摘要:

ANALYST, JULY 1993, VOL. 118 753 Determination of Anatoxin-a and Homoanatoxin in Blue-Green Algal - Extracts by High-performance Liquid Chromatography and Gas Chromatography-Mass Spectrometry Anastasia Zotou" and Terry M. Jefferiest School of Pharmacy and Pharmacology, University of Bath, Claverton Down, Bath, UK BA2 7AY Paul A. Brough and Timothy Gallagher School of Chemistry, University of Bath, Claverton Down, Bath, UK BA2 7AY The production of cyanobacterial toxins as anatoxin-a in the UK by blue-green algae such as Oscillatoria and Anabaena flos-aquae is a potential health hazard, especially to animals, birds and fish. A sensitive reversed-phase ion-pair high-performance liquid chromatographic (RP-HPLC) method is described for the measurement of the neurotoxins anatoxin-a (AnTx) and homoanatoxin (HomoAnTx) in the presence of lyophilized algal extract.A base-deactivated CI8 column material was used with acetonitrile-phosphate buffer (pH 3) and sodium dodecyl sulfate as ion-pair reagent with ultraviolet detection a t 227 nm. Linear calibrations were obtained between 2 and 93 ng on-column for AnTx-HCI and 2-112 ng on-column for HomoAnTx.HCI with limits of detection of 1 and 2 ng on-column, respectively. A sample of Oscillatoria bloom material collected from Loch Insh, Scotland, in 1991 was found by this method to contain approximately 0.8 mg of AnTx-HCI per gram of lyophilized extract. Homoanatoxin was not present. The identity of AnTx was confirmed by the isolation of the AnTx peak by RP-HPLC, its derivatization to N-butylAnTx and its analysis by RP-HPLC and gas chromatography-mass spectrometry.Keywords: Anatoxin-a; homoanatoxin; Ana baena flos-aquae and Oscillatoria; high-performance liquid chromatography; gas chromatograph y-mass spectrometry Anatoxin-a (AnTx), 2-acetyl-9-azabicyclo[4.2.l]non-2-ene, is a bicyclic secondary amine, relative molecular mass 165, pK, 9.4, incorporating an a,fi-unsaturated enone moiety.' Most reports of AnTx occurrence are associated with Anabaena jZos-aquae, A. spiroides or A. circinalis.2 Recently, it has been shown that Oscillatoria strains also produce AnTx.3-4 Ana- toxin-a is a potent nicotinic acetylcholine agonist, which acts as a post-synaptic, depolarizing neuromuscular blocking agent, having an LDSo value [lethal dose (median); intraperi- toneal mouse] of approximately 200 pg kg-I body mass.2 Animals injected at that dose level can die within a few minutes, with death preceded by symptoms of ataxia and convulsions. No human deaths have been attributed to AnTx poisoning but cases of acutc gastrointestinal disease, allergic dermatitis and general malaise have been attributed to A.flos-aquae. Blooms of A. flos-aquae producing AnTx have been found in lakes of the northwestern USA, western Canada and Finland.5 Homoanatoxin is a methylene homol- ogue of AnTx, produced by Oscillatoria species,6 having a similar, but less toxic, neuromuscular blocking activity to AnTx.2.7 Currently, the most common cyanbacteria cited in fresh water poisoning incidents are Microcystis ueruginosa, Anabaena flos-aquae, Aphanizomenon and Oscillatoria. In the UK during the summer of 1989 about 20 sheep and 15 dogs died after drinking contaminated water, showing symptoms consistent with poisoning by microcystins, which are algal- derived hepatotoxins.8 Deaths of four dogs occurred also during the following summers of 1990 and 1991 due to cyanobacterial poisoning at Loch Insh, near Kingussie in the Grampian region of Scotland.Oscillatoria species were found at the water's edge and AnTx was identified in the bloom material and in the stomach contents of two of the poisoned dogs.3 Literature methods for the measurement of AnTx in algae and waters include high-performance liquid chromatography ' Present address: Laboratory of Analytical Chemistry, Chemistry Department. University of Thessaloniki, 54006 Thessaloniki, Greece.1 To whom correspondence should be addressed. (HPLC),s-'~310 high-performance thin-layer chromato- graphy,ll thin-layer chrornatography,IZ gas chromatography with electron-capture detection, 13 gas chromatography-mass spectrometry (GC-MS)14,15 and mass spectrometry. l6 Although most of these methods present a complete scheme for the isolation, purification and determination of AnTx, they either propose tedious and time-comsuming clean-up procedures or lack adequate sensitivity for the detection and determination of sub-lethal levels of AnTx. Two of the most sensitive methods proposed'"1S achieve their sensitivity by means of derivatization and only one of all the methods published makes use of an internal standard.13 In this work, an isocratic ion-pair reversed-phase (RP) HPLC method was used for the determination of low levels (approximately 2 ng on-column) of AnTx and HomoAnTx.It has also been used for the isolation of AnTx from algal material collected from Loch Insh. The identity of this AnTx was confirmed by GC-MS as its N-butyl derivative. This derivative was also examined by the same RP-HPLC system as used for AnTx. Experimental Materials Anatoxin-a is not readily isolated in reasonable amounts from Anabaena, which is also not a reliable source of supply. Homoanatoxin has only recently been identified in Oscilla- toriah and is not commercially available. Anatoxin-a and HomoAnTx occur naturally as the (+ )-enantiomers, but as the analytical methods employed in this study do not distinguish between the (+)- and (-)-enantiomers, the (+)-racemic standards are equally acceptable.The compounds (+)-AnTx-HCI and (+)-HomoAnTx.HC1 were synthesized on the milligram scale at the University of Bath by methods published elsewhere7.17-19 and their high purity was confirmed by KP-HPLC of aqueous solutions. The solutions were then lyophilized and their residues dried over P205 under vacuum at room temperature to constant mass754 ANALYST, JULY 1993, VOL. 118 (about 3 mg). The dried residues were then reconstituted in 0.5 mmol 1 - 1 HCI to give concentrations of 161.4 and 370 pg ml-1 as the hydrochloride salts of AnTx and HomoAnTx, respectively. The solutions were stored at -20 "C. Working standard solutions for the calibration graphs were prepared by suitable dilutions with mobile phase, when required.The accuracy of all dilutions was ensured by using a five-place decimal balance to weigh all volumes of calibration solutions. Lyophilized bloom material (7.23 mg) identified as Oscilla- toria was collected from Loch Insh in 1991. Reagents All solvents were of HPLC grade; acetonitrile and acetone were obtained from Fisons (Loughborough, Leicestershire, UK) and ethanol from Aldrich (Gillingham, Dorset, UK). Other chemicals were of analytical-reagent grade. Sodium dodecyl sulfate (SDS), atropine sulfate, pethidine hydro- chloride and anhydrous potassium carbonate were obtained from BDH (Poole, Dorset, UK). The buffer components, potassium dihydrogenphosphate and orthophosphoric acid, were obtained from Fisons; 1-iodobutane (99%) was obtained from Aldrich.Apparatus The HPLC instrumentation was assembled from a Model 300 constant-flow rate pump (Scientific Systems State College, PA, USA) a Rheodyne (Cotati, CA, USA) Model 7125 injection valve with a 20 or 50 pl loop, a Model 500 ultraviolet/visible spectrophotometric detector (Scientific Systems) and a BBC Servoscript SE-120 recorder (LDC, Stone, Staffordshire, UK). A Hypersil BDS (base-deactivated silica) reversed-phase C18 ( 5 pm) column (150 X 2.1 mm i.d.) (Capital HPLC Specialists, Bathgate, West Lothian, UK) was used for the analysis and purification of AnTx. An LDC Analytical Spectromonitor 5000 photodiode-array detector connected to the HPLC column was used to determine the A,,, of AnTx. A Hewlett-Packard (Avondale, PA, USA) GC-MS system consisting of a Model 5890A gas chromatograph with a 50 m x 0.2 mm i.d.HP-1 fused-silica capillary column (0.33 pm film thickness), a Model 5970 Series mass-selective detector and a Model 7946 data system was used for the confirmation of peak identities. Hydrogen was used as the carrier gas. For the determination of AnTx and N-butyl AnTx the oven tempera- ture was programmed from 60 to 240°C at 20°C min-1. The volume injected was 2 p1 (splitless) at an injection temperature of 100 "C. A Model 5414 Eppendorf centrifuge (BDH, Poole, Dorset, UK) was used for the centrifugation of the algal material suspension and a Mettler (Highstown, NJ, USA) AE 163 analytical balance with an accuracy of +0.00001 g was used for accurate weighing. RP-HPLC Conditions Anatoxin-a, HomoAnTx and N-butylAnTx were monitored at their A,,, of 227 nm, produced by the a,p-unsaturated ketone group.A detector sensitivity setting of 0.02 or 0.05 a.u.f.s. was used throughout this study. A 20 p1 loop was used for the analytical part of the work and a 50 pl loop for the collection of AnTx fractions from the Loch Insh sample. The mobile phase consisted of mixtures of acetonitrile and phosphate buffer (pH 3) with SDS as ion-pair agent. The following mobile phase systems were used: (i) for the determination of AnTx and HomoAnTx, acetonitrile-0.5 mmol l-1 SDS in 0.005 moll-' phosphate buffer (pH 3) (30 + 70) at a flow rate of 0.3 ml min-1; (ii) for the determination of N-butylAnTx, acetonitrile-0.5 mmol 1- SDS in 0.033 mol 1-1 phosphate buffer (pH 3) (30 + 70) at a flow rate of 0.4 ml min- 1; and (iii) for the purification of AnTx extracted from the algal suspension, acetonitrile-0.5 mmol 1-1 SDS in 0.005 moll-1 phosphate buffer (pH 3) (25 + 75) at a flow rate of 0.3 ml min-1.The phosphate buffer was prepared from KH2P04 solution adjusted to pH 3 by the addition of orthophosphoric acid after dissolution of the SDS. The mobile phases were filtered through a 0.45 pm pore-size nylon membrane filter and de-gassed with helium prior to use. The above-mentioned concentrations of phosphate buffer, SDS and contents of acetonitrile were selected from among several others exam- ined as giving the best chromatograms in terms of short retention times and good separation of the analytes. All solutions subject to analysis were dissolved in the appropriate mobile phase prior to injection onto the RP-HPLC column.Extraction of AnTx From Algal Material For the extraction of AnTx, the lyophilized and homogenized algal material (7.23 mg) was placed in an Eppendorf vial and suspended in 1 ml of acetonitrile4.005 mol 1-1 KH2 PO4 (pH 3) (10 + 90). After addition of 100 pl of 0.01 moll-1 HCI, the suspension was vortex mixed, sonicated for 30 min and centrifuged at 12 000 rev min-1 (8800g) in an Eppendorf centrifuge for 10 min. The yellowish supernatant was filtered through a 0.2 pm Acro LC13 filter (Gelman Sciences, Ann Arbor, MI, USA) to remove particulate matter and injected onto the RP-HPLC column for the purification of AnTx from other components of the algal suspension.Fractions of peaks, with a retention time identical with that of AnTx, were collected from a series of 50 pl injections, pooled and lyophilized. The residue was reconstituted with 1 ml of ethanol. Derivatization Procedure The derivatization of AnTx, as free base, was performed in acetone, over KzC03,") using 1-iodobutane to form N-butyl- AnTx. Ethanolic AnTx.HCI solutions containing 2.51-32.4 pg were placed in Reacti-vials (Pierce and Warriner, Chester, UK) and evaporated to dryness with a stream of nitrogen. To the dry residue of AnTx-HCI were added 10 mg of dried K2C03, 180 pl of acetone (dried over molecular sieves) and 30 pl of 1-iodobutane. The Reacti-vials were closed tightly and the reaction mixture was vortex mixed for 30 s and then allowed to stand at 60°C for 3 h. After cooling, the acetone was evaporated using nitrogen and the residue was washed three times with 180 p1 of acetone and vortex mixed for 30 s between each washing.The acetone fractions that contained the N-butylAnTx were combined in a clean container for GC-MS analysis. For RP-HPLC analysis, some of the acetone fraction was evaporated to dryness with nitrogen and the residue reconstituted with mobile phase. Analytical Procedure for Standard and Sample Solutions Analysis of An Tx standard solutions Seven working standards of concentrations between 0.102 and 4.67 pg ml-1 of AnTx.HC1 (corresponding to 0.084-3.82 pg m1-l of AnTx free base) were prepared by appropriate dilutions of the AnTxSHCl stock solution (161.4 pg ml-1) with mobile phase. Appropriate volumes of a 51 pg ml-1 solution of Fig.1 Structures of (u) anatoxin-a and ( h ) homoanatoxinANALYST. JULY 1993, VOL. 118 I I I I 16 12 8 4 755 I L 0 t - m C 0: v) .- Atropine Ho m oA nTx I 20 16 12 8 4 0 Time/mi n Fig. 2 Chromatogram of AnTx.HCI (0.395 pg ml-I), HomoAnTx.HCI (0.575 pg ml-1) and atropine sulfate (internal standard) (15.3 pg ml I ) . Conditions as dcscribed under RP-HPLC conditions (i), with a 20 pl injection volume and ultraviolet sensitivity at 0.02 a.u.f.s. Pet hidine Butyl- AnTx 20 16 12 8 4 0 Fig. 3 Chromatogram of the N-butyl derivative of a 2.91 pg ml-1 standard AnTx solution with pethidine hydrochloride (19.2 pg ml-1) as internal standard. Conditions as described under RP-HPLC conditions (ii), with a 20 pl injection volume and ultraviolet scnsitivity at 0.02 a.u.f.s.Time/min atropine sulfate (internal standard) in mobile phase were added to the working standards during dilution, to give a final concentration of 15.3 pg ml-1 of atropine sulfate. The solutions were analysed by RP-HPLC in duplicate using the conditions described earlier. Analysis of HomoAn Tx standard solutions Five working standards of concentrations between 0.113 and 5.60 pg ml-1 of HomoAnTx-HCI (corresponding to 0.094- 4.65 pg ml-1 of free base), containing atropine sulfate (15.3 pg ml-1) as internal standard, were prepared and analysed as described for AnTx. t - m C 0, v) .- Ir -1 I I I 20 16 12 8 4 0 :b) AnTx Analysis of N-butylAn T x standard solutions A 1 ml volume of the AnTxeHCl stock solution (161.4 pg ml-1) was lyophilized and the residue reconstituted with 1 ml of ethanol.From this ethanolic solution, amounts of 2.51-32.4 pg of AnTx-HCI (which correspond to 2.05-26.5 pg of AnTx as free base) were derivatized according to the procedure described earlier. To aliquots of 50 pl of the acetone fractions, which contained the N-butyl derivative, volumes of 80 p1 of a 120 pg ml-1 ethanolic solution of pethidine hydrochloride (chromatographic internal standard) were added and the mixtures evaporated to dryness with a stream of nitrogen. The residues were reconstituted with 500 pl of mobile phase, yielding final concentrations of N-butyl- AnTx between 0.380 and 4.91 pg ml-1 and a final concentra- tion of pethidine hydrochloride equal to 19.2 pg ml-1. The solutions were analysed by RP-HPLC using the conditions described earlier.A seven-point calibration graph was con- structed using duplicate injections. Determination o f A n T x in extract f r o m algal material Aliquots of 50 and 100 pl of the ethanolic solution of the Loch Insh fractions that were obtained from RP-HPLC separation756 t - m C m v) .- a) Atropine AnTx I I I I 16 12 8 4 Pethidine I I I I 1 I 20 16 12 8 4 0 Time/mi n Fig. 5 Chromatograms of ( a ) extracted AnTx from Loch Insh sample after RP-HPLC purification (conditions as in Fig. 2) and ( h ) the N-butyl derivative of this AnTx (conditions as in Fig. 3) (as described under Extraction of AnTx from Algal Material) were evaporated to dryness with a stream of nitrogen. The residues were reconstituted with 350 p1 of mobile phase and 150 pl of a 51 pg ml-1 atropine sulfate solution in mobile phase.The resulting solutions were analysed for AnTx by RP-HPLC, making five replicate injections of each between standard AnTx-HCI solutions. Another 500 p1 aliquot of the same ethanolic solution of the Loch Insh fractions was evaporated to dryness with nitrogen and derivatized as described earlier. The acetone solution of the derivative was injected into the GC-MS system for confirmation of the presence of AnTx in the Loch Insh sample. To an 80 pl aliquot of the acetone solution of the N-butyl derivative, an 80 pl volume of a 120 pg ml-1 ethanolic solution of pethidine hydrochloride (internal standard) was added and the mixture was evaporated to dryness. The residue was reconstituted with 500 pl of mobile phase and analysed by RP-HPLC using five replicate injections between standard N-butylAnTx solutions.Results and Discussion Anatoxin-a [Fig. l(a)] is a polar, water-soluble amine that readily forms ion pairs in acidic mobile phases with reagents such as SDS. In this way, the chromatographic retention ( k ' ) of AnTx may be increased to remove it from interferences caused by the presence of other compounds in the algal extracts. The peak shape was improved by the selection of a base-deactivated column material and the choice of a 2 mm i.d. column to enhance peak heights (Fig. 2). The presence of an additional methylene group in HomoAnTx [Fig. l(b)] 8.0 x 106 6.0 x 106 4.0 x 106 2.0 x 106 0, C m -0 C 2 ANALYST, JULY 1993, VOL. I- 2 4 6 8 10 Time/min I 12* ( b) 2 1.2 x 106 8.0 x 105 'i' 100 110 1'20 130 140 150 160 m/z - I6! 118 Fig.6 gram; and ( b ) mass spectrum. Conditions as under Apparatus GC-MS of standard AnTx. ( a ) Total ion current chromato- increases the hydrophobicity of this compound compared with AnTx and so causes a further increase in retention. The resolution between AnTx and HomoAnTx is good (Fig. 2), and it is probable that these conditions would also be suitable for other natural or synthetic analogues of AnTx. AnTx Calibration Graph By using peak-height ratios of AnTx to atropine sulfate, excellent linearity for concentrations of AnTx-HCI between 0.102 and 4.67 pg ml-1 (corresponding to 0.084-3.82 pg ml-1 of free base) was obtained with a correlation coefficient of 0.999. The limit of detection was 1 ng on-column of AnTx-HCI (corresponding to 0.8 ng as free base).The least-squares straight-line equation with the standard deviations (SDs) of the intercept and the slope was y = 0.0125 (k0.0366) + 1.234 (+ 0.0156) x, where x is expressed in pg ml-1 of AnTx-HCI. The precision of the method as measured by the relative standard deviation (RSD) of replicate measurements was 1.4 and 1.1% for eight replicate analyses of a 0.533 and a 1.45 pg ml-1 solution of AnTx-HCI, respectively. The between- day precision was assessed by the repeated analysis of the 1.45 pg ml-1 solution of AnTx-HCI over 5 d and the RSD was found to be 2.4%. HomoAnTx Calibration Graph The linear working range of HomoAnTx extended between 0.113 and 5.60 pg ml-1 of HomoAnTx-HCI (corresponding to 0.094-4.65 pg ml-1 of free base) using peak-height ratios of HomoAnTx to atropine sulfate.The correlation coefficient was 0.999 and the least-squares straight-line equation with the SD of the intercept and the slope was y = 0.0193 (k0.0335) +ANALYST, JULY 1993, VOL. 118 a 2.5 x 106 757 178 - (b) 221 192 206 8.0 x 106 6.0 x 106 I I I I 1 4.0 x 106 12 000 8 000 2.0 x 106 I 6.0 8.0 10.0 12.0 Time/mi n 1 2.0 x 106 17742 (d) I 1 I 1 I 1 1.5 x 106 t 17 742 1.0 x 106 1 5.0 x 105 0 17742 - c I L I I 6.0 8.0 10.0 12.0 Time/m in Fig. 7 GC-MS of AnTx isolated from Loch Insh sample, derivatizcd to N-butylAnTx and analysed using SIM: monitored at ( a ) rnlz 164 and ( b ) mlz 221; monitored at (c) rnlz 122, (d) mlz 136, (e) rnlz 150 and (f) mlz 164.Conditions as in Fig. 6 120 140 160 180 200 220 rnlz Fig. 8 chromatogram; and ( h ) mass spectrum. Conditions as in Fig. 6 GC-MS of standard N-butylAnTx. (a) Total ion current having the same retention time as AnTx [Fig. 4(b)]. As mentioned earlier, this peak was then collected from a series of 50 pl injections and the fractions were lyophilized and reconstituted in 1 ml of ethanol. Aliquots of this solution, when analysed by RP-HPLC, after the appropriate treatment, gave chromatograms free from any impurities and a peak with a retention time identical with that of the AnTx standard [Fig. 5(a)]. Another aliquot of the same ethanolic solution of the Loch Insh fractions was used to form the N-butyl derivative, if present, and injected onto the RP-HPLC column, after the appropriate treatment.The chromatograms obtained [Fig. 5(b)] were similar to those previously obtained from N-butyl- AnTx standards (Fig. 3), and thus provided good evidence for the presence of AnTx in the algal material. There was no evidence for the presence of HomoAnTx. The final confirmation of peak identities was established by GC-MS analysis. For AnTx standard, the retention time was 8.6 min [Fig. 6(a)] and principal peaks occurred at rnlz 165 (M+), 122, 136, 150, 108, 105, 109 and 132 [Fig. 6(b)]. The AnTx isolated from the Loch Insh sample was derivatized and the acetone fraction containing the N-butyl- AnTx was reduced to a very small volume (about 5 pI) for GC-MS analysis. Gas chromatography-mass spectrometry of this solution under selected ion monitoring (STM) conditions at rnlz 221 (M+) and 164 gave two peaks with identical retention times of 10.2 min, indicating the presence of N-butylAnTx [Fig.7(a) and (b)]. A second injection moni- tored at rnlz 164, 150, 136 and 122 gave four peaks with identical retention times of 10.2 min, indicating the AnTx component of the N-butylAnTx structure [Fig. 7(c)-@]. Gas chromatography-mass spectrometry of standard N-butylAnTx gave a retention time of 10.2 min [Fig. 8(a)] and the mass spectrum [Fig. 8(b)] confirmed the presence of the ions of mlz 221,164,150,136 and 122. The mass spectrum also 1.019 (k0.0124)~~ where x is the concentration of HomoAnTx-HCI in pg ml-1. The limit of detection was 2 ng on-column of HomoAnTx-HC1 or 1.7 ng of free base. The RSD of eight replicate analyses of a 1.05 pg ml-l HomoAnTx.HC1 solution was 1.9%.N-ButylAnTx Calibration Graph A linear relationship between concentrations of 0.380 and 4.91 pg ml-1 and peak-height ratios of N-butylAnTx to the internal standard pethidine hydrochloride was observed. No suitable recovery internal standard was found for the N-butyl derivative of AnTx, nor could atropine sulfate be used as a chromatographic standard with the present mobile phase because its retention time was almost identical with that of N-butyl AnTx. With respect to satisfactory separation from the N-butylAnTx peak, pethidine hydrochloride was found to be the most suitable chromatographic internal standard among several others examined. The least-squares straight- line equation with the SD of the intercept and the slope was y = 0.0972 (k0.0224) + 0.4109 ( k 0 .0 0 7 7 ) ~ ~ where x is the concentration of N-butylAnTx in pg ml-1. The correlation coefficient was 0.999. A typical chromatogram of N-butyl- AnTx is shown in Fig. 3. The early-eluting peaks are due to the excess reagent and reaction by-products. No peaks occur in the region of the N-butylAnTx peak. The N-butyl derivatives are stable in acetone and are suitable for both RP-HPLC and GC-MS analysis. Identification and Determination of AnTx in Algal Material An injection of 50 pl of the extract from the Loch Insh algal material [Fig. 4(a)] indicated the presence of a compound(s)758 ANALYST, JULY 1993, VOL. 118 shows the progressive loss of methylene units from the N-butyl group in the series mlz 221,206, 192, 178, 164.The AnTX content (as the free base) of the Loch Tnsh sample was calculated using the calibration graphs for both AnTx and N-butylAnTx as a double-check procedure and was found to be 0.8 mg of AnTx.HC1 per gram of lyophilized material. Lyophilized and homogenized algal material that had been shown by RP-HPLC not to contain AnTx was used to test the effectiveness of the AnTx extraction step. Duplicate samples ( 5 mg) of algal material were placed in Eppendorf vials and spiked with either 0.5 or 4 pg of AnTx, and then extracted as described under Experimental. Duplicate samples of algal material were also processed as blank controls. The extracts were analysed by RP-HPLC and the peak heights were compared directly with those obtained from the RP-HPLC of 0.5 and 4 pg of AnTx standard solutions that had not been extracted.Each solution was injected in duplicate, with sample and standard solutions alternating in injection order. For the 0.5 pg spike, the recoveries were 90.9 and 90.9% and for the 4 pg spike they were 98.7 and loo%, respectively. No peaks were present in the blank controls at the retention time of AnTx. The amount of AnTx found in the Loch Insh sample is therefore likely to be a reasonable estimate. Conclusions The proposed RP-HPLC method is sensitive, precise and accurate, and allows the determination of low levels (approxi- mately 2 ng on-column) of AnTx and its homologue HomoAnTx, both separately and in mixtures. The method has also been demonstrated to be suitable for the purification of AnTx from other compounds in algal material and has been applied to the purification and subsequent determination of AnTx in algal material from Loch Insh.The final confirmation of the presence of AnTx in the algal material tested was made by GC-MS, using the N-butyl derivative of AnTx as a means of confirmation. The proposed derivative is stable, relatively free from reaction by-products and suitable for both RP- HPLC and GC-MS analysis. There was no evidence of the presence of HomoAnTx in the algal material tested. The authors thank Professor G. A. Codd (Department of Biological Sciences, University of Dundee) for providing the lyophilized algal material from Loch Insh and K. Smith and Dr. I. M. Roy (both at the School of Pharmacy and Pharmacology, University of Bath) for technical assistance and for carrying out the recovery tests on AnTx from algal material, respectively.1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 References Koskinen, A. M. P . , and Rapoport, H., J. Med. Chem., 1985, 28, 1301. Carmichacl, W. W., J. Appl. Bacteriol., 1992, 72, 445. Gunn, G. J., Rafferty, A. G.. Rafferty. G. C., Cockburn, N., Edwards, C., Beattie, K. A., and Codd, C;. A., Vet. Rec., 1992, 130, 301. Sivonen, K., Himberg, K., Luukkaincn, K., Nicmcla. S . , Poon, G. K.. and Codd, G. A., Toxic. Assess., 1989, 4. 339. Harada, K.-I., Kimura, Y., Ogawa, K . , Suzuki, M., Dahlem, A. M., Beasley, V. R., and Carmichael, W. W., Toxicon, 1989, 27, 1289. Skulberg, 0. M., Carmichael, W. W., Andersen, R. A., Matsunaga, S . , Moorc, R. E., and Skulbcrg. R., Environ. Toxicol. Chem., 1992, 11, 32 I . Wonnacott, S . . Swanson, K. L., Albuqucrque, E. X., Huby, N. J . S., Thompson, P., and Gallagher, T., Biochem. Phar- macol., 1992, 43, 419. Reynolds, C. S . , Freshwater Forum, 1991, 1, 29. Astrahan, N. B., and Archer, B. G., in The Water Environment: Algal Toxins and Health, ed. Carmichael, W. W., Plenum, New York, 1981, pp. 437-446. Wong, S. H., and Hindin, E . , J. Am. Water Works Assoc., 1982, 74, 528. Al-Layal, K. J . , Poon, G. K . , and Codd, G. A., J. Microbiol, Methods, 1988, 7, 251. Ojanpcra, I., Vuori, E., Himbcrg, K., Waris, M., and Niini- vaara, K., Analyst, 1991, 116, 265. Stevens, D. K., and Krieger, R. I . , J. Anal. Toxicol., 1988, 12, 126. Smith, R. A., and Lewis, D., Vet. Hum. 'Ioxicol., 1987,29,153. Himberg, K., J. Chrornatogr., 1989, 481, 358. Ross, M. M., Kidwell, D. A.. and Callahan, J . H., J . Anal. Toxicol., 1989, 13, 317. Lindgren, B., Stjernlof. P., and Trogen, L,. , Acfa Chem. Scund., 1987, 41, 180. Sardina, F. J . , Howard, M. H., Morningstar, M., and Rapo- port, H., J. Org. Chem., 1990, 55, 5025. Huby, N. J . , Ph.D. Thcsis, University of Bath, 1990. Ortuno, J . , de la Torre, R., Segura, J., and Cami, J . , J. Pharm. Biorned. Anal., 1990, 8, 911. Paper 2i04353J Received August I I , I992 Accepted January 6, 1993

ISSN:0003-2654    

DOI:10.1039/AN9931800753

出版商:RSC    

年代:1993

数据来源: RSC

摘要:

ANALYST. JULY 1993, VOL. 118 759 4-( N,N=Dimethylaminosulfonyl)-7-(2-chloroformylpyrrolidin~l -yl)- 2,1,3- benzoxadiazole : Novel Fluorescent C h iral Derivatization Reagents for the Resolution of Alcohol Enantiomers by High-performance Liquid Chromatography Toshimasa Toyo’oka, Mumio lshibashi and Tadao Terao Division of Drugs, National Institute of Hygienic Sciences, 1-18-1 Kamiyoga, Setaga ya-ku, Tokyo 158, Japan Kazuhiro lmai Branch Hospital Pharmacy, University of Tokyo, 3-28-6 Mejirodai, Bunkyo-ku, Tokyo I 12, Japan 0 p t i c a I I y a ct i ve d e r i vat i za t i o n re a g e n t s , 4- ( N, N- d i m e t h y I a m i n o s u If o n y I ) -7 - ( 2-c h I o r of o r m y I p y r r o I i d i n - 1 - y I ) - 2,1,3-benzoxadiazole [(R)-( +)-DBD-Pro-COCI and (S)-( -)-DBD-Pro-COCI], were synthesized to permit the separation of alcohol enantiomers by high-performance liquid chromatography.The reagents react with hydroxyl groups in the presence of pyridine, which functions as a catalyst and reacts with hydrogen chloride. The maximum excitation and emission wavelengths of the diastereomers derived from the alcohols and the reagents were approximately 450 and 560 nm, respectively. The emission wavelengths of the derivatives shifted slightly towards the blue with increasing acetonitrile concentration in the medium; however, the excitation wavelengths remained constant. The diastereomers derived from some aliphatic alcohols were efficiently resolved by normal-phase chromatography with hexane-ethyl acetate as the eluent. Incomplete separation was realized with a reversed-phase column with water-acetonitrile as the eluent. When (R)-( +)-DBD-Pro-COCI was used as the derivatization reagent, alcohols corresponding to the R-configuration were eluted faster than those corresponding to the S-configuration.As expected, the elution order of the alcohols was reversed when the diastereomers were prepared with (S)-( -)-DBD-Pro-COCI. The Rs values of diastereomers derived from hydrophobic alcohols are larger than those from hydrophilic alcohols. Keywords: Pre-column derivatization; chiral derivatization reagent; fluorescence detection; alcohol enan tiomers; high -performance liquid chroma tograph y Many biologically active compounds, including drugs, have one or more asymmetric carbons. As enantiomers sometimes exhibit different pharmacological activities, the precise and sensitive resolution of racemic compounds is important in many fields such as the pharmacokinetic study of drugs.Various chiral separations of alcohol compounds have been achieved with gas chromatography (GC) ,1-4 high-perfor- mance liquid chromatography (HPLC)s--7 and high-perfor- mance capillary electrophoresis (HPCE) .8-10 Two general approaches to the chromatographic separation of the enantiomers have been developed. The indirect diastereomeric method involving a derivatization step with a suitable chiral reagent has the advantage of adding a func- tional group that permits both sensitive detection and resolu- tion of the racemates, although the diastereometric method requires more manipulations than the direct method using a chiral stationary phase (CSP) .5,6 The hydroxyl group is one of the most difficult to derivatize owing to its limited reactivity with the derivatization reagent and the relatively low stability of the reagent.Drimanoyl chloride, chrysanthemoyl chloride1I and a-trifluoromethyl-a-methoxyphenylacetyl chloride [( +)- and (-)-enantiomers]12 have been successfully used with alcohol enantiomers as derivatization reagents for GC analysis. However, the GC method is not applicable to heat-labile compounds. High-performance liquid chromatography has been widely employed for the resolution of chiral molecules. Derivatiza- tion reagents for HPLC analysis include (-)-1-(1-naphthyl- ethyl) isocyanate,’3 2-methyl-l,l’-binaphthalene-2’-carbonyl cyanide [ (+)- and (-)-forms] 14 and (aS)-2-methoxy-l, 1 ’- binaphthalene-2’-carbonyl cyanide (a chiral axis reagent).1s We have reported that 14-( N , N-dimethylaminosulfonyl)-7- (3-aminopyrrolidine- l-yl)-2,1,3-benzoxadiazole (DBD-APy), 4 - nitro - 7 - (3 - aminopyrrolidine - 1 - yl) - 2,1,3 - benzoxadiazole (NBD-APy) and 4-(aminosulfonyl)-7-(3-aminopyrrolidine-l- yl)-2,1,3-benzoxadiazole (ABD-APy)]1”19 are fluorescent chiral derivatization reagents suitable for carboxylic acid enantiomers. The resulting diastereomers exhibit excitation and emission maxima in the long-wavelength region. In the course of our studies on fluorescent chiral derivatization reagents, we have developed optically active 4-(N, N-dimethy- laminosulfonyl)-7-(2-carboxypyrrolidin-l -yl)-2,1,3-benzoxa- diazole (DBD-Pro), which is obtained easily by the one-step reaction of proline with 4-(N, N-dimethylaminosulfonyl)-7- fluoro-2,1,3-benzoxadiazole (DBD-F) .20 This substance was considered as a plausible fluorophore for the detection of alcohol enantiomers because of its structural similarity to DBD- APy, which was reported previously. 16 However, DBD-Pro itself does not react directly with alcohol com- pounds without modification.The carboxylic acid functional group in the structure of DBD-Pro can be converted into a carboxylic acid chloride group, which readily esterifies the hydroxyl group. This paper describes the synthesis of chiral derivatization reagents [(S)-( -)- and (R)-( +)-enantiomers of DBD-Pro- COCI] that react with alcohol enantiomers to form corre- sponding fluorescent diastereomers. Their reactivity towards the alcohol functional group, the fluorescence characteristics of the diastereomers derived from the reagents and the separation of the diastereomers by HPLC were also investi- gated.Experimental Materials and Reagents The reagent DBD-F was purchased from Tokyo Kasei (Tokyo, Japan). 4-( N , N-Dime thylaminosulfonyl)-7-(2-~ar- boxypyrrolidine-l-yl)-2,1,3-benzoxadiazole [(R)-( +)- and (S)-( -)-DBD-Pro] were synthesized as described pre- viously.20 Proline [(R)-( +)- and (S)-( -)-enantiomers] were760 ANALYST, .JULY 1993. VOL. 118 obtained from Sigma (St. Louis, Mo, USA). The (S)-(+)- and (R)-( -)-enantiomen of hexan-2-01, heptan-2-01 and nonan-2- 01 were purchased from Wako (Osaka, Japan).1-Phenyl- ethanol [(S)-( -)- and (R)-( +)-enantiomers], methylamine (30% in water) and pyridine (Wako) werc used as received. Ethyl acetate, hexane, benzene, acetonitrile and water were of HPLC grade (Wako). All 'other chemicals were of analytical-reagcnt grade and were used without further purification. Apparatus Proton nuclear magnetic resonance ('H NMR) spectra were recorded on a Varian (Palo Alto, CA, USA) Jemini-300 instrument at 300 MHz using tetramethylsilane (0.00 ppm) as the internal standard. For describing NMR characteristics, the following abbreviations are used: s = singlet, d = doublet and m = multiplet. Mass spectrometry (MS) was carried out on a Jeol (Tokyo, Japan) DX-300 [70eV, electron impact (El) ionization] mass Spectrometer. Infrared (IR) spectra were measured using potassium bromide discs with a Shimadzu (Kyoto, Japan) Model IR-460 spectrometer.For measure- ment of excitation and emission spectra, a Hitachi (Tokyo, Japan), Model 650-60 spectrofluorimeter with a 1 cm quartz cell was employed without spectral correction. Optical rota- tionc; were measured on a DIP-370 digital polarimeter (Jasco, Tokyo, Japan) with a 50 X 3.5 mm diameter cylindrical cell. Melting-points (m.p.) were measured with a Yanagimoto (Tokyo, Japan) micro melting-point apparatus. The high-performance liquid chromatograph consisted of two LC-9A pumps (Shimadzu) and an SCL-6B system controller (Shimadzu). Sample solutions were injected with a SIL-6R autoinjector (Shimadzu). The analytical columns were Inertsil ODs-2 (150 X 4.6mm i.d.; 5 um) and Inertsil ODs-80A (150 x 4.6 mm i.d.; 5 pm) for reversed-phase chromatography and Inertsil SIL (150 x 4.6 mm i.d.; 5 pm) (GL Sciences, Tokyo, Japan) for normal-phase chromato- graphy.The columns were maintained at 40 "C with a Model 655A-52 column oven (Hitachi). A Shimadzu RF-550 fluores- cence monitor equipped with a 12 1.11 flow cell was employed for detection. The excitation and emission wavelengths were fixed at 450 and 560 nm, respectively. The peak areas obtained from the fluorescence monitor were determined with a C-R4A Chromatopac integrator (Shimadzu). All mobile phases were de-gassed with a DGU-3A on-line de-gasser (Shimadzu). The flow rate of the eluent was 1.0 ml min-I. Syntheses of Chiral Derivatization Reagents To (S)-( -)-DBD-Pro (55 mg, 0.16 mmol) suspended in 55 ml of anhydrous diethyl ether at 0°C was added phosphorus pentachloride (PClS) (110 mg, 0.53 mmol) and the mixture was stirred for 60 min in ice-water (approximately 5 "C).The crystalline precipitate was quickly filtered off and the filtrate solution was evaporated to dryness under reduced pressure. The residue was dried for 12 h in a vacuum desiccator over phosphorus pentoxide (P205). formylpyrrolidin- l-y1)-2,1,3- benzoxadiazole [ (S)-( - )-DBD- Pro-COCI]: yellow crystals, m.p. 116-117 "C (decomp.); yield 33 mg (56%); NMR in CDC13 (ppm), 7.90 (1 H, d, J a b = 8.1 Hz, a), 6.13 (1 H, d, Jab = 8.1 Hz, b), 5.59-5.63 (1 H, m, c), 3.74-3.84 (2 H, m, d), 2.88 (6 H, s, g), 2.54-2.63 (2 H, m, e), 2.18-2.28 (2 H, m, f); EI-MS, mlz 358 (M+); IR (KBr), 1794,1601,1555,1416,1337,1148,968 and 716 cm-1; [a]g = -44.7" (c = 0.89 in CHC13).(Found: C, 43.39; H, 4.21; N, 15.42. Calc. for CI3H15N404SCI: C, 43.52; H, 4.21; N, ( K ) - ( +)-DBD-Pro-COCI was obtained from the reaction of (R)-( +)-DBD-Pro and PC15 in the same manner as described above. (S)-( -)-4-(N, N-Dimethyaminosulfonyl)-7-(2-chloro- 15.62% .) (R)-(+)-DBD-Pro-COCI: yield 34 mg (59%); [a]$' = +44.7" (c = 0.89 in CHC13). (Found: C, 43.55; H, 4.23; N, 15.55. Calc. for C13H1sN404SC1: C, 43.52; H, 4.21; N, 15.62% .) Other instrumental data were the same as those for (S)-(-)-DBD-Pro-COCI. f e DBD-Pro-COCI Reactivity of Optically Active DBD-Pro-COCl With Heptan-2-01 Enantiomers A 30 p1 volume of 10 mmol 1 - 1 DBD-Pro-COCI [(R)-(+)- or (S)-( -)-enantiomer] in anhydrous benzene and 30 pl solution of heptan-2-01 (1 mmol I-' of one enantiomer) in anhydrous benzene containing 2% of pyridine were mixed in a 1.5 ml mini-vial (GL Scicnces).The vials were tightly capped and heated at 80 "C for 4 h. At fixed time intervals, one vial was removed from the dry heat block, and cooled in ice-water (0-5 "C). Then, 690 p1 of a 1% solution of methylamine in acetonitrile were added to the reaction mixture to stop the derivatization reaction. An aliquot (10 1.11) of the diluted solution was automatically injected into the Inertsil ODs-2 column and the fluorescence peak area of the resulting diastereomer was calculated with the integrator. The reagent blanks without heptan-2-01 were treated in the same manner. For the fluorescence spectral measurements, 50 p1 of the diluted solution were injected onto the column and the peak corresponding to the heptan-2-01 derivative was collected from the outlet of the detector (an approximately 2 ml portion).HPLC Separation of the Diastereomers Derived From the Alcohol Enantiomers and (R)-( +)- or (S)-( -)-DBD-Pro-COCl Alcohol enantiomers (approximately 1 mg each) were reacted at 80°C with DBD-Pro-COCI [l mmol I-' ( K ) - ( + ) - or (S)- (-)-enantiomer] in 1 ml of anhydrous benzene in the presence of 1% of pyridine. After a 3 h reaction time, and aliquot ( 5 1.11) of the solution was injected into the Incrtsil ODs-80A (reversed-phase) and Inertsil SIL (normal-phase) columns. The eluen ts for reversed- and normal-phase chromatography were water-acetonitrile and hexane-ethyl acetate mixtures, respectively.The capacity factor ( k ' ) , separation factor (a) and the resolution (R,) were calculated from the following equations: k' = (tn - to)/to; a = k;/k;; R, = 2 ( t ~ ~ - ~ R , ) / ( W I + W2) respectively, where tR, tR, and tR2 are the retention times of the peaks, to is the void volume of the column ( t = 1 .0 min) and w1 and w2 are the widths of the bases formed by triangulation of the peaks. Results and Discussion Synthesis of the Chiral Derivatization Reagents The structural features required of a chiral derivatization reagent for the separation and detection of enantiomers are as follows: (1) chiral centre close to reactive functional group; (2) good reactivity towards the target functional group; (3) excellent stability of the diastereomeric product; and (4)ANALYST, JULY 1093, VOL.118 76 1 Fig. 1 Synthetic route of the chiral derivatization reagents and preparation of diastereomers spectral properties that permit sensitive detection techniques such as fluorescence or chemiluminescence to be applied. In previous work, lc*') we developed optically active chiral derivatization reagents for carboxylic acids, DBD-APys, NBD-APys and ABD- APys, which have the benzofurazan structure (4,7-substituted 2,1,3-benzoxadiazoles). The dia- stereomers derived from these reagents exhibited characteris- tic fluorescence at long wavelengths. 16 All carboxylic acid enantiomers tested were successfully resolved by both reversed- and normal-phase chromatography. 19 The pyrroli- dine structure at the 4-position of the reagents seems to enhance the resolution because the conformations of the resulting diastereomers are probably fixed by the ring structure. In addition, the amido formation in the diaster- eomers may play a dominant role in the chromatographic separation owing to the formation of hydrogen bonds with the stationary phase.DBD-Pro, which has a structure similar to DBD-APy, is readily synthesized as described previously.20 DBD-Pro [ (R)- (+)- or (S)-( -)-enantiomer] is an optically active compound that exhibits excellent fluorescence characteristics. The excita- tion and emission wavelengths (approximately 450 and 560 nm, respectively) are in a spectral region that is relatively free from interferences. Although DBD-Pro is promising as a derivatization reagent for alcohols, it does not react with the alcoholic hydroxyl.The carboxylic acid group (-COOH) in the proline structure was therefore converted into an acid chloride group (-COCl) with various chlorination reagents such as phosphorus pentachloride (PCIs),21,22 thionyl chloride (SOCI?) and triphosgene.23 Among the reagents tested, PCIs yielded the best results. The chiral reagents are easily obtained by reaction of DBD-Pro enantiomers (S- or R-configuration) with PCIs at low temperature (0-5°C). Fig. 1 shows the synthetic route for the chiral derivatization reagents and the subsequent reaction with alcohol enantiomers. The direction of the optical rotation of the chiral reagent is the same as that of the starting material.No racemization occurs in the synthetic pathway, as indicated by the fact that the reaction of the alcohol enantiomer with the chiral reagent yielded a single diastereomer. Apparently, racemization does not occur dur- ing the synthesis of the reagents or the derivatization reaction. The yield (approximately 60%) of the halogenation reaction was less than ideal, perhaps owing to decomposition of PClS by trace amounts of water in the reaction medium. Further, the use of poor quality PCIS (old reagent) reduced the yield. Although S0C12 reacts rapidly with DBD-Pro under mild conditions, side-reactions occur that produce various com- pounds. Fluorescence Characteristics of the Derivatives The fluorescence excitation and emission spectra of the diastereomers were measured in acetonitrile-water (1 + 1).The excitation and emission maxima of the three diastereomers are essentially the same (appoximately 450 and 560 nm, respectively). The emission maxima shift slightly towards the blue as the concentration of acetonitrile in water is increased. The excitation spectrum does not change with variation in the acetonitrile concentration in the medium. The 0 60 120 180 240 Time/min Fig. 2 Time coursc of derivatization of heptan-2-01 enantiomers with the chiral reagents at 80 "C in the presence of pyridine. A , Reaction of (R)-( -j-heptan-2-ol with (K)-( -t)-DBD-Pro-COCI; B, reaction of (S)-( +)-heptan-2-01 with (R)-( +)-DBD-Pro-COCI; C, reaction of (I?)-( -)-heptan-2-01 with (Sj-( -)-DBD-Pro-COCI; and D, reaction of (S)-( +)-heptan-2-01 with (S)-( -)-DBD-Pro-COCI.Elucnt for HPLC consisting of H20-CH3CN (3 + 7) with fluorcscencc detection excitation and emission at long wavelengths are a distinct advantage with biological samples because there is negligible interference from compounds co-extracted with the sample. Derivatization Reaction of Alcohol Enantiomers With the Chiral Reagents The derivatization of alcohol compounds is more difficult than the derivatization of amines, thiols, carboxylic acid, etc., owing to the relatively low reactivity of the hydroxyl function group with electrophiles. Therefore, many alcohol tagging reagents include an acyl halide group as the reactive functional group. Although the acyl chloride group exhibits excellent reactivity with hydroxyls, the reagents are highly reactive with moisture.The proposed chiral tagging reagents are fairly stable as solids. The reaction of the chiral reagents described in this paper requires a hydrogen chloride scavenger such as pyridine or triethylamine. The yield of the derivatization reaction without such a reagent was very low (<lo%) compared with the reaction in the presence of pyridine. Among various reagents, pyridine provides an optimum combination of low nucleophilicity and adequate basicity (pK, = 5.19). Good solubility in non-polar solvents such as benzene is another advantage. Therefore, pyridine was added to the reaction medium in the following experiments. It was necessary to test the reactivity of each enantiomer of the reagent towards each enantiomer of the alcohol because differences in reactivity could give mixtures of diastereomers that would not accurately reflect the isomeric composition.Therefore, the reactivity of the optically active reagents [(S)-( -)- and (R)-( +)-DBD-Pro-COCI] towards (I?)-(-)- and (S)-( +)-heptan-2-01, which were selected as representative enantiomers, was examined in benzene at 80 "C. As shown in Fig. 2, the formation of the derivatives increased with time.762 ANALYST, JULY 1993, VOL. 118 The derivatization reactions with both enantiomers were almost complete after 150 min. The peak height derived from (S)-( -)-DBD-Pro-COCI and (S)-( +)-heptan-2-01 was slightly higher than that of the other diastereomer, derived from (S)-( -)-DBD-Pro-COCI and (I?)-( -)-heptan-2-01. The same slight difference in the rate of formation of the diastereomers with (R)-( +)-DBD-Pro-COC1 was confirmed.Judging from the reaction curves in Fig. 2, the reactivities of DBD-Pro- COCl enantiomers are essentially the same for both enantio- mers of heptan-2-01. Therefore, a 3 h reaction period at 80 "C in benzene was selected for the derivatization of alcohol enantiomers with the chiral reagents. Resolution of Alcohol Enantiomers by HPLC The applicability of the chiral derivatization reagents to the HPLC separation of alcohol enantiomers was investigated. As intermolecular hydrogen bonding between the derivative and the stationary phase not only contributes to fixation of the Table 1 HPLC separation of diastercomers derived from (S)-( -)- DBD-Pro-COCl by normal-phase chromatography.Column, Inertsil SIL (150 x 4.6 mm i.d.; 5 pm) at 40°C; eluent, (A) hexane-ethyl acetate (80 + 20) and (B) hexane-ethyl acetate (85 + 15); flow rate, 1.0 ml min-l, fluorcsccnce detection, he, = 450 nm, he, = 560 nm (S)-Enantiomer (R)-Enantiomer Alcohol tRlmin k' tR/min k' a R,T Eluent Hexan-2-01 10.90 19.81 Heptan-2-01 10.00 17.81 Nonan-2-01 9.07 14.49 1-Phenyl- 15.85 ethanol 30.08 9.90 12.83 18.81 23.81 9.00 12.08 16.81 21.99 8.07 11.33 13.49 18.69 14.85 19.04 29.08 37.13 11.83 22.81 11.08 20.99 10.33 17.69 18.04 36.13 1.20 2.56 A 1.21 3.34 B 1.23 2.86 A 1.25 3.49 B 1.28 3.12 A 1.38 4.01 B 1.22 3.49 A 1.24 4.48 B Table 2 HPLC separation of diastereomers derived from (R)-( +)- DBD-Pro-COCI by normal-phase chromatography. Conditions as in Table 1 (S)-Enantiomer (R)-Enantiomcr Alcohol tRlmin k' tRlmin k' a R,T Eluent Hexan-2-01 12.87 23.86 Heptan-2-01 12.13 22.13 Nonan-2-01 11.37 19.26 1-Phenyl- 19.58 ethanol 37.27 11.87 22.86 11.13 21.13 10.37 18.26 18.58 36.27 10.93 19.81 10.05 17.93 9.10 14.99 16.26 30.19 9.93 18.81 9.05 16.93 8.10 13.99 15.26 29.19 1.20 1.22 1.23 1.25 1.28 1.31 1.22 1.24 2.67 A 3.37 B 2.86 A 3.50 B 3.13 A 3.97 B 3.63 A 4.43 B 2 0 10 20 0 10 20 conformation but also is important for the efficient resolution of the diastereomers, the normal-phase column is usually employed together with organic solvents as the mobile phase.Therefore, the separation of each pair of alcohols was attempted by normal-phase chromatography with non-polar solvents. The capacity factors ( k ' ) , separation factor (a) and resolution (R,) for each pair of diastereomers are given in Tables 1 and 2.As can be seen, the four alcohols tested were well resolved by the Inertsil SIL column with hexane-ethyl acetate eluent. The R, values obtained from the alcohols having higher hydrophobicity, e . g . , nonan-2-01, were larger than those for the alcohols having higher hydrophilicity, e.g., hexan-2-01. The polar compounds, including the hydrolysate of the derivatization reagent, eluted later than the diastere- omers. When (S)-( -)-DBD-Pro-COCl was used as the chiral derivatization regent, the corresponding diastereomers of the (S)-enantiomers of the alcohols eluted more rapidly than the (R)-enantiomers. As expected, the opposite results were observed with the use of (R)-(+)-DBD-Pro-COCI. No excep- tions were observed among the pairs of enantiomers.Typical normal-phase chromatograms of the resulting diastereomers formed with (S)-(-)-DBD-Pro-COCI are depicted in Fig. 3. The results described above suggest that the formation of hydrogen bonds between the stationary phases and the diastereomers, derived from alcohol enantiomers and the derivatization reagents, plays an important role in the separations. Although the complete resolution of alcohol enantiomers was achieved by normal-phase chromatography, this technique may not be suitable for biological specimens because of sample handling difficulties and the use of harmful organic solvents. Therefore, analysis by reversed-phase chro- matography with an aqueous solvent system was investigated. The diastereomers formed with nonan-2-01 and l-phenyl- ethanol are not separated, whereas there is some resolution of the diastereomers from hexan-2-01 and heptan-2-01 by re- versed-phase chromatography with water-acetonitrile (Table 3).The elution orders were the same as those obtained by normal-phase chromatography; the (S)-enantiomers eluted faster than the (R)-enantiomers with the use of (S)-(-)-DBD- Pro-COCI and the (R)-enantiomers eluted faster than the (S)-enantiomers with the use of (R)-( +)-DBD-Pro-COCl. The proposed chiral derivatization reagents provided excel- lent resolution of alcohol enantiomers by normal-phase HPLC. The detection limits of the alcohols are in the sub-picomole range. The proposed method is satisfactory for the determination of alcohol enantiomers because no racemi- zation occurs. As the elution order of enantiomers can be changed using different enantiomers of the chiral reagent (Tables 1 and 2), the determination of trace amounts of one enantiomer in the presence of a much greater amount of the other is easily accomplished.Moreover, the long-wavelength excitation and emission maxima of the derivatives might be an 5 6 7 8 0 10 20 0 10 20 Time/m in Fig. 3 Chromatograms obtained from the reaction with (S)-( -)-DBD-Pro-COCl by normal-phase chromatography. Separation of the resulting diastercomers: (a), hcxan-2-01; ( b ) , he tan-2-01; (c), nonan-2-01; (d), 1-phenylethanol. Peak, derivative from: ( 1 ) (S)-(+)-hexan-2-01, [2] (R)-( -)-hexan-2-01, (3) (S)-( + 7-h eptan-2-01, (4) (R)-( -)-heptan-2-01, (5) (S)-( +)-nonan-2-01, (6) (R)-( -)-nonan-2-o11 7 (S)-( -)-1-phenylethanol, (8) (R)-( +)-1-phenylethanol.Eluent for HPLC consisting of hexane-benzene (80 + 20) with fluorescence detectionANALYST, JULY 1993. VOL. 118 763 Table 3 HPLC separation of diastereomers dcrivcd from (R)-(+)- DBD-Pro-COCI by reversed-phase chromatography. Column: Incrt- sil ODS-80A (150 X 4.6 mm i d . ; 5 pm) at 40°C; eluent, (A) CH3CN (25 + 75) and (D) H20-CH3CN (30 + 70); flow rate, 1 .0 ml min-I. fluorescence detection, he, = 450 nm, he, = 560 nm H20-CH3CN (40 + 60), (B) HzO-CH~CN (45 + 55), (C) H20- (S)-Enantiomer (I?)-Enantiomer Alcohol tRlmin k’ tRlmin k’ a R, Eluent Hexan-2-01 14.64 22.95 Heptan-2-01 20.96 35.10 Nonan-2-01 12.18 18.61 1-Phenyl- 8.71 ethanol 13.09 13.64 21.95 19.96 34.10 11.18 17.61 7.71 12.09 14.11 22.00 20.43 34.02 12.18 18.61 8.71 13.09 13.11 21 .00 19.43 33.02 11.18 17.61 7.71 12.09 1.04 0.65 A 1.05 0.86 B 1.03 0.46 A 1.03 0.64 B 1.0 0 C 1.0 0 D 1.0 0 A 1.0 0 B advantage for detection with biological samples, because the likelihood of interferences from intrinsic compounds, which emit at shorter wavelengths, would be reduced.As the excitation wavelengths of the derivatives are close to that of the argon ion laser light emission, the determination of alcohol enantiomers at the attomole level should be possible with laser-induced fluorescence detection. Further, the reagents may also be preferable for more sensitive detection by peroxyoxalate chemiluminescence, hence attomole detection limits could be achieved for alcohol enantiomers.The proposed method might therefore serve for the resolution of alcohol enantiomers in biological fluids. As various types of reagents with acyl chloride groups are easily synthesized with the optically active DBD-amino acid (DBD-AA), excellent reagents such as DBD-Pro-COCI may be developed by screening various DBD-AA-COCI compounds. Further studies concerning the resolution of racemic drugs such as P-blockers are in progress. The authors thank Dr. C. R. Warner, Food and Drug Administration, Washington, DC, for reviewing the manus- cript. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 References Nambara, T., and Goto. J . , Runseki Kagaku, 1974, 23, 704. Oi, N., Bunseki Kagaku, 1984, 12, E401. Armstrong, D. W., Li, W., and Pitha, J., Anal. Chem., 1990, 62, 214. Armstrong, D. W., Li, W., Chang, C . D., and Pitha, J . , Anal. Chem., 1990, 62,914. Chromatographic Chiral Separation, eds. Zief, M., and Crane, L. J., Marcel Dckker, New York, 1988. Chiral Separation by HPLC, ed. Krstulovic, A. M., Ellis Horwood, Chichester, 1989. Pirklc, W. H., and Pochopsky. T. C., Chem. Rev., 1989, 89, 347. Nishi, H., Fukuyama, T., and Terabe, S . , J. Chromatogr., 1991, 553, 503. Soini, H., Riekkola, M.-L., and Novotony, M. V., J. Chroma- togr.. 1992, 608, 265. Cruzado, I. D., and Vigh, Gy., J. Chromatogr., 1992,608,421. Brooks, C. J. W., Gilbert, T . , and Gilbert, J. D., Anal. Chem., 1973, 45, 896. Takasu, A . , and Ohya, K., J . Chromatogr., 1987, 389, 251. Sasaki, K., and Hirata, H., J. Chromarogr., 1991, 585, 117. Goto, J., Goto, N., Shao, G., Ito, M., Hongo, A . , Nakamura, S . , and Nambara, T., Anal. Sci., 1990, 6, 261. Goto, J . , Shao, G . , Ito, M., Kuriki, T., and Nambara, T., Anal. Sci., 1991, 7, 723. Toyo’oka, T., Ishibashi, M.. and Terao, T., Analyst, 1992,117, 727. Toyo’oka, T., Ishibashi, M., and Terao, T., J. Chromatogr., 1992, 625,357. ‘Toyo’oka, T., Ishibashi, M., and Terao, T . , J. Chromatogr., 1992, 627,75. Toyo’oka, T., Ishibashi. M., and Tcrao. T., Anal. Chim. Acta, 1993, 278, 71. Toyo’oka, T.. Suzuki, T., Saito, Y.. Uzu, S . . and Imai, K., Analyst, 1989, 114, 1233. Popenoe, E. A., and du Vigneaud, V.. J. A m . Chem. SOC., 1954, 76, 6202. Hashimoto, S., Kase, S . , Suzuki, A., Yanagiya, Y . , and Ikegami, S . , Synth. Commun., 1991, 21, 833. Eckert, H., and Forster, B., Angew. Chem., Znt. Ed. Engl., 1987, 26, 894. Paper 3100743 J Received February 8, 1993 Accepted March 26, 1993

ISSN:0003-2654    

DOI:10.1039/AN9931800759

出版商:RSC    

年代:1993

数据来源: RSC

摘要:

ANALYST. JULY 1993, VOL. 118 765 Determination of Carboxylic Acids by High-performance Liquid Chromatography With 2-(2,3-Anthracenedicarboximido)ethyl Trifluoromethanesulfonate as a Highly Sensitive Fluorescent Labelling Reagent Kazuaki Akasaka, Hiroshi Ohrui" and Hiroshi Meguro Department of Applied Biological Chemistry, Faculty of Agriculture, Tohoku University, I-? Tsutsumidori-Amamiyamachi, Aoba-ku, Sendai 981, Japan 2-(2,3-Anthracenedicarboximido)ethyl trifluoromethanesulfonate (AE-OTf) is a highly sensitive fluorescent labelling reagent for carboxylic acids for use in liquid chromatography. The labelling reaction of carboxylic acids with AE-OTf was completed within 10 min at room temperature in acetonitrile in the presence of tetraethylammonium carbonate as a base. The 2-(2,3-anthracenedicarboximido)ethyl esters of 18 kinds of fatty acids including polyunsaturated fatty acids were separated from each other on an ODS column with an isocratic solvent system within 30 min.The relative standard deviations of their peak areas were within 2% ( n = 8, 1.4-3.3 pmol on-column). Their relative peak areas were almost 1.0 and their detection limits were 0.8-2.7 fmol (signal-to-noise ratio = 3). These fatty acids were successfully determined at least over the range 20 fmol-6 pmol (correlation coefficient r > 0.999 for all fatty acids tested). Keywords: 2-(2,3-Anthracenedicarboximido)eth yl trifluoromethanesulfonate; high-performance liquid chromatography; fluorimetric detection; carboxylic acid Carboxylic acids are widely distributed in nature and play important roles in living organisms.As the carboxyl functional group is a weak chromophore, it is necessary to derivatize carboxylic acids for their sensitive determination by high- performance liquid chromatography (HPLC). Ultraviolet labelling reagents have been reported for this purpose,l-4 but their sensitivities were unsatisfactory. Recently some fluor- escent labelling reagents, e.g., 4-bromomethyl-7-methoxy- coumarin,s 4-bromomethyl-7-acetoxycoumarin,6 9-bromo- methylacridine,7 I-bromoacetylpyrene,X 9-anthryldiazo- methane,g 3-bromomethyl-6,7-dimethoxy-l-methyl-2(lH)- quinoxalinone10 and 2-(2,3-naphthalimido)ethyl trifluoro- methanesulfonate (NE-OTf),ll have been developed for the highly sensitive and selective determination of carboxylic acids by HPLC.The recent development of NE-OTf by Yasaka et al. 1 1 prompted us to synthesize an analogous reagent with one more aromatic ring in its structure in the expectation of obtaining a more sensitive reagent. In this work , 2-(2,3-anthracenedicarboximido)ethyl tri- fluoromethanesulfonate (AE-OTf) (Fig. 1) was developed as a highly reactive and sensitive fluorescent labelling reagent for carboxylic acids for use in HPLC. Palmitic acid was chosen as a representative carboxylic acid and the reactivity of AE-OTf towards this acid was studied. Eighteen kinds of fatty acid derivatives were separated by HPLC with an ODS column. Experimental Apparatus The HPLC system consisted of a CCPD pump (Tosoh, Tokyo, Japan), a Rheodyne (Cotato, CA, USA) Model 7125 injection valve, an FS 8010 spectrofluorimeter (Tosoh) and a Chromat- ocorder 12 integrator (System Instrument, Tokyo, Japan).The analytical column (100 X 4.6 mm i.d.) contained Develo- sil ODs-K3 (3 pm) (Nomura Chemical, Aichi, Japan). A Paratherm U2 electronic water-bath (Juchheim Labortechnik, Schwarzwald, Germany) was used to control the column temperature. A JASCO (Tokyo, Japan) FP-550A spectroflu- * To whom correspondence should be addressed. orimeter was used for the measurement of fluorescence spectra. Chemicals Solvents such as methanol, water and acetonitrile were of HPLC grade from Kanto Chemical (Tokyo, Japan). Other chemicals were purchased from Wako (Osaka, Japan). Fatty acids used as standards were purchased from Sigma (St. Louis, MO, USA). Tetraethylammonium carbonate (TEAC) was prepared as described previously .7 Preparation of AE-OTf The AE-OTf was synthesized by modifying the method for NE-OTf1 1 from 2,3-anthracenedicarboxylic anhydride, which was prepared from 2,3-anthracenedicarboxylic acid by reflux- ing in acetic anhydride for 2 h (90% from 2,3-anthracenedicar- boxylic acid). 2,3-Anthracenedicarboxylic acid was prepared from 1,2,4-trimethylbenzene and benzoyl chloride according to the reported method.12 2,3-Anthracenedicarboxylic anhydride (0.2 g), 2-amino- ethanol (0.2 g), dry toluene (60 ml) and butanol (30 ml) were placed in a 200ml flask fitted with a Dean-Stark trap and reflux condenser and the mixture was refluxed vigorously for 1.5 h.The solvent was removed under reduced pressure until crystallization took place.The crystals were dissolved com- pletely by warming and then cooled to give further crystals. Recrystallization gave N-(hydroxyethyl)-2,3-anthracenedicar- boximide as orange crystals from toluene-butanol; yield 88% ; m.p. 292-294 "C; mass spectrum, mlz = 291; analysis, calc. for CI8Hl3NO3, C 74.2, H 4.50, N 4.81; found, C 74.0, H 4.67, N 4.77%. A mixture of dry pyridine (1 ml) and a suspension of the imide (200 mg) in dichloromethane (50 ml) was carefully added dropwise to a solution of trifluoromethanesulfonic anhydride (0.4 g) in dichloromethane (30 ml) at a rate such as to keep the reaction temperature below -5°C. The mixture was stirred for 3 h below -5 "C, then washed with 200 ml of cold water, separated and the organic layer was dried over anhydrous magnesium sulfate.The solvent was evaporated under reduced pressure to give a crystalline mass that was766 ANALYST, JULY 1993, VOL. 118 AE-QTf 8 6 -0Tf = -0-S-CCF3 Fig. 1 Reaction of a carboxylic acid with AE-OTf Table 1 Concentrations of AE-OTf and TEAC solutions Concentration of fatty acids/pmolI-l AE-OTf/mmol1-1 TEAClmmol 1-1 =30 1 .o 0.5 ~ 3 . 0 0.5 0.25 4 . 3 0.2 0.1 recrystallized from dichloromethane at -20 "C to give pale yellow crystals; yield 74% ; m.p. >300 "C; mass spectrum, mlz = 423; IR, 1200 and 1400 cm-1 (-O-S02-); analysis, calc. for C19H12NS05F3, C 53.9, H 2.86, N 3.31, S 7.57; found, C 53.6, H 3.09, N 3.21, S 7.44%. The total yield from 1,2,4- trimethylbenzenc was 16.7%. These crystals were used without further purification.Derivatization Procedure A typical derivatization procedure was as follows. A 0.05 ml volume of TEAC in acetonitrile was added to 0.1 ml of test solution in acetonitrile in a brown reaction vial. Then, 0.05 ml of AE-OTf, which was freshly prepared as an acetonitrile solution just before use, was added and the mixture was vortex mixed for 10 s and then allowed to stand for 10 min at room temperature. A 1 pl aliquot was injected directly into the HPLC system. Table 1 shows the concentrations of fatty acids, AE-OTf and TEAC solutions. It should be noted that the molar ratio of AE-OTf to TEAC should be kept at 2 : 1. Separation of Fatty Acid Derivatives The HPLC separation of 18 kinds of fatty acid esters was performed on a Develosil ODs-K3 column with methanol- acetonitrile-water (135 + 45 + 20 v/v/v) as eluent at a flow rate of 0.8 ml min-1.The column temperature was kept at 55 "C with an isothermal water-bath. The fluorescence inten- sity at 456 nm (excitation at 298 nm) was monitored. Results and Discussion The AE-OTf was prepared from 1,2,4-trimethyIbenzene and benzoyl chloride in 16.7% yield. This crystalline reagent was stable for more than 6 months at -20 "C. Fig. 2 shows the excitation and emission spectra of 2-(2,3- anthracenedicarboximido)ethyl palmitate in methanol. The effects of solvents on the fluorescence spectra and intensities are shown in Table 2. The fluorescence intensity in aceto- nitrile was 15% stronger than that in methanol. The emission wavelength in acetonitrile was shorter than that in methanol.An increase in the water content in both methanol-water and acetonitrile-water solutions caused a red shift of the emission wavelength and a decrease in fluorescence intensity. As there is only a 20% difference in the fluorescence intensities in acetonitrile and in methanol-water (6 + 4 v/v), these solvents were used as mobile phase components. Potassium carbonate-crown ether or potassium fluoride- crown ether is usually used in the derivatization reactions of 220 260 300 340 390 420 450 480 510 540 Wavelengthlnm Fig. 2 Excitation spectrum and (b) emission spectrum Fluorescence spectra of A E palmitate in methanol. (a) Table 2 Effect of solvents on fluorcsccncc of AE palmitate Solvent CHSOH CH30H-H20 (v/v)- 90+ 10 80 + 20 70 + 30 60 + 40 CHSCN CHRCN-H20 (v/v)- 90+ 10 70 + 30 (135 + 45 + 20 v/v/v) CH3OH-CH3CN-HZO Maximum wavelengthhm Excitation Emission 298 45 5 298 460 298 462 298 465 298 467 295 437 298 450 298 459 298 456 Relative fluorescence intensity (n = 5) 1 .oo 0.99 0.98 0.93 0.90 1.15 1.12 1.08 1.01 carboxylic acids with benzyl halide reager,ts"b.8>10 and NE- OTf.11 We have reported previously that TEAC was as good as the crown ether systems in the derivatization reaction of free fatty acids with benzyl halide reagents.' In this study also, TEAC was used as a base instead of the crown ether systems.Fig. 3 shows the effects of the concentrations of TEAC and AE-OTf solutions on the peak area of the palmitate. The TEAC and AE-OTf were added so as to keep their molar ratio at 1 : 2 throughout the experiments. Solutions of 1.0,0.5 and 0.2 mmol I--' AE-OTf were satisfactory for labelling less than 30, 3.0 and 0.3 pmol 1-1 of fatty acid solutions, respec- tively.Fig. 4 shows the time courses of the reaction of AE-OTf with palmitic acid. In each instance, the peak areas reached a plateau within 10 min at room temperature in acetonitrile. The reactivity of AE-OTf was almost identical with those of NE-OTf and 9-bromomethylacridine, which are very reactive reagents among the available fluorescence labelling reagents for carboxylic acids. It is well known that polar aprotic solvents such as dimethyl sulfoxide and dimethylformamide accelerate the reactivity in SN2 reactions. However, these solvents were not suitable for AE-OTf because they produced many unknown by-products, probably owing to the high reactivity of AE-OTf in these solvents.ANALYST, JULY 1993, VOL.118 767 100 8 I m (D m a a, rn L Y 75 .- + 0 0.1 0.2 A 0 0.25 0.5 B 0 0.5 1.0 c AE-OTf/mmol I 1 Fig. 3 Effects of the concentrations of AE-OTf and TEAC on peak area. Concentrations of palmitic acid solutions werc: A, 0.3; B, 3.0; and C, 30 pmol I-'. TEAC was addcd to kccp the AE-OTf to TEAC molar ratio to 1 : 2 100 I s ? m~ 80 - m Y m a, a a, ._ c.' 60 K 40 0 10 20 30 Time/m in Fig. 4 Reaction time courses of AE-OTf with palmitic acid. Concentrations of palmitic acid, AE-OTf and TEAC solutions were: 30 pmol 1-1, 1 mmol 1-1 and 0.5 mmol 1-1, 0 3.0 urnoll-1, 0.5 mmol 1-I and 0.25 mmol 1-1; and A, 0.3 pmol 1-1, 0.2 mmol 1-1 and 0.1 mmol 1-1, respectively The calibration graph for palmitic acid showed good linearity at lcast between 15 fmol and 60 pmol [y = 2560x - 30.3, where x = concentration of the acid (fmol) and y = peak area; r = 0.99991.The detection limit of palmitate was 0.5 fmol (signal-to-noise ratio = 3, separation on Develosil ODS-K3 at 50 "C with methanol as eluent at 0.6 ml min-1, detection by fluorescence intensity at 458 nm with excitation at 298 nm). Under these conditions, palmitate was eluted at 6.7 min. The result shows that AE-OTf is a very sensitive fluorescence reagent for carboxylic acids. The resultant ester was stable for more than 7 d at room temperature. Fig. 5 shows a typical HPLC trace for 18 derivatives of fatty acids, most of which are important components in many biological materials. All of them were separated within 30 min.Table 3 shows their peak areas relative to that of myristate, detection limits and relative standard deviations of their peak areas for eight runs. As the peaks of linolenate and docosahexanoate were not separated completely from that of myristate, first their peak areas relative to that of caprate were obtained and then they were corrected to myristate. Their relative peak areas were almost the same as that of myristate (mean k s = 0.988 k 0.07). This means that the AE derivatives of fatty acids have almost the same fluorescence intensity, and this allows us to calculate the fatty acid compositions from the ratios of their peak areas. The concentrations of fatty acids were also calculated from their peak areas using a suitable internal standard such as hcpta- 19 t - m C rn v) .- 0 7 14 21 28 Time/mi n Fig.5 Chromatogram of fatty acid derivatives. HPLC conditions as described under Expcrimental. Thc peaks wcre: 1. AE caprylatc (23 fmol); 2, caprate (31 fmol); 3, laurate (26 fmol); 4, myristoleate (30 fmol); 5, cis-5,8,11,14,17-cicosapentacnoatc (24 fmol); 6, linolen- ate (31 fmol); 7, myristate (26 fmol); 8, ci~-4,7,10,13.16,19-doco- sahexaenoate (27 fmol); 9, palmitoleatc (25 fmol); 10, arachidonate (27 fmol); 11, linoleate (27 fmol); 12, cis-8,11,14-eicosatrienoate (26 fmol); 13, palmitate (27 fmol); 14, oleatc (29 fmol); 15. cis-11,14- eicosadienoate (21 fmol); 16, heptadecanoate (22 fmol); 17, stearate (22 fmol); and gondoate (33 fmol). Peak 19 was derived from AE-OTf Table 3 Relative peak areas, detection limits and reproducibilities for fatty acid derivatives Peak No.Fatty acid RPA* 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Capry 1 ic Capric Lauric Myristoleic cis-5 ,s, 1 1 ,14,17- Linolenic Myristic Eicosapentaenoic ~i.~-4,7,10,13.16,19- Docosahexaenoic Palmitoleic Arachidonic Linoleic cis-8,11,14-Eicosatrienoic Palmitic Oleic cis-l1,14-Eicosadienoic Heptadecanoic Stearic Gondoic 0.91 0.98 0.92 1.09 0.88 0.99 I .00 0.92 1.00 1.09 0.94 0.99 1.06 0.97 1.06 0.91 I .08 1 .00 Relative peak area (n = 9). + Detection limit. f Relative standard deviation ( a = 8). D L+/ fmol 0.8 1 .0 1.1 1.2 1.3 1.4 1.4 1.5 1.6 1.6 1.6 1.8 1.8 2.0 2.2 2.5 2.7 2.7 Amount RSD injected/ (%)t pmol 1.04 3.2 1.29 1.9 0.92 1.6 1.03 1.4 0.97 2.2 1.20 1.8 1.04 1.5 1.27 1.4 0.59 2.3 0.75 3.8 0.68 1.7 1.02 2.7 0.66 1.7 0.90 1.5 1.77 2.1 0.90 1.8 0.43 1.6 0.74 1.6 decanoic acid.Using the proposed method, sub-femtomole to femtomole levels of fatty acid derivatives could be detected, and the correlation coefficients of their calibration graphs were >0.999 between 20 fmol and 6 pmol with good repro- duci bili t y . Conclusions The AE-OTf was developed as a highly reactive fluorescence labelling reagent for carboxylic acids. As was expected, it is more sensitive than NE-OTf and most other labelling reagents.768 ANALYST, JULY 1993. VOL. 118 This work was supported in part by a Grant-in-Aid for Scientific Research from the Ministry of Education, Science and Culture of Japan. References 1 Miller, R. A.. Bussell. N. E., and Ricketts. C., J . Liq. Chrornatogr., 1978, 1. 291. 2 Knapp, D. R., and Krueger, S., Anal. Lett., 1975, 8, 603. 3 Matthess, D. E., and Purdy, W. C., Anal. Chirn. Actrr, 1979, 109, 61. 4 Ingalls, S. T., Minkler, P. E., Hoppel. C. L., and Nordlander. J . E . , J . Chrornatogr., 1984, 299. 365. 5 Lam, S . , and Grushka, E., J . Chrornatogr., 1978, 158, 207. 6 Tsuchiya. H . , Hayashi, T.. Naruse, H., and Takagi, N., J . Chrornatogr.. 1982, 234, 121. 7 8 9 10 11 12 Akasaka, K., Suzuki. T., Ohrui, H., Meguro, H . , Shindo, Y . , and ‘Takahashi, H . , Anul. Lett., 1987, 20, 1581. Kamada, S . , Maeda, M., and Tuji, A., J . Chrornutogr., 1983, 272, 29. Nimura, N., and Kinoshita, T., Anul. Left.. 1980, 13, 191. Yamaguchi. M., Hara. S . , Matsunaga, R.. Nakamura, M.. and Ohkura, Y., J . Chrornutogr., 1985, 346, 227. Yasaka, Y . , Tanaka, M., Shono, T., Tctsumi, T., and Kata- kawa, J . , J . Chrornatogr., 1990, 508, 133. Hallman, J . L., and Bartsch, R . A., J . Org. Chem., 1991, 56, 6243. Paper 2106108B Received November 17, I992 Accepted January 19, 1993

ISSN:0003-2654    

DOI:10.1039/AN9931800765

出版商:RSC    

年代:1993

数据来源: RSC

摘要:

ANALYST, JULY 1993, VOL. 118 769 Highly Sensitive Detection of Non-reducing Carbohydrates by Liquid Chromatography Shuji Yamauchi Central Research Laboratories, SS Pharmaceutical Co. Ltd., I 143 Nanpeidai, Narita-shi, Chiba 286, Japan Chie Nakai, Noriyuki Nimura and Toshio Kinoshita School of Pharmaceutical Sciences, Kitasato University, 9- I Shirokane-5, Minato-ku, Tokyo 108, Japan Toshihiko Hanai International Institute of Technological Analysis, Health Research Foundation, lnstitut Pasteur de Kyoto, 5F, Hyakumanben, Sakyo-ku, Kyoto 606, Japan A highly sensitive post-column reaction detection system was developed for the liquid chromatographic determination of non-reducing carbohydrates. Non-reducing Carbohydrates were oxidized by periodate prior to forming fluorescent compounds by reaction with guanidine in alkaline solution.The calibration graphs were linear between 50 pmol and 10 nmol. Keywords: Non-reducing carbohydrates; guanidine; high-performance liquid chromatography; periodate oxidation; fluorimetric detection Non-reducing carbohydrates, such as sugar alcohols and glycosides , possess few reactive groups and have been detected by measuring their refractive indexl-5 or ultraviolet absorption in the short-wavelength range .+* These methods are non-destructive, but they gave only poor sensitivity. Although gas chromatography”10 has achieved the sensitive detection of these carbohydrates, it requires their preliminary derivatization to give volatile compounds. In comparison with derivatization for liquid chromatography, that for gas chro- matography is complicated because it requires removal of water prior to the reaction and the selection of the reagent is far more restricted because it should add volatility to the substance to be analysed.Pre-column derivatization in liquid chromatography has therefore been applied to these carbohy- drates using perbenzoates”l1-14 or per@-bromobenzoate). 15 Another technique has been reported for the analysis of non-reducing oligosaccharides. Non-reducing oligosac- charides are hydrolysed by passing the effluent from the chromatographic column through a small reaction column packed with a strongly acidic cation exchanger and subse- quently mixed with p-hydroxybenzoid acid hydrazide to give colour, which was measured with a spectrophotometric detector.16 However, the selection of the eluent was limited and the sensitivity was low. Moreover, sugar alcohols could not be detected by this procedure.Non-reducing carbohy- drate could also be detected by colour reactions with formaldehyde, produced by periodate oxidation of sac- charides. 17-19 However, as periodate oxidation and the colour reaction were carried out separately, the procedure was complicated and the sensitivity of this method is only around 1 nmol. A pulse amperometric detector has been used as a highly sensitive saccharide detector. However, the requirement20 for a strongly alkaline mobile phase restricts the separation column, and reducing mono- and oligosuccharides are liable to undergo isomerization and degradation under these condi- tions.In addition, proteins and amino acids are often adsorbed on the electrode and interfere with detection. Guanidine has been used as a fluorimetric detection reagent for reducing carbohydrates with high sensitivity.21 This reagent is very stable in the presence of periodate and seemed promising for the detection of non-reducing carbohydrates. In this work, sodium metaperiodate was added to the guanidine reagent and the optimum conditions for the fluorimetric detection of non-reducing saccharides were studied. Experimental Reagents and Materials Samples of carbohydrates, guanidine hydrochloride and other reagents were purchased from Nakalai (Kyoto, Japan). The guanidine-periodate reagent was prepared by dissolving 50 mmol of guanidine hydrochloride and 0.5 mmol of sodium metaperiodate in 0.1 mol I-’ of potassium tetraborate to make 1 1 and the pH was adjusted to 11.5 with sodium hydroxide.Chemically bonded 5 pm propylamine-bonded silica gel (Nucleosil 5-NH2) and an octadecyl-bonded silica gel (Develosil ODs-5) obtained from Macherey-Nagel (Diirer, Germany) and Nomura Kagaku (Seto, Japan), respectively, were packed in 100 and 200 x 6 mm i.d. stainless-steel columns, respectively, by a conventional slurry packing technique. A pre-packed Shodex Ionpak KS-801 column (300 x 8 mm i.d.) was purchased from Showa Denko (Tokyo, Japan). Apparatus This system was assembled from a Shimadzu (Tokyo, Japan) Model LC-5A high-performance liquid chromatography (HPLC) pump, a Rheodyne Model 7125 injector, a Shimadzu CRB-3A reaction bath, a Shimadzu RF-530 fluorescence detector and a Sanuki (Tokyo, Japan) high-pressure reagent pump.The flow diagram for the liquid chromatography for the guanidine-periodate system was the same as that for reducing saccharides reported previously.21 The Nucleosil and Develo- sil columns were eluted with aqueous acetonitrile at a flow rate of 1.0 ml min-* at ambient temperature. The Shodex Ionpak column was thermostated in a water-bath at 70°C, and the flow rate of the eluent (water) was 0.6 ml min-1. The flow rate of guanidine-periodate was 1.0 ml min-l for the Nucleosil and Develosil columns and 1.2 ml min-’ for the Shodex Ionpak column. The stainless-steel reaction tube was 10 m x 0.8 mm i.d., and the stainless-steel cooling tube was 3 m X 0.25 mm i.d. Results and Discussion Excitation and fluorescence spectra were measured for the sorbitol-guanidine-periodate reaction product.Sorbitol was added to the guanidine-periodate reagent to yield a 1 mmol l-1 solution. This mixture (4 ml) was heated in a sealed770 ANALYST, JULY 1993, VOL. 118 test-tube for 30 min on a boiling water-bath. The tube was rapidly cooled and the spectra were measured. The excitation and fluorescence maxima were at 314 and 433 nm, respec- tively. These spectra were almost identical with those obtained for the reaction of glucose with the guanidine reagent without periodate. Therefore, 31.4 and 433 nm were used as excitation and fluorescence wavelengths in further studies. The effect of the periodate concentration in the guanidine- periodate reagent was studied for several non-reducing carbohydrates (sucrose, raffinose, stachyose, sorbitol, arabi- to1 and a-methylglycoside). The fluorescence intensities reached their maxima between 0.5 and 1.0 mmol 1-I.Arabitol was easily reacted at low periodate concentrations and a- methylglycoside showed poor fluorescence intensity at low periodate concentrations, reaching a plateau at a concentra- tion of 2 mmol I - ' . In order to ensure sensitivity for aldoses, 0.5 mmol I- of sodium metaperiodate was selected for further studies. Periodate proved to be effective at far lower concentrations compared with the method for the generation of formaldehyde to be submitted to colour reactions.17-19 The chromatogram showed only a small blank fluorescence and a flat baseline, hence ethanol or acetonitrile can be used as a component of the eluent.The effect of the pH of the guanidine reagent on the fluorescence intensity of non-reducing carbohydrates was examined using 0.5 mmol I-I metaperiodate and 50 mmol I-' guanidine. The fluorescence intensities of the non-reducing oligosaccharides and sugar alcohols were maximum at pH 11.5, except for a-methylglycoside, which showed a maximum at pH 9.5. This result differed from that obtained for reducing saccharides where the relative peak height increased with increasing pH and was maximum at pH 12, except for maltopentaose and N-acetylneuraminic acid, whose intensity increased further at higher pH. From these results, a pH of 11.5 was adopted for the simultaneous determination of reducing and non-reducing carbohydrates, periodate oxida- tion being allowed to proceed under alkaline conditions.The fluorescence intensity of non-reducing carbohydrates associated with guanidine concentrations up to 0.2 mol I-] was examined. The fluorescence intensity gradually increased and reached a plateau at a guanidine concentration of 50 mmol I-'. This concentration was about twice that required for reducing saccharides. The guanidine concentration did not affect the Table 1 Relative fluorescence intcnsitics (RF) of carbohydrates measured by thc flow injection method. Carrier, 0.6 ml min-l of water; reagent solution, 1.0 ml min-' of 50 mmoll-' guanidine, (pH 11.5) containing 0.5 mmol I-' sodium metapcriodatc and 0.1 rnol I-' potassium tetraborate; reaction temperature, 1 50 "C Carbohydrate RF Carbohydrate RF D-XyloSe D-Ribose D- Arabinose D-Glucose D-Galactose D-Mannose L-Fucose L-Rhamnose 2-Dcoxy-D-glucose 2-Dcoxy-~-ribose D-GIucosamine.HCI N-Acetyl-D-glucosamine D-Glucuronic acid D-Gluconic acid D-Sorbito 1 Inositol Arabitol D-Fructose 100 23 52 77 85 6.5 109 43 38 544 190 20 1 91 125 171 63 29 23 Mannitol 32 Maltose 323 Isomaltose 167 Gentiobiosc 219 Lactose 253 Maltotriose 475 Maltotetraose 620 Maltopentaose 632 Maltohcxaose 677 Sucrose 62 Trehalosc 47 Raffinose 149 Stach yose 230 N- Aceylneuraminic acid 772 a- Methylglucoside 152 p-Nitrophcny1-a- 196 Phenyl-a-glucoside 57 glucop yranoside sensitivity in the concentration range 25-75 mmol I-' for reducing saccharides.The optimum concentration is about 50 mmol I-', except for a-methyl glycoside, which required lower concentrations than the others. The effect of borate concentration was examined up to 0.2 rnol 1 - I .Non-reducing saccharides required lower concentra- tions of borate; the fluorescence intensity gradually reached a maximum in 0.05 rnol I-' borate, and 0.1 rnol was sufficient to optimize the detection system. Reducing saccharides required a minimum of 0.1 rnol of borate. Periodate did not affect the concentration profiles of guanidine and borate. This suggests that guanidine is not consumed by periodate. Colorimetric detection of sugars involving periodate oxida- tion has so far required a dual-pump system to deliver the periodate and chromogenic reagent separately; the column effluent was first oxidized with periodate and then mixed with the chromogenic reagent.In contrast to this complexity, a single delivery of the post-column reagent sufficed for detection in the present method. Hence the rapid and sensitivc detection of non-reducing carbohydrates with a simple detec- tion system was accomplished . The effect of temperature on the chemical reaction was examined in order to obtain the maximum fluorescence intensity. The optimum temperature for reducing saccharides was about 135 "C, except for N-acetylneuraminic acid and maltopentaose. The fluorescence intensity of all non-reducing carbohydrates increased with increasing reaction tempera- ture, except for sorbitol and arabitol, for which the optimum temperature was about 120 "C. The sensitivity of other non- reducing saccharides increased with the increase in the reaction temperature.The reaction temperature was, how- ever, set at 150°C for ease of operation of the system. The calibration graphs showed excellent linearity for sample concentrations between 50 pmol and 10 nmol and C A 0 10 Time/min Fig. 1 Chromatogram of a mixture of 2 nmol each of: A, a- cyclodextrin; B, (3-cyclodextrin; and C, y-cyclodextrin, separated on a Nucleosil S-NHz column in water-acetonitrile (40 + 60)ANALYST, JULY 1993, VOL. 118 77 1 passed through the origin. The relative standard deviations for 1 nmol per sample of sorbitol, sucrose and stachyose were 3.17, 3.36 and 1.64% (n = lo), respectively. The relative fluorescence intensities of various carbo- hydrates measured by flow injection are given in Table 1. N- Acetylneuraminic acid, 2-deoxy-~-glucose and 2-deoxy-~- ribose showed high fluorescence intensities among the mono- saccharides tested.2-Deoxy-~-glucose and 2-deoxy-~-ribose showed poor fluorescence intensity when periodate was removed from the reagent. These sugars almost always give poor sensitivity in conventional colour or fluorescence reac- tions. The proposed method resolved this problem. Although N-acetylneuraminic acid gave a considerable fluorescence intensity even without periodate, the intensity was greatly enhanced by its addition. These three carbohydrates gave dicarbonylmethane derivatives whereas these could not be formed with the other saccharides tested. These derivatives might be related to the development of fluorescence but further investigations will be necessary for the elucidation of the reaction mechanism.In addition, non-reducing oligosac- charides, sugar alcohols and D-gluconic acid showed peak heights as high as those of aldoses. On the other hand, 6- deoxyhexaoses and pentaoses showed lower peak heights. The detection system was applied to the chromatography of standard non-reducing oligosaccharides, sugar alcohols, a- methylglucoside, phenylglucosides and cyclodextrins. Non- reducing oligosaccharides and sugar alcohols were separated on a Nucleosil 5NH2 column with water-acetonitrile (30 + 70) as eluent, a mixture of reducing and non-reducing saccharides was separated on a Shodex Ionpak KS-801 column with water as eluent and phenylglycosides were separated on a Develosil ODs-5 column with water-acetonitrile (90 + 10) as eluent, with detection using the proposed guanidine-periodate system.A chromatogram of a mixture of 2 nmol each of a-, p- and y-cyclodextrins separated on a Nucleosil 5-NH2 column with water-acetonitrile (40 + 60) as eluent is shown in Fig. 1. 1 2 3 4 9 10 11 12 13 14 15 16 17 18 19 20 21 References Ghias-Ud-Din, M., Smith, A. E. and Phillips, D. V., J. Cheetham, N. W. H., and Sirimanne, P., J. Chromatogr., 1981, McGinnis, G. D., and Fang, P., J. Chromatogr., 1978,153,107. Koizumi, K., Odaka, Y., Horiyama, S . , Utamura, T., Hisa- matsu, M., and Amemura, A., J. Chromatogr., 1983,265, 89. Chromatogr., 1981, 211, 295. 208, 100. Hokuse, H., J . Chromatogr., 1980, 189, 98. Shukla, A. K., and Schauer, R., J. Chromatogr., 1982,244,81. Rajakyla, E., J. Chromatogr., 1981, 218, 695. Laakso, E. I., Tokola, R. A., and Hirvisalo, E. L., J . Sawardeker, J. S . , Sloneker, J . H., and Jeans, A. R., Anal. Haga, H., Imanari, T., Tamura, Z., and Momosc, A., Chem. Daniel, P. F., De Feudis, D . F., Lott, I . T., and McCluer, White, C. A., Kennedy, J . F., and Golding, B. T., Carbohydr. Oshima, R., and Kumanotani, J., J. Chromatogr.. 1983, 265, Lehrfeld, J . , J. Chromatogr., 1976, 120, 141. Golik, J . , Liu, H. W., Dinovi, M., Furukawa, J . , and Vratny, P., and Ouhrabkava, J . , J. Chromatogr., 1980, 210, Carlson, B., Isakssom, T., and Samuelson, O., Anal. Chim. Honda, S., Takahashi, M., Shimada, S . , Kakehi, K., and Aminoff, D., Gathmann, W. D., McLean, C. M., and Lee, Y. C., Kagaku to Kogyo, 1990,43, 953. Yamauchi, S . , Nakai, C., Nimura, N., Kinoshita, T., and Hanai, T., Analyst, 1993, 118, 773. Paper 2106835 D Received December 24, I992 Accepted February 15, 1993 Chromatogr., 1983, 278,406. Chem., 1965,37, 1602. Pharm. Bull., 1972, 20, 1805. R. G., Carbohydr. Res., 1981, 97, 51. Res., 1979, 76, 1. 335. Nakanishi, K., Carbohydr. Res., 1983, 118, 135. 313. Acta, 1968, 43, 47. Ganno, S., Anal. Biochem., 1983, 128, 429. Yadomae, T., Anal. Biochem., 1980, 101, 44.

ISSN:0003-2654    

DOI:10.1039/AN9931800769

出版商:RSC    

年代:1993

数据来源: RSC

摘要:

ANALYST, JULY 1993, VOL. 118 773 Development of a Highly Sensitive Fluorescence Reaction Detection System for Liquid Chromatographic Analysis of Reducing Carbohydrates Shuji Yamauchi Central Research Laboratory, SS Pharmaceutical Co. Ltd., I 143 Nanpeidai, Narita-shi, Chiba 286, Japan Chie Nakai, Noriyuki Nimura and Toshio Kinoshita School of Pharmaceutical Sciences, Kitasato University, 9-1 Shirokane-5, Minato-ku, Tokyo 108, Japan Toshihiko Hanai International Institute of Technological Analysis, Health Research Foundation, lnstitut Pasteur de Kyoto, 5F, Hyakumanben, Sakyo-ku, Kyoto 606, Japan A highly sensitive post-column detection system for reducing carbohydrates in liquid chromatography was developed using guanidine as the reagent. The limit of determination for pentose, 6-deoxyhexose and reducing saccharides was 5 pmol and that for hexose, N-acetylglucosamine and N-acetylneuraminic acid was 10 pmol.The calibration graphs were linear up t o 1 nmol. Keywords: Reducing carbohydrate; guanidine; high-performance liquid chromatography; post-column fluorimetric detection Various fluorescence reagents, such as ethylenediamine, 1 ethanolamine,2 2-cyanoacetamide3 and taurine,4 have been reported for the microdetection of carbohydrates in high- performance liquid chromatography (HPLC) .5,6 However, the sensitivities of these reagents were poor and of approxi- mately the same magnitude as those of spectrophotometric methods using orcinol,7 phenol,x anthrone,g 2,2'-bicinchoni- nate10 or Tetrazolium Blue." This might be partly owing to the recent improvements of ultraviolet-visible detectors which now provide excellent sensitivity.In addition, 2-cyanoacet- amide is unstable in the presence of borate buffer and the reagent solution must be prepared insitu by mixing an aqueous solution of 2-cyanoacetamide with the buffer by using a dual-pump system. Mikami and Ishidal2 demonstrated that arginine provides far higher sensitivity than other fluorescence reagents when reacted with reducing carbohydrates, down to about 25 pmol. However, this reagent gave high blank values. This blank fluorescence might be due to the a-amino group of arginine, which seemed essential in the fluorescence reaction. Guanidine was, therefore, examined as it seemed to be more suitable as a fluorescence reagent in liquid chromato- graphy because it is more stable than arginine in solution.This paper describes the application of guanidine hydrochloride to the detection of reducing carbohydrates in liquid chromato- graphy. Experimental Reagents and Materials Samples of carbohydrates, guanidine hydrochloride and other reagents were purchased from Nakalai Tesque (Kyoto, Japan). Guanidine reagent standard solution was prepared by dissolving 50 mmol of guanidine hydrochloride in 0.1 mol 1-1 potassium tetraborate solution and diluting with the latter to 1 1, and the pH was adjusted to 11.5 with sodium hydroxide. Chemically bonded propylamine silica gel (Nucleosil 5NH2, particle size 5 pm) and octadecyl silica (ODS) gel (Develosil ODs-5, particle size 5 pm) were obtained from Macherey- Nagcl (Duren, Germany) and Nomura Kagaku (Seto, Japan), respectively.Nucleosil 5NH2 and Develosil ODs-5 were packed in 100 x 6.0 mm i.d. and 200 x 6.0 mm i.d. stainless- steel columns, respectively, by the conventional slurry pac- king technique in our laboratory. A pre-packed Shodex Ionpak KS-801 column (300 x 8 mm i.d.) was purchased from Showa Denko (Tokyo, Japan). Apparatus A flow diagram of the liquid chromatography set up is shown in Fig. 1. The system was assembled from an HPLC pump (Shimadzu Model LC-5A) , an injector (Rheodyne Model 7125), a reaction bath (Shimadzu Model CRB-3A) and a fluorescence detector (Shimadzu Model RF-530) from Shi- madzu (Tokyo, Japan) and a high-pressure reagent pump from Sanuki Industrial (Tokyo, Japan). The Nucleosil and Develosil columns for the separation of glucose and malto- oligosaccharides were eluted with acetonitrile-water (65 + 35) and distilled water, respectively, at a flow rate of 1.0 mi min-* at ambient temperature.The Shodex column for the separa- tion of monosaccharides and N-acetylneuraminic acid was dipped in a water-bath kept at 70 "C. The eluent was distilled water at a flow rate of 0.6ml min-I. The stainless-steel reaction tube was 10 m x 0.8 mm i.d. and the cooling tube was 3 m x 0.24 mm i.d. The guanidine reagent was delivered at a flow rate of 1 .0 ml min-l when the Nucleosil and Develosil columns were used and at 1.2 ml min-l when the Shodex column was used. Results and Discussion Excitation and fluorescence spectra were measured for the glucose-guanidine reaction product.Glucose solution was added to the guanidine reagent to yield a solution of Fig. 1 Flow diagram of liquid chromatograph for the guanidine method. R1, eluent; R2, guanidine reagent; P, pump; I, injector: C, column; RC, reaction coil in oven; CC, cooling coil; and FP, fluorescence detector774 ANALYST, JULY 1993, VOL. 118 - 100 c .- C 3 B A I T u 'D 25 50 75 Guanidine concentration/mmol I -' Fig. 2 Guanidine concentration effect in the post-column reagent. A, Galactose; B, arabinose; C, N-acetylneuraminic acid; D, glucose; E, maltose; and F, maltopentaose (a mixture of 500 pmol of each compound) were separated on a Shodex Ionpak KS-801 column in water at 70°C. For components of the standard reagent see text. Reaction temperature, 135 "C Y L a, z P, 50 .- Y m a, Q a, .- .I- - a, U 0 10.0 11.0 12.0 13.0 PH Fig.3 pH effect of the guanidine reagent. Symbols, chromato- graphic and standard reaction conditions as in Fig. 2 concentration 1 mmol 1-l. This mixture (4 ml) was heated in a sealed test-tube for 30 min on a boiling water-bath. The tube was then quickly cooled and the spectra were recorded. The wavelengths of the excitation and fluorescence maxima were 314 and 433 nm, respectively. These wavelengths were used in the following experiments. The effect of guanidine concentration was examined for several monosaccharides, maltooligosaccharides and N-ace- tylneuraminic acid. As can be seen in Fig. 2, plateaux of fluorescence intensity were reached at a concentration of 25 mmol 1- * guanidine hydrochloride for monosaccharides and at 50 mmol l-l for the other saccharides.Consequently, 50 mmol I-' guanidine hydrochloride was selected for further studies. This result indicated that guanidine was >lo0 times more effective at far lower concentrations than taurine,4 ethanolaminez or 2-cyanoacetamide.6 Fig. 3 shows the effect of pH on the fluorescence intensity. The optimum pH range of the guanidine reagent for most carbohydrates was between 11.5 and 12.5. Kai et a1.13 usedp- methoxybenzamide (MBA) as a post-labelling reagent with good sensitivity. However, MBA was susceptible to degrada- tion by alkali while the highest fluorescence intensity was obtained in alkaline media. Accordingly, the reagent should be delivered to the effluent prior to the alkali solution. In contrast, guanidine is highly stable in an alkaline buffer for at least 1 month at ambient temperature.In the present method, addition of boric acid was essential to increase the sensitivity, similarly to reagents such as taurine,4 ethanolaminez and 2-cyanoacetamide.6 Borate could also be conveniently utilized for adjusting the pH of the reaction mixture. Fig. 4 shows the effect of borate concentra- 0.1 0.2 Borate concentration/mol I-' Fig. 4 Borate concentration effect in reaction reagent. Symbols, chromatographic and standard reaction conditions as in Fig. 2 - 100 A rn F c .- C 3 2 2 f! 0) 50 .I- .- Y c L a, z Y m Q a, .- .- c - $ 0 100 110 120 130 140 150 Reaction temperaturePC Fig. 5 standard reaction conditions as in Fig. 2 Reaction temperature effect. Symbols, chromatographic and Table 1 Relative fluorescence intensities of carbohydrates measured by the flow injection method.Carrier, 0.6 ml min-l of water; reagent solution, 1.0 ml min-l of 50 mmol l-l guanidine (pH 11.5) containing 0.1 moll-' potassium tetraborate; reaction temperature, 135 "C Relative Relative fluorescence fluorescence Carbohydrate intensity Carbohydrate intensity D-Xylose 100 D-Ribose 99 D- Arabinose 92 D-Glucose 34 D-Galactose 59 D-Mannose 37 D-Fructose 30 L-Fucose 129 L-Rhamnose 127 2-Deoxy-~-glucose 6 D-Glucosamine . HCl 1 N-Acetyl-D glucosamine 28 D-Glucuronic acid 5 D-Gluconic acid 0 D-Sorbitol 0 2-Deoxy-~-ribose 5 Inositol Arabitol Maltose Isomaltose Gentiobiose Lactose Sucrose Tre halose Maltotriose Raffinose Maltotetraose Stach yose Maltopentaose Maltohexaose Maltoheptaose N- Acetylneuraminic acid 0 0 99 76 78 138 0 0 136 0 141 0 149 149 138 56 tion in the guanidine reagent on the fluorescence intensity obtained from monosaccharides, maltooligosaccharides and N-acetylneuraminic acid.A guanidine concentration of 50 mmol 1-1 and a pH of 11.5 were used in this experiment. The maximum fluorescence intensity for all carbohydrates was observed when 0.1 mol 1-1 boric acid was used. The reaction temperature was also examined and the results are shown in Fig. 5. The maximum fluorescence intensity wasANALYST, JULY 1993, VOL. 118 77s 1 2 6 u 0 10 20 Time/mi n Fig. 6 Chromatogram of a mixture of 500 pmol each of: 1, N- acetylneuraminic acid; 2, lactose; 3, N-acetyl-D-glucosamine; 4, glucose; 5 , galactose; 6 arabinose; and 7, ribose, separated on a Shodex Ionpak KS-801 column in water at 70°C.FP: hex = 314 nm: he, = 433 nm observed for most of the saccharides between 130 and 140 "C. The maxima shifted toward higher reaction temperatures for N-acetylneuraminic acid and maltopentaose, maltohexaose and maltoheptaose. The 2-cyanoacetamide method6 was normally carried out at a lower temperature of 100 "C, but the linear range for the calibration graph was only 0.5-50 nmol and different sensitivities among carbohydrates were ob- served. Schlabach and Robinson14 reported that higher temperatures, of up to 160"C, accelerated the formation of fluorescence products with carbohydrates and 2-cyanoacetam- ide, but variable sensitivity among carbohydrates occurred, presumably owing to the use of temperatures far below the plateau of the fluorescence intensity. In the present system, the chemical reaction oven facilitated heating of the effluent even to 150°C without any problem; heating at 135 "C was adopted to ensure sensitivity and reproducibility.The more severe the reaction conditions, the more degradation of saccharides must have occurred, because larger oligosac- charides exhibited higher sensitivity. The relative fluorescence intensities of various carbo- hydrates measured by flow injection are given in Table 1. 6- Deoxyhexoses showed the highest fluorescence intensity, and that of pentoses was slightly lower than that of 6-deoxy- hexoses, and those of hexoses, N-acetylglucosamine and N- acetylneuraminic acid were about one third to two thirds of that of pentoses.2-Deoxy sugars, which do not react well with other fluorescence reagents, gave poor fluorescence also in the present method. Glucuronic acid gave only a low peak height. Sugar alcohols and non-reducing oligosaccharides showed larger peak heights. The present reagent and MBA possess a guanidine group, and Kai etal.13 assumed that the formation of an imidazole ring was essential for the development of fluorescence. lmidazole formation was substantiated by the fact that 2- t - m C (5: v) .- 4 3 2 3 0 10 20 Ti me/m i n Fig. 7 Chromatogram of 500 pmol each of: 1, glucose; 2, maltose; 3, maltotriose; 4, maltotetraose; 5 , maltopentaose; 6, maltohexaose; and 7, maltoheptaose, separated on a Nucleosil SNH2 column in acetonitrile-water (65 + 35). FP: hex = 314 nm; A,,, = 433 nm eoxyaldoses gave little fluorescence.This reaction might proceed more readily than the reaction of ethyleneamino derivatives to give higher sensitivity. In addition, the fluores- cence intensity of 6-deoxyhexoses was three times higher than that of ordinary hexoses. Pentoses gave more intense fluores- cence than hexoses. These results suggest the formation of some aromatic ring fused with the imidazole ring to enhance the fluorescence intensity. The sensitivity of the present method is assumed to be valid for reducing carbohydrates having more than three carbon atoms. However, further study might be required to understand the reaction mechanism. The calibration graphs showed excellent linearity for sample concentrations from less than 10 pmol up to 1 nmol, and passed through the origin.The limit of determination for pentose, 6-deoxyhexose and reducing oligosaccharides was 5 pmol and that for hexose, N-acetylglucosamine and N- acetylneuraminic acid was 10 pmol. The relative standard deviations for 500 pmol per sample of glucose, maltopentaose and N-acetylneuraminic acid were 1.86, 1.45 and 3.01% (n = 10) , respectively. The present method proved to be far more sensitive than other fluorimetric methods employing reagents having an ethyleneamino structure, such as ethylenediamine,' ethanol- amine,2 taurine4 and 2-cyanoacetamide,6 even though the structure of the fluorescent product has not yet been elucidated. A chromatogram of a mixture of standard monosaccharides, lactose and N-acetylneuraminic acid is shown in Fig.6, and those of maltooligosaccharides are shown in Figs. 7 and 8. An undesirable resolution of anomers of the individual oligosaccharides was observed on the ODS column, hence propylamine-bonded silica gel was better for oligosac- charide analysis. This method was more sensitive than other fluorimetric methods, reproducible and gave only a small blank value,776 3 - ANALYST, JULY 1993, VOL. 118 0 10 20 Tirne/rn in Fig. 8 Chromatogram of 500 pmol each of: 1, glucose; 2, maltose; 3, maltotriose; 4, maltotetraose; 5, maltopentaose; 6 , maltohexaose; and 7, maltoheptaose, se arated on a Develosil ODs-5 column in water. FP: he, = 314 nm; re, = 433 nm 4 0 10 20 30 Tirne/min Fig. 9 Chromatogram of 500 pmol each of: 1, p-nitrophenyl-a- galactopyranoside; 2, p-nitrophenyl-a-glucopyranoside; 3 , p-nitro- phenyl-a-mannopyranoside; 4, y-nitrophenyf-a-xylopyranoside; and 5 , p-nitrophenyl-a-fucopyranoside, separated on a Develosil ODs-5 column in acetonitrile-water (20 + 80).FP: he, = 314nm; he, = 433 nm presumably owing to the purity and stability of guanidine. Contrary to other fluorimetric methods, reducing oligosac- charides showed a higher fluorescence than those of monosac- charides. Moreover, the guanidine reagent exhibited excellent sensitivity for N-acetylneuraminic acid. Sialic acids have conventionally been detected spectrophotometrically follow- ing post-column derivatizaton by periodate oxidation and reaction with either thiobarbituric acid15 or resorcinol. 16 Honda and co-workersl7~18 reported a method for converting sialic acid into N-acetylmannosamines by the combined use of neuraminidase and N-acetylneuraminate pyruvate lyase and then detecting the product by the 2-cyanoacetamide method.In addition, p-nitrophenylglycosides were detected by the present method as shown in Fig. 9, presumably owing to the instability of these glycosides to the action of alkali. Recently, anion-exchange chromatography with sodium hydroxide solution as the mobile phase in combination with pulsed amperometric detection permitted the separation of carbohydrates with high sensitivity. 19-21 The requirement22 for a strongly alkaline mobile phase restricts the separation column, and reducing mono- and oligosaccharides are liable to undergo isomerization and degradation under the conditions used.In addition, proteins and amino acids are often adsorbed on the electrode and interfere with detection. On the other hand, the present post-column fluorimetric detection can be applied using acetonitrile-water, borate buffer or highly alkaline solutions as mobile phases and these separation conditions do not affect the detection of carbohydrates. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 References Mopper, K., Liebezeit, G., and Hansen, H. F., Anal. Chem., Kato, T., and Kinoshita, T., Anal. Biochem., 1980, 106, 238. Honda, S . , Natsuda, Y., Terao, M., and Kakei, K.. Anal. Chim. Acfa, 1979, 108, 421. Kato, T., and Kinoshita, T., Chem. Phurm. Bull., 1978, 26, 1291. Honda, S . , Takahashi, M., Kakehi, K., and Ganno, S . , Anal. Biuchem., 1981, 113, 130. Honda, S . , Matsuda, Y., Takahashi, M., Kakehi, K., and Ganno, S . , Anal. Chem.. 1980, 52, 1079. Simatupang, M. H., J . Chromatogr., 1979, 180, 177. Conchie, J., and Hay, A. J . , Curhohydr. Res., 1983, 112, 261. Kramrner, K. J . , Speirs, R. D., and Childs, C. N., Anal. Shukla, A. K., Scholz, N., Reimerdes, E. H., and Schauer, R., D'Amboise, M., Hanai, T., and Noel, D., Clin. Chem. Mikami, H., and Ishida. Y., Bunseki Kagaku, 1983, 32, E207. Kai, M . , Tamura, K., Yamaguchi, M., and Ohkura, Y., Anal. Schlabach, T. D., and Robinson, J., J . Chromatogr., 1983,282, Krantz, M. J., and Lee, Y. C., Anal. Biochem., 1975, 63,464. Tsuji, T., Yamamoto, K., Konami, Y., Irimura, T., and Kakehi, K., Maeda, K . , Teramae, M., Honda, S . , and Takai, Honda, S., and Suzuki, S . , Anal. Biochem., 1984, 142, 167. Johnson, D. C., and LaCourse, W. R., Anal. Chem., 1990,62. Martens, D. A . , and Frankenberger, W. T., Jr., Chromato- Rocklin, R. D., and Pohl, C., J. Liq. Chrumurogr., 1983, 6, Lee, Y. C., Kagaku to Kogyo, 1990, 43, 953. 1980, 52, 2018. Biochem., 1978. 86, 692. Anal. Biochem., 1982, 123, 78. (Winston-Salem), 1980, 26, 1348. Sci., 1985, 1, 59. 169. Ohsawa, T., Carbohydr. Res., 1982, 109, 259. T., J. Chromatogr., 1983, 272, 1. 589. gruphia, 1990, 29, 7. 1577. Paper 2106377H Received November 30, 1992 Accepted January 26, 1993

ISSN:0003-2654    

DOI:10.1039/AN9931800773

出版商:RSC    

年代:1993

数据来源: RSC

摘要:

ANALYST, JULY 1993, VOL. 118 777 Thin-layer Chromatographic Detection of Pyrethroid Insecticides Using o-Tolidine Akmal Pasha and Yadathora N. Vijayashankar Infestation Control and Protectants, Central Food Technological Research Institute, Mysore 570 0 13, India A method for the detection of pyrethroid insecticides by thin-layer chromatography is described. These insecticides, on bromination and treatment with o-tolidine, yield an intensely blue product. Some of the pyrethroids can be resolved into individual isomers by this method. The limit of detection is 0.25-1.0 pg. Keywords: Thin-la yer chromatography; bromination; o-tolidine; pyrethroid insecticide Pyrethroids are amongst the most potent insecticides that have low mammalian toxicity.1J They are effective against a wide spectrum of pests both in pre- and post-harvest infestation.Owing to their wide usage they find their way into agricultural produce. Consequently, their characterization becomes necessary to ascertain the level of contamination. There are several gas-chromatographic and high-perform- ance liquid-chromatographic separation methods available.3 A number of chromogenic reagents have also been reported in the literature for detecting some of the pyrethroids by a thin- layer chromatography (TLC) method. For example, 2% ethanolic phosphomolybdic acid has been used for the detection of permethrin, cypermethrin and deltamethrin,4 0.5% palladium chloride in HCI (12 mol dm-3) for the detection of deltamethrin ,5 and alkaline hydrolysis of the nitrile group followed by reaction with copper(n) acetate and o-tolidine for cypermethrin, fenvalerate and deltamethrin.6 Visualizing, under ultraviolet (UV) radiation, TLC plates having layers of adsorbent impregnated with silver nitrate is a general method.7.8 In this paper, a simple and sensitive method of detecting pyrethroids by TLC is described.After spotting and elution by solvent, the TLC plate is exposed to bromine vapour and subsequently sprayed with 0.1% o-tolidine in acetone. On exposure to sunlight or UV radiation, intense blue spots appear on a colourless background. The method is applicable to established pyrethroids, some of which separate into distinct spots resulting from their constituent isomers. The sensitivity of the method is 0.25-1.00 pg. o-Tulidine Reagent. Dissolve 0.1 g of o-tolidine in 100 cm3 [Caution: o-tolidine is a known carcinogen.] of acetone. Procedure A 20 x 20 cm glass plate was coated with the silica gel slurry to a uniform thickness of 0.2 mm.The plate was activated by heating in an oven at 110 "C for 1 h and subsequently stored in a desiccator. A 10 mm3 aliquot of each standard pyrethroid solution in acetone (1 mg ~ m - ~ ) was spotted on to the plate. Development was carried out in a TLC chamber saturated with light petroleum (60-8OoC)-diethyl ether (9 + 1) or cyclohexane-toluene (7 + 3) as the mobile phase. After the mobile phase had travelled about 15 cm, the solvent front was marked and the plate was removed from the chamber, air dried and kept in another chamber containing bromine vapour for about 15 s.The residual bromine was removed from the plate using a fan, and 0.1% o-tolidine chromogenic reagent was sprayed uniformly. The plate was exposed to UV radiation for 5 min or to sunlight for 2 min. Brominated pyrethroids appeared as intense blue spots on a colourless background. The RF values are presented in Table 1. Results Deltamethrin could be differentiated from the other pyre- throids as it yielded a greenish-blue spot, whereas the spots Experimental Apparatus A Camag (Muttenz, Switzerland) autocoater was used to prepare the TLC plates. A UV lamp of 325 nm wavelength was used to irradiate the TLC plates. Reagents All the reagents used were of analytical-reagent grade. Distilled water was used throughout. Standard pyrethroid solutions. Prepare separately, 1 mg ~ m - ~ solutions of allethrin, alphacypermethrin, cyper- methrin, deltamethrin, fenvalerate, indothrin and permethrin in acetone.Silica gel G. Prepare a slurry of Silica gel G of particle size 10-40 pm with 13% CaS04 binder by mixing thoroughly 40 g in 100 cm3 of distilled water. Bromine. Place 5 cm3 of bromine in a 25 X 25 X 5 cm glass chamber and allow the chamber to saturate with bromine vapour at 30 "C. Table 1 RF values of brominated pyrethroid insecticides and their detection limits RF* Detection Pyrethroid limit/yg Solvent If Solvent 2* Allethrin 0.2s 0.18 0.00 Alphacyper- methrin 0.25 0.35 0.11 0.49 0.10 0.45 Deltamethrin 0.25 0.44 0.12 Fenvalcrate 1 .oo 0.42 0.08 0.40 Indot hrin 1 .00 0.86 0.60 0.66 0.22 0.58 Permethrin 0.50 0.68 0.30 0.59 0.20 Cypermethri n 1 .oo 0.54 0.13 * Individual isomers not identified.+ Light petroleum (6&80"C)-diethyl ether (9 3- 1). * Cyclohexane-toluene (7 + 3).778 due to the others were an intense blue. The colour of the spots is stable indefinitely, but the background slowly develops a dark-yellow tinge after about 1 h, thus reducing the clarity. This background colour development was also observed when plates were sprayed with a high concentration of o-tolidine (>0.1%), when the TLC plate was over-exposed to bromine vapour (>30 s) and when remdval of the excess bromine vapour was incomplete before spraying the plate with o-tolidine. Chlorination of pyrethroids, instead of bromination after developing the chromatogram, afforded similar results. How- ever, the quality of the chromatogram in terms of sensitivity and visualization of the spots against the background colour was many times diminished.Out of many solvent mixtures screened, light petroleum (60-8O0C)-diethyl ether (9 + 1) was found to be the best mobile phase when constituent isomers of the individual pyrethroid insecticides were required to be separated on the chromatogram. However, since the RF values were close for some pyrethroids, cyclohexane-toluene (7 + 3) was used to achieve a better separation of pyrethroid mixtures. The limit of detection of the method is 0.25-1.00 pg. ANALYST, JULY 1993, VOL. 118 Rajagopala Rao, Chairman, Food Science Division, and Dr. S. R. Bhowmik, Director, CFTRT, Mysore, for their encour- agement. The authors thank Dr. M. K. Krishnakumari, Area Coordina- tor, Infestation Control and Protectants Department, Dr. D. References Elliott, M., Pestic. Sci., 1980, 1 1 , 119. Elliott, M. , in Synthetic Pyrethroids, ed. Elliott, M., American Chemical Society, Washington, DC. 1977, p. 1. Mourkidou, E. P . , in Residue Reviews, eds. Gunther, F . A. , and Gunther, J. D . , Springer Verlag, New York, vol. 89, 1983, pp. 179. Shono, T., Ohsawa, K., and Casida, J . E., J . Agric. Food Chem., 1979, 27, 316. Ruzo, L. O., Engel, J . L., and Casida, J . E., J . Agric. Food Chem., 1979, 27, 725. Patil, V. B . , Sevalkar, M. T., and Padalikar, S. V., Analyst, 1992, 117, 75. Klisenko, M. A . , and Girenko, D . B . , Zh. Anal. Khim., 1984, 39, 1132; Anal. A bstr., 1985,47, 9H72. Sundararajan, R., and Chawla, R. P., J. Assoc. Off. Anal. Chem., 1983, 66, 1009. Paper 210451 01 Received August 20, 1992 Accepted February 11, 1993

ISSN:0003-2654    

DOI:10.1039/AN9931800777

出版商:RSC    

年代:1993

数据来源: RSC

摘要:

ANALYST, JULY 1993, VOL. 118 779 Use of Chemometric Factor Analysis for Chromatographic Integration: Application to Diode-array Hig h-performance Liquid Chromatography of Mixtures of Chlorophyll a Degradation Products Yi-Zeng Liang Department of Chemistry, University of Bergen, N-5007 Bergen, Norway Richard G. Brereton" School of Chemistry, University of Bristol, Bristol, UK BS8 ITS Olav M. Kvalheim School of Chemistry, University of Bergen, N-5007 Bergen, Norway Ali Rahmani School of Chemistry, University of Bristol, Bristol, UK BS8 ITS Quantification errors of using the vertical divisor method for single wavelength chromatographic integration of closely eluting peaks are discussed. A new approach of using the chemometric factor analysis method, heuristic evolving latent projections, prior to integration is discussed, and typical single-wavelength errors in estimated integrals are less than 1% on simulated data.The method was applied to three mixtures of six closely eluting chlorophyll degradation products. Single-wavelength quantification is good, provided that the wavelengths chosen are close to the maxima of the peaks. Further improvements can be made by summing over all wavelengths. Finally, an approach to normalization of chromatographic integrals is discussed, which is particularly valuable when pure standards are not available. Keywords: Chemometrics; factor analysis; chlorophyll a; high-performance liquid chromatography; in teg ra ti0 n Interest in chlorophyll degradation reactions is widespread. The Treibs hypothesis that petroporphyrins originate from chlorophylls is attributed as the origin of organic geo- chemistry.' Browning of leaves during the autumn is of substantial interest to biochemists,2 and this major global phenomenon can be followed from space as waves of brown replace the greenery.Environmental scientists use chlorophyll degradation products to monitor the fate of algae3 and hence the history of natural waters; databanks spreading over many decades have been accumulated from environmental monitor- ing programmes. Even archaeologists use chlorophyll degra- dation products to deduce the nature of plant material in ancient sites.4 Despite this continuing interest in the process of chlorophyll degradation, spreading over more than half a century,5,6 there is virtually no quantitative information available on the nature of such processes.The turnover rate of chlorophylls in healthy leaves is unknown; the global turnover of chlorophyll in natural waters is not known; even geochemists cannot determine whether the porphyrins observed in sediments represent 10 or 0.01% or even less of the total chlorophyll content. These difficulties present serious problems in inter- preting distributions of chlorophyll degradation products. We have shown previously that chlorophyll a rapidly degrades to two- and three-carbon fragments on illumination ,798 demon- strating that intact porphyrins certainly do not represent the main remains of the chlorophylls in many instances. Another observation9 is that the degradation rates of different chlo- rophylls vary substantially according to structure, demonstrat- ing that a 1 + 1 mixture of two chlorin structures will not yield a 1 + 1 mixture of derived products and so an analysis of sedimentary porphyrins will not necessarily provide informa- tion on the relative algal composition of the original material.The difficulty with studying chlorophyll degradation pro- cesses is one of finding robust analytical techniques for the quantification and identification of structurally similar com- pounds. Because of the obvious chromophoric differences * To whom correspondence should be addressed. between the major chlorophylls, early investigators used a variety of ultraviolet (UV)-visible spectrometric methods to determine the relative concentrations of chlorins.l(k-12 These involve setting up equations of the relative absorption at different wavelengths. For example, if pure compound C has absorption at wavelength j of xc,, then where Y is the ratio of absorbance at wavelength 1 to that at wavelength 2 in a given mixture where the true ratio of concentrations is [A]/[B]. The weakness of this approach is four-fold.13.13 First, it does not take into account noise. Second, and more seriously, it only allows for two compounds. A third compound with a different absorbance will invalidate eqn. (1). It is possible, of course, to increase the number of compounds, and so wavelengths, monitored, but this involves isolating pure compounds, and the equations are still restric- ted to precisely these pure compounds. Third, if compounds have very similar absorbances, very small differences in ratios can dramatically change the estimates of relative concentra- tions.As will be shown below, the systems under study here involve determining a series of compounds with very similar structural features, and so similar spectra. Fourth, these equations are critically dependent on accurate wavelength calibration. Most UV-visible spectrometers exhibit some errors here, as can be shown by recording spectra of reference standardsl5; these errors are not always exactly linear. If peaks exhibit sharp maxima, an inaccuracy of only a few nanometres can make very large differences to the determination of relative concentrations of different compounds, particularly when absorption spectra differ by only a few nanometres in turn.In many instances, published equations lead to nonsensical or even negative estimates of concentrations of chlorophylls for many of the reasons outlined above, yet changes in concentration with time, depth and geographical location provide crucial evidence for the fate of chlorophylls, and are frequently interpreted in great detail.780 ANALYST, JULY 1993, VOL. 118 With improvements in analytical methodology, much atten- tion has been focused on the high-performance liquid chromatography (HPLC) of chlorophylls. 16-18 Chromato- graphic separation increases the information about relative concentrations in the mixtures. Quantification is possible by looking at the relative areas of different peaks. There are, however, relatively few descriptions of the use of HPLC for the determination of chlorophylls.A critical difficulty is that chlorophylls exhibit subtle differences in spectrometry, so the relative peak areas are not equal to relative concentrations. Spectra of pure components are required for quantification. A second difficulty is when peaks elute close to each other. Although some improvements can be made in the chromato- graphy, complete resolution is difficult to achieve under conditions reported in this paper. In studies we have reported, up to nine compounds may elute over a period of 2-3 min19-21; pure standards are not available for most compounds, making quantification of closely overlapping peaks difficult because the spectra and elution profiles of each compound are not known in advance.The applications in this paper differ from the more commonly reported pharmaceutical applications, where large amounts of pure standards are frequently available. The advent of HPLC with diode-array detection (DAD) over the last few years has increased the potential for the quantification of chlorophylls. Most early work was per- formed in the area of pharmaceutical chemistry, where the problems were different. Often large amounts of pure standards, with substantially differing chromophores, are available, and the main objective was isolation as opposed to analysis. Recent advances in HPLC-DAD methodology include extending the wavelength range to high wavelengths, of interest in the study of pigments. Chemometric approaches for the deconvolution of mixtures by factor analysis have also been developed by various groups.22-27 Most applications, however, have been to fairly tractable problems where large amounts of standards with substantially different spectro- scopic properties have been available.In many instances, chromatographic optimization methods will help to resolve the peaks better, making the chemometrics largely redundant, or at any rate a complementary, but not necessary, approach. In the work discussed in this paper, we concentrate on the difficult analytical problem of the quantitative analysis of mixtures of closely similar chlorophyll degradation products. Experimental Preparation of Chlorophyll a Chlorophyll a (1) was extracted from spinach as described in greater detail elsewhere. 19 Purity was monitored by HPLC- DAD and nuclear magnetic resonance (NMR) spectrometry. A very small amount of epimer (1') was present in the starting material. Degradation of Chlorophyll a to Produce Mixtures Several degradation experiments were performed, involving heating chlorophyll a between 20 and 50 "C, in both methanol and acetone.Some experiments were performed with a stream of oxygen constantly bubbling through the solution and others with nitrogen bubbling through. Some were performed in the dark and others with light of varying intensities illuminating the solution. It was found that the various conditions can be used to control the rate and nature of the degradation reactions. For the purpose of this study, the following conditions were employed. Chlorophyll a (approximately 0.072 mmol 1- as determined by measuring the molar absorptivity) dissolved in methanol (BDH, HiPerSolv grade) was transferred into three 50 ml three-necked reaction flasks.Two flasks were placed in a water-bath at 30°C with a constant stream of oxygen bubbling through in the dark. Sleeved condensers with poly(tetrafluoroethy1ene) (PTFE) tubing were placed on one neck of the reaction flasks, and ethanol at -40°C was circulated throughout to prevent evaporation. The flasks were sampled at 30 min and 6 h (to give samples 1 and 2, respectively). The third sample was placed in a water-bath at 30°C with cooled ethanol circulating as above, but with a constant stream of nitrogen circulating and illuminated by a lamp placed at 30 cm away from the reaction vessel, giving an illuminance of 7800 lux as measured by a Macam Quantum/ Radiometer/Photometer (Model Q.101). The conditions used to obtain the three samples are summarized in Table 1. HPLC-DAD A Waters Model 990 diode-array detector equipped with a Model 600E multi-solvent delivery system and an RP-C18 column (300 x 3.9 mm i.d.) was used. Acetone and methanol (BDH, HiPerSolv grade) and water purified with a Milli-Q system (Millipore) were employed for the mobile phase. The flow rate was 1 ml min-l. In order to compare samples, two different gradients were used (Table 2). Most analyses re- ported in this paper are from gradient 1, although the results from gradient 2 are used where necessary. Each spectrum was recorded over 38 ms and 20 successive scans were averaged, making each point in time an average of spectra over 0.76 s.Chromatograms were sampled every 2 s. Spectra were recorded between 350 and 800 nm at 5 nm intervals. Isolating and calculating the molar absorptivities for all pure components would be an impracticable and time-consuming task. In some instances the chemical structures are not unambiguously known. Traditionally, chromatographers integrate chromatograms at individual wavelengths, and use these to obtain relative concentrations. The integration can be performed by a variety of methods, including vertical lines, triangulation and tangen- tial skimming. One major weakness of this is that if two compounds have different spectra, as usually occurs, it is necessary to calibrate peak areas at each wavelength to relative concentrations. This can be performed by recording, or extracting, the pure spectra of each compound, and then Table 1 Samples uscd in this study Sample Tempera- NO.Time/h ture/"C N2 or O2 Illumination 1 0.5 30 0 2 Dark 2 6 30 0 2 Dark 3 5 30 N2 7800 lux Table 2 Chromatographic gradients used in this study Time/min* Methanol (%) Acetone (%) Water (YO) Gradient 1- 0 80 0 20 5 80 0 20 35 60 40 0 5 0 60 40 0 60 80 0 20 70 80 0 20 0 90 0 10 5 90 0 1 0 15 10 85 5 30 to 88 2 35 90 0 10 40 90 0 1 0 * Time = 0 is the injection time; the next injection is immediately after the end of each run. The last few minutes of each run allow time for cquilibration. Gradient 2-ANALYST, JULY 1993. VOL. 118 78 1 relating the relative absorbances at the chosen wavelength to the relative concentrations.In this paper we shall restrict the work to comparing the vertical divisor approach to chemo- metric factor analysis. An experimental difficulty involves choosing the best wavelengths. A more serious problem arises from choosing the vertical divisor between two peaks. Most chromatographic integration packages allow the operator to state where he or she wishes to divide up the areas, and assumes that the region to the left of the divisor belongs to peak 1 and that to the right to peak 2, in the case of two peaks, 1 and 2. At different wavelengths the two peaks will appear to have different relative intensities. It is easy to show that the relative area of a small peak in the presence of a large peak will be distorted more than if two peaks are of equal intensity. Chemometrics Details of the HELP (heuristic evolving latent projections) factor analysis method, employed in this paper have been described elsewhere,20,25 so only a brief outline will be given here.It is one of a family of approaches to evolutionary factor analysis developed over the last decade. Evolving factor analysis (EFA), the most commonly cited approach, will not work in the presence of a structured background, heterosce- dastic noise and non-linear response behaviour of the components in the sample.26 The data were transferred to a VAXstation 2000, running under VMS Workstation Software (released June 1989). Programs were written in VAX FORTRAN version 5.0 (released June 1990). The HYLC-DAD trace can be con- sidered as an I x J matrix, where Z is the number of rows or points in time and J is the number of columns or spectroscopic frequencies.The rank of a matrix is the number of linearly independent variables that are needed to construct the matrix.25 In a noise-free HPLC-DAD trace, the rank of the matrix should correspond to the number of chemical species present, provided that there are differences in their absorption spectra: this is called the ‘chemical’ rank. Because of noise, the actual rank of the matrix exceeds the chemical rank. The HELP method first takes a submatrix of the chromatogram of dimensions i X 1, where i < J in a region where there are known to be no compounds eluting (zero-component region). This matrix is analysed by principal components analysis (PCA).28-3() The size of the first eigenvalue in this region is called the zero-component eigenvalue.The chemical rank of other regions of the chromatogram can be estimated by comparing the first few eigenvalues of submatrices of size il with the zero-component eigenvalue. The ratios of these eigenvalues to the zero-component eigenvalue can be tested using an F-test.31.32 The larger the value of this ratio, the more significant the component. If the F-test predicts that the component is significant then it is assumed that the component represents a compound. The number of significant com- ponents equals the chemical rank of the submatrix and so, C 0 ’= 0.6 2 +.. C 2 0.4 0 0.2 0 Fig. 1 Time Simulated elution profiles of peaks A and I3 ideally, the number of chemical species.The HELP method looks at submatrices of various sizes, and especially tries to detect regions in the cluster that have a chemical rank of 1. These are regions of chromatographic selectivity and are critical to the success of the resolution achieved by this method. In these small regions, it is possible to use PCA to decompose the data matrix into scores (corresponding to concentration profiles with time) and loadings (corresponding to spectra) of the pure component. By using the region of selectivity in combination with the so-called ‘zero concentra- tion window’, the principal component scores correspond to the concentration profile of the compound. For the rest of the chromatogram, it is possible to rotate the principal component scores on the factor scores.Results A simple numerical example illustrating the potential serious- ness of these errors will be given. Consider monitoring the elution of two peaks over 40 intervals in time. Both peaks have equal intensities and widths, but peak 1 is centred at time 13 and peak 2 at time 20 (Table 3 and Fig. 1). The spectra are identical in shape but shifted relative to each other by four datapoints (Table 4). These are illustrated in Fig. 2. Single wavelength profiles are illustrated in Fig. 3. The full mathe- matical theory is given in Appendix 1, and the reader is referred there for the mathematical notation used below. In the example, it is difficult to determine, visually, exactly where one peak starts and the other ends, although it is easy to show that the cluster consists of two peaks using a variety of graphical approaches such as derivatives.In practice, this can cause large differences in estimates of relative integrals. However, a reasonable approach would be to choose Z12 = 16, as the mid-point between the maxima of thc peaks is 16.5. 0 5 10 15 20 Wavelength Fig. 2 Simulated spectra of peaks A and B Table 3 Gaussian parameters for elution profiles of peaks 1 and 2 Parameter Peak 1* Peak 2* A 1 1 S 4 4 C 13 20 * Peaks are characterized A exp[-(c - i)2/s2]. where i correspondc to the time. Table 4 Spectra of peaks 1 and 2 Peak 1* Peak 2‘ Parameter Gaussian 1 Gaussian 2 Gaussian 1 Gaussian 2 A 1 2 1 2 S 4 4 4 4 C 9 19 5 15 * Each spectrum is characterized by a sum of two Gaussians each, in turn characterized by A exp[ - ( c - j)2/s2].where j corresponds to the wavelength.782 ANALYST, JULY 1993, VOL. 118 In Table 5 , the values of t,12, F,12 and Ej12 for all 20 wavelengths are listed. The percentage error can be substan- tial. As can be seen from Table 5, wavelengths where the error is least are those at which ?,12 = FjI2, i . e . , wavelengths with equal absorption for the two compounds. Most chromatographers prefer. to work at one wavelength for the purpose of integration, often calibrating the data to the spectra of pure standards, for the purpose of converting from peak intensities into concentrations. Although the optimum wavelengths for quantification appear to be j = 7, 11 and 17, according to Table 5, a fresh simulation (Table 6) where peaks 1 and 2 are in the ratios 1 : 3 (the parameters are identical with those in Tables 3 and 4, except the value of A for the time profile of peak 1 is 0.333), gives a different error profile, with the optimum between wavelengths 3 and 4.Another important factor influencing the quantification errors is the elution profile. The simulations in Table 7 provide a sharper elution profile than Table 3, although both peaks are centred at the same wavelengths and also have the same spectra. Table 8 is the result of using the same vertical divisor as in Table5. It can be seen that, although the optimum wavelengths remain the same, the magnitude of the errors decreases. Hence, for single-wavelength monitoring, there is no ‘best’ wavelength for quantification. The optimum wavelength depends on the spectral characteristics and relative amounts of each compound.As the relative proportion of each compound Table 5 Results of using the vertical divisor to quantify two peaks, 1 and 2, as discussed in the text J 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 fj12 0.17 0.20 0.25 0.33 0.47 0.67 0.96 1.31 1.54 1.38 0.98 0.64 0.47 0.44 0.51 0.70 1.01 1.48 2.14 3.00 rj12 0.05 0.08 0.14 0.22 0.37 0.60 0.96 1.42 1.73 1.51 0.97 0.56 0.37 0.33 0.42 0.63 1.01 1.65 2.72 4.48 Ej12 238.06 143.23 85.34 49.64 27.05 11.96 0.98 -7.28 -11.47 -8.64 0.66 13.77 26.98 30.62 22.30 10.59 -0.25 - 10.48 -21.23 -33.15 2.0 a, C m + d 1.0 Q 0 10 20 30 40 Time Fig. 3 Single-wavelength profiles of peaks A and B at wavelengths I, 4,6,7, 10,13, 16 and 19 for the the peaks simulated in Tables 3 and 4.The vertical divisor is indicated is unknown in advance, it is not easy to select an optimum wavelength. Another approach is to sum !he integral over all wave- lengths, so giving the value X , [eqn. (AlO)]. For the simulations in Tables 3 and 4, these are 123.50 and 142.87 for peaks 1 and 2, respectively, compared with the true values of 120.88 and 145.49, representing errors of 2.17% and - 1 .SO%. Normalizing the concentrations so that the integrals add up to 1 [eqns. (A12) and (A13)] provides estimated values of 0.4636 and 0.5364, compared with true values of 0.4538 and 0.5462 for peaks 1 and 2, respectively. This multi-wavelength integration is much more reliable than single-wavelength integration. Further factors influencing the performance of quantifica- tion methods include the difference in spectroscopic proper- ties, baseline roll and the rate of sampling with time.There is not space in this paper to discuss all such factors in detail, but it is evident that single-wavelength quantification of complex mixtures is a hazardous process. With clusters of several closely eluting peaks, conventional approaches to chromato- graphic integration can at best yield a qualitative picture of relative concentrations of compounds, unless a great deal is known about the system. Hence there is a need to develop multivariate chemometric methods for integration, as dis- cussed below. In order to develop a chemometric method, the chromato- grams are first resolved into their components using approaches for evolving factor analysis.In this paper we apply the HELP method, summarized briefly above, and in more detail elsewhere.20.24 This method is one of a family of methods for evolutionary factor analysis.22-27 The estimated concentrations now come from the estimated elution profiles. It is important to recognize that factor analysis methods alone cannot provide absolute quantification for the simple reason that these approaches estimate the product of the spectrum with the concentration profile, and external information calibrating the spectra to true concentrations is required. However, relative concentration profiles can be easily esti- mated. Table 6 Results of using the vertical divisor to quantify two peaks, 1 and 2 in the ratio 1 : 3, as discussed in the text i 1 2 3 4 5 6 7 8 9 10 11 22 13 14 15 16 17 18 19 20 fj12 0.14 0.15 0.16 0.19 0.24 0.31 0.42 0.56 0.65 0.59 0.43 0.30 0.24 0.23 0.26 0.32 0.44 0.63 0.93 1.37 Fj12 0.05 0.08 0.14 0.22 0.37 0.60 0.96 1.42 1.73 1.51 0.97 0.56 0.37 0.33 0.42 0.63 1.01 1.65 2.72 4.48 Ej12 172.86 78.37 21.02 -13.81 -35.00 -47.94 -55.84 -60.47 -62.38 -61.12 -56.04 -46.48 -35.07 -31.74 -39.24 -49.02 -56.60 -61.95 -65.96 -69.46 Table 7 Elution profile for sharp peaks Parameter Peak 1‘ Peak 2* A 1 1 ‘i 3 3 C 13 20 * Peaks are characterized by A exp(-(c - i)2/s*], where i corresponds to the time.ANALYST, JULY 1993, VOL.118 783 Table 8 Results of using the vertical divisor to quantify two peaks, 1 and 2. whose elution profiles are given in Table 7 i 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 fJ12 0.10 0.13 0.18 0.27 0.41 0.63 0.96 1.37 1.64 1.45 0.97 0.59 0.41 0.38 0.46 0.66 1.01 1.57 2.44 3.70 r;12 0.05 0.0s 0.14 0.22 0.37 0.60 0.96 I .42 1.73 1.51 0.97 0.56 0.37 0.33 0.42 0.63 1.01 1.65 2.72 4.4s E,12 100.58 60.65 36.27 21.22 11.67 5.24 0.44 -3.35 -5.36 -3.99 0.30 6.02 11.64 13.19 9.66 4.65 -0.11 -4.88 -10.42 - 17.49 Table 9 Results of using HELP to quantify the two peaks in Tables 3 and 4 i 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 512 0.0498 0.0821 0.1353 0.2229 0.3665 0.5985 0.9554 1.4178 1.7348 1.5140 0.9694 0.5591 0.3672 0.3336 0.4202 0.6313 1.0117 1.6542 2.7209 4.4829 41 0.1274 0.3331 0.7443 1.4882 2.6087 4.0410 5.5227 6.6642 7.1179 6.7493 5.7780 4.7004 4.1048 4.4593 5.9623 8.4082 11.1732 13.3638 14.1933 13.3212 X;2 2.6261 4.0674 5.5591 6.7081 7.1448 6.7938 5.8161 4.7314 4.1319 4.4887 6.0017 8.4637 11 4916 13.4521 14.2870 13.4092 11.1185 8.1354 5.2523 2.9901 r;12 0.0485 0.0819 0.1339 0.2219 0.3641 0.5948 0.9496 1.4085 1 .7227 1.5036 0.9627 0.5554 0.3701 0.3315 0.4173 0.6270 1.0049 1.6427 2.7023 4.4551 E -2.65 -0.22 - 1.04 -0.44 -0.65 -0.63 -0.61 -0.66 -0.70 -0.69 -0.70 -0.67 0.78 -0.63 -0.69 -0.68 -0.68 -0.70 -0.69 -0.62 Fig.6 1, at 400 nm, indicated by shaded regions Selective regions for compounds A-F, sample 3 and gradient Table 10 Identities of peaks in the three samples 2 Sample No. Peak 1 A B C D E F A B C D E F A B C D E F 3 Identity* 2 3 2 3 1 1 2 3 4 3 1 1 2 3 4 3 1 1 * Compounds 1 4 are identified elsewhere. Each of the compounds 1, 2 and 3 has corresponding stereoisomers (epimers) with identical UV-visible spectra but different chromatographic properties.In samples 2 and 3, there is no selective information concerning the epimer of 2, which is probably present in low concentration. Fig. 4 1 , at 400 nm, indicated by shaded rcgions Selective regions for compounds A-F, sample 1 and gradient The application of HELP to the simulations in Tables 3 and 4 is given in Table 9. It can be seen that the errors are substantially smaller than in Table 5 , and that there is no systematic variation with wavelength. It is important to note that all the errors except one are small negative numbers. This is caused by the method used to calculate the two components. First, the left-hand peak (1) is estimated using HELP, to give fijl. The second concentration profile is estimated by subtrac- tion, so that xij2 = XiJ - X r j l ensuring that the two components add up exactly to the true data.The errors are accounted for by computer rounding, and would be in the other direction if component 2 was estimated first. The errors are, however, in most instances very much smaller than for the vertical divisor method. It is also possible to find X k for the two comp?unds, being equal to 120.86 and 145.49, respectively, and Zk, which is (2)784 ANALYST, JULY 1993, VQL,. 118 I I Elution time - Fig. 7 Elution profiles of (from left to right) the six com ounds A-F, as estimated by HELP using sample 1 and gradient 1 at (LIf400, (b) 440 and ( c ) 665 nm equal to 0.4538 and 0.5462, respectively, or almost exactly the true values.This gives us high confidence in the chemometric approach to integration, and solves many of the problems of the vertical divisor method. The methods discussed above were then applied to the chromatograms of mixtures of chlorophyll degradation pro- ducts obtained as described under Experimental. As des- cribed elsewhere ,20721 six regions of selectivity for samples 1-3 run under gradient 1 are determined. These are summarized in Figs. 4-6. These six compounds are labelled A-F in each chromatogram, with compound A being the fastest eluting. The spectra of compounds A, B, D, E and F are virtually identical throughout all three samples, as can be demonstrated by superimposing the spectra.21 The spectrum of compound C in chromatogram 1 differs from that of compound C in samples 2 and 3.This can be interpreted chemically as follows. Compounds E and F are epimeric pairs (1 and 1’). They degrade first to A and C (sample 1) (2 and 2’) and then to B and D (3 and 3’). Finally, B and D degrade to C (samples 2 and 3) (4). Compound C in samples 2 and 3 does not possess epimers. In sample 1 , A is the major epimer compared with C, I 1 Elution time -+ Fig. 8 Elution profiles of (from Ieft to right) the six com ounds A-F, as estimated by HELP using sample 1 and gradient 2 at (~7400, (b) 440 and ( c ) 665 nm I Elution time - Fig. 9 Elution profilcs of (from lcft to right) the six com ounds A-F, as estimated by HELP using sample 1 and gradient 3 at (~$400, ( b ) 440 and ( c ) 665 nm so small amounts of C (sample 1) may still be present in samples 2 and 3, but in insufficient amounts for detection.It is possible to isolate four pure compounds referred to in this paper as 1, 2, 3 and 421 and obtain the UV-visible spectra under identical chromatographic conditions to those used forANALYST, JULY 1993, VOL. 118 785 the mixtures analysed above. The identities of the various peaks, A-F, are summarized in Table 10. Three wavelengths, namely 400, 440 and 66.5nm, were chosen for testing the method of quantification. The elution profiles determined by HELP for samples 1-3 are given in Figs. 7-9, showing the six peaks for whichJhere is*selective information, in each sample. The.values of Xjk and Zjk for the six resolved compounds at these three wavelengths are given in Table 11. The relative absorbances (not normalized) of the four pure compounds are given in Table 12.These numbers are useful when comparing ratios. Table 11 Results of quantification of samples 1-3 using chromato- graphic condition 1 and HELP to resolve six components Wave- length (j)/nm 400 440 665 k A B C D E F A B C D E F A B C D E F k z j k Sample Sample Sample Sample Sample Sample 1 2 3 1 2 3 0.3700 0.5464 0.6587 0.0515 0.1177 0.1735 0.1698 0.4347 0.2864 0.0236 0.0936 0.0725 0.2320 0.6347 0.3731 0.0323 0.1367 0.0944 0.3199 1 .8103 1.3506 0.0445 0.3898 0.3418 5.9878 0.9597 0.9743 0.8327 0.2066 0.2466 0.1111 0.2583 0.2811 0.0155 0.0556 0.0711 0.4324 0.6669 0.7235 0.0562 0.2090 0.2551 0.0927 0.2017 0.1287 0.0121 0.0632 0.0454 0.2360 0.6514 0.3233 0.0307 0.2042 0.1140 0.0831 0.4310 0.2968 0.0108 0.1351 0.1047 6.7293 1.0268 1.0918 0.8753 0.3219 0.3850 0.1144 0.2125 0.2719 0.0149 0.0666 0.0959 0.3785 0.5399 0.5974 0.0553 0.1335 0.1787 0.1.561 0.3844 0.2000 0.0228 0.0951 0.0598 0.2076 0.3626 0.2034 0.0303 0.0897 0.0609 0.2926 1.6311 1.1892 0.0427 0.4034 0.3558 5.7171 0.8964 0.8983 0.8349 0.2217 0.2688 0.0958 0.2289 0.2541 0.0140 0.0566 0.0760 Table 12 Relative absorbances for the four purified compounds.Absolute numbers have no particular significance Wavelengthlnm Compound 400 440 665 1 0.3195 0.3567 0.3239 2 0.3171 0.3581 0.3248 3 0.5379 0.1633 0.2988 4 0.4758 0.4299 0.2124 I As in Tables 5 , 6 , 8 and 9, the information on the spectra of the pure standards can be used to examine the errors in the estimation of relative integrals at given wavelengths using HELP. Because there are six compounds in each mixture, there are 4.5 possible ratios of the form rnmj over all thr_ee wavel_engths.In order to simplify the task, the values of Yjk and Y j k for all compounds and wavelengths are listed in Table 13, where k = 3 and corresponds only to three wavelengths considered below. Examination of Table 13 suggests a further weakness of the single-wavelength approach to quantification even when HELP is employed to resolve the spectra. The UV-visible spectra of the 18 pure components are given in Fig. 10, with wavelengths 400,440 and 66.5 nm marked. As can be seen, 440 and 665 nm are close to the bottom of sharp peaks for compound 3. This means that only very small inaccuracies in wavelength calibration, of 2-3 nm, can make dramatic differ- ences to the relative estimation of integrals.The absolute values of the errors are small, of course, but frequently the Table 13 Errors in estimates of absorbances. k = 3 in this instance, corresponding to the three wavelengths Sample I- A B C D E F A B C D E F A B C D E F Sample 2- Sample 3- 0.2997 0.4501 0.3337 0.5369 0.3176 0.3042 0.3000 0.4992 0.3813 0.5772 0.3047 0.3312 0.3138 0.5300 0.4153 0.5798 0.3068 0.3199 0.3171 0.5379 0.3178 0.5379 0.3195 0.3195 0.3171 0.5379 0.4255 0.5379 0.3195 0.3195 0.3171 0.5379 0.4255 0.5379 0.3195 0.3195 0.3494 0.2442 0.3410 0.1561 0.3570 0.3332 0.3620 0.2179 0.4085 0.1269 0.3434 0.3153 0.3513 0.2210 0.3764 0.1240 0.3406 0.3244 0.3581 0.1633 0.3575 0.1633 0.3567 0.3567 0.3581 0.1633 0.3845 0.1633 0.3567 0.3567 0.3581 0.1633 0.3845 0.1633 0.3567 0.3567 0.3508 0.30.57 0.3253 0.3070 0.3254 0.3626 0.3380 0.2828 0.2101 0.2959 0.3519 0.3535 0.3349 0.2490 0.2083 0.2963 0.3526 0.3557 0.3248 0.2988 0.3246 0.2988 0.3239 0.3239 0.3248 0.2988 0.1900 0.2988 0.3239 0.3239 0.3248 0.2988 0.1900 0.2988 0.3239 0.3239 800 800 350 350 800 350 800 350 800 350 800 350 Wavelengthlnm Fig.10 Extracted electronic absor tion spectra of sample 1 compounds ( a ) A (b) B (c) C (d) D ( e ) E and (f, F, sample 2 compounds (g) A ( h ) B (i) C 0') D ( k ) E (8 F and sample 3 compounds (m) A (n) B (0) C 01) D (4) E (Y) F. The wavelengths 400, 440 and 665 nm are indicated on the figures786 ANALYST, JULY 1993, VOL. 118 Table 14 Values of ri, and Z,, for thc six peaks in the three samples recorded using gradient 1 Compound rix Z k Sample I- A 11.98 0.054 B 3.76 0.01 7 C 6.21 0.028 D 6.45 0.030 E 189.20 0.855 F 2.96 0.014 A 18.04 0.152 B 9.55 0.080 C 16.82 0.142 D 37.70 0.317 E 30.28 0.256 F 7.03 0.061 A 21.26 0.2 10 B 6.11 0.059 C 9.26 0.09 1 D 26.62 0.263 E 31.13 0.308 F 8.57 0.085 Sample 2- Sample 3- ratios of integrals at given wavelengths are used for the purpose of quantification, and this information can be very misleading, even when chemometric approaches are employed, unless the wavelengths are carefully chosen.The errors are much less when the wavelengths chosen are closer to the maxima of the peaks, which are smoother towards the top. This is typical behaviour of Gaussian peaks. The errors in reconstructing the spectra using HELP are less than the digital resolution ( 5 nm) in the spectral dimension.Hence each compound has different optimum wavelengths depending on the spectral properties. A much better approach is to use the values of Xk and z k ; these average out the errors at different wavelengths. As these parameters weight each wavelength according to the absolute value of absorbance, more intense and more accurate wavelengths will assume greater significance. These values are given in Table 14. Some interesting interpretations are possible. For example, the E : F ratio represents the ratio of the epimer of chlorophyll a to chlorophyll a. This stereoisomer should not normally exceed about 25% of chlorophyll a at equilibrium. In sample 1 the ratio is only 1.6%. This suggests that the starting material is substantially free from epimer and so is very pure chlorophyll.As the reaction proceeds (samples 2 and 3) the ratio becomes 23.2% (sample 2) and 27.5% (sample 3), as the result of thermodynamic equilibration. Peaks D and €3 arise from 3 and its epimer. In this instance, the fastest eluting compound is the major compound. The ratios of areas D to B in samples 2 and 3 are 25.2 and 22.4%, respectively, close to the equilibrium epimeric ratios for chlorophyll a . These ratios are not the same in sample 1, but the ratios of C to A (51.9%) (2 and its epimer) and B to A (56.7%) (3 and its epimer) are very close. This may imply that 2 undergoes a reaction that has equal probabilities of both stereoisomers, and the equilibration back to a thermodynamic mixture occurs subsequently, although the failure of HELP to detect the epimer of 2 in samples 2 and 3 requires explanation.Further work must be performed on the chemistry of allomerization reactions of chlorophylls. In order to verify the quantification method, the same samples were analysed using gradient 2. This gradient results in poorer resolution than gradient 1, and there are only four areas of selectivity, corresponding to peaks A, D, E and F for each of the three samples, as illustrated in Figs. 11-13. Because only some of the compounds are resolved it is not possible to compare values of Z k anbd z,k. However, it is possible to calculate the corresponding unnormalized values I A D E F I Elution time - Fig. 11 Selective regions for compounds A, B, E and F, as estimated by HELP using samplc 1 and gradient 2, at 400nm, indicated by shaded regions A D E F Elution time - Fig.12 Selective regions for compounds A, B, E and F, as cstimated by HELP using sample 2 and gradient 2, at 400 nm, indicated by shaded regions I A D E F Elution time - Fig. 13 Selective regions for compounds A, B, E and F, as estimated by HELP using sample 3 and gradient 2, at 400nm, indicated by shaded regions 0 5 10 15 20 25 Time ( i ) Fig. 14 Illustration of vertical divisor method, and parameters mentioned in Appendix 1ANALYST, JULY 1993, VOL. 118 787 of x k and x ; k for E and F and these are given in Table 15, using wavelengths of 400, 440 and 665 nm. Both compounds have identical UV-visible spectra and so the ratios of absorbances should be constant at every wavelength and not require further corrections; these are given in Table 16.As can be seen, there is reasonable agreement between samples and between wavelengths, and also for the over-all ratios (of x k ) . The exception is sample 1 between gradients 1 and 2. This is probably because a small amount of equilibration takes place during analysis, as samples were dissolved in acetone in the HPLC autoinjector for a few hours. However, overall, the comparison of quantification for both gradients gives us confidence in the method and is encouraging. It is important to be able to quantify samples using two independent chromato- graphic conditions. Conclusion It has been shown that there are major weaknesses in the conventional approaches for chromatographic integration using vertical divisors, when peaks are partially overlapping and have differing electronic absorption spectra.Some improvement comes from using multi-wavelength integration, which is rare in chromatography. There are a variety of methods for the resolution of HPLC-DAD spectra using factor analysis, but the approach of HELP is especially useful when there are substantial similarities between spectra. The errors in reconstruction are normally very small , although occasionally, if the spectral resolution is low and there are inaccuracies in wavelength calibration , single-wavelength integration can still result in significant errors if the wave- lengths chosen are in a region where the peak shape changes rapidly. A study of the influence of similarity in spectra and chromatographic separation on the effectiveness of approaches for evolutionary factor analysis is under way.33 However, this paper has shown that prior resolution of overlapping HPLC peaks by factor analysis can result in a dramatic improvement in chromatographic resolution.Such methodology is an essential development for the quantitative study of chlorophyll degradation pathways, which is of major interest in geochemistry, environmental chemistry and bio- chemistry. Table 15 Values of integration parameters €or peaks E and F recorded using gradient 2 2 j k Compound 400nm 440nm 665nm k k Sample I- E 5.87 6.54 6.11 185.40 F 0.22 0.23 0.26 7.06 E 1.04 0.93 1.03 31.04 F 0.26 0.21 0.26 8.07 E 1.16 1.21 1.23 36.26 F 0.30 0.29 0.33 9.88 Sample 2- Sample 3- We are grateful to the Science and Engineering Research Council (SERC GR/E 73376) for support of A.R. and finance for HPLC-DAD facilities, and to the Royal Norwegian Council for Scientific and Industrial Research (NTNF) for support of Y.-2. L. Appendix 1 Consider an observed two-way chromatogram X which consists of I observations in time and J wavelengths. This chromatogram is the sum of chromatograms due to K compounds: K where x k is the true chromatogram of component k and E is an error matrix (‘true’ experimental error). The intensity of each chromatogram is related to the concentration of each compound by x k = &,A, (A21 Where is the chromatogram of unit concentration of compound k and &k the concentration. In this paper, we assume a k is unknown, although it can be obtained from Xk if Ak and hence the molar absorptivities at all wavelengths are known.We assume that the concentration profile of each com- pound, k , is given by the column vector i& whose elements are i & k . The spectrum is given by the row vector G ’ k whose elements are z ; k . Hence the elements of the two-way chromatograms Xk are given by Hence the elements of X are given by K K where E ~ , is an error term. Using conventional approaches to the estimation of concen- trations of closely eluting peaks, a wavelengthj is chosen. If a vertical cursor is placed between two peaks 1 and 2 between times I,, and II2 + 1 (Fig. 14), then the estimated integral of peak 1 at wavelength j is given by andpeak2by I k;* = c xi; (A61 i = f 1 2 + 1 so the estimated ratio of integrals at wavelength j is given by Table 16 Percentage ratios of integrated concentrations for compounds E and F.These are calculated by 1002jd&;E for the single- wavelength case and 100kF/2E for the multiple wavelength case Gradient 1 Gradient 2 Single wavelengthlnm Single wavelengthhm Multiple Multiple 400 440 665 wavelength 400 440 665 wavelength Sample 1 1.8 1.7 1.7 1.6 3.8 3.5 4.2 3.8 Sample2 26.9 20.7 25.5 23.2 25.0 22.6 25.2 26.0 Sample3 28.9 25.9 28.3 27.5 25.9 23.9 26.8 27.2788 ANALYST, JULY 1993, VOL. 118 This compares with the true ratio, which is given by can also be found for each component and K i = 1 The percentage error between the true and observed ratio is given by E;12 = lOO(Pj12 - Fj&Fj12 (A91 It is sometimes better to estimate the integral by summing over all wavelengths, so that Using HELP, there is a small error in the estimated chromatogram, so % = X + F where F is the model error introduced by HELP and E is the ‘true’/experimental noise. We can set up an over-all error G = E + F (A201 which is the error between true and reconstructed data, There is some analogy to Malinowski’s theory of errors, in which the total error is the sum of embedded and extracted error, but the terms are added in quadrature to give an over-all root-mean- square error (which is a single number) rather than in this paper where we co9sider the error matrices.The elements of X are given by i i j and the elements of 2, by The integral for compound k can be estimated at wave- Rijk.length j : which can be compared with the true integral over all wavelengths: Often, it is useful to normalize the data over all compounds, in which case we can define parameters c z k k = l and compared with the true integral: k = l An alternative approach for dealing with single-wavelength data is to normalize over all wavelengths observed, giving (A141 2 j k c &k p. =- Ik J ;= 1 It is sometimes preferable to estimate the integral over all wavelengths, in which case J J 1 and and, similarly, 2 q k ;= 1 This implies that the normalized spectra add up to 1 over the wavelengths considered. Eqns. (A14) and (A15) can be modified to take into account cases where a few wavelengths only are selected, in which event the sum is over these restricted wavelengths.As an alternative approach, it is possible to use methods for factor analysis such as HELP, and to estimate the spectrum of compound k at wavelength j given by the scalar &;k, and the estimated spectrum for compound k is given by the vector &Ik. Normally there is a scaling factor introduced, for example normalizing the maximum value of &jk to 1, but the results below are completely general and do not require this assumption. The HELP method further provides estimated i k k concentration pf;ofiles with elements i&k. We now define XS’ as a matrix consisting of,K rows and J columns, each row being equal to i h f k and XC is a matrix consisting of I rows and K columns; each column is a vector &k consisting of the estimated concentration profiles for each compound.A new matrix However, as discussed above, the relationship between Xk and A k is unknown in the absence of standards. Therefore, it is proposed that normalized relative integrals can be used as preference in some instances. The integral is defined, for a single wavelength, by and and for the multi-wavelength case by I K J I K J is the estimated two-way chromatogram using HELP.ANALYST, JULY 1993, VOL. 118 789 (A30) Similarly, equations can be set up for $k and F j k as in Eqns. (A14) and (A15), and also the ratios Y as described above. In order to examine the errors, we assume that the over-all error arises from errors in the estimation of spectral and chromatographic profiles, so that f i i k = G i k + g c i k (A31) giving so that In practice, the second-order error terms can be neglected in Eqns.(A33)-(A35). The the over-all error at wavelength j for compound k : is given by and the error over all wavelengths is given for compound k by x k = x k + G k (A371 so In order to convert integrals into absolute concentrations, it is best to use the normalized integrals of Eqns. (A26) and (A27). If we have the reference two-way chromatogram Xk of compound k of known concentration a k , then and where Z k is the normalized two-way chromatogram from the mixture for compound k and Ak is the chromatogram of the compound at unit concentration. Then it can be shown that I I provided that the reference chromatogram is measured under identical conditions. The analysis can, of course, be extended to single-wavelength quantification to give estimates of c k , Appendix Structures Discussed in This Paper 2 3 Note: Epimers involve inversion of the stereochemistry at the steric centre on ring V, and are referred to by a prime symbol (’), so structure 1’ is the epimer of chlorophyll a, for example. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 References Treibs, A., Angew.Chem., 1936, 49, 682. Hendry, G. A. F., and Stobart, A. F., Phytochemistry, 1986,25, 2735. Newton Downs, J., and Lorenzen, C. J., Limnol. Oceanogr., 1985, 30, 1024. Hendry, G. A. F., in Willsford Shaft Excavations: English Heritage Archaeological Report Number 11, eds. Ashbee, P., Bell, M., and Proudfoot, E., Historical Buildings and Monuments Commission, London, 1989, pp.96-97. Hendry, G. A. F., Houghton, J. D., and Brown, S. B., New Phytol. , 1987, 107, 255. Brown, S. B., Houghton, J . D., and Hendry, G. A. F., in Chlorophylls, ed. Scheer, H., CRC Press, Boca Raton, FL, 1991, pp. 456-489. Llewellyn, C. A., Mantoura, R. F. C., and Brereton, R. G., Photochem. Photobiol., 1990, 52, 1037. Llewellyn, C. A . , Mantoura, R. F. C., and Brereton, R. G., Photochem. Photobiol. , 1990, 52, 1043. Llewellyn, C. A., MSc. Thesis, University of Bristol, 1989. Smith, J. H. C., and Benitez, A., in Modern Methods of Plant Analysis, eds. Peach, K., and Tracey, M. V., Springer- Verlag, Berlin, 1955, vol. 4, p. 42. Porra, R. J . , and Gimme, L. H., Anal. Biochem., 1974,57,255. Mackinney, G., J. Biol. Chem., 1941, 140, 315. Lichtenthaler, H. K . , Methods Enzymol., 1987, 148, 350. Brereton, R. G., Chemometrics: Applications of Mathematics and Statistics to Laboratory Systems, Ellis Honvood, Chichester, 1990, pp. 20-22. de Figueiredo, J., M.Sc. Thesis, University of Bristol, 1991. Schaber, P. M., Hunt, J . E., Fries, R., and Katz, J. J . , Suzuki, N., Saitoh, K., and Adachi, K., J. Chromatogr., 1987, Mantoura, R. F. C . , and Llewellyn, C. A . , Anal. Chim. Acra, Rahmani, A., Eckardt, C. B., Brereton, R. G., and Maxwell, Liang, Y.-Z., Kvalheim, 0. M., Rahmani, A., and Brereton, Brereton, R. G., Rahmani, A., Liang, Y.-Z., and Kvalheim, Malinowski, E., Factor Analysis in Chemistry, Wiley, New Gemperline, P. J., J. Chemometr., 1989, 3, 549. Maeder, M., Anal. Chem., 1987, 59, 527. Kvalheim, 0. M., and Liang, Y.-Z., Anal. Chem., 1992, 64, Keller, H. R., and Massart, D. L., Chemometr. Intell. Lab. Windig, W., Chemometr. Intell. Lab. Syst., 1992, 16, 1. J. Chromatogr., 1984, 316,25. 408, 181. 1983, 151, 297. J. R., Photochem. Photobiol., in the press. R. G., J. Chemometr., 1993, 7, 15. 0. M., Photochem. Photobiol. , in the press. York, 2nd edn., 1991. 936. Syst., 1992, 12,209.790 ANALYST, JULY 1993, VOL. 118 28 Bratchell, N . , in Multivariate Pattern Recognition in Chemo- metrics, Illustrated by Case Studies, ed. Brereton, R. G., Elsevier, Amsterdam, 1992, ch. 3. Lewi, P. J., in Multivariate Pattern Recognition in Chemo- metrics, Illustrated by Case Studies, ed. Brereton, R. G., Elsevier, Amsterdam, 1992, ch. 2. 30 Wold, S . , Esbensen, K . , and Geladi, P. J., Chemometr. Intell. Lab. Syst., 1987, 2, 37. 29 31 Miller, J . C., and Miller, J . N., Statistics for Analytical 32 33 Chemistry, Ellis Horwood, Chichester, 2nd edn., 1988. Malinowski, E. R., J . Chemornetr., 1988, 3, 49. Elbergali, A., and Brereton, R. G., unpublished work. Paper 21047158 Received September 2, 1992 Accepted February 5, 1993

ISSN:0003-2654    

DOI:10.1039/AN9931800779

出版商:RSC    

年代:1993

数据来源: RSC

摘要:

ANALYST, JULY 1993, VOL. 118 791 Characterization of Acidic Dyes Using Euclidean Distance Measurements of Variables Derived From Hig h-performance Liquid Chromatography-Visible Multi-Wavelength Detection Data Peter C. White* and Tirnoth'y Catterickt Metropolitan Police Forensic Science Laboratory, 109 Lambeth Road, London, UK SE I 7LP A multi-wavelength analytical procedure is described for identifying the structural features of <I00 ng of acidic monoazo dyes using data generated by high-performance liquid chromatography with ultraviolet (UV)-visible diode-array multi-wavelength detection. From a single chromatographic analysis at least 14 variables, including absorbance ratio, peak purity parameters, complementary chromaticity coordinates, UV- visible h,,, values and retention time data can be determined for an eluted dye.By employing a group centroid Euclidean distance measurement (GCEDM) multivariate analytical procedure, these dyes can be classified based on similarities in their chemical structures. The procedure was evaluated by using a group of test dyes and other classes of acidic dyes. The advantages of using GCEDM in preference t o either principal component analysis or cluster analysis for obtaining structural information are discussed. Keywords: Acidic dye; hig h-performance liquid chromatography; ultra violet-visible detection; multi- wavelength detection; Euclidean distance Dyes are used extensively for the colouring of many items, including inks, drinks, foods, textiles and plastics, and can provide evidence in forensic examinations.In many instances qualitative analyses are used to compare the dyes extracted from a control and a suspect sample. However, there are occasions when a positive identification of a dye and hence structural information are required. Nuclear magnetic reso- nance (NMR) spectrometry, mass spectrometry (MS) and Fourier transform infrared (FTIR) spectrometry can assist in the elucidation of chemical structures of certain classes of dyes, but difficulties are encountered with acidic dyes and in particular with the very small amounts of sample available for analysis. Some progress has been achieved by using a fast atom bombardment (FAB) MS technique for acidic dyes but the detection levels are still a major problem.' In previous studies it was established that detection levels below 1 ng can be achieved for dyes using high-performance liquid chromatography (HPLC) with a multi-wavelength ultraviolet (UV)-visible diode-array detector.2 Furthermore, parameters including absorbance ratio, peak purity para- meters (PPP) and complementary chromaticity coordinates (CCC) have been applied to reduce spectral data to just a few numerical values (typically 3-7) and these have been employed to improve dye discrimination and identification.3--5 The chromatographic characteristics [e.g., relative reten- tion time (RRT)] and UV-visible spectral characteristics (e.g., absorbance ratios, PPP, CCC and A,,, values) of a compound are related to its chemical structure. Therefore, it was postulated that if these data were subjected to multi- variate analysis this approach might provide structural infor- mation.In order to test this hypothesis, the monoazo sub-group of the acidic class of dye was selected for this study. Data obtained from 22 of these dyes were subjected to several mu1 tivariate analytical procedures, including group centroid Euclidean distance measurement (GCEDM) , principal com- ponent analysis (PCA) and cluster analysis (CA). A further 18 dyes from this sub-group together with 12 dyes from other acidic dye sub-groups were analysed to confirm that this procedure could generate structural information. The results * Present address: Forensic Science Unit, University of Strathclyde, t Present address: Laboratory of the Government Chemist, Queens Royal College, 204 George Street, Glasgow, UK G1 1XW.Road, Teddington, Middlesex, UK TWll 0LY. obtained are presented and the advantages of employing the GCEDM multivariate analysis technique are discussed. Experimental Dye Samples Authentic samples of the dyes used were mainly obtained from different sources including BDH (Poole, Dorset, UK), Ciba-Geigy (Manchester, UK), Cory European Colour (Lon- don, UK), Hoechst (Halifax, UK), Holliday Dyes and Chemical (Huddersfield, UK), ICI (Manchester, UK), Man- Chester Polytechnic (Manchester, UK) and Pointing (Prud- hoe, Northumberland, UK). Other dyes required to test the procedure were prepared as their sodium salt as described by Vogelh and their structures were confirmed by NMR or MS at Manchester Polytechnic. Individual solutions of these dyes were prepared in the HPLC eluent, which also contained the dye of Colour Index (CI) number 16250 at a concentration of 50 pg ml-I.This dye was included for the purpose of calculating RRT data. The concentrations of the individual dyes were such that they gave a peak height absorbance of not less than 0.02 at 500 nm. Chromatographic Conditions The HPLC analyses of the acidic dyes were performed by using a system based on a polymeric packing material reported previously.4.5 An ACS pump (Model 400; Applied Chromato- graphy Systems, Macclesfield, Cheshire, UK) was used to deliver eluent at 0.7 ml min-1 through a 150 x 4.9 mm i.d. stainless-steel tube packed with PLRP-S ( 5 pm) (Polymer Laboratories, Shrewsbury, Shropshire, UK). The eluent was acetonitrile-water (50 + 50) containing 0.7 g 1-1 of citric acid and 3.396 g 1-1 (0.01 mol I-I) of tetrabutylammonium hydro- gensulfate and was adjusted to pH9.0 with concentrated ammonia solution.Samples were introduced onto the column via an injection valve (Negretti and Zambra, Southampton, Hampshire, UK) fitted with a 5 p1 loop. Detection Conditions Samples were monitored with a linear diode-array multi- wavelength (190-600 nm) UV-visible detector (Model HP1040A: Hewlett-Packard. Waldbron, Germany). Details of this system are available elsewhere.7 The detector was set to792 Group 2 ANALYST, JULY 1993, VOL. 118 monitor wavelengths of 590, 550, 500, 450, 400, 350,300 and 250 nm simultaneously, each with a bandwidth of 20 nm (i.e. , +lo nm). During the chromatographic run the 500 nm signal was selected as the pilot wavelength and was monitored on the plotter.In order to obtain absorbance ratio data, the pilot wavelength was used as the refkrence wavelength and the ratios were determined using a computer program described previously.7 Seven absorbance ratios were obtained, i. e. , A5W:A3W and A5W:A250, where A is the absorb'ance as determined by peak-height measurement of a component at the particular wavelength. Spectra were recorded automatically at the peak apex of an eluting component if the absorbance at 500 nm was greater than 0.001. Absorbance versus wavelength data were obtained over the spectral range 190400nm using a wavelength interval of 4 nm. From these data, PPP values were computed over the wavelength ranges 220-600 nm (UV-visible) and 380-600 nm (visible), and the CCC x and y values were determined over the wavelength range 380-600 nm.Full details for computing these data have been described pre- viously.5 A500 A s g o , A500 A S S O , ASMI : A450, A 5 0 A400, A5C)0 A350, Multivariate Analysis All multivariate analytical procedures were executed on a microcomputer (Elonex Model 286M) fitted with a maths coprocessor board. The PCA and CA techniques were performed using the commercially available computer pro- gram Einsight (Info Metrix, Seattle, WA, USA). Euclidean Group 1 15985 distance measurement data were obtained by using a com- puter program developed specifically for this task. Results and Discussion Hydroxylated monoazo acidic dyes are used as colorants for foods, drinks, water-soluble inks and certain types of fibres, and were therefore selected for this study.Initially 22 dyes were obtained to evaluate the multivariate analysis proce- dures and this group of dyes is referred to as the control group. Their structures, CI numbers and the reference number assigned to each dye for this study (shown in bold) are shown in Fig. 1 and grouped according to structural similarities, viz. , Group 1 phenyl-N=N-monosulfonated naphthol Group 2 nap h th y I-N=N-nap h t hol Group 3 phenyl-N=N-p-hydroxylated naphthyl Group 4 monosulfonated phenyl-N=N-naphthol Group 5 phenyl-N=N-disulfonated naphthol Following HPLC of each dye, their RRT, absorbance ratio, PPP, CCC and UV-visible A, values were computed and the results obtained are given in Table 1.From these data it can be observed that in a few instances no values could be obtained for particular variables. In order to perform multivariate analysis on this set of data, an accepted procedure of assigning a value was adopted. With the absorbance ratio data the missing values were ascribed a value of 100 and for dye 10 a UV A, value of 230 nm was used, as this was the mean of the UV A, for all the dyes. For the purposes of the multivariate analysis these data were normalized by determining the mean value (n), subtracting this from the value of the variable ( x i ) and dividing by the standard deviation (on - HO H03S 0 - N=N 8 S03H 2 15980 H03S HO 16035 OCH, HO S03H 5 SOBH 1601 1 17 HO 16230 '""8 H03S HO S03H 9 N=N 19 8 - S03H S03H 3 11 22 Group 3 14615 14600 15 10 Group 4 15510 15575 HO CH3 HO H03S 0 - N=N 8 H03S a N=N 18 - 16 18 15670 OH HO H03S 8 15690 OZN OH HO H03S p="8 13 Group 5 16100 16150 HO S03H S03H 6 S03H 20 Fig.1 assigned to each dye in this study Chemical structures of the control group of dyes together with their corresponding CT number and the reference number (shown in bold)ANALYST, JULY 1993, VOL. 118 793 Table 1 HPLC and spectral data obtained for the control dyes DY e No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 RRT with respect to 14 0.56 0.56 0.61 0.64 0.65 0.66 0.68 0.70 0.70 0.71 0.73 0.80 0.90 1.00 1.08 1.19 1.12 1.35 1.43 1.62 1.82 2.54 A,,, (UV) 236 236 220 236 220 240 220 220 248 220 236 232 220 220 228 236 232 240 244 220 220 1- Peak purity Chromaticity parameters coordinates Absorbance ratios with respect to 500 nm Amax .220- 380- (visible) 590 512 484 524 480 508 488 508 536 484 480 512 504 572 512 504 484 480 492 516 5 16 >600 5 12 * - - 5.56 50.66 28.60 0.87 16.80 11.61 0.38 23.11 10.02 - - - - - - - - - 0.46 - 550 6.64 19.52 1.08 29.93 1.65 12.45 1.66 0.79 2.26 1.39 1.93 0.40 1.58 1.56 16.60 21.73 6.91 9.01 1.94 0.63 1.43 - * - Indicates absorbance below threshold of 0.001. + No UV A,,,. 450 2.07 1.40 2.42 1.23 2.50 1.74 2.43 2.43 1.47 1.17 2.04 2.27 2.49 2.41 1.43 1.42 1.32 1.82 1.70 2.56 1.40 2.11 400 4.40 2.20 4.18 1.75 4.04 2.81 4.74 2.54 2.70 4.11 3.65 3.10 2.43 4.50 3.01 2.26 1.74 2.86 2.38 3.63 2.02 3.72 350 6.32 5.41 3.91 4.04 8.87 4.67 3.32 3 $9 3.71 7.12 5.13 3.98 3.32 3.01 4.43 6.43 3.59 7.69 4.82 2.13 1.70 6.26 300 3.33 2.16 2.17 1.89 3.15 3.10 2.45 1.92 2.75 2.30 2.18 3.61 1.16 2.37 2.26 2.51 2.03 3.15 2.64 3.92 0.90 2.52 250 0.95 0.96 0.70 0.79 1.24 0.74 0.84 0.72 0.78 2.12 0.80 0.79 0.41 0.77 1.78 1.47 0.74 1.65 0.84 1.38 0.30 1.16 X 0.1232 0.1242 0.2000 0.1262 0.1551 0.1213 0.1594 0.2792 0.1214 0.1551 0.1760 0.1448 0.3396 0.1629 0.1700 0.1244 0.1255 0.1255 0.1244 0.1469 0.3181 0.1751 Y 0.2778 0.1802 0.4205 0.1546 0.3724 0.2274 0.3703 0.4468 0.1529 0.2443 0.3664 0.3473 0.5016 0.3728 0.3080 0.1847 0.1727 0.2471 0.2331 0.3651 0.4483 0.3676 596 nm 361.03 353.37 345.93 333.04 394.65 341.24 350.59 370.95 350.64 441.67 342.08 363.87 404.30 345 .00 424.65 360.75 331.99 382.81 352.54 425.36 318.11 352.53 596 nm 491.67 472.94 513.52 466.21 503.01 482.21 503.29 531.44 472.94 483.09 502.93 498.16 554.15 503.65 492.71 473.75 468.21 484.32 480.66 500.72 540.03 503.09 Table 2 Loadings of the variables and variances for the first three principal components Principal component Loading of variables Variance number (in order of importance)" ("/I 1 380-596 nm, x , y , Lax (visible), A590 41 2 A4m, 220-596 nm 21 3 A450, A300 11 * 380-596 and 220-596 nm = PPP values for the selected wavelength ranges.Aso0, Am), etc., = absorbance ratios ASm : A590, A5m :Am, etc. Principal Component Analysis Principal component analysis was performed on these data and the loading and variances for the first three principal components ( E l , E2 and E3) are given in Table 2. From these results there were some indications that PCA was not an ideal method for classifying this set of data, as the variance for each principal component was low and their sum was only 73% of the total variance.The principal component plots of E l versus E2, El versus E3 and E2 versus E3 are shown in Figs. 2, 3 and 4, respectively. With the knowledge of the dye structures it was possible to ascertain from these principal component plots four groups containing dyes with similar structural features. However, it was difficult to establish clearly the boundaries for these groups, and various attempts to improve the situation by using only the most highly correlated variables proved to be unsuccessful. Owing to the subjectivity in placing of the boundaries and the low variances of the principal components, this PCA technique was considered to be unsuitable for this particular application.Cluster Analysis The software package used permitted CA to be performed by any of the following agglomerative hierarchical techniques: ( a ) nearest neighbour; (b) furthest neighbour; ( c ) centroid; (d) incremental sum of squares; ( e ) median; v> group average; and (g) Lance and Williams flexible. Only one of these, -4.0 L . t I -4.0 0 6.0 Principal component El Fig. 2 PCA of the 22 control dyes using plots of El versus E2. The numbers are those used to identify the dyes. Boundaries of a dye group are indicated incremental sum of squares (ISOS), produced a dendrogram where groups of samples with similar structural features could be identified with confidence and a high degree of discrimina- tion.The ISOS dendrogram is presented in Fig. 5. When a similarity value of 0.47 was selected as the criterion for the detection of sample clusters, this bisected vertically five nodes and therefore produced five groups of samples labelled A-E. Examination of the structures of the dyes within each group confirmed that CA had classified samples on the basis of similarities in their chemical structure. The common basic structures for these groups of dyes are detailed in Table 3. It is interesting that further discrimination within these main groups can be achieved. For example, the o,o'-dihydroxylated dyes in Group D can be divided into phenyl-N=N-naphthyl (8, 13) and naphthyl-N=N-naphthyl (21) dyes. Although CA appeared to provide the desired information, any attempt to classify an unknown completely distorted the groupings in the794 ANALYST, JULY 1993, VOL.118 HO HO -4.0 0 Principal component El 6.0 Fig. 3 other details as described in Fig. 2 PCA of the 22 control dyes using plots of El versus E3. All -4.0 0 Principal component E2 4.0 Fig. 4 PCA of the 22 control dyes using plots of E2 versus E3. All other details as described in Fig. 2 dendrogram. This observation is in agreement with that noted by Romberg,* who proposed the use of CA to classify initially the original set of data and then use discrimination analysis to classify or identify unknown samples. As this procedure requires a much larger number of samples to generate an original set of data, the CA multivariate technique was considered to be unsuitable for this particular study.Euclidean Distance Measurements With the failure to obtain acceptable results with the PCA and CA multivariate techniques, our attention was directed to the feasibility of using Euclidean distance measurement (EDM) as a classification system. A recognized method for determin- ing the similarity between species is to measure and compare their spatial separations, i.e., Euclidean distance. By exten- sion of the Pythagoras theorem the separation between species (e.g., A and B) for which x, y or any number of variables ( n ) have been determined can be computed by the following algorithm: AB = V ( X ~ - XI)^ + (y2 - ~ 1 ) ' + . . . + (112 - n1)2 where xl, yl nd nl are the variables for A and x2, y 2 and n2 are the variables for B. Initially EDMs were calculated from the set of data presented in Table 1 by using a computer program based on the above algorithm.The program was extended to generate a listing of measurements between any individual dye and all of the other dyes in the control group. Analysis of Control Dye Data This approach indicated that it was possible to identify three groups of dyes, viz . , p-hydroxy- (10, 15), o,o'-dihydroxy- (8, 18, 21) and the naphthyl-naphthol dyes (3, 7, 11, 14, 22), but further refinements were required because of (a) the large number of distance measurement listings produced and (b) the difficulties in distinguishing between dyes in a group and the other dyes due to similar distance measurements. Considerable advances were made first by identifying and using only those variables which isolated a group of dyes from the others. This not only improved the separation and discrimination of the three groups identified already, but when extended to the remaining dyes two further groups could be identified.These included the monosubstituted phenyl dyes (1, 18, 19) and the di- and trisubstituted phenyl dyes ( 5 , 12, 1 .O 0.8 I 0.6 0.4 Similarity value 0.2 0 Fig. 5 CA dendrogram obtained from the control dye data using the ISOS distance measurement algorithm. A, B, C, D and E are the five groups generated when a similarity value of 0.47 is usedANALYST, JULY 1993, VOL. 118 795 Table 3 Structures of the dye groups identified by CA Group Common dye structures B H03S 0 - N=N 8 OH HO HO S03H HO Phenyl ring c H o 3 s 0 N = N 9 - / / - + - ( - J - N = N ~ - + m N = N 8 - substitution mono, di and tri - - - SOBH S03H HO HO E g N = N e + H o 3 s o N = N Q - Phenyl ring di-substituted - - S03H 20).(For the purposes of this investigation, the number of substituted groups excludes the sulfonic acid group.) Further improvements were also achieved by altering the normaliza- tion procedure used for variables containing a number of missing values, e . g . , the A 5 M : As90 absorbance ratio. The procedure of using a default value of 100 in this situation created a non-normal distribution and therefore statistical data were only calculated €or those dye samples that produced a measurable value. Finally, significant benefits were gained by using a GCEDM procedure. This approach improved discrimination because it produced a tighter cluster of measurements €or a particular dye classification and, furthermore, simplified the presenta- tion of results as only one set of measurements is produced for any group identification.With this GCEDM procedure the mean value (centroid) of each selected variable €or a given dye classification is calculated and measurements for all the samples being investigated are taken and compared from this reference point. A summary of the results obtained by using this procedure is presented in Table4, which shows the variables selected for the purpose of identifying each group containing dyes of similar structure, together with their GCEDM values. The GCEDM for the next dye not belonging to the group and the largest GCEDM value obtained for each group are also given in Table 4.From these results it can be seen that if the classification procedure is followed whereby each dye group is tested sequentially as shown in Table 4, five classes of dye structure could be identified. Of the six dyes remaining in this control group no other information could be obtained about their structures. Within each class the GCEDM measurements have, with the possible exception of the di- and trisubstituted phenyl dyes, clearly discriminated between dyes within a class and those not of a similar structure. It can be observed that only six of the original variables are required and one of these, the AsOO:Asso absorbance ratio, can be used on its own to split the monosubstituted phenyl dyes from the di- and trisubstituted compounds. This aspect of using a single variable to classify groups of compounds is possible with this GCEDM procedure because of the differ- ence between the centroid values obtained €or the two groups of dyes.A further point to note was that the six variables required were all generated from spectral data. Therefore, it would appear that although the RRT can often be used for discriminating between compounds it is of no value in this classification procedure.796 ANALYST, JULY 1993, VOL. 118 Table 4 Summary of the GCEDM values obtained using selected variables for the groups of dyes with similar chemical structures. { ) Identifies the next dye in GCEDM listing that does not belong to the dye group and [ ] signifies the dye with the largest GCEDM.( ) = Variables used to identify the dye groups Group centroid Euclidean Dye group Dye No. distance (GCEDM) p-Hydroxy-(Asm : A ~ w , PPP 220-596 nm) 15 0.130 10 0.130 (5 0.839) t21 1.9881 o,~’-Dihydroxy-(A~~~ : Asyo) Naphthyl-naphthol [As00 : A3m, As00 : A4007 ccc (Y>l 21 0.002 13 0.004 8 0.007 2.2271 0.122) (3 14 0.064 22 0.096 3 0.104 11 0.106 7 0.110 0.305) 0.7171 ( 5 [9 Monosubstituted phenyl (Asm : ASSO) 18 0.028 1 0.041 19 0.069 0.229) 4.3101 (6 [9 Di- and trisubstituted phenyl (ASOO : AS501 12 0.004 20 0.00s 7* 0.008 5 0.009 0.012} 4.5751 (14 19 * Already classified above as a naphthyl-naphthol dye and can therefore be excluded. S03H 23 HO S03H H3C 0 - N=N 8 - S03H 25 CI a - N = N H g O ’ ” - \ S03H 27 32 H03S 0 N = N H g 0 3 H S03H 33 H03S N H : 3 9 OH H03S 34 CH3 HO S03H H03S a N=N 9 S03H 35 CH3 H03S H03S 6 N=N 8 OH H03S 36 Na02C H03S Analysis of Test Dye Data In order to test this classification procedure further, another 18 structurally related dyes were either prepared or obtained from commercial sources. Some dyes based on chromotropic acid were also included in this group because they are dihydroxylated compounds, which are used extensively but are not directly related to the same class of dihydroxylated dyes used in the above study.Unfortunately, no additional naphthyl-N=N-naphthol or di- and trisubstituted phenyl dyes could be obtained. For the purpose of this investigation this additional group of dyes are referred to as the test group and their structures and dye reference numbers are given in Fig.6. Analyses were performed on these dyes by using HPLC and the results obtained are presented in Table 5. This multivariate procedure was tested by predicting the particular group into which each of the 18 dyes should fall when their data were subjected to CCEDM analysis. Starting with the p-hydroxylated dyes it was predicted that the dyes 31, 32,34,36,37 and 38 should be identified with this group. The first ten distance measurements recorded contained the control dyes 10 and 15 and the six test dyes as predicted. However, two of the chromotropic acid (dihydroxynaphthyl) dyes, 29 and 30, were also included in this this group. In the original study of the control dyes the classification of the p-hydroxylated group was based on just two dyes (10 and 15). However, inspection of the data from this larger group of samples indicated that the inclusion of an additional variable , the y chromaticity coordinate, might remove the dyes 29 and 30, thereby providing a clearly defined classification far this group of dyes.This proved to be the case. 28 H03S 37 y 3 OH OH H2N 0 N=N H03S 0 N=N OH ’ \ 1 , ‘ H03S S03H 29 38 OH OH OH CI N=N a H03S S03H H3C0 0 N=N - - ‘ \ I - HO S03H - 30 39 S03H H03S HO OnN N=N H03S 0 - N=N 8 OH - / \ - H03S S03H 31 40 Fig. 6 Chemical structures and dye reference numbers of the test dye samplesANALYST, JULY 1993, VOL. 128 797 Table 5 HPLC and spectral data obtained for the test dyes RRT with Absorbance ratios with respect to 500 nm Dye respect Amax Amax , No. to14 (UV) (visible) 590 550 450 400 350 300 250 23 0.73 240 496 - * 6.10 2.14 3.35 5.05 3.66 0.80 24 0.74 240 492 - 10.22 1.87 3.02 4.63 3.36 0.76 25 0.75 240 496 - 4.88 1.89 2.66 3.73 3.18 0.72 26 0.59 240 488 - 14.17 1.76 3.13 6.15 3.19 0.88 27 0.81 240 488 - 9.68 1.71 2.73 4.78 2.99 0.76 28 0.73 232 512 38.41 1.41 3.55 5.43 4.44 2.08 1.54 29 0.62 244 576 0.40 0.42 4.29 3.03 2.91 1.20 0.68 30 0.82 228 556 1.45 1.66 3.55 5.52 3.17 2.03 1.00 31 0.56 244 488 - 7.55 1.84 4.16 2.98 1.57 1.27 32 0.77 220 484 42.19 2.83 1.19 3.26 6.23 2.26 1.82 33 0.56 236 484 - 20.48 1.57 2.60 4.70 2.47 0.69 34 0.58 240 488 - 11.20 1.73 4.15 2.74 1.61 1.36 36 0.64 244 496 - 5.24 2.21 4.40 3.47 1.78 1.34 37 0.55 244 492 - 6.97 2.17 6.02 3.56 1.74 1.40 38 0.80 220 488 28.45 2.68 1.11 1.82 4.32 2.02 1.40 39 0.62 224 512 46.95 1.52 1.40 1.67 2.37 2.57 0.57 40 0.73 236 492 - 12.53 1.84 3.62 3.96 3.64 1.35 * - Indicates absorbance below threshold of 0.001. 35 0.56 236 496 - 7.72 1.89 2.91 4.43 2.33 - Peak purity Chromaticity parameters coordinates X 0.1243 0.1220 0.1292 0.1209 0.1236 0.1619 0.3458 0.2801 0.1246 0.1452 0.1223 0.1219 0.1242 0.1253 0.1221 0.1491 0.1570 0.1210 220- y 596nm 0.2840 355.38 0.2452 347.24 0.2748 347.50 0.2235 356.21 0.2307 347.37 0.4375 429.78 0.5242 490.56 0.4597 398.38 0.2480 394.88 0.2299 427.74 0.1949 324.58 0.2250 389.86 0.2602 266.17 0.2940 396.92 0.2819 398.93 0.2248 401.46 0.2693 340.02 0.2305 385.88 380- 596 nm 490.65 485.10 487.01 482.89 481.99 510.43 555.55 535.77 486.39 480.14 477.25 483.78 486.97 492.57 492.81 472.79 483.50 484.84 For the o,o’-dihydroxylated group of dyes, it was impossible to obtain any more dyes based on this structure.However, it was predicted that the dihydroxylated chromotropic acid dyes 28,29 and 30 would be identified with this group. Based on the As~:As90 variable used previously, these dyes could be identified in this group but unfortunately contained another dye (3), and no discrimination could be obtained between the two types of dihydroxylated dyes. However, it was possible to overcome both of these problems by inclusion of two additional variables (AsW : A4s0 and As00 : A4aI). This pro- duced two distinct groups, one for the o,o’-dihydroxylated dyes 8, 13 and 21 with GCEDM values of 0.176, 0.195 and 0.373, respectively, and the other for the dihydroxylated chromotropic dyes 28, 29 and 30, which were grouped at the extreme of the GCEDM values obtained for all the dyes and were in the range 1.126-1.436.A further interesting point with the o,o‘-dihydroxylated dye is that it would appear that it is possible to discriminate between the phenyl-N=N-naphthol dyes (8, 13) and the naphthol-N=N-naphthol dye (21) owing to the difference between their GCEDM values. Unfortunately, no additional dyes of the naphthyl-N=N- naphthol or di- and trisubstituted phenyl classes could be obtained, but GCEDM analyses were still performed. This was essential because it was important to show that the additional data from all the dyes did not interfere with the classification of these dye groups. The results of these studies proved that the inclusion of additional data did not have any adverse effects on the original dye classifications. For the monosubstituted phenyl class of dye it was predicted that the test dyes 23, 24, 25, 27, 35, 39 and 40 would be classified with the control dyes 1, 18 and 19.Analysis of their data by GCEDM resulted in these dyes being grouped in the first 11 measurements (range 0.012-0.365), but this included a non-substituted phenyl dye (3). The structure of the remaining non-classified dyes (both control and test), e.g., 2,4,9,16,17, 26 and 33, were also found to be non-substituted phenyl dyes (sulfonic acid groups excluded). It would be very informative if the mono- and non- substituted dyes could be classified into separate groups. Originally only one variable (AsOO:AssO) had been used to identify the monosubstituted group represented by the control dyes (1,18 and 19). Other variables were now considered in an effort to discriminate between the non- and monosubstituted species.By inspection of all the data it was established that when only the y chromaticity coordinate was used all the non- ubstituted dyes were grouped in a GCEDM range of 0.015- 0.460 whereas the monosubstituted dyes fell within another range, 0.531-1.080. No other dyes were found within these ranges. Overall, the classifications for these dyes were as predicted, but it was evident that with the original control samples there were insufficient numbers of dyes within certain classes to be able to select the variable(s) that could give the best discrimination. However, with the combined control-test dye data this situation was improved considerably, which resulted in the identification of additional groups of dyes, and hence further structural information.With the change in selection of some of these variables it is interesting that still only seven of the original 14 variables are used to identify the various groups of dyes by GCEDM analysis, as illustrated in Table 6. Although there are some indications that additional structu- ral information might be obtained from these results (e.g., the discrimination between the two classes of o, 0’-dihydroxylated dye), it would appear that there are some limitations to the method. For example, the type of position of substituent(s) cannot be determined and the presence of any sulfonic acid group cannot be ascertained. These limitations are due primarily to the weak chromophoric nature and hence small spectral changes often associated with substituents.It must be emphasized, however, that the RRT is an excellent parameter for discriminating between compounds, especially when there are changes in the type, position or number of substituents.9 Therefore, this variable, although excluded in the GCEDM classification scheme, can still be of considerable value in the examination of the structure of a molecule. Analysis of Blind Trial Samples In previous studies it was found that the chromatographic and spectral data recorded with the HP 1040A multi-wavelength diode-array detector can be obtained with very high precision for very small amounts (typically 4 0 0 ng) of dyes and other materials.3 Nevertheless, a blind trial was undertaken to confirm the precision of the spectral and GCEDM data, and furthermore to confirm that overall the proposed scheme could identify the dyes correctly.Six orange dye samples previously analysed during these investigations were submit- ted as solids for this trial and no other information was supplied.798 ANALYST, JULY 1993, VOL. 118 These samples were treated as described under Experimen- tal. Aliquots of these dyes both with and without the RRT dye standard (14) were analysed by HPLC. Following GCEDM analysis all the dye samples were correctly classified and also the correct structures were assigned to each dye after comparing their GCEDM value3 and HPLC data against the data recorded previously in this study. Analysis of Other Acidic Dyes As stated earlier, the monoazo acidic dyes are used exten- sively.However, there are other classes of acidic dye and Table 6 Summary of variables used in the GCEDM procedure for the classification of acidic monoazo dyes Dye class Variables used Comments p-Hydroxylated AS00 : A300 Permits discrimination CCCO) between naphthol and PPP (220-596 nm) 3.6-disulfonated l-naphthol Dihydrox ylated ASOO : A590 Permits discrimination AS00 : A450 between o,o'-dihydroxy- A500 :A400 and 1,s-dihydroxylated napht h yl Naphthyl-naphthol Asm : AmO A5m: A300 CCCb) PPP (220-596 nm) PPP (380-596 nm) Di- and trisubstituted Non- and mono- phenyl As00 : Ass0 substituted phenyl CCC ( y ) Permits discrimination between the two groups of dyes Table 7 Other classes of acidic dyes used to test the GCEDM procedure Dye class Bisazo Trisazo Triarylmethane Xanthene Anthraquinone Azine Indigoid DY e C1 No.No. 26905 27615 30235 34180 42045 44090 45350 45380 60730 63010 50090 73015 41 42 43 44 45 46 47 48 49 50 51 52 these can be chromatographed successfully under the HPLC conditions used in these investigations.4 It was important, therefore, to establish whether any of these would interfere with the monoazo dye classification scheme. The dyes selected as representative of the other classes of acidic dye together with their CI and reference numbers are given in Table 7. The HPLC and spectral data recorded for these dyes are presented in Table 8. These data were combined with the results obtained for the control (Table 1) and test (Table 5 ) dyes and GCEDM analyses were performed.The results for each of the monoazo dye classifications, e.g. , p-hydroxy-, dihydroxy-, were exam- ined and, as shown in Table 9, some of these other acidic dyes were incorrectly classified. The results from this particular study indicated that there were some difficulties in attempting to discriminate between the monoazo dye classifications and other acidic dyes. However, with most of the incorrectly classified acidic dyes these could be distinguished clearly from the monoazo dyes by comparison of the other variables not used in the GCEDM classification procedure. For example, with 53 the values for virtually all the other variables are very different to those for the non- or monosubstituted phenyl dyes, and for dyes 51 and 57 these can be excluded from the dihydroxylated group because their CCC (x) values are outside the range observed for this group of monoazo dye samples.Unfortunately, the procedure proposed above failed to provide any discrimination between the bisazo dyes (47, 48) and the di- and trisubstituted phenyl monoazo class of dyes owing to the very similar values for all of the variables. Overall, these investigations have illustrated that GCEDM analysis of data generated using HPLC with a multi- wavelength detector can considerably improve the character- ization of the monoazo acidic dyes. Furthermore, these results can be achieved with very small amounts of dye, and therefore this method offers considerable advantages over other tech- niques that can normally yield structural information.Table 9 Summary of GCEDM results obtained for the other classes of acidic dyes Other acidic dye classes: dyes incorrectly classified GCEDM dye classification p-H ydrox ylated None Dih ydroxylated 45 (triarylmethane) 51 (azine) Naphthyl-naphthol None Di- and trisubstituted phenyl 41 (bisazo) 42 (bisazo) Non- and monosubstituted phenyl 47 (xanthene) Table 8 HPLC and spectral data obtained for the other classes of acidic dyes RRT with Absorbance ratios with respect to 500 nm Dye respect A,,,, h,,, No. to14 (UV) (visible) 590 550 450 400 350 300 41 1.43 228 516 42 1.54 244 512 43 4.16 220 600 44 1.24 220 600 45 0.67 220 600 46 0.60 240 600 47 0.50 240 496 48 0.90 256 524 49 2.62 252 568 50 0.54 244 600 51 0.66 220 520 52 0.52 286 600 67.50 84.71 0.88 0.56 0.10 0.32 - 0.42 0.29 2.80 0.22 1.93 2.34 2.01 2.56 0.57 0.95 0.37 0.82 0.03 2.58 0.10 0.85 - 4.95 3.62 9.80 0.38 5.56 0.16 2.88 1.06 2.45 0.10 1.21 3.51 3.65 1.28 1.07 0.10 0.54 40.38 23.56 3.42 5.84 3.64 2.01 2.38 3.46 2.14 3.92 1.70 1.70 1.17 0.42 0.39 9.31 0.75 0.25 24.63 7.10 9.17 3.23 0.74 0.43 2.88 0.40 0.64 0.77 0.20 0.07 Chromaticity coordinates 250 2.71 1.38 1.27 0.29 0.16 0.11 2.82 1.27 0.15 0.11 0.55 0.01 X 0.1484 0.1455 0.2731 0.2998 0.4561 0.4043 0.0925 0.1107 0.3511 0.3995 0.2193 0.4084 Y 0.3568 0.3641 0.3252 0.3683 0.4677 0.4304 0.2253 0.6082 0.5338 0.5106 0.4259 0.4789 Peak purity parameters 220- 380- 596nm 596nm 430.22 425.75 465.39 336.22 491.35 350.23 441.27 482.28 314.34 311.29 329.63 338.15 499.72 500.69 516.05 531.11 562.71 570.55 491.67 519.27 555.32 571,12 514.24 574.25ANALYST, JULY 1993, VOL.118 799 Conclusions Information about the chemical structure of a small amount (<lo0 ng) of a monoazo acidic dye can be obtained by performing a GCEDM multivariate analysis on HPLC data generated with a diode-array multi-wavelength detector. With the GCEDM technique it is possiblk to classify these dyes as either p-hydroxylated, o,o’-dihydroxylated, 1,2-dihydroxy- naphthyl, naphthyl-N=N-naphthol or phenyl-N=N-naphthol dyes. Additionally with the latter group discrimination between non-, mono- and the di- and trisubstituted phenyl compounds can be achieved. To obtain this information requires only seven of the original 14 variables computed from the HPLC data and includes the five absorbance ratios (Aso0 : As9(), ASo0 : ASSO, A500 : A450, As00 : A400 and Asoo : A300), the CCC y coordinate and the UV-visible PPP value.With the exception of the bisazo dyes, which are indistin- guishable from the di- and trisubstituted phenyl-N=N-naph- tho1 compounds, discrimination between the monoazo and other classes of acidic dyes, e.g., triazo, triarylmethane, xanthene, azine, anthraquinone and indigoid, can be achieved. Additional structural information regarding the type or position of substituents cannot be achieved with this proce- dure. However, as no other analytical technique can currently provide the required information, this HPLC-GCEDM procedure is invaluable for the characterization of acidic monoazo dyes. Although the GCEDM scheme required refinement in the light of additional data from extra samples, other popular multivariate analytical techniques including PCA and CA are unsuitable for this particular application. With these latter techniques structurally related samples could not be identified clearly, and also data from additional samples were found to distort the original classification scheme. As reduced spectral data proved successful with the GCEDM method, this suggests that the failure of the PCA and CA procedures is due to limitations of the software. It is thought that a Fourier transform of the full spectral data is unlikely to overcome this problem and therefore this approach has not been investigated. We are grateful to Dr. V. Garner at Manchester Polytechnic for preparing some of the dye samples and for confirming the structures of the dyes used in this study. We also express our thanks to B. B. Wheals for his helpful advice and guidance in this project. References Garner, V., personal communication. Wheals, B. B . , White, P. C., and Paterson, M. D., J . Chroma- togr., 1985, 350, 205. White, P. C., Analyst, 1988, 113, 1625. White, P. C., and Harbin. A.M., Analyst, 1989, 114, 877. White, P. C., and Catterick. T., Analyst, 1990, 115, 919. Vogel, A., Textbook of Practical Organic Chemistry, Longman, Harlow, 1978. White, P. C., and Catterick, T., J. Chromatogr., 1987,402,135. Romberg, H. C., Cluster Analysis for Researchers, Lifetime Learning Publications. CA, 1984. White, P. C., Ph.D. Thesis, Brunel University, Middlesex, 1992. Paper 2/06195C Received November 20, 1992 Accepted February 16, 1993

ISSN:0003-2654    

DOI:10.1039/AN9931800791

出版商:RSC    

年代:1993

数据来源: RSC

摘要:

ANALYST, JULY 1993, VOL. 118 80 1 Typification of Alcoholic Distillates by Multivariate Techniques Using Data From Chromatographic Analyses M. Cruz Ortiz, Jose A. Saez and Jesus Lopez Palacios Department of Analytical Chemistry, Colegio Universitario of Burgos, University of Valladolid, Apdo. 231,09080 Burgos, Spain Multivariate chemometric techniques were used t o classify alcoholic distillates and t o develop a typification model for Galician liquors, on the basis of percentage data obtained from nine chromatographic peaks. By using the Bayesian model, the probability of a genuine Galician liquor being rejected is 0.1 1 and that of a false one being accepted is practically nil. Partial least squares was used as a modelling method, taking the liquor category as response variable.This method enables a confidence interval (95%) t o be constructed that does not include any of the other distillates. Keywords: Distillate chromatographic typification; varimax rotation, Bayes model, partial least-squares regression; cluster analysis The use of specific sensors for characterizing foodstuffs is gradually being replaced by a trend to draw on the wealth of information available from the large number of data provided by current analytical instrumentation. The optimum use of all this analytical information is one of the main objectives of chemometrics. 172 In this work such an approach was applied to a specific problem, viz., identifying genuine Galician spirit (on the basis of percentage contents of some volatile compounds as determined by gas chromatography).For this purpose we chromatographed 30 samples: 7 Galician spirits and 23 of different origins (Catalonian, French, Portuguese and Ital- ian). Various multivariate classification [ K nearest neighbour (KNN)] and modelling techniques [partial least squares (PLS), Bayes] (see later for explanation of Bayes method) were applied to the 120 chromatograms recorded (4 per sample) in order to determine whether the origin of the liquor concerned could be ascertained by using a rapid analytical procedure. Also identified were the variables (chromato- graphic peaks) that contributed most significantly to the characterization and which distillates were the most similar to Galician spirits on the basis of the chemical information obtained. The results were fairly consistent and allow the development of a model for characterization of Galician distillates; also, by using the Bayes model, the probability of rejecting a genuine Galician liquor or accepting a false one could be assessed.The study reported in this work was performed on commercially available distillates, which do not comprise a complete sample but include a significantly wide range of products. The origins certified on their labels were always accepted as true. Experimental Experimental data were obtained by using a Perkin-Elmer Model 8600 gas chromatograph equipped with a flame-ioniza- tion detector and interfaced to a computer for data acquisition and processing. Distillate samples (0.4 pl) were injected into the Chromatograph with no pre-treatment.3 The chromato- graphic column used was 4 m x 0.32 mm i.d., and was packed with 15% Carbowax 1500 on Chromosorb W (80-100 mesh).Nitrogen at a flow rate of 20 ml min-l was used as a carrier. The furnace, injector and detector temperatures were kept at 93, 150 and 200°C, respectively. Some preliminary trials were carried out in order to determine the optimum values for various experimental parameters. Variations arising from the way the chromato- graphic peaks were integrated were checked as was the width of the data observation window. It was also checked that the replicate considered had no significant influence on the end results. After the optimum working conditions has been estab- lished, each of the samples was chromatographed four times. The chromatographic peaks used in the analyses were those corresponding to the following individual standards: acetal- dehyde, methyl acetate, ethyl acetate, methanol, ethanol, propanol , butan-2-01, 2-methylpropan-1-01 and amyl alcohols (mixture of 3- and 2-methylbutan-1-01). Experimental data were processed with the aid of the PARVUS statistical package4 and some graphs were treated with the program STATGRAPHICS5 on a Tandon MCS 486/ 33 microcomputer.Results and Discussion Data A matrix whose rows (120) represented the different analysed samples (objects) and whose columns (9) corresponded to the percentage areas of the chromatographic peaks (variables) was constructed. Data were classified as of either category 1 (Galician) or category 2 (all other origins). Univariate Analysis Before the multivariate typification proper was addressed, it was checked that none of the variables could be used by itself to distinguish clearly between the two liquor categories.The separation power was assessed on the basis of Fisher weight4 (those variables with the greatest weight would allow for sharper separations), however, none of the variables con- cerned had a significantly greater weight than the rest. In order to avoid the possibility of a given type of liquor differing from another in the variance of one of the variables rather than its mean value, the Fisher modified weight was also used, which takes into account the different variance of each variable in each category, and found 2-methylpropan-1-01 to have the greatest discriminatory power in this respect.This was confirmed by the Box-Whisker diagrams5 constructed for each variable and liquor category; even for 2-methylpropan- 2-01 (Fig. l), nearly 50% of the values in category 1 were possible values for category 2. In summary, none of the variables was specific, i.e., none allowed Galician distillates to be distinguished from the others. It should be noted that the medians of ethyl acetate, methanol, butan-2-01,2-methylpro- pan-1-01 and amyl alcohols were all clearly smaller for the Galician distillates, which, however, featured higher ethanol contents.802 ANALYST, JULY 1993, VOL. 118 Cluster Analysis of the Objects Searching for 'natural' groupings among the samples is one way of studying the structure of data tables. Because of its unsupervised character, cluster analysis can only be used to perform preliminary, essentially descriptive, scans of the data to be analysed;6,7 however, it is of great potential for typification problems such as that addressed in this work.8 The information provided by cluster analysis is largely com- plementary to that supplied by factor analysis.Factor analysis deals with the covariance matrix, i.e., with mutual relation- ships between variables, whereas cluster analysis describes the nearness between the samples (objects) themselves. In this work we used a matrix consisting of the Euclidean distances between objects (objects were in turn represented by nine-dimensional vectors) as the chromatogram similarity matrix. Thus, we constructed an S12" 120 matrix, the ele- ments of which were the Euclidean distances of each object from the rest.In order to establish clusters we used a hierarchical agglomeration method, viz., a weighted average linkage method.8 This method ensured that individual objects in a cluster were differently weighted in calculating the distances between clusters. The last added element is assigned a greater weight than the previously clustered elements, so this 1 2 Category Fig. 1 Multiple Box-Whisker plot for the percent of the 2-methyl- propan-1-01. Category 1, Galician distillates; and category 2, non- Galician distillates 0 0.1 0.2 0.3 0.4 0.5 .E 0.6 v) 0.7 > .r .- - .- method is particularly suitable whenever the numbers of elements in the categories to be resolved are different. The results obtained in these analyses showed the occur- rence of clusters of objects, i.e., the data table itself contained useful information for classification into the two categories.The results of the cluster analysis are shown as a dendogram in Fig. 2. The first graphical evidence is that the similarity between different chromatographic injections of the same sample was greater than that found between different samples and that it did not interfere with the typification of the distillates. Specifically, at a similarity level of 0.7, six clusters were found, which can be identified as follows. The first cluster, which consisted of the largest number of objects, includes five Italian distillates, together with one Portuguese, one Catalonian, one French and two Galician liquors.However, at a slightly higher similarity level (0.75), the Galician and the French distillates made up a different cluster. The second cluster was essentially composed of Portuguese distillates. The third cluster was made up of the Galician liquors plus one French and one Catalonian distillate. The fourth cluster consisted solely of a French distillate, which was later found to be different from all the others. Finally, the fifth and sixth clusters were composed of Portuguese liquor. Factor Analysis of the Internal Structure of the Chromatograms After the above-described analyses, the nine previously standardized variables were dealt with because the variance differs widely as they are related to their means. For this purpose the number of principal components were deter- mined9 from the correlation matrix.Three principal com- ponents, which accounted for 85.8% of the variance, were considered to be sufficient for such data. Application of Kaiser's criterion,lO viz., taking eigenvalues of greater than unity, also allows three principal components to be chosen. Each principal component is a linear combination of the original variables, the coefficients (loadings) of which are given in Table 1. According to the projection of the objects on the plane formed by the first two components (Fig. 3), variations between injections did not mask the specificity of 0.8 0.9 i n ---c..-.----. .. ' - - - I '--.-------L-C..-?--- -- . .- I P I P C F I G C G P F P G F G C G C F P Distillate origin Dendogram showing the results of cluster analysis.Origin of distillate: G, Galician; C, Catalonian; F, French; P, Portuguese; and Fig. 2 I , Italian Table 1 Loadings of the first three principal components Variables (chromatographic peaks) Principal component PI p2 9 3 p4 p5 P6 P? p x p9 1 -0.34 -0.38 -0.37 -0.39 0.44 -0.22 -0.16 -0.31 -0.30 2 -0.38 -0.28 -0.36 -0.10 -0.07 0.49 0.42 0.36 0.30 3 -0.08 0.15 0.09 0.16 -0.05 0.27 0.60 -0.45 -0.54ANALYST, JULY 1993, VOL. 118 803 each type of distillate. Object groups make up triangles with one of the vertices (a high score in the first principle component) occupied by the Galician distillates and the facing side occupied by the Portuguese liquors. In this factor plane, Galician distillates make up a group that includes two non- Galician liquors, viz., a French and a Catalonian liquor. The principal components are okthogonal. However, latent factors with a chemical significance cannot always be assumed to be orthogonal; the chromatographic peaks involved in the problem must obviously bear internal relationships to one another and to the samples being analysed. The search for a latent structure should take this fact into account; accordingly, an Arthur-Varimax rotationll.12 was carried out in order to endow the structure revealed by the principal components with chemical significance. The rotated components (varivec- tors) had closer variances than the original values. The chemical interpretation sought can be read from Table 2: clearly, the first varivector consists of the proportion of the more volatile chemical components in the distillates (peaks 1- 4), whereas the second varivector corresponds to the less volatile components (peaks 8 and 9) and the third represents those of intermediate volatility (peaks 6 and 7).The virtually identical contribution of ethanol (peak 5) to three varivectors shows that this alcohol is specific to none. The scores for the Galician liquors are the lowest (and negative) for the second rotated component; as the eth has a negative coefficient and occurs in a high proportion P C no1 md Eigenvector 1 Fig. 3 Eigenvector projection of alcoholic distillates. G, Galician; C, Catalonian; F, French; P, Portuguese; and I, Italian the coefficients of peaks 8 and 9 are positive, the differential feature of Galician distillates is the lowest proportion of the less volatile components.The other coefficients contribute nothing significant to this component. Hence the latent structure can be used as a quality index for the Galician distillates. Classification: the K Nearest Neighbour (KNN) Method The KNN method assigns each object to the predominant class among the nearest K neighbours. It is a non-parametric method13 inasmuch as it does not formulate a hypothesis on the distribution of the variables used. The Galician distillate category is obviously included in the other and the number of objects is much smaller. It is, therefore, necessary to use a weighting criterion based on the reciprocal distance. The nearest 8,lO and 12 points were used, and the distances were calculated by using the normalized variables.The results obtained with the nine variables are given in Table 3. The overall percentage of successes was higher than 95%. The only misclassified objects were three samples of the same Catalonian distillate. However, it is significant that only two chromatograms of Galician distillates were misclassified. These two misclassifications can be ascribed to experimental errors as the remainder of the replicates of the same samples were classified correctly. Modelling: the Bayes Method There is a marked conceptual difference between a classifica- tion method, such as that described above, and a modelling method.l The former simply divides the variable space into two disjoint zones in such a way that each distillate belongs to either one zone or the other. On the other hand, modelling involves constructing an enclosure for each category in the space formed by the nine chromatographic peaks.This might reveal the occurrence of outliers that cannot be assigned to either category. Application of the Bayesian analysis4 involves constructing one ellipsoid per category, whose centroid is that of the contained objects, whose directions are those of the axes of the covariance matrix and whose boundary encloses 95% of the overall probability. Once the models have been construc- ted, application of the classification rule entails assigning one object to the class whose centroid lies closest to it. Unlike other classification techniques ( e . g . , linear discriminant analy- sis), the Bayes method takes into account the relationships between the variables in each category considered separately, which is relevant to our case.The classification matrix obtained by applying the Bayes method to the standardized data is shown in Table 4; the Table 2 Loadings of the first three ‘rotated’ principal components Variables (chromatographic peaks) Principal component P I p2 p3 p4 PS p6 p7 P8 p9 1 0.44 0.47 0.48 0.41 -0.39 0.04 0.05 0.12 0.12 2 0.14 0.10 0.08 0.17 -0.33 0.24 0.00 0.61 0.62 3 -0.09 0.12 0.03 0.24 -0.32 0.57 0.68 0.16 0.07 Table 3 Classification matrix (KNN method) KNN assigned category K = 8 K= 10 K= 12 True category G NG Yo G NG Yo G NG Yo Galician (G) 26 2 92.86 27 1 96.43 27 1 96.43 Non-Galician (NG) 3 89 96.74 3 89 96.74 4 88 95.65 Overall 95.83 96.67 95.83804 Table 4 Classification matrix (BAYES method) ANALYST, JULY 1993, VOL.118 1 2.3 Assigned category True category G NG Hits (%) Galician (G) 22 * 6 78.57 Non-Galician (NG) 0 92 100.00 Overall 95.00 p 95% Distance from class (Galician) Fig. 4 distillates. d , Galician and 0, non-Galician Coomans diagram for BAYES model. Two categories overall percentage of successes was 95%. It should be noted that the six misclassified Galician samples also belonged to the correct class, but were more distant from the true centroid. The Coomans diagram14 in Fig. 4 is highly descriptive of the situation. The x-axis represents the distance of the model centroid for the Galician distillates category while the y-axis represents the distance to the centroid of the non-Galician category. Parallels to both axes delineate rectangular zones representing schematically the model of each category.The two rectangles intersect as a square (lower left-hand corner of the graph) that corresponds to the intersection of the two classes. Finally, the upper right-hand corner contains the samples that lie outside both categories. The classification rule is represented by the diagonal: objects lying above it are classified as Galician and objects lying below it as non- Galician. As can be seen, the model is highly specific to the Galician category, as non-Galician objects are not included in it. Even more importantly, all the other distillates lie at a long distance. Only three replicates of Galician liquors lie outside the Galician category but remain closer to the centroid of the Galician model than to the other centroid.The features of the Galician distillates model can be assessed by estimating the actual percentage probabilities of first- and second-class errors using the model as the deciding rule of a hypothesis test. These values were used as percentages under the denominations ‘selectivity’ and ‘speci- ficity’ elsewhere.15 For a given Galician distillates model, the selectivity is given by the proportion of correctly classified Galician objects, and the specificity is the proportion of non- Galician objects that are actually classified as non-Galician. The above indices can be expressed in the form of a hypothesis test: (i) null hypothesis (No) + the distillate is Galician; and (ii) alternative hypothesis (Ha) + the distillate is non-Galician .The critical test region is made up of the chromatograms that fall outside the BAYES ellipsoid constructed for the 1.1 L 6 i II 1 2 Category Fig. 5 Multiple Box-Whisker plot for PLS calculated values. Category 1, Galician distillates; and category 2, non-Galician distillatcs Galician distillate category. The first- and second-class errors associated with the test are as follows: a = pr{rejecting Hd associated with the test are as follows: a = pr{rejecting H,,/H,, being true} = 1 - selectivity = 0.11; and p = pr{accepting Ho/ Ho being false} = 1 - specificity = 0. Modelling: the Partial Least-squares Method In order to confirm the possibility of distinguishing the Galician samples from the others, a new model was construc- ted by using the partial least-squares (PLS) method as a normalizing multivariate transformation. The PLS method combines some useful features for describing one or several response variables Yi by means of a block made up of several predicting variables Xi.Notwithstanding its fairly recent inception, it is already widely used as one of the best regression tools available.16 In practice, the PLS method encompasses various procedures, of which the different variants developed by Martens and Naes‘7-19 are the most frequently used in chemometric applications. In this work a version based on the non-linear iterative partial least squares algorithm was used, which is described in detail elsewhere.19 The PLS method has previously been used for modelling,2o although in a different way.The procedure used in this work involved the following steps. (z) The binary response variable was defined as Ydistlllate = 1 for Galician distillates and Ydistillate = 2 for non-Galician distillates. (ii) Variable Y was subjected to PLS regression on the nine typified variables. The model thus constructed consisted of three components that were determined by cross-validation with three cancellation groups by using an improved procedure reported elsewhere.3 The response variable accounted for 50.6% of the variance. Fig. 5 shows the Box-Whisker graph for the PLS calculated values, Ycal. As can be clearly seen, the Galician distillates have different scores from those of the non-Galician distillates. It is interesting to compare this graph with Fig. 1, which represents the best possible univariate resolution.As noted earlier in commenting on the BAYES model and the structure of the first factor plane of the principal components, Galician distillates make up a sub-set of the distillates as a whole, which is also reflected in the PLS values. (iii) The distribution of Ycal values obtained for each category was analysed. Galician liquors were found to conform to a normal distribution at a significance level of 0.05. ( i v ) The model for the Galician category was the confidence interval for the mean Ycal value on the normality hypothesis, The constructed interval was [1.327, 1.4121 for a significance level, a = 0.05. In the previous exploratory analysis of the principal components, summarized above, a Catalonian distillate and a French distillate were found togelther with the Galician liquors.Careful analysis of the samples in the non-Galician category reveals that the four replicates of one Catalonian distillate had values between 1.164 and 1.229 and those of one French distillate were between 1.245 and 1.320, i.e., both lie outside the constructed interval for the Galician distillates.ANALYST, JULY 1993, VOL. 118 805 Again, the Galician model is specific: it includes none of the non-Galician samples and forms a sub-set within the other distillates. Conclusions None of the chromatographic peaks studied was found to be specific to the Galician distillates studied according to the univariate analyses. Chromatogram variations conformed to a latent chemical structure that can be accounted for by the first three factors and interpreted by performing a varimax rotation.The first rotated factor is related to the more volatile components of the distillates (acetaldehyde, methyl acetate, ethyl acetate and methanol), whereas the second is associated with the less volatile components (2-methylpropan-1-01 and amyl alcohols) and the third is related to the compounds of intermediate volatility (propanol and butan-2-01}. Cluster analysis revealed the occurrence of groupings between the analysed samples. Groups were identified by the predominance of Galician, Portuguese and Italian distillates. The KNN classification method provided an overall per- centage of successes above 95%. The constructed Bayes model is selective and specific to the Galician distillates; the probability of a non-Galician distillate being categorized as Galician is virtually nil.Also, the probability of a genuine Galician distillate being rejected is about 10%. We used a multivariate PLS method and the variable corresponding to the Galician liquor category as the response in order to design a model in the form of a confidence interval for the fitted values. The consistency between the results provided by the BAYES and PLS models and the absolute independence of these methods support their reliability for the typification of Galician distillates on the basis of percentage data from a very simple chromatographic analysis where only nine peaks are required. 1 2 References Forina, M., Armanino, C., and Lanteri, S . , Top. Curr. Chem., 1987,141,91.Massart, D. L., Vandeginste, B. G. M., Deming, S. N., Michotte, M., and Kaufman. L., Chemometrics: a Textbook, Elsevier, Amsterdam, 1988, ch. 20, pp. 329-338. 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Ribkreau-Gayon, J., Peynaud, E., Sudraud, P., and RibCreau- Gayon P., Ciencias y Tkcnicas del Vino. Tomo I: Analisis y Control de 10s Vinos, Hemisferio Sur, Buenos Aires, 1980. ch. Forina, M., Leardi, R., Armanino, C., and Lanteri, S., PARVUS. An Extendable Package of Programs for Data Exploration, Classification and Correlation, Elsevier, Amster- dam, 1990. Statgraphics, Version 5, STSC Inc., Rockville, MD, 1991. Anderberg, M. R., Cluster Analysis for Applications, Academic Press, New York, 1973. Spath, H., Cluster Analysis Algorithms for Data Reduction and ClassiJcation of Objects, Ellis Horwood, Chichester, 1982. Massart, D. L., and Kaufman, L., The Interpretation of Analytical Chemical Data by Use of Cluster Analysis, Wiley, New York, 1983. Jackson, J. E., A User’s Guide To Principal Components, Wiley, New York, 1991, ch. 3-4. Mallo, F., Analisis de Componentes Principales y Tkcnicas Factoriales Relacionadas, Universidad de Lebn, Le6n, 1985, ch. 3, p. 154. Forina, M., Armanino, C., Lanteri, S . , and Leardi, R., J . Chernometr., 1988,3,115. Malinowski, E. R., and Howery, D. G., Factor Analysis in Chemistry, Wiley, New York, 1980, ch. 3, p. 49. Wold, S . , Albano, C., Dunn, W. J . , 111, Edlund, U., Esbensen, K., Geladi, P., Hellberg, S., Johansson, E., Lindberg, W., and Sjostrom, M., in Chemometrics, Mathematics and Statistics in Chemistry, ed. Kowalski, B. R., Reidel, Dordrecht, 1984, Coomans, D., P1i.D. Thesis, Vrije Universiteit Brussel, 1982. Derde, M. P., Kaufman, L., and Massart, D. L., J . Chemometr., 1989,3,375. Martens, H., and Naes, T., Multivariate Calibration, Wiley, Chichester, 1989. Martens, H., Ph.D. Thesis, University of Trondheim, 1985. Unscrambler II. User’s Guide. Version ex 4.0, CAMO, Trond- heim, 1992. Forina, M., Frank, I . , and Lanteri, S . , Regressione, Progetto COMETT per la Chemiometria, Universita di Genova, Genoa, 1990, ch. 3, p. 52. Frank, I., and Kowalski, B. R., Anal. Chim. Acta, 1984, 162, 241. 10, pp. 341-391. pp. 66-68. Paper 2106626 B Received December 14,1992 Accepted January 7,1993

ISSN:0003-2654    

DOI:10.1039/AN9931800801

出版商:RSC    

年代:1993

数据来源: RSC

摘要:

ANALYST, JULY 1993, VOL. 118 807 Multicomponent Determination of Flavour Enhancers in Food Preparations by Partial Least Squares and Principal Component Regression Modelling of Spectrophotometric Data Isabel Duran-Meras, Arsenio Munoz de la Pena, Anunciacion Espinosa-Mansilla and Francisco Salinas Department of Analytical Chemistry, University of Extremadura, 06071, Badajoz, Spain Three multivariate calibration methods, partial least squares (PLS-1 and PLS-2) and principal component regression (PCR), were applied to the simultaneous determination of three flavour enhancers (inosine 5'- monophosphate, guanosine 5'-monophosphate and monosodium glutamate), in mixtures by ultraviolet/ visible absorption spectrophotometry. The absorption and first- and second-derivative absorption spectra of the ternary mixtures were used to perform the optimization of the calibration matrices by the PLS and PCR methods.The results obtained by the application of the different chemometric approaches are discussed and compared. No significant advantages were found for the prior differentiation step. The proposed method was applied satisfactorily to the determination of inosine 5'-monophosphate and guanosine 5'-monophosphate in several food preparations, in the presence of monosodium glutamate. Keywords: Spectrophotometry; multivariate analysis; flavour enhancer; inosine 5-monophosphate; guanosine 5'-monophosphate Under computer-controlled instrumentation, derivative tech- niques and multivariate calibration methods are playing a very important role in the multicomponent analysis of mixtures by ultraviolet (UV)/visible molecular absorption spectropho- tometry.1-3 Both approaches are useful for the resolution of band overlapping in quantitative analysis. Derivative tech- niques have proved to be useful in the resolution of simple binary mixtures and/or turbid background samples ,&* where- as multivariate calibration has been found to be the method of choice for more complex mixtures.9-12 The advantage of multicomponent analysis using multivariate calibration is the speed of the method of determination for the components of interest in a mixture, as a separation step can be avoided. The application of quantitative chemometrics, particularly classical least squares (CLS), inverse least squares (ILS), principal component regression (PCR) and partial least squares (PLS), to multivariate chemical data is becoming more widespread owing to the availability of digitized spectroscopic data and commercial software for laboratory computers.Each method needs a calibration step where the relationship between the spectra and the component concen- trations is deduced from a set of reference samples, followed by a prediction step in which the results of the calibration are used to determine the component concentrations from the sample spectrum. The full-spectrum multivariate calibration methods (CLS, PCR and PLS) are all known to reduce the calibration spectral-intensity data at many frequencies to a relatively small number of intensities in a transformed full-spectrum coordinate system.Both PLS and PCR are factor analysis based methods, which exhibit many of the full-spectrum advantages of the CLS method without suffering the disadvan- tages of this more classical statistical tool. In addition, they retain the ILS advantage of being able to perform the analysis one chemical component at a time while avoiding the ILS frequency selection problems. Principal component regres- sion is simply principal component analysis followed by a regression step. 13 The basic concept of PLS regression was originally de- veloped by Wold,14 and the use of the PLS method for chemical applications was pioneered by Wold and co-work- ers.15316 Partial least squares is related to PCR in that spectral decomposition is also performed, but this decomposition step is performed differently. In PCR, the spectra are decomposed on the basis of the maximum variance between spectral data without using information about the concentrations.Partial least squares differs from PCR in that it uses both the spectral data and concentration data in modelling. Hence, PLS sacrifices some fit of the spectral data relative to PCR in order to achieve better correlations to concentrations during predic- tion. The PLS-2 method differs from PLS-1 in that type 1 is used to perform the decomposition and regression for only one component at a time, whereas type 2 is used to calculate the loading on the basis of all of the concentrations simul- taneously, and only one calibration matrix is necessary. Also, PLS-2 is faster and slightly simpler to use than PLS-1.16 The multivariate calibration techniques are discussed in more detail elsewhere.9.14-24 The PLS-2 method has recently been applied in our laboratory to the determination of the pesticides carbaryl and chlorpyrifoss and to the simultaneous determination of 2- furaldehyde, 5-hydroxymethyl-2-furaldehyde and malonal- dehyde in mixtures.'* Recently, the combination of derivative techniques with multivariate calibration methods has been proposed, and the convenience of such an approach has been evaluated by several workers with contradictory results.2"26 Both PCR and PLS calibrations have been applied to the absorption and first- and second-derivative spectra of mixtures of up to five inorganic components,ZS and the combination of conven- tional, synchronous and derivative synchronous spectroflu- orimetry with PCR has been described for the determination of some nitrogen heterocycles.26 This paper reports on the resolution of ternary mixtures of inosine 5'-monophosphate (IMP), guanosine 5'-monophos- phate (GMP) and monosodium glutamate (MSG).These compounds are widely used as flavour enhancers in food preparations.27 By themselves, these compounds do not have a pronounced flavour effect and, in consequence, high concentrations need to be used to detect them clearly. These high concentrations, mainly for MSG , have been associated with a nervous system sickness.27 Because of this, alternative methods have been pursued with the aim of diminishing the MSG concentration used in food preparations. Monosodium glutamate is currently used as a flavour enhancer in dehy- drated soups and related products, together with small amounts of IMP or GMP or their mixtures, because of its synergistic flavour action .*8 The mixture analysis has been accomplished by application of the multivariate PLS-1, PLS-2808 ANALYST, JULY 1993, VOL.118 and PCR methods to the absorption and first- and second- derivative absorption spectra. These procedures have been applied to the resolution of mixtures of IMP, GMP and MSG, and to the simultaneous determination of IMP and GMP in food samples in the presence of MSG. The results obtained are compared and discussed. , Experimental Apparatus A Beckman (Fullerton, CA, USA) DU-64 spectropho- tometer, connected via an RS-232 connector to an Olivetti (Ivrea, Italy) PCS-286 microcomputer equipped with Beck- man Data Leader software (version 3.0,29) was used.The software was used for spectra acquisition, and storage, manipulation and analysis of the spectrophotometric data. The calculation of the first- and second-derivative absorption spectra was performed by the Savitzky-Golay simplified least- squares method of spectral smoothing and differentiation.30~31 An Olivetti PCS-386SX microcomputer provided with an 80386SX microprocessor and an Intel 387SX co-processor, and equipped with the Lab Calc software package (version A 1 .01) with the PLS plus version 2.0 application software,32 was used. The software allows statistical treatment of the data and the application of the PLS-1, PLS-2 and PCR methods. The software incorporates matrix handling routines, allowing manipulation of files in the Beckman standard format.Reagents The reagents IMP, GMP and L-glutamic acid (monosodium salt) were obtained from Sigma (St. Louis, MO, USA), and standard solutions were prepared by exact weighing and dissolution in demineralized, de-ionized water. A buffer solution of pH 4.0 was prepared from acetic acid and 0.5 mol I-’ sodium acetate. Procedure for Analysing Mixtures of IMP, GMP and MSG In a 25 ml calibrated flask, introduce an aliquot of the sample containing between 125 and 800 pg of IMP, between 125 and 800 pg of GMP and between 11 and 35 mg of MSG. Add 3 ml of acetate buffer solution (pH 4.0) and dilute to the mark with de-ionized water. Record the absorption spectrum between 200 and 350 nm.Measure the spectra of all the solutions against a blank of de-ionized water with 3 ml of acetate buffer solution (pH 4.0). Apply the optimized calibration matrices, calculated by application of the PLS-1, PLS-2 and/or PCR methods, to analyse the spectra of the samples and calculate the concentrations of IMP, GMP and MSG in the mixture. Smooth the spectra obtained through seven experimental points and calculate the first-derivative spectra with a A1 = 10 nm and the second-derivative spectra with a A1 = 14 nm by using the Savitzky-Golay procedure .30,31 Perform the optimi- zation of the calibration matrices by the application of the PLS-1, PLS-2 and/or PCR methods, using the first- and second-derivative spectra, apply it to the analyses of the first- and second-derivative spectra of the samples and calculate the concentrations of IMP, GMP and MSG in the mixture.Procedure for Determining IMP and GMP in Food Preparations in the Presence of MSG For products in dry form, reduce 0.5 g to powder in a mortar and mix in a beaker with 25 ml of de-ionized water. For liquid products, weigh 0.5 g into a 50 ml beaker and mix with 25 ml of de-ionized water. Stir the contents of the beaker for 15 min and centrifuge. Transfer 15 ml into a separating funnel, add 5 ml of water and 20 ml of hexane. Shake vigorously for 5 min and allow the phases to separate. Re-extract the aqueous phase with 20 ml of hexane and centrifuge the aqueous phase. Transfer 17 ml of the supernatant solution into a 25 ml calibrated flask, add 3 ml of acetate buffer (pH 4.0) and dilute to volume with de-ionized water.Record the absorption spectra of these sample solutions, and calculate the concentra- tions of IMP and GMP by analysis of the spectra with use of the optimized PLS-1 , PLS-2 and/or PCR calibration matrices. Results and Discussion In a previous paper,33 derivative spectrophotometric methods were developed for the determination of IMP and GMP, either alone or in mixtures, and in the presence of MSG. For that purpose, chemical and instrumental parameters influenc- ing the determination were optimized. Both IMP and GMP are highly absorbing substances in the UV region of the spectrum: IMP shows an absorption maximum at 248 nm, and GMP exhibits maximum absorption at 253 nm and a shoulder at about 275 nm.Monosodium glutamate absorbs slightly at about 220 nm [Fig. l(a)]. Because of the highly overlapping peaks of IMP and GMP, conventional spectrophotometry cannot be applied satisfactorily to the determination of IMP and GMP, and the mixture has previously been resolved by application of the ‘zero-crossing’ measurement technique over the first-derivative spectra .33 GMP (a) 0.80 a, c m n I 0.40 s: 2 0 200 250 300 350 0.040 IMP GMP -0.040 L I I 200 250 300 350 -0.03 I I I 1 200 250 300 350 Wavelengthlnm Fig. 1 spectra of a ternary mixture of IMP, GMP and MSG (a) Absorbance, ( b ) first-derivative and ( c ) second-derivativeANALYST, JULY 1993, VOL. 118 809 With the aim of improving the analyses for these commonly used food additives in authentic samples, several different chemometric approaches were evaluated.Multivariate cali- brations are useful in spectral analyses because the simul- taneous inclusion of multiple spectral intensities can greatly improve the precision and predic,tive ability. Haaland and Thomas11 made a comparison of the different multivariate calibration methods for quantitative spectral analysis. They concluded that it is very difficult to generalize about the superiority of one method over another, because the relative performances of the methods are often dependent on the particular data set being analysed. They also recommended the use of PLS, in the absence of specific information about the data set. The PLS-1, PLS-2 and PCR methods were evaluated for the resolution of mixtures and a comparative study of the prediction capabilities of the three chemometric approaches in our particular work was undertaken.The methods were evaluated using of three different data sets for the analyses. The absorption spectra and the first- and second-derivative absorption spectra of the mixtures under consideration were used (Fig. 1). Table 1 Concentration data for the different mixtures used in the calibration set for the determination of IMP, GMP and MSG Mixture M1 M2 M3 M4 M5 M6 M7 M8 M9 M10 MI1 M12 M13 M 14 M15 M 16 M 17 MI8 M19 IMP/ pg ml-I 5.58 5.58 5.58 5.58 5.58 5.58 11.16 16.73 22.31 27.89 33.47 27.89 22.3 1 16.73 11.16 11.16 22.31 11.16 15.18 GMPI pg ml-I 32.72 27.27 21.81 16.36 10.91 5.45 27.27 21.81 16.36 10.91 5.45 5.45 5.45 5.45 5.45 21.81 10.91 10.91 14.82 MSG/ pgml-1 451.90 632.66 813.42 994.17 1174.93 1355.69 451.90 451.90 451.90 451.90 451.90 632.66 813.42 994.17 1174.93 632.66 632.66 994.17 795.34 M I 1 A M I M2 M3 M4 M5 M6 GMP Fig. 2 Mixture design for the three-component mixtures used in the data set for the PCR, PLS-1 and PLS-2 methods.The concentrations of the different mixtures are given in Table 1 Experimental Design of the Calibration Matrix and Selection of the Spectral Zone for the Analysis A mixture design was used to maximize statistically the information content in the spectra.9,11,34 A training set of 19 samples was taken. The concentrations of IMP and GMP were between 5 and 34 pg ml-l for both compounds, and the concentration of MSG was varied between 450 and 1400 pg ml-1 through the calibration matrix.In Table 1 the compositions of the ternary mixtures used in the calibration matrices are summarized, and a diagrammatic representation of the mixture design is shown in Fig. 2. The spectral region between 227 and 315 nm, which implies working with 177 experimental points per spectra (as the spectra are digitized each 0.5 nm), was selected for analysis, because this is the zone with the maximum spectral informa- tion from the mixture components of interest. Selection of the Optimum Number of Factors To select the number of factors in the PLS and PCR algorithms, in order to model the system without overfitting the concentration data, a cross-validation method, leaving out one sample at a time, was used.35 Given the set of 19 calibration spectra, the PCR and PLS calibrations on 18 calibration spectra were performed, and using this calibration, the concentrations of the compounds in the sample left out during calibration was predicted.This process was repeated 19 times until each calibration sample had been left out once. The predicted concentrations of the compounds in each sample were compared with the known concentrations of the com- pounds in this reference sample and the prediction error sum of squares (PRESS) was calculated. The PRESS was calcu- lated in the same manner each time a new factor was added to the PLS or PCR models. The maximum number of factors used to calculate the optimum PRESS was selected as 11 (half the number of standards plus one). One reasonable choice for the optimum number of factors would be that number which yielded the minimum PRESS. However, using the number of factors (h*) that yields a minimum PRESS usually leads to some overfitting.A better criterion for selecting the optimum number of factors involves the comparison of PRESS from models with fewer than h* factors. The model selected is that with the fewest number of factors such that PRESS for that model is not significantly greater than PRESS from the model with h* factors. The F statistic was used to make the significance determination. Haaland and Thomas9 empirically determined that an F-ratio probability of 0.75 is a good choice. We selected as the optimum the number of factors for the first PRESS value the F-ratio probability of which drops below 0.75. In Fig. 3, the PRESS obtained by optimizing the calibration matrix of the absorption spectra, and the PRESS obtained by optimizing the calibration matrix of the first- and second- derivative absorption spectra with the PLS-2 method, are shown. When using the absorption spectra data set, the optimum number of factors was found to be three for the PLS-2 and PCR methods.In PLS-1, cross-validation is performed with respect to the number of factors affecting the prediction of each of the compounds individually, and two factors for IMP, three for GMP and three for MSG were found. For the first-derivative spectral data set, the optimum number of factors was found to be four for the PLS-2 and PCR methods, and four for IMP, four for GMP and four for MSG, for the PLS-1 method. For the second-derivative data set, the optimum number of factors was found to be five for the PLS-2 method and six for the PCR method.For PLS-1, values of seven for IMP, five for GMP and five for MSG were found as the optimum number of factors. It can be observed that the optimum number of factors increases as the order of differentiation increases (Fig. 3). The810 ANALYST, JULY 1993, VOL. 118 increase in the number of factors found when using derivative spectra instead of the direct absorption spectra has been reported by other workers.25 This increment in the number of factors affecting the variability of the data can be attributed to the smaller signal-to-noise (S/N) ratio of the derivative signals when compared with the original absorption data. It is well established that the S/N ratio degrades as the order of differentiation increases.1.2J6 It is also known that, when the differentiation is digitally performed, the S/N ratio can be improved by choosing wider filters on the differentiation process but, at the same time, thc peaks are broader with a resulting loss of resolution. Consequently, the smoothing of the data was selected to decrease the amount of noise while maintaining minimal broadening of the peaks.Statistical Parameters The values of the root mean square difference (RMSD), which is an indication of the average error in the analysis, for each component, is N I = 1 RMSD = [1/N,C (2i - ~ i ) 2 ] " . 5 and the square of the correlation coefficient ( R 2 ) , which is an indication of the quality of fit of all the data to a straight line, is N 2 ( X i - X ) 2 i = 1 140 000 120 000 100 000 40 000 30 000 20 000 10000 Optimum factor 6 0 20000 - 0 2 4 where xi is the true concentration of the analyte in the sample i, represents the estimated concentration of the analyte in the sample i, X is the mean of the true concentrations in the prediction set, and N is the total number of samples used in the prediction set, were calculated.In Table 2 the values found for the absorbance spectral data and in Tables 3 and 4 those for the first- and second-derivative spectral data, are summarized. In Table 2, it can be seen that the values of the RMSD are similar for PLS-1, PLS-2 and PCR in all instances. With respect to the values of R 2 , they are virtually the same in the three approaches used and for the three components.In all instances the values of R 2 obtained are satisfactory. In Tables 3 and 4, it can be observed that the RMSD and R2 values for each component, obtained with the first- and second-derivative spectral data, are similar for PLS-1, PLS-2, and PCR methods and not significantly different from the values found when using the absorption spectral data. The RMSD values are an estimate of the absolute error of prediction for each component. The predictive ability of each method and for each component can also be described in terms of the relative error of prediction (REP), which is the square root of the mean square of the error in the prediction for each component, expressed as a percentage of the mean of the true concentrations: N I = 1 REP (%) = lOO/X [1/N 2 (it - ~ , ) 2 ] 0 .5 In Table 5 , the REPS for the three components of the mixture, using absorbance, first- and second-derivative spec- t I I I 80 000 70000 - ( cl 60000 - 50000 - 40000 - Optimum factor 10000 I I I I I 2 4 6 8 0 2 4 6 8 10 Number of Factors Fig. 3 number of factors used in the calibration, for (a) absorbance, ($1 first-derivative and ( c ) sccond-derivative spectral data Representation of PRESS values gcneratcd from thc rcdiction of IMP, GMP and MSG by the PLS-2 method, as a function of the Table 2 Statistical parameters of the PLS-1, PLS-2 and PCR methods with use of the absorption spectral data set PLS- 1 PLS-2 PCR Component RMSD* R2 RMSD* R2 RMSD* R2 IMP 0.3374 (2) 0.9985 0.3623 (3) 0.9982 0.3619 (3) 0.9980 GMP 0.3634 (3) 0.9982 0.3730 (3) 0.9981 0.3757 (3) 0.9981 MSG 14.4680 (3) 0.9977 14.4673 (3) 0.9977 14.3362 (3) 0.9978 * Values in parentheses correspond to the number of factors used for prediction.~- Table 3 Statistical parameters of the PLS-1, PLS-2 and PCR methods with U S ~ of the first-dcrivativc absorption spectral data set PLS- 1 PLS-2 PCR Component RMSD* R2 KMSD* R2 RMSD* R2 IMP 0.2624 (4) 0.9906 0.2622 (4) 0.9991 0.2609 (4) 0.9991 GMP 0.3635 (4) 0.9982 0.3769 (4) 0.9980 0.3709 (4) 0.9981 MSG 16.6666 (4) 0.9970 16.6684 (4) 0.9970 17.1122 (4) 0.9968 * Values in parcnthcses correspond to the number of factors used for prediction.ANALYST, JULY 1993, VOL. 118 81 1 Table 4 Statistical parameters of the PLS-1, PLS-2 and PCR methods with use of the second-derivative absorption spectral data set PLS- 1 PLS-2 PCR Component RMSD* R? RMSD* R2 RMSD* R2 IMP * 0.3303 (7) 0.9985 0.3640 ( 5 ) 0.9781 0.3889 (6) 0.9979 GMP 0.2930 ( 5 ) 0.9988 0.2943 ( 5 ) 0.9988 0.2868 (6) 0.9989 MSG 17.1106 ( 5 ) 0.9968 17.1173 ( 5 ) 0.9968 17.5265 (6) 0.9967 * Values in parentheses correspond to the number of factors used for prediction.Table 5 Relative error of prediction values (%) for IMP, GMP and MSG by PLS and PCR methods. A, lD and *D represent absorbance, first-derivative and second-derivative data, respectively PLS-1 PLS-2 PCR A ID 2D A ID 2D A ID 2D IMP 2.42 1.88 2.34 2.60 1.88 2.44 2.69 1.87 2.79 GMP 2.51 2.52 2.02 2.58 2.61 2.04 2.60 2.57 1.98 MSG 1.83 2.11 2.17 1.83 2.11 2.17 1.82 2.17 2.23 Table 6 Composition of the synthetic mixtures of IMP, GMP and MSG for its resolution by PLS and PCR methods IMP/ GMPI MSGI Mixture pgml-1 pg ml-1 pg ml- 1 P1 16.73 10.91 813.42 P2 22.31 10.91 632.66 P3 19.35 16.36 542.28 P4 25.13 10.91 542.28 P5 13.97 8.20 994.17 tral data, are reported.The values found for the REPS are, in all instances, at about 2%, and there is no significant difference between the precision of prediction for the absorbance and for the first- and second-derivative data in either PLS-1, PLS-2 or PCR. Application to Synthetic Mixtures The proposed PLS-1, PLS-2 and PCR methods, applied to both the absorption spectra and to the first- and second- derivative spectra, allow the resolution of synthetic mixtures of the three components. In Table 6, the composition of the ternary mixtures assayed is shown.In Fig. 4, the results obtained by the application of the three methods are represented by using the absorbance and first- and second-derivative absorption data. The results obtained by application of PLS-1, PLS-2 and PCR are not significantly different from each other, in agreement with findings of other workers.25.26 In spite of the statistical analysis performed on the predicted ability of the methods, showing no significant differences in the REP values, when using absorbance or first- or second- derivative data, slightly better results were obtained for the prediction of the five synthetic mixtures assayed, when using the absorbance data set. Contradictory results about the convenience of applying differentiation techniques prior to the use of multicomponent calibration methods can be found in the literature.Jones et ~ 1 . 2 6 applied factor-analysis multicomponent methods to the determination of acyclovir and its major degradation product, guanine, by using several luminescence analytical signals. They found that use of synchronous spectral data was the best choice for the determination of acyclovir, followed by second- derivative synchronous, excitation and emission spectra, respectively, whereas for the determination of guanine, use of second-derivative synchronous spectra was the best choice, followed by excitation and synchronous spectra. On the other hand, MacLaurin et aZ.25 applied several multivariate calibration methods to UV/visible spectra for the simultaneous multicomponent determination of Cr, Fe, Co, Ni and Cu in mixtures.They made a comparative study applying the methods, on the use of absorbance, and first- and second-derivative data. They did not find significant differ- ences in the predictions from the absorbance and first- derivative data with PCR and PLS. The second-derivative data yielded much less precise prediction and they rationalized the fact taking into account the much poorer S/N ratio of the second-derivative signal compared with that of the direct absorbance signal. No significant difference was observed by these workers in the precision of prediction between the PCR and PLS methods in any of the experiments, in accordance with our own observations. Our findings mainly agree with those of MacLaurin et aZ.25 Simultaneous Determination of IMP and GMP in Food Samples by the PLS and PCR Calibration Methods As already stated, no significant advantages were found with the application of the differentiation technique in our particular work.Because of this, for the analysis of authentic samples, it was decided to apply the multicomponent methods of calibration directly under the absorption spectra, without ‘reworking’ the data. Although the proposed method allows the determination of IMP, GMP and MSG in ternary mixtures, the low absorbance from MSG in the samples analysed did not allow the quantification of this component. Tt is well established that, in the chromatographic determination of the three components in food preparations, MSG is usually determined by refrac- tive-index detection instead of UV detection .37 The three chemometric factor analysis-based methods were applied to several food preparations as described under Experimental. The standard additions method was used to check the procedure, and a calculation of the recoveries obtained was effected.The predicted concentrations and the percentage recoveries obtained with the three procedures are summarized in Table 7. Satisfactory results were found in all instances, with recoveries ranging from 80 to 124%. Conclusions Chemometrics has generated much interest in analytical molecular spectroscopy . UI traviole t/vi si ble spectra contain non-specific data, which can be converted into useful informa- tion by multivariate calibration methods. Clear explanations of the different chemometric methods and properly designed user-friendly software should provide a bridge between chemometricians/mathematicians and potential spectroscopic users, enabling them to make successful use of these powerful tools.Superior performance for the analysis performed with the full-spectrum methods has been demonstrated, when compar- ing the results with those found by applying the ‘zero-crossing’ first-derivative method, in which only a single wavelength signal is used for the analysis.12 As already stated, different workers have reported contra- dictory results regarding the convenience of using differen- tiation techniques as a prior step in the application of multivariate calibration methods.25.26 It is well known that the resolution increment achieved by differentiation of the absorption spectra is due to the ability to discriminate between812 ANALYST, JULY 1993, VOL.118 120 100 80 60 40 20 0 120 7 120 100 80 60 40 20 0 120 100 80 60 40 20 0 120 100 80 60 40 20 0 100 80 60 40 20 0 P I P I P2 P3 P4 P3 P2 P4 P I P2 P3 P4 P5 120 I 120 100 80 60 40 20 0 ( ( P4 P5 P I P2 P5 P3 P2 P4 P1 P3 P2 P3 P I P4 P5 120 100 80 60 40 20 0 1 120 100 80 60 40 20 0 P5 P1 P2 P I P2 P3 P4 P3 P4 P2 P I P3 P4 P5 WPLS-ICSIPLS-2 O P C R Fig. 4 Diagrammatic representation of the percentage recovery found in the analysis of five mixtures of IMP, GMP and MSG by PLS-1, PLS-2 and PCR methods, using (a)-(.) direct absorbance, (d)-Cf) first-derivative and (g)-(i) second-derivative spectral data. ( a ) , ( d ) and ( g ) IMP; ( b ) , (e)and ( h ) GMP; (c), Cf) and (i) MSG Table 7 Recovery of IMP and GMP in food preparations by PLS-1, PLS-2 and PCR methods with use of the absorption spectra data set (values given in kg ml-1 except where indicated) Added Found PLS-1 PLS-2 PCR Recovery Recovery Recovery Recover y Recovery Recovery IMP GMP IMP +SD*(%) GMP +SD(%) IMP f S D ( % ) GMP +-SD(%) IMP +SO(%) GMP +SD(%) Instant - - 2.50 - 3.33 - 2.92 - 3.27 - 2.92 - 3.23 - noodle 18.00 18.00 21.37 105 2 2 25.65 124k3 21.79 105 t 2 25.57 1 2 4 f 2 21.79 105 +-2 25.31 122k2 soup 9.00 32.40 10.91 9 3 k 3 36.03 I01 5 2 11.38 94+4 36.01 101 f l 11.39 9 4 k 4 36.59 9 9 k 2 36.00 5.00 38.19 9 9 f 3 9.16 1 1 6 f S 38.71 99+3 9.09 1 1 6 f 5 38.72 9 9 f 3 9.04 116+5 Chicken - - 0.70 - 3.15 - 0.53 - 3.65 - 0.74 - 3.02 - vegetable 4.08 4.08 4.45 92L2 7.48 1 0 6 f 9 4.28 9 2 k 4 7.68 99+- 1 4.42 9 0 k 3 7.43 108f 10 soup 17.00 20.00 19.61 109f3 23.11 99+ 1 17.05 97-t 1 23.04 97+ 1 20.17 114+-1 22.99 100+1 30.60 11.22 32.35 103 + 3 11.55 8 0 k 2 29.60 95-1 1 14.95 101 + 1 32.73 t 0 4 f 3 10.06 8 4 f 3 ' SD = standard deviation.ANALYST, JULY 1993, VOL.118 813 broad and sharp bands.36 From the results obtained, it is clear that the application of differentiation techniques in combina- tion with multicomponent calibration methods can be advan- tangeous in those instances in which significant differences in the spectral widths of the mixture components exist. How- ever, in mixtures with overlapped components with similar spectral bandwidths, the application of differentiation tech- niques is not likely to produce a better resolution and, in consequence, their use is not recommended.Hence, it follows that it is not possible to generalize about the convenience of applying this technique as a prior step to the application of multicomponent calibration methods; on the contrary, each mixture to be analysed should be investigated separately. On the other hand, it is also known that the degradation increment in the S/N ratio increases as the order of differen- tiation increases. In consequence, in those instances in which the differentiation process appears convenient for improving resolution, a trade-off between resolution improvement and S/N degradation must be taken into account. The authors acknowledge the CICYT of the Ministry of Education and Science of Spain (Project PB91-0856) for financial support of this work.1 2 3 4 5 6 7 8 9 10 1 1 12 13 References O’Haver, T. C., and Green, G. L., Anal. Chem., 1976,48,312. O’Haver, T. C., Anal. Chem., 1979, 51, 91A. Geladi, P., and Kowalski, B. R., Anal. Chim. Acta, 1986, 185, 1. Salinas, F., Berzas Nevado, J . J . , and Espinosa, A., Analyst. 1989, 114, 1141. Espinosa Mansilla, A., Muiioz de la Pefia, A., Salinas, F., and Zamoro, A., Anal. Chim. Acta, 1992, 258, 47. Tu. D., Xue, S . , Meng, C., Espinosa Mansilla, A . , Miinoz de la Pefia, A . , and Salinas. F., J . Agric. Food Chem., 1992,10,1022. Espinosa Mansilla, A., Berzas Nevado, J. J., and Salinas, F.. J. Assoc. Off. Anal. Chem.. 1992, 75, 678. Guiberteau Cabanillas, A., Galeano Diaz, T., and Salinas, F., Analusis, 1991, 19, 262.Haaland, D. M., and Thomas, E. V., Anal. Chem., 1988, 60, 1193. Haaland, D. M., and Thomas, E. V., Anal. Chem., 1988, 60, 1202. Haaland. D. M., and Thomas, E. V., Anal. Chem., 1990, 62, 109 1. Espinosa Mansilla, A., Muiioz dc la Pciia, A., Salinas. F., and Martinez-Galera, M., Anal. Chim. Acta, 1993, 276, 141. Jolliffe, 1. T., Principal Component Analysis, Springer-Verlag, New York, 1986. 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 Wold, H., in Research Papers in Statistics, ed. David, F., Wiley, Ncw York, 1966, pp. 411-444. Wold, H., in Systems Under Indirect Observation, cds. Jorcs- kog, H., and Wold, H., North-Holland. Amsterdam, 1982, vol. Wold. S . , Martens, H., and Wold, H . in The Multivariate Calibration Problem in Chemistry Solved by PLS, eds. Ruhe, A . , and Kagstrom, B., ‘Matrix Pencils’ (Lecture Notes in Mathematics). Springer, Heidelberg, 1983, pp. 286-293. Wold, S . , in Food Research and Data Analysis, eds. Martens, H., and Russwurm, H . , Applied Scicnce Publishcrs, London, 1983. Lindberg, W., Persson, J . A., and Wold, S . , Anal. Chem., 1983. Martens, H., and Nacs, T., TrAC Trends Anal. Chem., 1984,3, 204. Naes, T., and Martens, H., TrAC Trend3 Anal. Chem., 1984,3, 266. Geladi. P., and Kowalski, B. R., Anal. Chim. Acta, 1986, 185, 19. Beebe, K. R., and Kowalski, B. R., Anal. Chem., 1987. 59. 1007A. Martens, H., and Naes, T., Multivariate Calibration, Wiley, Chichester, 1989. Sanchez, E . , and Kowalski, B. R., J. Chemomctr., 1988, 2, 247. MacLaurin, P., Worsfold, P. J . , Crane, M., and Norman, P., Anal. Proc., 1992. 29, 65. Jones, R . , Coomber, T. J . , McCormick, J. P., Fell, A. F., and Clark, B. J., Anal. Proc., 1988, 25, 381. Expert Pancl on Food Safety & Nutrition, Institute of Food Tcchnologists, USA, Food Technol., 1980. 34, 49. Coultate, T. P., Food: The Chemistry of its Components, Royal Society of Chemiqtry, London, 1984. Data Leader Software Package, Beckman. Fullerton, CA, 1989. Savitzky, A.. and Golay, M. J . E., Anal. Chem., 1964,36,1627. Steiner, J., Termonia, Y., and Dcltour, J . , Anal. Chem.. 1972, 44, 1906. Lab Calc Software Package, Galactic Industries. Salem, NH, 1989. Duran Meras, I., Mufioz de la Peiia, A., Salinas, F., and Lopez Rosas, M., J. Assoc. Off. Anal. Chem., 1993. in the press. Cornell, J . A., Experiments with Mixtures, Wiley, New York, 1981. Stone, M. J. R., Statist. Sue., 1974, 36. 111. O’Haver. T. C., Anal. Proc., 1982, 19, 22. Nguyen, T. T., and Sporns, P., J. A~soc. Off. Anal. Chem., 1984, 67, 747. 2, pp. 1-54. 55, 643. Paper 210591 I H Received November 5, 1992 Accepted February 9, 1993

ISSN:0003-2654    

DOI:10.1039/AN9931800807

出版商:RSC    

年代:1993

数据来源: RSC