Some statistical aspects of pyrolysis-GLC in the identification of alkaloids

Abstract

A study of twenty-one alkaloids shows that they can be differentiated by means of the lower hydrocarbon pyrolysis products and trimethylamine, except for morphine and heroin, which cannot be distinguished with these products. Heroin can be differentiated from morphine as well as the other alkaloids studied by the pyrolytic formation of acetic acid, which is separated on a diisodecyl phthalate column. A squalane column was used for separation of the former products. Multivariate statistical methods are briefly discussed and utilized for evaluating the data.

Details

Author: Charles R. KINGSTON, Paul L. KIRK
Pages: 19 to 26
Creation Date: 1965/01/01

Some statistical aspects of pyrolysis-GLC in the identification of alkaloids

Charles R. KINGSTON
Paul L. KIRK
School of Criminology, University of California, Berkeley, California

A study of twenty-one alkaloids shows that they can be differentiated by means of the lower hydrocarbon pyrolysis products and trimethylamine, except for morphine and heroin, which cannot be distinguished with these products. Heroin can be differentiated from morphine as well as the other alkaloids studied by the pyrolytic formation of acetic acid, which is separated on a diisodecyl phthalate column. A squalane column was used for separation of the former products. Multivariate statistical methods are briefly discussed and utilized for evaluating the data.

Pyrolysis-GLC, which is the process of the pyrolysis of compounds in the carrier gas stream of a gas chromatograph so that the thermal decomposition products directly enter the column, has been proven of value in the identification of different types of drugs and other related compounds. Janik [ 5] [ 6] demonstrated the application of pyrolysis-GLC to barbiturates and alkaloids along with other types of compounds. Nelson and Kirk [ 8] [ 9] studied the identification of the barbiturates by pyrolysis-GLC in some detail, and reported the presence of various nitriles which were a characteristic aid to the identification of the 5,5-disubstituted barbituric acids. Fontan, Jain and Kirk [ 4] reported a study of the identification of some of the phenothiazines by pyrolysis-GLC. In the case of the phenothiazines, there were apparently no characteristic major products as was the case with the barbiturates. The authors report that the observed products did, however, serve to differentiate among the phenothiazines studied.

The application of pyrolysis-GLC is extended to the identification of the alkaloids in this paper, with the emphasis being placed on the relative amounts of the lower hydrocarbons and trimethylamine. The identification was viewed as a problem in classification, and multivariate statistical methods were used to evaluate some of the results.

Experimental

Apparatus and Conditions. An Aerograph A-90-P2 gas chromatograph was used. This instrument normally contains a thermal conductivity detector system, but was modified by the addition of a flame ionization detector placed in the detector oven. A stream splitting arrangement was incorporated, which allowed a selected proportion of the carrier stream to be routed through the flame detector, while letting the main flow continue through the catharometer. The necessary equipment for the modification was obtained from Wilkins Instrument and Research, Inc., Walnut Creek, California. A Leeds and Northrup Speedomax H, O- to 1-mV recorder was fitted with a switch to connect it to either the thermal conductivity or the flame ionization detector. The Wilkins pyrolysis unit described by Jennings and Dimick [ 7] was used, and a Wilkins Model 650 hydrogen generator was used for the flame unit hydrogen supply.

Four columns were used for examining the pyrolysis products:

  1. Squalane, 30% by weight on firebrick, 100/120 mesh in a 27'x1/4" column of aluminium tubing. This was the column used for all of the peak measurements made to obtain the classification data. Operating conditions for this column were as follows: column -50°C; injector-100°C; detector oven-80°C; the pyrolysis unit was heated by conduction from the injector block; TC flow rate (helium)-60 cc/min.; FI flow rate-20 cc/min. (helium) + 20 cc/min. (hydrogen); inlet helium pressure - 50 psi.

  2. Carbowax 600, 10% by weight on Celite, 100/120 mesh in a 5'x1/4" column of aluminium tubing. Operating conditions: column·- 75°C; injector-125°C; detector oven-100°C; TC flow rate (helium)-75 cc/min.; FI flow rate -25 cc/min. (helium)+5- +15 cc/min. (hydrogen); inlet helium pressure -40 psi.

  3. Hallcomide, 10% by weight on Chromosorb G, 70/80 mesh in a 5'x6 mm glass column. Operating conditions same as column 2.

  4. Diisodecyl phthalate, 5% by weight on Chromosorb G, 70/80 mesh in a ,5'x6 mm glass column. Operating conditions: column-80°C; injector -130°C; detector oven-110°C; TC flow rate (helium)-75 cc/min.; FI flow rate -25 cc/min. (helium) + 20 cc/min. (hydrogen); inlet helium pressure -40 psi.

The FI electrometer (Aerograph Model 500-C) was operated at a range setting of 10 and an attenuation of IX. The pyrolyses were made with the platinum coil in the pyrolysis unit at a temperature of 1000° to 1200°C for 15 seconds with column 1 and for 10 seconds with the other three columns. Melting points were determined with a Köfler hotstage.

Reagents. The following ten alkaloids, which are all opium alkaloids or alkaloids structurally related to opium alkaloids, were recrystallized until the melting point behaviour indicated that the crystals were pure: morphine, codeine, heroin, hydromorphone, hydrastine, hydrocodone, oxycodone, thebaine, narcotine and papaverine. A second set of eleven alkaloids were recrystallized once. These were aconitine, theobromine, caffeine, phenacetin, quinine, strychnine, cinchonine, atropine, cocaine, brucine, and colchicine. These eleven alkaloids were pyrolyzed to determine what range of quantitative results might be expected from a variety of alkaloid compounds. Other available alkaloids could not be crystallized and were not used.

The first ten alkaloids were dissolved in 2% H 3aPO 4 to make up a 4% (4±0.2%) solution of each just prior to the first series of runs on that alkaloid. The other eleven were dissolved in 2 % H 3PO 4 to make up 2% to 10% solutions approximately. The quantity of the products that were detected was considerably less if the alkaloids were pyrolyzed as hydrochlorides or free bases. For the basic work done, 6µl samples of the 4% solutions were applied to the platinum coil; thus about 240 µgm of each of the first ten alkaloids were pyrolyzed.

Procedures. The initial measurements included the peak height and the base width of the best visually fitted triangle for the recorded peak. From these measurements the areas under the peaks were estimated, and these areas used to obtain the relative ratios used. Later it was decided to use the peak heights alone. These heights were measured to the nearest half millimeter, and the results in this paper are in terms of, or calculated from, them.

Identification of the products were made by means of retention time comparisons on different columns of standards and products. Standard hydrocarbons were obtained from the Matheson Company, Newark, California, and their comparative retention times determined on the squalane column and an adiponitrile column on a different instrument, which was not used for any classification runs. The identification of trimethylamine was made by comparing retention times of the pyrolysis product and trimethylamine produced by pyrolysis of tetramethylammonium hydroxide (as a free base and in H 3PO 4 solution) on the four columns listed above. The characteristic tailing on the squalane column helped identify the trimethylamine peak from the other nearby peaks when it was present in small quantities. It is of interest to note that pyrolysis of tetramethylammonium hydroxide produced C 2 to C 4 hydrocarbons, as well as (CH 3) 3N and CH 3OH.

Calculations of the mean and dispersion matrices in the original and discriminant spaces were made using the computer programmes DISCRIM and RSPACE as described by Cooley and Lohnes (1), and the classification chi-square values were determined using the computer programme CLASSIF described by the same authors (2). All computer runs were made using the IBM 7090 computer at the University of California Computer Center in Berkeley.

Results and discussion

Initial work on the identification of the alkaloids with pyrolysis-GLC indicated that there were essentially no major characteristic products that would serve to differentiate closely related compounds, such as morphine, heroin, codeine, thebaine, and oxycodone, whereas all alkaloids tested did show relatively large amounts of the lower hydrocarbons. Therefore, it was decided to examine the possibility of using the relative amounts of these hydrocarbons to differentiate the alkaloids. Also, it appears that a large majority of compounds give off quantities of the lower hydrocarbons when pyrolyzed, and it is possible that identifications of compounds other than the alkaloids could be effected by means of these same products. Thus a single set of instrument operating conditions might suffice for general identifications using pyrolysis-GLC.

Initial Classification Data. In order to establish classification groups for each of the first ten alkaloids listed under reagents, and to determine what variability might obtain in different runs of the same alkaloid, a series of runs for each alkaloid was made using freshly prepared solutions in H 3PO 4 as previously described. Ten runs were made for morphine, codeine, and hydromorphone, and five for the other seven alkaloids, using the squalane column and 6 µl of the solution evaporated on the pyrolysis coil. A typical chromatogram is illustrated in figure 1. The mean ratios of the last five peak heights with respect to the height of the second peak (ethene) and the standard deviations (to 3 decimal places) of these ratios are listed in table la, as well as the levels (to be defined later). The identifications and retention times of the products are listed in table 1 b. The FI output was used for all data.

FIGURE 1 Minutes Chromatogram of the pyrolysis products of thebaine with column 1 (squalane).

Full size image: 15 kB, FIGURE 1 MinutesChromatogram of the pyrolysis products of thebaine with column 1 (squalane).

After obtaining the above data, a second series of pyrolysis-GLC runs on each of the ten alkaloids was made using the previously prepared solutions. At first, the ratios of the peak areas with respect to the first peak (methane) were determined for both series of runs. An examination of the data indicated that the results clearly showed too much variation for any classification procedure. Then it was noticed that the variability was significantly less if the second peak were used as the reference rather than the first. Consequently, all ratios were recalculated with respect to the second peak, and only peak heights were used for convenience. The ratio related to the first peak now showed considerable variability, and was considered unusable without modification. However, it could be used to divide the ten alkaloids into discrete "levels" on the basis of gross differences in the ratios. The three levels are as follows:

TABLE 1 a

Classification levels, and mean values and standard deviations of peak height ratios with respect to peak 2 (Ethene)

 

1

2

3

4

5

Alkaloid
Level
Mean
S.D.
Mean
S.D.
Mean
S.D.
Mean
S.D.
Mean
S.D.
Morphine
1
.163
.020
.157
.003
.025
.004
.000
.000
.016
.001
Codeine
1
.222
.014
.200
.003
.037
.003
.013
.003
.032
.002
Heroin
1
.157
.011
.142
.007
.017
.002
.000
.000
.013
.002
Ocycodone
1
.166
.011
.195
.006
.030
.003
.011
.001
.022
.002
Thebaine
1
.284
.034
.211
.018
.058
.006
.031
.004
.031
.004
Hydromorphone
0
.045
.003
.091
.006
.006
.000
.000
.000
.009
.001
Hydrocodone
0
.039
.004
.070
.007
.004
.000
.003
.000
.009
.001
Hydrastine
2
.207
.024
.146
.016
.003
.000
.077
.006
.000
.000
Narcotine
2
.747
.105
.114
.034
.008
.018
1.433
.123
.000
.000
Papaverine
2
.593
.051
.131
.025
.000
.000
10.477 2.216
.000
.000

TABLE 1 b

Peak identification and retention times

   
1
2
3
4
5
6
7
Identification
Peak
Methane
Ethene
Ethane
Propene
Propane
Trimethyl-amine
C 4 hydro-carbon
Retention time *
(min.)
4.0 5.0 5.7 9.0 9.8 17.3 19.2

* "Air" retention time - 3.8 min.

Level 0 ... The ratio of the heights of peak 1 to peak 2 is less than 1:3; i.e., the first peak is small compared to the second.

Level 1 ... The ratio is between 1:3 and 3:1.

Level 2 ... The ratio is greater than 3:1.

The dividing points were arbitrarily chosen by a consideration of the actual ratios.

Using the level values and the ratios for the last five peaks, a simple classification test was run using the 7090 computer. This test was a modification of the nearest neighbour procedure discussed by Fix and Hodges [ 3] , and used the local majority rule mentioned by Sebestyen [ 11] . The results from this classification procedure were very poor.

Sources of Error. One possible cause of error in the classification that was considered was that the alkaloids had undergone some change or decomposition in the H 3PO 4 solutions. To check this, a few runs were made using the old solution of morphine along with a few runs using a freshly prepared solution. The combined results had about the same variability of ratios as did the original runs, but exhibited a shift of the mean value point. Different sets of runs on different days behaved similarly. The same type of experiment was made with codeine with similar results. When several runs of morphine and codeine were made in sequence on the same day, an interesting correlation of the mean value shift was observed. The same linear transformation that would bring the mean value of the morphine group back to the originally observed value would also bring the mean value of the codeine group back to a position near the original mean value for the first series of runs. This was a repeatable phenomenon for different sets of combined runs, and indicated that the observed differences were not due to changes in the alkaloids, but rather to minor variations in the operating conditions of the gas chromatograph.

Other considerations relative to the applicability of the local majority test suggested that the distance function used would have to depend on the direction in which the distance was measured and upon the group to which a standard point belonged. This is due to the different variances shown by the different ratios, which give each group an ellipsoidal type appearance with greatly different major diameters and the ratio-dependent variability of the distances between the groups. The desirable properties of the nearest neighbour type of test using ordinary Euclidean distance seem to diminish under such conditions.

Modified Classification Procedures. In order to avoid the above-mentioned sources of error, a new set of test runs for the level 1 alkaloids were made. In this set, a standard morphine pyrolysis was made immediately prior to each simulated unknown. Twenty-four total runs were made, twelve of which were the morphine standards, with three each of the other four alkaloids. The morphine vectors (points) were translated to coincide with the mean value vector of the original group, and the same translation applied to the following run. The resulting values are listed in table 2.

The original (standard) group vectors were projected onto a 3-dimensional discriminant space by means of computer programmes DISCRIM and RSPACE, and programme CLASSIF was used to classify the twelve runs, using the translated values and the chi-square scores. This resulted in five errors in the classifications.

TABLE 2

Translated test run results

 
ID
1
2
3
4
5
Codeine
11
.258
.227
.044
.016
.036
Heroin
12
.222
.158
.021
.000
.014
Thebaine
13
.421
.271
.084
.033
.045
Oxycodone
14
.155
.093
.029
.018
.018
Thebaine
21
.264
.218
.049
.032
.035
Codeine
22
.230
.213
.036
.020
.031
Heroin
23
.164
.148
.022
.000
.020
Oxycodone
24
.166
.178
.026
.014
.018
Codeine
31
.241
.215
.049
.019
.037
Oxycodone
32
.169
.190
.034
.011
.023
Heroin
33
.176
.174
.036
.000
.018
Thebaine
34
.279
.223
.057
.044
.032

FIGURE 2 Ratio I Scatter diagram of ratios 1 and 4 for thebaine, codeine and oxycodone. Test run is ID No. 14 (table 2).

Full size image: 11 kB, FIGURE 2 Ratio IScatter diagram of ratios 1 and 4 for thebaine, codeine and oxycodone. Test run is ID No. 14 (table 2).

On examining the data closer, it appeared that the normality assumptions made in the programmes are not valid - that is, the points in the individual groups are not distributed according to a normal (Gaussian) distribution.

On re-examining the data and discriminant vectors, it seemed that ratios 1 and 4 (for peaks 3 and 6) might be suited alone (but not necessarily optimally) for the classification. Ratio 4 will serve to separate out morphine and heroin from the other three level 1 alkaloids, since no peak is seen with the former, whereas a definite peak is seen with the latter. Utilizing this information, CLASSIF was used in the 2-dimensional space generated by ratios 1 and 4 and resulted in only one misclassification, but with no differentiation between morphine and heroin. Figure 2 illustrates the distribution of the group points for thebaine, oxycodone and codeine, with the position of the misclassified run indicated (this run was oxycodone but was classified as thebaine, a result which is reasonable for the data under the normal assumptions made in the programmes).

In table 1, it can be noted that the standard deviations are approximately proportional to the mean values. This indicates that a log transformation would tend to equalize the variances, and thus might tend also to make the data conform better to the assumption of normality. Figure 3 shows the data from figure 2 plotted on 1 x 1 cycle log paper, and it can be observed that the variances have been equalized to a considerable extent. The indicated test run that was misclassified with respect to the data in figure 2 is correctly classified (using ratios 1 and 4) when the log values are used.

It is of interest to note that the simple (nonparametric) nearest neighbour rule [ 3] makes one misclassification of a codeine test run for the data in figure 2, and one misclassification (the indicated test run) for the data in figure 3 using the log values. The ratio-4 values tended to be consistently high for all test runs as compared to the original group data. This could not be corrected for with the morphine correction runs since the ratio there was zero. If the ratio-4 values could be corrected with respect to instrument variations, the test runs would probably all be correctly classified by CLASSIF (with ratios 1 and 4) and the nearest neighbour rule for both the ratio and the log values. This suggests that codeine would be a better standard than morphine for the correction or standardizing runs.

FIGURE 3

Ratio I

Scatter diagram of ratios 1 and 4 for thebaine, codeine and oxycodone plotted on 1 x 1 cycle log paper.

Full size image: 19 kB, FIGURE 3

TABLE 3

Classification levels and peak height ratios with respect to peak 2

 

Ratio

Alkaloid
Level
1
2
3
4
5
Aconitine
0
.028
.044
.003
.010
.006
Phenacetin
0
Ethene was the only product seen after methane
" " "
"
Theobromine
2          
Caffeine
2 111 167
-
-
-
Atropine
1
.144
.253
.011
.000
.043
Brucine
1
.192
.178
.030
.022
.030
Cinchonine
1
.583
.500
.092
.000
.150
Cocaine
1
.183
.366
.037
.079
.061
Colchicine
1
.050
.100
.000
1.100
.000
Quinine
1
.982
.440
.133
.000
.114
Strychnine
1
.433
.161
.078
.000
.022

Distinction between morphine and heroin was not possible using the squalane column. Minor differences were observed on the hallcomide column, but were not considered sufficient for distinguishing between the two alkaloids. The possibility that the acetyl groups on heroin were split off at the initial stage of the pyrolysis, leaving essentially morphine to be more completely degraded, was considered. In this case it might be expected that one of the products for heroin would be acetic acid, which would not be produced from morphine. The diisodecyl phthalate column, which will effectively chromatograph acetic acid, indicated that this was indeed the case, and presented a clear distinction in the pyrolysis products of morphine and heroin. Aconitine, which also has an acetyl group, showed a possible acetic acid peak on column 4.

The alkaloids in levels 0 and 2 are readily distinguished with the additional information given by ratios 1 and 4. A single pyrolysis run for each of the additional eleven alkaloids was made on the squalane column. The peak ratios are listed in table 3. Assuming a similar trend of the variability of the ratios as seen with the first ten alkaloids, it appears that reasonably accurate identifications can be made on all twenty-one, with the exception of morphine and heroin, by ratios involving peaks 1, 2, 3 and 6. The alkaloids can be separated into six discrete and easily distinguishable groups with peaks 1, 2 and 6, and the members of each group can be classified according to ratios 1 and 4. The six groups are listed in table 4.

TABLE 4

Six discrete alkaloid groups classified by level and trimethylamine

 
(CH3) 3N present
(CH3) 3N absent
Level 0
Hydrocodone
Hydromorphone
 
Aconotine
Phenacetin
Level 1
Codeine
Morphine
 
Oxycodone
Heroin
 
Thebaine
Quinine
 
Cocaine
Strychnine
 
Brucine
Cinchonine
 
Colchicine
Atropine
Level 2
Hydrastine
Caffeine
 
Papaverine
Theobromine
 
Narcotine
 

Some General Results. Chromatograms of the pyrolysis products of all twenty-one alkaloids were also made on columns 2, 3 and 4 using the FI detector. Some of them have characteristic patterns, with columns, 2 and 3 showing the most differentiation. The relative ratios of peaks which probably represent acetone and methanol would be a possible distinguishing feature for the morphine type structured group (except morphine and heroin). Toluene and acetonitrile, as well as trimethylamine, appear to be other products that would have utility in differentiating among some of the other alkaloids on columns 2 and 3. Column 4 showed only a few characteristic patterns, but was quite distinctive for heroin, which was the only alkaloid studied that showed a relatively large acetic acid peak with respect to the other observed peaks.

General Discussion. The results that have been obtained by several workers using pyrolysis-GLC for identification indicates that it may have several advantages over some other techniques which are widely used at the present time, such as ultra-violet spectrophotometry. In pyrolysis-GLC, the same instrument that is used in the pyrolysis process can also be used for the initial separation or purification of the substance to be identified if necessary; the additional parameter of the retention time of the substance is also obtained during this initial step. Pyrolysis-GLC is useful with a wider variety of substances than UV-spectrophotometry - the identification of the barbituric acid derivatives offers one example of this - and gives a greater number of parameters for identification. One criticism of pyrolysis-GLC is that the results are likely to be too variable for practical use. Janik [ 5] demonstrated that the results were consistent, and the statistical study in this paper indicates that the variability (within the scope of this study) is within practical limits. Another question that is raised relates to the effects of small impurities on the results. Commercially available U.S.P. grade morphine was not distinguishable from the purified morphine, and thebaine was correctly classified even after the solution showed considerable darkening from decomposition.

Other general results indicate that the necessity of a pure substance is apparently no more critical for pyrolysis-GLC than it is for UV-or IR-spectrophotometry, or X-ray diffraction. The lower hydrocarbons considered here probably consist of original products from the pyrolitic splitting plus quantities from the random recombinations of free radical products. It is therefore to be expected that the amounts finally seen would be statistically stable and relatively unaffected by small amounts of impurities. Since the statistical study was directed toward the group of alkaloids that showed the closest clustering of points, error possibilities from impurities in classifying alkaloids within other groups should be considerably less.

The use of multivariate methods of analysis with the type of experimental data presented above offers a more sensitive and informative evaluation than individual comparisons of variables or an overall visual comparison of results. It can also point out what aspects or combinations of the data contain the desired information, and thus often reduce the amount of "filing" information needed. One disadvantage of multivariate methods is that the computations quickly become too numerous to handle without using a computer. However, as can be seen in this study, a computer may only be necessary for the initial evaluation. If the classification boundaries are computed and drawn on the graphs for ratios 1 and 4 for each of the six discrete groups, a classification could be made quite easily without any calculating equipment. On the other hand, the use of computers is becoming an accepted fact, and it should not be very long before even the smaller laboratories have access to a computer capable of handling such calculations.

The utilization of efficient computer oriented data processing methods combined with modern statistical methodology offers the modern analyst a means of developing quicker and more sensitive analytical procedures, especially for initial screening type problems and general unknowns that typically confront the toxicologist, and for problems of origin determinations of opium, marijuana, and other materials that are a complex mixture of substances. Savitzky [ 10] has made several interesting comments on the general applications of related techniques. A more detailed discussion of the statistical analysis and procedures used in this paper will be reported separately.

Acknowledgements

This work was supported by grants from the National Institutes of Health, U.S. Public Health Service (AC 00185), and from the Committee on Research, University of California.

We wish to acknowledge with thanks the donation of thebaine by S. B. Penick and Co., and dihydrohydroxycodeinone by Endo Laboratories, Inc.

References

001

COOLEY, W. W. and LOHNES, P. R. Multivariate Procedures for the Behavioral Sciences. Chap. 6, John Wiley and Sons, Inc., New York, 1962.

002

COOLEY, W. W. and LOHNES, P. R. Multivariate Procedures for the Behavioral Sciences. Chap. 7, John Wiley and Sons, Inc., New York, 1962.

003

FIX, E. and HODGES, J. L., Jr. Discriminatory Analysis - Nonparametric Discriminations: Small Sample Performance. USAF School of Aviation Medicine, Project Number 21-49-004, Report No. 11, August 1952.

004

FONTAN, C. R., JAIN, N. C. and KIRK, P. L. Mikro-chimica Acta 2-4, 326; 1964.

005

JANAK, J. Collection of Czech. Chem. Commun. 25, 1780; 1960.

006

JANAK, J. Nature, 185, 684; 1960.

007

JENNINGS, E. C., Jr. and DIMICK, K. P. Anal. Chem., 34, 1543; 1962.

008

NELSON, D. F. and KIRK, P. L. Anal. Chem., 34, 899; 1962.

009

NELSON, D. F. and KIRK, P. L. Anal. Chem., 36, 875; 1964.

010

SAVITZKY, A. Anal. Chem., 33, 25A; 1961.

011

SEBESTYEN, G. S. Decision-Making Processes in Pattern Recognition, p. 97, The Macmillan Co., New York, 1962.