h Analyte Sample 1 Sample 2 (CTOTAL/ng ml21)a Sample 1 Sample 2 Naphthalene 2931.31 2660.2 3027.8 0.968 0.879 Acenaphthylene 1180.35 970.6 1257.5 0.939 0.772 Fluorene 607.4 456.8 647.8 0.938 0.705 Phenanthrene 10062.4 8864.2 10374.4 0.970 0.854 Anthracene 2780.8 2446.8 3039.0 0.915 0.805 Fluoranthene 10976.6 9901.8 11384.9 0.964 0.870 Pyrene 8980.7 8161.9 9320.3 0.964 0.876 Benzo[a]anthracene 1553.1 3304.4 4033.1 0.385 0.819 Chrysene 4111.5 3767.4 4354.8 0.944 0.865 Benzo[b]fluoranthene 3253.2 8143.8 3500.7 0.929 2.326 Benzo[k]fluoranthene 1265.4 1007.8 1667.8 0.759 0.604 Benzo[a]pyrene 3125.5 2632.1 3349.7 0.933 0.786 Indeno[1.2.3-cd]pyrene 2755.3 2280.4 2990.0 0.922 0.763 Dibenz[a,h]anthracene 674.8 671.9 725.9 0.930 0.926 Benzo[g,h,i]perylene 2269.8 2107.6 2530.7 0.897 0.833 a All concentrations corrected to 10 g sample.Table 7 Comparison of the concentration of analyte extracted after 6 h with the predicted total concentration Concentration after 6 h extraction (C6 h/ng ml21) C6 h/m0 Analyte Sample 1 Sample 2 Sample 1 Sample2 Naphthalene 3039.0 2406.1 1.04 0.90 Acenaphthylene 1192.7 767.9 1.01 0.79 Fluorene 626.8 418.0 1.03 0.92 Phenanthrene 10263.8 8077.0 1.02 0.91 Anthracene 2924.4 2220.5 1.05 0.91 Fluoranthene 11318.2 8957.6 1.03 0.90 Pyrene 9326.5 7214.6 1.04 0.88 Benzo[a]anthracene 3962.7 2963.3 2.55 0.90 Chrysene 4378.4 3226.0 1.06 0.86 Benzo[b]fluoranthene 3266.8 2553.4 1.00 0.31 Benzo[k]fluoranthene 1286.7 1146.8 1.02 1.14 Benzo[a]pyrene 3159.7 2348.5 1.01 0.89 Indeno[1.2.3-cd]pyrene 2769.4 1982.4 1.01 0.87 Dibenz[a,h]anthracene 680.3 580.0 1.01 0.86 Benzo[g,h,i]perylene 2235.0 1693.5 0.98 0.80 Table 8 Uncertainties associated with m0 Sample 1 Sample 2 Analyte u(m0)/ ng ml21 u(m0)/ m0 u(m0)/ ng ml21 u(m0)/ m0 Naphthalene 299.2 0.10 606.0 0.23 Acenaphthylene 164.8 0.14 604.9 0.63 Fluorene 38.2 0.11 76.2 0.17 Phenanthrene 1 217.5 0.12 1 838.9 0.21 Anthracene 225.2 0.08 582.3 0.24 Fluoranthene 1 093.7 0.10 2 353.7 0.24 Pyrene 926.8 0.10 3 079.2 0.38 Benzo[a]anthracene 235 884 151.9 840.7 0.25 Chrysene 326.8 0.08 2 570.4 0.68 Benzo[b]fluoranthene 388.5 0.12 382 747 47.0 Benzo[k]fluoranthene 192.0 0.15 77.6 0.08 Benzo[a]pyrene 467.7 0.15 751.7 0.29 Indeno[1.2.3-cd]pyrene 350.4 0.13 998.9 0.44 Dibenz[a,h]anthracene 90.3 0.13 323.2 0.48 Benzo[g,h,i]perylene 215.8 0.10 2 097.7 1.00 Analyst, 1999, 124, 981–990 989other words, general implementation of traceability through reference materials will require improvements in methodology for characterising recovery and matrix effects if traceability is to be a generally useful means of ensuring comparability in analytical chemistry.Acknowledgement Production of this paper was supported under contract with the Department of Trade and Industry as part of the National Measurement System Valid Analytical Measurement Programme. References 1 M. Thompson, Analyst, 1995, 120, 117N. 2 W. Horwitz and A. Albert, Analyst, 1997, 122, 615. 3 S. L. R. Ellison, W. Wegsheider and A. Williams, Anal. Chem., 1997, 69, 607A. 4 J. S. Kane, Analyst, 1997, 122, 1283. 5 S. L. R. Ellison and A. Williams, Accred. Qual. Assur., 1998, 3, 6. 6 S. L. R. Ellison, Accred. Qual. Assur., 1998, 3, 95. 7 ISO, Guide to the Expression of Uncertainty in Measurement, International Standards Organisation, Geneva, 1993. 8 EURACHEM GUIDE: Quantifying Uncertainty in Analytical Measurement, Laboratory of the Government Chemist, London, 1995. 9 M. Pueyo, J. Obiols and E. Vilalta, Anal. Commun., 1996, 33, 205. 10 A. Williams, Anal. Proc., 1993, 30, 248. 11 Analytical Methods Committee, Analyst, 1995, 120, 2303. 12 S. L. R. Ellison, Accred. Qual. Assur. 1998, 3, 95 13 S. L. R. Ellison and A. Williams, Accred. Qual. Assur., 1998, 3, 6. 14 W. Horwitz, Pure Appl. Chem., 1988, 60, 885. 15 AOAC recommendation, J. Assoc. Off. Anal. Chem., 1989, 72, 694. 16 ISO 5725:1994, Accuracy (Trueness and Precision) of Measurement Methods and Results, International Standards Organisation, Geneva, 1995. 17 S. L. R. Ellison and V. J. Barwick, Accred. Qual. Assur., 1998, 3, 101. 18 S. L. R. Ellison and V. J. Barwick, Analyst, 1998, 123, 1387. 19 V. J. Barwick and S. L. R. Ellison, Anal. Commun., 1998, 35 377 20 K. D. Bartle, A. A. Clifford, S. B. Hawthorne, J. J. Langenfeld, D. J. Miller and R. Robinson, J. Supercrit. Fluids, 1990, 3, 143. 21 S. L. R. Ellison and A. Williams, in The use of recovery factors in trace analysis, ed. M. Parkany, Royal Society of Chemistry, Cambridge, UK, 1996. 22 Soil Quality—Determination of polycyclic aromatic hydrocarbons (PAH), Draft British Standard, 1998. 23 T. J. Farrant, Practical Statistics for the Analytical Scientist, A Bench Guide, Royal Society of Chemistry, Cambridge, UK, 1997. Paper 9/01845J 990 Analyst, 1999, 124, 981–990

ISSN:0003-2654    

DOI:10.1039/a901845j

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Perspective A natural history of analytical methods† Michael Thompson Department of Chemistry, Birkbeck College, London, UK WC1H 0PP Received 29th April 1999, Accepted 19th May 1999 It is postulated that, since analytical methods adapt to user requirements by an evolutionary process, the Horwitz function, which predicts reproducibility precision, must be regarded as a fitness-for-purpose criterion in appropriate sectors of analytical chemistry. Analytical methods once they are devised are, like lifeforms, subject to evolution.Natural selection, here mediated by analytical chemists, ensures that only the fittest analytical methods survive. Methods that provide false information with too high a probability, or true information at too high a price, become extinct, while their modified descendants may survive. We can confidently expect, after many generations and extinctions, that analytical methods will automatically have adapted to their environment. In short, they will approach fitness for purpose (in the sense defined as loss minimisation1).We can even expect that a general pattern of performance will emerge naturally for a whole sector of analysis, just as such patterns emerge for lifeforms. The Horwitz function,2 an empirical relationship between reproducibility precision and concentration of the analyte, is just such a manifestation of this evolutionary process. The Horwitz function, observed in data collected in the analysis of food and drugs, is a remarkable generalisation.Over a wide range of concentration it describes the properties of an analytical method well, apparently without regard to the nature of the analyte or the type of test material, and regardless of the physical principle underlying the analytical method or the complexity of the procedure. It applies equally well to data collected in 1920 or in 1990. It is true (in the sense of unbiased) for analyte concentrations spanning a concentration range from about 5 ppb to 10%m/m, about seven orders of magnitude.Perhaps most remarkably, the Horwitz function can be expressed simply, although, at first sight, rather oddly: the predicted reproducibility standard deviation is proportional to the concentration of the analyte raised to a power of about 0.85, specifically, sH = 0.02c0.85, when the variables are expressed as mass fractions. At the moment we cannot explain why the function takes this particular form. This type of fractional dimension is, however, seen in many examples from biological evolution.For example, we see it in a plot of brain mass versus body mass in mammals. This too is a manifestation of evolution towards fitness for purpose in the mammalian class. The nature of the Horwitz function is clearly not determined by chemistry or physics, but by the use to which the analytical data is being put. Accordingly, we are almost compelled to regard the function as more than a mere description of interlaboratory uncertainty.We may think of it, at least for the present and where similar evolutionary pressures exist, as a general purpose criterion of fitness for purpose. The Horwitz function seems moderately transportable; for example, although it is derived from data drawn from the food and drugs sector, the function applies rather well to the analysis of rocks. Of course it cannot be universally applicable: there are analytical sectors that demand better reproducibility.An obvious example is in the assaying of precious metals, a more hostile environment than food and drugs in some respects. A less obvious example is in the determination of residues of toxic compounds at sub-ppb concentrations. Regardless of its current success, we must exercise caution with the Horwitz function in the future. It may represent fitness for purpose now, but purposes change and analytical methods will have to adapt to the new environments. Such evolution may well take place in a more consciously directed manner, for instance, in the context of loss minimisation when the joint costs of sampling uncertainty and analytical uncertainty are considered. References 1 M. Thompson and T. Fearn, Analyst, 1996, 121, 275. 2 R. Albert and W. Horwitz, Anal. Chem., 1997, 69, 789. Paper 9/03450A † The opinions expressed in the following article are entirely those of the authors and do not necessarily represent the views of either The Royal Society of Chemistry or the Editor of The Analyst Analyst, 1999, 124, 991 991

ISSN:0003-2654    

DOI:10.1039/a903450a

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

OH OH CH2 (CH2OH)n Separation and characterisation of phenol–formaldehyde (resol) prepolymers using packed-column supercritical fluid chromatography with APCI mass spectrometric detection Michael J. Carrott and George Davidson* School of Chemistry, University of Nottingham, University Park, Nottingham, UK NG7 2RD. E-mail: George.Davidson@nottingham.ac.uk Received 2nd March 1999, Accepted 17th May 1999 Packed-column supercritical fluid chromatography (pSFC), with negative-ion atmospheric-pressure chemical-ionisation (APCI) mass spectrometric (MS) detection has been used to analyse a commercial phenol–formaldehyde (resol) prepolymer.After initial optimisation of the system using a standard dimer species [bis(2-hydroxyphenyl)methane], it was possible to identify positively 34 components in the resin sample, including the starting reagents (phenol and cresol), and a range of dimers, trimers, tetramers and pentamers, with varying phenol/cresol ratios and amounts of methylol substitution.The method requires no pre-separation derivatisation of the resin sample, and a chromatographic run time of ca. 10 min. Introduction Phenolic resins are some of the oldest synthetic polymers, and include materials such as Bakelite.1 Their manufacture involves two stages: (i) prepolymer production by an acid or base catalysed condensation reaction between phenol (or substituted phenols) and formaldehyde; and (ii) curing of these low molecular weight intermediates by the use of heat or a crosslinking agent.There are three basic types of prepolymer (random novolac, high-ortho novolac and resol), depending on the conditions used for the condensation reaction. Acid catalysed condensations produce novolac type prepolymers, whereas base catalysed conditions result in the formation of the resol prepolymer. We will concentrate in this study on resol prepolymers. A schematic representation of the compound types for the resol dimer is shown below. The formula enclosed in square brackets denotes the basic structure, and the number of methylol substituents associated with the oligomer (n = 0–4) is given in parentheses.Positional isomers for both methylol groups and the methylene linkage are associated with each of the structures, thus giving rise to a highly complex mixture. The compositions of the prepolymers are variable and depend upon the conditions of the condensation reaction. Since the properties of the cured resin are dependent upon the prepolymer composition, characterisation of the prepolymer is desirable.Several analytical techniques have been applied to the determination of the chemical composition and structure of molecular species in these mixtures, including GC-MS,2–5 HPLC,6 gel permeation chromatography (GPC),7 NMR8 and MS.9 Direct GC analysis of the prepolymers is impossible without derivatisation, due to the involatility and thermal instability of the compounds.GC separation of the isomers of novolac2 and resol3 prepolymers was achieved by formation of the trimethylsilyl derivatives. Subsequent detection by MS provided structural information for the identification of the components, and even enabled differentiation between positional isomers. However, GC-MS could not be applied to oligomers with more than three monomer units, due to the involatility of the compounds even after derivatisation. Pyrolysis GC-MS has been applied to the determination of the sequence of phenolic units and the position of the methylene linkage in the oligomers of a novolac resin.10 GPC has been applied to resol and novolac resins, but identification was difficult due to poor separation of the components.HPLC has been the most suitable technique thus far, as it can be applied to higher molecular weight compounds than can GC and provides higher resolution than GPC. However, identification of the components still proved to be difficult.Direct MS analysis using soft ionisation techniques, including field desorption (FD), fast atom bombardment (FAB), thermospray (TSP) and desorption chemical ionisation (DCI), has been applied to the characterisation of novolac and resol prepolymers.9 Soft ionisation methods produce simple spectra containing molecular or quasi-molecular ions, thus allowing characterisation of the resin in terms of the chemical species present, molecular weight distribution and average molecular weights.FD proved to be the most successful method for direct MS analysis of the prepolymers. However, resols still exhibited fragmentation due to thermal decomposition, which made the interpretation of the spectra difficult. Therefore, analysis of the acetylated resol derivative was performed to generate spectra without fragmentation. 1H-NMR has been used for the general characterisation of the condensation products,8 but this is unsuitable for the determination of molecular species in a complex mixture.Packed-column supercritical fluid chromatography (pSFC) is suitable for the analysis of high molecular weight compounds, and is capable of providing faster analysis times and improved resolution compared with HPLC. Application of pSFC, using carbon dioxide with a methanol modifier, has already been demonstrated for the separation of the oligomers of novolac and resol prepolymers.11 However, in both cases, UV detection was used and identification of individual components was difficult.Therefore coupling SFC with atmospheric-pressure chemicalionisation mass spectrometry (APCI-MS) would allow separation and identification of the components in the prepolymer. By employing a soft ionisation technique, such as APCI, it is possible to obtain molecular weight information, which is Analyst, 1999, 124, 993–997 993invaluable for the identification of components in such a complex mixture. We have recently demonstrated the successful application of pSFC-APCI-MS to the analysis of cannabis samples,12 polymer additives13 and explosive substances.14 Experimental All SFC analyses were performed using a Gilson packedcolumn SFC system (Anachem, Luton, UK) coupled to the APCI source of a Trio 2000 quadrupole mass spectrometer (VG Biotech, Altrincham, UK).The SFC mobile phase was delivered using two Gilson piston pumps. A microprocessorcontrolled Gilson 308 pump, fitted with a chiller unit (Anachem) to cool the pump head to 210 °C, was used to deliver SFC-grade CO2 (99.99%, BOC, Guildford, UK).A Gilson 306 pump was used for the programmed addition of methanol modifier to the mobile phase. The pumps were connected to a Gilson 311C dynamic mixer to ensure homogeneity of the mobile phase. UV detection was carried out at 230 nm, using a Jasco 875-CE UV detector (Jasco, Tokyo, Japan). Separations of the resol standards and mixtures were achieved using a 25 cm 3 4.6 mm id column with a C18 stationary phase. Samples were introduced using a 10 ml injection loop and eluted using a CO2 mobile phase with a methanol modifier gradient.Chromatographic details are given in the appropriate Results section. The packed-column SFC system was interfaced to APCI-MS using a tapered 75 mm id restrictor, with a heated tip, inserted into the APCI probe. The position of the restrictor and probe, the gas flow rates and probe temperature were all separately optimised.15 Comparison with UV chromatograms showed that the interface had some effect on the chromatographic resolution.There was therefore some evidence for mass-transfer problems leading to a two-phase system and impaired resolution. 16 However, using mass-selected chromatograms, effective separations were possible, as will be discussed below. The mass spectrometer was operated in both positive- and negative-ion modes. BP International Ltd. (Sunbury-on-Thames, UK) provided a resol sample for characterisation.The resol was produced from a condensation reaction using phenol, cresol and formaldehyde, which results in a mixture containing a range of oligomers substituted with one or more methylol groups. In addition each oligomer may contain a varying number of cresol and phenol units. A 1.5% (w/v) solution was prepared in methanol for analysis by SFC-MS. Reference mass spectra were obtained for a resol dimer standard using a 0.04 mg ml21 solution of bis(2-hydroxyphenyl) methane (Mw = 200) (Aldrich, Gillingham, Dorset, UK) in methanol.Mass spectra were recorded in both positiveand negative-ionisation modes. Tuning and optimisation of the MS parameters were also performed using replicate injections of the standard solution. Details of the optimisation processes are available as Electronic Supplementary Information.† Results and discussion Optimisation of chromatographic conditions SFC-MS analysis of bis(2-hydroxyphenyl)methane was used to carry out the optimisation.It was first necessary to determine whether positive- or negative-ion APCI conditions were more suitable for the detection of such compounds. A positive-ion CI reference spectrum for bis(2-hydroxyphenyl)methane was obtained using the following conditions: elution using 5% methanol in CO2 at a flow rate of 2 ml min21, and a back pressure of 205 bar measured at the column inlet. Ionisation was achieved using methanol reagent ions, typically [MeOH]+ and [(MeOH)2H]+, generated from the mobile phase via a 3.0 kV discharge at the corona pin.The source temperature was maintained at 120 °C and the probe temperature at 300 °C. Nitrogen, at 60 psi, was used as the bath and sheath gas at flow rates of 50 and 100 l h21, respectively. Ions were sampled into the mass spectrometer using a sampling cone voltage of 30 V, and full scan spectra recorded from 100 to 300 u in 1.0 s. The mass spectrum obtained for the standard compound showed major ions in the spectrum at m/z 181 and 197; unfortunately, there was no molecular ion or protonated molecular ion at m/z 200 or 201.Therefore, identification of the components in the resol sample would be difficult due to the lack of molecular weight information in the positive ion spectra. These compounds may not be particularly amenable to protonation due to the acidity of the phenolic OH group, thus producing the spectrum observed. A negative-ion CI spectrum for a resol dimer was obtained by eluting the standard compound with 5% methanol in CO2 at a flow rate of 2 ml min21, and a back pressure of 195 bar measured at the column inlet.Ionisation was achieved using reagent ions generated from the mobile phase via a 2.5 kV discharge at the corona pin. The source temperature was maintained at 120 °C and the probe temperature at 300 °C. Compressed air, at 60 psi, was used as the bath and sheath gas, at flow rates of 50 and 100 l h21, respectively. Ions were sampled into the mass spectrometer using a sampling cone voltage of 230 V, and full scan spectra recorded from 100 to 300 u in 1.0 s. The mass spectrum of the standard compound in negative-ion mode contains a single mass peak at m/z 199 corresponding to the [M 2 H]2ion.Therefore, negative-ion APCI is ideal for characterising the complex resol resin, as it is capable of providing molecular weight information without extensive fragmentation.Optimisation of the bath and sheath gas flow rates and the source temperature was performed for negative-ion APCI. Replicate injections of the standard compound were made, and the MS response monitored using the mass chromatogram for the [M 2 H]2 ion at m/z 199. The SFC and MS conditions used were identical to those stated previously in this section, except for the bath and sheath gas flow rates and the source temperature. The optimum value for the source temperature was found to be 150 °C, for the bath gas flow rate 100 l h21 and the highest sensitivity was obtained with no sheath gas.These values were then used in the analysis of the resol prepolymer, all other conditions being identical to those stated previously. Characterisation of a resol prepolymer In order to achieve separation of the components of the resol resin, pressure and modifier programmes were required. As a fixed restrictor was used to control the pressure, a flow programme had to be used to create the desired pressure programme.The flow and modifier programmes used and the corresponding increase in pressure are shown in Table 1. Fig. 1 shows the UV and MS total ion current (TIC) chromatograms obtained for the resol sample. The composition of the sample is so complex that identification of components from either the TIC chromatogram or the UV trace alone is impossible as many of the components are co-eluting. This mixture can, however, be resolved and the components characterised by using the selectivity of the mass spectrometer to look at the [M 2 H]2 ions of individual mass species.† Available as Electronic Supplementary material; see http://www.rsc.org/ supdata/an/1999/993. 994 Analyst, 1999, 124, 993–997The presence of starting material residues in the resol was confirmed by the mass chromatograms for [M 2 H]2 ions at m/z 93 and 107 (Fig. 2), which correspond to phenol and cresol, respectively. Fig. 2(a) shows the mass chromatograms for phenol and the methylol- and dimethylol-substituted products at m/z 123 and m/z 153, respectively. Dimethylol-substituted phenol is a very minor product, as shown by comparison of the peak intensities on the mass chromatogram corresponding to m/z 93 + 123 + 153 in Fig. 2(a). The presence of methylolsubstituted cresol is confirmed by the [M 2 H]2 ion at m/z 137 in Fig. 2(b); however, in this case, the dimethylol-substituted product is not observed.Resol oligomers, comprising phenol units only, from n = 2 to n = 5, were identified from the mass chromatograms of [M 2 H]2 ions at m/z 199, 305, 411 and 517, shown in Fig. 3. All the mass chromatograms are shown with the peaks at full scale on the y axis; therefore, these do not show the difference in relative abundances for the components. Comparison of the relative peak areas (trimer = 100%) for each oligomer in Table 2 indicates that the dimer and trimer are the major components in the resin, whereas the pentamer is only a minor component.Phenol and cresol were both used in the condensation reaction to produce this particular resol; therefore, each oligomer may be composed of varying numbers of phenol and cresol units. Hence, there are three possible structures for the dimer, containing zero, one or two cresol units, with molecular weights of 200, 214 and 228, respectively. For the trimer, there Table 1 Flow and modifier programmes Time/min Modifier (%) Flow/ml min21 Pressure/bara 0 2 1.25 156 2 2 1.25 156 15 30 3.00 385 25 30 3.00 385 a Pressure at column inlet.Fig. 1 SFC separation of a resol prepolymer resin with (a) UV and (b) APCI-MS detection (TIC chromatogram in the latter case). Fig. 2 Mass chromatograms for starting meterial residues in a resol prepolymer: (a) phenol and methylol-substituted products; (b) cresol and methylol-substituted products. Fig. 3 Mass chromatograms for phenol oligomers of resol.Analyst, 1999, 124, 993–997 995OH OH CH2 (CH2OH)n H3C Where n = 0,1,2,3 or 4 OH H3C OH OH CH2 CH2 CH3 are four possible structures, five for the tetramer and six for the pentamer. The three dimers may be identified from their characteristic [M 2 H]2 ions, and Fig. 4 shows the mass chromatograms for m/z 199, 213 and 227, thus confirming the presence of all three dimers in the resol prepolymer. Similarly, all four possible trimers are observed with mass chromatograms at m/z 305, 319, 333 and 347; however, only four structures are seen for the tetramer at m/z 411, 425, 439 and 453, and two structures are observed for the pentamer at m/z 517 and 531.The basic conditions employed in the condensation reaction also produce oligomers with methylol substituents on the aromatic ring, similar to those shown previously for phenol and cresol monomers (Fig. 2). Fig. 5 shows the mass chromatograms for a dimer composed of one phenol and one cresol unit, containing zero, one, two and three methylol substituents: The dimer and its methylol-substituted analogues can easily be identified from their corresponding [M 2 H]2 ions at m/z 213, 243, 273, 303 and 333.However, the mass chromatogram for the tetramethylol-substituted dimer at m/z 333 also coincides with the mass of the following trimer: The coincidence of masses for certain components is due to the presence of phenol and cresol analogues for each oligomer in addition to the methylol substituents. Identical masses occurred only when oligomers possessed four or more methylol substituents.The small proportion of dimethylol-substituted monomers (see above) suggests that such highly substituted oligomers are unlikely to be present in significant amounts. In addition, the trimer, tetramer and pentamer do not show such a high degree of substitution; the trimer was found to have a maximum of two methylol substituents, the tetramer only has one methylol substituent, and methylol substituted pentamer oligomers could not be detected, and no ambiguity exists.The compounds present in the resol identified from their [M 2 H]2 ions are listed in Table 3. Conclusion Packed-column SFC has been shown to be a powerful technique for the analysis of phenol-formaldehyde resin prepolymers, with negative-ion APCI-MS being capable of providing molecular weight information without extensive fragmentation. Table 2 Relative peak areas of oligomers Phenol oligomer m/z Relative peak area Dimer 199 51% Trimer 305 100% Tetramer 411 27% Pentamer 517 5% Fig. 4 Mass chromatograms for phenol and cresol dimers in resol prepolymer. Fig. 5 Mass chromatograms of methylol-substituted dimers. (The tetramethylol- substituted dimer at m/z 333 coincides with the mass of the trimer). Table 3 Summary of components identified by SFC-APCI-MS in the resol prepolymer resin Number of cresol units Resin Methylol component substituents 0 1 2 3 4 5 Phenol 0 93 1 123 2 153 Cresol 0 107 1 137 2 3* Dimer 0 199 213 227 1 229 243 257 2 259 273 287 6 303 317 Trimer 0 305 319 333 347 1 335 349 363 3 2 3 379 393 3 3 3 3 3 3 Tetramer 0 411 425 439 453 3 1 441 455 469 3 3 2 3 3 3 3 3 Pentamer 0 517 531 3 3 3 3 1 3 3 3 3 3 3 2 3 3 3 3 3 3 * 3, not detected. 996 Analyst, 1999, 124, 993–997Additionally, no sample derivatisation is required, either for the separation or detection of the components of the resol. The combination of these two techniques has enabled the identification of thirty four components, including starting material residues, from the characteristic [M 2 H]2 ions of components in a complex resol prepolymer. The identification of chemical species and the determination of the molecular weight distribution by SFC-APCI-MS is complementary to the information obtained from other techniques such as NMR, thus enabling the analyst to produce a more complete picture of the prepolymer composition.Acknowledgements The authors thank their colleagues in the School of Chemistry at the University of Nottingham, especially Dr M.W. George, for stimulating discussions and helpful advice, and the technical staff for invaluable help in the maintenance of the equipment, Mr T. Lynch, now of BP Chemicals, and Dr S. Bajic of VG Biotech. They are also grateful to BP for the supply of a resin sample, to VG Biotech for financial support, and to the EPSRC for a Research Studentship (to M. J. C.). References 1 A. Knop and L. A. Pilato, Phenolic Resins, Springer, Berlin, 1985. 2 L. Prokai, J. Chromatogr., 1985, 329, 290. 3 L. Prokai, J. Chromatogr., 1985, 331, 98. 4 L. Prokai, J. Chromatogr., 1985, 333, 161. 5 L. Prokai, J. Chromatogr., 1986, 356, 331. 6 L. Lai and L. Sangermo, J. Chromatogr., 1985, 321, 325. 7 M. Duval, B. Bloch and S. Kohn, J. Appl. Polym. Sci., 1972, 16. 1585. 8 T. H. Fisher, P. Chao, C. G. Upton and A. J. Dai, Magn. Res. Chem., 1995, 33, 717. 9 L. Prokai and W. Simonsick, Macromolecules, 1992. 25, 6532. 10 M. Blazso and T. Toth, J. Anal. Appl. Pyrolysis, 1991, 19, 251. 11 S. Mori, J. Chromatogr., 1989, 478, 181. 12 B. Backstrom, M. D. Cole, M. J. Carrott, D. C. Jones, G. Davidson and K. Coleman, Science and Justice, 1997, 37, 91. 13 M. J. Carrott, D. C. Jones and G. Davidson, Analyst, 1998, 123, 1827. 14 Y. McAvoy, K. Dost, D. C. Jones, M. D. Cole, M. W. George and G. Davidson, Forensic Sci. International, 1999, 99, 123. 15 M. J. Carrott, PhD Thesis, University of Nottingham, 1996. 16 T. L. Chester and J. D. Pinkston, J. Chromatogr. A, 1998, 807, 265. Paper 9/01683J Analyst, 1999, 124, 993–997 997

ISSN:0003-2654    

DOI:10.1039/a901683j

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Determination of the arabica/robusta composition of roasted coffee according to their sterolic content M. S. Valdenebro,b M. León-Camacho,b F. Pablos,a A. G. Gonzáleza and M. J. Martín*a a Department of Analytical Chemistry, University of Seville, 41012 Seville, Spain b Instituto de la Grasa, CSIC, 41012 Seville, Spain Received 22nd March 1999, Accepted 24th May 1999 A method for the determination of the percentage of the arabica coffee in mixtures of roasted coffee is proposed. The sterol content of roasted coffee blends was determined by extracting the coffee oil, saponifying the lipids and the sterols present in the unsaponifiable fraction were separated by thin layer chromatography.Then, they were converted into trimethyl silyl derivatives and analysed by gas chromatography. Twelve sterols were determined in roasted coffee samples which were mixtures of the arabica and robusta classes. Considering the sterols as chemical descriptors, principal component regression was applied.D5avenasterol was found to be a very adequate variable to establish the arabica percentage in roasted coffee blends. The method was applied to the determination of the arabica–robusta composition of commercial roasted coffee samples. Introduction The two species of coffee with greatest commercial significance are Coffea arabica and Coffea canephora, that are known in the trade, respectively, as arabica and robusta, and will be referred to by these names throughout.There are other species like Coffea excelsa and Coffea liberica but they are of minor interest for the market.1 Roasted coffee is commercially available as one of the two first classes, arabica and robusta, or blends of them. In the case of mixtures, the percentage of robusta is usually lower than arabica’s. Due to its more pronounced and finer flavour2 the arabica coffees are considered of better quality and consequently they command for higher prices. To take this situation into account, it is important to have appropriate methods to discriminate between the two classes mentioned and to determine the composition of their mixtures. The chemical analysis is a very useful tool to differentiate between these categories.Several studies3–5 about the chemical composition of coffee beans have been carried out. Some of the parameters used to characterise the arabica and robusta coffees are 16-O-methylcafestol,6,7 the metal content8,9 the volatile components. 3,10 The lipid composition of the coffee seeds has also been analysed.11 Particularly, the sterolic fraction of the lipids present in the coffee oil is a very interesting approach. Within this realm, there are several reports in the literature. 12–16 This paper deals with a method to determine the arabica percentage in mixtures of roasted coffee. A study of the sterolic fraction of arabica–robusta roasted coffee blends has been carried out. After the extraction of the coffee oil, the lipids have been saponified and the sterols present in the unsaponifiable fraction have been separated by thin layer chromatography (TLC), converted into trimethyl silyl (TMS) derivatives and analysed by gas chromatography (GC) with flame ionisation detection (FID).Considering the sterols studied in the blends as chemical descriptors and applying principal component regression (PCR), the composition of mixtures of roasted commercial coffees has been determined.Experimental Apparatus A Hewlett Packard (Palo Alto, CA, USA) 5890 II gas chromatograph equipped with a flame ionisation detector, a fused silica capillary column of 30 m 3 0.32 mm coated with a 0.2 mm film of HP-5 stationary phase and a Hewlett Packard 7673 automatic injector was used. The oven was isothermally operated at 265 °C and the injector and detector were held at 280 and 300 °C, respectively. Hydrogen was used as carrier gas at a flow rate of 0.7 ml min21 through the column with a split ratio of 1+80.Air and hydrogen with flow rates of 430 and 30 ml min21, respectively, were used for the detector, which had an auxiliary flow of 30 ml min21 of nitrogen. Reagents and solutions Cholesterol, campesterol, stigmasterol and b-sitosterol were obtained from Fluka (Buchs, Switzerland). Campestanol, D7campesterol, chlerosterol, sitostanol, D5avenasterol, D5,24stigmastadienol, D7stigmastenol and D7avenasterol were obtained and purified from a mixture of lard and olive, rapeseed and sunflower oils. 5a-Cholestane-3b-ol (Fluka) in a 0.2% (m/v) solution of isopropyl ether was used as an internal standard in the gas chromatography determinations. Acetone (Panreac, Barcelona, Spain), chloroform (Riedel-de Haën, Seelze, Germany), diethyl ether (Romil, Cambridge, UK), isopropyl ether (Fluka), ethanol (Riedel-de Haën), n-hexane (Romil), anhydrous sodium sulfate (Fluka) and potassium hydroxide (Merck, Darmstadt, Germany) used were of analytical reagent grade.A mixture (9:3:1,v/v/v) of anhydrous pyridine (Fluka), hexamethyldisilazane (Fluka) and trimethylchlorosilane (Fluka) was used as derivatization reagent. A 0.2% (m/v) solution of 2’,7’-dichlorofluorescein (Fluka) in ethanol was used to display the bands in the TLC separations. Samples Arabica and robusta green coffee samples were supplied by Kraft Jacobs Suchard. They were laboratory roasted, ground Analyst, 1999, 124, 999–1002 999and then stored in polyethylene flasks.Rossell et al.17 pointed out that the temperature treatments do not alter the sterol content of vegetable oil samples. Accordingly, slight differences in roasting from one lot to another should not be considered as a source of spreading with respect to the analytical procedure. In order to prepare the different arabica–robusta mixtures, two pools of arabica and robusta coffees were prepared by blending nine different lots of each class of coffee.These lots differed both in their origin and harvest year. Thus, for the arabica pool, the origins were: Brazil (2), Nicaragua (2), Honduras, Salvador, Colombia (2) and Guatemala; and for the robusta one: Vietnam, Indonesia, Cameroon (2), Uganda (3) and Ivory Coast (2). The harvest year for the individual lots, according to the information provided by Kraft Jacobs Suchard, ranged between 1992–1995. This randomization procedure may overcome the dependency on origin and crop to some extent.From these two pools, 13 blends were prepared. The arabica content varied in the range 100–40% (m/m). A single 100% pure robusta sample was also considered for comparison purposes. Commercial roasted coffee samples were purchased from the market. Before analysis, the samples were dried at 103 °C [ref. (18)] until constant weight to determine their moisture. Extraction of the coffee oil The extraction of the coffee oil and its saponification was performed according to the Directive 91/2568/CEE.19 Thus, about 8.0 g, exactly weighed, of roasted coffee sample, or blend, were extracted with hexane16,19 in a Soxhlet for 8 h, siphoning six times per hour. The extract was dried over anhydrous sodium sulfate and placed in a 250 ml round bottom flask.Using a vacuum rotary evaporator the solvent was evaporated and the residue was dried at 105 °C to obtain the exact amount of coffee oil. Then, 1.5 ml of internal standard solution was added.Saponification of the lipids The coffee oil obtained was treated with 50 ml of 2 M solution of potassium hydroxide in ethanol–water (80+20, v/v) and this mixture refluxed, with constant stirring, for 30 min. This way, the esters present in the oil were transformed into potassium salts that are soluble in water. The sterols did not react and could be extracted into non-polar solvents like ethyl ether. Next, 50 ml of water was added and extractions with 3 portions of 80, 70 and 60 ml of diethyl ether were carried out.The organic extract was separated and washed with 5 portions of 100 ml of water. Then, it was dried over anhydrous sodium sulfate, filtered into a 250 ml round bottom flask and concentrated on a rotary evaporator under reduced pressure at room temperature to distil the diethyl ether. At this step, the unsaponifiable fraction was dissolved into chloroform to get a solution of ca. 5% (m/v). Separation of the sterolic fraction from the unsaponifiable one by TLC The separation of the sterolic fraction from the unsaponifiable one by TLC presents a minor modification with respect to the CEE method.19 The eluent used for the TLC separation was a mixture hexane–diethyl ether–acetone (60+40+1, v/v/v) instead of benzene–acetone (95+5 v/v) or hexane–ethyl ether (65+35 v/v).This modification was recently proposed and established. 20 Accordingly, the solution obtained in the previous section was spotted on a potassium hydroxide-impregnated silica gel TLC plate. 5 ml of the internal standard solution was also spotted with the solution of the sample. The plate was developed using hexane–diethyl ether–acetone (60+40+1, v/v/v). Once dried, it was sprayed with the 2A,7A-dichlorofluorescein solution and the pink band of the sterols can be observed under UV light together with the spot of the internal standard. After identifying it, the band was scraped off and the sterols were dissolved with a portion of 10 ml of trichloromethane and two portions of 10 ml of diethyl ether.The solution obtained was filtered through a paper filter, the solvent was evaporated under reduced pressure and the sterolic fraction was dried in an oven at 105 °C. Determination of the sterols by GC The procedure used for the determination of sterols in the extracted coffee oil was an international standard one developed for determining the content and composition of sterols in animal and vegetable fats by gas chromatography.21 The reliability of the method was established already by an international interlaboratory test including 14 laboratories, organised by the ISO German member body in 1996/97 on samples of olive oil, sunflower seed and oil rapeseed in accordance with ISO 5725 standard.Accordingly, the sterolic fraction isolated from TLC, as indicated above, was treated with the derivatisation reagent to obtain the trimethylsilyl (TMS) derivatives, which are much more volatile.A volume of 0.05 ml of reagent for each mg of sterol was added. Aliquots of 2 ml of this solution were injected into the gas chromatograph and the ratio of the peak areas of the analyte and internal standard was used as analytical signal. The content of individual sterols was expressed as the percentage of the sterolic fraction obtained. Albeit the method is sound and well established, a previous study on the accuracy based on recovery assays from spiked samples was done. Four additions of standard mixtures of the studied sterols were spiked on samples of extracted coffee oils, and then, the analytical method was applied. Average recoveries (in %) for each sterol were calculated therefrom and are presented as follows: cholesterol (106.2%), campesterol (98.0%), campestanol (91.2%), stigmasterol (96.1%), D7campesterol (99.0%), chlerosterol (97.9%), b-sitosterol (99.7%), sitostanol (93.9%), D5avenasterol (94.6%), D5,24stigmastadienol (98.5%), D7stigmastenol (98.0%) and D7avenasterol (100.9%).These results show that the analytical procedure can be considered accurate according to the AOAC guidelines.22 Data analysis Twelve sterols have been analysed in the roasted coffee mixtures and considered as chemical descriptors. These sterols will be viewed as follows: cholesterol (COL), campesterol (CPR), campestanol (CPN), stigmasterol (STR), D7campesterol (D7C), chlerosterol (CLE), b-sitosterol (BSIT), sitostanol (SIT), D5avenasterol (D5), D5,24stigmastadienol (D524), D7stigmastenol (D7S) and D7avenasterol (D7A).Table 1 shows the content of the sterols found in the analysed blends, indicating the percentage of arabica in each of the mixtures. PCR calculations were made for the compositional analysis of the blends, using the statistical package CSS: STATISTICA from StafsoftTM (Tulsa, OK, USA). Results and discussion The analysis of the sterolic content by GC after TLC separation of the unsaponifiable fraction of the oil present in coffee samples has been proved to be very adequate to differentiate between arabica and robusta.15 Fig. 1 shows the chromatograms corresponding to 100% arabica, 100% robusta and 50% arabica roasted coffee samples, in which the peaks of the TMS derivatives of the sterols can be observed. As can be seen, the profile of the chromatogram is distinct and the major differences appear for peaks 9, 11 and 12, that correspond to D5, D7S and D7A, respectively.Resolution of arabica–robusta mixtures by PCR As indicated above, 12 descriptors feature each coffee mixture, which are the contents (in % m/m) of 12 different sterols. Thus 1000 Analyst, 1999, 124, 999–1002we have 13 mixtures (within 100–40% arabica) that can be arranged as a column vector y (column 1 of Table 1 except for the last row ) and a descriptor data matrix X of dimension 13 3 12 (from columns 2 to 13 of Table 1). The last row of Table 1 refers to a pure robusta sample for the sake of comparison, but cannot be considered in the modelling of blends.In multivariate analysis we often have to model a response y, in this case the % arabica of coffee mixtures, as function of several X-variables (here, the selected descriptors). Instead of considering a multiple linear regression to model y as a function on X, it is better to perform a principal component analysis (PCA) on the X matrix and use the issued principal component (PC) scores as the basis of fitting a multiple regression model.This approach is called principal component regression (PCR) and has found application in areas such as multivariate calibration, structure–activity relationships and modelling chemical properties of molecules.23 PCR joins PCA and regression techniques, and it is particularly appropriate to avoid multi-collinearity problems among X-variables.24 First, PCA is carried out on the autoscaled X matrix to derive a low dimension representation of the samples through a few PCs (which are orthogonal and cannot exhibit collinearity features).Then the yproperty can be modelled as a function of the scores of these PCs as follows: y = b0 + b1 (PC1 scores) + b2 (PC2 scores) + ... + bf (PCf scores) (1) f being the established dimensionality of the factor space. Thus, after PCA of our X autoscaled data matrix, two PCs (f = 2) were selected according to the Kaiser criterion,25 the Malinowski indicator function26 and the method of cross validation.27 Kaiser criterion, also called the average eigenvalue criterion is based upon retaining only those PC dimensions whose eigenvalues are above the average eigenvalue. If the data are autoscaled, then any component having an eigenvalue higher than unity is retained.Malinowski devised an indicator function called IND that is very sensitive to identify the true dimensionality of data matrices. IND is an empirical function based on the residual standard deviation of the data matrix and reaches a minimum for the right number of factors.Cross validation aims to identify those dimensions (first f PCs) with best predictive ability to rebuild the data matrix once some data groups were deleted. As indicated above, the three methods lead to two underlying factors, that is, a two PC model. The y property (% arabica) was then modelled with the corresponding PC scores giving a first linear fit % arabica = 70 + 19.3 (PC1 scores) 2 0.8 (PC2 scores) (2) with a correlation coefficient of 0.992.However, according to the t-test, the coefficient for the PC2 scores was found without statistical significance and the corresponding term dropped out from the model. The new fit was fairly linear %arabica = 70 + 19.3 (PC1 scores) (3) with a correlation coefficient of 0.991. The predicted values were plotted against the actual % arabica in Fig. 2. In order to detect possible outliers and for checking linearity, both analyses of residuals and regression ANOVA were used.A plot of the residuals on normal probability paper shows model adequacy. Residuals are also used to detect outliers. A very straightforward way is to consider as outlier any point whose residual is greater than twice the value of the standard deviation of the regression line.28 In our case, outliers were absent according to this criterion. The ANOVA of the regression shows that the F ratio between the Table 1 Composition of the sterolic fraction (%) of roasted coffee blendsa % Arabica COL CPR CPN STR D7C CLE BSIT SIT D5 D524 D7S D7A 100 1.2(1) 15.5(4) 0.73(2) 18.9(7) 0.6(2) 0.87(1) 52.7(2) 2.41(7) 2.84(4) 0.6(2) 2.04(3) 1.74(5) 95 1.2(1) 15.6(4) 0.65(2) 18.0(8) 0.6(4) 0.93(1) 53.0(8) 2.01(9) 3.05(8) 0.4(2) 1.76(9) 1.59(9) 90 0.9(3) 15.8(2) 0.68(1) 18.8(5) 0.5(3) 0.84(4) 52.5(7) 2.02(9) 3.72(8) 0.5(1) 1.75(5) 1.57(5) 85 1.1(2) 15.7(2) 0.67(1) 18.7(3) 0.6(3) 0.96(7) 52.2(2) 1.96(3) 4.27(6) 0.6(2) 1.71(8) 1.56(6) 80 1.1(2) 15.9(1) 0.65(3) 18.7(2) 0.5(5) 0.86(4) 52.3(4) 1.87(7) 4.59(7) 0.5(1) 1.54(2) 1.46(2) 75 1.0(1) 16.2(3) 0.67(3) 19.1(4) 0.6(1) 0.94(8) 51.6(2) 1.89(3) 5.07(6) 0.6(2) 1.49(6) 1.39(8) 70 0.9(2) 16.3(1) 0.56(9) 19.4(1) 0.5(1) 0.86(8) 50.5(4) 1.91(4) 5.70(6) 0.6(1) 1.42(6) 1.38(4) 65 1.0(1) 16.2(1) 0.62(3) 19.0(1) 0.6(2) 0.99(9) 50.2(6) 1.75(6) 6.09(8) 0.8(1) 1.41(9) 1.34(9) 60 0.8(1) 16.4(2) 0.73(9) 19.2(1) 0.7(2) 1.08(9) 49.7(6) 1.85(9) 6.80(4) 0.9(3) 1.40(9) 1.33(7) 55 — 16.6(1) 0.72(9) 19.1(1) 0.6(1) 0.99(8) 50.3(5) 1.83(9) 6.91(5) 0.6(1) 1.20(8) 1.21(3) 50 0.8(1) 16.6(4) 0.61(4) 19.1(4) 0.6(1) 0.95(7) 50.1(8) 1.80(4) 7.54(7) 0.7(3) 1.09(6) 1.14(3) 45 0.8(2) 16.5(1) 0.59(5) 18.8(2) 0.7(2) 0.96(8) 49.1(4) 1.67(8) 7.87(3) 0.8(1) 1.06(7) 1.11(9) 40 1.0(1) 16.7(1) 0.56(2) 19.1(2) 0.5(1) 0.98(6) 49.0(3) 1.53(1) 8.38(3) 0.7(1) 0.84(4) 0.99(2) 0 1.3(1) 17.2(1) 0.51(1) 18.1(2) 0.4(2) 0.84(8) 47.9(1) 1.34(4) 11.07(5) 0.6(1) 0.25(7) 0.55(1) a Average of triplicate determinations.Values between parentheses refer to the error corresponding to the last significant figure. Fig. 1 Chromatograms of the sterolic fraction of roasted coffee samples: (a) 100% arabica, (b) 100% robusta, (c) 50% arabica. (1) Cholesterol (COL), (I.S.) 5a-cholestane-3b-ol, (2) campesterol (CPR), (3) campestanol (CPN), (4) stigmasterol (STR), (5) D7campesterol (D7C), (6) chlerosterol (CLE), (7) b-sitosterol (BSIT), (8) sitostanol (SIT), (9) D5avenasterol (D5), (10) D5,24stigmastadienol (D524), (11) D7stigmastenol (D7S) and (12) D7avenasterol (D7A).Analyst, 1999, 124, 999–1002 1001regression variance and residual variance was above 600, indicating that linearity was fair. However, in spite of these findings, the prediction model was built by using PCs, that is, ‘abstract factors’ without a proper chemical definition. Our principal aim was to find relationships between the y variable and the chemical descriptors rather than PCs.Anyway, after a glance of eqn. 3, PC1 scores is the predictor variable. If the corresponding PC1 loadings are considered, one can find that the descriptors with highest contributions ( > 0.92) are CPR, BSIT, SIT, D5, D7S and D7A. A further study of correlation shows that all these descriptors are mutually correlated, with correlation coefficients ranging from 0.88 (CPR&SIT) to 1.00 (D7S&D7A). This indicates that all the variables contributing to PC1 are redundant and consequently, any of them gives the same information as the complete set.PC1 is the capital factor for modelling the % arabica. Besides, all the descriptors contributing to PC1 are redundant, and so this factor could be associated to one descriptor alone. Following this deductive way, the next step was the study of the prediction of % arabica by using as unique descriptor one of the set (CPR, BSIT, SIT, D5, D7S and D7A).The best fit was achieved with D5: % arabica = 129 2 10.5 D5 with a correlation coefficient of 0.998. In this case, the plot of residuals on normal probability paper was linear and again outliers were absent. The ANOVA of regression leads to a F ratio of about 2850 which indicates a very good linearity. The predicted values were plotted against % arabica in Fig. 3. The standard error of prediction was less than 1.1%. The result obtained is outstanding: By using the descriptor D5 alone, it is possible to predict the % arabica in coffee mixtures with better reliability than using the first PC.The possible explanation may arise from the fact that the majority of the contributing variables involved in the fundamental factor PC1, are mutually correlated and give the same information than one of them. The use of D5 as a key descriptor filters the possible noise provoked by the remaining descriptors and gives a good linear fit. In order to validate the results so obtained in the modelling study, a test set consisting of real coffee samples purchased in the market was considered. All the test samples were certified for the content in arabica or robusta that appeared on the label.The sterolic profile of seven samples 100% arabica, a mixture of 75% arabica and 100% robusta was obtained. From the D5 content, the % arabica was calculated. The results obtained were excellent as can be seen in Table 2. Conclusion The analysis of the sterolic fraction of the oil of roasted coffee blends provides a very useful tool for establishing the percentage of arabica–robusta.D5 is the most adequate chemical descriptor for predicting the arabica–robusta content in coffee samples. References 1 A. W. Smith, in Coffee, ed. R. J. Clarke and R. Macrae, Elsevier, London, 1985, vol. 1, pp. 3–6. 2 R. Briandet, E. K. Kemsley and R. H. Wilson, J. Agric. Food Chem., 1996, 44, 170. 3 C. P. Bicchi, O. M. Panero, G. M. Pellegrino and A.C. Vanni, J. Agric. Food Chem., 1997, 45, 4680. 4 M. J. Martín, F. Pablos and A. G. González, Anal. Chim. Acta, 1996, 320, 191. 5 M. Suchánek, H. Filipova, K. Volka, I. Delgadillo and A. N. Davies, Fresenius J. Anal. Chem., 1996, 354, 327. 6 K. Speer, R. Tewis and A. Montag, Z. Lebensm. Unters. Forsch., 1991, 192, 451. 7 M. D. Trouche, M. Derbesy and J. Estienne, Ann. Falsif. Expert. Chim., 1997, 90, 121. 8 M. J. Martín, F. Pablos and A. G. González, Anal. Chim. Acta, 1998, 358, 177. 9 S. J. Haswell and A. D. Walmsley, J. Anal. At. Spectrom., 1998, 13, 131. 10 A. Murota, Biosci. Biotechnol. Biochem., 1993, 57, 1043. 11 G. Muratore, M. C. Cataldi-Lupo, F. Fiorenza and C. N. Asmundo, Ind. Aliment., 1998, 37, 161. 12 E. Tiscornia, M. Centi-Grossi, C. Tassi-Micco and F. Evangelisti, Riv. Ital. Sostanze Grasse, 1979, 56, 283. 13 N. Frega, F. Bocci and G. Lercker, J. High Resolut. Chromatogr., 1994, 17, 303. 14 G. Lercker, M. F. Caboni, G. Bertacco, E.Turchetto, A. Lucci, R. Bortolomeazzi, E. Pagani, N. Frega and F. Bocci, Ind. Aliment., 1996, 35, 1186. 15 F. Carrera, M. León-Camacho, F. Pablos and A. G. González, Anal. Chim. Acta, 1998, 370, 131. 16 G. Lercker, N. Frega, F. Bocci and M. T. Rodríguez-Estrada, Chromatographia, 1995, 41, 29. 17 J. B. Rossell, S. P. Kochhar and I. M. Jawad, Chemical changes in soy oil during high temperature processing. Proceedings of the 2nd ASA Symposium on Soybean Processing, Antwerp, June 1981, p. 367. 18 International Organization of Standardization ISO 11294, 1994. 19 Directive 91/2568/CEE, Official journal L-248/15, September 1991. 20 F. Carrera, M. León-Camacho, F. Pablos and A. G. González, Anal. Chim. Acta 1998, 370, 131-139 21 ISO/DIS 12228: 1997. 22 AOAC, Peer verified method program, manual on policies and procedures, Arlington, VA, November, 1993. 23 W. P. Gardiner, Statistical Analysis Methods for Chemists, The Royal Society of Chemistry, Cambridge, 1997, p. 311. 24 R. Henrion and G. Henrion, Multivariate Datenanalyse, Springer- Verlag, Berlin, 1995, p. 129. 25 H. F. Kaiser, Educ. Psychol. Meas., 1966, 20, 141. 26 E. R. Malinowski, Factor analysis in Chemistry, 2nd edn., Wiley, New York, 1991. 27 S. Wold, Technometrics, 1978, 20, 397. 28 A. G. González, M. A. Herrador and A. G. Asuero, Talanta, 1999, 48, 729. Paper 9/02245G Fig. 2 PCR predicted vs. actual of % arabica in arabica–robusta mixtures. Fig. 3 D5avenasterol predicted vs. actual of % arabica in arabica–robusta mixtures. Table 2 Percentage of arabica in commercial coffees Claimed 100 100 100 100 100 100 100 75 0 (100% Robusta) Found 104 103 103 95 101 105 100 75 3 (97% Robusta) 1002 Analyst, 1999, 124, 999–1002

ISSN:0003-2654    

DOI:10.1039/a902245g

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Chiral determination of various adrenergic drugs by thin-layer chromatography using molecularly imprinted chiral stationary phases prepared with a-agonists Roongnapa Suedee,*a Teerapol Srichana,b Juraiporn Saelima and Thitirat Thavornpibulbuta a Department of Pharmaceutical Chemistry, Faculty of Pharmaceutical Sciences, Prince of Songkla University, Hatyai, Songkhla, Thailand 90110. E-mail: sroongna@ratree.psu.ac.th b Department of Pharmaceutical Technology, Faculty of Pharmaceutical Sciences, Prince of Songkla University, Hatyai, Songkhla, Thailand 90110 Received 22nd March 1999, Accepted 2nd June 1999 Thin-layer chromatography (TLC) based on molecularly imprinted polymers (MIPs) of a-agonists as chiral stationary phases was applied to the determination of enantiomers of various adrenergic drugs including a- and b-agonists and b-antagonists (b-blockers). In this study, three MIPs imprinted with (+)-ephedrine, (+)-pseudoephedrine and (+)-norephedrine plus a non-imprinted polymer (non-MIP) were prepared, processed and coated on a glass support as thin layers.Then enantiomeric determination of adrenergic drugs was carried out by development of their racemates on the TLC plates, using established conditions. From the results, the racemates of the compounds used as print molecules were well separated into two isomers on the MIP-plates, except on the plate based on MIP of (+)-norephedrine. Most adrenergic drugs structurally related to print molecules were completely resolved into two spots with the MIP plates.In general, the retention of (+)-isomers (or 1S-isomers) was greater than that of (2)-isomers (or 1R-isomers), indicating the stereoselectivity of the MIPs with the former isomers. Moreover, the role between the chemical structures of the analytes with chiral recognition of the MIPs has been investigated. The proposed method enables rapid determination of enantiomers and screening of large numbers for optical purity of adrenergic drugs.Introduction The chiral determination for verifying enantiomeric purity/ composition of pharmaceutical drugs is a growing concern to analytical chemists and the pharmaceutical regulatory agencies. The reason for this is that optical isomers of pharmaceutical drugs can have differences in pharmacological activities, side effects and even toxic effects. In this context, it is necessary to find reliable, sensitive and rapid methods for analysis and characterization of enantiomers of optically active compounds.Most enantiomeric determinations are performed using spectroscopic and chromatographic methods. Spectroscopic methods are extremely valuable in the stereochemical analysis of drugs, however, they present difficulties during the analysis of a pure enantiomeric drug.1 Chromatographic methods such as HPLC, GC and TLC provide satisfactory determination of enantiomeric composition or purity.2 Chromatographic procedures have been developed with both direct methods using chiral stationary phases (CSPs) or chiral mobile phase additives (CMAs), and indirect methods using a chiral derivatizing reagent.3,4 Both GC and HPLC methods are sensitive but time consuming, costly and restricted to a certain class of compounds, 3,5,6 whereas TLC possesses several advantages over other methods such as simplicity, rapidity, low cost and simultaneous detectability.Molecular imprinting is a technique for the preparation of tailor-made chiral selectors, which is rapidly growing in the field of chiral separation.During the last decade, MIPs have been prepared for numerous classes of either achiral or chiral compounds, e.g., amino acid,7–9 sugars,10 and a number of pharmaceutical drugs,11–16 mostly for HPLC application. There are several advantages of the employment of MIP in chiral discrimination. Firstly, the enantiomeric order of elution is foreseen by predetermining of the enantiomer selected as the print molecule. A MIP permits molecular recognition for several types of compounds varying according to chiral template, whilst other chiral selectors have enantioselectivity only with certain types of compounds.MIP can be reused after removing the print molecule from such a polymer. Finally, the need to screen a range of CSPs to find one that affects a given separation can be dismissed when a MIP is used and subsequently the cost of analysis is reduced.From the advantages stated above of TLC separation and imprinting technique, the application of MIP in TLC will be a useful tool for enantiomeric determination of chiral drugs. The first report on the employment of MIP as a TLC stationary phase was made by Kriz et al.,17 involving the separation of enantiomers of amino acid derivatives. Our group further applied this approach for enantiomeric separations of a number of pharmaceutical drugs.18–20 Adrenergic drugs are agents that exert pharmacological and therapeutic effects on the autonomic nervous system producing either stimulation (adrenergic agonist) or decrease (adrenergic antagonist) in sympathetic activity.They contain at least one asymmetric carbon which provides the molecule in more than two stereoisomeric forms. As the pharmacological and/or pharmacokinetic action of each form of these drugs are different,21–24 the separation of enantiomers of these drugs is therefore important.The most simple and rapid method used for the determination of optical purity of adrenergic drugs is thinlayer chromatography; one which can be performed by an indirect method with the use of a chiral derivatizing reagent25–28 or complex–ligand exchange plate29 or a chiral ion pair reagent as mobile phase additive.30,31 Recently, we reported the direct enantioseparation of four adrenergic agonists including nor- Analyst, 1999, 124, 1003–1009 1003ephedrine, pseudoephedrine, ephedrine and epinephrine by TLC based on MIPs of both (2)-norephedrine and (2)-pseudoephedrine as CSPs.19 The result shows the potential of this method for chiral resolution of such compounds.The objective of this work was to separate and determine enantiomers of a number of adrenergic drugs by applying the previous work. The adrenergic drugs studied included four aagonists, two b-agonists and four b-blockers, widely used in clinical therapeutics. (Their chemical structures are given in Tables 2 and 3.) In order to extend the previous work,19 we imprinted the polymers with (+)-pseudoephedrine, (+)-norephedrine and (+)-ephedrine by the use of a thermal polymerization method.Note that the latter compound has not yet been subjected to this approach. The preparation procedure of the TLC plate of MIPs was entirely adopted from our previous work.18–20 The suitable chromatographic conditions for chiral determination by TLC of each compound were examined.Experimental Reagents Ethylene glycol dimethacrylate (EDMA) and methacrylic acid (MAA) were obtained from Aldrich (Milwaukee, WI, USA). 2,2A-Azobis (butyronitrile) (AIBN) was purchased from Janssen Chemica (Geel, Belgium). (+)-Ephedrine and racemic ephedrine were obtained from Sigma (St. Louis, MO, USA). (+)-Pseudoephedrine, (2)-pseudoephedrine, (2)-ephedrine HCl, (2)-norephedrine and (+)-norephedrine HCl were obtained from Aldrich. Racemic norephedrine, epinephrine, isoproterenol, propranolol HCl, oxprenolol HCl and pindolol were obtained from Aldrich.Nadolol was a generous gift from Schwartz (Monheim, Germany). The free bases were prepared by neutralization of the aqueous solution of the salts with 1 M NaOH. All other reagents (ACS certified reagent grade) were used without further purification. The preparation of polymeric materials The preparation procedure of the polymers (Table 1) was similar to that outlined by Vlatakis et al.12 Methacrylic acid and ethylene glycol dimethacrylate were used as functional monomer and cross-linking monomer, respectively.A print molecule (3 mmol) was dissolved in dichloromethane (40 ml) and then MAA (12 mmol), EDMA (0.31 mol) and initiator (AIBN) (0.1 mmol) were added. The mixture was degassed under vacuum in an ultrasonic bath for 5 min and then purged with nitrogen for 5 min. The polymerization was carried out by heating of the mixture at 40 °C for 16 h. The bulk polymer was ground to a fine powder in an agate pestle and mortar and sieved through a 100 mm mesh.The sifted particles were collected and the remaining particles were reground. Then, the polymer was stirred in a solvent mixture of acetic acid and methanol (1 + 9 v/v) for 24 h, and filtered. The precipitate was washed with methanol. Finally, the polymer was dried under vacuum. Non-MIP included as the control was prepared in the absence of the print molecule. Particle size data of each polymer were acquired using an Aerosizer (Amherst, MA, USA).A Jeol 5700 scanning electron microscope (JSM 5800 LV; Palo Alto, CA, USA) was used to determine particle morphology. Thin-layer chromatography TLC plates were prepared according to the method described in previous work.18–20 Each polymer (1 g) and CaSO4 (1 g) were carefully mixed with distilled water (3 ml) by means of a pestle and mortar. The slurry was poured on standard glass microscope slides (76 3 26 mm), which then spread as a thin layer with a layer thickness of 0.25 mm.The plates were dried for 24 h at room temperature. The polymeric materials of the same batch were exploited to maintain consistency of their physical properties. The analytes were four a-agonists; pseudoephedrine, norephedrine, ephedrine and epinephrine, two b-agonists; salbutamol, isoproterenol and four b-antagonists; propranolol, oxprenolol, pindolol and nadolol. An analyte was dissolved in methanol (2–3 mg ml21) and 1 ml of solution was applied manually to TLC plates by use of a Hamilton syringe (Altech, Deerfield, IL, USA).The plate was then dried in air and developed with eluent in which a satisfactory retention and/or resolution can be obtained (see Table 2 for eluents of a- and bagonists and Table 3 for eluents of b-blockers). The solvent fronts were left to migrate approximately 60 mm and development time was typically 5 min. To detect a spot, after being sprayed with detection reagent (listed in Tables 2 and 3 for each analyte), the plate was heated gently with the hot air of a blower until the spot intensely appeared.The racemates of ephedrine, pseudoephedrine, norephedrine, epinephrine and propranolol were analyzed for each enantiomer by spotting of a pure enantiomer of these compounds alongside. RFs of separated enantiomers (or racemate) were measured and presented in the Tables as hRF values (RF3100). For unseparated racemate, hRF of each enantiomer was displayed in equal value as that of one spot.Chiral separation factor (a) was calculated according to previous work.19 Duplicate determinations for each separation were performed. Results and discussion Molecularly imprinted polymer-based chiral stationary phases for TLC Although imprinted polymer can be obtained either by photopolymerization methods or thermal polymerization methods, in this study the preparation of polymers was based on a thermal polymerization method. The polymerization was carried out at 40 °C, which is lower than that (60 °C) described by Vlatakis et al.12 with the purpose of reducing the degradation of the print Table 1 Polymer preparation and physical data Polymer No.Print molecule Solventb Particle sizec/ mm Surface aread/ m2 g21 Swellinge/ ml ml21 Bulk densityg/ g ml21 1a — CH2Cl2 31 0.22 1.63 0.46 2 (+)-ephedrine CH2Cl2 37 0.19 1.89 0.45 3 (+)-pseudoephedrine CH2Cl2 27 0.26 1.63 0.42 4 (+)-norephedrine CH2Cl2 40 0.16 1.83 0.46 5 (+)-isoproterenol THF 17 0.38 2.20f 0.33 a Polymer 1 was prepared in the absence of print molecule.b Solvent used in polymerizing process. c Mean value of particle size. d The surface area was determined from equivalent geometric diameter of the product using the Aerosizer® software. e Volume of dried polymer/volume of swollen polymer in acetonitrile determined as described elsewhere.32 f The polymer also swelled in water. g Determined as described elsewhere32. 1004 Analyst, 1999, 124, 1003–1009molecules during polymerization. The use of methacrylic acid as functional monomer with dichloromethane as solvent in the polymerization process gave the polymeric materials the capability of coating as a stationary thin layer on a glass support. Besides a-agonists, we tried to use a b-agonist, (2)-isoproterenol, as a print molecule. In preparation of this polymer, THF was employed as a solvent in the polymerization process due to low solubility of (2)-isoproterenol in a non-polar solvent.It was found that the stationary layer obtained with this polymer was easily rubbed off. This observation was also found in earlier work made by the authors20 for preparation of the polymers with the same method of polymerization but different types of print molecules and functional monomer. However, this result was not found when THF was used as a solvent in preparation of the polymer by photo-initiation at low temperature (4 °C), as observed by our group.19 From this finding in both the present and previous work, we can conclude that the type of solvent employed in the polymerization process has a significant influence on the characteristics of the MIP-stationary layer. Physical properties of the polymeric materials SEM of all the polymers was carried out.Fig. 1 shows SEM images obtained from non-MIP (A) and MIP of (+)-ephedrine (B). It can be seen from Fig. 1, non-MIP was composed of irregularly shaped particles having a smooth surface and MIP imprinted with (+)-ephedrine was mostly agglomerates of random irregular particles, which was similar to other MIPs (their electron micrographs are not shown).These results show the difference between the morphologies of non-MIP and MIPs, which were prepared in the absence and presence of print molecule, respectively, under given conditions. Physical properties of TLC–adsorbent often reflect on the adhesion of adsorbent on a support or the characteristic of the TLC layer obtained or the chromatographic efficacy, i.e., the retention of analyte on a stationary phase.The physical data of polymer particles used is shown in Table 1. The size distributions of the polymer particles were 30–40 mm. Typically, the polymers had a surface area approximately 0.20 m2 g21 and a bulk density of 0.45 g ml21. The polymers having Table 2 Eluent and detection conditions used for a- and b-adrenergic agonists Compound Eluent Detection reagent Color of spot a-Agonist— Ephedrine 10% Acetic acid in acetonitrile Ninhydrin reagent Purple–blue Pseudoephedrine 5% Acetic acid in acetonitrile Acidified potassium permanganate reagent Brown Norephedrine 10% Acetic acid in acetonitrile Ninhydrin reagent Purple–blue Epinephrine 10% Acetic acid in acetonitrile Ninhydrin reagent Purple–blue b-Agonist— Isoproterenol 5% Acetic acid in acetonitrile Acidified potassium permanganate reagent White (on brown background) Salbutamol 10% Acetic acid in acetonitrile Acidified potassium permanganate reagent White (on brown background) Analyst, 1999, 124, 1003–1009 1005such physical properties produced desirable layers, except the MIP imprinted with (+)-isoproterenol, which gave a brittle layer.It should be noted that the polymer imprinted with (2)-isoproterenol had swelling ability and low bulk density, which caused layer-coating problems. TLC chromatographic conditions We examined several solvent systems to use as a chromatographic eluent and found that acetonitrile was appropriate for elution-development of a- and b-agonists, while a non-polar solvent such as dichloromethane was preferred for that of bblockers which are less polar than a- and b-agonists.In addition, we found that modification of both solvents with acetic acid in the range of 5–10% resulted in the analytes being less retained and enabled better separation of the enantiomers as well as improving the spot shape.Although the analytes were UV absorbing, most of this activity was lost on the MIP-plate surface due to the strong absorbance of MIP in UV light. Moreover, the residual print molecule in the polymer (determined quantitatively by an IR method described elsewhere18–20) may interfere in the spot visualization, hence its amount would have to be kept low so as not to cause such interference. In this study, the spot detection of analytes was based on the use of detection reagents.Nonetheless, the MIPs also responded themselves by giving color reactions with the detection reagents. Thus, a sensitive and selective detection reagent was required. In the search for detection reagents of adrenergic drugs, three reagents were found. The ninhydrin reagent was suitable for detection of ephedrine, norephedrine and epinephrine with the greatest sensitivities. The acidified potassium permanganate reagent enabled the detection of pseudoephedrine, isoproterenol and sulbutamol.The anisaldehyde reagent enabled the visualization of all b-blockers. The potassium permanganate reagent gave a brown background with white spots for salbutamol and isoproterenol, or in a reverse manner in the case of pseudoephedrine. The reaction spots by the anisaldehyde reagent were unstable with the chromatographic zone fading quickly after 5 min. Of all the detection reagents used the ninhydrin reagent gave the most stable spots.Separation of adrenergic drugs on MIP prepared against (+)-ephedrine Table 4 gives the retention and resolution data of ten adrenergic drugs for CSP based on MIP of (+)-ephedrine. This CSP could enantiomerically resolve seven out of ten racemates of adrenergic drugs and the rest were not separated into individual isomers. Seven racemates that separated into isomers were ephedrine, pseudoephedrine norephedrine, sabutamol, pindolol, propranolol and oxprenolol.In these resolutions, the hRF values of (+)-isomers were lower than those of (2)-isomers, indicating greater affinity of the former isomers. Also, the racemates of b- Table 3 Eluent and detection conditions used for b-blockers Compound Eluent Detection reagent Color of spot Nadolol 7% Acetic acid in dichloromethane Anisaldehyde reagent Pink Pindolol 7% Acetic acid in dichloromethane Anisaldehyde reagent Purple–blue Propranolol 5% Acetic acid in dichloromethane Anisaldehyde reagent Blue Oxprenolol 5% Acetic acid in dichloromethane Anisaldehyde reagent Purple–blue 1006 Analyst, 1999, 124, 1003–1009blockers except nadolol were completely separated into two spots.When the resolution of racemic propranolol occurred, the hRF value of the (+)-R-enantiomer was however higher than that of the (2)-S-enantiomer (as the two spots obtained from racemic propranolol were identified). Generally, for the enantiomerically separated case, this CSP produced the a values > 1.3 with tailing of the spot up to 10–12 mm.No enantiomeric resolution was observed for racemic epinephrine, isoproterenol and nadolol on this CSP. It should be noticed that the first two racemates possess the catechol substituent in their structures and the latter racemate has two hydroxyl groups on the ring extended from the benzene ring (see Tables 2 and 3). The catechol group increases hydrophilicity of the molecule, resulting in the change in retention of an analyte.It is reasonable for the retention results that were obtained for catechol derivatives because the CSP contains the polar carboxyl groups; racemic epinephrine and isoproterenol moderately or less retained with the polar eluent, while with a non-polar solvent, racemic nadolol was significantly retained. However, racemic salbutamol was completely resolved into two spots with this CSP, although its structure is closely related to those of the catechol derivatives. Separation of adrenergic drugs on MIP prepared against (+)-pseudoephedrine Table 5 shows the hRF data and the a values of ten adrenergic drugs for CSP based on MIP of (+)-pseudoephedrine. Chromatographic behaviors of adrenergic drugs on this CSP were similar to those on CSP based on MIP of (+)-ephedrine.On this CSP, the retention values of the enantiomers of adrenergic drugs, omitting those of isoproterenol and propranolol, were highest compared with those obtained on other CSPs. As expected, this CSP was able to resolve the print molecule into two spots, providing a higher hRF value of the (2)-isomer than that of the (+)-isomer. The racemates of all a-agonists were completely resolved into two spots but the tailing spots occurred in the case of epinephrine.However, the highest a value was achieved for racemic epinephrine, which was hardly separated on other CSPs. This CSP did not afford the enantiomeric resolution for racemic salbutamol, whilst the CSP based on MIP of (+)-ephedrine exhibited moderate enantiomeric resolving capability for this compound.Like CSP based on MIP of ephedrine, this CSP enabled resolution of the enantiomers of b-blockers but not in the case of nadolol, and again, the (2)-isomer of propranolol was more retarded than the (+)-isomer. Generally, with this CSP, the a values of the separated spots were more than 1.2, this value being close to that obtained with CSP based on MIP of (+)-ephedrine.Separation of adrenergic drugs on MIP prepared against (+)-norephedrine Table 6 displays the retention and resolution data of ten adrenergic drugs for CSP based on MIP of (+)-norephedrine. From the results, this CSP showed chiral resolving ability with Fig. 1 Scanning electron micrographs of (A) non-MIP, (B) MIP imprinted with (+)-ephedrine. Table 4 Retention and resolution data of adrenergic drugs on MIP prepared against (+)-ephedrine hRF valuea Compound Isomer 1 Isomer 2 ab a-Agonist— Ephedrine 22 34 1.54 (+) Pseudoephedrine 31 51 1.64 (+) Norephedrine 16 25 1.56 (+) Epinephrine 67 67 1.00 (+) b-Agonist— Isoproterenol 42 42 1.00 Salbutamol 34 51 1.50 b-Blocker— Nadolol 4 4 1.00 Pindolol 14 22 1.57 Propranolol 56 73 1.30 (2) Oxprenolol 40 53 1.32 a The hRF values at room temperature are averages of two determinations, the standard deviation being less than 0.5.b a = RF(isomer 2)/RF(isomer 1). The sign of the optical rotation of the retained isomer is shown in parentheses. Table 5 Retention and resolution data of adrenergic drugs on MIP prepared against (+)-pseudoephedrine hRF valuea Compound Isomer 1 Isomer 2 ab a-Agonist— Ephedrine 34 47 1.38 (+) Pseudoephedrine 44 58 1.32 (+) Norephedrine 24 42 1.75 (+) Epinephrine 34 64 1.88 (+) b-Agonist— Isoproterenol 25 25 1.00 Salbutamol 60 60 1.00 b-Blocker— Nadolol 33 33 1.00 Pindolol 24 36 1.50 Propranolol 45 54 1.20 (2) Oxprenolol 44 58 1.32 a The hRF values at room temperature are averages of two determinations, the standard deviation being less than 0.5.b a = RF(isomer 2)/RF(isomer 1). The sign of the optical rotation of the retained isomer is shown in parentheses. Analyst, 1999, 124, 1003–1009 1007b-agonists and b-blockers other than a-agonists. Rather surprisingly, and contrary to our anticipation, this CSP could not separate the enantiomers of compounds corresponding to the print molecule but enabled the separation of the enantiomers of related compounds such as pseudoephedrine.In the previous work,19 the CSP based on MIP of (2)-norephedrine, prepared by use of the photo-polymerization method, resolved the enantiomer of print molecules rather than other a-agonists. This suggests that MIPs prepared with different methods may produce the difference in enantiomeric resolving capability of MIP. Not only was this CSP able to resolve racemic isoproterenol into two spots, but it also provided the best enantiomeric resolution; that was achieved for the separation of racemic salbutamol (a value = 2.3).Again, b-blockers except racemic nadolol were completely separated into two spots with the same elution order of propranolol enantiomers as that obtained on other CSPs. With this CSP, when the resolution occurred, the a values of 1.3–2.3 were produced. The examination of TLC behavior of ten adrenergic drugs on the control plate based on non-MIP was also performed. It was found that none of the adrenergic drugs were resolved into two spots on this type of plate (data not shown).This result confirmed the enantiomeric discrimination contributed by the MIPs. In this work, the MIPs demonstrate good TLC chromatographic characteristics and efficient enantiomeric resolution with various adrenergic drugs under given conditions. In addition, the resolution results of some adrenergic drugs such as ephedrine, pseudoephedrine and propranolol are similar to those in the previous work on HPLC,14 regarding the determination of mimicked adrenoceptor binding evaluated from recognition properties of MIPs prepared against (+)-ephedrine and (+)-pseudoephedrine.The results have shown that the chiral determination for optical purity of three classes of adrenergic drugs is feasible with the employment of TLC based on MIPs of three aagonists. In addition, it would be very interesting to determine the enantiomeric compositions of these drugs; however, there will have to be a further search for a detection reagent providing a highly stable reaction spot.The TLC chromatograms of the high degree of stereoselectivity obtainable with this method are presented in Fig. 2. The plates A, B, C, D and E illustrate the separation of the enantiomers of norephedrine, pseudoephedrine, salbutamol, propranolol and oxprenolol, respectively, on CSPs described in the figure caption. The role of the chemical structures of the analytes in enantiomeric recognition of the MIPs We have investigated the role of the chemical structures of the analytes with chiral recognition of MIPs for the results obtained on TLC separation. As the MIPs could separate the enantiomers of compounds corresponding to the print molecules, the recognition sites formed after polymerization were very similar to the structure of print molecules which are the (+)-isomer (or 1S-isomer) of two stereogenic center-containing compounds.Generally, the resolution results indicate the stereoselectivity of MIPs with (+)-isomers of print molecules and the related compounds.This means that geometry around the b-carbon which is similar to that of the print molecule, was probably essential for enantiomeric recognition of the MIPs. Similarly, previous work14,19 has pointed out the critical role of the bchiral carbon to enantiomeric recognition of MIPs prepared against either (+)- or (2)-isomer of a-agonists. In this study, the enantiomeric determination was performed with a wider range of structures of adrenergic drugs than in the previous study,19 giving additional information concerning chiral recognition of the MIPs imprinted with the a-agonists. All of the compounds studied have three functionalities in common: the aromatic ring, b-hydroxyl group and a-amine, but the aryl and amine subtituents vary. In addition, there is a –O– CH2-group situated between the aromatic ring and the hydroxyl carbon in b-blockers (see the chemical structures in Tables 2 and 3).The results show that besides the print molecule, the MIPs can enantiomerically resolve closely related compounds with different chiral separation factors, indicating the role of other substituents of the molecule to chiral recognition of the MIPs. To investigate this, the a values obtained for the MIPs were compared and elucidated. The a value > 1 was obtained for several compounds having either a symmetric or asymmetric carbon atom at the a-position and the amine substituents differ from those of the print molecules, particularly in the case of b-blockers, indicating that the a-carbon and amine side chain are less important for the chiral recognition of the MIPs.The enantiomeric resolution of b-blockers with all the MIPs implies that the –O–CH2-linkage, or on the other hand the extension of distance between the aromatic ring and the amine group did not affect the enantiomeric resolution of the MIPs.The conclusions described above are in agreement with those on a MIP-column.14 As exemplified by b-blockers, the substituents on the aromatic region, particularly the lipophilic group, did not influence the chiral recognition of the MIPs. However, the catechol group renders negative resolution on MIPs, for example in the case of racemic epinephrine and isoproterenol. This behavior might be a result of the steric hindrance of two hydroxyl groups in catechol on the interaction of the aryl group Table 6 Retention and resolution data of adrenergic drugs on MIP prepared against (+)-norephedrine hRF valuea Compound Isomer 1 Isomer 2 ab a-Agonist— Ephedrine 22 22 1.00 (+) Pseudoephedrine 40 60 1.50 (+) Norephedrine 18 18 1.00 (+) Epinephrine 31 31 1.00 (+) b-Agonist— Isoproterenol 37 57 1.54 (+) Salbutamol 18 42 2.33 b-Blocker— Nadolol 11 11 1.00 Pindolol 31 40 1.29 Propranolol 57 71 1.25 (2) Oxprenolol 42 60 1.43 a The hRFvalues at room temperature are averages of two determinations, the standard deviation being less than 0.5.b a = RF(isomer 2)/RF(isomer 1). The sign of the optical rotation of the retained isomer is shown in parentheses. Fig. 2 Representative TLC chromatograms of the high degree of separations for (A) racemic norephedrine on MIP-plate of pseudoephedrine, (B) racemic pseudoephedrine on MIP-plate of ephedrine, (C) racemic salbutamol on MIP-plate of norephedrine, (D) racemic propranolol on MIPplate of ephedrine and (E) racemic oxprenolol on MIP-plate of pseudoephedrine. The signs of + , 2 and ± are (+)-isomer , (2)-isomer and racemate, respectively. 1008 Analyst, 1999, 124, 1003–1009with the enantiomeric binding site. Furthermore, the highly hydrophilic characteristics of compounds owing to the polar hydroxyl groups is virtually considered as a concomitant effect. With regard to the hydrophilic effect, the polar molecule may be quickly eluted with the polar solvent and consequently decrease the binding in favor of the antipode with receptor.On the other hand, the steric effect is expressed with nadolol eluted with a non-polar solvent due to the presence of the polar hydroxyl groups on the aryl moiety. The observed effect is also decreased by a change in functional group hydrophilicity to a less polar group, such as the case of salbutamol; methyl alcohol substituting for the m-hydroxyl group of cathecol. In spite of that, there are some exceptions to this behavior of catechol derivatives, for example, racemic epinephrine was resolved on CSP based on MIP of (+)-pseudoephedrine.The reason for this is unclear, but may have been due to the more favorable influence of other functionalities on chiral recognition of the MIPs. Conclusions The employment of MIPs as stationary phases in TLC is a substantially useful method of determining optical isomers for enantiomeric purity. Although the previous chiral determinations of adrenergic drugs, using indirect methods in TLC have been reported in the literature,25–31 the chiral determinations obtained with the proposed method are rapid and effective with many classes of adrenergic drugs.With this method, the screening for optical purity of a large number of adrenergic drugs will be possible. The observed correlation between structurally related compounds could also be useful in predicting whether it is possible to determine the enantiomers of optically active compounds on MIP.References 1 T. J. Wozniak, R. J. Bopp and E. C. Jensen, J. Pharm. Biomed. Anal., 1991, 9, 363. 2 W. H. Pirkle and D. J. Hoover, in Topics in stereochemistry, ed. N. L. Allinger, E. L. Eliel and S. H. Wilen, Interscience Publishers, New York, 1982, vol. 13, p. 263. 3 R. Dennis, Pharm. Int., 1986, 7, 246. 4 S. G. Allenmark, Chromatographic enantioseparation methods and application, Ellis Horwood, Chichester, 1988. 5 D. B. Campbell, Eur. J.Drug Metab. Pharmacokinet., 1990, 15, 109. 6 A. MacKirdy, Lab. Pract., 1990, 39,13. 7 L. I. Andersson, D. J. O’Shanessy and K. Mosbach, J. Chromatogr., 1990, 516, 167. 8 M. Kempe and K. Mosbach, J. Chromatogr., 1995, 691, 317. 9 M. Kempe and K. Mosbach, J. Chromatogr., 1995, 694, 3. 10 G. Wulff and J. Haarer, Makromol. Chem., 1991, 192, 1329. 11 L. Fischer, R. Muller, B. Ekberg and K. Mosbach, J. Am. Chem. Soc., 1991, 113, 9358. 12 G. Vlatakis, L. I. Andersson, R. Muller and K. Mosbach, Nature (London), 1993, 361, 645. 13 M. Kempe and K. Mosbach, J. Chromatogr. A, 1994, 664, 276. 14 O. Ramstrom, Y. Cong and K. Mosbach, J. Mol. Recognit., 1996, 9, 691. 15 B. Sellergren, J. Chromatogr., 1994, 673, 133. 16 J. Haginaka, H. Takahira, K. Hosaya and N. Tanaka, J. Chromatogr. A., 1998, 816, 113. 17 D. Kriz, C. K. Berggren, L. I. Andersson and K. Mosbach, Anal. Chem., 1994, 66, 2636. 18 R. Suedee, C. Songkram, A. Petmoreekul, S. Sangkunakup, S. Sankasa and N. Kongyarit, J. Planar Chromatogr., 1998, 11, 272. 19 R. Suedee, C. Songkram, A. Petmoreekul, S. Sangkunakup, S. Sankasa and N. Kongyarit, J. Pharm. Biomed. Anal., 1999, 19, 519. 20 R. Suedee, T. Srichana, J. Saelim and T. Thavornpibulbut, J. Planar Chromatogr., in the press. 21 A. M. Barret and V. C. Cullum, Br. J. Pharmacol., 1968, 34, 43. 22 T. Walle, J. G. Webb, E. E. Bagwell, U. K. Walle, H. B. Daniell and T. E. Gaffney, Biochem. Pharmacol., 1988, 37, 115. 23 G. Egginger, W. Lindner, G. Brunner and K. Stoschitzky, J. Pharm. Biomed. Anal., 1994, 12, 1537. 24 W. A. Clementi, T. Q. Garvey, G. D. Clifton, R. A. McCoy, S. Brandt and S. Schwartz, Chirality, 1994, 6, 169. 25 H. Weber, H. Spahn, E. Mutschler and W. Mohrke, J. Chromatogr., 1984, 307, 145. 26 G. Gubitz and S. Mihellyes, J. Chromatogr., 1984, 314, 462. 27 D. Heuser and P. Meads, J. Planar Chromatogr., 1993, 6, 324. 28 J. C. Spell and J. T. Stewart, J. Planar Chromatogr., 1997, 10, 222. 29 P. E. Wall, J. Planar Chromatogr., 1989, 2, 228. 30 A. M. Tivert and A. Backman, J. Planar Chromatogr., 1989, 2, 472. 31 J. D. Duncan, D. W. Armstrong and A. M. Stalcup, J. Liq. Chromatogr., 1990, 13, 1091. 32 B. Sellergen and K. J. Shea, J. Chromatogr., 1993, 635, 31. Paper 9/02257K Analyst, 1999, 124, 1003–1009 1009

ISSN:0003-2654    

DOI:10.1039/a902257k

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Determination of residues of the plant growth regulator chlormequat in pears by ion-exchange high performance liquid chromatography-electrospray mass spectrometry† James R. Startin,* Simon J. Hird, Mark D. Sykes, John C. Taylor and Alan R. C. Hill Central Science Laboratory, Sand Hutton, York, UK, YO41 1LZ. Tel: +44 (0)1904 462587; Fax: +44 (0)1904 462111; E-mail: j.startin@csl.gov.uk Received 6th April 1999, Accepted 13th May 1999 We report a method which we have used routinely for the determination of chlormequat residues in pears.After extraction with methanol, determination was performed, without clean-up, by ion-exchange HPLC using an SCX column eluted with aqueous ammonium formate–methanol, and HPLC-MS with an electrospray interface. MS and MS-MS were employed concurrently, using selected ion monitoring and selected reaction monitoring, respectively, of the 35Cl and 37Cl isotopes of the chlormequat cation and the CID transitions of each of these precursors to the common product ion at m/z 58. The method was suitable for determinations at concentrations of chlormequat cation of 0.04 mg kg21.Concentrations determined using the four signals were in good agreement (mean RSD 3%). The mean recovery of chlormequat cation at 0.16 mg kg21, measured using the m/z 122?58 signal, was 86% (RSD 7%) under repeatability conditions and 88% (RSD 15%) in routine application of the method over a 3 month period. Analysis of an in-house reference sample of pears, similarly analysed over the 3 month period, gave an RSD of 10% with a mean of 0.14 mg kg21.Mean recovery at 0.016 mg kg21, under repeatability conditions on two occasions, was 101% (RSD 6%) and 56% (RSD 12%). Introduction Chlormequat is the ISO common name for the 2-chloroethyltrimethylammonium cation, ClCH2CH2N+(CH3)3. This is an important plant growth regulator, most commonly applied as the highly water-soluble chloride (also known as chlorocholine chloride, CCC, or Cycocel) and widely used in many parts of the world in the culture of cereals, fruit and ornamental plants.1 In the UK, chlormequat, alone or mixed with other growth regulators, is approved for use only on cereals, ornamental plants and certain oilseeds;2 other uses are not currently permitted.Chlormequat is, however, approved for use on pears and other fruit in some other countries, and the Codex Alimentarius and European Union maximum residue limit (MRL) of 3 mg kg21 (of chlormequat cation) is applicable to produce from these countries.Various methods for the determination of chlormequat residues have been reported but these have generally been based on non-specific determinative techniques, necessitating timeconsuming procedures for clean-up and for separation of chlormequat from naturally occurring or co-formulated choline and other choline derivatives. Ion-exchange and alumina columns have generally been employed prior to determination, 3–11 although thin-layer chromatography (TLC) has been used for separation prior to colorimetry.4 The determinative steps have included colorimetry after chemical derivatisation or complexation,3–6 TLC with colorimetric detection and visual comparison7 or densitometry,8,9 and GC determination as the derivative formed by treatment with sodium thiophenolate.10,11 The use of potassium pentafluorothiophenolate to form an electron-capturing derivative has also been reported,12 but the derivative formed was subsequently shown to be non-specific and the method to be potentially prone to interferences in the presence of a variety of methylating agents.13 A GC approach based on determination of methyl chloride produced by pyrolysis14 has been shown to be prone to interferences and poor yield, leading to unreliable quantification.15 An alternative pyrolysis method based on formation of acetylene under alkaline conditions has been reported,15 but acetylene is nonspecific and this method, again, requires extensive clean-up.The ionic nature of chlormequat makes it an excellent candidate for ion-exchange high performance liquid chromatography (HPLC) and also for detection by electrospray (ES) mass spectrometry (MS). Although the use of thermospray ionisation for the detection of chlormequat and other quaternary ammonium pesticides has been demonstrated,16 and ES ionisation has been used for the detection of chlormequat and other quaternary ammonium pesticides separated by capillary electrophoresis, 17,18 residues monitoring by these techniques has not been reported.There are few references to the use of ion-exchange HPLC with ES, possibly because the concentrations of electrolytes needed in the mobile phase to provide counter-ions, raise ionic strength, and control pH are generally expected to suppress the formation of gas-phase ions from the analyte.19 We have previously presented20 a method for the determination of chlormequat in apples and pears that was based on a simple extraction procedure, adapted with little modification from that described by Pasarela and Orloski,6 followed by HPLC-ES-MSMS using a mobile phase containing 0.05 M ammonium acetate.We report here a revised version of the method which, in this form, has been in routine use in our laboratory for several months. Experimental Materials Methanol and water of HPLC grade and ammonium acetate of analytical-reagent grade were obtained from Fisher Scientific (Loughborough, UK).Celite 545 was obtained from BDH † Crown copyright. Analyst, 1999, 124, 1011–1015 1011(Poole, Dorset, UK). Chlormequat chloride (99.5% purity) was purchased from QMx (Halstead, UK); a stock standard solution of 1 mg ml21 was prepared in methanol, and working standards of 100, 10 and 1 mg ml21 were prepared by dilution in methanol. Pears for use as analytical blanks, in recovery determinations, and for preparation of matrix-matched calibration solutions were organically-produced and were shown not to contain detectable residues of chlormequat. For use as an in-house reference material, a 2 kg sample of pears containing a field-incurred residue of chlormequat was milled as described below, and the resulting powder stored at 220°C.Matrix-matched calibration solutions with concentrations of chlormequat chloride of between 0.004 and 1.0 mg ml21 (equivalent to crop concentrations of chlormequat cation of between 0.016 and 3.9 mg kg21) were prepared by dilution of the working standard solutions, taking aliquots of 0.02–0.1 ml, with extracts of blank samples to give a final volume of 10 ml.Sample processing Pears were frozen at 218 ± 2 °C, overnight. A minimum of 10 fruits or 1 kg was taken for processing. Without thawing, approximately 10%, by weight, of dry ice was added to the sample which was then ground to a fine powder in a Stephan (Hameln, Germany) UM12 mill and mixed well.If the dry ice appeared likely to dissipate completely during this procedure, further small quantities were added. Prior to extraction the remaining dry ice was allowed to dissipate while the material was kept at 218 ± 2 °C for a minimum of 16 h. Extraction Without thawing, 20 g of milled sample material was placed in a 100 ml wide-necked, screw-capped bottle. For the determination of recovery, chlormequat chloride was added as 40 ml of a 10 mg ml21 or 100 mg ml21 solution, to give 0.02 mg kg21 and 0.2 mg kg21, respectively.Methanol (20 ml) was added and the mixture dispersed thoroughly with an Ultra-Turrax (Janke & Kunkel, Staufen, Germany) T-25 mixer at 20 500 rpm for 0.5 min. The bottle was capped and agitated for 4 h using a rotary shaker. Celite 545 (10 ml) was added and mixed well. The mixture was filtered under vacuum through a pre-wetted type 541 filter paper (Whatman, Maidstone, UK) bearing a 3–5 mm bed of Celite 545.The bottle and filter were rinsed 3 times with water–methanol (3 3 10 ml; 1+1 v/v). The combined filtrate was made up to 100 ml with water–methanol (1+1 v/v). A 1 ml aliquot was filtered through a 0.45 mm 3 13 mm nylon syringe filter (Rainin Inst. Co., Woburn, USA) prior to analysis by HPLC-MS. HPLC-MS-MS determination HPLC-MS-MS was performed using a Micromass (Manchester, UK) Quattro I tandem quadrupole mass spectrometer with ES interface, a Gilson (Villiers le Bel, France) 231 autosampler and a Waters (Milford, MA, USA) 600MS HPLC pump.The injection volume was 20 ml (filled loop method). The HPLC column was 150 mm 3 2.1 mm packed with Partisil 10 SCX (Capital HPLC, Broxburn, UK). For use with previously unused columns, the mobile phase was prepared by dissolving 12.6 g ammonium formate in 1 l of 1+1 v/v water–methanol (to give 0.2 M) and filtering though a 0.45 mm filter. After extended use of the column, which resulted in decreased retention times under these conditions, the ammonium formate concentration was reduced to 0.1 M, and subsequently to 0.05 M.The mobile phase was supplied at a flow rate of 0.2 ml min21 and was degassed with helium continuously during use. The mass spectrometer was operated in positive ES mode. The source temperature was 120 °C, nebuliser gas flow 20 l h21 and bath gas flow 400 l h21. After an initial tuning using background ions, the instrument operating conditions were adjusted for maximum sensitivity while making repeated injections of standards with the HPLC column removed.The optimum cone voltage was 25 V. Typical settings of capillary and HV lens voltages were 3.6 and 0.15 kV, respectively. For HPLC-MS-MS, collisionally induced dissociation (CID) was performed with argon collision gas at 5 3 1023 mbar and a collision energy of 35 eV. MS and MS-MS were employed concurrently, using selected ion monitoring (SIM) at the detector of MS-1, and selected reaction monitoring (SRM), respectively, for the 35Cl and 37Cl isotopes of the chlormequat cation (m/z 122.07 and 124.07), and the CID transitions of each of these precursors to the common product ion at m/z 58.The dwell time for each of the four channels was 0.2 s, the interchannel delay was 0.02 s and the mass span was 0.2. Quantification Quantification was based on peak area measurements made using Masslynx 2.22 software (Micromass, Manchester, UK).Data were smoothed with 2 passes of a moving average smoothing algorithm with an 8 point window. Peak baselines and peak assignments made by the software were checked and adjusted manually where necessary. For normal application of the method, matrix-matched calibration solutions at 5 or more concentrations were injected prior to analysis of samples, and again after a maximum of 15 sample injections. Quadratic calibration coefficients were calculated using data from the calibration analyses bracketing each group of samples.In addition, injections of a chlormequat chloride solution in methanol (0.02 mg ml21) were made at regular intervals during each batch of analyses. Extracts of samples found to contain more than 0.2 mg ml21 of chlormequat chloride (corresponding to a crop concentration, as cation, of 0.78 mg kg21) were re-injected after 10-fold dilution with an extract, prepared as above, of blank pears. Note: In this paper concentrations of solutions are given in terms of chlormequat chloride.Equivalent crop concentrations are given in terms of the chlormequat cation. Results and discussion Sample processing and extraction The techniques for sample processing and extraction were adapted without significant change from those described by Pasarela and Orloski,6 because of the extensive validation cited. Although these authors reported that chlormequat is minimally metabolised in plants, it has subsequently been noted that hydrolysis to choline can occur.1 Milling of frozen samples in dry ice should minimise the possibility of this occurring during sample processing. Mass spectrometry The positive ion ES mass spectrum of chlormequat (Fig. 1), introduced as a solution of the chloride, was characterised by ions at m/z 122 and m/z 124, corresponding to [35ClCH2CH2N(CH3)3]+ and [37ClCH2CH2N(CH3)3]+, respectively, and the virtual absence of fragment or adduct ions. CID 1012 Analyst, 1999, 124, 1011–1015of either ion (Fig. 2) gave [(CH3)3N]+ (m/z 59), [ClC2H4]+ (m/z 63 or 65), and the dimethylimmonium cation [(CH3)2NCH2]+; m/z 58) as previously reported for tetraalkylammonium salts.21 Although these product ions were obtained only in low yield with the instrument used, even after careful optimisation of the collision conditions, SRM of m/z 122 ? 58 and m/z 124 ? 58 afforded sufficient sensitivity for measurement at 0.01 mg ml21. As no clean-up was employed, the aqueous methanolic extracts contained substantial concentrations of co-extractives.With the standard ES interface fitted to the Micromass Quattro a total of up to 60 extracts (including matrix-matched calibration solutions and control samples) could be analysed before sensitivity loss necessitated source cleaning. Washing of the high voltage lens and sample cone successively with water and methanol, using a pipe cleaner to remove deposits from the apertures of the high voltage lens, was normally sufficient to restore sensitivity.We also evaluated the use of a Finnigan (Thermoquest, Hemel Hempstead, UK) Navigator single-quadrupole instrument for the determination of chlormequat in pear extracts. Although the general design of the standard ES source is very similar to that of the Quattro, we found that sensitivity fell noticeably with each injection of extract so that quantification was almost impossible. However, when a prototype ‘Matrixflow’ ES source was fitted to the instrument, a stable response was obtained for extended periods.22 This suggests that more modern ES designs using orthogonal ion extraction would be of considerable value.HPLC Since our earlier presentation,19 we have made a number of changes to the chromatographic conditions. The use of ammonium formate as a source of ammonium counter-ions for ion exchange HPLC, rather than the acetate previously employed, produced a sharper and more symmetrical chromatographic peak shape.An increase in the concentration of ammonium salt, from 0.05 to 0.2 M, allowed a reduction in the HPLC flow rate from 0.3 to 0.2 ml min21 without excessively prolonging the retention time. The increased concentration of ammonium ions was not accompanied by any noticeable decrease in ES-MS sensitivity for chlormequat. On the other hand, the reduction in flow rate provided an improvement in ESMS sensitivity, allowing the injected volume to be reduced from 30 to 20 ml.The latter change allowed a larger number of samples to be analysed before cleaning of the mass spectrometer high-voltage lens and skimmer cone became necessary. With previously unused HPLC columns, operated under the chromatographic conditions described, chlormequat gave a retention time of approximately 9 min. The retention time decreased gradually with use of the column in this application, but this did not affect the performance of the method as long as the peak shape and separation from potential interferences remained adequate, as discussed below.To extend the operating life of the column the ammonium formate concentration in the mobile phase was reduced after a period of column use, thus increasing the retention time of chlormequat and restoring separation from potential interferences. Several hundred injections could be performed before column replacement became necessary. Interferences and limit of detection Most pear samples analysed produced potentially interfering peaks at m/z 122, as is evident in Fig. 3, and, to a lesser extent, at m/z 124, although these were significant mainly for the determination of relatively low concentrations. Interferences at m/z 122 were considerably diminished by MS-MS (m/z 122 ? 58), and no interference was observed with m/z 124 ? 58, The latter, however, afforded the least sensitivity and therefore m/z 122 ? 58 was adopted as the primary quantification channel. Typically, analysis of pear extracts containing chlormequat chloride at 0.004 mg ml21 (0.016 mg cation kg21) gave S/N for Fig. 1 Positive ion ES mass spectrum of chlormequat. Fig. 2 MS-MS product ion spectra of chlormequat from (top) m/z 122 and (bottom) m/z 124. Fig. 3 HPLC-MS and MS-MS chromatograms (unsmoothed) from analysis of blank pear extract. Traces are (top to bottom) m/z 122, 124, 122 ? 58 and 124 ? 58. The arrows mark the retention time of chlormequat. Analyst, 1999, 124, 1011–1015 1013m/z 122, 124, 122 ? 58 and 124 ? 58, respectively, of approximately 20, 6, 2 and 1 without smoothing (Fig. 4). The S/N for m/z 124? 58 improved to approximately 3 after data smoothing (Fig. 5), which was used routinely. At a concentration of 0.01 mg ml21 (0.04 mg cation kg21) chlormequat could be reliably detected and quantified in all four channels. At m/z 124 ion suppression was also frequently apparent, manifested as a dip in the baseline of the chromatogram prior to elution of chlormequat, as seen in Figs. 3 and 4. The reduced retention that accompanied extended column use tended to reduce the separation between this negative interference and the chlormequat peak, and maintenance of baseline separation was a key criterion for reduction of the ammonium formate concentration, or replacement of the column. Fig. 6 shows a chromatogram from a well used column, illustrating the least separation that we judged to be acceptable. Calibration Matrix-matched solutions were used for calibration, as recommended in EU guidelines.23 A typical calibration graph for chlormequat in pear extract, using the m/z 122 ?58 channel, is shown in Fig. 7; similar calibration graphs were obtained from the other channels monitored. Below about 0.2 mg ml21 (0.78 mg cation kg21) response was a linear function of concentration, but at greater concentrations curvature was evident, giving a decrease in slope with increasing concentration. Consequently, to improve the accuracy of determination, extracts of samples found to contain more than 0.78 mg cation kg21 were routinely re-injected after 10-fold dilution in blank pear extract.As noted above, suppression of the signal was often evident in chromatograms for m/z 124. Using a column and an ammonium formate concentration that gave a separation between this region and the chlormequat peak that was the minimum we would accept for normal use, as shown in Fig. 6, we compared the peak areas produced by matrix-matched and solvent-based calibration solutions, keeping the solutions paired but injecting in random concentration order.Over the concentration range 0.02–0.2 mg ml21 the mean of the ratio of the peak area from matrix-matched solutions to those obtained from methanolic solutions was 1.00 (RSD 18%) for m/z 122, 0.98 (15%) for m/z 124, 0.95 (14%) for m/z 122 ? 58, and 0.97 (13%) for m/z 124 ? 58, showing that the accuracy of determinations was not compromised with chromatographic performance as illustrated.Fig. 4 HPLC-MS and MS-MS chromatograms (unsmoothed) from analysis of pear extract spiked at an equivalent of 0.016 mg cation kg21. Traces are (top to bottom) m/z 122, 124, 122 ? 58 and 124 ? 58. The arrows mark the chlormequat peak. Fig. 5 Smoothed (2 passes of 8 point moving average) HPLC-MS and MS-MS chromatograms from analysis of pear extract spiked at equivalent of 0.016 mg cation kg21. Traces are (top to bottom) m/z 122, 124, 122 ? 58 and 124 ? 58.The arrows mark the chlormequat peak. Fig. 6 HPLC-MS chromatogram (m/z 124) from analysis of pear extract at 0.016 mg cation kg21 using heavily used column. The arrow marks the chlormequat peak. Fig. 7 Calibration curve for determination of chlormequat in pear extracts using m/z 122 ? 58. 1014 Analyst, 1999, 124, 1011–1015Accuracy and precision The measured recovery from pears spiked with chlormequat chloride at 0.02 and 0.2 mg kg21 (0.016 and 0.16 mg cation kg21) prior to extraction are presented in Table 1.Repeatability data were each obtained on a single occasion, whereas internal reproducibility data were obtained during routine application of the method over a 3 month period. The mean recovery under repeatability conditions was approximately 80% at both concentrations. During routine use of the method over a 3 month period the mean recovery was 88% at 0.16 mg cation kg21. We consider the precision, under both repeatability (about 7–15% RSD) and internal reproducibility (about 15% RSD) conditions, to be typical of determinations based on electrospray HPLCMS, and acceptable for routine monitoring purposes.Data from analysis of the in-house reference material under internal reproducibility conditions are presented in Table 2. The accuracy of analysis is unknown but the precision was 10% RSD. Confirmation of residues The presence and ratios of the four signals measured provide the basis for qualitative and quantitative confirmation of residues.The ratios of concentrations determined from the four separately calibrated signals were highly consistent, with a mean RSD between channels of 3% (range 1–7%), and provide good confidence in both identity and quantity of residues. Application to other matrices The complete method has also been applied successfully to apples and to oriental pears, although we have not validated performance for these matrices in such detail. HPLC-MS-MS also appears to be suitable for the determination of chlormequat in a variety of fruit juices and fruit juice concentrates, following dilution with methanol and filtration.Acknowledgements We thank the UK Pesticides Safety Directorate for funding and S. Brewin, E. T. Griffiths, G. A. Keenan, R. E. Oliver, E. Patel, and E. Roberts for their contributions to the development and application of this method. References 1 C. Tomlin, The Pesticide Manual, British Crop Protection Council, Farnham, and The Royal Society of Chemistry, Cambridge, 11th edn., 1997, pp. 220–222. 2 R. Whitehead, The UK Pesticide Guide 1999, British Crop Protection Council, Farnham, and CABI Publishing, Wallingford, 1999, pp. 272–280. 3 R. P. Mooney and N. R. Pasarela, J. Agric. Food Chem., 1967, 15, 989. 4 J. Jung and G. Henjes, Z. Pflanzenernaehr Bodenkd., 1969, 124, 97. 5 J. Sachse, Z. Lebensm.–Unters.-Forsch., 1977, 163, 274. 6 N. R. Pasarela and E. J.Orloski, in Analytical Methods for Pesticides and Plant Growth Regulators, ed. G. Zweig and J. Sherma, Academic Press, New York, 1978, vol. VII, pp. 523–544. 7 J. Jung and G. Henjes, Z. Pflanzenernaehr Dueng. Boedenkd., 1964, 106, 108. 8 G. Puchwein, G. Schmidinger, S. Hain and D. Kruetzen, Z. Lebensm.–Unters. Forsch., 1979, 169, 339. 9 T. Stijve, Dtsch. Lebensm.-Rundsch., 1980, 76, 234. 10 F. Tafuri, M. Businelli and P. L. Giusquiani, Analyst, 1970, 95, 675. 11 F. Tafuri, M.Businelli, L. Scarponi and P. L. Giusquiani, J. Agric. Food Chem., 1970, 18, 869. 12 W. J. Allender, Pestic. Sci., 1992, 35, 265. 13 R. D. Mortimer and D. F. Weber, Pestic. Sci., 1994, 40, 31. 14 K. Pfeilsticker and F. Marx, Getreide Mehl Brot, 1978, 32, 268. 15 P. A. Greve and E. A. Hogendoorn, Med. Fac. Landbouww. Rijksuniv. Gent, 1987, 52, 695. 16 D. Barcelo, G. Durand and R. J. Vreeken, J. Chromatogr., 1993, 647, 271. 17 E. Moyano, D. E. Games and M. T. Galceran, Rapid Commun.Mass Spectrom., 1996, 10, 1379. 18 D. Wycherley, M. E. Rose, K. Giles, T. M. Hutton and D. A. Rimmer, J. Chromatogr. A, 1996, 734, 339. 19 P. Kebarle and L. Tang, Anal. Chem., 1993, 65, 972A. 20 A. R. C. Hill, S. Brewin, G. Keenan and R. E. Oliver, 1st European Pesticide Residue Workshop Program, Inspectorate for Health Protection, Alkmaar, The Netherlands, 1996. 21 G. J. C. Paul, I. Marcotte, J. Anastassopoulou, T. Theophanides, M. Arkas, C. M. Paleos and M. J. Bertrand, J. Mass Spectrom., 1996, 31, 95. 22 J. R. Startin, S. Hird, S. Brewin, A. Hill, A. Jones, S. Bajic, N. J. Loftus and D. Little, presented at 45th ASMS Conference on Mass Spectrometry and Allied Topics, Palm Springs, 2–6 June, 1997. 23 A. R. C. Hill, Quality Control Procedures for Pesticide Residue Analysis - Guidelines for Residues Monitoring in the European Union, European Commission, Brussels, 1997, Document 7826/VI/97. Paper 9/02712B Table 1 Recovery of chlormequat from pears under repeatability and internal reproducibilitya conditions m/z 122 m/z 124 m/z 122 ? 58 m/z 124 ? 58 Replication conditions Concentration/ mg cation kg21 n Mean recovery (%) RSD (%) Mean recovery (%) RSD (%) Mean recovery (%) RSD (%) Mean recovery (%) RSD (%) Repeatability 0.16 7 83.9 7 85.1 9 85.6 7 85.9 8 Repeatability 0.16 6 81.3 10 81.7 7 76.1 12 79.2 12 Repeatability 0.016 7 57.1 13 59.3 10 55.7 12 —a —a Repeatability 0.016 7 102.1 6 99.3 7 100.7 6 —a —a Internal reproducibilityb 0.16 38 88.7 14 89.3 15 88.2 15 88.0 15 a Below limit of determination. b Internal reproducibility data derive from 21 batches, analysed over a 3-month period, in which recovery was determined either singly or in duplicate. Table 2 Mean (n = 21) concentration of chlormequat cation determined in in-house reference materiala m/z Concentration measured/mg kg21 RSD (%) 122 0.143 10 124 0.143 10 122 ? 58 0.144 10 124 ? 58 0.144 9 a The reference material was produced in-house by homogenisation of a sample of pears found to contain chlormequat. In each of the 21 batches analysed over the 3 month period, 1 sub-sample of reference material was analysed. Analyst, 1999, 124, 1011–1015 1015

ISSN:0003-2654    

DOI:10.1039/a902712b

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

Determination of ammonia and aliphatic amines in environmental aqueous samples utilizing pre-column derivatization to their phenylthioureas and high performance liquid chromatography Bhushan Sahasrabuddhey, Archana Jain and Krishna K. Verma* Department of Chemistry, Rani Durgavati University, Jabalpur 482001, Madhya Pradesh, India Received 30th March 1999, Accepted 18th May 1999 Pre-column conversion of ammonia and a number of aliphatic amines into phenylthiourea or its derivatives by reaction with phenyl isothiocyanate, followed by HPLC, has been used for their determination in environmental waters.Optimum conversion was found when the reaction was carried out in sodium hydrogencarbonate– carbonate medium at 40 °C for 15 min. Well separated peaks were obtained on a C18 column with an acetonitrile–water gradient (1 ml min21) of 30% acetonitrile for an initial 5 min which was increased linearly to 100% over 15 min and then maintained isocratic for 5 min, the acetonitrile ratio finally being returned to 30% in 5 min.The derivatized analytes were subjected to off-line solid phase extraction on C18 sorbent. A linear calibration graph was obtained for 0.01–10 mg l21 analytes with a correlation coefficient of 0.9954 for ammonia and in the range 0.9982–0.9996 for amines. The limit of detection for ammonia was 0.2 mg l21 and for amines in the range 0.3–0.6 mg l21. The method was applied to tap, underground, river and aquarium waters, the recovery being in the range 97–106% (RSD 1.8–4.5%).Many of the samples were found to contain more than the permissible limit of ammonia. Phenyl isothiocyanate is stable for long periods in aqueous medium over wide ranges of pH and temperature, and the resulting phenylthioureas have adequate retention on C18 sorbent and strong UV absorption, making this reagent suitable for the determination of amines in water. Most aquatic species excrete ammonia and urea, the rate of excretion and level of efflux being dependent upon the individual physiological status of the species and the modifying influences of their environment.1 Ammonia is more toxic and water soluble than urea and this has important implications for the successful husbandry of commercially important, cultured aquatic species, or any that are held in fixed volumes of water.The quantitative determination of the principal metabolite endproducts excreted to the external medium are therefore important analytical procedures.Short-chain aliphatic amines are emitted into atmosphere from anthropogenic sources such as cattle feedlot operations, waste incineration, sewage treatment and various industries.2–4 Amines are also emitted in car exhausts.5 A natural background level of aliphatic amines can also be assumed to originate from animal wastes and microbiological activities.3 Aliphatic amines are industrial chemicals with a wide range of applications. They are used as raw materials or as chemical intermediates in the production of other chemicals, pharmaceuticals, polymers, pesticides, dyestuffs and corrosion inhibitors.Aliphatic amines and polyamines are well known as odorous substances and as precursors of N-nitrosamines, which are carcinogenic.6,7 Dimethylamine is present in untreated waste water discharges from aramide polymer manufacturing facilities, where it is produced by the decomposition of N,N-dimethylacetamide in the solvent stripping step.Environmental protection authorities demand analytical monitoring of unconverted dimethylamine in the aquatic environment close to waste treatment facilities. The monitoring of alkylamines is of considerable interest as most of them are toxic, sensitizers of and irritants to the skin, mucous membrane and respiratory tract, through all routes of exposure, i.e., inhalation, ingestion and contact. The American Conference of Government Industrial Hygienists (ACGIH) has adopted threshold limit values–time weighted average (TLV– TWA) in the range 5–10 mg l21 for various alkylamines and 0.5 mg l21 for ammonia.8 Most primary and secondary amines exhibit poor chromatographic performance via direct HPLC approaches, making quantitative analysis difficult.9 Methods for their determination require a high degree of specificity and sensitivity as they do not exhibit any structural feature that allows detection without derivatization.All existing liquid chromatographic methods for amine determination require at least two steps, separation from potential interferents in the sample and pre- or post-column formation of derivatives with better detectability.Chemical derivatization in solution has long been accepted as an effective modification technique in HPLC, improving the overall specificity, chromatographic performance and sensitivity for trace analysis.10–12 The diverse reagents and conditions for derivatization of ammonia and aliphatic amines, as available in the literature, are summarized in Table 1.With secondary amines there is no reaction of o-phthalaldehyde, fluorescamine gives a non-fluorescent product and Lumarin 1 has a long reaction time. Many reagents require high temperatures and a prolonged period for derivatization, and the volatility of aliphatic amines may require special handling. Phenyl isothiocyanate has been utilized for the determination of dimethylamine in waste waters but the method was reported to show poor linearity and a poor limit of detection.29 In this work, phenyl isothiocyanate was used for the determination of ammonia and a number of aliphatic amines in environmental waters, involving their conversion into phenylthioureas and HPLC.Phenyl isothiocyanate is stable for long periods in aqueous medium over a wide range of pH and temperature and the resulting phenylthioureas have adequate retention on C18 sorbent and strong UV absorption, making this reagent suitable for the determination of amines in water.Analyst, 1999, 124, 1017–1021 1017Experimental Equipment The chromatographic system consisted of a Beckman System Gold 127 binary gradient pump and Model 166 UV-Vis spectrophotometric detector (8 ml flow-through cell) (Beckman, Fullerton, CA, USA). Detection was carried out at 240 nm. A Rheodyne Model 7010 valve (Alltech, Deerfield, IL, USA) equipped with a 10 ml sample loop was used for sample injection.Data processing was carried out with an HP 3395 integrator (Hewlett-Packard, Palo Alto, CA, USA). The analytical column was 25 cm 3 4.6 mm id ODS2 (5 mm particle size) (Anachem, Luton, UK). Solid-phase extraction cartridges (2.8 ml) containing 500 mg of C18 sorbent were obtained from Alltech. Before analysis, all environmental aqueous samples were filtered through a 0.45 mm membrane filter (Millipore- India, Mumbai, India). Quantification was effected by measuring both peak height and area; peak height measurements gave better results.Reagents and standard solutions Phenyl isothiocyanate (PITC) was obtained from Merck (Darmstadt, Germany) and a standard solution was prepared by dissolving 4 ml of PITC in 100 ml of acetonitrile. Acetonitrile and HPLC-grade water were obtained from Merck (Mumbai, India). Stock standard solutions (1000 mg l21), of ammonia (as ammonium chloride; Qualigens, Mumbai, India), methylamine, ethylamine (BDH, Poole, Dorset, UK), dimethylamine, isopropylamine, and diethylamine (Merck) were prepared in methanol, and standardized by titration with acid or by the dithiocarbamic acid formation method.30 The stock standard solutions were stored refrigerated when not in use. Test samples were prepared freshly by spiking with known aliquots of suitably diluted stock standard solution before analysis.Solutions of sodium hydrogencarbonate and sodium carbonate (5%) were prepared in water.Mobile phase and HPLC gradient programme Acetonitrile–water at a flow rate of 1 ml min21 was used for elution. The optimum gradient programme consisted of an initial 30% acetonitrile for 5 min that was increased linearly to 100% over 15 min and maintained isocratic for 5 min. Finally, the acetonitrile concentration was returned to 30% in 5 min. Table 1 Conditions for the derivatization of ammonia and aliphatic amines Reagenta NH3/amineb Time Temperature/°C Mediumc Ref.OPA 1° 1 min 18 pH 9.5 13,14 2° OPA does not react with 2° amines 14 1° On-line 40 pH 9 15 1° Post-column 23 pH 10.5 16 FMOC 1°/2° 40 s 18 pH 7.7 10 FMOC NH3 2 min 61 pH 6.8 17 FMOC 2° 2 min 18 pH 8.0 18 FMOC-tagged silica 1°/2° 10–15 min 60 MeCN/Py 19 Polymer activated FMOC 1°/2° On-line 60 pH 10 20 Dansyl chloride 1°/2° 10 min 40 pH 8.5 13 4-Chloro-7-nitrobenzo-1,2,5-oxadiazole 1°/2° 60 min 55 pH 8.5 13 Lumarin 1 1° 20 min 50 THF–DMSO 21 2° 180 min 70 THF–DMSO 21 Fluorescamine 1° 5–30 min 18 pH 8–8.5 10 1° 1 min 18 pH 10 22 2° Non-fluorescent derivative is formed 14 3-Toluoyl chloride 1°/2° 10 min 18 MeCN–NaOH 23 2-Naphthyloxycarbonyl chloride 1°/2° 3 min 18 pH 9 24 8-Quinolinesulfonyl chloride 1°/2° 20 min 65 pH 8.5 25 1-Fluoro-2,4-dinitrobenzene 1°/2° ? 20 pH 10.5 26 1-Naphthyl isocyanate 1°/2° Immediate Ice-bath Hexane 27 1-Naphthyl isothiocyanate 1°/2° Exposure of impregnated reagent to air 28 Phenyl isothiocyanate 1°/2° 15 min 40 pH 8.5 This work a OPA = o-phthalaldehyde; FMOC = 9-fluorenylmethyl chloroformate. b Amines: 1° = primary; 2° = secondary.c THF = tetrahydrofuran; DMSO = dimethyl sulfoxide; MeCN = acetonitrile; Py = pyridine. Table 2 Calibration and other statistical data for the determination of ammonia and aliphatic amines (range 0.01–10 mg l21 each) after derivatization with phenyl isothiocyanate Analyte ra Intercept (IU)b Slope (IU)b Conversionc (%) LODd/ mg l21 Recovery (%) SPEe Ammonia 0.9874 6564 514035 98.7 0.2 80 Methylamine 0.9994 286 1144255 — 0.3 90 Dimethylamine 0.9992 2257 436408 — 0.6 93 Ethylamine 0.9996 299 790337 99.0 0.4 88 Isopropylamine 0.9982 3786 801363 — 0.5 94 Diethylamine 0.9988 5530 304008 99.6 0.5 71 a Average of six replicate analyses.b Integrator units; AFS = 0.008. c Conversion to thiourea derivative takes into account the peak area produced by equimolar amount of corresponding authentic compound; a dash indicates that conversion was not determined.d LOD = limit of detection.31 e Average of three replicate analyses. Recovery of analytes from 25 ml of derivatization mixture after SPE on C18 sorbent. 1018 Analyst, 1999, 124, 1017–1021All compounds were eluted within 20 min. A 20 min flushing of the column with 30% acetonitrile is recommended before the next injection. Sampling The performance of the method was tested with unspiked and spiked domestic, surface and river water samples. Once environmental water samples had been collected, they were immediately acidified with 1 ml of 0.1 M hydrochloric acid to avoid volatilization of amines and filtered through a 0.45 mm membrane filter.A 4 ml portion of real water sample, unspiked and spiked with amine standard, was subjected to derivatization and analysis by HPLC. Derivatization procedure and analysis A 0.1–2 ml aliquot of sample solution was mixed with 0.5 ml of 4% phenyl isothiocyanate and 0.5 ml of 5% sodium hydrogencarbonate in a 10 ml calibrated flask and the flask was capped, shaken well, and heated at 40 °C in a water-bath for 10 min.Then, 0.5 ml of 5% sodium carbonate solution was added and the flask again heated at the same temperature for 5 min. Subsequently, one of the following methods was used. (1) The contents were cooled to room temperature, diluted to the mark with acetonitrile–water (30 + 70 v/v) and a 10 ml aliquot of derivatized amine mixture was injected into the chromatographic system. (2) The derivatized amine solution, after cooling to room temperature, was passed through a C18 cartridge that had previously been activated with 2 ml of acetonitrile and equilibrated with 2 ml of de-ionized, distilled water.The sorbent was washed with 1 ml of distilled water and the retained derivatives were eluted with 2 ml of acetonitrile. A 10 ml aliquot of eluate was injected into the liquid chromatograph. Fig. 1 Chromatogram obtained for (A) standard solution (1 mg l21) of ammonia and five aliphatic amines derivatized with phenyl isothiocyanate and (B) reagent blank.Peaks (as phenylthiourea and its derivatives): 1 = ammonia; 2 = methylamine; 3 = dimethylamine; 4 = ethylamine; 5 = isopropylamine; 6 = diethylamine; R = phenyl isothiocyanate reagent; I = unknown impurity. Column, C18, 25 cm 3 4.6 mm id (5 mm particle size); detection wavelength, 240 nm; mobile phase, acetonitrile–water, gradient elution; absorbance full-scale (AFS), 0.05; flow rate, 1 ml min21.Fig. 2 Chromatograms obtained for aliphatic amines spiked at the 2 mg l21 level in two river water samples: Narmada water, (A) spiked and (B) unspiked; and Ganga water, (C) spiked and (D) unspiked. C18 column, 25 cm 3 4.6 mm id (5 mm particle size); detection wavelength, 240 nm; mobile phase, acetonitrile–water, gradient elution; AFS, 0.05; flow rate, 1 ml min21. Peak designation as for Fig. 1. Analyst, 1999, 124, 1017–1021 1019Results and discussion Optimization of the chromatographic separation Conventional HPLC with aqueous organic mobile phases was initially tested for the separation of ammonia and different amine derivatives.Isocratic elutions were unsuccessful because the peaks remained almost unresolved and the reagent peak eluted 10 min after all peaks of derivatives. The latter problem resulted in an unnecessarily long analysis time. Studies were performed with different ratios of methanol–water and acetonitrile –water as eluents but without success.Adjustment of the mobile phase pH also did not improve the resolution. These observations indicated the necessity for gradient elution. Well defined, sharp and reproducible peaks were obtained when acetonitrile–water was used as the mobile phase, the flow rate being 1.0 ml min21. For the initial 5 min of the chromatographic run, acetonitrile–water (30 + 70, v/v) was used, to acetonitrile concentration being increased linearly to 100% over 15 min and then maintained isocratic for another 5 min.Finally, the acetonitrile concentration was returned to 30% in 5 min, and the column was flushed for about 20 min before the next injection. The detector was set at 240 nm as at this wavelength all thioureas absorbed strongly. Under these optimum conditions of HPLC, all peaks were baseline separated (Fig. 1). Amino acids are also known to react with phenyl isothiocyanate; 10 however, glycine and alanine did not produce any interfering peaks in the working chromatogram for amines.Optimization of derivatization reaction In acidic solutions, the derivatization reaction was incomplete since the protonated amines are only weak nucleophiles towards their addition to the thiocyanato group of the derivatizing agent. To increase the percentage conversion, attempts were made to carry the reaction at higher temperatures and in the presence of buffering agents. All amines responded to higher conversion in sodium acetate or hydrogencarbonate medium but reaction with ammonia was still incomplete even after allowing the reaction mixture to stand for 1 h at room temperature.Reaction in the presence of sodium hydrogencarbonate at elevated temperatures (the range tested was 35–60 °C) for 10 min served to increase the conversion for all amines but the effect levelled off at 40 °C except for ammonia, which had an optimum peak height at 60 °C. The effect was almost the same in sodium carbonate for all amines but ammonia showed a significantly different behaviour, a lower levelling off temperature (40 °C), and poor precision (RSD 5–10%).In sodium carbonate medium a pale yellow colour developed during heating and some additional peaks appeared especially close to the derivative of dimethylamine and with which it merged at lower amine concentrations. It appeared that all amines had optimum conversion in hydrogencarbonate but ammonia required a higher pH (carbonate medium).Optimum peak heights and the best precision were obtained when the derivatization reaction was carried out at 40–45 °C first in the presence of hydrogencarbonate for 10 min, and then sodium carbonate for an additional 5 min. Under these conditions, the reaction mixture Table 3 Determination of ammonia and aliphatic amines in real samples Ammonia Recovery of 2 mg l21 spike (%)b founda/ Sample mg l21 1 2 3 4 5 6 Narmada river water 0.17 98.2 104.8 105.6 100.9 97.9 96.9 Ganga river water 1.96 97.8 101.6 100.7 101.9 98.7 97.9 Underground water no. 1 0.53 104.4 103.9 101.2 100.7 97.9 97.6 Underground water no. 2 0.95 Tap water no. 1 0.42 106.0 107.9 103.9 104.4 100.0 97.8 Tap water no. 2 0.44 Aquarium water 1.01 101.6 99.3 98.4 99.0 102.5 103.0 Average RSD of recovery (%) 4.1 3.7 3.8 3.3 3.6 3.9 a The results are averages of three determinations; RSD = 1.2–3.5%. b The results are the average of three determinations; RSD = 1.8–4.5%. 1 = Ammonia; 2 = methylamine; 3 = dimethylamine; 4 = ethylamine; 5 = isopropylamine; 6 = diethylamine. The recovery for ammonia takes into account the concentration already present in the real sample. No aliphatic amine was found in any of the samples analysed. Fig. 3 Chromatograms obtained for aliphatic amines spiked at the 2 mg l21 level in different environmental water matrices. Laboratory tap water, (A) spiked and (B) unspiked; aquarium water, (C) spiked and (D) unspiked; underground water, (E) spiked and (F) unspiked; and city tap water, (G) spiked and (H) unspiked.C18 column, 25 cm 3 4.6 mm id (5 mm particle size); detection wavelength, 240 nm; mobile phase, acetonitrile–water, gradient elution; AFS, 0.05; flow rate, 1 ml min21; sample volume, 10 ml. Peak designation as for Fig. 1. 1020 Analyst, 1999, 124, 1017–1021did not turn yellow, nor were there unwanted additional peaks. The optimum concentrations of reagents for derivatization were 0.5 ml each of 5% sodium hydrogencarbonate, 5% sodium carbonate and 4% phenyl isothiocyanate.Calibration data Calibration graphs were obtained using a series of six standard solutions of ammonia and amines. Three replicate derivatizations at each concentration level were performed and their average response was plotted against the concentration of the corresponding analyte. Rectilinear calibration graphs were obtained over the range 0.01–10 mg l21 of analytes. Calibration and other statistical data for the determination of ammonia and aliphatic amines are given in Table 2.Analysis of real samples The method was validated by spiking natural samples with known amounts of ammonia and amines (range sub-mg l21 to low mg l21 level) and evaluating the recovery. All chromatographic peaks of interest were well separated from peaks due to extraneous matter. Typical chromatograms for a 2 mg l21 spike are given in Fig. 2 and 3. Ammonia was found in all samples but none of the samples analysed showed the presence of any amine.The results for real water samples are presented in Table 3. From the recovery results we observe that there is no significant matrix effect and the recoveries are within acceptable limits. Hence, this method can be used for real water samples. Narmada river water and Jabalpur city tap water showed acceptable ammonia levels, but Ganga water (collected from Kanpur) and both underground (well) waters of Jabalpur city exceeded the prescribed limit for ammonia.The use of untreated river and underground (well) waters for drinking purposes is a common practice in India, and therefore caution is advisable. Conclusions Conversion of ammonia and aliphatic amines into their corresponding thiourea derivatives followed by HPLC with gradient elution with acetonitrile–water and UV detection at 240 nm is an elegant method for their determination in environmental waters. In comparison with other reagents available for the synthesis of derivatives of amines, phenyl isothiocyanate is a fast reacting and stable reagent over a wide range of pH, and it can react with both primary and secondary amines.Sample clean-up and analyte enrichment by solid-phase extraction on C18 sorbent were feasible, and this technique has potential for still better detection when used on-line with HPLC. Although not tested, the reducing property of the thiourea group makes possible the detection of derivatives by an electrochemical method.We gratefully acknowledge financial support for this research by the European Union (grant No. Cl1*-CT94-0049). B.S. thanks the Council of Scientific and Industrial Research, New Delhi, for a senior research fellowship. References 1 M. J. Dagg, Int. Rev. Gesamt. Hydrobiol., 1976, 61, 267. 2 S. E. Mnahan, Environmental Chemistry, Lewis, Chelsea, MI, 4th edn., 1990, p. 422. 3 B. J. Finlayson-Pitts and J. N. Pitts, Jr., Atmospheric Chemistry, Wiley-Interscience, New York, 1986, p. 561. 4 S. Fuselli, S. Cerquiglini and E. Chiacchierini, Chim. Ind. (Milan), 1978, 60, 711. 5 S. H. Cadle and P. A. Mulawa, Environ. Sci. Technol., 1980, 14, 718. 6 P. Simon and C. Lemacon. Anal. Chem., 1987, 59, 480. 7 Kirk-Othmer Encyclopedia of Chemical Technology, Wiley-Interscience, New York, 3rd edn., 1978, vol. 2, pp. 272–283. 8 Threshold Limit Values and Biological Exposure Indices for 1988–1989, American Conference of Government Industrial Hygienists, Cincinnati, OH, 1988. 9 R.Wills and J. Silalahi, J. Liq. Chromatogr., 1987, 10, 3183. 10 N. Seiler, Handbook of Derivatives for Chromatography, ed. K. Blau and J. Halket, Wiley, Chichester, 2nd edn., 1993, pp. 175–213. 11 J. W. Lawrence and R. W. Frei, Chemical Derivatization in Liquid Chromatography, Elsevier, Amsterdam, 1976. 12 H. Lingeman and W. J. M. Underberg, Detection-Oriented Derivatization Techniques in Liquid Chromatography, Marcel Dekker, New York, 1990. 13 G. Mellbin and B. E. F. Smith, J. Chromatogr., 1984, 312, 203. 14 K. Imai, T. Toyo’oka and H. Miyano, Analyst, 1984, 109, 1365. 15 O. Busto, M. Miracle, J. Guasch and F. Borrull, J. Chromatogr. A, 1997, 757, 311. 16 S. R. Vale and M. B. A. Gloria, J. AOAC Int., 1997, 80, 1006. 17 S. S. Goyal, D. W. Rains and R. C. Huffaker, Anal. Chem., 1988, 60, 175. 18 M. R. Lopez, M. J. G. Alvarez, A. J. M. Ordieres and P. T. Blanco, J. Chromatogr., 1996, 721, 231. 19 H.-M. Zhang, F.-X. Zhou and I. S. Krull, J. Pharm. Biomed. Anal., 1992, 10, 577. 20 C. X. Gao, I. S. Krull and T. M. Trainor, J. Chromatogr., 1989, 463, 192. 21 H. Kouwatli, J. Chalom, M. Tod, R. Farinotti and G. Mahuzier, Anal. Chim. Acta, 1992, 266, 243. 22 Dj. Djozan and M. A. Faraj-Zadeh, J. High Resolut. Chromatogr., 1996, 19, 633. 23 P. Simon and C. Lemacon, Anal. Chem., 1987, 59, 480. 24 J. Kirschbaum, I. Busch and H. Bruckner, Chromatographia, 1997, 45, 263. 25 M. I. Saleh and F. W. Pok, J. Chromatogr. A, 1997, 763, 173. 26 F. A. L. Van Der Horst and J. J. M. Holthuis, J. Chromatogr., 1988, 426, 267. 27 K. Andersson, C. Hallgren, J.-O. Levin and C.-A. Nilsson, J. Chromatogr., 1984, 312, 482. 28 R. Lindahl, J.-O. Levin and K. Andersson, J. Chromatogr., 1993, 643, 35. 29 L. Lehotav and D. Oktavec, J. Liq. Chromatogr., 1992, 15, 307. 30 S. Siggia and J. G. Hanna, Functional Group Analysis, Wiley, New York, 4th edn., 1979, pp. 545 and 572. 31 J. C. Miller and J. N. Miller, Statistics for Analytical Chemistry, Ellis Horwood, Chichester, 3rd edn., 1993, p. 115. Paper 9/02587A Analyst, 1999, 124, 1017–1021 1021

ISSN:0003-2654    

DOI:10.1039/a902587a

出版商:RSC    

年代:1999

数据来源: RSC

摘要:

A method for the separation of residues of nine compounds in cattle liver related to treatment with oxfendazole Martin D. Rose CSL Food Science Laboratory, Norwich Research Park, Colney, Norwich, UK NR4 7UQ. E-mail: m.rose@csl.gov.uk Received 19th November 1998, Accepted 1st April 1999 A method for the determination of nine compounds closely related to oxfendazole has been developed for the monitoring of residues in food. The method is based on a multi-residue procedure for basic drug residues and used strong cation exchange solid phase extraction for sample clean-up.These nine compounds include fenbendazole, which is itself a licensed veterinary product. The pro-drug febantel converts quickly to fenbendazole or oxfendazole soon after administration. The method is therefore suitable for monitoring residues following the use of any of these compounds. Some of these analytes have been shown to be present as residues following the treatment of farm animals with oxfendazole.Average recoveries for the nine compounds from tissue fortified with 100 mg kg21 were between 34% and 96% with relative standard deviations between 3% and 22%. Anthelmintic agents, including benzimidazoles such as oxfendazole, are used in animal husbandry for the prevention and control of internal worm parasites. A number of these compounds have been shown to cause teratogenic and embryotoxic effects in some species.1 Their use with farm animals raises the possibility that residues may be found in food produced for human consumption.Maximum residue limits (MRLs) for these compounds are set in European legislation2 and range from 10 to 1000 mg kg21 depending upon the compound and food type. Details of these limits are summarised in Table 1. Several methods exist in the literature for the analysis of one or more benzimidazole compounds as residues in a variety of food types. For example, a method for the determination of fenbendazole, oxfendazole, thiabendazole and 5-hydroxythiabendazole in milk was described3 with a limit of detection at 5 mg kg21.This procedure was based on partition between organic and aqueous phases with pH adjustment followed by solid phase extraction (SPE) clean-up on silica. Butylated hydroxytoluene (BHT) was added during the extraction to prevent further oxidation of residues. A method for the determination of six benzimidazoles in sheep or chicken liver was described using sequential SPE clean-up, first loading an acidic alumina cartridge with a hexane–chloroform (25 + 75) extract, eluting with methanol, adding water and applying to a C18 SPE cartridge.The final eluant was acetonitrile.4 Matrix Table 1 MRLs for benzimidazole anthelmintics2 Pharmacologically Animal MRLs/ Target active substance(s) Marker residue species mg kg21 tissues Febante Sum of extractable residues Bovine, 500 Liver Fenbendazole and which may be oxidised to ovine, 50 Muscle, kidney, fat Oxfendazole oxfendazole sulfone porcine, 10 Milk Equidae Thiabendazole Sum of extractable residues that may be oxidised to ketotriclabendazole Bovine, ovine 100 Muscle, liver, kidney Thiabendazole Sum of thiabendazole and 5-hydroxythiabendazole Bovine 100 Muscle, liver, kidney, fat, milk Flubendazole Sum of flubendazole and Poultry and 400 Liver (2-amino-1H-benzimidazol- game birds, 50 Muscle, skin, fat 5-yl)(4 fluorophenyl) porcine 300 Kidney methanone Flubendazole Chicken 400 Eggs Netobimin Sum of netobimin and Bovine, 1000 Liver albendazole and metabolites ovine, caprine 500 Kidney of albendazole measured as 100 Muscle, fat 2-aminobenzimidazole 100 Milk sufone Oxibendazole Oxibendazole Porcine 100 Muscle, kidney 500 Skin, fat 200 Liver Albendazole Sum of albendazole, Bovine 1000 Liver sulfoxide albendazole sulfoxide, ovine, 500 Kidney albendazole sulfone, and pheasant 100 Muscle, fat, milk albendazole 2-amino sulfone, expressed as Bovine, ovine 100 Milk albendazole Analyst, 1999, 124, 1023–1026 1023S N NHCOCH2OCH3 NHCOOCH3 NHCOOCH3 S N NHCOCH2OCH3 NHCOOCH3 NHCOOCH3 O febantel febantel sulfoxide S N N NHCOOCH3 H S N N NH2 H S N N NHCOOCH3 H OH S N N NHCOOCH3 H S N N NH2 H S N N NHCOOCH3 H HO fenbendazole 4–hydroxyfenbendazole fenbendazole amine oxfendazole 4–hydroxyoxfendazole oxfendazole amine O O O SO2 N N NHCOOCH3 H SO2 N N NH2 H SO2 N N NHCOOCH3 H HO oxfendazole sulfone 4–hydroxyoxfendazole sulfone oxfendazole sulfone amine O solid phase dispersion has also been used as part of the analysis procedure for these compounds in cattle liver5 and milk.6 During a study on the effect of cooking on residues of oxfendazole in food, conducted in this laboratory,7 several metabolites and breakdown products were identified in the tissue.These arose from the treatment of farm animals with oxfendazole, and there was some evidence of the presence of other metabolites which were not characterised.There was also evidence of an ‘unstable equilibrium’ between oxfendazole, oxfendazole sulfone and fenbendazole in incurred tissue: an overall instability of these compounds in tissue during frozen storage, an uneven distribution of residues within the tissue and the possibility of an effect of protein binding on extractability of residues from tissue. None of the available methods in the literature was suitable for studying the interaction of the various metabolites and breakdown products found in this study.The structures and relationship of 11 compounds known to be part of the metabolic pathway of oxfendazole are shown in Fig. 1. The aim of the work described here was to develop a method capable of the determination of all 11 of these compounds, which could potentially be present as residues in cattle treated with oxfendazole, fenbendazole or febantel. Febantel is a pro-drug, i.e., a product known to convert into an active compound soon after administration.8 It is converted either directly to fenbendazole or to oxfendazole, which is arrived at via febantel sulfoxide as an intermediate. During preliminary investigations febantel and its sulfoxide were both found to be unstable as they converted to fenbendazole during the developed extraction and clean-up procedure.Because of their mode of action, residues of these compounds are unlikely to be found in animal tissue. Methodology which excluded these compounds was therefore deemed acceptable.A method for the remaining nine compounds in raw and cooked tissue at normal residue levels was developed and validated. The approach was based on the multi-residue procedure for basic drugs developed in this laboratory.9 Experimental Standards Analytical standards for all compounds identified as potential metabolites and breakdown products were either purchased, given by manufacturers or synthesised at the University of East Anglia (UEA).The origins of the range of standards are shown in Table 2. Standard purity checks were performed on each compound with satisfactory results. These included melting point, 1H NMR (270 MHz), 13C NMR (270 MHz), IR, TLC, MS and elemental analysis. Principle Tissue was extracted with acetonitrile. The extract was dried with sodium sulfate and acidified with acetic acid before loading onto a conditioned strong cation exchange (SCX) solid phase extraction (SPE) cartridge.The cartridge was washed successively with acetone, methanol and finally acetonitrile Fig. 1 1024 Analyst, 1999, 124, 1023–1026before elution with acetonitrile–35% aqueous ammonia (95 + 5). Reagents All chemicals were of analytical grade. Solvents were HPLC or glass-distilled grade. Water was obtained from an in-house Elga water purification system. Extraction and clean-up Finely sliced liver tissue (5 g) was weighed into a polypropylene centrifuge tube (100 ml) and homogenised for 30 s in acetonitrile (50 ml).Sodium sulfate (5g) was added and the tubes were centrifuged (1700g, 5 min). A Bond-Elut SCX cartridge (500 mg per 3 ml; Varian, Walton-on-Thames, Surrey, UK) was conditioned with acetonitrile–glacial acetic acid (95 + 5, 5 ml) and a reservoir (100 ml) fitted. The sample extract was acidified with glacial acetic acid (5 ml), transferred to the reservoir and passed through the cartridge. The cartridge was washed sequentially with acetone (2.5 ml), methanol (5 ml) and acetonitrile (5ml).The analytes were eluted with acetonitrile– 35% aqueous ammmonia (95 + 5, 5 ml) into a test-tube. HPLC conditions A gradient HPLC system was required to separate all nine compounds in a convenient run time. The column eluate was monitored for UV absorbance at 290 nm. The flow rate was constant at 0.45 ml min21. A Chromspher 5C8 25 cm 3 3 mm id column was used. The gradient profile is shown in Table 3. A flow diagram of the method is shown in Fig. 2. Validation protocol Tissue for use as a control was purchased locally and analysed to verify the absence of residues. Batches consisting of at least one blank control sample and at least four fortified blank samples were analysed on three separate days. The solution Table 2 Oxfendazole related compounds and their source Abbreviation Procured from (i) Febantel FBT Bayer (ii) Febantel sulfoxide FBT-SO UEA (iii) 4-Hydroxyfenendazole OH-FEN UEA (iv) Fenbendazole FEN Sigma (v) Fenbendazole amine FEN-A UEA (vi) 4-Hydroxyoxfendazole OH-OXF UEA (vii) Oxfendazole OXF Syntex (viii) Oxfendazole amine OXF-A UEA (ix) 4-Hydroxyoxfendazole sulfone OH-OXF-S UEA (x) Oxfendazole sulfone OXF-S Syntex (xi) Oxfendazole sulfone amine OXF-S-A UEA Table 3 Gradient elution profile Time/min A (%)a B (%)a 0 75 25 18 60 40 23 0 100 30 0 100 31 75 25 a Mobile phase A: 0.1 m ammonium carbonate–methanol (80 + 20); mobile phase B: 0.1 m ammonium carbonate–methanol (20 + 80).Fig. 2 Fig. 3 Analyst, 1999, 124, 1023–1026 1025used for fortification was added to the tissue and allowed to equilibrate for 5–10 min prior to the addition of the acetonitrile used for extraction. Linearity of standards was checked for the range equivalent to 100–1000 mg kg21. Results and discussion Although there is some chemical similarity between the analytes, there is a wide range of polarity between the first and last eluting compounds. Their chemical similarity (they are all basic compounds) was exploited by applying cation exchange solid phase extraction for clean-up and variation in polarity was used to separate these compounds on a gradient HPLC system.None of the metabolites measured were found to co-elute with other benzimidazole compounds tested (thiabendazole, albendazole sulfone, cambendazole) using these HPLC conditions. The method developed separates all nine compounds associated with oxfendazole that are likely to be found as residues in food.Chromatograms of a standard mixture, a blank sample extract and an extract from a sample fortified at 100 mg kg21 are shown in Fig. 3. Thiabendazole was incorporated as an internal standard. Because of the variation in recovery between analytes, external standards were used for recovery calculations for individual compounds. Validation data for the method on three separate days for samples fortified at 100 mg kg21 are shown in Table 4. Injections of standards demonstrated that the response was linear over a range 100–1000 mg kg21 with correlation coefficients 40.98.Normal quality targets in this laboratory for recovery and CV of analytes determined by a method are to give an average recovery greater than 40% with an RSD of less than 20%. These values were generally achieved with this method, although there was some compromise with some analytes, notably fenbendazole amine, where recoveries less than the target were permitted in order to allow the same method to be used for all compounds. This compromise was allowed because of the versatility of the multi-analyte method.Conclusion A multi-residue method for the determination of residues of compounds associated with oxfendazole treatment of farm animals has been developed. The same residues are likely to be associated with the treatment of farm animals with fenbendazole and the pro-drug febantel due to the reaction pathways and metabolism of these compounds.The method is suitable for application to surveillance programmes to detect residues arising from the use of these compounds. Acknowledgement This work was funded by the Veterinary Medicines Directorate of the UK Ministry of Agriculture, Fisheries and Food. References 1 P. Delatour and R. Parish, in Drug Residues in Animals, ed. A. G. Rico, Academic Press, New York, USA, 1986, pp. 175–192. 2 Council Regulation (EEC) No. 2377/90, Off. J., L 224, 1990 (as amended). 3 S. S.-C. Tai, N. Cargile and C. J. Barnes, J. Ass. Off. Anal Chem., 1990, 73, 368. 4 W. H. H. Farrington, S. Chapman and D. Tyler, in Proceedings of Euroresidues Conference on Residues of Veterinary Drugs in Food, ed. N. Haagsma, A. Ruiter and P. B. Czedik-Eysenberg, University of Utrecht, The Netherlands, 1990, pp. 185-188. 5 A. R. Long, M. S. Marlbrough, L. C. Hsieh, C. R. Short and S. R. Barker, J. Ass. Off. Anal. Chem., 1990 73, 860. 6 A. R. Long, L. C. Hsieh, M. S. Marlbrough, C. R. Short and S. R. Barker, J. Ass. Off. Anal. Chem., 1989, 72, 739. 7 M. D. Rose, G. Shearer and W. H. H. Farrington, Food Addit. Contam, 1997, 14, 15. 8 Toxicological evaluation of certain veterinary drug residues in food; World Health Organisation Additives Series: No. 29, World Health Organization, Geneva, Switzerland, pp. 79–106. 9 G. W. F. Stubbings, in preparation. Paper 8/09058K Table 4 Method validation data—samples fortified at 100 mg kg21. A blank sample was analysed on each day and found to contain no trace of any analyte Recovery (%) OH-OXF-S OH-OXF OXF-A OXF-S-A OXF OXF-S OH-FEN FEN-A FEN Day 1 n = 6 Mean (%) 53 74 48 51 77 66 40 28 41 RSD (%) 19 14 12 12 11 17 19 19 21 Day 2 n = 4 Mean 59 93 77 82 84 105 48 37 49 RSD (%) 22 18 2.9 3.2 16 11 13 17 14 Day 3 n = 4 Mean 63 105 81 85 93 117 49 36 46 RSD (%) 21 20 2.6 3.8 15 12 22 8.4 16 1026 Analyst, 1999, 124, 1023–1026

ISSN:0003-2654    

DOI:10.1039/a809058k

出版商:RSC    

年代:1999

数据来源: RSC