Determination of Asphalt Molecular Weight Distributions by Gel Permeation Chromatography L. R. Snyder Union Oil Co. of California, Research Center, Brea, Calif. 92621 The direct determination of asphalt molecular weight distributions by gel permeation chromatography (without obtaining independent molecular weights on separated fractions) presents a number of problems which appear to have been overlooked by previous workers. These include unexpected solution aggregation effects, adsorption of asphalt on the polystyrene gel, and a peculiar dependence of RI or UV response data on fraction molecular weight. Attention to these various effects allows the rapid, accurate determination of an asphalt molecular weight distribution at minimum cost, using high efficiency gel permeation chromatography in an automated unit.
SEVERAL WORKERS (1-5) have reported the use of gel permeation chromatography (GPC) in the determination of molecular weight distribution for residual petroleum samples-e.g., asphalts. Such data have been used in attempts to define the molecular structure of these materials and in correlations of their performance or physical properties. A number of petroleum laboratories are carrying out these analytical GPC separations on a more or less routine basis, using one of two possible procedures. A preparative separation can be carried out in a large column (usually of low efficiency), the molecular weights of individual fractions can be determined by an independent method, and the resulting data then yield a molecular weight distribution for the original sample (2, 3, 5). This method is reliable and straightforward, but time consuming and tedious. Alternatively, similar GPC separations can be carried out on much smaller samples in high efficiency, automated units (4). The resulting elution curve (usually a trace of eluate refractive index us. volume) can be translated into a molecular weight distribution curve, assuming (a) a constant response factor for the detector (differential refractometer), and (b) a standard sample calibration curve of elution volume us. molecular weight. The second procedure is rapid and applicable to automatic data handling, but its reliability is limited by assumptions (a) and (b). However, similar procedures are widely used in the analysis of synthetic polymers, with generally adequate accuracy. In the course of setting up a GPC procedure of the second type (small scale, automated operation) in our own laboratory, it became apparent that a number of previously ignored complications exist in the GPC analysis of asphalts. The present paper describes these complications and the measures that are necessary to insure accurate molecular weight distributions by GPC. Similar care may be required in the quantitative GPC analysis of other heterogeneous polymeric materials, such as reaction tars, certain natural products, shale oils or coal tars. ~
(1) J. P. Dickie and T. F. Yen, presented before the Div. of Petro-
leum Chem., ACS Meeting, Miami Beach, Fla., April, 1967; Petrol. Diu. Preprints, 12, NO. 2 B-117, 1967. (2) K. H. Altgelt, J. Appl. Polymer Sci., 9, 3389 (1965). (3) K. H. Altgelt, Mukromol. Chem., 88, 75 (1965). (4) W. B. Richman, Proc. Ass. Asphalt Paving Technol., 36, 106
(1967). ( 5 ) C . A. Stout and S. W. Nicksic, SOC.Petroleum Eng. J., 8, 253 (1968).
EXPERIMENTAL
GPC Separations. All separations were carried out in using four 4-ft X */& the unit described previously columns (0.30-in. i.d.) connected together in series. The columns (i-iv) were packed with Styragel (Water Associates, Framingham, M!ss.) of varying pore size: (i) a 50/50 mixture of 100 and 250 A gel, (ii) and (iu) lo3 gel, and (iv) lo4 8, gel. The overall efficiency of the four column unit was equal to 9000 theoretical plates at the standard solvent flow rate of 2.0 ml/min (0-xylene solute, 5 v methanol-chloroform solvent). In the procedure eventually used for the routine determination of asphalt molecular weight distributions, 9 mg of sample was charged as a 20-gram/liter solution in the elution solvent, elution was carried out with 5 vol methanol-chloroform (reagent grade), and 10 ml fractions were collected. The absorbance of these fractions at 370 nm was determined spectrophotometrically. This manual procedure has since been replaced by use of a flow-through spectrophotometer set at 370 nm. Molecular weights were determined by vapor phase osmometry, using chloroform solvent.
(a,
A
z
z
METHOD DEVELOPMENT
The following discussion is divided between the problems of setting up a satisfactory experimental system and the subsequent interpretation of experimental data from that system. Experimental Conditions. Previous workers who have attempted the automated GPC analysis of asphalts and related fractions ( I , 4) have generally used tetrahydrofuran (THF) as solvent and a refractometer as detector. In the present study it was found that T H F as solvent gives incomplete recovery of asphalt samples: the total UV absorbance (370 nm) of collected GPC fractions (0-1 column volumes) averaged 20-40 low, relative to the value calculated from the original sample. The strongly retained asphalt components were slowly eluted from the column in several column volumes of eluate. Several other solvents were investigated, including chloroform and various mixtures of chloroform, THF, and/or various alcohols. Of these solvents, only 5 vol methanol-chloroform gave sample recoveries which averaged higher than 95 (UV absorbance), in agreement with (2, 3). However, with the latter solvent it proved impossible to obtain a satisfactory output from the refractometer detector. Sample chromatograms were ragged and irreproducible, although baseline stability in the absence of sample injection was satisfactory. We accordingly changed our detection system to ultraviolet monitoring at 370 nm. Some interesting sample size effects were noted in the GPC analysis of asphalt samples. Earlier it had been reported (4) that the performance qualities of road asphalts appear to be related to the relative concentrations of high molecular weight components present. This is illustrated in Figure l a (THF solvent, refractometer output) for two asphalt samples from our laboratory ( A , “bad,” B, “good”). At a sample concen-
z
z
z
(6) L. R. Snyder, ANAL.CHEM.,39,698 (1967). VOL. 41, NO. 10,AUGUST 1969
1223
\
en
q
-500 1000 2000 6000 l0,OOO MW (uncorrected)
Figure 1. Sample size effects in GPC separation of asphalts
(a) THF solvent, (b) THF solvent, sample B, (c) 5 vol. methanol-chloroform solvent, sample C.
I .
I
I20
I
I40
I
I
I60
ELUATE VOLUME ( m l )
Figure 2. Detector response (1 gram/ liter) for narrow GPC fractions as a function of average elution volume
Refractive index (---) and UV (-) samples Cand D. tration of 30 mg, asphalt B shows a well defined hump in the 4000-6000 molecular weight region, and this hump is much less pronounced in the case of asphalt A . These results parallel those of (4). Upon decreasing the sample size to 4 mg, however, the high molecular weight hump in asphalt B is much depressed (Figure lb), and it is then difficult to see significant differences between “good” and “bad” asphalts. Similarly, by increasing sample size to 150 mg (Figure lb), a large hump at very high molecular weights (lO,OOO+) appears in sample B. The latter sample size effects suggest increasing association of sample molecules at higher concentrations, and this is not unexpected (7). Thus the difference between “good” and “bad” road asphalts appears to be more a function of intermolecular associations than of relative molecular weight distribution. This does not detract from the possible use of GPC in the study of asphalt quality, but it does emphasize the importance of sample size in carrying out such studies. At very small sample sizes in 5 vol % methanol-chloroform, another interesting effect was observed (Figure IC). Whereas a sample size of 10 mg or larger gave a simple monomodal distribution curve for all samples, at sample sizes less than 1 mg many asphalts show a definite bimodal distribution, with a new band appearing of very high apparent molecular weight (6000-10,000). This effect is not expected on the basis of simple intermolecular association. The relative size of the high molecular weight band is somewhat exaggerated in Figure IC, because the UV absorbance of this band is about five times greater than that of the total sample. In no case did the estimated concentration of the high molecular weight band exceed 10 wt of the original asphalt. This new band at small sample concentrations does not appear to be a chromatographic artifact or the result of adsorption on the gel. Repeated separation of the sample of Figure I C at low sample concentrations (0.8 mg) gave two fractions, corresponding to each band. Reseparation of each fraction (0.8-mg charge) gave an elution band that coincided with the band collected in the original separation (except for minor tailing). Thus the high molecular weight band does not reequilibrate to form lower molecular weight material in the course of GPC separa(7) R.S.Winniford, J. Inst. Petrol, 49, 215 (1963). 1224
ANALYTICAL CHEMISTRY
for
tion. The relative concentrations of this high molecular fraction in different asphalts correlated roughly with their asphaltene content. One possible explanation for the effect of Figure ICis that at very low sample concentrations in 5 wt % methanol-chloroform there is a dissociation of certain asphalt micelles, with subsequent recombination of the fragments into irreversibly bound high molecular weight aggregates. In any event, the various results summarized in Figure 1 emphasize the importance of controlling sample size within narrow limits, if the resulting molecular weight distribution data are to be meaningful. The use of a 9-mg sample in the routine separation avoids the production of a high molecular weight band as in Figure IC, and avoids normal association effects which occur at still higher sample concentrations. Data Interpretation. The reduction of a raw chromatogram (as generated by the routine procedure of the Experimental section) into a final molecular weight distribution requires three operations: conversion of the detector output into mass units, correction of the resulting mass os. volume chromatogram for the band spreading of individual asphalt components during separation, and a correlation between sample elution volume and molecular weight. We will discuss each of these items in sequence. In the GPC analysis of typical polymer samples, it can often be assumed that detector response-Le., refractometer-for different sample components is constant on a mass concentration basis. This is not the case for typical asphalt samples, as shown in Figure 2. Here relative weight response for both refractometer (dashed curve) and UV detection (solid curves) is plotted us. the elution volume of narrow asphalt fractions prepared by preparative GPC. Data on the UV response of two samples (C, a deasphaltened asphalt; and D, a total asphalt) are shown. Both detectors give a high response for low and high molecular weight components, with minimum response for molecular weights of 600-1000 (140-150 ml). Furthermore, the UV response curve appears to vary from one asphalt sample to another. The method of the Experimental section is based on the UV curve for sample D of Figure 2, but this is seen to be approximate at best. The preferred solution to this problem is the use of a detector which responds directly to mass, rather than a physical property
c‘t
IO0
IO00
80 -
I
I
60 -
40 -
20 I
I I
I
1000 MOLECULAR WEIGHT
400 600 ELUATE VOLUME (ml.)
Figure 3. Calibration curve for molecular length L , or molecular weight (mw) us. sample elution volume 0 standard polymer fractions and n-alkanes (15,)
asphalt GPC fractions (mw) V deasphaltened asphalt GPC fractions (mw) such as refractive index or spectral absorptivity. Commercial liquid chromatography detectors based on solvent stripping and detection by flame ionization come close to meeting this requirement. An alternative is simple fraction collection, with a gravimetric finish on the solvent-free fractions (using a micro balance). In the case of GPC polymer analyses, band broadening of individual sample components during separation can be handled in a variety of ways (8). In the case of asphalt separations, it can be estimated that normal band broadening as a result of eddy diffusion and mass transfer effects is of minor importance. Thus the data of Hendrickson (9) show an inverse correlation between plate height and sample elution volume; this is reasonable in terms of the slower mass transfer of larger sample molecules. Application of this correlation in the present study suggested that band broadening was of minor significance, because of the relatively low molecular weights of major asphalt components (500-2000) and the high column efficiency (9000 theoretical plates). However, it was observed that higher molecular weight asphalt components exhibit slowly reversible adsorption with band tailing, even in the presence of the highly polar solvent 5 vol methanolchloroform. This effect was shown by separating a typical asphalt into narrow fractions by GPC, then rerunning each fraction under the conditions of our routine analysis. In each case some tailing of an original fraction into later eluted fractions was evident. Thus 1 8 z of the band initially eluted between 110-120 ml was found in the 130-140 ml fraction of the second separation. Even larger overlap into the 120-130r d fraction occurred, but this could not be distinguished from the normal spreading of the 110-120 ml band in the second separation. Similarly, 18 of the 120-130-ml fraction from the first separation was found in the 140-150-ml fraction of the second separation. This band tailing dropped to 6z for the 130-140-ml fraction, and 4z for later fractions. The correction of the chromatogram for band tailing is important if use is made of UV or refractive index detection, because the initially eluting (strongly tailing) fractiocs have higher detector response factors. Correction for tailing becomes less important with gravimetric detection of the column effluent. One approach to the correlation of sample elution volume
z
(8) K. H. Altgelt, Adu. Chromatography, 7, 3 (1968).
4000
Figure 4. Experimental asphalt molecular weight distribution - routine GPC method (iFnequal 768) routine GPC method ( M nequal 837) 0 preparative scale GPC with molecular weight determinations on each fraction
___
in GPC with molecular weight is to measure a calibration curve for available standards, in terms of effective molecular length L, us. elution volume V. Such a curve is shown in Figure 3 (closed circles, solid line), based on commercial polystyrene and polypropyleneglycol standards (Waters Associates) and the n-alkanes (9). We can assume a proportional relationship between t, and the molecular weight ( M W )of a sample component: MW
=
CL,
Cis normally constant for a narrow range of sample molecular structures-e.g., synthetic polymers of a given type-so that a measurement of GPC elution volume for a narrow molecular weight sample fraction provides a value of C from a plot of the type shown in Figure 3. Equation 1 seems to adequately describe the relationship between molecular weight and elution volume for narrow GPC fractions from a single asphalt, as shown in Figure 3 (open squares and triangles, for two different original asphalts). However, the value of the constant C can vary by a factor of three among different asphalt samples, particularly for solvent extracted asphalts. This reflects the heterogeneity of typical asphalts in terms of the types of molecular components present. Bulky molecules have smaller molecular lengths per unit molecular weight, and give larger values of C in Equation 1. Consequently, it is necessary to calculate C for each asphalt to be analyzed by GPC. This can be done easily, providing we measure the number-average molecular weight of the original asphalt sample in each case: Vapor phase osmometry (VPO) was used in the present study, as described in the Experimental section. Use of a “standard” value of C (equal to 20, from Figure 3) provides an initial, uncorrected molecular weight distribution, from which an initial estimate of &Vncan be calculated. The true value of C is then equal to 20 (A%‘n)t,,,/(L%‘n)est. With a final value of C, the true molecular weight distribution can be calculated. This is most conveniently expressed as an accumulative plot of wt of sample us. molecular weight, as in Figure 4, or as a tabulation of maximum molecular weight us. accumulative sample percentages (Table I).
Lvn.
z
(9) J. G . Hendrickson and J . C . Moore, J . Polymer Sci., A1(4), 167 (1966). VOL. 41, NO. 10,AUGUST 1969
1225
DISCUSSION OF RESULTS
Table I provides some typical asphalt molecular weight distribution data, as calculated by the present procedure. These are based on UV detection, and are to some extent inaccurate because of the variable UV response curves for different asphalt samples-e.g., C and D in Figure 3). However it
should be noted that the procedure for calculating a true value of C in Equation 1 partially compensates for this infor the calculated molecular weight accuracy, because distribution is set equal to the experimental value of Some internal consistency checks given below suggest that this is not a serious practical problem. The data of Figure 4 and Tables 1-111 are corrected for normal band broadening and
n,
a,.
Table I. Typical Asphalt Molecular Weight Distributions as Determined by GPC Molecular weight “Good” asphalts “Bad” asphalts F E G H 400 380 450 506 490 455 430 535 660 515 680 550 820 595 815 635 lo00 660 715 945 730 1190 810 1090 835 1380 930 1280 1640 995 lo00 1540 2050 1260 1330 1930 1810 3 300 2950 1900 2900 5300 2850 4900 6.7 4.2 4.2 5.1 0.93 0.95 0.82 1.os 720 986 768 994
x
Wt of sample falling below indicated mol wt. 5
10 20 30 40 50 60 70 80 90 95 Rso,io
(C/20) M n
asphaltenes 1110 1310 1660 1990
2300 2700 3150 3700 4500 6200 8900 4.7 2.3
2400
Table 11. Comparison of Molecular Weight Distributions on Isopropanol Extracted Fractions with Original Sample Distribution Molecular weight Wt Z of sample falling below Fractions 1 2 3 Composite Original asphalt indicated mol wt. 530 560 510 490 5 415 580 640 565 555 10 465 655 760 655 670 20 540 720 860 735 770 30 605 785 945 800 860 40 640 845 1040 870 960 50 685 915 1190 960 1050 60 725 1000 1380 1100 1160 70 785 1330 1360 80 860 1130 1640 1750 1760 90 1020 1390 2100 2200 2300 95 1160 1690 2600 2.4 3.3 3.1 Reom 2.2 c/20 0.98 1.06 1.06 834 1009 E n 653 Table 111. Repeatability of Present GPC Asphalt Analysis
z
Wt of sample falling below indicated mol wt. 5 10 20 30 40
Molecular weightD (sample E of Table I) 2 3 4 5 440 440 450 450 450 530 530 540 535 535 675 670 685 680 690 820 805 820 820 825 935 955 920 940 965 1110 1100 1080 1080 50 1090 1310 1300 1260 1270 60 1260 1590 1570 1520 1500 70 1500 2030 1970 1940 1850 80 1850 3000 3020 3350 2560 90 2750 4900 6000 4900 95 4500 4200 a Numbers 1-6 represent run numbers, uncertainty figures are std. deviations.
1226
1
ANALYTICAL CHEMISTRY
6 450 535 670 795 930 1100 1290 1540 1940 3000 5000
Av.
447 534 678 814 946 1093 1282 1537 1930 2940 4900
f 5 f 4 f.8 &12
+13
f.12 f21 *37 f.70 f270 f600
the tailing of high molecular weight fractions into lower molecular weight fractions. Figure 4 compares a molecular weight distribution curve (solid line) obtained by the present GPC method with corresponding data (squares) from preparative GPC separation and molecular weight determinations on individual fractions. The differences in these two sets of data arise mainly from errors in the corresponding molecular weight determination for the original sample is by VPO. If the VPO value of obtained from the composited GPC fractions (squares), a value of 837 is obtained, us. a value of 768 in the original sample. Use of the 837 value (rather than the 768 value) in the routine GPC calculation yields the dashed curve of Figure 4. This is seen to be in reasonable agreement with the data by preparative scale GPC. A series of fractions from the successive extraction of an asphalt at two temperatures with isopropanol are shown in Table 11. Comparison of the original feed analysis with the composited fraction data shows good agreement. These two examples indicate that the present method is reasonably reliable, despite the limitations of UV detection. The repeatability of the present GPC method is illustrated in Table 111, for replicate analyses of the same sample using a single experimental value of ATn. The resulting molecular weight distributions from the present GPC procedure furnish three useful parameters for characterizing the sample. The ratio of molecular weights at
90 and 10 wt % ( R W , ~is~a) good measure of the molecular weight spread of the sample. The quantity C/20 provides a measure of the deviation of the sample molecular weight from the calibration of Figure 3, and is an indicator of average molecular shape. Large values of C/20 correspond to relatively bulky molecules, while small values of C/20 mean long, narrow molecules. The number-average molecular weight is the third characteristic parameter of interest. These various parameters are tabulated in Table I and I1 for the samples shown there. The difference between these particular “good” and “bad” asphalts seems to be related to molecular weight spread (Roo,lo)and to mean molecular weight, but these may be accidental correlations based on different boiling ranges in the original crude oils. The molecular shape factor C/20 does not appear to be related to asphalt performance. Asphaltenes have higher average molecular weights and large values of C/20. As expected, asphaltene molecules are bulkier and more compact than other asphalt components. ACKNOWLEDGMENT
Several asphalt GPC fractions (triangles in Figure 3, sample C of Figure 2) were kindly supplied by S. W. Nicksic of the Chevron Research Corp., La Habra, Calif. RECEIVED for review April 4, 1969. Accepted June 2, 1969.
Rapid Analysis of Ribonucleosides and Bases at the Picomole Level Using Pellicular Cation Exchange Resin in Narrow Bore Columns Csaba Horvathl and S . R. Lipsky Section of Physical Sciences, Yale University School of Medicine, New Haven, Conn.
The separation of the four major bases and nucleosides of RNA has been investigated under conditions utilizing wide ranges of pH, temperature, and flow rates. Smallbore columns packed with pellicular cation-exchange resin were used in a high-pressure liquid chromatograph equipped with a micro UV detector. Dilute acidic potassium or ammonium phosphate solutions served as eluents. The four bases or nucleosides were analyzed at the subnanomole level in less than 6 minutes using 3-meter colunms and high-flow velocities. With a shorter column operated at relatively lowflow velocities, a few picomoles of ribonucleosides can be separated and determined without resorting to radioactive techniques.
THE DETERMINATION of the base composition of nucleic acids is one of the most important steps in sequence analysis. Further progress in this field largely depends on the development of highly sensitive and rapid techniques for quantitative analysis of nucleic acid constituents. High sensitivity is often required for the minute quantities of nucleic acids or nucleic acid fragments that may be available-e.g., by zonal centrifugation or single cell isolation. On the other hand, the need for a large number of analyses can only be overcome by using a fast and reliable method that is amenable to automation. Thus far, attempts to utilize gas chromatography to analyze nucleic acid hydrolyzates, although promising 1 Also Department of Engineering & Applied Science, Yale University.
06510
( I , 2), have not been wholly successful. Thin-layer and paper chromatography (3, 4) are widely used, but quantitative analysis is time consuming and generally inaccurate below the nanomole level (5). Using an elaborate radioactive tracer technique, however, the Randeraths (6, 7) recently achieved base composition determination at the picomole level. Higher sensitivity has been obtained only in the quantitation of ATP by the luciferin-luciferase enzyme system (8). Various column chromatographic methods with gels (9, IO), with a ligand exchange resin ( I I ) , with a liquid ion (1) C. W. Gehrke and C . D. Ruyle, J. Chromatog., 38, 473 (1968). (2) M. Jacobson, J. F. O’Brien, and C . Hedgcoth, Anal. Biochem., 25,363 (1968). (3) K. Randerath and E. Randerath, in “Methods in Enzymology,” Vol. XII, “Nucleic Acids,” Part A., L. Grossman and K. Moldave, Eds., Academic Press, New York, 1967, pp 323-50. (4) G. R. Wyatt, in “The Nucleic Acids Chemistry and Biology,” Vol. I, E. Chargaff and J. N. Davidson, Eds., Academic Press, New York, 1955, pp 243-65. (5) J. M. Gebicki and S . Freed, Anal. Biochem., 14,253 (1966). (6) K. Randerath and E. Randerath, ibid., in press. (7) E. Randerath, J. W. Ten Broeke, and K . Randerath, FEBS Letters, 2, 10 (1968). (8) G. E. Lyman and J. P. DeVincenzo, Anal. Biochem., 21, 435 (1967). (9) I. Mezzasoma and B. Farina, Boll. SOC.Ital. Biol. Sper., 42,1449 (1966). (10) M. Carrara and G. Bernardi, Biochim. Biophys. Acta, 155 (1968). (11) G. Goldstein, Anal. Biochem., 20,477 (1967). VOL. 41, NO. 10,AUGUST 1969
1227
