Anal. Chem. 1989, 6 1 , 831-838
831
Determination of Mono- and Oligosaccharides in Fermentation Broths by Liquid Chromatographic Separation and Amperometric Detection Using Immobilized Enzyme Reactors and a Chemically Modified Electrode Gyorgy A. Marko-Varga Department of Analytical Chemistry, University of Lund, P.O. Box 124,S-221 01 Lund, Sweden
Liquid chromatographic (LC) determlnations of mono- and disaccharides In complex fermentation broths and beverage samples are disturbed by the presence of interfering matrix components. High selectivity can be performed by coupilng of LC to immobilized enzyme reactors and amperometric detectkn. The carbohydrates eluting from the column are fhst introduced into a reactor contalnlng immobilized amyioglucosidase which hydrolyses the oilgosaccharides into glucose. A second reactor follows the first one and contains colmmoMiized giucosedehydrogenase and mutarotase. The monosaccharkles eMng from the fkst reactor are mbed with a make-up flow consisting of a nlcotlneamide adenine dinucleotide (NAD') buffer. The carbohydrates are oxldlzed In an equivalent amount of reduced coenzyme (NADH) which is detected eiectrochemkaily by using an electrode modlfled with a phenoxazine derlvative. The postcolumn system was applied to a high-energy sofl drink, malt beer, and fermentation broths from the penicillin industry.
INTRODUCTION Determinations of poly-, oligo-, and monosaccharides are of great importance in biotechnology and food science. It has been known for a long time that in fermentation processes the production of, e.g., penicillin is dependent on the carbon source. In chemically defied media starch, dextrins, lactose, saccharose, or glucose are used (I). The cost of the carbon source is important to the overall economy of the fermentation process making the analysis of the sugars a tool to control the process and process yield. However, analysis of carbohydrates in such a complex media is not an easy task. A variety of chromatographic systems can be used for the separation and analysis of oligo- and monosaccharides. Most common is normal-phase partition using an amine column with acetonitrile/water as the mobile phase (2,3), although the toxicity and costs of acetonitrile are drawbacks of the method. Reversed-phase columns with water as the eluent also provide for separation of oligosaccharides. Elution in these systems takes place in the order of increasing molecular weight (4,5).The separation at ambient temperature of the various anomeric forms complicates the chromatogram (3,6) and is thus disadvantageous. New types of polymeric gel permeation columns have been developed for the separation of smaller molecules and proved useful for the separation of oligosaccharides (7). Polymeric cation exchange columns modified by binding of metal ions like Ca2+or Ag+ with water as the mobile phase are also frequently used for the separation of oligosaccharides (8,9).The interaction with the stationary phase is weak and the binding kinetics are slow. The optimized separation is therefore performed at elevated temperatures (10). 0003-2700/89/0361-0831$01.50/0
Refractive index (RI) and UV (20
5.0 4.1 4.2 3.5 5.6 3.5 1.2
100
300
Table 111. Determination of Carbohydrates in a High-Energy Soft Drink (SD) and in Malt Beer (MB) carbohydrate
nn
dilution
SD meanb
RSD'
n"
dilution
mead
RSD'
DP4 DP3 maltose glucose
6 6 6 6
50 50 50 50
14 22 40 128
5.3 4.8 5.1 3.6
4 4
3 3
55 100
4.6 3.1
MB
" Number of replicates. Concentration, pM. Relative standard deviation, %. favorable pH was found to be 6.0 and was used continuously throughout all experiments reported below. The linearity of the detector response to the various substrates shows strictly linear calibration curves (correlation coefficients equal to at least 0.99961,reaching almost 3 decades. The slopes, the upper and lower linear ranges, as well as the detection limits of the various maltosaccharides will vary according to their respective glucose contents and turnover rates with AMG. This can be illustrated by comparing DP6 and DPI. The splitting efficiency of DPs to form glucose was 56% compared to 60% for DPI. As DP6 contains six glucose monomers, whereas DP, contains four, the slope of the calibration curve for DPs will be higher. Xylose is unaffected by AMG but is oxidized by GDH at a much lower turnover rate than that for glucose (11). MT is also active in the transformation of a-D-xylose to D-pxylose. With the operating conditions the response for xylose was found to be 28% of that for glucose. The detection limits for the various substrates are found in Table 11. For glucose the lowest detectable amount was 100 ng. Corresponding values are found for the oligosaccharides, maltose and xylose. No further attmepts were made to improve the detection limit to a lower region as the sugar contents in the applied samples are well above this level; see below. The noise level of the graphitic material used as the working electrode is higher compared with those of glassy carbon and carbon paste electrodes (38,39). The background current was typically A4 nA, see Figure 2, and varied somewhat between different graphite electrodes. At high concentrations of the substrates the calibration curves will deviate from linearity. The deviation will start at lower concentrations for the larger oligosaccharides falling with decreasing chain length. Recalculated into glucose units they will all start to deviate at equal concentrations, and the limiting step for linearity is thus found in the MT/GDH reactor. An increase in the NAD+ content in the buffer could increase the upper linear range, cf. reaction 2, but was also found to increase the noise level of the background current (30). The reproducibility of the system was investigated by injecting 25 repetitive injections of one standard, B3; see Table 11. The relative standard deviation (RSD) values for the six substrates were around 4% except for maltose and xylose, the
5
IOnA
I
I
-
0
10
20
t/rnin
B
1
1OOnAI
I
I
I
0
10
20
t/min Figure 5. (A) Chromatographic recordings for a higknergy soft drink diluted 50-fdd and (B) a fermented malt beer sample diluted Sfold: (1) oligosaccharides > DP,; (2) mabtetraose; (3) maltotriose; (4) maltose; (5) glucose. For concentrations see Table 11.
substrates with the lowest response factors for which values of 5.6% and 6.2% were obtained. These RSD values are within acceptable limits for sugar determinations in biological samples (8). Application to Real Samples. Figure 5 shows the resulting chromatogramsobtained for injected samples from the high-energy soft drink, PULSE (A) and the beer (B). The various peaks were identified by their respective retention times as well as by the addition of standards; see Table 111. The soft drink (A) proved to contain a fraction of maltosaccharides with a chain length larger than five (an early eluting peak) as well as DP,, DP3, maltose, and glucose. The beer sample also contained a larger sugar fraction (>DP,), as well as maltose and glucose. The reproducibility of the
838
ANALYTICAL CHEMISTRY, VOL. 61, NO. 8, APRIL 15, 1989 3
I
I
I
I
20
10
Time / min
0
4
2
6
log(m/rg)
Flgure 7. Chromatogram of a spiked penicillin fermentation broth diluted lO-fA and sample pretreated as described h the Experimental Section: (1) and (2) starch fractions; (3)7.4 pg of makohexose (DP,); (4)maltotetrose (DP,); (5)3.7 pg of maltotriose (DP& (6) 1.25 pg of maltose; (7)1.2 pg of glucose; (8) 0.33pg of xylose. Two Aminex HPX-87C columns were coupled in series with a column flow rate of 0.55 mL/min and a make-up volume of 0.15 mL/min at 65 O C and an applied potential of 0 mV vs Ag/AgCI.
Table IV. Determination of Glucose in Fusarium oxysporum Broth
Flgure 6. Breakthrough curves of 10-folddiluted penicillin broth (white dots) and FusariWn oxysporum broth (black dots). Adsorbent materials
sample time/h
no
glucose/pM
RSD/%
are (circles)benzenesulfonic acid silica, (triangles) aminopropyl silica, and (rectangk) quarternary amine silica. Observed concentration (C) was relative to steady-state plateau concentration C, vs amount m of broths.
42 61.5 68
8 8 8
62
390
3.5 5.6 3.3
(I
analysis was good and repetitive injections (n = 70) affected neither the separation column efficiency nor the detection system. When samples taken out of fermentation broths were analyzed, pretreatment was necessary mostly to prevent clogging in the separation column. Centrifugation and membrane filtering of the samples turned out to be insufficient. The sample liquid was still strongly dark brown. As shown in a previous report, strongly colored compounds can be completely removed by passing fermentation samples through solid-phase extraction columns (11). The three demands of such cleanup columns are as follows; the adsorption of strongly brown colored componentsmust be essentially complete, the capacity of adsorption of each column must be sufficient so that an appreciable amount of filtered sample may pass before breakthrough of interferents occurs, and the sugars to be analysed should not be significantly adsorbed. Separate analyses were undertaken when standard solutions were added to the sample before and after passing the cleanup columns. No loss of sugars could be found, revealed by equal responses and by RSD values not higher than for standard solutions. The capacities of adsorbing interferents of the various cleanup columns were determined with breakthrough experiments; see Figure 6. The capacity was calculated as the amount of interferents, absorbing at 225 nm, that could be loaded onto 1 g of the adsorbent until breakthrough was noticed. As is seen in Figure 6, the capacities depend strongly on the type and function on the adsorbent. The largest capacity was found for the benzenesulfonic acid-silica column. The other two, with amine and quartenary amine functions, were insufficient for this purpose. One part of the content of the fermentation broths that absorbs at 225 nm was only weakly adsorbed by the sulfonic resin while the major part was strongly retarded. The parts are referred to in Figure 6, curves A and D, as fractions 1 (weakly adsorbed) and 2 (strongly adsorbed). As fraction 1 is not strongly adsorbed in the cleanup column, it was investigated whether this fraction would adsorb on the separation column or otherwise interfere with the detection system. No adsorption could be noticed with this fraction. Under the conditions used for separation of the sugars, it was found that fraction 1 moves with the solvent front through the separation column as expected and will thus cause no interference.
800
Number of replicates
Consequently only fraction 2 will disturb the chromatographic analysis and must therefore be removed. As argued above this is best accomplished with the benzenesulfonic-containingcleanup columns. It was found that approximately 100 mg of the Penicillium broth per gram of adsorbent could pass before breakthrough occurred. With the Fusarium broth 11.2 mg could pass before saturation. The actual penicillin content of the Penicillium broth was 2.1 mM, most of which was effectively removed in the sample pretreatment. Elemental analyses of broths were made on the supernatants obtained after centrifugation. After evaporation to dryness the Penicillium broth was found to contain 43.1% C, 4.0% H, 1%N, and 47.5% 0. These values are in accordance with a high content of sugars. The Fusarium broth contained 31.4% C, 4.7% H, 7.4% N, and 30% 0. The much higher nitrogen content in this broth suggests that this broth contains both sugar and proteins. Determination of Sugars in Fermentation Broths. The Penicillium broth was chosen as a model reference matrix. It was diluted 10 times with water in all model solutions. A separate analysis of the 10 times diluted fermentation broth was found to contain 50 pM glucose. No indications of any remaining lactose could be traced. This was considered as the background level. In order to compare the recovery (concentration found/concentration expected) of saccharides in the broth, standard addition of a series of model solutions were made; see Experimental Section. The model solutions were treated by using the cleanup procedure. The calibration curves for the various sugars were obtained from the solutions without the addition of Penicillium broth. They were equally pretreated as the broth samples. Standard addition of the model solutions to the broth prior to pretreament and analysis revealed recoveries close to loo%, where each value is the mean of six determinations;see Figure 7 . However, lower sugar levels generally tend to yield higher recoveries and higher oligosaccharide levels tend to yield somewhat lower than expected. Two equal samples were prepared and three injections were made of each RSD, within each set of six values, was typically 3-7% for all sugars. The sugar contents in the Fusarium oxysporum fermentation broth are found in Table IV and Figure 8. The sample taken out after 42 h from the fermentor was highly viscous and the chromatogram revealed an early eluting peak corre-
ANALYTICAL CHEMISTRY, VOL. 61, NO. 8, APRIL 15, 1989 3
a
3
2 lOnA
20
I
I
l 10
i 0
t/min
Flgure 8. Chromatographic recordings from the Fusarium oxyspwum broth taken at two different tlmes during the process: (1) starch fradon(s), (2) transient peak, (3) glucose; (a) peak 3, 112 ng (sample diluted 2-fold); (b) peak 3, 360 ng of glucose (sample diluted &fold); (c) sample b with omkted reactors.
sponding to a high molecular weight fraction; see Figure 8a. Chromatogramsof samples taken out later in the fermentation process reveal lower peaks of early eluting species. In all three samples glucose was found in various concentrations but no oligosaccharides could be found present at any time. In the sample taken out last, an unknown transient appeared in the chromatogram, first a negative peak followed by a positive, see Figure 8b. Similar phenomena are observed in ion chromatography known to be a disturbance in buffer protonation equilibrium created by the injection of aqueous samples (40-41). The transient observed here was not seen when injecting blanks of water. Attempts were made to verify the assumption that the transient was due to a chromatographed substance which adsorbs and desorbs at the electrode surface, thus contributing to a nonfaradaic current response, as observed in Figure 8b. When the same sample was injected into the system, where the reactors were omitted, a simii behavior was noticed; see Figure 8c. The k’ value obtained for the transient was the same as when the reactors were incorporated in the system, taking into account the dead volumes of the reactors. This verifies the assumption of a species that interferes with the electrode detection system. This species omits at present the possibility to analyze the presence of oligosaccharidesin this sample as the retention time is equal to those of some oligosaccharides. A change in the applied potential of the working electrode will change the appearance of the transient but it could not be fully eliminated. A further pretreatment of the sample could possibly remove the substance. CONCLUSIONS Starch, hydrolyzed starch, and starch syrups are used as the main carbon source in variety of fermentation processes as well as in brewery and nutrition industries. The degradation of the polysaccharide into smaller units can be followed and analyzed with the combination of LC, immobilized enzyme reactors, and electrochemicaldetection. The selectivity of the detection system was proved for samples like the soft drink and the beer. It was shown that the beer contains high molecular fractions, larger than DP5 as well as maltose and glucose. The naturally present a-
837
amylase in the barley corn splits starch into smaller fractions; the smallest product formed is maltose. By the addition of yeast in the fermentation process, maltose is then further split into glucose. Fermentation broths with complex matrices can be analyzed after a cleanup procedure using ion exchange cleanup columns. There were constitutents remaining in only one sample that interfered with the detection system but only in a totally reversible manner and a more elaborate cleanup procedure may well solve this problem. More than 300 injections into the LC system did not affect the chromatographic properties indicating a stable system. In the Fusarium fermentation broth no maltooligosaccharides could be detected but rather the fast degradation of starch directly into glucose can be followed. This information is of vital interest in the production process for the understanding of the metabolic pathway that occurs in the fermentor. It has neither been known nor proved in earlier studies (42). Chemically modified electrodes have only recently been applied as sensors in analytical devices (43). Combining the selectivity of enzymes with that of a chemically modified electrode is here demonstrated as a successful detector in LC compared to existing refractive index and UV detectors. Drawbacks of the method are the need of an extra pump as well as the limited number of sugars that can be analyzed and limitations in using organic solvents in the mobile phase (21). ACKNOWLEDGMENT The author thanks Lo Gorton for constructive discussions and valuable help with the manuscript and Istvan Csiky, Fermenta Products, for the fermentation broths and valuable discussions. LITERATURE CITED Van Damme, E. J. Blotechndogy of IndusMel Antibiotics; Marcel Dekker: New York, 1984. Hounsell, E. F.; RMeout, J. M.; Pickering, N. J.; Llm, C. K. J . Liq. Chromatogr. 1984, 7 , 661-674. Kainuma, K.; Nakakuki, T.; Ogowa, T. J . Chromatogr. 1081, 212, 126-131. Verhaar, L. A. Th.; Kuster, B. F. M.; Claessens, H. A. J . Chromatogr. 1984, 284, 1-11, Mc Glnnis, G. D.; Prince, S.; Lowrimore, J. J . Carbohydr. Chem. 1088, 5 , 63-97. Vratny, P.; Coupek, J.; Vozka, S.; Hostomska, 2. J . Chromatogr. 1083, 254, 143-155. Goso-Kato, K.; Iwase, H.; Slshihara, K.; Hotta, K. J . Chromatogr. 1988, 380, 374-378. Kuo, T. N.; van Middlesworth, J. F.; Wolf, W. J. J . Agric. Food Chem. 1088, 3 6 , 32-36. Bonn, G.; Pecina, R.; Burtscher, E.; Bobletter, 0.J . Chromarogr. 1984, 287, 245-258. Barker, S . A.; Hatt, B. W.; Kennedy, J. F.; Somers. P. J. Carbohydr. Res. 1989, 9 , 327-334. Marko-Varga, G. J . Chromatogr. 1987, 408. 157-170. Femia, R. A.; Weinberger. R. J . Chromatogr. 1087, 402, 127-134. Hughes, S.; Johnson, D. C. Anal. Chim. Acta 1083, 149, 1-10. Neuburger, 0 . G.; Johnson, D. C. Anal. Chem. 1087, 5 9 , 150-154. Reim, R. E.; van Effen, M. Anal. Chem. 1988. 5 8 , 3203-3207. Santos, L. M.; Baidwin, R. P. Anal. Chem. 1087, 59, 1766-1770. Santos, L. M.; Baldwin, R. P. Anal. Chim. Acta 1088, 206. 85-96. Vratny, P.; Brinkman, U. A. Th.; Frel, R. W. Anal. Chem. 1085, 5 7 , 224-229. Watanabe, N.; Toda, G.; Ikeda, Y. SunsekiKagaku 1084, 33. E241E246. Honda, S.; Konishi, T.; Suzuki, S. J . Chromatogr. 1984, 299. 245-251. Krull, I. S. Reaction Detection in LiquM Chromatography; Marcel Dekker: New York, 1986. Bowers, L. D. Anal. Chem. 1988, 5 8 , 513A-530A. Brinkman, U. A. Th. Chromatographia 1987, 2 4 , 190-200. Daigaard, L. Prog. HPLC, 1087, 2 , 219-233. Carr, P. W.; Bowers, L. D. Immobilized Enzymes in Analytical and Clinical Chemistry; John Wiley and Sons: New York, 1980. Van Zoonen, P.; Gooiler, C.; Velthorst, N. H.; Frei, R. W.; Wolf, J. H.; Gerrits, J.; Flentge, F. J . Pharmaceut. Sbmed. Anal. 1087, 5 , 485-492. Kiopf, L. L.; Nieman, T. A. Anal. Chem. 1085. 5 7 , 46-51. Oamsma, G.; Lammerts van Bueren. D.; Westerink, D. H. C.; Horn, A. S. Chromatographia 1087, 24 827-831. Weetall, H. H.; Hersch, L. S. Siochim. Siophys. Acta 1080, 185. 464-465. ~
a38
Anal. Chem. wag, 61,
(30) Appelqvlst, R.; Marko-Varga, G.; Gortoo, L.; Torstensson. A.; Johnsson, G. Anal. Chim. Acta 1986, 169. 237-247. (31) Appelqvist, R. Doctoral Thesis, Lund, 1987. (32) Hoffmann, E.; Marko-Varga. G.; Csiky, I.; Jonsson, J. A. Int. J . Envlron. Anal. Chem. 1986, 2 5 , 161-171. (33) Aspinall, 0.0. The PolysaccharMes; Academic Press: New York, 1985; Vol. 3. (34) W o n . L. J . Chem. Soc., Faraday Trans. 1986, 82, 1245-1258. (35) Marko-Varga, G., Lund, unpublished results, Jan 1988. (36) Leyden, D.E.; Collins, W. Sllylafed Surfaces; Gordon and Breach Publishers, Inc.: New York, 1980. (37) Ernn6us. J.; Appelqvist, R.; Marko-Varga, G.; W o n , L.; Johnson, G. Anal. Chlm. Acta 1986, 180, 3-8. Heineman, W. R. Leboratory Techniques in Ebctrmn(38) Klsslnger, P. 1.; a!Ytical chemistry; Marcel Dekker: New York, 1984.
838-843
(39) Urbaniczky, C.; Lundstrom, K. J . E l e c m n a l . Chem. 1984, 176, 169-182. (40) Stevens, T. S. Ind. Res. Dev. 1983, 25, 98. (41) Marko-Varga. G.; Cslky, I.; JBnsson. J. A. Anal. Chem. 1984, 56, 2066-2089. (42) Csiky, I., Fermenta Products AB, Striingnas, Sweden, personal corn munlcatlons. March 1988. (43) Murray, R. W.; Ewlng, A. G.; Durst, R. A. Anal. Chem. 1987, 59, 379A-390A.
RECEIVED for review September 23,1988. Accepted January 11, 1989. Financial support from the Swedish Board for Technical Development (STUF)is gratefully acknowledged.
Identification of Atmospheric Organic Sources Using the Carbon Hollow Tube-Gas Chromatography Method and Factor Analysis G. P. Cobb*J and R. S. Braman Department of Chemistry, University of South Florida, Tampa, Florida 33620
R. A. Gilbert Chemical Engineering Department, University of South Florida, Tampa, Florida 33620
Atmospherlc organlcs were sampled and analyzed by using the carbon hollow tube-gas chromatography method. Chromatograms from splce mlxtures, clgarettes, and ambient air were analyzed. Prlnclpal factor analysis of row order chromatographlc data produces factors which are eigenchromatograms of the components in the samples. Component sources are identifled from the elgenchromatograms in ail experlments and the lndlvlduai eigenchromatogram corresponding to a partkular source Is determlned In most cases. Organic sources In amblent alr and in clgarettes are Mentifled wlth 87 % certainty. Analysls of clove clgarettes allows the determinatlon of the relatlve amount of clove in different cigarettes. A new MMdestructhre quality control method using the hollow tube-gas chromatography analysls Is dlscussed.
INTRODUCTION The carbon hollow tube-gas chromatography (CHT-GC) method has proven applicable in studies of atmospheric organics (1,2). Complex systems may be sampled with the CHT and the components separated and quantitated by using GC. Dozens of peaks are routinely obtained in chromatograms of atmospheric samples (2-4). Data of this complexity presents special data interpretation problems. The number of peaks present in environmentalsamples makes visual peak matching to determine sources contributing to organic patterns difficult at best. A statistical method is needed for pattern recognition in these complex systems. Chemometrics, one of the most rapidly growing fields of chemistry, encompassesthe various methods of complex data interpretation. One versatile chemometric technique, factor analysis (FA), provides insight into the chemical nature of our environment. Psychologists developed FA during the period 'Present address: Institute of Wildlife Toxicology, Western Washington University, Bellingham, WA 98225. 0003-2700/89/0361-0838$01.50/0
between 1900 and 1905 (3). Their original goal was to determine the number of factors that contribute to an observed response (5). The technique also has the advantage of removing bias from the observations ( 3 , 5 , 6 ) . Until the advent of modern computers, the laborious matrix diagonalization and inversions required for FA restricted useful applications. With the current accessibility of computers,FA can be applied to large data sets within seconds. Chemists have used FA to study various physical properties of systems (6-81, model reaction intermediates (9, IO), correlate toxicity of chemicals based on structure ( I I ) , and identify molecules by using chromatographic or spectroscopicdata (4, 6,12-14). Biennial reviews of chemometric studies have appeared since 1980 (15-18). This paper describes uses of principal factor analysis (PFA) for identifying the contribution of specific, complex organic sources to an air sample. The computations involved are centered on five steps: [ l ] matrix arrangement of data, [2] data preprocessing, [3] data matrix deconvolution into weighting and factor matricies, [4]error removal, [5] target testing. The reader is referred elsewhere (6) for basic details of PFA. One primary attribute of PFA is extraction of factors (eigenvectors) in order of importance to the data set. The first factor calculated for a data set will describe more of the data than will any other single factor. If these factors were assigned to specific organic sources, individual components of the sample could be identified. Unfortunately, eigenvectors do not directly correspond to unique molecules represented by the data. Despite this fact, PFA of CHT-GC samples can lead to identification of characteristic patterns from volatile organic sources. In order to give eigenvalues physical significance, a different view of factor determination is useful. The n X m data matrix (D) to be analyzed is constructed of n CHT sample chromatograms with peak areas (retention areas) at m retention times. In all cases discussed, the number of retention areas is greater than the number of experiments. Analysis of D constructed @ 1989 American Chemical Soclety
