Anal. Chem. 1990, 62,2114-2122
2114
perta Med. Found., H u m n Growth Hormone: A Family of Proteins: Pecile, A., Muller, E. E., E&.; Malan: Amsterdam, 1972; p 64. (23) Town&, R. J.; Weinbrger, L.F T l m s h f f , S. N. J . Am. C M . sot. 1960, 82,3175.
RECEIVED for review March 7 , 1990. Accepted June 26, 1990.
This majority of this work was supported by LDC AnalytiInstruments Systems' funding was provided, in part, by Pfizer, Inc., and Merck Sharp & Dohme Research Laboratories, Inc. This is contribution number 427 from The Barnett Institute at Northeastern Univerisity.
Determination of Warfarin-Human Serum Albumin Protein Binding Parameters by an Improved Hummel-Dreyer High-Performance Liquid Chromatographic Method Using Internal Surface Reversed-Phase Columns Thomas C. Pinkerton* and Kenneth A. Koeplinger
Control Division, Building 259, Mail Stop 12, The Upjohn Company, Kalamazoo, Michigan 49001
The Hummel-Dreyer slzaexcluslon hlgh-performance liquid chromatographic method for the determination of proteln Mndlng parameters has been improved and autmated by use of an Internal surface reversed-phase (ISRP) column (5 cm X 4.6 mm) and a computercontrolkdmob#ophase dellvery system wRh low volume syrlnge mixlng. The high-etflclency ISRP columns, whlch are nonadsorptlve and excluslonary to serum protdns but aliow partWlng of small molecules with an Internal peptkle bonded phase, maintain htgh performance after many InJectkns d human sefum albumin (HSA), enable the use of short columns, and provlde for the resdutlon of primary llgand from protein binding dlsplacers. The modifled Hummel-Dreyer high-performance llquld chromatographlc method was demonstrated by the determlnatlon of blndlng parameters for warfarln-HSA In phosphate buffer, whkh were found to be n , = 1.0, n 2 = 2.1, K , = 3.30 X lo* W', and K , = 2.03 X lo4 M-l. The necessary sequence of chromatographic experiments was repeated 4 tlmes at 18 separate warfarin moble phase concentrations. Each automated sequence required 8 h to complete. The parameters were measured with a predsslon of less than 10% relative standard devlatlon.
INTRODUCTION Drugs can be highly bound to blood plasma proteins, and the extent of binding can have important consequences relative to the biological distribution and clearance of such drugs (1-4). This is of particular concern with toxic drugs that have narrow therapeutic indices and are bound to plasma proteins in excess of 90%. In such cases, the protein binding parameters must be determined with confidence. Further, susceptibility of the drug to displacement by other substances must be well established. When a compound binds to only one protein binding site, generally, a binding constant can be determined in a straightforward fashion with a variety of methods. However, when a small molecule binds to more than one class of binding sites on a protein, the drug bound becomes a complex function of the amount of drug present. The determination of the binding parameters can be complicated by the means of controlling the drug concentration, the number of data points collected over the drug concentration range, the procedures used for measuring the amount of drug bound, and the data analysis methodology.
Binding of warfarin to human serum albumin (HSA) is a classical example of a multiclass model which has been studied by a variety of methods, yet comparisons of reported results illustrate inconsistencies (5-9). Warfarin is an anticoagulant which decreases prothrombinogenic activity by inhibiting the vitamin K cycle (10). Warfarin can be administered in maximum doses of 15 mg per day, yielding total plasma concentrations of 3 to 8 pM in paitents 1-10 h after administration (11). Because of the toxicity of warfarin, the narrow therapeutic index, and the ease with which it can be displaced from protein by other substances, it is important to know the percent of warfarin bound to HSA under physiological conditions. Because of variations in HSA concentrations among patients and disease states, it is sometimes desirable to theoretically predict changes in unbound (free) warfarin concentration. For this to be done, warfarin-HSA binding constants must be accurately known. Human serum albumin (HSA) has two classes of binding sites for acidic drugs, such as warfarin (I, 2). It is assumed that warfarin binds to HSA in two separate binding regions in which each ligand associates independently to individual sites of a given class. This means there can be a different total number of individual binding sites within each class. The independent model assumes that binding at one site does not affect binding at any other site. The model can be represented by the following expression: n,KIPDf n2K2PDf Db = (1) 1 + K1Df 1 KzDf where Db is the concentration of drug bound to protein, Df is the free drug concentration, P is the total protein concentration, n j is the number of drug molecules bound in each binding site class, and Kiis the equilibrium association constant (i.e., affinity constant) for each class of sites. Derivation of fundamental binding relationships can be found elsewhere
+ +
(12).
The primary objective in studying any two class drugprotein binding is to determine the binding parameters n,, n2,K1,and K2 with acceptable accuracy and precision, under some set of fixed conditions (i.e., temperature, pH, ionic strength, and protein concentration). Ideally, one wishes to measure the bound drug concentration (Db) directly as a function of the free drug concentration (Of),which should be controlled as the independent variable. Since the relationship between Db and Df is nonlinear, a sufficient number of appropriately spaced data points must be taken over a drug
0003-2700/90/0362-2114$02.50/00 1990 American Chemical Society
ANALYTICAL CHEMISTRY, VOL. 62, NO. 19, OCTOBER 1, 1990
concentration range approaching protein saturation. Reliable cruve-fitting procedures must be employed to ascertain the binding parameters. Of course, the protein must be pure and its concentration must be accurately known. Lastly, the type of analytical technique used to discriminate the bound drug from protein and drug-protein complexes must be free of interactions which affect the equilibria. A variety of analytical methods have been used to study protein binding of small molecules to large proteins, including equilibrium dialysis, ultrafiltration, gel-filtration, and spectroscopic techniques. In the absence of membrane adsorption and Donnan effects, equilibrium dialysis and ultrafiltration can provide accurate measurements of binding parameters with inexpensive equipment. However, to acquire a large number of data points, the techniques can be very laborious. In addition, since each point represents a separate experiment, involving solution preparations and an additional membrane, the resulting binding parameters can be imprecise. Spectroscopic techniques can be used to monitor protein binding by titration, if the small molecule being studied undergoes a shift in spectral properties on binding (13). Spectroscopic methods enable larger numbers of data points to be collected more efficiently than membrane dialysis and ultrafiltration, but spectroscopic techniques are limited by the inherent physical properties of the molecules being studied. Also, spectroscopic equipment is often expensive or inaccessible. Gel-filtration chromatographic techniques include the direct injection method, frontal analysis, and the Hummel-Dreyer method. Although useful in some cases, direct injection of an equilibrium mixture onto a size-exclusion column cannot be considered an equilibrium method, because on-column dilution dissociates the drug-protein complex. On the other hand, frontal analysis and the Hummel-Dreyer method are considered equilibrium techniques (12). In frontal analysis an equilibrium mixture of drug and protein is pumped into a size-exclusion column until the column is saturated (14). Because of differential migration between the protein, drugprotein complex, and free drug, plateaus form as the large sample elutes from the column. With a detector responding to the drug, the free drug can be measured from the last eluting plateau. With low efficiencies, low peak capacities, and tailing inherent in conventional size-exclusion columns, frontal plateaus are difficult to quantify, thus the method is used infrequently. In addition, frontal analysis can consume large amounts of protein. In the classical Hummel-Dreyer method (15),a size-exclusion column is equilibrated with mobile phase containing the drug (or ligand). Samples with protein at a fixed concentration and drug at the same concentration as the mobile phase are injected onto the column. The detector response monitors the level of drug in the mobile phase. A peak containing the free protein and all drug-protein complexes elutes a t the interstitial (extraparticular) volume of the column. After the elution of this peak, the detector response returns to baseline. Later in the chromatogram, a trough or vacancy appears that can be related to the concentration of the drug that is bound to protein at equilibrium with the drug in the mobile phase. Since equilibrium is established rapidly on introduction of the sample to the mobile phase and the vacancy is completely separated from the protein complex peak, the drug in the mobile phase represents the free drug concentration at equilibrium. This particular feature of the Hummel-Dreyer method is unique among binding methods, in that it enables the free drug concentration to be controlled as the true independent variable. All other protein binding methods require that the total drug concentration be controlled and the free drug concentration measured. In methods
2115
other than the Hummel-Dreyer method, the bound concentration is then calculated by subtraction of free from total concentration. This contributes to imprecision in the results, because uncertainty in measuring the free drug concentration is compounded with uncertainty in preparing the samples. With highly bound drugs, the ones of most interest, this uncertainty can be significant. The error associated with measuring very low free drug concentration can bias the estimates of primary binding, while error in discriminating free drug concentration at protein saturation can bias estimates of secondary binding. Inclusion of this measurement error in the independent variable violates fundamental nonlinear curve-fitting assumptions (16). Although the Hummel-Dreyer method has the fundamental advantage over other binding techniques of being able to control the free drug concentration as a true independent variable, it has seen limited use, because of inadequate chromatographic technology. First, conventional size-exclusion columns have exhibited low efficiencies, poor protein recoveries, and short column lives. Second, because of the low peak capacity inherent in size-exclusion columns, long columns have been required to separate protein peaks from drug vacancies. This has meant long elution times and large consumption of drug. Third, preparation of mobile phases, column equilibration times, and unreliable chromatographic pump performance meant the method has been very cumbersome to use. In recent years, the advancement of chromatographic technology has led to significant improvements in mobile phase delivery systems and size-exclusion columns. Highprecision, low-volume syringe metering pumps now enable the continuous mixing of solutions for isocratic elution with greater reproducibility than was previously possible. This, coupled with computer-controlled high-performance liquid chromatography (HPLC) systems, now makes it possible to conduct an entire series of Hummel-Dreyer experiments (i.e., change in mobile phase drug concentration, column equilibration, sample injection, and detection) under complete automation. More importantly, the new multimechanism HPLC columns, which separate proteins by size exclusion but resolve small molecules by partitioning, have been designed to yield high protein recoveries, high efficiencies, and long column lifetimes. This means the Hummel-Dreyer method can be conducted with high resolution on small columns, thus allowing shorter analysis times while consuming less drug. These advances in chromatographic hardware have provided an opportunity for the Hummel-Dreyer method to become a viable, efficient method for the determination of multiclass binding parameters. The HPLC columns used in this study are the so-called internal surface reversed-phase (ISRP) type "restricted access" columns. The internal surface reversed-phase columns were specifically designed to facilitate the HPLC analysis of drugs in blood plasma or serum by direct injection (17,18). The modified spherical porous silica (5 Mm) packing material functions through the size exclusion of proteins and the bonded-phase partitioning of small molecules on the internal surface of the packing. The ISRP columns make use of a glycerylpropyl (diol) bonded phase to eliminate protein adsorption. In addition, the ISRP packing has glycine residues on its external surface to aid in repulsion of negatively charged proteins, such as albumin. Unlike conventional size-exclusion columns, a tripeptide (Gly-Phe-Phe) is covalently coupled to the diol phase through the N-terminus. The phenylalanine groups bound to the external surface are cleaved from the packing material using carboxypeptidase A, leaving the hydrophilic glycine on the external surface. The hydrophobic Gly-Phe-Phe phase is left intact in the internal pores of the
2116
ANALYTICAL CHEMISTRY. VOL. 62. NO. 19. OCTOBER 1. 1990
packing material, since the enzyme is size-excluded from the small pore silica. Drugs of low molecular weight (typically less that 2000) are able to enter the pores of the ISRP packing and partition with the tripeptide bonded phase. Large serum proteins elute in the column interstitial void volume and are not adsorbed to the hydrophilic external surface of the packing. Studies have demonstrated that plasma protein can he recovered from ISRP columns on the order of 99 3% at neutral pH (19). The ISRP columns have been used for HPLC assays with direct injection of serum or plasma in a variety of clinical applications (20-25). The nature of ISRP columns offers some unique advantages for studying drug-protein binding equilibria by HPLC methods. The utilization of ISRP columns for the study of drug-protein binding, thus far, has included the use of frontal analysis (2627) and an unusual split-peak phenomena (28-31). Encouraging results in the above applications have prompted an evaluation of ISRP column technology for the determination of drug-protein binding parameters using the Hummel-Dreyer method. Potential advantages over the use of conventional size-exclusion columns include better column performance, decreased analysis time, control of drug elution, and chromatographic resolution of displacer from drug in competitive binding experiments. The object of this study was to utilize short ISRP columns and a computer-controlled HPLC system to investigate whether the Hummel-Dreyer method could he modified and automated to provide a more precise means of determining multiclass drug-protein binding parameters in less time than has been previously possible. EXPERIMENTAL SECTION Chromatographic Apparatus. The Hummel-Dreyer experiments were performed with a Hewlett-Packard HP 1090M liquid chromatograph with an integrated ternary mobile phase delivery system, autosampler, column oven, and photodiode-array detector. Chromatographic conditions, mobile phase delivery, solution mixing, and data acquisition were controlled by an HP 9000 Model 310 computer (HP ChemStation). Internal surface reversed-phase (ISRP) columns (GFF-S5-80) and guard cartridges were obtained from the Regis Chemical Co., Morton Grove, IL. Coupled ISRP guard cartridges (each 1 cm X 3.0 mm id.) were used initially to establish appropriate column length. A 5 cm X 4.6 mm i.d. ISRP column (no. 049272) was used for actual binding measurements. Unless stated otherwise, Chromatographicparameters included an injection volume of 50.0 rL and a flow rate of 2.00 mL/min. Experiments were conducted at ambient temperatures of 23 i 1 "C or at a constant temperature of 25.0 0.5 OC by control of column temperature with the HP 1090M oven and external refrigerated water. Water from the external circulator was passed through the HP 1090M column oven thermal exchange block. Chromatograms were acquired with the HP 1090M photodiode-array detector set at a wavelength of either 310 or 214 nm with bandwidths of 4 nm. A "reference" wavelength of 550 nm with a bandwidth of 100 nm was used for noise reduction. The photodiode-array detector was automatically balanced against the column eluent at the start of each run. Data were stored on 20MB removable disks run on a Bering 5840-RM Bernoulli disk drive. Reagents and Solutions. Racemic warfarin (rac-3-(aacetonylbenzyl)-4-hydroxycoumarin)(catalog no. A-2250, lot 16F-0026)was purchased from the Sigma Chemical Co.. St. Louis, MO. The aqumus insoluble warfarin was converted to the soluble sodium salt by reaction with 50% aqueous solution of sodium hydroxide (analytical reagent, Mallinckrodt Inc.). A 8.12 mM sodium warfarin stock solution was used to prepare 81.2 pM warfarin and 16.2 pM warfarin by dilution with a 0.W7 M sodium phosphate buffer (pH 7.4, ionic strength 0.15). The sodium phosphate buffer was prepared with anhydrous dibasic sodium phosphate (99.7%,J. T. Baker, Inc.) and monohydrate monbasic sodium phosphate (99.1%. Baker), and pH adjusted with 85% phosphoric acid (Mallinckrodt, Inc.). Water used in preparation
*
and dellvery &stem. Reproducec
wth
(Copyright 1984 Hewlen-Packard Cornpan;
permission )
of the phosphate buffer was purified by a Milli-Q Water system (Millipore, Bedford. MA) using house deionized feed water. The phosphate buffer was filtered through 0.45-pm Nylon 66 membranes (Alltech Associates, Inc., Deerfield, IL) prior to use. Racemic ibuprofen (rac-ol-methyl-4-(2-methylpropyl)benzeneacetic acid) was obtained from within The Upjohn Company. Ibuprofen was converted to the aqueous soluble sodium salt in the Same manner as warfarin. A 12.1 mM sodium ibuprofen stuck solution was used to prepare 121 pM ibuprofen by dilution with 0.067 M sodium phosphate buffer (pH 7.4). Human serum albumin (HSA) was obtained from two sources. "Essentially" globulin free and fatty acid free (99% pure by electrophoresis), was purchased from Sigma. Fatty acid free (0.02%) HSA, Cohn Fraction V (catalog no. 82-323-4. lot no. 19; lot analysis, 99.9% pure hy nitrogen assay), was purchased from ICN Biomedicals Inc., Costa Mesa, CA. The HSA distributed by ICN was Pentex Brand "Fatty Acid Free" HSA (catalog no. 82-323-4) manufactured by Miles Scentific, Naperville, IL. A fresh HSA sample solution was prepared for each series of binding experiments, by dissolviong 39 mg of HSA in 10 mL of 0.067 M phosphate buffer (pH 7.4). The HSA concentration in each solution was determined spectrophotometrically. Determination of Protein Concentration. The HSA concentration in each sample (ca. 54 pM) was accurately measured spectrophotometrically. The absorbance of an HSA sample solution, diluted with 0.067 M phosphate (pH 7.4), was measured at 280 nm with an HP 8451A photodiode array spectrophotometer using the diluent as a reference. The molar absorptivity of the HSA at 280 nm (3.300 X 10' A V ' M-') was determined from absorhance measurements on an identical standard solution, whose absolute HSA concentration had been ascertained from quantitative amino acid analysis. Quantification of the HSA concentration by amino acid analysis eliminates the need to correct for nonprotein substances (Le., hydrated water, salts, etc.) hut requires that the protein he free of proteinaceous impurities. The amino acid analysis on the standard HSA solution was run on a Beckman Model 7300 amino acid analyzer using four hydrolyzed replicates. Three amino acid types GIX (Glu + Gln), Leu, and Lys were used to calculate the HSA concentration and the results were pooled. Chromatographic Procedure for Hnmmel-Dreyer Method. The 81.2 pM and 16.2 pM warfarin solutions, and 0.067 M (pH 7.4) phosphate buffer were placed in reservoir positions A, B, and C of the HP 1090M liquid chromatograph, respectively. The solutions in A and C or B and C were mixed throughout each chromatographic run with low pressure syringe metering by the HP 1090M delivery system (Figure 1)to produce mobile phases containing various concentrations of warfarin in 0.067 M (pH 7.4) phosphate buffer. The chromatographic sequence began with the most concentrated mobile phase by metering 100% A with 1% C, followed hy 90% A with 10% C, then 80% A with 20% C, etc. for 10 runs. This was followed by 10 more runs with a similar sequence metering solutions B and C. Since two concentrations
ANALYTICAL CHEMISTRY, VOL. 62, NO. 19, OCTOBER 1, 1990
between the sequences were degenerate, 18 different mobile phase drug concentrations were produced for each set of experiments. Mobile phase at each warfarin concentration was pumped isocratically for 15 min to equilibrate the 5-cm ISRP column. A run began by automatically balancing the detector against the warfarin mobile phase. Next, a blank injection was performed with 50 pL of buffer. Then, at the same mobile phase composition, a 50 p L sample containing only HSA in buffer was injected. Paired chromatographic injections (Le., blank and protein) were run at the different warfarin mobile phase concentrations in an automated sequence, controlled by the HP ChemStation. The vacancies were integrated and reports generated, automatically, for each set of experiments. For the binding displacement experiments with ibuprofen, 81.2 pM warfarin, 121 pM ibuprofen, and 0.067 M (pH 7.4) phosphate buffer were placed in reservoirs A, B, and C of the HP 1090M liquid chromatograph, respectively. The solutions were mixed to give a mobile phase containing warfarin and ibuprofen in phosphate buffer. Data Analysis. Subtraction of buffer blanks, construction of linear regression calibration plots from blank injections, and calculation of warfarin bound at each free drug concentration were performed with Lotus 1-2-3software run on an IBM AT computer. Binding parameten (nl, n2,K1, and K2)were determined by means of nonlinear least-squares curve fitting of warfarin bound versus free warfarin concentration plots (eq 1) using the simplex algorithm. The nonlinear least-squares curve fitting program was written in BASIC and run on an IBM A T computer. The program, described in detail elsewhere (32),can use either the simplex or damping Gauss-Newton algorithms.
RESULTS AND DISCUSSION Previous Hummel-Dreyer Methods. In a classical Hummel-Dreyer experiment a size-exclusion column is equilibrated with a mobile phase containing a given concentration of ligand in buffer (15). An aliquot of the mobile phase is then used to prepare a sample by dissolving a known amount of protein. The ligand-protein sample is injected onto the column and equilibrium between the ligand and protein is rapidly reestablished in accordance with the free ligand concentration in the mobile phase. The protein and ligandprotein complexes elute in the column interstitial void volume, while the ligand migrates a t a velocity dictated by its pore volume penetration. Since the detector monitors the ligand in the mobile phase, the chromatographic baseline later dips to create a vacancy because the unbound drug in the sample was less than that in the mobile phase. If the ligand had not bound to the protein at all, the detector response would not have changed because the amount of ligand per unit volume in the sample was the same as the mobile phase. If the ligand binds to the protein by a finite amount, then the vacancy represents that amount bound at equilibrium with the free ligand concentration in the mobile phase. It is important to emphasize that the state of equilibrium is established in the mobile phase and not in the sample, so the total ligand concentration, which is not controlled in the Hummel-Dreyer method, is the sum of the measured amount bound and the free ligand concentration in the mobile phase. To determine binding parameters, these experiments can be repeated over a range of free ligand concentrations in the mobile phase, always using the mobile phase to prepare the protein sample. Therefore, the amount of ligand bound is measured directly as a function of the free ligand in the mobile phase. A variation of the Hummel-Dreyer method ( 6 , 1 5 , 3 3 )involves using the same sample preparation described above except adding an excess of ligand to the sample, compared to the mobile phase. A collection of samples are prepared with incremental increases in the excess of ligand added to the sample. When the samples are chromatographed, if the unbound ligand concentration in the sample is less than the mobile phase, then a vacancy will appear, or if the unbound ligand concentration in the sample is greater than the mobile
2117
phase, then a peak will occur (see figure 1 of ref 33). The amount of ligand required to fill the vacancy is determined by plotting the absorbance response (negative for vacancy and positive for peak) against the excess of ligand added to the sample and extrapolating to zero absorbance. It has been assumed by previous investigators that this variation renders better precision, because of the more experimental points involved in the determination (34, 35). However, one should use this method with caution, for their is no guarantee that the excess ligand added to the protein sample will necessarily be linear over a wide range. If plots are extrapolated beyond the range of data collection,erroneous results could be obtained. In order to determine binding parameters with this method variation, similar sets of experiments are repeated at different mobile phase concentrations. The large number of sample injections is a particular problem when determining binding parameters for multiclass binding, where many points are desired. Typically, five sample solutions and 18 mobile phase concentrations require 90 experiments. Without automated chromatographic equipment, this could take as much as a month to collect enough data to calculate one set of binding parameters. Modified Hummel-Dreyer Method. In an effort to simplify the Hummel-Dreyer method, it was recognized that if the protein ligand binding equilibria were "instantaneous" relative to the chromatographic process, then a sample of protein alone could be injected rather than ligand-protein mixtures. This assumption was found to be valid for warfarin-HSA binding, described below. In principle, this greatly improves the precision of the experiments by requiring preparation of only one sample solution, whose protein concentration can be measured independently. Further, it reduces the number of experiments required. Introduction of a protein solution into a mobile phase, which has been equilibrated to some free ligand concentration, results in removal of ligand from the mobile phase, because of the protein binding, and creates in the elution profile a trough whose area is the linear sum of a vacancy buffer blank and the bound ligand. Thus, the bound ligand concentration can be measured directly, by simple blank subtraction, while the free ligand concentration is controlled in the mobile phase as the true independent variable. Further it was surmised that if the ligand could undergo partitioning on a high-performance chromatographic column, while the protein is size-excluded, then potentially more efficient separations could be achieved in less time with shorter columns. The internal surface reversed-phase (ISRP) columns meet the requirements of nonadsorptive, size exclusion of HSA with concurrent chromatographic partitioning of the ligand. Selection of Protein Concentration. The sample protein concentration to be used in the modified Hummel-Dreyer experiment is governed by the dynamic range of the detector and the characteristics of the drug-protein binding system. Since the detector responds to the drug in the mobile phase and provides the baseline from which the vacancy response is measured, the upper limit will be determined by detector saturation. Therefore, the mobile phase free drug concentration must be kept within the linear range of the detector. Once the available free drug concentration range is ascertained, the protein concentration of the sample is chosen such that the ratio of drug bound per unit protein approaches protein binding site saturation a t high drug concentrations, while, on the other hand, sufficient protein is available for an adequate response at low drug concentrations. After preliminary experiments, the optimum amount of protein to be injected can be established.
ANALYTICAL CHEMISTRY, VOL. 62, NO. 19, OCTOBER 1, 1990
2118
6003
?
L 5-
100;
u
.
0
5
5
0
0
Figure 3. Four representative Hummel-Dreyer elution proflles on 5 cm ISRP column for warfarin-HSA bindlng as a function of the change in warfarin mobile phase concentration: (A) 81.1 pM, (8) 32.5 pM, (C) 8.1 pM, and (D) 1.6 (other conditions same as Figure 1). Top trace is injection of buffer blank and bottom trace is from injection of HSA.
experiments using 18 different mobile phase drug concentrations was conducted in an unattended fashion, starting with the highest drug concentration. With 5 min for each injection and 15 min for column equilibration of each mobile phase, a set of experiments took 8 h to complete. What previously might have taken weeks to acomplish has now been reduced to an overnight run. Determination of Warfarin-HSA Binding Parameters. The modified Hummel-Dreyer experiments were conducted over a free drug concentration in the mobile phase of 1.6 to 81.1 pM using 18 separate mobile phases, for the determination of the warfarin-HSA binding parameters. A 50-pL solution of 54 pM HSA in 0.067 M phosphate buffer and a buffer blank was injected for each experiment. Four representative pairs of chromatograms (blank and HSA) obtained from a sequence of experiments are shown in Figure 3. The vacancy areas obtained from the blank injections were regressed to provide a warfarin linear response calibration. In order to verify that the vacancy area integration was comparable to normal HPLC peak integration, a calibration curve was created by injecting varying warfarin concentrations into a phosphate buffer mobile phase. The peak and vacancy calibrations were found to be coincident. Linear regression of a typical vacancy calibration gave the following values: Y = (12.3area units/pM) X + (-4.221, R2= 0.9998. The vacancy calibration data for each separate set of experiments were used to calculate the amount of drug bound from the vacancy areas of the HSA injections, after the buffer blank areas had been subtracted. The removal of the blank area (Figure 2, trace A) from the overall measured response is important because, obviously, no protein binding is necessary to generate this portion of the vacancy, which is created by sample volume displacement. The sequence of Hummel-Dreyer experiments was repeated four times: twice with Sigma HSA at ambient temperature (expt set 1 and 2); once with Sigma HSA at 25 O C (expt set 3); once with Miles HSA at 25 "C (expt set 4). The parameters (nl, n2,ICl, and K,) were found by iterative nonlinear leastsquares regression of eq 1, using the simplex algorithm to find the best fit of the warfarin bound versus the free warfarin concentration a t a constant HSA Concentration. A typical curve is illustrated in Figure 4. The residuals for this regression are shown in Figure 5. Akaike's information criteria (AIC) (32) and root mean square (RMS) errors were used to evaluate goodness of fit. The AIC and RMS values are defined as
AIC = N In (SS)
+ 2M
RMS = ( S S / ( N - M))1/2
(2)
(3)
where N is the number of points regressed, SS is the final sum
ANALYTICAL CHEMISTRY, VOL. 62, NO. 19, OCTOBER 1, 1990
2119
Table I. Protein Binding Parameters for Warfarin-HSA in 0.067 M Phosphate (pH 7.4) Determined by Modified Hummel-Dreyer Method expt set
nl
n2
10-’K1, M-’
10-’Kz, M-’
N
AIC
RMS’
rmb
temp, “C
HSA
1 2 3 4
1.09 1.03 0.97 1.09
1.92 2.21 2.09 2.12
3.13 3.63 3.50 2.94
2.14 1.96 2.10 1.93
16 18 19 17
45 33 69 69
0.91 0.54 1.27 1.67
2.3 2.4 2.2 2.3
23c 23c 25d 25d
Sigma Sigma Sigma
mean sd
1.05 0.06 5.7%
2.09 0.12 5.7%
3.30 0.32 9.7%
2.03 0.10 4.9%
RSD
Miles
1.10
‘Root mean square errors in pM. *Maximum drug bound per unit protein. CAmbienttemperature f l “C. dControlled column temperature f0.5 “C.
1503 i
0
rrr,Tm-rrrTTT,
20.0
0.0
I
r
l
,
~
1
~
40.0 60.0 Warfarin free, uM
I
1
~
f
f
80.0
l
t
’
1
I
~
r
I
I
~
1
100.0
Figure 4. Simplex nonlinear regression of warfarin bound to HSA versus free warfarin concentration at equilibrium in the mobile phase (expt set 3). “X”s are experimental data points, while curve is regressed fit.
x
4 x
Flgure 5. Plot of residuals from simplex nonlinear regression (Figure 4) versus free warfarin concentration.
of squares, and M is the number of parameters determined from the fitted curve. The lower the AIC and RMS values, the better the fit. For this study N is equal to the number of data points plus one to account for the origin. The binding parameters determined from each set of experiments are given in Table I. The damping Gauss-Newton algorithm (determinant of Jacobian = 0.001) was used in an attempt to obtain better estimates; however, a significant decrease in AIC values was not observed. In addition, the variance between the experimental sets increased, suggesting the Gauss-Newton algorithm might have been distorting the parameters by converging to local minima; thus only the simplex algorithm was used to estimate the parameters (Table I). The average RMS of 1.1p M between the experiments is about 17’0 of the range of drug bound on the ordinate (Figure 4). All four sets of experiments yielded consistent parameters regardless of the slight temperature differences or source of
1
fatty acid free protein. The maximum amount of drug bound per unit protein (r) for these experiments was 2.4. The mean parameters indicate that the binding association constants and molecules bound per class can be determined with a precision of less than 10% RSD (Table I). Fundamental Kinetic Considerations. When any chromatographic approach for measuring binding parameters is used, the consideration of the kinetis of binding is important. Chromatographic methods require that equilibria be continually reestablished as the protein-drug complex separates from free drug during migration down the column. In the case of the modified Hummel-Dreyer method, described here, the protein alone is injected into a mobile phase containing the binding drug. In order for the vacancy areas to yield valid binding parameters, the half-life required for the drug-protein binding to reach equilibrium must be short with respect to the elution time. It has been estimated from kinetic modeling of size-exclusion chromatography of protein interacting substances that “instantaneous equilibration” can be assumed, if the protein binding dissociation half-life is less than 5% of the elution time (36). The dissociation process is rate limiting, because protein binding association rates are very rapid, on the order of lo5 M-’ s-l. Kinetic simulations under typical size-exclusion chromatographic conditions, exhibit chromatographic peak distortion of two coincident components in equilibrium with one another when the dissociation rate constants are less than 0.02 s-l (36). The HSA dissociation rate constants for warfarin a t 6 and 37 “C are 0.7 and 9.8 s-l, respectively (1). This means a t ambient temperatures the dissociation rate constant would be approximately 7 s-l. This very rapid dissociation corresponds to a half-life of 0.1 s. In the warfarin-HSA binding experiments conducted in this study, the protein complex elutes at about 12 s; thus, the dissociation half-life is 0.8% of the elution time, which is very much less than the 5% limit. It is sufficient to assume that “instantaneous equilibration” is achieved throughout elution. In the absence of kinetic information, one can observe the shape of the vacancy. If the vacancy is asymmetric, particularly fronting, then one can surmise that kinetics of dissociation may be too slow. An increase in elution time should decrease the asymmetry. Curve Fitting Concerns. When two class binding models are studied, the means of evaluating eq 1 can become critical. Often parameters are estimated from Scatchard plots. A Scatchard transformation can be obtained by dividing each side of eq 1 by the total protein concentration (P)to generate a term ( r ) on the left side of the equation, which represents the moles of drug bound per mole of protein. Plotting of the ratio r/(free drug concentration) versus r for a two-site binding model yields a nonlinear Scatchard plot. If done properly, graphic analysis of nonlinear Scatchard plots can yield reasonable approximations (37); however, more often than not, nonlinear Scatchard plots have been grossly misinterpreted. Zierler has recently described how some Scatchard plots are
2120
ANALYTICAL CHEMISTRY, VOL. 62, NO. 19, OCTOBER 1, 1990
Table 11. Comparison of Protein Binding Parameters for Warfarin-HSA Determined in Phosphate Buffer method
site class (i)
ni
frontal anal. frontal anal. Hummel-Dreyer (conventional) equil dialysis
primary primary primary secondary primary secondary tertiary primary secondary primary secondary primary secondary
1.4 1.2 1.3 3.8 1
equil dialysis fluorescent titration Hummel-Dreyer (modified)
1 10 1 2 1.0 2.0 1.0 2.1
Ki 2.4 X 2.1 x 2.2 X 4.2 X 4.8 x 1.5 x 2.0 x 1.4 X 1.8x 2.5 X 1.1 x 3.3 X 2.0 x
lo5 105 lo5 los 105 104
Pt 0"
rm
16 (S) 15 (s) 17 (S)
3.1 2.5 4.4
(s)
21
temp, "C
pH
note
ref
25 37 37
7.4 7.4 7.4
C C
5 6
C
6
3.5
25
6.1
d
7
103 lo6 104
12 (S)
1.9
4
7.4
C
8
lo5
24 (S)
1.5
22
7.4
e
9
104 lo5 104
66 (D)
2.4
23-25
7.4
C
f
Opt is number of data points for each study; (S) means Scatchard plots were used; (D) means direct nonlinear regression of data. *rm is maximum apparent drug bound per unit protein. c0.067 M phosphate buffer (ionic strength 0.15). dHSA contained fatty acids. e O . l M Dhowhate buffer with 0.9% NaCI. 'This work. misused (38). In addition to graphical problems, nonlinear regression programs cannot be used with Scatchard transformations because the transformation places experimental error in the abscissa; thus the orientation of the error changes throughout the curve. This violates the fundamentals of nonlinear least-squares curve fitting, which assume that experimental error resides only in the ordinate (16).To avoid transformation errors, when studying multiclass binding, bound drug versus free drug concentrations should be curve fitted directly with nonlinear least-squares programs. Protein Dilution. Another fundamental issue that arises in protein binding methods is the uniform distribution of protein within an apparatus. With many binding methods, such as membrane techniques, protein dilution or gradients can be of concern because the free drug concentration is being measured in a separate portion of the sample, and it must reflect the true protein concentration of the solution at equilibrium. This is important because the free drug concentration is not linearly related to protein concentration (eq 1). It is often noted with gel-filtration techniques that the sample is diluted on-column, so one questions whether the diluted protein concentration on-column truly reflects the protein concentration of the sample. With the short high-performance ISRP columns used in this study it can be shown from monitoring at 280 nm that the protein peak elutes with a bandwidth of about 0.12 min. At a flow rate of 2 mL/min this encompasses 240 pL; less 30 pL for off-column peak spreading yields 210 pL. Since the ratio of pore volume to interstitial volume for ISRP columns is 0.5 (18),the interstitial volume would be 67% of the 210 pL or 140 pL. Since the protein is confined to the interstitial space, this means the 50-pL protein sample would be diluted roughly by a factor of 3. In the nonlinear regression of eq 1, two of the coefficients generated equate to the products of nlP and n2P. The n values are then calculated from the protein concentration. If the protein concentration was diluted by a factor of 3, then the n values would be one-third of that expected, based on the sample protein concentration. It is well established for warfarin-HSA primary binding that nl equals one (5-9). The modified Hummel-Dreyer method produces the same results (Table I), thus the on-column dilution of protein is of no consequence. The above consluion is reasonable, if one recognizes that the Hummel-Dreyer method measures the bound drug concentration directly. The protein throughout the sample band may differ in concentration, but each segment is in instantaneous equilibrium with free concentration of drug in the mobile phase. Since the bound concentration is linearly related to the protein concentration (eq l),the total concentration of drug bound in the band is the linear sum of all
infinitesimally small band segments. The concentration of drug bound is measured from the integrated area of the vacancy, which truly reflects the concentration of drug bound to protein at equilibrium. Therefore, it is inconsequential how much the protein is diluted on column with the HummelDreyer method. One must remember, however, that the protein peak must be baseline resolved from the vacancy and the vacancy should be symmetrical, as an indication good exchange kinetics. Comparison of Binding Parameters from Different Studies. As noted by Connors (12),considerable variation between results determined by different techniques is often observed for drug-protein binding. I t is not always easy to identify the reasons why data from different methods disagree. Larger than normal variations can be noted between the binding parameters reported for warfarin-HSA in phosphate buffer (Table 11) (5-9). The binding parameters for the warfarin-HSA primary site determined under similar conditions are in reasonable agreement. On the other hand, the warfarin-HSA parameters for the secondary binding site vary significantly. It is tempting to ascribe variations in secondary binding parameters between studies to differences in drug concentration range; however, it is insufficient to argue that differences in concentration range alone account for these discrepancies. The n2 of 3.8 accompanied by a low K 2value of ref 6 appears to result from measurement error and curve distortion from nonlinear curve fitting of Scatchard transformations. At higher drug concentrations it becomes more difficult to discriminate the change in free drug concentration as protein approaches saturation. This uncertainty can be compounded by saturation of ultraviolet-visible detectors. The Scatchard transformation places this experimental error in the abscissa and significantly distorts the shape of the curve in the region of the secondary binding, thus overestimating n2 and underestimating K1. The high n value of 10 from ref 7 using a three class model is due tq the intentional use of HSA containing fatty acids. It is well accepted that the presence of fatty acids increases nonspecific binding; thus most studies employ fatty acid free protein. Also, temperature has been shown to have an effect on warfarin HSA binding, exhibited by an increase in the binding constants with decrease in temperature (5). Although the Kl of ref 8 appears inordinately high, the value to some degree reflects the lower temperature of the study. Although it is often impossible to compare results between studies because of differences in experimental conditions and methodology, the results of the modified Hummel-Dreyer method of this study strongly support the findings of several other investigators (Table 11)that warfarin undergoes specific binding to fatty acid free HSA at primary and secondary sites
ANALYTICAL CHEMISTRY, VOL. 62, NO. 19, OCTOBER 1, 1990
2121
1000 900
2
800
m
400
B
W
G 0 . .
.
.
.
.
.
.
.
.
20
10 Tlms
.
.
. 30
- - - - - - - -40
-
(mln)
Flgure 6. Chromatographic profile obtained from frontal analysis on
a 5-cm ISRP column by the injection of 6 mL of solution containing 56.6 pM HSA and 81.2 pM warfarin in 0.067 M phosphate buffer (pH 7.4): mobile phase, 0.067 phosphate buffer: flow rate, 0.3 mL/min; temperature, 25 "C. Plateaus are (A) a,(B) 0, and (0)y. Free drug concentration measured from y plateau.
with 1 and 2 mol of drug bound per mole of protein, respectively. The binding constants determined in this study over a drug concentration range up to 80% protein saturation are just slightly higher than those determined by fluorescent titration (9) a t a comparable pH and temperature, but measured a t a higher ionic strength (Table 11). Frontal Analysis. The ISRP columns have been used to study the binding of warfarin to bovine serum albumin (BSA) (26) and HSA (27) by frontal analysis. The primary binding site parameters for warfarin-HSA were nl = 1.24 and K1 = 1.96 X lo5M-' at 37 "C by high-performance frontal analysis (HPFA) using an ISRP column (27). One might ask whether the Hummel-Dreyer method and frontal analysis yield comparable responses under the conditions of this study. One experiment was conducted to address this question. A 6.0-mL sample containing 56.6 pM HSA and 81.2 pM warfarin was pumped at a flow rate of 0.3 mL/min through an ISRP column (5 cm X 4.6 mm i.d.), equilibrated with a mobile phase consisting of only 0.067 M, pH 7.4 phosphate buffer. The resulting frontal profile exhibits the classical a, p, and y migrational zones (Figure 6). The absorbance response in the flat region of the last plateau was found to correspond to a free warfarin concentration of 12.0 pM; thus the measured amount bound was 69.2 pM. By use of the mean binding parameters determined from the Hummel-Dreyer method (Table I) and the free warfarin concentration of 12.0 pM measured by frontal analysis, the predicted drug bound would be 70.6 pM. This represents a 2% difference from the drug bound value measured by frontal analysis. This suggests that the methods yield comparable results; however, further work will be required to confirm this hypothesis. Of course, which method is best for a given study depends on the practical constrains of each circumstance. Frontal analysis requires the preparation of many solutions and consumes more protein, while the modified Hummel-Dreyer method requires only one protein solution, but consumes more drug. However, the use of short ISRP columns minimizes the consumption of binding ligand. For this study, 15 mg of warfarin was used per set of experiments. Of course, the amount of drug consumed could be further reduced by the use of microbore columns. Also, from a fundamental standpoint, the Hummel-Dreyer method has the unique advantage of controlling the free drug concentration as the true independent variable. Resolution of Drug from Displacer in HummelDreyer ISRP Experiment. Often it is desirable to study the binding of a substance in the presence of a displacer. If a detection method can discriminate the displacer from the primary component, then binding studies can be conducted easily. This has been demonstrated by the discrimination of warfarin from furosemide by UV detection a t different
L
0
"
v
-2004
.
.
.
B
.
. 1
.
.
Tlme
.
.
B
. 2
.
. 3
(mln)
Figure 7. Hummel-Dreyer chromatographic profile obtained with a 5 t m ISRP column from the injection of 50 pL of 54.8 pM HSA (trace B) into a mobile phase containing 40.5 pM warfarln (W) and 60.6 pM ibuprofen ( I ) in 0.067 M phosphate buffer: flow rate, 2.0 mL/min; temperature, 25 "C; detection, 214 nm with 4 nm bandwidth. Trace A is from injection of 50 pL of 0.067 M phosphate buffer blank.
wavelengths in conventional gel-fitration binding study with HSA (6). If such discrimination cannot be made, then the species must be separated. In selected cases, such as tryptophan derivatives binding with BSA, secondary retention mechanisms on conventional size-exclusion columns have resolved vacancies of competing substances with classical Hummel-Dreyer experiments (39). The same type of resolution of primary component and displacer can readily be achieved with high efficiency during Hummel-Dreyer experiments using the ISRP columns, because of the internal partitioning phase on the packing. This is illustrated by the separation of ibuprofen and warfarin vacancies (Figure 7), produced from the injection of HSA into a mobile phase containing both compounds. Although the elution profiles were not optimized and the protein binding displacement was not treated in this study, a variety of experimental approaches could be conceived where the resolving power of the ISRP columns could be exploited with the modified Hummel-Dreyer method.
CONCLUSION The modified Hummel-Dreyer method, using an internal surface reversed-phase HPLC column and an automated system, provides a rigorous, efficient means of determining accurate protein binding parameters with high precision for protein-ligand binding equilibria occurring concomitantly a t more than one type of binding site. The decreased analysis time and increased precision compared to conventional methods suggest that the technique should prove useful in measuring changes in binding parameters as a function of experimental conditions. Such studies could involve changes in temperature to acquire thermodynamic constants, control of pH and ionic strength to understand changes in protein conformation, and the use of optically pure compounds to understand chiral binding discrimination. In addition, the internal surface reversed-phase HPLC columns enable the use of shorter columns, the control of drug elution, the resolution of protein binding displacers, and the maintenance of high performance after many sample injections of HSA. ACKNOWLEDGMENT The authors wish to thank Ferenc J. Kezdy for providing the nonlinear least-squares program and for engaging in many helpful discussions concerning protein binding. Appreciation is also expressed to John J. Dougherty and Lynn M. Snyder for conducting the amino acid analyses. LITERATURE CITED (1) Progress in Drug Protein Binding; Rietbrock, N., Woodcock, B. G., Eds.; Methods in Clinical Pharmacology No. 2; Friedr. Vleweg 8 Sohn: Braunschwelg, FRG, 1981; pp 5-29. (2) Kwong, T. C. Ciin. Cbim. Acta 1985, 151, 193-216.
2122
Anal. Chem. 1990, 62, 2122-2130 Meyer. M. C.; Guttman, D. E. J. fharm. Scl. 1988, 57(6),895-918. Banker, G. S.;Rhodes. C. T. Modern fharmaceufks; Marcel Dekker: New York, 1979;pp 202-207. Oester, Y. T.; Keresztes-Nagy, S.;Mais, R. F.; Beckkel, J.; Zaroslinski, J. F. J. fharm. Scl. 1978, 65(11),1673-1677. Sebille. 8.; Thaud, N.; Tillement, P. J. Chromafogr. 1978, 167,
159- 170. Wilting, J.; VanderGiesen, W. F.; Janssen, L. H. M.; Weideman, M. M.; Otagiri, M.; Perrin, J. H. J. Biol. Chem. 1980, 255 (7),3032-3037. Veronlch, K.: White, G.; Kapoor, A. J. fharm. Sci. 1979, 68 (12),
15 15- 1518. Sundlow, G.; Brikett, D. J.; Wade, N. N. Clln. Exp. fharmacol. Physiol. 1975, 2 , 129-140. Holford, N. H. G. Clin. fharmacoklnet. 1988, 7 7 , 483-504. Chu, Y.-Q.; Wainer, I. W. Pharm. Res. 1988, 5(10),680-683. Connors. K. A. Blndlng Consfanfs: The Measurements of Molecular Complex Stability; John Wiiey 8. Sons: New York, 1987;p 315. Chignell, C. F. Drug Fate and Metabolism, Methods and Techniques; Marcel Dekker: New York, 1977;Chapter 5. Cooper, P. F.; Wood, G. C. J. Pharm. fharmacol. 1988, 20, Suppi.
150s-156s. Hummel, J. P.; Dreyer, W. J. Biochim. Biophys. Acta 1982, 6 3 ,
530-532. Johnson, M. L.; Frasier, S. G. I n Methods in Enzymology; Academic: New York, 1985,Vol. 117,Chapter 16, p 325. Pinkerton. T. C.; Hagestam, I. H. Anal. Chem. 1985, 5 7 , 1757-1763. Cook, S. E.; Pinkerton, T. C. J. Chromafogr. 1988, 368, 233-248. Pinkerton, T. C.; Miller, T. D.; Cook, S.E.; Perry, J. A.; Rateike, J. D.; Szczerba. T. J. BloChromafogr. 1988, 7 (2),96-105. Dawson, C. M.; Wang. T. W.; Rainbow, S.J.; Tickner, T. R. Ann. Clln. Blochem. 1988, 25,661-667. Toshimitsu, N.; Takeda, N.; Tatematsu, A.; Maeda, K. Clin. Chem. 1988, 34 ( I l),2264-2267.
(22) Atherton, N. D. Clin. Chem. 1988, 35(6),975-978. (23) Tamlsier-Karolak, L.; Farinotti, R.; Bossant, M. J.; Dauphin, A. J. fharm. Clin. 1988, 7(4),543-556. (24) Oshima, T.; Johno, I.; Hasegawa, T.; Kitazawa, S. J. Llq. Chromatogr. 1988, 1 1 (16),3457-3470. (25) Nakagawa, T.; Shibukawa, A.; Shimono, N.; Kawashlma, T.; Tanaka, H. J. Chromafogr. 1987, 420, 297-311. (26) Shlbukawa, A.: Nakagawa, T.; Nishimura, N.; Miyake, M.; Tanaka, H. Chem. fharm. Bull. 1989, 37(3). 702-706. (27)Shlbukawa, A.; Nago, M.; Kuroda, Y.; Nakagawa, T. Anal. Chem. 1990, 62, 712-716. (28) Pinkerton, T. c.; Miller, T. D.; Janls. L. J. Anal. Chem. 1989, 67, 1171-1174. (29) Shibukawa, A.; Nakagawa, T.; Miyake, M.; Tanaka, H. Chem. fharm, Bull. 1988, 36 (5),1930-1933. (30) Shibukawa, A.; Nakagawa, T.; Mlyake, M.; Nlshimura, N.; Tanaka, H. Chem. fharm. Bull. 1989, 37(5), 1311-1115. (31) Ohshima, T.; Johno, I.: Hasegawa, T.; Kitazawa, S. J . fharm. Sci. 1990, 79 (l),77-81. (32) Yamaoka, K.; Tanlgawara, Y.; Nakagawa, T.; Uno, T. J. fharm. Dyn. 1981, 4 , 879-885. (33) Soites, L.; Bree, F.; Sebille, 6.; Tillement, J. P.; Durisva, M.; Trnovec, T. Biochem. fharm. 1985, 34 (2),4331-4334. (34) Sun. S. F.; Kuo. S . W.; Nash, R. A. J . Chromatcgr. 1984, 288, 377-388. (35) Sun, S. F.; Wong, F. Chromafcgraph/a 1985, 20 (E),495-499. (36) Stevens, F. J. Blophys. J . 1989, 55, 1155-1167. (37) Rosenthal, H. Anal. Biochem. 1987, 2 0 , 525-532. (38) Zlerler, K. Trends Biochem. Sci. 1989, 14, 314-317. (39) Fairclough, G. F.; Fruton. J. S . Biochemistry 1988, 5. 673-683.
RECEIVED for review March 6, 1990. Accepted July 11,1990.
Molecular Ion Imaging and Dynamic Secondary Ion Mass Spectrometry of Organic Compounds Greg Gillen* a n d David S. Simons
Center for Analytical Chemistry, National Institute of Standards and Technology, Gaithersburg, Maryland 20899
Peter Williams Department of Chemistry, Arizona State University, Tempe, Arizona 85287
An ion microscope equipped with a reslstive anode encoder imaglng system has been used to acquire molecular secondary ion images, with lateral resolutions on the order of 1 pm, from several quaternary ammonium salts, an amino acid, and a polynuclear aromatlc hydrocarbon which were deposHed onto copper transmlsaion electron microscope grids. Ail images were generated by using the secondary ion signal of the parent molecular species. The variation of parent and fragment moiecuiar ion signals with primary ion dose indicates that, for many bulk organic compounds, bombardment-induced fragmentation of parent molecules saturates at prlmary ion doses of (1-8) X 10'' ions/cm2. Subsequent ion impacts cause lmle further accumulation of damage in the sample, and intact parent molecular ions are sputtered even after prolonged ion bombardment (i.e. primary ion doses >1 X 10'' ions/cm2). This saturation process allows molecular images to be obtained at high primary ion doses and allows depth profiles to be obtained from slmple mdecuiar solld/metai test structures.
INTRODUCTION In recent years there has been intense interest, both from analytical and fundamental standpoints, in the application of secondary ion mass spectrometry (SIMS) to the analysis
of organic molecules. This interest followed the demonstrations by Benninghoven, Cooks, and their co-workers in 1976-1977 that kiloelectronvolt ion bombardment could eject intact parent molecular ions from compounds deposited onto metal surfaces (1,2). This technique has proven particularly useful for analytical and structural studies of involatile, thermally labile, organic compounds that were difficult to analyze by electron impact mass spectrometry. With the secondary ion imaging capabilities found in many secondary ion mass spectrometers, it should be possible to generate molecular secondary ion images using the same techniques developed for imaging elemental species (3-5). Imaging of molecular species in biological materials might have a major impact on a wide variety of studies of molecular processes in biological systems where previously only gross assays were possible. Progress toward this goal has recently been reported (6-14). A major difficulty in generating molecular secondary ion images using conventional (high ion dose) SIMS is that each primary ion, while desorbing several atoms or molecules from the surface layer, penetrates below the surface undergoing numerous collisions that can rupture bonds in many more molecules. Intuitively, it might be expected that after the outer few monolayers of the sample have been removed by sputtering, few intact molecules would remain below the sputtered surface for a depth on the order of the primary ion
0003-2700/90/0362-2122$02.50/00 1990 American Chemical Society
